Loading...
1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * stackglue.c
5 *
6 * Code which implements an OCFS2 specific interface to underlying
7 * cluster stacks.
8 *
9 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
10 *
11 * This program is free software; you can redistribute it and/or
12 * modify it under the terms of the GNU General Public
13 * License as published by the Free Software Foundation, version 2.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU
18 * General Public License for more details.
19 */
20
21#include <linux/list.h>
22#include <linux/spinlock.h>
23#include <linux/module.h>
24#include <linux/slab.h>
25#include <linux/kmod.h>
26#include <linux/fs.h>
27#include <linux/kobject.h>
28#include <linux/sysfs.h>
29#include <linux/sysctl.h>
30
31#include "ocfs2_fs.h"
32
33#include "stackglue.h"
34
35#define OCFS2_STACK_PLUGIN_O2CB "o2cb"
36#define OCFS2_STACK_PLUGIN_USER "user"
37#define OCFS2_MAX_HB_CTL_PATH 256
38
39static struct ocfs2_protocol_version locking_max_version;
40static DEFINE_SPINLOCK(ocfs2_stack_lock);
41static LIST_HEAD(ocfs2_stack_list);
42static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
43static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
44
45/*
46 * The stack currently in use. If not null, active_stack->sp_count > 0,
47 * the module is pinned, and the locking protocol cannot be changed.
48 */
49static struct ocfs2_stack_plugin *active_stack;
50
51inline int ocfs2_is_o2cb_active(void)
52{
53 return !strcmp(active_stack->sp_name, OCFS2_STACK_PLUGIN_O2CB);
54}
55EXPORT_SYMBOL_GPL(ocfs2_is_o2cb_active);
56
57static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
58{
59 struct ocfs2_stack_plugin *p;
60
61 assert_spin_locked(&ocfs2_stack_lock);
62
63 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
64 if (!strcmp(p->sp_name, name))
65 return p;
66 }
67
68 return NULL;
69}
70
71static int ocfs2_stack_driver_request(const char *stack_name,
72 const char *plugin_name)
73{
74 int rc;
75 struct ocfs2_stack_plugin *p;
76
77 spin_lock(&ocfs2_stack_lock);
78
79 /*
80 * If the stack passed by the filesystem isn't the selected one,
81 * we can't continue.
82 */
83 if (strcmp(stack_name, cluster_stack_name)) {
84 rc = -EBUSY;
85 goto out;
86 }
87
88 if (active_stack) {
89 /*
90 * If the active stack isn't the one we want, it cannot
91 * be selected right now.
92 */
93 if (!strcmp(active_stack->sp_name, plugin_name))
94 rc = 0;
95 else
96 rc = -EBUSY;
97 goto out;
98 }
99
100 p = ocfs2_stack_lookup(plugin_name);
101 if (!p || !try_module_get(p->sp_owner)) {
102 rc = -ENOENT;
103 goto out;
104 }
105
106 active_stack = p;
107 rc = 0;
108
109out:
110 /* If we found it, pin it */
111 if (!rc)
112 active_stack->sp_count++;
113
114 spin_unlock(&ocfs2_stack_lock);
115 return rc;
116}
117
118/*
119 * This function looks up the appropriate stack and makes it active. If
120 * there is no stack, it tries to load it. It will fail if the stack still
121 * cannot be found. It will also fail if a different stack is in use.
122 */
123static int ocfs2_stack_driver_get(const char *stack_name)
124{
125 int rc;
126 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
127
128 /*
129 * Classic stack does not pass in a stack name. This is
130 * compatible with older tools as well.
131 */
132 if (!stack_name || !*stack_name)
133 stack_name = OCFS2_STACK_PLUGIN_O2CB;
134
135 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
136 printk(KERN_ERR
137 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
138 stack_name);
139 return -EINVAL;
140 }
141
142 /* Anything that isn't the classic stack is a user stack */
143 if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
144 plugin_name = OCFS2_STACK_PLUGIN_USER;
145
146 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
147 if (rc == -ENOENT) {
148 request_module("ocfs2_stack_%s", plugin_name);
149 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
150 }
151
152 if (rc == -ENOENT) {
153 printk(KERN_ERR
154 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
155 plugin_name);
156 } else if (rc == -EBUSY) {
157 printk(KERN_ERR
158 "ocfs2: A different cluster stack is in use\n");
159 }
160
161 return rc;
162}
163
164static void ocfs2_stack_driver_put(void)
165{
166 spin_lock(&ocfs2_stack_lock);
167 BUG_ON(active_stack == NULL);
168 BUG_ON(active_stack->sp_count == 0);
169
170 active_stack->sp_count--;
171 if (!active_stack->sp_count) {
172 module_put(active_stack->sp_owner);
173 active_stack = NULL;
174 }
175 spin_unlock(&ocfs2_stack_lock);
176}
177
178int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
179{
180 int rc;
181
182 spin_lock(&ocfs2_stack_lock);
183 if (!ocfs2_stack_lookup(plugin->sp_name)) {
184 plugin->sp_count = 0;
185 plugin->sp_max_proto = locking_max_version;
186 list_add(&plugin->sp_list, &ocfs2_stack_list);
187 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
188 plugin->sp_name);
189 rc = 0;
190 } else {
191 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
192 plugin->sp_name);
193 rc = -EEXIST;
194 }
195 spin_unlock(&ocfs2_stack_lock);
196
197 return rc;
198}
199EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
200
201void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
202{
203 struct ocfs2_stack_plugin *p;
204
205 spin_lock(&ocfs2_stack_lock);
206 p = ocfs2_stack_lookup(plugin->sp_name);
207 if (p) {
208 BUG_ON(p != plugin);
209 BUG_ON(plugin == active_stack);
210 BUG_ON(plugin->sp_count != 0);
211 list_del_init(&plugin->sp_list);
212 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
213 plugin->sp_name);
214 } else {
215 printk(KERN_ERR "Stack \"%s\" is not registered\n",
216 plugin->sp_name);
217 }
218 spin_unlock(&ocfs2_stack_lock);
219}
220EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
221
222void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
223{
224 struct ocfs2_stack_plugin *p;
225
226 spin_lock(&ocfs2_stack_lock);
227 if (memcmp(max_proto, &locking_max_version,
228 sizeof(struct ocfs2_protocol_version))) {
229 BUG_ON(locking_max_version.pv_major != 0);
230
231 locking_max_version = *max_proto;
232 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
233 p->sp_max_proto = locking_max_version;
234 }
235 }
236 spin_unlock(&ocfs2_stack_lock);
237}
238EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
239
240
241/*
242 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
243 * for the ast and bast functions. They will pass the lksb to the ast
244 * and bast. The caller can wrap the lksb with their own structure to
245 * get more information.
246 */
247int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
248 int mode,
249 struct ocfs2_dlm_lksb *lksb,
250 u32 flags,
251 void *name,
252 unsigned int namelen)
253{
254 if (!lksb->lksb_conn)
255 lksb->lksb_conn = conn;
256 else
257 BUG_ON(lksb->lksb_conn != conn);
258 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
259 name, namelen);
260}
261EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
262
263int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
264 struct ocfs2_dlm_lksb *lksb,
265 u32 flags)
266{
267 BUG_ON(lksb->lksb_conn == NULL);
268
269 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
270}
271EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
272
273int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
274{
275 return active_stack->sp_ops->lock_status(lksb);
276}
277EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
278
279int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
280{
281 return active_stack->sp_ops->lvb_valid(lksb);
282}
283EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
284
285void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
286{
287 return active_stack->sp_ops->lock_lvb(lksb);
288}
289EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
290
291void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
292{
293 active_stack->sp_ops->dump_lksb(lksb);
294}
295EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
296
297int ocfs2_stack_supports_plocks(void)
298{
299 return active_stack && active_stack->sp_ops->plock;
300}
301EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
302
303/*
304 * ocfs2_plock() can only be safely called if
305 * ocfs2_stack_supports_plocks() returned true
306 */
307int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
308 struct file *file, int cmd, struct file_lock *fl)
309{
310 WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
311 if (active_stack->sp_ops->plock)
312 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
313 return -EOPNOTSUPP;
314}
315EXPORT_SYMBOL_GPL(ocfs2_plock);
316
317int ocfs2_cluster_connect(const char *stack_name,
318 const char *cluster_name,
319 int cluster_name_len,
320 const char *group,
321 int grouplen,
322 struct ocfs2_locking_protocol *lproto,
323 void (*recovery_handler)(int node_num,
324 void *recovery_data),
325 void *recovery_data,
326 struct ocfs2_cluster_connection **conn)
327{
328 int rc = 0;
329 struct ocfs2_cluster_connection *new_conn;
330
331 BUG_ON(group == NULL);
332 BUG_ON(conn == NULL);
333 BUG_ON(recovery_handler == NULL);
334
335 if (grouplen > GROUP_NAME_MAX) {
336 rc = -EINVAL;
337 goto out;
338 }
339
340 if (memcmp(&lproto->lp_max_version, &locking_max_version,
341 sizeof(struct ocfs2_protocol_version))) {
342 rc = -EINVAL;
343 goto out;
344 }
345
346 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
347 GFP_KERNEL);
348 if (!new_conn) {
349 rc = -ENOMEM;
350 goto out;
351 }
352
353 strlcpy(new_conn->cc_name, group, GROUP_NAME_MAX + 1);
354 new_conn->cc_namelen = grouplen;
355 if (cluster_name_len)
356 strlcpy(new_conn->cc_cluster_name, cluster_name,
357 CLUSTER_NAME_MAX + 1);
358 new_conn->cc_cluster_name_len = cluster_name_len;
359 new_conn->cc_recovery_handler = recovery_handler;
360 new_conn->cc_recovery_data = recovery_data;
361
362 new_conn->cc_proto = lproto;
363 /* Start the new connection at our maximum compatibility level */
364 new_conn->cc_version = lproto->lp_max_version;
365
366 /* This will pin the stack driver if successful */
367 rc = ocfs2_stack_driver_get(stack_name);
368 if (rc)
369 goto out_free;
370
371 rc = active_stack->sp_ops->connect(new_conn);
372 if (rc) {
373 ocfs2_stack_driver_put();
374 goto out_free;
375 }
376
377 *conn = new_conn;
378
379out_free:
380 if (rc)
381 kfree(new_conn);
382
383out:
384 return rc;
385}
386EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
387
388/* The caller will ensure all nodes have the same cluster stack */
389int ocfs2_cluster_connect_agnostic(const char *group,
390 int grouplen,
391 struct ocfs2_locking_protocol *lproto,
392 void (*recovery_handler)(int node_num,
393 void *recovery_data),
394 void *recovery_data,
395 struct ocfs2_cluster_connection **conn)
396{
397 char *stack_name = NULL;
398
399 if (cluster_stack_name[0])
400 stack_name = cluster_stack_name;
401 return ocfs2_cluster_connect(stack_name, NULL, 0, group, grouplen,
402 lproto, recovery_handler, recovery_data,
403 conn);
404}
405EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
406
407/* If hangup_pending is 0, the stack driver will be dropped */
408int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
409 int hangup_pending)
410{
411 int ret;
412
413 BUG_ON(conn == NULL);
414
415 ret = active_stack->sp_ops->disconnect(conn);
416
417 /* XXX Should we free it anyway? */
418 if (!ret) {
419 kfree(conn);
420 if (!hangup_pending)
421 ocfs2_stack_driver_put();
422 }
423
424 return ret;
425}
426EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
427
428/*
429 * Leave the group for this filesystem. This is executed by a userspace
430 * program (stored in ocfs2_hb_ctl_path).
431 */
432static void ocfs2_leave_group(const char *group)
433{
434 int ret;
435 char *argv[5], *envp[3];
436
437 argv[0] = ocfs2_hb_ctl_path;
438 argv[1] = "-K";
439 argv[2] = "-u";
440 argv[3] = (char *)group;
441 argv[4] = NULL;
442
443 /* minimal command environment taken from cpu_run_sbin_hotplug */
444 envp[0] = "HOME=/";
445 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
446 envp[2] = NULL;
447
448 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
449 if (ret < 0) {
450 printk(KERN_ERR
451 "ocfs2: Error %d running user helper "
452 "\"%s %s %s %s\"\n",
453 ret, argv[0], argv[1], argv[2], argv[3]);
454 }
455}
456
457/*
458 * Hangup is a required post-umount. ocfs2-tools software expects the
459 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
460 * regardless of whether the DLM got started, so we can't do it
461 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
462 * the actual work.
463 */
464void ocfs2_cluster_hangup(const char *group, int grouplen)
465{
466 BUG_ON(group == NULL);
467 BUG_ON(group[grouplen] != '\0');
468
469 ocfs2_leave_group(group);
470
471 /* cluster_disconnect() was called with hangup_pending==1 */
472 ocfs2_stack_driver_put();
473}
474EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
475
476int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
477 unsigned int *node)
478{
479 return active_stack->sp_ops->this_node(conn, node);
480}
481EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
482
483
484/*
485 * Sysfs bits
486 */
487
488static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
489 struct kobj_attribute *attr,
490 char *buf)
491{
492 ssize_t ret = 0;
493
494 spin_lock(&ocfs2_stack_lock);
495 if (locking_max_version.pv_major)
496 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
497 locking_max_version.pv_major,
498 locking_max_version.pv_minor);
499 spin_unlock(&ocfs2_stack_lock);
500
501 return ret;
502}
503
504static struct kobj_attribute ocfs2_attr_max_locking_protocol =
505 __ATTR(max_locking_protocol, S_IRUGO,
506 ocfs2_max_locking_protocol_show, NULL);
507
508static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
509 struct kobj_attribute *attr,
510 char *buf)
511{
512 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
513 struct ocfs2_stack_plugin *p;
514
515 spin_lock(&ocfs2_stack_lock);
516 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
517 ret = snprintf(buf, remain, "%s\n",
518 p->sp_name);
519 if (ret < 0) {
520 total = ret;
521 break;
522 }
523 if (ret == remain) {
524 /* snprintf() didn't fit */
525 total = -E2BIG;
526 break;
527 }
528 total += ret;
529 remain -= ret;
530 }
531 spin_unlock(&ocfs2_stack_lock);
532
533 return total;
534}
535
536static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
537 __ATTR(loaded_cluster_plugins, S_IRUGO,
538 ocfs2_loaded_cluster_plugins_show, NULL);
539
540static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
541 struct kobj_attribute *attr,
542 char *buf)
543{
544 ssize_t ret = 0;
545
546 spin_lock(&ocfs2_stack_lock);
547 if (active_stack) {
548 ret = snprintf(buf, PAGE_SIZE, "%s\n",
549 active_stack->sp_name);
550 if (ret == PAGE_SIZE)
551 ret = -E2BIG;
552 }
553 spin_unlock(&ocfs2_stack_lock);
554
555 return ret;
556}
557
558static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
559 __ATTR(active_cluster_plugin, S_IRUGO,
560 ocfs2_active_cluster_plugin_show, NULL);
561
562static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
563 struct kobj_attribute *attr,
564 char *buf)
565{
566 ssize_t ret;
567 spin_lock(&ocfs2_stack_lock);
568 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
569 spin_unlock(&ocfs2_stack_lock);
570
571 return ret;
572}
573
574static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
575 struct kobj_attribute *attr,
576 const char *buf, size_t count)
577{
578 size_t len = count;
579 ssize_t ret;
580
581 if (len == 0)
582 return len;
583
584 if (buf[len - 1] == '\n')
585 len--;
586
587 if ((len != OCFS2_STACK_LABEL_LEN) ||
588 (strnlen(buf, len) != len))
589 return -EINVAL;
590
591 spin_lock(&ocfs2_stack_lock);
592 if (active_stack) {
593 if (!strncmp(buf, cluster_stack_name, len))
594 ret = count;
595 else
596 ret = -EBUSY;
597 } else {
598 memcpy(cluster_stack_name, buf, len);
599 ret = count;
600 }
601 spin_unlock(&ocfs2_stack_lock);
602
603 return ret;
604}
605
606
607static struct kobj_attribute ocfs2_attr_cluster_stack =
608 __ATTR(cluster_stack, S_IRUGO | S_IWUSR,
609 ocfs2_cluster_stack_show,
610 ocfs2_cluster_stack_store);
611
612
613
614static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
615 struct kobj_attribute *attr,
616 char *buf)
617{
618 return snprintf(buf, PAGE_SIZE, "1\n");
619}
620
621static struct kobj_attribute ocfs2_attr_dlm_recover_support =
622 __ATTR(dlm_recover_callback_support, S_IRUGO,
623 ocfs2_dlm_recover_show, NULL);
624
625static struct attribute *ocfs2_attrs[] = {
626 &ocfs2_attr_max_locking_protocol.attr,
627 &ocfs2_attr_loaded_cluster_plugins.attr,
628 &ocfs2_attr_active_cluster_plugin.attr,
629 &ocfs2_attr_cluster_stack.attr,
630 &ocfs2_attr_dlm_recover_support.attr,
631 NULL,
632};
633
634static const struct attribute_group ocfs2_attr_group = {
635 .attrs = ocfs2_attrs,
636};
637
638struct kset *ocfs2_kset;
639EXPORT_SYMBOL_GPL(ocfs2_kset);
640
641static void ocfs2_sysfs_exit(void)
642{
643 kset_unregister(ocfs2_kset);
644}
645
646static int ocfs2_sysfs_init(void)
647{
648 int ret;
649
650 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
651 if (!ocfs2_kset)
652 return -ENOMEM;
653
654 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
655 if (ret)
656 goto error;
657
658 return 0;
659
660error:
661 kset_unregister(ocfs2_kset);
662 return ret;
663}
664
665/*
666 * Sysctl bits
667 *
668 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
669 * make as much sense in a multiple cluster stack world, but it's safer
670 * and easier to preserve the name.
671 */
672
673#define FS_OCFS2_NM 1
674
675static struct ctl_table ocfs2_nm_table[] = {
676 {
677 .procname = "hb_ctl_path",
678 .data = ocfs2_hb_ctl_path,
679 .maxlen = OCFS2_MAX_HB_CTL_PATH,
680 .mode = 0644,
681 .proc_handler = proc_dostring,
682 },
683 { }
684};
685
686static struct ctl_table ocfs2_mod_table[] = {
687 {
688 .procname = "nm",
689 .data = NULL,
690 .maxlen = 0,
691 .mode = 0555,
692 .child = ocfs2_nm_table
693 },
694 { }
695};
696
697static struct ctl_table ocfs2_kern_table[] = {
698 {
699 .procname = "ocfs2",
700 .data = NULL,
701 .maxlen = 0,
702 .mode = 0555,
703 .child = ocfs2_mod_table
704 },
705 { }
706};
707
708static struct ctl_table ocfs2_root_table[] = {
709 {
710 .procname = "fs",
711 .data = NULL,
712 .maxlen = 0,
713 .mode = 0555,
714 .child = ocfs2_kern_table
715 },
716 { }
717};
718
719static struct ctl_table_header *ocfs2_table_header;
720
721
722/*
723 * Initialization
724 */
725
726static int __init ocfs2_stack_glue_init(void)
727{
728 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
729
730 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
731 if (!ocfs2_table_header) {
732 printk(KERN_ERR
733 "ocfs2 stack glue: unable to register sysctl\n");
734 return -ENOMEM; /* or something. */
735 }
736
737 return ocfs2_sysfs_init();
738}
739
740static void __exit ocfs2_stack_glue_exit(void)
741{
742 memset(&locking_max_version, 0,
743 sizeof(struct ocfs2_protocol_version));
744 ocfs2_sysfs_exit();
745 if (ocfs2_table_header)
746 unregister_sysctl_table(ocfs2_table_header);
747}
748
749MODULE_AUTHOR("Oracle");
750MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
751MODULE_LICENSE("GPL");
752module_init(ocfs2_stack_glue_init);
753module_exit(ocfs2_stack_glue_exit);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * stackglue.c
4 *
5 * Code which implements an OCFS2 specific interface to underlying
6 * cluster stacks.
7 *
8 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
9 */
10
11#include <linux/list.h>
12#include <linux/spinlock.h>
13#include <linux/module.h>
14#include <linux/slab.h>
15#include <linux/kmod.h>
16#include <linux/fs.h>
17#include <linux/kobject.h>
18#include <linux/sysfs.h>
19#include <linux/sysctl.h>
20
21#include "ocfs2_fs.h"
22
23#include "stackglue.h"
24
25#define OCFS2_STACK_PLUGIN_O2CB "o2cb"
26#define OCFS2_STACK_PLUGIN_USER "user"
27#define OCFS2_MAX_HB_CTL_PATH 256
28
29static struct ocfs2_protocol_version locking_max_version;
30static DEFINE_SPINLOCK(ocfs2_stack_lock);
31static LIST_HEAD(ocfs2_stack_list);
32static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
33static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
34
35/*
36 * The stack currently in use. If not null, active_stack->sp_count > 0,
37 * the module is pinned, and the locking protocol cannot be changed.
38 */
39static struct ocfs2_stack_plugin *active_stack;
40
41static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
42{
43 struct ocfs2_stack_plugin *p;
44
45 assert_spin_locked(&ocfs2_stack_lock);
46
47 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
48 if (!strcmp(p->sp_name, name))
49 return p;
50 }
51
52 return NULL;
53}
54
55static int ocfs2_stack_driver_request(const char *stack_name,
56 const char *plugin_name)
57{
58 int rc;
59 struct ocfs2_stack_plugin *p;
60
61 spin_lock(&ocfs2_stack_lock);
62
63 /*
64 * If the stack passed by the filesystem isn't the selected one,
65 * we can't continue.
66 */
67 if (strcmp(stack_name, cluster_stack_name)) {
68 rc = -EBUSY;
69 goto out;
70 }
71
72 if (active_stack) {
73 /*
74 * If the active stack isn't the one we want, it cannot
75 * be selected right now.
76 */
77 if (!strcmp(active_stack->sp_name, plugin_name))
78 rc = 0;
79 else
80 rc = -EBUSY;
81 goto out;
82 }
83
84 p = ocfs2_stack_lookup(plugin_name);
85 if (!p || !try_module_get(p->sp_owner)) {
86 rc = -ENOENT;
87 goto out;
88 }
89
90 active_stack = p;
91 rc = 0;
92
93out:
94 /* If we found it, pin it */
95 if (!rc)
96 active_stack->sp_count++;
97
98 spin_unlock(&ocfs2_stack_lock);
99 return rc;
100}
101
102/*
103 * This function looks up the appropriate stack and makes it active. If
104 * there is no stack, it tries to load it. It will fail if the stack still
105 * cannot be found. It will also fail if a different stack is in use.
106 */
107static int ocfs2_stack_driver_get(const char *stack_name)
108{
109 int rc;
110 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
111
112 /*
113 * Classic stack does not pass in a stack name. This is
114 * compatible with older tools as well.
115 */
116 if (!stack_name || !*stack_name)
117 stack_name = OCFS2_STACK_PLUGIN_O2CB;
118
119 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
120 printk(KERN_ERR
121 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
122 stack_name);
123 return -EINVAL;
124 }
125
126 /* Anything that isn't the classic stack is a user stack */
127 if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
128 plugin_name = OCFS2_STACK_PLUGIN_USER;
129
130 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
131 if (rc == -ENOENT) {
132 request_module("ocfs2_stack_%s", plugin_name);
133 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
134 }
135
136 if (rc == -ENOENT) {
137 printk(KERN_ERR
138 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
139 plugin_name);
140 } else if (rc == -EBUSY) {
141 printk(KERN_ERR
142 "ocfs2: A different cluster stack is in use\n");
143 }
144
145 return rc;
146}
147
148static void ocfs2_stack_driver_put(void)
149{
150 spin_lock(&ocfs2_stack_lock);
151 BUG_ON(active_stack == NULL);
152 BUG_ON(active_stack->sp_count == 0);
153
154 active_stack->sp_count--;
155 if (!active_stack->sp_count) {
156 module_put(active_stack->sp_owner);
157 active_stack = NULL;
158 }
159 spin_unlock(&ocfs2_stack_lock);
160}
161
162int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
163{
164 int rc;
165
166 spin_lock(&ocfs2_stack_lock);
167 if (!ocfs2_stack_lookup(plugin->sp_name)) {
168 plugin->sp_count = 0;
169 plugin->sp_max_proto = locking_max_version;
170 list_add(&plugin->sp_list, &ocfs2_stack_list);
171 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
172 plugin->sp_name);
173 rc = 0;
174 } else {
175 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
176 plugin->sp_name);
177 rc = -EEXIST;
178 }
179 spin_unlock(&ocfs2_stack_lock);
180
181 return rc;
182}
183EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
184
185void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
186{
187 struct ocfs2_stack_plugin *p;
188
189 spin_lock(&ocfs2_stack_lock);
190 p = ocfs2_stack_lookup(plugin->sp_name);
191 if (p) {
192 BUG_ON(p != plugin);
193 BUG_ON(plugin == active_stack);
194 BUG_ON(plugin->sp_count != 0);
195 list_del_init(&plugin->sp_list);
196 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
197 plugin->sp_name);
198 } else {
199 printk(KERN_ERR "Stack \"%s\" is not registered\n",
200 plugin->sp_name);
201 }
202 spin_unlock(&ocfs2_stack_lock);
203}
204EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
205
206void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
207{
208 struct ocfs2_stack_plugin *p;
209
210 spin_lock(&ocfs2_stack_lock);
211 if (memcmp(max_proto, &locking_max_version,
212 sizeof(struct ocfs2_protocol_version))) {
213 BUG_ON(locking_max_version.pv_major != 0);
214
215 locking_max_version = *max_proto;
216 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
217 p->sp_max_proto = locking_max_version;
218 }
219 }
220 spin_unlock(&ocfs2_stack_lock);
221}
222EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
223
224
225/*
226 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
227 * for the ast and bast functions. They will pass the lksb to the ast
228 * and bast. The caller can wrap the lksb with their own structure to
229 * get more information.
230 */
231int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
232 int mode,
233 struct ocfs2_dlm_lksb *lksb,
234 u32 flags,
235 void *name,
236 unsigned int namelen)
237{
238 if (!lksb->lksb_conn)
239 lksb->lksb_conn = conn;
240 else
241 BUG_ON(lksb->lksb_conn != conn);
242 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
243 name, namelen);
244}
245EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
246
247int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
248 struct ocfs2_dlm_lksb *lksb,
249 u32 flags)
250{
251 BUG_ON(lksb->lksb_conn == NULL);
252
253 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
254}
255EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
256
257int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
258{
259 return active_stack->sp_ops->lock_status(lksb);
260}
261EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
262
263int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
264{
265 return active_stack->sp_ops->lvb_valid(lksb);
266}
267EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
268
269void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
270{
271 return active_stack->sp_ops->lock_lvb(lksb);
272}
273EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
274
275void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
276{
277 active_stack->sp_ops->dump_lksb(lksb);
278}
279EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
280
281int ocfs2_stack_supports_plocks(void)
282{
283 return active_stack && active_stack->sp_ops->plock;
284}
285EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
286
287/*
288 * ocfs2_plock() can only be safely called if
289 * ocfs2_stack_supports_plocks() returned true
290 */
291int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
292 struct file *file, int cmd, struct file_lock *fl)
293{
294 WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
295 if (active_stack->sp_ops->plock)
296 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
297 return -EOPNOTSUPP;
298}
299EXPORT_SYMBOL_GPL(ocfs2_plock);
300
301int ocfs2_cluster_connect(const char *stack_name,
302 const char *cluster_name,
303 int cluster_name_len,
304 const char *group,
305 int grouplen,
306 struct ocfs2_locking_protocol *lproto,
307 void (*recovery_handler)(int node_num,
308 void *recovery_data),
309 void *recovery_data,
310 struct ocfs2_cluster_connection **conn)
311{
312 int rc = 0;
313 struct ocfs2_cluster_connection *new_conn;
314
315 BUG_ON(group == NULL);
316 BUG_ON(conn == NULL);
317 BUG_ON(recovery_handler == NULL);
318
319 if (grouplen > GROUP_NAME_MAX) {
320 rc = -EINVAL;
321 goto out;
322 }
323
324 if (memcmp(&lproto->lp_max_version, &locking_max_version,
325 sizeof(struct ocfs2_protocol_version))) {
326 rc = -EINVAL;
327 goto out;
328 }
329
330 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
331 GFP_KERNEL);
332 if (!new_conn) {
333 rc = -ENOMEM;
334 goto out;
335 }
336
337 strscpy(new_conn->cc_name, group, GROUP_NAME_MAX + 1);
338 new_conn->cc_namelen = grouplen;
339 if (cluster_name_len)
340 strscpy(new_conn->cc_cluster_name, cluster_name,
341 CLUSTER_NAME_MAX + 1);
342 new_conn->cc_cluster_name_len = cluster_name_len;
343 new_conn->cc_recovery_handler = recovery_handler;
344 new_conn->cc_recovery_data = recovery_data;
345
346 new_conn->cc_proto = lproto;
347 /* Start the new connection at our maximum compatibility level */
348 new_conn->cc_version = lproto->lp_max_version;
349
350 /* This will pin the stack driver if successful */
351 rc = ocfs2_stack_driver_get(stack_name);
352 if (rc)
353 goto out_free;
354
355 rc = active_stack->sp_ops->connect(new_conn);
356 if (rc) {
357 ocfs2_stack_driver_put();
358 goto out_free;
359 }
360
361 *conn = new_conn;
362
363out_free:
364 if (rc)
365 kfree(new_conn);
366
367out:
368 return rc;
369}
370EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
371
372/* The caller will ensure all nodes have the same cluster stack */
373int ocfs2_cluster_connect_agnostic(const char *group,
374 int grouplen,
375 struct ocfs2_locking_protocol *lproto,
376 void (*recovery_handler)(int node_num,
377 void *recovery_data),
378 void *recovery_data,
379 struct ocfs2_cluster_connection **conn)
380{
381 char *stack_name = NULL;
382
383 if (cluster_stack_name[0])
384 stack_name = cluster_stack_name;
385 return ocfs2_cluster_connect(stack_name, NULL, 0, group, grouplen,
386 lproto, recovery_handler, recovery_data,
387 conn);
388}
389EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
390
391/* If hangup_pending is 0, the stack driver will be dropped */
392int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
393 int hangup_pending)
394{
395 int ret;
396
397 BUG_ON(conn == NULL);
398
399 ret = active_stack->sp_ops->disconnect(conn);
400
401 /* XXX Should we free it anyway? */
402 if (!ret) {
403 kfree(conn);
404 if (!hangup_pending)
405 ocfs2_stack_driver_put();
406 }
407
408 return ret;
409}
410EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
411
412/*
413 * Leave the group for this filesystem. This is executed by a userspace
414 * program (stored in ocfs2_hb_ctl_path).
415 */
416static void ocfs2_leave_group(const char *group)
417{
418 int ret;
419 char *argv[5], *envp[3];
420
421 argv[0] = ocfs2_hb_ctl_path;
422 argv[1] = "-K";
423 argv[2] = "-u";
424 argv[3] = (char *)group;
425 argv[4] = NULL;
426
427 /* minimal command environment taken from cpu_run_sbin_hotplug */
428 envp[0] = "HOME=/";
429 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
430 envp[2] = NULL;
431
432 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
433 if (ret < 0) {
434 printk(KERN_ERR
435 "ocfs2: Error %d running user helper "
436 "\"%s %s %s %s\"\n",
437 ret, argv[0], argv[1], argv[2], argv[3]);
438 }
439}
440
441/*
442 * Hangup is a required post-umount. ocfs2-tools software expects the
443 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
444 * regardless of whether the DLM got started, so we can't do it
445 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
446 * the actual work.
447 */
448void ocfs2_cluster_hangup(const char *group, int grouplen)
449{
450 BUG_ON(group == NULL);
451 BUG_ON(group[grouplen] != '\0');
452
453 ocfs2_leave_group(group);
454
455 /* cluster_disconnect() was called with hangup_pending==1 */
456 ocfs2_stack_driver_put();
457}
458EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
459
460int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
461 unsigned int *node)
462{
463 return active_stack->sp_ops->this_node(conn, node);
464}
465EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
466
467
468/*
469 * Sysfs bits
470 */
471
472static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
473 struct kobj_attribute *attr,
474 char *buf)
475{
476 ssize_t ret = 0;
477
478 spin_lock(&ocfs2_stack_lock);
479 if (locking_max_version.pv_major)
480 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
481 locking_max_version.pv_major,
482 locking_max_version.pv_minor);
483 spin_unlock(&ocfs2_stack_lock);
484
485 return ret;
486}
487
488static struct kobj_attribute ocfs2_attr_max_locking_protocol =
489 __ATTR(max_locking_protocol, S_IRUGO,
490 ocfs2_max_locking_protocol_show, NULL);
491
492static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
493 struct kobj_attribute *attr,
494 char *buf)
495{
496 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
497 struct ocfs2_stack_plugin *p;
498
499 spin_lock(&ocfs2_stack_lock);
500 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
501 ret = snprintf(buf, remain, "%s\n",
502 p->sp_name);
503 if (ret >= remain) {
504 /* snprintf() didn't fit */
505 total = -E2BIG;
506 break;
507 }
508 total += ret;
509 remain -= ret;
510 }
511 spin_unlock(&ocfs2_stack_lock);
512
513 return total;
514}
515
516static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
517 __ATTR(loaded_cluster_plugins, S_IRUGO,
518 ocfs2_loaded_cluster_plugins_show, NULL);
519
520static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
521 struct kobj_attribute *attr,
522 char *buf)
523{
524 ssize_t ret = 0;
525
526 spin_lock(&ocfs2_stack_lock);
527 if (active_stack) {
528 ret = snprintf(buf, PAGE_SIZE, "%s\n",
529 active_stack->sp_name);
530 if (ret >= PAGE_SIZE)
531 ret = -E2BIG;
532 }
533 spin_unlock(&ocfs2_stack_lock);
534
535 return ret;
536}
537
538static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
539 __ATTR(active_cluster_plugin, S_IRUGO,
540 ocfs2_active_cluster_plugin_show, NULL);
541
542static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
543 struct kobj_attribute *attr,
544 char *buf)
545{
546 ssize_t ret;
547 spin_lock(&ocfs2_stack_lock);
548 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
549 spin_unlock(&ocfs2_stack_lock);
550
551 return ret;
552}
553
554static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
555 struct kobj_attribute *attr,
556 const char *buf, size_t count)
557{
558 size_t len = count;
559 ssize_t ret;
560
561 if (len == 0)
562 return len;
563
564 if (buf[len - 1] == '\n')
565 len--;
566
567 if ((len != OCFS2_STACK_LABEL_LEN) ||
568 (strnlen(buf, len) != len))
569 return -EINVAL;
570
571 spin_lock(&ocfs2_stack_lock);
572 if (active_stack) {
573 if (!strncmp(buf, cluster_stack_name, len))
574 ret = count;
575 else
576 ret = -EBUSY;
577 } else {
578 memcpy(cluster_stack_name, buf, len);
579 ret = count;
580 }
581 spin_unlock(&ocfs2_stack_lock);
582
583 return ret;
584}
585
586
587static struct kobj_attribute ocfs2_attr_cluster_stack =
588 __ATTR(cluster_stack, S_IRUGO | S_IWUSR,
589 ocfs2_cluster_stack_show,
590 ocfs2_cluster_stack_store);
591
592
593
594static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
595 struct kobj_attribute *attr,
596 char *buf)
597{
598 return snprintf(buf, PAGE_SIZE, "1\n");
599}
600
601static struct kobj_attribute ocfs2_attr_dlm_recover_support =
602 __ATTR(dlm_recover_callback_support, S_IRUGO,
603 ocfs2_dlm_recover_show, NULL);
604
605static struct attribute *ocfs2_attrs[] = {
606 &ocfs2_attr_max_locking_protocol.attr,
607 &ocfs2_attr_loaded_cluster_plugins.attr,
608 &ocfs2_attr_active_cluster_plugin.attr,
609 &ocfs2_attr_cluster_stack.attr,
610 &ocfs2_attr_dlm_recover_support.attr,
611 NULL,
612};
613
614static const struct attribute_group ocfs2_attr_group = {
615 .attrs = ocfs2_attrs,
616};
617
618struct kset *ocfs2_kset;
619EXPORT_SYMBOL_GPL(ocfs2_kset);
620
621static void ocfs2_sysfs_exit(void)
622{
623 kset_unregister(ocfs2_kset);
624}
625
626static int ocfs2_sysfs_init(void)
627{
628 int ret;
629
630 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
631 if (!ocfs2_kset)
632 return -ENOMEM;
633
634 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
635 if (ret)
636 goto error;
637
638 return 0;
639
640error:
641 kset_unregister(ocfs2_kset);
642 return ret;
643}
644
645/*
646 * Sysctl bits
647 *
648 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
649 * make as much sense in a multiple cluster stack world, but it's safer
650 * and easier to preserve the name.
651 */
652
653static struct ctl_table ocfs2_nm_table[] = {
654 {
655 .procname = "hb_ctl_path",
656 .data = ocfs2_hb_ctl_path,
657 .maxlen = OCFS2_MAX_HB_CTL_PATH,
658 .mode = 0644,
659 .proc_handler = proc_dostring,
660 },
661 { }
662};
663
664static struct ctl_table_header *ocfs2_table_header;
665
666/*
667 * Initialization
668 */
669
670static int __init ocfs2_stack_glue_init(void)
671{
672 int ret;
673
674 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
675
676 ocfs2_table_header = register_sysctl("fs/ocfs2/nm", ocfs2_nm_table);
677 if (!ocfs2_table_header) {
678 printk(KERN_ERR
679 "ocfs2 stack glue: unable to register sysctl\n");
680 return -ENOMEM; /* or something. */
681 }
682
683 ret = ocfs2_sysfs_init();
684 if (ret)
685 unregister_sysctl_table(ocfs2_table_header);
686
687 return ret;
688}
689
690static void __exit ocfs2_stack_glue_exit(void)
691{
692 memset(&locking_max_version, 0,
693 sizeof(struct ocfs2_protocol_version));
694 ocfs2_sysfs_exit();
695 if (ocfs2_table_header)
696 unregister_sysctl_table(ocfs2_table_header);
697}
698
699MODULE_AUTHOR("Oracle");
700MODULE_DESCRIPTION("ocfs2 cluster stack glue layer");
701MODULE_LICENSE("GPL");
702module_init(ocfs2_stack_glue_init);
703module_exit(ocfs2_stack_glue_exit);