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