drivers/staging/lustre/lnet/klnds/socklnd/socklnd_proto.c

   1 /*
   2  * Copyright (c) 2009, 2010, Oracle and/or its affiliates. All rights reserved.
   3  *
   4  * Copyright (c) 2012, Intel Corporation.
   5  *
   6  *   Author: Zach Brown <zab@zabbo.net>
   7  *   Author: Peter J. Braam <braam@clusterfs.com>
   8  *   Author: Phil Schwan <phil@clusterfs.com>
   9  *   Author: Eric Barton <eric@bartonsoftware.com>
  10  *
  11  *   This file is part of Portals, http://www.sf.net/projects/sandiaportals/
  12  *
  13  *   Portals is free software; you can redistribute it and/or
  14  *   modify it under the terms of version 2 of the GNU General Public
  15  *   License as published by the Free Software Foundation.
  16  *
  17  *   Portals is distributed in the hope that it will be useful,
  18  *   but WITHOUT ANY WARRANTY; without even the implied warranty of
  19  *   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
  20  *   GNU General Public License for more details.
  21  *
  22  *   You should have received a copy of the GNU General Public License
  23  *   along with Portals; if not, write to the Free Software
  24  *   Foundation, Inc., 675 Mass Ave, Cambridge, MA 02139, USA.
  25  */
  26
  27 #include "socklnd.h"
  28
  29 /*
  30  * Protocol entries :
  31  *   pro_send_hello       : send hello message
  32  *   pro_recv_hello       : receive hello message
  33  *   pro_pack        : pack message header
  34  *   pro_unpack    : unpack message header
  35  *   pro_queue_tx_zcack() : Called holding BH lock: kss_lock
  36  *                        return 1 if ACK is piggybacked, otherwise return 0
  37  *   pro_queue_tx_msg()   : Called holding BH lock: kss_lock
  38  *                        return the ACK that piggybacked by my message, or NULL
  39  *   pro_handle_zcreq()   : handler of incoming ZC-REQ
  40  *   pro_handle_zcack()   : handler of incoming ZC-ACK
  41  *   pro_match_tx()       : Called holding glock
  42  */
  43
  44 static ksock_tx_t *
  45 ksocknal_queue_tx_msg_v1(ksock_conn_t *conn, ksock_tx_t *tx_msg)
  46 {
  47         /* V1.x, just enqueue it */
  48         list_add_tail(&tx_msg->tx_list, &conn->ksnc_tx_queue);
  49         return NULL;
  50 }
  51
  52 void
  53 ksocknal_next_tx_carrier(ksock_conn_t *conn)
  54 {
  55         ksock_tx_t     *tx = conn->ksnc_tx_carrier;
  56
  57         /* Called holding BH lock: conn->ksnc_scheduler->kss_lock */
  58         LASSERT (!list_empty(&conn->ksnc_tx_queue));
  59         LASSERT (tx != NULL);
  60
  61         /* Next TX that can carry ZC-ACK or LNet message */
  62         if (tx->tx_list.next == &conn->ksnc_tx_queue) {
  63                 /* no more packets queued */
  64                 conn->ksnc_tx_carrier = NULL;
  65         } else {
  66                 conn->ksnc_tx_carrier = list_entry(tx->tx_list.next,
  67                                                        ksock_tx_t, tx_list);
  68                 LASSERT (conn->ksnc_tx_carrier->tx_msg.ksm_type == tx->tx_msg.ksm_type);
  69         }
  70 }
  71
  72 static int
  73 ksocknal_queue_tx_zcack_v2(ksock_conn_t *conn,
  74                            ksock_tx_t *tx_ack, __u64 cookie)
  75 {
  76         ksock_tx_t *tx = conn->ksnc_tx_carrier;
  77
  78         LASSERT (tx_ack == NULL ||
  79                  tx_ack->tx_msg.ksm_type == KSOCK_MSG_NOOP);
  80
  81         /*
  82          * Enqueue or piggyback tx_ack / cookie
  83          * . no tx can piggyback cookie of tx_ack (or cookie), just
  84          *   enqueue the tx_ack (if tx_ack != NUL) and return NULL.
  85          * . There is tx can piggyback cookie of tx_ack (or cookie),
  86          *   piggyback the cookie and return the tx.
  87          */
  88         if (tx == NULL) {
  89                 if (tx_ack != NULL) {
  90                         list_add_tail(&tx_ack->tx_list,
  91                                           &conn->ksnc_tx_queue);
  92                         conn->ksnc_tx_carrier = tx_ack;
  93                 }
  94                 return 0;
  95         }
  96
  97         if (tx->tx_msg.ksm_type == KSOCK_MSG_NOOP) {
  98                 /* tx is noop zc-ack, can't piggyback zc-ack cookie */
  99                 if (tx_ack != NULL)
 100                         list_add_tail(&tx_ack->tx_list,
 101                                           &conn->ksnc_tx_queue);
 102                 return 0;
 103         }
 104
 105         LASSERT(tx->tx_msg.ksm_type == KSOCK_MSG_LNET);
 106         LASSERT(tx->tx_msg.ksm_zc_cookies[1] == 0);
 107
 108         if (tx_ack != NULL)
 109                 cookie = tx_ack->tx_msg.ksm_zc_cookies[1];
 110
 111         /* piggyback the zc-ack cookie */
 112         tx->tx_msg.ksm_zc_cookies[1] = cookie;
 113         /* move on to the next TX which can carry cookie */
 114         ksocknal_next_tx_carrier(conn);
 115
 116         return 1;
 117 }
 118
 119 static ksock_tx_t *
 120 ksocknal_queue_tx_msg_v2(ksock_conn_t *conn, ksock_tx_t *tx_msg)
 121 {
 122         ksock_tx_t  *tx  = conn->ksnc_tx_carrier;
 123
 124         /*
 125          * Enqueue tx_msg:
 126          * . If there is no NOOP on the connection, just enqueue
 127          *   tx_msg and return NULL
 128          * . If there is NOOP on the connection, piggyback the cookie
 129          *   and replace the NOOP tx, and return the NOOP tx.
 130          */
 131         if (tx == NULL) { /* nothing on queue */
 132                 list_add_tail(&tx_msg->tx_list, &conn->ksnc_tx_queue);
 133                 conn->ksnc_tx_carrier = tx_msg;
 134                 return NULL;
 135         }
 136
 137         if (tx->tx_msg.ksm_type == KSOCK_MSG_LNET) { /* nothing to carry */
 138                 list_add_tail(&tx_msg->tx_list, &conn->ksnc_tx_queue);
 139                 return NULL;
 140         }
 141
 142         LASSERT (tx->tx_msg.ksm_type == KSOCK_MSG_NOOP);
 143
 144         /* There is a noop zc-ack can be piggybacked */
 145         tx_msg->tx_msg.ksm_zc_cookies[1] = tx->tx_msg.ksm_zc_cookies[1];
 146         ksocknal_next_tx_carrier(conn);
 147
 148         /* use new_tx to replace the noop zc-ack packet */
 149         list_add(&tx_msg->tx_list, &tx->tx_list);
 150         list_del(&tx->tx_list);
 151
 152         return tx;
 153 }
 154
 155 static int
 156 ksocknal_queue_tx_zcack_v3(ksock_conn_t *conn,
 157                            ksock_tx_t *tx_ack, __u64 cookie)
 158 {
 159         ksock_tx_t *tx;
 160
 161         if (conn->ksnc_type != SOCKLND_CONN_ACK)
 162                 return ksocknal_queue_tx_zcack_v2(conn, tx_ack, cookie);
 163
 164         /* non-blocking ZC-ACK (to router) */
 165         LASSERT (tx_ack == NULL ||
 166                  tx_ack->tx_msg.ksm_type == KSOCK_MSG_NOOP);
 167
 168         if ((tx = conn->ksnc_tx_carrier) == NULL) {
 169                 if (tx_ack != NULL) {
 170                         list_add_tail(&tx_ack->tx_list,
 171                                           &conn->ksnc_tx_queue);
 172                         conn->ksnc_tx_carrier = tx_ack;
 173                 }
 174                 return 0;
 175         }
 176
 177         /* conn->ksnc_tx_carrier != NULL */
 178
 179         if (tx_ack != NULL)
 180                 cookie = tx_ack->tx_msg.ksm_zc_cookies[1];
 181
 182         if (cookie == SOCKNAL_KEEPALIVE_PING) /* ignore keepalive PING */
 183                 return 1;
 184
 185         if (tx->tx_msg.ksm_zc_cookies[1] == SOCKNAL_KEEPALIVE_PING) {
 186                 /* replace the keepalive PING with a real ACK */
 187                 LASSERT (tx->tx_msg.ksm_zc_cookies[0] == 0);
 188                 tx->tx_msg.ksm_zc_cookies[1] = cookie;
 189                 return 1;
 190         }
 191
 192         if (cookie == tx->tx_msg.ksm_zc_cookies[0] ||
 193             cookie == tx->tx_msg.ksm_zc_cookies[1]) {
 194                 CWARN("%s: duplicated ZC cookie: "LPU64"\n",
 195                       libcfs_id2str(conn->ksnc_peer->ksnp_id), cookie);
 196                 return 1; /* XXX return error in the future */
 197         }
 198
 199         if (tx->tx_msg.ksm_zc_cookies[0] == 0) {
 200                 /* NOOP tx has only one ZC-ACK cookie, can carry at least one more */
 201                 if (tx->tx_msg.ksm_zc_cookies[1] > cookie) {
 202                         tx->tx_msg.ksm_zc_cookies[0] = tx->tx_msg.ksm_zc_cookies[1];
 203                         tx->tx_msg.ksm_zc_cookies[1] = cookie;
 204                 } else {
 205                         tx->tx_msg.ksm_zc_cookies[0] = cookie;
 206                 }
 207
 208                 if (tx->tx_msg.ksm_zc_cookies[0] - tx->tx_msg.ksm_zc_cookies[1] > 2) {
 209                         /* not likely to carry more ACKs, skip it to simplify logic */
 210                         ksocknal_next_tx_carrier(conn);
 211                 }
 212
 213                 return 1;
 214         }
 215
 216         /* takes two or more cookies already */
 217
 218         if (tx->tx_msg.ksm_zc_cookies[0] > tx->tx_msg.ksm_zc_cookies[1]) {
 219                 __u64   tmp = 0;
 220
 221                 /* two separated cookies: (a+2, a) or (a+1, a) */
 222                 LASSERT (tx->tx_msg.ksm_zc_cookies[0] -
 223                          tx->tx_msg.ksm_zc_cookies[1] <= 2);
 224
 225                 if (tx->tx_msg.ksm_zc_cookies[0] -
 226                     tx->tx_msg.ksm_zc_cookies[1] == 2) {
 227                         if (cookie == tx->tx_msg.ksm_zc_cookies[1] + 1)
 228                                 tmp = cookie;
 229                 } else if (cookie == tx->tx_msg.ksm_zc_cookies[1] - 1) {
 230                         tmp = tx->tx_msg.ksm_zc_cookies[1];
 231                 } else if (cookie == tx->tx_msg.ksm_zc_cookies[0] + 1) {
 232                         tmp = tx->tx_msg.ksm_zc_cookies[0];
 233                 }
 234
 235                 if (tmp != 0) {
 236                         /* range of cookies */
 237                         tx->tx_msg.ksm_zc_cookies[0] = tmp - 1;
 238                         tx->tx_msg.ksm_zc_cookies[1] = tmp + 1;
 239                         return 1;
 240                 }
 241
 242         } else {
 243                 /* ksm_zc_cookies[0] < ksm_zc_cookies[1], it is range of cookies */
 244                 if (cookie >= tx->tx_msg.ksm_zc_cookies[0] &&
 245                     cookie <= tx->tx_msg.ksm_zc_cookies[1]) {
 246                         CWARN("%s: duplicated ZC cookie: "LPU64"\n",
 247                               libcfs_id2str(conn->ksnc_peer->ksnp_id), cookie);
 248                         return 1; /* XXX: return error in the future */
 249                 }
 250
 251                 if (cookie == tx->tx_msg.ksm_zc_cookies[1] + 1) {
 252                         tx->tx_msg.ksm_zc_cookies[1] = cookie;
 253                         return 1;
 254                 }
 255
 256                 if (cookie == tx->tx_msg.ksm_zc_cookies[0] - 1) {
 257                         tx->tx_msg.ksm_zc_cookies[0] = cookie;
 258                         return 1;
 259                 }
 260         }
 261
 262         /* failed to piggyback ZC-ACK */
 263         if (tx_ack != NULL) {
 264                 list_add_tail(&tx_ack->tx_list, &conn->ksnc_tx_queue);
 265                 /* the next tx can piggyback at least 1 ACK */
 266                 ksocknal_next_tx_carrier(conn);
 267         }
 268
 269         return 0;
 270 }
 271
 272 static int
 273 ksocknal_match_tx(ksock_conn_t *conn, ksock_tx_t *tx, int nonblk)
 274 {
 275         int nob;
 276
 277 #if SOCKNAL_VERSION_DEBUG
 278         if (!*ksocknal_tunables.ksnd_typed_conns)
 279                 return SOCKNAL_MATCH_YES;
 280 #endif
 281
 282         if (tx == NULL || tx->tx_lnetmsg == NULL) {
 283                 /* noop packet */
 284                 nob = offsetof(ksock_msg_t, ksm_u);
 285         } else {
 286                 nob = tx->tx_lnetmsg->msg_len +
 287                       ((conn->ksnc_proto == &ksocknal_protocol_v1x) ?
 288                        sizeof(lnet_hdr_t) : sizeof(ksock_msg_t));
 289         }
 290
 291         /* default checking for typed connection */
 292         switch (conn->ksnc_type) {
 293         default:
 294                 CERROR("ksnc_type bad: %u\n", conn->ksnc_type);
 295                 LBUG();
 296         case SOCKLND_CONN_ANY:
 297                 return SOCKNAL_MATCH_YES;
 298
 299         case SOCKLND_CONN_BULK_IN:
 300                 return SOCKNAL_MATCH_MAY;
 301
 302         case SOCKLND_CONN_BULK_OUT:
 303                 if (nob < *ksocknal_tunables.ksnd_min_bulk)
 304                         return SOCKNAL_MATCH_MAY;
 305                 else
 306                         return SOCKNAL_MATCH_YES;
 307
 308         case SOCKLND_CONN_CONTROL:
 309                 if (nob >= *ksocknal_tunables.ksnd_min_bulk)
 310                         return SOCKNAL_MATCH_MAY;
 311                 else
 312                         return SOCKNAL_MATCH_YES;
 313         }
 314 }
 315
 316 static int
 317 ksocknal_match_tx_v3(ksock_conn_t *conn, ksock_tx_t *tx, int nonblk)
 318 {
 319         int nob;
 320
 321         if (tx == NULL || tx->tx_lnetmsg == NULL)
 322                 nob = offsetof(ksock_msg_t, ksm_u);
 323         else
 324                 nob = tx->tx_lnetmsg->msg_len + sizeof(ksock_msg_t);
 325
 326         switch (conn->ksnc_type) {
 327         default:
 328                 CERROR("ksnc_type bad: %u\n", conn->ksnc_type);
 329                 LBUG();
 330         case SOCKLND_CONN_ANY:
 331                 return SOCKNAL_MATCH_NO;
 332
 333         case SOCKLND_CONN_ACK:
 334                 if (nonblk)
 335                         return SOCKNAL_MATCH_YES;
 336                 else if (tx == NULL || tx->tx_lnetmsg == NULL)
 337                         return SOCKNAL_MATCH_MAY;
 338                 else
 339                         return SOCKNAL_MATCH_NO;
 340
 341         case SOCKLND_CONN_BULK_OUT:
 342                 if (nonblk)
 343                         return SOCKNAL_MATCH_NO;
 344                 else if (nob < *ksocknal_tunables.ksnd_min_bulk)
 345                         return SOCKNAL_MATCH_MAY;
 346                 else
 347                         return SOCKNAL_MATCH_YES;
 348
 349         case SOCKLND_CONN_CONTROL:
 350                 if (nonblk)
 351                         return SOCKNAL_MATCH_NO;
 352                 else if (nob >= *ksocknal_tunables.ksnd_min_bulk)
 353                         return SOCKNAL_MATCH_MAY;
 354                 else
 355                         return SOCKNAL_MATCH_YES;
 356         }
 357 }
 358
 359 /* (Sink) handle incoming ZC request from sender */
 360 static int
 361 ksocknal_handle_zcreq(ksock_conn_t *c, __u64 cookie, int remote)
 362 {
 363         ksock_peer_t   *peer = c->ksnc_peer;
 364         ksock_conn_t   *conn;
 365         ksock_tx_t     *tx;
 366         int          rc;
 367
 368         read_lock(&ksocknal_data.ksnd_global_lock);
 369
 370         conn = ksocknal_find_conn_locked(peer, NULL, !!remote);
 371         if (conn != NULL) {
 372                 ksock_sched_t *sched = conn->ksnc_scheduler;
 373
 374                 LASSERT(conn->ksnc_proto->pro_queue_tx_zcack != NULL);
 375
 376                 spin_lock_bh(&sched->kss_lock);
 377
 378                 rc = conn->ksnc_proto->pro_queue_tx_zcack(conn, NULL, cookie);
 379
 380                 spin_unlock_bh(&sched->kss_lock);
 381
 382                 if (rc) { /* piggybacked */
 383                         read_unlock(&ksocknal_data.ksnd_global_lock);
 384                         return 0;
 385                 }
 386         }
 387
 388         read_unlock(&ksocknal_data.ksnd_global_lock);
 389
 390         /* ACK connection is not ready, or can't piggyback the ACK */
 391         tx = ksocknal_alloc_tx_noop(cookie, !!remote);
 392         if (tx == NULL)
 393                 return -ENOMEM;
 394
 395         if ((rc = ksocknal_launch_packet(peer->ksnp_ni, tx, peer->ksnp_id)) == 0)
 396                 return 0;
 397
 398         ksocknal_free_tx(tx);
 399         return rc;
 400 }
 401
 402 /* (Sender) handle ZC_ACK from sink */
 403 static int
 404 ksocknal_handle_zcack(ksock_conn_t *conn, __u64 cookie1, __u64 cookie2)
 405 {
 406         ksock_peer_t      *peer = conn->ksnc_peer;
 407         ksock_tx_t      *tx;
 408         ksock_tx_t      *tmp;
 409         LIST_HEAD     (zlist);
 410         int             count;
 411
 412         if (cookie1 == 0)
 413                 cookie1 = cookie2;
 414
 415         count = (cookie1 > cookie2) ? 2 : (cookie2 - cookie1 + 1);
 416
 417         if (cookie2 == SOCKNAL_KEEPALIVE_PING &&
 418             conn->ksnc_proto == &ksocknal_protocol_v3x) {
 419                 /* keepalive PING for V3.x, just ignore it */
 420                 return count == 1 ? 0 : -EPROTO;
 421         }
 422
 423         spin_lock(&peer->ksnp_lock);
 424
 425         list_for_each_entry_safe(tx, tmp,
 426                                      &peer->ksnp_zc_req_list, tx_zc_list) {
 427                 __u64 c = tx->tx_msg.ksm_zc_cookies[0];
 428
 429                 if (c == cookie1 || c == cookie2 || (cookie1 < c && c < cookie2)) {
 430                         tx->tx_msg.ksm_zc_cookies[0] = 0;
 431                         list_del(&tx->tx_zc_list);
 432                         list_add(&tx->tx_zc_list, &zlist);
 433
 434                         if (--count == 0)
 435                                 break;
 436                 }
 437         }
 438
 439         spin_unlock(&peer->ksnp_lock);
 440
 441         while (!list_empty(&zlist)) {
 442                 tx = list_entry(zlist.next, ksock_tx_t, tx_zc_list);
 443                 list_del(&tx->tx_zc_list);
 444                 ksocknal_tx_decref(tx);
 445         }
 446
 447         return count == 0 ? 0 : -EPROTO;
 448 }
 449
 450 static int
 451 ksocknal_send_hello_v1 (ksock_conn_t *conn, ksock_hello_msg_t *hello)
 452 {
 453         socket_t        *sock = conn->ksnc_sock;
 454         lnet_hdr_t        *hdr;
 455         lnet_magicversion_t *hmv;
 456         int               rc;
 457         int               i;
 458
 459         CLASSERT(sizeof(lnet_magicversion_t) == offsetof(lnet_hdr_t, src_nid));
 460
 461         LIBCFS_ALLOC(hdr, sizeof(*hdr));
 462         if (hdr == NULL) {
 463                 CERROR("Can't allocate lnet_hdr_t\n");
 464                 return -ENOMEM;
 465         }
 466
 467         hmv = (lnet_magicversion_t *)&hdr->dest_nid;
 468
 469         /* Re-organize V2.x message header to V1.x (lnet_hdr_t)
 470          * header and send out */
 471         hmv->magic       = cpu_to_le32 (LNET_PROTO_TCP_MAGIC);
 472         hmv->version_major = cpu_to_le16 (KSOCK_PROTO_V1_MAJOR);
 473         hmv->version_minor = cpu_to_le16 (KSOCK_PROTO_V1_MINOR);
 474
 475         if (the_lnet.ln_testprotocompat != 0) {
 476                 /* single-shot proto check */
 477                 LNET_LOCK();
 478                 if ((the_lnet.ln_testprotocompat & 1) != 0) {
 479                         hmv->version_major++;   /* just different! */
 480                         the_lnet.ln_testprotocompat &= ~1;
 481                 }
 482                 if ((the_lnet.ln_testprotocompat & 2) != 0) {
 483                         hmv->magic = LNET_PROTO_MAGIC;
 484                         the_lnet.ln_testprotocompat &= ~2;
 485                 }
 486                 LNET_UNLOCK();
 487         }
 488
 489         hdr->src_nid    = cpu_to_le64 (hello->kshm_src_nid);
 490         hdr->src_pid    = cpu_to_le32 (hello->kshm_src_pid);
 491         hdr->type          = cpu_to_le32 (LNET_MSG_HELLO);
 492         hdr->payload_length = cpu_to_le32 (hello->kshm_nips * sizeof(__u32));
 493         hdr->msg.hello.type = cpu_to_le32 (hello->kshm_ctype);
 494         hdr->msg.hello.incarnation = cpu_to_le64 (hello->kshm_src_incarnation);
 495
 496         rc = libcfs_sock_write(sock, hdr, sizeof(*hdr),lnet_acceptor_timeout());
 497
 498         if (rc != 0) {
 499                 CNETERR("Error %d sending HELLO hdr to %pI4h/%d\n",
 500                         rc, &conn->ksnc_ipaddr, conn->ksnc_port);
 501                 goto out;
 502         }
 503
 504         if (hello->kshm_nips == 0)
 505                 goto out;
 506
 507         for (i = 0; i < (int) hello->kshm_nips; i++) {
 508                 hello->kshm_ips[i] = __cpu_to_le32 (hello->kshm_ips[i]);
 509         }
 510
 511         rc = libcfs_sock_write(sock, hello->kshm_ips,
 512                                hello->kshm_nips * sizeof(__u32),
 513                                lnet_acceptor_timeout());
 514         if (rc != 0) {
 515                 CNETERR("Error %d sending HELLO payload (%d)"
 516                         " to %pI4h/%d\n", rc, hello->kshm_nips,
 517                         &conn->ksnc_ipaddr, conn->ksnc_port);
 518         }
 519 out:
 520         LIBCFS_FREE(hdr, sizeof(*hdr));
 521
 522         return rc;
 523 }
 524
 525 static int
 526 ksocknal_send_hello_v2 (ksock_conn_t *conn, ksock_hello_msg_t *hello)
 527 {
 528         socket_t   *sock = conn->ksnc_sock;
 529         int          rc;
 530
 531         hello->kshm_magic   = LNET_PROTO_MAGIC;
 532         hello->kshm_version = conn->ksnc_proto->pro_version;
 533
 534         if (the_lnet.ln_testprotocompat != 0) {
 535                 /* single-shot proto check */
 536                 LNET_LOCK();
 537                 if ((the_lnet.ln_testprotocompat & 1) != 0) {
 538                         hello->kshm_version++;   /* just different! */
 539                         the_lnet.ln_testprotocompat &= ~1;
 540                 }
 541                 LNET_UNLOCK();
 542         }
 543
 544         rc = libcfs_sock_write(sock, hello, offsetof(ksock_hello_msg_t, kshm_ips),
 545                                lnet_acceptor_timeout());
 546
 547         if (rc != 0) {
 548                 CNETERR("Error %d sending HELLO hdr to %pI4h/%d\n",
 549                         rc, &conn->ksnc_ipaddr, conn->ksnc_port);
 550                 return rc;
 551         }
 552
 553         if (hello->kshm_nips == 0)
 554                 return 0;
 555
 556         rc = libcfs_sock_write(sock, hello->kshm_ips,
 557                                hello->kshm_nips * sizeof(__u32),
 558                                lnet_acceptor_timeout());
 559         if (rc != 0) {
 560                 CNETERR("Error %d sending HELLO payload (%d)"
 561                         " to %pI4h/%d\n", rc, hello->kshm_nips,
 562                         &conn->ksnc_ipaddr, conn->ksnc_port);
 563         }
 564
 565         return rc;
 566 }
 567
 568 static int
 569 ksocknal_recv_hello_v1(ksock_conn_t *conn, ksock_hello_msg_t *hello,int timeout)
 570 {
 571         socket_t        *sock = conn->ksnc_sock;
 572         lnet_hdr_t        *hdr;
 573         int               rc;
 574         int               i;
 575
 576         LIBCFS_ALLOC(hdr, sizeof(*hdr));
 577         if (hdr == NULL) {
 578                 CERROR("Can't allocate lnet_hdr_t\n");
 579                 return -ENOMEM;
 580         }
 581
 582         rc = libcfs_sock_read(sock, &hdr->src_nid,
 583                               sizeof (*hdr) - offsetof (lnet_hdr_t, src_nid),
 584                               timeout);
 585         if (rc != 0) {
 586                 CERROR("Error %d reading rest of HELLO hdr from %pI4h\n",
 587                         rc, &conn->ksnc_ipaddr);
 588                 LASSERT (rc < 0 && rc != -EALREADY);
 589                 goto out;
 590         }
 591
 592         /* ...and check we got what we expected */
 593         if (hdr->type != cpu_to_le32 (LNET_MSG_HELLO)) {
 594                 CERROR("Expecting a HELLO hdr,"
 595                         " but got type %d from %pI4h\n",
 596                         le32_to_cpu (hdr->type),
 597                         &conn->ksnc_ipaddr);
 598                 rc = -EPROTO;
 599                 goto out;
 600         }
 601
 602         hello->kshm_src_nid      = le64_to_cpu (hdr->src_nid);
 603         hello->kshm_src_pid      = le32_to_cpu (hdr->src_pid);
 604         hello->kshm_src_incarnation = le64_to_cpu (hdr->msg.hello.incarnation);
 605         hello->kshm_ctype          = le32_to_cpu (hdr->msg.hello.type);
 606         hello->kshm_nips            = le32_to_cpu (hdr->payload_length) /
 607                                          sizeof (__u32);
 608
 609         if (hello->kshm_nips > LNET_MAX_INTERFACES) {
 610                 CERROR("Bad nips %d from ip %pI4h\n",
 611                        hello->kshm_nips, &conn->ksnc_ipaddr);
 612                 rc = -EPROTO;
 613                 goto out;
 614         }
 615
 616         if (hello->kshm_nips == 0)
 617                 goto out;
 618
 619         rc = libcfs_sock_read(sock, hello->kshm_ips,
 620                               hello->kshm_nips * sizeof(__u32), timeout);
 621         if (rc != 0) {
 622                 CERROR("Error %d reading IPs from ip %pI4h\n",
 623                         rc, &conn->ksnc_ipaddr);
 624                 LASSERT(rc < 0 && rc != -EALREADY);
 625                 goto out;
 626         }
 627
 628         for (i = 0; i < (int) hello->kshm_nips; i++) {
 629                 hello->kshm_ips[i] = __le32_to_cpu(hello->kshm_ips[i]);
 630
 631                 if (hello->kshm_ips[i] == 0) {
 632                         CERROR("Zero IP[%d] from ip %pI4h\n",
 633                                i, &conn->ksnc_ipaddr);
 634                         rc = -EPROTO;
 635                         break;
 636                 }
 637         }
 638 out:
 639         LIBCFS_FREE(hdr, sizeof(*hdr));
 640
 641         return rc;
 642 }
 643
 644 static int
 645 ksocknal_recv_hello_v2 (ksock_conn_t *conn, ksock_hello_msg_t *hello, int timeout)
 646 {
 647         socket_t      *sock = conn->ksnc_sock;
 648         int             rc;
 649         int             i;
 650
 651         if (hello->kshm_magic == LNET_PROTO_MAGIC)
 652                 conn->ksnc_flip = 0;
 653         else
 654                 conn->ksnc_flip = 1;
 655
 656         rc = libcfs_sock_read(sock, &hello->kshm_src_nid,
 657                               offsetof(ksock_hello_msg_t, kshm_ips) -
 658                                        offsetof(ksock_hello_msg_t, kshm_src_nid),
 659                               timeout);
 660         if (rc != 0) {
 661                 CERROR("Error %d reading HELLO from %pI4h\n",
 662                         rc, &conn->ksnc_ipaddr);
 663                 LASSERT(rc < 0 && rc != -EALREADY);
 664                 return rc;
 665         }
 666
 667         if (conn->ksnc_flip) {
 668                 __swab32s(&hello->kshm_src_pid);
 669                 __swab64s(&hello->kshm_src_nid);
 670                 __swab32s(&hello->kshm_dst_pid);
 671                 __swab64s(&hello->kshm_dst_nid);
 672                 __swab64s(&hello->kshm_src_incarnation);
 673                 __swab64s(&hello->kshm_dst_incarnation);
 674                 __swab32s(&hello->kshm_ctype);
 675                 __swab32s(&hello->kshm_nips);
 676         }
 677
 678         if (hello->kshm_nips > LNET_MAX_INTERFACES) {
 679                 CERROR("Bad nips %d from ip %pI4h\n",
 680                        hello->kshm_nips, &conn->ksnc_ipaddr);
 681                 return -EPROTO;
 682         }
 683
 684         if (hello->kshm_nips == 0)
 685                 return 0;
 686
 687         rc = libcfs_sock_read(sock, hello->kshm_ips,
 688                               hello->kshm_nips * sizeof(__u32), timeout);
 689         if (rc != 0) {
 690                 CERROR("Error %d reading IPs from ip %pI4h\n",
 691                         rc, &conn->ksnc_ipaddr);
 692                 LASSERT(rc < 0 && rc != -EALREADY);
 693                 return rc;
 694         }
 695
 696         for (i = 0; i < (int) hello->kshm_nips; i++) {
 697                 if (conn->ksnc_flip)
 698                         __swab32s(&hello->kshm_ips[i]);
 699
 700                 if (hello->kshm_ips[i] == 0) {
 701                         CERROR("Zero IP[%d] from ip %pI4h\n",
 702                                i, &conn->ksnc_ipaddr);
 703                         return -EPROTO;
 704                 }
 705         }
 706
 707         return 0;
 708 }
 709
 710 static void
 711 ksocknal_pack_msg_v1(ksock_tx_t *tx)
 712 {
 713         /* V1.x has no KSOCK_MSG_NOOP */
 714         LASSERT(tx->tx_msg.ksm_type != KSOCK_MSG_NOOP);
 715         LASSERT(tx->tx_lnetmsg != NULL);
 716
 717         tx->tx_iov[0].iov_base = (void *)&tx->tx_lnetmsg->msg_hdr;
 718         tx->tx_iov[0].iov_len  = sizeof(lnet_hdr_t);
 719
 720         tx->tx_resid = tx->tx_nob = tx->tx_lnetmsg->msg_len + sizeof(lnet_hdr_t);
 721 }
 722
 723 static void
 724 ksocknal_pack_msg_v2(ksock_tx_t *tx)
 725 {
 726         tx->tx_iov[0].iov_base = (void *)&tx->tx_msg;
 727
 728         if (tx->tx_lnetmsg != NULL) {
 729                 LASSERT(tx->tx_msg.ksm_type != KSOCK_MSG_NOOP);
 730
 731                 tx->tx_msg.ksm_u.lnetmsg.ksnm_hdr = tx->tx_lnetmsg->msg_hdr;
 732                 tx->tx_iov[0].iov_len = sizeof(ksock_msg_t);
 733                 tx->tx_resid = tx->tx_nob = sizeof(ksock_msg_t) + tx->tx_lnetmsg->msg_len;
 734         } else {
 735                 LASSERT(tx->tx_msg.ksm_type == KSOCK_MSG_NOOP);
 736
 737                 tx->tx_iov[0].iov_len = offsetof(ksock_msg_t, ksm_u.lnetmsg.ksnm_hdr);
 738                 tx->tx_resid = tx->tx_nob = offsetof(ksock_msg_t,  ksm_u.lnetmsg.ksnm_hdr);
 739         }
 740         /* Don't checksum before start sending, because packet can be piggybacked with ACK */
 741 }
 742
 743 static void
 744 ksocknal_unpack_msg_v1(ksock_msg_t *msg)
 745 {
 746         msg->ksm_csum      = 0;
 747         msg->ksm_type      = KSOCK_MSG_LNET;
 748         msg->ksm_zc_cookies[0]  = msg->ksm_zc_cookies[1]  = 0;
 749 }
 750
 751 static void
 752 ksocknal_unpack_msg_v2(ksock_msg_t *msg)
 753 {
 754         return;  /* Do nothing */
 755 }
 756
 757 ksock_proto_t  ksocknal_protocol_v1x =
 758 {
 759         .pro_version        = KSOCK_PROTO_V1,
 760         .pro_send_hello  = ksocknal_send_hello_v1,
 761         .pro_recv_hello  = ksocknal_recv_hello_v1,
 762         .pro_pack              = ksocknal_pack_msg_v1,
 763         .pro_unpack          = ksocknal_unpack_msg_v1,
 764         .pro_queue_tx_msg       = ksocknal_queue_tx_msg_v1,
 765         .pro_handle_zcreq       = NULL,
 766         .pro_handle_zcack       = NULL,
 767         .pro_queue_tx_zcack     = NULL,
 768         .pro_match_tx      = ksocknal_match_tx
 769 };
 770
 771 ksock_proto_t  ksocknal_protocol_v2x =
 772 {
 773         .pro_version        = KSOCK_PROTO_V2,
 774         .pro_send_hello  = ksocknal_send_hello_v2,
 775         .pro_recv_hello  = ksocknal_recv_hello_v2,
 776         .pro_pack              = ksocknal_pack_msg_v2,
 777         .pro_unpack          = ksocknal_unpack_msg_v2,
 778         .pro_queue_tx_msg       = ksocknal_queue_tx_msg_v2,
 779         .pro_queue_tx_zcack     = ksocknal_queue_tx_zcack_v2,
 780         .pro_handle_zcreq       = ksocknal_handle_zcreq,
 781         .pro_handle_zcack       = ksocknal_handle_zcack,
 782         .pro_match_tx      = ksocknal_match_tx
 783 };
 784
 785 ksock_proto_t  ksocknal_protocol_v3x =
 786 {
 787         .pro_version        = KSOCK_PROTO_V3,
 788         .pro_send_hello  = ksocknal_send_hello_v2,
 789         .pro_recv_hello  = ksocknal_recv_hello_v2,
 790         .pro_pack              = ksocknal_pack_msg_v2,
 791         .pro_unpack          = ksocknal_unpack_msg_v2,
 792         .pro_queue_tx_msg       = ksocknal_queue_tx_msg_v2,
 793         .pro_queue_tx_zcack     = ksocknal_queue_tx_zcack_v3,
 794         .pro_handle_zcreq       = ksocknal_handle_zcreq,
 795         .pro_handle_zcack       = ksocknal_handle_zcack,
 796         .pro_match_tx      = ksocknal_match_tx_v3
 797 };