xref: /linux/fs/dlm/user.c (revision 8ccd54fe45713cd458015b5b08d6098545e70543)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (C) 2006-2010 Red Hat, Inc.  All rights reserved.
4  */
5 
6 #include <linux/miscdevice.h>
7 #include <linux/init.h>
8 #include <linux/wait.h>
9 #include <linux/file.h>
10 #include <linux/fs.h>
11 #include <linux/poll.h>
12 #include <linux/signal.h>
13 #include <linux/spinlock.h>
14 #include <linux/dlm.h>
15 #include <linux/dlm_device.h>
16 #include <linux/slab.h>
17 #include <linux/sched/signal.h>
18 
19 #include <trace/events/dlm.h>
20 
21 #include "dlm_internal.h"
22 #include "lockspace.h"
23 #include "lock.h"
24 #include "lvb_table.h"
25 #include "user.h"
26 #include "ast.h"
27 #include "config.h"
28 #include "memory.h"
29 
30 static const char name_prefix[] = "dlm";
31 static const struct file_operations device_fops;
32 static atomic_t dlm_monitor_opened;
33 static int dlm_monitor_unused = 1;
34 
35 #ifdef CONFIG_COMPAT
36 
37 struct dlm_lock_params32 {
38 	__u8 mode;
39 	__u8 namelen;
40 	__u16 unused;
41 	__u32 flags;
42 	__u32 lkid;
43 	__u32 parent;
44 	__u64 xid;
45 	__u64 timeout;
46 	__u32 castparam;
47 	__u32 castaddr;
48 	__u32 bastparam;
49 	__u32 bastaddr;
50 	__u32 lksb;
51 	char lvb[DLM_USER_LVB_LEN];
52 	char name[];
53 };
54 
55 struct dlm_write_request32 {
56 	__u32 version[3];
57 	__u8 cmd;
58 	__u8 is64bit;
59 	__u8 unused[2];
60 
61 	union  {
62 		struct dlm_lock_params32 lock;
63 		struct dlm_lspace_params lspace;
64 		struct dlm_purge_params purge;
65 	} i;
66 };
67 
68 struct dlm_lksb32 {
69 	__u32 sb_status;
70 	__u32 sb_lkid;
71 	__u8 sb_flags;
72 	__u32 sb_lvbptr;
73 };
74 
75 struct dlm_lock_result32 {
76 	__u32 version[3];
77 	__u32 length;
78 	__u32 user_astaddr;
79 	__u32 user_astparam;
80 	__u32 user_lksb;
81 	struct dlm_lksb32 lksb;
82 	__u8 bast_mode;
83 	__u8 unused[3];
84 	/* Offsets may be zero if no data is present */
85 	__u32 lvb_offset;
86 };
87 
88 static void compat_input(struct dlm_write_request *kb,
89 			 struct dlm_write_request32 *kb32,
90 			 int namelen)
91 {
92 	kb->version[0] = kb32->version[0];
93 	kb->version[1] = kb32->version[1];
94 	kb->version[2] = kb32->version[2];
95 
96 	kb->cmd = kb32->cmd;
97 	kb->is64bit = kb32->is64bit;
98 	if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
99 	    kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
100 		kb->i.lspace.flags = kb32->i.lspace.flags;
101 		kb->i.lspace.minor = kb32->i.lspace.minor;
102 		memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
103 	} else if (kb->cmd == DLM_USER_PURGE) {
104 		kb->i.purge.nodeid = kb32->i.purge.nodeid;
105 		kb->i.purge.pid = kb32->i.purge.pid;
106 	} else {
107 		kb->i.lock.mode = kb32->i.lock.mode;
108 		kb->i.lock.namelen = kb32->i.lock.namelen;
109 		kb->i.lock.flags = kb32->i.lock.flags;
110 		kb->i.lock.lkid = kb32->i.lock.lkid;
111 		kb->i.lock.parent = kb32->i.lock.parent;
112 		kb->i.lock.xid = kb32->i.lock.xid;
113 		kb->i.lock.timeout = kb32->i.lock.timeout;
114 		kb->i.lock.castparam = (__user void *)(long)kb32->i.lock.castparam;
115 		kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr;
116 		kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam;
117 		kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr;
118 		kb->i.lock.lksb = (__user void *)(long)kb32->i.lock.lksb;
119 		memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
120 		memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
121 	}
122 }
123 
124 static void compat_output(struct dlm_lock_result *res,
125 			  struct dlm_lock_result32 *res32)
126 {
127 	memset(res32, 0, sizeof(*res32));
128 
129 	res32->version[0] = res->version[0];
130 	res32->version[1] = res->version[1];
131 	res32->version[2] = res->version[2];
132 
133 	res32->user_astaddr = (__u32)(__force long)res->user_astaddr;
134 	res32->user_astparam = (__u32)(__force long)res->user_astparam;
135 	res32->user_lksb = (__u32)(__force long)res->user_lksb;
136 	res32->bast_mode = res->bast_mode;
137 
138 	res32->lvb_offset = res->lvb_offset;
139 	res32->length = res->length;
140 
141 	res32->lksb.sb_status = res->lksb.sb_status;
142 	res32->lksb.sb_flags = res->lksb.sb_flags;
143 	res32->lksb.sb_lkid = res->lksb.sb_lkid;
144 	res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
145 }
146 #endif
147 
148 /* Figure out if this lock is at the end of its life and no longer
149    available for the application to use.  The lkb still exists until
150    the final ast is read.  A lock becomes EOL in three situations:
151      1. a noqueue request fails with EAGAIN
152      2. an unlock completes with EUNLOCK
153      3. a cancel of a waiting request completes with ECANCEL/EDEADLK
154    An EOL lock needs to be removed from the process's list of locks.
155    And we can't allow any new operation on an EOL lock.  This is
156    not related to the lifetime of the lkb struct which is managed
157    entirely by refcount. */
158 
159 static int lkb_is_endoflife(int mode, int status)
160 {
161 	switch (status) {
162 	case -DLM_EUNLOCK:
163 		return 1;
164 	case -DLM_ECANCEL:
165 	case -ETIMEDOUT:
166 	case -EDEADLK:
167 	case -EAGAIN:
168 		if (mode == DLM_LOCK_IV)
169 			return 1;
170 		break;
171 	}
172 	return 0;
173 }
174 
175 /* we could possibly check if the cancel of an orphan has resulted in the lkb
176    being removed and then remove that lkb from the orphans list and free it */
177 
178 void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
179 		      int status, uint32_t sbflags)
180 {
181 	struct dlm_ls *ls;
182 	struct dlm_user_args *ua;
183 	struct dlm_user_proc *proc;
184 	int rv;
185 
186 	if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
187 	    test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
188 		return;
189 
190 	ls = lkb->lkb_resource->res_ls;
191 	spin_lock(&ls->ls_clear_proc_locks);
192 
193 	/* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
194 	   can't be delivered.  For ORPHAN's, dlm_clear_proc_locks() freed
195 	   lkb->ua so we can't try to use it.  This second check is necessary
196 	   for cases where a completion ast is received for an operation that
197 	   began before clear_proc_locks did its cancel/unlock. */
198 
199 	if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
200 	    test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
201 		goto out;
202 
203 	DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
204 	ua = lkb->lkb_ua;
205 	proc = ua->proc;
206 
207 	if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
208 		goto out;
209 
210 	if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
211 		set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags);
212 
213 	spin_lock(&proc->asts_spin);
214 
215 	rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
216 	switch (rv) {
217 	case DLM_ENQUEUE_CALLBACK_FAILURE:
218 		spin_unlock(&proc->asts_spin);
219 		WARN_ON_ONCE(1);
220 		goto out;
221 	case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
222 		kref_get(&lkb->lkb_ref);
223 		list_add_tail(&lkb->lkb_cb_list, &proc->asts);
224 		wake_up_interruptible(&proc->wait);
225 		break;
226 	case DLM_ENQUEUE_CALLBACK_SUCCESS:
227 		break;
228 	default:
229 		WARN_ON_ONCE(1);
230 		break;
231 	}
232 	spin_unlock(&proc->asts_spin);
233 
234 	if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
235 		/* N.B. spin_lock locks_spin, not asts_spin */
236 		spin_lock(&proc->locks_spin);
237 		if (!list_empty(&lkb->lkb_ownqueue)) {
238 			list_del_init(&lkb->lkb_ownqueue);
239 			dlm_put_lkb(lkb);
240 		}
241 		spin_unlock(&proc->locks_spin);
242 	}
243  out:
244 	spin_unlock(&ls->ls_clear_proc_locks);
245 }
246 
247 static int device_user_lock(struct dlm_user_proc *proc,
248 			    struct dlm_lock_params *params)
249 {
250 	struct dlm_ls *ls;
251 	struct dlm_user_args *ua;
252 	uint32_t lkid;
253 	int error = -ENOMEM;
254 
255 	ls = dlm_find_lockspace_local(proc->lockspace);
256 	if (!ls)
257 		return -ENOENT;
258 
259 	if (!params->castaddr || !params->lksb) {
260 		error = -EINVAL;
261 		goto out;
262 	}
263 
264 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
265 	if (!ua)
266 		goto out;
267 	ua->proc = proc;
268 	ua->user_lksb = params->lksb;
269 	ua->castparam = params->castparam;
270 	ua->castaddr = params->castaddr;
271 	ua->bastparam = params->bastparam;
272 	ua->bastaddr = params->bastaddr;
273 	ua->xid = params->xid;
274 
275 	if (params->flags & DLM_LKF_CONVERT) {
276 		error = dlm_user_convert(ls, ua,
277 					 params->mode, params->flags,
278 					 params->lkid, params->lvb);
279 	} else if (params->flags & DLM_LKF_ORPHAN) {
280 		error = dlm_user_adopt_orphan(ls, ua,
281 					 params->mode, params->flags,
282 					 params->name, params->namelen,
283 					 &lkid);
284 		if (!error)
285 			error = lkid;
286 	} else {
287 		error = dlm_user_request(ls, ua,
288 					 params->mode, params->flags,
289 					 params->name, params->namelen);
290 		if (!error)
291 			error = ua->lksb.sb_lkid;
292 	}
293  out:
294 	dlm_put_lockspace(ls);
295 	return error;
296 }
297 
298 static int device_user_unlock(struct dlm_user_proc *proc,
299 			      struct dlm_lock_params *params)
300 {
301 	struct dlm_ls *ls;
302 	struct dlm_user_args *ua;
303 	int error = -ENOMEM;
304 
305 	ls = dlm_find_lockspace_local(proc->lockspace);
306 	if (!ls)
307 		return -ENOENT;
308 
309 	ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
310 	if (!ua)
311 		goto out;
312 	ua->proc = proc;
313 	ua->user_lksb = params->lksb;
314 	ua->castparam = params->castparam;
315 	ua->castaddr = params->castaddr;
316 
317 	if (params->flags & DLM_LKF_CANCEL)
318 		error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
319 	else
320 		error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
321 					params->lvb);
322  out:
323 	dlm_put_lockspace(ls);
324 	return error;
325 }
326 
327 static int device_user_deadlock(struct dlm_user_proc *proc,
328 				struct dlm_lock_params *params)
329 {
330 	struct dlm_ls *ls;
331 	int error;
332 
333 	ls = dlm_find_lockspace_local(proc->lockspace);
334 	if (!ls)
335 		return -ENOENT;
336 
337 	error = dlm_user_deadlock(ls, params->flags, params->lkid);
338 
339 	dlm_put_lockspace(ls);
340 	return error;
341 }
342 
343 static int dlm_device_register(struct dlm_ls *ls, char *name)
344 {
345 	int error, len;
346 
347 	/* The device is already registered.  This happens when the
348 	   lockspace is created multiple times from userspace. */
349 	if (ls->ls_device.name)
350 		return 0;
351 
352 	error = -ENOMEM;
353 	len = strlen(name) + strlen(name_prefix) + 2;
354 	ls->ls_device.name = kzalloc(len, GFP_NOFS);
355 	if (!ls->ls_device.name)
356 		goto fail;
357 
358 	snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
359 		 name);
360 	ls->ls_device.fops = &device_fops;
361 	ls->ls_device.minor = MISC_DYNAMIC_MINOR;
362 
363 	error = misc_register(&ls->ls_device);
364 	if (error) {
365 		kfree(ls->ls_device.name);
366 		/* this has to be set to NULL
367 		 * to avoid a double-free in dlm_device_deregister
368 		 */
369 		ls->ls_device.name = NULL;
370 	}
371 fail:
372 	return error;
373 }
374 
375 int dlm_device_deregister(struct dlm_ls *ls)
376 {
377 	/* The device is not registered.  This happens when the lockspace
378 	   was never used from userspace, or when device_create_lockspace()
379 	   calls dlm_release_lockspace() after the register fails. */
380 	if (!ls->ls_device.name)
381 		return 0;
382 
383 	misc_deregister(&ls->ls_device);
384 	kfree(ls->ls_device.name);
385 	return 0;
386 }
387 
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 
404 static int device_create_lockspace(struct dlm_lspace_params *params)
405 {
406 	dlm_lockspace_t *lockspace;
407 	struct dlm_ls *ls;
408 	int error;
409 
410 	if (!capable(CAP_SYS_ADMIN))
411 		return -EPERM;
412 
413 	error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name,
414 				       params->flags, DLM_USER_LVB_LEN, NULL,
415 				       NULL, NULL, &lockspace);
416 	if (error)
417 		return error;
418 
419 	ls = dlm_find_lockspace_local(lockspace);
420 	if (!ls)
421 		return -ENOENT;
422 
423 	error = dlm_device_register(ls, params->name);
424 	dlm_put_lockspace(ls);
425 
426 	if (error)
427 		dlm_release_lockspace(lockspace, 0);
428 	else
429 		error = ls->ls_device.minor;
430 
431 	return error;
432 }
433 
434 static int device_remove_lockspace(struct dlm_lspace_params *params)
435 {
436 	dlm_lockspace_t *lockspace;
437 	struct dlm_ls *ls;
438 	int error, force = 0;
439 
440 	if (!capable(CAP_SYS_ADMIN))
441 		return -EPERM;
442 
443 	ls = dlm_find_lockspace_device(params->minor);
444 	if (!ls)
445 		return -ENOENT;
446 
447 	if (params->flags & DLM_USER_LSFLG_FORCEFREE)
448 		force = 2;
449 
450 	lockspace = ls->ls_local_handle;
451 	dlm_put_lockspace(ls);
452 
453 	/* The final dlm_release_lockspace waits for references to go to
454 	   zero, so all processes will need to close their device for the
455 	   ls before the release will proceed.  release also calls the
456 	   device_deregister above.  Converting a positive return value
457 	   from release to zero means that userspace won't know when its
458 	   release was the final one, but it shouldn't need to know. */
459 
460 	error = dlm_release_lockspace(lockspace, force);
461 	if (error > 0)
462 		error = 0;
463 	return error;
464 }
465 
466 /* Check the user's version matches ours */
467 static int check_version(struct dlm_write_request *req)
468 {
469 	if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
470 	    (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
471 	     req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
472 
473 		printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
474 		       "user (%d.%d.%d) kernel (%d.%d.%d)\n",
475 		       current->comm,
476 		       task_pid_nr(current),
477 		       req->version[0],
478 		       req->version[1],
479 		       req->version[2],
480 		       DLM_DEVICE_VERSION_MAJOR,
481 		       DLM_DEVICE_VERSION_MINOR,
482 		       DLM_DEVICE_VERSION_PATCH);
483 		return -EINVAL;
484 	}
485 	return 0;
486 }
487 
488 /*
489  * device_write
490  *
491  *   device_user_lock
492  *     dlm_user_request -> request_lock
493  *     dlm_user_convert -> convert_lock
494  *
495  *   device_user_unlock
496  *     dlm_user_unlock -> unlock_lock
497  *     dlm_user_cancel -> cancel_lock
498  *
499  *   device_create_lockspace
500  *     dlm_new_lockspace
501  *
502  *   device_remove_lockspace
503  *     dlm_release_lockspace
504  */
505 
506 /* a write to a lockspace device is a lock or unlock request, a write
507    to the control device is to create/remove a lockspace */
508 
509 static ssize_t device_write(struct file *file, const char __user *buf,
510 			    size_t count, loff_t *ppos)
511 {
512 	struct dlm_user_proc *proc = file->private_data;
513 	struct dlm_write_request *kbuf;
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 
523 	/*
524 	 * can't compare against COMPAT/dlm_write_request32 because
525 	 * we don't yet know if is64bit is zero
526 	 */
527 	if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
528 		return -EINVAL;
529 
530 	kbuf = memdup_user_nul(buf, count);
531 	if (IS_ERR(kbuf))
532 		return PTR_ERR(kbuf);
533 
534 	if (check_version(kbuf)) {
535 		error = -EBADE;
536 		goto out_free;
537 	}
538 
539 #ifdef CONFIG_COMPAT
540 	if (!kbuf->is64bit) {
541 		struct dlm_write_request32 *k32buf;
542 		int namelen = 0;
543 
544 		if (count > sizeof(struct dlm_write_request32))
545 			namelen = count - sizeof(struct dlm_write_request32);
546 
547 		k32buf = (struct dlm_write_request32 *)kbuf;
548 
549 		/* add 1 after namelen so that the name string is terminated */
550 		kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
551 			       GFP_NOFS);
552 		if (!kbuf) {
553 			kfree(k32buf);
554 			return -ENOMEM;
555 		}
556 
557 		if (proc)
558 			set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
559 
560 		compat_input(kbuf, k32buf, namelen);
561 		kfree(k32buf);
562 	}
563 #endif
564 
565 	/* do we really need this? can a write happen after a close? */
566 	if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
567 	    (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
568 		error = -EINVAL;
569 		goto out_free;
570 	}
571 
572 	error = -EINVAL;
573 
574 	switch (kbuf->cmd)
575 	{
576 	case DLM_USER_LOCK:
577 		if (!proc) {
578 			log_print("no locking on control device");
579 			goto out_free;
580 		}
581 		error = device_user_lock(proc, &kbuf->i.lock);
582 		break;
583 
584 	case DLM_USER_UNLOCK:
585 		if (!proc) {
586 			log_print("no locking on control device");
587 			goto out_free;
588 		}
589 		error = device_user_unlock(proc, &kbuf->i.lock);
590 		break;
591 
592 	case DLM_USER_DEADLOCK:
593 		if (!proc) {
594 			log_print("no locking on control device");
595 			goto out_free;
596 		}
597 		error = device_user_deadlock(proc, &kbuf->i.lock);
598 		break;
599 
600 	case DLM_USER_CREATE_LOCKSPACE:
601 		if (proc) {
602 			log_print("create/remove only on control device");
603 			goto out_free;
604 		}
605 		error = device_create_lockspace(&kbuf->i.lspace);
606 		break;
607 
608 	case DLM_USER_REMOVE_LOCKSPACE:
609 		if (proc) {
610 			log_print("create/remove only on control device");
611 			goto out_free;
612 		}
613 		error = device_remove_lockspace(&kbuf->i.lspace);
614 		break;
615 
616 	case DLM_USER_PURGE:
617 		if (!proc) {
618 			log_print("no locking on control device");
619 			goto out_free;
620 		}
621 		error = device_user_purge(proc, &kbuf->i.purge);
622 		break;
623 
624 	default:
625 		log_print("Unknown command passed to DLM device : %d\n",
626 			  kbuf->cmd);
627 	}
628 
629  out_free:
630 	kfree(kbuf);
631 	return error;
632 }
633 
634 /* Every process that opens the lockspace device has its own "proc" structure
635    hanging off the open file that's used to keep track of locks owned by the
636    process and asts that need to be delivered to the process. */
637 
638 static int device_open(struct inode *inode, struct file *file)
639 {
640 	struct dlm_user_proc *proc;
641 	struct dlm_ls *ls;
642 
643 	ls = dlm_find_lockspace_device(iminor(inode));
644 	if (!ls)
645 		return -ENOENT;
646 
647 	proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
648 	if (!proc) {
649 		dlm_put_lockspace(ls);
650 		return -ENOMEM;
651 	}
652 
653 	proc->lockspace = ls->ls_local_handle;
654 	INIT_LIST_HEAD(&proc->asts);
655 	INIT_LIST_HEAD(&proc->locks);
656 	INIT_LIST_HEAD(&proc->unlocking);
657 	spin_lock_init(&proc->asts_spin);
658 	spin_lock_init(&proc->locks_spin);
659 	init_waitqueue_head(&proc->wait);
660 	file->private_data = proc;
661 
662 	return 0;
663 }
664 
665 static int device_close(struct inode *inode, struct file *file)
666 {
667 	struct dlm_user_proc *proc = file->private_data;
668 	struct dlm_ls *ls;
669 
670 	ls = dlm_find_lockspace_local(proc->lockspace);
671 	if (!ls)
672 		return -ENOENT;
673 
674 	set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
675 
676 	dlm_clear_proc_locks(ls, proc);
677 
678 	/* at this point no more lkb's should exist for this lockspace,
679 	   so there's no chance of dlm_user_add_ast() being called and
680 	   looking for lkb->ua->proc */
681 
682 	kfree(proc);
683 	file->private_data = NULL;
684 
685 	dlm_put_lockspace(ls);
686 	dlm_put_lockspace(ls);  /* for the find in device_open() */
687 
688 	/* FIXME: AUTOFREE: if this ls is no longer used do
689 	   device_remove_lockspace() */
690 
691 	return 0;
692 }
693 
694 static int copy_result_to_user(struct dlm_user_args *ua, int compat,
695 			       uint32_t flags, int mode, int copy_lvb,
696 			       char __user *buf, size_t count)
697 {
698 #ifdef CONFIG_COMPAT
699 	struct dlm_lock_result32 result32;
700 #endif
701 	struct dlm_lock_result result;
702 	void *resultptr;
703 	int error=0;
704 	int len;
705 	int struct_len;
706 
707 	memset(&result, 0, sizeof(struct dlm_lock_result));
708 	result.version[0] = DLM_DEVICE_VERSION_MAJOR;
709 	result.version[1] = DLM_DEVICE_VERSION_MINOR;
710 	result.version[2] = DLM_DEVICE_VERSION_PATCH;
711 	memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr));
712 	result.user_lksb = ua->user_lksb;
713 
714 	/* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
715 	   in a conversion unless the conversion is successful.  See code
716 	   in dlm_user_convert() for updating ua from ua_tmp.  OpenVMS, though,
717 	   notes that a new blocking AST address and parameter are set even if
718 	   the conversion fails, so maybe we should just do that. */
719 
720 	if (flags & DLM_CB_BAST) {
721 		result.user_astaddr = ua->bastaddr;
722 		result.user_astparam = ua->bastparam;
723 		result.bast_mode = mode;
724 	} else {
725 		result.user_astaddr = ua->castaddr;
726 		result.user_astparam = ua->castparam;
727 	}
728 
729 #ifdef CONFIG_COMPAT
730 	if (compat)
731 		len = sizeof(struct dlm_lock_result32);
732 	else
733 #endif
734 		len = sizeof(struct dlm_lock_result);
735 	struct_len = len;
736 
737 	/* copy lvb to userspace if there is one, it's been updated, and
738 	   the user buffer has space for it */
739 
740 	if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
741 		if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
742 				 DLM_USER_LVB_LEN)) {
743 			error = -EFAULT;
744 			goto out;
745 		}
746 
747 		result.lvb_offset = len;
748 		len += DLM_USER_LVB_LEN;
749 	}
750 
751 	result.length = len;
752 	resultptr = &result;
753 #ifdef CONFIG_COMPAT
754 	if (compat) {
755 		compat_output(&result, &result32);
756 		resultptr = &result32;
757 	}
758 #endif
759 
760 	if (copy_to_user(buf, resultptr, struct_len))
761 		error = -EFAULT;
762 	else
763 		error = len;
764  out:
765 	return error;
766 }
767 
768 static int copy_version_to_user(char __user *buf, size_t count)
769 {
770 	struct dlm_device_version ver;
771 
772 	memset(&ver, 0, sizeof(struct dlm_device_version));
773 	ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
774 	ver.version[1] = DLM_DEVICE_VERSION_MINOR;
775 	ver.version[2] = DLM_DEVICE_VERSION_PATCH;
776 
777 	if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
778 		return -EFAULT;
779 	return sizeof(struct dlm_device_version);
780 }
781 
782 /* a read returns a single ast described in a struct dlm_lock_result */
783 
784 static ssize_t device_read(struct file *file, char __user *buf, size_t count,
785 			   loff_t *ppos)
786 {
787 	struct dlm_user_proc *proc = file->private_data;
788 	struct dlm_lkb *lkb;
789 	DECLARE_WAITQUEUE(wait, current);
790 	struct dlm_callback *cb;
791 	int rv, copy_lvb = 0;
792 	int old_mode, new_mode;
793 
794 	if (count == sizeof(struct dlm_device_version)) {
795 		rv = copy_version_to_user(buf, count);
796 		return rv;
797 	}
798 
799 	if (!proc) {
800 		log_print("non-version read from control device %zu", count);
801 		return -EINVAL;
802 	}
803 
804 #ifdef CONFIG_COMPAT
805 	if (count < sizeof(struct dlm_lock_result32))
806 #else
807 	if (count < sizeof(struct dlm_lock_result))
808 #endif
809 		return -EINVAL;
810 
811  try_another:
812 
813 	/* do we really need this? can a read happen after a close? */
814 	if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
815 		return -EINVAL;
816 
817 	spin_lock(&proc->asts_spin);
818 	if (list_empty(&proc->asts)) {
819 		if (file->f_flags & O_NONBLOCK) {
820 			spin_unlock(&proc->asts_spin);
821 			return -EAGAIN;
822 		}
823 
824 		add_wait_queue(&proc->wait, &wait);
825 
826 	repeat:
827 		set_current_state(TASK_INTERRUPTIBLE);
828 		if (list_empty(&proc->asts) && !signal_pending(current)) {
829 			spin_unlock(&proc->asts_spin);
830 			schedule();
831 			spin_lock(&proc->asts_spin);
832 			goto repeat;
833 		}
834 		set_current_state(TASK_RUNNING);
835 		remove_wait_queue(&proc->wait, &wait);
836 
837 		if (signal_pending(current)) {
838 			spin_unlock(&proc->asts_spin);
839 			return -ERESTARTSYS;
840 		}
841 	}
842 
843 	/* if we empty lkb_callbacks, we don't want to unlock the spinlock
844 	   without removing lkb_cb_list; so empty lkb_cb_list is always
845 	   consistent with empty lkb_callbacks */
846 
847 	lkb = list_first_entry(&proc->asts, struct dlm_lkb, lkb_cb_list);
848 
849 	/* rem_lkb_callback sets a new lkb_last_cast */
850 	old_mode = lkb->lkb_last_cast->mode;
851 
852 	rv = dlm_dequeue_lkb_callback(lkb, &cb);
853 	switch (rv) {
854 	case DLM_DEQUEUE_CALLBACK_EMPTY:
855 		/* this shouldn't happen; lkb should have been removed from
856 		 * list when last item was dequeued
857 		 */
858 		log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
859 		list_del_init(&lkb->lkb_cb_list);
860 		spin_unlock(&proc->asts_spin);
861 		/* removes ref for proc->asts, may cause lkb to be freed */
862 		dlm_put_lkb(lkb);
863 		WARN_ON_ONCE(1);
864 		goto try_another;
865 	case DLM_DEQUEUE_CALLBACK_LAST:
866 		list_del_init(&lkb->lkb_cb_list);
867 		clear_bit(DLM_IFL_CB_PENDING_BIT, &lkb->lkb_iflags);
868 		break;
869 	case DLM_DEQUEUE_CALLBACK_SUCCESS:
870 		break;
871 	default:
872 		WARN_ON_ONCE(1);
873 		break;
874 	}
875 	spin_unlock(&proc->asts_spin);
876 
877 	if (cb->flags & DLM_CB_BAST) {
878 		trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb->mode);
879 	} else if (cb->flags & DLM_CB_CAST) {
880 		new_mode = cb->mode;
881 
882 		if (!cb->sb_status && lkb->lkb_lksb->sb_lvbptr &&
883 		    dlm_lvb_operations[old_mode + 1][new_mode + 1])
884 			copy_lvb = 1;
885 
886 		lkb->lkb_lksb->sb_status = cb->sb_status;
887 		lkb->lkb_lksb->sb_flags = cb->sb_flags;
888 		trace_dlm_ast(lkb->lkb_resource->res_ls, lkb);
889 	}
890 
891 	rv = copy_result_to_user(lkb->lkb_ua,
892 				 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
893 				 cb->flags, cb->mode, copy_lvb, buf, count);
894 
895 	kref_put(&cb->ref, dlm_release_callback);
896 
897 	/* removes ref for proc->asts, may cause lkb to be freed */
898 	if (rv == DLM_DEQUEUE_CALLBACK_LAST)
899 		dlm_put_lkb(lkb);
900 
901 	return rv;
902 }
903 
904 static __poll_t device_poll(struct file *file, poll_table *wait)
905 {
906 	struct dlm_user_proc *proc = file->private_data;
907 
908 	poll_wait(file, &proc->wait, wait);
909 
910 	spin_lock(&proc->asts_spin);
911 	if (!list_empty(&proc->asts)) {
912 		spin_unlock(&proc->asts_spin);
913 		return EPOLLIN | EPOLLRDNORM;
914 	}
915 	spin_unlock(&proc->asts_spin);
916 	return 0;
917 }
918 
919 int dlm_user_daemon_available(void)
920 {
921 	/* dlm_controld hasn't started (or, has started, but not
922 	   properly populated configfs) */
923 
924 	if (!dlm_our_nodeid())
925 		return 0;
926 
927 	/* This is to deal with versions of dlm_controld that don't
928 	   know about the monitor device.  We assume that if the
929 	   dlm_controld was started (above), but the monitor device
930 	   was never opened, that it's an old version.  dlm_controld
931 	   should open the monitor device before populating configfs. */
932 
933 	if (dlm_monitor_unused)
934 		return 1;
935 
936 	return atomic_read(&dlm_monitor_opened) ? 1 : 0;
937 }
938 
939 static int ctl_device_open(struct inode *inode, struct file *file)
940 {
941 	file->private_data = NULL;
942 	return 0;
943 }
944 
945 static int ctl_device_close(struct inode *inode, struct file *file)
946 {
947 	return 0;
948 }
949 
950 static int monitor_device_open(struct inode *inode, struct file *file)
951 {
952 	atomic_inc(&dlm_monitor_opened);
953 	dlm_monitor_unused = 0;
954 	return 0;
955 }
956 
957 static int monitor_device_close(struct inode *inode, struct file *file)
958 {
959 	if (atomic_dec_and_test(&dlm_monitor_opened))
960 		dlm_stop_lockspaces();
961 	return 0;
962 }
963 
964 static const struct file_operations device_fops = {
965 	.open    = device_open,
966 	.release = device_close,
967 	.read    = device_read,
968 	.write   = device_write,
969 	.poll    = device_poll,
970 	.owner   = THIS_MODULE,
971 	.llseek  = noop_llseek,
972 };
973 
974 static const struct file_operations ctl_device_fops = {
975 	.open    = ctl_device_open,
976 	.release = ctl_device_close,
977 	.read    = device_read,
978 	.write   = device_write,
979 	.owner   = THIS_MODULE,
980 	.llseek  = noop_llseek,
981 };
982 
983 static struct miscdevice ctl_device = {
984 	.name  = "dlm-control",
985 	.fops  = &ctl_device_fops,
986 	.minor = MISC_DYNAMIC_MINOR,
987 };
988 
989 static const struct file_operations monitor_device_fops = {
990 	.open    = monitor_device_open,
991 	.release = monitor_device_close,
992 	.owner   = THIS_MODULE,
993 	.llseek  = noop_llseek,
994 };
995 
996 static struct miscdevice monitor_device = {
997 	.name  = "dlm-monitor",
998 	.fops  = &monitor_device_fops,
999 	.minor = MISC_DYNAMIC_MINOR,
1000 };
1001 
1002 int __init dlm_user_init(void)
1003 {
1004 	int error;
1005 
1006 	atomic_set(&dlm_monitor_opened, 0);
1007 
1008 	error = misc_register(&ctl_device);
1009 	if (error) {
1010 		log_print("misc_register failed for control device");
1011 		goto out;
1012 	}
1013 
1014 	error = misc_register(&monitor_device);
1015 	if (error) {
1016 		log_print("misc_register failed for monitor device");
1017 		misc_deregister(&ctl_device);
1018 	}
1019  out:
1020 	return error;
1021 }
1022 
1023 void dlm_user_exit(void)
1024 {
1025 	misc_deregister(&ctl_device);
1026 	misc_deregister(&monitor_device);
1027 }
1028 
1029