fs/dlm/midcomms.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /******************************************************************************
   3 *******************************************************************************
   4 **
   5 **  Copyright (C) Sistina Software, Inc.  1997-2003  All rights reserved.
   6 **  Copyright (C) 2004-2021 Red Hat, Inc.  All rights reserved.
   7 **
   8 **
   9 *******************************************************************************
  10 ******************************************************************************/
  11
  12 /*
  13  * midcomms.c
  14  *
  15  * This is the appallingly named "mid-level" comms layer. It takes care about
  16  * deliver an on application layer "reliable" communication above the used
  17  * lowcomms transport layer.
  18  *
  19  * How it works:
  20  *
  21  * Each nodes keeps track of all send DLM messages in send_queue with a sequence
  22  * number. The receive will send an DLM_ACK message back for every DLM message
  23  * received at the other side. If a reconnect happens in lowcomms we will send
  24  * all unacknowledged dlm messages again. The receiving side might drop any already
  25  * received message by comparing sequence numbers.
  26  *
  27  * How version detection works:
  28  *
  29  * Due the fact that dlm has pre-configured node addresses on every side
  30  * it is in it's nature that every side connects at starts to transmit
  31  * dlm messages which ends in a race. However DLM_RCOM_NAMES, DLM_RCOM_STATUS
  32  * and their replies are the first messages which are exchanges. Due backwards
  33  * compatibility these messages are not covered by the midcomms re-transmission
  34  * layer. These messages have their own re-transmission handling in the dlm
  35  * application layer. The version field of every node will be set on these RCOM
  36  * messages as soon as they arrived and the node isn't yet part of the nodes
  37  * hash. There exists also logic to detect version mismatched if something weird
  38  * going on or the first messages isn't an expected one.
  39  *
  40  * Termination:
  41  *
  42  * The midcomms layer does a 4 way handshake for termination on DLM protocol
  43  * like TCP supports it with half-closed socket support. SCTP doesn't support
  44  * half-closed socket, so we do it on DLM layer. Also socket shutdown() can be
  45  * interrupted by .e.g. tcp reset itself. Additional there exists the othercon
  46  * paradigm in lowcomms which cannot be easily without breaking backwards
  47  * compatibility. A node cannot send anything to another node when a DLM_FIN
  48  * message was send. There exists additional logic to print a warning if
  49  * DLM wants to do it. There exists a state handling like RFC 793 but reduced
  50  * to termination only. The event "member removal event" describes the cluster
  51  * manager removed the node from internal lists, at this point DLM does not
  52  * send any message to the other node. There exists two cases:
  53  *
  54  * 1. The cluster member was removed and we received a FIN
  55  * OR
  56  * 2. We received a FIN but the member was not removed yet
  57  *
  58  * One of these cases will do the CLOSE_WAIT to LAST_ACK change.
  59  *
  60  *
  61  *                              +---------+
  62  *                              | CLOSED  |
  63  *                              +---------+
  64  *                                   | add member/receive RCOM version
  65  *                                   |            detection msg
  66  *                                   V
  67  *                              +---------+
  68  *                              |  ESTAB  |
  69  *                              +---------+
  70  *                       CLOSE    |     |    rcv FIN
  71  *                      -------   |     |    -------
  72  * +---------+          snd FIN  /       \   snd ACK          +---------+
  73  * |  FIN    |<-----------------           ------------------>|  CLOSE  |
  74  * | WAIT-1  |------------------                              |   WAIT  |
  75  * +---------+          rcv FIN  \                            +---------+
  76  * | rcv ACK of FIN   -------   |                            CLOSE  | member
  77  * | --------------   snd ACK   |                           ------- | removal
  78  * V        x                   V                           snd FIN V event
  79  * +---------+                  +---------+                   +---------+
  80  * |FINWAIT-2|                  | CLOSING |                   | LAST-ACK|
  81  * +---------+                  +---------+                   +---------+
  82  * |                rcv ACK of FIN |                 rcv ACK of FIN |
  83  * |  rcv FIN       -------------- |                 -------------- |
  84  * |  -------              x       V                        x       V
  85  *  \ snd ACK                 +---------+                   +---------+
  86  *   ------------------------>| CLOSED  |                   | CLOSED  |
  87  *                            +---------+                   +---------+
  88  *
  89  * NOTE: any state can interrupted by midcomms_close() and state will be
  90  * switched to CLOSED in case of fencing. There exists also some timeout
  91  * handling when we receive the version detection RCOM messages which is
  92  * made by observation.
  93  *
  94  * Future improvements:
  95  *
  96  * There exists some known issues/improvements of the dlm handling. Some
  97  * of them should be done in a next major dlm version bump which makes
  98  * it incompatible with previous versions.
  99  *
 100  * Unaligned memory access:
 101  *
 102  * There exists cases when the dlm message buffer length is not aligned
 103  * to 8 byte. However seems nobody detected any problem with it. This
 104  * can be fixed in the next major version bump of dlm.
 105  *
 106  * Version detection:
 107  *
 108  * The version detection and how it's done is related to backwards
 109  * compatibility. There exists better ways to make a better handling.
 110  * However this should be changed in the next major version bump of dlm.
 111  *
 112  * Tail Size checking:
 113  *
 114  * There exists a message tail payload in e.g. DLM_MSG however we don't
 115  * check it against the message length yet regarding to the receive buffer
 116  * length. That need to be validated.
 117  *
 118  * Fencing bad nodes:
 119  *
 120  * At timeout places or weird sequence number behaviours we should send
 121  * a fencing request to the cluster manager.
 122  */
 123
 124 /* Debug switch to enable a 5 seconds sleep waiting of a termination.
 125  * This can be useful to test fencing while termination is running.
 126  * This requires a setup with only gfs2 as dlm user, so that the
 127  * last umount will terminate the connection.
 128  *
 129  * However it became useful to test, while the 5 seconds block in umount
 130  * just press the reset button. In a lot of dropping the termination
 131  * process can could take several seconds.
 132  */
 133 #define DLM_DEBUG_FENCE_TERMINATION     0
 134
 135 #include <net/tcp.h>
 136
 137 #include "dlm_internal.h"
 138 #include "lowcomms.h"
 139 #include "config.h"
 140 #include "memory.h"
 141 #include "lock.h"
 142 #include "util.h"
 143 #include "midcomms.h"
 144
 145 /* init value for sequence numbers for testing purpose only e.g. overflows */
 146 #define DLM_SEQ_INIT            0
 147 /* 3 minutes wait to sync ending of dlm */
 148 #define DLM_SHUTDOWN_TIMEOUT    msecs_to_jiffies(3 * 60 * 1000)
 149 #define DLM_VERSION_NOT_SET     0
 150
 151 struct midcomms_node {
 152         int nodeid;
 153         uint32_t version;
 154         uint32_t seq_send;
 155         uint32_t seq_next;
 156         /* These queues are unbound because we cannot drop any message in dlm.
 157          * We could send a fence signal for a specific node to the cluster
 158          * manager if queues hits some maximum value, however this handling
 159          * not supported yet.
 160          */
 161         struct list_head send_queue;
 162         spinlock_t send_queue_lock;
 163         atomic_t send_queue_cnt;
 164 #define DLM_NODE_FLAG_CLOSE     1
 165 #define DLM_NODE_FLAG_STOP_TX   2
 166 #define DLM_NODE_FLAG_STOP_RX   3
 167 #define DLM_NODE_ULP_DELIVERED  4
 168         unsigned long flags;
 169         wait_queue_head_t shutdown_wait;
 170
 171         /* dlm tcp termination state */
 172 #define DLM_CLOSED      1
 173 #define DLM_ESTABLISHED 2
 174 #define DLM_FIN_WAIT1   3
 175 #define DLM_FIN_WAIT2   4
 176 #define DLM_CLOSE_WAIT  5
 177 #define DLM_LAST_ACK    6
 178 #define DLM_CLOSING     7
 179         int state;
 180         spinlock_t state_lock;
 181
 182         /* counts how many lockspaces are using this node
 183          * this refcount is necessary to determine if the
 184          * node wants to disconnect.
 185          */
 186         int users;
 187
 188         /* not protected by srcu, node_hash lifetime */
 189         void *debugfs;
 190
 191         struct hlist_node hlist;
 192         struct rcu_head rcu;
 193 };
 194
 195 struct dlm_mhandle {
 196         const struct dlm_header *inner_hd;
 197         struct midcomms_node *node;
 198         struct dlm_opts *opts;
 199         struct dlm_msg *msg;
 200         bool committed;
 201         uint32_t seq;
 202
 203         void (*ack_rcv)(struct midcomms_node *node);
 204
 205         /* get_mhandle/commit srcu idx exchange */
 206         int idx;
 207
 208         struct list_head list;
 209         struct rcu_head rcu;
 210 };
 211
 212 static struct hlist_head node_hash[CONN_HASH_SIZE];
 213 static DEFINE_SPINLOCK(nodes_lock);
 214 DEFINE_STATIC_SRCU(nodes_srcu);
 215
 216 /* This mutex prevents that midcomms_close() is running while
 217  * stop() or remove(). As I experienced invalid memory access
 218  * behaviours when DLM_DEBUG_FENCE_TERMINATION is enabled and
 219  * resetting machines. I will end in some double deletion in nodes
 220  * datastructure.
 221  */
 222 static DEFINE_MUTEX(close_lock);
 223
 224 struct kmem_cache *dlm_midcomms_cache_create(void)
 225 {
 226         return kmem_cache_create("dlm_mhandle", sizeof(struct dlm_mhandle),
 227                                  0, 0, NULL);
 228 }
 229
 230 static inline const char *dlm_state_str(int state)
 231 {
 232         switch (state) {
 233         case DLM_CLOSED:
 234                 return "CLOSED";
 235         case DLM_ESTABLISHED:
 236                 return "ESTABLISHED";
 237         case DLM_FIN_WAIT1:
 238                 return "FIN_WAIT1";
 239         case DLM_FIN_WAIT2:
 240                 return "FIN_WAIT2";
 241         case DLM_CLOSE_WAIT:
 242                 return "CLOSE_WAIT";
 243         case DLM_LAST_ACK:
 244                 return "LAST_ACK";
 245         case DLM_CLOSING:
 246                 return "CLOSING";
 247         default:
 248                 return "UNKNOWN";
 249         }
 250 }
 251
 252 const char *dlm_midcomms_state(struct midcomms_node *node)
 253 {
 254         return dlm_state_str(node->state);
 255 }
 256
 257 unsigned long dlm_midcomms_flags(struct midcomms_node *node)
 258 {
 259         return node->flags;
 260 }
 261
 262 int dlm_midcomms_send_queue_cnt(struct midcomms_node *node)
 263 {
 264         return atomic_read(&node->send_queue_cnt);
 265 }
 266
 267 uint32_t dlm_midcomms_version(struct midcomms_node *node)
 268 {
 269         return node->version;
 270 }
 271
 272 static struct midcomms_node *__find_node(int nodeid, int r)
 273 {
 274         struct midcomms_node *node;
 275
 276         hlist_for_each_entry_rcu(node, &node_hash[r], hlist) {
 277                 if (node->nodeid == nodeid)
 278                         return node;
 279         }
 280
 281         return NULL;
 282 }
 283
 284 static void dlm_mhandle_release(struct rcu_head *rcu)
 285 {
 286         struct dlm_mhandle *mh = container_of(rcu, struct dlm_mhandle, rcu);
 287
 288         dlm_lowcomms_put_msg(mh->msg);
 289         dlm_free_mhandle(mh);
 290 }
 291
 292 static void dlm_mhandle_delete(struct midcomms_node *node,
 293                                struct dlm_mhandle *mh)
 294 {
 295         list_del_rcu(&mh->list);
 296         atomic_dec(&node->send_queue_cnt);
 297         call_rcu(&mh->rcu, dlm_mhandle_release);
 298 }
 299
 300 static void dlm_send_queue_flush(struct midcomms_node *node)
 301 {
 302         struct dlm_mhandle *mh;
 303
 304         pr_debug("flush midcomms send queue of node %d\n", node->nodeid);
 305
 306         rcu_read_lock();
 307         spin_lock(&node->send_queue_lock);
 308         list_for_each_entry_rcu(mh, &node->send_queue, list) {
 309                 dlm_mhandle_delete(node, mh);
 310         }
 311         spin_unlock(&node->send_queue_lock);
 312         rcu_read_unlock();
 313 }
 314
 315 static void midcomms_node_reset(struct midcomms_node *node)
 316 {
 317         pr_debug("reset node %d\n", node->nodeid);
 318
 319         node->seq_next = DLM_SEQ_INIT;
 320         node->seq_send = DLM_SEQ_INIT;
 321         node->version = DLM_VERSION_NOT_SET;
 322         node->flags = 0;
 323
 324         dlm_send_queue_flush(node);
 325         node->state = DLM_CLOSED;
 326         wake_up(&node->shutdown_wait);
 327 }
 328
 329 static struct midcomms_node *nodeid2node(int nodeid, gfp_t alloc)
 330 {
 331         struct midcomms_node *node, *tmp;
 332         int r = nodeid_hash(nodeid);
 333
 334         node = __find_node(nodeid, r);
 335         if (node || !alloc)
 336                 return node;
 337
 338         node = kmalloc(sizeof(*node), alloc);
 339         if (!node)
 340                 return NULL;
 341
 342         node->nodeid = nodeid;
 343         spin_lock_init(&node->state_lock);
 344         spin_lock_init(&node->send_queue_lock);
 345         atomic_set(&node->send_queue_cnt, 0);
 346         INIT_LIST_HEAD(&node->send_queue);
 347         init_waitqueue_head(&node->shutdown_wait);
 348         node->users = 0;
 349         midcomms_node_reset(node);
 350
 351         spin_lock(&nodes_lock);
 352         /* check again if there was somebody else
 353          * earlier here to add the node
 354          */
 355         tmp = __find_node(nodeid, r);
 356         if (tmp) {
 357                 spin_unlock(&nodes_lock);
 358                 kfree(node);
 359                 return tmp;
 360         }
 361
 362         hlist_add_head_rcu(&node->hlist, &node_hash[r]);
 363         spin_unlock(&nodes_lock);
 364
 365         node->debugfs = dlm_create_debug_comms_file(nodeid, node);
 366         return node;
 367 }
 368
 369 static int dlm_send_ack(int nodeid, uint32_t seq)
 370 {
 371         int mb_len = sizeof(struct dlm_header);
 372         struct dlm_header *m_header;
 373         struct dlm_msg *msg;
 374         char *ppc;
 375
 376         msg = dlm_lowcomms_new_msg(nodeid, mb_len, GFP_NOFS, &ppc,
 377                                    NULL, NULL);
 378         if (!msg)
 379                 return -ENOMEM;
 380
 381         m_header = (struct dlm_header *)ppc;
 382
 383         m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
 384         m_header->h_nodeid = dlm_our_nodeid();
 385         m_header->h_length = mb_len;
 386         m_header->h_cmd = DLM_ACK;
 387         m_header->u.h_seq = seq;
 388
 389         header_out(m_header);
 390         dlm_lowcomms_commit_msg(msg);
 391         dlm_lowcomms_put_msg(msg);
 392
 393         return 0;
 394 }
 395
 396 static int dlm_send_fin(struct midcomms_node *node,
 397                         void (*ack_rcv)(struct midcomms_node *node))
 398 {
 399         int mb_len = sizeof(struct dlm_header);
 400         struct dlm_header *m_header;
 401         struct dlm_mhandle *mh;
 402         char *ppc;
 403
 404         mh = dlm_midcomms_get_mhandle(node->nodeid, mb_len, GFP_NOFS, &ppc);
 405         if (!mh)
 406                 return -ENOMEM;
 407
 408         mh->ack_rcv = ack_rcv;
 409
 410         m_header = (struct dlm_header *)ppc;
 411
 412         m_header->h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
 413         m_header->h_nodeid = dlm_our_nodeid();
 414         m_header->h_length = mb_len;
 415         m_header->h_cmd = DLM_FIN;
 416
 417         header_out(m_header);
 418
 419         pr_debug("sending fin msg to node %d\n", node->nodeid);
 420         dlm_midcomms_commit_mhandle(mh);
 421         set_bit(DLM_NODE_FLAG_STOP_TX, &node->flags);
 422
 423         return 0;
 424 }
 425
 426 static void dlm_receive_ack(struct midcomms_node *node, uint32_t seq)
 427 {
 428         struct dlm_mhandle *mh;
 429
 430         rcu_read_lock();
 431         list_for_each_entry_rcu(mh, &node->send_queue, list) {
 432                 if (before(mh->seq, seq)) {
 433                         if (mh->ack_rcv)
 434                                 mh->ack_rcv(node);
 435                 } else {
 436                         /* send queue should be ordered */
 437                         break;
 438                 }
 439         }
 440
 441         spin_lock(&node->send_queue_lock);
 442         list_for_each_entry_rcu(mh, &node->send_queue, list) {
 443                 if (before(mh->seq, seq)) {
 444                         dlm_mhandle_delete(node, mh);
 445                 } else {
 446                         /* send queue should be ordered */
 447                         break;
 448                 }
 449         }
 450         spin_unlock(&node->send_queue_lock);
 451         rcu_read_unlock();
 452 }
 453
 454 static void dlm_pas_fin_ack_rcv(struct midcomms_node *node)
 455 {
 456         spin_lock(&node->state_lock);
 457         pr_debug("receive passive fin ack from node %d with state %s\n",
 458                  node->nodeid, dlm_state_str(node->state));
 459
 460         switch (node->state) {
 461         case DLM_LAST_ACK:
 462                 /* DLM_CLOSED */
 463                 midcomms_node_reset(node);
 464                 break;
 465         case DLM_CLOSED:
 466                 /* not valid but somehow we got what we want */
 467                 wake_up(&node->shutdown_wait);
 468                 break;
 469         default:
 470                 spin_unlock(&node->state_lock);
 471                 log_print("%s: unexpected state: %d\n",
 472                           __func__, node->state);
 473                 WARN_ON(1);
 474                 return;
 475         }
 476         spin_unlock(&node->state_lock);
 477 }
 478
 479 static void dlm_midcomms_receive_buffer(union dlm_packet *p,
 480                                         struct midcomms_node *node,
 481                                         uint32_t seq)
 482 {
 483         if (seq == node->seq_next) {
 484                 node->seq_next++;
 485
 486                 switch (p->header.h_cmd) {
 487                 case DLM_FIN:
 488                         /* send ack before fin */
 489                         dlm_send_ack(node->nodeid, node->seq_next);
 490
 491                         spin_lock(&node->state_lock);
 492                         pr_debug("receive fin msg from node %d with state %s\n",
 493                                  node->nodeid, dlm_state_str(node->state));
 494
 495                         switch (node->state) {
 496                         case DLM_ESTABLISHED:
 497                                 node->state = DLM_CLOSE_WAIT;
 498                                 pr_debug("switch node %d to state %s\n",
 499                                          node->nodeid, dlm_state_str(node->state));
 500                                 /* passive shutdown DLM_LAST_ACK case 1
 501                                  * additional we check if the node is used by
 502                                  * cluster manager events at all.
 503                                  */
 504                                 if (node->users == 0) {
 505                                         node->state = DLM_LAST_ACK;
 506                                         pr_debug("switch node %d to state %s case 1\n",
 507                                                  node->nodeid, dlm_state_str(node->state));
 508                                         spin_unlock(&node->state_lock);
 509                                         goto send_fin;
 510                                 }
 511                                 break;
 512                         case DLM_FIN_WAIT1:
 513                                 node->state = DLM_CLOSING;
 514                                 pr_debug("switch node %d to state %s\n",
 515                                          node->nodeid, dlm_state_str(node->state));
 516                                 break;
 517                         case DLM_FIN_WAIT2:
 518                                 midcomms_node_reset(node);
 519                                 pr_debug("switch node %d to state %s\n",
 520                                          node->nodeid, dlm_state_str(node->state));
 521                                 wake_up(&node->shutdown_wait);
 522                                 break;
 523                         case DLM_LAST_ACK:
 524                                 /* probably remove_member caught it, do nothing */
 525                                 break;
 526                         default:
 527                                 spin_unlock(&node->state_lock);
 528                                 log_print("%s: unexpected state: %d\n",
 529                                           __func__, node->state);
 530                                 WARN_ON(1);
 531                                 return;
 532                         }
 533                         spin_unlock(&node->state_lock);
 534
 535                         set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
 536                         break;
 537                 default:
 538                         WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
 539                         dlm_receive_buffer(p, node->nodeid);
 540                         set_bit(DLM_NODE_ULP_DELIVERED, &node->flags);
 541                         break;
 542                 }
 543         } else {
 544                 /* retry to ack message which we already have by sending back
 545                  * current node->seq_next number as ack.
 546                  */
 547                 if (seq < node->seq_next)
 548                         dlm_send_ack(node->nodeid, node->seq_next);
 549
 550                 log_print_ratelimited("ignore dlm msg because seq mismatch, seq: %u, expected: %u, nodeid: %d",
 551                                       seq, node->seq_next, node->nodeid);
 552         }
 553
 554         return;
 555
 556 send_fin:
 557         set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
 558         dlm_send_fin(node, dlm_pas_fin_ack_rcv);
 559 }
 560
 561 static struct midcomms_node *
 562 dlm_midcomms_recv_node_lookup(int nodeid, const union dlm_packet *p,
 563                               uint16_t msglen, int (*cb)(struct midcomms_node *node))
 564 {
 565         struct midcomms_node *node = NULL;
 566         gfp_t allocation = 0;
 567         int ret;
 568
 569         switch (p->header.h_cmd) {
 570         case DLM_RCOM:
 571                 if (msglen < sizeof(struct dlm_rcom)) {
 572                         log_print("rcom msg too small: %u, will skip this message from node %d",
 573                                   msglen, nodeid);
 574                         return NULL;
 575                 }
 576
 577                 switch (le32_to_cpu(p->rcom.rc_type)) {
 578                 case DLM_RCOM_NAMES:
 579                         fallthrough;
 580                 case DLM_RCOM_NAMES_REPLY:
 581                         fallthrough;
 582                 case DLM_RCOM_STATUS:
 583                         fallthrough;
 584                 case DLM_RCOM_STATUS_REPLY:
 585                         node = nodeid2node(nodeid, 0);
 586                         if (node) {
 587                                 spin_lock(&node->state_lock);
 588                                 if (node->state != DLM_ESTABLISHED)
 589                                         pr_debug("receive begin RCOM msg from node %d with state %s\n",
 590                                                  node->nodeid, dlm_state_str(node->state));
 591
 592                                 switch (node->state) {
 593                                 case DLM_CLOSED:
 594                                         node->state = DLM_ESTABLISHED;
 595                                         pr_debug("switch node %d to state %s\n",
 596                                                  node->nodeid, dlm_state_str(node->state));
 597                                         break;
 598                                 case DLM_ESTABLISHED:
 599                                         break;
 600                                 default:
 601                                         /* some invalid state passive shutdown
 602                                          * was failed, we try to reset and
 603                                          * hope it will go on.
 604                                          */
 605                                         log_print("reset node %d because shutdown stuck",
 606                                                   node->nodeid);
 607
 608                                         midcomms_node_reset(node);
 609                                         node->state = DLM_ESTABLISHED;
 610                                         break;
 611                                 }
 612                                 spin_unlock(&node->state_lock);
 613                         }
 614
 615                         allocation = GFP_NOFS;
 616                         break;
 617                 default:
 618                         break;
 619                 }
 620
 621                 break;
 622         default:
 623                 break;
 624         }
 625
 626         node = nodeid2node(nodeid, allocation);
 627         if (!node) {
 628                 switch (p->header.h_cmd) {
 629                 case DLM_OPTS:
 630                         if (msglen < sizeof(struct dlm_opts)) {
 631                                 log_print("opts msg too small: %u, will skip this message from node %d",
 632                                           msglen, nodeid);
 633                                 return NULL;
 634                         }
 635
 636                         log_print_ratelimited("received dlm opts message nextcmd %d from node %d in an invalid sequence",
 637                                               p->opts.o_nextcmd, nodeid);
 638                         break;
 639                 default:
 640                         log_print_ratelimited("received dlm message cmd %d from node %d in an invalid sequence",
 641                                               p->header.h_cmd, nodeid);
 642                         break;
 643                 }
 644
 645                 return NULL;
 646         }
 647
 648         ret = cb(node);
 649         if (ret < 0)
 650                 return NULL;
 651
 652         return node;
 653 }
 654
 655 static int dlm_midcomms_version_check_3_2(struct midcomms_node *node)
 656 {
 657         switch (node->version) {
 658         case DLM_VERSION_NOT_SET:
 659                 node->version = DLM_VERSION_3_2;
 660                 log_print("version 0x%08x for node %d detected", DLM_VERSION_3_2,
 661                           node->nodeid);
 662                 break;
 663         case DLM_VERSION_3_2:
 664                 break;
 665         default:
 666                 log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
 667                                       DLM_VERSION_3_2, node->nodeid, node->version);
 668                 return -1;
 669         }
 670
 671         return 0;
 672 }
 673
 674 static int dlm_opts_check_msglen(union dlm_packet *p, uint16_t msglen, int nodeid)
 675 {
 676         int len = msglen;
 677
 678         /* we only trust outer header msglen because
 679          * it's checked against receive buffer length.
 680          */
 681         if (len < sizeof(struct dlm_opts))
 682                 return -1;
 683         len -= sizeof(struct dlm_opts);
 684
 685         if (len < le16_to_cpu(p->opts.o_optlen))
 686                 return -1;
 687         len -= le16_to_cpu(p->opts.o_optlen);
 688
 689         switch (p->opts.o_nextcmd) {
 690         case DLM_FIN:
 691                 if (len < sizeof(struct dlm_header)) {
 692                         log_print("fin too small: %d, will skip this message from node %d",
 693                                   len, nodeid);
 694                         return -1;
 695                 }
 696
 697                 break;
 698         case DLM_MSG:
 699                 if (len < sizeof(struct dlm_message)) {
 700                         log_print("msg too small: %d, will skip this message from node %d",
 701                                   msglen, nodeid);
 702                         return -1;
 703                 }
 704
 705                 break;
 706         case DLM_RCOM:
 707                 if (len < sizeof(struct dlm_rcom)) {
 708                         log_print("rcom msg too small: %d, will skip this message from node %d",
 709                                   len, nodeid);
 710                         return -1;
 711                 }
 712
 713                 break;
 714         default:
 715                 log_print("unsupported o_nextcmd received: %u, will skip this message from node %d",
 716                           p->opts.o_nextcmd, nodeid);
 717                 return -1;
 718         }
 719
 720         return 0;
 721 }
 722
 723 static void dlm_midcomms_receive_buffer_3_2(union dlm_packet *p, int nodeid)
 724 {
 725         uint16_t msglen = le16_to_cpu(p->header.h_length);
 726         struct midcomms_node *node;
 727         uint32_t seq;
 728         int ret, idx;
 729
 730         idx = srcu_read_lock(&nodes_srcu);
 731         node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
 732                                              dlm_midcomms_version_check_3_2);
 733         if (!node)
 734                 goto out;
 735
 736         switch (p->header.h_cmd) {
 737         case DLM_RCOM:
 738                 /* these rcom message we use to determine version.
 739                  * they have their own retransmission handling and
 740                  * are the first messages of dlm.
 741                  *
 742                  * length already checked.
 743                  */
 744                 switch (le32_to_cpu(p->rcom.rc_type)) {
 745                 case DLM_RCOM_NAMES:
 746                         fallthrough;
 747                 case DLM_RCOM_NAMES_REPLY:
 748                         fallthrough;
 749                 case DLM_RCOM_STATUS:
 750                         fallthrough;
 751                 case DLM_RCOM_STATUS_REPLY:
 752                         break;
 753                 default:
 754                         log_print("unsupported rcom type received: %u, will skip this message from node %d",
 755                                   le32_to_cpu(p->rcom.rc_type), nodeid);
 756                         goto out;
 757                 }
 758
 759                 WARN_ON(test_bit(DLM_NODE_FLAG_STOP_RX, &node->flags));
 760                 dlm_receive_buffer(p, nodeid);
 761                 break;
 762         case DLM_OPTS:
 763                 seq = le32_to_cpu(p->header.u.h_seq);
 764
 765                 ret = dlm_opts_check_msglen(p, msglen, nodeid);
 766                 if (ret < 0) {
 767                         log_print("opts msg too small: %u, will skip this message from node %d",
 768                                   msglen, nodeid);
 769                         goto out;
 770                 }
 771
 772                 p = (union dlm_packet *)((unsigned char *)p->opts.o_opts +
 773                                          le16_to_cpu(p->opts.o_optlen));
 774
 775                 /* recheck inner msglen just if it's not garbage */
 776                 msglen = le16_to_cpu(p->header.h_length);
 777                 switch (p->header.h_cmd) {
 778                 case DLM_RCOM:
 779                         if (msglen < sizeof(struct dlm_rcom)) {
 780                                 log_print("inner rcom msg too small: %u, will skip this message from node %d",
 781                                           msglen, nodeid);
 782                                 goto out;
 783                         }
 784
 785                         break;
 786                 case DLM_MSG:
 787                         if (msglen < sizeof(struct dlm_message)) {
 788                                 log_print("inner msg too small: %u, will skip this message from node %d",
 789                                           msglen, nodeid);
 790                                 goto out;
 791                         }
 792
 793                         break;
 794                 case DLM_FIN:
 795                         if (msglen < sizeof(struct dlm_header)) {
 796                                 log_print("inner fin too small: %u, will skip this message from node %d",
 797                                           msglen, nodeid);
 798                                 goto out;
 799                         }
 800
 801                         break;
 802                 default:
 803                         log_print("unsupported inner h_cmd received: %u, will skip this message from node %d",
 804                                   msglen, nodeid);
 805                         goto out;
 806                 }
 807
 808                 dlm_midcomms_receive_buffer(p, node, seq);
 809                 break;
 810         case DLM_ACK:
 811                 seq = le32_to_cpu(p->header.u.h_seq);
 812                 dlm_receive_ack(node, seq);
 813                 break;
 814         default:
 815                 log_print("unsupported h_cmd received: %u, will skip this message from node %d",
 816                           p->header.h_cmd, nodeid);
 817                 break;
 818         }
 819
 820 out:
 821         srcu_read_unlock(&nodes_srcu, idx);
 822 }
 823
 824 static int dlm_midcomms_version_check_3_1(struct midcomms_node *node)
 825 {
 826         switch (node->version) {
 827         case DLM_VERSION_NOT_SET:
 828                 node->version = DLM_VERSION_3_1;
 829                 log_print("version 0x%08x for node %d detected", DLM_VERSION_3_1,
 830                           node->nodeid);
 831                 break;
 832         case DLM_VERSION_3_1:
 833                 break;
 834         default:
 835                 log_print_ratelimited("version mismatch detected, assumed 0x%08x but node %d has 0x%08x",
 836                                       DLM_VERSION_3_1, node->nodeid, node->version);
 837                 return -1;
 838         }
 839
 840         return 0;
 841 }
 842
 843 static void dlm_midcomms_receive_buffer_3_1(union dlm_packet *p, int nodeid)
 844 {
 845         uint16_t msglen = le16_to_cpu(p->header.h_length);
 846         struct midcomms_node *node;
 847         int idx;
 848
 849         idx = srcu_read_lock(&nodes_srcu);
 850         node = dlm_midcomms_recv_node_lookup(nodeid, p, msglen,
 851                                              dlm_midcomms_version_check_3_1);
 852         if (!node) {
 853                 srcu_read_unlock(&nodes_srcu, idx);
 854                 return;
 855         }
 856         srcu_read_unlock(&nodes_srcu, idx);
 857
 858         switch (p->header.h_cmd) {
 859         case DLM_RCOM:
 860                 /* length already checked */
 861                 break;
 862         case DLM_MSG:
 863                 if (msglen < sizeof(struct dlm_message)) {
 864                         log_print("msg too small: %u, will skip this message from node %d",
 865                                   msglen, nodeid);
 866                         return;
 867                 }
 868
 869                 break;
 870         default:
 871                 log_print("unsupported h_cmd received: %u, will skip this message from node %d",
 872                           p->header.h_cmd, nodeid);
 873                 return;
 874         }
 875
 876         dlm_receive_buffer(p, nodeid);
 877 }
 878
 879 /*
 880  * Called from the low-level comms layer to process a buffer of
 881  * commands.
 882  */
 883
 884 int dlm_process_incoming_buffer(int nodeid, unsigned char *buf, int len)
 885 {
 886         const unsigned char *ptr = buf;
 887         const struct dlm_header *hd;
 888         uint16_t msglen;
 889         int ret = 0;
 890
 891         while (len >= sizeof(struct dlm_header)) {
 892                 hd = (struct dlm_header *)ptr;
 893
 894                 /* no message should be more than DLM_MAX_SOCKET_BUFSIZE or
 895                  * less than dlm_header size.
 896                  *
 897                  * Some messages does not have a 8 byte length boundary yet
 898                  * which can occur in a unaligned memory access of some dlm
 899                  * messages. However this problem need to be fixed at the
 900                  * sending side, for now it seems nobody run into architecture
 901                  * related issues yet but it slows down some processing.
 902                  * Fixing this issue should be scheduled in future by doing
 903                  * the next major version bump.
 904                  */
 905                 msglen = le16_to_cpu(hd->h_length);
 906                 if (msglen > DLM_MAX_SOCKET_BUFSIZE ||
 907                     msglen < sizeof(struct dlm_header)) {
 908                         log_print("received invalid length header: %u from node %d, will abort message parsing",
 909                                   msglen, nodeid);
 910                         return -EBADMSG;
 911                 }
 912
 913                 /* caller will take care that leftover
 914                  * will be parsed next call with more data
 915                  */
 916                 if (msglen > len)
 917                         break;
 918
 919                 switch (hd->h_version) {
 920                 case cpu_to_le32(DLM_VERSION_3_1):
 921                         dlm_midcomms_receive_buffer_3_1((union dlm_packet *)ptr, nodeid);
 922                         break;
 923                 case cpu_to_le32(DLM_VERSION_3_2):
 924                         dlm_midcomms_receive_buffer_3_2((union dlm_packet *)ptr, nodeid);
 925                         break;
 926                 default:
 927                         log_print("received invalid version header: %u from node %d, will skip this message",
 928                                   le32_to_cpu(hd->h_version), nodeid);
 929                         break;
 930                 }
 931
 932                 ret += msglen;
 933                 len -= msglen;
 934                 ptr += msglen;
 935         }
 936
 937         return ret;
 938 }
 939
 940 void dlm_midcomms_receive_done(int nodeid)
 941 {
 942         struct midcomms_node *node;
 943         int idx;
 944
 945         idx = srcu_read_lock(&nodes_srcu);
 946         node = nodeid2node(nodeid, 0);
 947         if (!node) {
 948                 srcu_read_unlock(&nodes_srcu, idx);
 949                 return;
 950         }
 951
 952         /* old protocol, we do nothing */
 953         switch (node->version) {
 954         case DLM_VERSION_3_2:
 955                 break;
 956         default:
 957                 srcu_read_unlock(&nodes_srcu, idx);
 958                 return;
 959         }
 960
 961         /* do nothing if we didn't delivered stateful to ulp */
 962         if (!test_and_clear_bit(DLM_NODE_ULP_DELIVERED,
 963                                 &node->flags)) {
 964                 srcu_read_unlock(&nodes_srcu, idx);
 965                 return;
 966         }
 967
 968         spin_lock(&node->state_lock);
 969         /* we only ack if state is ESTABLISHED */
 970         switch (node->state) {
 971         case DLM_ESTABLISHED:
 972                 spin_unlock(&node->state_lock);
 973                 dlm_send_ack(node->nodeid, node->seq_next);
 974                 break;
 975         default:
 976                 spin_unlock(&node->state_lock);
 977                 /* do nothing FIN has it's own ack send */
 978                 break;
 979         }
 980         srcu_read_unlock(&nodes_srcu, idx);
 981 }
 982
 983 void dlm_midcomms_unack_msg_resend(int nodeid)
 984 {
 985         struct midcomms_node *node;
 986         struct dlm_mhandle *mh;
 987         int idx, ret;
 988
 989         idx = srcu_read_lock(&nodes_srcu);
 990         node = nodeid2node(nodeid, 0);
 991         if (!node) {
 992                 srcu_read_unlock(&nodes_srcu, idx);
 993                 return;
 994         }
 995
 996         /* old protocol, we don't support to retransmit on failure */
 997         switch (node->version) {
 998         case DLM_VERSION_3_2:
 999                 break;
1000         default:
1001                 srcu_read_unlock(&nodes_srcu, idx);
1002                 return;
1003         }
1004
1005         rcu_read_lock();
1006         list_for_each_entry_rcu(mh, &node->send_queue, list) {
1007                 if (!mh->committed)
1008                         continue;
1009
1010                 ret = dlm_lowcomms_resend_msg(mh->msg);
1011                 if (!ret)
1012                         log_print_ratelimited("retransmit dlm msg, seq %u, nodeid %d",
1013                                               mh->seq, node->nodeid);
1014         }
1015         rcu_read_unlock();
1016         srcu_read_unlock(&nodes_srcu, idx);
1017 }
1018
1019 static void dlm_fill_opts_header(struct dlm_opts *opts, uint16_t inner_len,
1020                                  uint32_t seq)
1021 {
1022         opts->o_header.h_cmd = DLM_OPTS;
1023         opts->o_header.h_version = (DLM_HEADER_MAJOR | DLM_HEADER_MINOR);
1024         opts->o_header.h_nodeid = dlm_our_nodeid();
1025         opts->o_header.h_length = DLM_MIDCOMMS_OPT_LEN + inner_len;
1026         opts->o_header.u.h_seq = seq;
1027         header_out(&opts->o_header);
1028 }
1029
1030 static void midcomms_new_msg_cb(void *data)
1031 {
1032         struct dlm_mhandle *mh = data;
1033
1034         atomic_inc(&mh->node->send_queue_cnt);
1035
1036         spin_lock(&mh->node->send_queue_lock);
1037         list_add_tail_rcu(&mh->list, &mh->node->send_queue);
1038         spin_unlock(&mh->node->send_queue_lock);
1039
1040         mh->seq = mh->node->seq_send++;
1041 }
1042
1043 static struct dlm_msg *dlm_midcomms_get_msg_3_2(struct dlm_mhandle *mh, int nodeid,
1044                                                 int len, gfp_t allocation, char **ppc)
1045 {
1046         struct dlm_opts *opts;
1047         struct dlm_msg *msg;
1048
1049         msg = dlm_lowcomms_new_msg(nodeid, len + DLM_MIDCOMMS_OPT_LEN,
1050                                    allocation, ppc, midcomms_new_msg_cb, mh);
1051         if (!msg)
1052                 return NULL;
1053
1054         opts = (struct dlm_opts *)*ppc;
1055         mh->opts = opts;
1056
1057         /* add possible options here */
1058         dlm_fill_opts_header(opts, len, mh->seq);
1059
1060         *ppc += sizeof(*opts);
1061         mh->inner_hd = (const struct dlm_header *)*ppc;
1062         return msg;
1063 }
1064
1065 struct dlm_mhandle *dlm_midcomms_get_mhandle(int nodeid, int len,
1066                                              gfp_t allocation, char **ppc)
1067 {
1068         struct midcomms_node *node;
1069         struct dlm_mhandle *mh;
1070         struct dlm_msg *msg;
1071         int idx;
1072
1073         idx = srcu_read_lock(&nodes_srcu);
1074         node = nodeid2node(nodeid, 0);
1075         if (!node) {
1076                 WARN_ON_ONCE(1);
1077                 goto err;
1078         }
1079
1080         /* this is a bug, however we going on and hope it will be resolved */
1081         WARN_ON(test_bit(DLM_NODE_FLAG_STOP_TX, &node->flags));
1082
1083         mh = dlm_allocate_mhandle();
1084         if (!mh)
1085                 goto err;
1086
1087         mh->committed = false;
1088         mh->ack_rcv = NULL;
1089         mh->idx = idx;
1090         mh->node = node;
1091
1092         switch (node->version) {
1093         case DLM_VERSION_3_1:
1094                 msg = dlm_lowcomms_new_msg(nodeid, len, allocation, ppc,
1095                                            NULL, NULL);
1096                 if (!msg) {
1097                         dlm_free_mhandle(mh);
1098                         goto err;
1099                 }
1100
1101                 break;
1102         case DLM_VERSION_3_2:
1103                 msg = dlm_midcomms_get_msg_3_2(mh, nodeid, len, allocation,
1104                                                ppc);
1105                 if (!msg) {
1106                         dlm_free_mhandle(mh);
1107                         goto err;
1108                 }
1109
1110                 break;
1111         default:
1112                 dlm_free_mhandle(mh);
1113                 WARN_ON(1);
1114                 goto err;
1115         }
1116
1117         mh->msg = msg;
1118
1119         /* keep in mind that is a must to call
1120          * dlm_midcomms_commit_msg() which releases
1121          * nodes_srcu using mh->idx which is assumed
1122          * here that the application will call it.
1123          */
1124         return mh;
1125
1126 err:
1127         srcu_read_unlock(&nodes_srcu, idx);
1128         return NULL;
1129 }
1130
1131 static void dlm_midcomms_commit_msg_3_2(struct dlm_mhandle *mh)
1132 {
1133         /* nexthdr chain for fast lookup */
1134         mh->opts->o_nextcmd = mh->inner_hd->h_cmd;
1135         mh->committed = true;
1136         dlm_lowcomms_commit_msg(mh->msg);
1137 }
1138
1139 void dlm_midcomms_commit_mhandle(struct dlm_mhandle *mh)
1140 {
1141         switch (mh->node->version) {
1142         case DLM_VERSION_3_1:
1143                 srcu_read_unlock(&nodes_srcu, mh->idx);
1144
1145                 dlm_lowcomms_commit_msg(mh->msg);
1146                 dlm_lowcomms_put_msg(mh->msg);
1147                 /* mh is not part of rcu list in this case */
1148                 dlm_free_mhandle(mh);
1149                 break;
1150         case DLM_VERSION_3_2:
1151                 dlm_midcomms_commit_msg_3_2(mh);
1152                 srcu_read_unlock(&nodes_srcu, mh->idx);
1153                 break;
1154         default:
1155                 srcu_read_unlock(&nodes_srcu, mh->idx);
1156                 WARN_ON(1);
1157                 break;
1158         }
1159 }
1160
1161 int dlm_midcomms_start(void)
1162 {
1163         int i;
1164
1165         for (i = 0; i < CONN_HASH_SIZE; i++)
1166                 INIT_HLIST_HEAD(&node_hash[i]);
1167
1168         return dlm_lowcomms_start();
1169 }
1170
1171 static void dlm_act_fin_ack_rcv(struct midcomms_node *node)
1172 {
1173         spin_lock(&node->state_lock);
1174         pr_debug("receive active fin ack from node %d with state %s\n",
1175                  node->nodeid, dlm_state_str(node->state));
1176
1177         switch (node->state) {
1178         case DLM_FIN_WAIT1:
1179                 node->state = DLM_FIN_WAIT2;
1180                 pr_debug("switch node %d to state %s\n",
1181                          node->nodeid, dlm_state_str(node->state));
1182                 break;
1183         case DLM_CLOSING:
1184                 midcomms_node_reset(node);
1185                 pr_debug("switch node %d to state %s\n",
1186                          node->nodeid, dlm_state_str(node->state));
1187                 wake_up(&node->shutdown_wait);
1188                 break;
1189         case DLM_CLOSED:
1190                 /* not valid but somehow we got what we want */
1191                 wake_up(&node->shutdown_wait);
1192                 break;
1193         default:
1194                 spin_unlock(&node->state_lock);
1195                 log_print("%s: unexpected state: %d\n",
1196                           __func__, node->state);
1197                 WARN_ON(1);
1198                 return;
1199         }
1200         spin_unlock(&node->state_lock);
1201 }
1202
1203 void dlm_midcomms_add_member(int nodeid)
1204 {
1205         struct midcomms_node *node;
1206         int idx;
1207
1208         if (nodeid == dlm_our_nodeid())
1209                 return;
1210
1211         idx = srcu_read_lock(&nodes_srcu);
1212         node = nodeid2node(nodeid, GFP_NOFS);
1213         if (!node) {
1214                 srcu_read_unlock(&nodes_srcu, idx);
1215                 return;
1216         }
1217
1218         spin_lock(&node->state_lock);
1219         if (!node->users) {
1220                 pr_debug("receive add member from node %d with state %s\n",
1221                          node->nodeid, dlm_state_str(node->state));
1222                 switch (node->state) {
1223                 case DLM_ESTABLISHED:
1224                         break;
1225                 case DLM_CLOSED:
1226                         node->state = DLM_ESTABLISHED;
1227                         pr_debug("switch node %d to state %s\n",
1228                                  node->nodeid, dlm_state_str(node->state));
1229                         break;
1230                 default:
1231                         /* some invalid state passive shutdown
1232                          * was failed, we try to reset and
1233                          * hope it will go on.
1234                          */
1235                         log_print("reset node %d because shutdown stuck",
1236                                   node->nodeid);
1237
1238                         midcomms_node_reset(node);
1239                         node->state = DLM_ESTABLISHED;
1240                         break;
1241                 }
1242         }
1243
1244         node->users++;
1245         pr_debug("node %d users inc count %d\n", nodeid, node->users);
1246         spin_unlock(&node->state_lock);
1247
1248         srcu_read_unlock(&nodes_srcu, idx);
1249 }
1250
1251 void dlm_midcomms_remove_member(int nodeid)
1252 {
1253         struct midcomms_node *node;
1254         int idx;
1255
1256         if (nodeid == dlm_our_nodeid())
1257                 return;
1258
1259         idx = srcu_read_lock(&nodes_srcu);
1260         node = nodeid2node(nodeid, 0);
1261         if (!node) {
1262                 srcu_read_unlock(&nodes_srcu, idx);
1263                 return;
1264         }
1265
1266         spin_lock(&node->state_lock);
1267         node->users--;
1268         pr_debug("node %d users dec count %d\n", nodeid, node->users);
1269
1270         /* hitting users count to zero means the
1271          * other side is running dlm_midcomms_stop()
1272          * we meet us to have a clean disconnect.
1273          */
1274         if (node->users == 0) {
1275                 pr_debug("receive remove member from node %d with state %s\n",
1276                          node->nodeid, dlm_state_str(node->state));
1277                 switch (node->state) {
1278                 case DLM_ESTABLISHED:
1279                         break;
1280                 case DLM_CLOSE_WAIT:
1281                         /* passive shutdown DLM_LAST_ACK case 2 */
1282                         node->state = DLM_LAST_ACK;
1283                         spin_unlock(&node->state_lock);
1284
1285                         pr_debug("switch node %d to state %s case 2\n",
1286                                  node->nodeid, dlm_state_str(node->state));
1287                         goto send_fin;
1288                 case DLM_LAST_ACK:
1289                         /* probably receive fin caught it, do nothing */
1290                         break;
1291                 case DLM_CLOSED:
1292                         /* already gone, do nothing */
1293                         break;
1294                 default:
1295                         log_print("%s: unexpected state: %d\n",
1296                                   __func__, node->state);
1297                         break;
1298                 }
1299         }
1300         spin_unlock(&node->state_lock);
1301
1302         srcu_read_unlock(&nodes_srcu, idx);
1303         return;
1304
1305 send_fin:
1306         set_bit(DLM_NODE_FLAG_STOP_RX, &node->flags);
1307         dlm_send_fin(node, dlm_pas_fin_ack_rcv);
1308         srcu_read_unlock(&nodes_srcu, idx);
1309 }
1310
1311 static void midcomms_node_release(struct rcu_head *rcu)
1312 {
1313         struct midcomms_node *node = container_of(rcu, struct midcomms_node, rcu);
1314
1315         WARN_ON(atomic_read(&node->send_queue_cnt));
1316         kfree(node);
1317 }
1318
1319 static void midcomms_shutdown(struct midcomms_node *node)
1320 {
1321         int ret;
1322
1323         /* old protocol, we don't wait for pending operations */
1324         switch (node->version) {
1325         case DLM_VERSION_3_2:
1326                 break;
1327         default:
1328                 return;
1329         }
1330
1331         spin_lock(&node->state_lock);
1332         pr_debug("receive active shutdown for node %d with state %s\n",
1333                  node->nodeid, dlm_state_str(node->state));
1334         switch (node->state) {
1335         case DLM_ESTABLISHED:
1336                 node->state = DLM_FIN_WAIT1;
1337                 pr_debug("switch node %d to state %s case 2\n",
1338                          node->nodeid, dlm_state_str(node->state));
1339                 break;
1340         case DLM_CLOSED:
1341                 /* we have what we want */
1342                 spin_unlock(&node->state_lock);
1343                 return;
1344         default:
1345                 /* busy to enter DLM_FIN_WAIT1, wait until passive
1346                  * done in shutdown_wait to enter DLM_CLOSED.
1347                  */
1348                 break;
1349         }
1350         spin_unlock(&node->state_lock);
1351
1352         if (node->state == DLM_FIN_WAIT1) {
1353                 dlm_send_fin(node, dlm_act_fin_ack_rcv);
1354
1355                 if (DLM_DEBUG_FENCE_TERMINATION)
1356                         msleep(5000);
1357         }
1358
1359         /* wait for other side dlm + fin */
1360         ret = wait_event_timeout(node->shutdown_wait,
1361                                  node->state == DLM_CLOSED ||
1362                                  test_bit(DLM_NODE_FLAG_CLOSE, &node->flags),
1363                                  DLM_SHUTDOWN_TIMEOUT);
1364         if (!ret || test_bit(DLM_NODE_FLAG_CLOSE, &node->flags)) {
1365                 pr_debug("active shutdown timed out for node %d with state %s\n",
1366                          node->nodeid, dlm_state_str(node->state));
1367                 midcomms_node_reset(node);
1368                 return;
1369         }
1370
1371         pr_debug("active shutdown done for node %d with state %s\n",
1372                  node->nodeid, dlm_state_str(node->state));
1373 }
1374
1375 void dlm_midcomms_shutdown(void)
1376 {
1377         struct midcomms_node *node;
1378         int i, idx;
1379
1380         mutex_lock(&close_lock);
1381         idx = srcu_read_lock(&nodes_srcu);
1382         for (i = 0; i < CONN_HASH_SIZE; i++) {
1383                 hlist_for_each_entry_rcu(node, &node_hash[i], hlist) {
1384                         midcomms_shutdown(node);
1385
1386                         dlm_delete_debug_comms_file(node->debugfs);
1387
1388                         spin_lock(&nodes_lock);
1389                         hlist_del_rcu(&node->hlist);
1390                         spin_unlock(&nodes_lock);
1391
1392                         call_srcu(&nodes_srcu, &node->rcu, midcomms_node_release);
1393                 }
1394         }
1395         srcu_read_unlock(&nodes_srcu, idx);
1396         mutex_unlock(&close_lock);
1397
1398         dlm_lowcomms_shutdown();
1399 }
1400
1401 int dlm_midcomms_close(int nodeid)
1402 {
1403         struct midcomms_node *node;
1404         int idx, ret;
1405
1406         if (nodeid == dlm_our_nodeid())
1407                 return 0;
1408
1409         idx = srcu_read_lock(&nodes_srcu);
1410         /* Abort pending close/remove operation */
1411         node = nodeid2node(nodeid, 0);
1412         if (node) {
1413                 /* let shutdown waiters leave */
1414                 set_bit(DLM_NODE_FLAG_CLOSE, &node->flags);
1415                 wake_up(&node->shutdown_wait);
1416         }
1417         srcu_read_unlock(&nodes_srcu, idx);
1418
1419         synchronize_srcu(&nodes_srcu);
1420
1421         idx = srcu_read_lock(&nodes_srcu);
1422         mutex_lock(&close_lock);
1423         node = nodeid2node(nodeid, 0);
1424         if (!node) {
1425                 mutex_unlock(&close_lock);
1426                 srcu_read_unlock(&nodes_srcu, idx);
1427                 return dlm_lowcomms_close(nodeid);
1428         }
1429
1430         ret = dlm_lowcomms_close(nodeid);
1431         spin_lock(&node->state_lock);
1432         midcomms_node_reset(node);
1433         spin_unlock(&node->state_lock);
1434         srcu_read_unlock(&nodes_srcu, idx);
1435         mutex_unlock(&close_lock);
1436
1437         return ret;
1438 }
1439
1440 /* debug functionality to send raw dlm msg from user space */
1441 struct dlm_rawmsg_data {
1442         struct midcomms_node *node;
1443         void *buf;
1444 };
1445
1446 static void midcomms_new_rawmsg_cb(void *data)
1447 {
1448         struct dlm_rawmsg_data *rd = data;
1449         struct dlm_header *h = rd->buf;
1450
1451         switch (h->h_version) {
1452         case cpu_to_le32(DLM_VERSION_3_1):
1453                 break;
1454         default:
1455                 switch (h->h_cmd) {
1456                 case DLM_OPTS:
1457                         if (!h->u.h_seq)
1458                                 h->u.h_seq = rd->node->seq_send++;
1459                         break;
1460                 default:
1461                         break;
1462                 }
1463                 break;
1464         }
1465 }
1466
1467 int dlm_midcomms_rawmsg_send(struct midcomms_node *node, void *buf,
1468                              int buflen)
1469 {
1470         struct dlm_rawmsg_data rd;
1471         struct dlm_msg *msg;
1472         char *msgbuf;
1473
1474         rd.node = node;
1475         rd.buf = buf;
1476
1477         msg = dlm_lowcomms_new_msg(node->nodeid, buflen, GFP_NOFS,
1478                                    &msgbuf, midcomms_new_rawmsg_cb, &rd);
1479         if (!msg)
1480                 return -ENOMEM;
1481
1482         memcpy(msgbuf, buf, buflen);
1483         dlm_lowcomms_commit_msg(msg);
1484         return 0;
1485 }
1486