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