mlock: avoid dirtying pages and triggering writeback
[linux-2.6/next.git] / fs / ocfs2 / stackglue.c
blob39abf89697edfc296173dce40a3e5aab1575202b
1 /* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
4 * stackglue.c
6 * Code which implements an OCFS2 specific interface to underlying
7 * cluster stacks.
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>
26 #include <linux/fs.h>
27 #include <linux/kobject.h>
28 #include <linux/sysfs.h>
29 #include <linux/sysctl.h>
31 #include "ocfs2_fs.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))
59 return p;
62 return NULL;
65 static int ocfs2_stack_driver_request(const char *stack_name,
66 const char *plugin_name)
68 int rc;
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,
75 * we can't continue.
77 if (strcmp(stack_name, cluster_stack_name)) {
78 rc = -EBUSY;
79 goto out;
82 if (active_stack) {
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))
88 rc = 0;
89 else
90 rc = -EBUSY;
91 goto out;
94 p = ocfs2_stack_lookup(plugin_name);
95 if (!p || !try_module_get(p->sp_owner)) {
96 rc = -ENOENT;
97 goto out;
100 active_stack = p;
101 rc = 0;
103 out:
104 /* If we found it, pin it */
105 if (!rc)
106 active_stack->sp_count++;
108 spin_unlock(&ocfs2_stack_lock);
109 return rc;
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)
119 int rc;
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) {
130 printk(KERN_ERR
131 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
132 stack_name);
133 return -EINVAL;
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);
141 if (rc == -ENOENT) {
142 request_module("ocfs2_stack_%s", plugin_name);
143 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
146 if (rc == -ENOENT) {
147 printk(KERN_ERR
148 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
149 plugin_name);
150 } else if (rc == -EBUSY) {
151 printk(KERN_ERR
152 "ocfs2: A different cluster stack is in use\n");
155 return rc;
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);
167 active_stack = NULL;
169 spin_unlock(&ocfs2_stack_lock);
172 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
174 int rc;
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",
182 plugin->sp_name);
183 rc = 0;
184 } else {
185 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
186 plugin->sp_name);
187 rc = -EEXIST;
189 spin_unlock(&ocfs2_stack_lock);
191 return rc;
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);
201 if (p) {
202 BUG_ON(p != plugin);
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",
207 plugin->sp_name);
208 } else {
209 printk(KERN_ERR "Stack \"%s\" is not registered\n",
210 plugin->sp_name);
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,
242 int mode,
243 struct ocfs2_dlm_lksb *lksb,
244 u32 flags,
245 void *name,
246 unsigned int namelen)
248 if (!lksb->lksb_conn)
249 lksb->lksb_conn = conn;
250 else
251 BUG_ON(lksb->lksb_conn != conn);
252 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
253 name, namelen);
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
257 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
258 struct ocfs2_dlm_lksb *lksb,
259 u32 flags)
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);
307 return -EOPNOTSUPP;
309 EXPORT_SYMBOL_GPL(ocfs2_plock);
311 int ocfs2_cluster_connect(const char *stack_name,
312 const char *group,
313 int grouplen,
314 struct ocfs2_locking_protocol *lproto,
315 void (*recovery_handler)(int node_num,
316 void *recovery_data),
317 void *recovery_data,
318 struct ocfs2_cluster_connection **conn)
320 int rc = 0;
321 struct ocfs2_cluster_connection *new_conn;
323 BUG_ON(group == NULL);
324 BUG_ON(conn == NULL);
325 BUG_ON(recovery_handler == NULL);
327 if (grouplen > GROUP_NAME_MAX) {
328 rc = -EINVAL;
329 goto out;
332 if (memcmp(&lproto->lp_max_version, &locking_max_version,
333 sizeof(struct ocfs2_protocol_version))) {
334 rc = -EINVAL;
335 goto out;
338 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
339 GFP_KERNEL);
340 if (!new_conn) {
341 rc = -ENOMEM;
342 goto out;
345 memcpy(new_conn->cc_name, group, grouplen);
346 new_conn->cc_namelen = grouplen;
347 new_conn->cc_recovery_handler = recovery_handler;
348 new_conn->cc_recovery_data = recovery_data;
350 new_conn->cc_proto = lproto;
351 /* Start the new connection at our maximum compatibility level */
352 new_conn->cc_version = lproto->lp_max_version;
354 /* This will pin the stack driver if successful */
355 rc = ocfs2_stack_driver_get(stack_name);
356 if (rc)
357 goto out_free;
359 rc = active_stack->sp_ops->connect(new_conn);
360 if (rc) {
361 ocfs2_stack_driver_put();
362 goto out_free;
365 *conn = new_conn;
367 out_free:
368 if (rc)
369 kfree(new_conn);
371 out:
372 return rc;
374 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
376 /* The caller will ensure all nodes have the same cluster stack */
377 int ocfs2_cluster_connect_agnostic(const char *group,
378 int grouplen,
379 struct ocfs2_locking_protocol *lproto,
380 void (*recovery_handler)(int node_num,
381 void *recovery_data),
382 void *recovery_data,
383 struct ocfs2_cluster_connection **conn)
385 char *stack_name = NULL;
387 if (cluster_stack_name[0])
388 stack_name = cluster_stack_name;
389 return ocfs2_cluster_connect(stack_name, group, grouplen, lproto,
390 recovery_handler, recovery_data, conn);
392 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
394 /* If hangup_pending is 0, the stack driver will be dropped */
395 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
396 int hangup_pending)
398 int ret;
400 BUG_ON(conn == NULL);
402 ret = active_stack->sp_ops->disconnect(conn);
404 /* XXX Should we free it anyway? */
405 if (!ret) {
406 kfree(conn);
407 if (!hangup_pending)
408 ocfs2_stack_driver_put();
411 return ret;
413 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
416 * Leave the group for this filesystem. This is executed by a userspace
417 * program (stored in ocfs2_hb_ctl_path).
419 static void ocfs2_leave_group(const char *group)
421 int ret;
422 char *argv[5], *envp[3];
424 argv[0] = ocfs2_hb_ctl_path;
425 argv[1] = "-K";
426 argv[2] = "-u";
427 argv[3] = (char *)group;
428 argv[4] = NULL;
430 /* minimal command environment taken from cpu_run_sbin_hotplug */
431 envp[0] = "HOME=/";
432 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
433 envp[2] = NULL;
435 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
436 if (ret < 0) {
437 printk(KERN_ERR
438 "ocfs2: Error %d running user helper "
439 "\"%s %s %s %s\"\n",
440 ret, argv[0], argv[1], argv[2], argv[3]);
445 * Hangup is a required post-umount. ocfs2-tools software expects the
446 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
447 * regardless of whether the DLM got started, so we can't do it
448 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
449 * the actual work.
451 void ocfs2_cluster_hangup(const char *group, int grouplen)
453 BUG_ON(group == NULL);
454 BUG_ON(group[grouplen] != '\0');
456 ocfs2_leave_group(group);
458 /* cluster_disconnect() was called with hangup_pending==1 */
459 ocfs2_stack_driver_put();
461 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
463 int ocfs2_cluster_this_node(unsigned int *node)
465 return active_stack->sp_ops->this_node(node);
467 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
471 * Sysfs bits
474 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
475 struct kobj_attribute *attr,
476 char *buf)
478 ssize_t ret = 0;
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);
487 return ret;
490 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
491 __ATTR(max_locking_protocol, S_IFREG | 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,
496 char *buf)
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",
504 p->sp_name);
505 if (ret < 0) {
506 total = ret;
507 break;
509 if (ret == remain) {
510 /* snprintf() didn't fit */
511 total = -E2BIG;
512 break;
514 total += ret;
515 remain -= ret;
517 spin_unlock(&ocfs2_stack_lock);
519 return total;
522 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
523 __ATTR(loaded_cluster_plugins, S_IFREG | 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,
528 char *buf)
530 ssize_t ret = 0;
532 spin_lock(&ocfs2_stack_lock);
533 if (active_stack) {
534 ret = snprintf(buf, PAGE_SIZE, "%s\n",
535 active_stack->sp_name);
536 if (ret == PAGE_SIZE)
537 ret = -E2BIG;
539 spin_unlock(&ocfs2_stack_lock);
541 return ret;
544 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
545 __ATTR(active_cluster_plugin, S_IFREG | 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,
550 char *buf)
552 ssize_t ret;
553 spin_lock(&ocfs2_stack_lock);
554 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
555 spin_unlock(&ocfs2_stack_lock);
557 return ret;
560 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
561 struct kobj_attribute *attr,
562 const char *buf, size_t count)
564 size_t len = count;
565 ssize_t ret;
567 if (len == 0)
568 return len;
570 if (buf[len - 1] == '\n')
571 len--;
573 if ((len != OCFS2_STACK_LABEL_LEN) ||
574 (strnlen(buf, len) != len))
575 return -EINVAL;
577 spin_lock(&ocfs2_stack_lock);
578 if (active_stack) {
579 if (!strncmp(buf, cluster_stack_name, len))
580 ret = count;
581 else
582 ret = -EBUSY;
583 } else {
584 memcpy(cluster_stack_name, buf, len);
585 ret = count;
587 spin_unlock(&ocfs2_stack_lock);
589 return ret;
593 static struct kobj_attribute ocfs2_attr_cluster_stack =
594 __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
595 ocfs2_cluster_stack_show,
596 ocfs2_cluster_stack_store);
598 static struct attribute *ocfs2_attrs[] = {
599 &ocfs2_attr_max_locking_protocol.attr,
600 &ocfs2_attr_loaded_cluster_plugins.attr,
601 &ocfs2_attr_active_cluster_plugin.attr,
602 &ocfs2_attr_cluster_stack.attr,
603 NULL,
606 static struct attribute_group ocfs2_attr_group = {
607 .attrs = ocfs2_attrs,
610 static struct kset *ocfs2_kset;
612 static void ocfs2_sysfs_exit(void)
614 kset_unregister(ocfs2_kset);
617 static int ocfs2_sysfs_init(void)
619 int ret;
621 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
622 if (!ocfs2_kset)
623 return -ENOMEM;
625 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
626 if (ret)
627 goto error;
629 return 0;
631 error:
632 kset_unregister(ocfs2_kset);
633 return ret;
637 * Sysctl bits
639 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
640 * make as much sense in a multiple cluster stack world, but it's safer
641 * and easier to preserve the name.
644 #define FS_OCFS2_NM 1
646 static ctl_table ocfs2_nm_table[] = {
648 .procname = "hb_ctl_path",
649 .data = ocfs2_hb_ctl_path,
650 .maxlen = OCFS2_MAX_HB_CTL_PATH,
651 .mode = 0644,
652 .proc_handler = proc_dostring,
657 static ctl_table ocfs2_mod_table[] = {
659 .procname = "nm",
660 .data = NULL,
661 .maxlen = 0,
662 .mode = 0555,
663 .child = ocfs2_nm_table
668 static ctl_table ocfs2_kern_table[] = {
670 .procname = "ocfs2",
671 .data = NULL,
672 .maxlen = 0,
673 .mode = 0555,
674 .child = ocfs2_mod_table
679 static ctl_table ocfs2_root_table[] = {
681 .procname = "fs",
682 .data = NULL,
683 .maxlen = 0,
684 .mode = 0555,
685 .child = ocfs2_kern_table
690 static struct ctl_table_header *ocfs2_table_header = NULL;
694 * Initialization
697 static int __init ocfs2_stack_glue_init(void)
699 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
701 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
702 if (!ocfs2_table_header) {
703 printk(KERN_ERR
704 "ocfs2 stack glue: unable to register sysctl\n");
705 return -ENOMEM; /* or something. */
708 return ocfs2_sysfs_init();
711 static void __exit ocfs2_stack_glue_exit(void)
713 memset(&locking_max_version, 0,
714 sizeof(struct ocfs2_protocol_version));
715 locking_max_version.pv_major = 0;
716 locking_max_version.pv_minor = 0;
717 ocfs2_sysfs_exit();
718 if (ocfs2_table_header)
719 unregister_sysctl_table(ocfs2_table_header);
722 MODULE_AUTHOR("Oracle");
723 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
724 MODULE_LICENSE("GPL");
725 module_init(ocfs2_stack_glue_init);
726 module_exit(ocfs2_stack_glue_exit);