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 <trace/events/dlm.h>
20
21#include "dlm_internal.h"
22#include "lockspace.h"
23#include "lock.h"
24#include "lvb_table.h"
25#include "user.h"
26#include "ast.h"
27#include "config.h"
28#include "memory.h"
29
30static const char name_prefix[] = "dlm";
31static const struct file_operations device_fops;
32static atomic_t dlm_monitor_opened;
33static int dlm_monitor_unused = 1;
34
35#ifdef CONFIG_COMPAT
36
37struct dlm_lock_params32 {
38 __u8 mode;
39 __u8 namelen;
40 __u16 unused;
41 __u32 flags;
42 __u32 lkid;
43 __u32 parent;
44 __u64 xid;
45 __u64 timeout;
46 __u32 castparam;
47 __u32 castaddr;
48 __u32 bastparam;
49 __u32 bastaddr;
50 __u32 lksb;
51 char lvb[DLM_USER_LVB_LEN];
52 char name[];
53};
54
55struct dlm_write_request32 {
56 __u32 version[3];
57 __u8 cmd;
58 __u8 is64bit;
59 __u8 unused[2];
60
61 union {
62 struct dlm_lock_params32 lock;
63 struct dlm_lspace_params lspace;
64 struct dlm_purge_params purge;
65 } i;
66};
67
68struct dlm_lksb32 {
69 __u32 sb_status;
70 __u32 sb_lkid;
71 __u8 sb_flags;
72 __u32 sb_lvbptr;
73};
74
75struct dlm_lock_result32 {
76 __u32 version[3];
77 __u32 length;
78 __u32 user_astaddr;
79 __u32 user_astparam;
80 __u32 user_lksb;
81 struct dlm_lksb32 lksb;
82 __u8 bast_mode;
83 __u8 unused[3];
84 /* Offsets may be zero if no data is present */
85 __u32 lvb_offset;
86};
87
88static void compat_input(struct dlm_write_request *kb,
89 struct dlm_write_request32 *kb32,
90 int namelen)
91{
92 kb->version[0] = kb32->version[0];
93 kb->version[1] = kb32->version[1];
94 kb->version[2] = kb32->version[2];
95
96 kb->cmd = kb32->cmd;
97 kb->is64bit = kb32->is64bit;
98 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
99 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
100 kb->i.lspace.flags = kb32->i.lspace.flags;
101 kb->i.lspace.minor = kb32->i.lspace.minor;
102 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
103 } else if (kb->cmd == DLM_USER_PURGE) {
104 kb->i.purge.nodeid = kb32->i.purge.nodeid;
105 kb->i.purge.pid = kb32->i.purge.pid;
106 } else {
107 kb->i.lock.mode = kb32->i.lock.mode;
108 kb->i.lock.namelen = kb32->i.lock.namelen;
109 kb->i.lock.flags = kb32->i.lock.flags;
110 kb->i.lock.lkid = kb32->i.lock.lkid;
111 kb->i.lock.parent = kb32->i.lock.parent;
112 kb->i.lock.xid = kb32->i.lock.xid;
113 kb->i.lock.timeout = kb32->i.lock.timeout;
114 kb->i.lock.castparam = (__user void *)(long)kb32->i.lock.castparam;
115 kb->i.lock.castaddr = (__user void *)(long)kb32->i.lock.castaddr;
116 kb->i.lock.bastparam = (__user void *)(long)kb32->i.lock.bastparam;
117 kb->i.lock.bastaddr = (__user void *)(long)kb32->i.lock.bastaddr;
118 kb->i.lock.lksb = (__user void *)(long)kb32->i.lock.lksb;
119 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
120 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
121 }
122}
123
124static void compat_output(struct dlm_lock_result *res,
125 struct dlm_lock_result32 *res32)
126{
127 memset(res32, 0, sizeof(*res32));
128
129 res32->version[0] = res->version[0];
130 res32->version[1] = res->version[1];
131 res32->version[2] = res->version[2];
132
133 res32->user_astaddr = (__u32)(__force long)res->user_astaddr;
134 res32->user_astparam = (__u32)(__force long)res->user_astparam;
135 res32->user_lksb = (__u32)(__force long)res->user_lksb;
136 res32->bast_mode = res->bast_mode;
137
138 res32->lvb_offset = res->lvb_offset;
139 res32->length = res->length;
140
141 res32->lksb.sb_status = res->lksb.sb_status;
142 res32->lksb.sb_flags = res->lksb.sb_flags;
143 res32->lksb.sb_lkid = res->lksb.sb_lkid;
144 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
145}
146#endif
147
148/* Figure out if this lock is at the end of its life and no longer
149 available for the application to use. The lkb still exists until
150 the final ast is read. A lock becomes EOL in three situations:
151 1. a noqueue request fails with EAGAIN
152 2. an unlock completes with EUNLOCK
153 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
154 An EOL lock needs to be removed from the process's list of locks.
155 And we can't allow any new operation on an EOL lock. This is
156 not related to the lifetime of the lkb struct which is managed
157 entirely by refcount. */
158
159static int lkb_is_endoflife(int mode, int status)
160{
161 switch (status) {
162 case -DLM_EUNLOCK:
163 return 1;
164 case -DLM_ECANCEL:
165 case -ETIMEDOUT:
166 case -EDEADLK:
167 case -EAGAIN:
168 if (mode == DLM_LOCK_IV)
169 return 1;
170 break;
171 }
172 return 0;
173}
174
175/* we could possibly check if the cancel of an orphan has resulted in the lkb
176 being removed and then remove that lkb from the orphans list and free it */
177
178void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
179 int status, uint32_t sbflags)
180{
181 struct dlm_ls *ls;
182 struct dlm_user_args *ua;
183 struct dlm_user_proc *proc;
184 struct dlm_callback *cb;
185 int rv, copy_lvb;
186
187 if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
188 test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
189 return;
190
191 ls = lkb->lkb_resource->res_ls;
192 spin_lock_bh(&ls->ls_clear_proc_locks);
193
194 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
195 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
196 lkb->ua so we can't try to use it. This second check is necessary
197 for cases where a completion ast is received for an operation that
198 began before clear_proc_locks did its cancel/unlock. */
199
200 if (test_bit(DLM_DFL_ORPHAN_BIT, &lkb->lkb_dflags) ||
201 test_bit(DLM_IFL_DEAD_BIT, &lkb->lkb_iflags))
202 goto out;
203
204 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
205 ua = lkb->lkb_ua;
206 proc = ua->proc;
207
208 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
209 goto out;
210
211 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
212 set_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags);
213
214 spin_lock_bh(&proc->asts_spin);
215
216 if (!dlm_may_skip_callback(lkb, flags, mode, status, sbflags,
217 ©_lvb)) {
218 rv = dlm_get_cb(lkb, flags, mode, status, sbflags, &cb);
219 if (!rv) {
220 cb->copy_lvb = copy_lvb;
221 cb->ua = *ua;
222 cb->lkb_lksb = &cb->ua.lksb;
223 if (copy_lvb) {
224 memcpy(cb->lvbptr, ua->lksb.sb_lvbptr,
225 DLM_USER_LVB_LEN);
226 cb->lkb_lksb->sb_lvbptr = cb->lvbptr;
227 }
228
229 list_add_tail(&cb->list, &proc->asts);
230 wake_up_interruptible(&proc->wait);
231 }
232 }
233 spin_unlock_bh(&proc->asts_spin);
234
235 if (test_bit(DLM_IFL_ENDOFLIFE_BIT, &lkb->lkb_iflags)) {
236 /* N.B. spin_lock locks_spin, not asts_spin */
237 spin_lock_bh(&proc->locks_spin);
238 if (!list_empty(&lkb->lkb_ownqueue)) {
239 list_del_init(&lkb->lkb_ownqueue);
240 dlm_put_lkb(lkb);
241 }
242 spin_unlock_bh(&proc->locks_spin);
243 }
244 out:
245 spin_unlock_bh(&ls->ls_clear_proc_locks);
246}
247
248static int device_user_lock(struct dlm_user_proc *proc,
249 struct dlm_lock_params *params)
250{
251 struct dlm_ls *ls;
252 struct dlm_user_args *ua;
253 uint32_t lkid;
254 int error = -ENOMEM;
255
256 ls = dlm_find_lockspace_local(proc->lockspace);
257 if (!ls)
258 return -ENOENT;
259
260 if (!params->castaddr || !params->lksb) {
261 error = -EINVAL;
262 goto out;
263 }
264
265 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
266 if (!ua)
267 goto out;
268 ua->proc = proc;
269 ua->user_lksb = params->lksb;
270 ua->castparam = params->castparam;
271 ua->castaddr = params->castaddr;
272 ua->bastparam = params->bastparam;
273 ua->bastaddr = params->bastaddr;
274 ua->xid = params->xid;
275
276 if (params->flags & DLM_LKF_CONVERT) {
277 error = dlm_user_convert(ls, ua,
278 params->mode, params->flags,
279 params->lkid, params->lvb);
280 } else if (params->flags & DLM_LKF_ORPHAN) {
281 error = dlm_user_adopt_orphan(ls, ua,
282 params->mode, params->flags,
283 params->name, params->namelen,
284 &lkid);
285 if (!error)
286 error = lkid;
287 } else {
288 error = dlm_user_request(ls, ua,
289 params->mode, params->flags,
290 params->name, params->namelen);
291 if (!error)
292 error = ua->lksb.sb_lkid;
293 }
294 out:
295 dlm_put_lockspace(ls);
296 return error;
297}
298
299static int device_user_unlock(struct dlm_user_proc *proc,
300 struct dlm_lock_params *params)
301{
302 struct dlm_ls *ls;
303 struct dlm_user_args *ua;
304 int error = -ENOMEM;
305
306 ls = dlm_find_lockspace_local(proc->lockspace);
307 if (!ls)
308 return -ENOENT;
309
310 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
311 if (!ua)
312 goto out;
313 ua->proc = proc;
314 ua->user_lksb = params->lksb;
315 ua->castparam = params->castparam;
316 ua->castaddr = params->castaddr;
317
318 if (params->flags & DLM_LKF_CANCEL)
319 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
320 else
321 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
322 params->lvb);
323 out:
324 dlm_put_lockspace(ls);
325 return error;
326}
327
328static int device_user_deadlock(struct dlm_user_proc *proc,
329 struct dlm_lock_params *params)
330{
331 struct dlm_ls *ls;
332 int error;
333
334 ls = dlm_find_lockspace_local(proc->lockspace);
335 if (!ls)
336 return -ENOENT;
337
338 error = dlm_user_deadlock(ls, params->flags, params->lkid);
339
340 dlm_put_lockspace(ls);
341 return error;
342}
343
344static int dlm_device_register(struct dlm_ls *ls, char *name)
345{
346 int error, len;
347
348 /* The device is already registered. This happens when the
349 lockspace is created multiple times from userspace. */
350 if (ls->ls_device.name)
351 return 0;
352
353 error = -ENOMEM;
354 len = strlen(name) + strlen(name_prefix) + 2;
355 ls->ls_device.name = kzalloc(len, GFP_NOFS);
356 if (!ls->ls_device.name)
357 goto fail;
358
359 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
360 name);
361 ls->ls_device.fops = &device_fops;
362 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
363
364 error = misc_register(&ls->ls_device);
365 if (error) {
366 kfree(ls->ls_device.name);
367 /* this has to be set to NULL
368 * to avoid a double-free in dlm_device_deregister
369 */
370 ls->ls_device.name = NULL;
371 }
372fail:
373 return error;
374}
375
376int dlm_device_deregister(struct dlm_ls *ls)
377{
378 /* The device is not registered. This happens when the lockspace
379 was never used from userspace, or when device_create_lockspace()
380 calls dlm_release_lockspace() after the register fails. */
381 if (!ls->ls_device.name)
382 return 0;
383
384 misc_deregister(&ls->ls_device);
385 kfree(ls->ls_device.name);
386 return 0;
387}
388
389static int device_user_purge(struct dlm_user_proc *proc,
390 struct dlm_purge_params *params)
391{
392 struct dlm_ls *ls;
393 int error;
394
395 ls = dlm_find_lockspace_local(proc->lockspace);
396 if (!ls)
397 return -ENOENT;
398
399 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
400
401 dlm_put_lockspace(ls);
402 return error;
403}
404
405static int device_create_lockspace(struct dlm_lspace_params *params)
406{
407 dlm_lockspace_t *lockspace;
408 struct dlm_ls *ls;
409 int error;
410
411 if (!capable(CAP_SYS_ADMIN))
412 return -EPERM;
413
414 error = dlm_new_user_lockspace(params->name, dlm_config.ci_cluster_name,
415 params->flags, DLM_USER_LVB_LEN, NULL,
416 NULL, NULL, &lockspace);
417 if (error)
418 return error;
419
420 ls = dlm_find_lockspace_local(lockspace);
421 if (!ls)
422 return -ENOENT;
423
424 error = dlm_device_register(ls, params->name);
425 dlm_put_lockspace(ls);
426
427 if (error)
428 dlm_release_lockspace(lockspace, 0);
429 else
430 error = ls->ls_device.minor;
431
432 return error;
433}
434
435static int device_remove_lockspace(struct dlm_lspace_params *params)
436{
437 dlm_lockspace_t *lockspace;
438 struct dlm_ls *ls;
439 int error, force = 0;
440
441 if (!capable(CAP_SYS_ADMIN))
442 return -EPERM;
443
444 ls = dlm_find_lockspace_device(params->minor);
445 if (!ls)
446 return -ENOENT;
447
448 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
449 force = 2;
450
451 lockspace = ls;
452 dlm_put_lockspace(ls);
453
454 /* The final dlm_release_lockspace waits for references to go to
455 zero, so all processes will need to close their device for the
456 ls before the release will proceed. release also calls the
457 device_deregister above. Converting a positive return value
458 from release to zero means that userspace won't know when its
459 release was the final one, but it shouldn't need to know. */
460
461 error = dlm_release_lockspace(lockspace, force);
462 if (error > 0)
463 error = 0;
464 return error;
465}
466
467/* Check the user's version matches ours */
468static int check_version(struct dlm_write_request *req)
469{
470 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
471 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
472 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
473
474 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
475 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
476 current->comm,
477 task_pid_nr(current),
478 req->version[0],
479 req->version[1],
480 req->version[2],
481 DLM_DEVICE_VERSION_MAJOR,
482 DLM_DEVICE_VERSION_MINOR,
483 DLM_DEVICE_VERSION_PATCH);
484 return -EINVAL;
485 }
486 return 0;
487}
488
489/*
490 * device_write
491 *
492 * device_user_lock
493 * dlm_user_request -> request_lock
494 * dlm_user_convert -> convert_lock
495 *
496 * device_user_unlock
497 * dlm_user_unlock -> unlock_lock
498 * dlm_user_cancel -> cancel_lock
499 *
500 * device_create_lockspace
501 * dlm_new_lockspace
502 *
503 * device_remove_lockspace
504 * dlm_release_lockspace
505 */
506
507/* a write to a lockspace device is a lock or unlock request, a write
508 to the control device is to create/remove a lockspace */
509
510static ssize_t device_write(struct file *file, const char __user *buf,
511 size_t count, loff_t *ppos)
512{
513 struct dlm_user_proc *proc = file->private_data;
514 struct dlm_write_request *kbuf;
515 int error;
516
517#ifdef CONFIG_COMPAT
518 if (count < sizeof(struct dlm_write_request32))
519#else
520 if (count < sizeof(struct dlm_write_request))
521#endif
522 return -EINVAL;
523
524 /*
525 * can't compare against COMPAT/dlm_write_request32 because
526 * we don't yet know if is64bit is zero
527 */
528 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
529 return -EINVAL;
530
531 kbuf = memdup_user_nul(buf, count);
532 if (IS_ERR(kbuf))
533 return PTR_ERR(kbuf);
534
535 if (check_version(kbuf)) {
536 error = -EBADE;
537 goto out_free;
538 }
539
540#ifdef CONFIG_COMPAT
541 if (!kbuf->is64bit) {
542 struct dlm_write_request32 *k32buf;
543 int namelen = 0;
544
545 if (count > sizeof(struct dlm_write_request32))
546 namelen = count - sizeof(struct dlm_write_request32);
547
548 k32buf = (struct dlm_write_request32 *)kbuf;
549
550 /* add 1 after namelen so that the name string is terminated */
551 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
552 GFP_NOFS);
553 if (!kbuf) {
554 kfree(k32buf);
555 return -ENOMEM;
556 }
557
558 if (proc)
559 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
560
561 compat_input(kbuf, k32buf, namelen);
562 kfree(k32buf);
563 }
564#endif
565
566 /* do we really need this? can a write happen after a close? */
567 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
568 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
569 error = -EINVAL;
570 goto out_free;
571 }
572
573 error = -EINVAL;
574
575 switch (kbuf->cmd)
576 {
577 case DLM_USER_LOCK:
578 if (!proc) {
579 log_print("no locking on control device");
580 goto out_free;
581 }
582 error = device_user_lock(proc, &kbuf->i.lock);
583 break;
584
585 case DLM_USER_UNLOCK:
586 if (!proc) {
587 log_print("no locking on control device");
588 goto out_free;
589 }
590 error = device_user_unlock(proc, &kbuf->i.lock);
591 break;
592
593 case DLM_USER_DEADLOCK:
594 if (!proc) {
595 log_print("no locking on control device");
596 goto out_free;
597 }
598 error = device_user_deadlock(proc, &kbuf->i.lock);
599 break;
600
601 case DLM_USER_CREATE_LOCKSPACE:
602 if (proc) {
603 log_print("create/remove only on control device");
604 goto out_free;
605 }
606 error = device_create_lockspace(&kbuf->i.lspace);
607 break;
608
609 case DLM_USER_REMOVE_LOCKSPACE:
610 if (proc) {
611 log_print("create/remove only on control device");
612 goto out_free;
613 }
614 error = device_remove_lockspace(&kbuf->i.lspace);
615 break;
616
617 case DLM_USER_PURGE:
618 if (!proc) {
619 log_print("no locking on control device");
620 goto out_free;
621 }
622 error = device_user_purge(proc, &kbuf->i.purge);
623 break;
624
625 default:
626 log_print("Unknown command passed to DLM device : %d\n",
627 kbuf->cmd);
628 }
629
630 out_free:
631 kfree(kbuf);
632 return error;
633}
634
635/* Every process that opens the lockspace device has its own "proc" structure
636 hanging off the open file that's used to keep track of locks owned by the
637 process and asts that need to be delivered to the process. */
638
639static int device_open(struct inode *inode, struct file *file)
640{
641 struct dlm_user_proc *proc;
642 struct dlm_ls *ls;
643
644 ls = dlm_find_lockspace_device(iminor(inode));
645 if (!ls)
646 return -ENOENT;
647
648 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
649 if (!proc) {
650 dlm_put_lockspace(ls);
651 return -ENOMEM;
652 }
653
654 proc->lockspace = ls;
655 INIT_LIST_HEAD(&proc->asts);
656 INIT_LIST_HEAD(&proc->locks);
657 INIT_LIST_HEAD(&proc->unlocking);
658 spin_lock_init(&proc->asts_spin);
659 spin_lock_init(&proc->locks_spin);
660 init_waitqueue_head(&proc->wait);
661 file->private_data = proc;
662
663 return 0;
664}
665
666static int device_close(struct inode *inode, struct file *file)
667{
668 struct dlm_user_proc *proc = file->private_data;
669 struct dlm_ls *ls;
670
671 ls = dlm_find_lockspace_local(proc->lockspace);
672 if (!ls)
673 return -ENOENT;
674
675 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
676
677 dlm_clear_proc_locks(ls, proc);
678
679 /* at this point no more lkb's should exist for this lockspace,
680 so there's no chance of dlm_user_add_ast() being called and
681 looking for lkb->ua->proc */
682
683 kfree(proc);
684 file->private_data = NULL;
685
686 dlm_put_lockspace(ls);
687 dlm_put_lockspace(ls); /* for the find in device_open() */
688
689 /* FIXME: AUTOFREE: if this ls is no longer used do
690 device_remove_lockspace() */
691
692 return 0;
693}
694
695static int copy_result_to_user(struct dlm_user_args *ua, int compat,
696 uint32_t flags, int mode, int copy_lvb,
697 char __user *buf, size_t count)
698{
699#ifdef CONFIG_COMPAT
700 struct dlm_lock_result32 result32;
701#endif
702 struct dlm_lock_result result;
703 void *resultptr;
704 int error=0;
705 int len;
706 int struct_len;
707
708 memset(&result, 0, sizeof(struct dlm_lock_result));
709 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
710 result.version[1] = DLM_DEVICE_VERSION_MINOR;
711 result.version[2] = DLM_DEVICE_VERSION_PATCH;
712 memcpy(&result.lksb, &ua->lksb, offsetof(struct dlm_lksb, sb_lvbptr));
713 result.user_lksb = ua->user_lksb;
714
715 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
716 in a conversion unless the conversion is successful. See code
717 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
718 notes that a new blocking AST address and parameter are set even if
719 the conversion fails, so maybe we should just do that. */
720
721 if (flags & DLM_CB_BAST) {
722 result.user_astaddr = ua->bastaddr;
723 result.user_astparam = ua->bastparam;
724 result.bast_mode = mode;
725 } else {
726 result.user_astaddr = ua->castaddr;
727 result.user_astparam = ua->castparam;
728 }
729
730#ifdef CONFIG_COMPAT
731 if (compat)
732 len = sizeof(struct dlm_lock_result32);
733 else
734#endif
735 len = sizeof(struct dlm_lock_result);
736 struct_len = len;
737
738 /* copy lvb to userspace if there is one, it's been updated, and
739 the user buffer has space for it */
740
741 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
742 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
743 DLM_USER_LVB_LEN)) {
744 error = -EFAULT;
745 goto out;
746 }
747
748 result.lvb_offset = len;
749 len += DLM_USER_LVB_LEN;
750 }
751
752 result.length = len;
753 resultptr = &result;
754#ifdef CONFIG_COMPAT
755 if (compat) {
756 compat_output(&result, &result32);
757 resultptr = &result32;
758 }
759#endif
760
761 if (copy_to_user(buf, resultptr, struct_len))
762 error = -EFAULT;
763 else
764 error = len;
765 out:
766 return error;
767}
768
769static int copy_version_to_user(char __user *buf, size_t count)
770{
771 struct dlm_device_version ver;
772
773 memset(&ver, 0, sizeof(struct dlm_device_version));
774 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
775 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
776 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
777
778 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
779 return -EFAULT;
780 return sizeof(struct dlm_device_version);
781}
782
783/* a read returns a single ast described in a struct dlm_lock_result */
784
785static ssize_t device_read(struct file *file, char __user *buf, size_t count,
786 loff_t *ppos)
787{
788 struct dlm_user_proc *proc = file->private_data;
789 DECLARE_WAITQUEUE(wait, current);
790 struct dlm_callback *cb;
791 int rv, ret;
792
793 if (count == sizeof(struct dlm_device_version)) {
794 rv = copy_version_to_user(buf, count);
795 return rv;
796 }
797
798 if (!proc) {
799 log_print("non-version read from control device %zu", count);
800 return -EINVAL;
801 }
802
803#ifdef CONFIG_COMPAT
804 if (count < sizeof(struct dlm_lock_result32))
805#else
806 if (count < sizeof(struct dlm_lock_result))
807#endif
808 return -EINVAL;
809
810 /* do we really need this? can a read happen after a close? */
811 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
812 return -EINVAL;
813
814 spin_lock_bh(&proc->asts_spin);
815 if (list_empty(&proc->asts)) {
816 if (file->f_flags & O_NONBLOCK) {
817 spin_unlock_bh(&proc->asts_spin);
818 return -EAGAIN;
819 }
820
821 add_wait_queue(&proc->wait, &wait);
822
823 repeat:
824 set_current_state(TASK_INTERRUPTIBLE);
825 if (list_empty(&proc->asts) && !signal_pending(current)) {
826 spin_unlock_bh(&proc->asts_spin);
827 schedule();
828 spin_lock_bh(&proc->asts_spin);
829 goto repeat;
830 }
831 set_current_state(TASK_RUNNING);
832 remove_wait_queue(&proc->wait, &wait);
833
834 if (signal_pending(current)) {
835 spin_unlock_bh(&proc->asts_spin);
836 return -ERESTARTSYS;
837 }
838 }
839
840 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
841 without removing lkb_cb_list; so empty lkb_cb_list is always
842 consistent with empty lkb_callbacks */
843
844 cb = list_first_entry(&proc->asts, struct dlm_callback, list);
845 list_del(&cb->list);
846 spin_unlock_bh(&proc->asts_spin);
847
848 if (cb->flags & DLM_CB_BAST) {
849 trace_dlm_bast(cb->ls_id, cb->lkb_id, cb->mode, cb->res_name,
850 cb->res_length);
851 } else if (cb->flags & DLM_CB_CAST) {
852 cb->lkb_lksb->sb_status = cb->sb_status;
853 cb->lkb_lksb->sb_flags = cb->sb_flags;
854 trace_dlm_ast(cb->ls_id, cb->lkb_id, cb->sb_status,
855 cb->sb_flags, cb->res_name, cb->res_length);
856 }
857
858 ret = copy_result_to_user(&cb->ua,
859 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
860 cb->flags, cb->mode, cb->copy_lvb, buf, count);
861 dlm_free_cb(cb);
862 return ret;
863}
864
865static __poll_t device_poll(struct file *file, poll_table *wait)
866{
867 struct dlm_user_proc *proc = file->private_data;
868
869 poll_wait(file, &proc->wait, wait);
870
871 spin_lock_bh(&proc->asts_spin);
872 if (!list_empty(&proc->asts)) {
873 spin_unlock_bh(&proc->asts_spin);
874 return EPOLLIN | EPOLLRDNORM;
875 }
876 spin_unlock_bh(&proc->asts_spin);
877 return 0;
878}
879
880int dlm_user_daemon_available(void)
881{
882 /* dlm_controld hasn't started (or, has started, but not
883 properly populated configfs) */
884
885 if (!dlm_our_nodeid())
886 return 0;
887
888 /* This is to deal with versions of dlm_controld that don't
889 know about the monitor device. We assume that if the
890 dlm_controld was started (above), but the monitor device
891 was never opened, that it's an old version. dlm_controld
892 should open the monitor device before populating configfs. */
893
894 if (dlm_monitor_unused)
895 return 1;
896
897 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
898}
899
900static int ctl_device_open(struct inode *inode, struct file *file)
901{
902 file->private_data = NULL;
903 return 0;
904}
905
906static int ctl_device_close(struct inode *inode, struct file *file)
907{
908 return 0;
909}
910
911static int monitor_device_open(struct inode *inode, struct file *file)
912{
913 atomic_inc(&dlm_monitor_opened);
914 dlm_monitor_unused = 0;
915 return 0;
916}
917
918static int monitor_device_close(struct inode *inode, struct file *file)
919{
920 if (atomic_dec_and_test(&dlm_monitor_opened))
921 dlm_stop_lockspaces();
922 return 0;
923}
924
925static const struct file_operations device_fops = {
926 .open = device_open,
927 .release = device_close,
928 .read = device_read,
929 .write = device_write,
930 .poll = device_poll,
931 .owner = THIS_MODULE,
932 .llseek = noop_llseek,
933};
934
935static const struct file_operations ctl_device_fops = {
936 .open = ctl_device_open,
937 .release = ctl_device_close,
938 .read = device_read,
939 .write = device_write,
940 .owner = THIS_MODULE,
941 .llseek = noop_llseek,
942};
943
944static struct miscdevice ctl_device = {
945 .name = "dlm-control",
946 .fops = &ctl_device_fops,
947 .minor = MISC_DYNAMIC_MINOR,
948};
949
950static const struct file_operations monitor_device_fops = {
951 .open = monitor_device_open,
952 .release = monitor_device_close,
953 .owner = THIS_MODULE,
954 .llseek = noop_llseek,
955};
956
957static struct miscdevice monitor_device = {
958 .name = "dlm-monitor",
959 .fops = &monitor_device_fops,
960 .minor = MISC_DYNAMIC_MINOR,
961};
962
963int __init dlm_user_init(void)
964{
965 int error;
966
967 atomic_set(&dlm_monitor_opened, 0);
968
969 error = misc_register(&ctl_device);
970 if (error) {
971 log_print("misc_register failed for control device");
972 goto out;
973 }
974
975 error = misc_register(&monitor_device);
976 if (error) {
977 log_print("misc_register failed for monitor device");
978 misc_deregister(&ctl_device);
979 }
980 out:
981 return error;
982}
983
984void dlm_user_exit(void)
985{
986 misc_deregister(&ctl_device);
987 misc_deregister(&monitor_device);
988}
989
1/*
2 * Copyright (C) 2006-2010 Red Hat, Inc. All rights reserved.
3 *
4 * This copyrighted material is made available to anyone wishing to use,
5 * modify, copy, or redistribute it subject to the terms and conditions
6 * of the GNU General Public License v.2.
7 */
8
9#include <linux/miscdevice.h>
10#include <linux/init.h>
11#include <linux/wait.h>
12#include <linux/file.h>
13#include <linux/fs.h>
14#include <linux/poll.h>
15#include <linux/signal.h>
16#include <linux/spinlock.h>
17#include <linux/dlm.h>
18#include <linux/dlm_device.h>
19#include <linux/slab.h>
20#include <linux/sched/signal.h>
21
22#include "dlm_internal.h"
23#include "lockspace.h"
24#include "lock.h"
25#include "lvb_table.h"
26#include "user.h"
27#include "ast.h"
28
29static const char name_prefix[] = "dlm";
30static const struct file_operations device_fops;
31static atomic_t dlm_monitor_opened;
32static int dlm_monitor_unused = 1;
33
34#ifdef CONFIG_COMPAT
35
36struct dlm_lock_params32 {
37 __u8 mode;
38 __u8 namelen;
39 __u16 unused;
40 __u32 flags;
41 __u32 lkid;
42 __u32 parent;
43 __u64 xid;
44 __u64 timeout;
45 __u32 castparam;
46 __u32 castaddr;
47 __u32 bastparam;
48 __u32 bastaddr;
49 __u32 lksb;
50 char lvb[DLM_USER_LVB_LEN];
51 char name[0];
52};
53
54struct dlm_write_request32 {
55 __u32 version[3];
56 __u8 cmd;
57 __u8 is64bit;
58 __u8 unused[2];
59
60 union {
61 struct dlm_lock_params32 lock;
62 struct dlm_lspace_params lspace;
63 struct dlm_purge_params purge;
64 } i;
65};
66
67struct dlm_lksb32 {
68 __u32 sb_status;
69 __u32 sb_lkid;
70 __u8 sb_flags;
71 __u32 sb_lvbptr;
72};
73
74struct dlm_lock_result32 {
75 __u32 version[3];
76 __u32 length;
77 __u32 user_astaddr;
78 __u32 user_astparam;
79 __u32 user_lksb;
80 struct dlm_lksb32 lksb;
81 __u8 bast_mode;
82 __u8 unused[3];
83 /* Offsets may be zero if no data is present */
84 __u32 lvb_offset;
85};
86
87static void compat_input(struct dlm_write_request *kb,
88 struct dlm_write_request32 *kb32,
89 int namelen)
90{
91 kb->version[0] = kb32->version[0];
92 kb->version[1] = kb32->version[1];
93 kb->version[2] = kb32->version[2];
94
95 kb->cmd = kb32->cmd;
96 kb->is64bit = kb32->is64bit;
97 if (kb->cmd == DLM_USER_CREATE_LOCKSPACE ||
98 kb->cmd == DLM_USER_REMOVE_LOCKSPACE) {
99 kb->i.lspace.flags = kb32->i.lspace.flags;
100 kb->i.lspace.minor = kb32->i.lspace.minor;
101 memcpy(kb->i.lspace.name, kb32->i.lspace.name, namelen);
102 } else if (kb->cmd == DLM_USER_PURGE) {
103 kb->i.purge.nodeid = kb32->i.purge.nodeid;
104 kb->i.purge.pid = kb32->i.purge.pid;
105 } else {
106 kb->i.lock.mode = kb32->i.lock.mode;
107 kb->i.lock.namelen = kb32->i.lock.namelen;
108 kb->i.lock.flags = kb32->i.lock.flags;
109 kb->i.lock.lkid = kb32->i.lock.lkid;
110 kb->i.lock.parent = kb32->i.lock.parent;
111 kb->i.lock.xid = kb32->i.lock.xid;
112 kb->i.lock.timeout = kb32->i.lock.timeout;
113 kb->i.lock.castparam = (void *)(long)kb32->i.lock.castparam;
114 kb->i.lock.castaddr = (void *)(long)kb32->i.lock.castaddr;
115 kb->i.lock.bastparam = (void *)(long)kb32->i.lock.bastparam;
116 kb->i.lock.bastaddr = (void *)(long)kb32->i.lock.bastaddr;
117 kb->i.lock.lksb = (void *)(long)kb32->i.lock.lksb;
118 memcpy(kb->i.lock.lvb, kb32->i.lock.lvb, DLM_USER_LVB_LEN);
119 memcpy(kb->i.lock.name, kb32->i.lock.name, namelen);
120 }
121}
122
123static void compat_output(struct dlm_lock_result *res,
124 struct dlm_lock_result32 *res32)
125{
126 memset(res32, 0, sizeof(*res32));
127
128 res32->version[0] = res->version[0];
129 res32->version[1] = res->version[1];
130 res32->version[2] = res->version[2];
131
132 res32->user_astaddr = (__u32)(long)res->user_astaddr;
133 res32->user_astparam = (__u32)(long)res->user_astparam;
134 res32->user_lksb = (__u32)(long)res->user_lksb;
135 res32->bast_mode = res->bast_mode;
136
137 res32->lvb_offset = res->lvb_offset;
138 res32->length = res->length;
139
140 res32->lksb.sb_status = res->lksb.sb_status;
141 res32->lksb.sb_flags = res->lksb.sb_flags;
142 res32->lksb.sb_lkid = res->lksb.sb_lkid;
143 res32->lksb.sb_lvbptr = (__u32)(long)res->lksb.sb_lvbptr;
144}
145#endif
146
147/* Figure out if this lock is at the end of its life and no longer
148 available for the application to use. The lkb still exists until
149 the final ast is read. A lock becomes EOL in three situations:
150 1. a noqueue request fails with EAGAIN
151 2. an unlock completes with EUNLOCK
152 3. a cancel of a waiting request completes with ECANCEL/EDEADLK
153 An EOL lock needs to be removed from the process's list of locks.
154 And we can't allow any new operation on an EOL lock. This is
155 not related to the lifetime of the lkb struct which is managed
156 entirely by refcount. */
157
158static int lkb_is_endoflife(int mode, int status)
159{
160 switch (status) {
161 case -DLM_EUNLOCK:
162 return 1;
163 case -DLM_ECANCEL:
164 case -ETIMEDOUT:
165 case -EDEADLK:
166 case -EAGAIN:
167 if (mode == DLM_LOCK_IV)
168 return 1;
169 break;
170 }
171 return 0;
172}
173
174/* we could possibly check if the cancel of an orphan has resulted in the lkb
175 being removed and then remove that lkb from the orphans list and free it */
176
177void dlm_user_add_ast(struct dlm_lkb *lkb, uint32_t flags, int mode,
178 int status, uint32_t sbflags, uint64_t seq)
179{
180 struct dlm_ls *ls;
181 struct dlm_user_args *ua;
182 struct dlm_user_proc *proc;
183 int rv;
184
185 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
186 return;
187
188 ls = lkb->lkb_resource->res_ls;
189 mutex_lock(&ls->ls_clear_proc_locks);
190
191 /* If ORPHAN/DEAD flag is set, it means the process is dead so an ast
192 can't be delivered. For ORPHAN's, dlm_clear_proc_locks() freed
193 lkb->ua so we can't try to use it. This second check is necessary
194 for cases where a completion ast is received for an operation that
195 began before clear_proc_locks did its cancel/unlock. */
196
197 if (lkb->lkb_flags & (DLM_IFL_ORPHAN | DLM_IFL_DEAD))
198 goto out;
199
200 DLM_ASSERT(lkb->lkb_ua, dlm_print_lkb(lkb););
201 ua = lkb->lkb_ua;
202 proc = ua->proc;
203
204 if ((flags & DLM_CB_BAST) && ua->bastaddr == NULL)
205 goto out;
206
207 if ((flags & DLM_CB_CAST) && lkb_is_endoflife(mode, status))
208 lkb->lkb_flags |= DLM_IFL_ENDOFLIFE;
209
210 spin_lock(&proc->asts_spin);
211
212 rv = dlm_add_lkb_callback(lkb, flags, mode, status, sbflags, seq);
213 if (rv < 0) {
214 spin_unlock(&proc->asts_spin);
215 goto out;
216 }
217
218 if (list_empty(&lkb->lkb_cb_list)) {
219 kref_get(&lkb->lkb_ref);
220 list_add_tail(&lkb->lkb_cb_list, &proc->asts);
221 wake_up_interruptible(&proc->wait);
222 }
223 spin_unlock(&proc->asts_spin);
224
225 if (lkb->lkb_flags & DLM_IFL_ENDOFLIFE) {
226 /* N.B. spin_lock locks_spin, not asts_spin */
227 spin_lock(&proc->locks_spin);
228 if (!list_empty(&lkb->lkb_ownqueue)) {
229 list_del_init(&lkb->lkb_ownqueue);
230 dlm_put_lkb(lkb);
231 }
232 spin_unlock(&proc->locks_spin);
233 }
234 out:
235 mutex_unlock(&ls->ls_clear_proc_locks);
236}
237
238static int device_user_lock(struct dlm_user_proc *proc,
239 struct dlm_lock_params *params)
240{
241 struct dlm_ls *ls;
242 struct dlm_user_args *ua;
243 uint32_t lkid;
244 int error = -ENOMEM;
245
246 ls = dlm_find_lockspace_local(proc->lockspace);
247 if (!ls)
248 return -ENOENT;
249
250 if (!params->castaddr || !params->lksb) {
251 error = -EINVAL;
252 goto out;
253 }
254
255 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
256 if (!ua)
257 goto out;
258 ua->proc = proc;
259 ua->user_lksb = params->lksb;
260 ua->castparam = params->castparam;
261 ua->castaddr = params->castaddr;
262 ua->bastparam = params->bastparam;
263 ua->bastaddr = params->bastaddr;
264 ua->xid = params->xid;
265
266 if (params->flags & DLM_LKF_CONVERT) {
267 error = dlm_user_convert(ls, ua,
268 params->mode, params->flags,
269 params->lkid, params->lvb,
270 (unsigned long) params->timeout);
271 } else if (params->flags & DLM_LKF_ORPHAN) {
272 error = dlm_user_adopt_orphan(ls, ua,
273 params->mode, params->flags,
274 params->name, params->namelen,
275 (unsigned long) params->timeout,
276 &lkid);
277 if (!error)
278 error = lkid;
279 } else {
280 error = dlm_user_request(ls, ua,
281 params->mode, params->flags,
282 params->name, params->namelen,
283 (unsigned long) params->timeout);
284 if (!error)
285 error = ua->lksb.sb_lkid;
286 }
287 out:
288 dlm_put_lockspace(ls);
289 return error;
290}
291
292static int device_user_unlock(struct dlm_user_proc *proc,
293 struct dlm_lock_params *params)
294{
295 struct dlm_ls *ls;
296 struct dlm_user_args *ua;
297 int error = -ENOMEM;
298
299 ls = dlm_find_lockspace_local(proc->lockspace);
300 if (!ls)
301 return -ENOENT;
302
303 ua = kzalloc(sizeof(struct dlm_user_args), GFP_NOFS);
304 if (!ua)
305 goto out;
306 ua->proc = proc;
307 ua->user_lksb = params->lksb;
308 ua->castparam = params->castparam;
309 ua->castaddr = params->castaddr;
310
311 if (params->flags & DLM_LKF_CANCEL)
312 error = dlm_user_cancel(ls, ua, params->flags, params->lkid);
313 else
314 error = dlm_user_unlock(ls, ua, params->flags, params->lkid,
315 params->lvb);
316 out:
317 dlm_put_lockspace(ls);
318 return error;
319}
320
321static int device_user_deadlock(struct dlm_user_proc *proc,
322 struct dlm_lock_params *params)
323{
324 struct dlm_ls *ls;
325 int error;
326
327 ls = dlm_find_lockspace_local(proc->lockspace);
328 if (!ls)
329 return -ENOENT;
330
331 error = dlm_user_deadlock(ls, params->flags, params->lkid);
332
333 dlm_put_lockspace(ls);
334 return error;
335}
336
337static int dlm_device_register(struct dlm_ls *ls, char *name)
338{
339 int error, len;
340
341 /* The device is already registered. This happens when the
342 lockspace is created multiple times from userspace. */
343 if (ls->ls_device.name)
344 return 0;
345
346 error = -ENOMEM;
347 len = strlen(name) + strlen(name_prefix) + 2;
348 ls->ls_device.name = kzalloc(len, GFP_NOFS);
349 if (!ls->ls_device.name)
350 goto fail;
351
352 snprintf((char *)ls->ls_device.name, len, "%s_%s", name_prefix,
353 name);
354 ls->ls_device.fops = &device_fops;
355 ls->ls_device.minor = MISC_DYNAMIC_MINOR;
356
357 error = misc_register(&ls->ls_device);
358 if (error) {
359 kfree(ls->ls_device.name);
360 /* this has to be set to NULL
361 * to avoid a double-free in dlm_device_deregister
362 */
363 ls->ls_device.name = NULL;
364 }
365fail:
366 return error;
367}
368
369int dlm_device_deregister(struct dlm_ls *ls)
370{
371 /* The device is not registered. This happens when the lockspace
372 was never used from userspace, or when device_create_lockspace()
373 calls dlm_release_lockspace() after the register fails. */
374 if (!ls->ls_device.name)
375 return 0;
376
377 misc_deregister(&ls->ls_device);
378 kfree(ls->ls_device.name);
379 return 0;
380}
381
382static int device_user_purge(struct dlm_user_proc *proc,
383 struct dlm_purge_params *params)
384{
385 struct dlm_ls *ls;
386 int error;
387
388 ls = dlm_find_lockspace_local(proc->lockspace);
389 if (!ls)
390 return -ENOENT;
391
392 error = dlm_user_purge(ls, proc, params->nodeid, params->pid);
393
394 dlm_put_lockspace(ls);
395 return error;
396}
397
398static int device_create_lockspace(struct dlm_lspace_params *params)
399{
400 dlm_lockspace_t *lockspace;
401 struct dlm_ls *ls;
402 int error;
403
404 if (!capable(CAP_SYS_ADMIN))
405 return -EPERM;
406
407 error = dlm_new_lockspace(params->name, NULL, params->flags,
408 DLM_USER_LVB_LEN, NULL, NULL, NULL,
409 &lockspace);
410 if (error)
411 return error;
412
413 ls = dlm_find_lockspace_local(lockspace);
414 if (!ls)
415 return -ENOENT;
416
417 error = dlm_device_register(ls, params->name);
418 dlm_put_lockspace(ls);
419
420 if (error)
421 dlm_release_lockspace(lockspace, 0);
422 else
423 error = ls->ls_device.minor;
424
425 return error;
426}
427
428static int device_remove_lockspace(struct dlm_lspace_params *params)
429{
430 dlm_lockspace_t *lockspace;
431 struct dlm_ls *ls;
432 int error, force = 0;
433
434 if (!capable(CAP_SYS_ADMIN))
435 return -EPERM;
436
437 ls = dlm_find_lockspace_device(params->minor);
438 if (!ls)
439 return -ENOENT;
440
441 if (params->flags & DLM_USER_LSFLG_FORCEFREE)
442 force = 2;
443
444 lockspace = ls->ls_local_handle;
445 dlm_put_lockspace(ls);
446
447 /* The final dlm_release_lockspace waits for references to go to
448 zero, so all processes will need to close their device for the
449 ls before the release will proceed. release also calls the
450 device_deregister above. Converting a positive return value
451 from release to zero means that userspace won't know when its
452 release was the final one, but it shouldn't need to know. */
453
454 error = dlm_release_lockspace(lockspace, force);
455 if (error > 0)
456 error = 0;
457 return error;
458}
459
460/* Check the user's version matches ours */
461static int check_version(struct dlm_write_request *req)
462{
463 if (req->version[0] != DLM_DEVICE_VERSION_MAJOR ||
464 (req->version[0] == DLM_DEVICE_VERSION_MAJOR &&
465 req->version[1] > DLM_DEVICE_VERSION_MINOR)) {
466
467 printk(KERN_DEBUG "dlm: process %s (%d) version mismatch "
468 "user (%d.%d.%d) kernel (%d.%d.%d)\n",
469 current->comm,
470 task_pid_nr(current),
471 req->version[0],
472 req->version[1],
473 req->version[2],
474 DLM_DEVICE_VERSION_MAJOR,
475 DLM_DEVICE_VERSION_MINOR,
476 DLM_DEVICE_VERSION_PATCH);
477 return -EINVAL;
478 }
479 return 0;
480}
481
482/*
483 * device_write
484 *
485 * device_user_lock
486 * dlm_user_request -> request_lock
487 * dlm_user_convert -> convert_lock
488 *
489 * device_user_unlock
490 * dlm_user_unlock -> unlock_lock
491 * dlm_user_cancel -> cancel_lock
492 *
493 * device_create_lockspace
494 * dlm_new_lockspace
495 *
496 * device_remove_lockspace
497 * dlm_release_lockspace
498 */
499
500/* a write to a lockspace device is a lock or unlock request, a write
501 to the control device is to create/remove a lockspace */
502
503static ssize_t device_write(struct file *file, const char __user *buf,
504 size_t count, loff_t *ppos)
505{
506 struct dlm_user_proc *proc = file->private_data;
507 struct dlm_write_request *kbuf;
508 int error;
509
510#ifdef CONFIG_COMPAT
511 if (count < sizeof(struct dlm_write_request32))
512#else
513 if (count < sizeof(struct dlm_write_request))
514#endif
515 return -EINVAL;
516
517 /*
518 * can't compare against COMPAT/dlm_write_request32 because
519 * we don't yet know if is64bit is zero
520 */
521 if (count > sizeof(struct dlm_write_request) + DLM_RESNAME_MAXLEN)
522 return -EINVAL;
523
524 kbuf = memdup_user_nul(buf, count);
525 if (IS_ERR(kbuf))
526 return PTR_ERR(kbuf);
527
528 if (check_version(kbuf)) {
529 error = -EBADE;
530 goto out_free;
531 }
532
533#ifdef CONFIG_COMPAT
534 if (!kbuf->is64bit) {
535 struct dlm_write_request32 *k32buf;
536 int namelen = 0;
537
538 if (count > sizeof(struct dlm_write_request32))
539 namelen = count - sizeof(struct dlm_write_request32);
540
541 k32buf = (struct dlm_write_request32 *)kbuf;
542
543 /* add 1 after namelen so that the name string is terminated */
544 kbuf = kzalloc(sizeof(struct dlm_write_request) + namelen + 1,
545 GFP_NOFS);
546 if (!kbuf) {
547 kfree(k32buf);
548 return -ENOMEM;
549 }
550
551 if (proc)
552 set_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags);
553
554 compat_input(kbuf, k32buf, namelen);
555 kfree(k32buf);
556 }
557#endif
558
559 /* do we really need this? can a write happen after a close? */
560 if ((kbuf->cmd == DLM_USER_LOCK || kbuf->cmd == DLM_USER_UNLOCK) &&
561 (proc && test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))) {
562 error = -EINVAL;
563 goto out_free;
564 }
565
566 error = -EINVAL;
567
568 switch (kbuf->cmd)
569 {
570 case DLM_USER_LOCK:
571 if (!proc) {
572 log_print("no locking on control device");
573 goto out_free;
574 }
575 error = device_user_lock(proc, &kbuf->i.lock);
576 break;
577
578 case DLM_USER_UNLOCK:
579 if (!proc) {
580 log_print("no locking on control device");
581 goto out_free;
582 }
583 error = device_user_unlock(proc, &kbuf->i.lock);
584 break;
585
586 case DLM_USER_DEADLOCK:
587 if (!proc) {
588 log_print("no locking on control device");
589 goto out_free;
590 }
591 error = device_user_deadlock(proc, &kbuf->i.lock);
592 break;
593
594 case DLM_USER_CREATE_LOCKSPACE:
595 if (proc) {
596 log_print("create/remove only on control device");
597 goto out_free;
598 }
599 error = device_create_lockspace(&kbuf->i.lspace);
600 break;
601
602 case DLM_USER_REMOVE_LOCKSPACE:
603 if (proc) {
604 log_print("create/remove only on control device");
605 goto out_free;
606 }
607 error = device_remove_lockspace(&kbuf->i.lspace);
608 break;
609
610 case DLM_USER_PURGE:
611 if (!proc) {
612 log_print("no locking on control device");
613 goto out_free;
614 }
615 error = device_user_purge(proc, &kbuf->i.purge);
616 break;
617
618 default:
619 log_print("Unknown command passed to DLM device : %d\n",
620 kbuf->cmd);
621 }
622
623 out_free:
624 kfree(kbuf);
625 return error;
626}
627
628/* Every process that opens the lockspace device has its own "proc" structure
629 hanging off the open file that's used to keep track of locks owned by the
630 process and asts that need to be delivered to the process. */
631
632static int device_open(struct inode *inode, struct file *file)
633{
634 struct dlm_user_proc *proc;
635 struct dlm_ls *ls;
636
637 ls = dlm_find_lockspace_device(iminor(inode));
638 if (!ls)
639 return -ENOENT;
640
641 proc = kzalloc(sizeof(struct dlm_user_proc), GFP_NOFS);
642 if (!proc) {
643 dlm_put_lockspace(ls);
644 return -ENOMEM;
645 }
646
647 proc->lockspace = ls->ls_local_handle;
648 INIT_LIST_HEAD(&proc->asts);
649 INIT_LIST_HEAD(&proc->locks);
650 INIT_LIST_HEAD(&proc->unlocking);
651 spin_lock_init(&proc->asts_spin);
652 spin_lock_init(&proc->locks_spin);
653 init_waitqueue_head(&proc->wait);
654 file->private_data = proc;
655
656 return 0;
657}
658
659static int device_close(struct inode *inode, struct file *file)
660{
661 struct dlm_user_proc *proc = file->private_data;
662 struct dlm_ls *ls;
663
664 ls = dlm_find_lockspace_local(proc->lockspace);
665 if (!ls)
666 return -ENOENT;
667
668 set_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags);
669
670 dlm_clear_proc_locks(ls, proc);
671
672 /* at this point no more lkb's should exist for this lockspace,
673 so there's no chance of dlm_user_add_ast() being called and
674 looking for lkb->ua->proc */
675
676 kfree(proc);
677 file->private_data = NULL;
678
679 dlm_put_lockspace(ls);
680 dlm_put_lockspace(ls); /* for the find in device_open() */
681
682 /* FIXME: AUTOFREE: if this ls is no longer used do
683 device_remove_lockspace() */
684
685 return 0;
686}
687
688static int copy_result_to_user(struct dlm_user_args *ua, int compat,
689 uint32_t flags, int mode, int copy_lvb,
690 char __user *buf, size_t count)
691{
692#ifdef CONFIG_COMPAT
693 struct dlm_lock_result32 result32;
694#endif
695 struct dlm_lock_result result;
696 void *resultptr;
697 int error=0;
698 int len;
699 int struct_len;
700
701 memset(&result, 0, sizeof(struct dlm_lock_result));
702 result.version[0] = DLM_DEVICE_VERSION_MAJOR;
703 result.version[1] = DLM_DEVICE_VERSION_MINOR;
704 result.version[2] = DLM_DEVICE_VERSION_PATCH;
705 memcpy(&result.lksb, &ua->lksb, sizeof(struct dlm_lksb));
706 result.user_lksb = ua->user_lksb;
707
708 /* FIXME: dlm1 provides for the user's bastparam/addr to not be updated
709 in a conversion unless the conversion is successful. See code
710 in dlm_user_convert() for updating ua from ua_tmp. OpenVMS, though,
711 notes that a new blocking AST address and parameter are set even if
712 the conversion fails, so maybe we should just do that. */
713
714 if (flags & DLM_CB_BAST) {
715 result.user_astaddr = ua->bastaddr;
716 result.user_astparam = ua->bastparam;
717 result.bast_mode = mode;
718 } else {
719 result.user_astaddr = ua->castaddr;
720 result.user_astparam = ua->castparam;
721 }
722
723#ifdef CONFIG_COMPAT
724 if (compat)
725 len = sizeof(struct dlm_lock_result32);
726 else
727#endif
728 len = sizeof(struct dlm_lock_result);
729 struct_len = len;
730
731 /* copy lvb to userspace if there is one, it's been updated, and
732 the user buffer has space for it */
733
734 if (copy_lvb && ua->lksb.sb_lvbptr && count >= len + DLM_USER_LVB_LEN) {
735 if (copy_to_user(buf+len, ua->lksb.sb_lvbptr,
736 DLM_USER_LVB_LEN)) {
737 error = -EFAULT;
738 goto out;
739 }
740
741 result.lvb_offset = len;
742 len += DLM_USER_LVB_LEN;
743 }
744
745 result.length = len;
746 resultptr = &result;
747#ifdef CONFIG_COMPAT
748 if (compat) {
749 compat_output(&result, &result32);
750 resultptr = &result32;
751 }
752#endif
753
754 if (copy_to_user(buf, resultptr, struct_len))
755 error = -EFAULT;
756 else
757 error = len;
758 out:
759 return error;
760}
761
762static int copy_version_to_user(char __user *buf, size_t count)
763{
764 struct dlm_device_version ver;
765
766 memset(&ver, 0, sizeof(struct dlm_device_version));
767 ver.version[0] = DLM_DEVICE_VERSION_MAJOR;
768 ver.version[1] = DLM_DEVICE_VERSION_MINOR;
769 ver.version[2] = DLM_DEVICE_VERSION_PATCH;
770
771 if (copy_to_user(buf, &ver, sizeof(struct dlm_device_version)))
772 return -EFAULT;
773 return sizeof(struct dlm_device_version);
774}
775
776/* a read returns a single ast described in a struct dlm_lock_result */
777
778static ssize_t device_read(struct file *file, char __user *buf, size_t count,
779 loff_t *ppos)
780{
781 struct dlm_user_proc *proc = file->private_data;
782 struct dlm_lkb *lkb;
783 DECLARE_WAITQUEUE(wait, current);
784 struct dlm_callback cb;
785 int rv, resid, copy_lvb = 0;
786 int old_mode, new_mode;
787
788 if (count == sizeof(struct dlm_device_version)) {
789 rv = copy_version_to_user(buf, count);
790 return rv;
791 }
792
793 if (!proc) {
794 log_print("non-version read from control device %zu", count);
795 return -EINVAL;
796 }
797
798#ifdef CONFIG_COMPAT
799 if (count < sizeof(struct dlm_lock_result32))
800#else
801 if (count < sizeof(struct dlm_lock_result))
802#endif
803 return -EINVAL;
804
805 try_another:
806
807 /* do we really need this? can a read happen after a close? */
808 if (test_bit(DLM_PROC_FLAGS_CLOSING, &proc->flags))
809 return -EINVAL;
810
811 spin_lock(&proc->asts_spin);
812 if (list_empty(&proc->asts)) {
813 if (file->f_flags & O_NONBLOCK) {
814 spin_unlock(&proc->asts_spin);
815 return -EAGAIN;
816 }
817
818 add_wait_queue(&proc->wait, &wait);
819
820 repeat:
821 set_current_state(TASK_INTERRUPTIBLE);
822 if (list_empty(&proc->asts) && !signal_pending(current)) {
823 spin_unlock(&proc->asts_spin);
824 schedule();
825 spin_lock(&proc->asts_spin);
826 goto repeat;
827 }
828 set_current_state(TASK_RUNNING);
829 remove_wait_queue(&proc->wait, &wait);
830
831 if (signal_pending(current)) {
832 spin_unlock(&proc->asts_spin);
833 return -ERESTARTSYS;
834 }
835 }
836
837 /* if we empty lkb_callbacks, we don't want to unlock the spinlock
838 without removing lkb_cb_list; so empty lkb_cb_list is always
839 consistent with empty lkb_callbacks */
840
841 lkb = list_entry(proc->asts.next, struct dlm_lkb, lkb_cb_list);
842
843 /* rem_lkb_callback sets a new lkb_last_cast */
844 old_mode = lkb->lkb_last_cast.mode;
845
846 rv = dlm_rem_lkb_callback(lkb->lkb_resource->res_ls, lkb, &cb, &resid);
847 if (rv < 0) {
848 /* this shouldn't happen; lkb should have been removed from
849 list when resid was zero */
850 log_print("dlm_rem_lkb_callback empty %x", lkb->lkb_id);
851 list_del_init(&lkb->lkb_cb_list);
852 spin_unlock(&proc->asts_spin);
853 /* removes ref for proc->asts, may cause lkb to be freed */
854 dlm_put_lkb(lkb);
855 goto try_another;
856 }
857 if (!resid)
858 list_del_init(&lkb->lkb_cb_list);
859 spin_unlock(&proc->asts_spin);
860
861 if (cb.flags & DLM_CB_SKIP) {
862 /* removes ref for proc->asts, may cause lkb to be freed */
863 if (!resid)
864 dlm_put_lkb(lkb);
865 goto try_another;
866 }
867
868 if (cb.flags & DLM_CB_CAST) {
869 new_mode = cb.mode;
870
871 if (!cb.sb_status && lkb->lkb_lksb->sb_lvbptr &&
872 dlm_lvb_operations[old_mode + 1][new_mode + 1])
873 copy_lvb = 1;
874
875 lkb->lkb_lksb->sb_status = cb.sb_status;
876 lkb->lkb_lksb->sb_flags = cb.sb_flags;
877 }
878
879 rv = copy_result_to_user(lkb->lkb_ua,
880 test_bit(DLM_PROC_FLAGS_COMPAT, &proc->flags),
881 cb.flags, cb.mode, copy_lvb, buf, count);
882
883 /* removes ref for proc->asts, may cause lkb to be freed */
884 if (!resid)
885 dlm_put_lkb(lkb);
886
887 return rv;
888}
889
890static __poll_t device_poll(struct file *file, poll_table *wait)
891{
892 struct dlm_user_proc *proc = file->private_data;
893
894 poll_wait(file, &proc->wait, wait);
895
896 spin_lock(&proc->asts_spin);
897 if (!list_empty(&proc->asts)) {
898 spin_unlock(&proc->asts_spin);
899 return EPOLLIN | EPOLLRDNORM;
900 }
901 spin_unlock(&proc->asts_spin);
902 return 0;
903}
904
905int dlm_user_daemon_available(void)
906{
907 /* dlm_controld hasn't started (or, has started, but not
908 properly populated configfs) */
909
910 if (!dlm_our_nodeid())
911 return 0;
912
913 /* This is to deal with versions of dlm_controld that don't
914 know about the monitor device. We assume that if the
915 dlm_controld was started (above), but the monitor device
916 was never opened, that it's an old version. dlm_controld
917 should open the monitor device before populating configfs. */
918
919 if (dlm_monitor_unused)
920 return 1;
921
922 return atomic_read(&dlm_monitor_opened) ? 1 : 0;
923}
924
925static int ctl_device_open(struct inode *inode, struct file *file)
926{
927 file->private_data = NULL;
928 return 0;
929}
930
931static int ctl_device_close(struct inode *inode, struct file *file)
932{
933 return 0;
934}
935
936static int monitor_device_open(struct inode *inode, struct file *file)
937{
938 atomic_inc(&dlm_monitor_opened);
939 dlm_monitor_unused = 0;
940 return 0;
941}
942
943static int monitor_device_close(struct inode *inode, struct file *file)
944{
945 if (atomic_dec_and_test(&dlm_monitor_opened))
946 dlm_stop_lockspaces();
947 return 0;
948}
949
950static const struct file_operations device_fops = {
951 .open = device_open,
952 .release = device_close,
953 .read = device_read,
954 .write = device_write,
955 .poll = device_poll,
956 .owner = THIS_MODULE,
957 .llseek = noop_llseek,
958};
959
960static const struct file_operations ctl_device_fops = {
961 .open = ctl_device_open,
962 .release = ctl_device_close,
963 .read = device_read,
964 .write = device_write,
965 .owner = THIS_MODULE,
966 .llseek = noop_llseek,
967};
968
969static struct miscdevice ctl_device = {
970 .name = "dlm-control",
971 .fops = &ctl_device_fops,
972 .minor = MISC_DYNAMIC_MINOR,
973};
974
975static const struct file_operations monitor_device_fops = {
976 .open = monitor_device_open,
977 .release = monitor_device_close,
978 .owner = THIS_MODULE,
979 .llseek = noop_llseek,
980};
981
982static struct miscdevice monitor_device = {
983 .name = "dlm-monitor",
984 .fops = &monitor_device_fops,
985 .minor = MISC_DYNAMIC_MINOR,
986};
987
988int __init dlm_user_init(void)
989{
990 int error;
991
992 atomic_set(&dlm_monitor_opened, 0);
993
994 error = misc_register(&ctl_device);
995 if (error) {
996 log_print("misc_register failed for control device");
997 goto out;
998 }
999
1000 error = misc_register(&monitor_device);
1001 if (error) {
1002 log_print("misc_register failed for monitor device");
1003 misc_deregister(&ctl_device);
1004 }
1005 out:
1006 return error;
1007}
1008
1009void dlm_user_exit(void)
1010{
1011 misc_deregister(&ctl_device);
1012 misc_deregister(&monitor_device);
1013}
1014