kmemtrace: SLOB hooks.
[linux-2.6/kmemtrace.git] / fs / ocfs2 / stackglue.c
blob10e149ae5e3aad289f3070714ef5b8530078acf0
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 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_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))
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 /* Ok, the stack is pinned */
101 p->sp_count++;
102 active_stack = p;
104 rc = 0;
106 out:
107 spin_unlock(&ocfs2_stack_lock);
108 return rc;
112 * This function looks up the appropriate stack and makes it active. If
113 * there is no stack, it tries to load it. It will fail if the stack still
114 * cannot be found. It will also fail if a different stack is in use.
116 static int ocfs2_stack_driver_get(const char *stack_name)
118 int rc;
119 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
122 * Classic stack does not pass in a stack name. This is
123 * compatible with older tools as well.
125 if (!stack_name || !*stack_name)
126 stack_name = OCFS2_STACK_PLUGIN_O2CB;
128 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
129 printk(KERN_ERR
130 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
131 stack_name);
132 return -EINVAL;
135 /* Anything that isn't the classic stack is a user stack */
136 if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
137 plugin_name = OCFS2_STACK_PLUGIN_USER;
139 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
140 if (rc == -ENOENT) {
141 request_module("ocfs2_stack_%s", plugin_name);
142 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
145 if (rc == -ENOENT) {
146 printk(KERN_ERR
147 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
148 plugin_name);
149 } else if (rc == -EBUSY) {
150 printk(KERN_ERR
151 "ocfs2: A different cluster stack is in use\n");
154 return rc;
157 static void ocfs2_stack_driver_put(void)
159 spin_lock(&ocfs2_stack_lock);
160 BUG_ON(active_stack == NULL);
161 BUG_ON(active_stack->sp_count == 0);
163 active_stack->sp_count--;
164 if (!active_stack->sp_count) {
165 module_put(active_stack->sp_owner);
166 active_stack = NULL;
168 spin_unlock(&ocfs2_stack_lock);
171 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
173 int rc;
175 spin_lock(&ocfs2_stack_lock);
176 if (!ocfs2_stack_lookup(plugin->sp_name)) {
177 plugin->sp_count = 0;
178 plugin->sp_proto = lproto;
179 list_add(&plugin->sp_list, &ocfs2_stack_list);
180 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
181 plugin->sp_name);
182 rc = 0;
183 } else {
184 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
185 plugin->sp_name);
186 rc = -EEXIST;
188 spin_unlock(&ocfs2_stack_lock);
190 return rc;
192 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
194 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
196 struct ocfs2_stack_plugin *p;
198 spin_lock(&ocfs2_stack_lock);
199 p = ocfs2_stack_lookup(plugin->sp_name);
200 if (p) {
201 BUG_ON(p != plugin);
202 BUG_ON(plugin == active_stack);
203 BUG_ON(plugin->sp_count != 0);
204 list_del_init(&plugin->sp_list);
205 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
206 plugin->sp_name);
207 } else {
208 printk(KERN_ERR "Stack \"%s\" is not registered\n",
209 plugin->sp_name);
211 spin_unlock(&ocfs2_stack_lock);
213 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
215 void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
217 struct ocfs2_stack_plugin *p;
219 BUG_ON(proto == NULL);
221 spin_lock(&ocfs2_stack_lock);
222 BUG_ON(active_stack != NULL);
224 lproto = proto;
225 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
226 p->sp_proto = lproto;
229 spin_unlock(&ocfs2_stack_lock);
231 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_locking_protocol);
235 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take
236 * "struct ocfs2_lock_res *astarg" instead of "void *astarg" because the
237 * underlying stack plugins need to pilfer the lksb off of the lock_res.
238 * If some other structure needs to be passed as an astarg, the plugins
239 * will need to be given a different avenue to the lksb.
241 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
242 int mode,
243 union ocfs2_dlm_lksb *lksb,
244 u32 flags,
245 void *name,
246 unsigned int namelen,
247 struct ocfs2_lock_res *astarg)
249 BUG_ON(lproto == NULL);
251 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
252 name, namelen, astarg);
254 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
256 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
257 union ocfs2_dlm_lksb *lksb,
258 u32 flags,
259 struct ocfs2_lock_res *astarg)
261 BUG_ON(lproto == NULL);
263 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags, astarg);
265 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
267 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
269 return active_stack->sp_ops->lock_status(lksb);
271 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
274 * Why don't we cast to ocfs2_meta_lvb? The "clean" answer is that we
275 * don't cast at the glue level. The real answer is that the header
276 * ordering is nigh impossible.
278 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
280 return active_stack->sp_ops->lock_lvb(lksb);
282 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
284 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
286 active_stack->sp_ops->dump_lksb(lksb);
288 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
290 int ocfs2_cluster_connect(const char *stack_name,
291 const char *group,
292 int grouplen,
293 void (*recovery_handler)(int node_num,
294 void *recovery_data),
295 void *recovery_data,
296 struct ocfs2_cluster_connection **conn)
298 int rc = 0;
299 struct ocfs2_cluster_connection *new_conn;
301 BUG_ON(group == NULL);
302 BUG_ON(conn == NULL);
303 BUG_ON(recovery_handler == NULL);
305 if (grouplen > GROUP_NAME_MAX) {
306 rc = -EINVAL;
307 goto out;
310 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
311 GFP_KERNEL);
312 if (!new_conn) {
313 rc = -ENOMEM;
314 goto out;
317 memcpy(new_conn->cc_name, group, grouplen);
318 new_conn->cc_namelen = grouplen;
319 new_conn->cc_recovery_handler = recovery_handler;
320 new_conn->cc_recovery_data = recovery_data;
322 /* Start the new connection at our maximum compatibility level */
323 new_conn->cc_version = lproto->lp_max_version;
325 /* This will pin the stack driver if successful */
326 rc = ocfs2_stack_driver_get(stack_name);
327 if (rc)
328 goto out_free;
330 rc = active_stack->sp_ops->connect(new_conn);
331 if (rc) {
332 ocfs2_stack_driver_put();
333 goto out_free;
336 *conn = new_conn;
338 out_free:
339 if (rc)
340 kfree(new_conn);
342 out:
343 return rc;
345 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
347 /* If hangup_pending is 0, the stack driver will be dropped */
348 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
349 int hangup_pending)
351 int ret;
353 BUG_ON(conn == NULL);
355 ret = active_stack->sp_ops->disconnect(conn);
357 /* XXX Should we free it anyway? */
358 if (!ret) {
359 kfree(conn);
360 if (!hangup_pending)
361 ocfs2_stack_driver_put();
364 return ret;
366 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
369 * Leave the group for this filesystem. This is executed by a userspace
370 * program (stored in ocfs2_hb_ctl_path).
372 static void ocfs2_leave_group(const char *group)
374 int ret;
375 char *argv[5], *envp[3];
377 argv[0] = ocfs2_hb_ctl_path;
378 argv[1] = "-K";
379 argv[2] = "-u";
380 argv[3] = (char *)group;
381 argv[4] = NULL;
383 /* minimal command environment taken from cpu_run_sbin_hotplug */
384 envp[0] = "HOME=/";
385 envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
386 envp[2] = NULL;
388 ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
389 if (ret < 0) {
390 printk(KERN_ERR
391 "ocfs2: Error %d running user helper "
392 "\"%s %s %s %s\"\n",
393 ret, argv[0], argv[1], argv[2], argv[3]);
398 * Hangup is a required post-umount. ocfs2-tools software expects the
399 * filesystem to call "ocfs2_hb_ctl" during unmount. This happens
400 * regardless of whether the DLM got started, so we can't do it
401 * in ocfs2_cluster_disconnect(). The ocfs2_leave_group() function does
402 * the actual work.
404 void ocfs2_cluster_hangup(const char *group, int grouplen)
406 BUG_ON(group == NULL);
407 BUG_ON(group[grouplen] != '\0');
409 ocfs2_leave_group(group);
411 /* cluster_disconnect() was called with hangup_pending==1 */
412 ocfs2_stack_driver_put();
414 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
416 int ocfs2_cluster_this_node(unsigned int *node)
418 return active_stack->sp_ops->this_node(node);
420 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
424 * Sysfs bits
427 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
428 struct kobj_attribute *attr,
429 char *buf)
431 ssize_t ret = 0;
433 spin_lock(&ocfs2_stack_lock);
434 if (lproto)
435 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
436 lproto->lp_max_version.pv_major,
437 lproto->lp_max_version.pv_minor);
438 spin_unlock(&ocfs2_stack_lock);
440 return ret;
443 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
444 __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
445 ocfs2_max_locking_protocol_show, NULL);
447 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
448 struct kobj_attribute *attr,
449 char *buf)
451 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
452 struct ocfs2_stack_plugin *p;
454 spin_lock(&ocfs2_stack_lock);
455 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
456 ret = snprintf(buf, remain, "%s\n",
457 p->sp_name);
458 if (ret < 0) {
459 total = ret;
460 break;
462 if (ret == remain) {
463 /* snprintf() didn't fit */
464 total = -E2BIG;
465 break;
467 total += ret;
468 remain -= ret;
470 spin_unlock(&ocfs2_stack_lock);
472 return total;
475 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
476 __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
477 ocfs2_loaded_cluster_plugins_show, NULL);
479 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
480 struct kobj_attribute *attr,
481 char *buf)
483 ssize_t ret = 0;
485 spin_lock(&ocfs2_stack_lock);
486 if (active_stack) {
487 ret = snprintf(buf, PAGE_SIZE, "%s\n",
488 active_stack->sp_name);
489 if (ret == PAGE_SIZE)
490 ret = -E2BIG;
492 spin_unlock(&ocfs2_stack_lock);
494 return ret;
497 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
498 __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
499 ocfs2_active_cluster_plugin_show, NULL);
501 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
502 struct kobj_attribute *attr,
503 char *buf)
505 ssize_t ret;
506 spin_lock(&ocfs2_stack_lock);
507 ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
508 spin_unlock(&ocfs2_stack_lock);
510 return ret;
513 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
514 struct kobj_attribute *attr,
515 const char *buf, size_t count)
517 size_t len = count;
518 ssize_t ret;
520 if (len == 0)
521 return len;
523 if (buf[len - 1] == '\n')
524 len--;
526 if ((len != OCFS2_STACK_LABEL_LEN) ||
527 (strnlen(buf, len) != len))
528 return -EINVAL;
530 spin_lock(&ocfs2_stack_lock);
531 if (active_stack) {
532 if (!strncmp(buf, cluster_stack_name, len))
533 ret = count;
534 else
535 ret = -EBUSY;
536 } else {
537 memcpy(cluster_stack_name, buf, len);
538 ret = count;
540 spin_unlock(&ocfs2_stack_lock);
542 return ret;
546 static struct kobj_attribute ocfs2_attr_cluster_stack =
547 __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
548 ocfs2_cluster_stack_show,
549 ocfs2_cluster_stack_store);
551 static struct attribute *ocfs2_attrs[] = {
552 &ocfs2_attr_max_locking_protocol.attr,
553 &ocfs2_attr_loaded_cluster_plugins.attr,
554 &ocfs2_attr_active_cluster_plugin.attr,
555 &ocfs2_attr_cluster_stack.attr,
556 NULL,
559 static struct attribute_group ocfs2_attr_group = {
560 .attrs = ocfs2_attrs,
563 static struct kset *ocfs2_kset;
565 static void ocfs2_sysfs_exit(void)
567 kset_unregister(ocfs2_kset);
570 static int ocfs2_sysfs_init(void)
572 int ret;
574 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
575 if (!ocfs2_kset)
576 return -ENOMEM;
578 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
579 if (ret)
580 goto error;
582 return 0;
584 error:
585 kset_unregister(ocfs2_kset);
586 return ret;
590 * Sysctl bits
592 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
593 * make as much sense in a multiple cluster stack world, but it's safer
594 * and easier to preserve the name.
597 #define FS_OCFS2_NM 1
599 static ctl_table ocfs2_nm_table[] = {
601 .ctl_name = 1,
602 .procname = "hb_ctl_path",
603 .data = ocfs2_hb_ctl_path,
604 .maxlen = OCFS2_MAX_HB_CTL_PATH,
605 .mode = 0644,
606 .proc_handler = &proc_dostring,
607 .strategy = &sysctl_string,
609 { .ctl_name = 0 }
612 static ctl_table ocfs2_mod_table[] = {
614 .ctl_name = FS_OCFS2_NM,
615 .procname = "nm",
616 .data = NULL,
617 .maxlen = 0,
618 .mode = 0555,
619 .child = ocfs2_nm_table
621 { .ctl_name = 0}
624 static ctl_table ocfs2_kern_table[] = {
626 .ctl_name = FS_OCFS2,
627 .procname = "ocfs2",
628 .data = NULL,
629 .maxlen = 0,
630 .mode = 0555,
631 .child = ocfs2_mod_table
633 { .ctl_name = 0}
636 static ctl_table ocfs2_root_table[] = {
638 .ctl_name = CTL_FS,
639 .procname = "fs",
640 .data = NULL,
641 .maxlen = 0,
642 .mode = 0555,
643 .child = ocfs2_kern_table
645 { .ctl_name = 0 }
648 static struct ctl_table_header *ocfs2_table_header = NULL;
652 * Initialization
655 static int __init ocfs2_stack_glue_init(void)
657 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
659 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
660 if (!ocfs2_table_header) {
661 printk(KERN_ERR
662 "ocfs2 stack glue: unable to register sysctl\n");
663 return -ENOMEM; /* or something. */
666 return ocfs2_sysfs_init();
669 static void __exit ocfs2_stack_glue_exit(void)
671 lproto = NULL;
672 ocfs2_sysfs_exit();
673 if (ocfs2_table_header)
674 unregister_sysctl_table(ocfs2_table_header);
677 MODULE_AUTHOR("Oracle");
678 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
679 MODULE_LICENSE("GPL");
680 module_init(ocfs2_stack_glue_init);
681 module_exit(ocfs2_stack_glue_exit);