ocfs2: Pass lksbs back from stackglue ast/bast functions.
[safe/jmp/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, 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_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 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                    union ocfs2_dlm_lksb *lksb,
244                    u32 flags,
245                    void *name,
246                    unsigned int namelen)
247 {
248         BUG_ON(lproto == NULL);
249
250         return active_stack->sp_ops->dlm_lock(conn, mode, lksb, flags,
251                                               name, namelen);
252 }
253 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock);
254
255 int ocfs2_dlm_unlock(struct ocfs2_cluster_connection *conn,
256                      union ocfs2_dlm_lksb *lksb,
257                      u32 flags)
258 {
259         BUG_ON(lproto == NULL);
260
261         return active_stack->sp_ops->dlm_unlock(conn, lksb, flags);
262 }
263 EXPORT_SYMBOL_GPL(ocfs2_dlm_unlock);
264
265 int ocfs2_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
266 {
267         return active_stack->sp_ops->lock_status(lksb);
268 }
269 EXPORT_SYMBOL_GPL(ocfs2_dlm_lock_status);
270
271 int ocfs2_dlm_lvb_valid(union ocfs2_dlm_lksb *lksb)
272 {
273         return active_stack->sp_ops->lvb_valid(lksb);
274 }
275 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb_valid);
276
277 void *ocfs2_dlm_lvb(union ocfs2_dlm_lksb *lksb)
278 {
279         return active_stack->sp_ops->lock_lvb(lksb);
280 }
281 EXPORT_SYMBOL_GPL(ocfs2_dlm_lvb);
282
283 void ocfs2_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
284 {
285         active_stack->sp_ops->dump_lksb(lksb);
286 }
287 EXPORT_SYMBOL_GPL(ocfs2_dlm_dump_lksb);
288
289 int ocfs2_stack_supports_plocks(void)
290 {
291         return active_stack && active_stack->sp_ops->plock;
292 }
293 EXPORT_SYMBOL_GPL(ocfs2_stack_supports_plocks);
294
295 /*
296  * ocfs2_plock() can only be safely called if
297  * ocfs2_stack_supports_plocks() returned true
298  */
299 int ocfs2_plock(struct ocfs2_cluster_connection *conn, u64 ino,
300                 struct file *file, int cmd, struct file_lock *fl)
301 {
302         WARN_ON_ONCE(active_stack->sp_ops->plock == NULL);
303         if (active_stack->sp_ops->plock)
304                 return active_stack->sp_ops->plock(conn, ino, file, cmd, fl);
305         return -EOPNOTSUPP;
306 }
307 EXPORT_SYMBOL_GPL(ocfs2_plock);
308
309 int ocfs2_cluster_connect(const char *stack_name,
310                           const char *group,
311                           int grouplen,
312                           void (*recovery_handler)(int node_num,
313                                                    void *recovery_data),
314                           void *recovery_data,
315                           struct ocfs2_cluster_connection **conn)
316 {
317         int rc = 0;
318         struct ocfs2_cluster_connection *new_conn;
319
320         BUG_ON(group == NULL);
321         BUG_ON(conn == NULL);
322         BUG_ON(recovery_handler == NULL);
323
324         if (grouplen > GROUP_NAME_MAX) {
325                 rc = -EINVAL;
326                 goto out;
327         }
328
329         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
330                            GFP_KERNEL);
331         if (!new_conn) {
332                 rc = -ENOMEM;
333                 goto out;
334         }
335
336         memcpy(new_conn->cc_name, group, grouplen);
337         new_conn->cc_namelen = grouplen;
338         new_conn->cc_recovery_handler = recovery_handler;
339         new_conn->cc_recovery_data = recovery_data;
340
341         /* Start the new connection at our maximum compatibility level */
342         new_conn->cc_version = lproto->lp_max_version;
343
344         /* This will pin the stack driver if successful */
345         rc = ocfs2_stack_driver_get(stack_name);
346         if (rc)
347                 goto out_free;
348
349         rc = active_stack->sp_ops->connect(new_conn);
350         if (rc) {
351                 ocfs2_stack_driver_put();
352                 goto out_free;
353         }
354
355         *conn = new_conn;
356
357 out_free:
358         if (rc)
359                 kfree(new_conn);
360
361 out:
362         return rc;
363 }
364 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
365
366 /* If hangup_pending is 0, the stack driver will be dropped */
367 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
368                              int hangup_pending)
369 {
370         int ret;
371
372         BUG_ON(conn == NULL);
373
374         ret = active_stack->sp_ops->disconnect(conn);
375
376         /* XXX Should we free it anyway? */
377         if (!ret) {
378                 kfree(conn);
379                 if (!hangup_pending)
380                         ocfs2_stack_driver_put();
381         }
382
383         return ret;
384 }
385 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
386
387 /*
388  * Leave the group for this filesystem.  This is executed by a userspace
389  * program (stored in ocfs2_hb_ctl_path).
390  */
391 static void ocfs2_leave_group(const char *group)
392 {
393         int ret;
394         char *argv[5], *envp[3];
395
396         argv[0] = ocfs2_hb_ctl_path;
397         argv[1] = "-K";
398         argv[2] = "-u";
399         argv[3] = (char *)group;
400         argv[4] = NULL;
401
402         /* minimal command environment taken from cpu_run_sbin_hotplug */
403         envp[0] = "HOME=/";
404         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
405         envp[2] = NULL;
406
407         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
408         if (ret < 0) {
409                 printk(KERN_ERR
410                        "ocfs2: Error %d running user helper "
411                        "\"%s %s %s %s\"\n",
412                        ret, argv[0], argv[1], argv[2], argv[3]);
413         }
414 }
415
416 /*
417  * Hangup is a required post-umount.  ocfs2-tools software expects the
418  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
419  * regardless of whether the DLM got started, so we can't do it
420  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
421  * the actual work.
422  */
423 void ocfs2_cluster_hangup(const char *group, int grouplen)
424 {
425         BUG_ON(group == NULL);
426         BUG_ON(group[grouplen] != '\0');
427
428         ocfs2_leave_group(group);
429
430         /* cluster_disconnect() was called with hangup_pending==1 */
431         ocfs2_stack_driver_put();
432 }
433 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
434
435 int ocfs2_cluster_this_node(unsigned int *node)
436 {
437         return active_stack->sp_ops->this_node(node);
438 }
439 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
440
441
442 /*
443  * Sysfs bits
444  */
445
446 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
447                                                struct kobj_attribute *attr,
448                                                char *buf)
449 {
450         ssize_t ret = 0;
451
452         spin_lock(&ocfs2_stack_lock);
453         if (lproto)
454                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
455                                lproto->lp_max_version.pv_major,
456                                lproto->lp_max_version.pv_minor);
457         spin_unlock(&ocfs2_stack_lock);
458
459         return ret;
460 }
461
462 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
463         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
464                ocfs2_max_locking_protocol_show, NULL);
465
466 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
467                                                  struct kobj_attribute *attr,
468                                                  char *buf)
469 {
470         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
471         struct ocfs2_stack_plugin *p;
472
473         spin_lock(&ocfs2_stack_lock);
474         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
475                 ret = snprintf(buf, remain, "%s\n",
476                                p->sp_name);
477                 if (ret < 0) {
478                         total = ret;
479                         break;
480                 }
481                 if (ret == remain) {
482                         /* snprintf() didn't fit */
483                         total = -E2BIG;
484                         break;
485                 }
486                 total += ret;
487                 remain -= ret;
488         }
489         spin_unlock(&ocfs2_stack_lock);
490
491         return total;
492 }
493
494 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
495         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
496                ocfs2_loaded_cluster_plugins_show, NULL);
497
498 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
499                                                 struct kobj_attribute *attr,
500                                                 char *buf)
501 {
502         ssize_t ret = 0;
503
504         spin_lock(&ocfs2_stack_lock);
505         if (active_stack) {
506                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
507                                active_stack->sp_name);
508                 if (ret == PAGE_SIZE)
509                         ret = -E2BIG;
510         }
511         spin_unlock(&ocfs2_stack_lock);
512
513         return ret;
514 }
515
516 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
517         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
518                ocfs2_active_cluster_plugin_show, NULL);
519
520 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
521                                         struct kobj_attribute *attr,
522                                         char *buf)
523 {
524         ssize_t ret;
525         spin_lock(&ocfs2_stack_lock);
526         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
527         spin_unlock(&ocfs2_stack_lock);
528
529         return ret;
530 }
531
532 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
533                                          struct kobj_attribute *attr,
534                                          const char *buf, size_t count)
535 {
536         size_t len = count;
537         ssize_t ret;
538
539         if (len == 0)
540                 return len;
541
542         if (buf[len - 1] == '\n')
543                 len--;
544
545         if ((len != OCFS2_STACK_LABEL_LEN) ||
546             (strnlen(buf, len) != len))
547                 return -EINVAL;
548
549         spin_lock(&ocfs2_stack_lock);
550         if (active_stack) {
551                 if (!strncmp(buf, cluster_stack_name, len))
552                         ret = count;
553                 else
554                         ret = -EBUSY;
555         } else {
556                 memcpy(cluster_stack_name, buf, len);
557                 ret = count;
558         }
559         spin_unlock(&ocfs2_stack_lock);
560
561         return ret;
562 }
563
564
565 static struct kobj_attribute ocfs2_attr_cluster_stack =
566         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
567                ocfs2_cluster_stack_show,
568                ocfs2_cluster_stack_store);
569
570 static struct attribute *ocfs2_attrs[] = {
571         &ocfs2_attr_max_locking_protocol.attr,
572         &ocfs2_attr_loaded_cluster_plugins.attr,
573         &ocfs2_attr_active_cluster_plugin.attr,
574         &ocfs2_attr_cluster_stack.attr,
575         NULL,
576 };
577
578 static struct attribute_group ocfs2_attr_group = {
579         .attrs = ocfs2_attrs,
580 };
581
582 static struct kset *ocfs2_kset;
583
584 static void ocfs2_sysfs_exit(void)
585 {
586         kset_unregister(ocfs2_kset);
587 }
588
589 static int ocfs2_sysfs_init(void)
590 {
591         int ret;
592
593         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
594         if (!ocfs2_kset)
595                 return -ENOMEM;
596
597         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
598         if (ret)
599                 goto error;
600
601         return 0;
602
603 error:
604         kset_unregister(ocfs2_kset);
605         return ret;
606 }
607
608 /*
609  * Sysctl bits
610  *
611  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
612  * make as much sense in a multiple cluster stack world, but it's safer
613  * and easier to preserve the name.
614  */
615
616 #define FS_OCFS2_NM             1
617
618 static ctl_table ocfs2_nm_table[] = {
619         {
620                 .procname       = "hb_ctl_path",
621                 .data           = ocfs2_hb_ctl_path,
622                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
623                 .mode           = 0644,
624                 .proc_handler   = proc_dostring,
625         },
626         { }
627 };
628
629 static ctl_table ocfs2_mod_table[] = {
630         {
631                 .procname       = "nm",
632                 .data           = NULL,
633                 .maxlen         = 0,
634                 .mode           = 0555,
635                 .child          = ocfs2_nm_table
636         },
637         { }
638 };
639
640 static ctl_table ocfs2_kern_table[] = {
641         {
642                 .procname       = "ocfs2",
643                 .data           = NULL,
644                 .maxlen         = 0,
645                 .mode           = 0555,
646                 .child          = ocfs2_mod_table
647         },
648         { }
649 };
650
651 static ctl_table ocfs2_root_table[] = {
652         {
653                 .procname       = "fs",
654                 .data           = NULL,
655                 .maxlen         = 0,
656                 .mode           = 0555,
657                 .child          = ocfs2_kern_table
658         },
659         { }
660 };
661
662 static struct ctl_table_header *ocfs2_table_header = NULL;
663
664
665 /*
666  * Initialization
667  */
668
669 static int __init ocfs2_stack_glue_init(void)
670 {
671         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
672
673         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
674         if (!ocfs2_table_header) {
675                 printk(KERN_ERR
676                        "ocfs2 stack glue: unable to register sysctl\n");
677                 return -ENOMEM; /* or something. */
678         }
679
680         return ocfs2_sysfs_init();
681 }
682
683 static void __exit ocfs2_stack_glue_exit(void)
684 {
685         lproto = NULL;
686         ocfs2_sysfs_exit();
687         if (ocfs2_table_header)
688                 unregister_sysctl_table(ocfs2_table_header);
689 }
690
691 MODULE_AUTHOR("Oracle");
692 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
693 MODULE_LICENSE("GPL");
694 module_init(ocfs2_stack_glue_init);
695 module_exit(ocfs2_stack_glue_exit);