Linux Audio

Check our new training course

Loading...
v5.4
   1// SPDX-License-Identifier: GPL-2.0-only
   2/* -*- mode: c; c-basic-offset: 8; -*-
   3 * vim: noexpandtab sw=8 ts=8 sts=0:
   4 *
   5 * stack_user.c
   6 *
   7 * Code which interfaces ocfs2 with fs/dlm and a userspace stack.
   8 *
   9 * Copyright (C) 2007 Oracle.  All rights reserved.
  10 */
  11
  12#include <linux/module.h>
  13#include <linux/fs.h>
  14#include <linux/miscdevice.h>
  15#include <linux/mutex.h>
  16#include <linux/slab.h>
  17#include <linux/reboot.h>
  18#include <linux/sched.h>
  19#include <linux/uaccess.h>
  20
  21#include "stackglue.h"
  22
  23#include <linux/dlm_plock.h>
  24
  25/*
  26 * The control protocol starts with a handshake.  Until the handshake
  27 * is complete, the control device will fail all write(2)s.
  28 *
  29 * The handshake is simple.  First, the client reads until EOF.  Each line
  30 * of output is a supported protocol tag.  All protocol tags are a single
  31 * character followed by a two hex digit version number.  Currently the
  32 * only things supported is T01, for "Text-base version 0x01".  Next, the
  33 * client writes the version they would like to use, including the newline.
  34 * Thus, the protocol tag is 'T01\n'.  If the version tag written is
  35 * unknown, -EINVAL is returned.  Once the negotiation is complete, the
  36 * client can start sending messages.
  37 *
  38 * The T01 protocol has three messages.  First is the "SETN" message.
  39 * It has the following syntax:
  40 *
  41 *  SETN<space><8-char-hex-nodenum><newline>
  42 *
  43 * This is 14 characters.
  44 *
  45 * The "SETN" message must be the first message following the protocol.
  46 * It tells ocfs2_control the local node number.
  47 *
  48 * Next comes the "SETV" message.  It has the following syntax:
  49 *
  50 *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
  51 *
  52 * This is 11 characters.
  53 *
  54 * The "SETV" message sets the filesystem locking protocol version as
  55 * negotiated by the client.  The client negotiates based on the maximum
  56 * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
  57 * number from the "SETV" message must match
  58 * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
  59 * must be less than or equal to ...sp_max_version.pv_minor.
  60 *
  61 * Once this information has been set, mounts will be allowed.  From this
  62 * point on, the "DOWN" message can be sent for node down notification.
  63 * It has the following syntax:
  64 *
  65 *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
  66 *
  67 * eg:
  68 *
  69 *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
  70 *
  71 * This is 47 characters.
  72 */
  73
  74/*
  75 * Whether or not the client has done the handshake.
  76 * For now, we have just one protocol version.
  77 */
  78#define OCFS2_CONTROL_PROTO			"T01\n"
  79#define OCFS2_CONTROL_PROTO_LEN			4
  80
  81/* Handshake states */
  82#define OCFS2_CONTROL_HANDSHAKE_INVALID		(0)
  83#define OCFS2_CONTROL_HANDSHAKE_READ		(1)
  84#define OCFS2_CONTROL_HANDSHAKE_PROTOCOL	(2)
  85#define OCFS2_CONTROL_HANDSHAKE_VALID		(3)
  86
  87/* Messages */
  88#define OCFS2_CONTROL_MESSAGE_OP_LEN		4
  89#define OCFS2_CONTROL_MESSAGE_SETNODE_OP	"SETN"
  90#define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN	14
  91#define OCFS2_CONTROL_MESSAGE_SETVERSION_OP	"SETV"
  92#define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN	11
  93#define OCFS2_CONTROL_MESSAGE_DOWN_OP		"DOWN"
  94#define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN	47
  95#define OCFS2_TEXT_UUID_LEN			32
  96#define OCFS2_CONTROL_MESSAGE_VERNUM_LEN	2
  97#define OCFS2_CONTROL_MESSAGE_NODENUM_LEN	8
  98#define VERSION_LOCK				"version_lock"
  99
 100enum ocfs2_connection_type {
 101	WITH_CONTROLD,
 102	NO_CONTROLD
 103};
 104
 105/*
 106 * ocfs2_live_connection is refcounted because the filesystem and
 107 * miscdevice sides can detach in different order.  Let's just be safe.
 108 */
 109struct ocfs2_live_connection {
 110	struct list_head		oc_list;
 111	struct ocfs2_cluster_connection	*oc_conn;
 112	enum ocfs2_connection_type	oc_type;
 113	atomic_t                        oc_this_node;
 114	int                             oc_our_slot;
 115	struct dlm_lksb                 oc_version_lksb;
 116	char                            oc_lvb[DLM_LVB_LEN];
 117	struct completion               oc_sync_wait;
 118	wait_queue_head_t		oc_wait;
 119};
 120
 121struct ocfs2_control_private {
 122	struct list_head op_list;
 123	int op_state;
 124	int op_this_node;
 125	struct ocfs2_protocol_version op_proto;
 126};
 127
 128/* SETN<space><8-char-hex-nodenum><newline> */
 129struct ocfs2_control_message_setn {
 130	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 131	char	space;
 132	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
 133	char	newline;
 134};
 135
 136/* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
 137struct ocfs2_control_message_setv {
 138	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 139	char	space1;
 140	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
 141	char	space2;
 142	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
 143	char	newline;
 144};
 145
 146/* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
 147struct ocfs2_control_message_down {
 148	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 149	char	space1;
 150	char	uuid[OCFS2_TEXT_UUID_LEN];
 151	char	space2;
 152	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
 153	char	newline;
 154};
 155
 156union ocfs2_control_message {
 157	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 158	struct ocfs2_control_message_setn	u_setn;
 159	struct ocfs2_control_message_setv	u_setv;
 160	struct ocfs2_control_message_down	u_down;
 161};
 162
 163static struct ocfs2_stack_plugin ocfs2_user_plugin;
 164
 165static atomic_t ocfs2_control_opened;
 166static int ocfs2_control_this_node = -1;
 167static struct ocfs2_protocol_version running_proto;
 168
 169static LIST_HEAD(ocfs2_live_connection_list);
 170static LIST_HEAD(ocfs2_control_private_list);
 171static DEFINE_MUTEX(ocfs2_control_lock);
 172
 173static inline void ocfs2_control_set_handshake_state(struct file *file,
 174						     int state)
 175{
 176	struct ocfs2_control_private *p = file->private_data;
 177	p->op_state = state;
 178}
 179
 180static inline int ocfs2_control_get_handshake_state(struct file *file)
 181{
 182	struct ocfs2_control_private *p = file->private_data;
 183	return p->op_state;
 184}
 185
 186static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
 187{
 188	size_t len = strlen(name);
 189	struct ocfs2_live_connection *c;
 190
 191	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
 192
 193	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
 194		if ((c->oc_conn->cc_namelen == len) &&
 195		    !strncmp(c->oc_conn->cc_name, name, len))
 196			return c;
 197	}
 198
 199	return NULL;
 200}
 201
 202/*
 203 * ocfs2_live_connection structures are created underneath the ocfs2
 204 * mount path.  Since the VFS prevents multiple calls to
 205 * fill_super(), we can't get dupes here.
 206 */
 207static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
 208				     struct ocfs2_live_connection *c)
 209{
 210	int rc = 0;
 211
 212	mutex_lock(&ocfs2_control_lock);
 213	c->oc_conn = conn;
 214
 215	if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
 216		list_add(&c->oc_list, &ocfs2_live_connection_list);
 217	else {
 218		printk(KERN_ERR
 219		       "ocfs2: Userspace control daemon is not present\n");
 220		rc = -ESRCH;
 221	}
 222
 223	mutex_unlock(&ocfs2_control_lock);
 224	return rc;
 225}
 226
 227/*
 228 * This function disconnects the cluster connection from ocfs2_control.
 229 * Afterwards, userspace can't affect the cluster connection.
 230 */
 231static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
 232{
 233	mutex_lock(&ocfs2_control_lock);
 234	list_del_init(&c->oc_list);
 235	c->oc_conn = NULL;
 236	mutex_unlock(&ocfs2_control_lock);
 237
 238	kfree(c);
 239}
 240
 241static int ocfs2_control_cfu(void *target, size_t target_len,
 242			     const char __user *buf, size_t count)
 243{
 244	/* The T01 expects write(2) calls to have exactly one command */
 245	if ((count != target_len) ||
 246	    (count > sizeof(union ocfs2_control_message)))
 247		return -EINVAL;
 248
 249	if (copy_from_user(target, buf, target_len))
 250		return -EFAULT;
 251
 252	return 0;
 253}
 254
 255static ssize_t ocfs2_control_validate_protocol(struct file *file,
 256					       const char __user *buf,
 257					       size_t count)
 258{
 259	ssize_t ret;
 260	char kbuf[OCFS2_CONTROL_PROTO_LEN];
 261
 262	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
 263				buf, count);
 264	if (ret)
 265		return ret;
 266
 267	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
 268		return -EINVAL;
 269
 270	ocfs2_control_set_handshake_state(file,
 271					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
 272
 273	return count;
 274}
 275
 276static void ocfs2_control_send_down(const char *uuid,
 277				    int nodenum)
 278{
 279	struct ocfs2_live_connection *c;
 280
 281	mutex_lock(&ocfs2_control_lock);
 282
 283	c = ocfs2_connection_find(uuid);
 284	if (c) {
 285		BUG_ON(c->oc_conn == NULL);
 286		c->oc_conn->cc_recovery_handler(nodenum,
 287						c->oc_conn->cc_recovery_data);
 288	}
 289
 290	mutex_unlock(&ocfs2_control_lock);
 291}
 292
 293/*
 294 * Called whenever configuration elements are sent to /dev/ocfs2_control.
 295 * If all configuration elements are present, try to set the global
 296 * values.  If there is a problem, return an error.  Skip any missing
 297 * elements, and only bump ocfs2_control_opened when we have all elements
 298 * and are successful.
 299 */
 300static int ocfs2_control_install_private(struct file *file)
 301{
 302	int rc = 0;
 303	int set_p = 1;
 304	struct ocfs2_control_private *p = file->private_data;
 305
 306	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
 307
 308	mutex_lock(&ocfs2_control_lock);
 309
 310	if (p->op_this_node < 0) {
 311		set_p = 0;
 312	} else if ((ocfs2_control_this_node >= 0) &&
 313		   (ocfs2_control_this_node != p->op_this_node)) {
 314		rc = -EINVAL;
 315		goto out_unlock;
 316	}
 317
 318	if (!p->op_proto.pv_major) {
 319		set_p = 0;
 320	} else if (!list_empty(&ocfs2_live_connection_list) &&
 321		   ((running_proto.pv_major != p->op_proto.pv_major) ||
 322		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
 323		rc = -EINVAL;
 324		goto out_unlock;
 325	}
 326
 327	if (set_p) {
 328		ocfs2_control_this_node = p->op_this_node;
 329		running_proto.pv_major = p->op_proto.pv_major;
 330		running_proto.pv_minor = p->op_proto.pv_minor;
 331	}
 332
 333out_unlock:
 334	mutex_unlock(&ocfs2_control_lock);
 335
 336	if (!rc && set_p) {
 337		/* We set the global values successfully */
 338		atomic_inc(&ocfs2_control_opened);
 339		ocfs2_control_set_handshake_state(file,
 340					OCFS2_CONTROL_HANDSHAKE_VALID);
 341	}
 342
 343	return rc;
 344}
 345
 346static int ocfs2_control_get_this_node(void)
 347{
 348	int rc;
 349
 350	mutex_lock(&ocfs2_control_lock);
 351	if (ocfs2_control_this_node < 0)
 352		rc = -EINVAL;
 353	else
 354		rc = ocfs2_control_this_node;
 355	mutex_unlock(&ocfs2_control_lock);
 356
 357	return rc;
 358}
 359
 360static int ocfs2_control_do_setnode_msg(struct file *file,
 361					struct ocfs2_control_message_setn *msg)
 362{
 363	long nodenum;
 364	char *ptr = NULL;
 365	struct ocfs2_control_private *p = file->private_data;
 366
 367	if (ocfs2_control_get_handshake_state(file) !=
 368	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
 369		return -EINVAL;
 370
 371	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
 372		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 373		return -EINVAL;
 374
 375	if ((msg->space != ' ') || (msg->newline != '\n'))
 376		return -EINVAL;
 377	msg->space = msg->newline = '\0';
 378
 379	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
 380	if (!ptr || *ptr)
 381		return -EINVAL;
 382
 383	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
 384	    (nodenum > INT_MAX) || (nodenum < 0))
 385		return -ERANGE;
 386	p->op_this_node = nodenum;
 387
 388	return ocfs2_control_install_private(file);
 389}
 390
 391static int ocfs2_control_do_setversion_msg(struct file *file,
 392					   struct ocfs2_control_message_setv *msg)
 393{
 394	long major, minor;
 395	char *ptr = NULL;
 396	struct ocfs2_control_private *p = file->private_data;
 397	struct ocfs2_protocol_version *max =
 398		&ocfs2_user_plugin.sp_max_proto;
 399
 400	if (ocfs2_control_get_handshake_state(file) !=
 401	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
 402		return -EINVAL;
 403
 404	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
 405		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 406		return -EINVAL;
 407
 408	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
 409	    (msg->newline != '\n'))
 410		return -EINVAL;
 411	msg->space1 = msg->space2 = msg->newline = '\0';
 412
 413	major = simple_strtol(msg->major, &ptr, 16);
 414	if (!ptr || *ptr)
 415		return -EINVAL;
 416	minor = simple_strtol(msg->minor, &ptr, 16);
 417	if (!ptr || *ptr)
 418		return -EINVAL;
 419
 420	/*
 421	 * The major must be between 1 and 255, inclusive.  The minor
 422	 * must be between 0 and 255, inclusive.  The version passed in
 423	 * must be within the maximum version supported by the filesystem.
 424	 */
 425	if ((major == LONG_MIN) || (major == LONG_MAX) ||
 426	    (major > (u8)-1) || (major < 1))
 427		return -ERANGE;
 428	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
 429	    (minor > (u8)-1) || (minor < 0))
 430		return -ERANGE;
 431	if ((major != max->pv_major) ||
 432	    (minor > max->pv_minor))
 433		return -EINVAL;
 434
 435	p->op_proto.pv_major = major;
 436	p->op_proto.pv_minor = minor;
 437
 438	return ocfs2_control_install_private(file);
 439}
 440
 441static int ocfs2_control_do_down_msg(struct file *file,
 442				     struct ocfs2_control_message_down *msg)
 443{
 444	long nodenum;
 445	char *p = NULL;
 446
 447	if (ocfs2_control_get_handshake_state(file) !=
 448	    OCFS2_CONTROL_HANDSHAKE_VALID)
 449		return -EINVAL;
 450
 451	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
 452		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 453		return -EINVAL;
 454
 455	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
 456	    (msg->newline != '\n'))
 457		return -EINVAL;
 458	msg->space1 = msg->space2 = msg->newline = '\0';
 459
 460	nodenum = simple_strtol(msg->nodestr, &p, 16);
 461	if (!p || *p)
 462		return -EINVAL;
 463
 464	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
 465	    (nodenum > INT_MAX) || (nodenum < 0))
 466		return -ERANGE;
 467
 468	ocfs2_control_send_down(msg->uuid, nodenum);
 469
 470	return 0;
 471}
 472
 473static ssize_t ocfs2_control_message(struct file *file,
 474				     const char __user *buf,
 475				     size_t count)
 476{
 477	ssize_t ret;
 478	union ocfs2_control_message msg;
 479
 480	/* Try to catch padding issues */
 481	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
 482		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
 483
 484	memset(&msg, 0, sizeof(union ocfs2_control_message));
 485	ret = ocfs2_control_cfu(&msg, count, buf, count);
 486	if (ret)
 487		goto out;
 488
 489	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
 490	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
 491		     OCFS2_CONTROL_MESSAGE_OP_LEN))
 492		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
 493	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
 494		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
 495			  OCFS2_CONTROL_MESSAGE_OP_LEN))
 496		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
 497	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
 498		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
 499			  OCFS2_CONTROL_MESSAGE_OP_LEN))
 500		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
 501	else
 502		ret = -EINVAL;
 503
 504out:
 505	return ret ? ret : count;
 506}
 507
 508static ssize_t ocfs2_control_write(struct file *file,
 509				   const char __user *buf,
 510				   size_t count,
 511				   loff_t *ppos)
 512{
 513	ssize_t ret;
 514
 515	switch (ocfs2_control_get_handshake_state(file)) {
 516		case OCFS2_CONTROL_HANDSHAKE_INVALID:
 517			ret = -EINVAL;
 518			break;
 519
 520		case OCFS2_CONTROL_HANDSHAKE_READ:
 521			ret = ocfs2_control_validate_protocol(file, buf,
 522							      count);
 523			break;
 524
 525		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
 526		case OCFS2_CONTROL_HANDSHAKE_VALID:
 527			ret = ocfs2_control_message(file, buf, count);
 528			break;
 529
 530		default:
 531			BUG();
 532			ret = -EIO;
 533			break;
 534	}
 535
 536	return ret;
 537}
 538
 539/*
 540 * This is a naive version.  If we ever have a new protocol, we'll expand
 541 * it.  Probably using seq_file.
 542 */
 543static ssize_t ocfs2_control_read(struct file *file,
 544				  char __user *buf,
 545				  size_t count,
 546				  loff_t *ppos)
 547{
 548	ssize_t ret;
 549
 550	ret = simple_read_from_buffer(buf, count, ppos,
 551			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
 552
 553	/* Have we read the whole protocol list? */
 554	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
 555		ocfs2_control_set_handshake_state(file,
 556						  OCFS2_CONTROL_HANDSHAKE_READ);
 557
 558	return ret;
 559}
 560
 561static int ocfs2_control_release(struct inode *inode, struct file *file)
 562{
 563	struct ocfs2_control_private *p = file->private_data;
 564
 565	mutex_lock(&ocfs2_control_lock);
 566
 567	if (ocfs2_control_get_handshake_state(file) !=
 568	    OCFS2_CONTROL_HANDSHAKE_VALID)
 569		goto out;
 570
 571	if (atomic_dec_and_test(&ocfs2_control_opened)) {
 572		if (!list_empty(&ocfs2_live_connection_list)) {
 573			/* XXX: Do bad things! */
 574			printk(KERN_ERR
 575			       "ocfs2: Unexpected release of ocfs2_control!\n"
 576			       "       Loss of cluster connection requires "
 577			       "an emergency restart!\n");
 578			emergency_restart();
 579		}
 580		/*
 581		 * Last valid close clears the node number and resets
 582		 * the locking protocol version
 583		 */
 584		ocfs2_control_this_node = -1;
 585		running_proto.pv_major = 0;
 586		running_proto.pv_minor = 0;
 587	}
 588
 589out:
 590	list_del_init(&p->op_list);
 591	file->private_data = NULL;
 592
 593	mutex_unlock(&ocfs2_control_lock);
 594
 595	kfree(p);
 596
 597	return 0;
 598}
 599
 600static int ocfs2_control_open(struct inode *inode, struct file *file)
 601{
 602	struct ocfs2_control_private *p;
 603
 604	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
 605	if (!p)
 606		return -ENOMEM;
 607	p->op_this_node = -1;
 608
 609	mutex_lock(&ocfs2_control_lock);
 610	file->private_data = p;
 611	list_add(&p->op_list, &ocfs2_control_private_list);
 612	mutex_unlock(&ocfs2_control_lock);
 613
 614	return 0;
 615}
 616
 617static const struct file_operations ocfs2_control_fops = {
 618	.open    = ocfs2_control_open,
 619	.release = ocfs2_control_release,
 620	.read    = ocfs2_control_read,
 621	.write   = ocfs2_control_write,
 622	.owner   = THIS_MODULE,
 623	.llseek  = default_llseek,
 624};
 625
 626static struct miscdevice ocfs2_control_device = {
 627	.minor		= MISC_DYNAMIC_MINOR,
 628	.name		= "ocfs2_control",
 629	.fops		= &ocfs2_control_fops,
 630};
 631
 632static int ocfs2_control_init(void)
 633{
 634	int rc;
 635
 636	atomic_set(&ocfs2_control_opened, 0);
 637
 638	rc = misc_register(&ocfs2_control_device);
 639	if (rc)
 640		printk(KERN_ERR
 641		       "ocfs2: Unable to register ocfs2_control device "
 642		       "(errno %d)\n",
 643		       -rc);
 644
 645	return rc;
 646}
 647
 648static void ocfs2_control_exit(void)
 649{
 650	misc_deregister(&ocfs2_control_device);
 651}
 652
 653static void fsdlm_lock_ast_wrapper(void *astarg)
 654{
 655	struct ocfs2_dlm_lksb *lksb = astarg;
 656	int status = lksb->lksb_fsdlm.sb_status;
 657
 658	/*
 659	 * For now we're punting on the issue of other non-standard errors
 660	 * where we can't tell if the unlock_ast or lock_ast should be called.
 661	 * The main "other error" that's possible is EINVAL which means the
 662	 * function was called with invalid args, which shouldn't be possible
 663	 * since the caller here is under our control.  Other non-standard
 664	 * errors probably fall into the same category, or otherwise are fatal
 665	 * which means we can't carry on anyway.
 666	 */
 667
 668	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
 669		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
 670	else
 671		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
 672}
 673
 674static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
 675{
 676	struct ocfs2_dlm_lksb *lksb = astarg;
 677
 678	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
 679}
 680
 681static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
 682			 int mode,
 683			 struct ocfs2_dlm_lksb *lksb,
 684			 u32 flags,
 685			 void *name,
 686			 unsigned int namelen)
 687{
 688	int ret;
 689
 690	if (!lksb->lksb_fsdlm.sb_lvbptr)
 691		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
 692					     sizeof(struct dlm_lksb);
 693
 694	ret = dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
 695		       flags|DLM_LKF_NODLCKWT, name, namelen, 0,
 696		       fsdlm_lock_ast_wrapper, lksb,
 697		       fsdlm_blocking_ast_wrapper);
 698	return ret;
 699}
 700
 701static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
 702			   struct ocfs2_dlm_lksb *lksb,
 703			   u32 flags)
 704{
 705	int ret;
 706
 707	ret = dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
 708			 flags, &lksb->lksb_fsdlm, lksb);
 709	return ret;
 710}
 711
 712static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
 713{
 714	return lksb->lksb_fsdlm.sb_status;
 715}
 716
 717static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
 718{
 719	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
 720
 721	return !invalid;
 722}
 723
 724static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
 725{
 726	if (!lksb->lksb_fsdlm.sb_lvbptr)
 727		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
 728					     sizeof(struct dlm_lksb);
 729	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
 730}
 731
 732static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
 733{
 734}
 735
 736static int user_plock(struct ocfs2_cluster_connection *conn,
 737		      u64 ino,
 738		      struct file *file,
 739		      int cmd,
 740		      struct file_lock *fl)
 741{
 742	/*
 743	 * This more or less just demuxes the plock request into any
 744	 * one of three dlm calls.
 745	 *
 746	 * Internally, fs/dlm will pass these to a misc device, which
 747	 * a userspace daemon will read and write to.
 748	 *
 749	 * For now, cancel requests (which happen internally only),
 750	 * are turned into unlocks. Most of this function taken from
 751	 * gfs2_lock.
 752	 */
 753
 754	if (cmd == F_CANCELLK) {
 755		cmd = F_SETLK;
 756		fl->fl_type = F_UNLCK;
 757	}
 758
 759	if (IS_GETLK(cmd))
 760		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
 761	else if (fl->fl_type == F_UNLCK)
 762		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
 763	else
 764		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
 765}
 766
 767/*
 768 * Compare a requested locking protocol version against the current one.
 769 *
 770 * If the major numbers are different, they are incompatible.
 771 * If the current minor is greater than the request, they are incompatible.
 772 * If the current minor is less than or equal to the request, they are
 773 * compatible, and the requester should run at the current minor version.
 774 */
 775static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
 776			       struct ocfs2_protocol_version *request)
 777{
 778	if (existing->pv_major != request->pv_major)
 779		return 1;
 780
 781	if (existing->pv_minor > request->pv_minor)
 782		return 1;
 783
 784	if (existing->pv_minor < request->pv_minor)
 785		request->pv_minor = existing->pv_minor;
 786
 787	return 0;
 788}
 789
 790static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
 791{
 792	struct ocfs2_protocol_version *pv =
 793		(struct ocfs2_protocol_version *)lvb;
 794	/*
 795	 * ocfs2_protocol_version has two u8 variables, so we don't
 796	 * need any endian conversion.
 797	 */
 798	ver->pv_major = pv->pv_major;
 799	ver->pv_minor = pv->pv_minor;
 800}
 801
 802static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
 803{
 804	struct ocfs2_protocol_version *pv =
 805		(struct ocfs2_protocol_version *)lvb;
 806	/*
 807	 * ocfs2_protocol_version has two u8 variables, so we don't
 808	 * need any endian conversion.
 809	 */
 810	pv->pv_major = ver->pv_major;
 811	pv->pv_minor = ver->pv_minor;
 812}
 813
 814static void sync_wait_cb(void *arg)
 815{
 816	struct ocfs2_cluster_connection *conn = arg;
 817	struct ocfs2_live_connection *lc = conn->cc_private;
 818	complete(&lc->oc_sync_wait);
 819}
 820
 821static int sync_unlock(struct ocfs2_cluster_connection *conn,
 822		struct dlm_lksb *lksb, char *name)
 823{
 824	int error;
 825	struct ocfs2_live_connection *lc = conn->cc_private;
 826
 827	error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
 828	if (error) {
 829		printk(KERN_ERR "%s lkid %x error %d\n",
 830				name, lksb->sb_lkid, error);
 831		return error;
 832	}
 833
 834	wait_for_completion(&lc->oc_sync_wait);
 835
 836	if (lksb->sb_status != -DLM_EUNLOCK) {
 837		printk(KERN_ERR "%s lkid %x status %d\n",
 838				name, lksb->sb_lkid, lksb->sb_status);
 839		return -1;
 840	}
 841	return 0;
 842}
 843
 844static int sync_lock(struct ocfs2_cluster_connection *conn,
 845		int mode, uint32_t flags,
 846		struct dlm_lksb *lksb, char *name)
 847{
 848	int error, status;
 849	struct ocfs2_live_connection *lc = conn->cc_private;
 850
 851	error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
 852			name, strlen(name),
 853			0, sync_wait_cb, conn, NULL);
 854	if (error) {
 855		printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
 856				name, lksb->sb_lkid, flags, mode, error);
 857		return error;
 858	}
 859
 860	wait_for_completion(&lc->oc_sync_wait);
 861
 862	status = lksb->sb_status;
 863
 864	if (status && status != -EAGAIN) {
 865		printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
 866				name, lksb->sb_lkid, flags, mode, status);
 867	}
 868
 869	return status;
 870}
 871
 872
 873static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
 874		int flags)
 875{
 876	struct ocfs2_live_connection *lc = conn->cc_private;
 877	return sync_lock(conn, mode, flags,
 878			&lc->oc_version_lksb, VERSION_LOCK);
 879}
 880
 881static int version_unlock(struct ocfs2_cluster_connection *conn)
 882{
 883	struct ocfs2_live_connection *lc = conn->cc_private;
 884	return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
 885}
 886
 887/* get_protocol_version()
 888 *
 889 * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
 890 * The algorithm is:
 891 * 1. Attempt to take the lock in EX mode (non-blocking).
 892 * 2. If successful (which means it is the first mount), write the
 893 *    version number and downconvert to PR lock.
 894 * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
 895 *    taking the PR lock.
 896 */
 897
 898static int get_protocol_version(struct ocfs2_cluster_connection *conn)
 899{
 900	int ret;
 901	struct ocfs2_live_connection *lc = conn->cc_private;
 902	struct ocfs2_protocol_version pv;
 903
 904	running_proto.pv_major =
 905		ocfs2_user_plugin.sp_max_proto.pv_major;
 906	running_proto.pv_minor =
 907		ocfs2_user_plugin.sp_max_proto.pv_minor;
 908
 909	lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
 910	ret = version_lock(conn, DLM_LOCK_EX,
 911			DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
 912	if (!ret) {
 913		conn->cc_version.pv_major = running_proto.pv_major;
 914		conn->cc_version.pv_minor = running_proto.pv_minor;
 915		version_to_lvb(&running_proto, lc->oc_lvb);
 916		version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
 917	} else if (ret == -EAGAIN) {
 918		ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
 919		if (ret)
 920			goto out;
 921		lvb_to_version(lc->oc_lvb, &pv);
 922
 923		if ((pv.pv_major != running_proto.pv_major) ||
 924				(pv.pv_minor > running_proto.pv_minor)) {
 925			ret = -EINVAL;
 926			goto out;
 927		}
 928
 929		conn->cc_version.pv_major = pv.pv_major;
 930		conn->cc_version.pv_minor = pv.pv_minor;
 931	}
 932out:
 933	return ret;
 934}
 935
 936static void user_recover_prep(void *arg)
 937{
 938}
 939
 940static void user_recover_slot(void *arg, struct dlm_slot *slot)
 941{
 942	struct ocfs2_cluster_connection *conn = arg;
 943	printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
 944			slot->nodeid, slot->slot);
 945	conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
 946
 947}
 948
 949static void user_recover_done(void *arg, struct dlm_slot *slots,
 950		int num_slots, int our_slot,
 951		uint32_t generation)
 952{
 953	struct ocfs2_cluster_connection *conn = arg;
 954	struct ocfs2_live_connection *lc = conn->cc_private;
 955	int i;
 956
 957	for (i = 0; i < num_slots; i++)
 958		if (slots[i].slot == our_slot) {
 959			atomic_set(&lc->oc_this_node, slots[i].nodeid);
 960			break;
 961		}
 962
 963	lc->oc_our_slot = our_slot;
 964	wake_up(&lc->oc_wait);
 965}
 966
 967static const struct dlm_lockspace_ops ocfs2_ls_ops = {
 968	.recover_prep = user_recover_prep,
 969	.recover_slot = user_recover_slot,
 970	.recover_done = user_recover_done,
 971};
 972
 973static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 974{
 975	version_unlock(conn);
 976	dlm_release_lockspace(conn->cc_lockspace, 2);
 977	conn->cc_lockspace = NULL;
 978	ocfs2_live_connection_drop(conn->cc_private);
 979	conn->cc_private = NULL;
 980	return 0;
 981}
 982
 983static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
 984{
 985	dlm_lockspace_t *fsdlm;
 986	struct ocfs2_live_connection *lc;
 987	int rc, ops_rv;
 988
 989	BUG_ON(conn == NULL);
 990
 991	lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
 992	if (!lc)
 993		return -ENOMEM;
 994
 995	init_waitqueue_head(&lc->oc_wait);
 996	init_completion(&lc->oc_sync_wait);
 997	atomic_set(&lc->oc_this_node, 0);
 998	conn->cc_private = lc;
 999	lc->oc_type = NO_CONTROLD;
1000
1001	rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
1002			       DLM_LSFL_FS | DLM_LSFL_NEWEXCL, DLM_LVB_LEN,
1003			       &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
1004	if (rc) {
1005		if (rc == -EEXIST || rc == -EPROTO)
1006			printk(KERN_ERR "ocfs2: Unable to create the "
1007				"lockspace %s (%d), because a ocfs2-tools "
1008				"program is running on this file system "
1009				"with the same name lockspace\n",
1010				conn->cc_name, rc);
1011		goto out;
1012	}
1013
1014	if (ops_rv == -EOPNOTSUPP) {
1015		lc->oc_type = WITH_CONTROLD;
1016		printk(KERN_NOTICE "ocfs2: You seem to be using an older "
1017				"version of dlm_controld and/or ocfs2-tools."
1018				" Please consider upgrading.\n");
1019	} else if (ops_rv) {
1020		rc = ops_rv;
1021		goto out;
1022	}
1023	conn->cc_lockspace = fsdlm;
1024
1025	rc = ocfs2_live_connection_attach(conn, lc);
1026	if (rc)
1027		goto out;
1028
1029	if (lc->oc_type == NO_CONTROLD) {
1030		rc = get_protocol_version(conn);
1031		if (rc) {
1032			printk(KERN_ERR "ocfs2: Could not determine"
1033					" locking version\n");
1034			user_cluster_disconnect(conn);
1035			goto out;
1036		}
1037		wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
1038	}
1039
1040	/*
1041	 * running_proto must have been set before we allowed any mounts
1042	 * to proceed.
1043	 */
1044	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
1045		printk(KERN_ERR
1046		       "Unable to mount with fs locking protocol version "
1047		       "%u.%u because negotiated protocol is %u.%u\n",
1048		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
1049		       running_proto.pv_major, running_proto.pv_minor);
1050		rc = -EPROTO;
1051		ocfs2_live_connection_drop(lc);
1052		lc = NULL;
1053	}
1054
1055out:
1056	if (rc)
1057		kfree(lc);
1058	return rc;
1059}
1060
1061
1062static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
1063				  unsigned int *this_node)
1064{
1065	int rc;
1066	struct ocfs2_live_connection *lc = conn->cc_private;
1067
1068	if (lc->oc_type == WITH_CONTROLD)
1069		rc = ocfs2_control_get_this_node();
1070	else if (lc->oc_type == NO_CONTROLD)
1071		rc = atomic_read(&lc->oc_this_node);
1072	else
1073		rc = -EINVAL;
1074
1075	if (rc < 0)
1076		return rc;
1077
1078	*this_node = rc;
1079	return 0;
1080}
1081
1082static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
1083	.connect	= user_cluster_connect,
1084	.disconnect	= user_cluster_disconnect,
1085	.this_node	= user_cluster_this_node,
1086	.dlm_lock	= user_dlm_lock,
1087	.dlm_unlock	= user_dlm_unlock,
1088	.lock_status	= user_dlm_lock_status,
1089	.lvb_valid	= user_dlm_lvb_valid,
1090	.lock_lvb	= user_dlm_lvb,
1091	.plock		= user_plock,
1092	.dump_lksb	= user_dlm_dump_lksb,
1093};
1094
1095static struct ocfs2_stack_plugin ocfs2_user_plugin = {
1096	.sp_name	= "user",
1097	.sp_ops		= &ocfs2_user_plugin_ops,
1098	.sp_owner	= THIS_MODULE,
1099};
1100
1101
1102static int __init ocfs2_user_plugin_init(void)
1103{
1104	int rc;
1105
1106	rc = ocfs2_control_init();
1107	if (!rc) {
1108		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
1109		if (rc)
1110			ocfs2_control_exit();
1111	}
1112
1113	return rc;
1114}
1115
1116static void __exit ocfs2_user_plugin_exit(void)
1117{
1118	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
1119	ocfs2_control_exit();
1120}
1121
1122MODULE_AUTHOR("Oracle");
1123MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
1124MODULE_LICENSE("GPL");
1125module_init(ocfs2_user_plugin_init);
1126module_exit(ocfs2_user_plugin_exit);
v6.2
   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/miscdevice.h>
  13#include <linux/mutex.h>
  14#include <linux/slab.h>
  15#include <linux/reboot.h>
  16#include <linux/sched.h>
  17#include <linux/uaccess.h>
  18
  19#include "stackglue.h"
  20
  21#include <linux/dlm_plock.h>
  22
  23/*
  24 * The control protocol starts with a handshake.  Until the handshake
  25 * is complete, the control device will fail all write(2)s.
  26 *
  27 * The handshake is simple.  First, the client reads until EOF.  Each line
  28 * of output is a supported protocol tag.  All protocol tags are a single
  29 * character followed by a two hex digit version number.  Currently the
  30 * only things supported is T01, for "Text-base version 0x01".  Next, the
  31 * client writes the version they would like to use, including the newline.
  32 * Thus, the protocol tag is 'T01\n'.  If the version tag written is
  33 * unknown, -EINVAL is returned.  Once the negotiation is complete, the
  34 * client can start sending messages.
  35 *
  36 * The T01 protocol has three messages.  First is the "SETN" message.
  37 * It has the following syntax:
  38 *
  39 *  SETN<space><8-char-hex-nodenum><newline>
  40 *
  41 * This is 14 characters.
  42 *
  43 * The "SETN" message must be the first message following the protocol.
  44 * It tells ocfs2_control the local node number.
  45 *
  46 * Next comes the "SETV" message.  It has the following syntax:
  47 *
  48 *  SETV<space><2-char-hex-major><space><2-char-hex-minor><newline>
  49 *
  50 * This is 11 characters.
  51 *
  52 * The "SETV" message sets the filesystem locking protocol version as
  53 * negotiated by the client.  The client negotiates based on the maximum
  54 * version advertised in /sys/fs/ocfs2/max_locking_protocol.  The major
  55 * number from the "SETV" message must match
  56 * ocfs2_user_plugin.sp_max_proto.pv_major, and the minor number
  57 * must be less than or equal to ...sp_max_version.pv_minor.
  58 *
  59 * Once this information has been set, mounts will be allowed.  From this
  60 * point on, the "DOWN" message can be sent for node down notification.
  61 * It has the following syntax:
  62 *
  63 *  DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline>
  64 *
  65 * eg:
  66 *
  67 *  DOWN 632A924FDD844190BDA93C0DF6B94899 00000001\n
  68 *
  69 * This is 47 characters.
  70 */
  71
  72/*
  73 * Whether or not the client has done the handshake.
  74 * For now, we have just one protocol version.
  75 */
  76#define OCFS2_CONTROL_PROTO			"T01\n"
  77#define OCFS2_CONTROL_PROTO_LEN			4
  78
  79/* Handshake states */
  80#define OCFS2_CONTROL_HANDSHAKE_INVALID		(0)
  81#define OCFS2_CONTROL_HANDSHAKE_READ		(1)
  82#define OCFS2_CONTROL_HANDSHAKE_PROTOCOL	(2)
  83#define OCFS2_CONTROL_HANDSHAKE_VALID		(3)
  84
  85/* Messages */
  86#define OCFS2_CONTROL_MESSAGE_OP_LEN		4
  87#define OCFS2_CONTROL_MESSAGE_SETNODE_OP	"SETN"
  88#define OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN	14
  89#define OCFS2_CONTROL_MESSAGE_SETVERSION_OP	"SETV"
  90#define OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN	11
  91#define OCFS2_CONTROL_MESSAGE_DOWN_OP		"DOWN"
  92#define OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN	47
  93#define OCFS2_TEXT_UUID_LEN			32
  94#define OCFS2_CONTROL_MESSAGE_VERNUM_LEN	2
  95#define OCFS2_CONTROL_MESSAGE_NODENUM_LEN	8
  96#define VERSION_LOCK				"version_lock"
  97
  98enum ocfs2_connection_type {
  99	WITH_CONTROLD,
 100	NO_CONTROLD
 101};
 102
 103/*
 104 * ocfs2_live_connection is refcounted because the filesystem and
 105 * miscdevice sides can detach in different order.  Let's just be safe.
 106 */
 107struct ocfs2_live_connection {
 108	struct list_head		oc_list;
 109	struct ocfs2_cluster_connection	*oc_conn;
 110	enum ocfs2_connection_type	oc_type;
 111	atomic_t                        oc_this_node;
 112	int                             oc_our_slot;
 113	struct dlm_lksb                 oc_version_lksb;
 114	char                            oc_lvb[DLM_LVB_LEN];
 115	struct completion               oc_sync_wait;
 116	wait_queue_head_t		oc_wait;
 117};
 118
 119struct ocfs2_control_private {
 120	struct list_head op_list;
 121	int op_state;
 122	int op_this_node;
 123	struct ocfs2_protocol_version op_proto;
 124};
 125
 126/* SETN<space><8-char-hex-nodenum><newline> */
 127struct ocfs2_control_message_setn {
 128	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 129	char	space;
 130	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
 131	char	newline;
 132};
 133
 134/* SETV<space><2-char-hex-major><space><2-char-hex-minor><newline> */
 135struct ocfs2_control_message_setv {
 136	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 137	char	space1;
 138	char	major[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
 139	char	space2;
 140	char	minor[OCFS2_CONTROL_MESSAGE_VERNUM_LEN];
 141	char	newline;
 142};
 143
 144/* DOWN<space><32-char-cap-hex-uuid><space><8-char-hex-nodenum><newline> */
 145struct ocfs2_control_message_down {
 146	char	tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 147	char	space1;
 148	char	uuid[OCFS2_TEXT_UUID_LEN];
 149	char	space2;
 150	char	nodestr[OCFS2_CONTROL_MESSAGE_NODENUM_LEN];
 151	char	newline;
 152};
 153
 154union ocfs2_control_message {
 155	char					tag[OCFS2_CONTROL_MESSAGE_OP_LEN];
 156	struct ocfs2_control_message_setn	u_setn;
 157	struct ocfs2_control_message_setv	u_setv;
 158	struct ocfs2_control_message_down	u_down;
 159};
 160
 161static struct ocfs2_stack_plugin ocfs2_user_plugin;
 162
 163static atomic_t ocfs2_control_opened;
 164static int ocfs2_control_this_node = -1;
 165static struct ocfs2_protocol_version running_proto;
 166
 167static LIST_HEAD(ocfs2_live_connection_list);
 168static LIST_HEAD(ocfs2_control_private_list);
 169static DEFINE_MUTEX(ocfs2_control_lock);
 170
 171static inline void ocfs2_control_set_handshake_state(struct file *file,
 172						     int state)
 173{
 174	struct ocfs2_control_private *p = file->private_data;
 175	p->op_state = state;
 176}
 177
 178static inline int ocfs2_control_get_handshake_state(struct file *file)
 179{
 180	struct ocfs2_control_private *p = file->private_data;
 181	return p->op_state;
 182}
 183
 184static struct ocfs2_live_connection *ocfs2_connection_find(const char *name)
 185{
 186	size_t len = strlen(name);
 187	struct ocfs2_live_connection *c;
 188
 189	BUG_ON(!mutex_is_locked(&ocfs2_control_lock));
 190
 191	list_for_each_entry(c, &ocfs2_live_connection_list, oc_list) {
 192		if ((c->oc_conn->cc_namelen == len) &&
 193		    !strncmp(c->oc_conn->cc_name, name, len))
 194			return c;
 195	}
 196
 197	return NULL;
 198}
 199
 200/*
 201 * ocfs2_live_connection structures are created underneath the ocfs2
 202 * mount path.  Since the VFS prevents multiple calls to
 203 * fill_super(), we can't get dupes here.
 204 */
 205static int ocfs2_live_connection_attach(struct ocfs2_cluster_connection *conn,
 206				     struct ocfs2_live_connection *c)
 207{
 208	int rc = 0;
 209
 210	mutex_lock(&ocfs2_control_lock);
 211	c->oc_conn = conn;
 212
 213	if ((c->oc_type == NO_CONTROLD) || atomic_read(&ocfs2_control_opened))
 214		list_add(&c->oc_list, &ocfs2_live_connection_list);
 215	else {
 216		printk(KERN_ERR
 217		       "ocfs2: Userspace control daemon is not present\n");
 218		rc = -ESRCH;
 219	}
 220
 221	mutex_unlock(&ocfs2_control_lock);
 222	return rc;
 223}
 224
 225/*
 226 * This function disconnects the cluster connection from ocfs2_control.
 227 * Afterwards, userspace can't affect the cluster connection.
 228 */
 229static void ocfs2_live_connection_drop(struct ocfs2_live_connection *c)
 230{
 231	mutex_lock(&ocfs2_control_lock);
 232	list_del_init(&c->oc_list);
 233	c->oc_conn = NULL;
 234	mutex_unlock(&ocfs2_control_lock);
 235
 236	kfree(c);
 237}
 238
 239static int ocfs2_control_cfu(void *target, size_t target_len,
 240			     const char __user *buf, size_t count)
 241{
 242	/* The T01 expects write(2) calls to have exactly one command */
 243	if ((count != target_len) ||
 244	    (count > sizeof(union ocfs2_control_message)))
 245		return -EINVAL;
 246
 247	if (copy_from_user(target, buf, target_len))
 248		return -EFAULT;
 249
 250	return 0;
 251}
 252
 253static ssize_t ocfs2_control_validate_protocol(struct file *file,
 254					       const char __user *buf,
 255					       size_t count)
 256{
 257	ssize_t ret;
 258	char kbuf[OCFS2_CONTROL_PROTO_LEN];
 259
 260	ret = ocfs2_control_cfu(kbuf, OCFS2_CONTROL_PROTO_LEN,
 261				buf, count);
 262	if (ret)
 263		return ret;
 264
 265	if (strncmp(kbuf, OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN))
 266		return -EINVAL;
 267
 268	ocfs2_control_set_handshake_state(file,
 269					  OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
 270
 271	return count;
 272}
 273
 274static void ocfs2_control_send_down(const char *uuid,
 275				    int nodenum)
 276{
 277	struct ocfs2_live_connection *c;
 278
 279	mutex_lock(&ocfs2_control_lock);
 280
 281	c = ocfs2_connection_find(uuid);
 282	if (c) {
 283		BUG_ON(c->oc_conn == NULL);
 284		c->oc_conn->cc_recovery_handler(nodenum,
 285						c->oc_conn->cc_recovery_data);
 286	}
 287
 288	mutex_unlock(&ocfs2_control_lock);
 289}
 290
 291/*
 292 * Called whenever configuration elements are sent to /dev/ocfs2_control.
 293 * If all configuration elements are present, try to set the global
 294 * values.  If there is a problem, return an error.  Skip any missing
 295 * elements, and only bump ocfs2_control_opened when we have all elements
 296 * and are successful.
 297 */
 298static int ocfs2_control_install_private(struct file *file)
 299{
 300	int rc = 0;
 301	int set_p = 1;
 302	struct ocfs2_control_private *p = file->private_data;
 303
 304	BUG_ON(p->op_state != OCFS2_CONTROL_HANDSHAKE_PROTOCOL);
 305
 306	mutex_lock(&ocfs2_control_lock);
 307
 308	if (p->op_this_node < 0) {
 309		set_p = 0;
 310	} else if ((ocfs2_control_this_node >= 0) &&
 311		   (ocfs2_control_this_node != p->op_this_node)) {
 312		rc = -EINVAL;
 313		goto out_unlock;
 314	}
 315
 316	if (!p->op_proto.pv_major) {
 317		set_p = 0;
 318	} else if (!list_empty(&ocfs2_live_connection_list) &&
 319		   ((running_proto.pv_major != p->op_proto.pv_major) ||
 320		    (running_proto.pv_minor != p->op_proto.pv_minor))) {
 321		rc = -EINVAL;
 322		goto out_unlock;
 323	}
 324
 325	if (set_p) {
 326		ocfs2_control_this_node = p->op_this_node;
 327		running_proto.pv_major = p->op_proto.pv_major;
 328		running_proto.pv_minor = p->op_proto.pv_minor;
 329	}
 330
 331out_unlock:
 332	mutex_unlock(&ocfs2_control_lock);
 333
 334	if (!rc && set_p) {
 335		/* We set the global values successfully */
 336		atomic_inc(&ocfs2_control_opened);
 337		ocfs2_control_set_handshake_state(file,
 338					OCFS2_CONTROL_HANDSHAKE_VALID);
 339	}
 340
 341	return rc;
 342}
 343
 344static int ocfs2_control_get_this_node(void)
 345{
 346	int rc;
 347
 348	mutex_lock(&ocfs2_control_lock);
 349	if (ocfs2_control_this_node < 0)
 350		rc = -EINVAL;
 351	else
 352		rc = ocfs2_control_this_node;
 353	mutex_unlock(&ocfs2_control_lock);
 354
 355	return rc;
 356}
 357
 358static int ocfs2_control_do_setnode_msg(struct file *file,
 359					struct ocfs2_control_message_setn *msg)
 360{
 361	long nodenum;
 362	char *ptr = NULL;
 363	struct ocfs2_control_private *p = file->private_data;
 364
 365	if (ocfs2_control_get_handshake_state(file) !=
 366	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
 367		return -EINVAL;
 368
 369	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
 370		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 371		return -EINVAL;
 372
 373	if ((msg->space != ' ') || (msg->newline != '\n'))
 374		return -EINVAL;
 375	msg->space = msg->newline = '\0';
 376
 377	nodenum = simple_strtol(msg->nodestr, &ptr, 16);
 378	if (!ptr || *ptr)
 379		return -EINVAL;
 380
 381	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
 382	    (nodenum > INT_MAX) || (nodenum < 0))
 383		return -ERANGE;
 384	p->op_this_node = nodenum;
 385
 386	return ocfs2_control_install_private(file);
 387}
 388
 389static int ocfs2_control_do_setversion_msg(struct file *file,
 390					   struct ocfs2_control_message_setv *msg)
 391{
 392	long major, minor;
 393	char *ptr = NULL;
 394	struct ocfs2_control_private *p = file->private_data;
 395	struct ocfs2_protocol_version *max =
 396		&ocfs2_user_plugin.sp_max_proto;
 397
 398	if (ocfs2_control_get_handshake_state(file) !=
 399	    OCFS2_CONTROL_HANDSHAKE_PROTOCOL)
 400		return -EINVAL;
 401
 402	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
 403		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 404		return -EINVAL;
 405
 406	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
 407	    (msg->newline != '\n'))
 408		return -EINVAL;
 409	msg->space1 = msg->space2 = msg->newline = '\0';
 410
 411	major = simple_strtol(msg->major, &ptr, 16);
 412	if (!ptr || *ptr)
 413		return -EINVAL;
 414	minor = simple_strtol(msg->minor, &ptr, 16);
 415	if (!ptr || *ptr)
 416		return -EINVAL;
 417
 418	/*
 419	 * The major must be between 1 and 255, inclusive.  The minor
 420	 * must be between 0 and 255, inclusive.  The version passed in
 421	 * must be within the maximum version supported by the filesystem.
 422	 */
 423	if ((major == LONG_MIN) || (major == LONG_MAX) ||
 424	    (major > (u8)-1) || (major < 1))
 425		return -ERANGE;
 426	if ((minor == LONG_MIN) || (minor == LONG_MAX) ||
 427	    (minor > (u8)-1) || (minor < 0))
 428		return -ERANGE;
 429	if ((major != max->pv_major) ||
 430	    (minor > max->pv_minor))
 431		return -EINVAL;
 432
 433	p->op_proto.pv_major = major;
 434	p->op_proto.pv_minor = minor;
 435
 436	return ocfs2_control_install_private(file);
 437}
 438
 439static int ocfs2_control_do_down_msg(struct file *file,
 440				     struct ocfs2_control_message_down *msg)
 441{
 442	long nodenum;
 443	char *p = NULL;
 444
 445	if (ocfs2_control_get_handshake_state(file) !=
 446	    OCFS2_CONTROL_HANDSHAKE_VALID)
 447		return -EINVAL;
 448
 449	if (strncmp(msg->tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
 450		    OCFS2_CONTROL_MESSAGE_OP_LEN))
 451		return -EINVAL;
 452
 453	if ((msg->space1 != ' ') || (msg->space2 != ' ') ||
 454	    (msg->newline != '\n'))
 455		return -EINVAL;
 456	msg->space1 = msg->space2 = msg->newline = '\0';
 457
 458	nodenum = simple_strtol(msg->nodestr, &p, 16);
 459	if (!p || *p)
 460		return -EINVAL;
 461
 462	if ((nodenum == LONG_MIN) || (nodenum == LONG_MAX) ||
 463	    (nodenum > INT_MAX) || (nodenum < 0))
 464		return -ERANGE;
 465
 466	ocfs2_control_send_down(msg->uuid, nodenum);
 467
 468	return 0;
 469}
 470
 471static ssize_t ocfs2_control_message(struct file *file,
 472				     const char __user *buf,
 473				     size_t count)
 474{
 475	ssize_t ret;
 476	union ocfs2_control_message msg;
 477
 478	/* Try to catch padding issues */
 479	WARN_ON(offsetof(struct ocfs2_control_message_down, uuid) !=
 480		(sizeof(msg.u_down.tag) + sizeof(msg.u_down.space1)));
 481
 482	memset(&msg, 0, sizeof(union ocfs2_control_message));
 483	ret = ocfs2_control_cfu(&msg, count, buf, count);
 484	if (ret)
 485		goto out;
 486
 487	if ((count == OCFS2_CONTROL_MESSAGE_SETNODE_TOTAL_LEN) &&
 488	    !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETNODE_OP,
 489		     OCFS2_CONTROL_MESSAGE_OP_LEN))
 490		ret = ocfs2_control_do_setnode_msg(file, &msg.u_setn);
 491	else if ((count == OCFS2_CONTROL_MESSAGE_SETVERSION_TOTAL_LEN) &&
 492		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_SETVERSION_OP,
 493			  OCFS2_CONTROL_MESSAGE_OP_LEN))
 494		ret = ocfs2_control_do_setversion_msg(file, &msg.u_setv);
 495	else if ((count == OCFS2_CONTROL_MESSAGE_DOWN_TOTAL_LEN) &&
 496		 !strncmp(msg.tag, OCFS2_CONTROL_MESSAGE_DOWN_OP,
 497			  OCFS2_CONTROL_MESSAGE_OP_LEN))
 498		ret = ocfs2_control_do_down_msg(file, &msg.u_down);
 499	else
 500		ret = -EINVAL;
 501
 502out:
 503	return ret ? ret : count;
 504}
 505
 506static ssize_t ocfs2_control_write(struct file *file,
 507				   const char __user *buf,
 508				   size_t count,
 509				   loff_t *ppos)
 510{
 511	ssize_t ret;
 512
 513	switch (ocfs2_control_get_handshake_state(file)) {
 514		case OCFS2_CONTROL_HANDSHAKE_INVALID:
 515			ret = -EINVAL;
 516			break;
 517
 518		case OCFS2_CONTROL_HANDSHAKE_READ:
 519			ret = ocfs2_control_validate_protocol(file, buf,
 520							      count);
 521			break;
 522
 523		case OCFS2_CONTROL_HANDSHAKE_PROTOCOL:
 524		case OCFS2_CONTROL_HANDSHAKE_VALID:
 525			ret = ocfs2_control_message(file, buf, count);
 526			break;
 527
 528		default:
 529			BUG();
 530			ret = -EIO;
 531			break;
 532	}
 533
 534	return ret;
 535}
 536
 537/*
 538 * This is a naive version.  If we ever have a new protocol, we'll expand
 539 * it.  Probably using seq_file.
 540 */
 541static ssize_t ocfs2_control_read(struct file *file,
 542				  char __user *buf,
 543				  size_t count,
 544				  loff_t *ppos)
 545{
 546	ssize_t ret;
 547
 548	ret = simple_read_from_buffer(buf, count, ppos,
 549			OCFS2_CONTROL_PROTO, OCFS2_CONTROL_PROTO_LEN);
 550
 551	/* Have we read the whole protocol list? */
 552	if (ret > 0 && *ppos >= OCFS2_CONTROL_PROTO_LEN)
 553		ocfs2_control_set_handshake_state(file,
 554						  OCFS2_CONTROL_HANDSHAKE_READ);
 555
 556	return ret;
 557}
 558
 559static int ocfs2_control_release(struct inode *inode, struct file *file)
 560{
 561	struct ocfs2_control_private *p = file->private_data;
 562
 563	mutex_lock(&ocfs2_control_lock);
 564
 565	if (ocfs2_control_get_handshake_state(file) !=
 566	    OCFS2_CONTROL_HANDSHAKE_VALID)
 567		goto out;
 568
 569	if (atomic_dec_and_test(&ocfs2_control_opened)) {
 570		if (!list_empty(&ocfs2_live_connection_list)) {
 571			/* XXX: Do bad things! */
 572			printk(KERN_ERR
 573			       "ocfs2: Unexpected release of ocfs2_control!\n"
 574			       "       Loss of cluster connection requires "
 575			       "an emergency restart!\n");
 576			emergency_restart();
 577		}
 578		/*
 579		 * Last valid close clears the node number and resets
 580		 * the locking protocol version
 581		 */
 582		ocfs2_control_this_node = -1;
 583		running_proto.pv_major = 0;
 584		running_proto.pv_minor = 0;
 585	}
 586
 587out:
 588	list_del_init(&p->op_list);
 589	file->private_data = NULL;
 590
 591	mutex_unlock(&ocfs2_control_lock);
 592
 593	kfree(p);
 594
 595	return 0;
 596}
 597
 598static int ocfs2_control_open(struct inode *inode, struct file *file)
 599{
 600	struct ocfs2_control_private *p;
 601
 602	p = kzalloc(sizeof(struct ocfs2_control_private), GFP_KERNEL);
 603	if (!p)
 604		return -ENOMEM;
 605	p->op_this_node = -1;
 606
 607	mutex_lock(&ocfs2_control_lock);
 608	file->private_data = p;
 609	list_add(&p->op_list, &ocfs2_control_private_list);
 610	mutex_unlock(&ocfs2_control_lock);
 611
 612	return 0;
 613}
 614
 615static const struct file_operations ocfs2_control_fops = {
 616	.open    = ocfs2_control_open,
 617	.release = ocfs2_control_release,
 618	.read    = ocfs2_control_read,
 619	.write   = ocfs2_control_write,
 620	.owner   = THIS_MODULE,
 621	.llseek  = default_llseek,
 622};
 623
 624static struct miscdevice ocfs2_control_device = {
 625	.minor		= MISC_DYNAMIC_MINOR,
 626	.name		= "ocfs2_control",
 627	.fops		= &ocfs2_control_fops,
 628};
 629
 630static int ocfs2_control_init(void)
 631{
 632	int rc;
 633
 634	atomic_set(&ocfs2_control_opened, 0);
 635
 636	rc = misc_register(&ocfs2_control_device);
 637	if (rc)
 638		printk(KERN_ERR
 639		       "ocfs2: Unable to register ocfs2_control device "
 640		       "(errno %d)\n",
 641		       -rc);
 642
 643	return rc;
 644}
 645
 646static void ocfs2_control_exit(void)
 647{
 648	misc_deregister(&ocfs2_control_device);
 649}
 650
 651static void fsdlm_lock_ast_wrapper(void *astarg)
 652{
 653	struct ocfs2_dlm_lksb *lksb = astarg;
 654	int status = lksb->lksb_fsdlm.sb_status;
 655
 656	/*
 657	 * For now we're punting on the issue of other non-standard errors
 658	 * where we can't tell if the unlock_ast or lock_ast should be called.
 659	 * The main "other error" that's possible is EINVAL which means the
 660	 * function was called with invalid args, which shouldn't be possible
 661	 * since the caller here is under our control.  Other non-standard
 662	 * errors probably fall into the same category, or otherwise are fatal
 663	 * which means we can't carry on anyway.
 664	 */
 665
 666	if (status == -DLM_EUNLOCK || status == -DLM_ECANCEL)
 667		lksb->lksb_conn->cc_proto->lp_unlock_ast(lksb, 0);
 668	else
 669		lksb->lksb_conn->cc_proto->lp_lock_ast(lksb);
 670}
 671
 672static void fsdlm_blocking_ast_wrapper(void *astarg, int level)
 673{
 674	struct ocfs2_dlm_lksb *lksb = astarg;
 675
 676	lksb->lksb_conn->cc_proto->lp_blocking_ast(lksb, level);
 677}
 678
 679static int user_dlm_lock(struct ocfs2_cluster_connection *conn,
 680			 int mode,
 681			 struct ocfs2_dlm_lksb *lksb,
 682			 u32 flags,
 683			 void *name,
 684			 unsigned int namelen)
 685{
 
 
 686	if (!lksb->lksb_fsdlm.sb_lvbptr)
 687		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
 688					     sizeof(struct dlm_lksb);
 689
 690	return dlm_lock(conn->cc_lockspace, mode, &lksb->lksb_fsdlm,
 691			flags|DLM_LKF_NODLCKWT, name, namelen, 0,
 692			fsdlm_lock_ast_wrapper, lksb,
 693			fsdlm_blocking_ast_wrapper);
 
 694}
 695
 696static int user_dlm_unlock(struct ocfs2_cluster_connection *conn,
 697			   struct ocfs2_dlm_lksb *lksb,
 698			   u32 flags)
 699{
 700	return dlm_unlock(conn->cc_lockspace, lksb->lksb_fsdlm.sb_lkid,
 701			  flags, &lksb->lksb_fsdlm, lksb);
 
 
 
 702}
 703
 704static int user_dlm_lock_status(struct ocfs2_dlm_lksb *lksb)
 705{
 706	return lksb->lksb_fsdlm.sb_status;
 707}
 708
 709static int user_dlm_lvb_valid(struct ocfs2_dlm_lksb *lksb)
 710{
 711	int invalid = lksb->lksb_fsdlm.sb_flags & DLM_SBF_VALNOTVALID;
 712
 713	return !invalid;
 714}
 715
 716static void *user_dlm_lvb(struct ocfs2_dlm_lksb *lksb)
 717{
 718	if (!lksb->lksb_fsdlm.sb_lvbptr)
 719		lksb->lksb_fsdlm.sb_lvbptr = (char *)lksb +
 720					     sizeof(struct dlm_lksb);
 721	return (void *)(lksb->lksb_fsdlm.sb_lvbptr);
 722}
 723
 724static void user_dlm_dump_lksb(struct ocfs2_dlm_lksb *lksb)
 725{
 726}
 727
 728static int user_plock(struct ocfs2_cluster_connection *conn,
 729		      u64 ino,
 730		      struct file *file,
 731		      int cmd,
 732		      struct file_lock *fl)
 733{
 734	/*
 735	 * This more or less just demuxes the plock request into any
 736	 * one of three dlm calls.
 737	 *
 738	 * Internally, fs/dlm will pass these to a misc device, which
 739	 * a userspace daemon will read and write to.
 740	 *
 741	 * For now, cancel requests (which happen internally only),
 742	 * are turned into unlocks. Most of this function taken from
 743	 * gfs2_lock.
 744	 */
 745
 746	if (cmd == F_CANCELLK) {
 747		cmd = F_SETLK;
 748		fl->fl_type = F_UNLCK;
 749	}
 750
 751	if (IS_GETLK(cmd))
 752		return dlm_posix_get(conn->cc_lockspace, ino, file, fl);
 753	else if (fl->fl_type == F_UNLCK)
 754		return dlm_posix_unlock(conn->cc_lockspace, ino, file, fl);
 755	else
 756		return dlm_posix_lock(conn->cc_lockspace, ino, file, cmd, fl);
 757}
 758
 759/*
 760 * Compare a requested locking protocol version against the current one.
 761 *
 762 * If the major numbers are different, they are incompatible.
 763 * If the current minor is greater than the request, they are incompatible.
 764 * If the current minor is less than or equal to the request, they are
 765 * compatible, and the requester should run at the current minor version.
 766 */
 767static int fs_protocol_compare(struct ocfs2_protocol_version *existing,
 768			       struct ocfs2_protocol_version *request)
 769{
 770	if (existing->pv_major != request->pv_major)
 771		return 1;
 772
 773	if (existing->pv_minor > request->pv_minor)
 774		return 1;
 775
 776	if (existing->pv_minor < request->pv_minor)
 777		request->pv_minor = existing->pv_minor;
 778
 779	return 0;
 780}
 781
 782static void lvb_to_version(char *lvb, struct ocfs2_protocol_version *ver)
 783{
 784	struct ocfs2_protocol_version *pv =
 785		(struct ocfs2_protocol_version *)lvb;
 786	/*
 787	 * ocfs2_protocol_version has two u8 variables, so we don't
 788	 * need any endian conversion.
 789	 */
 790	ver->pv_major = pv->pv_major;
 791	ver->pv_minor = pv->pv_minor;
 792}
 793
 794static void version_to_lvb(struct ocfs2_protocol_version *ver, char *lvb)
 795{
 796	struct ocfs2_protocol_version *pv =
 797		(struct ocfs2_protocol_version *)lvb;
 798	/*
 799	 * ocfs2_protocol_version has two u8 variables, so we don't
 800	 * need any endian conversion.
 801	 */
 802	pv->pv_major = ver->pv_major;
 803	pv->pv_minor = ver->pv_minor;
 804}
 805
 806static void sync_wait_cb(void *arg)
 807{
 808	struct ocfs2_cluster_connection *conn = arg;
 809	struct ocfs2_live_connection *lc = conn->cc_private;
 810	complete(&lc->oc_sync_wait);
 811}
 812
 813static int sync_unlock(struct ocfs2_cluster_connection *conn,
 814		struct dlm_lksb *lksb, char *name)
 815{
 816	int error;
 817	struct ocfs2_live_connection *lc = conn->cc_private;
 818
 819	error = dlm_unlock(conn->cc_lockspace, lksb->sb_lkid, 0, lksb, conn);
 820	if (error) {
 821		printk(KERN_ERR "%s lkid %x error %d\n",
 822				name, lksb->sb_lkid, error);
 823		return error;
 824	}
 825
 826	wait_for_completion(&lc->oc_sync_wait);
 827
 828	if (lksb->sb_status != -DLM_EUNLOCK) {
 829		printk(KERN_ERR "%s lkid %x status %d\n",
 830				name, lksb->sb_lkid, lksb->sb_status);
 831		return -1;
 832	}
 833	return 0;
 834}
 835
 836static int sync_lock(struct ocfs2_cluster_connection *conn,
 837		int mode, uint32_t flags,
 838		struct dlm_lksb *lksb, char *name)
 839{
 840	int error, status;
 841	struct ocfs2_live_connection *lc = conn->cc_private;
 842
 843	error = dlm_lock(conn->cc_lockspace, mode, lksb, flags,
 844			name, strlen(name),
 845			0, sync_wait_cb, conn, NULL);
 846	if (error) {
 847		printk(KERN_ERR "%s lkid %x flags %x mode %d error %d\n",
 848				name, lksb->sb_lkid, flags, mode, error);
 849		return error;
 850	}
 851
 852	wait_for_completion(&lc->oc_sync_wait);
 853
 854	status = lksb->sb_status;
 855
 856	if (status && status != -EAGAIN) {
 857		printk(KERN_ERR "%s lkid %x flags %x mode %d status %d\n",
 858				name, lksb->sb_lkid, flags, mode, status);
 859	}
 860
 861	return status;
 862}
 863
 864
 865static int version_lock(struct ocfs2_cluster_connection *conn, int mode,
 866		int flags)
 867{
 868	struct ocfs2_live_connection *lc = conn->cc_private;
 869	return sync_lock(conn, mode, flags,
 870			&lc->oc_version_lksb, VERSION_LOCK);
 871}
 872
 873static int version_unlock(struct ocfs2_cluster_connection *conn)
 874{
 875	struct ocfs2_live_connection *lc = conn->cc_private;
 876	return sync_unlock(conn, &lc->oc_version_lksb, VERSION_LOCK);
 877}
 878
 879/* get_protocol_version()
 880 *
 881 * To exchange ocfs2 versioning, we use the LVB of the version dlm lock.
 882 * The algorithm is:
 883 * 1. Attempt to take the lock in EX mode (non-blocking).
 884 * 2. If successful (which means it is the first mount), write the
 885 *    version number and downconvert to PR lock.
 886 * 3. If unsuccessful (returns -EAGAIN), read the version from the LVB after
 887 *    taking the PR lock.
 888 */
 889
 890static int get_protocol_version(struct ocfs2_cluster_connection *conn)
 891{
 892	int ret;
 893	struct ocfs2_live_connection *lc = conn->cc_private;
 894	struct ocfs2_protocol_version pv;
 895
 896	running_proto.pv_major =
 897		ocfs2_user_plugin.sp_max_proto.pv_major;
 898	running_proto.pv_minor =
 899		ocfs2_user_plugin.sp_max_proto.pv_minor;
 900
 901	lc->oc_version_lksb.sb_lvbptr = lc->oc_lvb;
 902	ret = version_lock(conn, DLM_LOCK_EX,
 903			DLM_LKF_VALBLK|DLM_LKF_NOQUEUE);
 904	if (!ret) {
 905		conn->cc_version.pv_major = running_proto.pv_major;
 906		conn->cc_version.pv_minor = running_proto.pv_minor;
 907		version_to_lvb(&running_proto, lc->oc_lvb);
 908		version_lock(conn, DLM_LOCK_PR, DLM_LKF_CONVERT|DLM_LKF_VALBLK);
 909	} else if (ret == -EAGAIN) {
 910		ret = version_lock(conn, DLM_LOCK_PR, DLM_LKF_VALBLK);
 911		if (ret)
 912			goto out;
 913		lvb_to_version(lc->oc_lvb, &pv);
 914
 915		if ((pv.pv_major != running_proto.pv_major) ||
 916				(pv.pv_minor > running_proto.pv_minor)) {
 917			ret = -EINVAL;
 918			goto out;
 919		}
 920
 921		conn->cc_version.pv_major = pv.pv_major;
 922		conn->cc_version.pv_minor = pv.pv_minor;
 923	}
 924out:
 925	return ret;
 926}
 927
 928static void user_recover_prep(void *arg)
 929{
 930}
 931
 932static void user_recover_slot(void *arg, struct dlm_slot *slot)
 933{
 934	struct ocfs2_cluster_connection *conn = arg;
 935	printk(KERN_INFO "ocfs2: Node %d/%d down. Initiating recovery.\n",
 936			slot->nodeid, slot->slot);
 937	conn->cc_recovery_handler(slot->nodeid, conn->cc_recovery_data);
 938
 939}
 940
 941static void user_recover_done(void *arg, struct dlm_slot *slots,
 942		int num_slots, int our_slot,
 943		uint32_t generation)
 944{
 945	struct ocfs2_cluster_connection *conn = arg;
 946	struct ocfs2_live_connection *lc = conn->cc_private;
 947	int i;
 948
 949	for (i = 0; i < num_slots; i++)
 950		if (slots[i].slot == our_slot) {
 951			atomic_set(&lc->oc_this_node, slots[i].nodeid);
 952			break;
 953		}
 954
 955	lc->oc_our_slot = our_slot;
 956	wake_up(&lc->oc_wait);
 957}
 958
 959static const struct dlm_lockspace_ops ocfs2_ls_ops = {
 960	.recover_prep = user_recover_prep,
 961	.recover_slot = user_recover_slot,
 962	.recover_done = user_recover_done,
 963};
 964
 965static int user_cluster_disconnect(struct ocfs2_cluster_connection *conn)
 966{
 967	version_unlock(conn);
 968	dlm_release_lockspace(conn->cc_lockspace, 2);
 969	conn->cc_lockspace = NULL;
 970	ocfs2_live_connection_drop(conn->cc_private);
 971	conn->cc_private = NULL;
 972	return 0;
 973}
 974
 975static int user_cluster_connect(struct ocfs2_cluster_connection *conn)
 976{
 977	dlm_lockspace_t *fsdlm;
 978	struct ocfs2_live_connection *lc;
 979	int rc, ops_rv;
 980
 981	BUG_ON(conn == NULL);
 982
 983	lc = kzalloc(sizeof(struct ocfs2_live_connection), GFP_KERNEL);
 984	if (!lc)
 985		return -ENOMEM;
 986
 987	init_waitqueue_head(&lc->oc_wait);
 988	init_completion(&lc->oc_sync_wait);
 989	atomic_set(&lc->oc_this_node, 0);
 990	conn->cc_private = lc;
 991	lc->oc_type = NO_CONTROLD;
 992
 993	rc = dlm_new_lockspace(conn->cc_name, conn->cc_cluster_name,
 994			       DLM_LSFL_NEWEXCL, DLM_LVB_LEN,
 995			       &ocfs2_ls_ops, conn, &ops_rv, &fsdlm);
 996	if (rc) {
 997		if (rc == -EEXIST || rc == -EPROTO)
 998			printk(KERN_ERR "ocfs2: Unable to create the "
 999				"lockspace %s (%d), because a ocfs2-tools "
1000				"program is running on this file system "
1001				"with the same name lockspace\n",
1002				conn->cc_name, rc);
1003		goto out;
1004	}
1005
1006	if (ops_rv == -EOPNOTSUPP) {
1007		lc->oc_type = WITH_CONTROLD;
1008		printk(KERN_NOTICE "ocfs2: You seem to be using an older "
1009				"version of dlm_controld and/or ocfs2-tools."
1010				" Please consider upgrading.\n");
1011	} else if (ops_rv) {
1012		rc = ops_rv;
1013		goto out;
1014	}
1015	conn->cc_lockspace = fsdlm;
1016
1017	rc = ocfs2_live_connection_attach(conn, lc);
1018	if (rc)
1019		goto out;
1020
1021	if (lc->oc_type == NO_CONTROLD) {
1022		rc = get_protocol_version(conn);
1023		if (rc) {
1024			printk(KERN_ERR "ocfs2: Could not determine"
1025					" locking version\n");
1026			user_cluster_disconnect(conn);
1027			goto out;
1028		}
1029		wait_event(lc->oc_wait, (atomic_read(&lc->oc_this_node) > 0));
1030	}
1031
1032	/*
1033	 * running_proto must have been set before we allowed any mounts
1034	 * to proceed.
1035	 */
1036	if (fs_protocol_compare(&running_proto, &conn->cc_version)) {
1037		printk(KERN_ERR
1038		       "Unable to mount with fs locking protocol version "
1039		       "%u.%u because negotiated protocol is %u.%u\n",
1040		       conn->cc_version.pv_major, conn->cc_version.pv_minor,
1041		       running_proto.pv_major, running_proto.pv_minor);
1042		rc = -EPROTO;
1043		ocfs2_live_connection_drop(lc);
1044		lc = NULL;
1045	}
1046
1047out:
1048	if (rc)
1049		kfree(lc);
1050	return rc;
1051}
1052
1053
1054static int user_cluster_this_node(struct ocfs2_cluster_connection *conn,
1055				  unsigned int *this_node)
1056{
1057	int rc;
1058	struct ocfs2_live_connection *lc = conn->cc_private;
1059
1060	if (lc->oc_type == WITH_CONTROLD)
1061		rc = ocfs2_control_get_this_node();
1062	else if (lc->oc_type == NO_CONTROLD)
1063		rc = atomic_read(&lc->oc_this_node);
1064	else
1065		rc = -EINVAL;
1066
1067	if (rc < 0)
1068		return rc;
1069
1070	*this_node = rc;
1071	return 0;
1072}
1073
1074static struct ocfs2_stack_operations ocfs2_user_plugin_ops = {
1075	.connect	= user_cluster_connect,
1076	.disconnect	= user_cluster_disconnect,
1077	.this_node	= user_cluster_this_node,
1078	.dlm_lock	= user_dlm_lock,
1079	.dlm_unlock	= user_dlm_unlock,
1080	.lock_status	= user_dlm_lock_status,
1081	.lvb_valid	= user_dlm_lvb_valid,
1082	.lock_lvb	= user_dlm_lvb,
1083	.plock		= user_plock,
1084	.dump_lksb	= user_dlm_dump_lksb,
1085};
1086
1087static struct ocfs2_stack_plugin ocfs2_user_plugin = {
1088	.sp_name	= "user",
1089	.sp_ops		= &ocfs2_user_plugin_ops,
1090	.sp_owner	= THIS_MODULE,
1091};
1092
1093
1094static int __init ocfs2_user_plugin_init(void)
1095{
1096	int rc;
1097
1098	rc = ocfs2_control_init();
1099	if (!rc) {
1100		rc = ocfs2_stack_glue_register(&ocfs2_user_plugin);
1101		if (rc)
1102			ocfs2_control_exit();
1103	}
1104
1105	return rc;
1106}
1107
1108static void __exit ocfs2_user_plugin_exit(void)
1109{
1110	ocfs2_stack_glue_unregister(&ocfs2_user_plugin);
1111	ocfs2_control_exit();
1112}
1113
1114MODULE_AUTHOR("Oracle");
1115MODULE_DESCRIPTION("ocfs2 driver for userspace cluster stacks");
1116MODULE_LICENSE("GPL");
1117module_init(ocfs2_user_plugin_init);
1118module_exit(ocfs2_user_plugin_exit);