Loading...
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 "dlm_internal.h"
20#include "lockspace.h"
21#include "lock.h"
22#include "lvb_table.h"
23#include "user.h"
24#include "ast.h"
25#include "config.h"
26
27static const char name_prefix[] = "dlm";
28static const struct file_operations device_fops;
29static atomic_t dlm_monitor_opened;
30static int dlm_monitor_unused = 1;
31
32#ifdef CONFIG_COMPAT
33
34struct dlm_lock_params32 {
35 __u8 mode;
36 __u8 namelen;
37 __u16 unused;
38 __u32 flags;
39 __u32 lkid;
40 __u32 parent;
41 __u64 xid;
42 __u64 timeout;
43 __u32 castparam;
44 __u32 castaddr;
45 __u32 bastparam;
46 __u32 bastaddr;
47 __u32 lksb;
48 char lvb[DLM_USER_LVB_LEN];
49 char name[];
50};
51
52struct dlm_write_request32 {
53 __u32 version[3];
54 __u8 cmd;
55 __u8 is64bit;
56 __u8 unused[2];
57
58 union {
59 struct dlm_lock_params32 lock;
60 struct dlm_lspace_params lspace;
61 struct dlm_purge_params purge;
62 } i;
63};
64
65struct dlm_lksb32 {
66 __u32 sb_status;
67 __u32 sb_lkid;
68 __u8 sb_flags;
69 __u32 sb_lvbptr;
70};
71
72struct dlm_lock_result32 {
73 __u32 version[3];
74 __u32 length;
75 __u32 user_astaddr;
76 __u32 user_astparam;
77 __u32 user_lksb;
78 struct dlm_lksb32 lksb;
79 __u8 bast_mode;
80 __u8 unused[3];
81 /* Offsets may be zero if no data is present */
82 __u32 lvb_offset;
83};
84
85static void compat_input(struct dlm_write_request *kb,
86 struct dlm_write_request32 *kb32,
87 int namelen)
88{
89 kb->version[0] = kb32->version[0];
90 kb->version[1] = kb32->version[1];
91 kb->version[2] = kb32->version[2];
92
93 kb->cmd = kb32->cmd;
94 kb->is64bit = kb32->is64bit;
95 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
96 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
97 kb->i.lspace.flags = kb32->i.lspace.flags;
98 kb->i.lspace.minor = kb32->i.lspace.minor;
99 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
100 } else if (kb->cmd == DLM_USER_PURGE) {
101 kb->i.purge.nodeid = kb32->i.purge.nodeid;
102 kb->i.purge.pid = kb32->i.purge.pid;
103 } else {
104 kb->i.lock.mode = kb32->i.lock.mode;
105 kb->i.lock.namelen = kb32->i.lock.namelen;
106 kb->i.lock.flags = kb32->i.lock.flags;
107 kb->i.lock.lkid = kb32->i.lock.lkid;
108 kb->i.lock.parent = kb32->i.lock.parent;
109 kb->i.lock.xid = kb32->i.lock.xid;
110 kb->i.lock.timeout = kb32->i.lock.timeout;
111 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
112 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
113 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
114 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
115 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
116 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
117 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
118 }
119}
120
121static void compat_output(struct dlm_lock_result *res,
122 struct dlm_lock_result32 *res32)
123{
124 memset(res32, 0, sizeof(*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 uint32_t lkid;
242 int error = -ENOMEM;
243
244 ls = dlm_find_lockspace_local(proc->lockspace);
245 if (!ls)
246 return -ENOENT;
247
248 if (!params->castaddr || !params->lksb) {
249 error = -EINVAL;
250 goto out;
251 }
252
253 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
254 if (!ua)
255 goto out;
256 ua->proc = proc;
257 ua->user_lksb = params->lksb;
258 ua->castparam = params->castparam;
259 ua->castaddr = params->castaddr;
260 ua->bastparam = params->bastparam;
261 ua->bastaddr = params->bastaddr;
262 ua->xid = params->xid;
263
264 if (params->flags & DLM_LKF_CONVERT) {
265 error = dlm_user_convert(ls, ua,
266 params->mode, params->flags,
267 params->lkid, params->lvb,
268 (unsigned long) params->timeout);
269 } else if (params->flags & DLM_LKF_ORPHAN) {
270 error = dlm_user_adopt_orphan(ls, ua,
271 params->mode, params->flags,
272 params->name, params->namelen,
273 (unsigned long) params->timeout,
274 &lkid);
275 if (!error)
276 error = lkid;
277 } else {
278 error = dlm_user_request(ls, ua,
279 params->mode, params->flags,
280 params->name, params->namelen,
281 (unsigned long) params->timeout);
282 if (!error)
283 error = ua->lksb.sb_lkid;
284 }
285 out:
286 dlm_put_lockspace(ls);
287 return error;
288}
289
290static int device_user_unlock(struct dlm_user_proc *proc,
291 struct dlm_lock_params *params)
292{
293 struct dlm_ls *ls;
294 struct dlm_user_args *ua;
295 int error = -ENOMEM;
296
297 ls = dlm_find_lockspace_local(proc->lockspace);
298 if (!ls)
299 return -ENOENT;
300
301 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
302 if (!ua)
303 goto out;
304 ua->proc = proc;
305 ua->user_lksb = params->lksb;
306 ua->castparam = params->castparam;
307 ua->castaddr = params->castaddr;
308
309 if (params->flags & DLM_LKF_CANCEL)
310 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
311 else
312 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
313 params->lvb);
314 out:
315 dlm_put_lockspace(ls);
316 return error;
317}
318
319static int device_user_deadlock(struct dlm_user_proc *proc,
320 struct dlm_lock_params *params)
321{
322 struct dlm_ls *ls;
323 int error;
324
325 ls = dlm_find_lockspace_local(proc->lockspace);
326 if (!ls)
327 return -ENOENT;
328
329 error = dlm_user_deadlock(ls, params->flags, params->lkid);
330
331 dlm_put_lockspace(ls);
332 return error;
333}
334
335static int dlm_device_register(struct dlm_ls *ls, char *name)
336{
337 int error, len;
338
339 /* The device is already registered. This happens when the
340 lockspace is created multiple times from userspace. */
341 if (ls->ls_device.name)
342 return 0;
343
344 error = -ENOMEM;
345 len = strlen(name) + strlen(name_prefix) + 2;
346 ls->ls_device.name = kzalloc(len, GFP_NOFS);
347 if (!ls->ls_device.name)
348 goto fail;
349
350 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
351 name);
352 ls->ls_device.fops = &device_fops;
353 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
354
355 error = misc_register(&ls->ls_device);
356 if (error) {
357 kfree(ls->ls_device.name);
358 /* this has to be set to NULL
359 * to avoid a double-free in dlm_device_deregister
360 */
361 ls->ls_device.name = NULL;
362 }
363fail:
364 return error;
365}
366
367int dlm_device_deregister(struct dlm_ls *ls)
368{
369 /* The device is not registered. This happens when the lockspace
370 was never used from userspace, or when device_create_lockspace()
371 calls dlm_release_lockspace() after the register fails. */
372 if (!ls->ls_device.name)
373 return 0;
374
375 misc_deregister(&ls->ls_device);
376 kfree(ls->ls_device.name);
377 return 0;
378}
379
380static int device_user_purge(struct dlm_user_proc *proc,
381 struct dlm_purge_params *params)
382{
383 struct dlm_ls *ls;
384 int error;
385
386 ls = dlm_find_lockspace_local(proc->lockspace);
387 if (!ls)
388 return -ENOENT;
389
390 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
391
392 dlm_put_lockspace(ls);
393 return error;
394}
395
396static int device_create_lockspace(struct dlm_lspace_params *params)
397{
398 dlm_lockspace_t *lockspace;
399 struct dlm_ls *ls;
400 int error;
401
402 if (!capable(CAP_SYS_ADMIN))
403 return -EPERM;
404
405 error = dlm_new_lockspace(params->name, dlm_config.ci_cluster_name, params->flags,
406 DLM_USER_LVB_LEN, NULL, NULL, NULL,
407 &lockspace);
408 if (error)
409 return error;
410
411 ls = dlm_find_lockspace_local(lockspace);
412 if (!ls)
413 return -ENOENT;
414
415 error = dlm_device_register(ls, params->name);
416 dlm_put_lockspace(ls);
417
418 if (error)
419 dlm_release_lockspace(lockspace, 0);
420 else
421 error = ls->ls_device.minor;
422
423 return error;
424}
425
426static int device_remove_lockspace(struct dlm_lspace_params *params)
427{
428 dlm_lockspace_t *lockspace;
429 struct dlm_ls *ls;
430 int error, force = 0;
431
432 if (!capable(CAP_SYS_ADMIN))
433 return -EPERM;
434
435 ls = dlm_find_lockspace_device(params->minor);
436 if (!ls)
437 return -ENOENT;
438
439 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
440 force = 2;
441
442 lockspace = ls->ls_local_handle;
443 dlm_put_lockspace(ls);
444
445 /* The final dlm_release_lockspace waits for references to go to
446 zero, so all processes will need to close their device for the
447 ls before the release will proceed. release also calls the
448 device_deregister above. Converting a positive return value
449 from release to zero means that userspace won't know when its
450 release was the final one, but it shouldn't need to know. */
451
452 error = dlm_release_lockspace(lockspace, force);
453 if (error > 0)
454 error = 0;
455 return error;
456}
457
458/* Check the user's version matches ours */
459static int check_version(struct dlm_write_request *req)
460{
461 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
462 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
463 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
464
465 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
466 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
467 current->comm,
468 task_pid_nr(current),
469 req->version[0],
470 req->version[1],
471 req->version[2],
472 DLM_DEVICE_VERSION_MAJOR,
473 DLM_DEVICE_VERSION_MINOR,
474 DLM_DEVICE_VERSION_PATCH);
475 return -EINVAL;
476 }
477 return 0;
478}
479
480/*
481 * device_write
482 *
483 * device_user_lock
484 * dlm_user_request -> request_lock
485 * dlm_user_convert -> convert_lock
486 *
487 * device_user_unlock
488 * dlm_user_unlock -> unlock_lock
489 * dlm_user_cancel -> cancel_lock
490 *
491 * device_create_lockspace
492 * dlm_new_lockspace
493 *
494 * device_remove_lockspace
495 * dlm_release_lockspace
496 */
497
498/* a write to a lockspace device is a lock or unlock request, a write
499 to the control device is to create/remove a lockspace */
500
501static ssize_t device_write(struct file *file, const char __user *buf,
502 size_t count, loff_t *ppos)
503{
504 struct dlm_user_proc *proc = file->private_data;
505 struct dlm_write_request *kbuf;
506 int error;
507
508#ifdef CONFIG_COMPAT
509 if (count < sizeof(struct dlm_write_request32))
510#else
511 if (count < sizeof(struct dlm_write_request))
512#endif
513 return -EINVAL;
514
515 /*
516 * can't compare against COMPAT/dlm_write_request32 because
517 * we don't yet know if is64bit is zero
518 */
519 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
520 return -EINVAL;
521
522 kbuf = memdup_user_nul(buf, count);
523 if (IS_ERR(kbuf))
524 return PTR_ERR(kbuf);
525
526 if (check_version(kbuf)) {
527 error = -EBADE;
528 goto out_free;
529 }
530
531#ifdef CONFIG_COMPAT
532 if (!kbuf->is64bit) {
533 struct dlm_write_request32 *k32buf;
534 int namelen = 0;
535
536 if (count > sizeof(struct dlm_write_request32))
537 namelen = count - sizeof(struct dlm_write_request32);
538
539 k32buf = (struct dlm_write_request32 *)kbuf;
540
541 /* add 1 after namelen so that the name string is terminated */
542 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
543 GFP_NOFS);
544 if (!kbuf) {
545 kfree(k32buf);
546 return -ENOMEM;
547 }
548
549 if (proc)
550 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
551
552 compat_input(kbuf, k32buf, namelen);
553 kfree(k32buf);
554 }
555#endif
556
557 /* do we really need this? can a write happen after a close? */
558 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
559 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
560 error = -EINVAL;
561 goto out_free;
562 }
563
564 error = -EINVAL;
565
566 switch (kbuf->cmd)
567 {
568 case DLM_USER_LOCK:
569 if (!proc) {
570 log_print("no locking on control device");
571 goto out_free;
572 }
573 error = device_user_lock(proc, &kbuf->i.lock);
574 break;
575
576 case DLM_USER_UNLOCK:
577 if (!proc) {
578 log_print("no locking on control device");
579 goto out_free;
580 }
581 error = device_user_unlock(proc, &kbuf->i.lock);
582 break;
583
584 case DLM_USER_DEADLOCK:
585 if (!proc) {
586 log_print("no locking on control device");
587 goto out_free;
588 }
589 error = device_user_deadlock(proc, &kbuf->i.lock);
590 break;
591
592 case DLM_USER_CREATE_LOCKSPACE:
593 if (proc) {
594 log_print("create/remove only on control device");
595 goto out_free;
596 }
597 error = device_create_lockspace(&kbuf->i.lspace);
598 break;
599
600 case DLM_USER_REMOVE_LOCKSPACE:
601 if (proc) {
602 log_print("create/remove only on control device");
603 goto out_free;
604 }
605 error = device_remove_lockspace(&kbuf->i.lspace);
606 break;
607
608 case DLM_USER_PURGE:
609 if (!proc) {
610 log_print("no locking on control device");
611 goto out_free;
612 }
613 error = device_user_purge(proc, &kbuf->i.purge);
614 break;
615
616 default:
617 log_print("Unknown command passed to DLM device : %d\n",
618 kbuf->cmd);
619 }
620
621 out_free:
622 kfree(kbuf);
623 return error;
624}
625
626/* Every process that opens the lockspace device has its own "proc" structure
627 hanging off the open file that's used to keep track of locks owned by the
628 process and asts that need to be delivered to the process. */
629
630static int device_open(struct inode *inode, struct file *file)
631{
632 struct dlm_user_proc *proc;
633 struct dlm_ls *ls;
634
635 ls = dlm_find_lockspace_device(iminor(inode));
636 if (!ls)
637 return -ENOENT;
638
639 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
640 if (!proc) {
641 dlm_put_lockspace(ls);
642 return -ENOMEM;
643 }
644
645 proc->lockspace = ls->ls_local_handle;
646 INIT_LIST_HEAD(&proc->asts);
647 INIT_LIST_HEAD(&proc->locks);
648 INIT_LIST_HEAD(&proc->unlocking);
649 spin_lock_init(&proc->asts_spin);
650 spin_lock_init(&proc->locks_spin);
651 init_waitqueue_head(&proc->wait);
652 file->private_data = proc;
653
654 return 0;
655}
656
657static int device_close(struct inode *inode, struct file *file)
658{
659 struct dlm_user_proc *proc = file->private_data;
660 struct dlm_ls *ls;
661
662 ls = dlm_find_lockspace_local(proc->lockspace);
663 if (!ls)
664 return -ENOENT;
665
666 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
667
668 dlm_clear_proc_locks(ls, proc);
669
670 /* at this point no more lkb's should exist for this lockspace,
671 so there's no chance of dlm_user_add_ast() being called and
672 looking for lkb->ua->proc */
673
674 kfree(proc);
675 file->private_data = NULL;
676
677 dlm_put_lockspace(ls);
678 dlm_put_lockspace(ls); /* for the find in device_open() */
679
680 /* FIXME: AUTOFREE: if this ls is no longer used do
681 device_remove_lockspace() */
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, offsetof(struct dlm_lksb, sb_lvbptr));
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 int old_mode, new_mode;
785
786 if (count == sizeof(struct dlm_device_version)) {
787 rv = copy_version_to_user(buf, count);
788 return rv;
789 }
790
791 if (!proc) {
792 log_print("non-version read from control device %zu", count);
793 return -EINVAL;
794 }
795
796#ifdef CONFIG_COMPAT
797 if (count < sizeof(struct dlm_lock_result32))
798#else
799 if (count < sizeof(struct dlm_lock_result))
800#endif
801 return -EINVAL;
802
803 try_another:
804
805 /* do we really need this? can a read happen after a close? */
806 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
807 return -EINVAL;
808
809 spin_lock(&proc->asts_spin);
810 if (list_empty(&proc->asts)) {
811 if (file->f_flags & O_NONBLOCK) {
812 spin_unlock(&proc->asts_spin);
813 return -EAGAIN;
814 }
815
816 add_wait_queue(&proc->wait, &wait);
817
818 repeat:
819 set_current_state(TASK_INTERRUPTIBLE);
820 if (list_empty(&proc->asts) && !signal_pending(current)) {
821 spin_unlock(&proc->asts_spin);
822 schedule();
823 spin_lock(&proc->asts_spin);
824 goto repeat;
825 }
826 set_current_state(TASK_RUNNING);
827 remove_wait_queue(&proc->wait, &wait);
828
829 if (signal_pending(current)) {
830 spin_unlock(&proc->asts_spin);
831 return -ERESTARTSYS;
832 }
833 }
834
835 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
836 without removing lkb_cb_list; so empty lkb_cb_list is always
837 consistent with empty lkb_callbacks */
838
839 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
840
841 /* rem_lkb_callback sets a new lkb_last_cast */
842 old_mode = lkb->lkb_last_cast.mode;
843
844 rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
845 if (rv < 0) {
846 /* this shouldn't happen; lkb should have been removed from
847 list when resid was zero */
848 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
849 list_del_init(&lkb->lkb_cb_list);
850 spin_unlock(&proc->asts_spin);
851 /* removes ref for proc->asts, may cause lkb to be freed */
852 dlm_put_lkb(lkb);
853 goto try_another;
854 }
855 if (!resid)
856 list_del_init(&lkb->lkb_cb_list);
857 spin_unlock(&proc->asts_spin);
858
859 if (cb.flags & DLM_CB_SKIP) {
860 /* removes ref for proc->asts, may cause lkb to be freed */
861 if (!resid)
862 dlm_put_lkb(lkb);
863 goto try_another;
864 }
865
866 if (cb.flags & DLM_CB_CAST) {
867 new_mode = cb.mode;
868
869 if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
870 dlm_lvb_operations[old_mode + 1][new_mode + 1])
871 copy_lvb = 1;
872
873 lkb->lkb_lksb->sb_status = cb.sb_status;
874 lkb->lkb_lksb->sb_flags = cb.sb_flags;
875 }
876
877 rv = copy_result_to_user(lkb->lkb_ua,
878 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
879 cb.flags, cb.mode, copy_lvb, buf, count);
880
881 /* removes ref for proc->asts, may cause lkb to be freed */
882 if (!resid)
883 dlm_put_lkb(lkb);
884
885 return rv;
886}
887
888static __poll_t device_poll(struct file *file, poll_table *wait)
889{
890 struct dlm_user_proc *proc = file->private_data;
891
892 poll_wait(file, &proc->wait, wait);
893
894 spin_lock(&proc->asts_spin);
895 if (!list_empty(&proc->asts)) {
896 spin_unlock(&proc->asts_spin);
897 return EPOLLIN | EPOLLRDNORM;
898 }
899 spin_unlock(&proc->asts_spin);
900 return 0;
901}
902
903int dlm_user_daemon_available(void)
904{
905 /* dlm_controld hasn't started (or, has started, but not
906 properly populated configfs) */
907
908 if (!dlm_our_nodeid())
909 return 0;
910
911 /* This is to deal with versions of dlm_controld that don't
912 know about the monitor device. We assume that if the
913 dlm_controld was started (above), but the monitor device
914 was never opened, that it's an old version. dlm_controld
915 should open the monitor device before populating configfs. */
916
917 if (dlm_monitor_unused)
918 return 1;
919
920 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
921}
922
923static int ctl_device_open(struct inode *inode, struct file *file)
924{
925 file->private_data = NULL;
926 return 0;
927}
928
929static int ctl_device_close(struct inode *inode, struct file *file)
930{
931 return 0;
932}
933
934static int monitor_device_open(struct inode *inode, struct file *file)
935{
936 atomic_inc(&dlm_monitor_opened);
937 dlm_monitor_unused = 0;
938 return 0;
939}
940
941static int monitor_device_close(struct inode *inode, struct file *file)
942{
943 if (atomic_dec_and_test(&dlm_monitor_opened))
944 dlm_stop_lockspaces();
945 return 0;
946}
947
948static const struct file_operations device_fops = {
949 .open = device_open,
950 .release = device_close,
951 .read = device_read,
952 .write = device_write,
953 .poll = device_poll,
954 .owner = THIS_MODULE,
955 .llseek = noop_llseek,
956};
957
958static const struct file_operations ctl_device_fops = {
959 .open = ctl_device_open,
960 .release = ctl_device_close,
961 .read = device_read,
962 .write = device_write,
963 .owner = THIS_MODULE,
964 .llseek = noop_llseek,
965};
966
967static struct miscdevice ctl_device = {
968 .name = "dlm-control",
969 .fops = &ctl_device_fops,
970 .minor = MISC_DYNAMIC_MINOR,
971};
972
973static const struct file_operations monitor_device_fops = {
974 .open = monitor_device_open,
975 .release = monitor_device_close,
976 .owner = THIS_MODULE,
977 .llseek = noop_llseek,
978};
979
980static struct miscdevice monitor_device = {
981 .name = "dlm-monitor",
982 .fops = &monitor_device_fops,
983 .minor = MISC_DYNAMIC_MINOR,
984};
985
986int __init dlm_user_init(void)
987{
988 int error;
989
990 atomic_set(&dlm_monitor_opened, 0);
991
992 error = misc_register(&ctl_device);
993 if (error) {
994 log_print("misc_register failed for control device");
995 goto out;
996 }
997
998 error = misc_register(&monitor_device);
999 if (error) {
1000 log_print("misc_register failed for monitor device");
1001 misc_deregister(&ctl_device);
1002 }
1003 out:
1004 return error;
1005}
1006
1007void dlm_user_exit(void)
1008{
1009 misc_deregister(&ctl_device);
1010 misc_deregister(&monitor_device);
1011}
1012
1/*
2 * Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved.
3 *
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License v.2.
7 */
8
9#include <linux/miscdevice.h>
10#include <linux/init.h>
11#include <linux/wait.h>
12#include <linux/file.h>
13#include <linux/fs.h>
14#include <linux/poll.h>
15#include <linux/signal.h>
16#include <linux/spinlock.h>
17#include <linux/dlm.h>
18#include <linux/dlm_device.h>
19#include <linux/slab.h>
20
21#include "dlm_internal.h"
22#include "lockspace.h"
23#include "lock.h"
24#include "lvb_table.h"
25#include "user.h"
26#include "ast.h"
27
28static const char name_prefix[] = "dlm";
29static const struct file_operations device_fops;
30static atomic_t dlm_monitor_opened;
31static int dlm_monitor_unused = 1;
32
33#ifdef CONFIG_COMPAT
34
35struct dlm_lock_params32 {
36 __u8 mode;
37 __u8 namelen;
38 __u16 unused;
39 __u32 flags;
40 __u32 lkid;
41 __u32 parent;
42 __u64 xid;
43 __u64 timeout;
44 __u32 castparam;
45 __u32 castaddr;
46 __u32 bastparam;
47 __u32 bastaddr;
48 __u32 lksb;
49 char lvb[DLM_USER_LVB_LEN];
50 char name[0];
51};
52
53struct dlm_write_request32 {
54 __u32 version[3];
55 __u8 cmd;
56 __u8 is64bit;
57 __u8 unused[2];
58
59 union {
60 struct dlm_lock_params32 lock;
61 struct dlm_lspace_params lspace;
62 struct dlm_purge_params purge;
63 } i;
64};
65
66struct dlm_lksb32 {
67 __u32 sb_status;
68 __u32 sb_lkid;
69 __u8 sb_flags;
70 __u32 sb_lvbptr;
71};
72
73struct dlm_lock_result32 {
74 __u32 version[3];
75 __u32 length;
76 __u32 user_astaddr;
77 __u32 user_astparam;
78 __u32 user_lksb;
79 struct dlm_lksb32 lksb;
80 __u8 bast_mode;
81 __u8 unused[3];
82 /* Offsets may be zero if no data is present */
83 __u32 lvb_offset;
84};
85
86static void compat_input(struct dlm_write_request *kb,
87 struct dlm_write_request32 *kb32,
88 int namelen)
89{
90 kb->version[0] = kb32->version[0];
91 kb->version[1] = kb32->version[1];
92 kb->version[2] = kb32->version[2];
93
94 kb->cmd = kb32->cmd;
95 kb->is64bit = kb32->is64bit;
96 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
97 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
98 kb->i.lspace.flags = kb32->i.lspace.flags;
99 kb->i.lspace.minor = kb32->i.lspace.minor;
100 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
101 } else if (kb->cmd == DLM_USER_PURGE) {
102 kb->i.purge.nodeid = kb32->i.purge.nodeid;
103 kb->i.purge.pid = kb32->i.purge.pid;
104 } else {
105 kb->i.lock.mode = kb32->i.lock.mode;
106 kb->i.lock.namelen = kb32->i.lock.namelen;
107 kb->i.lock.flags = kb32->i.lock.flags;
108 kb->i.lock.lkid = kb32->i.lock.lkid;
109 kb->i.lock.parent = kb32->i.lock.parent;
110 kb->i.lock.xid = kb32->i.lock.xid;
111 kb->i.lock.timeout = kb32->i.lock.timeout;
112 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
113 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
114 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
115 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
116 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
117 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
118 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
119 }
120}
121
122static void compat_output(struct dlm_lock_result *res,
123 struct dlm_lock_result32 *res32)
124{
125 res32->version[0] = res->version[0];
126 res32->version[1] = res->version[1];
127 res32->version[2] = res->version[2];
128
129 res32->user_astaddr = (__u32)(long)res->user_astaddr;
130 res32->user_astparam = (__u32)(long)res->user_astparam;
131 res32->user_lksb = (__u32)(long)res->user_lksb;
132 res32->bast_mode = res->bast_mode;
133
134 res32->lvb_offset = res->lvb_offset;
135 res32->length = res->length;
136
137 res32->lksb.sb_status = res->lksb.sb_status;
138 res32->lksb.sb_flags = res->lksb.sb_flags;
139 res32->lksb.sb_lkid = res->lksb.sb_lkid;
140 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
141}
142#endif
143
144/* Figure out if this lock is at the end of its life and no longer
145 available for the application to use. The lkb still exists until
146 the final ast is read. A lock becomes EOL in three situations:
147 1. a noqueue request fails with EAGAIN
148 2. an unlock completes with EUNLOCK
149 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
150 An EOL lock needs to be removed from the process's list of locks.
151 And we can't allow any new operation on an EOL lock. This is
152 not related to the lifetime of the lkb struct which is managed
153 entirely by refcount. */
154
155static int lkb_is_endoflife(int mode, int status)
156{
157 switch (status) {
158 case -DLM_EUNLOCK:
159 return 1;
160 case -DLM_ECANCEL:
161 case -ETIMEDOUT:
162 case -EDEADLK:
163 case -EAGAIN:
164 if (mode == DLM_LOCK_IV)
165 return 1;
166 break;
167 }
168 return 0;
169}
170
171/* we could possibly check if the cancel of an orphan has resulted in the lkb
172 being removed and then remove that lkb from the orphans list and free it */
173
174void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
175 int status, uint32_t sbflags, uint64_t seq)
176{
177 struct dlm_ls *ls;
178 struct dlm_user_args *ua;
179 struct dlm_user_proc *proc;
180 int rv;
181
182 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
183 return;
184
185 ls = lkb->lkb_resource->res_ls;
186 mutex_lock(&ls->ls_clear_proc_locks);
187
188 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
189 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
190 lkb->ua so we can't try to use it. This second check is necessary
191 for cases where a completion ast is received for an operation that
192 began before clear_proc_locks did its cancel/unlock. */
193
194 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
195 goto out;
196
197 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
198 ua = lkb->lkb_ua;
199 proc = ua->proc;
200
201 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
202 goto out;
203
204 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
205 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
206
207 spin_lock(&proc->asts_spin);
208
209 rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
210 if (rv < 0) {
211 spin_unlock(&proc->asts_spin);
212 goto out;
213 }
214
215 if (list_empty(&lkb->lkb_cb_list)) {
216 kref_get(&lkb->lkb_ref);
217 list_add_tail(&lkb->lkb_cb_list, &proc->asts);
218 wake_up_interruptible(&proc->wait);
219 }
220 spin_unlock(&proc->asts_spin);
221
222 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
223 /* N.B. spin_lock locks_spin, not asts_spin */
224 spin_lock(&proc->locks_spin);
225 if (!list_empty(&lkb->lkb_ownqueue)) {
226 list_del_init(&lkb->lkb_ownqueue);
227 dlm_put_lkb(lkb);
228 }
229 spin_unlock(&proc->locks_spin);
230 }
231 out:
232 mutex_unlock(&ls->ls_clear_proc_locks);
233}
234
235static int device_user_lock(struct dlm_user_proc *proc,
236 struct dlm_lock_params *params)
237{
238 struct dlm_ls *ls;
239 struct dlm_user_args *ua;
240 uint32_t lkid;
241 int error = -ENOMEM;
242
243 ls = dlm_find_lockspace_local(proc->lockspace);
244 if (!ls)
245 return -ENOENT;
246
247 if (!params->castaddr || !params->lksb) {
248 error = -EINVAL;
249 goto out;
250 }
251
252 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
253 if (!ua)
254 goto out;
255 ua->proc = proc;
256 ua->user_lksb = params->lksb;
257 ua->castparam = params->castparam;
258 ua->castaddr = params->castaddr;
259 ua->bastparam = params->bastparam;
260 ua->bastaddr = params->bastaddr;
261 ua->xid = params->xid;
262
263 if (params->flags & DLM_LKF_CONVERT) {
264 error = dlm_user_convert(ls, ua,
265 params->mode, params->flags,
266 params->lkid, params->lvb,
267 (unsigned long) params->timeout);
268 } else if (params->flags & DLM_LKF_ORPHAN) {
269 error = dlm_user_adopt_orphan(ls, ua,
270 params->mode, params->flags,
271 params->name, params->namelen,
272 (unsigned long) params->timeout,
273 &lkid);
274 if (!error)
275 error = lkid;
276 } else {
277 error = dlm_user_request(ls, ua,
278 params->mode, params->flags,
279 params->name, params->namelen,
280 (unsigned long) params->timeout);
281 if (!error)
282 error = ua->lksb.sb_lkid;
283 }
284 out:
285 dlm_put_lockspace(ls);
286 return error;
287}
288
289static int device_user_unlock(struct dlm_user_proc *proc,
290 struct dlm_lock_params *params)
291{
292 struct dlm_ls *ls;
293 struct dlm_user_args *ua;
294 int error = -ENOMEM;
295
296 ls = dlm_find_lockspace_local(proc->lockspace);
297 if (!ls)
298 return -ENOENT;
299
300 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
301 if (!ua)
302 goto out;
303 ua->proc = proc;
304 ua->user_lksb = params->lksb;
305 ua->castparam = params->castparam;
306 ua->castaddr = params->castaddr;
307
308 if (params->flags & DLM_LKF_CANCEL)
309 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
310 else
311 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
312 params->lvb);
313 out:
314 dlm_put_lockspace(ls);
315 return error;
316}
317
318static int device_user_deadlock(struct dlm_user_proc *proc,
319 struct dlm_lock_params *params)
320{
321 struct dlm_ls *ls;
322 int error;
323
324 ls = dlm_find_lockspace_local(proc->lockspace);
325 if (!ls)
326 return -ENOENT;
327
328 error = dlm_user_deadlock(ls, params->flags, params->lkid);
329
330 dlm_put_lockspace(ls);
331 return error;
332}
333
334static int dlm_device_register(struct dlm_ls *ls, char *name)
335{
336 int error, len;
337
338 /* The device is already registered. This happens when the
339 lockspace is created multiple times from userspace. */
340 if (ls->ls_device.name)
341 return 0;
342
343 error = -ENOMEM;
344 len = strlen(name) + strlen(name_prefix) + 2;
345 ls->ls_device.name = kzalloc(len, GFP_NOFS);
346 if (!ls->ls_device.name)
347 goto fail;
348
349 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
350 name);
351 ls->ls_device.fops = &device_fops;
352 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
353
354 error = misc_register(&ls->ls_device);
355 if (error) {
356 kfree(ls->ls_device.name);
357 }
358fail:
359 return error;
360}
361
362int dlm_device_deregister(struct dlm_ls *ls)
363{
364 /* The device is not registered. This happens when the lockspace
365 was never used from userspace, or when device_create_lockspace()
366 calls dlm_release_lockspace() after the register fails. */
367 if (!ls->ls_device.name)
368 return 0;
369
370 misc_deregister(&ls->ls_device);
371 kfree(ls->ls_device.name);
372 return 0;
373}
374
375static int device_user_purge(struct dlm_user_proc *proc,
376 struct dlm_purge_params *params)
377{
378 struct dlm_ls *ls;
379 int error;
380
381 ls = dlm_find_lockspace_local(proc->lockspace);
382 if (!ls)
383 return -ENOENT;
384
385 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
386
387 dlm_put_lockspace(ls);
388 return error;
389}
390
391static int device_create_lockspace(struct dlm_lspace_params *params)
392{
393 dlm_lockspace_t *lockspace;
394 struct dlm_ls *ls;
395 int error;
396
397 if (!capable(CAP_SYS_ADMIN))
398 return -EPERM;
399
400 error = dlm_new_lockspace(params->name, NULL, params->flags,
401 DLM_USER_LVB_LEN, NULL, NULL, NULL,
402 &lockspace);
403 if (error)
404 return error;
405
406 ls = dlm_find_lockspace_local(lockspace);
407 if (!ls)
408 return -ENOENT;
409
410 error = dlm_device_register(ls, params->name);
411 dlm_put_lockspace(ls);
412
413 if (error)
414 dlm_release_lockspace(lockspace, 0);
415 else
416 error = ls->ls_device.minor;
417
418 return error;
419}
420
421static int device_remove_lockspace(struct dlm_lspace_params *params)
422{
423 dlm_lockspace_t *lockspace;
424 struct dlm_ls *ls;
425 int error, force = 0;
426
427 if (!capable(CAP_SYS_ADMIN))
428 return -EPERM;
429
430 ls = dlm_find_lockspace_device(params->minor);
431 if (!ls)
432 return -ENOENT;
433
434 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
435 force = 2;
436
437 lockspace = ls->ls_local_handle;
438 dlm_put_lockspace(ls);
439
440 /* The final dlm_release_lockspace waits for references to go to
441 zero, so all processes will need to close their device for the
442 ls before the release will proceed. release also calls the
443 device_deregister above. Converting a positive return value
444 from release to zero means that userspace won't know when its
445 release was the final one, but it shouldn't need to know. */
446
447 error = dlm_release_lockspace(lockspace, force);
448 if (error > 0)
449 error = 0;
450 return error;
451}
452
453/* Check the user's version matches ours */
454static int check_version(struct dlm_write_request *req)
455{
456 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
457 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
458 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
459
460 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
461 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
462 current->comm,
463 task_pid_nr(current),
464 req->version[0],
465 req->version[1],
466 req->version[2],
467 DLM_DEVICE_VERSION_MAJOR,
468 DLM_DEVICE_VERSION_MINOR,
469 DLM_DEVICE_VERSION_PATCH);
470 return -EINVAL;
471 }
472 return 0;
473}
474
475/*
476 * device_write
477 *
478 * device_user_lock
479 * dlm_user_request -> request_lock
480 * dlm_user_convert -> convert_lock
481 *
482 * device_user_unlock
483 * dlm_user_unlock -> unlock_lock
484 * dlm_user_cancel -> cancel_lock
485 *
486 * device_create_lockspace
487 * dlm_new_lockspace
488 *
489 * device_remove_lockspace
490 * dlm_release_lockspace
491 */
492
493/* a write to a lockspace device is a lock or unlock request, a write
494 to the control device is to create/remove a lockspace */
495
496static ssize_t device_write(struct file *file, const char __user *buf,
497 size_t count, loff_t *ppos)
498{
499 struct dlm_user_proc *proc = file->private_data;
500 struct dlm_write_request *kbuf;
501 int error;
502
503#ifdef CONFIG_COMPAT
504 if (count < sizeof(struct dlm_write_request32))
505#else
506 if (count < sizeof(struct dlm_write_request))
507#endif
508 return -EINVAL;
509
510 /*
511 * can't compare against COMPAT/dlm_write_request32 because
512 * we don't yet know if is64bit is zero
513 */
514 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
515 return -EINVAL;
516
517 kbuf = memdup_user_nul(buf, count);
518 if (IS_ERR(kbuf))
519 return PTR_ERR(kbuf);
520
521 if (check_version(kbuf)) {
522 error = -EBADE;
523 goto out_free;
524 }
525
526#ifdef CONFIG_COMPAT
527 if (!kbuf->is64bit) {
528 struct dlm_write_request32 *k32buf;
529 int namelen = 0;
530
531 if (count > sizeof(struct dlm_write_request32))
532 namelen = count - sizeof(struct dlm_write_request32);
533
534 k32buf = (struct dlm_write_request32 *)kbuf;
535
536 /* add 1 after namelen so that the name string is terminated */
537 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
538 GFP_NOFS);
539 if (!kbuf) {
540 kfree(k32buf);
541 return -ENOMEM;
542 }
543
544 if (proc)
545 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
546
547 compat_input(kbuf, k32buf, namelen);
548 kfree(k32buf);
549 }
550#endif
551
552 /* do we really need this? can a write happen after a close? */
553 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
554 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
555 error = -EINVAL;
556 goto out_free;
557 }
558
559 error = -EINVAL;
560
561 switch (kbuf->cmd)
562 {
563 case DLM_USER_LOCK:
564 if (!proc) {
565 log_print("no locking on control device");
566 goto out_free;
567 }
568 error = device_user_lock(proc, &kbuf->i.lock);
569 break;
570
571 case DLM_USER_UNLOCK:
572 if (!proc) {
573 log_print("no locking on control device");
574 goto out_free;
575 }
576 error = device_user_unlock(proc, &kbuf->i.lock);
577 break;
578
579 case DLM_USER_DEADLOCK:
580 if (!proc) {
581 log_print("no locking on control device");
582 goto out_free;
583 }
584 error = device_user_deadlock(proc, &kbuf->i.lock);
585 break;
586
587 case DLM_USER_CREATE_LOCKSPACE:
588 if (proc) {
589 log_print("create/remove only on control device");
590 goto out_free;
591 }
592 error = device_create_lockspace(&kbuf->i.lspace);
593 break;
594
595 case DLM_USER_REMOVE_LOCKSPACE:
596 if (proc) {
597 log_print("create/remove only on control device");
598 goto out_free;
599 }
600 error = device_remove_lockspace(&kbuf->i.lspace);
601 break;
602
603 case DLM_USER_PURGE:
604 if (!proc) {
605 log_print("no locking on control device");
606 goto out_free;
607 }
608 error = device_user_purge(proc, &kbuf->i.purge);
609 break;
610
611 default:
612 log_print("Unknown command passed to DLM device : %d\n",
613 kbuf->cmd);
614 }
615
616 out_free:
617 kfree(kbuf);
618 return error;
619}
620
621/* Every process that opens the lockspace device has its own "proc" structure
622 hanging off the open file that's used to keep track of locks owned by the
623 process and asts that need to be delivered to the process. */
624
625static int device_open(struct inode *inode, struct file *file)
626{
627 struct dlm_user_proc *proc;
628 struct dlm_ls *ls;
629
630 ls = dlm_find_lockspace_device(iminor(inode));
631 if (!ls)
632 return -ENOENT;
633
634 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
635 if (!proc) {
636 dlm_put_lockspace(ls);
637 return -ENOMEM;
638 }
639
640 proc->lockspace = ls->ls_local_handle;
641 INIT_LIST_HEAD(&proc->asts);
642 INIT_LIST_HEAD(&proc->locks);
643 INIT_LIST_HEAD(&proc->unlocking);
644 spin_lock_init(&proc->asts_spin);
645 spin_lock_init(&proc->locks_spin);
646 init_waitqueue_head(&proc->wait);
647 file->private_data = proc;
648
649 return 0;
650}
651
652static int device_close(struct inode *inode, struct file *file)
653{
654 struct dlm_user_proc *proc = file->private_data;
655 struct dlm_ls *ls;
656
657 ls = dlm_find_lockspace_local(proc->lockspace);
658 if (!ls)
659 return -ENOENT;
660
661 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
662
663 dlm_clear_proc_locks(ls, proc);
664
665 /* at this point no more lkb's should exist for this lockspace,
666 so there's no chance of dlm_user_add_ast() being called and
667 looking for lkb->ua->proc */
668
669 kfree(proc);
670 file->private_data = NULL;
671
672 dlm_put_lockspace(ls);
673 dlm_put_lockspace(ls); /* for the find in device_open() */
674
675 /* FIXME: AUTOFREE: if this ls is no longer used do
676 device_remove_lockspace() */
677
678 return 0;
679}
680
681static int copy_result_to_user(struct dlm_user_args *ua, int compat,
682 uint32_t flags, int mode, int copy_lvb,
683 char __user *buf, size_t count)
684{
685#ifdef CONFIG_COMPAT
686 struct dlm_lock_result32 result32;
687#endif
688 struct dlm_lock_result result;
689 void *resultptr;
690 int error=0;
691 int len;
692 int struct_len;
693
694 memset(&result, 0, sizeof(struct dlm_lock_result));
695 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
696 result.version[1] = DLM_DEVICE_VERSION_MINOR;
697 result.version[2] = DLM_DEVICE_VERSION_PATCH;
698 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
699 result.user_lksb = ua->user_lksb;
700
701 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
702 in a conversion unless the conversion is successful. See code
703 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
704 notes that a new blocking AST address and parameter are set even if
705 the conversion fails, so maybe we should just do that. */
706
707 if (flags & DLM_CB_BAST) {
708 result.user_astaddr = ua->bastaddr;
709 result.user_astparam = ua->bastparam;
710 result.bast_mode = mode;
711 } else {
712 result.user_astaddr = ua->castaddr;
713 result.user_astparam = ua->castparam;
714 }
715
716#ifdef CONFIG_COMPAT
717 if (compat)
718 len = sizeof(struct dlm_lock_result32);
719 else
720#endif
721 len = sizeof(struct dlm_lock_result);
722 struct_len = len;
723
724 /* copy lvb to userspace if there is one, it's been updated, and
725 the user buffer has space for it */
726
727 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
728 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
729 DLM_USER_LVB_LEN)) {
730 error = -EFAULT;
731 goto out;
732 }
733
734 result.lvb_offset = len;
735 len += DLM_USER_LVB_LEN;
736 }
737
738 result.length = len;
739 resultptr = &result;
740#ifdef CONFIG_COMPAT
741 if (compat) {
742 compat_output(&result, &result32);
743 resultptr = &result32;
744 }
745#endif
746
747 if (copy_to_user(buf, resultptr, struct_len))
748 error = -EFAULT;
749 else
750 error = len;
751 out:
752 return error;
753}
754
755static int copy_version_to_user(char __user *buf, size_t count)
756{
757 struct dlm_device_version ver;
758
759 memset(&ver, 0, sizeof(struct dlm_device_version));
760 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
761 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
762 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
763
764 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
765 return -EFAULT;
766 return sizeof(struct dlm_device_version);
767}
768
769/* a read returns a single ast described in a struct dlm_lock_result */
770
771static ssize_t device_read(struct file *file, char __user *buf, size_t count,
772 loff_t *ppos)
773{
774 struct dlm_user_proc *proc = file->private_data;
775 struct dlm_lkb *lkb;
776 DECLARE_WAITQUEUE(wait, current);
777 struct dlm_callback cb;
778 int rv, resid, copy_lvb = 0;
779 int old_mode, new_mode;
780
781 if (count == sizeof(struct dlm_device_version)) {
782 rv = copy_version_to_user(buf, count);
783 return rv;
784 }
785
786 if (!proc) {
787 log_print("non-version read from control device %zu", count);
788 return -EINVAL;
789 }
790
791#ifdef CONFIG_COMPAT
792 if (count < sizeof(struct dlm_lock_result32))
793#else
794 if (count < sizeof(struct dlm_lock_result))
795#endif
796 return -EINVAL;
797
798 try_another:
799
800 /* do we really need this? can a read happen after a close? */
801 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
802 return -EINVAL;
803
804 spin_lock(&proc->asts_spin);
805 if (list_empty(&proc->asts)) {
806 if (file->f_flags & O_NONBLOCK) {
807 spin_unlock(&proc->asts_spin);
808 return -EAGAIN;
809 }
810
811 add_wait_queue(&proc->wait, &wait);
812
813 repeat:
814 set_current_state(TASK_INTERRUPTIBLE);
815 if (list_empty(&proc->asts) && !signal_pending(current)) {
816 spin_unlock(&proc->asts_spin);
817 schedule();
818 spin_lock(&proc->asts_spin);
819 goto repeat;
820 }
821 set_current_state(TASK_RUNNING);
822 remove_wait_queue(&proc->wait, &wait);
823
824 if (signal_pending(current)) {
825 spin_unlock(&proc->asts_spin);
826 return -ERESTARTSYS;
827 }
828 }
829
830 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
831 without removing lkb_cb_list; so empty lkb_cb_list is always
832 consistent with empty lkb_callbacks */
833
834 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
835
836 /* rem_lkb_callback sets a new lkb_last_cast */
837 old_mode = lkb->lkb_last_cast.mode;
838
839 rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
840 if (rv < 0) {
841 /* this shouldn't happen; lkb should have been removed from
842 list when resid was zero */
843 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
844 list_del_init(&lkb->lkb_cb_list);
845 spin_unlock(&proc->asts_spin);
846 /* removes ref for proc->asts, may cause lkb to be freed */
847 dlm_put_lkb(lkb);
848 goto try_another;
849 }
850 if (!resid)
851 list_del_init(&lkb->lkb_cb_list);
852 spin_unlock(&proc->asts_spin);
853
854 if (cb.flags & DLM_CB_SKIP) {
855 /* removes ref for proc->asts, may cause lkb to be freed */
856 if (!resid)
857 dlm_put_lkb(lkb);
858 goto try_another;
859 }
860
861 if (cb.flags & DLM_CB_CAST) {
862 new_mode = cb.mode;
863
864 if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
865 dlm_lvb_operations[old_mode + 1][new_mode + 1])
866 copy_lvb = 1;
867
868 lkb->lkb_lksb->sb_status = cb.sb_status;
869 lkb->lkb_lksb->sb_flags = cb.sb_flags;
870 }
871
872 rv = copy_result_to_user(lkb->lkb_ua,
873 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
874 cb.flags, cb.mode, copy_lvb, buf, count);
875
876 /* removes ref for proc->asts, may cause lkb to be freed */
877 if (!resid)
878 dlm_put_lkb(lkb);
879
880 return rv;
881}
882
883static unsigned int device_poll(struct file *file, poll_table *wait)
884{
885 struct dlm_user_proc *proc = file->private_data;
886
887 poll_wait(file, &proc->wait, wait);
888
889 spin_lock(&proc->asts_spin);
890 if (!list_empty(&proc->asts)) {
891 spin_unlock(&proc->asts_spin);
892 return POLLIN | POLLRDNORM;
893 }
894 spin_unlock(&proc->asts_spin);
895 return 0;
896}
897
898int dlm_user_daemon_available(void)
899{
900 /* dlm_controld hasn't started (or, has started, but not
901 properly populated configfs) */
902
903 if (!dlm_our_nodeid())
904 return 0;
905
906 /* This is to deal with versions of dlm_controld that don't
907 know about the monitor device. We assume that if the
908 dlm_controld was started (above), but the monitor device
909 was never opened, that it's an old version. dlm_controld
910 should open the monitor device before populating configfs. */
911
912 if (dlm_monitor_unused)
913 return 1;
914
915 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
916}
917
918static int ctl_device_open(struct inode *inode, struct file *file)
919{
920 file->private_data = NULL;
921 return 0;
922}
923
924static int ctl_device_close(struct inode *inode, struct file *file)
925{
926 return 0;
927}
928
929static int monitor_device_open(struct inode *inode, struct file *file)
930{
931 atomic_inc(&dlm_monitor_opened);
932 dlm_monitor_unused = 0;
933 return 0;
934}
935
936static int monitor_device_close(struct inode *inode, struct file *file)
937{
938 if (atomic_dec_and_test(&dlm_monitor_opened))
939 dlm_stop_lockspaces();
940 return 0;
941}
942
943static const struct file_operations device_fops = {
944 .open = device_open,
945 .release = device_close,
946 .read = device_read,
947 .write = device_write,
948 .poll = device_poll,
949 .owner = THIS_MODULE,
950 .llseek = noop_llseek,
951};
952
953static const struct file_operations ctl_device_fops = {
954 .open = ctl_device_open,
955 .release = ctl_device_close,
956 .read = device_read,
957 .write = device_write,
958 .owner = THIS_MODULE,
959 .llseek = noop_llseek,
960};
961
962static struct miscdevice ctl_device = {
963 .name = "dlm-control",
964 .fops = &ctl_device_fops,
965 .minor = MISC_DYNAMIC_MINOR,
966};
967
968static const struct file_operations monitor_device_fops = {
969 .open = monitor_device_open,
970 .release = monitor_device_close,
971 .owner = THIS_MODULE,
972 .llseek = noop_llseek,
973};
974
975static struct miscdevice monitor_device = {
976 .name = "dlm-monitor",
977 .fops = &monitor_device_fops,
978 .minor = MISC_DYNAMIC_MINOR,
979};
980
981int __init dlm_user_init(void)
982{
983 int error;
984
985 atomic_set(&dlm_monitor_opened, 0);
986
987 error = misc_register(&ctl_device);
988 if (error) {
989 log_print("misc_register failed for control device");
990 goto out;
991 }
992
993 error = misc_register(&monitor_device);
994 if (error) {
995 log_print("misc_register failed for monitor device");
996 misc_deregister(&ctl_device);
997 }
998 out:
999 return error;
1000}
1001
1002void dlm_user_exit(void)
1003{
1004 misc_deregister(&ctl_device);
1005 misc_deregister(&monitor_device);
1006}
1007