1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * stackglue.c
4 *
5 * Code which implements an OCFS2 specific interface to underlying
6 * cluster stacks.
7 *
8 * Copyright (C) 2007, 2009 Oracle.  All rights reserved.
9 */
10
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>
16#include <linux/fs.h>
17#include <linux/kobject.h>
18#include <linux/sysfs.h>
19#include <linux/sysctl.h>
20
21#include "ocfs2_fs.h"
22
23#include "stackglue.h"
24
25#define OCFS2_STACK_PLUGIN_O2CB		"o2cb"
26#define OCFS2_STACK_PLUGIN_USER		"user"
27#define OCFS2_MAX_HB_CTL_PATH		256
28
29static struct ocfs2_protocol_version locking_max_version;
30static DEFINE_SPINLOCK(ocfs2_stack_lock);
31static LIST_HEAD(ocfs2_stack_list);
32static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
33static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
34
35/*
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.
38 */
39static struct ocfs2_stack_plugin *active_stack;
40
41static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
42{
43	struct ocfs2_stack_plugin *p;
44
45	assert_spin_locked(&ocfs2_stack_lock);
46
47	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
48		if (!strcmp(p->sp_name, name))
49			return p;
50	}
51
52	return NULL;
53}
54
55static int ocfs2_stack_driver_request(const char *stack_name,
56				      const char *plugin_name)
57{
58	int rc;
59	struct ocfs2_stack_plugin *p;
60
61	spin_lock(&ocfs2_stack_lock);
62
63	/*
64	 * If the stack passed by the filesystem isn't the selected one,
65	 * we can't continue.
66	 */
67	if (strcmp(stack_name, cluster_stack_name)) {
68		rc = -EBUSY;
69		goto out;
70	}
71
72	if (active_stack) {
73		/*
74		 * If the active stack isn't the one we want, it cannot
75		 * be selected right now.
76		 */
77		if (!strcmp(active_stack->sp_name, plugin_name))
78			rc = 0;
79		else
80			rc = -EBUSY;
81		goto out;
82	}
83
84	p = ocfs2_stack_lookup(plugin_name);
85	if (!p || !try_module_get(p->sp_owner)) {
86		rc = -ENOENT;
87		goto out;
88	}
89
90	active_stack = p;
91	rc = 0;
92
93out:
94	/* If we found it, pin it */
95	if (!rc)
96		active_stack->sp_count++;
97
98	spin_unlock(&ocfs2_stack_lock);
99	return rc;
100}
101
102/*
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.
106 */
107static int ocfs2_stack_driver_get(const char *stack_name)
108{
109	int rc;
110	char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
111
112	/*
113	 * Classic stack does not pass in a stack name.  This is
114	 * compatible with older tools as well.
115	 */
116	if (!stack_name || !*stack_name)
117		stack_name = OCFS2_STACK_PLUGIN_O2CB;
118
119	if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
120		printk(KERN_ERR
121		       "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
122		       stack_name);
123		return -EINVAL;
124	}
125
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;
129
130	rc = ocfs2_stack_driver_request(stack_name, plugin_name);
131	if (rc == -ENOENT) {
132		request_module("ocfs2_stack_%s", plugin_name);
133		rc = ocfs2_stack_driver_request(stack_name, plugin_name);
134	}
135
136	if (rc == -ENOENT) {
137		printk(KERN_ERR
138		       "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
139		       plugin_name);
140	} else if (rc == -EBUSY) {
141		printk(KERN_ERR
142		       "ocfs2: A different cluster stack is in use\n");
143	}
144
145	return rc;
146}
147
148static void ocfs2_stack_driver_put(void)
149{
150	spin_lock(&ocfs2_stack_lock);
151	BUG_ON(active_stack == NULL);
152	BUG_ON(active_stack->sp_count == 0);
153
154	active_stack->sp_count--;
155	if (!active_stack->sp_count) {
156		module_put(active_stack->sp_owner);
157		active_stack = NULL;
158	}
159	spin_unlock(&ocfs2_stack_lock);
160}
161
162int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
163{
164	int rc;
165
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",
172		       plugin->sp_name);
173		rc = 0;
174	} else {
175		printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
176		       plugin->sp_name);
177		rc = -EEXIST;
178	}
179	spin_unlock(&ocfs2_stack_lock);
180
181	return rc;
182}
183EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
184
185void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
186{
187	struct ocfs2_stack_plugin *p;
188
189	spin_lock(&ocfs2_stack_lock);
190	p = ocfs2_stack_lookup(plugin->sp_name);
191	if (p) {
192		BUG_ON(p != plugin);
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",
197		       plugin->sp_name);
198	} else {
199		printk(KERN_ERR "Stack \"%s\" is not registered\n",
200		       plugin->sp_name);
201	}
202	spin_unlock(&ocfs2_stack_lock);
203}
204EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
205
206void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
207{
208	struct ocfs2_stack_plugin *p;
209
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);
214
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;
218		}
219	}
220	spin_unlock(&ocfs2_stack_lock);
221}
222EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
223
224
225/*
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.
230 */
231int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
232		   int mode,
233		   struct ocfs2_dlm_lksb *lksb,
234		   u32 flags,
235		   void *name,
236		   unsigned int namelen)
237{
238	if (!lksb->lksb_conn)
239		lksb->lksb_conn = conn;
240	else
241		BUG_ON(lksb->lksb_conn != conn);
242	return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
243					      name, namelen);
244}
245EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
246
247int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
248		     struct ocfs2_dlm_lksb *lksb,
249		     u32 flags)
250{
251	BUG_ON(lksb->lksb_conn == NULL);
252
253	return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
254}
255EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
256
257int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
258{
259	return active_stack->sp_ops->lock_status(lksb);
260}
261EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
262
263int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
264{
265	return active_stack->sp_ops->lvb_valid(lksb);
266}
267EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
268
269void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
270{
271	return active_stack->sp_ops->lock_lvb(lksb);
272}
273EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
274
275void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
276{
277	active_stack->sp_ops->dump_lksb(lksb);
278}
279EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
280
281int ocfs2_stack_supports_plocks(void)
282{
283	return active_stack && active_stack->sp_ops->plock;
284}
285EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
286
287/*
288 * ocfs2_plock() can only be safely called if
289 * ocfs2_stack_supports_plocks() returned true
290 */
291int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
292		struct file *file, int cmd, struct file_lock *fl)
293{
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);
297	return -EOPNOTSUPP;
298}
299EXPORT_SYMBOL_GPL(ocfs2_plock);
300
301int ocfs2_cluster_connect(const char *stack_name,
302			  const char *cluster_name,
303			  int cluster_name_len,
304			  const char *group,
305			  int grouplen,
306			  struct ocfs2_locking_protocol *lproto,
307			  void (*recovery_handler)(int node_num,
308						   void *recovery_data),
309			  void *recovery_data,
310			  struct ocfs2_cluster_connection **conn)
311{
312	int rc = 0;
313	struct ocfs2_cluster_connection *new_conn;
314
315	BUG_ON(group == NULL);
316	BUG_ON(conn == NULL);
317	BUG_ON(recovery_handler == NULL);
318
319	if (grouplen > GROUP_NAME_MAX) {
320		rc = -EINVAL;
321		goto out;
322	}
323
324	if (memcmp(&lproto->lp_max_version, &locking_max_version,
325		   sizeof(struct ocfs2_protocol_version))) {
326		rc = -EINVAL;
327		goto out;
328	}
329
330	new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
331			   GFP_KERNEL);
332	if (!new_conn) {
333		rc = -ENOMEM;
334		goto out;
335	}
336
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;
345
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;
349
350	/* This will pin the stack driver if successful */
351	rc = ocfs2_stack_driver_get(stack_name);
352	if (rc)
353		goto out_free;
354
355	rc = active_stack->sp_ops->connect(new_conn);
356	if (rc) {
357		ocfs2_stack_driver_put();
358		goto out_free;
359	}
360
361	*conn = new_conn;
362
363out_free:
364	if (rc)
365		kfree(new_conn);
366
367out:
368	return rc;
369}
370EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
371
372/* The caller will ensure all nodes have the same cluster stack */
373int ocfs2_cluster_connect_agnostic(const char *group,
374				   int grouplen,
375				   struct ocfs2_locking_protocol *lproto,
376				   void (*recovery_handler)(int node_num,
377							    void *recovery_data),
378				   void *recovery_data,
379				   struct ocfs2_cluster_connection **conn)
380{
381	char *stack_name = NULL;
382
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,
387				     conn);
388}
389EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
390
391/* If hangup_pending is 0, the stack driver will be dropped */
392int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
393			     int hangup_pending)
394{
395	int ret;
396
397	BUG_ON(conn == NULL);
398
399	ret = active_stack->sp_ops->disconnect(conn);
400
401	/* XXX Should we free it anyway? */
402	if (!ret) {
403		kfree(conn);
404		if (!hangup_pending)
405			ocfs2_stack_driver_put();
406	}
407
408	return ret;
409}
410EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
411
412/*
413 * Leave the group for this filesystem.  This is executed by a userspace
414 * program (stored in ocfs2_hb_ctl_path).
415 */
416static void ocfs2_leave_group(const char *group)
417{
418	int ret;
419	char *argv[5], *envp[3];
420
421	argv[0] = ocfs2_hb_ctl_path;
422	argv[1] = "-K";
423	argv[2] = "-u";
424	argv[3] = (char *)group;
425	argv[4] = NULL;
426
427	/* minimal command environment taken from cpu_run_sbin_hotplug */
428	envp[0] = "HOME=/";
429	envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
430	envp[2] = NULL;
431
432	ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
433	if (ret < 0) {
434		printk(KERN_ERR
435		       "ocfs2: Error %d running user helper "
436		       "\"%s %s %s %s\"\n",
437		       ret, argv[0], argv[1], argv[2], argv[3]);
438	}
439}
440
441/*
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
446 * the actual work.
447 */
448void ocfs2_cluster_hangup(const char *group, int grouplen)
449{
450	BUG_ON(group == NULL);
451	BUG_ON(group[grouplen] != '\0');
452
453	ocfs2_leave_group(group);
454
455	/* cluster_disconnect() was called with hangup_pending==1 */
456	ocfs2_stack_driver_put();
457}
458EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
459
460int ocfs2_cluster_this_node(struct ocfs2_cluster_connection *conn,
461			    unsigned int *node)
462{
463	return active_stack->sp_ops->this_node(conn, node);
464}
465EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
466
467
468/*
469 * Sysfs bits
470 */
471
472static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
473					       struct kobj_attribute *attr,
474					       char *buf)
475{
476	ssize_t ret = 0;
477
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);
484
485	return ret;
486}
487
488static struct kobj_attribute ocfs2_attr_max_locking_protocol =
489	__ATTR(max_locking_protocol, S_IRUGO,
490	       ocfs2_max_locking_protocol_show, NULL);
491
492static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
493						 struct kobj_attribute *attr,
494						 char *buf)
495{
496	ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
497	struct ocfs2_stack_plugin *p;
498
499	spin_lock(&ocfs2_stack_lock);
500	list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
501		ret = snprintf(buf, remain, "%s\n",
502			       p->sp_name);
503		if (ret >= remain) {
504			/* snprintf() didn't fit */
505			total = -E2BIG;
506			break;
507		}
508		total += ret;
509		remain -= ret;
510	}
511	spin_unlock(&ocfs2_stack_lock);
512
513	return total;
514}
515
516static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
517	__ATTR(loaded_cluster_plugins, S_IRUGO,
518	       ocfs2_loaded_cluster_plugins_show, NULL);
519
520static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
521						struct kobj_attribute *attr,
522						char *buf)
523{
524	ssize_t ret = 0;
525
526	spin_lock(&ocfs2_stack_lock);
527	if (active_stack) {
528		ret = snprintf(buf, PAGE_SIZE, "%s\n",
529			       active_stack->sp_name);
530		if (ret >= PAGE_SIZE)
531			ret = -E2BIG;
532	}
533	spin_unlock(&ocfs2_stack_lock);
534
535	return ret;
536}
537
538static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
539	__ATTR(active_cluster_plugin, S_IRUGO,
540	       ocfs2_active_cluster_plugin_show, NULL);
541
542static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
543					struct kobj_attribute *attr,
544					char *buf)
545{
546	ssize_t ret;
547	spin_lock(&ocfs2_stack_lock);
548	ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
549	spin_unlock(&ocfs2_stack_lock);
550
551	return ret;
552}
553
554static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
555					 struct kobj_attribute *attr,
556					 const char *buf, size_t count)
557{
558	size_t len = count;
559	ssize_t ret;
560
561	if (len == 0)
562		return len;
563
564	if (buf[len - 1] == '\n')
565		len--;
566
567	if ((len != OCFS2_STACK_LABEL_LEN) ||
568	    (strnlen(buf, len) != len))
569		return -EINVAL;
570
571	spin_lock(&ocfs2_stack_lock);
572	if (active_stack) {
573		if (!strncmp(buf, cluster_stack_name, len))
574			ret = count;
575		else
576			ret = -EBUSY;
577	} else {
578		memcpy(cluster_stack_name, buf, len);
579		ret = count;
580	}
581	spin_unlock(&ocfs2_stack_lock);
582
583	return ret;
584}
585
586
587static 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);
591
592
593
594static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
595					struct kobj_attribute *attr,
596					char *buf)
597{
598	return snprintf(buf, PAGE_SIZE, "1\n");
599}
600
601static struct kobj_attribute ocfs2_attr_dlm_recover_support =
602	__ATTR(dlm_recover_callback_support, S_IRUGO,
603	       ocfs2_dlm_recover_show, NULL);
604
605static 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,
611	NULL,
612};
613
614static const struct attribute_group ocfs2_attr_group = {
615	.attrs = ocfs2_attrs,
616};
617
618struct kset *ocfs2_kset;
619EXPORT_SYMBOL_GPL(ocfs2_kset);
620
621static void ocfs2_sysfs_exit(void)
622{
623	kset_unregister(ocfs2_kset);
624}
625
626static int ocfs2_sysfs_init(void)
627{
628	int ret;
629
630	ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
631	if (!ocfs2_kset)
632		return -ENOMEM;
633
634	ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
635	if (ret)
636		goto error;
637
638	return 0;
639
640error:
641	kset_unregister(ocfs2_kset);
642	return ret;
643}
644
645/*
646 * Sysctl bits
647 *
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.
651 */
652
653static struct ctl_table ocfs2_nm_table[] = {
654	{
655		.procname	= "hb_ctl_path",
656		.data		= ocfs2_hb_ctl_path,
657		.maxlen		= OCFS2_MAX_HB_CTL_PATH,
658		.mode		= 0644,
659		.proc_handler	= proc_dostring,
660	},
661};
662
663static struct ctl_table_header *ocfs2_table_header;
664
665/*
666 * Initialization
667 */
668
669static int __init ocfs2_stack_glue_init(void)
670{
671	int ret;
672
673	strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
674
675	ocfs2_table_header = register_sysctl("fs/ocfs2/nm", ocfs2_nm_table);
676	if (!ocfs2_table_header) {
677		printk(KERN_ERR
678		       "ocfs2 stack glue: unable to register sysctl\n");
679		return -ENOMEM; /* or something. */
680	}
681
682	ret = ocfs2_sysfs_init();
683	if (ret)
684		unregister_sysctl_table(ocfs2_table_header);
685
686	return ret;
687}
688
689static void __exit ocfs2_stack_glue_exit(void)
690{
691	memset(&locking_max_version, 0,
692	       sizeof(struct ocfs2_protocol_version));
693	ocfs2_sysfs_exit();
694	if (ocfs2_table_header)
695		unregister_sysctl_table(ocfs2_table_header);
696}
697
698MODULE_AUTHOR("Oracle");
699MODULE_DESCRIPTION("ocfs2 cluster stack glue layer");
700MODULE_LICENSE("GPL");
701module_init(ocfs2_stack_glue_init);
702module_exit(ocfs2_stack_glue_exit);
703