Merge commit 'v2.6.27-rc5' into tip/oprofile
[linux-2.6] / fs / ocfs2 / stackglue.c
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 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
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";
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  */
49 static struct ocfs2_stack_plugin *active_stack;
50
51 static 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
65 static 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
103 out:
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  */
117 static 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
158 static 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
172 int 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_proto = lproto;
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 }
193 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
194
195 void 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 }
214 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
215
216 void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
217 {
218         struct ocfs2_stack_plugin *p;
219
220         BUG_ON(proto == NULL);
221
222         spin_lock(&ocfs2_stack_lock);
223         BUG_ON(active_stack != NULL);
224
225         lproto = proto;
226         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
227                 p->sp_proto = lproto;
228         }
229
230         spin_unlock(&ocfs2_stack_lock);
231 }
232 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_locking_protocol);
233
234
235 /*
236  * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take
237  * "struct ocfs2_lock_res *astarg" instead of "void *astarg" because the
238  * underlying stack plugins need to pilfer the lksb off of the lock_res.
239  * If some other structure needs to be passed as an astarg, the plugins
240  * will need to be given a different avenue to the lksb.
241  */
242 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
243                    int mode,
244                    union ocfs2_dlm_lksb *lksb,
245                    u32 flags,
246                    void *name,
247                    unsigned int namelen,
248                    struct ocfs2_lock_res *astarg)
249 {
250         BUG_ON(lproto == NULL);
251
252         return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
253                                               name, namelen, astarg);
254 }
255 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
256
257 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
258                      union ocfs2_dlm_lksb *lksb,
259                      u32 flags,
260                      struct ocfs2_lock_res *astarg)
261 {
262         BUG_ON(lproto == NULL);
263
264         return active_stack->sp_ops->dlm_unlock(conn, lksb, flags, astarg);
265 }
266 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
267
268 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
269 {
270         return active_stack->sp_ops->lock_status(lksb);
271 }
272 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
273
274 /*
275  * Why don't we cast to ocfs2_meta_lvb?  The "clean" answer is that we
276  * don't cast at the glue level.  The real answer is that the header
277  * ordering is nigh impossible.
278  */
279 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
280 {
281         return active_stack->sp_ops->lock_lvb(lksb);
282 }
283 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
284
285 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
286 {
287         active_stack->sp_ops->dump_lksb(lksb);
288 }
289 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
290
291 int ocfs2_cluster_connect(const char *stack_name,
292                           const char *group,
293                           int grouplen,
294                           void (*recovery_handler)(int node_num,
295                                                    void *recovery_data),
296                           void *recovery_data,
297                           struct ocfs2_cluster_connection **conn)
298 {
299         int rc = 0;
300         struct ocfs2_cluster_connection *new_conn;
301
302         BUG_ON(group == NULL);
303         BUG_ON(conn == NULL);
304         BUG_ON(recovery_handler == NULL);
305
306         if (grouplen > GROUP_NAME_MAX) {
307                 rc = -EINVAL;
308                 goto out;
309         }
310
311         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
312                            GFP_KERNEL);
313         if (!new_conn) {
314                 rc = -ENOMEM;
315                 goto out;
316         }
317
318         memcpy(new_conn->cc_name, group, grouplen);
319         new_conn->cc_namelen = grouplen;
320         new_conn->cc_recovery_handler = recovery_handler;
321         new_conn->cc_recovery_data = recovery_data;
322
323         /* Start the new connection at our maximum compatibility level */
324         new_conn->cc_version = lproto->lp_max_version;
325
326         /* This will pin the stack driver if successful */
327         rc = ocfs2_stack_driver_get(stack_name);
328         if (rc)
329                 goto out_free;
330
331         rc = active_stack->sp_ops->connect(new_conn);
332         if (rc) {
333                 ocfs2_stack_driver_put();
334                 goto out_free;
335         }
336
337         *conn = new_conn;
338
339 out_free:
340         if (rc)
341                 kfree(new_conn);
342
343 out:
344         return rc;
345 }
346 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
347
348 /* If hangup_pending is 0, the stack driver will be dropped */
349 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
350                              int hangup_pending)
351 {
352         int ret;
353
354         BUG_ON(conn == NULL);
355
356         ret = active_stack->sp_ops->disconnect(conn);
357
358         /* XXX Should we free it anyway? */
359         if (!ret) {
360                 kfree(conn);
361                 if (!hangup_pending)
362                         ocfs2_stack_driver_put();
363         }
364
365         return ret;
366 }
367 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
368
369 /*
370  * Leave the group for this filesystem.  This is executed by a userspace
371  * program (stored in ocfs2_hb_ctl_path).
372  */
373 static void ocfs2_leave_group(const char *group)
374 {
375         int ret;
376         char *argv[5], *envp[3];
377
378         argv[0] = ocfs2_hb_ctl_path;
379         argv[1] = "-K";
380         argv[2] = "-u";
381         argv[3] = (char *)group;
382         argv[4] = NULL;
383
384         /* minimal command environment taken from cpu_run_sbin_hotplug */
385         envp[0] = "HOME=/";
386         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
387         envp[2] = NULL;
388
389         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
390         if (ret < 0) {
391                 printk(KERN_ERR
392                        "ocfs2: Error %d running user helper "
393                        "\"%s %s %s %s\"\n",
394                        ret, argv[0], argv[1], argv[2], argv[3]);
395         }
396 }
397
398 /*
399  * Hangup is a required post-umount.  ocfs2-tools software expects the
400  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
401  * regardless of whether the DLM got started, so we can't do it
402  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
403  * the actual work.
404  */
405 void ocfs2_cluster_hangup(const char *group, int grouplen)
406 {
407         BUG_ON(group == NULL);
408         BUG_ON(group[grouplen] != '\0');
409
410         ocfs2_leave_group(group);
411
412         /* cluster_disconnect() was called with hangup_pending==1 */
413         ocfs2_stack_driver_put();
414 }
415 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
416
417 int ocfs2_cluster_this_node(unsigned int *node)
418 {
419         return active_stack->sp_ops->this_node(node);
420 }
421 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
422
423
424 /*
425  * Sysfs bits
426  */
427
428 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
429                                                struct kobj_attribute *attr,
430                                                char *buf)
431 {
432         ssize_t ret = 0;
433
434         spin_lock(&ocfs2_stack_lock);
435         if (lproto)
436                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
437                                lproto->lp_max_version.pv_major,
438                                lproto->lp_max_version.pv_minor);
439         spin_unlock(&ocfs2_stack_lock);
440
441         return ret;
442 }
443
444 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
445         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
446                ocfs2_max_locking_protocol_show, NULL);
447
448 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
449                                                  struct kobj_attribute *attr,
450                                                  char *buf)
451 {
452         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
453         struct ocfs2_stack_plugin *p;
454
455         spin_lock(&ocfs2_stack_lock);
456         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
457                 ret = snprintf(buf, remain, "%s\n",
458                                p->sp_name);
459                 if (ret < 0) {
460                         total = ret;
461                         break;
462                 }
463                 if (ret == remain) {
464                         /* snprintf() didn't fit */
465                         total = -E2BIG;
466                         break;
467                 }
468                 total += ret;
469                 remain -= ret;
470         }
471         spin_unlock(&ocfs2_stack_lock);
472
473         return total;
474 }
475
476 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
477         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
478                ocfs2_loaded_cluster_plugins_show, NULL);
479
480 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
481                                                 struct kobj_attribute *attr,
482                                                 char *buf)
483 {
484         ssize_t ret = 0;
485
486         spin_lock(&ocfs2_stack_lock);
487         if (active_stack) {
488                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
489                                active_stack->sp_name);
490                 if (ret == PAGE_SIZE)
491                         ret = -E2BIG;
492         }
493         spin_unlock(&ocfs2_stack_lock);
494
495         return ret;
496 }
497
498 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
499         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
500                ocfs2_active_cluster_plugin_show, NULL);
501
502 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
503                                         struct kobj_attribute *attr,
504                                         char *buf)
505 {
506         ssize_t ret;
507         spin_lock(&ocfs2_stack_lock);
508         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
509         spin_unlock(&ocfs2_stack_lock);
510
511         return ret;
512 }
513
514 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
515                                          struct kobj_attribute *attr,
516                                          const char *buf, size_t count)
517 {
518         size_t len = count;
519         ssize_t ret;
520
521         if (len == 0)
522                 return len;
523
524         if (buf[len - 1] == '\n')
525                 len--;
526
527         if ((len != OCFS2_STACK_LABEL_LEN) ||
528             (strnlen(buf, len) != len))
529                 return -EINVAL;
530
531         spin_lock(&ocfs2_stack_lock);
532         if (active_stack) {
533                 if (!strncmp(buf, cluster_stack_name, len))
534                         ret = count;
535                 else
536                         ret = -EBUSY;
537         } else {
538                 memcpy(cluster_stack_name, buf, len);
539                 ret = count;
540         }
541         spin_unlock(&ocfs2_stack_lock);
542
543         return ret;
544 }
545
546
547 static struct kobj_attribute ocfs2_attr_cluster_stack =
548         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
549                ocfs2_cluster_stack_show,
550                ocfs2_cluster_stack_store);
551
552 static struct attribute *ocfs2_attrs[] = {
553         &ocfs2_attr_max_locking_protocol.attr,
554         &ocfs2_attr_loaded_cluster_plugins.attr,
555         &ocfs2_attr_active_cluster_plugin.attr,
556         &ocfs2_attr_cluster_stack.attr,
557         NULL,
558 };
559
560 static struct attribute_group ocfs2_attr_group = {
561         .attrs = ocfs2_attrs,
562 };
563
564 static struct kset *ocfs2_kset;
565
566 static void ocfs2_sysfs_exit(void)
567 {
568         kset_unregister(ocfs2_kset);
569 }
570
571 static int ocfs2_sysfs_init(void)
572 {
573         int ret;
574
575         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
576         if (!ocfs2_kset)
577                 return -ENOMEM;
578
579         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
580         if (ret)
581                 goto error;
582
583         return 0;
584
585 error:
586         kset_unregister(ocfs2_kset);
587         return ret;
588 }
589
590 /*
591  * Sysctl bits
592  *
593  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
594  * make as much sense in a multiple cluster stack world, but it's safer
595  * and easier to preserve the name.
596  */
597
598 #define FS_OCFS2_NM             1
599
600 static ctl_table ocfs2_nm_table[] = {
601         {
602                 .ctl_name       = 1,
603                 .procname       = "hb_ctl_path",
604                 .data           = ocfs2_hb_ctl_path,
605                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
606                 .mode           = 0644,
607                 .proc_handler   = &proc_dostring,
608                 .strategy       = &sysctl_string,
609         },
610         { .ctl_name = 0 }
611 };
612
613 static ctl_table ocfs2_mod_table[] = {
614         {
615                 .ctl_name       = FS_OCFS2_NM,
616                 .procname       = "nm",
617                 .data           = NULL,
618                 .maxlen         = 0,
619                 .mode           = 0555,
620                 .child          = ocfs2_nm_table
621         },
622         { .ctl_name = 0}
623 };
624
625 static ctl_table ocfs2_kern_table[] = {
626         {
627                 .ctl_name       = FS_OCFS2,
628                 .procname       = "ocfs2",
629                 .data           = NULL,
630                 .maxlen         = 0,
631                 .mode           = 0555,
632                 .child          = ocfs2_mod_table
633         },
634         { .ctl_name = 0}
635 };
636
637 static ctl_table ocfs2_root_table[] = {
638         {
639                 .ctl_name       = CTL_FS,
640                 .procname       = "fs",
641                 .data           = NULL,
642                 .maxlen         = 0,
643                 .mode           = 0555,
644                 .child          = ocfs2_kern_table
645         },
646         { .ctl_name = 0 }
647 };
648
649 static struct ctl_table_header *ocfs2_table_header = NULL;
650
651
652 /*
653  * Initialization
654  */
655
656 static int __init ocfs2_stack_glue_init(void)
657 {
658         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
659
660         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
661         if (!ocfs2_table_header) {
662                 printk(KERN_ERR
663                        "ocfs2 stack glue: unable to register sysctl\n");
664                 return -ENOMEM; /* or something. */
665         }
666
667         return ocfs2_sysfs_init();
668 }
669
670 static void __exit ocfs2_stack_glue_exit(void)
671 {
672         lproto = NULL;
673         ocfs2_sysfs_exit();
674         if (ocfs2_table_header)
675                 unregister_sysctl_table(ocfs2_table_header);
676 }
677
678 MODULE_AUTHOR("Oracle");
679 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
680 MODULE_LICENSE("GPL");
681 module_init(ocfs2_stack_glue_init);
682 module_exit(ocfs2_stack_glue_exit);