1 // SPDX-License-Identifier: GPL-2.0-only
2 /* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
7 * Code which implements an OCFS2 specific interface to underlying
10 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
13 #include <linux/list.h>
14 #include <linux/spinlock.h>
15 #include <linux/module.h>
16 #include <linux/slab.h>
17 #include <linux/kmod.h>
19 #include <linux/kobject.h>
20 #include <linux/sysfs.h>
21 #include <linux/sysctl.h>
25 #include "stackglue.h"
27 #define OCFS2_STACK_PLUGIN_O2CB "o2cb"
28 #define OCFS2_STACK_PLUGIN_USER "user"
29 #define OCFS2_MAX_HB_CTL_PATH 256
31 static struct ocfs2_protocol_version locking_max_version;
32 static DEFINE_SPINLOCK(ocfs2_stack_lock);
33 static LIST_HEAD(ocfs2_stack_list);
34 static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
35 static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
38 * The stack currently in use. If not null, active_stack->sp_count > 0,
39 * the module is pinned, and the locking protocol cannot be changed.
41 static struct ocfs2_stack_plugin *active_stack;
43 static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
45 struct ocfs2_stack_plugin *p;
47 assert_spin_locked(&ocfs2_stack_lock);
49 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
50 if (!strcmp(p->sp_name, name))
57 static int ocfs2_stack_driver_request(const char *stack_name,
58 const char *plugin_name)
61 struct ocfs2_stack_plugin *p;
63 spin_lock(&ocfs2_stack_lock);
66 * If the stack passed by the filesystem isn't the selected one,
69 if (strcmp(stack_name, cluster_stack_name)) {
76 * If the active stack isn't the one we want, it cannot
77 * be selected right now.
79 if (!strcmp(active_stack->sp_name, plugin_name))
86 p = ocfs2_stack_lookup(plugin_name);
87 if (!p || !try_module_get(p->sp_owner)) {
96 /* If we found it, pin it */
98 active_stack->sp_count++;
100 spin_unlock(&ocfs2_stack_lock);
105 * This function looks up the appropriate stack and makes it active. If
106 * there is no stack, it tries to load it. It will fail if the stack still
107 * cannot be found. It will also fail if a different stack is in use.
109 static int ocfs2_stack_driver_get(const char *stack_name)
112 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
115 * Classic stack does not pass in a stack name. This is
116 * compatible with older tools as well.
118 if (!stack_name || !*stack_name)
119 stack_name = OCFS2_STACK_PLUGIN_O2CB;
121 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
123 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
128 /* Anything that isn't the classic stack is a user stack */
129 if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
130 plugin_name = OCFS2_STACK_PLUGIN_USER;
132 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
134 request_module("ocfs2_stack_%s", plugin_name);
135 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
140 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
142 } else if (rc == -EBUSY) {
144 "ocfs2: A different cluster stack is in use\n");
150 static void ocfs2_stack_driver_put(void)
152 spin_lock(&ocfs2_stack_lock);
153 BUG_ON(active_stack == NULL);
154 BUG_ON(active_stack->sp_count == 0);
156 active_stack->sp_count--;
157 if (!active_stack->sp_count) {
158 module_put(active_stack->sp_owner);
161 spin_unlock(&ocfs2_stack_lock);
164 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
168 spin_lock(&ocfs2_stack_lock);
169 if (!ocfs2_stack_lookup(plugin->sp_name)) {
170 plugin->sp_count = 0;
171 plugin->sp_max_proto = locking_max_version;
172 list_add(&plugin->sp_list, &ocfs2_stack_list);
173 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
177 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
181 spin_unlock(&ocfs2_stack_lock);
185 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
187 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
189 struct ocfs2_stack_plugin *p;
191 spin_lock(&ocfs2_stack_lock);
192 p = ocfs2_stack_lookup(plugin->sp_name);
195 BUG_ON(plugin == active_stack);
196 BUG_ON(plugin->sp_count != 0);
197 list_del_init(&plugin->sp_list);
198 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
201 printk(KERN_ERR "Stack \"%s\" is not registered\n",
204 spin_unlock(&ocfs2_stack_lock);
206 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
208 void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
210 struct ocfs2_stack_plugin *p;
212 spin_lock(&ocfs2_stack_lock);
213 if (memcmp(max_proto, &locking_max_version,
214 sizeof(struct ocfs2_protocol_version))) {
215 BUG_ON(locking_max_version.pv_major != 0);
217 locking_max_version = *max_proto;
218 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
219 p->sp_max_proto = locking_max_version;
222 spin_unlock(&ocfs2_stack_lock);
224 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
228 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
229 * for the ast and bast functions. They will pass the lksb to the ast
230 * and bast. The caller can wrap the lksb with their own structure to
231 * get more information.
233 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
235 struct ocfs2_dlm_lksb *lksb,
238 unsigned int namelen)
240 if (!lksb->lksb_conn)
241 lksb->lksb_conn = conn;
243 BUG_ON(lksb->lksb_conn != conn);
244 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
247 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
249 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
250 struct ocfs2_dlm_lksb *lksb,
253 BUG_ON(lksb->lksb_conn == NULL);
255 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
257 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
259 int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
261 return active_stack->sp_ops->lock_status(lksb);
263 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
265 int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
267 return active_stack->sp_ops->lvb_valid(lksb);
269 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
271 void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
273 return active_stack->sp_ops->lock_lvb(lksb);
275 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
277 void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
279 active_stack->sp_ops->dump_lksb(lksb);
281 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
283 int ocfs2_stack_supports_plocks(void)
285 return active_stack && active_stack->sp_ops->plock;
287 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
290 * ocfs2_plock() can only be safely called if
291 * ocfs2_stack_supports_plocks() returned true
293 int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
294 struct file *file, int cmd, struct file_lock *fl)
296 WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
297 if (active_stack->sp_ops->plock)
298 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
301 EXPORT_SYMBOL_GPL(ocfs2_plock);
303 int ocfs2_cluster_connect(const char *stack_name,
304 const char *cluster_name,
305 int cluster_name_len,
308 struct ocfs2_locking_protocol *lproto,
309 void (*recovery_handler)(int node_num,
310 void *recovery_data),
312 struct ocfs2_cluster_connection **conn)
315 struct ocfs2_cluster_connection *new_conn;
317 BUG_ON(group == NULL);
318 BUG_ON(conn == NULL);
319 BUG_ON(recovery_handler == NULL);
321 if (grouplen > GROUP_NAME_MAX) {
326 if (memcmp(&lproto->lp_max_version, &locking_max_version,
327 sizeof(struct ocfs2_protocol_version))) {
332 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
339 strlcpy(new_conn->cc_name, group, GROUP_NAME_MAX + 1);
340 new_conn->cc_namelen = grouplen;
341 if (cluster_name_len)
342 strlcpy(new_conn->cc_cluster_name, cluster_name,
343 CLUSTER_NAME_MAX + 1);
344 new_conn->cc_cluster_name_len = cluster_name_len;
345 new_conn->cc_recovery_handler = recovery_handler;
346 new_conn->cc_recovery_data = recovery_data;
348 new_conn->cc_proto = lproto;
349 /* Start the new connection at our maximum compatibility level */
350 new_conn->cc_version = lproto->lp_max_version;
352 /* This will pin the stack driver if successful */
353 rc = ocfs2_stack_driver_get(stack_name);
357 rc = active_stack->sp_ops->connect(new_conn);
359 ocfs2_stack_driver_put();
372 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
374 /* The caller will ensure all nodes have the same cluster stack */
375 int ocfs2_cluster_connect_agnostic(const char *group,
377 struct ocfs2_locking_protocol *lproto,
378 void (*recovery_handler)(int node_num,
379 void *recovery_data),
381 struct ocfs2_cluster_connection **conn)
383 char *stack_name = NULL;
385 if (cluster_stack_name[0])
386 stack_name = cluster_stack_name;
387 return ocfs2_cluster_connect(stack_name, NULL, 0, group, grouplen,
388 lproto, recovery_handler, recovery_data,
391 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
393 /* If hangup_pending is 0, the stack driver will be dropped */
394 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
399 BUG_ON(conn == NULL);
401 ret = active_stack->sp_ops->disconnect(conn);
403 /* XXX Should we free it anyway? */
407 ocfs2_stack_driver_put();
412 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
415 * Leave the group for this filesystem. This is executed by a userspace
416 * program (stored in ocfs2_hb_ctl_path).
418 static void ocfs2_leave_group(const char *group)
421 char *argv[5], *envp[3];
423 argv[0] = ocfs2_hb_ctl_path;
426 argv[3] = (char *)group;
429 /* minimal command environment taken from cpu_run_sbin_hotplug */
431 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
434 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
437 "ocfs2: Error %d running user helper "
439 ret, argv[0], argv[1], argv[2], argv[3]);
444 * Hangup is a required post-umount. ocfs2-tools software expects the
445 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
446 * regardless of whether the DLM got started, so we can't do it
447 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
450 void ocfs2_cluster_hangup(const char *group, int grouplen)
452 BUG_ON(group == NULL);
453 BUG_ON(group[grouplen] != '\0');
455 ocfs2_leave_group(group);
457 /* cluster_disconnect() was called with hangup_pending==1 */
458 ocfs2_stack_driver_put();
460 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
462 int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
465 return active_stack->sp_ops->this_node(conn, node);
467 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
474 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
475 struct kobj_attribute *attr,
480 spin_lock(&ocfs2_stack_lock);
481 if (locking_max_version.pv_major)
482 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
483 locking_max_version.pv_major,
484 locking_max_version.pv_minor);
485 spin_unlock(&ocfs2_stack_lock);
490 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
491 __ATTR(max_locking_protocol, S_IRUGO,
492 ocfs2_max_locking_protocol_show, NULL);
494 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
495 struct kobj_attribute *attr,
498 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
499 struct ocfs2_stack_plugin *p;
501 spin_lock(&ocfs2_stack_lock);
502 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
503 ret = snprintf(buf, remain, "%s\n",
510 /* snprintf() didn't fit */
517 spin_unlock(&ocfs2_stack_lock);
522 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
523 __ATTR(loaded_cluster_plugins, S_IRUGO,
524 ocfs2_loaded_cluster_plugins_show, NULL);
526 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
527 struct kobj_attribute *attr,
532 spin_lock(&ocfs2_stack_lock);
534 ret = snprintf(buf, PAGE_SIZE, "%s\n",
535 active_stack->sp_name);
536 if (ret == PAGE_SIZE)
539 spin_unlock(&ocfs2_stack_lock);
544 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
545 __ATTR(active_cluster_plugin, S_IRUGO,
546 ocfs2_active_cluster_plugin_show, NULL);
548 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
549 struct kobj_attribute *attr,
553 spin_lock(&ocfs2_stack_lock);
554 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
555 spin_unlock(&ocfs2_stack_lock);
560 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
561 struct kobj_attribute *attr,
562 const char *buf, size_t count)
570 if (buf[len - 1] == '\n')
573 if ((len != OCFS2_STACK_LABEL_LEN) ||
574 (strnlen(buf, len) != len))
577 spin_lock(&ocfs2_stack_lock);
579 if (!strncmp(buf, cluster_stack_name, len))
584 memcpy(cluster_stack_name, buf, len);
587 spin_unlock(&ocfs2_stack_lock);
593 static struct kobj_attribute ocfs2_attr_cluster_stack =
594 __ATTR(cluster_stack, S_IRUGO | S_IWUSR,
595 ocfs2_cluster_stack_show,
596 ocfs2_cluster_stack_store);
600 static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
601 struct kobj_attribute *attr,
604 return snprintf(buf, PAGE_SIZE, "1\n");
607 static struct kobj_attribute ocfs2_attr_dlm_recover_support =
608 __ATTR(dlm_recover_callback_support, S_IRUGO,
609 ocfs2_dlm_recover_show, NULL);
611 static struct attribute *ocfs2_attrs[] = {
612 &ocfs2_attr_max_locking_protocol.attr,
613 &ocfs2_attr_loaded_cluster_plugins.attr,
614 &ocfs2_attr_active_cluster_plugin.attr,
615 &ocfs2_attr_cluster_stack.attr,
616 &ocfs2_attr_dlm_recover_support.attr,
620 static const struct attribute_group ocfs2_attr_group = {
621 .attrs = ocfs2_attrs,
624 struct kset *ocfs2_kset;
625 EXPORT_SYMBOL_GPL(ocfs2_kset);
627 static void ocfs2_sysfs_exit(void)
629 kset_unregister(ocfs2_kset);
632 static int ocfs2_sysfs_init(void)
636 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
640 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
647 kset_unregister(ocfs2_kset);
654 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
655 * make as much sense in a multiple cluster stack world, but it's safer
656 * and easier to preserve the name.
659 static struct ctl_table ocfs2_nm_table[] = {
661 .procname = "hb_ctl_path",
662 .data = ocfs2_hb_ctl_path,
663 .maxlen = OCFS2_MAX_HB_CTL_PATH,
665 .proc_handler = proc_dostring,
670 static struct ctl_table ocfs2_mod_table[] = {
676 .child = ocfs2_nm_table
681 static struct ctl_table ocfs2_kern_table[] = {
687 .child = ocfs2_mod_table
692 static struct ctl_table ocfs2_root_table[] = {
698 .child = ocfs2_kern_table
703 static struct ctl_table_header *ocfs2_table_header;
710 static int __init ocfs2_stack_glue_init(void)
712 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
714 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
715 if (!ocfs2_table_header) {
717 "ocfs2 stack glue: unable to register sysctl\n");
718 return -ENOMEM; /* or something. */
721 return ocfs2_sysfs_init();
724 static void __exit ocfs2_stack_glue_exit(void)
726 memset(&locking_max_version, 0,
727 sizeof(struct ocfs2_protocol_version));
729 if (ocfs2_table_header)
730 unregister_sysctl_table(ocfs2_table_header);
733 MODULE_AUTHOR("Oracle");
734 MODULE_DESCRIPTION("ocfs2 cluster stack glue layer");
735 MODULE_LICENSE("GPL");
736 module_init(ocfs2_stack_glue_init);
737 module_exit(ocfs2_stack_glue_exit);