xref: /linux/fs/ocfs2/stack_user.c (revision c532de5a67a70f8533d495f8f2aaa9a0491c3ad0)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * stack_user.c
4  *
5  * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
6  *
7  * Copyright (C) 2007 Oracle.  All rights reserved.
8  */
9 
10 #include <linux/module.h>
11 #include <linux/fs.h>
12 #include <linux/filelock.h>
13 #include <linux/miscdevice.h>
14 #include <linux/mutex.h>
15 #include <linux/slab.h>
16 #include <linux/reboot.h>
17 #include <linux/sched.h>
18 #include <linux/uaccess.h>
19 
20 #include "stackglue.h"
21 
22 #include <linux/dlm_plock.h>
23 
24 /*
25  * The control protocol starts with a handshake.  Until the handshake
26  * is complete, the control device will fail all write(2)s.
27  *
28  * The handshake is simple.  First, the client reads until EOF.  Each line
29  * of output is a supported protocol tag.  All protocol tags are a single
30  * character followed by a two hex digit version number.  Currently the
31  * only things supported is T01, for "Text-base version 0x01".  Next, the
32  * client writes the version they would like to use, including the newline.
33  * Thus, the protocol tag is 'T01\n'.  If the version tag written is
34  * unknown, -EINVAL is returned.  Once the negotiation is complete, the
35  * client can start sending messages.
36  *
37  * The T01 protocol has three messages.  First is the "SETN" message.
38  * It has the following syntax:
39  *
40  *  SETN<space><8-char-hex-nodenum><newline>
41  *
42  * This is 14 characters.
43  *
44  * The "SETN" message must be the first message following the protocol.
45  * It tells ocfs2_control the local node number.
46  *
47  * Next comes the "SETV" message.  It has the following syntax:
48  *
49  *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
50  *
51  * This is 11 characters.
52  *
53  * The "SETV" message sets the filesystem locking protocol version as
54  * negotiated by the client.  The client negotiates based on the maximum
55  * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
56  * number from the "SETV" message must match
57  * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
58  * must be less than or equal to ...sp_max_version.pv_minor.
59  *
60  * Once this information has been set, mounts will be allowed.  From this
61  * point on, the "DOWN" message can be sent for node down notification.
62  * It has the following syntax:
63  *
64  *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
65  *
66  * eg:
67  *
68  *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
69  *
70  * This is 47 characters.
71  */
72 
73 /*
74  * Whether or not the client has done the handshake.
75  * For now, we have just one protocol version.
76  */
77 #define OCFS2_CONTROL_PROTO			"T01\n"
78 #define OCFS2_CONTROL_PROTO_LEN			4
79 
80 /* Handshake states */
81 #define OCFS2_CONTROL_HANDSHAKE_INVALID		(0)
82 #define OCFS2_CONTROL_HANDSHAKE_READ		(1)
83 #define OCFS2_CONTROL_HANDSHAKE_PROTOCOL	(2)
84 #define OCFS2_CONTROL_HANDSHAKE_VALID		(3)
85 
86 /* Messages */
87 #define OCFS2_CONTROL_MESSAGE_OP_LEN		4
88 #define OCFS2_CONTROL_MESSAGE_SETNODE_OP	"SETN"
89 #define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN	14
90 #define OCFS2_CONTROL_MESSAGE_SETVERSION_OP	"SETV"
91 #define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN	11
92 #define OCFS2_CONTROL_MESSAGE_DOWN_OP		"DOWN"
93 #define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN	47
94 #define OCFS2_TEXT_UUID_LEN			32
95 #define OCFS2_CONTROL_MESSAGE_VERNUM_LEN	2
96 #define OCFS2_CONTROL_MESSAGE_NODENUM_LEN	8
97 #define VERSION_LOCK				"version_lock"
98 
99 enum ocfs2_connection_type {
100 	WITH_CONTROLD,
101 	NO_CONTROLD
102 };
103 
104 /*
105  * ocfs2_live_connection is refcounted because the filesystem and
106  * miscdevice sides can detach in different order.  Let's just be safe.
107  */
108 struct ocfs2_live_connection {
109 	struct list_head		oc_list;
110 	struct ocfs2_cluster_connection	*oc_conn;
111 	enum ocfs2_connection_type	oc_type;
112 	atomic_t                        oc_this_node;
113 	int                             oc_our_slot;
114 	struct dlm_lksb                 oc_version_lksb;
115 	char                            oc_lvb[DLM_LVB_LEN];
116 	struct completion               oc_sync_wait;
117 	wait_queue_head_t		oc_wait;
118 };
119 
120 struct ocfs2_control_private {
121 	struct list_head op_list;
122 	int op_state;
123 	int op_this_node;
124 	struct ocfs2_protocol_version op_proto;
125 };
126 
127 /* SETN<space><8-char-hex-nodenum><newline> */
128 struct ocfs2_control_message_setn {
129 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
130 	char	space;
131 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
132 	char	newline;
133 };
134 
135 /* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
136 struct ocfs2_control_message_setv {
137 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
138 	char	space1;
139 	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
140 	char	space2;
141 	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
142 	char	newline;
143 };
144 
145 /* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
146 struct ocfs2_control_message_down {
147 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
148 	char	space1;
149 	char	uuid[OCFS2_TEXT_UUID_LEN];
150 	char	space2;
151 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
152 	char	newline;
153 };
154 
155 union ocfs2_control_message {
156 	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
157 	struct ocfs2_control_message_setn	u_setn;
158 	struct ocfs2_control_message_setv	u_setv;
159 	struct ocfs2_control_message_down	u_down;
160 };
161 
162 static struct ocfs2_stack_plugin ocfs2_user_plugin;
163 
164 static atomic_t ocfs2_control_opened;
165 static int ocfs2_control_this_node = -1;
166 static struct ocfs2_protocol_version running_proto;
167 
168 static LIST_HEAD(ocfs2_live_connection_list);
169 static LIST_HEAD(ocfs2_control_private_list);
170 static DEFINE_MUTEX(ocfs2_control_lock);
171 
172 static inline void ocfs2_control_set_handshake_state(struct file *file,
173 						     int state)
174 {
175 	struct ocfs2_control_private *p = file->private_data;
176 	p->op_state = state;
177 }
178 
179 static inline int ocfs2_control_get_handshake_state(struct file *file)
180 {
181 	struct ocfs2_control_private *p = file->private_data;
182 	return p->op_state;
183 }
184 
185 static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
186 {
187 	size_t len = strlen(name);
188 	struct ocfs2_live_connection *c;
189 
190 	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
191 
192 	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
193 		if ((c->oc_conn->cc_namelen == len) &&
194 		    !strncmp(c->oc_conn->cc_name, name, len))
195 			return c;
196 	}
197 
198 	return NULL;
199 }
200 
201 /*
202  * ocfs2_live_connection structures are created underneath the ocfs2
203  * mount path.  Since the VFS prevents multiple calls to
204  * fill_super(), we can't get dupes here.
205  */
206 static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
207 				     struct ocfs2_live_connection *c)
208 {
209 	int rc = 0;
210 
211 	mutex_lock(&ocfs2_control_lock);
212 	c->oc_conn = conn;
213 
214 	if ((c->oc_type == NO_CONTROLD) || 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 	return rc;
224 }
225 
226 /*
227  * This function disconnects the cluster connection from ocfs2_control.
228  * Afterwards, userspace can't affect the cluster connection.
229  */
230 static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
231 {
232 	mutex_lock(&ocfs2_control_lock);
233 	list_del_init(&c->oc_list);
234 	c->oc_conn = NULL;
235 	mutex_unlock(&ocfs2_control_lock);
236 
237 	kfree(c);
238 }
239 
240 static int ocfs2_control_cfu(void *target, size_t target_len,
241 			     const char __user *buf, size_t count)
242 {
243 	/* The T01 expects write(2) calls to have exactly one command */
244 	if ((count != target_len) ||
245 	    (count > sizeof(union ocfs2_control_message)))
246 		return -EINVAL;
247 
248 	if (copy_from_user(target, buf, target_len))
249 		return -EFAULT;
250 
251 	return 0;
252 }
253 
254 static ssize_t ocfs2_control_validate_protocol(struct file *file,
255 					       const char __user *buf,
256 					       size_t count)
257 {
258 	ssize_t ret;
259 	char kbuf[OCFS2_CONTROL_PROTO_LEN];
260 
261 	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
262 				buf, count);
263 	if (ret)
264 		return ret;
265 
266 	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
267 		return -EINVAL;
268 
269 	ocfs2_control_set_handshake_state(file,
270 					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
271 
272 	return count;
273 }
274 
275 static void ocfs2_control_send_down(const char *uuid,
276 				    int nodenum)
277 {
278 	struct ocfs2_live_connection *c;
279 
280 	mutex_lock(&ocfs2_control_lock);
281 
282 	c = ocfs2_connection_find(uuid);
283 	if (c) {
284 		BUG_ON(c->oc_conn == NULL);
285 		c->oc_conn->cc_recovery_handler(nodenum,
286 						c->oc_conn->cc_recovery_data);
287 	}
288 
289 	mutex_unlock(&ocfs2_control_lock);
290 }
291 
292 /*
293  * Called whenever configuration elements are sent to /dev/ocfs2_control.
294  * If all configuration elements are present, try to set the global
295  * values.  If there is a problem, return an error.  Skip any missing
296  * elements, and only bump ocfs2_control_opened when we have all elements
297  * and are successful.
298  */
299 static int ocfs2_control_install_private(struct file *file)
300 {
301 	int rc = 0;
302 	int set_p = 1;
303 	struct ocfs2_control_private *p = file->private_data;
304 
305 	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
306 
307 	mutex_lock(&ocfs2_control_lock);
308 
309 	if (p->op_this_node < 0) {
310 		set_p = 0;
311 	} else if ((ocfs2_control_this_node >= 0) &&
312 		   (ocfs2_control_this_node != p->op_this_node)) {
313 		rc = -EINVAL;
314 		goto out_unlock;
315 	}
316 
317 	if (!p->op_proto.pv_major) {
318 		set_p = 0;
319 	} else if (!list_empty(&ocfs2_live_connection_list) &&
320 		   ((running_proto.pv_major != p->op_proto.pv_major) ||
321 		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
322 		rc = -EINVAL;
323 		goto out_unlock;
324 	}
325 
326 	if (set_p) {
327 		ocfs2_control_this_node = p->op_this_node;
328 		running_proto.pv_major = p->op_proto.pv_major;
329 		running_proto.pv_minor = p->op_proto.pv_minor;
330 	}
331 
332 out_unlock:
333 	mutex_unlock(&ocfs2_control_lock);
334 
335 	if (!rc && set_p) {
336 		/* We set the global values successfully */
337 		atomic_inc(&ocfs2_control_opened);
338 		ocfs2_control_set_handshake_state(file,
339 					OCFS2_CONTROL_HANDSHAKE_VALID);
340 	}
341 
342 	return rc;
343 }
344 
345 static int ocfs2_control_get_this_node(void)
346 {
347 	int rc;
348 
349 	mutex_lock(&ocfs2_control_lock);
350 	if (ocfs2_control_this_node < 0)
351 		rc = -EINVAL;
352 	else
353 		rc = ocfs2_control_this_node;
354 	mutex_unlock(&ocfs2_control_lock);
355 
356 	return rc;
357 }
358 
359 static int ocfs2_control_do_setnode_msg(struct file *file,
360 					struct ocfs2_control_message_setn *msg)
361 {
362 	long nodenum;
363 	char *ptr = NULL;
364 	struct ocfs2_control_private *p = file->private_data;
365 
366 	if (ocfs2_control_get_handshake_state(file) !=
367 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
368 		return -EINVAL;
369 
370 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
371 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
372 		return -EINVAL;
373 
374 	if ((msg->space != ' ') || (msg->newline != '\n'))
375 		return -EINVAL;
376 	msg->space = msg->newline = '\0';
377 
378 	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
379 	if (!ptr || *ptr)
380 		return -EINVAL;
381 
382 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
383 	    (nodenum > INT_MAX) || (nodenum < 0))
384 		return -ERANGE;
385 	p->op_this_node = nodenum;
386 
387 	return ocfs2_control_install_private(file);
388 }
389 
390 static int ocfs2_control_do_setversion_msg(struct file *file,
391 					   struct ocfs2_control_message_setv *msg)
392 {
393 	long major, minor;
394 	char *ptr = NULL;
395 	struct ocfs2_control_private *p = file->private_data;
396 	struct ocfs2_protocol_version *max =
397 		&ocfs2_user_plugin.sp_max_proto;
398 
399 	if (ocfs2_control_get_handshake_state(file) !=
400 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
401 		return -EINVAL;
402 
403 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
404 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
405 		return -EINVAL;
406 
407 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
408 	    (msg->newline != '\n'))
409 		return -EINVAL;
410 	msg->space1 = msg->space2 = msg->newline = '\0';
411 
412 	major = simple_strtol(msg->major, &ptr, 16);
413 	if (!ptr || *ptr)
414 		return -EINVAL;
415 	minor = simple_strtol(msg->minor, &ptr, 16);
416 	if (!ptr || *ptr)
417 		return -EINVAL;
418 
419 	/*
420 	 * The major must be between 1 and 255, inclusive.  The minor
421 	 * must be between 0 and 255, inclusive.  The version passed in
422 	 * must be within the maximum version supported by the filesystem.
423 	 */
424 	if ((major == LONG_MIN) || (major == LONG_MAX) ||
425 	    (major > (u8)-1) || (major < 1))
426 		return -ERANGE;
427 	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
428 	    (minor > (u8)-1) || (minor < 0))
429 		return -ERANGE;
430 	if ((major != max->pv_major) ||
431 	    (minor > max->pv_minor))
432 		return -EINVAL;
433 
434 	p->op_proto.pv_major = major;
435 	p->op_proto.pv_minor = minor;
436 
437 	return ocfs2_control_install_private(file);
438 }
439 
440 static int ocfs2_control_do_down_msg(struct file *file,
441 				     struct ocfs2_control_message_down *msg)
442 {
443 	long nodenum;
444 	char *p = NULL;
445 
446 	if (ocfs2_control_get_handshake_state(file) !=
447 	    OCFS2_CONTROL_HANDSHAKE_VALID)
448 		return -EINVAL;
449 
450 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
451 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
452 		return -EINVAL;
453 
454 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
455 	    (msg->newline != '\n'))
456 		return -EINVAL;
457 	msg->space1 = msg->space2 = msg->newline = '\0';
458 
459 	nodenum = simple_strtol(msg->nodestr, &p, 16);
460 	if (!p || *p)
461 		return -EINVAL;
462 
463 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
464 	    (nodenum > INT_MAX) || (nodenum < 0))
465 		return -ERANGE;
466 
467 	ocfs2_control_send_down(msg->uuid, nodenum);
468 
469 	return 0;
470 }
471 
472 static ssize_t ocfs2_control_message(struct file *file,
473 				     const char __user *buf,
474 				     size_t count)
475 {
476 	ssize_t ret;
477 	union ocfs2_control_message msg;
478 
479 	/* Try to catch padding issues */
480 	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
481 		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
482 
483 	memset(&msg, 0, sizeof(union ocfs2_control_message));
484 	ret = ocfs2_control_cfu(&msg, count, buf, count);
485 	if (ret)
486 		goto out;
487 
488 	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
489 	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
490 		     OCFS2_CONTROL_MESSAGE_OP_LEN))
491 		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
492 	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
493 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
494 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
495 		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
496 	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
497 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
498 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
499 		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
500 	else
501 		ret = -EINVAL;
502 
503 out:
504 	return ret ? ret : count;
505 }
506 
507 static ssize_t ocfs2_control_write(struct file *file,
508 				   const char __user *buf,
509 				   size_t count,
510 				   loff_t *ppos)
511 {
512 	ssize_t ret;
513 
514 	switch (ocfs2_control_get_handshake_state(file)) {
515 		case OCFS2_CONTROL_HANDSHAKE_INVALID:
516 			ret = -EINVAL;
517 			break;
518 
519 		case OCFS2_CONTROL_HANDSHAKE_READ:
520 			ret = ocfs2_control_validate_protocol(file, buf,
521 							      count);
522 			break;
523 
524 		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
525 		case OCFS2_CONTROL_HANDSHAKE_VALID:
526 			ret = ocfs2_control_message(file, buf, count);
527 			break;
528 
529 		default:
530 			BUG();
531 			ret = -EIO;
532 			break;
533 	}
534 
535 	return ret;
536 }
537 
538 /*
539  * This is a naive version.  If we ever have a new protocol, we'll expand
540  * it.  Probably using seq_file.
541  */
542 static ssize_t ocfs2_control_read(struct file *file,
543 				  char __user *buf,
544 				  size_t count,
545 				  loff_t *ppos)
546 {
547 	ssize_t ret;
548 
549 	ret = simple_read_from_buffer(buf, count, ppos,
550 			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
551 
552 	/* Have we read the whole protocol list? */
553 	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
554 		ocfs2_control_set_handshake_state(file,
555 						  OCFS2_CONTROL_HANDSHAKE_READ);
556 
557 	return ret;
558 }
559 
560 static int ocfs2_control_release(struct inode *inode, struct file *file)
561 {
562 	struct ocfs2_control_private *p = file->private_data;
563 
564 	mutex_lock(&ocfs2_control_lock);
565 
566 	if (ocfs2_control_get_handshake_state(file) !=
567 	    OCFS2_CONTROL_HANDSHAKE_VALID)
568 		goto out;
569 
570 	if (atomic_dec_and_test(&ocfs2_control_opened)) {
571 		if (!list_empty(&ocfs2_live_connection_list)) {
572 			/* XXX: Do bad things! */
573 			printk(KERN_ERR
574 			       "ocfs2: Unexpected release of ocfs2_control!\n"
575 			       "       Loss of cluster connection requires "
576 			       "an emergency restart!\n");
577 			emergency_restart();
578 		}
579 		/*
580 		 * Last valid close clears the node number and resets
581 		 * the locking protocol version
582 		 */
583 		ocfs2_control_this_node = -1;
584 		running_proto.pv_major = 0;
585 		running_proto.pv_minor = 0;
586 	}
587 
588 out:
589 	list_del_init(&p->op_list);
590 	file->private_data = NULL;
591 
592 	mutex_unlock(&ocfs2_control_lock);
593 
594 	kfree(p);
595 
596 	return 0;
597 }
598 
599 static int ocfs2_control_open(struct inode *inode, struct file *file)
600 {
601 	struct ocfs2_control_private *p;
602 
603 	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
604 	if (!p)
605 		return -ENOMEM;
606 	p->op_this_node = -1;
607 
608 	mutex_lock(&ocfs2_control_lock);
609 	file->private_data = p;
610 	list_add(&p->op_list, &ocfs2_control_private_list);
611 	mutex_unlock(&ocfs2_control_lock);
612 
613 	return 0;
614 }
615 
616 static const struct file_operations ocfs2_control_fops = {
617 	.open    = ocfs2_control_open,
618 	.release = ocfs2_control_release,
619 	.read    = ocfs2_control_read,
620 	.write   = ocfs2_control_write,
621 	.owner   = THIS_MODULE,
622 	.llseek  = default_llseek,
623 };
624 
625 static struct miscdevice ocfs2_control_device = {
626 	.minor		= MISC_DYNAMIC_MINOR,
627 	.name		= "ocfs2_control",
628 	.fops		= &ocfs2_control_fops,
629 };
630 
631 static int ocfs2_control_init(void)
632 {
633 	int rc;
634 
635 	atomic_set(&ocfs2_control_opened, 0);
636 
637 	rc = misc_register(&ocfs2_control_device);
638 	if (rc)
639 		printk(KERN_ERR
640 		       "ocfs2: Unable to register ocfs2_control device "
641 		       "(errno %d)\n",
642 		       -rc);
643 
644 	return rc;
645 }
646 
647 static void ocfs2_control_exit(void)
648 {
649 	misc_deregister(&ocfs2_control_device);
650 }
651 
652 static void fsdlm_lock_ast_wrapper(void *astarg)
653 {
654 	struct ocfs2_dlm_lksb *lksb = astarg;
655 	int status = lksb->lksb_fsdlm.sb_status;
656 
657 	/*
658 	 * For now we're punting on the issue of other non-standard errors
659 	 * where we can't tell if the unlock_ast or lock_ast should be called.
660 	 * The main "other error" that's possible is EINVAL which means the
661 	 * function was called with invalid args, which shouldn't be possible
662 	 * since the caller here is under our control.  Other non-standard
663 	 * errors probably fall into the same category, or otherwise are fatal
664 	 * which means we can't carry on anyway.
665 	 */
666 
667 	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
668 		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
669 	else
670 		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
671 }
672 
673 static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
674 {
675 	struct ocfs2_dlm_lksb *lksb = astarg;
676 
677 	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
678 }
679 
680 static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
681 			 int mode,
682 			 struct ocfs2_dlm_lksb *lksb,
683 			 u32 flags,
684 			 void *name,
685 			 unsigned int namelen)
686 {
687 	if (!lksb->lksb_fsdlm.sb_lvbptr)
688 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
689 					     sizeof(struct dlm_lksb);
690 
691 	return dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
692 			flags|DLM_LKF_NODLCKWT, name, namelen, 0,
693 			fsdlm_lock_ast_wrapper, lksb,
694 			fsdlm_blocking_ast_wrapper);
695 }
696 
697 static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
698 			   struct ocfs2_dlm_lksb *lksb,
699 			   u32 flags)
700 {
701 	return dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
702 			  flags, &lksb->lksb_fsdlm, lksb);
703 }
704 
705 static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
706 {
707 	return lksb->lksb_fsdlm.sb_status;
708 }
709 
710 static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
711 {
712 	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
713 
714 	return !invalid;
715 }
716 
717 static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
718 {
719 	if (!lksb->lksb_fsdlm.sb_lvbptr)
720 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
721 					     sizeof(struct dlm_lksb);
722 	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
723 }
724 
725 static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
726 {
727 }
728 
729 static int user_plock(struct ocfs2_cluster_connection *conn,
730 		      u64 ino,
731 		      struct file *file,
732 		      int cmd,
733 		      struct file_lock *fl)
734 {
735 	/*
736 	 * This more or less just demuxes the plock request into any
737 	 * one of three dlm calls.
738 	 *
739 	 * Internally, fs/dlm will pass these to a misc device, which
740 	 * a userspace daemon will read and write to.
741 	 */
742 
743 	if (cmd == F_CANCELLK)
744 		return dlm_posix_cancel(conn->cc_lockspace, ino, file, fl);
745 	else if (IS_GETLK(cmd))
746 		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
747 	else if (lock_is_unlock(fl))
748 		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
749 	else
750 		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
751 }
752 
753 /*
754  * Compare a requested locking protocol version against the current one.
755  *
756  * If the major numbers are different, they are incompatible.
757  * If the current minor is greater than the request, they are incompatible.
758  * If the current minor is less than or equal to the request, they are
759  * compatible, and the requester should run at the current minor version.
760  */
761 static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
762 			       struct ocfs2_protocol_version *request)
763 {
764 	if (existing->pv_major != request->pv_major)
765 		return 1;
766 
767 	if (existing->pv_minor > request->pv_minor)
768 		return 1;
769 
770 	if (existing->pv_minor < request->pv_minor)
771 		request->pv_minor = existing->pv_minor;
772 
773 	return 0;
774 }
775 
776 static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
777 {
778 	struct ocfs2_protocol_version *pv =
779 		(struct ocfs2_protocol_version *)lvb;
780 	/*
781 	 * ocfs2_protocol_version has two u8 variables, so we don't
782 	 * need any endian conversion.
783 	 */
784 	ver->pv_major = pv->pv_major;
785 	ver->pv_minor = pv->pv_minor;
786 }
787 
788 static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
789 {
790 	struct ocfs2_protocol_version *pv =
791 		(struct ocfs2_protocol_version *)lvb;
792 	/*
793 	 * ocfs2_protocol_version has two u8 variables, so we don't
794 	 * need any endian conversion.
795 	 */
796 	pv->pv_major = ver->pv_major;
797 	pv->pv_minor = ver->pv_minor;
798 }
799 
800 static void sync_wait_cb(void *arg)
801 {
802 	struct ocfs2_cluster_connection *conn = arg;
803 	struct ocfs2_live_connection *lc = conn->cc_private;
804 	complete(&lc->oc_sync_wait);
805 }
806 
807 static int sync_unlock(struct ocfs2_cluster_connection *conn,
808 		struct dlm_lksb *lksb, char *name)
809 {
810 	int error;
811 	struct ocfs2_live_connection *lc = conn->cc_private;
812 
813 	error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
814 	if (error) {
815 		printk(KERN_ERR "%s lkid %x error %d\n",
816 				name, lksb->sb_lkid, error);
817 		return error;
818 	}
819 
820 	wait_for_completion(&lc->oc_sync_wait);
821 
822 	if (lksb->sb_status != -DLM_EUNLOCK) {
823 		printk(KERN_ERR "%s lkid %x status %d\n",
824 				name, lksb->sb_lkid, lksb->sb_status);
825 		return -1;
826 	}
827 	return 0;
828 }
829 
830 static int sync_lock(struct ocfs2_cluster_connection *conn,
831 		int mode, uint32_t flags,
832 		struct dlm_lksb *lksb, char *name)
833 {
834 	int error, status;
835 	struct ocfs2_live_connection *lc = conn->cc_private;
836 
837 	error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
838 			name, strlen(name),
839 			0, sync_wait_cb, conn, NULL);
840 	if (error) {
841 		printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
842 				name, lksb->sb_lkid, flags, mode, error);
843 		return error;
844 	}
845 
846 	wait_for_completion(&lc->oc_sync_wait);
847 
848 	status = lksb->sb_status;
849 
850 	if (status && status != -EAGAIN) {
851 		printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
852 				name, lksb->sb_lkid, flags, mode, status);
853 	}
854 
855 	return status;
856 }
857 
858 
859 static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
860 		int flags)
861 {
862 	struct ocfs2_live_connection *lc = conn->cc_private;
863 	return sync_lock(conn, mode, flags,
864 			&lc->oc_version_lksb, VERSION_LOCK);
865 }
866 
867 static int version_unlock(struct ocfs2_cluster_connection *conn)
868 {
869 	struct ocfs2_live_connection *lc = conn->cc_private;
870 	return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
871 }
872 
873 /* get_protocol_version()
874  *
875  * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
876  * The algorithm is:
877  * 1. Attempt to take the lock in EX mode (non-blocking).
878  * 2. If successful (which means it is the first mount), write the
879  *    version number and downconvert to PR lock.
880  * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
881  *    taking the PR lock.
882  */
883 
884 static int get_protocol_version(struct ocfs2_cluster_connection *conn)
885 {
886 	int ret;
887 	struct ocfs2_live_connection *lc = conn->cc_private;
888 	struct ocfs2_protocol_version pv;
889 
890 	running_proto.pv_major =
891 		ocfs2_user_plugin.sp_max_proto.pv_major;
892 	running_proto.pv_minor =
893 		ocfs2_user_plugin.sp_max_proto.pv_minor;
894 
895 	lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
896 	ret = version_lock(conn, DLM_LOCK_EX,
897 			DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
898 	if (!ret) {
899 		conn->cc_version.pv_major = running_proto.pv_major;
900 		conn->cc_version.pv_minor = running_proto.pv_minor;
901 		version_to_lvb(&running_proto, lc->oc_lvb);
902 		version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
903 	} else if (ret == -EAGAIN) {
904 		ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
905 		if (ret)
906 			goto out;
907 		lvb_to_version(lc->oc_lvb, &pv);
908 
909 		if ((pv.pv_major != running_proto.pv_major) ||
910 				(pv.pv_minor > running_proto.pv_minor)) {
911 			ret = -EINVAL;
912 			goto out;
913 		}
914 
915 		conn->cc_version.pv_major = pv.pv_major;
916 		conn->cc_version.pv_minor = pv.pv_minor;
917 	}
918 out:
919 	return ret;
920 }
921 
922 static void user_recover_prep(void *arg)
923 {
924 }
925 
926 static void user_recover_slot(void *arg, struct dlm_slot *slot)
927 {
928 	struct ocfs2_cluster_connection *conn = arg;
929 	printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
930 			slot->nodeid, slot->slot);
931 	conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
932 
933 }
934 
935 static void user_recover_done(void *arg, struct dlm_slot *slots,
936 		int num_slots, int our_slot,
937 		uint32_t generation)
938 {
939 	struct ocfs2_cluster_connection *conn = arg;
940 	struct ocfs2_live_connection *lc = conn->cc_private;
941 	int i;
942 
943 	for (i = 0; i < num_slots; i++)
944 		if (slots[i].slot == our_slot) {
945 			atomic_set(&lc->oc_this_node, slots[i].nodeid);
946 			break;
947 		}
948 
949 	lc->oc_our_slot = our_slot;
950 	wake_up(&lc->oc_wait);
951 }
952 
953 static const struct dlm_lockspace_ops ocfs2_ls_ops = {
954 	.recover_prep = user_recover_prep,
955 	.recover_slot = user_recover_slot,
956 	.recover_done = user_recover_done,
957 };
958 
959 static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
960 {
961 	version_unlock(conn);
962 	dlm_release_lockspace(conn->cc_lockspace, 2);
963 	conn->cc_lockspace = NULL;
964 	ocfs2_live_connection_drop(conn->cc_private);
965 	conn->cc_private = NULL;
966 	return 0;
967 }
968 
969 static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
970 {
971 	dlm_lockspace_t *fsdlm;
972 	struct ocfs2_live_connection *lc;
973 	int rc, ops_rv;
974 
975 	BUG_ON(conn == NULL);
976 
977 	lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
978 	if (!lc)
979 		return -ENOMEM;
980 
981 	init_waitqueue_head(&lc->oc_wait);
982 	init_completion(&lc->oc_sync_wait);
983 	atomic_set(&lc->oc_this_node, 0);
984 	conn->cc_private = lc;
985 	lc->oc_type = NO_CONTROLD;
986 
987 	rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
988 			       DLM_LSFL_NEWEXCL, DLM_LVB_LEN,
989 			       &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
990 	if (rc) {
991 		if (rc == -EEXIST || rc == -EPROTO)
992 			printk(KERN_ERR "ocfs2: Unable to create the "
993 				"lockspace %s (%d), because a ocfs2-tools "
994 				"program is running on this file system "
995 				"with the same name lockspace\n",
996 				conn->cc_name, rc);
997 		goto out;
998 	}
999 
1000 	if (ops_rv == -EOPNOTSUPP) {
1001 		lc->oc_type = WITH_CONTROLD;
1002 		printk(KERN_NOTICE "ocfs2: You seem to be using an older "
1003 				"version of dlm_controld and/or ocfs2-tools."
1004 				" Please consider upgrading.\n");
1005 	} else if (ops_rv) {
1006 		rc = ops_rv;
1007 		goto out;
1008 	}
1009 	conn->cc_lockspace = fsdlm;
1010 
1011 	rc = ocfs2_live_connection_attach(conn, lc);
1012 	if (rc)
1013 		goto out;
1014 
1015 	if (lc->oc_type == NO_CONTROLD) {
1016 		rc = get_protocol_version(conn);
1017 		if (rc) {
1018 			printk(KERN_ERR "ocfs2: Could not determine"
1019 					" locking version\n");
1020 			user_cluster_disconnect(conn);
1021 			goto out;
1022 		}
1023 		wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
1024 	}
1025 
1026 	/*
1027 	 * running_proto must have been set before we allowed any mounts
1028 	 * to proceed.
1029 	 */
1030 	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
1031 		printk(KERN_ERR
1032 		       "Unable to mount with fs locking protocol version "
1033 		       "%u.%u because negotiated protocol is %u.%u\n",
1034 		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
1035 		       running_proto.pv_major, running_proto.pv_minor);
1036 		rc = -EPROTO;
1037 		ocfs2_live_connection_drop(lc);
1038 		lc = NULL;
1039 	}
1040 
1041 out:
1042 	if (rc)
1043 		kfree(lc);
1044 	return rc;
1045 }
1046 
1047 
1048 static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
1049 				  unsigned int *this_node)
1050 {
1051 	int rc;
1052 	struct ocfs2_live_connection *lc = conn->cc_private;
1053 
1054 	if (lc->oc_type == WITH_CONTROLD)
1055 		rc = ocfs2_control_get_this_node();
1056 	else if (lc->oc_type == NO_CONTROLD)
1057 		rc = atomic_read(&lc->oc_this_node);
1058 	else
1059 		rc = -EINVAL;
1060 
1061 	if (rc < 0)
1062 		return rc;
1063 
1064 	*this_node = rc;
1065 	return 0;
1066 }
1067 
1068 static const struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
1069 	.connect	= user_cluster_connect,
1070 	.disconnect	= user_cluster_disconnect,
1071 	.this_node	= user_cluster_this_node,
1072 	.dlm_lock	= user_dlm_lock,
1073 	.dlm_unlock	= user_dlm_unlock,
1074 	.lock_status	= user_dlm_lock_status,
1075 	.lvb_valid	= user_dlm_lvb_valid,
1076 	.lock_lvb	= user_dlm_lvb,
1077 	.plock		= user_plock,
1078 	.dump_lksb	= user_dlm_dump_lksb,
1079 };
1080 
1081 static struct ocfs2_stack_plugin ocfs2_user_plugin = {
1082 	.sp_name	= "user",
1083 	.sp_ops		= &ocfs2_user_plugin_ops,
1084 	.sp_owner	= THIS_MODULE,
1085 };
1086 
1087 
1088 static int __init ocfs2_user_plugin_init(void)
1089 {
1090 	int rc;
1091 
1092 	rc = ocfs2_control_init();
1093 	if (!rc) {
1094 		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
1095 		if (rc)
1096 			ocfs2_control_exit();
1097 	}
1098 
1099 	return rc;
1100 }
1101 
1102 static void __exit ocfs2_user_plugin_exit(void)
1103 {
1104 	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
1105 	ocfs2_control_exit();
1106 }
1107 
1108 MODULE_AUTHOR("Oracle");
1109 MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
1110 MODULE_LICENSE("GPL");
1111 module_init(ocfs2_user_plugin_init);
1112 module_exit(ocfs2_user_plugin_exit);
1113