Loading...
1/*
2 * Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved.
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/file.h>
13#include <linux/fs.h>
14#include <linux/poll.h>
15#include <linux/signal.h>
16#include <linux/spinlock.h>
17#include <linux/dlm.h>
18#include <linux/dlm_device.h>
19#include <linux/slab.h>
20#include <linux/sched/signal.h>
21
22#include "dlm_internal.h"
23#include "lockspace.h"
24#include "lock.h"
25#include "lvb_table.h"
26#include "user.h"
27#include "ast.h"
28
29static const char name_prefix[] = "dlm";
30static const struct file_operations device_fops;
31static atomic_t dlm_monitor_opened;
32static int dlm_monitor_unused = 1;
33
34#ifdef CONFIG_COMPAT
35
36struct dlm_lock_params32 {
37 __u8 mode;
38 __u8 namelen;
39 __u16 unused;
40 __u32 flags;
41 __u32 lkid;
42 __u32 parent;
43 __u64 xid;
44 __u64 timeout;
45 __u32 castparam;
46 __u32 castaddr;
47 __u32 bastparam;
48 __u32 bastaddr;
49 __u32 lksb;
50 char lvb[DLM_USER_LVB_LEN];
51 char name[0];
52};
53
54struct dlm_write_request32 {
55 __u32 version[3];
56 __u8 cmd;
57 __u8 is64bit;
58 __u8 unused[2];
59
60 union {
61 struct dlm_lock_params32 lock;
62 struct dlm_lspace_params lspace;
63 struct dlm_purge_params purge;
64 } i;
65};
66
67struct dlm_lksb32 {
68 __u32 sb_status;
69 __u32 sb_lkid;
70 __u8 sb_flags;
71 __u32 sb_lvbptr;
72};
73
74struct dlm_lock_result32 {
75 __u32 version[3];
76 __u32 length;
77 __u32 user_astaddr;
78 __u32 user_astparam;
79 __u32 user_lksb;
80 struct dlm_lksb32 lksb;
81 __u8 bast_mode;
82 __u8 unused[3];
83 /* Offsets may be zero if no data is present */
84 __u32 lvb_offset;
85};
86
87static void compat_input(struct dlm_write_request *kb,
88 struct dlm_write_request32 *kb32,
89 int namelen)
90{
91 kb->version[0] = kb32->version[0];
92 kb->version[1] = kb32->version[1];
93 kb->version[2] = kb32->version[2];
94
95 kb->cmd = kb32->cmd;
96 kb->is64bit = kb32->is64bit;
97 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
98 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
99 kb->i.lspace.flags = kb32->i.lspace.flags;
100 kb->i.lspace.minor = kb32->i.lspace.minor;
101 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
102 } else if (kb->cmd == DLM_USER_PURGE) {
103 kb->i.purge.nodeid = kb32->i.purge.nodeid;
104 kb->i.purge.pid = kb32->i.purge.pid;
105 } else {
106 kb->i.lock.mode = kb32->i.lock.mode;
107 kb->i.lock.namelen = kb32->i.lock.namelen;
108 kb->i.lock.flags = kb32->i.lock.flags;
109 kb->i.lock.lkid = kb32->i.lock.lkid;
110 kb->i.lock.parent = kb32->i.lock.parent;
111 kb->i.lock.xid = kb32->i.lock.xid;
112 kb->i.lock.timeout = kb32->i.lock.timeout;
113 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
114 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
115 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
116 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
117 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
118 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
119 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
120 }
121}
122
123static void compat_output(struct dlm_lock_result *res,
124 struct dlm_lock_result32 *res32)
125{
126 memset(res32, 0, sizeof(*res32));
127
128 res32->version[0] = res->version[0];
129 res32->version[1] = res->version[1];
130 res32->version[2] = res->version[2];
131
132 res32->user_astaddr = (__u32)(long)res->user_astaddr;
133 res32->user_astparam = (__u32)(long)res->user_astparam;
134 res32->user_lksb = (__u32)(long)res->user_lksb;
135 res32->bast_mode = res->bast_mode;
136
137 res32->lvb_offset = res->lvb_offset;
138 res32->length = res->length;
139
140 res32->lksb.sb_status = res->lksb.sb_status;
141 res32->lksb.sb_flags = res->lksb.sb_flags;
142 res32->lksb.sb_lkid = res->lksb.sb_lkid;
143 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
144}
145#endif
146
147/* Figure out if this lock is at the end of its life and no longer
148 available for the application to use. The lkb still exists until
149 the final ast is read. A lock becomes EOL in three situations:
150 1. a noqueue request fails with EAGAIN
151 2. an unlock completes with EUNLOCK
152 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
153 An EOL lock needs to be removed from the process's list of locks.
154 And we can't allow any new operation on an EOL lock. This is
155 not related to the lifetime of the lkb struct which is managed
156 entirely by refcount. */
157
158static int lkb_is_endoflife(int mode, int status)
159{
160 switch (status) {
161 case -DLM_EUNLOCK:
162 return 1;
163 case -DLM_ECANCEL:
164 case -ETIMEDOUT:
165 case -EDEADLK:
166 case -EAGAIN:
167 if (mode == DLM_LOCK_IV)
168 return 1;
169 break;
170 }
171 return 0;
172}
173
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 */
176
177void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
178 int status, uint32_t sbflags, uint64_t seq)
179{
180 struct dlm_ls *ls;
181 struct dlm_user_args *ua;
182 struct dlm_user_proc *proc;
183 int rv;
184
185 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
186 return;
187
188 ls = lkb->lkb_resource->res_ls;
189 mutex_lock(&ls->ls_clear_proc_locks);
190
191 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
192 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
193 lkb->ua so we can't try to use it. This second check is necessary
194 for cases where a completion ast is received for an operation that
195 began before clear_proc_locks did its cancel/unlock. */
196
197 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
198 goto out;
199
200 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
201 ua = lkb->lkb_ua;
202 proc = ua->proc;
203
204 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
205 goto out;
206
207 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
208 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
209
210 spin_lock(&proc->asts_spin);
211
212 rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
213 if (rv < 0) {
214 spin_unlock(&proc->asts_spin);
215 goto out;
216 }
217
218 if (list_empty(&lkb->lkb_cb_list)) {
219 kref_get(&lkb->lkb_ref);
220 list_add_tail(&lkb->lkb_cb_list, &proc->asts);
221 wake_up_interruptible(&proc->wait);
222 }
223 spin_unlock(&proc->asts_spin);
224
225 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
226 /* N.B. spin_lock locks_spin, not asts_spin */
227 spin_lock(&proc->locks_spin);
228 if (!list_empty(&lkb->lkb_ownqueue)) {
229 list_del_init(&lkb->lkb_ownqueue);
230 dlm_put_lkb(lkb);
231 }
232 spin_unlock(&proc->locks_spin);
233 }
234 out:
235 mutex_unlock(&ls->ls_clear_proc_locks);
236}
237
238static int device_user_lock(struct dlm_user_proc *proc,
239 struct dlm_lock_params *params)
240{
241 struct dlm_ls *ls;
242 struct dlm_user_args *ua;
243 uint32_t lkid;
244 int error = -ENOMEM;
245
246 ls = dlm_find_lockspace_local(proc->lockspace);
247 if (!ls)
248 return -ENOENT;
249
250 if (!params->castaddr || !params->lksb) {
251 error = -EINVAL;
252 goto out;
253 }
254
255 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
256 if (!ua)
257 goto out;
258 ua->proc = proc;
259 ua->user_lksb = params->lksb;
260 ua->castparam = params->castparam;
261 ua->castaddr = params->castaddr;
262 ua->bastparam = params->bastparam;
263 ua->bastaddr = params->bastaddr;
264 ua->xid = params->xid;
265
266 if (params->flags & DLM_LKF_CONVERT) {
267 error = dlm_user_convert(ls, ua,
268 params->mode, params->flags,
269 params->lkid, params->lvb,
270 (unsigned long) params->timeout);
271 } else if (params->flags & DLM_LKF_ORPHAN) {
272 error = dlm_user_adopt_orphan(ls, ua,
273 params->mode, params->flags,
274 params->name, params->namelen,
275 (unsigned long) params->timeout,
276 &lkid);
277 if (!error)
278 error = lkid;
279 } else {
280 error = dlm_user_request(ls, ua,
281 params->mode, params->flags,
282 params->name, params->namelen,
283 (unsigned long) params->timeout);
284 if (!error)
285 error = ua->lksb.sb_lkid;
286 }
287 out:
288 dlm_put_lockspace(ls);
289 return error;
290}
291
292static int device_user_unlock(struct dlm_user_proc *proc,
293 struct dlm_lock_params *params)
294{
295 struct dlm_ls *ls;
296 struct dlm_user_args *ua;
297 int error = -ENOMEM;
298
299 ls = dlm_find_lockspace_local(proc->lockspace);
300 if (!ls)
301 return -ENOENT;
302
303 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
304 if (!ua)
305 goto out;
306 ua->proc = proc;
307 ua->user_lksb = params->lksb;
308 ua->castparam = params->castparam;
309 ua->castaddr = params->castaddr;
310
311 if (params->flags & DLM_LKF_CANCEL)
312 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
313 else
314 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
315 params->lvb);
316 out:
317 dlm_put_lockspace(ls);
318 return error;
319}
320
321static int device_user_deadlock(struct dlm_user_proc *proc,
322 struct dlm_lock_params *params)
323{
324 struct dlm_ls *ls;
325 int error;
326
327 ls = dlm_find_lockspace_local(proc->lockspace);
328 if (!ls)
329 return -ENOENT;
330
331 error = dlm_user_deadlock(ls, params->flags, params->lkid);
332
333 dlm_put_lockspace(ls);
334 return error;
335}
336
337static int dlm_device_register(struct dlm_ls *ls, char *name)
338{
339 int error, len;
340
341 /* The device is already registered. This happens when the
342 lockspace is created multiple times from userspace. */
343 if (ls->ls_device.name)
344 return 0;
345
346 error = -ENOMEM;
347 len = strlen(name) + strlen(name_prefix) + 2;
348 ls->ls_device.name = kzalloc(len, GFP_NOFS);
349 if (!ls->ls_device.name)
350 goto fail;
351
352 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
353 name);
354 ls->ls_device.fops = &device_fops;
355 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
356
357 error = misc_register(&ls->ls_device);
358 if (error) {
359 kfree(ls->ls_device.name);
360 /* this has to be set to NULL
361 * to avoid a double-free in dlm_device_deregister
362 */
363 ls->ls_device.name = NULL;
364 }
365fail:
366 return error;
367}
368
369int dlm_device_deregister(struct dlm_ls *ls)
370{
371 /* The device is not registered. This happens when the lockspace
372 was never used from userspace, or when device_create_lockspace()
373 calls dlm_release_lockspace() after the register fails. */
374 if (!ls->ls_device.name)
375 return 0;
376
377 misc_deregister(&ls->ls_device);
378 kfree(ls->ls_device.name);
379 return 0;
380}
381
382static int device_user_purge(struct dlm_user_proc *proc,
383 struct dlm_purge_params *params)
384{
385 struct dlm_ls *ls;
386 int error;
387
388 ls = dlm_find_lockspace_local(proc->lockspace);
389 if (!ls)
390 return -ENOENT;
391
392 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
393
394 dlm_put_lockspace(ls);
395 return error;
396}
397
398static int device_create_lockspace(struct dlm_lspace_params *params)
399{
400 dlm_lockspace_t *lockspace;
401 struct dlm_ls *ls;
402 int error;
403
404 if (!capable(CAP_SYS_ADMIN))
405 return -EPERM;
406
407 error = dlm_new_lockspace(params->name, NULL, params->flags,
408 DLM_USER_LVB_LEN, NULL, NULL, NULL,
409 &lockspace);
410 if (error)
411 return error;
412
413 ls = dlm_find_lockspace_local(lockspace);
414 if (!ls)
415 return -ENOENT;
416
417 error = dlm_device_register(ls, params->name);
418 dlm_put_lockspace(ls);
419
420 if (error)
421 dlm_release_lockspace(lockspace, 0);
422 else
423 error = ls->ls_device.minor;
424
425 return error;
426}
427
428static int device_remove_lockspace(struct dlm_lspace_params *params)
429{
430 dlm_lockspace_t *lockspace;
431 struct dlm_ls *ls;
432 int error, force = 0;
433
434 if (!capable(CAP_SYS_ADMIN))
435 return -EPERM;
436
437 ls = dlm_find_lockspace_device(params->minor);
438 if (!ls)
439 return -ENOENT;
440
441 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
442 force = 2;
443
444 lockspace = ls->ls_local_handle;
445 dlm_put_lockspace(ls);
446
447 /* The final dlm_release_lockspace waits for references to go to
448 zero, so all processes will need to close their device for the
449 ls before the release will proceed. release also calls the
450 device_deregister above. Converting a positive return value
451 from release to zero means that userspace won't know when its
452 release was the final one, but it shouldn't need to know. */
453
454 error = dlm_release_lockspace(lockspace, force);
455 if (error > 0)
456 error = 0;
457 return error;
458}
459
460/* Check the user's version matches ours */
461static int check_version(struct dlm_write_request *req)
462{
463 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
464 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
465 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
466
467 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
468 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
469 current->comm,
470 task_pid_nr(current),
471 req->version[0],
472 req->version[1],
473 req->version[2],
474 DLM_DEVICE_VERSION_MAJOR,
475 DLM_DEVICE_VERSION_MINOR,
476 DLM_DEVICE_VERSION_PATCH);
477 return -EINVAL;
478 }
479 return 0;
480}
481
482/*
483 * device_write
484 *
485 * device_user_lock
486 * dlm_user_request -> request_lock
487 * dlm_user_convert -> convert_lock
488 *
489 * device_user_unlock
490 * dlm_user_unlock -> unlock_lock
491 * dlm_user_cancel -> cancel_lock
492 *
493 * device_create_lockspace
494 * dlm_new_lockspace
495 *
496 * device_remove_lockspace
497 * dlm_release_lockspace
498 */
499
500/* a write to a lockspace device is a lock or unlock request, a write
501 to the control device is to create/remove a lockspace */
502
503static ssize_t device_write(struct file *file, const char __user *buf,
504 size_t count, loff_t *ppos)
505{
506 struct dlm_user_proc *proc = file->private_data;
507 struct dlm_write_request *kbuf;
508 int error;
509
510#ifdef CONFIG_COMPAT
511 if (count < sizeof(struct dlm_write_request32))
512#else
513 if (count < sizeof(struct dlm_write_request))
514#endif
515 return -EINVAL;
516
517 /*
518 * can't compare against COMPAT/dlm_write_request32 because
519 * we don't yet know if is64bit is zero
520 */
521 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
522 return -EINVAL;
523
524 kbuf = memdup_user_nul(buf, count);
525 if (IS_ERR(kbuf))
526 return PTR_ERR(kbuf);
527
528 if (check_version(kbuf)) {
529 error = -EBADE;
530 goto out_free;
531 }
532
533#ifdef CONFIG_COMPAT
534 if (!kbuf->is64bit) {
535 struct dlm_write_request32 *k32buf;
536 int namelen = 0;
537
538 if (count > sizeof(struct dlm_write_request32))
539 namelen = count - sizeof(struct dlm_write_request32);
540
541 k32buf = (struct dlm_write_request32 *)kbuf;
542
543 /* add 1 after namelen so that the name string is terminated */
544 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
545 GFP_NOFS);
546 if (!kbuf) {
547 kfree(k32buf);
548 return -ENOMEM;
549 }
550
551 if (proc)
552 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
553
554 compat_input(kbuf, k32buf, namelen);
555 kfree(k32buf);
556 }
557#endif
558
559 /* do we really need this? can a write happen after a close? */
560 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
561 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
562 error = -EINVAL;
563 goto out_free;
564 }
565
566 error = -EINVAL;
567
568 switch (kbuf->cmd)
569 {
570 case DLM_USER_LOCK:
571 if (!proc) {
572 log_print("no locking on control device");
573 goto out_free;
574 }
575 error = device_user_lock(proc, &kbuf->i.lock);
576 break;
577
578 case DLM_USER_UNLOCK:
579 if (!proc) {
580 log_print("no locking on control device");
581 goto out_free;
582 }
583 error = device_user_unlock(proc, &kbuf->i.lock);
584 break;
585
586 case DLM_USER_DEADLOCK:
587 if (!proc) {
588 log_print("no locking on control device");
589 goto out_free;
590 }
591 error = device_user_deadlock(proc, &kbuf->i.lock);
592 break;
593
594 case DLM_USER_CREATE_LOCKSPACE:
595 if (proc) {
596 log_print("create/remove only on control device");
597 goto out_free;
598 }
599 error = device_create_lockspace(&kbuf->i.lspace);
600 break;
601
602 case DLM_USER_REMOVE_LOCKSPACE:
603 if (proc) {
604 log_print("create/remove only on control device");
605 goto out_free;
606 }
607 error = device_remove_lockspace(&kbuf->i.lspace);
608 break;
609
610 case DLM_USER_PURGE:
611 if (!proc) {
612 log_print("no locking on control device");
613 goto out_free;
614 }
615 error = device_user_purge(proc, &kbuf->i.purge);
616 break;
617
618 default:
619 log_print("Unknown command passed to DLM device : %d\n",
620 kbuf->cmd);
621 }
622
623 out_free:
624 kfree(kbuf);
625 return error;
626}
627
628/* Every process that opens the lockspace device has its own "proc" structure
629 hanging off the open file that's used to keep track of locks owned by the
630 process and asts that need to be delivered to the process. */
631
632static int device_open(struct inode *inode, struct file *file)
633{
634 struct dlm_user_proc *proc;
635 struct dlm_ls *ls;
636
637 ls = dlm_find_lockspace_device(iminor(inode));
638 if (!ls)
639 return -ENOENT;
640
641 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
642 if (!proc) {
643 dlm_put_lockspace(ls);
644 return -ENOMEM;
645 }
646
647 proc->lockspace = ls->ls_local_handle;
648 INIT_LIST_HEAD(&proc->asts);
649 INIT_LIST_HEAD(&proc->locks);
650 INIT_LIST_HEAD(&proc->unlocking);
651 spin_lock_init(&proc->asts_spin);
652 spin_lock_init(&proc->locks_spin);
653 init_waitqueue_head(&proc->wait);
654 file->private_data = proc;
655
656 return 0;
657}
658
659static int device_close(struct inode *inode, struct file *file)
660{
661 struct dlm_user_proc *proc = file->private_data;
662 struct dlm_ls *ls;
663
664 ls = dlm_find_lockspace_local(proc->lockspace);
665 if (!ls)
666 return -ENOENT;
667
668 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
669
670 dlm_clear_proc_locks(ls, proc);
671
672 /* at this point no more lkb's should exist for this lockspace,
673 so there's no chance of dlm_user_add_ast() being called and
674 looking for lkb->ua->proc */
675
676 kfree(proc);
677 file->private_data = NULL;
678
679 dlm_put_lockspace(ls);
680 dlm_put_lockspace(ls); /* for the find in device_open() */
681
682 /* FIXME: AUTOFREE: if this ls is no longer used do
683 device_remove_lockspace() */
684
685 return 0;
686}
687
688static int copy_result_to_user(struct dlm_user_args *ua, int compat,
689 uint32_t flags, int mode, int copy_lvb,
690 char __user *buf, size_t count)
691{
692#ifdef CONFIG_COMPAT
693 struct dlm_lock_result32 result32;
694#endif
695 struct dlm_lock_result result;
696 void *resultptr;
697 int error=0;
698 int len;
699 int struct_len;
700
701 memset(&result, 0, sizeof(struct dlm_lock_result));
702 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
703 result.version[1] = DLM_DEVICE_VERSION_MINOR;
704 result.version[2] = DLM_DEVICE_VERSION_PATCH;
705 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
706 result.user_lksb = ua->user_lksb;
707
708 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
709 in a conversion unless the conversion is successful. See code
710 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
711 notes that a new blocking AST address and parameter are set even if
712 the conversion fails, so maybe we should just do that. */
713
714 if (flags & DLM_CB_BAST) {
715 result.user_astaddr = ua->bastaddr;
716 result.user_astparam = ua->bastparam;
717 result.bast_mode = mode;
718 } else {
719 result.user_astaddr = ua->castaddr;
720 result.user_astparam = ua->castparam;
721 }
722
723#ifdef CONFIG_COMPAT
724 if (compat)
725 len = sizeof(struct dlm_lock_result32);
726 else
727#endif
728 len = sizeof(struct dlm_lock_result);
729 struct_len = len;
730
731 /* copy lvb to userspace if there is one, it's been updated, and
732 the user buffer has space for it */
733
734 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
735 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
736 DLM_USER_LVB_LEN)) {
737 error = -EFAULT;
738 goto out;
739 }
740
741 result.lvb_offset = len;
742 len += DLM_USER_LVB_LEN;
743 }
744
745 result.length = len;
746 resultptr = &result;
747#ifdef CONFIG_COMPAT
748 if (compat) {
749 compat_output(&result, &result32);
750 resultptr = &result32;
751 }
752#endif
753
754 if (copy_to_user(buf, resultptr, struct_len))
755 error = -EFAULT;
756 else
757 error = len;
758 out:
759 return error;
760}
761
762static int copy_version_to_user(char __user *buf, size_t count)
763{
764 struct dlm_device_version ver;
765
766 memset(&ver, 0, sizeof(struct dlm_device_version));
767 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
768 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
769 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
770
771 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
772 return -EFAULT;
773 return sizeof(struct dlm_device_version);
774}
775
776/* a read returns a single ast described in a struct dlm_lock_result */
777
778static ssize_t device_read(struct file *file, char __user *buf, size_t count,
779 loff_t *ppos)
780{
781 struct dlm_user_proc *proc = file->private_data;
782 struct dlm_lkb *lkb;
783 DECLARE_WAITQUEUE(wait, current);
784 struct dlm_callback cb;
785 int rv, resid, copy_lvb = 0;
786 int old_mode, new_mode;
787
788 if (count == sizeof(struct dlm_device_version)) {
789 rv = copy_version_to_user(buf, count);
790 return rv;
791 }
792
793 if (!proc) {
794 log_print("non-version read from control device %zu", count);
795 return -EINVAL;
796 }
797
798#ifdef CONFIG_COMPAT
799 if (count < sizeof(struct dlm_lock_result32))
800#else
801 if (count < sizeof(struct dlm_lock_result))
802#endif
803 return -EINVAL;
804
805 try_another:
806
807 /* do we really need this? can a read happen after a close? */
808 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
809 return -EINVAL;
810
811 spin_lock(&proc->asts_spin);
812 if (list_empty(&proc->asts)) {
813 if (file->f_flags & O_NONBLOCK) {
814 spin_unlock(&proc->asts_spin);
815 return -EAGAIN;
816 }
817
818 add_wait_queue(&proc->wait, &wait);
819
820 repeat:
821 set_current_state(TASK_INTERRUPTIBLE);
822 if (list_empty(&proc->asts) && !signal_pending(current)) {
823 spin_unlock(&proc->asts_spin);
824 schedule();
825 spin_lock(&proc->asts_spin);
826 goto repeat;
827 }
828 set_current_state(TASK_RUNNING);
829 remove_wait_queue(&proc->wait, &wait);
830
831 if (signal_pending(current)) {
832 spin_unlock(&proc->asts_spin);
833 return -ERESTARTSYS;
834 }
835 }
836
837 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
838 without removing lkb_cb_list; so empty lkb_cb_list is always
839 consistent with empty lkb_callbacks */
840
841 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
842
843 /* rem_lkb_callback sets a new lkb_last_cast */
844 old_mode = lkb->lkb_last_cast.mode;
845
846 rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
847 if (rv < 0) {
848 /* this shouldn't happen; lkb should have been removed from
849 list when resid was zero */
850 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
851 list_del_init(&lkb->lkb_cb_list);
852 spin_unlock(&proc->asts_spin);
853 /* removes ref for proc->asts, may cause lkb to be freed */
854 dlm_put_lkb(lkb);
855 goto try_another;
856 }
857 if (!resid)
858 list_del_init(&lkb->lkb_cb_list);
859 spin_unlock(&proc->asts_spin);
860
861 if (cb.flags & DLM_CB_SKIP) {
862 /* removes ref for proc->asts, may cause lkb to be freed */
863 if (!resid)
864 dlm_put_lkb(lkb);
865 goto try_another;
866 }
867
868 if (cb.flags & DLM_CB_CAST) {
869 new_mode = cb.mode;
870
871 if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
872 dlm_lvb_operations[old_mode + 1][new_mode + 1])
873 copy_lvb = 1;
874
875 lkb->lkb_lksb->sb_status = cb.sb_status;
876 lkb->lkb_lksb->sb_flags = cb.sb_flags;
877 }
878
879 rv = copy_result_to_user(lkb->lkb_ua,
880 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
881 cb.flags, cb.mode, copy_lvb, buf, count);
882
883 /* removes ref for proc->asts, may cause lkb to be freed */
884 if (!resid)
885 dlm_put_lkb(lkb);
886
887 return rv;
888}
889
890static __poll_t device_poll(struct file *file, poll_table *wait)
891{
892 struct dlm_user_proc *proc = file->private_data;
893
894 poll_wait(file, &proc->wait, wait);
895
896 spin_lock(&proc->asts_spin);
897 if (!list_empty(&proc->asts)) {
898 spin_unlock(&proc->asts_spin);
899 return EPOLLIN | EPOLLRDNORM;
900 }
901 spin_unlock(&proc->asts_spin);
902 return 0;
903}
904
905int dlm_user_daemon_available(void)
906{
907 /* dlm_controld hasn't started (or, has started, but not
908 properly populated configfs) */
909
910 if (!dlm_our_nodeid())
911 return 0;
912
913 /* This is to deal with versions of dlm_controld that don't
914 know about the monitor device. We assume that if the
915 dlm_controld was started (above), but the monitor device
916 was never opened, that it's an old version. dlm_controld
917 should open the monitor device before populating configfs. */
918
919 if (dlm_monitor_unused)
920 return 1;
921
922 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
923}
924
925static int ctl_device_open(struct inode *inode, struct file *file)
926{
927 file->private_data = NULL;
928 return 0;
929}
930
931static int ctl_device_close(struct inode *inode, struct file *file)
932{
933 return 0;
934}
935
936static int monitor_device_open(struct inode *inode, struct file *file)
937{
938 atomic_inc(&dlm_monitor_opened);
939 dlm_monitor_unused = 0;
940 return 0;
941}
942
943static int monitor_device_close(struct inode *inode, struct file *file)
944{
945 if (atomic_dec_and_test(&dlm_monitor_opened))
946 dlm_stop_lockspaces();
947 return 0;
948}
949
950static const struct file_operations device_fops = {
951 .open = device_open,
952 .release = device_close,
953 .read = device_read,
954 .write = device_write,
955 .poll = device_poll,
956 .owner = THIS_MODULE,
957 .llseek = noop_llseek,
958};
959
960static const struct file_operations ctl_device_fops = {
961 .open = ctl_device_open,
962 .release = ctl_device_close,
963 .read = device_read,
964 .write = device_write,
965 .owner = THIS_MODULE,
966 .llseek = noop_llseek,
967};
968
969static struct miscdevice ctl_device = {
970 .name = "dlm-control",
971 .fops = &ctl_device_fops,
972 .minor = MISC_DYNAMIC_MINOR,
973};
974
975static const struct file_operations monitor_device_fops = {
976 .open = monitor_device_open,
977 .release = monitor_device_close,
978 .owner = THIS_MODULE,
979 .llseek = noop_llseek,
980};
981
982static struct miscdevice monitor_device = {
983 .name = "dlm-monitor",
984 .fops = &monitor_device_fops,
985 .minor = MISC_DYNAMIC_MINOR,
986};
987
988int __init dlm_user_init(void)
989{
990 int error;
991
992 atomic_set(&dlm_monitor_opened, 0);
993
994 error = misc_register(&ctl_device);
995 if (error) {
996 log_print("misc_register failed for control device");
997 goto out;
998 }
999
1000 error = misc_register(&monitor_device);
1001 if (error) {
1002 log_print("misc_register failed for monitor device");
1003 misc_deregister(&ctl_device);
1004 }
1005 out:
1006 return error;
1007}
1008
1009void dlm_user_exit(void)
1010{
1011 misc_deregister(&ctl_device);
1012 misc_deregister(&monitor_device);
1013}
1014
1/*
2 * Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved.
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#include <linux/slab.h>
21
22#include "dlm_internal.h"
23#include "lockspace.h"
24#include "lock.h"
25#include "lvb_table.h"
26#include "user.h"
27#include "ast.h"
28
29static const char name_prefix[] = "dlm";
30static const struct file_operations device_fops;
31static atomic_t dlm_monitor_opened;
32static int dlm_monitor_unused = 1;
33
34#ifdef CONFIG_COMPAT
35
36struct dlm_lock_params32 {
37 __u8 mode;
38 __u8 namelen;
39 __u16 unused;
40 __u32 flags;
41 __u32 lkid;
42 __u32 parent;
43 __u64 xid;
44 __u64 timeout;
45 __u32 castparam;
46 __u32 castaddr;
47 __u32 bastparam;
48 __u32 bastaddr;
49 __u32 lksb;
50 char lvb[DLM_USER_LVB_LEN];
51 char name[0];
52};
53
54struct dlm_write_request32 {
55 __u32 version[3];
56 __u8 cmd;
57 __u8 is64bit;
58 __u8 unused[2];
59
60 union {
61 struct dlm_lock_params32 lock;
62 struct dlm_lspace_params lspace;
63 struct dlm_purge_params purge;
64 } i;
65};
66
67struct dlm_lksb32 {
68 __u32 sb_status;
69 __u32 sb_lkid;
70 __u8 sb_flags;
71 __u32 sb_lvbptr;
72};
73
74struct dlm_lock_result32 {
75 __u32 version[3];
76 __u32 length;
77 __u32 user_astaddr;
78 __u32 user_astparam;
79 __u32 user_lksb;
80 struct dlm_lksb32 lksb;
81 __u8 bast_mode;
82 __u8 unused[3];
83 /* Offsets may be zero if no data is present */
84 __u32 lvb_offset;
85};
86
87static void compat_input(struct dlm_write_request *kb,
88 struct dlm_write_request32 *kb32,
89 int namelen)
90{
91 kb->version[0] = kb32->version[0];
92 kb->version[1] = kb32->version[1];
93 kb->version[2] = kb32->version[2];
94
95 kb->cmd = kb32->cmd;
96 kb->is64bit = kb32->is64bit;
97 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
98 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
99 kb->i.lspace.flags = kb32->i.lspace.flags;
100 kb->i.lspace.minor = kb32->i.lspace.minor;
101 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
102 } else if (kb->cmd == DLM_USER_PURGE) {
103 kb->i.purge.nodeid = kb32->i.purge.nodeid;
104 kb->i.purge.pid = kb32->i.purge.pid;
105 } else {
106 kb->i.lock.mode = kb32->i.lock.mode;
107 kb->i.lock.namelen = kb32->i.lock.namelen;
108 kb->i.lock.flags = kb32->i.lock.flags;
109 kb->i.lock.lkid = kb32->i.lock.lkid;
110 kb->i.lock.parent = kb32->i.lock.parent;
111 kb->i.lock.xid = kb32->i.lock.xid;
112 kb->i.lock.timeout = kb32->i.lock.timeout;
113 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
114 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
115 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
116 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
117 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
118 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
119 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
120 }
121}
122
123static void compat_output(struct dlm_lock_result *res,
124 struct dlm_lock_result32 *res32)
125{
126 res32->version[0] = res->version[0];
127 res32->version[1] = res->version[1];
128 res32->version[2] = res->version[2];
129
130 res32->user_astaddr = (__u32)(long)res->user_astaddr;
131 res32->user_astparam = (__u32)(long)res->user_astparam;
132 res32->user_lksb = (__u32)(long)res->user_lksb;
133 res32->bast_mode = res->bast_mode;
134
135 res32->lvb_offset = res->lvb_offset;
136 res32->length = res->length;
137
138 res32->lksb.sb_status = res->lksb.sb_status;
139 res32->lksb.sb_flags = res->lksb.sb_flags;
140 res32->lksb.sb_lkid = res->lksb.sb_lkid;
141 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
142}
143#endif
144
145/* Figure out if this lock is at the end of its life and no longer
146 available for the application to use. The lkb still exists until
147 the final ast is read. A lock becomes EOL in three situations:
148 1. a noqueue request fails with EAGAIN
149 2. an unlock completes with EUNLOCK
150 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
151 An EOL lock needs to be removed from the process's list of locks.
152 And we can't allow any new operation on an EOL lock. This is
153 not related to the lifetime of the lkb struct which is managed
154 entirely by refcount. */
155
156static int lkb_is_endoflife(int mode, int status)
157{
158 switch (status) {
159 case -DLM_EUNLOCK:
160 return 1;
161 case -DLM_ECANCEL:
162 case -ETIMEDOUT:
163 case -EDEADLK:
164 case -EAGAIN:
165 if (mode == DLM_LOCK_IV)
166 return 1;
167 break;
168 }
169 return 0;
170}
171
172/* we could possibly check if the cancel of an orphan has resulted in the lkb
173 being removed and then remove that lkb from the orphans list and free it */
174
175void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
176 int status, uint32_t sbflags, uint64_t seq)
177{
178 struct dlm_ls *ls;
179 struct dlm_user_args *ua;
180 struct dlm_user_proc *proc;
181 int rv;
182
183 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
184 return;
185
186 ls = lkb->lkb_resource->res_ls;
187 mutex_lock(&ls->ls_clear_proc_locks);
188
189 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
190 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
191 lkb->ua so we can't try to use it. This second check is necessary
192 for cases where a completion ast is received for an operation that
193 began before clear_proc_locks did its cancel/unlock. */
194
195 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
196 goto out;
197
198 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
199 ua = lkb->lkb_ua;
200 proc = ua->proc;
201
202 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
203 goto out;
204
205 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
206 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
207
208 spin_lock(&proc->asts_spin);
209
210 rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
211 if (rv < 0) {
212 spin_unlock(&proc->asts_spin);
213 goto out;
214 }
215
216 if (list_empty(&lkb->lkb_cb_list)) {
217 kref_get(&lkb->lkb_ref);
218 list_add_tail(&lkb->lkb_cb_list, &proc->asts);
219 wake_up_interruptible(&proc->wait);
220 }
221 spin_unlock(&proc->asts_spin);
222
223 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
224 /* N.B. spin_lock locks_spin, not asts_spin */
225 spin_lock(&proc->locks_spin);
226 if (!list_empty(&lkb->lkb_ownqueue)) {
227 list_del_init(&lkb->lkb_ownqueue);
228 dlm_put_lkb(lkb);
229 }
230 spin_unlock(&proc->locks_spin);
231 }
232 out:
233 mutex_unlock(&ls->ls_clear_proc_locks);
234}
235
236static int device_user_lock(struct dlm_user_proc *proc,
237 struct dlm_lock_params *params)
238{
239 struct dlm_ls *ls;
240 struct dlm_user_args *ua;
241 int error = -ENOMEM;
242
243 ls = dlm_find_lockspace_local(proc->lockspace);
244 if (!ls)
245 return -ENOENT;
246
247 if (!params->castaddr || !params->lksb) {
248 error = -EINVAL;
249 goto out;
250 }
251
252 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
253 if (!ua)
254 goto out;
255 ua->proc = proc;
256 ua->user_lksb = params->lksb;
257 ua->castparam = params->castparam;
258 ua->castaddr = params->castaddr;
259 ua->bastparam = params->bastparam;
260 ua->bastaddr = params->bastaddr;
261 ua->xid = params->xid;
262
263 if (params->flags & DLM_LKF_CONVERT)
264 error = dlm_user_convert(ls, ua,
265 params->mode, params->flags,
266 params->lkid, params->lvb,
267 (unsigned long) params->timeout);
268 else {
269 error = dlm_user_request(ls, ua,
270 params->mode, params->flags,
271 params->name, params->namelen,
272 (unsigned long) params->timeout);
273 if (!error)
274 error = ua->lksb.sb_lkid;
275 }
276 out:
277 dlm_put_lockspace(ls);
278 return error;
279}
280
281static int device_user_unlock(struct dlm_user_proc *proc,
282 struct dlm_lock_params *params)
283{
284 struct dlm_ls *ls;
285 struct dlm_user_args *ua;
286 int error = -ENOMEM;
287
288 ls = dlm_find_lockspace_local(proc->lockspace);
289 if (!ls)
290 return -ENOENT;
291
292 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
293 if (!ua)
294 goto out;
295 ua->proc = proc;
296 ua->user_lksb = params->lksb;
297 ua->castparam = params->castparam;
298 ua->castaddr = params->castaddr;
299
300 if (params->flags & DLM_LKF_CANCEL)
301 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
302 else
303 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
304 params->lvb);
305 out:
306 dlm_put_lockspace(ls);
307 return error;
308}
309
310static int device_user_deadlock(struct dlm_user_proc *proc,
311 struct dlm_lock_params *params)
312{
313 struct dlm_ls *ls;
314 int error;
315
316 ls = dlm_find_lockspace_local(proc->lockspace);
317 if (!ls)
318 return -ENOENT;
319
320 error = dlm_user_deadlock(ls, params->flags, params->lkid);
321
322 dlm_put_lockspace(ls);
323 return error;
324}
325
326static int dlm_device_register(struct dlm_ls *ls, char *name)
327{
328 int error, len;
329
330 /* The device is already registered. This happens when the
331 lockspace is created multiple times from userspace. */
332 if (ls->ls_device.name)
333 return 0;
334
335 error = -ENOMEM;
336 len = strlen(name) + strlen(name_prefix) + 2;
337 ls->ls_device.name = kzalloc(len, GFP_NOFS);
338 if (!ls->ls_device.name)
339 goto fail;
340
341 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
342 name);
343 ls->ls_device.fops = &device_fops;
344 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
345
346 error = misc_register(&ls->ls_device);
347 if (error) {
348 kfree(ls->ls_device.name);
349 }
350fail:
351 return error;
352}
353
354int dlm_device_deregister(struct dlm_ls *ls)
355{
356 int error;
357
358 /* The device is not registered. This happens when the lockspace
359 was never used from userspace, or when device_create_lockspace()
360 calls dlm_release_lockspace() after the register fails. */
361 if (!ls->ls_device.name)
362 return 0;
363
364 error = misc_deregister(&ls->ls_device);
365 if (!error)
366 kfree(ls->ls_device.name);
367 return error;
368}
369
370static int device_user_purge(struct dlm_user_proc *proc,
371 struct dlm_purge_params *params)
372{
373 struct dlm_ls *ls;
374 int error;
375
376 ls = dlm_find_lockspace_local(proc->lockspace);
377 if (!ls)
378 return -ENOENT;
379
380 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
381
382 dlm_put_lockspace(ls);
383 return error;
384}
385
386static int device_create_lockspace(struct dlm_lspace_params *params)
387{
388 dlm_lockspace_t *lockspace;
389 struct dlm_ls *ls;
390 int error;
391
392 if (!capable(CAP_SYS_ADMIN))
393 return -EPERM;
394
395 error = dlm_new_lockspace(params->name, strlen(params->name),
396 &lockspace, params->flags, DLM_USER_LVB_LEN);
397 if (error)
398 return error;
399
400 ls = dlm_find_lockspace_local(lockspace);
401 if (!ls)
402 return -ENOENT;
403
404 error = dlm_device_register(ls, params->name);
405 dlm_put_lockspace(ls);
406
407 if (error)
408 dlm_release_lockspace(lockspace, 0);
409 else
410 error = ls->ls_device.minor;
411
412 return error;
413}
414
415static int device_remove_lockspace(struct dlm_lspace_params *params)
416{
417 dlm_lockspace_t *lockspace;
418 struct dlm_ls *ls;
419 int error, force = 0;
420
421 if (!capable(CAP_SYS_ADMIN))
422 return -EPERM;
423
424 ls = dlm_find_lockspace_device(params->minor);
425 if (!ls)
426 return -ENOENT;
427
428 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
429 force = 2;
430
431 lockspace = ls->ls_local_handle;
432 dlm_put_lockspace(ls);
433
434 /* The final dlm_release_lockspace waits for references to go to
435 zero, so all processes will need to close their device for the
436 ls before the release will proceed. release also calls the
437 device_deregister above. Converting a positive return value
438 from release to zero means that userspace won't know when its
439 release was the final one, but it shouldn't need to know. */
440
441 error = dlm_release_lockspace(lockspace, force);
442 if (error > 0)
443 error = 0;
444 return error;
445}
446
447/* Check the user's version matches ours */
448static int check_version(struct dlm_write_request *req)
449{
450 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
451 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
452 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
453
454 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
455 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
456 current->comm,
457 task_pid_nr(current),
458 req->version[0],
459 req->version[1],
460 req->version[2],
461 DLM_DEVICE_VERSION_MAJOR,
462 DLM_DEVICE_VERSION_MINOR,
463 DLM_DEVICE_VERSION_PATCH);
464 return -EINVAL;
465 }
466 return 0;
467}
468
469/*
470 * device_write
471 *
472 * device_user_lock
473 * dlm_user_request -> request_lock
474 * dlm_user_convert -> convert_lock
475 *
476 * device_user_unlock
477 * dlm_user_unlock -> unlock_lock
478 * dlm_user_cancel -> cancel_lock
479 *
480 * device_create_lockspace
481 * dlm_new_lockspace
482 *
483 * device_remove_lockspace
484 * dlm_release_lockspace
485 */
486
487/* a write to a lockspace device is a lock or unlock request, a write
488 to the control device is to create/remove a lockspace */
489
490static ssize_t device_write(struct file *file, const char __user *buf,
491 size_t count, loff_t *ppos)
492{
493 struct dlm_user_proc *proc = file->private_data;
494 struct dlm_write_request *kbuf;
495 sigset_t tmpsig, allsigs;
496 int error;
497
498#ifdef CONFIG_COMPAT
499 if (count < sizeof(struct dlm_write_request32))
500#else
501 if (count < sizeof(struct dlm_write_request))
502#endif
503 return -EINVAL;
504
505 kbuf = kzalloc(count + 1, GFP_NOFS);
506 if (!kbuf)
507 return -ENOMEM;
508
509 if (copy_from_user(kbuf, buf, count)) {
510 error = -EFAULT;
511 goto out_free;
512 }
513
514 if (check_version(kbuf)) {
515 error = -EBADE;
516 goto out_free;
517 }
518
519#ifdef CONFIG_COMPAT
520 if (!kbuf->is64bit) {
521 struct dlm_write_request32 *k32buf;
522 int namelen = 0;
523
524 if (count > sizeof(struct dlm_write_request32))
525 namelen = count - sizeof(struct dlm_write_request32);
526
527 k32buf = (struct dlm_write_request32 *)kbuf;
528
529 /* add 1 after namelen so that the name string is terminated */
530 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
531 GFP_NOFS);
532 if (!kbuf) {
533 kfree(k32buf);
534 return -ENOMEM;
535 }
536
537 if (proc)
538 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
539
540 compat_input(kbuf, k32buf, namelen);
541 kfree(k32buf);
542 }
543#endif
544
545 /* do we really need this? can a write happen after a close? */
546 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
547 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
548 error = -EINVAL;
549 goto out_free;
550 }
551
552 sigfillset(&allsigs);
553 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
554
555 error = -EINVAL;
556
557 switch (kbuf->cmd)
558 {
559 case DLM_USER_LOCK:
560 if (!proc) {
561 log_print("no locking on control device");
562 goto out_sig;
563 }
564 error = device_user_lock(proc, &kbuf->i.lock);
565 break;
566
567 case DLM_USER_UNLOCK:
568 if (!proc) {
569 log_print("no locking on control device");
570 goto out_sig;
571 }
572 error = device_user_unlock(proc, &kbuf->i.lock);
573 break;
574
575 case DLM_USER_DEADLOCK:
576 if (!proc) {
577 log_print("no locking on control device");
578 goto out_sig;
579 }
580 error = device_user_deadlock(proc, &kbuf->i.lock);
581 break;
582
583 case DLM_USER_CREATE_LOCKSPACE:
584 if (proc) {
585 log_print("create/remove only on control device");
586 goto out_sig;
587 }
588 error = device_create_lockspace(&kbuf->i.lspace);
589 break;
590
591 case DLM_USER_REMOVE_LOCKSPACE:
592 if (proc) {
593 log_print("create/remove only on control device");
594 goto out_sig;
595 }
596 error = device_remove_lockspace(&kbuf->i.lspace);
597 break;
598
599 case DLM_USER_PURGE:
600 if (!proc) {
601 log_print("no locking on control device");
602 goto out_sig;
603 }
604 error = device_user_purge(proc, &kbuf->i.purge);
605 break;
606
607 default:
608 log_print("Unknown command passed to DLM device : %d\n",
609 kbuf->cmd);
610 }
611
612 out_sig:
613 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
614 out_free:
615 kfree(kbuf);
616 return error;
617}
618
619/* Every process that opens the lockspace device has its own "proc" structure
620 hanging off the open file that's used to keep track of locks owned by the
621 process and asts that need to be delivered to the process. */
622
623static int device_open(struct inode *inode, struct file *file)
624{
625 struct dlm_user_proc *proc;
626 struct dlm_ls *ls;
627
628 ls = dlm_find_lockspace_device(iminor(inode));
629 if (!ls)
630 return -ENOENT;
631
632 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
633 if (!proc) {
634 dlm_put_lockspace(ls);
635 return -ENOMEM;
636 }
637
638 proc->lockspace = ls->ls_local_handle;
639 INIT_LIST_HEAD(&proc->asts);
640 INIT_LIST_HEAD(&proc->locks);
641 INIT_LIST_HEAD(&proc->unlocking);
642 spin_lock_init(&proc->asts_spin);
643 spin_lock_init(&proc->locks_spin);
644 init_waitqueue_head(&proc->wait);
645 file->private_data = proc;
646
647 return 0;
648}
649
650static int device_close(struct inode *inode, struct file *file)
651{
652 struct dlm_user_proc *proc = file->private_data;
653 struct dlm_ls *ls;
654 sigset_t tmpsig, allsigs;
655
656 ls = dlm_find_lockspace_local(proc->lockspace);
657 if (!ls)
658 return -ENOENT;
659
660 sigfillset(&allsigs);
661 sigprocmask(SIG_BLOCK, &allsigs, &tmpsig);
662
663 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
664
665 dlm_clear_proc_locks(ls, proc);
666
667 /* at this point no more lkb's should exist for this lockspace,
668 so there's no chance of dlm_user_add_ast() being called and
669 looking for lkb->ua->proc */
670
671 kfree(proc);
672 file->private_data = NULL;
673
674 dlm_put_lockspace(ls);
675 dlm_put_lockspace(ls); /* for the find in device_open() */
676
677 /* FIXME: AUTOFREE: if this ls is no longer used do
678 device_remove_lockspace() */
679
680 sigprocmask(SIG_SETMASK, &tmpsig, NULL);
681 recalc_sigpending();
682
683 return 0;
684}
685
686static int copy_result_to_user(struct dlm_user_args *ua, int compat,
687 uint32_t flags, int mode, int copy_lvb,
688 char __user *buf, size_t count)
689{
690#ifdef CONFIG_COMPAT
691 struct dlm_lock_result32 result32;
692#endif
693 struct dlm_lock_result result;
694 void *resultptr;
695 int error=0;
696 int len;
697 int struct_len;
698
699 memset(&result, 0, sizeof(struct dlm_lock_result));
700 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
701 result.version[1] = DLM_DEVICE_VERSION_MINOR;
702 result.version[2] = DLM_DEVICE_VERSION_PATCH;
703 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
704 result.user_lksb = ua->user_lksb;
705
706 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
707 in a conversion unless the conversion is successful. See code
708 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
709 notes that a new blocking AST address and parameter are set even if
710 the conversion fails, so maybe we should just do that. */
711
712 if (flags & DLM_CB_BAST) {
713 result.user_astaddr = ua->bastaddr;
714 result.user_astparam = ua->bastparam;
715 result.bast_mode = mode;
716 } else {
717 result.user_astaddr = ua->castaddr;
718 result.user_astparam = ua->castparam;
719 }
720
721#ifdef CONFIG_COMPAT
722 if (compat)
723 len = sizeof(struct dlm_lock_result32);
724 else
725#endif
726 len = sizeof(struct dlm_lock_result);
727 struct_len = len;
728
729 /* copy lvb to userspace if there is one, it's been updated, and
730 the user buffer has space for it */
731
732 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
733 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
734 DLM_USER_LVB_LEN)) {
735 error = -EFAULT;
736 goto out;
737 }
738
739 result.lvb_offset = len;
740 len += DLM_USER_LVB_LEN;
741 }
742
743 result.length = len;
744 resultptr = &result;
745#ifdef CONFIG_COMPAT
746 if (compat) {
747 compat_output(&result, &result32);
748 resultptr = &result32;
749 }
750#endif
751
752 if (copy_to_user(buf, resultptr, struct_len))
753 error = -EFAULT;
754 else
755 error = len;
756 out:
757 return error;
758}
759
760static int copy_version_to_user(char __user *buf, size_t count)
761{
762 struct dlm_device_version ver;
763
764 memset(&ver, 0, sizeof(struct dlm_device_version));
765 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
766 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
767 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
768
769 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
770 return -EFAULT;
771 return sizeof(struct dlm_device_version);
772}
773
774/* a read returns a single ast described in a struct dlm_lock_result */
775
776static ssize_t device_read(struct file *file, char __user *buf, size_t count,
777 loff_t *ppos)
778{
779 struct dlm_user_proc *proc = file->private_data;
780 struct dlm_lkb *lkb;
781 DECLARE_WAITQUEUE(wait, current);
782 struct dlm_callback cb;
783 int rv, resid, copy_lvb = 0;
784
785 if (count == sizeof(struct dlm_device_version)) {
786 rv = copy_version_to_user(buf, count);
787 return rv;
788 }
789
790 if (!proc) {
791 log_print("non-version read from control device %zu", count);
792 return -EINVAL;
793 }
794
795#ifdef CONFIG_COMPAT
796 if (count < sizeof(struct dlm_lock_result32))
797#else
798 if (count < sizeof(struct dlm_lock_result))
799#endif
800 return -EINVAL;
801
802 try_another:
803
804 /* do we really need this? can a read happen after a close? */
805 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
806 return -EINVAL;
807
808 spin_lock(&proc->asts_spin);
809 if (list_empty(&proc->asts)) {
810 if (file->f_flags & O_NONBLOCK) {
811 spin_unlock(&proc->asts_spin);
812 return -EAGAIN;
813 }
814
815 add_wait_queue(&proc->wait, &wait);
816
817 repeat:
818 set_current_state(TASK_INTERRUPTIBLE);
819 if (list_empty(&proc->asts) && !signal_pending(current)) {
820 spin_unlock(&proc->asts_spin);
821 schedule();
822 spin_lock(&proc->asts_spin);
823 goto repeat;
824 }
825 set_current_state(TASK_RUNNING);
826 remove_wait_queue(&proc->wait, &wait);
827
828 if (signal_pending(current)) {
829 spin_unlock(&proc->asts_spin);
830 return -ERESTARTSYS;
831 }
832 }
833
834 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
835 without removing lkb_cb_list; so empty lkb_cb_list is always
836 consistent with empty lkb_callbacks */
837
838 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
839
840 rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
841 if (rv < 0) {
842 /* this shouldn't happen; lkb should have been removed from
843 list when resid was zero */
844 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
845 list_del_init(&lkb->lkb_cb_list);
846 spin_unlock(&proc->asts_spin);
847 /* removes ref for proc->asts, may cause lkb to be freed */
848 dlm_put_lkb(lkb);
849 goto try_another;
850 }
851 if (!resid)
852 list_del_init(&lkb->lkb_cb_list);
853 spin_unlock(&proc->asts_spin);
854
855 if (cb.flags & DLM_CB_SKIP) {
856 /* removes ref for proc->asts, may cause lkb to be freed */
857 if (!resid)
858 dlm_put_lkb(lkb);
859 goto try_another;
860 }
861
862 if (cb.flags & DLM_CB_CAST) {
863 int old_mode, new_mode;
864
865 old_mode = lkb->lkb_last_cast.mode;
866 new_mode = cb.mode;
867
868 if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
869 dlm_lvb_operations[old_mode + 1][new_mode + 1])
870 copy_lvb = 1;
871
872 lkb->lkb_lksb->sb_status = cb.sb_status;
873 lkb->lkb_lksb->sb_flags = cb.sb_flags;
874 }
875
876 rv = copy_result_to_user(lkb->lkb_ua,
877 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
878 cb.flags, cb.mode, copy_lvb, buf, count);
879
880 /* removes ref for proc->asts, may cause lkb to be freed */
881 if (!resid)
882 dlm_put_lkb(lkb);
883
884 return rv;
885}
886
887static unsigned int device_poll(struct file *file, poll_table *wait)
888{
889 struct dlm_user_proc *proc = file->private_data;
890
891 poll_wait(file, &proc->wait, wait);
892
893 spin_lock(&proc->asts_spin);
894 if (!list_empty(&proc->asts)) {
895 spin_unlock(&proc->asts_spin);
896 return POLLIN | POLLRDNORM;
897 }
898 spin_unlock(&proc->asts_spin);
899 return 0;
900}
901
902int dlm_user_daemon_available(void)
903{
904 /* dlm_controld hasn't started (or, has started, but not
905 properly populated configfs) */
906
907 if (!dlm_our_nodeid())
908 return 0;
909
910 /* This is to deal with versions of dlm_controld that don't
911 know about the monitor device. We assume that if the
912 dlm_controld was started (above), but the monitor device
913 was never opened, that it's an old version. dlm_controld
914 should open the monitor device before populating configfs. */
915
916 if (dlm_monitor_unused)
917 return 1;
918
919 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
920}
921
922static int ctl_device_open(struct inode *inode, struct file *file)
923{
924 file->private_data = NULL;
925 return 0;
926}
927
928static int ctl_device_close(struct inode *inode, struct file *file)
929{
930 return 0;
931}
932
933static int monitor_device_open(struct inode *inode, struct file *file)
934{
935 atomic_inc(&dlm_monitor_opened);
936 dlm_monitor_unused = 0;
937 return 0;
938}
939
940static int monitor_device_close(struct inode *inode, struct file *file)
941{
942 if (atomic_dec_and_test(&dlm_monitor_opened))
943 dlm_stop_lockspaces();
944 return 0;
945}
946
947static const struct file_operations device_fops = {
948 .open = device_open,
949 .release = device_close,
950 .read = device_read,
951 .write = device_write,
952 .poll = device_poll,
953 .owner = THIS_MODULE,
954 .llseek = noop_llseek,
955};
956
957static const struct file_operations ctl_device_fops = {
958 .open = ctl_device_open,
959 .release = ctl_device_close,
960 .read = device_read,
961 .write = device_write,
962 .owner = THIS_MODULE,
963 .llseek = noop_llseek,
964};
965
966static struct miscdevice ctl_device = {
967 .name = "dlm-control",
968 .fops = &ctl_device_fops,
969 .minor = MISC_DYNAMIC_MINOR,
970};
971
972static const struct file_operations monitor_device_fops = {
973 .open = monitor_device_open,
974 .release = monitor_device_close,
975 .owner = THIS_MODULE,
976 .llseek = noop_llseek,
977};
978
979static struct miscdevice monitor_device = {
980 .name = "dlm-monitor",
981 .fops = &monitor_device_fops,
982 .minor = MISC_DYNAMIC_MINOR,
983};
984
985int __init dlm_user_init(void)
986{
987 int error;
988
989 atomic_set(&dlm_monitor_opened, 0);
990
991 error = misc_register(&ctl_device);
992 if (error) {
993 log_print("misc_register failed for control device");
994 goto out;
995 }
996
997 error = misc_register(&monitor_device);
998 if (error) {
999 log_print("misc_register failed for monitor device");
1000 misc_deregister(&ctl_device);
1001 }
1002 out:
1003 return error;
1004}
1005
1006void dlm_user_exit(void)
1007{
1008 misc_deregister(&ctl_device);
1009 misc_deregister(&monitor_device);
1010}
1011