Merge git://git.kernel.org/pub/scm/linux/kernel/git/davem/net-2.6
[pandora-kernel.git] / 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         /* Ok, the stack is pinned */
101         p->sp_count++;
102         active_stack = p;
103
104         rc = 0;
105
106 out:
107         spin_unlock(&ocfs2_stack_lock);
108         return rc;
109 }
110
111 /*
112  * This function looks up the appropriate stack and makes it active.  If
113  * there is no stack, it tries to load it.  It will fail if the stack still
114  * cannot be found.  It will also fail if a different stack is in use.
115  */
116 static int ocfs2_stack_driver_get(const char *stack_name)
117 {
118         int rc;
119         char *plugin_name = OCFS2_STACK_PLUGIN_O2CB;
120
121         /*
122          * Classic stack does not pass in a stack name.  This is
123          * compatible with older tools as well.
124          */
125         if (!stack_name || !*stack_name)
126                 stack_name = OCFS2_STACK_PLUGIN_O2CB;
127
128         if (strlen(stack_name) != OCFS2_STACK_LABEL_LEN) {
129                 printk(KERN_ERR
130                        "ocfs2 passed an invalid cluster stack label: \"%s\"\n",
131                        stack_name);
132                 return -EINVAL;
133         }
134
135         /* Anything that isn't the classic stack is a user stack */
136         if (strcmp(stack_name, OCFS2_STACK_PLUGIN_O2CB))
137                 plugin_name = OCFS2_STACK_PLUGIN_USER;
138
139         rc = ocfs2_stack_driver_request(stack_name, plugin_name);
140         if (rc == -ENOENT) {
141                 request_module("ocfs2_stack_%s", plugin_name);
142                 rc = ocfs2_stack_driver_request(stack_name, plugin_name);
143         }
144
145         if (rc == -ENOENT) {
146                 printk(KERN_ERR
147                        "ocfs2: Cluster stack driver \"%s\" cannot be found\n",
148                        plugin_name);
149         } else if (rc == -EBUSY) {
150                 printk(KERN_ERR
151                        "ocfs2: A different cluster stack is in use\n");
152         }
153
154         return rc;
155 }
156
157 static void ocfs2_stack_driver_put(void)
158 {
159         spin_lock(&ocfs2_stack_lock);
160         BUG_ON(active_stack == NULL);
161         BUG_ON(active_stack->sp_count == 0);
162
163         active_stack->sp_count--;
164         if (!active_stack->sp_count) {
165                 module_put(active_stack->sp_owner);
166                 active_stack = NULL;
167         }
168         spin_unlock(&ocfs2_stack_lock);
169 }
170
171 int ocfs2_stack_glue_register(struct ocfs2_stack_plugin *plugin)
172 {
173         int rc;
174
175         spin_lock(&ocfs2_stack_lock);
176         if (!ocfs2_stack_lookup(plugin->sp_name)) {
177                 plugin->sp_count = 0;
178                 plugin->sp_proto = lproto;
179                 list_add(&plugin->sp_list, &ocfs2_stack_list);
180                 printk(KERN_INFO "ocfs2: Registered cluster interface %s\n",
181                        plugin->sp_name);
182                 rc = 0;
183         } else {
184                 printk(KERN_ERR "ocfs2: Stack \"%s\" already registered\n",
185                        plugin->sp_name);
186                 rc = -EEXIST;
187         }
188         spin_unlock(&ocfs2_stack_lock);
189
190         return rc;
191 }
192 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_register);
193
194 void ocfs2_stack_glue_unregister(struct ocfs2_stack_plugin *plugin)
195 {
196         struct ocfs2_stack_plugin *p;
197
198         spin_lock(&ocfs2_stack_lock);
199         p = ocfs2_stack_lookup(plugin->sp_name);
200         if (p) {
201                 BUG_ON(p != plugin);
202                 BUG_ON(plugin == active_stack);
203                 BUG_ON(plugin->sp_count != 0);
204                 list_del_init(&plugin->sp_list);
205                 printk(KERN_INFO "ocfs2: Unregistered cluster interface %s\n",
206                        plugin->sp_name);
207         } else {
208                 printk(KERN_ERR "Stack \"%s\" is not registered\n",
209                        plugin->sp_name);
210         }
211         spin_unlock(&ocfs2_stack_lock);
212 }
213 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_unregister);
214
215 void ocfs2_stack_glue_set_locking_protocol(struct ocfs2_locking_protocol *proto)
216 {
217         struct ocfs2_stack_plugin *p;
218
219         BUG_ON(proto == NULL);
220
221         spin_lock(&ocfs2_stack_lock);
222         BUG_ON(active_stack != NULL);
223
224         lproto = proto;
225         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
226                 p->sp_proto = lproto;
227         }
228
229         spin_unlock(&ocfs2_stack_lock);
230 }
231 EXPORT_SYMBOL_GPL(ocfs2_stack_glue_set_locking_protocol);
232
233
234 /*
235  * The ocfs2_dlm_lock() and ocfs2_dlm_unlock() functions take
236  * "struct ocfs2_lock_res *astarg" instead of "void *astarg" because the
237  * underlying stack plugins need to pilfer the lksb off of the lock_res.
238  * If some other structure needs to be passed as an astarg, the plugins
239  * will need to be given a different avenue to the lksb.
240  */
241 int ocfs2_dlm_lock(struct ocfs2_cluster_connection *conn,
242                    int mode,
243                    union ocfs2_dlm_lksb *lksb,
244                    u32 flags,
245                    void *name,
246                    unsigned int namelen,
247                    struct ocfs2_lock_res *astarg)
248 {
249         BUG_ON(lproto == NULL);
250
251         return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
252                                               name, namelen, astarg);
253 }
254 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
255
256 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
257                      union ocfs2_dlm_lksb *lksb,
258                      u32 flags,
259                      struct ocfs2_lock_res *astarg)
260 {
261         BUG_ON(lproto == NULL);
262
263         return active_stack->sp_ops->dlm_unlock(conn, lksb, flags, astarg);
264 }
265 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
266
267 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
268 {
269         return active_stack->sp_ops->lock_status(lksb);
270 }
271 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
272
273 /*
274  * Why don't we cast to ocfs2_meta_lvb?  The "clean" answer is that we
275  * don't cast at the glue level.  The real answer is that the header
276  * ordering is nigh impossible.
277  */
278 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
279 {
280         return active_stack->sp_ops->lock_lvb(lksb);
281 }
282 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
283
284 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
285 {
286         active_stack->sp_ops->dump_lksb(lksb);
287 }
288 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
289
290 int ocfs2_cluster_connect(const char *stack_name,
291                           const char *group,
292                           int grouplen,
293                           void (*recovery_handler)(int node_num,
294                                                    void *recovery_data),
295                           void *recovery_data,
296                           struct ocfs2_cluster_connection **conn)
297 {
298         int rc = 0;
299         struct ocfs2_cluster_connection *new_conn;
300
301         BUG_ON(group == NULL);
302         BUG_ON(conn == NULL);
303         BUG_ON(recovery_handler == NULL);
304
305         if (grouplen > GROUP_NAME_MAX) {
306                 rc = -EINVAL;
307                 goto out;
308         }
309
310         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
311                            GFP_KERNEL);
312         if (!new_conn) {
313                 rc = -ENOMEM;
314                 goto out;
315         }
316
317         memcpy(new_conn->cc_name, group, grouplen);
318         new_conn->cc_namelen = grouplen;
319         new_conn->cc_recovery_handler = recovery_handler;
320         new_conn->cc_recovery_data = recovery_data;
321
322         /* Start the new connection at our maximum compatibility level */
323         new_conn->cc_version = lproto->lp_max_version;
324
325         /* This will pin the stack driver if successful */
326         rc = ocfs2_stack_driver_get(stack_name);
327         if (rc)
328                 goto out_free;
329
330         rc = active_stack->sp_ops->connect(new_conn);
331         if (rc) {
332                 ocfs2_stack_driver_put();
333                 goto out_free;
334         }
335
336         *conn = new_conn;
337
338 out_free:
339         if (rc)
340                 kfree(new_conn);
341
342 out:
343         return rc;
344 }
345 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
346
347 /* If hangup_pending is 0, the stack driver will be dropped */
348 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
349                              int hangup_pending)
350 {
351         int ret;
352
353         BUG_ON(conn == NULL);
354
355         ret = active_stack->sp_ops->disconnect(conn);
356
357         /* XXX Should we free it anyway? */
358         if (!ret) {
359                 kfree(conn);
360                 if (!hangup_pending)
361                         ocfs2_stack_driver_put();
362         }
363
364         return ret;
365 }
366 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
367
368 /*
369  * Leave the group for this filesystem.  This is executed by a userspace
370  * program (stored in ocfs2_hb_ctl_path).
371  */
372 static void ocfs2_leave_group(const char *group)
373 {
374         int ret;
375         char *argv[5], *envp[3];
376
377         argv[0] = ocfs2_hb_ctl_path;
378         argv[1] = "-K";
379         argv[2] = "-u";
380         argv[3] = (char *)group;
381         argv[4] = NULL;
382
383         /* minimal command environment taken from cpu_run_sbin_hotplug */
384         envp[0] = "HOME=/";
385         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
386         envp[2] = NULL;
387
388         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
389         if (ret < 0) {
390                 printk(KERN_ERR
391                        "ocfs2: Error %d running user helper "
392                        "\"%s %s %s %s\"\n",
393                        ret, argv[0], argv[1], argv[2], argv[3]);
394         }
395 }
396
397 /*
398  * Hangup is a required post-umount.  ocfs2-tools software expects the
399  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
400  * regardless of whether the DLM got started, so we can't do it
401  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
402  * the actual work.
403  */
404 void ocfs2_cluster_hangup(const char *group, int grouplen)
405 {
406         BUG_ON(group == NULL);
407         BUG_ON(group[grouplen] != '\0');
408
409         ocfs2_leave_group(group);
410
411         /* cluster_disconnect() was called with hangup_pending==1 */
412         ocfs2_stack_driver_put();
413 }
414 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
415
416 int ocfs2_cluster_this_node(unsigned int *node)
417 {
418         return active_stack->sp_ops->this_node(node);
419 }
420 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
421
422
423 /*
424  * Sysfs bits
425  */
426
427 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
428                                                struct kobj_attribute *attr,
429                                                char *buf)
430 {
431         ssize_t ret = 0;
432
433         spin_lock(&ocfs2_stack_lock);
434         if (lproto)
435                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
436                                lproto->lp_max_version.pv_major,
437                                lproto->lp_max_version.pv_minor);
438         spin_unlock(&ocfs2_stack_lock);
439
440         return ret;
441 }
442
443 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
444         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
445                ocfs2_max_locking_protocol_show, NULL);
446
447 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
448                                                  struct kobj_attribute *attr,
449                                                  char *buf)
450 {
451         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
452         struct ocfs2_stack_plugin *p;
453
454         spin_lock(&ocfs2_stack_lock);
455         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
456                 ret = snprintf(buf, remain, "%s\n",
457                                p->sp_name);
458                 if (ret < 0) {
459                         total = ret;
460                         break;
461                 }
462                 if (ret == remain) {
463                         /* snprintf() didn't fit */
464                         total = -E2BIG;
465                         break;
466                 }
467                 total += ret;
468                 remain -= ret;
469         }
470         spin_unlock(&ocfs2_stack_lock);
471
472         return total;
473 }
474
475 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
476         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
477                ocfs2_loaded_cluster_plugins_show, NULL);
478
479 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
480                                                 struct kobj_attribute *attr,
481                                                 char *buf)
482 {
483         ssize_t ret = 0;
484
485         spin_lock(&ocfs2_stack_lock);
486         if (active_stack) {
487                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
488                                active_stack->sp_name);
489                 if (ret == PAGE_SIZE)
490                         ret = -E2BIG;
491         }
492         spin_unlock(&ocfs2_stack_lock);
493
494         return ret;
495 }
496
497 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
498         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
499                ocfs2_active_cluster_plugin_show, NULL);
500
501 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
502                                         struct kobj_attribute *attr,
503                                         char *buf)
504 {
505         ssize_t ret;
506         spin_lock(&ocfs2_stack_lock);
507         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
508         spin_unlock(&ocfs2_stack_lock);
509
510         return ret;
511 }
512
513 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
514                                          struct kobj_attribute *attr,
515                                          const char *buf, size_t count)
516 {
517         size_t len = count;
518         ssize_t ret;
519
520         if (len == 0)
521                 return len;
522
523         if (buf[len - 1] == '\n')
524                 len--;
525
526         if ((len != OCFS2_STACK_LABEL_LEN) ||
527             (strnlen(buf, len) != len))
528                 return -EINVAL;
529
530         spin_lock(&ocfs2_stack_lock);
531         if (active_stack) {
532                 if (!strncmp(buf, cluster_stack_name, len))
533                         ret = count;
534                 else
535                         ret = -EBUSY;
536         } else {
537                 memcpy(cluster_stack_name, buf, len);
538                 ret = count;
539         }
540         spin_unlock(&ocfs2_stack_lock);
541
542         return ret;
543 }
544
545
546 static struct kobj_attribute ocfs2_attr_cluster_stack =
547         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
548                ocfs2_cluster_stack_show,
549                ocfs2_cluster_stack_store);
550
551 static struct attribute *ocfs2_attrs[] = {
552         &ocfs2_attr_max_locking_protocol.attr,
553         &ocfs2_attr_loaded_cluster_plugins.attr,
554         &ocfs2_attr_active_cluster_plugin.attr,
555         &ocfs2_attr_cluster_stack.attr,
556         NULL,
557 };
558
559 static struct attribute_group ocfs2_attr_group = {
560         .attrs = ocfs2_attrs,
561 };
562
563 static struct kset *ocfs2_kset;
564
565 static void ocfs2_sysfs_exit(void)
566 {
567         kset_unregister(ocfs2_kset);
568 }
569
570 static int ocfs2_sysfs_init(void)
571 {
572         int ret;
573
574         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
575         if (!ocfs2_kset)
576                 return -ENOMEM;
577
578         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
579         if (ret)
580                 goto error;
581
582         return 0;
583
584 error:
585         kset_unregister(ocfs2_kset);
586         return ret;
587 }
588
589 /*
590  * Sysctl bits
591  *
592  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
593  * make as much sense in a multiple cluster stack world, but it's safer
594  * and easier to preserve the name.
595  */
596
597 #define FS_OCFS2_NM             1
598
599 static ctl_table ocfs2_nm_table[] = {
600         {
601                 .ctl_name       = 1,
602                 .procname       = "hb_ctl_path",
603                 .data           = ocfs2_hb_ctl_path,
604                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
605                 .mode           = 0644,
606                 .proc_handler   = &proc_dostring,
607                 .strategy       = &sysctl_string,
608         },
609         { .ctl_name = 0 }
610 };
611
612 static ctl_table ocfs2_mod_table[] = {
613         {
614                 .ctl_name       = FS_OCFS2_NM,
615                 .procname       = "nm",
616                 .data           = NULL,
617                 .maxlen         = 0,
618                 .mode           = 0555,
619                 .child          = ocfs2_nm_table
620         },
621         { .ctl_name = 0}
622 };
623
624 static ctl_table ocfs2_kern_table[] = {
625         {
626                 .ctl_name       = FS_OCFS2,
627                 .procname       = "ocfs2",
628                 .data           = NULL,
629                 .maxlen         = 0,
630                 .mode           = 0555,
631                 .child          = ocfs2_mod_table
632         },
633         { .ctl_name = 0}
634 };
635
636 static ctl_table ocfs2_root_table[] = {
637         {
638                 .ctl_name       = CTL_FS,
639                 .procname       = "fs",
640                 .data           = NULL,
641                 .maxlen         = 0,
642                 .mode           = 0555,
643                 .child          = ocfs2_kern_table
644         },
645         { .ctl_name = 0 }
646 };
647
648 static struct ctl_table_header *ocfs2_table_header = NULL;
649
650
651 /*
652  * Initialization
653  */
654
655 static int __init ocfs2_stack_glue_init(void)
656 {
657         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
658
659         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
660         if (!ocfs2_table_header) {
661                 printk(KERN_ERR
662                        "ocfs2 stack glue: unable to register sysctl\n");
663                 return -ENOMEM; /* or something. */
664         }
665
666         return ocfs2_sysfs_init();
667 }
668
669 static void __exit ocfs2_stack_glue_exit(void)
670 {
671         lproto = NULL;
672         ocfs2_sysfs_exit();
673         if (ocfs2_table_header)
674                 unregister_sysctl_table(ocfs2_table_header);
675 }
676
677 MODULE_AUTHOR("Oracle");
678 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
679 MODULE_LICENSE("GPL");
680 module_init(ocfs2_stack_glue_init);
681 module_exit(ocfs2_stack_glue_exit);