1/* -*- mode: c; c-basic-offset: 8; -*-
2 * vim: noexpandtab sw=8 ts=8 sts=0:
3 *
4 * stackglue.c
5 *
6 * Code which implements an OCFS2 specific interface to underlying
7 * cluster stacks.
8 *
9 * Copyright (C) 2007, 2009 Oracle.  All rights reserved.
10 *
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.
14 *
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.
19 */
20
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>
30
31#include "ocfs2_fs.h"
32
33#include "stackglue.h"
34
35#define OCFS2_STACK_PLUGIN_O2CB		"o2cb"
36#define OCFS2_STACK_PLUGIN_USER		"user"
37#define OCFS2_MAX_HB_CTL_PATH		256
38
39static struct ocfs2_protocol_version locking_max_version;
40static DEFINE_SPINLOCK(ocfs2_stack_lock);
41static LIST_HEAD(ocfs2_stack_list);
42static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
43static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
44
45/*
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.
48 */
49static struct ocfs2_stack_plugin *active_stack;
50
51static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
52{
53	struct ocfs2_stack_plugin *p;
54
55	assert_spin_locked(&ocfs2_stack_lock);
56
57	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
58		if (!strcmp(p->sp_name, name))
59			return p;
60	}
61
62	return NULL;
63}
64
65static int ocfs2_stack_driver_request(const char *stack_name,
66				      const char *plugin_name)
67{
68	int rc;
69	struct ocfs2_stack_plugin *p;
70
71	spin_lock(&ocfs2_stack_lock);
72
73	/*
74	 * If the stack passed by the filesystem isn't the selected one,
75	 * we can't continue.
76	 */
77	if (strcmp(stack_name, cluster_stack_name)) {
78		rc = -EBUSY;
79		goto out;
80	}
81
82	if (active_stack) {
83		/*
84		 * If the active stack isn't the one we want, it cannot
85		 * be selected right now.
86		 */
87		if (!strcmp(active_stack->sp_name, plugin_name))
88			rc = 0;
89		else
90			rc = -EBUSY;
91		goto out;
92	}
93
94	p = ocfs2_stack_lookup(plugin_name);
95	if (!p || !try_module_get(p->sp_owner)) {
96		rc = -ENOENT;
97		goto out;
98	}
99
100	active_stack = p;
101	rc = 0;
102
103out:
104	/* If we found it, pin it */
105	if (!rc)
106		active_stack->sp_count++;
107
108	spin_unlock(&ocfs2_stack_lock);
109	return rc;
110}
111
112/*
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.
116 */
117static int ocfs2_stack_driver_get(const char *stack_name)
118{
119	int rc;
120	char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
121
122	/*
123	 * Classic stack does not pass in a stack name.  This is
124	 * compatible with older tools as well.
125	 */
126	if (!stack_name || !*stack_name)
127		stack_name = OCFS2_STACK_PLUGIN_O2CB;
128
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;
134	}
135
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;
139
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);
144	}
145
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");
153	}
154
155	return rc;
156}
157
158static void ocfs2_stack_driver_put(void)
159{
160	spin_lock(&ocfs2_stack_lock);
161	BUG_ON(active_stack == NULL);
162	BUG_ON(active_stack->sp_count == 0);
163
164	active_stack->sp_count--;
165	if (!active_stack->sp_count) {
166		module_put(active_stack->sp_owner);
167		active_stack = NULL;
168	}
169	spin_unlock(&ocfs2_stack_lock);
170}
171
172int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
173{
174	int rc;
175
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;
188	}
189	spin_unlock(&ocfs2_stack_lock);
190
191	return rc;
192}
193EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
194
195void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
196{
197	struct ocfs2_stack_plugin *p;
198
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);
211	}
212	spin_unlock(&ocfs2_stack_lock);
213}
214EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
215
216void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
217{
218	struct ocfs2_stack_plugin *p;
219
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);
224
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;
228		}
229	}
230	spin_unlock(&ocfs2_stack_lock);
231}
232EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
233
234
235/*
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.
240 */
241int 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)
247{
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);
254}
255EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
256
257int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
258		     struct ocfs2_dlm_lksb *lksb,
259		     u32 flags)
260{
261	BUG_ON(lksb->lksb_conn == NULL);
262
263	return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
264}
265EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
266
267int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
268{
269	return active_stack->sp_ops->lock_status(lksb);
270}
271EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
272
273int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
274{
275	return active_stack->sp_ops->lvb_valid(lksb);
276}
277EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
278
279void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
280{
281	return active_stack->sp_ops->lock_lvb(lksb);
282}
283EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
284
285void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
286{
287	active_stack->sp_ops->dump_lksb(lksb);
288}
289EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
290
291int ocfs2_stack_supports_plocks(void)
292{
293	return active_stack && active_stack->sp_ops->plock;
294}
295EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
296
297/*
298 * ocfs2_plock() can only be safely called if
299 * ocfs2_stack_supports_plocks() returned true
300 */
301int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
302		struct file *file, int cmd, struct file_lock *fl)
303{
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;
308}
309EXPORT_SYMBOL_GPL(ocfs2_plock);
310
311int 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)
319{
320	int rc = 0;
321	struct ocfs2_cluster_connection *new_conn;
322
323	BUG_ON(group == NULL);
324	BUG_ON(conn == NULL);
325	BUG_ON(recovery_handler == NULL);
326
327	if (grouplen > GROUP_NAME_MAX) {
328		rc = -EINVAL;
329		goto out;
330	}
331
332	if (memcmp(&lproto->lp_max_version, &locking_max_version,
333		   sizeof(struct ocfs2_protocol_version))) {
334		rc = -EINVAL;
335		goto out;
336	}
337
338	new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
339			   GFP_KERNEL);
340	if (!new_conn) {
341		rc = -ENOMEM;
342		goto out;
343	}
344
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;
349
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;
353
354	/* This will pin the stack driver if successful */
355	rc = ocfs2_stack_driver_get(stack_name);
356	if (rc)
357		goto out_free;
358
359	rc = active_stack->sp_ops->connect(new_conn);
360	if (rc) {
361		ocfs2_stack_driver_put();
362		goto out_free;
363	}
364
365	*conn = new_conn;
366
367out_free:
368	if (rc)
369		kfree(new_conn);
370
371out:
372	return rc;
373}
374EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
375
376/* The caller will ensure all nodes have the same cluster stack */
377int 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)
384{
385	char *stack_name = NULL;
386
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);
391}
392EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
393
394/* If hangup_pending is 0, the stack driver will be dropped */
395int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
396			     int hangup_pending)
397{
398	int ret;
399
400	BUG_ON(conn == NULL);
401
402	ret = active_stack->sp_ops->disconnect(conn);
403
404	if (!ret) {
405		kfree(conn);
406		if (!hangup_pending)
407			ocfs2_stack_driver_put();
408	}
409
410	return ret;
411}
412EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
413
414/*
415 * Leave the group for this filesystem.  This is executed by a userspace
416 * program (stored in ocfs2_hb_ctl_path).
417 */
418static void ocfs2_leave_group(const char *group)
419{
420	int ret;
421	char *argv[5], *envp[3];
422
423	argv[0] = ocfs2_hb_ctl_path;
424	argv[1] = "-K";
425	argv[2] = "-u";
426	argv[3] = (char *)group;
427	argv[4] = NULL;
428
429	/* minimal command environment taken from cpu_run_sbin_hotplug */
430	envp[0] = "HOME=/";
431	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
432	envp[2] = NULL;
433
434	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
435	if (ret < 0) {
436		printk(KERN_ERR
437		       "ocfs2: Error %d running user helper "
438		       "\"%s %s %s %s\"\n",
439		       ret, argv[0], argv[1], argv[2], argv[3]);
440	}
441}
442
443/*
444 * Hangup is a required post-umount.  ocfs2-tools software expects the
445 * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
446 * regardless of whether the DLM got started, so we can't do it
447 * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
448 * the actual work.
449 */
450void ocfs2_cluster_hangup(const char *group, int grouplen)
451{
452	BUG_ON(group == NULL);
453	BUG_ON(group[grouplen] != '\0');
454
455	ocfs2_leave_group(group);
456
457	/* cluster_disconnect() was called with hangup_pending==1 */
458	ocfs2_stack_driver_put();
459}
460EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
461
462int ocfs2_cluster_this_node(unsigned int *node)
463{
464	return active_stack->sp_ops->this_node(node);
465}
466EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
467
468
469/*
470 * Sysfs bits
471 */
472
473static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
474					       struct kobj_attribute *attr,
475					       char *buf)
476{
477	ssize_t ret = 0;
478
479	spin_lock(&ocfs2_stack_lock);
480	if (locking_max_version.pv_major)
481		ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
482			       locking_max_version.pv_major,
483			       locking_max_version.pv_minor);
484	spin_unlock(&ocfs2_stack_lock);
485
486	return ret;
487}
488
489static struct kobj_attribute ocfs2_attr_max_locking_protocol =
490	__ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
491	       ocfs2_max_locking_protocol_show, NULL);
492
493static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
494						 struct kobj_attribute *attr,
495						 char *buf)
496{
497	ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
498	struct ocfs2_stack_plugin *p;
499
500	spin_lock(&ocfs2_stack_lock);
501	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
502		ret = snprintf(buf, remain, "%s\n",
503			       p->sp_name);
504		if (ret < 0) {
505			total = ret;
506			break;
507		}
508		if (ret == remain) {
509			/* snprintf() didn't fit */
510			total = -E2BIG;
511			break;
512		}
513		total += ret;
514		remain -= ret;
515	}
516	spin_unlock(&ocfs2_stack_lock);
517
518	return total;
519}
520
521static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
522	__ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
523	       ocfs2_loaded_cluster_plugins_show, NULL);
524
525static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
526						struct kobj_attribute *attr,
527						char *buf)
528{
529	ssize_t ret = 0;
530
531	spin_lock(&ocfs2_stack_lock);
532	if (active_stack) {
533		ret = snprintf(buf, PAGE_SIZE, "%s\n",
534			       active_stack->sp_name);
535		if (ret == PAGE_SIZE)
536			ret = -E2BIG;
537	}
538	spin_unlock(&ocfs2_stack_lock);
539
540	return ret;
541}
542
543static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
544	__ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
545	       ocfs2_active_cluster_plugin_show, NULL);
546
547static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
548					struct kobj_attribute *attr,
549					char *buf)
550{
551	ssize_t ret;
552	spin_lock(&ocfs2_stack_lock);
553	ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
554	spin_unlock(&ocfs2_stack_lock);
555
556	return ret;
557}
558
559static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
560					 struct kobj_attribute *attr,
561					 const char *buf, size_t count)
562{
563	size_t len = count;
564	ssize_t ret;
565
566	if (len == 0)
567		return len;
568
569	if (buf[len - 1] == '\n')
570		len--;
571
572	if ((len != OCFS2_STACK_LABEL_LEN) ||
573	    (strnlen(buf, len) != len))
574		return -EINVAL;
575
576	spin_lock(&ocfs2_stack_lock);
577	if (active_stack) {
578		if (!strncmp(buf, cluster_stack_name, len))
579			ret = count;
580		else
581			ret = -EBUSY;
582	} else {
583		memcpy(cluster_stack_name, buf, len);
584		ret = count;
585	}
586	spin_unlock(&ocfs2_stack_lock);
587
588	return ret;
589}
590
591
592static struct kobj_attribute ocfs2_attr_cluster_stack =
593	__ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
594	       ocfs2_cluster_stack_show,
595	       ocfs2_cluster_stack_store);
596
597static struct attribute *ocfs2_attrs[] = {
598	&ocfs2_attr_max_locking_protocol.attr,
599	&ocfs2_attr_loaded_cluster_plugins.attr,
600	&ocfs2_attr_active_cluster_plugin.attr,
601	&ocfs2_attr_cluster_stack.attr,
602	NULL,
603};
604
605static struct attribute_group ocfs2_attr_group = {
606	.attrs = ocfs2_attrs,
607};
608
609static struct kset *ocfs2_kset;
610
611static void ocfs2_sysfs_exit(void)
612{
613	kset_unregister(ocfs2_kset);
614}
615
616static int ocfs2_sysfs_init(void)
617{
618	int ret;
619
620	ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
621	if (!ocfs2_kset)
622		return -ENOMEM;
623
624	ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
625	if (ret)
626		goto error;
627
628	return 0;
629
630error:
631	kset_unregister(ocfs2_kset);
632	return ret;
633}
634
635/*
636 * Sysctl bits
637 *
638 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
639 * make as much sense in a multiple cluster stack world, but it's safer
640 * and easier to preserve the name.
641 */
642
643#define FS_OCFS2_NM		1
644
645static ctl_table ocfs2_nm_table[] = {
646	{
647		.procname	= "hb_ctl_path",
648		.data		= ocfs2_hb_ctl_path,
649		.maxlen		= OCFS2_MAX_HB_CTL_PATH,
650		.mode		= 0644,
651		.proc_handler	= proc_dostring,
652	},
653	{ }
654};
655
656static ctl_table ocfs2_mod_table[] = {
657	{
658		.procname	= "nm",
659		.data		= NULL,
660		.maxlen		= 0,
661		.mode		= 0555,
662		.child		= ocfs2_nm_table
663	},
664	{ }
665};
666
667static ctl_table ocfs2_kern_table[] = {
668	{
669		.procname	= "ocfs2",
670		.data		= NULL,
671		.maxlen		= 0,
672		.mode		= 0555,
673		.child		= ocfs2_mod_table
674	},
675	{ }
676};
677
678static ctl_table ocfs2_root_table[] = {
679	{
680		.procname	= "fs",
681		.data		= NULL,
682		.maxlen		= 0,
683		.mode		= 0555,
684		.child		= ocfs2_kern_table
685	},
686	{ }
687};
688
689static struct ctl_table_header *ocfs2_table_header = NULL;
690
691
692/*
693 * Initialization
694 */
695
696static int __init ocfs2_stack_glue_init(void)
697{
698	strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
699
700	ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
701	if (!ocfs2_table_header) {
702		printk(KERN_ERR
703		       "ocfs2 stack glue: unable to register sysctl\n");
704		return -ENOMEM; /* or something. */
705	}
706
707	return ocfs2_sysfs_init();
708}
709
710static void __exit ocfs2_stack_glue_exit(void)
711{
712	memset(&locking_max_version, 0,
713	       sizeof(struct ocfs2_protocol_version));
714	locking_max_version.pv_major = 0;
715	locking_max_version.pv_minor = 0;
716	ocfs2_sysfs_exit();
717	if (ocfs2_table_header)
718		unregister_sysctl_table(ocfs2_table_header);
719}
720
721MODULE_AUTHOR("Oracle");
722MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
723MODULE_LICENSE("GPL");
724module_init(ocfs2_stack_glue_init);
725module_exit(ocfs2_stack_glue_exit);
726