SAFE public projects git trees. - safe/jmp/linux-2.6/blob - fs/ocfs2/stack_user.c

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