xref: /openbmc/linux/fs/ocfs2/stack_user.c (revision 63dc02bd)
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/slab.h>
25 #include <linux/reboot.h>
26 #include <asm/uaccess.h>
27 
28 #include "stackglue.h"
29 
30 #include <linux/dlm_plock.h>
31 
32 /*
33  * The control protocol starts with a handshake.  Until the handshake
34  * is complete, the control device will fail all write(2)s.
35  *
36  * The handshake is simple.  First, the client reads until EOF.  Each line
37  * of output is a supported protocol tag.  All protocol tags are a single
38  * character followed by a two hex digit version number.  Currently the
39  * only things supported is T01, for "Text-base version 0x01".  Next, the
40  * client writes the version they would like to use, including the newline.
41  * Thus, the protocol tag is 'T01\n'.  If the version tag written is
42  * unknown, -EINVAL is returned.  Once the negotiation is complete, the
43  * client can start sending messages.
44  *
45  * The T01 protocol has three messages.  First is the "SETN" message.
46  * It has the following syntax:
47  *
48  *  SETN<space><8-char-hex-nodenum><newline>
49  *
50  * This is 14 characters.
51  *
52  * The "SETN" message must be the first message following the protocol.
53  * It tells ocfs2_control the local node number.
54  *
55  * Next comes the "SETV" message.  It has the following syntax:
56  *
57  *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
58  *
59  * This is 11 characters.
60  *
61  * The "SETV" message sets the filesystem locking protocol version as
62  * negotiated by the client.  The client negotiates based on the maximum
63  * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
64  * number from the "SETV" message must match
65  * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
66  * must be less than or equal to ...sp_max_version.pv_minor.
67  *
68  * Once this information has been set, mounts will be allowed.  From this
69  * point on, the "DOWN" message can be sent for node down notification.
70  * It has the following syntax:
71  *
72  *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
73  *
74  * eg:
75  *
76  *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
77  *
78  * This is 47 characters.
79  */
80 
81 /*
82  * Whether or not the client has done the handshake.
83  * For now, we have just one protocol version.
84  */
85 #define OCFS2_CONTROL_PROTO			"T01\n"
86 #define OCFS2_CONTROL_PROTO_LEN			4
87 
88 /* Handshake states */
89 #define OCFS2_CONTROL_HANDSHAKE_INVALID		(0)
90 #define OCFS2_CONTROL_HANDSHAKE_READ		(1)
91 #define OCFS2_CONTROL_HANDSHAKE_PROTOCOL	(2)
92 #define OCFS2_CONTROL_HANDSHAKE_VALID		(3)
93 
94 /* Messages */
95 #define OCFS2_CONTROL_MESSAGE_OP_LEN		4
96 #define OCFS2_CONTROL_MESSAGE_SETNODE_OP	"SETN"
97 #define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN	14
98 #define OCFS2_CONTROL_MESSAGE_SETVERSION_OP	"SETV"
99 #define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN	11
100 #define OCFS2_CONTROL_MESSAGE_DOWN_OP		"DOWN"
101 #define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN	47
102 #define OCFS2_TEXT_UUID_LEN			32
103 #define OCFS2_CONTROL_MESSAGE_VERNUM_LEN	2
104 #define OCFS2_CONTROL_MESSAGE_NODENUM_LEN	8
105 
106 /*
107  * ocfs2_live_connection is refcounted because the filesystem and
108  * miscdevice sides can detach in different order.  Let's just be safe.
109  */
110 struct ocfs2_live_connection {
111 	struct list_head		oc_list;
112 	struct ocfs2_cluster_connection	*oc_conn;
113 };
114 
115 struct ocfs2_control_private {
116 	struct list_head op_list;
117 	int op_state;
118 	int op_this_node;
119 	struct ocfs2_protocol_version op_proto;
120 };
121 
122 /* SETN<space><8-char-hex-nodenum><newline> */
123 struct ocfs2_control_message_setn {
124 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
125 	char	space;
126 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
127 	char	newline;
128 };
129 
130 /* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
131 struct ocfs2_control_message_setv {
132 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
133 	char	space1;
134 	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
135 	char	space2;
136 	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
137 	char	newline;
138 };
139 
140 /* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
141 struct ocfs2_control_message_down {
142 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
143 	char	space1;
144 	char	uuid[OCFS2_TEXT_UUID_LEN];
145 	char	space2;
146 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
147 	char	newline;
148 };
149 
150 union ocfs2_control_message {
151 	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
152 	struct ocfs2_control_message_setn	u_setn;
153 	struct ocfs2_control_message_setv	u_setv;
154 	struct ocfs2_control_message_down	u_down;
155 };
156 
157 static struct ocfs2_stack_plugin ocfs2_user_plugin;
158 
159 static atomic_t ocfs2_control_opened;
160 static int ocfs2_control_this_node = -1;
161 static struct ocfs2_protocol_version running_proto;
162 
163 static LIST_HEAD(ocfs2_live_connection_list);
164 static LIST_HEAD(ocfs2_control_private_list);
165 static DEFINE_MUTEX(ocfs2_control_lock);
166 
167 static inline void ocfs2_control_set_handshake_state(struct file *file,
168 						     int state)
169 {
170 	struct ocfs2_control_private *p = file->private_data;
171 	p->op_state = state;
172 }
173 
174 static inline int ocfs2_control_get_handshake_state(struct file *file)
175 {
176 	struct ocfs2_control_private *p = file->private_data;
177 	return p->op_state;
178 }
179 
180 static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
181 {
182 	size_t len = strlen(name);
183 	struct ocfs2_live_connection *c;
184 
185 	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
186 
187 	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
188 		if ((c->oc_conn->cc_namelen == len) &&
189 		    !strncmp(c->oc_conn->cc_name, name, len))
190 			return c;
191 	}
192 
193 	return NULL;
194 }
195 
196 /*
197  * ocfs2_live_connection structures are created underneath the ocfs2
198  * mount path.  Since the VFS prevents multiple calls to
199  * fill_super(), we can't get dupes here.
200  */
201 static int ocfs2_live_connection_new(struct ocfs2_cluster_connection *conn,
202 				     struct ocfs2_live_connection **c_ret)
203 {
204 	int rc = 0;
205 	struct ocfs2_live_connection *c;
206 
207 	c = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
208 	if (!c)
209 		return -ENOMEM;
210 
211 	mutex_lock(&ocfs2_control_lock);
212 	c->oc_conn = conn;
213 
214 	if (atomic_read(&ocfs2_control_opened))
215 		list_add(&c->oc_list, &ocfs2_live_connection_list);
216 	else {
217 		printk(KERN_ERR
218 		       "ocfs2: Userspace control daemon is not present\n");
219 		rc = -ESRCH;
220 	}
221 
222 	mutex_unlock(&ocfs2_control_lock);
223 
224 	if (!rc)
225 		*c_ret = c;
226 	else
227 		kfree(c);
228 
229 	return rc;
230 }
231 
232 /*
233  * This function disconnects the cluster connection from ocfs2_control.
234  * Afterwards, userspace can't affect the cluster connection.
235  */
236 static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
237 {
238 	mutex_lock(&ocfs2_control_lock);
239 	list_del_init(&c->oc_list);
240 	c->oc_conn = NULL;
241 	mutex_unlock(&ocfs2_control_lock);
242 
243 	kfree(c);
244 }
245 
246 static int ocfs2_control_cfu(void *target, size_t target_len,
247 			     const char __user *buf, size_t count)
248 {
249 	/* The T01 expects write(2) calls to have exactly one command */
250 	if ((count != target_len) ||
251 	    (count > sizeof(union ocfs2_control_message)))
252 		return -EINVAL;
253 
254 	if (copy_from_user(target, buf, target_len))
255 		return -EFAULT;
256 
257 	return 0;
258 }
259 
260 static ssize_t ocfs2_control_validate_protocol(struct file *file,
261 					       const char __user *buf,
262 					       size_t count)
263 {
264 	ssize_t ret;
265 	char kbuf[OCFS2_CONTROL_PROTO_LEN];
266 
267 	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
268 				buf, count);
269 	if (ret)
270 		return ret;
271 
272 	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
273 		return -EINVAL;
274 
275 	ocfs2_control_set_handshake_state(file,
276 					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
277 
278 	return count;
279 }
280 
281 static void ocfs2_control_send_down(const char *uuid,
282 				    int nodenum)
283 {
284 	struct ocfs2_live_connection *c;
285 
286 	mutex_lock(&ocfs2_control_lock);
287 
288 	c = ocfs2_connection_find(uuid);
289 	if (c) {
290 		BUG_ON(c->oc_conn == NULL);
291 		c->oc_conn->cc_recovery_handler(nodenum,
292 						c->oc_conn->cc_recovery_data);
293 	}
294 
295 	mutex_unlock(&ocfs2_control_lock);
296 }
297 
298 /*
299  * Called whenever configuration elements are sent to /dev/ocfs2_control.
300  * If all configuration elements are present, try to set the global
301  * values.  If there is a problem, return an error.  Skip any missing
302  * elements, and only bump ocfs2_control_opened when we have all elements
303  * and are successful.
304  */
305 static int ocfs2_control_install_private(struct file *file)
306 {
307 	int rc = 0;
308 	int set_p = 1;
309 	struct ocfs2_control_private *p = file->private_data;
310 
311 	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
312 
313 	mutex_lock(&ocfs2_control_lock);
314 
315 	if (p->op_this_node < 0) {
316 		set_p = 0;
317 	} else if ((ocfs2_control_this_node >= 0) &&
318 		   (ocfs2_control_this_node != p->op_this_node)) {
319 		rc = -EINVAL;
320 		goto out_unlock;
321 	}
322 
323 	if (!p->op_proto.pv_major) {
324 		set_p = 0;
325 	} else if (!list_empty(&ocfs2_live_connection_list) &&
326 		   ((running_proto.pv_major != p->op_proto.pv_major) ||
327 		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
328 		rc = -EINVAL;
329 		goto out_unlock;
330 	}
331 
332 	if (set_p) {
333 		ocfs2_control_this_node = p->op_this_node;
334 		running_proto.pv_major = p->op_proto.pv_major;
335 		running_proto.pv_minor = p->op_proto.pv_minor;
336 	}
337 
338 out_unlock:
339 	mutex_unlock(&ocfs2_control_lock);
340 
341 	if (!rc && set_p) {
342 		/* We set the global values successfully */
343 		atomic_inc(&ocfs2_control_opened);
344 		ocfs2_control_set_handshake_state(file,
345 					OCFS2_CONTROL_HANDSHAKE_VALID);
346 	}
347 
348 	return rc;
349 }
350 
351 static int ocfs2_control_get_this_node(void)
352 {
353 	int rc;
354 
355 	mutex_lock(&ocfs2_control_lock);
356 	if (ocfs2_control_this_node < 0)
357 		rc = -EINVAL;
358 	else
359 		rc = ocfs2_control_this_node;
360 	mutex_unlock(&ocfs2_control_lock);
361 
362 	return rc;
363 }
364 
365 static int ocfs2_control_do_setnode_msg(struct file *file,
366 					struct ocfs2_control_message_setn *msg)
367 {
368 	long nodenum;
369 	char *ptr = NULL;
370 	struct ocfs2_control_private *p = file->private_data;
371 
372 	if (ocfs2_control_get_handshake_state(file) !=
373 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
374 		return -EINVAL;
375 
376 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
377 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
378 		return -EINVAL;
379 
380 	if ((msg->space != ' ') || (msg->newline != '\n'))
381 		return -EINVAL;
382 	msg->space = msg->newline = '\0';
383 
384 	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
385 	if (!ptr || *ptr)
386 		return -EINVAL;
387 
388 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
389 	    (nodenum > INT_MAX) || (nodenum < 0))
390 		return -ERANGE;
391 	p->op_this_node = nodenum;
392 
393 	return ocfs2_control_install_private(file);
394 }
395 
396 static int ocfs2_control_do_setversion_msg(struct file *file,
397 					   struct ocfs2_control_message_setv *msg)
398  {
399 	long major, minor;
400 	char *ptr = NULL;
401 	struct ocfs2_control_private *p = file->private_data;
402 	struct ocfs2_protocol_version *max =
403 		&ocfs2_user_plugin.sp_max_proto;
404 
405 	if (ocfs2_control_get_handshake_state(file) !=
406 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
407 		return -EINVAL;
408 
409 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
410 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
411 		return -EINVAL;
412 
413 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
414 	    (msg->newline != '\n'))
415 		return -EINVAL;
416 	msg->space1 = msg->space2 = msg->newline = '\0';
417 
418 	major = simple_strtol(msg->major, &ptr, 16);
419 	if (!ptr || *ptr)
420 		return -EINVAL;
421 	minor = simple_strtol(msg->minor, &ptr, 16);
422 	if (!ptr || *ptr)
423 		return -EINVAL;
424 
425 	/*
426 	 * The major must be between 1 and 255, inclusive.  The minor
427 	 * must be between 0 and 255, inclusive.  The version passed in
428 	 * must be within the maximum version supported by the filesystem.
429 	 */
430 	if ((major == LONG_MIN) || (major == LONG_MAX) ||
431 	    (major > (u8)-1) || (major < 1))
432 		return -ERANGE;
433 	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
434 	    (minor > (u8)-1) || (minor < 0))
435 		return -ERANGE;
436 	if ((major != max->pv_major) ||
437 	    (minor > max->pv_minor))
438 		return -EINVAL;
439 
440 	p->op_proto.pv_major = major;
441 	p->op_proto.pv_minor = minor;
442 
443 	return ocfs2_control_install_private(file);
444 }
445 
446 static int ocfs2_control_do_down_msg(struct file *file,
447 				     struct ocfs2_control_message_down *msg)
448 {
449 	long nodenum;
450 	char *p = NULL;
451 
452 	if (ocfs2_control_get_handshake_state(file) !=
453 	    OCFS2_CONTROL_HANDSHAKE_VALID)
454 		return -EINVAL;
455 
456 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
457 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
458 		return -EINVAL;
459 
460 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
461 	    (msg->newline != '\n'))
462 		return -EINVAL;
463 	msg->space1 = msg->space2 = msg->newline = '\0';
464 
465 	nodenum = simple_strtol(msg->nodestr, &p, 16);
466 	if (!p || *p)
467 		return -EINVAL;
468 
469 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
470 	    (nodenum > INT_MAX) || (nodenum < 0))
471 		return -ERANGE;
472 
473 	ocfs2_control_send_down(msg->uuid, nodenum);
474 
475 	return 0;
476 }
477 
478 static ssize_t ocfs2_control_message(struct file *file,
479 				     const char __user *buf,
480 				     size_t count)
481 {
482 	ssize_t ret;
483 	union ocfs2_control_message msg;
484 
485 	/* Try to catch padding issues */
486 	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
487 		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
488 
489 	memset(&msg, 0, sizeof(union ocfs2_control_message));
490 	ret = ocfs2_control_cfu(&msg, count, buf, count);
491 	if (ret)
492 		goto out;
493 
494 	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
495 	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
496 		     OCFS2_CONTROL_MESSAGE_OP_LEN))
497 		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
498 	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
499 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
500 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
501 		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
502 	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
503 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
504 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
505 		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
506 	else
507 		ret = -EINVAL;
508 
509 out:
510 	return ret ? ret : count;
511 }
512 
513 static ssize_t ocfs2_control_write(struct file *file,
514 				   const char __user *buf,
515 				   size_t count,
516 				   loff_t *ppos)
517 {
518 	ssize_t ret;
519 
520 	switch (ocfs2_control_get_handshake_state(file)) {
521 		case OCFS2_CONTROL_HANDSHAKE_INVALID:
522 			ret = -EINVAL;
523 			break;
524 
525 		case OCFS2_CONTROL_HANDSHAKE_READ:
526 			ret = ocfs2_control_validate_protocol(file, buf,
527 							      count);
528 			break;
529 
530 		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
531 		case OCFS2_CONTROL_HANDSHAKE_VALID:
532 			ret = ocfs2_control_message(file, buf, count);
533 			break;
534 
535 		default:
536 			BUG();
537 			ret = -EIO;
538 			break;
539 	}
540 
541 	return ret;
542 }
543 
544 /*
545  * This is a naive version.  If we ever have a new protocol, we'll expand
546  * it.  Probably using seq_file.
547  */
548 static ssize_t ocfs2_control_read(struct file *file,
549 				  char __user *buf,
550 				  size_t count,
551 				  loff_t *ppos)
552 {
553 	ssize_t ret;
554 
555 	ret = simple_read_from_buffer(buf, count, ppos,
556 			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
557 
558 	/* Have we read the whole protocol list? */
559 	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
560 		ocfs2_control_set_handshake_state(file,
561 						  OCFS2_CONTROL_HANDSHAKE_READ);
562 
563 	return ret;
564 }
565 
566 static int ocfs2_control_release(struct inode *inode, struct file *file)
567 {
568 	struct ocfs2_control_private *p = file->private_data;
569 
570 	mutex_lock(&ocfs2_control_lock);
571 
572 	if (ocfs2_control_get_handshake_state(file) !=
573 	    OCFS2_CONTROL_HANDSHAKE_VALID)
574 		goto out;
575 
576 	if (atomic_dec_and_test(&ocfs2_control_opened)) {
577 		if (!list_empty(&ocfs2_live_connection_list)) {
578 			/* XXX: Do bad things! */
579 			printk(KERN_ERR
580 			       "ocfs2: Unexpected release of ocfs2_control!\n"
581 			       "       Loss of cluster connection requires "
582 			       "an emergency restart!\n");
583 			emergency_restart();
584 		}
585 		/*
586 		 * Last valid close clears the node number and resets
587 		 * the locking protocol version
588 		 */
589 		ocfs2_control_this_node = -1;
590 		running_proto.pv_major = 0;
591 		running_proto.pv_major = 0;
592 	}
593 
594 out:
595 	list_del_init(&p->op_list);
596 	file->private_data = NULL;
597 
598 	mutex_unlock(&ocfs2_control_lock);
599 
600 	kfree(p);
601 
602 	return 0;
603 }
604 
605 static int ocfs2_control_open(struct inode *inode, struct file *file)
606 {
607 	struct ocfs2_control_private *p;
608 
609 	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
610 	if (!p)
611 		return -ENOMEM;
612 	p->op_this_node = -1;
613 
614 	mutex_lock(&ocfs2_control_lock);
615 	file->private_data = p;
616 	list_add(&p->op_list, &ocfs2_control_private_list);
617 	mutex_unlock(&ocfs2_control_lock);
618 
619 	return 0;
620 }
621 
622 static const struct file_operations ocfs2_control_fops = {
623 	.open    = ocfs2_control_open,
624 	.release = ocfs2_control_release,
625 	.read    = ocfs2_control_read,
626 	.write   = ocfs2_control_write,
627 	.owner   = THIS_MODULE,
628 	.llseek  = default_llseek,
629 };
630 
631 static struct miscdevice ocfs2_control_device = {
632 	.minor		= MISC_DYNAMIC_MINOR,
633 	.name		= "ocfs2_control",
634 	.fops		= &ocfs2_control_fops,
635 };
636 
637 static int ocfs2_control_init(void)
638 {
639 	int rc;
640 
641 	atomic_set(&ocfs2_control_opened, 0);
642 
643 	rc = misc_register(&ocfs2_control_device);
644 	if (rc)
645 		printk(KERN_ERR
646 		       "ocfs2: Unable to register ocfs2_control device "
647 		       "(errno %d)\n",
648 		       -rc);
649 
650 	return rc;
651 }
652 
653 static void ocfs2_control_exit(void)
654 {
655 	int rc;
656 
657 	rc = misc_deregister(&ocfs2_control_device);
658 	if (rc)
659 		printk(KERN_ERR
660 		       "ocfs2: Unable to deregister ocfs2_control device "
661 		       "(errno %d)\n",
662 		       -rc);
663 }
664 
665 static void fsdlm_lock_ast_wrapper(void *astarg)
666 {
667 	struct ocfs2_dlm_lksb *lksb = astarg;
668 	int status = lksb->lksb_fsdlm.sb_status;
669 
670 	/*
671 	 * For now we're punting on the issue of other non-standard errors
672 	 * where we can't tell if the unlock_ast or lock_ast should be called.
673 	 * The main "other error" that's possible is EINVAL which means the
674 	 * function was called with invalid args, which shouldn't be possible
675 	 * since the caller here is under our control.  Other non-standard
676 	 * errors probably fall into the same category, or otherwise are fatal
677 	 * which means we can't carry on anyway.
678 	 */
679 
680 	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
681 		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
682 	else
683 		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
684 }
685 
686 static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
687 {
688 	struct ocfs2_dlm_lksb *lksb = astarg;
689 
690 	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
691 }
692 
693 static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
694 			 int mode,
695 			 struct ocfs2_dlm_lksb *lksb,
696 			 u32 flags,
697 			 void *name,
698 			 unsigned int namelen)
699 {
700 	int ret;
701 
702 	if (!lksb->lksb_fsdlm.sb_lvbptr)
703 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
704 					     sizeof(struct dlm_lksb);
705 
706 	ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
707 		       flags|DLM_LKF_NODLCKWT, name, namelen, 0,
708 		       fsdlm_lock_ast_wrapper, lksb,
709 		       fsdlm_blocking_ast_wrapper);
710 	return ret;
711 }
712 
713 static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
714 			   struct ocfs2_dlm_lksb *lksb,
715 			   u32 flags)
716 {
717 	int ret;
718 
719 	ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
720 			 flags, &lksb->lksb_fsdlm, lksb);
721 	return ret;
722 }
723 
724 static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
725 {
726 	return lksb->lksb_fsdlm.sb_status;
727 }
728 
729 static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
730 {
731 	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
732 
733 	return !invalid;
734 }
735 
736 static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
737 {
738 	if (!lksb->lksb_fsdlm.sb_lvbptr)
739 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
740 					     sizeof(struct dlm_lksb);
741 	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
742 }
743 
744 static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
745 {
746 }
747 
748 static int user_plock(struct ocfs2_cluster_connection *conn,
749 		      u64 ino,
750 		      struct file *file,
751 		      int cmd,
752 		      struct file_lock *fl)
753 {
754 	/*
755 	 * This more or less just demuxes the plock request into any
756 	 * one of three dlm calls.
757 	 *
758 	 * Internally, fs/dlm will pass these to a misc device, which
759 	 * a userspace daemon will read and write to.
760 	 *
761 	 * For now, cancel requests (which happen internally only),
762 	 * are turned into unlocks. Most of this function taken from
763 	 * gfs2_lock.
764 	 */
765 
766 	if (cmd == F_CANCELLK) {
767 		cmd = F_SETLK;
768 		fl->fl_type = F_UNLCK;
769 	}
770 
771 	if (IS_GETLK(cmd))
772 		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
773 	else if (fl->fl_type == F_UNLCK)
774 		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
775 	else
776 		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
777 }
778 
779 /*
780  * Compare a requested locking protocol version against the current one.
781  *
782  * If the major numbers are different, they are incompatible.
783  * If the current minor is greater than the request, they are incompatible.
784  * If the current minor is less than or equal to the request, they are
785  * compatible, and the requester should run at the current minor version.
786  */
787 static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
788 			       struct ocfs2_protocol_version *request)
789 {
790 	if (existing->pv_major != request->pv_major)
791 		return 1;
792 
793 	if (existing->pv_minor > request->pv_minor)
794 		return 1;
795 
796 	if (existing->pv_minor < request->pv_minor)
797 		request->pv_minor = existing->pv_minor;
798 
799 	return 0;
800 }
801 
802 static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
803 {
804 	dlm_lockspace_t *fsdlm;
805 	struct ocfs2_live_connection *uninitialized_var(control);
806 	int rc = 0;
807 
808 	BUG_ON(conn == NULL);
809 
810 	rc = ocfs2_live_connection_new(conn, &control);
811 	if (rc)
812 		goto out;
813 
814 	/*
815 	 * running_proto must have been set before we allowed any mounts
816 	 * to proceed.
817 	 */
818 	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
819 		printk(KERN_ERR
820 		       "Unable to mount with fs locking protocol version "
821 		       "%u.%u because the userspace control daemon has "
822 		       "negotiated %u.%u\n",
823 		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
824 		       running_proto.pv_major, running_proto.pv_minor);
825 		rc = -EPROTO;
826 		ocfs2_live_connection_drop(control);
827 		goto out;
828 	}
829 
830 	rc = dlm_new_lockspace(conn->cc_name, NULL, DLM_LSFL_FS, DLM_LVB_LEN,
831 			       NULL, NULL, NULL, &fsdlm);
832 	if (rc) {
833 		ocfs2_live_connection_drop(control);
834 		goto out;
835 	}
836 
837 	conn->cc_private = control;
838 	conn->cc_lockspace = fsdlm;
839 out:
840 	return rc;
841 }
842 
843 static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
844 {
845 	dlm_release_lockspace(conn->cc_lockspace, 2);
846 	conn->cc_lockspace = NULL;
847 	ocfs2_live_connection_drop(conn->cc_private);
848 	conn->cc_private = NULL;
849 	return 0;
850 }
851 
852 static int user_cluster_this_node(unsigned int *this_node)
853 {
854 	int rc;
855 
856 	rc = ocfs2_control_get_this_node();
857 	if (rc < 0)
858 		return rc;
859 
860 	*this_node = rc;
861 	return 0;
862 }
863 
864 static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
865 	.connect	= user_cluster_connect,
866 	.disconnect	= user_cluster_disconnect,
867 	.this_node	= user_cluster_this_node,
868 	.dlm_lock	= user_dlm_lock,
869 	.dlm_unlock	= user_dlm_unlock,
870 	.lock_status	= user_dlm_lock_status,
871 	.lvb_valid	= user_dlm_lvb_valid,
872 	.lock_lvb	= user_dlm_lvb,
873 	.plock		= user_plock,
874 	.dump_lksb	= user_dlm_dump_lksb,
875 };
876 
877 static struct ocfs2_stack_plugin ocfs2_user_plugin = {
878 	.sp_name	= "user",
879 	.sp_ops		= &ocfs2_user_plugin_ops,
880 	.sp_owner	= THIS_MODULE,
881 };
882 
883 
884 static int __init ocfs2_user_plugin_init(void)
885 {
886 	int rc;
887 
888 	rc = ocfs2_control_init();
889 	if (!rc) {
890 		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
891 		if (rc)
892 			ocfs2_control_exit();
893 	}
894 
895 	return rc;
896 }
897 
898 static void __exit ocfs2_user_plugin_exit(void)
899 {
900 	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
901 	ocfs2_control_exit();
902 }
903 
904 MODULE_AUTHOR("Oracle");
905 MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
906 MODULE_LICENSE("GPL");
907 module_init(ocfs2_user_plugin_init);
908 module_exit(ocfs2_user_plugin_exit);
909