1 // SPDX-License-Identifier: GPL-2.0-only
5 * Code which implements an OCFS2 specific interface to underlying
8 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
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>
17 #include <linux/kobject.h>
18 #include <linux/sysfs.h>
19 #include <linux/sysctl.h>
23 #include "stackglue.h"
25 #define OCFS2_STACK_PLUGIN_O2CB "o2cb"
26 #define OCFS2_STACK_PLUGIN_USER "user"
27 #define OCFS2_MAX_HB_CTL_PATH 256
29 static struct ocfs2_protocol_version locking_max_version
;
30 static DEFINE_SPINLOCK(ocfs2_stack_lock
);
31 static LIST_HEAD(ocfs2_stack_list
);
32 static char cluster_stack_name
[OCFS2_STACK_LABEL_LEN
+ 1];
33 static char ocfs2_hb_ctl_path
[OCFS2_MAX_HB_CTL_PATH
] = "/sbin/ocfs2_hb_ctl";
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.
39 static struct ocfs2_stack_plugin
*active_stack
;
41 static struct ocfs2_stack_plugin
*ocfs2_stack_lookup(const char *name
)
43 struct ocfs2_stack_plugin
*p
;
45 assert_spin_locked(&ocfs2_stack_lock
);
47 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
48 if (!strcmp(p
->sp_name
, name
))
55 static int ocfs2_stack_driver_request(const char *stack_name
,
56 const char *plugin_name
)
59 struct ocfs2_stack_plugin
*p
;
61 spin_lock(&ocfs2_stack_lock
);
64 * If the stack passed by the filesystem isn't the selected one,
67 if (strcmp(stack_name
, cluster_stack_name
)) {
74 * If the active stack isn't the one we want, it cannot
75 * be selected right now.
77 if (!strcmp(active_stack
->sp_name
, plugin_name
))
84 p
= ocfs2_stack_lookup(plugin_name
);
85 if (!p
|| !try_module_get(p
->sp_owner
)) {
94 /* If we found it, pin it */
96 active_stack
->sp_count
++;
98 spin_unlock(&ocfs2_stack_lock
);
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.
107 static int ocfs2_stack_driver_get(const char *stack_name
)
110 char *plugin_name
= OCFS2_STACK_PLUGIN_O2CB
;
113 * Classic stack does not pass in a stack name. This is
114 * compatible with older tools as well.
116 if (!stack_name
|| !*stack_name
)
117 stack_name
= OCFS2_STACK_PLUGIN_O2CB
;
119 if (strlen(stack_name
) != OCFS2_STACK_LABEL_LEN
) {
121 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
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
;
130 rc
= ocfs2_stack_driver_request(stack_name
, plugin_name
);
132 request_module("ocfs2_stack_%s", plugin_name
);
133 rc
= ocfs2_stack_driver_request(stack_name
, plugin_name
);
138 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
140 } else if (rc
== -EBUSY
) {
142 "ocfs2: A different cluster stack is in use\n");
148 static void ocfs2_stack_driver_put(void)
150 spin_lock(&ocfs2_stack_lock
);
151 BUG_ON(active_stack
== NULL
);
152 BUG_ON(active_stack
->sp_count
== 0);
154 active_stack
->sp_count
--;
155 if (!active_stack
->sp_count
) {
156 module_put(active_stack
->sp_owner
);
159 spin_unlock(&ocfs2_stack_lock
);
162 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin
*plugin
)
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",
175 printk(KERN_ERR
"ocfs2: Stack \"%s\" already registered\n",
179 spin_unlock(&ocfs2_stack_lock
);
183 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register
);
185 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin
*plugin
)
187 struct ocfs2_stack_plugin
*p
;
189 spin_lock(&ocfs2_stack_lock
);
190 p
= ocfs2_stack_lookup(plugin
->sp_name
);
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",
199 printk(KERN_ERR
"Stack \"%s\" is not registered\n",
202 spin_unlock(&ocfs2_stack_lock
);
204 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister
);
206 void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version
*max_proto
)
208 struct ocfs2_stack_plugin
*p
;
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);
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
;
220 spin_unlock(&ocfs2_stack_lock
);
222 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version
);
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.
231 int ocfs2_dlm_lock(struct ocfs2_cluster_connection
*conn
,
233 struct ocfs2_dlm_lksb
*lksb
,
236 unsigned int namelen
)
238 if (!lksb
->lksb_conn
)
239 lksb
->lksb_conn
= conn
;
241 BUG_ON(lksb
->lksb_conn
!= conn
);
242 return active_stack
->sp_ops
->dlm_lock(conn
, mode
, lksb
, flags
,
245 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock
);
247 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection
*conn
,
248 struct ocfs2_dlm_lksb
*lksb
,
251 BUG_ON(lksb
->lksb_conn
== NULL
);
253 return active_stack
->sp_ops
->dlm_unlock(conn
, lksb
, flags
);
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock
);
257 int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb
*lksb
)
259 return active_stack
->sp_ops
->lock_status(lksb
);
261 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status
);
263 int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb
*lksb
)
265 return active_stack
->sp_ops
->lvb_valid(lksb
);
267 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid
);
269 void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb
*lksb
)
271 return active_stack
->sp_ops
->lock_lvb(lksb
);
273 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb
);
275 void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb
*lksb
)
277 active_stack
->sp_ops
->dump_lksb(lksb
);
279 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb
);
281 int ocfs2_stack_supports_plocks(void)
283 return active_stack
&& active_stack
->sp_ops
->plock
;
285 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks
);
288 * ocfs2_plock() can only be safely called if
289 * ocfs2_stack_supports_plocks() returned true
291 int ocfs2_plock(struct ocfs2_cluster_connection
*conn
, u64 ino
,
292 struct file
*file
, int cmd
, struct file_lock
*fl
)
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
);
299 EXPORT_SYMBOL_GPL(ocfs2_plock
);
301 int ocfs2_cluster_connect(const char *stack_name
,
302 const char *cluster_name
,
303 int cluster_name_len
,
306 struct ocfs2_locking_protocol
*lproto
,
307 void (*recovery_handler
)(int node_num
,
308 void *recovery_data
),
310 struct ocfs2_cluster_connection
**conn
)
313 struct ocfs2_cluster_connection
*new_conn
;
315 BUG_ON(group
== NULL
);
316 BUG_ON(conn
== NULL
);
317 BUG_ON(recovery_handler
== NULL
);
319 if (grouplen
> GROUP_NAME_MAX
) {
324 if (memcmp(&lproto
->lp_max_version
, &locking_max_version
,
325 sizeof(struct ocfs2_protocol_version
))) {
330 new_conn
= kzalloc(sizeof(struct ocfs2_cluster_connection
),
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
;
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
;
350 /* This will pin the stack driver if successful */
351 rc
= ocfs2_stack_driver_get(stack_name
);
355 rc
= active_stack
->sp_ops
->connect(new_conn
);
357 ocfs2_stack_driver_put();
370 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect
);
372 /* The caller will ensure all nodes have the same cluster stack */
373 int ocfs2_cluster_connect_agnostic(const char *group
,
375 struct ocfs2_locking_protocol
*lproto
,
376 void (*recovery_handler
)(int node_num
,
377 void *recovery_data
),
379 struct ocfs2_cluster_connection
**conn
)
381 char *stack_name
= NULL
;
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
,
389 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic
);
391 /* If hangup_pending is 0, the stack driver will be dropped */
392 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection
*conn
,
397 BUG_ON(conn
== NULL
);
399 ret
= active_stack
->sp_ops
->disconnect(conn
);
401 /* XXX Should we free it anyway? */
405 ocfs2_stack_driver_put();
410 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect
);
413 * Leave the group for this filesystem. This is executed by a userspace
414 * program (stored in ocfs2_hb_ctl_path).
416 static void ocfs2_leave_group(const char *group
)
419 char *argv
[5], *envp
[3];
421 argv
[0] = ocfs2_hb_ctl_path
;
424 argv
[3] = (char *)group
;
427 /* minimal command environment taken from cpu_run_sbin_hotplug */
429 envp
[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
432 ret
= call_usermodehelper(argv
[0], argv
, envp
, UMH_WAIT_PROC
);
435 "ocfs2: Error %d running user helper "
437 ret
, argv
[0], argv
[1], argv
[2], argv
[3]);
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
448 void ocfs2_cluster_hangup(const char *group
, int grouplen
)
450 BUG_ON(group
== NULL
);
451 BUG_ON(group
[grouplen
] != '\0');
453 ocfs2_leave_group(group
);
455 /* cluster_disconnect() was called with hangup_pending==1 */
456 ocfs2_stack_driver_put();
458 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup
);
460 int ocfs2_cluster_this_node(struct ocfs2_cluster_connection
*conn
,
463 return active_stack
->sp_ops
->this_node(conn
, node
);
465 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node
);
472 static ssize_t
ocfs2_max_locking_protocol_show(struct kobject
*kobj
,
473 struct kobj_attribute
*attr
,
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
);
488 static struct kobj_attribute ocfs2_attr_max_locking_protocol
=
489 __ATTR(max_locking_protocol
, S_IRUGO
,
490 ocfs2_max_locking_protocol_show
, NULL
);
492 static ssize_t
ocfs2_loaded_cluster_plugins_show(struct kobject
*kobj
,
493 struct kobj_attribute
*attr
,
496 ssize_t ret
= 0, total
= 0, remain
= PAGE_SIZE
;
497 struct ocfs2_stack_plugin
*p
;
499 spin_lock(&ocfs2_stack_lock
);
500 list_for_each_entry(p
, &ocfs2_stack_list
, sp_list
) {
501 ret
= snprintf(buf
, remain
, "%s\n",
504 /* snprintf() didn't fit */
511 spin_unlock(&ocfs2_stack_lock
);
516 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins
=
517 __ATTR(loaded_cluster_plugins
, S_IRUGO
,
518 ocfs2_loaded_cluster_plugins_show
, NULL
);
520 static ssize_t
ocfs2_active_cluster_plugin_show(struct kobject
*kobj
,
521 struct kobj_attribute
*attr
,
526 spin_lock(&ocfs2_stack_lock
);
528 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n",
529 active_stack
->sp_name
);
530 if (ret
>= PAGE_SIZE
)
533 spin_unlock(&ocfs2_stack_lock
);
538 static struct kobj_attribute ocfs2_attr_active_cluster_plugin
=
539 __ATTR(active_cluster_plugin
, S_IRUGO
,
540 ocfs2_active_cluster_plugin_show
, NULL
);
542 static ssize_t
ocfs2_cluster_stack_show(struct kobject
*kobj
,
543 struct kobj_attribute
*attr
,
547 spin_lock(&ocfs2_stack_lock
);
548 ret
= snprintf(buf
, PAGE_SIZE
, "%s\n", cluster_stack_name
);
549 spin_unlock(&ocfs2_stack_lock
);
554 static ssize_t
ocfs2_cluster_stack_store(struct kobject
*kobj
,
555 struct kobj_attribute
*attr
,
556 const char *buf
, size_t count
)
564 if (buf
[len
- 1] == '\n')
567 if ((len
!= OCFS2_STACK_LABEL_LEN
) ||
568 (strnlen(buf
, len
) != len
))
571 spin_lock(&ocfs2_stack_lock
);
573 if (!strncmp(buf
, cluster_stack_name
, len
))
578 memcpy(cluster_stack_name
, buf
, len
);
581 spin_unlock(&ocfs2_stack_lock
);
587 static 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
);
594 static ssize_t
ocfs2_dlm_recover_show(struct kobject
*kobj
,
595 struct kobj_attribute
*attr
,
598 return snprintf(buf
, PAGE_SIZE
, "1\n");
601 static struct kobj_attribute ocfs2_attr_dlm_recover_support
=
602 __ATTR(dlm_recover_callback_support
, S_IRUGO
,
603 ocfs2_dlm_recover_show
, NULL
);
605 static 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
,
614 static const struct attribute_group ocfs2_attr_group
= {
615 .attrs
= ocfs2_attrs
,
618 struct kset
*ocfs2_kset
;
619 EXPORT_SYMBOL_GPL(ocfs2_kset
);
621 static void ocfs2_sysfs_exit(void)
623 kset_unregister(ocfs2_kset
);
626 static int ocfs2_sysfs_init(void)
630 ocfs2_kset
= kset_create_and_add("ocfs2", NULL
, fs_kobj
);
634 ret
= sysfs_create_group(&ocfs2_kset
->kobj
, &ocfs2_attr_group
);
641 kset_unregister(ocfs2_kset
);
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.
653 static struct ctl_table ocfs2_nm_table
[] = {
655 .procname
= "hb_ctl_path",
656 .data
= ocfs2_hb_ctl_path
,
657 .maxlen
= OCFS2_MAX_HB_CTL_PATH
,
659 .proc_handler
= proc_dostring
,
663 static struct ctl_table_header
*ocfs2_table_header
;
669 static int __init
ocfs2_stack_glue_init(void)
673 strcpy(cluster_stack_name
, OCFS2_STACK_PLUGIN_O2CB
);
675 ocfs2_table_header
= register_sysctl("fs/ocfs2/nm", ocfs2_nm_table
);
676 if (!ocfs2_table_header
) {
678 "ocfs2 stack glue: unable to register sysctl\n");
679 return -ENOMEM
; /* or something. */
682 ret
= ocfs2_sysfs_init();
684 unregister_sysctl_table(ocfs2_table_header
);
689 static void __exit
ocfs2_stack_glue_exit(void)
691 memset(&locking_max_version
, 0,
692 sizeof(struct ocfs2_protocol_version
));
694 if (ocfs2_table_header
)
695 unregister_sysctl_table(ocfs2_table_header
);
698 MODULE_AUTHOR("Oracle");
699 MODULE_DESCRIPTION("ocfs2 cluster stack glue layer");
700 MODULE_LICENSE("GPL");
701 module_init(ocfs2_stack_glue_init
);
702 module_exit(ocfs2_stack_glue_exit
);