Merge branch 'topic/hda' into for-linus
[linux-2.6] / fs / ocfs2 / stack_user.c
1 /* -*- mode: c; c-basic-offset: 8; -*-
2  * vim: noexpandtab sw=8 ts=8 sts=0:
3  *
4  * stack_user.c
5  *
6  * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
7  *
8  * Copyright (C) 2007 Oracle.  All rights reserved.
9  *
10  * This program is free software; you can redistribute it and/or
11  * modify it under the terms of the GNU General Public
12  * License as published by the Free Software Foundation, version 2.
13  *
14  * This program is distributed in the hope that it will be useful,
15  * but WITHOUT ANY WARRANTY; without even the implied warranty of
16  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
17  * General Public License for more details.
18  */
19
20 #include <linux/module.h>
21 #include <linux/fs.h>
22 #include <linux/miscdevice.h>
23 #include <linux/mutex.h>
24 #include <linux/smp_lock.h>
25 #include <linux/reboot.h>
26 #include <asm/uaccess.h>
27
28 #include "ocfs2.h"  /* For struct ocfs2_lock_res */
29 #include "stackglue.h"
30
31 #include <linux/dlm_plock.h>
32
33 /*
34  * The control protocol starts with a handshake.  Until the handshake
35  * is complete, the control device will fail all write(2)s.
36  *
37  * The handshake is simple.  First, the client reads until EOF.  Each line
38  * of output is a supported protocol tag.  All protocol tags are a single
39  * character followed by a two hex digit version number.  Currently the
40  * only things supported is T01, for "Text-base version 0x01".  Next, the
41  * client writes the version they would like to use, including the newline.
42  * Thus, the protocol tag is 'T01\n'.  If the version tag written is
43  * unknown, -EINVAL is returned.  Once the negotiation is complete, the
44  * client can start sending messages.
45  *
46  * The T01 protocol has three messages.  First is the "SETN" message.
47  * It has the following syntax:
48  *
49  *  SETN<space><8-char-hex-nodenum><newline>
50  *
51  * This is 14 characters.
52  *
53  * The "SETN" message must be the first message following the protocol.
54  * It tells ocfs2_control the local node number.
55  *
56  * Next comes the "SETV" message.  It has the following syntax:
57  *
58  *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
59  *
60  * This is 11 characters.
61  *
62  * The "SETV" message sets the filesystem locking protocol version as
63  * negotiated by the client.  The client negotiates based on the maximum
64  * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
65  * number from the "SETV" message must match
66  * ocfs2_user_plugin.sp_proto->lp_max_version.pv_major, and the minor number
67  * must be less than or equal to ...->lp_max_version.pv_minor.
68  *
69  * Once this information has been set, mounts will be allowed.  From this
70  * point on, the "DOWN" message can be sent for node down notification.
71  * It has the following syntax:
72  *
73  *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
74  *
75  * eg:
76  *
77  *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
78  *
79  * This is 47 characters.
80  */
81
82 /*
83  * Whether or not the client has done the handshake.
84  * For now, we have just one protocol version.
85  */
86 #define OCFS2_CONTROL_PROTO                     "T01\n"
87 #define OCFS2_CONTROL_PROTO_LEN                 4
88
89 /* Handshake states */
90 #define OCFS2_CONTROL_HANDSHAKE_INVALID         (0)
91 #define OCFS2_CONTROL_HANDSHAKE_READ            (1)
92 #define OCFS2_CONTROL_HANDSHAKE_PROTOCOL        (2)
93 #define OCFS2_CONTROL_HANDSHAKE_VALID           (3)
94
95 /* Messages */
96 #define OCFS2_CONTROL_MESSAGE_OP_LEN            4
97 #define OCFS2_CONTROL_MESSAGE_SETNODE_OP        "SETN"
98 #define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN 14
99 #define OCFS2_CONTROL_MESSAGE_SETVERSION_OP     "SETV"
100 #define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN      11
101 #define OCFS2_CONTROL_MESSAGE_DOWN_OP           "DOWN"
102 #define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN    47
103 #define OCFS2_TEXT_UUID_LEN                     32
104 #define OCFS2_CONTROL_MESSAGE_VERNUM_LEN        2
105 #define OCFS2_CONTROL_MESSAGE_NODENUM_LEN       8
106
107 /*
108  * ocfs2_live_connection is refcounted because the filesystem and
109  * miscdevice sides can detach in different order.  Let's just be safe.
110  */
111 struct ocfs2_live_connection {
112         struct list_head                oc_list;
113         struct ocfs2_cluster_connection *oc_conn;
114 };
115
116 struct ocfs2_control_private {
117         struct list_head op_list;
118         int op_state;
119         int op_this_node;
120         struct ocfs2_protocol_version op_proto;
121 };
122
123 /* SETN<space><8-char-hex-nodenum><newline> */
124 struct ocfs2_control_message_setn {
125         char    tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
126         char    space;
127         char    nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
128         char    newline;
129 };
130
131 /* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
132 struct ocfs2_control_message_setv {
133         char    tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
134         char    space1;
135         char    major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
136         char    space2;
137         char    minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
138         char    newline;
139 };
140
141 /* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
142 struct ocfs2_control_message_down {
143         char    tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
144         char    space1;
145         char    uuid[OCFS2_TEXT_UUID_LEN];
146         char    space2;
147         char    nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
148         char    newline;
149 };
150
151 union ocfs2_control_message {
152         char                                    tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
153         struct ocfs2_control_message_setn       u_setn;
154         struct ocfs2_control_message_setv       u_setv;
155         struct ocfs2_control_message_down       u_down;
156 };
157
158 static struct ocfs2_stack_plugin ocfs2_user_plugin;
159
160 static atomic_t ocfs2_control_opened;
161 static int ocfs2_control_this_node = -1;
162 static struct ocfs2_protocol_version running_proto;
163
164 static LIST_HEAD(ocfs2_live_connection_list);
165 static LIST_HEAD(ocfs2_control_private_list);
166 static DEFINE_MUTEX(ocfs2_control_lock);
167
168 static inline void ocfs2_control_set_handshake_state(struct file *file,
169                                                      int state)
170 {
171         struct ocfs2_control_private *p = file->private_data;
172         p->op_state = state;
173 }
174
175 static inline int ocfs2_control_get_handshake_state(struct file *file)
176 {
177         struct ocfs2_control_private *p = file->private_data;
178         return p->op_state;
179 }
180
181 static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
182 {
183         size_t len = strlen(name);
184         struct ocfs2_live_connection *c;
185
186         BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
187
188         list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
189                 if ((c->oc_conn->cc_namelen == len) &&
190                     !strncmp(c->oc_conn->cc_name, name, len))
191                         return c;
192         }
193
194         return c;
195 }
196
197 /*
198  * ocfs2_live_connection structures are created underneath the ocfs2
199  * mount path.  Since the VFS prevents multiple calls to
200  * fill_super(), we can't get dupes here.
201  */
202 static int ocfs2_live_connection_new(struct ocfs2_cluster_connection *conn,
203                                      struct ocfs2_live_connection **c_ret)
204 {
205         int rc = 0;
206         struct ocfs2_live_connection *c;
207
208         c = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
209         if (!c)
210                 return -ENOMEM;
211
212         mutex_lock(&ocfs2_control_lock);
213         c->oc_conn = conn;
214
215         if (atomic_read(&ocfs2_control_opened))
216                 list_add(&c->oc_list, &ocfs2_live_connection_list);
217         else {
218                 printk(KERN_ERR
219                        "ocfs2: Userspace control daemon is not present\n");
220                 rc = -ESRCH;
221         }
222
223         mutex_unlock(&ocfs2_control_lock);
224
225         if (!rc)
226                 *c_ret = c;
227         else
228                 kfree(c);
229
230         return rc;
231 }
232
233 /*
234  * This function disconnects the cluster connection from ocfs2_control.
235  * Afterwards, userspace can't affect the cluster connection.
236  */
237 static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
238 {
239         mutex_lock(&ocfs2_control_lock);
240         list_del_init(&c->oc_list);
241         c->oc_conn = NULL;
242         mutex_unlock(&ocfs2_control_lock);
243
244         kfree(c);
245 }
246
247 static int ocfs2_control_cfu(void *target, size_t target_len,
248                              const char __user *buf, size_t count)
249 {
250         /* The T01 expects write(2) calls to have exactly one command */
251         if ((count != target_len) ||
252             (count > sizeof(union ocfs2_control_message)))
253                 return -EINVAL;
254
255         if (copy_from_user(target, buf, target_len))
256                 return -EFAULT;
257
258         return 0;
259 }
260
261 static ssize_t ocfs2_control_validate_protocol(struct file *file,
262                                                const char __user *buf,
263                                                size_t count)
264 {
265         ssize_t ret;
266         char kbuf[OCFS2_CONTROL_PROTO_LEN];
267
268         ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
269                                 buf, count);
270         if (ret)
271                 return ret;
272
273         if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
274                 return -EINVAL;
275
276         ocfs2_control_set_handshake_state(file,
277                                           OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
278
279         return count;
280 }
281
282 static void ocfs2_control_send_down(const char *uuid,
283                                     int nodenum)
284 {
285         struct ocfs2_live_connection *c;
286
287         mutex_lock(&ocfs2_control_lock);
288
289         c = ocfs2_connection_find(uuid);
290         if (c) {
291                 BUG_ON(c->oc_conn == NULL);
292                 c->oc_conn->cc_recovery_handler(nodenum,
293                                                 c->oc_conn->cc_recovery_data);
294         }
295
296         mutex_unlock(&ocfs2_control_lock);
297 }
298
299 /*
300  * Called whenever configuration elements are sent to /dev/ocfs2_control.
301  * If all configuration elements are present, try to set the global
302  * values.  If there is a problem, return an error.  Skip any missing
303  * elements, and only bump ocfs2_control_opened when we have all elements
304  * and are successful.
305  */
306 static int ocfs2_control_install_private(struct file *file)
307 {
308         int rc = 0;
309         int set_p = 1;
310         struct ocfs2_control_private *p = file->private_data;
311
312         BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
313
314         mutex_lock(&ocfs2_control_lock);
315
316         if (p->op_this_node < 0) {
317                 set_p = 0;
318         } else if ((ocfs2_control_this_node >= 0) &&
319                    (ocfs2_control_this_node != p->op_this_node)) {
320                 rc = -EINVAL;
321                 goto out_unlock;
322         }
323
324         if (!p->op_proto.pv_major) {
325                 set_p = 0;
326         } else if (!list_empty(&ocfs2_live_connection_list) &&
327                    ((running_proto.pv_major != p->op_proto.pv_major) ||
328                     (running_proto.pv_minor != p->op_proto.pv_minor))) {
329                 rc = -EINVAL;
330                 goto out_unlock;
331         }
332
333         if (set_p) {
334                 ocfs2_control_this_node = p->op_this_node;
335                 running_proto.pv_major = p->op_proto.pv_major;
336                 running_proto.pv_minor = p->op_proto.pv_minor;
337         }
338
339 out_unlock:
340         mutex_unlock(&ocfs2_control_lock);
341
342         if (!rc && set_p) {
343                 /* We set the global values successfully */
344                 atomic_inc(&ocfs2_control_opened);
345                 ocfs2_control_set_handshake_state(file,
346                                         OCFS2_CONTROL_HANDSHAKE_VALID);
347         }
348
349         return rc;
350 }
351
352 static int ocfs2_control_get_this_node(void)
353 {
354         int rc;
355
356         mutex_lock(&ocfs2_control_lock);
357         if (ocfs2_control_this_node < 0)
358                 rc = -EINVAL;
359         else
360                 rc = ocfs2_control_this_node;
361         mutex_unlock(&ocfs2_control_lock);
362
363         return rc;
364 }
365
366 static int ocfs2_control_do_setnode_msg(struct file *file,
367                                         struct ocfs2_control_message_setn *msg)
368 {
369         long nodenum;
370         char *ptr = NULL;
371         struct ocfs2_control_private *p = file->private_data;
372
373         if (ocfs2_control_get_handshake_state(file) !=
374             OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
375                 return -EINVAL;
376
377         if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
378                     OCFS2_CONTROL_MESSAGE_OP_LEN))
379                 return -EINVAL;
380
381         if ((msg->space != ' ') || (msg->newline != '\n'))
382                 return -EINVAL;
383         msg->space = msg->newline = '\0';
384
385         nodenum = simple_strtol(msg->nodestr, &ptr, 16);
386         if (!ptr || *ptr)
387                 return -EINVAL;
388
389         if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
390             (nodenum > INT_MAX) || (nodenum < 0))
391                 return -ERANGE;
392         p->op_this_node = nodenum;
393
394         return ocfs2_control_install_private(file);
395 }
396
397 static int ocfs2_control_do_setversion_msg(struct file *file,
398                                            struct ocfs2_control_message_setv *msg)
399  {
400         long major, minor;
401         char *ptr = NULL;
402         struct ocfs2_control_private *p = file->private_data;
403         struct ocfs2_protocol_version *max =
404                 &ocfs2_user_plugin.sp_proto->lp_max_version;
405
406         if (ocfs2_control_get_handshake_state(file) !=
407             OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
408                 return -EINVAL;
409
410         if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
411                     OCFS2_CONTROL_MESSAGE_OP_LEN))
412                 return -EINVAL;
413
414         if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
415             (msg->newline != '\n'))
416                 return -EINVAL;
417         msg->space1 = msg->space2 = msg->newline = '\0';
418
419         major = simple_strtol(msg->major, &ptr, 16);
420         if (!ptr || *ptr)
421                 return -EINVAL;
422         minor = simple_strtol(msg->minor, &ptr, 16);
423         if (!ptr || *ptr)
424                 return -EINVAL;
425
426         /*
427          * The major must be between 1 and 255, inclusive.  The minor
428          * must be between 0 and 255, inclusive.  The version passed in
429          * must be within the maximum version supported by the filesystem.
430          */
431         if ((major == LONG_MIN) || (major == LONG_MAX) ||
432             (major > (u8)-1) || (major < 1))
433                 return -ERANGE;
434         if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
435             (minor > (u8)-1) || (minor < 0))
436                 return -ERANGE;
437         if ((major != max->pv_major) ||
438             (minor > max->pv_minor))
439                 return -EINVAL;
440
441         p->op_proto.pv_major = major;
442         p->op_proto.pv_minor = minor;
443
444         return ocfs2_control_install_private(file);
445 }
446
447 static int ocfs2_control_do_down_msg(struct file *file,
448                                      struct ocfs2_control_message_down *msg)
449 {
450         long nodenum;
451         char *p = NULL;
452
453         if (ocfs2_control_get_handshake_state(file) !=
454             OCFS2_CONTROL_HANDSHAKE_VALID)
455                 return -EINVAL;
456
457         if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
458                     OCFS2_CONTROL_MESSAGE_OP_LEN))
459                 return -EINVAL;
460
461         if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
462             (msg->newline != '\n'))
463                 return -EINVAL;
464         msg->space1 = msg->space2 = msg->newline = '\0';
465
466         nodenum = simple_strtol(msg->nodestr, &p, 16);
467         if (!p || *p)
468                 return -EINVAL;
469
470         if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
471             (nodenum > INT_MAX) || (nodenum < 0))
472                 return -ERANGE;
473
474         ocfs2_control_send_down(msg->uuid, nodenum);
475
476         return 0;
477 }
478
479 static ssize_t ocfs2_control_message(struct file *file,
480                                      const char __user *buf,
481                                      size_t count)
482 {
483         ssize_t ret;
484         union ocfs2_control_message msg;
485
486         /* Try to catch padding issues */
487         WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
488                 (sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
489
490         memset(&msg, 0, sizeof(union ocfs2_control_message));
491         ret = ocfs2_control_cfu(&msg, count, buf, count);
492         if (ret)
493                 goto out;
494
495         if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
496             !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
497                      OCFS2_CONTROL_MESSAGE_OP_LEN))
498                 ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
499         else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
500                  !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
501                           OCFS2_CONTROL_MESSAGE_OP_LEN))
502                 ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
503         else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
504                  !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
505                           OCFS2_CONTROL_MESSAGE_OP_LEN))
506                 ret = ocfs2_control_do_down_msg(file, &msg.u_down);
507         else
508                 ret = -EINVAL;
509
510 out:
511         return ret ? ret : count;
512 }
513
514 static ssize_t ocfs2_control_write(struct file *file,
515                                    const char __user *buf,
516                                    size_t count,
517                                    loff_t *ppos)
518 {
519         ssize_t ret;
520
521         switch (ocfs2_control_get_handshake_state(file)) {
522                 case OCFS2_CONTROL_HANDSHAKE_INVALID:
523                         ret = -EINVAL;
524                         break;
525
526                 case OCFS2_CONTROL_HANDSHAKE_READ:
527                         ret = ocfs2_control_validate_protocol(file, buf,
528                                                               count);
529                         break;
530
531                 case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
532                 case OCFS2_CONTROL_HANDSHAKE_VALID:
533                         ret = ocfs2_control_message(file, buf, count);
534                         break;
535
536                 default:
537                         BUG();
538                         ret = -EIO;
539                         break;
540         }
541
542         return ret;
543 }
544
545 /*
546  * This is a naive version.  If we ever have a new protocol, we'll expand
547  * it.  Probably using seq_file.
548  */
549 static ssize_t ocfs2_control_read(struct file *file,
550                                   char __user *buf,
551                                   size_t count,
552                                   loff_t *ppos)
553 {
554         ssize_t ret;
555
556         ret = simple_read_from_buffer(buf, count, ppos,
557                         OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
558
559         /* Have we read the whole protocol list? */
560         if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
561                 ocfs2_control_set_handshake_state(file,
562                                                   OCFS2_CONTROL_HANDSHAKE_READ);
563
564         return ret;
565 }
566
567 static int ocfs2_control_release(struct inode *inode, struct file *file)
568 {
569         struct ocfs2_control_private *p = file->private_data;
570
571         mutex_lock(&ocfs2_control_lock);
572
573         if (ocfs2_control_get_handshake_state(file) !=
574             OCFS2_CONTROL_HANDSHAKE_VALID)
575                 goto out;
576
577         if (atomic_dec_and_test(&ocfs2_control_opened)) {
578                 if (!list_empty(&ocfs2_live_connection_list)) {
579                         /* XXX: Do bad things! */
580                         printk(KERN_ERR
581                                "ocfs2: Unexpected release of ocfs2_control!\n"
582                                "       Loss of cluster connection requires "
583                                "an emergency restart!\n");
584                         emergency_restart();
585                 }
586                 /*
587                  * Last valid close clears the node number and resets
588                  * the locking protocol version
589                  */
590                 ocfs2_control_this_node = -1;
591                 running_proto.pv_major = 0;
592                 running_proto.pv_major = 0;
593         }
594
595 out:
596         list_del_init(&p->op_list);
597         file->private_data = NULL;
598
599         mutex_unlock(&ocfs2_control_lock);
600
601         kfree(p);
602
603         return 0;
604 }
605
606 static int ocfs2_control_open(struct inode *inode, struct file *file)
607 {
608         struct ocfs2_control_private *p;
609
610         p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
611         if (!p)
612                 return -ENOMEM;
613         p->op_this_node = -1;
614
615         lock_kernel();
616         mutex_lock(&ocfs2_control_lock);
617         file->private_data = p;
618         list_add(&p->op_list, &ocfs2_control_private_list);
619         mutex_unlock(&ocfs2_control_lock);
620         unlock_kernel();
621
622         return 0;
623 }
624
625 static const struct file_operations ocfs2_control_fops = {
626         .open    = ocfs2_control_open,
627         .release = ocfs2_control_release,
628         .read    = ocfs2_control_read,
629         .write   = ocfs2_control_write,
630         .owner   = THIS_MODULE,
631 };
632
633 static struct miscdevice ocfs2_control_device = {
634         .minor          = MISC_DYNAMIC_MINOR,
635         .name           = "ocfs2_control",
636         .fops           = &ocfs2_control_fops,
637 };
638
639 static int ocfs2_control_init(void)
640 {
641         int rc;
642
643         atomic_set(&ocfs2_control_opened, 0);
644
645         rc = misc_register(&ocfs2_control_device);
646         if (rc)
647                 printk(KERN_ERR
648                        "ocfs2: Unable to register ocfs2_control device "
649                        "(errno %d)\n",
650                        -rc);
651
652         return rc;
653 }
654
655 static void ocfs2_control_exit(void)
656 {
657         int rc;
658
659         rc = misc_deregister(&ocfs2_control_device);
660         if (rc)
661                 printk(KERN_ERR
662                        "ocfs2: Unable to deregister ocfs2_control device "
663                        "(errno %d)\n",
664                        -rc);
665 }
666
667 static struct dlm_lksb *fsdlm_astarg_to_lksb(void *astarg)
668 {
669         struct ocfs2_lock_res *res = astarg;
670         return &res->l_lksb.lksb_fsdlm;
671 }
672
673 static void fsdlm_lock_ast_wrapper(void *astarg)
674 {
675         struct dlm_lksb *lksb = fsdlm_astarg_to_lksb(astarg);
676         int status = lksb->sb_status;
677
678         BUG_ON(ocfs2_user_plugin.sp_proto == NULL);
679
680         /*
681          * For now we're punting on the issue of other non-standard errors
682          * where we can't tell if the unlock_ast or lock_ast should be called.
683          * The main "other error" that's possible is EINVAL which means the
684          * function was called with invalid args, which shouldn't be possible
685          * since the caller here is under our control.  Other non-standard
686          * errors probably fall into the same category, or otherwise are fatal
687          * which means we can't carry on anyway.
688          */
689
690         if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
691                 ocfs2_user_plugin.sp_proto->lp_unlock_ast(astarg, 0);
692         else
693                 ocfs2_user_plugin.sp_proto->lp_lock_ast(astarg);
694 }
695
696 static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
697 {
698         BUG_ON(ocfs2_user_plugin.sp_proto == NULL);
699
700         ocfs2_user_plugin.sp_proto->lp_blocking_ast(astarg, level);
701 }
702
703 static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
704                          int mode,
705                          union ocfs2_dlm_lksb *lksb,
706                          u32 flags,
707                          void *name,
708                          unsigned int namelen,
709                          void *astarg)
710 {
711         int ret;
712
713         if (!lksb->lksb_fsdlm.sb_lvbptr)
714                 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
715                                              sizeof(struct dlm_lksb);
716
717         ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
718                        flags|DLM_LKF_NODLCKWT, name, namelen, 0,
719                        fsdlm_lock_ast_wrapper, astarg,
720                        fsdlm_blocking_ast_wrapper);
721         return ret;
722 }
723
724 static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
725                            union ocfs2_dlm_lksb *lksb,
726                            u32 flags,
727                            void *astarg)
728 {
729         int ret;
730
731         ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
732                          flags, &lksb->lksb_fsdlm, astarg);
733         return ret;
734 }
735
736 static int user_dlm_lock_status(union ocfs2_dlm_lksb *lksb)
737 {
738         return lksb->lksb_fsdlm.sb_status;
739 }
740
741 static void *user_dlm_lvb(union ocfs2_dlm_lksb *lksb)
742 {
743         if (!lksb->lksb_fsdlm.sb_lvbptr)
744                 lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
745                                              sizeof(struct dlm_lksb);
746         return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
747 }
748
749 static void user_dlm_dump_lksb(union ocfs2_dlm_lksb *lksb)
750 {
751 }
752
753 static int user_plock(struct ocfs2_cluster_connection *conn,
754                       u64 ino,
755                       struct file *file,
756                       int cmd,
757                       struct file_lock *fl)
758 {
759         /*
760          * This more or less just demuxes the plock request into any
761          * one of three dlm calls.
762          *
763          * Internally, fs/dlm will pass these to a misc device, which
764          * a userspace daemon will read and write to.
765          *
766          * For now, cancel requests (which happen internally only),
767          * are turned into unlocks. Most of this function taken from
768          * gfs2_lock.
769          */
770
771         if (cmd == F_CANCELLK) {
772                 cmd = F_SETLK;
773                 fl->fl_type = F_UNLCK;
774         }
775
776         if (IS_GETLK(cmd))
777                 return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
778         else if (fl->fl_type == F_UNLCK)
779                 return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
780         else
781                 return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
782 }
783
784 /*
785  * Compare a requested locking protocol version against the current one.
786  *
787  * If the major numbers are different, they are incompatible.
788  * If the current minor is greater than the request, they are incompatible.
789  * If the current minor is less than or equal to the request, they are
790  * compatible, and the requester should run at the current minor version.
791  */
792 static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
793                                struct ocfs2_protocol_version *request)
794 {
795         if (existing->pv_major != request->pv_major)
796                 return 1;
797
798         if (existing->pv_minor > request->pv_minor)
799                 return 1;
800
801         if (existing->pv_minor < request->pv_minor)
802                 request->pv_minor = existing->pv_minor;
803
804         return 0;
805 }
806
807 static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
808 {
809         dlm_lockspace_t *fsdlm;
810         struct ocfs2_live_connection *control;
811         int rc = 0;
812
813         BUG_ON(conn == NULL);
814
815         rc = ocfs2_live_connection_new(conn, &control);
816         if (rc)
817                 goto out;
818
819         /*
820          * running_proto must have been set before we allowed any mounts
821          * to proceed.
822          */
823         if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
824                 printk(KERN_ERR
825                        "Unable to mount with fs locking protocol version "
826                        "%u.%u because the userspace control daemon has "
827                        "negotiated %u.%u\n",
828                        conn->cc_version.pv_major, conn->cc_version.pv_minor,
829                        running_proto.pv_major, running_proto.pv_minor);
830                 rc = -EPROTO;
831                 ocfs2_live_connection_drop(control);
832                 goto out;
833         }
834
835         rc = dlm_new_lockspace(conn->cc_name, strlen(conn->cc_name),
836                                &fsdlm, DLM_LSFL_FS, DLM_LVB_LEN);
837         if (rc) {
838                 ocfs2_live_connection_drop(control);
839                 goto out;
840         }
841
842         conn->cc_private = control;
843         conn->cc_lockspace = fsdlm;
844 out:
845         return rc;
846 }
847
848 static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
849 {
850         dlm_release_lockspace(conn->cc_lockspace, 2);
851         conn->cc_lockspace = NULL;
852         ocfs2_live_connection_drop(conn->cc_private);
853         conn->cc_private = NULL;
854         return 0;
855 }
856
857 static int user_cluster_this_node(unsigned int *this_node)
858 {
859         int rc;
860
861         rc = ocfs2_control_get_this_node();
862         if (rc < 0)
863                 return rc;
864
865         *this_node = rc;
866         return 0;
867 }
868
869 static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
870         .connect        = user_cluster_connect,
871         .disconnect     = user_cluster_disconnect,
872         .this_node      = user_cluster_this_node,
873         .dlm_lock       = user_dlm_lock,
874         .dlm_unlock     = user_dlm_unlock,
875         .lock_status    = user_dlm_lock_status,
876         .lock_lvb       = user_dlm_lvb,
877         .plock          = user_plock,
878         .dump_lksb      = user_dlm_dump_lksb,
879 };
880
881 static struct ocfs2_stack_plugin ocfs2_user_plugin = {
882         .sp_name        = "user",
883         .sp_ops         = &ocfs2_user_plugin_ops,
884         .sp_owner       = THIS_MODULE,
885 };
886
887
888 static int __init ocfs2_user_plugin_init(void)
889 {
890         int rc;
891
892         rc = ocfs2_control_init();
893         if (!rc) {
894                 rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
895                 if (rc)
896                         ocfs2_control_exit();
897         }
898
899         return rc;
900 }
901
902 static void __exit ocfs2_user_plugin_exit(void)
903 {
904         ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
905         ocfs2_control_exit();
906 }
907
908 MODULE_AUTHOR("Oracle");
909 MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
910 MODULE_LICENSE("GPL");
911 module_init(ocfs2_user_plugin_init);
912 module_exit(ocfs2_user_plugin_exit);