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 static struct ocfs2_stack_plugin
*ocfs2_stack_lookup(const char *name
)
53 struct ocfs2_stack_plugin
*p
;
55 assert_spin_locked(&ocfs2_stack_lock
);
57 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
58 if (!strcmp(p
->sp_name
, name
))
65 static int ocfs2_stack_driver_request(const char *stack_name
,
66 const char *plugin_name
)
69 struct ocfs2_stack_plugin
*p
;
71 spin_lock(&ocfs2_stack_lock
);
74 * If the stack passed by the filesystem isn't the selected one,
77 if (strcmp(stack_name
, cluster_stack_name
)) {
84 * If the active stack isn't the one we want, it cannot
85 * be selected right now.
87 if (!strcmp(active_stack
->sp_name
, plugin_name
))
94 p
= ocfs2_stack_lookup(plugin_name
);
95 if (!p
|| !try_module_get(p
->sp_owner
)) {
104 /* If we found it, pin it */
106 active_stack
->sp_count
++;
108 spin_unlock(&ocfs2_stack_lock
);
113 * This function looks up the appropriate stack and makes it active. If
114 * there is no stack, it tries to load it. It will fail if the stack still
115 * cannot be found. It will also fail if a different stack is in use.
117 static int ocfs2_stack_driver_get(const char *stack_name
)
120 char *plugin_name
= OCFS2_STACK_PLUGIN_O2CB
;
123 * Classic stack does not pass in a stack name. This is
124 * compatible with older tools as well.
126 if (!stack_name
|| !*stack_name
)
127 stack_name
= OCFS2_STACK_PLUGIN_O2CB
;
129 if (strlen(stack_name
) != OCFS2_STACK_LABEL_LEN
) {
131 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
136 /* Anything that isn't the classic stack is a user stack */
137 if (strcmp(stack_name
, OCFS2_STACK_PLUGIN_O2CB
))
138 plugin_name
= OCFS2_STACK_PLUGIN_USER
;
140 rc
= ocfs2_stack_driver_request(stack_name
, plugin_name
);
142 request_module("ocfs2_stack_%s", plugin_name
);
143 rc
= ocfs2_stack_driver_request(stack_name
, plugin_name
);
148 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
150 } else if (rc
== -EBUSY
) {
152 "ocfs2: A different cluster stack is in use\n");
158 static void ocfs2_stack_driver_put(void)
160 spin_lock(&ocfs2_stack_lock
);
161 BUG_ON(active_stack
== NULL
);
162 BUG_ON(active_stack
->sp_count
== 0);
164 active_stack
->sp_count
--;
165 if (!active_stack
->sp_count
) {
166 module_put(active_stack
->sp_owner
);
169 spin_unlock(&ocfs2_stack_lock
);
172 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin
*plugin
)
176 spin_lock(&ocfs2_stack_lock
);
177 if (!ocfs2_stack_lookup(plugin
->sp_name
)) {
178 plugin
->sp_count
= 0;
179 plugin
->sp_max_proto
= locking_max_version
;
180 list_add(&plugin
->sp_list
, &ocfs2_stack_list
);
181 printk(KERN_INFO
"ocfs2: Registered cluster interface %s\n",
185 printk(KERN_ERR
"ocfs2: Stack \"%s\" already registered\n",
189 spin_unlock(&ocfs2_stack_lock
);
193 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register
);
195 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin
*plugin
)
197 struct ocfs2_stack_plugin
*p
;
199 spin_lock(&ocfs2_stack_lock
);
200 p
= ocfs2_stack_lookup(plugin
->sp_name
);
203 BUG_ON(plugin
== active_stack
);
204 BUG_ON(plugin
->sp_count
!= 0);
205 list_del_init(&plugin
->sp_list
);
206 printk(KERN_INFO
"ocfs2: Unregistered cluster interface %s\n",
209 printk(KERN_ERR
"Stack \"%s\" is not registered\n",
212 spin_unlock(&ocfs2_stack_lock
);
214 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister
);
216 void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version
*max_proto
)
218 struct ocfs2_stack_plugin
*p
;
220 spin_lock(&ocfs2_stack_lock
);
221 if (memcmp(max_proto
, &locking_max_version
,
222 sizeof(struct ocfs2_protocol_version
))) {
223 BUG_ON(locking_max_version
.pv_major
!= 0);
225 locking_max_version
= *max_proto
;
226 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
227 p
->sp_max_proto
= locking_max_version
;
230 spin_unlock(&ocfs2_stack_lock
);
232 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version
);
236 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
237 * for the ast and bast functions. They will pass the lksb to the ast
238 * and bast. The caller can wrap the lksb with their own structure to
239 * get more information.
241 int ocfs2_dlm_lock(struct ocfs2_cluster_connection
*conn
,
243 struct ocfs2_dlm_lksb
*lksb
,
246 unsigned int namelen
)
248 if (!lksb
->lksb_conn
)
249 lksb
->lksb_conn
= conn
;
251 BUG_ON(lksb
->lksb_conn
!= conn
);
252 return active_stack
->sp_ops
->dlm_lock(conn
, mode
, lksb
, flags
,
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock
);
257 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection
*conn
,
258 struct ocfs2_dlm_lksb
*lksb
,
261 BUG_ON(lksb
->lksb_conn
== NULL
);
263 return active_stack
->sp_ops
->dlm_unlock(conn
, lksb
, flags
);
265 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock
);
267 int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb
*lksb
)
269 return active_stack
->sp_ops
->lock_status(lksb
);
271 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status
);
273 int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb
*lksb
)
275 return active_stack
->sp_ops
->lvb_valid(lksb
);
277 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid
);
279 void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb
*lksb
)
281 return active_stack
->sp_ops
->lock_lvb(lksb
);
283 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb
);
285 void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb
*lksb
)
287 active_stack
->sp_ops
->dump_lksb(lksb
);
289 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb
);
291 int ocfs2_stack_supports_plocks(void)
293 return active_stack
&& active_stack
->sp_ops
->plock
;
295 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks
);
298 * ocfs2_plock() can only be safely called if
299 * ocfs2_stack_supports_plocks() returned true
301 int ocfs2_plock(struct ocfs2_cluster_connection
*conn
, u64 ino
,
302 struct file
*file
, int cmd
, struct file_lock
*fl
)
304 WARN_ON_ONCE(active_stack
->sp_ops
->plock
== NULL
);
305 if (active_stack
->sp_ops
->plock
)
306 return active_stack
->sp_ops
->plock(conn
, ino
, file
, cmd
, fl
);
309 EXPORT_SYMBOL_GPL(ocfs2_plock
);
311 int ocfs2_cluster_connect(const char *stack_name
,
312 const char *cluster_name
,
313 int cluster_name_len
,
316 struct ocfs2_locking_protocol
*lproto
,
317 void (*recovery_handler
)(int node_num
,
318 void *recovery_data
),
320 struct ocfs2_cluster_connection
**conn
)
323 struct ocfs2_cluster_connection
*new_conn
;
325 BUG_ON(group
== NULL
);
326 BUG_ON(conn
== NULL
);
327 BUG_ON(recovery_handler
== NULL
);
329 if (grouplen
> GROUP_NAME_MAX
) {
334 if (memcmp(&lproto
->lp_max_version
, &locking_max_version
,
335 sizeof(struct ocfs2_protocol_version
))) {
340 new_conn
= kzalloc(sizeof(struct ocfs2_cluster_connection
),
347 strlcpy(new_conn
->cc_name
, group
, GROUP_NAME_MAX
+ 1);
348 new_conn
->cc_namelen
= grouplen
;
349 if (cluster_name_len
)
350 strlcpy(new_conn
->cc_cluster_name
, cluster_name
,
351 CLUSTER_NAME_MAX
+ 1);
352 new_conn
->cc_cluster_name_len
= cluster_name_len
;
353 new_conn
->cc_recovery_handler
= recovery_handler
;
354 new_conn
->cc_recovery_data
= recovery_data
;
356 new_conn
->cc_proto
= lproto
;
357 /* Start the new connection at our maximum compatibility level */
358 new_conn
->cc_version
= lproto
->lp_max_version
;
360 /* This will pin the stack driver if successful */
361 rc
= ocfs2_stack_driver_get(stack_name
);
365 rc
= active_stack
->sp_ops
->connect(new_conn
);
367 ocfs2_stack_driver_put();
380 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect
);
382 /* The caller will ensure all nodes have the same cluster stack */
383 int ocfs2_cluster_connect_agnostic(const char *group
,
385 struct ocfs2_locking_protocol
*lproto
,
386 void (*recovery_handler
)(int node_num
,
387 void *recovery_data
),
389 struct ocfs2_cluster_connection
**conn
)
391 char *stack_name
= NULL
;
393 if (cluster_stack_name
[0])
394 stack_name
= cluster_stack_name
;
395 return ocfs2_cluster_connect(stack_name
, NULL
, 0, group
, grouplen
,
396 lproto
, recovery_handler
, recovery_data
,
399 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic
);
401 /* If hangup_pending is 0, the stack driver will be dropped */
402 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection
*conn
,
407 BUG_ON(conn
== NULL
);
409 ret
= active_stack
->sp_ops
->disconnect(conn
);
411 /* XXX Should we free it anyway? */
415 ocfs2_stack_driver_put();
420 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect
);
423 * Leave the group for this filesystem. This is executed by a userspace
424 * program (stored in ocfs2_hb_ctl_path).
426 static void ocfs2_leave_group(const char *group
)
429 char *argv
[5], *envp
[3];
431 argv
[0] = ocfs2_hb_ctl_path
;
434 argv
[3] = (char *)group
;
437 /* minimal command environment taken from cpu_run_sbin_hotplug */
439 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
442 ret
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
445 "ocfs2: Error %d running user helper "
447 ret
, argv
[0], argv
[1], argv
[2], argv
[3]);
452 * Hangup is a required post-umount. ocfs2-tools software expects the
453 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
454 * regardless of whether the DLM got started, so we can't do it
455 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
458 void ocfs2_cluster_hangup(const char *group
, int grouplen
)
460 BUG_ON(group
== NULL
);
461 BUG_ON(group
[grouplen
] != '\0');
463 ocfs2_leave_group(group
);
465 /* cluster_disconnect() was called with hangup_pending==1 */
466 ocfs2_stack_driver_put();
468 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup
);
470 int ocfs2_cluster_this_node(struct ocfs2_cluster_connection
*conn
,
473 return active_stack
->sp_ops
->this_node(conn
, node
);
475 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node
);
482 static ssize_t
ocfs2_max_locking_protocol_show(struct kobject
*kobj
,
483 struct kobj_attribute
*attr
,
488 spin_lock(&ocfs2_stack_lock
);
489 if (locking_max_version
.pv_major
)
490 ret
= snprintf(buf
, PAGE_SIZE
, "%u.%u\n",
491 locking_max_version
.pv_major
,
492 locking_max_version
.pv_minor
);
493 spin_unlock(&ocfs2_stack_lock
);
498 static struct kobj_attribute ocfs2_attr_max_locking_protocol
=
499 __ATTR(max_locking_protocol
, S_IRUGO
,
500 ocfs2_max_locking_protocol_show
, NULL
);
502 static ssize_t
ocfs2_loaded_cluster_plugins_show(struct kobject
*kobj
,
503 struct kobj_attribute
*attr
,
506 ssize_t ret
= 0, total
= 0, remain
= PAGE_SIZE
;
507 struct ocfs2_stack_plugin
*p
;
509 spin_lock(&ocfs2_stack_lock
);
510 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
511 ret
= snprintf(buf
, remain
, "%s\n",
518 /* snprintf() didn't fit */
525 spin_unlock(&ocfs2_stack_lock
);
530 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins
=
531 __ATTR(loaded_cluster_plugins
, S_IRUGO
,
532 ocfs2_loaded_cluster_plugins_show
, NULL
);
534 static ssize_t
ocfs2_active_cluster_plugin_show(struct kobject
*kobj
,
535 struct kobj_attribute
*attr
,
540 spin_lock(&ocfs2_stack_lock
);
542 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n",
543 active_stack
->sp_name
);
544 if (ret
== PAGE_SIZE
)
547 spin_unlock(&ocfs2_stack_lock
);
552 static struct kobj_attribute ocfs2_attr_active_cluster_plugin
=
553 __ATTR(active_cluster_plugin
, S_IRUGO
,
554 ocfs2_active_cluster_plugin_show
, NULL
);
556 static ssize_t
ocfs2_cluster_stack_show(struct kobject
*kobj
,
557 struct kobj_attribute
*attr
,
561 spin_lock(&ocfs2_stack_lock
);
562 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n", cluster_stack_name
);
563 spin_unlock(&ocfs2_stack_lock
);
568 static ssize_t
ocfs2_cluster_stack_store(struct kobject
*kobj
,
569 struct kobj_attribute
*attr
,
570 const char *buf
, size_t count
)
578 if (buf
[len
- 1] == '\n')
581 if ((len
!= OCFS2_STACK_LABEL_LEN
) ||
582 (strnlen(buf
, len
) != len
))
585 spin_lock(&ocfs2_stack_lock
);
587 if (!strncmp(buf
, cluster_stack_name
, len
))
592 memcpy(cluster_stack_name
, buf
, len
);
595 spin_unlock(&ocfs2_stack_lock
);
601 static struct kobj_attribute ocfs2_attr_cluster_stack
=
602 __ATTR(cluster_stack
, S_IRUGO
| S_IWUSR
,
603 ocfs2_cluster_stack_show
,
604 ocfs2_cluster_stack_store
);
608 static ssize_t
ocfs2_dlm_recover_show(struct kobject
*kobj
,
609 struct kobj_attribute
*attr
,
612 return snprintf(buf
, PAGE_SIZE
, "1\n");
615 static struct kobj_attribute ocfs2_attr_dlm_recover_support
=
616 __ATTR(dlm_recover_callback_support
, S_IRUGO
,
617 ocfs2_dlm_recover_show
, NULL
);
619 static struct attribute
*ocfs2_attrs
[] = {
620 &ocfs2_attr_max_locking_protocol
.attr
,
621 &ocfs2_attr_loaded_cluster_plugins
.attr
,
622 &ocfs2_attr_active_cluster_plugin
.attr
,
623 &ocfs2_attr_cluster_stack
.attr
,
624 &ocfs2_attr_dlm_recover_support
.attr
,
628 static const struct attribute_group ocfs2_attr_group
= {
629 .attrs
= ocfs2_attrs
,
632 struct kset
*ocfs2_kset
;
633 EXPORT_SYMBOL_GPL(ocfs2_kset
);
635 static void ocfs2_sysfs_exit(void)
637 kset_unregister(ocfs2_kset
);
640 static int ocfs2_sysfs_init(void)
644 ocfs2_kset
= kset_create_and_add("ocfs2", NULL
, fs_kobj
);
648 ret
= sysfs_create_group(&ocfs2_kset
->kobj
, &ocfs2_attr_group
);
655 kset_unregister(ocfs2_kset
);
662 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
663 * make as much sense in a multiple cluster stack world, but it's safer
664 * and easier to preserve the name.
667 #define FS_OCFS2_NM 1
669 static struct ctl_table ocfs2_nm_table
[] = {
671 .procname
= "hb_ctl_path",
672 .data
= ocfs2_hb_ctl_path
,
673 .maxlen
= OCFS2_MAX_HB_CTL_PATH
,
675 .proc_handler
= proc_dostring
,
680 static struct ctl_table ocfs2_mod_table
[] = {
686 .child
= ocfs2_nm_table
691 static struct ctl_table ocfs2_kern_table
[] = {
697 .child
= ocfs2_mod_table
702 static struct ctl_table ocfs2_root_table
[] = {
708 .child
= ocfs2_kern_table
713 static struct ctl_table_header
*ocfs2_table_header
;
720 static int __init
ocfs2_stack_glue_init(void)
722 strcpy(cluster_stack_name
, OCFS2_STACK_PLUGIN_O2CB
);
724 ocfs2_table_header
= register_sysctl_table(ocfs2_root_table
);
725 if (!ocfs2_table_header
) {
727 "ocfs2 stack glue: unable to register sysctl\n");
728 return -ENOMEM
; /* or something. */
731 return ocfs2_sysfs_init();
734 static void __exit
ocfs2_stack_glue_exit(void)
736 memset(&locking_max_version
, 0,
737 sizeof(struct ocfs2_protocol_version
));
739 if (ocfs2_table_header
)
740 unregister_sysctl_table(ocfs2_table_header
);
743 MODULE_AUTHOR("Oracle");
744 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
745 MODULE_LICENSE("GPL");
746 module_init(ocfs2_stack_glue_init
);
747 module_exit(ocfs2_stack_glue_exit
);