kvm: fix kvm reboot crash when MAXSMP is used
[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_stack_supports_plocks(void)
292 {
293         return active_stack && active_stack->sp_ops->plock;
294 }
295 EXPORT_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  */
301 int 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 }
309 EXPORT_SYMBOL_GPL(ocfs2_plock);
310
311 int ocfs2_cluster_connect(const char *stack_name,
312                           const char *group,
313                           int grouplen,
314                           void (*recovery_handler)(int node_num,
315                                                    void *recovery_data),
316                           void *recovery_data,
317                           struct ocfs2_cluster_connection **conn)
318 {
319         int rc = 0;
320         struct ocfs2_cluster_connection *new_conn;
321
322         BUG_ON(group == NULL);
323         BUG_ON(conn == NULL);
324         BUG_ON(recovery_handler == NULL);
325
326         if (grouplen > GROUP_NAME_MAX) {
327                 rc = -EINVAL;
328                 goto out;
329         }
330
331         new_conn = kzalloc(sizeof(struct ocfs2_cluster_connection),
332                            GFP_KERNEL);
333         if (!new_conn) {
334                 rc = -ENOMEM;
335                 goto out;
336         }
337
338         memcpy(new_conn->cc_name, group, grouplen);
339         new_conn->cc_namelen = grouplen;
340         new_conn->cc_recovery_handler = recovery_handler;
341         new_conn->cc_recovery_data = recovery_data;
342
343         /* Start the new connection at our maximum compatibility level */
344         new_conn->cc_version = lproto->lp_max_version;
345
346         /* This will pin the stack driver if successful */
347         rc = ocfs2_stack_driver_get(stack_name);
348         if (rc)
349                 goto out_free;
350
351         rc = active_stack->sp_ops->connect(new_conn);
352         if (rc) {
353                 ocfs2_stack_driver_put();
354                 goto out_free;
355         }
356
357         *conn = new_conn;
358
359 out_free:
360         if (rc)
361                 kfree(new_conn);
362
363 out:
364         return rc;
365 }
366 EXPORT_SYMBOL_GPL(ocfs2_cluster_connect);
367
368 /* If hangup_pending is 0, the stack driver will be dropped */
369 int ocfs2_cluster_disconnect(struct ocfs2_cluster_connection *conn,
370                              int hangup_pending)
371 {
372         int ret;
373
374         BUG_ON(conn == NULL);
375
376         ret = active_stack->sp_ops->disconnect(conn);
377
378         /* XXX Should we free it anyway? */
379         if (!ret) {
380                 kfree(conn);
381                 if (!hangup_pending)
382                         ocfs2_stack_driver_put();
383         }
384
385         return ret;
386 }
387 EXPORT_SYMBOL_GPL(ocfs2_cluster_disconnect);
388
389 /*
390  * Leave the group for this filesystem.  This is executed by a userspace
391  * program (stored in ocfs2_hb_ctl_path).
392  */
393 static void ocfs2_leave_group(const char *group)
394 {
395         int ret;
396         char *argv[5], *envp[3];
397
398         argv[0] = ocfs2_hb_ctl_path;
399         argv[1] = "-K";
400         argv[2] = "-u";
401         argv[3] = (char *)group;
402         argv[4] = NULL;
403
404         /* minimal command environment taken from cpu_run_sbin_hotplug */
405         envp[0] = "HOME=/";
406         envp[1] = "PATH=/sbin:/bin:/usr/sbin:/usr/bin";
407         envp[2] = NULL;
408
409         ret = call_usermodehelper(argv[0], argv, envp, UMH_WAIT_PROC);
410         if (ret < 0) {
411                 printk(KERN_ERR
412                        "ocfs2: Error %d running user helper "
413                        "\"%s %s %s %s\"\n",
414                        ret, argv[0], argv[1], argv[2], argv[3]);
415         }
416 }
417
418 /*
419  * Hangup is a required post-umount.  ocfs2-tools software expects the
420  * filesystem to call "ocfs2_hb_ctl" during unmount.  This happens
421  * regardless of whether the DLM got started, so we can't do it
422  * in ocfs2_cluster_disconnect().  The ocfs2_leave_group() function does
423  * the actual work.
424  */
425 void ocfs2_cluster_hangup(const char *group, int grouplen)
426 {
427         BUG_ON(group == NULL);
428         BUG_ON(group[grouplen] != '\0');
429
430         ocfs2_leave_group(group);
431
432         /* cluster_disconnect() was called with hangup_pending==1 */
433         ocfs2_stack_driver_put();
434 }
435 EXPORT_SYMBOL_GPL(ocfs2_cluster_hangup);
436
437 int ocfs2_cluster_this_node(unsigned int *node)
438 {
439         return active_stack->sp_ops->this_node(node);
440 }
441 EXPORT_SYMBOL_GPL(ocfs2_cluster_this_node);
442
443
444 /*
445  * Sysfs bits
446  */
447
448 static ssize_t ocfs2_max_locking_protocol_show(struct kobject *kobj,
449                                                struct kobj_attribute *attr,
450                                                char *buf)
451 {
452         ssize_t ret = 0;
453
454         spin_lock(&ocfs2_stack_lock);
455         if (lproto)
456                 ret = snprintf(buf, PAGE_SIZE, "%u.%u\n",
457                                lproto->lp_max_version.pv_major,
458                                lproto->lp_max_version.pv_minor);
459         spin_unlock(&ocfs2_stack_lock);
460
461         return ret;
462 }
463
464 static struct kobj_attribute ocfs2_attr_max_locking_protocol =
465         __ATTR(max_locking_protocol, S_IFREG | S_IRUGO,
466                ocfs2_max_locking_protocol_show, NULL);
467
468 static ssize_t ocfs2_loaded_cluster_plugins_show(struct kobject *kobj,
469                                                  struct kobj_attribute *attr,
470                                                  char *buf)
471 {
472         ssize_t ret = 0, total = 0, remain = PAGE_SIZE;
473         struct ocfs2_stack_plugin *p;
474
475         spin_lock(&ocfs2_stack_lock);
476         list_for_each_entry(p, &ocfs2_stack_list, sp_list) {
477                 ret = snprintf(buf, remain, "%s\n",
478                                p->sp_name);
479                 if (ret < 0) {
480                         total = ret;
481                         break;
482                 }
483                 if (ret == remain) {
484                         /* snprintf() didn't fit */
485                         total = -E2BIG;
486                         break;
487                 }
488                 total += ret;
489                 remain -= ret;
490         }
491         spin_unlock(&ocfs2_stack_lock);
492
493         return total;
494 }
495
496 static struct kobj_attribute ocfs2_attr_loaded_cluster_plugins =
497         __ATTR(loaded_cluster_plugins, S_IFREG | S_IRUGO,
498                ocfs2_loaded_cluster_plugins_show, NULL);
499
500 static ssize_t ocfs2_active_cluster_plugin_show(struct kobject *kobj,
501                                                 struct kobj_attribute *attr,
502                                                 char *buf)
503 {
504         ssize_t ret = 0;
505
506         spin_lock(&ocfs2_stack_lock);
507         if (active_stack) {
508                 ret = snprintf(buf, PAGE_SIZE, "%s\n",
509                                active_stack->sp_name);
510                 if (ret == PAGE_SIZE)
511                         ret = -E2BIG;
512         }
513         spin_unlock(&ocfs2_stack_lock);
514
515         return ret;
516 }
517
518 static struct kobj_attribute ocfs2_attr_active_cluster_plugin =
519         __ATTR(active_cluster_plugin, S_IFREG | S_IRUGO,
520                ocfs2_active_cluster_plugin_show, NULL);
521
522 static ssize_t ocfs2_cluster_stack_show(struct kobject *kobj,
523                                         struct kobj_attribute *attr,
524                                         char *buf)
525 {
526         ssize_t ret;
527         spin_lock(&ocfs2_stack_lock);
528         ret = snprintf(buf, PAGE_SIZE, "%s\n", cluster_stack_name);
529         spin_unlock(&ocfs2_stack_lock);
530
531         return ret;
532 }
533
534 static ssize_t ocfs2_cluster_stack_store(struct kobject *kobj,
535                                          struct kobj_attribute *attr,
536                                          const char *buf, size_t count)
537 {
538         size_t len = count;
539         ssize_t ret;
540
541         if (len == 0)
542                 return len;
543
544         if (buf[len - 1] == '\n')
545                 len--;
546
547         if ((len != OCFS2_STACK_LABEL_LEN) ||
548             (strnlen(buf, len) != len))
549                 return -EINVAL;
550
551         spin_lock(&ocfs2_stack_lock);
552         if (active_stack) {
553                 if (!strncmp(buf, cluster_stack_name, len))
554                         ret = count;
555                 else
556                         ret = -EBUSY;
557         } else {
558                 memcpy(cluster_stack_name, buf, len);
559                 ret = count;
560         }
561         spin_unlock(&ocfs2_stack_lock);
562
563         return ret;
564 }
565
566
567 static struct kobj_attribute ocfs2_attr_cluster_stack =
568         __ATTR(cluster_stack, S_IFREG | S_IRUGO | S_IWUSR,
569                ocfs2_cluster_stack_show,
570                ocfs2_cluster_stack_store);
571
572 static struct attribute *ocfs2_attrs[] = {
573         &ocfs2_attr_max_locking_protocol.attr,
574         &ocfs2_attr_loaded_cluster_plugins.attr,
575         &ocfs2_attr_active_cluster_plugin.attr,
576         &ocfs2_attr_cluster_stack.attr,
577         NULL,
578 };
579
580 static struct attribute_group ocfs2_attr_group = {
581         .attrs = ocfs2_attrs,
582 };
583
584 static struct kset *ocfs2_kset;
585
586 static void ocfs2_sysfs_exit(void)
587 {
588         kset_unregister(ocfs2_kset);
589 }
590
591 static int ocfs2_sysfs_init(void)
592 {
593         int ret;
594
595         ocfs2_kset = kset_create_and_add("ocfs2", NULL, fs_kobj);
596         if (!ocfs2_kset)
597                 return -ENOMEM;
598
599         ret = sysfs_create_group(&ocfs2_kset->kobj, &ocfs2_attr_group);
600         if (ret)
601                 goto error;
602
603         return 0;
604
605 error:
606         kset_unregister(ocfs2_kset);
607         return ret;
608 }
609
610 /*
611  * Sysctl bits
612  *
613  * The sysctl lives at /proc/sys/fs/ocfs2/nm/hb_ctl_path.  The 'nm' doesn't
614  * make as much sense in a multiple cluster stack world, but it's safer
615  * and easier to preserve the name.
616  */
617
618 #define FS_OCFS2_NM             1
619
620 static ctl_table ocfs2_nm_table[] = {
621         {
622                 .ctl_name       = 1,
623                 .procname       = "hb_ctl_path",
624                 .data           = ocfs2_hb_ctl_path,
625                 .maxlen         = OCFS2_MAX_HB_CTL_PATH,
626                 .mode           = 0644,
627                 .proc_handler   = &proc_dostring,
628                 .strategy       = &sysctl_string,
629         },
630         { .ctl_name = 0 }
631 };
632
633 static ctl_table ocfs2_mod_table[] = {
634         {
635                 .ctl_name       = FS_OCFS2_NM,
636                 .procname       = "nm",
637                 .data           = NULL,
638                 .maxlen         = 0,
639                 .mode           = 0555,
640                 .child          = ocfs2_nm_table
641         },
642         { .ctl_name = 0}
643 };
644
645 static ctl_table ocfs2_kern_table[] = {
646         {
647                 .ctl_name       = FS_OCFS2,
648                 .procname       = "ocfs2",
649                 .data           = NULL,
650                 .maxlen         = 0,
651                 .mode           = 0555,
652                 .child          = ocfs2_mod_table
653         },
654         { .ctl_name = 0}
655 };
656
657 static ctl_table ocfs2_root_table[] = {
658         {
659                 .ctl_name       = CTL_FS,
660                 .procname       = "fs",
661                 .data           = NULL,
662                 .maxlen         = 0,
663                 .mode           = 0555,
664                 .child          = ocfs2_kern_table
665         },
666         { .ctl_name = 0 }
667 };
668
669 static struct ctl_table_header *ocfs2_table_header = NULL;
670
671
672 /*
673  * Initialization
674  */
675
676 static int __init ocfs2_stack_glue_init(void)
677 {
678         strcpy(cluster_stack_name, OCFS2_STACK_PLUGIN_O2CB);
679
680         ocfs2_table_header = register_sysctl_table(ocfs2_root_table);
681         if (!ocfs2_table_header) {
682                 printk(KERN_ERR
683                        "ocfs2 stack glue: unable to register sysctl\n");
684                 return -ENOMEM; /* or something. */
685         }
686
687         return ocfs2_sysfs_init();
688 }
689
690 static void __exit ocfs2_stack_glue_exit(void)
691 {
692         lproto = NULL;
693         ocfs2_sysfs_exit();
694         if (ocfs2_table_header)
695                 unregister_sysctl_table(ocfs2_table_header);
696 }
697
698 MODULE_AUTHOR("Oracle");
699 MODULE_DESCRIPTION("ocfs2 cluter stack glue layer");
700 MODULE_LICENSE("GPL");
701 module_init(ocfs2_stack_glue_init);
702 module_exit(ocfs2_stack_glue_exit);