Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/* -*- mode: c; c-basic-offset: 8; -*-
3 * vim: noexpandtab sw=8 ts=8 sts=0:
4 *
5 * stackglue.c
6 *
7 * Code which implements an OCFS2 specific interface to underlying
8 * cluster stacks.
9 *
10 * Copyright (C) 2007, 2009 Oracle. All rights reserved.
11 */
12
13#include <linux/list.h>
14#include <linux/spinlock.h>
15#include <linux/module.h>
16#include <linux/slab.h>
17#include <linux/kmod.h>
18#include <linux/fs.h>
19#include <linux/kobject.h>
20#include <linux/sysfs.h>
21#include <linux/sysctl.h>
22
23#include "ocfs2_fs.h"
24
25#include "stackglue.h"
26
27#define OCFS2_STACK_PLUGIN_O2CB "o2cb"
28#define OCFS2_STACK_PLUGIN_USER "user"
29#define OCFS2_MAX_HB_CTL_PATH 256
30
31static struct ocfs2_protocol_version locking_max_version;
32static DEFINE_SPINLOCK(ocfs2_stack_lock);
33static LIST_HEAD(ocfs2_stack_list);
34static char cluster_stack_name[OCFS2_STACK_LABEL_LEN + 1];
35static char ocfs2_hb_ctl_path[OCFS2_MAX_HB_CTL_PATH] = "/sbin/ocfs2_hb_ctl";
36
37/*
38 * The stack currently in use. If not null, active_stack->sp_count > 0,
39 * the module is pinned, and the locking protocol cannot be changed.
40 */
41static struct ocfs2_stack_plugin *active_stack;
42
43static struct ocfs2_stack_plugin *ocfs2_stack_lookup(const char *name)
44{
45 struct ocfs2_stack_plugin *p;
46
47 assert_spin_locked(&ocfs2_stack_lock);
48
49 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
50 if (!strcmp(p->sp_name, name))
51 return p;
52 }
53
54 return NULL;
55}
56
57static int ocfs2_stack_driver_request(const char *stack_name,
58 const char *plugin_name)
59{
60 int rc;
61 struct ocfs2_stack_plugin *p;
62
63 spin_lock(&ocfs2_stack_lock);
64
65 /*
66 * If the stack passed by the filesystem isn't the selected one,
67 * we can't continue.
68 */
69 if (strcmp(stack_name, cluster_stack_name)) {
70 rc = -EBUSY;
71 goto out;
72 }
73
74 if (active_stack) {
75 /*
76 * If the active stack isn't the one we want, it cannot
77 * be selected right now.
78 */
79 if (!strcmp(active_stack->sp_name, plugin_name))
80 rc = 0;
81 else
82 rc = -EBUSY;
83 goto out;
84 }
85
86 p = ocfs2_stack_lookup(plugin_name);
87 if (!p || !try_module_get(p->sp_owner)) {
88 rc = -ENOENT;
89 goto out;
90 }
91
92 active_stack = p;
93 rc = 0;
94
95out:
96 /* If we found it, pin it */
97 if (!rc)
98 active_stack->sp_count++;
99
100 spin_unlock(&ocfs2_stack_lock);
101 return rc;
102}
103
104/*
105 * This function looks up the appropriate stack and makes it active. If
106 * there is no stack, it tries to load it. It will fail if the stack still
107 * cannot be found. It will also fail if a different stack is in use.
108 */
109static int ocfs2_stack_driver_get(const char *stack_name)
110{
111 int rc;
112 char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
113
114 /*
115 * Classic stack does not pass in a stack name. This is
116 * compatible with older tools as well.
117 */
118 if (!stack_name || !*stack_name)
119 stack_name = OCFS2_STACK_PLUGIN_O2CB;
120
121 if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
122 printk(KERN_ERR
123 "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
124 stack_name);
125 return -EINVAL;
126 }
127
128 /* Anything that isn't the classic stack is a user stack */
129 if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
130 plugin_name = OCFS2_STACK_PLUGIN_USER;
131
132 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
133 if (rc == -ENOENT) {
134 request_module("ocfs2_stack_%s", plugin_name);
135 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
136 }
137
138 if (rc == -ENOENT) {
139 printk(KERN_ERR
140 "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
141 plugin_name);
142 } else if (rc == -EBUSY) {
143 printk(KERN_ERR
144 "ocfs2: A different cluster stack is in use\n");
145 }
146
147 return rc;
148}
149
150static void ocfs2_stack_driver_put(void)
151{
152 spin_lock(&ocfs2_stack_lock);
153 BUG_ON(active_stack == NULL);
154 BUG_ON(active_stack->sp_count == 0);
155
156 active_stack->sp_count--;
157 if (!active_stack->sp_count) {
158 module_put(active_stack->sp_owner);
159 active_stack = NULL;
160 }
161 spin_unlock(&ocfs2_stack_lock);
162}
163
164int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
165{
166 int rc;
167
168 spin_lock(&ocfs2_stack_lock);
169 if (!ocfs2_stack_lookup(plugin->sp_name)) {
170 plugin->sp_count = 0;
171 plugin->sp_max_proto = locking_max_version;
172 list_add(&plugin->sp_list, &ocfs2_stack_list);
173 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
174 plugin->sp_name);
175 rc = 0;
176 } else {
177 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
178 plugin->sp_name);
179 rc = -EEXIST;
180 }
181 spin_unlock(&ocfs2_stack_lock);
182
183 return rc;
184}
185EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
186
187void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
188{
189 struct ocfs2_stack_plugin *p;
190
191 spin_lock(&ocfs2_stack_lock);
192 p = ocfs2_stack_lookup(plugin->sp_name);
193 if (p) {
194 BUG_ON(p != plugin);
195 BUG_ON(plugin == active_stack);
196 BUG_ON(plugin->sp_count != 0);
197 list_del_init(&plugin->sp_list);
198 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
199 plugin->sp_name);
200 } else {
201 printk(KERN_ERR "Stack \"%s\" is not registered\n",
202 plugin->sp_name);
203 }
204 spin_unlock(&ocfs2_stack_lock);
205}
206EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
207
208void ocfs2_stack_glue_set_max_proto_version(struct ocfs2_protocol_version *max_proto)
209{
210 struct ocfs2_stack_plugin *p;
211
212 spin_lock(&ocfs2_stack_lock);
213 if (memcmp(max_proto, &locking_max_version,
214 sizeof(struct ocfs2_protocol_version))) {
215 BUG_ON(locking_max_version.pv_major != 0);
216
217 locking_max_version = *max_proto;
218 list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
219 p->sp_max_proto = locking_max_version;
220 }
221 }
222 spin_unlock(&ocfs2_stack_lock);
223}
224EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_max_proto_version);
225
226
227/*
228 * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take no argument
229 * for the ast and bast functions. They will pass the lksb to the ast
230 * and bast. The caller can wrap the lksb with their own structure to
231 * get more information.
232 */
233int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
234 int mode,
235 struct ocfs2_dlm_lksb *lksb,
236 u32 flags,
237 void *name,
238 unsigned int namelen)
239{
240 if (!lksb->lksb_conn)
241 lksb->lksb_conn = conn;
242 else
243 BUG_ON(lksb->lksb_conn != conn);
244 return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
245 name, namelen);
246}
247EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
248
249int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
250 struct ocfs2_dlm_lksb *lksb,
251 u32 flags)
252{
253 BUG_ON(lksb->lksb_conn == NULL);
254
255 return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
256}
257EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
258
259int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
260{
261 return active_stack->sp_ops->lock_status(lksb);
262}
263EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
264
265int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
266{
267 return active_stack->sp_ops->lvb_valid(lksb);
268}
269EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
270
271void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
272{
273 return active_stack->sp_ops->lock_lvb(lksb);
274}
275EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
276
277void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
278{
279 active_stack->sp_ops->dump_lksb(lksb);
280}
281EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
282
283int ocfs2_stack_supports_plocks(void)
284{
285 return active_stack && active_stack->sp_ops->plock;
286}
287EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
288
289/*
290 * ocfs2_plock() can only be safely called if
291 * ocfs2_stack_supports_plocks() returned true
292 */
293int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
294 struct file *file, int cmd, struct file_lock *fl)
295{
296 WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
297 if (active_stack->sp_ops->plock)
298 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
299 return -EOPNOTSUPP;
300}
301EXPORT_SYMBOL_GPL(ocfs2_plock);
302
303int ocfs2_cluster_connect(const char *stack_name,
304 const char *cluster_name,
305 int cluster_name_len,
306 const char *group,
307 int grouplen,
308 struct ocfs2_locking_protocol *lproto,
309 void (*recovery_handler)(int node_num,
310 void *recovery_data),
311 void *recovery_data,
312 struct ocfs2_cluster_connection **conn)
313{
314 int rc = 0;
315 struct ocfs2_cluster_connection *new_conn;
316
317 BUG_ON(group == NULL);
318 BUG_ON(conn == NULL);
319 BUG_ON(recovery_handler == NULL);
320
321 if (grouplen > GROUP_NAME_MAX) {
322 rc = -EINVAL;
323 goto out;
324 }
325
326 if (memcmp(&lproto->lp_max_version, &locking_max_version,
327 sizeof(struct ocfs2_protocol_version))) {
328 rc = -EINVAL;
329 goto out;
330 }
331
332 new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
333 GFP_KERNEL);
334 if (!new_conn) {
335 rc = -ENOMEM;
336 goto out;
337 }
338
339 strlcpy(new_conn->cc_name, group, GROUP_NAME_MAX + 1);
340 new_conn->cc_namelen = grouplen;
341 if (cluster_name_len)
342 strlcpy(new_conn->cc_cluster_name, cluster_name,
343 CLUSTER_NAME_MAX + 1);
344 new_conn->cc_cluster_name_len = cluster_name_len;
345 new_conn->cc_recovery_handler = recovery_handler;
346 new_conn->cc_recovery_data = recovery_data;
347
348 new_conn->cc_proto = lproto;
349 /* Start the new connection at our maximum compatibility level */
350 new_conn->cc_version = lproto->lp_max_version;
351
352 /* This will pin the stack driver if successful */
353 rc = ocfs2_stack_driver_get(stack_name);
354 if (rc)
355 goto out_free;
356
357 rc = active_stack->sp_ops->connect(new_conn);
358 if (rc) {
359 ocfs2_stack_driver_put();
360 goto out_free;
361 }
362
363 *conn = new_conn;
364
365out_free:
366 if (rc)
367 kfree(new_conn);
368
369out:
370 return rc;
371}
372EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
373
374/* The caller will ensure all nodes have the same cluster stack */
375int ocfs2_cluster_connect_agnostic(const char *group,
376 int grouplen,
377 struct ocfs2_locking_protocol *lproto,
378 void (*recovery_handler)(int node_num,
379 void *recovery_data),
380 void *recovery_data,
381 struct ocfs2_cluster_connection **conn)
382{
383 char *stack_name = NULL;
384
385 if (cluster_stack_name[0])
386 stack_name = cluster_stack_name;
387 return ocfs2_cluster_connect(stack_name, NULL, 0, group, grouplen,
388 lproto, recovery_handler, recovery_data,
389 conn);
390}
391EXPORT_SYMBOL_GPL(ocfs2_cluster_connect_agnostic);
392
393/* If hangup_pending is 0, the stack driver will be dropped */
394int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
395 int hangup_pending)
396{
397 int ret;
398
399 BUG_ON(conn == NULL);
400
401 ret = active_stack->sp_ops->disconnect(conn);
402
403 /* XXX Should we free it anyway? */
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(struct ocfs2_cluster_connection *conn,
463 unsigned int *node)
464{
465 return active_stack->sp_ops->this_node(conn, node);
466}
467EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
468
469
470/*
471 * Sysfs bits
472 */
473
474static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
475 struct kobj_attribute *attr,
476 char *buf)
477{
478 ssize_t ret = 0;
479
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);
486
487 return ret;
488}
489
490static struct kobj_attribute ocfs2_attr_max_locking_protocol =
491 __ATTR(max_locking_protocol, S_IRUGO,
492 ocfs2_max_locking_protocol_show, NULL);
493
494static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
495 struct kobj_attribute *attr,
496 char *buf)
497{
498 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
499 struct ocfs2_stack_plugin *p;
500
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;
508 }
509 if (ret == remain) {
510 /* snprintf() didn't fit */
511 total = -E2BIG;
512 break;
513 }
514 total += ret;
515 remain -= ret;
516 }
517 spin_unlock(&ocfs2_stack_lock);
518
519 return total;
520}
521
522static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
523 __ATTR(loaded_cluster_plugins, S_IRUGO,
524 ocfs2_loaded_cluster_plugins_show, NULL);
525
526static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
527 struct kobj_attribute *attr,
528 char *buf)
529{
530 ssize_t ret = 0;
531
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;
538 }
539 spin_unlock(&ocfs2_stack_lock);
540
541 return ret;
542}
543
544static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
545 __ATTR(active_cluster_plugin, S_IRUGO,
546 ocfs2_active_cluster_plugin_show, NULL);
547
548static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
549 struct kobj_attribute *attr,
550 char *buf)
551{
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);
556
557 return ret;
558}
559
560static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
561 struct kobj_attribute *attr,
562 const char *buf, size_t count)
563{
564 size_t len = count;
565 ssize_t ret;
566
567 if (len == 0)
568 return len;
569
570 if (buf[len - 1] == '\n')
571 len--;
572
573 if ((len != OCFS2_STACK_LABEL_LEN) ||
574 (strnlen(buf, len) != len))
575 return -EINVAL;
576
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;
586 }
587 spin_unlock(&ocfs2_stack_lock);
588
589 return ret;
590}
591
592
593static struct kobj_attribute ocfs2_attr_cluster_stack =
594 __ATTR(cluster_stack, S_IRUGO | S_IWUSR,
595 ocfs2_cluster_stack_show,
596 ocfs2_cluster_stack_store);
597
598
599
600static ssize_t ocfs2_dlm_recover_show(struct kobject *kobj,
601 struct kobj_attribute *attr,
602 char *buf)
603{
604 return snprintf(buf, PAGE_SIZE, "1\n");
605}
606
607static struct kobj_attribute ocfs2_attr_dlm_recover_support =
608 __ATTR(dlm_recover_callback_support, S_IRUGO,
609 ocfs2_dlm_recover_show, NULL);
610
611static struct attribute *ocfs2_attrs[] = {
612 &ocfs2_attr_max_locking_protocol.attr,
613 &ocfs2_attr_loaded_cluster_plugins.attr,
614 &ocfs2_attr_active_cluster_plugin.attr,
615 &ocfs2_attr_cluster_stack.attr,
616 &ocfs2_attr_dlm_recover_support.attr,
617 NULL,
618};
619
620static const struct attribute_group ocfs2_attr_group = {
621 .attrs = ocfs2_attrs,
622};
623
624struct kset *ocfs2_kset;
625EXPORT_SYMBOL_GPL(ocfs2_kset);
626
627static void ocfs2_sysfs_exit(void)
628{
629 kset_unregister(ocfs2_kset);
630}
631
632static int ocfs2_sysfs_init(void)
633{
634 int ret;
635
636 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
637 if (!ocfs2_kset)
638 return -ENOMEM;
639
640 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
641 if (ret)
642 goto error;
643
644 return 0;
645
646error:
647 kset_unregister(ocfs2_kset);
648 return ret;
649}
650
651/*
652 * Sysctl bits
653 *
654 * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path. The 'nm' doesn't
655 * make as much sense in a multiple cluster stack world, but it's safer
656 * and easier to preserve the name.
657 */
658
659#define FS_OCFS2_NM 1
660
661static struct ctl_table ocfs2_nm_table[] = {
662 {
663 .procname = "hb_ctl_path",
664 .data = ocfs2_hb_ctl_path,
665 .maxlen = OCFS2_MAX_HB_CTL_PATH,
666 .mode = 0644,
667 .proc_handler = proc_dostring,
668 },
669 { }
670};
671
672static struct ctl_table ocfs2_mod_table[] = {
673 {
674 .procname = "nm",
675 .data = NULL,
676 .maxlen = 0,
677 .mode = 0555,
678 .child = ocfs2_nm_table
679 },
680 { }
681};
682
683static struct ctl_table ocfs2_kern_table[] = {
684 {
685 .procname = "ocfs2",
686 .data = NULL,
687 .maxlen = 0,
688 .mode = 0555,
689 .child = ocfs2_mod_table
690 },
691 { }
692};
693
694static struct ctl_table ocfs2_root_table[] = {
695 {
696 .procname = "fs",
697 .data = NULL,
698 .maxlen = 0,
699 .mode = 0555,
700 .child = ocfs2_kern_table
701 },
702 { }
703};
704
705static struct ctl_table_header *ocfs2_table_header;
706
707
708/*
709 * Initialization
710 */
711
712static int __init ocfs2_stack_glue_init(void)
713{
714 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
715
716 ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
717 if (!ocfs2_table_header) {
718 printk(KERN_ERR
719 "ocfs2 stack glue: unable to register sysctl\n");
720 return -ENOMEM; /* or something. */
721 }
722
723 return ocfs2_sysfs_init();
724}
725
726static void __exit ocfs2_stack_glue_exit(void)
727{
728 memset(&locking_max_version, 0,
729 sizeof(struct ocfs2_protocol_version));
730 ocfs2_sysfs_exit();
731 if (ocfs2_table_header)
732 unregister_sysctl_table(ocfs2_table_header);
733}
734
735MODULE_AUTHOR("Oracle");
736MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
737MODULE_LICENSE("GPL");
738module_init(ocfs2_stack_glue_init);
739module_exit(ocfs2_stack_glue_exit);
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 /* XXX Should we free it anyway? */
405 if (!ret) {
406 kfree(conn);
407 if (!hangup_pending)
408 ocfs2_stack_driver_put();
409 }
410
411 return ret;
412}
413EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
414
415/*
416 * Leave the group for this filesystem. This is executed by a userspace
417 * program (stored in ocfs2_hb_ctl_path).
418 */
419static void ocfs2_leave_group(const char *group)
420{
421 int ret;
422 char *argv[5], *envp[3];
423
424 argv[0] = ocfs2_hb_ctl_path;
425 argv[1] = "-K";
426 argv[2] = "-u";
427 argv[3] = (char *)group;
428 argv[4] = NULL;
429
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;
434
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]);
441 }
442}
443
444/*
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.
450 */
451void ocfs2_cluster_hangup(const char *group, int grouplen)
452{
453 BUG_ON(group == NULL);
454 BUG_ON(group[grouplen] != '\0');
455
456 ocfs2_leave_group(group);
457
458 /* cluster_disconnect() was called with hangup_pending==1 */
459 ocfs2_stack_driver_put();
460}
461EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
462
463int ocfs2_cluster_this_node(unsigned int *node)
464{
465 return active_stack->sp_ops->this_node(node);
466}
467EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
468
469
470/*
471 * Sysfs bits
472 */
473
474static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
475 struct kobj_attribute *attr,
476 char *buf)
477{
478 ssize_t ret = 0;
479
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);
486
487 return ret;
488}
489
490static struct kobj_attribute ocfs2_attr_max_locking_protocol =
491 __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
492 ocfs2_max_locking_protocol_show, NULL);
493
494static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
495 struct kobj_attribute *attr,
496 char *buf)
497{
498 ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
499 struct ocfs2_stack_plugin *p;
500
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;
508 }
509 if (ret == remain) {
510 /* snprintf() didn't fit */
511 total = -E2BIG;
512 break;
513 }
514 total += ret;
515 remain -= ret;
516 }
517 spin_unlock(&ocfs2_stack_lock);
518
519 return total;
520}
521
522static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
523 __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
524 ocfs2_loaded_cluster_plugins_show, NULL);
525
526static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
527 struct kobj_attribute *attr,
528 char *buf)
529{
530 ssize_t ret = 0;
531
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;
538 }
539 spin_unlock(&ocfs2_stack_lock);
540
541 return ret;
542}
543
544static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
545 __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
546 ocfs2_active_cluster_plugin_show, NULL);
547
548static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
549 struct kobj_attribute *attr,
550 char *buf)
551{
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);
556
557 return ret;
558}
559
560static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
561 struct kobj_attribute *attr,
562 const char *buf, size_t count)
563{
564 size_t len = count;
565 ssize_t ret;
566
567 if (len == 0)
568 return len;
569
570 if (buf[len - 1] == '\n')
571 len--;
572
573 if ((len != OCFS2_STACK_LABEL_LEN) ||
574 (strnlen(buf, len) != len))
575 return -EINVAL;
576
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;
586 }
587 spin_unlock(&ocfs2_stack_lock);
588
589 return ret;
590}
591
592
593static 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);
597
598static 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,
604};
605
606static struct attribute_group ocfs2_attr_group = {
607 .attrs = ocfs2_attrs,
608};
609
610static struct kset *ocfs2_kset;
611
612static void ocfs2_sysfs_exit(void)
613{
614 kset_unregister(ocfs2_kset);
615}
616
617static int ocfs2_sysfs_init(void)
618{
619 int ret;
620
621 ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
622 if (!ocfs2_kset)
623 return -ENOMEM;
624
625 ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
626 if (ret)
627 goto error;
628
629 return 0;
630
631error:
632 kset_unregister(ocfs2_kset);
633 return ret;
634}
635
636/*
637 * Sysctl bits
638 *
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.
642 */
643
644#define FS_OCFS2_NM 1
645
646static ctl_table ocfs2_nm_table[] = {
647 {
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,
653 },
654 { }
655};
656
657static ctl_table ocfs2_mod_table[] = {
658 {
659 .procname = "nm",
660 .data = NULL,
661 .maxlen = 0,
662 .mode = 0555,
663 .child = ocfs2_nm_table
664 },
665 { }
666};
667
668static ctl_table ocfs2_kern_table[] = {
669 {
670 .procname = "ocfs2",
671 .data = NULL,
672 .maxlen = 0,
673 .mode = 0555,
674 .child = ocfs2_mod_table
675 },
676 { }
677};
678
679static ctl_table ocfs2_root_table[] = {
680 {
681 .procname = "fs",
682 .data = NULL,
683 .maxlen = 0,
684 .mode = 0555,
685 .child = ocfs2_kern_table
686 },
687 { }
688};
689
690static struct ctl_table_header *ocfs2_table_header = NULL;
691
692
693/*
694 * Initialization
695 */
696
697static int __init ocfs2_stack_glue_init(void)
698{
699 strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
700
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. */
706 }
707
708 return ocfs2_sysfs_init();
709}
710
711static void __exit ocfs2_stack_glue_exit(void)
712{
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);
720}
721
722MODULE_AUTHOR("Oracle");
723MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
724MODULE_LICENSE("GPL");
725module_init(ocfs2_stack_glue_init);
726module_exit(ocfs2_stack_glue_exit);