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 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_locking_protocol
*lproto
;
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_proto
= lproto
;
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_locking_protocol(struct ocfs2_locking_protocol
*proto
)
218 struct ocfs2_stack_plugin
*p
;
220 BUG_ON(proto
== NULL
);
222 spin_lock(&ocfs2_stack_lock
);
223 BUG_ON(active_stack
!= NULL
);
226 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
227 p
->sp_proto
= lproto
;
230 spin_unlock(&ocfs2_stack_lock
);
232 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_locking_protocol
);
236 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take
237 * "struct ocfs2_lock_res *astarg" instead of "void *astarg" because the
238 * underlying stack plugins need to pilfer the lksb off of the lock_res.
239 * If some other structure needs to be passed as an astarg, the plugins
240 * will need to be given a different avenue to the lksb.
242 int ocfs2_dlm_lock(struct ocfs2_cluster_connection
*conn
,
244 union ocfs2_dlm_lksb
*lksb
,
247 unsigned int namelen
,
248 struct ocfs2_lock_res
*astarg
)
250 BUG_ON(lproto
== NULL
);
252 return active_stack
->sp_ops
->dlm_lock(conn
, mode
, lksb
, flags
,
253 name
, namelen
, astarg
);
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock
);
257 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection
*conn
,
258 union ocfs2_dlm_lksb
*lksb
,
260 struct ocfs2_lock_res
*astarg
)
262 BUG_ON(lproto
== NULL
);
264 return active_stack
->sp_ops
->dlm_unlock(conn
, lksb
, flags
, astarg
);
266 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock
);
268 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb
*lksb
)
270 return active_stack
->sp_ops
->lock_status(lksb
);
272 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status
);
275 * Why don't we cast to ocfs2_meta_lvb? The "clean" answer is that we
276 * don't cast at the glue level. The real answer is that the header
277 * ordering is nigh impossible.
279 void *ocfs2_dlm_lvb(union 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(union 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
,
314 void (*recovery_handler
)(int node_num
,
315 void *recovery_data
),
317 struct ocfs2_cluster_connection
**conn
)
320 struct ocfs2_cluster_connection
*new_conn
;
322 BUG_ON(group
== NULL
);
323 BUG_ON(conn
== NULL
);
324 BUG_ON(recovery_handler
== NULL
);
326 if (grouplen
> GROUP_NAME_MAX
) {
331 new_conn
= kzalloc(sizeof(struct ocfs2_cluster_connection
),
338 memcpy(new_conn
->cc_name
, group
, grouplen
);
339 new_conn
->cc_namelen
= grouplen
;
340 new_conn
->cc_recovery_handler
= recovery_handler
;
341 new_conn
->cc_recovery_data
= recovery_data
;
343 /* Start the new connection at our maximum compatibility level */
344 new_conn
->cc_version
= lproto
->lp_max_version
;
346 /* This will pin the stack driver if successful */
347 rc
= ocfs2_stack_driver_get(stack_name
);
351 rc
= active_stack
->sp_ops
->connect(new_conn
);
353 ocfs2_stack_driver_put();
366 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect
);
368 /* If hangup_pending is 0, the stack driver will be dropped */
369 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection
*conn
,
374 BUG_ON(conn
== NULL
);
376 ret
= active_stack
->sp_ops
->disconnect(conn
);
378 /* XXX Should we free it anyway? */
382 ocfs2_stack_driver_put();
387 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect
);
390 * Leave the group for this filesystem. This is executed by a userspace
391 * program (stored in ocfs2_hb_ctl_path).
393 static void ocfs2_leave_group(const char *group
)
396 char *argv
[5], *envp
[3];
398 argv
[0] = ocfs2_hb_ctl_path
;
401 argv
[3] = (char *)group
;
404 /* minimal command environment taken from cpu_run_sbin_hotplug */
406 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
409 ret
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
412 "ocfs2: Error %d running user helper "
414 ret
, argv
[0], argv
[1], argv
[2], argv
[3]);
419 * Hangup is a required post-umount. ocfs2-tools software expects the
420 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
421 * regardless of whether the DLM got started, so we can't do it
422 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
425 void ocfs2_cluster_hangup(const char *group
, int grouplen
)
427 BUG_ON(group
== NULL
);
428 BUG_ON(group
[grouplen
] != '\0');
430 ocfs2_leave_group(group
);
432 /* cluster_disconnect() was called with hangup_pending==1 */
433 ocfs2_stack_driver_put();
435 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup
);
437 int ocfs2_cluster_this_node(unsigned int *node
)
439 return active_stack
->sp_ops
->this_node(node
);
441 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node
);
448 static ssize_t
ocfs2_max_locking_protocol_show(struct kobject
*kobj
,
449 struct kobj_attribute
*attr
,
454 spin_lock(&ocfs2_stack_lock
);
456 ret
= snprintf(buf
, PAGE_SIZE
, "%u.%u\n",
457 lproto
->lp_max_version
.pv_major
,
458 lproto
->lp_max_version
.pv_minor
);
459 spin_unlock(&ocfs2_stack_lock
);
464 static struct kobj_attribute ocfs2_attr_max_locking_protocol
=
465 __ATTR(max_locking_protocol
, S_IFREG
| S_IRUGO
,
466 ocfs2_max_locking_protocol_show
, NULL
);
468 static ssize_t
ocfs2_loaded_cluster_plugins_show(struct kobject
*kobj
,
469 struct kobj_attribute
*attr
,
472 ssize_t ret
= 0, total
= 0, remain
= PAGE_SIZE
;
473 struct ocfs2_stack_plugin
*p
;
475 spin_lock(&ocfs2_stack_lock
);
476 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
477 ret
= snprintf(buf
, remain
, "%s\n",
484 /* snprintf() didn't fit */
491 spin_unlock(&ocfs2_stack_lock
);
496 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins
=
497 __ATTR(loaded_cluster_plugins
, S_IFREG
| S_IRUGO
,
498 ocfs2_loaded_cluster_plugins_show
, NULL
);
500 static ssize_t
ocfs2_active_cluster_plugin_show(struct kobject
*kobj
,
501 struct kobj_attribute
*attr
,
506 spin_lock(&ocfs2_stack_lock
);
508 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n",
509 active_stack
->sp_name
);
510 if (ret
== PAGE_SIZE
)
513 spin_unlock(&ocfs2_stack_lock
);
518 static struct kobj_attribute ocfs2_attr_active_cluster_plugin
=
519 __ATTR(active_cluster_plugin
, S_IFREG
| S_IRUGO
,
520 ocfs2_active_cluster_plugin_show
, NULL
);
522 static ssize_t
ocfs2_cluster_stack_show(struct kobject
*kobj
,
523 struct kobj_attribute
*attr
,
527 spin_lock(&ocfs2_stack_lock
);
528 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n", cluster_stack_name
);
529 spin_unlock(&ocfs2_stack_lock
);
534 static ssize_t
ocfs2_cluster_stack_store(struct kobject
*kobj
,
535 struct kobj_attribute
*attr
,
536 const char *buf
, size_t count
)
544 if (buf
[len
- 1] == '\n')
547 if ((len
!= OCFS2_STACK_LABEL_LEN
) ||
548 (strnlen(buf
, len
) != len
))
551 spin_lock(&ocfs2_stack_lock
);
553 if (!strncmp(buf
, cluster_stack_name
, len
))
558 memcpy(cluster_stack_name
, buf
, len
);
561 spin_unlock(&ocfs2_stack_lock
);
567 static struct kobj_attribute ocfs2_attr_cluster_stack
=
568 __ATTR(cluster_stack
, S_IFREG
| S_IRUGO
| S_IWUSR
,
569 ocfs2_cluster_stack_show
,
570 ocfs2_cluster_stack_store
);
572 static struct attribute
*ocfs2_attrs
[] = {
573 &ocfs2_attr_max_locking_protocol
.attr
,
574 &ocfs2_attr_loaded_cluster_plugins
.attr
,
575 &ocfs2_attr_active_cluster_plugin
.attr
,
576 &ocfs2_attr_cluster_stack
.attr
,
580 static struct attribute_group ocfs2_attr_group
= {
581 .attrs
= ocfs2_attrs
,
584 static struct kset
*ocfs2_kset
;
586 static void ocfs2_sysfs_exit(void)
588 kset_unregister(ocfs2_kset
);
591 static int ocfs2_sysfs_init(void)
595 ocfs2_kset
= kset_create_and_add("ocfs2", NULL
, fs_kobj
);
599 ret
= sysfs_create_group(&ocfs2_kset
->kobj
, &ocfs2_attr_group
);
606 kset_unregister(ocfs2_kset
);
613 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
614 * make as much sense in a multiple cluster stack world, but it's safer
615 * and easier to preserve the name.
618 #define FS_OCFS2_NM 1
620 static ctl_table ocfs2_nm_table
[] = {
623 .procname
= "hb_ctl_path",
624 .data
= ocfs2_hb_ctl_path
,
625 .maxlen
= OCFS2_MAX_HB_CTL_PATH
,
627 .proc_handler
= &proc_dostring
,
628 .strategy
= &sysctl_string
,
633 static ctl_table ocfs2_mod_table
[] = {
635 .ctl_name
= FS_OCFS2_NM
,
640 .child
= ocfs2_nm_table
645 static ctl_table ocfs2_kern_table
[] = {
647 .ctl_name
= FS_OCFS2
,
652 .child
= ocfs2_mod_table
657 static ctl_table ocfs2_root_table
[] = {
664 .child
= ocfs2_kern_table
669 static struct ctl_table_header
*ocfs2_table_header
= NULL
;
676 static int __init
ocfs2_stack_glue_init(void)
678 strcpy(cluster_stack_name
, OCFS2_STACK_PLUGIN_O2CB
);
680 ocfs2_table_header
= register_sysctl_table(ocfs2_root_table
);
681 if (!ocfs2_table_header
) {
683 "ocfs2 stack glue: unable to register sysctl\n");
684 return -ENOMEM
; /* or something. */
687 return ocfs2_sysfs_init();
690 static void __exit
ocfs2_stack_glue_exit(void)
694 if (ocfs2_table_header
)
695 unregister_sysctl_table(ocfs2_table_header
);
698 MODULE_AUTHOR("Oracle");
699 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
700 MODULE_LICENSE("GPL");
701 module_init(ocfs2_stack_glue_init
);
702 module_exit(ocfs2_stack_glue_exit
);