fc184c762700f4ab1a6f7fc75e17e1001263b727
[linux-2.6.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, 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
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_max_proto = lproto->lp_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 }
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_max_proto = lproto->lp_max_version;
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 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  */
241 int 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         BUG_ON(lproto == NULL);
249
250         if (!lksb->lksb_conn)
251                 lksb->lksb_conn = conn;
252         else
253                 BUG_ON(lksb->lksb_conn != conn);
254         return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
255                                               name, namelen);
256 }
257 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
258
259 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
260                      struct ocfs2_dlm_lksb *lksb,
261                      u32 flags)
262 {
263         BUG_ON(lproto == NULL);
264         BUG_ON(lksb->lksb_conn == NULL);
265
266         return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
267 }
268 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
269
270 int ocfs2_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
271 {
272         return active_stack->sp_ops->lock_status(lksb);
273 }
274 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
275
276 int ocfs2_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
277 {
278         return active_stack->sp_ops->lvb_valid(lksb);
279 }
280 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
281
282 void *ocfs2_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
283 {
284         return active_stack->sp_ops->lock_lvb(lksb);
285 }
286 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
287
288 void ocfs2_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
289 {
290         active_stack->sp_ops->dump_lksb(lksb);
291 }
292 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
293
294 int ocfs2_stack_supports_plocks(void)
295 {
296         return active_stack && active_stack->sp_ops->plock;
297 }
298 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
299
300 /*
301  * ocfs2_plock() can only be safely called if
302  * ocfs2_stack_supports_plocks() returned true
303  */
304 int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
305                 struct file *file, int cmd, struct file_lock *fl)
306 {
307         WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
308         if (active_stack->sp_ops->plock)
309                 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
310         return -EOPNOTSUPP;
311 }
312 EXPORT_SYMBOL_GPL(ocfs2_plock);
313
314 int ocfs2_cluster_connect(const char *stack_name,
315                           const char *group,
316                           int grouplen,
317                           void (*recovery_handler)(int node_num,
318                                                    void *recovery_data),
319                           void *recovery_data,
320                           struct ocfs2_cluster_connection **conn)
321 {
322         int rc = 0;
323         struct ocfs2_cluster_connection *new_conn;
324
325         BUG_ON(group == NULL);
326         BUG_ON(conn == NULL);
327         BUG_ON(recovery_handler == NULL);
328
329         if (grouplen > GROUP_NAME_MAX) {
330                 rc = -EINVAL;
331                 goto out;
332         }
333
334         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
335                            GFP_KERNEL);
336         if (!new_conn) {
337                 rc = -ENOMEM;
338                 goto out;
339         }
340
341         memcpy(new_conn->cc_name, group, grouplen);
342         new_conn->cc_namelen = grouplen;
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
363 out_free:
364         if (rc)
365                 kfree(new_conn);
366
367 out:
368         return rc;
369 }
370 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
371
372 /* If hangup_pending is 0, the stack driver will be dropped */
373 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
374                              int hangup_pending)
375 {
376         int ret;
377
378         BUG_ON(conn == NULL);
379
380         ret = active_stack->sp_ops->disconnect(conn);
381
382         /* XXX Should we free it anyway? */
383         if (!ret) {
384                 kfree(conn);
385                 if (!hangup_pending)
386                         ocfs2_stack_driver_put();
387         }
388
389         return ret;
390 }
391 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
392
393 /*
394  * Leave the group for this filesystem.  This is executed by a userspace
395  * program (stored in ocfs2_hb_ctl_path).
396  */
397 static void ocfs2_leave_group(const char *group)
398 {
399         int ret;
400         char *argv[5], *envp[3];
401
402         argv[0] = ocfs2_hb_ctl_path;
403         argv[1] = "-K";
404         argv[2] = "-u";
405         argv[3] = (char *)group;
406         argv[4] = NULL;
407
408         /* minimal command environment taken from cpu_run_sbin_hotplug */
409         envp[0] = "HOME=/";
410         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
411         envp[2] = NULL;
412
413         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
414         if (ret < 0) {
415                 printk(KERN_ERR
416                        "ocfs2: Error %d running user helper "
417                        "\"%s %s %s %s\"\n",
418                        ret, argv[0], argv[1], argv[2], argv[3]);
419         }
420 }
421
422 /*
423  * Hangup is a required post-umount.  ocfs2-tools software expects the
424  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
425  * regardless of whether the DLM got started, so we can't do it
426  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
427  * the actual work.
428  */
429 void ocfs2_cluster_hangup(const char *group, int grouplen)
430 {
431         BUG_ON(group == NULL);
432         BUG_ON(group[grouplen] != '\0');
433
434         ocfs2_leave_group(group);
435
436         /* cluster_disconnect() was called with hangup_pending==1 */
437         ocfs2_stack_driver_put();
438 }
439 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
440
441 int ocfs2_cluster_this_node(unsigned int *node)
442 {
443         return active_stack->sp_ops->this_node(node);
444 }
445 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
446
447
448 /*
449  * Sysfs bits
450  */
451
452 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
453                                                struct kobj_attribute *attr,
454                                                char *buf)
455 {
456         ssize_t ret = 0;
457
458         spin_lock(&ocfs2_stack_lock);
459         if (lproto)
460                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
461                                lproto->lp_max_version.pv_major,
462                                lproto->lp_max_version.pv_minor);
463         spin_unlock(&ocfs2_stack_lock);
464
465         return ret;
466 }
467
468 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
469         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
470                ocfs2_max_locking_protocol_show, NULL);
471
472 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
473                                                  struct kobj_attribute *attr,
474                                                  char *buf)
475 {
476         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
477         struct ocfs2_stack_plugin *p;
478
479         spin_lock(&ocfs2_stack_lock);
480         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
481                 ret = snprintf(buf, remain, "%s\n",
482                                p->sp_name);
483                 if (ret < 0) {
484                         total = ret;
485                         break;
486                 }
487                 if (ret == remain) {
488                         /* snprintf() didn't fit */
489                         total = -E2BIG;
490                         break;
491                 }
492                 total += ret;
493                 remain -= ret;
494         }
495         spin_unlock(&ocfs2_stack_lock);
496
497         return total;
498 }
499
500 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
501         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
502                ocfs2_loaded_cluster_plugins_show, NULL);
503
504 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
505                                                 struct kobj_attribute *attr,
506                                                 char *buf)
507 {
508         ssize_t ret = 0;
509
510         spin_lock(&ocfs2_stack_lock);
511         if (active_stack) {
512                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
513                                active_stack->sp_name);
514                 if (ret == PAGE_SIZE)
515                         ret = -E2BIG;
516         }
517         spin_unlock(&ocfs2_stack_lock);
518
519         return ret;
520 }
521
522 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
523         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
524                ocfs2_active_cluster_plugin_show, NULL);
525
526 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
527                                         struct kobj_attribute *attr,
528                                         char *buf)
529 {
530         ssize_t ret;
531         spin_lock(&ocfs2_stack_lock);
532         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
533         spin_unlock(&ocfs2_stack_lock);
534
535         return ret;
536 }
537
538 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
539                                          struct kobj_attribute *attr,
540                                          const char *buf, size_t count)
541 {
542         size_t len = count;
543         ssize_t ret;
544
545         if (len == 0)
546                 return len;
547
548         if (buf[len - 1] == '\n')
549                 len--;
550
551         if ((len != OCFS2_STACK_LABEL_LEN) ||
552             (strnlen(buf, len) != len))
553                 return -EINVAL;
554
555         spin_lock(&ocfs2_stack_lock);
556         if (active_stack) {
557                 if (!strncmp(buf, cluster_stack_name, len))
558                         ret = count;
559                 else
560                         ret = -EBUSY;
561         } else {
562                 memcpy(cluster_stack_name, buf, len);
563                 ret = count;
564         }
565         spin_unlock(&ocfs2_stack_lock);
566
567         return ret;
568 }
569
570
571 static struct kobj_attribute ocfs2_attr_cluster_stack =
572         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
573                ocfs2_cluster_stack_show,
574                ocfs2_cluster_stack_store);
575
576 static struct attribute *ocfs2_attrs[] = {
577         &ocfs2_attr_max_locking_protocol.attr,
578         &ocfs2_attr_loaded_cluster_plugins.attr,
579         &ocfs2_attr_active_cluster_plugin.attr,
580         &ocfs2_attr_cluster_stack.attr,
581         NULL,
582 };
583
584 static struct attribute_group ocfs2_attr_group = {
585         .attrs = ocfs2_attrs,
586 };
587
588 static struct kset *ocfs2_kset;
589
590 static void ocfs2_sysfs_exit(void)
591 {
592         kset_unregister(ocfs2_kset);
593 }
594
595 static int ocfs2_sysfs_init(void)
596 {
597         int ret;
598
599         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
600         if (!ocfs2_kset)
601                 return -ENOMEM;
602
603         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
604         if (ret)
605                 goto error;
606
607         return 0;
608
609 error:
610         kset_unregister(ocfs2_kset);
611         return ret;
612 }
613
614 /*
615  * Sysctl bits
616  *
617  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
618  * make as much sense in a multiple cluster stack world, but it's safer
619  * and easier to preserve the name.
620  */
621
622 #define FS_OCFS2_NM             1
623
624 static ctl_table ocfs2_nm_table[] = {
625         {
626                 .procname       = "hb_ctl_path",
627                 .data           = ocfs2_hb_ctl_path,
628                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
629                 .mode           = 0644,
630                 .proc_handler   = proc_dostring,
631         },
632         { }
633 };
634
635 static ctl_table ocfs2_mod_table[] = {
636         {
637                 .procname       = "nm",
638                 .data           = NULL,
639                 .maxlen         = 0,
640                 .mode           = 0555,
641                 .child          = ocfs2_nm_table
642         },
643         { }
644 };
645
646 static ctl_table ocfs2_kern_table[] = {
647         {
648                 .procname       = "ocfs2",
649                 .data           = NULL,
650                 .maxlen         = 0,
651                 .mode           = 0555,
652                 .child          = ocfs2_mod_table
653         },
654         { }
655 };
656
657 static ctl_table ocfs2_root_table[] = {
658         {
659                 .procname       = "fs",
660                 .data           = NULL,
661                 .maxlen         = 0,
662                 .mode           = 0555,
663                 .child          = ocfs2_kern_table
664         },
665         { }
666 };
667
668 static struct ctl_table_header *ocfs2_table_header = NULL;
669
670
671 /*
672  * Initialization
673  */
674
675 static int __init ocfs2_stack_glue_init(void)
676 {
677         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
678
679         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
680         if (!ocfs2_table_header) {
681                 printk(KERN_ERR
682                        "ocfs2 stack glue: unable to register sysctl\n");
683                 return -ENOMEM; /* or something. */
684         }
685
686         return ocfs2_sysfs_init();
687 }
688
689 static void __exit ocfs2_stack_glue_exit(void)
690 {
691         lproto = NULL;
692         ocfs2_sysfs_exit();
693         if (ocfs2_table_header)
694                 unregister_sysctl_table(ocfs2_table_header);
695 }
696
697 MODULE_AUTHOR("Oracle");
698 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
699 MODULE_LICENSE("GPL");
700 module_init(ocfs2_stack_glue_init);
701 module_exit(ocfs2_stack_glue_exit);