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/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
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved.
4 */
5
6#include <linux/miscdevice.h>
7#include <linux/init.h>
8#include <linux/wait.h>
9#include <linux/file.h>
10#include <linux/fs.h>
11#include <linux/poll.h>
12#include <linux/signal.h>
13#include <linux/spinlock.h>
14#include <linux/dlm.h>
15#include <linux/dlm_device.h>
16#include <linux/slab.h>
17#include <linux/sched/signal.h>
18
19#include <trace/events/dlm.h>
20
21#include "dlm_internal.h"
22#include "lockspace.h"
23#include "lock.h"
24#include "lvb_table.h"
25#include "user.h"
26#include "ast.h"
27#include "config.h"
28#include "memory.h"
29
30static const char name_prefix[] = "dlm";
31static const struct file_operations device_fops;
32static atomic_t dlm_monitor_opened;
33static int dlm_monitor_unused = 1;
34
35#ifdef CONFIG_COMPAT
36
37struct dlm_lock_params32 {
38 __u8 mode;
39 __u8 namelen;
40 __u16 unused;
41 __u32 flags;
42 __u32 lkid;
43 __u32 parent;
44 __u64 xid;
45 __u64 timeout;
46 __u32 castparam;
47 __u32 castaddr;
48 __u32 bastparam;
49 __u32 bastaddr;
50 __u32 lksb;
51 char lvb[DLM_USER_LVB_LEN];
52 char name[];
53};
54
55struct dlm_write_request32 {
56 __u32 version[3];
57 __u8 cmd;
58 __u8 is64bit;
59 __u8 unused[2];
60
61 union {
62 struct dlm_lock_params32 lock;
63 struct dlm_lspace_params lspace;
64 struct dlm_purge_params purge;
65 } i;
66};
67
68struct dlm_lksb32 {
69 __u32 sb_status;
70 __u32 sb_lkid;
71 __u8 sb_flags;
72 __u32 sb_lvbptr;
73};
74
75struct dlm_lock_result32 {
76 __u32 version[3];
77 __u32 length;
78 __u32 user_astaddr;
79 __u32 user_astparam;
80 __u32 user_lksb;
81 struct dlm_lksb32 lksb;
82 __u8 bast_mode;
83 __u8 unused[3];
84 /* Offsets may be zero if no data is present */
85 __u32 lvb_offset;
86};
87
88static void compat_input(struct dlm_write_request *kb,
89 struct dlm_write_request32 *kb32,
90 int namelen)
91{
92 kb->version[0] = kb32->version[0];
93 kb->version[1] = kb32->version[1];
94 kb->version[2] = kb32->version[2];
95
96 kb->cmd = kb32->cmd;
97 kb->is64bit = kb32->is64bit;
98 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
99 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
100 kb->i.lspace.flags = kb32->i.lspace.flags;
101 kb->i.lspace.minor = kb32->i.lspace.minor;
102 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
103 } else if (kb->cmd == DLM_USER_PURGE) {
104 kb->i.purge.nodeid = kb32->i.purge.nodeid;
105 kb->i.purge.pid = kb32->i.purge.pid;
106 } else {
107 kb->i.lock.mode = kb32->i.lock.mode;
108 kb->i.lock.namelen = kb32->i.lock.namelen;
109 kb->i.lock.flags = kb32->i.lock.flags;
110 kb->i.lock.lkid = kb32->i.lock.lkid;
111 kb->i.lock.parent = kb32->i.lock.parent;
112 kb->i.lock.xid = kb32->i.lock.xid;
113 kb->i.lock.timeout = kb32->i.lock.timeout;
114 kb->i.lock.castparam = (__user void *)(long)kb32->i.lock.castparam;
115 kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr;
116 kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam;
117 kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr;
118 kb->i.lock.lksb = (__user void *)(long)kb32->i.lock.lksb;
119 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
120 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
121 }
122}
123
124static void compat_output(struct dlm_lock_result *res,
125 struct dlm_lock_result32 *res32)
126{
127 memset(res32, 0, sizeof(*res32));
128
129 res32->version[0] = res->version[0];
130 res32->version[1] = res->version[1];
131 res32->version[2] = res->version[2];
132
133 res32->user_astaddr = (__u32)(__force long)res->user_astaddr;
134 res32->user_astparam = (__u32)(__force long)res->user_astparam;
135 res32->user_lksb = (__u32)(__force long)res->user_lksb;
136 res32->bast_mode = res->bast_mode;
137
138 res32->lvb_offset = res->lvb_offset;
139 res32->length = res->length;
140
141 res32->lksb.sb_status = res->lksb.sb_status;
142 res32->lksb.sb_flags = res->lksb.sb_flags;
143 res32->lksb.sb_lkid = res->lksb.sb_lkid;
144 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
145}
146#endif
147
148/* Figure out if this lock is at the end of its life and no longer
149 available for the application to use. The lkb still exists until
150 the final ast is read. A lock becomes EOL in three situations:
151 1. a noqueue request fails with EAGAIN
152 2. an unlock completes with EUNLOCK
153 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
154 An EOL lock needs to be removed from the process's list of locks.
155 And we can't allow any new operation on an EOL lock. This is
156 not related to the lifetime of the lkb struct which is managed
157 entirely by refcount. */
158
159static int lkb_is_endoflife(int mode, int status)
160{
161 switch (status) {
162 case -DLM_EUNLOCK:
163 return 1;
164 case -DLM_ECANCEL:
165 case -ETIMEDOUT:
166 case -EDEADLK:
167 case -EAGAIN:
168 if (mode == DLM_LOCK_IV)
169 return 1;
170 break;
171 }
172 return 0;
173}
174
175/* we could possibly check if the cancel of an orphan has resulted in the lkb
176 being removed and then remove that lkb from the orphans list and free it */
177
178void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
179 int status, uint32_t sbflags)
180{
181 struct dlm_ls *ls;
182 struct dlm_user_args *ua;
183 struct dlm_user_proc *proc;
184 int rv;
185
186 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
187 return;
188
189 ls = lkb->lkb_resource->res_ls;
190 spin_lock(&ls->ls_clear_proc_locks);
191
192 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
193 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
194 lkb->ua so we can't try to use it. This second check is necessary
195 for cases where a completion ast is received for an operation that
196 began before clear_proc_locks did its cancel/unlock. */
197
198 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
199 goto out;
200
201 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
202 ua = lkb->lkb_ua;
203 proc = ua->proc;
204
205 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
206 goto out;
207
208 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
209 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
210
211 spin_lock(&proc->asts_spin);
212
213 rv = dlm_enqueue_lkb_callback(lkb, flags, mode, status, sbflags);
214 switch (rv) {
215 case DLM_ENQUEUE_CALLBACK_FAILURE:
216 spin_unlock(&proc->asts_spin);
217 WARN_ON_ONCE(1);
218 goto out;
219 case DLM_ENQUEUE_CALLBACK_NEED_SCHED:
220 kref_get(&lkb->lkb_ref);
221 list_add_tail(&lkb->lkb_cb_list, &proc->asts);
222 wake_up_interruptible(&proc->wait);
223 break;
224 case DLM_ENQUEUE_CALLBACK_SUCCESS:
225 break;
226 default:
227 WARN_ON_ONCE(1);
228 break;
229 }
230 spin_unlock(&proc->asts_spin);
231
232 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
233 /* N.B. spin_lock locks_spin, not asts_spin */
234 spin_lock(&proc->locks_spin);
235 if (!list_empty(&lkb->lkb_ownqueue)) {
236 list_del_init(&lkb->lkb_ownqueue);
237 dlm_put_lkb(lkb);
238 }
239 spin_unlock(&proc->locks_spin);
240 }
241 out:
242 spin_unlock(&ls->ls_clear_proc_locks);
243}
244
245static int device_user_lock(struct dlm_user_proc *proc,
246 struct dlm_lock_params *params)
247{
248 struct dlm_ls *ls;
249 struct dlm_user_args *ua;
250 uint32_t lkid;
251 int error = -ENOMEM;
252
253 ls = dlm_find_lockspace_local(proc->lockspace);
254 if (!ls)
255 return -ENOENT;
256
257 if (!params->castaddr || !params->lksb) {
258 error = -EINVAL;
259 goto out;
260 }
261
262#ifdef CONFIG_DLM_DEPRECATED_API
263 if (params->timeout)
264 pr_warn_once("========================================================\n"
265 "WARNING: the lkb timeout feature is being deprecated and\n"
266 " will be removed in v6.2!\n"
267 "========================================================\n");
268#endif
269
270 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
271 if (!ua)
272 goto out;
273 ua->proc = proc;
274 ua->user_lksb = params->lksb;
275 ua->castparam = params->castparam;
276 ua->castaddr = params->castaddr;
277 ua->bastparam = params->bastparam;
278 ua->bastaddr = params->bastaddr;
279 ua->xid = params->xid;
280
281 if (params->flags & DLM_LKF_CONVERT) {
282#ifdef CONFIG_DLM_DEPRECATED_API
283 error = dlm_user_convert(ls, ua,
284 params->mode, params->flags,
285 params->lkid, params->lvb,
286 (unsigned long) params->timeout);
287#else
288 error = dlm_user_convert(ls, ua,
289 params->mode, params->flags,
290 params->lkid, params->lvb);
291#endif
292 } else if (params->flags & DLM_LKF_ORPHAN) {
293 error = dlm_user_adopt_orphan(ls, ua,
294 params->mode, params->flags,
295 params->name, params->namelen,
296 &lkid);
297 if (!error)
298 error = lkid;
299 } else {
300#ifdef CONFIG_DLM_DEPRECATED_API
301 error = dlm_user_request(ls, ua,
302 params->mode, params->flags,
303 params->name, params->namelen,
304 (unsigned long) params->timeout);
305#else
306 error = dlm_user_request(ls, ua,
307 params->mode, params->flags,
308 params->name, params->namelen);
309#endif
310 if (!error)
311 error = ua->lksb.sb_lkid;
312 }
313 out:
314 dlm_put_lockspace(ls);
315 return error;
316}
317
318static int device_user_unlock(struct dlm_user_proc *proc,
319 struct dlm_lock_params *params)
320{
321 struct dlm_ls *ls;
322 struct dlm_user_args *ua;
323 int error = -ENOMEM;
324
325 ls = dlm_find_lockspace_local(proc->lockspace);
326 if (!ls)
327 return -ENOENT;
328
329 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
330 if (!ua)
331 goto out;
332 ua->proc = proc;
333 ua->user_lksb = params->lksb;
334 ua->castparam = params->castparam;
335 ua->castaddr = params->castaddr;
336
337 if (params->flags & DLM_LKF_CANCEL)
338 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
339 else
340 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
341 params->lvb);
342 out:
343 dlm_put_lockspace(ls);
344 return error;
345}
346
347static int device_user_deadlock(struct dlm_user_proc *proc,
348 struct dlm_lock_params *params)
349{
350 struct dlm_ls *ls;
351 int error;
352
353 ls = dlm_find_lockspace_local(proc->lockspace);
354 if (!ls)
355 return -ENOENT;
356
357 error = dlm_user_deadlock(ls, params->flags, params->lkid);
358
359 dlm_put_lockspace(ls);
360 return error;
361}
362
363static int dlm_device_register(struct dlm_ls *ls, char *name)
364{
365 int error, len;
366
367 /* The device is already registered. This happens when the
368 lockspace is created multiple times from userspace. */
369 if (ls->ls_device.name)
370 return 0;
371
372 error = -ENOMEM;
373 len = strlen(name) + strlen(name_prefix) + 2;
374 ls->ls_device.name = kzalloc(len, GFP_NOFS);
375 if (!ls->ls_device.name)
376 goto fail;
377
378 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
379 name);
380 ls->ls_device.fops = &device_fops;
381 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
382
383 error = misc_register(&ls->ls_device);
384 if (error) {
385 kfree(ls->ls_device.name);
386 /* this has to be set to NULL
387 * to avoid a double-free in dlm_device_deregister
388 */
389 ls->ls_device.name = NULL;
390 }
391fail:
392 return error;
393}
394
395int dlm_device_deregister(struct dlm_ls *ls)
396{
397 /* The device is not registered. This happens when the lockspace
398 was never used from userspace, or when device_create_lockspace()
399 calls dlm_release_lockspace() after the register fails. */
400 if (!ls->ls_device.name)
401 return 0;
402
403 misc_deregister(&ls->ls_device);
404 kfree(ls->ls_device.name);
405 return 0;
406}
407
408static int device_user_purge(struct dlm_user_proc *proc,
409 struct dlm_purge_params *params)
410{
411 struct dlm_ls *ls;
412 int error;
413
414 ls = dlm_find_lockspace_local(proc->lockspace);
415 if (!ls)
416 return -ENOENT;
417
418 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
419
420 dlm_put_lockspace(ls);
421 return error;
422}
423
424static int device_create_lockspace(struct dlm_lspace_params *params)
425{
426 dlm_lockspace_t *lockspace;
427 struct dlm_ls *ls;
428 int error;
429
430 if (!capable(CAP_SYS_ADMIN))
431 return -EPERM;
432
433 error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name,
434 params->flags, DLM_USER_LVB_LEN, NULL,
435 NULL, NULL, &lockspace);
436 if (error)
437 return error;
438
439 ls = dlm_find_lockspace_local(lockspace);
440 if (!ls)
441 return -ENOENT;
442
443 error = dlm_device_register(ls, params->name);
444 dlm_put_lockspace(ls);
445
446 if (error)
447 dlm_release_lockspace(lockspace, 0);
448 else
449 error = ls->ls_device.minor;
450
451 return error;
452}
453
454static int device_remove_lockspace(struct dlm_lspace_params *params)
455{
456 dlm_lockspace_t *lockspace;
457 struct dlm_ls *ls;
458 int error, force = 0;
459
460 if (!capable(CAP_SYS_ADMIN))
461 return -EPERM;
462
463 ls = dlm_find_lockspace_device(params->minor);
464 if (!ls)
465 return -ENOENT;
466
467 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
468 force = 2;
469
470 lockspace = ls->ls_local_handle;
471 dlm_put_lockspace(ls);
472
473 /* The final dlm_release_lockspace waits for references to go to
474 zero, so all processes will need to close their device for the
475 ls before the release will proceed. release also calls the
476 device_deregister above. Converting a positive return value
477 from release to zero means that userspace won't know when its
478 release was the final one, but it shouldn't need to know. */
479
480 error = dlm_release_lockspace(lockspace, force);
481 if (error > 0)
482 error = 0;
483 return error;
484}
485
486/* Check the user's version matches ours */
487static int check_version(struct dlm_write_request *req)
488{
489 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
490 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
491 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
492
493 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
494 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
495 current->comm,
496 task_pid_nr(current),
497 req->version[0],
498 req->version[1],
499 req->version[2],
500 DLM_DEVICE_VERSION_MAJOR,
501 DLM_DEVICE_VERSION_MINOR,
502 DLM_DEVICE_VERSION_PATCH);
503 return -EINVAL;
504 }
505 return 0;
506}
507
508/*
509 * device_write
510 *
511 * device_user_lock
512 * dlm_user_request -> request_lock
513 * dlm_user_convert -> convert_lock
514 *
515 * device_user_unlock
516 * dlm_user_unlock -> unlock_lock
517 * dlm_user_cancel -> cancel_lock
518 *
519 * device_create_lockspace
520 * dlm_new_lockspace
521 *
522 * device_remove_lockspace
523 * dlm_release_lockspace
524 */
525
526/* a write to a lockspace device is a lock or unlock request, a write
527 to the control device is to create/remove a lockspace */
528
529static ssize_t device_write(struct file *file, const char __user *buf,
530 size_t count, loff_t *ppos)
531{
532 struct dlm_user_proc *proc = file->private_data;
533 struct dlm_write_request *kbuf;
534 int error;
535
536#ifdef CONFIG_COMPAT
537 if (count < sizeof(struct dlm_write_request32))
538#else
539 if (count < sizeof(struct dlm_write_request))
540#endif
541 return -EINVAL;
542
543 /*
544 * can't compare against COMPAT/dlm_write_request32 because
545 * we don't yet know if is64bit is zero
546 */
547 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
548 return -EINVAL;
549
550 kbuf = memdup_user_nul(buf, count);
551 if (IS_ERR(kbuf))
552 return PTR_ERR(kbuf);
553
554 if (check_version(kbuf)) {
555 error = -EBADE;
556 goto out_free;
557 }
558
559#ifdef CONFIG_COMPAT
560 if (!kbuf->is64bit) {
561 struct dlm_write_request32 *k32buf;
562 int namelen = 0;
563
564 if (count > sizeof(struct dlm_write_request32))
565 namelen = count - sizeof(struct dlm_write_request32);
566
567 k32buf = (struct dlm_write_request32 *)kbuf;
568
569 /* add 1 after namelen so that the name string is terminated */
570 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
571 GFP_NOFS);
572 if (!kbuf) {
573 kfree(k32buf);
574 return -ENOMEM;
575 }
576
577 if (proc)
578 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
579
580 compat_input(kbuf, k32buf, namelen);
581 kfree(k32buf);
582 }
583#endif
584
585 /* do we really need this? can a write happen after a close? */
586 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
587 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
588 error = -EINVAL;
589 goto out_free;
590 }
591
592 error = -EINVAL;
593
594 switch (kbuf->cmd)
595 {
596 case DLM_USER_LOCK:
597 if (!proc) {
598 log_print("no locking on control device");
599 goto out_free;
600 }
601 error = device_user_lock(proc, &kbuf->i.lock);
602 break;
603
604 case DLM_USER_UNLOCK:
605 if (!proc) {
606 log_print("no locking on control device");
607 goto out_free;
608 }
609 error = device_user_unlock(proc, &kbuf->i.lock);
610 break;
611
612 case DLM_USER_DEADLOCK:
613 if (!proc) {
614 log_print("no locking on control device");
615 goto out_free;
616 }
617 error = device_user_deadlock(proc, &kbuf->i.lock);
618 break;
619
620 case DLM_USER_CREATE_LOCKSPACE:
621 if (proc) {
622 log_print("create/remove only on control device");
623 goto out_free;
624 }
625 error = device_create_lockspace(&kbuf->i.lspace);
626 break;
627
628 case DLM_USER_REMOVE_LOCKSPACE:
629 if (proc) {
630 log_print("create/remove only on control device");
631 goto out_free;
632 }
633 error = device_remove_lockspace(&kbuf->i.lspace);
634 break;
635
636 case DLM_USER_PURGE:
637 if (!proc) {
638 log_print("no locking on control device");
639 goto out_free;
640 }
641 error = device_user_purge(proc, &kbuf->i.purge);
642 break;
643
644 default:
645 log_print("Unknown command passed to DLM device : %d\n",
646 kbuf->cmd);
647 }
648
649 out_free:
650 kfree(kbuf);
651 return error;
652}
653
654/* Every process that opens the lockspace device has its own "proc" structure
655 hanging off the open file that's used to keep track of locks owned by the
656 process and asts that need to be delivered to the process. */
657
658static int device_open(struct inode *inode, struct file *file)
659{
660 struct dlm_user_proc *proc;
661 struct dlm_ls *ls;
662
663 ls = dlm_find_lockspace_device(iminor(inode));
664 if (!ls)
665 return -ENOENT;
666
667 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
668 if (!proc) {
669 dlm_put_lockspace(ls);
670 return -ENOMEM;
671 }
672
673 proc->lockspace = ls->ls_local_handle;
674 INIT_LIST_HEAD(&proc->asts);
675 INIT_LIST_HEAD(&proc->locks);
676 INIT_LIST_HEAD(&proc->unlocking);
677 spin_lock_init(&proc->asts_spin);
678 spin_lock_init(&proc->locks_spin);
679 init_waitqueue_head(&proc->wait);
680 file->private_data = proc;
681
682 return 0;
683}
684
685static int device_close(struct inode *inode, struct file *file)
686{
687 struct dlm_user_proc *proc = file->private_data;
688 struct dlm_ls *ls;
689
690 ls = dlm_find_lockspace_local(proc->lockspace);
691 if (!ls)
692 return -ENOENT;
693
694 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
695
696 dlm_clear_proc_locks(ls, proc);
697
698 /* at this point no more lkb's should exist for this lockspace,
699 so there's no chance of dlm_user_add_ast() being called and
700 looking for lkb->ua->proc */
701
702 kfree(proc);
703 file->private_data = NULL;
704
705 dlm_put_lockspace(ls);
706 dlm_put_lockspace(ls); /* for the find in device_open() */
707
708 /* FIXME: AUTOFREE: if this ls is no longer used do
709 device_remove_lockspace() */
710
711 return 0;
712}
713
714static int copy_result_to_user(struct dlm_user_args *ua, int compat,
715 uint32_t flags, int mode, int copy_lvb,
716 char __user *buf, size_t count)
717{
718#ifdef CONFIG_COMPAT
719 struct dlm_lock_result32 result32;
720#endif
721 struct dlm_lock_result result;
722 void *resultptr;
723 int error=0;
724 int len;
725 int struct_len;
726
727 memset(&result, 0, sizeof(struct dlm_lock_result));
728 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
729 result.version[1] = DLM_DEVICE_VERSION_MINOR;
730 result.version[2] = DLM_DEVICE_VERSION_PATCH;
731 memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr));
732 result.user_lksb = ua->user_lksb;
733
734 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
735 in a conversion unless the conversion is successful. See code
736 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
737 notes that a new blocking AST address and parameter are set even if
738 the conversion fails, so maybe we should just do that. */
739
740 if (flags & DLM_CB_BAST) {
741 result.user_astaddr = ua->bastaddr;
742 result.user_astparam = ua->bastparam;
743 result.bast_mode = mode;
744 } else {
745 result.user_astaddr = ua->castaddr;
746 result.user_astparam = ua->castparam;
747 }
748
749#ifdef CONFIG_COMPAT
750 if (compat)
751 len = sizeof(struct dlm_lock_result32);
752 else
753#endif
754 len = sizeof(struct dlm_lock_result);
755 struct_len = len;
756
757 /* copy lvb to userspace if there is one, it's been updated, and
758 the user buffer has space for it */
759
760 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
761 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
762 DLM_USER_LVB_LEN)) {
763 error = -EFAULT;
764 goto out;
765 }
766
767 result.lvb_offset = len;
768 len += DLM_USER_LVB_LEN;
769 }
770
771 result.length = len;
772 resultptr = &result;
773#ifdef CONFIG_COMPAT
774 if (compat) {
775 compat_output(&result, &result32);
776 resultptr = &result32;
777 }
778#endif
779
780 if (copy_to_user(buf, resultptr, struct_len))
781 error = -EFAULT;
782 else
783 error = len;
784 out:
785 return error;
786}
787
788static int copy_version_to_user(char __user *buf, size_t count)
789{
790 struct dlm_device_version ver;
791
792 memset(&ver, 0, sizeof(struct dlm_device_version));
793 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
794 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
795 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
796
797 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
798 return -EFAULT;
799 return sizeof(struct dlm_device_version);
800}
801
802/* a read returns a single ast described in a struct dlm_lock_result */
803
804static ssize_t device_read(struct file *file, char __user *buf, size_t count,
805 loff_t *ppos)
806{
807 struct dlm_user_proc *proc = file->private_data;
808 struct dlm_lkb *lkb;
809 DECLARE_WAITQUEUE(wait, current);
810 struct dlm_callback *cb;
811 int rv, copy_lvb = 0;
812 int old_mode, new_mode;
813
814 if (count == sizeof(struct dlm_device_version)) {
815 rv = copy_version_to_user(buf, count);
816 return rv;
817 }
818
819 if (!proc) {
820 log_print("non-version read from control device %zu", count);
821 return -EINVAL;
822 }
823
824#ifdef CONFIG_COMPAT
825 if (count < sizeof(struct dlm_lock_result32))
826#else
827 if (count < sizeof(struct dlm_lock_result))
828#endif
829 return -EINVAL;
830
831 try_another:
832
833 /* do we really need this? can a read happen after a close? */
834 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
835 return -EINVAL;
836
837 spin_lock(&proc->asts_spin);
838 if (list_empty(&proc->asts)) {
839 if (file->f_flags & O_NONBLOCK) {
840 spin_unlock(&proc->asts_spin);
841 return -EAGAIN;
842 }
843
844 add_wait_queue(&proc->wait, &wait);
845
846 repeat:
847 set_current_state(TASK_INTERRUPTIBLE);
848 if (list_empty(&proc->asts) && !signal_pending(current)) {
849 spin_unlock(&proc->asts_spin);
850 schedule();
851 spin_lock(&proc->asts_spin);
852 goto repeat;
853 }
854 set_current_state(TASK_RUNNING);
855 remove_wait_queue(&proc->wait, &wait);
856
857 if (signal_pending(current)) {
858 spin_unlock(&proc->asts_spin);
859 return -ERESTARTSYS;
860 }
861 }
862
863 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
864 without removing lkb_cb_list; so empty lkb_cb_list is always
865 consistent with empty lkb_callbacks */
866
867 lkb = list_first_entry(&proc->asts, struct dlm_lkb, lkb_cb_list);
868
869 /* rem_lkb_callback sets a new lkb_last_cast */
870 old_mode = lkb->lkb_last_cast->mode;
871
872 rv = dlm_dequeue_lkb_callback(lkb, &cb);
873 switch (rv) {
874 case DLM_DEQUEUE_CALLBACK_EMPTY:
875 /* this shouldn't happen; lkb should have been removed from
876 * list when last item was dequeued
877 */
878 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
879 list_del_init(&lkb->lkb_cb_list);
880 spin_unlock(&proc->asts_spin);
881 /* removes ref for proc->asts, may cause lkb to be freed */
882 dlm_put_lkb(lkb);
883 WARN_ON_ONCE(1);
884 goto try_another;
885 case DLM_DEQUEUE_CALLBACK_LAST:
886 list_del_init(&lkb->lkb_cb_list);
887 lkb->lkb_flags &= ~DLM_IFL_CB_PENDING;
888 break;
889 case DLM_DEQUEUE_CALLBACK_SUCCESS:
890 break;
891 default:
892 WARN_ON_ONCE(1);
893 break;
894 }
895 spin_unlock(&proc->asts_spin);
896
897 if (cb->flags & DLM_CB_BAST) {
898 trace_dlm_bast(lkb->lkb_resource->res_ls, lkb, cb->mode);
899 } else if (cb->flags & DLM_CB_CAST) {
900 new_mode = cb->mode;
901
902 if (!cb->sb_status && lkb->lkb_lksb->sb_lvbptr &&
903 dlm_lvb_operations[old_mode + 1][new_mode + 1])
904 copy_lvb = 1;
905
906 lkb->lkb_lksb->sb_status = cb->sb_status;
907 lkb->lkb_lksb->sb_flags = cb->sb_flags;
908 trace_dlm_ast(lkb->lkb_resource->res_ls, lkb);
909 }
910
911 rv = copy_result_to_user(lkb->lkb_ua,
912 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
913 cb->flags, cb->mode, copy_lvb, buf, count);
914
915 kref_put(&cb->ref, dlm_release_callback);
916
917 /* removes ref for proc->asts, may cause lkb to be freed */
918 if (rv == DLM_DEQUEUE_CALLBACK_LAST)
919 dlm_put_lkb(lkb);
920
921 return rv;
922}
923
924static __poll_t device_poll(struct file *file, poll_table *wait)
925{
926 struct dlm_user_proc *proc = file->private_data;
927
928 poll_wait(file, &proc->wait, wait);
929
930 spin_lock(&proc->asts_spin);
931 if (!list_empty(&proc->asts)) {
932 spin_unlock(&proc->asts_spin);
933 return EPOLLIN | EPOLLRDNORM;
934 }
935 spin_unlock(&proc->asts_spin);
936 return 0;
937}
938
939int dlm_user_daemon_available(void)
940{
941 /* dlm_controld hasn't started (or, has started, but not
942 properly populated configfs) */
943
944 if (!dlm_our_nodeid())
945 return 0;
946
947 /* This is to deal with versions of dlm_controld that don't
948 know about the monitor device. We assume that if the
949 dlm_controld was started (above), but the monitor device
950 was never opened, that it's an old version. dlm_controld
951 should open the monitor device before populating configfs. */
952
953 if (dlm_monitor_unused)
954 return 1;
955
956 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
957}
958
959static int ctl_device_open(struct inode *inode, struct file *file)
960{
961 file->private_data = NULL;
962 return 0;
963}
964
965static int ctl_device_close(struct inode *inode, struct file *file)
966{
967 return 0;
968}
969
970static int monitor_device_open(struct inode *inode, struct file *file)
971{
972 atomic_inc(&dlm_monitor_opened);
973 dlm_monitor_unused = 0;
974 return 0;
975}
976
977static int monitor_device_close(struct inode *inode, struct file *file)
978{
979 if (atomic_dec_and_test(&dlm_monitor_opened))
980 dlm_stop_lockspaces();
981 return 0;
982}
983
984static const struct file_operations device_fops = {
985 .open = device_open,
986 .release = device_close,
987 .read = device_read,
988 .write = device_write,
989 .poll = device_poll,
990 .owner = THIS_MODULE,
991 .llseek = noop_llseek,
992};
993
994static const struct file_operations ctl_device_fops = {
995 .open = ctl_device_open,
996 .release = ctl_device_close,
997 .read = device_read,
998 .write = device_write,
999 .owner = THIS_MODULE,
1000 .llseek = noop_llseek,
1001};
1002
1003static struct miscdevice ctl_device = {
1004 .name = "dlm-control",
1005 .fops = &ctl_device_fops,
1006 .minor = MISC_DYNAMIC_MINOR,
1007};
1008
1009static const struct file_operations monitor_device_fops = {
1010 .open = monitor_device_open,
1011 .release = monitor_device_close,
1012 .owner = THIS_MODULE,
1013 .llseek = noop_llseek,
1014};
1015
1016static struct miscdevice monitor_device = {
1017 .name = "dlm-monitor",
1018 .fops = &monitor_device_fops,
1019 .minor = MISC_DYNAMIC_MINOR,
1020};
1021
1022int __init dlm_user_init(void)
1023{
1024 int error;
1025
1026 atomic_set(&dlm_monitor_opened, 0);
1027
1028 error = misc_register(&ctl_device);
1029 if (error) {
1030 log_print("misc_register failed for control device");
1031 goto out;
1032 }
1033
1034 error = misc_register(&monitor_device);
1035 if (error) {
1036 log_print("misc_register failed for monitor device");
1037 misc_deregister(&ctl_device);
1038 }
1039 out:
1040 return error;
1041}
1042
1043void dlm_user_exit(void)
1044{
1045 misc_deregister(&ctl_device);
1046 misc_deregister(&monitor_device);
1047}
1048