1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
6 * Code which implements an OCFS2 specific interface to underlying
9 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
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.
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.
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>
27 #include <linux/kobject.h>
28 #include <linux/sysfs.h>
29 #include <linux/sysctl.h>
33 #include "stackglue.h"
35 #define OCFS2_STACK_PLUGIN_O2CB "o2cb"
36 #define OCFS2_STACK_PLUGIN_USER "user"
37 #define OCFS2_MAX_HB_CTL_PATH 256
39 static struct ocfs2_protocol_version locking_max_version;
40 static DEFINE_SPINLOCK(ocfs2_stack_lock);
41 static LIST_HEAD(ocfs2_stack_list);
42 static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
43 static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
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.
49 static struct ocfs2_stack_plugin *active_stack;
51 inline int ocfs2_is_o2cb_active(void)
53 return !strcmp(active_stack->sp_name, OCFS2_STACK_PLUGIN_O2CB);
55 EXPORT_SYMBOL_GPL(ocfs2_is_o2cb_active);
57 static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
59 struct ocfs2_stack_plugin *p;
61 assert_spin_locked(&ocfs2_stack_lock);
63 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
64 if (!strcmp(p->sp_name, name))
71 static int ocfs2_stack_driver_request(const char *stack_name,
72 const char *plugin_name)
75 struct ocfs2_stack_plugin *p;
77 spin_lock(&ocfs2_stack_lock);
80 * If the stack passed by the filesystem isn't the selected one,
83 if (strcmp(stack_name, cluster_stack_name)) {
90 * If the active stack isn't the one we want, it cannot
91 * be selected right now.
93 if (!strcmp(active_stack->sp_name, plugin_name))
100 p = ocfs2_stack_lookup(plugin_name);
101 if (!p || !try_module_get(p->sp_owner)) {
110 /* If we found it, pin it */
112 active_stack->sp_count++;
114 spin_unlock(&ocfs2_stack_lock);
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.
123 static int ocfs2_stack_driver_get(const char *stack_name)
126 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
129 * Classic stack does not pass in a stack name. This is
130 * compatible with older tools as well.
132 if (!stack_name || !*stack_name)
133 stack_name = OCFS2_STACK_PLUGIN_O2CB;
135 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
137 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
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;
146 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
148 request_module("ocfs2_stack_%s", plugin_name);
149 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
154 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
156 } else if (rc == -EBUSY) {
158 "ocfs2: A different cluster stack is in use\n");
164 static void ocfs2_stack_driver_put(void)
166 spin_lock(&ocfs2_stack_lock);
167 BUG_ON(active_stack == NULL);
168 BUG_ON(active_stack->sp_count == 0);
170 active_stack->sp_count--;
171 if (!active_stack->sp_count) {
172 module_put(active_stack->sp_owner);
175 spin_unlock(&ocfs2_stack_lock);
178 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
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",
191 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
195 spin_unlock(&ocfs2_stack_lock);
199 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
201 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
203 struct ocfs2_stack_plugin *p;
205 spin_lock(&ocfs2_stack_lock);
206 p = ocfs2_stack_lookup(plugin->sp_name);
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",
215 printk(KERN_ERR "Stack \"%s\" is not registered\n",
218 spin_unlock(&ocfs2_stack_lock);
220 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
222 void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
224 struct ocfs2_stack_plugin *p;
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);
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;
236 spin_unlock(&ocfs2_stack_lock);
238 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
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.
247 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
249 struct ocfs2_dlm_lksb *lksb,
252 unsigned int namelen)
254 if (!lksb->lksb_conn)
255 lksb->lksb_conn = conn;
257 BUG_ON(lksb->lksb_conn != conn);
258 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
261 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
263 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
264 struct ocfs2_dlm_lksb *lksb,
267 BUG_ON(lksb->lksb_conn == NULL);
269 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
271 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
273 int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
275 return active_stack->sp_ops->lock_status(lksb);
277 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
279 int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
281 return active_stack->sp_ops->lvb_valid(lksb);
283 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
285 void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
287 return active_stack->sp_ops->lock_lvb(lksb);
289 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
291 void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
293 active_stack->sp_ops->dump_lksb(lksb);
295 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
297 int ocfs2_stack_supports_plocks(void)
299 return active_stack && active_stack->sp_ops->plock;
301 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
304 * ocfs2_plock() can only be safely called if
305 * ocfs2_stack_supports_plocks() returned true
307 int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
308 struct file *file, int cmd, struct file_lock *fl)
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);
315 EXPORT_SYMBOL_GPL(ocfs2_plock);
317 int ocfs2_cluster_connect(const char *stack_name,
318 const char *cluster_name,
319 int cluster_name_len,
322 struct ocfs2_locking_protocol *lproto,
323 void (*recovery_handler)(int node_num,
324 void *recovery_data),
326 struct ocfs2_cluster_connection **conn)
329 struct ocfs2_cluster_connection *new_conn;
331 BUG_ON(group == NULL);
332 BUG_ON(conn == NULL);
333 BUG_ON(recovery_handler == NULL);
335 if (grouplen > GROUP_NAME_MAX) {
340 if (memcmp(&lproto->lp_max_version, &locking_max_version,
341 sizeof(struct ocfs2_protocol_version))) {
346 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
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;
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;
366 /* This will pin the stack driver if successful */
367 rc = ocfs2_stack_driver_get(stack_name);
371 rc = active_stack->sp_ops->connect(new_conn);
373 ocfs2_stack_driver_put();
386 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
388 /* The caller will ensure all nodes have the same cluster stack */
389 int ocfs2_cluster_connect_agnostic(const char *group,
391 struct ocfs2_locking_protocol *lproto,
392 void (*recovery_handler)(int node_num,
393 void *recovery_data),
395 struct ocfs2_cluster_connection **conn)
397 char *stack_name = NULL;
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,
405 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
407 /* If hangup_pending is 0, the stack driver will be dropped */
408 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
413 BUG_ON(conn == NULL);
415 ret = active_stack->sp_ops->disconnect(conn);
417 /* XXX Should we free it anyway? */
421 ocfs2_stack_driver_put();
426 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
429 * Leave the group for this filesystem. This is executed by a userspace
430 * program (stored in ocfs2_hb_ctl_path).
432 static void ocfs2_leave_group(const char *group)
435 char *argv[5], *envp[3];
437 argv[0] = ocfs2_hb_ctl_path;
440 argv[3] = (char *)group;
443 /* minimal command environment taken from cpu_run_sbin_hotplug */
445 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
448 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
451 "ocfs2: Error %d running user helper "
453 ret, argv[0], argv[1], argv[2], argv[3]);
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
464 void ocfs2_cluster_hangup(const char *group, int grouplen)
466 BUG_ON(group == NULL);
467 BUG_ON(group[grouplen] != '\0');
469 ocfs2_leave_group(group);
471 /* cluster_disconnect() was called with hangup_pending==1 */
472 ocfs2_stack_driver_put();
474 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
476 int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
479 return active_stack->sp_ops->this_node(conn, node);
481 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
488 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
489 struct kobj_attribute *attr,
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);
504 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
505 __ATTR(max_locking_protocol, S_IRUGO,
506 ocfs2_max_locking_protocol_show, NULL);
508 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
509 struct kobj_attribute *attr,
512 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
513 struct ocfs2_stack_plugin *p;
515 spin_lock(&ocfs2_stack_lock);
516 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
517 ret = snprintf(buf, remain, "%s\n",
524 /* snprintf() didn't fit */
531 spin_unlock(&ocfs2_stack_lock);
536 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
537 __ATTR(loaded_cluster_plugins, S_IRUGO,
538 ocfs2_loaded_cluster_plugins_show, NULL);
540 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
541 struct kobj_attribute *attr,
546 spin_lock(&ocfs2_stack_lock);
548 ret = snprintf(buf, PAGE_SIZE, "%s\n",
549 active_stack->sp_name);
550 if (ret == PAGE_SIZE)
553 spin_unlock(&ocfs2_stack_lock);
558 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
559 __ATTR(active_cluster_plugin, S_IRUGO,
560 ocfs2_active_cluster_plugin_show, NULL);
562 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
563 struct kobj_attribute *attr,
567 spin_lock(&ocfs2_stack_lock);
568 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
569 spin_unlock(&ocfs2_stack_lock);
574 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
575 struct kobj_attribute *attr,
576 const char *buf, size_t count)
584 if (buf[len - 1] == '\n')
587 if ((len != OCFS2_STACK_LABEL_LEN) ||
588 (strnlen(buf, len) != len))
591 spin_lock(&ocfs2_stack_lock);
593 if (!strncmp(buf, cluster_stack_name, len))
598 memcpy(cluster_stack_name, buf, len);
601 spin_unlock(&ocfs2_stack_lock);
607 static 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);
614 static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
615 struct kobj_attribute *attr,
618 return snprintf(buf, PAGE_SIZE, "1\n");
621 static struct kobj_attribute ocfs2_attr_dlm_recover_support =
622 __ATTR(dlm_recover_callback_support, S_IRUGO,
623 ocfs2_dlm_recover_show, NULL);
625 static 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,
634 static const struct attribute_group ocfs2_attr_group = {
635 .attrs = ocfs2_attrs,
638 struct kset *ocfs2_kset;
639 EXPORT_SYMBOL_GPL(ocfs2_kset);
641 static void ocfs2_sysfs_exit(void)
643 kset_unregister(ocfs2_kset);
646 static int ocfs2_sysfs_init(void)
650 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
654 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
661 kset_unregister(ocfs2_kset);
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.
673 #define FS_OCFS2_NM 1
675 static struct ctl_table ocfs2_nm_table[] = {
677 .procname = "hb_ctl_path",
678 .data = ocfs2_hb_ctl_path,
679 .maxlen = OCFS2_MAX_HB_CTL_PATH,
681 .proc_handler = proc_dostring,
686 static struct ctl_table ocfs2_mod_table[] = {
692 .child = ocfs2_nm_table
697 static struct ctl_table ocfs2_kern_table[] = {
703 .child = ocfs2_mod_table
708 static struct ctl_table ocfs2_root_table[] = {
714 .child = ocfs2_kern_table
719 static struct ctl_table_header *ocfs2_table_header;
726 static int __init ocfs2_stack_glue_init(void)
728 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
730 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
731 if (!ocfs2_table_header) {
733 "ocfs2 stack glue: unable to register sysctl\n");
734 return -ENOMEM; /* or something. */
737 return ocfs2_sysfs_init();
740 static void __exit ocfs2_stack_glue_exit(void)
742 memset(&locking_max_version, 0,
743 sizeof(struct ocfs2_protocol_version));
745 if (ocfs2_table_header)
746 unregister_sysctl_table(ocfs2_table_header);
749 MODULE_AUTHOR("Oracle");
750 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
751 MODULE_LICENSE("GPL");
752 module_init(ocfs2_stack_glue_init);
753 module_exit(ocfs2_stack_glue_exit);