xref: /openbmc/linux/fs/ocfs2/stack_user.c (revision 9cfc5c90)
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 <linux/sched.h>
27 #include <asm/uaccess.h>
28 
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_max_proto.pv_major, and the minor number
67  * must be less than or equal to ...sp_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 #define VERSION_LOCK				"version_lock"
107 
108 enum ocfs2_connection_type {
109 	WITH_CONTROLD,
110 	NO_CONTROLD
111 };
112 
113 /*
114  * ocfs2_live_connection is refcounted because the filesystem and
115  * miscdevice sides can detach in different order.  Let's just be safe.
116  */
117 struct ocfs2_live_connection {
118 	struct list_head		oc_list;
119 	struct ocfs2_cluster_connection	*oc_conn;
120 	enum ocfs2_connection_type	oc_type;
121 	atomic_t                        oc_this_node;
122 	int                             oc_our_slot;
123 	struct dlm_lksb                 oc_version_lksb;
124 	char                            oc_lvb[DLM_LVB_LEN];
125 	struct completion               oc_sync_wait;
126 	wait_queue_head_t		oc_wait;
127 };
128 
129 struct ocfs2_control_private {
130 	struct list_head op_list;
131 	int op_state;
132 	int op_this_node;
133 	struct ocfs2_protocol_version op_proto;
134 };
135 
136 /* SETN<space><8-char-hex-nodenum><newline> */
137 struct ocfs2_control_message_setn {
138 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
139 	char	space;
140 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
141 	char	newline;
142 };
143 
144 /* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
145 struct ocfs2_control_message_setv {
146 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
147 	char	space1;
148 	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
149 	char	space2;
150 	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
151 	char	newline;
152 };
153 
154 /* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
155 struct ocfs2_control_message_down {
156 	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
157 	char	space1;
158 	char	uuid[OCFS2_TEXT_UUID_LEN];
159 	char	space2;
160 	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
161 	char	newline;
162 };
163 
164 union ocfs2_control_message {
165 	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
166 	struct ocfs2_control_message_setn	u_setn;
167 	struct ocfs2_control_message_setv	u_setv;
168 	struct ocfs2_control_message_down	u_down;
169 };
170 
171 static struct ocfs2_stack_plugin ocfs2_user_plugin;
172 
173 static atomic_t ocfs2_control_opened;
174 static int ocfs2_control_this_node = -1;
175 static struct ocfs2_protocol_version running_proto;
176 
177 static LIST_HEAD(ocfs2_live_connection_list);
178 static LIST_HEAD(ocfs2_control_private_list);
179 static DEFINE_MUTEX(ocfs2_control_lock);
180 
181 static inline void ocfs2_control_set_handshake_state(struct file *file,
182 						     int state)
183 {
184 	struct ocfs2_control_private *p = file->private_data;
185 	p->op_state = state;
186 }
187 
188 static inline int ocfs2_control_get_handshake_state(struct file *file)
189 {
190 	struct ocfs2_control_private *p = file->private_data;
191 	return p->op_state;
192 }
193 
194 static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
195 {
196 	size_t len = strlen(name);
197 	struct ocfs2_live_connection *c;
198 
199 	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
200 
201 	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
202 		if ((c->oc_conn->cc_namelen == len) &&
203 		    !strncmp(c->oc_conn->cc_name, name, len))
204 			return c;
205 	}
206 
207 	return NULL;
208 }
209 
210 /*
211  * ocfs2_live_connection structures are created underneath the ocfs2
212  * mount path.  Since the VFS prevents multiple calls to
213  * fill_super(), we can't get dupes here.
214  */
215 static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
216 				     struct ocfs2_live_connection *c)
217 {
218 	int rc = 0;
219 
220 	mutex_lock(&ocfs2_control_lock);
221 	c->oc_conn = conn;
222 
223 	if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
224 		list_add(&c->oc_list, &ocfs2_live_connection_list);
225 	else {
226 		printk(KERN_ERR
227 		       "ocfs2: Userspace control daemon is not present\n");
228 		rc = -ESRCH;
229 	}
230 
231 	mutex_unlock(&ocfs2_control_lock);
232 	return rc;
233 }
234 
235 /*
236  * This function disconnects the cluster connection from ocfs2_control.
237  * Afterwards, userspace can't affect the cluster connection.
238  */
239 static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
240 {
241 	mutex_lock(&ocfs2_control_lock);
242 	list_del_init(&c->oc_list);
243 	c->oc_conn = NULL;
244 	mutex_unlock(&ocfs2_control_lock);
245 
246 	kfree(c);
247 }
248 
249 static int ocfs2_control_cfu(void *target, size_t target_len,
250 			     const char __user *buf, size_t count)
251 {
252 	/* The T01 expects write(2) calls to have exactly one command */
253 	if ((count != target_len) ||
254 	    (count > sizeof(union ocfs2_control_message)))
255 		return -EINVAL;
256 
257 	if (copy_from_user(target, buf, target_len))
258 		return -EFAULT;
259 
260 	return 0;
261 }
262 
263 static ssize_t ocfs2_control_validate_protocol(struct file *file,
264 					       const char __user *buf,
265 					       size_t count)
266 {
267 	ssize_t ret;
268 	char kbuf[OCFS2_CONTROL_PROTO_LEN];
269 
270 	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
271 				buf, count);
272 	if (ret)
273 		return ret;
274 
275 	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
276 		return -EINVAL;
277 
278 	ocfs2_control_set_handshake_state(file,
279 					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
280 
281 	return count;
282 }
283 
284 static void ocfs2_control_send_down(const char *uuid,
285 				    int nodenum)
286 {
287 	struct ocfs2_live_connection *c;
288 
289 	mutex_lock(&ocfs2_control_lock);
290 
291 	c = ocfs2_connection_find(uuid);
292 	if (c) {
293 		BUG_ON(c->oc_conn == NULL);
294 		c->oc_conn->cc_recovery_handler(nodenum,
295 						c->oc_conn->cc_recovery_data);
296 	}
297 
298 	mutex_unlock(&ocfs2_control_lock);
299 }
300 
301 /*
302  * Called whenever configuration elements are sent to /dev/ocfs2_control.
303  * If all configuration elements are present, try to set the global
304  * values.  If there is a problem, return an error.  Skip any missing
305  * elements, and only bump ocfs2_control_opened when we have all elements
306  * and are successful.
307  */
308 static int ocfs2_control_install_private(struct file *file)
309 {
310 	int rc = 0;
311 	int set_p = 1;
312 	struct ocfs2_control_private *p = file->private_data;
313 
314 	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
315 
316 	mutex_lock(&ocfs2_control_lock);
317 
318 	if (p->op_this_node < 0) {
319 		set_p = 0;
320 	} else if ((ocfs2_control_this_node >= 0) &&
321 		   (ocfs2_control_this_node != p->op_this_node)) {
322 		rc = -EINVAL;
323 		goto out_unlock;
324 	}
325 
326 	if (!p->op_proto.pv_major) {
327 		set_p = 0;
328 	} else if (!list_empty(&ocfs2_live_connection_list) &&
329 		   ((running_proto.pv_major != p->op_proto.pv_major) ||
330 		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
331 		rc = -EINVAL;
332 		goto out_unlock;
333 	}
334 
335 	if (set_p) {
336 		ocfs2_control_this_node = p->op_this_node;
337 		running_proto.pv_major = p->op_proto.pv_major;
338 		running_proto.pv_minor = p->op_proto.pv_minor;
339 	}
340 
341 out_unlock:
342 	mutex_unlock(&ocfs2_control_lock);
343 
344 	if (!rc && set_p) {
345 		/* We set the global values successfully */
346 		atomic_inc(&ocfs2_control_opened);
347 		ocfs2_control_set_handshake_state(file,
348 					OCFS2_CONTROL_HANDSHAKE_VALID);
349 	}
350 
351 	return rc;
352 }
353 
354 static int ocfs2_control_get_this_node(void)
355 {
356 	int rc;
357 
358 	mutex_lock(&ocfs2_control_lock);
359 	if (ocfs2_control_this_node < 0)
360 		rc = -EINVAL;
361 	else
362 		rc = ocfs2_control_this_node;
363 	mutex_unlock(&ocfs2_control_lock);
364 
365 	return rc;
366 }
367 
368 static int ocfs2_control_do_setnode_msg(struct file *file,
369 					struct ocfs2_control_message_setn *msg)
370 {
371 	long nodenum;
372 	char *ptr = NULL;
373 	struct ocfs2_control_private *p = file->private_data;
374 
375 	if (ocfs2_control_get_handshake_state(file) !=
376 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
377 		return -EINVAL;
378 
379 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
380 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
381 		return -EINVAL;
382 
383 	if ((msg->space != ' ') || (msg->newline != '\n'))
384 		return -EINVAL;
385 	msg->space = msg->newline = '\0';
386 
387 	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
388 	if (!ptr || *ptr)
389 		return -EINVAL;
390 
391 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
392 	    (nodenum > INT_MAX) || (nodenum < 0))
393 		return -ERANGE;
394 	p->op_this_node = nodenum;
395 
396 	return ocfs2_control_install_private(file);
397 }
398 
399 static int ocfs2_control_do_setversion_msg(struct file *file,
400 					   struct ocfs2_control_message_setv *msg)
401  {
402 	long major, minor;
403 	char *ptr = NULL;
404 	struct ocfs2_control_private *p = file->private_data;
405 	struct ocfs2_protocol_version *max =
406 		&ocfs2_user_plugin.sp_max_proto;
407 
408 	if (ocfs2_control_get_handshake_state(file) !=
409 	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
410 		return -EINVAL;
411 
412 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
413 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
414 		return -EINVAL;
415 
416 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
417 	    (msg->newline != '\n'))
418 		return -EINVAL;
419 	msg->space1 = msg->space2 = msg->newline = '\0';
420 
421 	major = simple_strtol(msg->major, &ptr, 16);
422 	if (!ptr || *ptr)
423 		return -EINVAL;
424 	minor = simple_strtol(msg->minor, &ptr, 16);
425 	if (!ptr || *ptr)
426 		return -EINVAL;
427 
428 	/*
429 	 * The major must be between 1 and 255, inclusive.  The minor
430 	 * must be between 0 and 255, inclusive.  The version passed in
431 	 * must be within the maximum version supported by the filesystem.
432 	 */
433 	if ((major == LONG_MIN) || (major == LONG_MAX) ||
434 	    (major > (u8)-1) || (major < 1))
435 		return -ERANGE;
436 	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
437 	    (minor > (u8)-1) || (minor < 0))
438 		return -ERANGE;
439 	if ((major != max->pv_major) ||
440 	    (minor > max->pv_minor))
441 		return -EINVAL;
442 
443 	p->op_proto.pv_major = major;
444 	p->op_proto.pv_minor = minor;
445 
446 	return ocfs2_control_install_private(file);
447 }
448 
449 static int ocfs2_control_do_down_msg(struct file *file,
450 				     struct ocfs2_control_message_down *msg)
451 {
452 	long nodenum;
453 	char *p = NULL;
454 
455 	if (ocfs2_control_get_handshake_state(file) !=
456 	    OCFS2_CONTROL_HANDSHAKE_VALID)
457 		return -EINVAL;
458 
459 	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
460 		    OCFS2_CONTROL_MESSAGE_OP_LEN))
461 		return -EINVAL;
462 
463 	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
464 	    (msg->newline != '\n'))
465 		return -EINVAL;
466 	msg->space1 = msg->space2 = msg->newline = '\0';
467 
468 	nodenum = simple_strtol(msg->nodestr, &p, 16);
469 	if (!p || *p)
470 		return -EINVAL;
471 
472 	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
473 	    (nodenum > INT_MAX) || (nodenum < 0))
474 		return -ERANGE;
475 
476 	ocfs2_control_send_down(msg->uuid, nodenum);
477 
478 	return 0;
479 }
480 
481 static ssize_t ocfs2_control_message(struct file *file,
482 				     const char __user *buf,
483 				     size_t count)
484 {
485 	ssize_t ret;
486 	union ocfs2_control_message msg;
487 
488 	/* Try to catch padding issues */
489 	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
490 		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
491 
492 	memset(&msg, 0, sizeof(union ocfs2_control_message));
493 	ret = ocfs2_control_cfu(&msg, count, buf, count);
494 	if (ret)
495 		goto out;
496 
497 	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
498 	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
499 		     OCFS2_CONTROL_MESSAGE_OP_LEN))
500 		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
501 	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
502 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
503 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
504 		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
505 	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
506 		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
507 			  OCFS2_CONTROL_MESSAGE_OP_LEN))
508 		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
509 	else
510 		ret = -EINVAL;
511 
512 out:
513 	return ret ? ret : count;
514 }
515 
516 static ssize_t ocfs2_control_write(struct file *file,
517 				   const char __user *buf,
518 				   size_t count,
519 				   loff_t *ppos)
520 {
521 	ssize_t ret;
522 
523 	switch (ocfs2_control_get_handshake_state(file)) {
524 		case OCFS2_CONTROL_HANDSHAKE_INVALID:
525 			ret = -EINVAL;
526 			break;
527 
528 		case OCFS2_CONTROL_HANDSHAKE_READ:
529 			ret = ocfs2_control_validate_protocol(file, buf,
530 							      count);
531 			break;
532 
533 		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
534 		case OCFS2_CONTROL_HANDSHAKE_VALID:
535 			ret = ocfs2_control_message(file, buf, count);
536 			break;
537 
538 		default:
539 			BUG();
540 			ret = -EIO;
541 			break;
542 	}
543 
544 	return ret;
545 }
546 
547 /*
548  * This is a naive version.  If we ever have a new protocol, we'll expand
549  * it.  Probably using seq_file.
550  */
551 static ssize_t ocfs2_control_read(struct file *file,
552 				  char __user *buf,
553 				  size_t count,
554 				  loff_t *ppos)
555 {
556 	ssize_t ret;
557 
558 	ret = simple_read_from_buffer(buf, count, ppos,
559 			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
560 
561 	/* Have we read the whole protocol list? */
562 	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
563 		ocfs2_control_set_handshake_state(file,
564 						  OCFS2_CONTROL_HANDSHAKE_READ);
565 
566 	return ret;
567 }
568 
569 static int ocfs2_control_release(struct inode *inode, struct file *file)
570 {
571 	struct ocfs2_control_private *p = file->private_data;
572 
573 	mutex_lock(&ocfs2_control_lock);
574 
575 	if (ocfs2_control_get_handshake_state(file) !=
576 	    OCFS2_CONTROL_HANDSHAKE_VALID)
577 		goto out;
578 
579 	if (atomic_dec_and_test(&ocfs2_control_opened)) {
580 		if (!list_empty(&ocfs2_live_connection_list)) {
581 			/* XXX: Do bad things! */
582 			printk(KERN_ERR
583 			       "ocfs2: Unexpected release of ocfs2_control!\n"
584 			       "       Loss of cluster connection requires "
585 			       "an emergency restart!\n");
586 			emergency_restart();
587 		}
588 		/*
589 		 * Last valid close clears the node number and resets
590 		 * the locking protocol version
591 		 */
592 		ocfs2_control_this_node = -1;
593 		running_proto.pv_major = 0;
594 		running_proto.pv_minor = 0;
595 	}
596 
597 out:
598 	list_del_init(&p->op_list);
599 	file->private_data = NULL;
600 
601 	mutex_unlock(&ocfs2_control_lock);
602 
603 	kfree(p);
604 
605 	return 0;
606 }
607 
608 static int ocfs2_control_open(struct inode *inode, struct file *file)
609 {
610 	struct ocfs2_control_private *p;
611 
612 	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
613 	if (!p)
614 		return -ENOMEM;
615 	p->op_this_node = -1;
616 
617 	mutex_lock(&ocfs2_control_lock);
618 	file->private_data = p;
619 	list_add(&p->op_list, &ocfs2_control_private_list);
620 	mutex_unlock(&ocfs2_control_lock);
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 	.llseek  = default_llseek,
632 };
633 
634 static struct miscdevice ocfs2_control_device = {
635 	.minor		= MISC_DYNAMIC_MINOR,
636 	.name		= "ocfs2_control",
637 	.fops		= &ocfs2_control_fops,
638 };
639 
640 static int ocfs2_control_init(void)
641 {
642 	int rc;
643 
644 	atomic_set(&ocfs2_control_opened, 0);
645 
646 	rc = misc_register(&ocfs2_control_device);
647 	if (rc)
648 		printk(KERN_ERR
649 		       "ocfs2: Unable to register ocfs2_control device "
650 		       "(errno %d)\n",
651 		       -rc);
652 
653 	return rc;
654 }
655 
656 static void ocfs2_control_exit(void)
657 {
658 	misc_deregister(&ocfs2_control_device);
659 }
660 
661 static void fsdlm_lock_ast_wrapper(void *astarg)
662 {
663 	struct ocfs2_dlm_lksb *lksb = astarg;
664 	int status = lksb->lksb_fsdlm.sb_status;
665 
666 	/*
667 	 * For now we're punting on the issue of other non-standard errors
668 	 * where we can't tell if the unlock_ast or lock_ast should be called.
669 	 * The main "other error" that's possible is EINVAL which means the
670 	 * function was called with invalid args, which shouldn't be possible
671 	 * since the caller here is under our control.  Other non-standard
672 	 * errors probably fall into the same category, or otherwise are fatal
673 	 * which means we can't carry on anyway.
674 	 */
675 
676 	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
677 		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
678 	else
679 		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
680 }
681 
682 static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
683 {
684 	struct ocfs2_dlm_lksb *lksb = astarg;
685 
686 	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
687 }
688 
689 static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
690 			 int mode,
691 			 struct ocfs2_dlm_lksb *lksb,
692 			 u32 flags,
693 			 void *name,
694 			 unsigned int namelen)
695 {
696 	int ret;
697 
698 	if (!lksb->lksb_fsdlm.sb_lvbptr)
699 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
700 					     sizeof(struct dlm_lksb);
701 
702 	ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
703 		       flags|DLM_LKF_NODLCKWT, name, namelen, 0,
704 		       fsdlm_lock_ast_wrapper, lksb,
705 		       fsdlm_blocking_ast_wrapper);
706 	return ret;
707 }
708 
709 static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
710 			   struct ocfs2_dlm_lksb *lksb,
711 			   u32 flags)
712 {
713 	int ret;
714 
715 	ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
716 			 flags, &lksb->lksb_fsdlm, lksb);
717 	return ret;
718 }
719 
720 static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
721 {
722 	return lksb->lksb_fsdlm.sb_status;
723 }
724 
725 static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
726 {
727 	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
728 
729 	return !invalid;
730 }
731 
732 static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
733 {
734 	if (!lksb->lksb_fsdlm.sb_lvbptr)
735 		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
736 					     sizeof(struct dlm_lksb);
737 	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
738 }
739 
740 static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
741 {
742 }
743 
744 static int user_plock(struct ocfs2_cluster_connection *conn,
745 		      u64 ino,
746 		      struct file *file,
747 		      int cmd,
748 		      struct file_lock *fl)
749 {
750 	/*
751 	 * This more or less just demuxes the plock request into any
752 	 * one of three dlm calls.
753 	 *
754 	 * Internally, fs/dlm will pass these to a misc device, which
755 	 * a userspace daemon will read and write to.
756 	 *
757 	 * For now, cancel requests (which happen internally only),
758 	 * are turned into unlocks. Most of this function taken from
759 	 * gfs2_lock.
760 	 */
761 
762 	if (cmd == F_CANCELLK) {
763 		cmd = F_SETLK;
764 		fl->fl_type = F_UNLCK;
765 	}
766 
767 	if (IS_GETLK(cmd))
768 		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
769 	else if (fl->fl_type == F_UNLCK)
770 		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
771 	else
772 		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
773 }
774 
775 /*
776  * Compare a requested locking protocol version against the current one.
777  *
778  * If the major numbers are different, they are incompatible.
779  * If the current minor is greater than the request, they are incompatible.
780  * If the current minor is less than or equal to the request, they are
781  * compatible, and the requester should run at the current minor version.
782  */
783 static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
784 			       struct ocfs2_protocol_version *request)
785 {
786 	if (existing->pv_major != request->pv_major)
787 		return 1;
788 
789 	if (existing->pv_minor > request->pv_minor)
790 		return 1;
791 
792 	if (existing->pv_minor < request->pv_minor)
793 		request->pv_minor = existing->pv_minor;
794 
795 	return 0;
796 }
797 
798 static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
799 {
800 	struct ocfs2_protocol_version *pv =
801 		(struct ocfs2_protocol_version *)lvb;
802 	/*
803 	 * ocfs2_protocol_version has two u8 variables, so we don't
804 	 * need any endian conversion.
805 	 */
806 	ver->pv_major = pv->pv_major;
807 	ver->pv_minor = pv->pv_minor;
808 }
809 
810 static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
811 {
812 	struct ocfs2_protocol_version *pv =
813 		(struct ocfs2_protocol_version *)lvb;
814 	/*
815 	 * ocfs2_protocol_version has two u8 variables, so we don't
816 	 * need any endian conversion.
817 	 */
818 	pv->pv_major = ver->pv_major;
819 	pv->pv_minor = ver->pv_minor;
820 }
821 
822 static void sync_wait_cb(void *arg)
823 {
824 	struct ocfs2_cluster_connection *conn = arg;
825 	struct ocfs2_live_connection *lc = conn->cc_private;
826 	complete(&lc->oc_sync_wait);
827 }
828 
829 static int sync_unlock(struct ocfs2_cluster_connection *conn,
830 		struct dlm_lksb *lksb, char *name)
831 {
832 	int error;
833 	struct ocfs2_live_connection *lc = conn->cc_private;
834 
835 	error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
836 	if (error) {
837 		printk(KERN_ERR "%s lkid %x error %d\n",
838 				name, lksb->sb_lkid, error);
839 		return error;
840 	}
841 
842 	wait_for_completion(&lc->oc_sync_wait);
843 
844 	if (lksb->sb_status != -DLM_EUNLOCK) {
845 		printk(KERN_ERR "%s lkid %x status %d\n",
846 				name, lksb->sb_lkid, lksb->sb_status);
847 		return -1;
848 	}
849 	return 0;
850 }
851 
852 static int sync_lock(struct ocfs2_cluster_connection *conn,
853 		int mode, uint32_t flags,
854 		struct dlm_lksb *lksb, char *name)
855 {
856 	int error, status;
857 	struct ocfs2_live_connection *lc = conn->cc_private;
858 
859 	error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
860 			name, strlen(name),
861 			0, sync_wait_cb, conn, NULL);
862 	if (error) {
863 		printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
864 				name, lksb->sb_lkid, flags, mode, error);
865 		return error;
866 	}
867 
868 	wait_for_completion(&lc->oc_sync_wait);
869 
870 	status = lksb->sb_status;
871 
872 	if (status && status != -EAGAIN) {
873 		printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
874 				name, lksb->sb_lkid, flags, mode, status);
875 	}
876 
877 	return status;
878 }
879 
880 
881 static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
882 		int flags)
883 {
884 	struct ocfs2_live_connection *lc = conn->cc_private;
885 	return sync_lock(conn, mode, flags,
886 			&lc->oc_version_lksb, VERSION_LOCK);
887 }
888 
889 static int version_unlock(struct ocfs2_cluster_connection *conn)
890 {
891 	struct ocfs2_live_connection *lc = conn->cc_private;
892 	return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
893 }
894 
895 /* get_protocol_version()
896  *
897  * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
898  * The algorithm is:
899  * 1. Attempt to take the lock in EX mode (non-blocking).
900  * 2. If successful (which means it is the first mount), write the
901  *    version number and downconvert to PR lock.
902  * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
903  *    taking the PR lock.
904  */
905 
906 static int get_protocol_version(struct ocfs2_cluster_connection *conn)
907 {
908 	int ret;
909 	struct ocfs2_live_connection *lc = conn->cc_private;
910 	struct ocfs2_protocol_version pv;
911 
912 	running_proto.pv_major =
913 		ocfs2_user_plugin.sp_max_proto.pv_major;
914 	running_proto.pv_minor =
915 		ocfs2_user_plugin.sp_max_proto.pv_minor;
916 
917 	lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
918 	ret = version_lock(conn, DLM_LOCK_EX,
919 			DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
920 	if (!ret) {
921 		conn->cc_version.pv_major = running_proto.pv_major;
922 		conn->cc_version.pv_minor = running_proto.pv_minor;
923 		version_to_lvb(&running_proto, lc->oc_lvb);
924 		version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
925 	} else if (ret == -EAGAIN) {
926 		ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
927 		if (ret)
928 			goto out;
929 		lvb_to_version(lc->oc_lvb, &pv);
930 
931 		if ((pv.pv_major != running_proto.pv_major) ||
932 				(pv.pv_minor > running_proto.pv_minor)) {
933 			ret = -EINVAL;
934 			goto out;
935 		}
936 
937 		conn->cc_version.pv_major = pv.pv_major;
938 		conn->cc_version.pv_minor = pv.pv_minor;
939 	}
940 out:
941 	return ret;
942 }
943 
944 static void user_recover_prep(void *arg)
945 {
946 }
947 
948 static void user_recover_slot(void *arg, struct dlm_slot *slot)
949 {
950 	struct ocfs2_cluster_connection *conn = arg;
951 	printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
952 			slot->nodeid, slot->slot);
953 	conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
954 
955 }
956 
957 static void user_recover_done(void *arg, struct dlm_slot *slots,
958 		int num_slots, int our_slot,
959 		uint32_t generation)
960 {
961 	struct ocfs2_cluster_connection *conn = arg;
962 	struct ocfs2_live_connection *lc = conn->cc_private;
963 	int i;
964 
965 	for (i = 0; i < num_slots; i++)
966 		if (slots[i].slot == our_slot) {
967 			atomic_set(&lc->oc_this_node, slots[i].nodeid);
968 			break;
969 		}
970 
971 	lc->oc_our_slot = our_slot;
972 	wake_up(&lc->oc_wait);
973 }
974 
975 static const struct dlm_lockspace_ops ocfs2_ls_ops = {
976 	.recover_prep = user_recover_prep,
977 	.recover_slot = user_recover_slot,
978 	.recover_done = user_recover_done,
979 };
980 
981 static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
982 {
983 	version_unlock(conn);
984 	dlm_release_lockspace(conn->cc_lockspace, 2);
985 	conn->cc_lockspace = NULL;
986 	ocfs2_live_connection_drop(conn->cc_private);
987 	conn->cc_private = NULL;
988 	return 0;
989 }
990 
991 static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
992 {
993 	dlm_lockspace_t *fsdlm;
994 	struct ocfs2_live_connection *lc;
995 	int rc, ops_rv;
996 
997 	BUG_ON(conn == NULL);
998 
999 	lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
1000 	if (!lc)
1001 		return -ENOMEM;
1002 
1003 	init_waitqueue_head(&lc->oc_wait);
1004 	init_completion(&lc->oc_sync_wait);
1005 	atomic_set(&lc->oc_this_node, 0);
1006 	conn->cc_private = lc;
1007 	lc->oc_type = NO_CONTROLD;
1008 
1009 	rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
1010 			       DLM_LSFL_FS, DLM_LVB_LEN,
1011 			       &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
1012 	if (rc)
1013 		goto out;
1014 
1015 	if (ops_rv == -EOPNOTSUPP) {
1016 		lc->oc_type = WITH_CONTROLD;
1017 		printk(KERN_NOTICE "ocfs2: You seem to be using an older "
1018 				"version of dlm_controld and/or ocfs2-tools."
1019 				" Please consider upgrading.\n");
1020 	} else if (ops_rv) {
1021 		rc = ops_rv;
1022 		goto out;
1023 	}
1024 	conn->cc_lockspace = fsdlm;
1025 
1026 	rc = ocfs2_live_connection_attach(conn, lc);
1027 	if (rc)
1028 		goto out;
1029 
1030 	if (lc->oc_type == NO_CONTROLD) {
1031 		rc = get_protocol_version(conn);
1032 		if (rc) {
1033 			printk(KERN_ERR "ocfs2: Could not determine"
1034 					" locking version\n");
1035 			user_cluster_disconnect(conn);
1036 			goto out;
1037 		}
1038 		wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
1039 	}
1040 
1041 	/*
1042 	 * running_proto must have been set before we allowed any mounts
1043 	 * to proceed.
1044 	 */
1045 	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
1046 		printk(KERN_ERR
1047 		       "Unable to mount with fs locking protocol version "
1048 		       "%u.%u because negotiated protocol is %u.%u\n",
1049 		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
1050 		       running_proto.pv_major, running_proto.pv_minor);
1051 		rc = -EPROTO;
1052 		ocfs2_live_connection_drop(lc);
1053 		lc = NULL;
1054 	}
1055 
1056 out:
1057 	if (rc)
1058 		kfree(lc);
1059 	return rc;
1060 }
1061 
1062 
1063 static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
1064 				  unsigned int *this_node)
1065 {
1066 	int rc;
1067 	struct ocfs2_live_connection *lc = conn->cc_private;
1068 
1069 	if (lc->oc_type == WITH_CONTROLD)
1070 		rc = ocfs2_control_get_this_node();
1071 	else if (lc->oc_type == NO_CONTROLD)
1072 		rc = atomic_read(&lc->oc_this_node);
1073 	else
1074 		rc = -EINVAL;
1075 
1076 	if (rc < 0)
1077 		return rc;
1078 
1079 	*this_node = rc;
1080 	return 0;
1081 }
1082 
1083 static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
1084 	.connect	= user_cluster_connect,
1085 	.disconnect	= user_cluster_disconnect,
1086 	.this_node	= user_cluster_this_node,
1087 	.dlm_lock	= user_dlm_lock,
1088 	.dlm_unlock	= user_dlm_unlock,
1089 	.lock_status	= user_dlm_lock_status,
1090 	.lvb_valid	= user_dlm_lvb_valid,
1091 	.lock_lvb	= user_dlm_lvb,
1092 	.plock		= user_plock,
1093 	.dump_lksb	= user_dlm_dump_lksb,
1094 };
1095 
1096 static struct ocfs2_stack_plugin ocfs2_user_plugin = {
1097 	.sp_name	= "user",
1098 	.sp_ops		= &ocfs2_user_plugin_ops,
1099 	.sp_owner	= THIS_MODULE,
1100 };
1101 
1102 
1103 static int __init ocfs2_user_plugin_init(void)
1104 {
1105 	int rc;
1106 
1107 	rc = ocfs2_control_init();
1108 	if (!rc) {
1109 		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
1110 		if (rc)
1111 			ocfs2_control_exit();
1112 	}
1113 
1114 	return rc;
1115 }
1116 
1117 static void __exit ocfs2_user_plugin_exit(void)
1118 {
1119 	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
1120 	ocfs2_control_exit();
1121 }
1122 
1123 MODULE_AUTHOR("Oracle");
1124 MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
1125 MODULE_LICENSE("GPL");
1126 module_init(ocfs2_user_plugin_init);
1127 module_exit(ocfs2_user_plugin_exit);
1128