Commit | Line | Data |
---|---|---|
597d0cae | 1 | /* |
ef0c2bb0 | 2 | * Copyright (C) 2006-2007 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 DT |
26 | |
27 | static const char *name_prefix="dlm"; | |
28 | static struct miscdevice ctl_device; | |
00977a59 | 29 | static const struct file_operations device_fops; |
597d0cae DT |
30 | |
31 | #ifdef CONFIG_COMPAT | |
32 | ||
33 | struct dlm_lock_params32 { | |
34 | __u8 mode; | |
35 | __u8 namelen; | |
d7db923e DT |
36 | __u16 unused; |
37 | __u32 flags; | |
597d0cae DT |
38 | __u32 lkid; |
39 | __u32 parent; | |
d7db923e DT |
40 | __u64 xid; |
41 | __u64 timeout; | |
597d0cae DT |
42 | __u32 castparam; |
43 | __u32 castaddr; | |
44 | __u32 bastparam; | |
45 | __u32 bastaddr; | |
46 | __u32 lksb; | |
597d0cae DT |
47 | char lvb[DLM_USER_LVB_LEN]; |
48 | char name[0]; | |
49 | }; | |
50 | ||
51 | struct dlm_write_request32 { | |
52 | __u32 version[3]; | |
53 | __u8 cmd; | |
54 | __u8 is64bit; | |
55 | __u8 unused[2]; | |
56 | ||
57 | union { | |
58 | struct dlm_lock_params32 lock; | |
59 | struct dlm_lspace_params lspace; | |
72c2be77 | 60 | struct dlm_purge_params purge; |
597d0cae DT |
61 | } i; |
62 | }; | |
63 | ||
64 | struct dlm_lksb32 { | |
65 | __u32 sb_status; | |
66 | __u32 sb_lkid; | |
67 | __u8 sb_flags; | |
68 | __u32 sb_lvbptr; | |
69 | }; | |
70 | ||
71 | struct dlm_lock_result32 { | |
d7db923e | 72 | __u32 version[3]; |
597d0cae DT |
73 | __u32 length; |
74 | __u32 user_astaddr; | |
75 | __u32 user_astparam; | |
76 | __u32 user_lksb; | |
77 | struct dlm_lksb32 lksb; | |
78 | __u8 bast_mode; | |
79 | __u8 unused[3]; | |
80 | /* Offsets may be zero if no data is present */ | |
81 | __u32 lvb_offset; | |
82 | }; | |
83 | ||
84 | static void compat_input(struct dlm_write_request *kb, | |
85 | struct dlm_write_request32 *kb32) | |
86 | { | |
87 | kb->version[0] = kb32->version[0]; | |
88 | kb->version[1] = kb32->version[1]; | |
89 | kb->version[2] = kb32->version[2]; | |
90 | ||
91 | kb->cmd = kb32->cmd; | |
92 | kb->is64bit = kb32->is64bit; | |
93 | if (kb->cmd == DLM_USER_CREATE_LOCKSPACE || | |
94 | kb->cmd == DLM_USER_REMOVE_LOCKSPACE) { | |
95 | kb->i.lspace.flags = kb32->i.lspace.flags; | |
96 | kb->i.lspace.minor = kb32->i.lspace.minor; | |
97 | strcpy(kb->i.lspace.name, kb32->i.lspace.name); | |
72c2be77 DT |
98 | } else if (kb->cmd == DLM_USER_PURGE) { |
99 | kb->i.purge.nodeid = kb32->i.purge.nodeid; | |
100 | kb->i.purge.pid = kb32->i.purge.pid; | |
597d0cae DT |
101 | } else { |
102 | kb->i.lock.mode = kb32->i.lock.mode; | |
103 | kb->i.lock.namelen = kb32->i.lock.namelen; | |
104 | kb->i.lock.flags = kb32->i.lock.flags; | |
105 | kb->i.lock.lkid = kb32->i.lock.lkid; | |
106 | kb->i.lock.parent = kb32->i.lock.parent; | |
d7db923e DT |
107 | kb->i.lock.xid = kb32->i.lock.xid; |
108 | kb->i.lock.timeout = kb32->i.lock.timeout; | |
597d0cae DT |
109 | kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam; |
110 | kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr; | |
111 | kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam; | |
112 | kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr; | |
113 | kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb; | |
114 | memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN); | |
115 | memcpy(kb->i.lock.name, kb32->i.lock.name, kb->i.lock.namelen); | |
116 | } | |
117 | } | |
118 | ||
119 | static void compat_output(struct dlm_lock_result *res, | |
120 | struct dlm_lock_result32 *res32) | |
121 | { | |
d7db923e DT |
122 | res32->version[0] = res->version[0]; |
123 | res32->version[1] = res->version[1]; | |
124 | res32->version[2] = res->version[2]; | |
125 | ||
597d0cae DT |
126 | res32->user_astaddr = (__u32)(long)res->user_astaddr; |
127 | res32->user_astparam = (__u32)(long)res->user_astparam; | |
128 | res32->user_lksb = (__u32)(long)res->user_lksb; | |
129 | res32->bast_mode = res->bast_mode; | |
130 | ||
131 | res32->lvb_offset = res->lvb_offset; | |
132 | res32->length = res->length; | |
133 | ||
134 | res32->lksb.sb_status = res->lksb.sb_status; | |
135 | res32->lksb.sb_flags = res->lksb.sb_flags; | |
136 | res32->lksb.sb_lkid = res->lksb.sb_lkid; | |
137 | res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr; | |
138 | } | |
139 | #endif | |
140 | ||
84d8cd69 DT |
141 | /* Figure out if this lock is at the end of its life and no longer |
142 | available for the application to use. The lkb still exists until | |
143 | the final ast is read. A lock becomes EOL in three situations: | |
144 | 1. a noqueue request fails with EAGAIN | |
145 | 2. an unlock completes with EUNLOCK | |
146 | 3. a cancel of a waiting request completes with ECANCEL/EDEADLK | |
147 | An EOL lock needs to be removed from the process's list of locks. | |
148 | And we can't allow any new operation on an EOL lock. This is | |
149 | not related to the lifetime of the lkb struct which is managed | |
150 | entirely by refcount. */ | |
151 | ||
152 | static int lkb_is_endoflife(struct dlm_lkb *lkb, int sb_status, int type) | |
153 | { | |
154 | switch (sb_status) { | |
155 | case -DLM_EUNLOCK: | |
156 | return 1; | |
157 | case -DLM_ECANCEL: | |
158 | case -ETIMEDOUT: | |
159 | if (lkb->lkb_grmode == DLM_LOCK_IV) | |
160 | return 1; | |
161 | break; | |
162 | case -EAGAIN: | |
163 | if (type == AST_COMP && lkb->lkb_grmode == DLM_LOCK_IV) | |
164 | return 1; | |
165 | break; | |
166 | } | |
167 | return 0; | |
168 | } | |
169 | ||
ef0c2bb0 DT |
170 | /* we could possibly check if the cancel of an orphan has resulted in the lkb |
171 | being removed and then remove that lkb from the orphans list and free it */ | |
597d0cae DT |
172 | |
173 | void dlm_user_add_ast(struct dlm_lkb *lkb, int type) | |
174 | { | |
175 | struct dlm_ls *ls; | |
176 | struct dlm_user_args *ua; | |
177 | struct dlm_user_proc *proc; | |
ef0c2bb0 | 178 | int eol = 0, ast_type; |
597d0cae | 179 | |
ef0c2bb0 | 180 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 181 | return; |
597d0cae DT |
182 | |
183 | ls = lkb->lkb_resource->res_ls; | |
184 | mutex_lock(&ls->ls_clear_proc_locks); | |
185 | ||
186 | /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast | |
187 | can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed | |
ef0c2bb0 DT |
188 | lkb->ua so we can't try to use it. This second check is necessary |
189 | for cases where a completion ast is received for an operation that | |
190 | began before clear_proc_locks did its cancel/unlock. */ | |
597d0cae | 191 | |
ef0c2bb0 | 192 | if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD)) |
597d0cae | 193 | goto out; |
597d0cae DT |
194 | |
195 | DLM_ASSERT(lkb->lkb_astparam, dlm_print_lkb(lkb);); | |
196 | ua = (struct dlm_user_args *)lkb->lkb_astparam; | |
197 | proc = ua->proc; | |
198 | ||
199 | if (type == AST_BAST && ua->bastaddr == NULL) | |
200 | goto out; | |
201 | ||
202 | spin_lock(&proc->asts_spin); | |
ef0c2bb0 DT |
203 | |
204 | ast_type = lkb->lkb_ast_type; | |
205 | lkb->lkb_ast_type |= type; | |
206 | ||
207 | if (!ast_type) { | |
597d0cae DT |
208 | kref_get(&lkb->lkb_ref); |
209 | list_add_tail(&lkb->lkb_astqueue, &proc->asts); | |
597d0cae DT |
210 | wake_up_interruptible(&proc->wait); |
211 | } | |
ef0c2bb0 DT |
212 | if (type == AST_COMP && (ast_type & AST_COMP)) |
213 | log_debug(ls, "ast overlap %x status %x %x", | |
214 | lkb->lkb_id, ua->lksb.sb_status, lkb->lkb_flags); | |
215 | ||
84d8cd69 | 216 | eol = lkb_is_endoflife(lkb, ua->lksb.sb_status, type); |
ef0c2bb0 DT |
217 | if (eol) { |
218 | lkb->lkb_ast_type &= ~AST_BAST; | |
219 | lkb->lkb_flags |= DLM_IFL_ENDOFLIFE; | |
220 | } | |
a1bc86e6 | 221 | |
597d0cae DT |
222 | /* We want to copy the lvb to userspace when the completion |
223 | ast is read if the status is 0, the lock has an lvb and | |
224 | lvb_ops says we should. We could probably have set_lvb_lock() | |
225 | set update_user_lvb instead and not need old_mode */ | |
226 | ||
227 | if ((lkb->lkb_ast_type & AST_COMP) && | |
228 | (lkb->lkb_lksb->sb_status == 0) && | |
229 | lkb->lkb_lksb->sb_lvbptr && | |
230 | dlm_lvb_operations[ua->old_mode + 1][lkb->lkb_grmode + 1]) | |
231 | ua->update_user_lvb = 1; | |
232 | else | |
233 | ua->update_user_lvb = 0; | |
234 | ||
235 | spin_unlock(&proc->asts_spin); | |
34e22bed | 236 | |
ef0c2bb0 | 237 | if (eol) { |
34e22bed | 238 | spin_lock(&ua->proc->locks_spin); |
ef0c2bb0 DT |
239 | if (!list_empty(&lkb->lkb_ownqueue)) { |
240 | list_del_init(&lkb->lkb_ownqueue); | |
241 | dlm_put_lkb(lkb); | |
242 | } | |
34e22bed | 243 | spin_unlock(&ua->proc->locks_spin); |
34e22bed | 244 | } |
597d0cae DT |
245 | out: |
246 | mutex_unlock(&ls->ls_clear_proc_locks); | |
247 | } | |
248 | ||
249 | static int device_user_lock(struct dlm_user_proc *proc, | |
250 | struct dlm_lock_params *params) | |
251 | { | |
252 | struct dlm_ls *ls; | |
253 | struct dlm_user_args *ua; | |
254 | int error = -ENOMEM; | |
255 | ||
256 | ls = dlm_find_lockspace_local(proc->lockspace); | |
257 | if (!ls) | |
258 | return -ENOENT; | |
259 | ||
260 | if (!params->castaddr || !params->lksb) { | |
261 | error = -EINVAL; | |
262 | goto out; | |
263 | } | |
264 | ||
265 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); | |
266 | if (!ua) | |
267 | goto out; | |
268 | ua->proc = proc; | |
269 | ua->user_lksb = params->lksb; | |
270 | ua->castparam = params->castparam; | |
271 | ua->castaddr = params->castaddr; | |
272 | ua->bastparam = params->bastparam; | |
273 | ua->bastaddr = params->bastaddr; | |
d7db923e | 274 | ua->xid = params->xid; |
597d0cae DT |
275 | |
276 | if (params->flags & DLM_LKF_CONVERT) | |
277 | error = dlm_user_convert(ls, ua, | |
278 | params->mode, params->flags, | |
d7db923e DT |
279 | params->lkid, params->lvb, |
280 | (unsigned long) params->timeout); | |
597d0cae DT |
281 | else { |
282 | error = dlm_user_request(ls, ua, | |
283 | params->mode, params->flags, | |
284 | params->name, params->namelen, | |
d7db923e | 285 | (unsigned long) params->timeout); |
597d0cae DT |
286 | if (!error) |
287 | error = ua->lksb.sb_lkid; | |
288 | } | |
289 | out: | |
290 | dlm_put_lockspace(ls); | |
291 | return error; | |
292 | } | |
293 | ||
294 | static int device_user_unlock(struct dlm_user_proc *proc, | |
295 | struct dlm_lock_params *params) | |
296 | { | |
297 | struct dlm_ls *ls; | |
298 | struct dlm_user_args *ua; | |
299 | int error = -ENOMEM; | |
300 | ||
301 | ls = dlm_find_lockspace_local(proc->lockspace); | |
302 | if (!ls) | |
303 | return -ENOENT; | |
304 | ||
305 | ua = kzalloc(sizeof(struct dlm_user_args), GFP_KERNEL); | |
306 | if (!ua) | |
307 | goto out; | |
308 | ua->proc = proc; | |
309 | ua->user_lksb = params->lksb; | |
310 | ua->castparam = params->castparam; | |
311 | ua->castaddr = params->castaddr; | |
312 | ||
313 | if (params->flags & DLM_LKF_CANCEL) | |
314 | error = dlm_user_cancel(ls, ua, params->flags, params->lkid); | |
315 | else | |
316 | error = dlm_user_unlock(ls, ua, params->flags, params->lkid, | |
317 | params->lvb); | |
318 | out: | |
319 | dlm_put_lockspace(ls); | |
320 | return error; | |
321 | } | |
322 | ||
254da030 PC |
323 | static int create_misc_device(struct dlm_ls *ls, char *name) |
324 | { | |
325 | int error, len; | |
326 | ||
327 | error = -ENOMEM; | |
328 | len = strlen(name) + strlen(name_prefix) + 2; | |
329 | ls->ls_device.name = kzalloc(len, GFP_KERNEL); | |
330 | if (!ls->ls_device.name) | |
331 | goto fail; | |
332 | ||
333 | snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix, | |
334 | name); | |
335 | ls->ls_device.fops = &device_fops; | |
336 | ls->ls_device.minor = MISC_DYNAMIC_MINOR; | |
337 | ||
338 | error = misc_register(&ls->ls_device); | |
339 | if (error) { | |
340 | kfree(ls->ls_device.name); | |
341 | } | |
342 | fail: | |
343 | return error; | |
344 | } | |
345 | ||
72c2be77 DT |
346 | static int device_user_purge(struct dlm_user_proc *proc, |
347 | struct dlm_purge_params *params) | |
348 | { | |
349 | struct dlm_ls *ls; | |
350 | int error; | |
351 | ||
352 | ls = dlm_find_lockspace_local(proc->lockspace); | |
353 | if (!ls) | |
354 | return -ENOENT; | |
355 | ||
356 | error = dlm_user_purge(ls, proc, params->nodeid, params->pid); | |
357 | ||
358 | dlm_put_lockspace(ls); | |
359 | return error; | |
360 | } | |
361 | ||
597d0cae DT |
362 | static int device_create_lockspace(struct dlm_lspace_params *params) |
363 | { | |
364 | dlm_lockspace_t *lockspace; | |
365 | struct dlm_ls *ls; | |
254da030 | 366 | int error; |
597d0cae DT |
367 | |
368 | if (!capable(CAP_SYS_ADMIN)) | |
369 | return -EPERM; | |
370 | ||
371 | error = dlm_new_lockspace(params->name, strlen(params->name), | |
3ae1acf9 | 372 | &lockspace, params->flags, DLM_USER_LVB_LEN); |
597d0cae DT |
373 | if (error) |
374 | return error; | |
375 | ||
376 | ls = dlm_find_lockspace_local(lockspace); | |
377 | if (!ls) | |
378 | return -ENOENT; | |
379 | ||
254da030 | 380 | error = create_misc_device(ls, params->name); |
597d0cae | 381 | dlm_put_lockspace(ls); |
597d0cae | 382 | |
254da030 PC |
383 | if (error) |
384 | dlm_release_lockspace(lockspace, 0); | |
385 | else | |
386 | error = ls->ls_device.minor; | |
387 | ||
597d0cae DT |
388 | return error; |
389 | } | |
390 | ||
391 | static int device_remove_lockspace(struct dlm_lspace_params *params) | |
392 | { | |
393 | dlm_lockspace_t *lockspace; | |
394 | struct dlm_ls *ls; | |
c6e6f0ba | 395 | int error, force = 0; |
597d0cae DT |
396 | |
397 | if (!capable(CAP_SYS_ADMIN)) | |
398 | return -EPERM; | |
399 | ||
400 | ls = dlm_find_lockspace_device(params->minor); | |
401 | if (!ls) | |
402 | return -ENOENT; | |
403 | ||
254da030 PC |
404 | /* Deregister the misc device first, so we don't have |
405 | * a device that's not attached to a lockspace. If | |
406 | * dlm_release_lockspace fails then we can recreate it | |
407 | */ | |
597d0cae DT |
408 | error = misc_deregister(&ls->ls_device); |
409 | if (error) { | |
410 | dlm_put_lockspace(ls); | |
411 | goto out; | |
412 | } | |
413 | kfree(ls->ls_device.name); | |
414 | ||
c6e6f0ba DT |
415 | if (params->flags & DLM_USER_LSFLG_FORCEFREE) |
416 | force = 2; | |
417 | ||
597d0cae DT |
418 | lockspace = ls->ls_local_handle; |
419 | ||
420 | /* dlm_release_lockspace waits for references to go to zero, | |
421 | so all processes will need to close their device for the ls | |
422 | before the release will procede */ | |
423 | ||
424 | dlm_put_lockspace(ls); | |
c6e6f0ba | 425 | error = dlm_release_lockspace(lockspace, force); |
254da030 PC |
426 | if (error) |
427 | create_misc_device(ls, ls->ls_name); | |
c6e6f0ba | 428 | out: |
597d0cae DT |
429 | return error; |
430 | } | |
431 | ||
432 | /* Check the user's version matches ours */ | |
433 | static int check_version(struct dlm_write_request *req) | |
434 | { | |
435 | if (req->version[0] != DLM_DEVICE_VERSION_MAJOR || | |
436 | (req->version[0] == DLM_DEVICE_VERSION_MAJOR && | |
437 | req->version[1] > DLM_DEVICE_VERSION_MINOR)) { | |
438 | ||
439 | printk(KERN_DEBUG "dlm: process %s (%d) version mismatch " | |
440 | "user (%d.%d.%d) kernel (%d.%d.%d)\n", | |
441 | current->comm, | |
442 | current->pid, | |
443 | req->version[0], | |
444 | req->version[1], | |
445 | req->version[2], | |
446 | DLM_DEVICE_VERSION_MAJOR, | |
447 | DLM_DEVICE_VERSION_MINOR, | |
448 | DLM_DEVICE_VERSION_PATCH); | |
449 | return -EINVAL; | |
450 | } | |
451 | return 0; | |
452 | } | |
453 | ||
454 | /* | |
455 | * device_write | |
456 | * | |
457 | * device_user_lock | |
458 | * dlm_user_request -> request_lock | |
459 | * dlm_user_convert -> convert_lock | |
460 | * | |
461 | * device_user_unlock | |
462 | * dlm_user_unlock -> unlock_lock | |
463 | * dlm_user_cancel -> cancel_lock | |
464 | * | |
465 | * device_create_lockspace | |
466 | * dlm_new_lockspace | |
467 | * | |
468 | * device_remove_lockspace | |
469 | * dlm_release_lockspace | |
470 | */ | |
471 | ||
472 | /* a write to a lockspace device is a lock or unlock request, a write | |
473 | to the control device is to create/remove a lockspace */ | |
474 | ||
475 | static ssize_t device_write(struct file *file, const char __user *buf, | |
476 | size_t count, loff_t *ppos) | |
477 | { | |
478 | struct dlm_user_proc *proc = file->private_data; | |
479 | struct dlm_write_request *kbuf; | |
480 | sigset_t tmpsig, allsigs; | |
481 | int error; | |
482 | ||
483 | #ifdef CONFIG_COMPAT | |
484 | if (count < sizeof(struct dlm_write_request32)) | |
485 | #else | |
486 | if (count < sizeof(struct dlm_write_request)) | |
487 | #endif | |
488 | return -EINVAL; | |
489 | ||
490 | kbuf = kmalloc(count, GFP_KERNEL); | |
491 | if (!kbuf) | |
492 | return -ENOMEM; | |
493 | ||
494 | if (copy_from_user(kbuf, buf, count)) { | |
495 | error = -EFAULT; | |
496 | goto out_free; | |
497 | } | |
498 | ||
499 | if (check_version(kbuf)) { | |
500 | error = -EBADE; | |
501 | goto out_free; | |
502 | } | |
503 | ||
504 | #ifdef CONFIG_COMPAT | |
505 | if (!kbuf->is64bit) { | |
506 | struct dlm_write_request32 *k32buf; | |
507 | k32buf = (struct dlm_write_request32 *)kbuf; | |
508 | kbuf = kmalloc(count + (sizeof(struct dlm_write_request) - | |
509 | sizeof(struct dlm_write_request32)), GFP_KERNEL); | |
510 | if (!kbuf) | |
511 | return -ENOMEM; | |
512 | ||
513 | if (proc) | |
514 | set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags); | |
515 | compat_input(kbuf, k32buf); | |
516 | kfree(k32buf); | |
517 | } | |
518 | #endif | |
519 | ||
520 | /* do we really need this? can a write happen after a close? */ | |
521 | if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) && | |
522 | test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) | |
523 | return -EINVAL; | |
524 | ||
525 | sigfillset(&allsigs); | |
526 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
527 | ||
528 | error = -EINVAL; | |
529 | ||
530 | switch (kbuf->cmd) | |
531 | { | |
532 | case DLM_USER_LOCK: | |
533 | if (!proc) { | |
534 | log_print("no locking on control device"); | |
535 | goto out_sig; | |
536 | } | |
537 | error = device_user_lock(proc, &kbuf->i.lock); | |
538 | break; | |
539 | ||
540 | case DLM_USER_UNLOCK: | |
541 | if (!proc) { | |
542 | log_print("no locking on control device"); | |
543 | goto out_sig; | |
544 | } | |
545 | error = device_user_unlock(proc, &kbuf->i.lock); | |
546 | break; | |
547 | ||
548 | case DLM_USER_CREATE_LOCKSPACE: | |
549 | if (proc) { | |
550 | log_print("create/remove only on control device"); | |
551 | goto out_sig; | |
552 | } | |
553 | error = device_create_lockspace(&kbuf->i.lspace); | |
554 | break; | |
555 | ||
556 | case DLM_USER_REMOVE_LOCKSPACE: | |
557 | if (proc) { | |
558 | log_print("create/remove only on control device"); | |
559 | goto out_sig; | |
560 | } | |
561 | error = device_remove_lockspace(&kbuf->i.lspace); | |
562 | break; | |
563 | ||
72c2be77 DT |
564 | case DLM_USER_PURGE: |
565 | if (!proc) { | |
566 | log_print("no locking on control device"); | |
567 | goto out_sig; | |
568 | } | |
569 | error = device_user_purge(proc, &kbuf->i.purge); | |
570 | break; | |
571 | ||
597d0cae DT |
572 | default: |
573 | log_print("Unknown command passed to DLM device : %d\n", | |
574 | kbuf->cmd); | |
575 | } | |
576 | ||
577 | out_sig: | |
578 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
579 | recalc_sigpending(); | |
580 | out_free: | |
581 | kfree(kbuf); | |
582 | return error; | |
583 | } | |
584 | ||
585 | /* Every process that opens the lockspace device has its own "proc" structure | |
586 | hanging off the open file that's used to keep track of locks owned by the | |
587 | process and asts that need to be delivered to the process. */ | |
588 | ||
589 | static int device_open(struct inode *inode, struct file *file) | |
590 | { | |
591 | struct dlm_user_proc *proc; | |
592 | struct dlm_ls *ls; | |
593 | ||
594 | ls = dlm_find_lockspace_device(iminor(inode)); | |
595 | if (!ls) | |
596 | return -ENOENT; | |
597 | ||
598 | proc = kzalloc(sizeof(struct dlm_user_proc), GFP_KERNEL); | |
599 | if (!proc) { | |
600 | dlm_put_lockspace(ls); | |
601 | return -ENOMEM; | |
602 | } | |
603 | ||
604 | proc->lockspace = ls->ls_local_handle; | |
605 | INIT_LIST_HEAD(&proc->asts); | |
606 | INIT_LIST_HEAD(&proc->locks); | |
a1bc86e6 | 607 | INIT_LIST_HEAD(&proc->unlocking); |
597d0cae DT |
608 | spin_lock_init(&proc->asts_spin); |
609 | spin_lock_init(&proc->locks_spin); | |
610 | init_waitqueue_head(&proc->wait); | |
611 | file->private_data = proc; | |
612 | ||
613 | return 0; | |
614 | } | |
615 | ||
616 | static int device_close(struct inode *inode, struct file *file) | |
617 | { | |
618 | struct dlm_user_proc *proc = file->private_data; | |
619 | struct dlm_ls *ls; | |
620 | sigset_t tmpsig, allsigs; | |
621 | ||
622 | ls = dlm_find_lockspace_local(proc->lockspace); | |
623 | if (!ls) | |
624 | return -ENOENT; | |
625 | ||
626 | sigfillset(&allsigs); | |
627 | sigprocmask(SIG_BLOCK, &allsigs, &tmpsig); | |
628 | ||
629 | set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags); | |
630 | ||
631 | dlm_clear_proc_locks(ls, proc); | |
632 | ||
633 | /* at this point no more lkb's should exist for this lockspace, | |
634 | so there's no chance of dlm_user_add_ast() being called and | |
635 | looking for lkb->ua->proc */ | |
636 | ||
637 | kfree(proc); | |
638 | file->private_data = NULL; | |
639 | ||
640 | dlm_put_lockspace(ls); | |
641 | dlm_put_lockspace(ls); /* for the find in device_open() */ | |
642 | ||
643 | /* FIXME: AUTOFREE: if this ls is no longer used do | |
644 | device_remove_lockspace() */ | |
645 | ||
646 | sigprocmask(SIG_SETMASK, &tmpsig, NULL); | |
647 | recalc_sigpending(); | |
648 | ||
649 | return 0; | |
650 | } | |
651 | ||
652 | static int copy_result_to_user(struct dlm_user_args *ua, int compat, int type, | |
653 | int bmode, char __user *buf, size_t count) | |
654 | { | |
655 | #ifdef CONFIG_COMPAT | |
656 | struct dlm_lock_result32 result32; | |
657 | #endif | |
658 | struct dlm_lock_result result; | |
659 | void *resultptr; | |
660 | int error=0; | |
661 | int len; | |
662 | int struct_len; | |
663 | ||
664 | memset(&result, 0, sizeof(struct dlm_lock_result)); | |
d7db923e DT |
665 | result.version[0] = DLM_DEVICE_VERSION_MAJOR; |
666 | result.version[1] = DLM_DEVICE_VERSION_MINOR; | |
667 | result.version[2] = DLM_DEVICE_VERSION_PATCH; | |
597d0cae DT |
668 | memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb)); |
669 | result.user_lksb = ua->user_lksb; | |
670 | ||
671 | /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated | |
672 | in a conversion unless the conversion is successful. See code | |
673 | in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though, | |
674 | notes that a new blocking AST address and parameter are set even if | |
675 | the conversion fails, so maybe we should just do that. */ | |
676 | ||
677 | if (type == AST_BAST) { | |
678 | result.user_astaddr = ua->bastaddr; | |
679 | result.user_astparam = ua->bastparam; | |
680 | result.bast_mode = bmode; | |
681 | } else { | |
682 | result.user_astaddr = ua->castaddr; | |
683 | result.user_astparam = ua->castparam; | |
684 | } | |
685 | ||
686 | #ifdef CONFIG_COMPAT | |
687 | if (compat) | |
688 | len = sizeof(struct dlm_lock_result32); | |
689 | else | |
690 | #endif | |
691 | len = sizeof(struct dlm_lock_result); | |
692 | struct_len = len; | |
693 | ||
694 | /* copy lvb to userspace if there is one, it's been updated, and | |
695 | the user buffer has space for it */ | |
696 | ||
697 | if (ua->update_user_lvb && ua->lksb.sb_lvbptr && | |
698 | count >= len + DLM_USER_LVB_LEN) { | |
699 | if (copy_to_user(buf+len, ua->lksb.sb_lvbptr, | |
700 | DLM_USER_LVB_LEN)) { | |
701 | error = -EFAULT; | |
702 | goto out; | |
703 | } | |
704 | ||
705 | result.lvb_offset = len; | |
706 | len += DLM_USER_LVB_LEN; | |
707 | } | |
708 | ||
709 | result.length = len; | |
710 | resultptr = &result; | |
711 | #ifdef CONFIG_COMPAT | |
712 | if (compat) { | |
713 | compat_output(&result, &result32); | |
714 | resultptr = &result32; | |
715 | } | |
716 | #endif | |
717 | ||
718 | if (copy_to_user(buf, resultptr, struct_len)) | |
719 | error = -EFAULT; | |
720 | else | |
721 | error = len; | |
722 | out: | |
723 | return error; | |
724 | } | |
725 | ||
d7db923e DT |
726 | static int copy_version_to_user(char __user *buf, size_t count) |
727 | { | |
728 | struct dlm_device_version ver; | |
729 | ||
730 | memset(&ver, 0, sizeof(struct dlm_device_version)); | |
731 | ver.version[0] = DLM_DEVICE_VERSION_MAJOR; | |
732 | ver.version[1] = DLM_DEVICE_VERSION_MINOR; | |
733 | ver.version[2] = DLM_DEVICE_VERSION_PATCH; | |
734 | ||
735 | if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version))) | |
736 | return -EFAULT; | |
737 | return sizeof(struct dlm_device_version); | |
738 | } | |
739 | ||
597d0cae DT |
740 | /* a read returns a single ast described in a struct dlm_lock_result */ |
741 | ||
742 | static ssize_t device_read(struct file *file, char __user *buf, size_t count, | |
743 | loff_t *ppos) | |
744 | { | |
745 | struct dlm_user_proc *proc = file->private_data; | |
746 | struct dlm_lkb *lkb; | |
747 | struct dlm_user_args *ua; | |
748 | DECLARE_WAITQUEUE(wait, current); | |
749 | int error, type=0, bmode=0, removed = 0; | |
750 | ||
d7db923e DT |
751 | if (count == sizeof(struct dlm_device_version)) { |
752 | error = copy_version_to_user(buf, count); | |
753 | return error; | |
754 | } | |
755 | ||
756 | if (!proc) { | |
757 | log_print("non-version read from control device %zu", count); | |
758 | return -EINVAL; | |
759 | } | |
760 | ||
597d0cae DT |
761 | #ifdef CONFIG_COMPAT |
762 | if (count < sizeof(struct dlm_lock_result32)) | |
763 | #else | |
764 | if (count < sizeof(struct dlm_lock_result)) | |
765 | #endif | |
766 | return -EINVAL; | |
767 | ||
768 | /* do we really need this? can a read happen after a close? */ | |
769 | if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags)) | |
770 | return -EINVAL; | |
771 | ||
772 | spin_lock(&proc->asts_spin); | |
773 | if (list_empty(&proc->asts)) { | |
774 | if (file->f_flags & O_NONBLOCK) { | |
775 | spin_unlock(&proc->asts_spin); | |
776 | return -EAGAIN; | |
777 | } | |
778 | ||
779 | add_wait_queue(&proc->wait, &wait); | |
780 | ||
781 | repeat: | |
782 | set_current_state(TASK_INTERRUPTIBLE); | |
783 | if (list_empty(&proc->asts) && !signal_pending(current)) { | |
784 | spin_unlock(&proc->asts_spin); | |
785 | schedule(); | |
786 | spin_lock(&proc->asts_spin); | |
787 | goto repeat; | |
788 | } | |
789 | set_current_state(TASK_RUNNING); | |
790 | remove_wait_queue(&proc->wait, &wait); | |
791 | ||
792 | if (signal_pending(current)) { | |
793 | spin_unlock(&proc->asts_spin); | |
794 | return -ERESTARTSYS; | |
795 | } | |
796 | } | |
797 | ||
597d0cae DT |
798 | /* there may be both completion and blocking asts to return for |
799 | the lkb, don't remove lkb from asts list unless no asts remain */ | |
800 | ||
801 | lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_astqueue); | |
802 | ||
803 | if (lkb->lkb_ast_type & AST_COMP) { | |
804 | lkb->lkb_ast_type &= ~AST_COMP; | |
805 | type = AST_COMP; | |
806 | } else if (lkb->lkb_ast_type & AST_BAST) { | |
807 | lkb->lkb_ast_type &= ~AST_BAST; | |
808 | type = AST_BAST; | |
809 | bmode = lkb->lkb_bastmode; | |
810 | } | |
811 | ||
812 | if (!lkb->lkb_ast_type) { | |
813 | list_del(&lkb->lkb_astqueue); | |
814 | removed = 1; | |
815 | } | |
816 | spin_unlock(&proc->asts_spin); | |
817 | ||
818 | ua = (struct dlm_user_args *)lkb->lkb_astparam; | |
819 | error = copy_result_to_user(ua, | |
820 | test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags), | |
821 | type, bmode, buf, count); | |
822 | ||
823 | /* removes reference for the proc->asts lists added by | |
824 | dlm_user_add_ast() and may result in the lkb being freed */ | |
825 | if (removed) | |
826 | dlm_put_lkb(lkb); | |
827 | ||
828 | return error; | |
829 | } | |
830 | ||
831 | static unsigned int device_poll(struct file *file, poll_table *wait) | |
832 | { | |
833 | struct dlm_user_proc *proc = file->private_data; | |
834 | ||
835 | poll_wait(file, &proc->wait, wait); | |
836 | ||
837 | spin_lock(&proc->asts_spin); | |
838 | if (!list_empty(&proc->asts)) { | |
839 | spin_unlock(&proc->asts_spin); | |
840 | return POLLIN | POLLRDNORM; | |
841 | } | |
842 | spin_unlock(&proc->asts_spin); | |
843 | return 0; | |
844 | } | |
845 | ||
846 | static int ctl_device_open(struct inode *inode, struct file *file) | |
847 | { | |
848 | file->private_data = NULL; | |
849 | return 0; | |
850 | } | |
851 | ||
852 | static int ctl_device_close(struct inode *inode, struct file *file) | |
853 | { | |
854 | return 0; | |
855 | } | |
856 | ||
00977a59 | 857 | static const struct file_operations device_fops = { |
597d0cae DT |
858 | .open = device_open, |
859 | .release = device_close, | |
860 | .read = device_read, | |
861 | .write = device_write, | |
862 | .poll = device_poll, | |
863 | .owner = THIS_MODULE, | |
864 | }; | |
865 | ||
00977a59 | 866 | static const struct file_operations ctl_device_fops = { |
597d0cae DT |
867 | .open = ctl_device_open, |
868 | .release = ctl_device_close, | |
d7db923e | 869 | .read = device_read, |
597d0cae DT |
870 | .write = device_write, |
871 | .owner = THIS_MODULE, | |
872 | }; | |
873 | ||
874 | int dlm_user_init(void) | |
875 | { | |
876 | int error; | |
877 | ||
878 | ctl_device.name = "dlm-control"; | |
879 | ctl_device.fops = &ctl_device_fops; | |
880 | ctl_device.minor = MISC_DYNAMIC_MINOR; | |
881 | ||
882 | error = misc_register(&ctl_device); | |
883 | if (error) | |
884 | log_print("misc_register failed for control device"); | |
885 | ||
886 | return error; | |
887 | } | |
888 | ||
889 | void dlm_user_exit(void) | |
890 | { | |
891 | misc_deregister(&ctl_device); | |
892 | } | |
893 |