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