1 // SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB 2 /* Copyright (c) 2015 - 2021 Intel Corporation */ 3 #include "main.h" 4 #include "trace.h" 5 6 static void irdma_cm_post_event(struct irdma_cm_event *event); 7 static void irdma_disconnect_worker(struct work_struct *work); 8 9 /** 10 * irdma_free_sqbuf - put back puda buffer if refcount is 0 11 * @vsi: The VSI structure of the device 12 * @bufp: puda buffer to free 13 */ 14 void irdma_free_sqbuf(struct irdma_sc_vsi *vsi, void *bufp) 15 { 16 struct irdma_puda_buf *buf = bufp; 17 struct irdma_puda_rsrc *ilq = vsi->ilq; 18 19 if (refcount_dec_and_test(&buf->refcount)) 20 irdma_puda_ret_bufpool(ilq, buf); 21 } 22 23 /** 24 * irdma_record_ird_ord - Record IRD/ORD passed in 25 * @cm_node: connection's node 26 * @conn_ird: connection IRD 27 * @conn_ord: connection ORD 28 */ 29 static void irdma_record_ird_ord(struct irdma_cm_node *cm_node, u32 conn_ird, 30 u32 conn_ord) 31 { 32 if (conn_ird > cm_node->dev->hw_attrs.max_hw_ird) 33 conn_ird = cm_node->dev->hw_attrs.max_hw_ird; 34 35 if (conn_ord > cm_node->dev->hw_attrs.max_hw_ord) 36 conn_ord = cm_node->dev->hw_attrs.max_hw_ord; 37 else if (!conn_ord && cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO) 38 conn_ord = 1; 39 cm_node->ird_size = conn_ird; 40 cm_node->ord_size = conn_ord; 41 } 42 43 /** 44 * irdma_copy_ip_ntohl - copy IP address from network to host 45 * @dst: IP address in host order 46 * @src: IP address in network order (big endian) 47 */ 48 void irdma_copy_ip_ntohl(u32 *dst, __be32 *src) 49 { 50 *dst++ = ntohl(*src++); 51 *dst++ = ntohl(*src++); 52 *dst++ = ntohl(*src++); 53 *dst = ntohl(*src); 54 } 55 56 /** 57 * irdma_copy_ip_htonl - copy IP address from host to network order 58 * @dst: IP address in network order (big endian) 59 * @src: IP address in host order 60 */ 61 void irdma_copy_ip_htonl(__be32 *dst, u32 *src) 62 { 63 *dst++ = htonl(*src++); 64 *dst++ = htonl(*src++); 65 *dst++ = htonl(*src++); 66 *dst = htonl(*src); 67 } 68 69 /** 70 * irdma_get_addr_info 71 * @cm_node: contains ip/tcp info 72 * @cm_info: to get a copy of the cm_node ip/tcp info 73 */ 74 static void irdma_get_addr_info(struct irdma_cm_node *cm_node, 75 struct irdma_cm_info *cm_info) 76 { 77 memset(cm_info, 0, sizeof(*cm_info)); 78 cm_info->ipv4 = cm_node->ipv4; 79 cm_info->vlan_id = cm_node->vlan_id; 80 memcpy(cm_info->loc_addr, cm_node->loc_addr, sizeof(cm_info->loc_addr)); 81 memcpy(cm_info->rem_addr, cm_node->rem_addr, sizeof(cm_info->rem_addr)); 82 cm_info->loc_port = cm_node->loc_port; 83 cm_info->rem_port = cm_node->rem_port; 84 } 85 86 /** 87 * irdma_fill_sockaddr4 - fill in addr info for IPv4 connection 88 * @cm_node: connection's node 89 * @event: upper layer's cm event 90 */ 91 static inline void irdma_fill_sockaddr4(struct irdma_cm_node *cm_node, 92 struct iw_cm_event *event) 93 { 94 struct sockaddr_in *laddr = (struct sockaddr_in *)&event->local_addr; 95 struct sockaddr_in *raddr = (struct sockaddr_in *)&event->remote_addr; 96 97 laddr->sin_family = AF_INET; 98 raddr->sin_family = AF_INET; 99 100 laddr->sin_port = htons(cm_node->loc_port); 101 raddr->sin_port = htons(cm_node->rem_port); 102 103 laddr->sin_addr.s_addr = htonl(cm_node->loc_addr[0]); 104 raddr->sin_addr.s_addr = htonl(cm_node->rem_addr[0]); 105 } 106 107 /** 108 * irdma_fill_sockaddr6 - fill in addr info for IPv6 connection 109 * @cm_node: connection's node 110 * @event: upper layer's cm event 111 */ 112 static inline void irdma_fill_sockaddr6(struct irdma_cm_node *cm_node, 113 struct iw_cm_event *event) 114 { 115 struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&event->local_addr; 116 struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)&event->remote_addr; 117 118 laddr6->sin6_family = AF_INET6; 119 raddr6->sin6_family = AF_INET6; 120 121 laddr6->sin6_port = htons(cm_node->loc_port); 122 raddr6->sin6_port = htons(cm_node->rem_port); 123 124 irdma_copy_ip_htonl(laddr6->sin6_addr.in6_u.u6_addr32, 125 cm_node->loc_addr); 126 irdma_copy_ip_htonl(raddr6->sin6_addr.in6_u.u6_addr32, 127 cm_node->rem_addr); 128 } 129 130 /** 131 * irdma_get_cmevent_info - for cm event upcall 132 * @cm_node: connection's node 133 * @cm_id: upper layers cm struct for the event 134 * @event: upper layer's cm event 135 */ 136 static inline void irdma_get_cmevent_info(struct irdma_cm_node *cm_node, 137 struct iw_cm_id *cm_id, 138 struct iw_cm_event *event) 139 { 140 memcpy(&event->local_addr, &cm_id->m_local_addr, 141 sizeof(event->local_addr)); 142 memcpy(&event->remote_addr, &cm_id->m_remote_addr, 143 sizeof(event->remote_addr)); 144 if (cm_node) { 145 event->private_data = cm_node->pdata_buf; 146 event->private_data_len = (u8)cm_node->pdata.size; 147 event->ird = cm_node->ird_size; 148 event->ord = cm_node->ord_size; 149 } 150 } 151 152 /** 153 * irdma_send_cm_event - upcall cm's event handler 154 * @cm_node: connection's node 155 * @cm_id: upper layer's cm info struct 156 * @type: Event type to indicate 157 * @status: status for the event type 158 */ 159 static int irdma_send_cm_event(struct irdma_cm_node *cm_node, 160 struct iw_cm_id *cm_id, 161 enum iw_cm_event_type type, int status) 162 { 163 struct iw_cm_event event = {}; 164 165 event.event = type; 166 event.status = status; 167 trace_irdma_send_cm_event(cm_node, cm_id, type, status, 168 __builtin_return_address(0)); 169 170 ibdev_dbg(&cm_node->iwdev->ibdev, 171 "CM: cm_node %p cm_id=%p state=%d accel=%d event_type=%d status=%d\n", 172 cm_node, cm_id, cm_node->accelerated, cm_node->state, type, 173 status); 174 175 switch (type) { 176 case IW_CM_EVENT_CONNECT_REQUEST: 177 if (cm_node->ipv4) 178 irdma_fill_sockaddr4(cm_node, &event); 179 else 180 irdma_fill_sockaddr6(cm_node, &event); 181 event.provider_data = cm_node; 182 event.private_data = cm_node->pdata_buf; 183 event.private_data_len = (u8)cm_node->pdata.size; 184 event.ird = cm_node->ird_size; 185 break; 186 case IW_CM_EVENT_CONNECT_REPLY: 187 irdma_get_cmevent_info(cm_node, cm_id, &event); 188 break; 189 case IW_CM_EVENT_ESTABLISHED: 190 event.ird = cm_node->ird_size; 191 event.ord = cm_node->ord_size; 192 break; 193 case IW_CM_EVENT_DISCONNECT: 194 case IW_CM_EVENT_CLOSE: 195 /* Wait if we are in RTS but havent issued the iwcm event upcall */ 196 if (!cm_node->accelerated) 197 wait_for_completion(&cm_node->establish_comp); 198 break; 199 default: 200 return -EINVAL; 201 } 202 203 return cm_id->event_handler(cm_id, &event); 204 } 205 206 /** 207 * irdma_timer_list_prep - add connection nodes to a list to perform timer tasks 208 * @cm_core: cm's core 209 * @timer_list: a timer list to which cm_node will be selected 210 */ 211 static void irdma_timer_list_prep(struct irdma_cm_core *cm_core, 212 struct list_head *timer_list) 213 { 214 struct irdma_cm_node *cm_node; 215 int bkt; 216 217 hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) { 218 if ((cm_node->close_entry || cm_node->send_entry) && 219 refcount_inc_not_zero(&cm_node->refcnt)) 220 list_add(&cm_node->timer_entry, timer_list); 221 } 222 } 223 224 /** 225 * irdma_create_event - create cm event 226 * @cm_node: connection's node 227 * @type: Event type to generate 228 */ 229 static struct irdma_cm_event *irdma_create_event(struct irdma_cm_node *cm_node, 230 enum irdma_cm_event_type type) 231 { 232 struct irdma_cm_event *event; 233 234 if (!cm_node->cm_id) 235 return NULL; 236 237 event = kzalloc(sizeof(*event), GFP_ATOMIC); 238 239 if (!event) 240 return NULL; 241 242 event->type = type; 243 event->cm_node = cm_node; 244 memcpy(event->cm_info.rem_addr, cm_node->rem_addr, 245 sizeof(event->cm_info.rem_addr)); 246 memcpy(event->cm_info.loc_addr, cm_node->loc_addr, 247 sizeof(event->cm_info.loc_addr)); 248 event->cm_info.rem_port = cm_node->rem_port; 249 event->cm_info.loc_port = cm_node->loc_port; 250 event->cm_info.cm_id = cm_node->cm_id; 251 ibdev_dbg(&cm_node->iwdev->ibdev, 252 "CM: node=%p event=%p type=%u dst=%pI4 src=%pI4\n", cm_node, 253 event, type, event->cm_info.loc_addr, 254 event->cm_info.rem_addr); 255 trace_irdma_create_event(cm_node, type, __builtin_return_address(0)); 256 irdma_cm_post_event(event); 257 258 return event; 259 } 260 261 /** 262 * irdma_free_retrans_entry - free send entry 263 * @cm_node: connection's node 264 */ 265 static void irdma_free_retrans_entry(struct irdma_cm_node *cm_node) 266 { 267 struct irdma_device *iwdev = cm_node->iwdev; 268 struct irdma_timer_entry *send_entry; 269 270 send_entry = cm_node->send_entry; 271 if (!send_entry) 272 return; 273 274 cm_node->send_entry = NULL; 275 irdma_free_sqbuf(&iwdev->vsi, send_entry->sqbuf); 276 kfree(send_entry); 277 refcount_dec(&cm_node->refcnt); 278 } 279 280 /** 281 * irdma_cleanup_retrans_entry - free send entry with lock 282 * @cm_node: connection's node 283 */ 284 static void irdma_cleanup_retrans_entry(struct irdma_cm_node *cm_node) 285 { 286 unsigned long flags; 287 288 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 289 irdma_free_retrans_entry(cm_node); 290 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags); 291 } 292 293 /** 294 * irdma_form_ah_cm_frame - get a free packet and build frame with address handle 295 * @cm_node: connection's node ionfo to use in frame 296 * @options: pointer to options info 297 * @hdr: pointer mpa header 298 * @pdata: pointer to private data 299 * @flags: indicates FIN or ACK 300 */ 301 static struct irdma_puda_buf *irdma_form_ah_cm_frame(struct irdma_cm_node *cm_node, 302 struct irdma_kmem_info *options, 303 struct irdma_kmem_info *hdr, 304 struct irdma_mpa_priv_info *pdata, 305 u8 flags) 306 { 307 struct irdma_puda_buf *sqbuf; 308 struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi; 309 u8 *buf; 310 struct tcphdr *tcph; 311 u16 pktsize; 312 u32 opts_len = 0; 313 u32 pd_len = 0; 314 u32 hdr_len = 0; 315 316 if (!cm_node->ah || !cm_node->ah->ah_info.ah_valid) { 317 ibdev_dbg(&cm_node->iwdev->ibdev, "CM: AH invalid\n"); 318 return NULL; 319 } 320 321 sqbuf = irdma_puda_get_bufpool(vsi->ilq); 322 if (!sqbuf) { 323 ibdev_dbg(&cm_node->iwdev->ibdev, "CM: SQ buf NULL\n"); 324 return NULL; 325 } 326 327 sqbuf->ah_id = cm_node->ah->ah_info.ah_idx; 328 buf = sqbuf->mem.va; 329 if (options) 330 opts_len = (u32)options->size; 331 332 if (hdr) 333 hdr_len = hdr->size; 334 335 if (pdata) 336 pd_len = pdata->size; 337 338 pktsize = sizeof(*tcph) + opts_len + hdr_len + pd_len; 339 340 memset(buf, 0, sizeof(*tcph)); 341 342 sqbuf->totallen = pktsize; 343 sqbuf->tcphlen = sizeof(*tcph) + opts_len; 344 sqbuf->scratch = cm_node; 345 346 tcph = (struct tcphdr *)buf; 347 buf += sizeof(*tcph); 348 349 tcph->source = htons(cm_node->loc_port); 350 tcph->dest = htons(cm_node->rem_port); 351 tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num); 352 353 if (flags & SET_ACK) { 354 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt; 355 tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num); 356 tcph->ack = 1; 357 } else { 358 tcph->ack_seq = 0; 359 } 360 361 if (flags & SET_SYN) { 362 cm_node->tcp_cntxt.loc_seq_num++; 363 tcph->syn = 1; 364 } else { 365 cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len; 366 } 367 368 if (flags & SET_FIN) { 369 cm_node->tcp_cntxt.loc_seq_num++; 370 tcph->fin = 1; 371 } 372 373 if (flags & SET_RST) 374 tcph->rst = 1; 375 376 tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2); 377 sqbuf->tcphlen = tcph->doff << 2; 378 tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd); 379 tcph->urg_ptr = 0; 380 381 if (opts_len) { 382 memcpy(buf, options->addr, opts_len); 383 buf += opts_len; 384 } 385 386 if (hdr_len) { 387 memcpy(buf, hdr->addr, hdr_len); 388 buf += hdr_len; 389 } 390 391 if (pdata && pdata->addr) 392 memcpy(buf, pdata->addr, pdata->size); 393 394 refcount_set(&sqbuf->refcount, 1); 395 396 print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET, 397 16, 8, sqbuf->mem.va, sqbuf->totallen, false); 398 399 return sqbuf; 400 } 401 402 /** 403 * irdma_form_uda_cm_frame - get a free packet and build frame full tcpip packet 404 * @cm_node: connection's node ionfo to use in frame 405 * @options: pointer to options info 406 * @hdr: pointer mpa header 407 * @pdata: pointer to private data 408 * @flags: indicates FIN or ACK 409 */ 410 static struct irdma_puda_buf *irdma_form_uda_cm_frame(struct irdma_cm_node *cm_node, 411 struct irdma_kmem_info *options, 412 struct irdma_kmem_info *hdr, 413 struct irdma_mpa_priv_info *pdata, 414 u8 flags) 415 { 416 struct irdma_puda_buf *sqbuf; 417 struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi; 418 u8 *buf; 419 420 struct tcphdr *tcph; 421 struct iphdr *iph; 422 struct ipv6hdr *ip6h; 423 struct ethhdr *ethh; 424 u16 pktsize; 425 u16 eth_hlen = ETH_HLEN; 426 u32 opts_len = 0; 427 u32 pd_len = 0; 428 u32 hdr_len = 0; 429 430 u16 vtag; 431 432 sqbuf = irdma_puda_get_bufpool(vsi->ilq); 433 if (!sqbuf) 434 return NULL; 435 436 buf = sqbuf->mem.va; 437 438 if (options) 439 opts_len = (u32)options->size; 440 441 if (hdr) 442 hdr_len = hdr->size; 443 444 if (pdata) 445 pd_len = pdata->size; 446 447 if (cm_node->vlan_id < VLAN_N_VID) 448 eth_hlen += 4; 449 450 if (cm_node->ipv4) 451 pktsize = sizeof(*iph) + sizeof(*tcph); 452 else 453 pktsize = sizeof(*ip6h) + sizeof(*tcph); 454 pktsize += opts_len + hdr_len + pd_len; 455 456 memset(buf, 0, eth_hlen + pktsize); 457 458 sqbuf->totallen = pktsize + eth_hlen; 459 sqbuf->maclen = eth_hlen; 460 sqbuf->tcphlen = sizeof(*tcph) + opts_len; 461 sqbuf->scratch = cm_node; 462 463 ethh = (struct ethhdr *)buf; 464 buf += eth_hlen; 465 466 if (cm_node->do_lpb) 467 sqbuf->do_lpb = true; 468 469 if (cm_node->ipv4) { 470 sqbuf->ipv4 = true; 471 472 iph = (struct iphdr *)buf; 473 buf += sizeof(*iph); 474 tcph = (struct tcphdr *)buf; 475 buf += sizeof(*tcph); 476 477 ether_addr_copy(ethh->h_dest, cm_node->rem_mac); 478 ether_addr_copy(ethh->h_source, cm_node->loc_mac); 479 if (cm_node->vlan_id < VLAN_N_VID) { 480 ((struct vlan_ethhdr *)ethh)->h_vlan_proto = 481 htons(ETH_P_8021Q); 482 vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) | 483 cm_node->vlan_id; 484 ((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag); 485 486 ((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto = 487 htons(ETH_P_IP); 488 } else { 489 ethh->h_proto = htons(ETH_P_IP); 490 } 491 492 iph->version = IPVERSION; 493 iph->ihl = 5; /* 5 * 4Byte words, IP headr len */ 494 iph->tos = cm_node->tos; 495 iph->tot_len = htons(pktsize); 496 iph->id = htons(++cm_node->tcp_cntxt.loc_id); 497 498 iph->frag_off = htons(0x4000); 499 iph->ttl = 0x40; 500 iph->protocol = IPPROTO_TCP; 501 iph->saddr = htonl(cm_node->loc_addr[0]); 502 iph->daddr = htonl(cm_node->rem_addr[0]); 503 } else { 504 sqbuf->ipv4 = false; 505 ip6h = (struct ipv6hdr *)buf; 506 buf += sizeof(*ip6h); 507 tcph = (struct tcphdr *)buf; 508 buf += sizeof(*tcph); 509 510 ether_addr_copy(ethh->h_dest, cm_node->rem_mac); 511 ether_addr_copy(ethh->h_source, cm_node->loc_mac); 512 if (cm_node->vlan_id < VLAN_N_VID) { 513 ((struct vlan_ethhdr *)ethh)->h_vlan_proto = 514 htons(ETH_P_8021Q); 515 vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) | 516 cm_node->vlan_id; 517 ((struct vlan_ethhdr *)ethh)->h_vlan_TCI = htons(vtag); 518 ((struct vlan_ethhdr *)ethh)->h_vlan_encapsulated_proto = 519 htons(ETH_P_IPV6); 520 } else { 521 ethh->h_proto = htons(ETH_P_IPV6); 522 } 523 ip6h->version = 6; 524 ip6h->priority = cm_node->tos >> 4; 525 ip6h->flow_lbl[0] = cm_node->tos << 4; 526 ip6h->flow_lbl[1] = 0; 527 ip6h->flow_lbl[2] = 0; 528 ip6h->payload_len = htons(pktsize - sizeof(*ip6h)); 529 ip6h->nexthdr = 6; 530 ip6h->hop_limit = 128; 531 irdma_copy_ip_htonl(ip6h->saddr.in6_u.u6_addr32, 532 cm_node->loc_addr); 533 irdma_copy_ip_htonl(ip6h->daddr.in6_u.u6_addr32, 534 cm_node->rem_addr); 535 } 536 537 tcph->source = htons(cm_node->loc_port); 538 tcph->dest = htons(cm_node->rem_port); 539 tcph->seq = htonl(cm_node->tcp_cntxt.loc_seq_num); 540 541 if (flags & SET_ACK) { 542 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt; 543 tcph->ack_seq = htonl(cm_node->tcp_cntxt.loc_ack_num); 544 tcph->ack = 1; 545 } else { 546 tcph->ack_seq = 0; 547 } 548 549 if (flags & SET_SYN) { 550 cm_node->tcp_cntxt.loc_seq_num++; 551 tcph->syn = 1; 552 } else { 553 cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len; 554 } 555 556 if (flags & SET_FIN) { 557 cm_node->tcp_cntxt.loc_seq_num++; 558 tcph->fin = 1; 559 } 560 561 if (flags & SET_RST) 562 tcph->rst = 1; 563 564 tcph->doff = (u16)((sizeof(*tcph) + opts_len + 3) >> 2); 565 sqbuf->tcphlen = tcph->doff << 2; 566 tcph->window = htons(cm_node->tcp_cntxt.rcv_wnd); 567 tcph->urg_ptr = 0; 568 569 if (opts_len) { 570 memcpy(buf, options->addr, opts_len); 571 buf += opts_len; 572 } 573 574 if (hdr_len) { 575 memcpy(buf, hdr->addr, hdr_len); 576 buf += hdr_len; 577 } 578 579 if (pdata && pdata->addr) 580 memcpy(buf, pdata->addr, pdata->size); 581 582 refcount_set(&sqbuf->refcount, 1); 583 584 print_hex_dump_debug("ILQ: TRANSMIT ILQ BUFFER", DUMP_PREFIX_OFFSET, 585 16, 8, sqbuf->mem.va, sqbuf->totallen, false); 586 return sqbuf; 587 } 588 589 /** 590 * irdma_send_reset - Send RST packet 591 * @cm_node: connection's node 592 */ 593 int irdma_send_reset(struct irdma_cm_node *cm_node) 594 { 595 struct irdma_puda_buf *sqbuf; 596 int flags = SET_RST | SET_ACK; 597 598 trace_irdma_send_reset(cm_node, 0, __builtin_return_address(0)); 599 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL, 600 flags); 601 if (!sqbuf) 602 return -ENOMEM; 603 604 ibdev_dbg(&cm_node->iwdev->ibdev, 605 "CM: caller: %pS cm_node %p cm_id=%p accel=%d state=%d rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n", 606 __builtin_return_address(0), cm_node, cm_node->cm_id, 607 cm_node->accelerated, cm_node->state, cm_node->rem_port, 608 cm_node->loc_port, cm_node->rem_addr, cm_node->loc_addr); 609 610 return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 0, 611 1); 612 } 613 614 /** 615 * irdma_active_open_err - send event for active side cm error 616 * @cm_node: connection's node 617 * @reset: Flag to send reset or not 618 */ 619 static void irdma_active_open_err(struct irdma_cm_node *cm_node, bool reset) 620 { 621 trace_irdma_active_open_err(cm_node, reset, 622 __builtin_return_address(0)); 623 irdma_cleanup_retrans_entry(cm_node); 624 cm_node->cm_core->stats_connect_errs++; 625 if (reset) { 626 ibdev_dbg(&cm_node->iwdev->ibdev, 627 "CM: cm_node=%p state=%d\n", cm_node, 628 cm_node->state); 629 refcount_inc(&cm_node->refcnt); 630 irdma_send_reset(cm_node); 631 } 632 633 cm_node->state = IRDMA_CM_STATE_CLOSED; 634 irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED); 635 } 636 637 /** 638 * irdma_passive_open_err - handle passive side cm error 639 * @cm_node: connection's node 640 * @reset: send reset or just free cm_node 641 */ 642 static void irdma_passive_open_err(struct irdma_cm_node *cm_node, bool reset) 643 { 644 irdma_cleanup_retrans_entry(cm_node); 645 cm_node->cm_core->stats_passive_errs++; 646 cm_node->state = IRDMA_CM_STATE_CLOSED; 647 ibdev_dbg(&cm_node->iwdev->ibdev, "CM: cm_node=%p state =%d\n", 648 cm_node, cm_node->state); 649 trace_irdma_passive_open_err(cm_node, reset, 650 __builtin_return_address(0)); 651 if (reset) 652 irdma_send_reset(cm_node); 653 else 654 irdma_rem_ref_cm_node(cm_node); 655 } 656 657 /** 658 * irdma_event_connect_error - to create connect error event 659 * @event: cm information for connect event 660 */ 661 static void irdma_event_connect_error(struct irdma_cm_event *event) 662 { 663 struct irdma_qp *iwqp; 664 struct iw_cm_id *cm_id; 665 666 cm_id = event->cm_node->cm_id; 667 if (!cm_id) 668 return; 669 670 iwqp = cm_id->provider_data; 671 672 if (!iwqp || !iwqp->iwdev) 673 return; 674 675 iwqp->cm_id = NULL; 676 cm_id->provider_data = NULL; 677 irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 678 -ECONNRESET); 679 irdma_rem_ref_cm_node(event->cm_node); 680 } 681 682 /** 683 * irdma_process_options - process options from TCP header 684 * @cm_node: connection's node 685 * @optionsloc: point to start of options 686 * @optionsize: size of all options 687 * @syn_pkt: flag if syn packet 688 */ 689 static int irdma_process_options(struct irdma_cm_node *cm_node, u8 *optionsloc, 690 u32 optionsize, u32 syn_pkt) 691 { 692 u32 tmp; 693 u32 offset = 0; 694 union all_known_options *all_options; 695 char got_mss_option = 0; 696 697 while (offset < optionsize) { 698 all_options = (union all_known_options *)(optionsloc + offset); 699 switch (all_options->base.optionnum) { 700 case OPTION_NUM_EOL: 701 offset = optionsize; 702 break; 703 case OPTION_NUM_NONE: 704 offset += 1; 705 continue; 706 case OPTION_NUM_MSS: 707 ibdev_dbg(&cm_node->iwdev->ibdev, 708 "CM: MSS Length: %d Offset: %d Size: %d\n", 709 all_options->mss.len, offset, optionsize); 710 got_mss_option = 1; 711 if (all_options->mss.len != 4) 712 return -EINVAL; 713 tmp = ntohs(all_options->mss.mss); 714 if ((cm_node->ipv4 && 715 (tmp + IRDMA_MTU_TO_MSS_IPV4) < IRDMA_MIN_MTU_IPV4) || 716 (!cm_node->ipv4 && 717 (tmp + IRDMA_MTU_TO_MSS_IPV6) < IRDMA_MIN_MTU_IPV6)) 718 return -EINVAL; 719 if (tmp < cm_node->tcp_cntxt.mss) 720 cm_node->tcp_cntxt.mss = tmp; 721 break; 722 case OPTION_NUM_WINDOW_SCALE: 723 cm_node->tcp_cntxt.snd_wscale = 724 all_options->windowscale.shiftcount; 725 break; 726 default: 727 ibdev_dbg(&cm_node->iwdev->ibdev, 728 "CM: Unsupported TCP Option: %x\n", 729 all_options->base.optionnum); 730 break; 731 } 732 offset += all_options->base.len; 733 } 734 if (!got_mss_option && syn_pkt) 735 cm_node->tcp_cntxt.mss = IRDMA_CM_DEFAULT_MSS; 736 737 return 0; 738 } 739 740 /** 741 * irdma_handle_tcp_options - setup TCP context info after parsing TCP options 742 * @cm_node: connection's node 743 * @tcph: pointer tcp header 744 * @optionsize: size of options rcvd 745 * @passive: active or passive flag 746 */ 747 static int irdma_handle_tcp_options(struct irdma_cm_node *cm_node, 748 struct tcphdr *tcph, int optionsize, 749 int passive) 750 { 751 u8 *optionsloc = (u8 *)&tcph[1]; 752 int ret; 753 754 if (optionsize) { 755 ret = irdma_process_options(cm_node, optionsloc, optionsize, 756 (u32)tcph->syn); 757 if (ret) { 758 ibdev_dbg(&cm_node->iwdev->ibdev, 759 "CM: Node %p, Sending Reset\n", cm_node); 760 if (passive) 761 irdma_passive_open_err(cm_node, true); 762 else 763 irdma_active_open_err(cm_node, true); 764 return ret; 765 } 766 } 767 768 cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->window) 769 << cm_node->tcp_cntxt.snd_wscale; 770 771 if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd) 772 cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd; 773 774 return 0; 775 } 776 777 /** 778 * irdma_build_mpa_v1 - build a MPA V1 frame 779 * @cm_node: connection's node 780 * @start_addr: address where to build frame 781 * @mpa_key: to do read0 or write0 782 */ 783 static void irdma_build_mpa_v1(struct irdma_cm_node *cm_node, void *start_addr, 784 u8 mpa_key) 785 { 786 struct ietf_mpa_v1 *mpa_frame = start_addr; 787 788 switch (mpa_key) { 789 case MPA_KEY_REQUEST: 790 memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE); 791 break; 792 case MPA_KEY_REPLY: 793 memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE); 794 break; 795 default: 796 break; 797 } 798 mpa_frame->flags = IETF_MPA_FLAGS_CRC; 799 mpa_frame->rev = cm_node->mpa_frame_rev; 800 mpa_frame->priv_data_len = htons(cm_node->pdata.size); 801 } 802 803 /** 804 * irdma_build_mpa_v2 - build a MPA V2 frame 805 * @cm_node: connection's node 806 * @start_addr: buffer start address 807 * @mpa_key: to do read0 or write0 808 */ 809 static void irdma_build_mpa_v2(struct irdma_cm_node *cm_node, void *start_addr, 810 u8 mpa_key) 811 { 812 struct ietf_mpa_v2 *mpa_frame = start_addr; 813 struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg; 814 u16 ctrl_ird, ctrl_ord; 815 816 /* initialize the upper 5 bytes of the frame */ 817 irdma_build_mpa_v1(cm_node, start_addr, mpa_key); 818 mpa_frame->flags |= IETF_MPA_V2_FLAG; 819 if (cm_node->iwdev->iw_ooo) { 820 mpa_frame->flags |= IETF_MPA_FLAGS_MARKERS; 821 cm_node->rcv_mark_en = true; 822 } 823 mpa_frame->priv_data_len = cpu_to_be16(be16_to_cpu(mpa_frame->priv_data_len) + 824 IETF_RTR_MSG_SIZE); 825 826 /* initialize RTR msg */ 827 if (cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) { 828 ctrl_ird = IETF_NO_IRD_ORD; 829 ctrl_ord = IETF_NO_IRD_ORD; 830 } else { 831 ctrl_ird = (cm_node->ird_size > IETF_NO_IRD_ORD) ? 832 IETF_NO_IRD_ORD : 833 cm_node->ird_size; 834 ctrl_ord = (cm_node->ord_size > IETF_NO_IRD_ORD) ? 835 IETF_NO_IRD_ORD : 836 cm_node->ord_size; 837 } 838 ctrl_ird |= IETF_PEER_TO_PEER; 839 840 switch (mpa_key) { 841 case MPA_KEY_REQUEST: 842 ctrl_ord |= IETF_RDMA0_WRITE; 843 ctrl_ord |= IETF_RDMA0_READ; 844 break; 845 case MPA_KEY_REPLY: 846 switch (cm_node->send_rdma0_op) { 847 case SEND_RDMA_WRITE_ZERO: 848 ctrl_ord |= IETF_RDMA0_WRITE; 849 break; 850 case SEND_RDMA_READ_ZERO: 851 ctrl_ord |= IETF_RDMA0_READ; 852 break; 853 } 854 break; 855 default: 856 break; 857 } 858 rtr_msg->ctrl_ird = htons(ctrl_ird); 859 rtr_msg->ctrl_ord = htons(ctrl_ord); 860 } 861 862 /** 863 * irdma_cm_build_mpa_frame - build mpa frame for mpa version 1 or version 2 864 * @cm_node: connection's node 865 * @mpa: mpa: data buffer 866 * @mpa_key: to do read0 or write0 867 */ 868 static int irdma_cm_build_mpa_frame(struct irdma_cm_node *cm_node, 869 struct irdma_kmem_info *mpa, u8 mpa_key) 870 { 871 int hdr_len = 0; 872 873 switch (cm_node->mpa_frame_rev) { 874 case IETF_MPA_V1: 875 hdr_len = sizeof(struct ietf_mpa_v1); 876 irdma_build_mpa_v1(cm_node, mpa->addr, mpa_key); 877 break; 878 case IETF_MPA_V2: 879 hdr_len = sizeof(struct ietf_mpa_v2); 880 irdma_build_mpa_v2(cm_node, mpa->addr, mpa_key); 881 break; 882 default: 883 break; 884 } 885 886 return hdr_len; 887 } 888 889 /** 890 * irdma_send_mpa_request - active node send mpa request to passive node 891 * @cm_node: connection's node 892 */ 893 static int irdma_send_mpa_request(struct irdma_cm_node *cm_node) 894 { 895 struct irdma_puda_buf *sqbuf; 896 897 cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame; 898 cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node, 899 &cm_node->mpa_hdr, 900 MPA_KEY_REQUEST); 901 if (!cm_node->mpa_hdr.size) { 902 ibdev_dbg(&cm_node->iwdev->ibdev, 903 "CM: mpa size = %d\n", cm_node->mpa_hdr.size); 904 return -EINVAL; 905 } 906 907 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, 908 &cm_node->mpa_hdr, 909 &cm_node->pdata, SET_ACK); 910 if (!sqbuf) 911 return -ENOMEM; 912 913 return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1, 914 0); 915 } 916 917 /** 918 * irdma_send_mpa_reject - 919 * @cm_node: connection's node 920 * @pdata: reject data for connection 921 * @plen: length of reject data 922 */ 923 static int irdma_send_mpa_reject(struct irdma_cm_node *cm_node, 924 const void *pdata, u8 plen) 925 { 926 struct irdma_puda_buf *sqbuf; 927 struct irdma_mpa_priv_info priv_info; 928 929 cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame; 930 cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node, 931 &cm_node->mpa_hdr, 932 MPA_KEY_REPLY); 933 934 cm_node->mpa_frame.flags |= IETF_MPA_FLAGS_REJECT; 935 priv_info.addr = pdata; 936 priv_info.size = plen; 937 938 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, 939 &cm_node->mpa_hdr, &priv_info, 940 SET_ACK | SET_FIN); 941 if (!sqbuf) 942 return -ENOMEM; 943 944 cm_node->state = IRDMA_CM_STATE_FIN_WAIT1; 945 946 return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1, 947 0); 948 } 949 950 /** 951 * irdma_negotiate_mpa_v2_ird_ord - negotiate MPAv2 IRD/ORD 952 * @cm_node: connection's node 953 * @buf: Data pointer 954 */ 955 static int irdma_negotiate_mpa_v2_ird_ord(struct irdma_cm_node *cm_node, 956 u8 *buf) 957 { 958 struct ietf_mpa_v2 *mpa_v2_frame; 959 struct ietf_rtr_msg *rtr_msg; 960 u16 ird_size; 961 u16 ord_size; 962 u16 ctrl_ord; 963 u16 ctrl_ird; 964 965 mpa_v2_frame = (struct ietf_mpa_v2 *)buf; 966 rtr_msg = &mpa_v2_frame->rtr_msg; 967 968 /* parse rtr message */ 969 ctrl_ord = ntohs(rtr_msg->ctrl_ord); 970 ctrl_ird = ntohs(rtr_msg->ctrl_ird); 971 ird_size = ctrl_ird & IETF_NO_IRD_ORD; 972 ord_size = ctrl_ord & IETF_NO_IRD_ORD; 973 974 if (!(ctrl_ird & IETF_PEER_TO_PEER)) 975 return -EOPNOTSUPP; 976 977 if (ird_size == IETF_NO_IRD_ORD || ord_size == IETF_NO_IRD_ORD) { 978 cm_node->mpav2_ird_ord = IETF_NO_IRD_ORD; 979 goto negotiate_done; 980 } 981 982 if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) { 983 /* responder */ 984 if (!ord_size && (ctrl_ord & IETF_RDMA0_READ)) 985 cm_node->ird_size = 1; 986 if (cm_node->ord_size > ird_size) 987 cm_node->ord_size = ird_size; 988 } else { 989 /* initiator */ 990 if (!ird_size && (ctrl_ord & IETF_RDMA0_READ)) 991 /* Remote peer doesn't support RDMA0_READ */ 992 return -EOPNOTSUPP; 993 994 if (cm_node->ord_size > ird_size) 995 cm_node->ord_size = ird_size; 996 997 if (cm_node->ird_size < ord_size) 998 /* no resources available */ 999 return -EINVAL; 1000 } 1001 1002 negotiate_done: 1003 if (ctrl_ord & IETF_RDMA0_READ) 1004 cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO; 1005 else if (ctrl_ord & IETF_RDMA0_WRITE) 1006 cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO; 1007 else 1008 /* Not supported RDMA0 operation */ 1009 return -EOPNOTSUPP; 1010 1011 ibdev_dbg(&cm_node->iwdev->ibdev, 1012 "CM: MPAV2 Negotiated ORD: %d, IRD: %d\n", 1013 cm_node->ord_size, cm_node->ird_size); 1014 trace_irdma_negotiate_mpa_v2(cm_node); 1015 return 0; 1016 } 1017 1018 /** 1019 * irdma_parse_mpa - process an IETF MPA frame 1020 * @cm_node: connection's node 1021 * @buf: Data pointer 1022 * @type: to return accept or reject 1023 * @len: Len of mpa buffer 1024 */ 1025 static int irdma_parse_mpa(struct irdma_cm_node *cm_node, u8 *buf, u32 *type, 1026 u32 len) 1027 { 1028 struct ietf_mpa_v1 *mpa_frame; 1029 int mpa_hdr_len, priv_data_len, ret; 1030 1031 *type = IRDMA_MPA_REQUEST_ACCEPT; 1032 1033 if (len < sizeof(struct ietf_mpa_v1)) { 1034 ibdev_dbg(&cm_node->iwdev->ibdev, 1035 "CM: ietf buffer small (%x)\n", len); 1036 return -EINVAL; 1037 } 1038 1039 mpa_frame = (struct ietf_mpa_v1 *)buf; 1040 mpa_hdr_len = sizeof(struct ietf_mpa_v1); 1041 priv_data_len = ntohs(mpa_frame->priv_data_len); 1042 1043 if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) { 1044 ibdev_dbg(&cm_node->iwdev->ibdev, 1045 "CM: private_data too big %d\n", priv_data_len); 1046 return -EOVERFLOW; 1047 } 1048 1049 if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) { 1050 ibdev_dbg(&cm_node->iwdev->ibdev, 1051 "CM: unsupported mpa rev = %d\n", mpa_frame->rev); 1052 return -EINVAL; 1053 } 1054 1055 if (mpa_frame->rev > cm_node->mpa_frame_rev) { 1056 ibdev_dbg(&cm_node->iwdev->ibdev, "CM: rev %d\n", 1057 mpa_frame->rev); 1058 return -EINVAL; 1059 } 1060 1061 cm_node->mpa_frame_rev = mpa_frame->rev; 1062 if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) { 1063 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ, 1064 IETF_MPA_KEY_SIZE)) { 1065 ibdev_dbg(&cm_node->iwdev->ibdev, 1066 "CM: Unexpected MPA Key received\n"); 1067 return -EINVAL; 1068 } 1069 } else { 1070 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REP, 1071 IETF_MPA_KEY_SIZE)) { 1072 ibdev_dbg(&cm_node->iwdev->ibdev, 1073 "CM: Unexpected MPA Key received\n"); 1074 return -EINVAL; 1075 } 1076 } 1077 1078 if (priv_data_len + mpa_hdr_len > len) { 1079 ibdev_dbg(&cm_node->iwdev->ibdev, 1080 "CM: ietf buffer len(%x + %x != %x)\n", 1081 priv_data_len, mpa_hdr_len, len); 1082 return -EOVERFLOW; 1083 } 1084 1085 if (len > IRDMA_MAX_CM_BUF) { 1086 ibdev_dbg(&cm_node->iwdev->ibdev, 1087 "CM: ietf buffer large len = %d\n", len); 1088 return -EOVERFLOW; 1089 } 1090 1091 switch (mpa_frame->rev) { 1092 case IETF_MPA_V2: 1093 mpa_hdr_len += IETF_RTR_MSG_SIZE; 1094 ret = irdma_negotiate_mpa_v2_ird_ord(cm_node, buf); 1095 if (ret) 1096 return ret; 1097 break; 1098 case IETF_MPA_V1: 1099 default: 1100 break; 1101 } 1102 1103 memcpy(cm_node->pdata_buf, buf + mpa_hdr_len, priv_data_len); 1104 cm_node->pdata.size = priv_data_len; 1105 1106 if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT) 1107 *type = IRDMA_MPA_REQUEST_REJECT; 1108 1109 if (mpa_frame->flags & IETF_MPA_FLAGS_MARKERS) 1110 cm_node->snd_mark_en = true; 1111 1112 return 0; 1113 } 1114 1115 /** 1116 * irdma_schedule_cm_timer 1117 * @cm_node: connection's node 1118 * @sqbuf: buffer to send 1119 * @type: if it is send or close 1120 * @send_retrans: if rexmits to be done 1121 * @close_when_complete: is cm_node to be removed 1122 * 1123 * note - cm_node needs to be protected before calling this. Encase in: 1124 * irdma_rem_ref_cm_node(cm_core, cm_node); 1125 * irdma_schedule_cm_timer(...) 1126 * refcount_inc(&cm_node->refcnt); 1127 */ 1128 int irdma_schedule_cm_timer(struct irdma_cm_node *cm_node, 1129 struct irdma_puda_buf *sqbuf, 1130 enum irdma_timer_type type, int send_retrans, 1131 int close_when_complete) 1132 { 1133 struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi; 1134 struct irdma_cm_core *cm_core = cm_node->cm_core; 1135 struct irdma_timer_entry *new_send; 1136 u32 was_timer_set; 1137 unsigned long flags; 1138 1139 new_send = kzalloc(sizeof(*new_send), GFP_ATOMIC); 1140 if (!new_send) { 1141 if (type != IRDMA_TIMER_TYPE_CLOSE) 1142 irdma_free_sqbuf(vsi, sqbuf); 1143 return -ENOMEM; 1144 } 1145 1146 new_send->retrycount = IRDMA_DEFAULT_RETRYS; 1147 new_send->retranscount = IRDMA_DEFAULT_RETRANS; 1148 new_send->sqbuf = sqbuf; 1149 new_send->timetosend = jiffies; 1150 new_send->type = type; 1151 new_send->send_retrans = send_retrans; 1152 new_send->close_when_complete = close_when_complete; 1153 1154 if (type == IRDMA_TIMER_TYPE_CLOSE) { 1155 new_send->timetosend += (HZ / 10); 1156 if (cm_node->close_entry) { 1157 kfree(new_send); 1158 ibdev_dbg(&cm_node->iwdev->ibdev, 1159 "CM: already close entry\n"); 1160 return -EINVAL; 1161 } 1162 1163 cm_node->close_entry = new_send; 1164 } else { /* type == IRDMA_TIMER_TYPE_SEND */ 1165 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 1166 cm_node->send_entry = new_send; 1167 refcount_inc(&cm_node->refcnt); 1168 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags); 1169 new_send->timetosend = jiffies + IRDMA_RETRY_TIMEOUT; 1170 1171 refcount_inc(&sqbuf->refcount); 1172 irdma_puda_send_buf(vsi->ilq, sqbuf); 1173 if (!send_retrans) { 1174 irdma_cleanup_retrans_entry(cm_node); 1175 if (close_when_complete) 1176 irdma_rem_ref_cm_node(cm_node); 1177 return 0; 1178 } 1179 } 1180 1181 spin_lock_irqsave(&cm_core->ht_lock, flags); 1182 was_timer_set = timer_pending(&cm_core->tcp_timer); 1183 1184 if (!was_timer_set) { 1185 cm_core->tcp_timer.expires = new_send->timetosend; 1186 add_timer(&cm_core->tcp_timer); 1187 } 1188 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 1189 1190 return 0; 1191 } 1192 1193 /** 1194 * irdma_retrans_expired - Could not rexmit the packet 1195 * @cm_node: connection's node 1196 */ 1197 static void irdma_retrans_expired(struct irdma_cm_node *cm_node) 1198 { 1199 enum irdma_cm_node_state state = cm_node->state; 1200 1201 cm_node->state = IRDMA_CM_STATE_CLOSED; 1202 switch (state) { 1203 case IRDMA_CM_STATE_SYN_RCVD: 1204 case IRDMA_CM_STATE_CLOSING: 1205 irdma_rem_ref_cm_node(cm_node); 1206 break; 1207 case IRDMA_CM_STATE_FIN_WAIT1: 1208 case IRDMA_CM_STATE_LAST_ACK: 1209 irdma_send_reset(cm_node); 1210 break; 1211 default: 1212 refcount_inc(&cm_node->refcnt); 1213 irdma_send_reset(cm_node); 1214 irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED); 1215 break; 1216 } 1217 } 1218 1219 /** 1220 * irdma_handle_close_entry - for handling retry/timeouts 1221 * @cm_node: connection's node 1222 * @rem_node: flag for remove cm_node 1223 */ 1224 static void irdma_handle_close_entry(struct irdma_cm_node *cm_node, 1225 u32 rem_node) 1226 { 1227 struct irdma_timer_entry *close_entry = cm_node->close_entry; 1228 struct irdma_qp *iwqp; 1229 unsigned long flags; 1230 1231 if (!close_entry) 1232 return; 1233 iwqp = (struct irdma_qp *)close_entry->sqbuf; 1234 if (iwqp) { 1235 spin_lock_irqsave(&iwqp->lock, flags); 1236 if (iwqp->cm_id) { 1237 iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED; 1238 iwqp->hw_iwarp_state = IRDMA_QP_STATE_ERROR; 1239 iwqp->last_aeq = IRDMA_AE_RESET_SENT; 1240 iwqp->ibqp_state = IB_QPS_ERR; 1241 spin_unlock_irqrestore(&iwqp->lock, flags); 1242 irdma_cm_disconn(iwqp); 1243 } else { 1244 spin_unlock_irqrestore(&iwqp->lock, flags); 1245 } 1246 } else if (rem_node) { 1247 /* TIME_WAIT state */ 1248 irdma_rem_ref_cm_node(cm_node); 1249 } 1250 1251 kfree(close_entry); 1252 cm_node->close_entry = NULL; 1253 } 1254 1255 /** 1256 * irdma_cm_timer_tick - system's timer expired callback 1257 * @t: Pointer to timer_list 1258 */ 1259 static void irdma_cm_timer_tick(struct timer_list *t) 1260 { 1261 unsigned long nexttimeout = jiffies + IRDMA_LONG_TIME; 1262 struct irdma_cm_node *cm_node; 1263 struct irdma_timer_entry *send_entry, *close_entry; 1264 struct list_head *list_core_temp; 1265 struct list_head *list_node; 1266 struct irdma_cm_core *cm_core = from_timer(cm_core, t, tcp_timer); 1267 struct irdma_sc_vsi *vsi; 1268 u32 settimer = 0; 1269 unsigned long timetosend; 1270 unsigned long flags; 1271 struct list_head timer_list; 1272 1273 INIT_LIST_HEAD(&timer_list); 1274 1275 rcu_read_lock(); 1276 irdma_timer_list_prep(cm_core, &timer_list); 1277 rcu_read_unlock(); 1278 1279 list_for_each_safe (list_node, list_core_temp, &timer_list) { 1280 cm_node = container_of(list_node, struct irdma_cm_node, 1281 timer_entry); 1282 close_entry = cm_node->close_entry; 1283 1284 if (close_entry) { 1285 if (time_after(close_entry->timetosend, jiffies)) { 1286 if (nexttimeout > close_entry->timetosend || 1287 !settimer) { 1288 nexttimeout = close_entry->timetosend; 1289 settimer = 1; 1290 } 1291 } else { 1292 irdma_handle_close_entry(cm_node, 1); 1293 } 1294 } 1295 1296 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 1297 1298 send_entry = cm_node->send_entry; 1299 if (!send_entry) 1300 goto done; 1301 if (time_after(send_entry->timetosend, jiffies)) { 1302 if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) { 1303 if (nexttimeout > send_entry->timetosend || 1304 !settimer) { 1305 nexttimeout = send_entry->timetosend; 1306 settimer = 1; 1307 } 1308 } else { 1309 irdma_free_retrans_entry(cm_node); 1310 } 1311 goto done; 1312 } 1313 1314 if (cm_node->state == IRDMA_CM_STATE_OFFLOADED || 1315 cm_node->state == IRDMA_CM_STATE_CLOSED) { 1316 irdma_free_retrans_entry(cm_node); 1317 goto done; 1318 } 1319 1320 if (!send_entry->retranscount || !send_entry->retrycount) { 1321 irdma_free_retrans_entry(cm_node); 1322 1323 spin_unlock_irqrestore(&cm_node->retrans_list_lock, 1324 flags); 1325 irdma_retrans_expired(cm_node); 1326 cm_node->state = IRDMA_CM_STATE_CLOSED; 1327 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 1328 goto done; 1329 } 1330 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags); 1331 1332 vsi = &cm_node->iwdev->vsi; 1333 if (!cm_node->ack_rcvd) { 1334 refcount_inc(&send_entry->sqbuf->refcount); 1335 irdma_puda_send_buf(vsi->ilq, send_entry->sqbuf); 1336 cm_node->cm_core->stats_pkt_retrans++; 1337 } 1338 1339 spin_lock_irqsave(&cm_node->retrans_list_lock, flags); 1340 if (send_entry->send_retrans) { 1341 send_entry->retranscount--; 1342 timetosend = (IRDMA_RETRY_TIMEOUT << 1343 (IRDMA_DEFAULT_RETRANS - 1344 send_entry->retranscount)); 1345 1346 send_entry->timetosend = jiffies + 1347 min(timetosend, IRDMA_MAX_TIMEOUT); 1348 if (nexttimeout > send_entry->timetosend || !settimer) { 1349 nexttimeout = send_entry->timetosend; 1350 settimer = 1; 1351 } 1352 } else { 1353 int close_when_complete; 1354 1355 close_when_complete = send_entry->close_when_complete; 1356 irdma_free_retrans_entry(cm_node); 1357 if (close_when_complete) 1358 irdma_rem_ref_cm_node(cm_node); 1359 } 1360 done: 1361 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags); 1362 irdma_rem_ref_cm_node(cm_node); 1363 } 1364 1365 if (settimer) { 1366 spin_lock_irqsave(&cm_core->ht_lock, flags); 1367 if (!timer_pending(&cm_core->tcp_timer)) { 1368 cm_core->tcp_timer.expires = nexttimeout; 1369 add_timer(&cm_core->tcp_timer); 1370 } 1371 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 1372 } 1373 } 1374 1375 /** 1376 * irdma_send_syn - send SYN packet 1377 * @cm_node: connection's node 1378 * @sendack: flag to set ACK bit or not 1379 */ 1380 int irdma_send_syn(struct irdma_cm_node *cm_node, u32 sendack) 1381 { 1382 struct irdma_puda_buf *sqbuf; 1383 int flags = SET_SYN; 1384 char optionsbuf[sizeof(struct option_mss) + 1385 sizeof(struct option_windowscale) + 1386 sizeof(struct option_base) + TCP_OPTIONS_PADDING]; 1387 struct irdma_kmem_info opts; 1388 int optionssize = 0; 1389 /* Sending MSS option */ 1390 union all_known_options *options; 1391 1392 opts.addr = optionsbuf; 1393 if (!cm_node) 1394 return -EINVAL; 1395 1396 options = (union all_known_options *)&optionsbuf[optionssize]; 1397 options->mss.optionnum = OPTION_NUM_MSS; 1398 options->mss.len = sizeof(struct option_mss); 1399 options->mss.mss = htons(cm_node->tcp_cntxt.mss); 1400 optionssize += sizeof(struct option_mss); 1401 1402 options = (union all_known_options *)&optionsbuf[optionssize]; 1403 options->windowscale.optionnum = OPTION_NUM_WINDOW_SCALE; 1404 options->windowscale.len = sizeof(struct option_windowscale); 1405 options->windowscale.shiftcount = cm_node->tcp_cntxt.rcv_wscale; 1406 optionssize += sizeof(struct option_windowscale); 1407 options = (union all_known_options *)&optionsbuf[optionssize]; 1408 options->eol = OPTION_NUM_EOL; 1409 optionssize += 1; 1410 1411 if (sendack) 1412 flags |= SET_ACK; 1413 1414 opts.size = optionssize; 1415 1416 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, &opts, NULL, NULL, 1417 flags); 1418 if (!sqbuf) 1419 return -ENOMEM; 1420 1421 return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1, 1422 0); 1423 } 1424 1425 /** 1426 * irdma_send_ack - Send ACK packet 1427 * @cm_node: connection's node 1428 */ 1429 void irdma_send_ack(struct irdma_cm_node *cm_node) 1430 { 1431 struct irdma_puda_buf *sqbuf; 1432 struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi; 1433 1434 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL, 1435 SET_ACK); 1436 if (sqbuf) 1437 irdma_puda_send_buf(vsi->ilq, sqbuf); 1438 } 1439 1440 /** 1441 * irdma_send_fin - Send FIN pkt 1442 * @cm_node: connection's node 1443 */ 1444 static int irdma_send_fin(struct irdma_cm_node *cm_node) 1445 { 1446 struct irdma_puda_buf *sqbuf; 1447 1448 sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL, 1449 SET_ACK | SET_FIN); 1450 if (!sqbuf) 1451 return -ENOMEM; 1452 1453 return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1, 1454 0); 1455 } 1456 1457 /** 1458 * irdma_find_listener - find a cm node listening on this addr-port pair 1459 * @cm_core: cm's core 1460 * @dst_addr: listener ip addr 1461 * @ipv4: flag indicating IPv4 when true 1462 * @dst_port: listener tcp port num 1463 * @vlan_id: virtual LAN ID 1464 * @listener_state: state to match with listen node's 1465 */ 1466 static struct irdma_cm_listener * 1467 irdma_find_listener(struct irdma_cm_core *cm_core, u32 *dst_addr, bool ipv4, 1468 u16 dst_port, u16 vlan_id, 1469 enum irdma_cm_listener_state listener_state) 1470 { 1471 struct irdma_cm_listener *listen_node; 1472 static const u32 ip_zero[4] = { 0, 0, 0, 0 }; 1473 u32 listen_addr[4]; 1474 u16 listen_port; 1475 unsigned long flags; 1476 1477 /* walk list and find cm_node associated with this session ID */ 1478 spin_lock_irqsave(&cm_core->listen_list_lock, flags); 1479 list_for_each_entry (listen_node, &cm_core->listen_list, list) { 1480 memcpy(listen_addr, listen_node->loc_addr, sizeof(listen_addr)); 1481 listen_port = listen_node->loc_port; 1482 if (listen_node->ipv4 != ipv4 || listen_port != dst_port || 1483 !(listener_state & listen_node->listener_state)) 1484 continue; 1485 /* compare node pair, return node handle if a match */ 1486 if (!memcmp(listen_addr, ip_zero, sizeof(listen_addr)) || 1487 (!memcmp(listen_addr, dst_addr, sizeof(listen_addr)) && 1488 vlan_id == listen_node->vlan_id)) { 1489 refcount_inc(&listen_node->refcnt); 1490 spin_unlock_irqrestore(&cm_core->listen_list_lock, 1491 flags); 1492 trace_irdma_find_listener(listen_node); 1493 return listen_node; 1494 } 1495 } 1496 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 1497 1498 return NULL; 1499 } 1500 1501 /** 1502 * irdma_del_multiple_qhash - Remove qhash and child listens 1503 * @iwdev: iWarp device 1504 * @cm_info: CM info for parent listen node 1505 * @cm_parent_listen_node: The parent listen node 1506 */ 1507 static int irdma_del_multiple_qhash(struct irdma_device *iwdev, 1508 struct irdma_cm_info *cm_info, 1509 struct irdma_cm_listener *cm_parent_listen_node) 1510 { 1511 struct irdma_cm_listener *child_listen_node; 1512 struct list_head *pos, *tpos; 1513 unsigned long flags; 1514 int ret = -EINVAL; 1515 1516 spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags); 1517 list_for_each_safe (pos, tpos, 1518 &cm_parent_listen_node->child_listen_list) { 1519 child_listen_node = list_entry(pos, struct irdma_cm_listener, 1520 child_listen_list); 1521 if (child_listen_node->ipv4) 1522 ibdev_dbg(&iwdev->ibdev, 1523 "CM: removing child listen for IP=%pI4, port=%d, vlan=%d\n", 1524 child_listen_node->loc_addr, 1525 child_listen_node->loc_port, 1526 child_listen_node->vlan_id); 1527 else 1528 ibdev_dbg(&iwdev->ibdev, 1529 "CM: removing child listen for IP=%pI6, port=%d, vlan=%d\n", 1530 child_listen_node->loc_addr, 1531 child_listen_node->loc_port, 1532 child_listen_node->vlan_id); 1533 trace_irdma_del_multiple_qhash(child_listen_node); 1534 list_del(pos); 1535 memcpy(cm_info->loc_addr, child_listen_node->loc_addr, 1536 sizeof(cm_info->loc_addr)); 1537 cm_info->vlan_id = child_listen_node->vlan_id; 1538 if (child_listen_node->qhash_set) { 1539 ret = irdma_manage_qhash(iwdev, cm_info, 1540 IRDMA_QHASH_TYPE_TCP_SYN, 1541 IRDMA_QHASH_MANAGE_TYPE_DELETE, 1542 NULL, false); 1543 child_listen_node->qhash_set = false; 1544 } else { 1545 ret = 0; 1546 } 1547 ibdev_dbg(&iwdev->ibdev, 1548 "CM: Child listen node freed = %p\n", 1549 child_listen_node); 1550 kfree(child_listen_node); 1551 cm_parent_listen_node->cm_core->stats_listen_nodes_destroyed++; 1552 } 1553 spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags); 1554 1555 return ret; 1556 } 1557 1558 static u8 irdma_iw_get_vlan_prio(u32 *loc_addr, u8 prio, bool ipv4) 1559 { 1560 struct net_device *ndev = NULL; 1561 1562 rcu_read_lock(); 1563 if (ipv4) { 1564 ndev = ip_dev_find(&init_net, htonl(loc_addr[0])); 1565 } else if (IS_ENABLED(CONFIG_IPV6)) { 1566 struct net_device *ip_dev; 1567 struct in6_addr laddr6; 1568 1569 irdma_copy_ip_htonl(laddr6.in6_u.u6_addr32, loc_addr); 1570 1571 for_each_netdev_rcu (&init_net, ip_dev) { 1572 if (ipv6_chk_addr(&init_net, &laddr6, ip_dev, 1)) { 1573 ndev = ip_dev; 1574 break; 1575 } 1576 } 1577 } 1578 1579 if (!ndev) 1580 goto done; 1581 if (is_vlan_dev(ndev)) 1582 prio = (vlan_dev_get_egress_qos_mask(ndev, prio) & VLAN_PRIO_MASK) 1583 >> VLAN_PRIO_SHIFT; 1584 if (ipv4) 1585 dev_put(ndev); 1586 1587 done: 1588 rcu_read_unlock(); 1589 1590 return prio; 1591 } 1592 1593 /** 1594 * irdma_get_vlan_mac_ipv6 - Gets the vlan and mac 1595 * @addr: local IPv6 address 1596 * @vlan_id: vlan id for the given IPv6 address 1597 * @mac: mac address for the given IPv6 address 1598 * 1599 * Returns the vlan id and mac for an IPv6 address. 1600 */ 1601 void irdma_get_vlan_mac_ipv6(u32 *addr, u16 *vlan_id, u8 *mac) 1602 { 1603 struct net_device *ip_dev = NULL; 1604 struct in6_addr laddr6; 1605 1606 if (!IS_ENABLED(CONFIG_IPV6)) 1607 return; 1608 1609 irdma_copy_ip_htonl(laddr6.in6_u.u6_addr32, addr); 1610 if (vlan_id) 1611 *vlan_id = 0xFFFF; /* Match rdma_vlan_dev_vlan_id() */ 1612 if (mac) 1613 eth_zero_addr(mac); 1614 1615 rcu_read_lock(); 1616 for_each_netdev_rcu (&init_net, ip_dev) { 1617 if (ipv6_chk_addr(&init_net, &laddr6, ip_dev, 1)) { 1618 if (vlan_id) 1619 *vlan_id = rdma_vlan_dev_vlan_id(ip_dev); 1620 if (ip_dev->dev_addr && mac) 1621 ether_addr_copy(mac, ip_dev->dev_addr); 1622 break; 1623 } 1624 } 1625 rcu_read_unlock(); 1626 } 1627 1628 /** 1629 * irdma_get_vlan_ipv4 - Returns the vlan_id for IPv4 address 1630 * @addr: local IPv4 address 1631 */ 1632 u16 irdma_get_vlan_ipv4(u32 *addr) 1633 { 1634 struct net_device *netdev; 1635 u16 vlan_id = 0xFFFF; 1636 1637 netdev = ip_dev_find(&init_net, htonl(addr[0])); 1638 if (netdev) { 1639 vlan_id = rdma_vlan_dev_vlan_id(netdev); 1640 dev_put(netdev); 1641 } 1642 1643 return vlan_id; 1644 } 1645 1646 /** 1647 * irdma_add_mqh_6 - Adds multiple qhashes for IPv6 1648 * @iwdev: iWarp device 1649 * @cm_info: CM info for parent listen node 1650 * @cm_parent_listen_node: The parent listen node 1651 * 1652 * Adds a qhash and a child listen node for every IPv6 address 1653 * on the adapter and adds the associated qhash filter 1654 */ 1655 static int irdma_add_mqh_6(struct irdma_device *iwdev, 1656 struct irdma_cm_info *cm_info, 1657 struct irdma_cm_listener *cm_parent_listen_node) 1658 { 1659 struct net_device *ip_dev; 1660 struct inet6_dev *idev; 1661 struct inet6_ifaddr *ifp, *tmp; 1662 struct irdma_cm_listener *child_listen_node; 1663 unsigned long flags; 1664 int ret = 0; 1665 1666 rtnl_lock(); 1667 for_each_netdev(&init_net, ip_dev) { 1668 if (!(ip_dev->flags & IFF_UP)) 1669 continue; 1670 1671 if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) || 1672 (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) && 1673 ip_dev != iwdev->netdev) 1674 continue; 1675 1676 idev = __in6_dev_get(ip_dev); 1677 if (!idev) { 1678 ibdev_dbg(&iwdev->ibdev, "CM: idev == NULL\n"); 1679 break; 1680 } 1681 list_for_each_entry_safe (ifp, tmp, &idev->addr_list, if_list) { 1682 ibdev_dbg(&iwdev->ibdev, "CM: IP=%pI6, vlan_id=%d, MAC=%pM\n", 1683 &ifp->addr, rdma_vlan_dev_vlan_id(ip_dev), 1684 ip_dev->dev_addr); 1685 child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL); 1686 ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n", 1687 child_listen_node); 1688 if (!child_listen_node) { 1689 ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n"); 1690 ret = -ENOMEM; 1691 goto exit; 1692 } 1693 1694 cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev); 1695 cm_parent_listen_node->vlan_id = cm_info->vlan_id; 1696 memcpy(child_listen_node, cm_parent_listen_node, 1697 sizeof(*child_listen_node)); 1698 irdma_copy_ip_ntohl(child_listen_node->loc_addr, 1699 ifp->addr.in6_u.u6_addr32); 1700 memcpy(cm_info->loc_addr, child_listen_node->loc_addr, 1701 sizeof(cm_info->loc_addr)); 1702 if (!iwdev->vsi.dscp_mode) 1703 cm_info->user_pri = 1704 irdma_iw_get_vlan_prio(child_listen_node->loc_addr, 1705 cm_info->user_pri, 1706 false); 1707 1708 ret = irdma_manage_qhash(iwdev, cm_info, 1709 IRDMA_QHASH_TYPE_TCP_SYN, 1710 IRDMA_QHASH_MANAGE_TYPE_ADD, 1711 NULL, true); 1712 if (ret) { 1713 kfree(child_listen_node); 1714 continue; 1715 } 1716 1717 trace_irdma_add_mqh_6(iwdev, child_listen_node, 1718 ip_dev->dev_addr); 1719 1720 child_listen_node->qhash_set = true; 1721 spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags); 1722 list_add(&child_listen_node->child_listen_list, 1723 &cm_parent_listen_node->child_listen_list); 1724 spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags); 1725 cm_parent_listen_node->cm_core->stats_listen_nodes_created++; 1726 } 1727 } 1728 exit: 1729 rtnl_unlock(); 1730 1731 return ret; 1732 } 1733 1734 /** 1735 * irdma_add_mqh_4 - Adds multiple qhashes for IPv4 1736 * @iwdev: iWarp device 1737 * @cm_info: CM info for parent listen node 1738 * @cm_parent_listen_node: The parent listen node 1739 * 1740 * Adds a qhash and a child listen node for every IPv4 address 1741 * on the adapter and adds the associated qhash filter 1742 */ 1743 static int irdma_add_mqh_4(struct irdma_device *iwdev, 1744 struct irdma_cm_info *cm_info, 1745 struct irdma_cm_listener *cm_parent_listen_node) 1746 { 1747 struct net_device *ip_dev; 1748 struct in_device *idev; 1749 struct irdma_cm_listener *child_listen_node; 1750 unsigned long flags; 1751 const struct in_ifaddr *ifa; 1752 int ret = 0; 1753 1754 rtnl_lock(); 1755 for_each_netdev(&init_net, ip_dev) { 1756 if (!(ip_dev->flags & IFF_UP)) 1757 continue; 1758 1759 if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) || 1760 (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) && 1761 ip_dev != iwdev->netdev) 1762 continue; 1763 1764 idev = in_dev_get(ip_dev); 1765 if (!idev) 1766 continue; 1767 1768 in_dev_for_each_ifa_rtnl(ifa, idev) { 1769 ibdev_dbg(&iwdev->ibdev, 1770 "CM: Allocating child CM Listener forIP=%pI4, vlan_id=%d, MAC=%pM\n", 1771 &ifa->ifa_address, rdma_vlan_dev_vlan_id(ip_dev), 1772 ip_dev->dev_addr); 1773 child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL); 1774 cm_parent_listen_node->cm_core->stats_listen_nodes_created++; 1775 ibdev_dbg(&iwdev->ibdev, "CM: Allocating child listener %p\n", 1776 child_listen_node); 1777 if (!child_listen_node) { 1778 ibdev_dbg(&iwdev->ibdev, "CM: listener memory allocation\n"); 1779 in_dev_put(idev); 1780 ret = -ENOMEM; 1781 goto exit; 1782 } 1783 1784 cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev); 1785 cm_parent_listen_node->vlan_id = cm_info->vlan_id; 1786 memcpy(child_listen_node, cm_parent_listen_node, 1787 sizeof(*child_listen_node)); 1788 child_listen_node->loc_addr[0] = 1789 ntohl(ifa->ifa_address); 1790 memcpy(cm_info->loc_addr, child_listen_node->loc_addr, 1791 sizeof(cm_info->loc_addr)); 1792 if (!iwdev->vsi.dscp_mode) 1793 cm_info->user_pri = 1794 irdma_iw_get_vlan_prio(child_listen_node->loc_addr, 1795 cm_info->user_pri, 1796 true); 1797 ret = irdma_manage_qhash(iwdev, cm_info, 1798 IRDMA_QHASH_TYPE_TCP_SYN, 1799 IRDMA_QHASH_MANAGE_TYPE_ADD, 1800 NULL, true); 1801 if (ret) { 1802 kfree(child_listen_node); 1803 cm_parent_listen_node->cm_core 1804 ->stats_listen_nodes_created--; 1805 continue; 1806 } 1807 1808 trace_irdma_add_mqh_4(iwdev, child_listen_node, 1809 ip_dev->dev_addr); 1810 1811 child_listen_node->qhash_set = true; 1812 spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, 1813 flags); 1814 list_add(&child_listen_node->child_listen_list, 1815 &cm_parent_listen_node->child_listen_list); 1816 spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags); 1817 } 1818 in_dev_put(idev); 1819 } 1820 exit: 1821 rtnl_unlock(); 1822 1823 return ret; 1824 } 1825 1826 /** 1827 * irdma_add_mqh - Adds multiple qhashes 1828 * @iwdev: iWarp device 1829 * @cm_info: CM info for parent listen node 1830 * @cm_listen_node: The parent listen node 1831 */ 1832 static int irdma_add_mqh(struct irdma_device *iwdev, 1833 struct irdma_cm_info *cm_info, 1834 struct irdma_cm_listener *cm_listen_node) 1835 { 1836 if (cm_info->ipv4) 1837 return irdma_add_mqh_4(iwdev, cm_info, cm_listen_node); 1838 else 1839 return irdma_add_mqh_6(iwdev, cm_info, cm_listen_node); 1840 } 1841 1842 /** 1843 * irdma_reset_list_prep - add connection nodes slated for reset to list 1844 * @cm_core: cm's core 1845 * @listener: pointer to listener node 1846 * @reset_list: a list to which cm_node will be selected 1847 */ 1848 static void irdma_reset_list_prep(struct irdma_cm_core *cm_core, 1849 struct irdma_cm_listener *listener, 1850 struct list_head *reset_list) 1851 { 1852 struct irdma_cm_node *cm_node; 1853 int bkt; 1854 1855 hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) { 1856 if (cm_node->listener == listener && 1857 !cm_node->accelerated && 1858 refcount_inc_not_zero(&cm_node->refcnt)) 1859 list_add(&cm_node->reset_entry, reset_list); 1860 } 1861 } 1862 1863 /** 1864 * irdma_dec_refcnt_listen - delete listener and associated cm nodes 1865 * @cm_core: cm's core 1866 * @listener: pointer to listener node 1867 * @free_hanging_nodes: to free associated cm_nodes 1868 * @apbvt_del: flag to delete the apbvt 1869 */ 1870 static int irdma_dec_refcnt_listen(struct irdma_cm_core *cm_core, 1871 struct irdma_cm_listener *listener, 1872 int free_hanging_nodes, bool apbvt_del) 1873 { 1874 int err; 1875 struct list_head *list_pos; 1876 struct list_head *list_temp; 1877 struct irdma_cm_node *cm_node; 1878 struct list_head reset_list; 1879 struct irdma_cm_info nfo; 1880 enum irdma_cm_node_state old_state; 1881 unsigned long flags; 1882 1883 trace_irdma_dec_refcnt_listen(listener, __builtin_return_address(0)); 1884 /* free non-accelerated child nodes for this listener */ 1885 INIT_LIST_HEAD(&reset_list); 1886 if (free_hanging_nodes) { 1887 rcu_read_lock(); 1888 irdma_reset_list_prep(cm_core, listener, &reset_list); 1889 rcu_read_unlock(); 1890 } 1891 1892 list_for_each_safe (list_pos, list_temp, &reset_list) { 1893 cm_node = container_of(list_pos, struct irdma_cm_node, 1894 reset_entry); 1895 if (cm_node->state >= IRDMA_CM_STATE_FIN_WAIT1) { 1896 irdma_rem_ref_cm_node(cm_node); 1897 continue; 1898 } 1899 1900 irdma_cleanup_retrans_entry(cm_node); 1901 err = irdma_send_reset(cm_node); 1902 if (err) { 1903 cm_node->state = IRDMA_CM_STATE_CLOSED; 1904 ibdev_dbg(&cm_node->iwdev->ibdev, 1905 "CM: send reset failed\n"); 1906 } else { 1907 old_state = cm_node->state; 1908 cm_node->state = IRDMA_CM_STATE_LISTENER_DESTROYED; 1909 if (old_state != IRDMA_CM_STATE_MPAREQ_RCVD) 1910 irdma_rem_ref_cm_node(cm_node); 1911 } 1912 } 1913 1914 if (refcount_dec_and_test(&listener->refcnt)) { 1915 spin_lock_irqsave(&cm_core->listen_list_lock, flags); 1916 list_del(&listener->list); 1917 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 1918 1919 if (apbvt_del) 1920 irdma_del_apbvt(listener->iwdev, 1921 listener->apbvt_entry); 1922 memcpy(nfo.loc_addr, listener->loc_addr, sizeof(nfo.loc_addr)); 1923 nfo.loc_port = listener->loc_port; 1924 nfo.ipv4 = listener->ipv4; 1925 nfo.vlan_id = listener->vlan_id; 1926 nfo.user_pri = listener->user_pri; 1927 nfo.qh_qpid = listener->iwdev->vsi.ilq->qp_id; 1928 1929 if (!list_empty(&listener->child_listen_list)) { 1930 irdma_del_multiple_qhash(listener->iwdev, &nfo, 1931 listener); 1932 } else { 1933 if (listener->qhash_set) 1934 irdma_manage_qhash(listener->iwdev, 1935 &nfo, 1936 IRDMA_QHASH_TYPE_TCP_SYN, 1937 IRDMA_QHASH_MANAGE_TYPE_DELETE, 1938 NULL, false); 1939 } 1940 1941 cm_core->stats_listen_destroyed++; 1942 cm_core->stats_listen_nodes_destroyed++; 1943 ibdev_dbg(&listener->iwdev->ibdev, 1944 "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d apbvt_del=%d\n", 1945 listener->loc_port, listener->loc_addr, listener, 1946 listener->cm_id, listener->qhash_set, 1947 listener->vlan_id, apbvt_del); 1948 kfree(listener); 1949 listener = NULL; 1950 return 0; 1951 } 1952 1953 return -EINVAL; 1954 } 1955 1956 /** 1957 * irdma_cm_del_listen - delete a listener 1958 * @cm_core: cm's core 1959 * @listener: passive connection's listener 1960 * @apbvt_del: flag to delete apbvt 1961 */ 1962 static int irdma_cm_del_listen(struct irdma_cm_core *cm_core, 1963 struct irdma_cm_listener *listener, 1964 bool apbvt_del) 1965 { 1966 listener->listener_state = IRDMA_CM_LISTENER_PASSIVE_STATE; 1967 listener->cm_id = NULL; 1968 1969 return irdma_dec_refcnt_listen(cm_core, listener, 1, apbvt_del); 1970 } 1971 1972 /** 1973 * irdma_addr_resolve_neigh - resolve neighbor address 1974 * @iwdev: iwarp device structure 1975 * @src_ip: local ip address 1976 * @dst_ip: remote ip address 1977 * @arpindex: if there is an arp entry 1978 */ 1979 static int irdma_addr_resolve_neigh(struct irdma_device *iwdev, u32 src_ip, 1980 u32 dst_ip, int arpindex) 1981 { 1982 struct rtable *rt; 1983 struct neighbour *neigh; 1984 int rc = arpindex; 1985 __be32 dst_ipaddr = htonl(dst_ip); 1986 __be32 src_ipaddr = htonl(src_ip); 1987 1988 rt = ip_route_output(&init_net, dst_ipaddr, src_ipaddr, 0, 0); 1989 if (IS_ERR(rt)) { 1990 ibdev_dbg(&iwdev->ibdev, "CM: ip_route_output fail\n"); 1991 return -EINVAL; 1992 } 1993 1994 neigh = dst_neigh_lookup(&rt->dst, &dst_ipaddr); 1995 if (!neigh) 1996 goto exit; 1997 1998 if (neigh->nud_state & NUD_VALID) 1999 rc = irdma_add_arp(iwdev->rf, &dst_ip, true, neigh->ha); 2000 else 2001 neigh_event_send(neigh, NULL); 2002 if (neigh) 2003 neigh_release(neigh); 2004 exit: 2005 ip_rt_put(rt); 2006 2007 return rc; 2008 } 2009 2010 /** 2011 * irdma_get_dst_ipv6 - get destination cache entry via ipv6 lookup 2012 * @src_addr: local ipv6 sock address 2013 * @dst_addr: destination ipv6 sock address 2014 */ 2015 static struct dst_entry *irdma_get_dst_ipv6(struct sockaddr_in6 *src_addr, 2016 struct sockaddr_in6 *dst_addr) 2017 { 2018 struct dst_entry *dst = NULL; 2019 2020 if ((IS_ENABLED(CONFIG_IPV6))) { 2021 struct flowi6 fl6 = {}; 2022 2023 fl6.daddr = dst_addr->sin6_addr; 2024 fl6.saddr = src_addr->sin6_addr; 2025 if (ipv6_addr_type(&fl6.daddr) & IPV6_ADDR_LINKLOCAL) 2026 fl6.flowi6_oif = dst_addr->sin6_scope_id; 2027 2028 dst = ip6_route_output(&init_net, NULL, &fl6); 2029 } 2030 2031 return dst; 2032 } 2033 2034 /** 2035 * irdma_addr_resolve_neigh_ipv6 - resolve neighbor ipv6 address 2036 * @iwdev: iwarp device structure 2037 * @src: local ip address 2038 * @dest: remote ip address 2039 * @arpindex: if there is an arp entry 2040 */ 2041 static int irdma_addr_resolve_neigh_ipv6(struct irdma_device *iwdev, u32 *src, 2042 u32 *dest, int arpindex) 2043 { 2044 struct neighbour *neigh; 2045 int rc = arpindex; 2046 struct dst_entry *dst; 2047 struct sockaddr_in6 dst_addr = {}; 2048 struct sockaddr_in6 src_addr = {}; 2049 2050 dst_addr.sin6_family = AF_INET6; 2051 irdma_copy_ip_htonl(dst_addr.sin6_addr.in6_u.u6_addr32, dest); 2052 src_addr.sin6_family = AF_INET6; 2053 irdma_copy_ip_htonl(src_addr.sin6_addr.in6_u.u6_addr32, src); 2054 dst = irdma_get_dst_ipv6(&src_addr, &dst_addr); 2055 if (!dst || dst->error) { 2056 if (dst) { 2057 dst_release(dst); 2058 ibdev_dbg(&iwdev->ibdev, 2059 "CM: ip6_route_output returned dst->error = %d\n", 2060 dst->error); 2061 } 2062 return -EINVAL; 2063 } 2064 2065 neigh = dst_neigh_lookup(dst, dst_addr.sin6_addr.in6_u.u6_addr32); 2066 if (!neigh) 2067 goto exit; 2068 2069 ibdev_dbg(&iwdev->ibdev, "CM: dst_neigh_lookup MAC=%pM\n", 2070 neigh->ha); 2071 2072 trace_irdma_addr_resolve(iwdev, neigh->ha); 2073 2074 if (neigh->nud_state & NUD_VALID) 2075 rc = irdma_add_arp(iwdev->rf, dest, false, neigh->ha); 2076 else 2077 neigh_event_send(neigh, NULL); 2078 if (neigh) 2079 neigh_release(neigh); 2080 exit: 2081 dst_release(dst); 2082 2083 return rc; 2084 } 2085 2086 /** 2087 * irdma_find_node - find a cm node that matches the reference cm node 2088 * @cm_core: cm's core 2089 * @rem_port: remote tcp port num 2090 * @rem_addr: remote ip addr 2091 * @loc_port: local tcp port num 2092 * @loc_addr: local ip addr 2093 * @vlan_id: local VLAN ID 2094 */ 2095 struct irdma_cm_node *irdma_find_node(struct irdma_cm_core *cm_core, 2096 u16 rem_port, u32 *rem_addr, u16 loc_port, 2097 u32 *loc_addr, u16 vlan_id) 2098 { 2099 struct irdma_cm_node *cm_node; 2100 u32 key = (rem_port << 16) | loc_port; 2101 2102 rcu_read_lock(); 2103 hash_for_each_possible_rcu(cm_core->cm_hash_tbl, cm_node, list, key) { 2104 if (cm_node->vlan_id == vlan_id && 2105 cm_node->loc_port == loc_port && cm_node->rem_port == rem_port && 2106 !memcmp(cm_node->loc_addr, loc_addr, sizeof(cm_node->loc_addr)) && 2107 !memcmp(cm_node->rem_addr, rem_addr, sizeof(cm_node->rem_addr))) { 2108 if (!refcount_inc_not_zero(&cm_node->refcnt)) 2109 goto exit; 2110 rcu_read_unlock(); 2111 trace_irdma_find_node(cm_node, 0, NULL); 2112 return cm_node; 2113 } 2114 } 2115 2116 exit: 2117 rcu_read_unlock(); 2118 2119 /* no owner node */ 2120 return NULL; 2121 } 2122 2123 /** 2124 * irdma_add_hte_node - add a cm node to the hash table 2125 * @cm_core: cm's core 2126 * @cm_node: connection's node 2127 */ 2128 static void irdma_add_hte_node(struct irdma_cm_core *cm_core, 2129 struct irdma_cm_node *cm_node) 2130 { 2131 unsigned long flags; 2132 u32 key = (cm_node->rem_port << 16) | cm_node->loc_port; 2133 2134 spin_lock_irqsave(&cm_core->ht_lock, flags); 2135 hash_add_rcu(cm_core->cm_hash_tbl, &cm_node->list, key); 2136 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 2137 } 2138 2139 /** 2140 * irdma_ipv4_is_lpb - check if loopback 2141 * @loc_addr: local addr to compare 2142 * @rem_addr: remote address 2143 */ 2144 bool irdma_ipv4_is_lpb(u32 loc_addr, u32 rem_addr) 2145 { 2146 return ipv4_is_loopback(htonl(rem_addr)) || (loc_addr == rem_addr); 2147 } 2148 2149 /** 2150 * irdma_ipv6_is_lpb - check if loopback 2151 * @loc_addr: local addr to compare 2152 * @rem_addr: remote address 2153 */ 2154 bool irdma_ipv6_is_lpb(u32 *loc_addr, u32 *rem_addr) 2155 { 2156 struct in6_addr raddr6; 2157 2158 irdma_copy_ip_htonl(raddr6.in6_u.u6_addr32, rem_addr); 2159 2160 return !memcmp(loc_addr, rem_addr, 16) || ipv6_addr_loopback(&raddr6); 2161 } 2162 2163 /** 2164 * irdma_cm_create_ah - create a cm address handle 2165 * @cm_node: The connection manager node to create AH for 2166 * @wait: Provides option to wait for ah creation or not 2167 */ 2168 static int irdma_cm_create_ah(struct irdma_cm_node *cm_node, bool wait) 2169 { 2170 struct irdma_ah_info ah_info = {}; 2171 struct irdma_device *iwdev = cm_node->iwdev; 2172 2173 ether_addr_copy(ah_info.mac_addr, iwdev->netdev->dev_addr); 2174 2175 ah_info.hop_ttl = 0x40; 2176 ah_info.tc_tos = cm_node->tos; 2177 ah_info.vsi = &iwdev->vsi; 2178 2179 if (cm_node->ipv4) { 2180 ah_info.ipv4_valid = true; 2181 ah_info.dest_ip_addr[0] = cm_node->rem_addr[0]; 2182 ah_info.src_ip_addr[0] = cm_node->loc_addr[0]; 2183 ah_info.do_lpbk = irdma_ipv4_is_lpb(ah_info.src_ip_addr[0], 2184 ah_info.dest_ip_addr[0]); 2185 } else { 2186 memcpy(ah_info.dest_ip_addr, cm_node->rem_addr, 2187 sizeof(ah_info.dest_ip_addr)); 2188 memcpy(ah_info.src_ip_addr, cm_node->loc_addr, 2189 sizeof(ah_info.src_ip_addr)); 2190 ah_info.do_lpbk = irdma_ipv6_is_lpb(ah_info.src_ip_addr, 2191 ah_info.dest_ip_addr); 2192 } 2193 2194 ah_info.vlan_tag = cm_node->vlan_id; 2195 if (cm_node->vlan_id < VLAN_N_VID) { 2196 ah_info.insert_vlan_tag = 1; 2197 ah_info.vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT; 2198 } 2199 2200 ah_info.dst_arpindex = 2201 irdma_arp_table(iwdev->rf, ah_info.dest_ip_addr, 2202 ah_info.ipv4_valid, NULL, IRDMA_ARP_RESOLVE); 2203 2204 if (irdma_puda_create_ah(&iwdev->rf->sc_dev, &ah_info, wait, 2205 IRDMA_PUDA_RSRC_TYPE_ILQ, cm_node, 2206 &cm_node->ah)) 2207 return -ENOMEM; 2208 2209 trace_irdma_create_ah(cm_node); 2210 return 0; 2211 } 2212 2213 /** 2214 * irdma_cm_free_ah - free a cm address handle 2215 * @cm_node: The connection manager node to create AH for 2216 */ 2217 static void irdma_cm_free_ah(struct irdma_cm_node *cm_node) 2218 { 2219 struct irdma_device *iwdev = cm_node->iwdev; 2220 2221 trace_irdma_cm_free_ah(cm_node); 2222 irdma_puda_free_ah(&iwdev->rf->sc_dev, cm_node->ah); 2223 cm_node->ah = NULL; 2224 } 2225 2226 /** 2227 * irdma_make_cm_node - create a new instance of a cm node 2228 * @cm_core: cm's core 2229 * @iwdev: iwarp device structure 2230 * @cm_info: quad info for connection 2231 * @listener: passive connection's listener 2232 */ 2233 static struct irdma_cm_node * 2234 irdma_make_cm_node(struct irdma_cm_core *cm_core, struct irdma_device *iwdev, 2235 struct irdma_cm_info *cm_info, 2236 struct irdma_cm_listener *listener) 2237 { 2238 struct irdma_cm_node *cm_node; 2239 int oldarpindex; 2240 int arpindex; 2241 struct net_device *netdev = iwdev->netdev; 2242 2243 /* create an hte and cm_node for this instance */ 2244 cm_node = kzalloc(sizeof(*cm_node), GFP_ATOMIC); 2245 if (!cm_node) 2246 return NULL; 2247 2248 /* set our node specific transport info */ 2249 cm_node->ipv4 = cm_info->ipv4; 2250 cm_node->vlan_id = cm_info->vlan_id; 2251 if (cm_node->vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode) 2252 cm_node->vlan_id = 0; 2253 cm_node->tos = cm_info->tos; 2254 cm_node->user_pri = cm_info->user_pri; 2255 if (listener) { 2256 if (listener->tos != cm_info->tos) 2257 ibdev_warn(&iwdev->ibdev, 2258 "application TOS[%d] and remote client TOS[%d] mismatch\n", 2259 listener->tos, cm_info->tos); 2260 if (iwdev->vsi.dscp_mode) { 2261 cm_node->user_pri = listener->user_pri; 2262 } else { 2263 cm_node->tos = max(listener->tos, cm_info->tos); 2264 cm_node->user_pri = rt_tos2priority(cm_node->tos); 2265 cm_node->user_pri = 2266 irdma_iw_get_vlan_prio(cm_info->loc_addr, 2267 cm_node->user_pri, 2268 cm_info->ipv4); 2269 } 2270 ibdev_dbg(&iwdev->ibdev, 2271 "DCB: listener: TOS:[%d] UP:[%d]\n", cm_node->tos, 2272 cm_node->user_pri); 2273 trace_irdma_listener_tos(iwdev, cm_node->tos, 2274 cm_node->user_pri); 2275 } 2276 memcpy(cm_node->loc_addr, cm_info->loc_addr, sizeof(cm_node->loc_addr)); 2277 memcpy(cm_node->rem_addr, cm_info->rem_addr, sizeof(cm_node->rem_addr)); 2278 cm_node->loc_port = cm_info->loc_port; 2279 cm_node->rem_port = cm_info->rem_port; 2280 2281 cm_node->mpa_frame_rev = IRDMA_CM_DEFAULT_MPA_VER; 2282 cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO; 2283 cm_node->iwdev = iwdev; 2284 cm_node->dev = &iwdev->rf->sc_dev; 2285 2286 cm_node->ird_size = cm_node->dev->hw_attrs.max_hw_ird; 2287 cm_node->ord_size = cm_node->dev->hw_attrs.max_hw_ord; 2288 2289 cm_node->listener = listener; 2290 cm_node->cm_id = cm_info->cm_id; 2291 ether_addr_copy(cm_node->loc_mac, netdev->dev_addr); 2292 spin_lock_init(&cm_node->retrans_list_lock); 2293 cm_node->ack_rcvd = false; 2294 2295 init_completion(&cm_node->establish_comp); 2296 refcount_set(&cm_node->refcnt, 1); 2297 /* associate our parent CM core */ 2298 cm_node->cm_core = cm_core; 2299 cm_node->tcp_cntxt.loc_id = IRDMA_CM_DEFAULT_LOCAL_ID; 2300 cm_node->tcp_cntxt.rcv_wscale = iwdev->rcv_wscale; 2301 cm_node->tcp_cntxt.rcv_wnd = iwdev->rcv_wnd >> cm_node->tcp_cntxt.rcv_wscale; 2302 if (cm_node->ipv4) { 2303 cm_node->tcp_cntxt.loc_seq_num = secure_tcp_seq(htonl(cm_node->loc_addr[0]), 2304 htonl(cm_node->rem_addr[0]), 2305 htons(cm_node->loc_port), 2306 htons(cm_node->rem_port)); 2307 cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV4; 2308 } else if (IS_ENABLED(CONFIG_IPV6)) { 2309 __be32 loc[4] = { 2310 htonl(cm_node->loc_addr[0]), htonl(cm_node->loc_addr[1]), 2311 htonl(cm_node->loc_addr[2]), htonl(cm_node->loc_addr[3]) 2312 }; 2313 __be32 rem[4] = { 2314 htonl(cm_node->rem_addr[0]), htonl(cm_node->rem_addr[1]), 2315 htonl(cm_node->rem_addr[2]), htonl(cm_node->rem_addr[3]) 2316 }; 2317 cm_node->tcp_cntxt.loc_seq_num = secure_tcpv6_seq(loc, rem, 2318 htons(cm_node->loc_port), 2319 htons(cm_node->rem_port)); 2320 cm_node->tcp_cntxt.mss = iwdev->vsi.mtu - IRDMA_MTU_TO_MSS_IPV6; 2321 } 2322 2323 if ((cm_node->ipv4 && 2324 irdma_ipv4_is_lpb(cm_node->loc_addr[0], cm_node->rem_addr[0])) || 2325 (!cm_node->ipv4 && 2326 irdma_ipv6_is_lpb(cm_node->loc_addr, cm_node->rem_addr))) { 2327 cm_node->do_lpb = true; 2328 arpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr, 2329 cm_node->ipv4, NULL, 2330 IRDMA_ARP_RESOLVE); 2331 } else { 2332 oldarpindex = irdma_arp_table(iwdev->rf, cm_node->rem_addr, 2333 cm_node->ipv4, NULL, 2334 IRDMA_ARP_RESOLVE); 2335 if (cm_node->ipv4) 2336 arpindex = irdma_addr_resolve_neigh(iwdev, 2337 cm_info->loc_addr[0], 2338 cm_info->rem_addr[0], 2339 oldarpindex); 2340 else if (IS_ENABLED(CONFIG_IPV6)) 2341 arpindex = irdma_addr_resolve_neigh_ipv6(iwdev, 2342 cm_info->loc_addr, 2343 cm_info->rem_addr, 2344 oldarpindex); 2345 else 2346 arpindex = -EINVAL; 2347 } 2348 2349 if (arpindex < 0) 2350 goto err; 2351 2352 ether_addr_copy(cm_node->rem_mac, 2353 iwdev->rf->arp_table[arpindex].mac_addr); 2354 irdma_add_hte_node(cm_core, cm_node); 2355 cm_core->stats_nodes_created++; 2356 return cm_node; 2357 2358 err: 2359 kfree(cm_node); 2360 2361 return NULL; 2362 } 2363 2364 static void irdma_destroy_connection(struct irdma_cm_node *cm_node) 2365 { 2366 struct irdma_cm_core *cm_core = cm_node->cm_core; 2367 struct irdma_qp *iwqp; 2368 struct irdma_cm_info nfo; 2369 2370 /* if the node is destroyed before connection was accelerated */ 2371 if (!cm_node->accelerated && cm_node->accept_pend) { 2372 ibdev_dbg(&cm_node->iwdev->ibdev, 2373 "CM: node destroyed before established\n"); 2374 atomic_dec(&cm_node->listener->pend_accepts_cnt); 2375 } 2376 if (cm_node->close_entry) 2377 irdma_handle_close_entry(cm_node, 0); 2378 if (cm_node->listener) { 2379 irdma_dec_refcnt_listen(cm_core, cm_node->listener, 0, true); 2380 } else { 2381 if (cm_node->apbvt_set) { 2382 irdma_del_apbvt(cm_node->iwdev, cm_node->apbvt_entry); 2383 cm_node->apbvt_set = 0; 2384 } 2385 irdma_get_addr_info(cm_node, &nfo); 2386 if (cm_node->qhash_set) { 2387 nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id; 2388 irdma_manage_qhash(cm_node->iwdev, &nfo, 2389 IRDMA_QHASH_TYPE_TCP_ESTABLISHED, 2390 IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL, 2391 false); 2392 cm_node->qhash_set = 0; 2393 } 2394 } 2395 2396 iwqp = cm_node->iwqp; 2397 if (iwqp) { 2398 cm_node->cm_id->rem_ref(cm_node->cm_id); 2399 cm_node->cm_id = NULL; 2400 iwqp->cm_id = NULL; 2401 irdma_qp_rem_ref(&iwqp->ibqp); 2402 cm_node->iwqp = NULL; 2403 } else if (cm_node->qhash_set) { 2404 irdma_get_addr_info(cm_node, &nfo); 2405 nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id; 2406 irdma_manage_qhash(cm_node->iwdev, &nfo, 2407 IRDMA_QHASH_TYPE_TCP_ESTABLISHED, 2408 IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL, false); 2409 cm_node->qhash_set = 0; 2410 } 2411 2412 cm_core->cm_free_ah(cm_node); 2413 } 2414 2415 /** 2416 * irdma_rem_ref_cm_node - destroy an instance of a cm node 2417 * @cm_node: connection's node 2418 */ 2419 void irdma_rem_ref_cm_node(struct irdma_cm_node *cm_node) 2420 { 2421 struct irdma_cm_core *cm_core = cm_node->cm_core; 2422 unsigned long flags; 2423 2424 trace_irdma_rem_ref_cm_node(cm_node, 0, __builtin_return_address(0)); 2425 spin_lock_irqsave(&cm_core->ht_lock, flags); 2426 2427 if (!refcount_dec_and_test(&cm_node->refcnt)) { 2428 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 2429 return; 2430 } 2431 if (cm_node->iwqp) { 2432 cm_node->iwqp->cm_node = NULL; 2433 cm_node->iwqp->cm_id = NULL; 2434 } 2435 hash_del_rcu(&cm_node->list); 2436 cm_node->cm_core->stats_nodes_destroyed++; 2437 2438 spin_unlock_irqrestore(&cm_core->ht_lock, flags); 2439 2440 irdma_destroy_connection(cm_node); 2441 2442 kfree_rcu(cm_node, rcu_head); 2443 } 2444 2445 /** 2446 * irdma_handle_fin_pkt - FIN packet received 2447 * @cm_node: connection's node 2448 */ 2449 static void irdma_handle_fin_pkt(struct irdma_cm_node *cm_node) 2450 { 2451 switch (cm_node->state) { 2452 case IRDMA_CM_STATE_SYN_RCVD: 2453 case IRDMA_CM_STATE_SYN_SENT: 2454 case IRDMA_CM_STATE_ESTABLISHED: 2455 case IRDMA_CM_STATE_MPAREJ_RCVD: 2456 cm_node->tcp_cntxt.rcv_nxt++; 2457 irdma_cleanup_retrans_entry(cm_node); 2458 cm_node->state = IRDMA_CM_STATE_LAST_ACK; 2459 irdma_send_fin(cm_node); 2460 break; 2461 case IRDMA_CM_STATE_MPAREQ_SENT: 2462 irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED); 2463 cm_node->tcp_cntxt.rcv_nxt++; 2464 irdma_cleanup_retrans_entry(cm_node); 2465 cm_node->state = IRDMA_CM_STATE_CLOSED; 2466 refcount_inc(&cm_node->refcnt); 2467 irdma_send_reset(cm_node); 2468 break; 2469 case IRDMA_CM_STATE_FIN_WAIT1: 2470 cm_node->tcp_cntxt.rcv_nxt++; 2471 irdma_cleanup_retrans_entry(cm_node); 2472 cm_node->state = IRDMA_CM_STATE_CLOSING; 2473 irdma_send_ack(cm_node); 2474 /* 2475 * Wait for ACK as this is simultaneous close. 2476 * After we receive ACK, do not send anything. 2477 * Just rm the node. 2478 */ 2479 break; 2480 case IRDMA_CM_STATE_FIN_WAIT2: 2481 cm_node->tcp_cntxt.rcv_nxt++; 2482 irdma_cleanup_retrans_entry(cm_node); 2483 cm_node->state = IRDMA_CM_STATE_TIME_WAIT; 2484 irdma_send_ack(cm_node); 2485 irdma_schedule_cm_timer(cm_node, NULL, IRDMA_TIMER_TYPE_CLOSE, 2486 1, 0); 2487 break; 2488 case IRDMA_CM_STATE_TIME_WAIT: 2489 cm_node->tcp_cntxt.rcv_nxt++; 2490 irdma_cleanup_retrans_entry(cm_node); 2491 cm_node->state = IRDMA_CM_STATE_CLOSED; 2492 irdma_rem_ref_cm_node(cm_node); 2493 break; 2494 case IRDMA_CM_STATE_OFFLOADED: 2495 default: 2496 ibdev_dbg(&cm_node->iwdev->ibdev, 2497 "CM: bad state node state = %d\n", cm_node->state); 2498 break; 2499 } 2500 } 2501 2502 /** 2503 * irdma_handle_rst_pkt - process received RST packet 2504 * @cm_node: connection's node 2505 * @rbuf: receive buffer 2506 */ 2507 static void irdma_handle_rst_pkt(struct irdma_cm_node *cm_node, 2508 struct irdma_puda_buf *rbuf) 2509 { 2510 ibdev_dbg(&cm_node->iwdev->ibdev, 2511 "CM: caller: %pS cm_node=%p state=%d rem_port=0x%04x loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n", 2512 __builtin_return_address(0), cm_node, cm_node->state, 2513 cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr, 2514 cm_node->loc_addr); 2515 2516 irdma_cleanup_retrans_entry(cm_node); 2517 switch (cm_node->state) { 2518 case IRDMA_CM_STATE_SYN_SENT: 2519 case IRDMA_CM_STATE_MPAREQ_SENT: 2520 switch (cm_node->mpa_frame_rev) { 2521 case IETF_MPA_V2: 2522 /* Drop down to MPA_V1*/ 2523 cm_node->mpa_frame_rev = IETF_MPA_V1; 2524 /* send a syn and goto syn sent state */ 2525 cm_node->state = IRDMA_CM_STATE_SYN_SENT; 2526 if (irdma_send_syn(cm_node, 0)) 2527 irdma_active_open_err(cm_node, false); 2528 break; 2529 case IETF_MPA_V1: 2530 default: 2531 irdma_active_open_err(cm_node, false); 2532 break; 2533 } 2534 break; 2535 case IRDMA_CM_STATE_MPAREQ_RCVD: 2536 atomic_inc(&cm_node->passive_state); 2537 break; 2538 case IRDMA_CM_STATE_ESTABLISHED: 2539 case IRDMA_CM_STATE_SYN_RCVD: 2540 case IRDMA_CM_STATE_LISTENING: 2541 irdma_passive_open_err(cm_node, false); 2542 break; 2543 case IRDMA_CM_STATE_OFFLOADED: 2544 irdma_active_open_err(cm_node, false); 2545 break; 2546 case IRDMA_CM_STATE_CLOSED: 2547 break; 2548 case IRDMA_CM_STATE_FIN_WAIT2: 2549 case IRDMA_CM_STATE_FIN_WAIT1: 2550 case IRDMA_CM_STATE_LAST_ACK: 2551 case IRDMA_CM_STATE_TIME_WAIT: 2552 cm_node->state = IRDMA_CM_STATE_CLOSED; 2553 irdma_rem_ref_cm_node(cm_node); 2554 break; 2555 default: 2556 break; 2557 } 2558 } 2559 2560 /** 2561 * irdma_handle_rcv_mpa - Process a recv'd mpa buffer 2562 * @cm_node: connection's node 2563 * @rbuf: receive buffer 2564 */ 2565 static void irdma_handle_rcv_mpa(struct irdma_cm_node *cm_node, 2566 struct irdma_puda_buf *rbuf) 2567 { 2568 int err; 2569 int datasize = rbuf->datalen; 2570 u8 *dataloc = rbuf->data; 2571 2572 enum irdma_cm_event_type type = IRDMA_CM_EVENT_UNKNOWN; 2573 u32 res_type; 2574 2575 err = irdma_parse_mpa(cm_node, dataloc, &res_type, datasize); 2576 if (err) { 2577 if (cm_node->state == IRDMA_CM_STATE_MPAREQ_SENT) 2578 irdma_active_open_err(cm_node, true); 2579 else 2580 irdma_passive_open_err(cm_node, true); 2581 return; 2582 } 2583 2584 switch (cm_node->state) { 2585 case IRDMA_CM_STATE_ESTABLISHED: 2586 if (res_type == IRDMA_MPA_REQUEST_REJECT) 2587 ibdev_dbg(&cm_node->iwdev->ibdev, 2588 "CM: state for reject\n"); 2589 cm_node->state = IRDMA_CM_STATE_MPAREQ_RCVD; 2590 type = IRDMA_CM_EVENT_MPA_REQ; 2591 irdma_send_ack(cm_node); /* ACK received MPA request */ 2592 atomic_set(&cm_node->passive_state, 2593 IRDMA_PASSIVE_STATE_INDICATED); 2594 break; 2595 case IRDMA_CM_STATE_MPAREQ_SENT: 2596 irdma_cleanup_retrans_entry(cm_node); 2597 if (res_type == IRDMA_MPA_REQUEST_REJECT) { 2598 type = IRDMA_CM_EVENT_MPA_REJECT; 2599 cm_node->state = IRDMA_CM_STATE_MPAREJ_RCVD; 2600 } else { 2601 type = IRDMA_CM_EVENT_CONNECTED; 2602 cm_node->state = IRDMA_CM_STATE_OFFLOADED; 2603 } 2604 irdma_send_ack(cm_node); 2605 break; 2606 default: 2607 ibdev_dbg(&cm_node->iwdev->ibdev, 2608 "CM: wrong cm_node state =%d\n", cm_node->state); 2609 break; 2610 } 2611 irdma_create_event(cm_node, type); 2612 } 2613 2614 /** 2615 * irdma_check_syn - Check for error on received syn ack 2616 * @cm_node: connection's node 2617 * @tcph: pointer tcp header 2618 */ 2619 static int irdma_check_syn(struct irdma_cm_node *cm_node, struct tcphdr *tcph) 2620 { 2621 if (ntohl(tcph->ack_seq) != cm_node->tcp_cntxt.loc_seq_num) { 2622 irdma_active_open_err(cm_node, true); 2623 return 1; 2624 } 2625 2626 return 0; 2627 } 2628 2629 /** 2630 * irdma_check_seq - check seq numbers if OK 2631 * @cm_node: connection's node 2632 * @tcph: pointer tcp header 2633 */ 2634 static int irdma_check_seq(struct irdma_cm_node *cm_node, struct tcphdr *tcph) 2635 { 2636 u32 seq; 2637 u32 ack_seq; 2638 u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num; 2639 u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt; 2640 u32 rcv_wnd; 2641 int err = 0; 2642 2643 seq = ntohl(tcph->seq); 2644 ack_seq = ntohl(tcph->ack_seq); 2645 rcv_wnd = cm_node->tcp_cntxt.rcv_wnd; 2646 if (ack_seq != loc_seq_num || 2647 !between(seq, rcv_nxt, (rcv_nxt + rcv_wnd))) 2648 err = -1; 2649 if (err) 2650 ibdev_dbg(&cm_node->iwdev->ibdev, 2651 "CM: seq number err\n"); 2652 2653 return err; 2654 } 2655 2656 void irdma_add_conn_est_qh(struct irdma_cm_node *cm_node) 2657 { 2658 struct irdma_cm_info nfo; 2659 2660 irdma_get_addr_info(cm_node, &nfo); 2661 nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id; 2662 irdma_manage_qhash(cm_node->iwdev, &nfo, 2663 IRDMA_QHASH_TYPE_TCP_ESTABLISHED, 2664 IRDMA_QHASH_MANAGE_TYPE_ADD, 2665 cm_node, false); 2666 cm_node->qhash_set = true; 2667 } 2668 2669 /** 2670 * irdma_handle_syn_pkt - is for Passive node 2671 * @cm_node: connection's node 2672 * @rbuf: receive buffer 2673 */ 2674 static void irdma_handle_syn_pkt(struct irdma_cm_node *cm_node, 2675 struct irdma_puda_buf *rbuf) 2676 { 2677 struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph; 2678 int err; 2679 u32 inc_sequence; 2680 int optionsize; 2681 2682 optionsize = (tcph->doff << 2) - sizeof(struct tcphdr); 2683 inc_sequence = ntohl(tcph->seq); 2684 2685 switch (cm_node->state) { 2686 case IRDMA_CM_STATE_SYN_SENT: 2687 case IRDMA_CM_STATE_MPAREQ_SENT: 2688 /* Rcvd syn on active open connection */ 2689 irdma_active_open_err(cm_node, 1); 2690 break; 2691 case IRDMA_CM_STATE_LISTENING: 2692 /* Passive OPEN */ 2693 if (atomic_read(&cm_node->listener->pend_accepts_cnt) > 2694 cm_node->listener->backlog) { 2695 cm_node->cm_core->stats_backlog_drops++; 2696 irdma_passive_open_err(cm_node, false); 2697 break; 2698 } 2699 err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1); 2700 if (err) { 2701 irdma_passive_open_err(cm_node, false); 2702 /* drop pkt */ 2703 break; 2704 } 2705 err = cm_node->cm_core->cm_create_ah(cm_node, false); 2706 if (err) { 2707 irdma_passive_open_err(cm_node, false); 2708 /* drop pkt */ 2709 break; 2710 } 2711 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1; 2712 cm_node->accept_pend = 1; 2713 atomic_inc(&cm_node->listener->pend_accepts_cnt); 2714 2715 cm_node->state = IRDMA_CM_STATE_SYN_RCVD; 2716 break; 2717 case IRDMA_CM_STATE_CLOSED: 2718 irdma_cleanup_retrans_entry(cm_node); 2719 refcount_inc(&cm_node->refcnt); 2720 irdma_send_reset(cm_node); 2721 break; 2722 case IRDMA_CM_STATE_OFFLOADED: 2723 case IRDMA_CM_STATE_ESTABLISHED: 2724 case IRDMA_CM_STATE_FIN_WAIT1: 2725 case IRDMA_CM_STATE_FIN_WAIT2: 2726 case IRDMA_CM_STATE_MPAREQ_RCVD: 2727 case IRDMA_CM_STATE_LAST_ACK: 2728 case IRDMA_CM_STATE_CLOSING: 2729 case IRDMA_CM_STATE_UNKNOWN: 2730 default: 2731 break; 2732 } 2733 } 2734 2735 /** 2736 * irdma_handle_synack_pkt - Process SYN+ACK packet (active side) 2737 * @cm_node: connection's node 2738 * @rbuf: receive buffer 2739 */ 2740 static void irdma_handle_synack_pkt(struct irdma_cm_node *cm_node, 2741 struct irdma_puda_buf *rbuf) 2742 { 2743 struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph; 2744 int err; 2745 u32 inc_sequence; 2746 int optionsize; 2747 2748 optionsize = (tcph->doff << 2) - sizeof(struct tcphdr); 2749 inc_sequence = ntohl(tcph->seq); 2750 switch (cm_node->state) { 2751 case IRDMA_CM_STATE_SYN_SENT: 2752 irdma_cleanup_retrans_entry(cm_node); 2753 /* active open */ 2754 if (irdma_check_syn(cm_node, tcph)) { 2755 ibdev_dbg(&cm_node->iwdev->ibdev, 2756 "CM: check syn fail\n"); 2757 return; 2758 } 2759 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); 2760 /* setup options */ 2761 err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 0); 2762 if (err) { 2763 ibdev_dbg(&cm_node->iwdev->ibdev, 2764 "CM: cm_node=%p tcp_options failed\n", 2765 cm_node); 2766 break; 2767 } 2768 irdma_cleanup_retrans_entry(cm_node); 2769 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1; 2770 irdma_send_ack(cm_node); /* ACK for the syn_ack */ 2771 err = irdma_send_mpa_request(cm_node); 2772 if (err) { 2773 ibdev_dbg(&cm_node->iwdev->ibdev, 2774 "CM: cm_node=%p irdma_send_mpa_request failed\n", 2775 cm_node); 2776 break; 2777 } 2778 cm_node->state = IRDMA_CM_STATE_MPAREQ_SENT; 2779 break; 2780 case IRDMA_CM_STATE_MPAREQ_RCVD: 2781 irdma_passive_open_err(cm_node, true); 2782 break; 2783 case IRDMA_CM_STATE_LISTENING: 2784 cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq); 2785 irdma_cleanup_retrans_entry(cm_node); 2786 cm_node->state = IRDMA_CM_STATE_CLOSED; 2787 irdma_send_reset(cm_node); 2788 break; 2789 case IRDMA_CM_STATE_CLOSED: 2790 cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->ack_seq); 2791 irdma_cleanup_retrans_entry(cm_node); 2792 refcount_inc(&cm_node->refcnt); 2793 irdma_send_reset(cm_node); 2794 break; 2795 case IRDMA_CM_STATE_ESTABLISHED: 2796 case IRDMA_CM_STATE_FIN_WAIT1: 2797 case IRDMA_CM_STATE_FIN_WAIT2: 2798 case IRDMA_CM_STATE_LAST_ACK: 2799 case IRDMA_CM_STATE_OFFLOADED: 2800 case IRDMA_CM_STATE_CLOSING: 2801 case IRDMA_CM_STATE_UNKNOWN: 2802 case IRDMA_CM_STATE_MPAREQ_SENT: 2803 default: 2804 break; 2805 } 2806 } 2807 2808 /** 2809 * irdma_handle_ack_pkt - process packet with ACK 2810 * @cm_node: connection's node 2811 * @rbuf: receive buffer 2812 */ 2813 static int irdma_handle_ack_pkt(struct irdma_cm_node *cm_node, 2814 struct irdma_puda_buf *rbuf) 2815 { 2816 struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph; 2817 u32 inc_sequence; 2818 int ret; 2819 int optionsize; 2820 u32 datasize = rbuf->datalen; 2821 2822 optionsize = (tcph->doff << 2) - sizeof(struct tcphdr); 2823 2824 if (irdma_check_seq(cm_node, tcph)) 2825 return -EINVAL; 2826 2827 inc_sequence = ntohl(tcph->seq); 2828 switch (cm_node->state) { 2829 case IRDMA_CM_STATE_SYN_RCVD: 2830 irdma_cleanup_retrans_entry(cm_node); 2831 ret = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1); 2832 if (ret) 2833 return ret; 2834 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); 2835 cm_node->state = IRDMA_CM_STATE_ESTABLISHED; 2836 if (datasize) { 2837 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 2838 irdma_handle_rcv_mpa(cm_node, rbuf); 2839 } 2840 break; 2841 case IRDMA_CM_STATE_ESTABLISHED: 2842 irdma_cleanup_retrans_entry(cm_node); 2843 if (datasize) { 2844 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 2845 irdma_handle_rcv_mpa(cm_node, rbuf); 2846 } 2847 break; 2848 case IRDMA_CM_STATE_MPAREQ_SENT: 2849 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->ack_seq); 2850 if (datasize) { 2851 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize; 2852 cm_node->ack_rcvd = false; 2853 irdma_handle_rcv_mpa(cm_node, rbuf); 2854 } else { 2855 cm_node->ack_rcvd = true; 2856 } 2857 break; 2858 case IRDMA_CM_STATE_LISTENING: 2859 irdma_cleanup_retrans_entry(cm_node); 2860 cm_node->state = IRDMA_CM_STATE_CLOSED; 2861 irdma_send_reset(cm_node); 2862 break; 2863 case IRDMA_CM_STATE_CLOSED: 2864 irdma_cleanup_retrans_entry(cm_node); 2865 refcount_inc(&cm_node->refcnt); 2866 irdma_send_reset(cm_node); 2867 break; 2868 case IRDMA_CM_STATE_LAST_ACK: 2869 case IRDMA_CM_STATE_CLOSING: 2870 irdma_cleanup_retrans_entry(cm_node); 2871 cm_node->state = IRDMA_CM_STATE_CLOSED; 2872 irdma_rem_ref_cm_node(cm_node); 2873 break; 2874 case IRDMA_CM_STATE_FIN_WAIT1: 2875 irdma_cleanup_retrans_entry(cm_node); 2876 cm_node->state = IRDMA_CM_STATE_FIN_WAIT2; 2877 break; 2878 case IRDMA_CM_STATE_SYN_SENT: 2879 case IRDMA_CM_STATE_FIN_WAIT2: 2880 case IRDMA_CM_STATE_OFFLOADED: 2881 case IRDMA_CM_STATE_MPAREQ_RCVD: 2882 case IRDMA_CM_STATE_UNKNOWN: 2883 default: 2884 irdma_cleanup_retrans_entry(cm_node); 2885 break; 2886 } 2887 2888 return 0; 2889 } 2890 2891 /** 2892 * irdma_process_pkt - process cm packet 2893 * @cm_node: connection's node 2894 * @rbuf: receive buffer 2895 */ 2896 static void irdma_process_pkt(struct irdma_cm_node *cm_node, 2897 struct irdma_puda_buf *rbuf) 2898 { 2899 enum irdma_tcpip_pkt_type pkt_type = IRDMA_PKT_TYPE_UNKNOWN; 2900 struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph; 2901 u32 fin_set = 0; 2902 int err; 2903 2904 if (tcph->rst) { 2905 pkt_type = IRDMA_PKT_TYPE_RST; 2906 } else if (tcph->syn) { 2907 pkt_type = IRDMA_PKT_TYPE_SYN; 2908 if (tcph->ack) 2909 pkt_type = IRDMA_PKT_TYPE_SYNACK; 2910 } else if (tcph->ack) { 2911 pkt_type = IRDMA_PKT_TYPE_ACK; 2912 } 2913 if (tcph->fin) 2914 fin_set = 1; 2915 2916 switch (pkt_type) { 2917 case IRDMA_PKT_TYPE_SYN: 2918 irdma_handle_syn_pkt(cm_node, rbuf); 2919 break; 2920 case IRDMA_PKT_TYPE_SYNACK: 2921 irdma_handle_synack_pkt(cm_node, rbuf); 2922 break; 2923 case IRDMA_PKT_TYPE_ACK: 2924 err = irdma_handle_ack_pkt(cm_node, rbuf); 2925 if (fin_set && !err) 2926 irdma_handle_fin_pkt(cm_node); 2927 break; 2928 case IRDMA_PKT_TYPE_RST: 2929 irdma_handle_rst_pkt(cm_node, rbuf); 2930 break; 2931 default: 2932 if (fin_set && 2933 (!irdma_check_seq(cm_node, (struct tcphdr *)rbuf->tcph))) 2934 irdma_handle_fin_pkt(cm_node); 2935 break; 2936 } 2937 } 2938 2939 /** 2940 * irdma_make_listen_node - create a listen node with params 2941 * @cm_core: cm's core 2942 * @iwdev: iwarp device structure 2943 * @cm_info: quad info for connection 2944 */ 2945 static struct irdma_cm_listener * 2946 irdma_make_listen_node(struct irdma_cm_core *cm_core, 2947 struct irdma_device *iwdev, 2948 struct irdma_cm_info *cm_info) 2949 { 2950 struct irdma_cm_listener *listener; 2951 unsigned long flags; 2952 2953 /* cannot have multiple matching listeners */ 2954 listener = 2955 irdma_find_listener(cm_core, cm_info->loc_addr, cm_info->ipv4, 2956 cm_info->loc_port, cm_info->vlan_id, 2957 IRDMA_CM_LISTENER_EITHER_STATE); 2958 if (listener && 2959 listener->listener_state == IRDMA_CM_LISTENER_ACTIVE_STATE) { 2960 refcount_dec(&listener->refcnt); 2961 return NULL; 2962 } 2963 2964 if (!listener) { 2965 /* create a CM listen node 2966 * 1/2 node to compare incoming traffic to 2967 */ 2968 listener = kzalloc(sizeof(*listener), GFP_KERNEL); 2969 if (!listener) 2970 return NULL; 2971 cm_core->stats_listen_nodes_created++; 2972 memcpy(listener->loc_addr, cm_info->loc_addr, 2973 sizeof(listener->loc_addr)); 2974 listener->loc_port = cm_info->loc_port; 2975 2976 INIT_LIST_HEAD(&listener->child_listen_list); 2977 2978 refcount_set(&listener->refcnt, 1); 2979 } else { 2980 listener->reused_node = 1; 2981 } 2982 2983 listener->cm_id = cm_info->cm_id; 2984 listener->ipv4 = cm_info->ipv4; 2985 listener->vlan_id = cm_info->vlan_id; 2986 atomic_set(&listener->pend_accepts_cnt, 0); 2987 listener->cm_core = cm_core; 2988 listener->iwdev = iwdev; 2989 2990 listener->backlog = cm_info->backlog; 2991 listener->listener_state = IRDMA_CM_LISTENER_ACTIVE_STATE; 2992 2993 if (!listener->reused_node) { 2994 spin_lock_irqsave(&cm_core->listen_list_lock, flags); 2995 list_add(&listener->list, &cm_core->listen_list); 2996 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 2997 } 2998 2999 return listener; 3000 } 3001 3002 /** 3003 * irdma_create_cm_node - make a connection node with params 3004 * @cm_core: cm's core 3005 * @iwdev: iwarp device structure 3006 * @conn_param: connection parameters 3007 * @cm_info: quad info for connection 3008 * @caller_cm_node: pointer to cm_node structure to return 3009 */ 3010 static int irdma_create_cm_node(struct irdma_cm_core *cm_core, 3011 struct irdma_device *iwdev, 3012 struct iw_cm_conn_param *conn_param, 3013 struct irdma_cm_info *cm_info, 3014 struct irdma_cm_node **caller_cm_node) 3015 { 3016 struct irdma_cm_node *cm_node; 3017 u16 private_data_len = conn_param->private_data_len; 3018 const void *private_data = conn_param->private_data; 3019 3020 /* create a CM connection node */ 3021 cm_node = irdma_make_cm_node(cm_core, iwdev, cm_info, NULL); 3022 if (!cm_node) 3023 return -ENOMEM; 3024 3025 /* set our node side to client (active) side */ 3026 cm_node->tcp_cntxt.client = 1; 3027 cm_node->tcp_cntxt.rcv_wscale = IRDMA_CM_DEFAULT_RCV_WND_SCALE; 3028 3029 irdma_record_ird_ord(cm_node, conn_param->ird, conn_param->ord); 3030 3031 cm_node->pdata.size = private_data_len; 3032 cm_node->pdata.addr = cm_node->pdata_buf; 3033 3034 memcpy(cm_node->pdata_buf, private_data, private_data_len); 3035 *caller_cm_node = cm_node; 3036 3037 return 0; 3038 } 3039 3040 /** 3041 * irdma_cm_reject - reject and teardown a connection 3042 * @cm_node: connection's node 3043 * @pdata: ptr to private data for reject 3044 * @plen: size of private data 3045 */ 3046 static int irdma_cm_reject(struct irdma_cm_node *cm_node, const void *pdata, 3047 u8 plen) 3048 { 3049 int ret; 3050 int passive_state; 3051 3052 if (cm_node->tcp_cntxt.client) 3053 return 0; 3054 3055 irdma_cleanup_retrans_entry(cm_node); 3056 3057 passive_state = atomic_add_return(1, &cm_node->passive_state); 3058 if (passive_state == IRDMA_SEND_RESET_EVENT) { 3059 cm_node->state = IRDMA_CM_STATE_CLOSED; 3060 irdma_rem_ref_cm_node(cm_node); 3061 return 0; 3062 } 3063 3064 if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) { 3065 irdma_rem_ref_cm_node(cm_node); 3066 return 0; 3067 } 3068 3069 ret = irdma_send_mpa_reject(cm_node, pdata, plen); 3070 if (!ret) 3071 return 0; 3072 3073 cm_node->state = IRDMA_CM_STATE_CLOSED; 3074 if (irdma_send_reset(cm_node)) 3075 ibdev_dbg(&cm_node->iwdev->ibdev, 3076 "CM: send reset failed\n"); 3077 3078 return ret; 3079 } 3080 3081 /** 3082 * irdma_cm_close - close of cm connection 3083 * @cm_node: connection's node 3084 */ 3085 static int irdma_cm_close(struct irdma_cm_node *cm_node) 3086 { 3087 switch (cm_node->state) { 3088 case IRDMA_CM_STATE_SYN_RCVD: 3089 case IRDMA_CM_STATE_SYN_SENT: 3090 case IRDMA_CM_STATE_ONE_SIDE_ESTABLISHED: 3091 case IRDMA_CM_STATE_ESTABLISHED: 3092 case IRDMA_CM_STATE_ACCEPTING: 3093 case IRDMA_CM_STATE_MPAREQ_SENT: 3094 case IRDMA_CM_STATE_MPAREQ_RCVD: 3095 irdma_cleanup_retrans_entry(cm_node); 3096 irdma_send_reset(cm_node); 3097 break; 3098 case IRDMA_CM_STATE_CLOSE_WAIT: 3099 cm_node->state = IRDMA_CM_STATE_LAST_ACK; 3100 irdma_send_fin(cm_node); 3101 break; 3102 case IRDMA_CM_STATE_FIN_WAIT1: 3103 case IRDMA_CM_STATE_FIN_WAIT2: 3104 case IRDMA_CM_STATE_LAST_ACK: 3105 case IRDMA_CM_STATE_TIME_WAIT: 3106 case IRDMA_CM_STATE_CLOSING: 3107 return -EINVAL; 3108 case IRDMA_CM_STATE_LISTENING: 3109 irdma_cleanup_retrans_entry(cm_node); 3110 irdma_send_reset(cm_node); 3111 break; 3112 case IRDMA_CM_STATE_MPAREJ_RCVD: 3113 case IRDMA_CM_STATE_UNKNOWN: 3114 case IRDMA_CM_STATE_INITED: 3115 case IRDMA_CM_STATE_CLOSED: 3116 case IRDMA_CM_STATE_LISTENER_DESTROYED: 3117 irdma_rem_ref_cm_node(cm_node); 3118 break; 3119 case IRDMA_CM_STATE_OFFLOADED: 3120 if (cm_node->send_entry) 3121 ibdev_dbg(&cm_node->iwdev->ibdev, 3122 "CM: CM send_entry in OFFLOADED state\n"); 3123 irdma_rem_ref_cm_node(cm_node); 3124 break; 3125 } 3126 3127 return 0; 3128 } 3129 3130 /** 3131 * irdma_receive_ilq - recv an ETHERNET packet, and process it 3132 * through CM 3133 * @vsi: VSI structure of dev 3134 * @rbuf: receive buffer 3135 */ 3136 void irdma_receive_ilq(struct irdma_sc_vsi *vsi, struct irdma_puda_buf *rbuf) 3137 { 3138 struct irdma_cm_node *cm_node; 3139 struct irdma_cm_listener *listener; 3140 struct iphdr *iph; 3141 struct ipv6hdr *ip6h; 3142 struct tcphdr *tcph; 3143 struct irdma_cm_info cm_info = {}; 3144 struct irdma_device *iwdev = vsi->back_vsi; 3145 struct irdma_cm_core *cm_core = &iwdev->cm_core; 3146 struct vlan_ethhdr *ethh; 3147 u16 vtag; 3148 3149 /* if vlan, then maclen = 18 else 14 */ 3150 iph = (struct iphdr *)rbuf->iph; 3151 print_hex_dump_debug("ILQ: RECEIVE ILQ BUFFER", DUMP_PREFIX_OFFSET, 3152 16, 8, rbuf->mem.va, rbuf->totallen, false); 3153 if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) { 3154 if (rbuf->vlan_valid) { 3155 vtag = rbuf->vlan_id; 3156 cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >> 3157 VLAN_PRIO_SHIFT; 3158 cm_info.vlan_id = vtag & VLAN_VID_MASK; 3159 } else { 3160 cm_info.vlan_id = 0xFFFF; 3161 } 3162 } else { 3163 ethh = rbuf->mem.va; 3164 3165 if (ethh->h_vlan_proto == htons(ETH_P_8021Q)) { 3166 vtag = ntohs(ethh->h_vlan_TCI); 3167 cm_info.user_pri = (vtag & VLAN_PRIO_MASK) >> 3168 VLAN_PRIO_SHIFT; 3169 cm_info.vlan_id = vtag & VLAN_VID_MASK; 3170 ibdev_dbg(&cm_core->iwdev->ibdev, 3171 "CM: vlan_id=%d\n", cm_info.vlan_id); 3172 } else { 3173 cm_info.vlan_id = 0xFFFF; 3174 } 3175 } 3176 tcph = (struct tcphdr *)rbuf->tcph; 3177 3178 if (rbuf->ipv4) { 3179 cm_info.loc_addr[0] = ntohl(iph->daddr); 3180 cm_info.rem_addr[0] = ntohl(iph->saddr); 3181 cm_info.ipv4 = true; 3182 cm_info.tos = iph->tos; 3183 } else { 3184 ip6h = (struct ipv6hdr *)rbuf->iph; 3185 irdma_copy_ip_ntohl(cm_info.loc_addr, 3186 ip6h->daddr.in6_u.u6_addr32); 3187 irdma_copy_ip_ntohl(cm_info.rem_addr, 3188 ip6h->saddr.in6_u.u6_addr32); 3189 cm_info.ipv4 = false; 3190 cm_info.tos = (ip6h->priority << 4) | (ip6h->flow_lbl[0] >> 4); 3191 } 3192 cm_info.loc_port = ntohs(tcph->dest); 3193 cm_info.rem_port = ntohs(tcph->source); 3194 cm_node = irdma_find_node(cm_core, cm_info.rem_port, cm_info.rem_addr, 3195 cm_info.loc_port, cm_info.loc_addr, cm_info.vlan_id); 3196 3197 if (!cm_node) { 3198 /* Only type of packet accepted are for the 3199 * PASSIVE open (syn only) 3200 */ 3201 if (!tcph->syn || tcph->ack) 3202 return; 3203 3204 listener = irdma_find_listener(cm_core, 3205 cm_info.loc_addr, 3206 cm_info.ipv4, 3207 cm_info.loc_port, 3208 cm_info.vlan_id, 3209 IRDMA_CM_LISTENER_ACTIVE_STATE); 3210 if (!listener) { 3211 cm_info.cm_id = NULL; 3212 ibdev_dbg(&cm_core->iwdev->ibdev, 3213 "CM: no listener found\n"); 3214 return; 3215 } 3216 3217 cm_info.cm_id = listener->cm_id; 3218 cm_node = irdma_make_cm_node(cm_core, iwdev, &cm_info, 3219 listener); 3220 if (!cm_node) { 3221 ibdev_dbg(&cm_core->iwdev->ibdev, 3222 "CM: allocate node failed\n"); 3223 refcount_dec(&listener->refcnt); 3224 return; 3225 } 3226 3227 if (!tcph->rst && !tcph->fin) { 3228 cm_node->state = IRDMA_CM_STATE_LISTENING; 3229 } else { 3230 irdma_rem_ref_cm_node(cm_node); 3231 return; 3232 } 3233 3234 refcount_inc(&cm_node->refcnt); 3235 } else if (cm_node->state == IRDMA_CM_STATE_OFFLOADED) { 3236 irdma_rem_ref_cm_node(cm_node); 3237 return; 3238 } 3239 3240 irdma_process_pkt(cm_node, rbuf); 3241 irdma_rem_ref_cm_node(cm_node); 3242 } 3243 3244 static int irdma_add_qh(struct irdma_cm_node *cm_node, bool active) 3245 { 3246 if (!active) 3247 irdma_add_conn_est_qh(cm_node); 3248 return 0; 3249 } 3250 3251 static void irdma_cm_free_ah_nop(struct irdma_cm_node *cm_node) 3252 { 3253 } 3254 3255 /** 3256 * irdma_setup_cm_core - setup top level instance of a cm core 3257 * @iwdev: iwarp device structure 3258 * @rdma_ver: HW version 3259 */ 3260 int irdma_setup_cm_core(struct irdma_device *iwdev, u8 rdma_ver) 3261 { 3262 struct irdma_cm_core *cm_core = &iwdev->cm_core; 3263 3264 cm_core->iwdev = iwdev; 3265 cm_core->dev = &iwdev->rf->sc_dev; 3266 3267 /* Handles CM event work items send to Iwarp core */ 3268 cm_core->event_wq = alloc_ordered_workqueue("iwarp-event-wq", 0); 3269 if (!cm_core->event_wq) 3270 return -ENOMEM; 3271 3272 INIT_LIST_HEAD(&cm_core->listen_list); 3273 3274 timer_setup(&cm_core->tcp_timer, irdma_cm_timer_tick, 0); 3275 3276 spin_lock_init(&cm_core->ht_lock); 3277 spin_lock_init(&cm_core->listen_list_lock); 3278 spin_lock_init(&cm_core->apbvt_lock); 3279 switch (rdma_ver) { 3280 case IRDMA_GEN_1: 3281 cm_core->form_cm_frame = irdma_form_uda_cm_frame; 3282 cm_core->cm_create_ah = irdma_add_qh; 3283 cm_core->cm_free_ah = irdma_cm_free_ah_nop; 3284 break; 3285 case IRDMA_GEN_2: 3286 default: 3287 cm_core->form_cm_frame = irdma_form_ah_cm_frame; 3288 cm_core->cm_create_ah = irdma_cm_create_ah; 3289 cm_core->cm_free_ah = irdma_cm_free_ah; 3290 } 3291 3292 return 0; 3293 } 3294 3295 /** 3296 * irdma_cleanup_cm_core - deallocate a top level instance of a 3297 * cm core 3298 * @cm_core: cm's core 3299 */ 3300 void irdma_cleanup_cm_core(struct irdma_cm_core *cm_core) 3301 { 3302 if (!cm_core) 3303 return; 3304 3305 del_timer_sync(&cm_core->tcp_timer); 3306 3307 destroy_workqueue(cm_core->event_wq); 3308 cm_core->dev->ws_reset(&cm_core->iwdev->vsi); 3309 } 3310 3311 /** 3312 * irdma_init_tcp_ctx - setup qp context 3313 * @cm_node: connection's node 3314 * @tcp_info: offload info for tcp 3315 * @iwqp: associate qp for the connection 3316 */ 3317 static void irdma_init_tcp_ctx(struct irdma_cm_node *cm_node, 3318 struct irdma_tcp_offload_info *tcp_info, 3319 struct irdma_qp *iwqp) 3320 { 3321 tcp_info->ipv4 = cm_node->ipv4; 3322 tcp_info->drop_ooo_seg = !iwqp->iwdev->iw_ooo; 3323 tcp_info->wscale = true; 3324 tcp_info->ignore_tcp_opt = true; 3325 tcp_info->ignore_tcp_uns_opt = true; 3326 tcp_info->no_nagle = false; 3327 3328 tcp_info->ttl = IRDMA_DEFAULT_TTL; 3329 tcp_info->rtt_var = IRDMA_DEFAULT_RTT_VAR; 3330 tcp_info->ss_thresh = IRDMA_DEFAULT_SS_THRESH; 3331 tcp_info->rexmit_thresh = IRDMA_DEFAULT_REXMIT_THRESH; 3332 3333 tcp_info->tcp_state = IRDMA_TCP_STATE_ESTABLISHED; 3334 tcp_info->snd_wscale = cm_node->tcp_cntxt.snd_wscale; 3335 tcp_info->rcv_wscale = cm_node->tcp_cntxt.rcv_wscale; 3336 3337 tcp_info->snd_nxt = cm_node->tcp_cntxt.loc_seq_num; 3338 tcp_info->snd_wnd = cm_node->tcp_cntxt.snd_wnd; 3339 tcp_info->rcv_nxt = cm_node->tcp_cntxt.rcv_nxt; 3340 tcp_info->snd_max = cm_node->tcp_cntxt.loc_seq_num; 3341 3342 tcp_info->snd_una = cm_node->tcp_cntxt.loc_seq_num; 3343 tcp_info->cwnd = 2 * cm_node->tcp_cntxt.mss; 3344 tcp_info->snd_wl1 = cm_node->tcp_cntxt.rcv_nxt; 3345 tcp_info->snd_wl2 = cm_node->tcp_cntxt.loc_seq_num; 3346 tcp_info->max_snd_window = cm_node->tcp_cntxt.max_snd_wnd; 3347 tcp_info->rcv_wnd = cm_node->tcp_cntxt.rcv_wnd 3348 << cm_node->tcp_cntxt.rcv_wscale; 3349 3350 tcp_info->flow_label = 0; 3351 tcp_info->snd_mss = (u32)cm_node->tcp_cntxt.mss; 3352 tcp_info->tos = cm_node->tos; 3353 if (cm_node->vlan_id < VLAN_N_VID) { 3354 tcp_info->insert_vlan_tag = true; 3355 tcp_info->vlan_tag = cm_node->vlan_id; 3356 tcp_info->vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT; 3357 } 3358 if (cm_node->ipv4) { 3359 tcp_info->src_port = cm_node->loc_port; 3360 tcp_info->dst_port = cm_node->rem_port; 3361 3362 tcp_info->dest_ip_addr[3] = cm_node->rem_addr[0]; 3363 tcp_info->local_ipaddr[3] = cm_node->loc_addr[0]; 3364 tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf, 3365 &tcp_info->dest_ip_addr[3], 3366 true, NULL, 3367 IRDMA_ARP_RESOLVE); 3368 } else { 3369 tcp_info->src_port = cm_node->loc_port; 3370 tcp_info->dst_port = cm_node->rem_port; 3371 memcpy(tcp_info->dest_ip_addr, cm_node->rem_addr, 3372 sizeof(tcp_info->dest_ip_addr)); 3373 memcpy(tcp_info->local_ipaddr, cm_node->loc_addr, 3374 sizeof(tcp_info->local_ipaddr)); 3375 3376 tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf, 3377 &tcp_info->dest_ip_addr[0], 3378 false, NULL, 3379 IRDMA_ARP_RESOLVE); 3380 } 3381 } 3382 3383 /** 3384 * irdma_cm_init_tsa_conn - setup qp for RTS 3385 * @iwqp: associate qp for the connection 3386 * @cm_node: connection's node 3387 */ 3388 static void irdma_cm_init_tsa_conn(struct irdma_qp *iwqp, 3389 struct irdma_cm_node *cm_node) 3390 { 3391 struct irdma_iwarp_offload_info *iwarp_info; 3392 struct irdma_qp_host_ctx_info *ctx_info; 3393 3394 iwarp_info = &iwqp->iwarp_info; 3395 ctx_info = &iwqp->ctx_info; 3396 3397 ctx_info->tcp_info = &iwqp->tcp_info; 3398 ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id; 3399 ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id; 3400 3401 iwarp_info->ord_size = cm_node->ord_size; 3402 iwarp_info->ird_size = cm_node->ird_size; 3403 iwarp_info->rd_en = true; 3404 iwarp_info->rdmap_ver = 1; 3405 iwarp_info->ddp_ver = 1; 3406 iwarp_info->pd_id = iwqp->iwpd->sc_pd.pd_id; 3407 3408 ctx_info->tcp_info_valid = true; 3409 ctx_info->iwarp_info_valid = true; 3410 ctx_info->user_pri = cm_node->user_pri; 3411 3412 irdma_init_tcp_ctx(cm_node, &iwqp->tcp_info, iwqp); 3413 if (cm_node->snd_mark_en) { 3414 iwarp_info->snd_mark_en = true; 3415 iwarp_info->snd_mark_offset = (iwqp->tcp_info.snd_nxt & SNDMARKER_SEQNMASK) + 3416 cm_node->lsmm_size; 3417 } 3418 3419 cm_node->state = IRDMA_CM_STATE_OFFLOADED; 3420 iwqp->tcp_info.tcp_state = IRDMA_TCP_STATE_ESTABLISHED; 3421 iwqp->tcp_info.src_mac_addr_idx = iwqp->iwdev->mac_ip_table_idx; 3422 3423 if (cm_node->rcv_mark_en) { 3424 iwarp_info->rcv_mark_en = true; 3425 iwarp_info->align_hdrs = true; 3426 } 3427 3428 irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info); 3429 3430 /* once tcp_info is set, no need to do it again */ 3431 ctx_info->tcp_info_valid = false; 3432 ctx_info->iwarp_info_valid = false; 3433 } 3434 3435 /** 3436 * irdma_cm_disconn - when a connection is being closed 3437 * @iwqp: associated qp for the connection 3438 */ 3439 void irdma_cm_disconn(struct irdma_qp *iwqp) 3440 { 3441 struct irdma_device *iwdev = iwqp->iwdev; 3442 struct disconn_work *work; 3443 unsigned long flags; 3444 3445 work = kzalloc(sizeof(*work), GFP_ATOMIC); 3446 if (!work) 3447 return; 3448 3449 spin_lock_irqsave(&iwdev->rf->qptable_lock, flags); 3450 if (!iwdev->rf->qp_table[iwqp->ibqp.qp_num]) { 3451 spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags); 3452 ibdev_dbg(&iwdev->ibdev, 3453 "CM: qp_id %d is already freed\n", 3454 iwqp->ibqp.qp_num); 3455 kfree(work); 3456 return; 3457 } 3458 irdma_qp_add_ref(&iwqp->ibqp); 3459 spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags); 3460 3461 work->iwqp = iwqp; 3462 INIT_WORK(&work->work, irdma_disconnect_worker); 3463 queue_work(iwdev->cleanup_wq, &work->work); 3464 } 3465 3466 /** 3467 * irdma_qp_disconnect - free qp and close cm 3468 * @iwqp: associate qp for the connection 3469 */ 3470 static void irdma_qp_disconnect(struct irdma_qp *iwqp) 3471 { 3472 struct irdma_device *iwdev = iwqp->iwdev; 3473 3474 iwqp->active_conn = 0; 3475 /* close the CM node down if it is still active */ 3476 ibdev_dbg(&iwdev->ibdev, "CM: Call close API\n"); 3477 irdma_cm_close(iwqp->cm_node); 3478 } 3479 3480 /** 3481 * irdma_cm_disconn_true - called by worker thread to disconnect qp 3482 * @iwqp: associate qp for the connection 3483 */ 3484 static void irdma_cm_disconn_true(struct irdma_qp *iwqp) 3485 { 3486 struct iw_cm_id *cm_id; 3487 struct irdma_device *iwdev; 3488 struct irdma_sc_qp *qp = &iwqp->sc_qp; 3489 u16 last_ae; 3490 u8 original_hw_tcp_state; 3491 u8 original_ibqp_state; 3492 int disconn_status = 0; 3493 int issue_disconn = 0; 3494 int issue_close = 0; 3495 int issue_flush = 0; 3496 unsigned long flags; 3497 int err; 3498 3499 iwdev = iwqp->iwdev; 3500 spin_lock_irqsave(&iwqp->lock, flags); 3501 if (rdma_protocol_roce(&iwdev->ibdev, 1)) { 3502 struct ib_qp_attr attr; 3503 3504 if (iwqp->flush_issued || iwqp->sc_qp.qp_uk.destroy_pending) { 3505 spin_unlock_irqrestore(&iwqp->lock, flags); 3506 return; 3507 } 3508 3509 spin_unlock_irqrestore(&iwqp->lock, flags); 3510 3511 attr.qp_state = IB_QPS_ERR; 3512 irdma_modify_qp_roce(&iwqp->ibqp, &attr, IB_QP_STATE, NULL); 3513 irdma_ib_qp_event(iwqp, qp->event_type); 3514 return; 3515 } 3516 3517 cm_id = iwqp->cm_id; 3518 original_hw_tcp_state = iwqp->hw_tcp_state; 3519 original_ibqp_state = iwqp->ibqp_state; 3520 last_ae = iwqp->last_aeq; 3521 3522 if (qp->term_flags) { 3523 issue_disconn = 1; 3524 issue_close = 1; 3525 iwqp->cm_id = NULL; 3526 irdma_terminate_del_timer(qp); 3527 if (!iwqp->flush_issued) { 3528 iwqp->flush_issued = 1; 3529 issue_flush = 1; 3530 } 3531 } else if ((original_hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT) || 3532 ((original_ibqp_state == IB_QPS_RTS) && 3533 (last_ae == IRDMA_AE_LLP_CONNECTION_RESET))) { 3534 issue_disconn = 1; 3535 if (last_ae == IRDMA_AE_LLP_CONNECTION_RESET) 3536 disconn_status = -ECONNRESET; 3537 } 3538 3539 if (original_hw_tcp_state == IRDMA_TCP_STATE_CLOSED || 3540 original_hw_tcp_state == IRDMA_TCP_STATE_TIME_WAIT || 3541 last_ae == IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE || 3542 last_ae == IRDMA_AE_BAD_CLOSE || 3543 last_ae == IRDMA_AE_LLP_CONNECTION_RESET || iwdev->rf->reset || !cm_id) { 3544 issue_close = 1; 3545 iwqp->cm_id = NULL; 3546 qp->term_flags = 0; 3547 if (!iwqp->flush_issued) { 3548 iwqp->flush_issued = 1; 3549 issue_flush = 1; 3550 } 3551 } 3552 3553 spin_unlock_irqrestore(&iwqp->lock, flags); 3554 if (issue_flush && !iwqp->sc_qp.qp_uk.destroy_pending) { 3555 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_FLUSH_RQ | 3556 IRDMA_FLUSH_WAIT); 3557 3558 if (qp->term_flags) 3559 irdma_ib_qp_event(iwqp, qp->event_type); 3560 } 3561 3562 if (!cm_id || !cm_id->event_handler) 3563 return; 3564 3565 spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags); 3566 if (!iwqp->cm_node) { 3567 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags); 3568 return; 3569 } 3570 refcount_inc(&iwqp->cm_node->refcnt); 3571 3572 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags); 3573 3574 if (issue_disconn) { 3575 err = irdma_send_cm_event(iwqp->cm_node, cm_id, 3576 IW_CM_EVENT_DISCONNECT, 3577 disconn_status); 3578 if (err) 3579 ibdev_dbg(&iwdev->ibdev, 3580 "CM: disconnect event failed: - cm_id = %p\n", 3581 cm_id); 3582 } 3583 if (issue_close) { 3584 cm_id->provider_data = iwqp; 3585 err = irdma_send_cm_event(iwqp->cm_node, cm_id, 3586 IW_CM_EVENT_CLOSE, 0); 3587 if (err) 3588 ibdev_dbg(&iwdev->ibdev, 3589 "CM: close event failed: - cm_id = %p\n", 3590 cm_id); 3591 irdma_qp_disconnect(iwqp); 3592 } 3593 irdma_rem_ref_cm_node(iwqp->cm_node); 3594 } 3595 3596 /** 3597 * irdma_disconnect_worker - worker for connection close 3598 * @work: points or disconn structure 3599 */ 3600 static void irdma_disconnect_worker(struct work_struct *work) 3601 { 3602 struct disconn_work *dwork = container_of(work, struct disconn_work, work); 3603 struct irdma_qp *iwqp = dwork->iwqp; 3604 3605 kfree(dwork); 3606 irdma_cm_disconn_true(iwqp); 3607 irdma_qp_rem_ref(&iwqp->ibqp); 3608 } 3609 3610 /** 3611 * irdma_free_lsmm_rsrc - free lsmm memory and deregister 3612 * @iwqp: associate qp for the connection 3613 */ 3614 void irdma_free_lsmm_rsrc(struct irdma_qp *iwqp) 3615 { 3616 struct irdma_device *iwdev; 3617 3618 iwdev = iwqp->iwdev; 3619 3620 if (iwqp->ietf_mem.va) { 3621 if (iwqp->lsmm_mr) 3622 iwdev->ibdev.ops.dereg_mr(iwqp->lsmm_mr, NULL); 3623 dma_free_coherent(iwdev->rf->sc_dev.hw->device, 3624 iwqp->ietf_mem.size, iwqp->ietf_mem.va, 3625 iwqp->ietf_mem.pa); 3626 iwqp->ietf_mem.va = NULL; 3627 } 3628 } 3629 3630 /** 3631 * irdma_accept - registered call for connection to be accepted 3632 * @cm_id: cm information for passive connection 3633 * @conn_param: accpet parameters 3634 */ 3635 int irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3636 { 3637 struct ib_qp *ibqp; 3638 struct irdma_qp *iwqp; 3639 struct irdma_device *iwdev; 3640 struct irdma_sc_dev *dev; 3641 struct irdma_cm_node *cm_node; 3642 struct ib_qp_attr attr = {}; 3643 int passive_state; 3644 struct ib_mr *ibmr; 3645 struct irdma_pd *iwpd; 3646 u16 buf_len = 0; 3647 struct irdma_kmem_info accept; 3648 u64 tagged_offset; 3649 int wait_ret; 3650 int ret = 0; 3651 3652 ibqp = irdma_get_qp(cm_id->device, conn_param->qpn); 3653 if (!ibqp) 3654 return -EINVAL; 3655 3656 iwqp = to_iwqp(ibqp); 3657 iwdev = iwqp->iwdev; 3658 dev = &iwdev->rf->sc_dev; 3659 cm_node = cm_id->provider_data; 3660 3661 if (((struct sockaddr_in *)&cm_id->local_addr)->sin_family == AF_INET) { 3662 cm_node->ipv4 = true; 3663 cm_node->vlan_id = irdma_get_vlan_ipv4(cm_node->loc_addr); 3664 } else { 3665 cm_node->ipv4 = false; 3666 irdma_get_vlan_mac_ipv6(cm_node->loc_addr, &cm_node->vlan_id, 3667 NULL); 3668 } 3669 ibdev_dbg(&iwdev->ibdev, "CM: Accept vlan_id=%d\n", 3670 cm_node->vlan_id); 3671 3672 trace_irdma_accept(cm_node, 0, NULL); 3673 3674 if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) { 3675 ret = -EINVAL; 3676 goto error; 3677 } 3678 3679 passive_state = atomic_add_return(1, &cm_node->passive_state); 3680 if (passive_state == IRDMA_SEND_RESET_EVENT) { 3681 ret = -ECONNRESET; 3682 goto error; 3683 } 3684 3685 buf_len = conn_param->private_data_len + IRDMA_MAX_IETF_SIZE; 3686 iwqp->ietf_mem.size = ALIGN(buf_len, 1); 3687 iwqp->ietf_mem.va = dma_alloc_coherent(dev->hw->device, 3688 iwqp->ietf_mem.size, 3689 &iwqp->ietf_mem.pa, GFP_KERNEL); 3690 if (!iwqp->ietf_mem.va) { 3691 ret = -ENOMEM; 3692 goto error; 3693 } 3694 3695 cm_node->pdata.size = conn_param->private_data_len; 3696 accept.addr = iwqp->ietf_mem.va; 3697 accept.size = irdma_cm_build_mpa_frame(cm_node, &accept, MPA_KEY_REPLY); 3698 memcpy((u8 *)accept.addr + accept.size, conn_param->private_data, 3699 conn_param->private_data_len); 3700 3701 if (cm_node->dev->ws_add(iwqp->sc_qp.vsi, cm_node->user_pri)) { 3702 ret = -ENOMEM; 3703 goto error; 3704 } 3705 iwqp->sc_qp.user_pri = cm_node->user_pri; 3706 irdma_qp_add_qos(&iwqp->sc_qp); 3707 /* setup our first outgoing iWarp send WQE (the IETF frame response) */ 3708 iwpd = iwqp->iwpd; 3709 tagged_offset = (uintptr_t)iwqp->ietf_mem.va; 3710 ibmr = irdma_reg_phys_mr(&iwpd->ibpd, iwqp->ietf_mem.pa, buf_len, 3711 IB_ACCESS_LOCAL_WRITE, &tagged_offset); 3712 if (IS_ERR(ibmr)) { 3713 ret = -ENOMEM; 3714 goto error; 3715 } 3716 3717 ibmr->pd = &iwpd->ibpd; 3718 ibmr->device = iwpd->ibpd.device; 3719 iwqp->lsmm_mr = ibmr; 3720 if (iwqp->page) 3721 iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page); 3722 3723 cm_node->lsmm_size = accept.size + conn_param->private_data_len; 3724 irdma_sc_send_lsmm(&iwqp->sc_qp, iwqp->ietf_mem.va, cm_node->lsmm_size, 3725 ibmr->lkey); 3726 3727 if (iwqp->page) 3728 kunmap_local(iwqp->sc_qp.qp_uk.sq_base); 3729 3730 iwqp->cm_id = cm_id; 3731 cm_node->cm_id = cm_id; 3732 3733 cm_id->provider_data = iwqp; 3734 iwqp->active_conn = 0; 3735 iwqp->cm_node = cm_node; 3736 cm_node->iwqp = iwqp; 3737 irdma_cm_init_tsa_conn(iwqp, cm_node); 3738 irdma_qp_add_ref(&iwqp->ibqp); 3739 cm_id->add_ref(cm_id); 3740 3741 attr.qp_state = IB_QPS_RTS; 3742 cm_node->qhash_set = false; 3743 cm_node->cm_core->cm_free_ah(cm_node); 3744 3745 irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL); 3746 if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) { 3747 wait_ret = wait_event_interruptible_timeout(iwqp->waitq, 3748 iwqp->rts_ae_rcvd, 3749 IRDMA_MAX_TIMEOUT); 3750 if (!wait_ret) { 3751 ibdev_dbg(&iwdev->ibdev, 3752 "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n", 3753 cm_node, cm_node->loc_port, 3754 cm_node->rem_port, cm_node->cm_id); 3755 ret = -ECONNRESET; 3756 goto error; 3757 } 3758 } 3759 3760 irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_ESTABLISHED, 0); 3761 cm_node->accelerated = true; 3762 complete(&cm_node->establish_comp); 3763 3764 if (cm_node->accept_pend) { 3765 atomic_dec(&cm_node->listener->pend_accepts_cnt); 3766 cm_node->accept_pend = 0; 3767 } 3768 3769 ibdev_dbg(&iwdev->ibdev, 3770 "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n", 3771 cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr, 3772 cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num); 3773 cm_node->cm_core->stats_accepts++; 3774 3775 return 0; 3776 error: 3777 irdma_free_lsmm_rsrc(iwqp); 3778 irdma_rem_ref_cm_node(cm_node); 3779 3780 return ret; 3781 } 3782 3783 /** 3784 * irdma_reject - registered call for connection to be rejected 3785 * @cm_id: cm information for passive connection 3786 * @pdata: private data to be sent 3787 * @pdata_len: private data length 3788 */ 3789 int irdma_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len) 3790 { 3791 struct irdma_device *iwdev; 3792 struct irdma_cm_node *cm_node; 3793 3794 cm_node = cm_id->provider_data; 3795 cm_node->pdata.size = pdata_len; 3796 3797 trace_irdma_reject(cm_node, 0, NULL); 3798 3799 iwdev = to_iwdev(cm_id->device); 3800 if (!iwdev) 3801 return -EINVAL; 3802 3803 cm_node->cm_core->stats_rejects++; 3804 3805 if (pdata_len + sizeof(struct ietf_mpa_v2) > IRDMA_MAX_CM_BUF) 3806 return -EINVAL; 3807 3808 return irdma_cm_reject(cm_node, pdata, pdata_len); 3809 } 3810 3811 /** 3812 * irdma_connect - registered call for connection to be established 3813 * @cm_id: cm information for passive connection 3814 * @conn_param: Information about the connection 3815 */ 3816 int irdma_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param) 3817 { 3818 struct ib_qp *ibqp; 3819 struct irdma_qp *iwqp; 3820 struct irdma_device *iwdev; 3821 struct irdma_cm_node *cm_node; 3822 struct irdma_cm_info cm_info; 3823 struct sockaddr_in *laddr; 3824 struct sockaddr_in *raddr; 3825 struct sockaddr_in6 *laddr6; 3826 struct sockaddr_in6 *raddr6; 3827 int ret = 0; 3828 3829 ibqp = irdma_get_qp(cm_id->device, conn_param->qpn); 3830 if (!ibqp) 3831 return -EINVAL; 3832 iwqp = to_iwqp(ibqp); 3833 if (!iwqp) 3834 return -EINVAL; 3835 iwdev = iwqp->iwdev; 3836 if (!iwdev) 3837 return -EINVAL; 3838 3839 laddr = (struct sockaddr_in *)&cm_id->m_local_addr; 3840 raddr = (struct sockaddr_in *)&cm_id->m_remote_addr; 3841 laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr; 3842 raddr6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr; 3843 3844 if (!(laddr->sin_port) || !(raddr->sin_port)) 3845 return -EINVAL; 3846 3847 iwqp->active_conn = 1; 3848 iwqp->cm_id = NULL; 3849 cm_id->provider_data = iwqp; 3850 3851 /* set up the connection params for the node */ 3852 if (cm_id->remote_addr.ss_family == AF_INET) { 3853 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4) 3854 return -EINVAL; 3855 3856 cm_info.ipv4 = true; 3857 memset(cm_info.loc_addr, 0, sizeof(cm_info.loc_addr)); 3858 memset(cm_info.rem_addr, 0, sizeof(cm_info.rem_addr)); 3859 cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr); 3860 cm_info.rem_addr[0] = ntohl(raddr->sin_addr.s_addr); 3861 cm_info.loc_port = ntohs(laddr->sin_port); 3862 cm_info.rem_port = ntohs(raddr->sin_port); 3863 cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr); 3864 } else { 3865 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6) 3866 return -EINVAL; 3867 3868 cm_info.ipv4 = false; 3869 irdma_copy_ip_ntohl(cm_info.loc_addr, 3870 laddr6->sin6_addr.in6_u.u6_addr32); 3871 irdma_copy_ip_ntohl(cm_info.rem_addr, 3872 raddr6->sin6_addr.in6_u.u6_addr32); 3873 cm_info.loc_port = ntohs(laddr6->sin6_port); 3874 cm_info.rem_port = ntohs(raddr6->sin6_port); 3875 irdma_get_vlan_mac_ipv6(cm_info.loc_addr, &cm_info.vlan_id, 3876 NULL); 3877 } 3878 cm_info.cm_id = cm_id; 3879 cm_info.qh_qpid = iwdev->vsi.ilq->qp_id; 3880 cm_info.tos = cm_id->tos; 3881 if (iwdev->vsi.dscp_mode) { 3882 cm_info.user_pri = 3883 iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(cm_info.tos)]; 3884 } else { 3885 cm_info.user_pri = rt_tos2priority(cm_id->tos); 3886 cm_info.user_pri = irdma_iw_get_vlan_prio(cm_info.loc_addr, 3887 cm_info.user_pri, 3888 cm_info.ipv4); 3889 } 3890 3891 if (iwqp->sc_qp.dev->ws_add(iwqp->sc_qp.vsi, cm_info.user_pri)) 3892 return -ENOMEM; 3893 iwqp->sc_qp.user_pri = cm_info.user_pri; 3894 irdma_qp_add_qos(&iwqp->sc_qp); 3895 ibdev_dbg(&iwdev->ibdev, "DCB: TOS:[%d] UP:[%d]\n", cm_id->tos, 3896 cm_info.user_pri); 3897 3898 trace_irdma_dcb_tos(iwdev, cm_id->tos, cm_info.user_pri); 3899 3900 ret = irdma_create_cm_node(&iwdev->cm_core, iwdev, conn_param, &cm_info, 3901 &cm_node); 3902 if (ret) 3903 return ret; 3904 ret = cm_node->cm_core->cm_create_ah(cm_node, true); 3905 if (ret) 3906 goto err; 3907 if (irdma_manage_qhash(iwdev, &cm_info, 3908 IRDMA_QHASH_TYPE_TCP_ESTABLISHED, 3909 IRDMA_QHASH_MANAGE_TYPE_ADD, NULL, true)) { 3910 ret = -EINVAL; 3911 goto err; 3912 } 3913 cm_node->qhash_set = true; 3914 3915 cm_node->apbvt_entry = irdma_add_apbvt(iwdev, cm_info.loc_port); 3916 if (!cm_node->apbvt_entry) { 3917 ret = -EINVAL; 3918 goto err; 3919 } 3920 3921 cm_node->apbvt_set = true; 3922 iwqp->cm_node = cm_node; 3923 cm_node->iwqp = iwqp; 3924 iwqp->cm_id = cm_id; 3925 irdma_qp_add_ref(&iwqp->ibqp); 3926 cm_id->add_ref(cm_id); 3927 3928 if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) { 3929 cm_node->state = IRDMA_CM_STATE_SYN_SENT; 3930 ret = irdma_send_syn(cm_node, 0); 3931 if (ret) 3932 goto err; 3933 } 3934 3935 ibdev_dbg(&iwdev->ibdev, 3936 "CM: rem_port=0x%04x, loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4 cm_node=%p cm_id=%p qp_id = %d\n\n", 3937 cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr, 3938 cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num); 3939 3940 trace_irdma_connect(cm_node, 0, NULL); 3941 3942 return 0; 3943 3944 err: 3945 if (cm_info.ipv4) 3946 ibdev_dbg(&iwdev->ibdev, 3947 "CM: connect() FAILED: dest addr=%pI4", 3948 cm_info.rem_addr); 3949 else 3950 ibdev_dbg(&iwdev->ibdev, 3951 "CM: connect() FAILED: dest addr=%pI6", 3952 cm_info.rem_addr); 3953 irdma_rem_ref_cm_node(cm_node); 3954 iwdev->cm_core.stats_connect_errs++; 3955 3956 return ret; 3957 } 3958 3959 /** 3960 * irdma_create_listen - registered call creating listener 3961 * @cm_id: cm information for passive connection 3962 * @backlog: to max accept pending count 3963 */ 3964 int irdma_create_listen(struct iw_cm_id *cm_id, int backlog) 3965 { 3966 struct irdma_device *iwdev; 3967 struct irdma_cm_listener *cm_listen_node; 3968 struct irdma_cm_info cm_info = {}; 3969 struct sockaddr_in *laddr; 3970 struct sockaddr_in6 *laddr6; 3971 bool wildcard = false; 3972 int err; 3973 3974 iwdev = to_iwdev(cm_id->device); 3975 if (!iwdev) 3976 return -EINVAL; 3977 3978 laddr = (struct sockaddr_in *)&cm_id->m_local_addr; 3979 laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr; 3980 cm_info.qh_qpid = iwdev->vsi.ilq->qp_id; 3981 3982 if (laddr->sin_family == AF_INET) { 3983 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4) 3984 return -EINVAL; 3985 3986 cm_info.ipv4 = true; 3987 cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr); 3988 cm_info.loc_port = ntohs(laddr->sin_port); 3989 3990 if (laddr->sin_addr.s_addr != htonl(INADDR_ANY)) { 3991 cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr); 3992 } else { 3993 cm_info.vlan_id = 0xFFFF; 3994 wildcard = true; 3995 } 3996 } else { 3997 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6) 3998 return -EINVAL; 3999 4000 cm_info.ipv4 = false; 4001 irdma_copy_ip_ntohl(cm_info.loc_addr, 4002 laddr6->sin6_addr.in6_u.u6_addr32); 4003 cm_info.loc_port = ntohs(laddr6->sin6_port); 4004 if (ipv6_addr_type(&laddr6->sin6_addr) != IPV6_ADDR_ANY) { 4005 irdma_get_vlan_mac_ipv6(cm_info.loc_addr, 4006 &cm_info.vlan_id, NULL); 4007 } else { 4008 cm_info.vlan_id = 0xFFFF; 4009 wildcard = true; 4010 } 4011 } 4012 4013 if (cm_info.vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode) 4014 cm_info.vlan_id = 0; 4015 cm_info.backlog = backlog; 4016 cm_info.cm_id = cm_id; 4017 4018 trace_irdma_create_listen(iwdev, &cm_info); 4019 4020 cm_listen_node = irdma_make_listen_node(&iwdev->cm_core, iwdev, 4021 &cm_info); 4022 if (!cm_listen_node) { 4023 ibdev_dbg(&iwdev->ibdev, 4024 "CM: cm_listen_node == NULL\n"); 4025 return -ENOMEM; 4026 } 4027 4028 cm_id->provider_data = cm_listen_node; 4029 4030 cm_listen_node->tos = cm_id->tos; 4031 if (iwdev->vsi.dscp_mode) 4032 cm_listen_node->user_pri = 4033 iwdev->vsi.dscp_map[irdma_tos2dscp(cm_id->tos)]; 4034 else 4035 cm_listen_node->user_pri = rt_tos2priority(cm_id->tos); 4036 cm_info.user_pri = cm_listen_node->user_pri; 4037 if (!cm_listen_node->reused_node) { 4038 if (wildcard) { 4039 err = irdma_add_mqh(iwdev, &cm_info, cm_listen_node); 4040 if (err) 4041 goto error; 4042 } else { 4043 if (!iwdev->vsi.dscp_mode) 4044 cm_listen_node->user_pri = 4045 irdma_iw_get_vlan_prio(cm_info.loc_addr, 4046 cm_info.user_pri, 4047 cm_info.ipv4); 4048 cm_info.user_pri = cm_listen_node->user_pri; 4049 err = irdma_manage_qhash(iwdev, &cm_info, 4050 IRDMA_QHASH_TYPE_TCP_SYN, 4051 IRDMA_QHASH_MANAGE_TYPE_ADD, 4052 NULL, true); 4053 if (err) 4054 goto error; 4055 4056 cm_listen_node->qhash_set = true; 4057 } 4058 4059 cm_listen_node->apbvt_entry = irdma_add_apbvt(iwdev, 4060 cm_info.loc_port); 4061 if (!cm_listen_node->apbvt_entry) 4062 goto error; 4063 } 4064 cm_id->add_ref(cm_id); 4065 cm_listen_node->cm_core->stats_listen_created++; 4066 ibdev_dbg(&iwdev->ibdev, 4067 "CM: loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d\n", 4068 cm_listen_node->loc_port, cm_listen_node->loc_addr, 4069 cm_listen_node, cm_listen_node->cm_id, 4070 cm_listen_node->qhash_set, cm_listen_node->vlan_id); 4071 4072 return 0; 4073 4074 error: 4075 4076 irdma_cm_del_listen(&iwdev->cm_core, cm_listen_node, false); 4077 4078 return -EINVAL; 4079 } 4080 4081 /** 4082 * irdma_destroy_listen - registered call to destroy listener 4083 * @cm_id: cm information for passive connection 4084 */ 4085 int irdma_destroy_listen(struct iw_cm_id *cm_id) 4086 { 4087 struct irdma_device *iwdev; 4088 4089 iwdev = to_iwdev(cm_id->device); 4090 if (cm_id->provider_data) 4091 irdma_cm_del_listen(&iwdev->cm_core, cm_id->provider_data, 4092 true); 4093 else 4094 ibdev_dbg(&iwdev->ibdev, 4095 "CM: cm_id->provider_data was NULL\n"); 4096 4097 cm_id->rem_ref(cm_id); 4098 4099 return 0; 4100 } 4101 4102 /** 4103 * irdma_teardown_list_prep - add conn nodes slated for tear down to list 4104 * @cm_core: cm's core 4105 * @teardown_list: a list to which cm_node will be selected 4106 * @ipaddr: pointer to ip address 4107 * @nfo: pointer to cm_info structure instance 4108 * @disconnect_all: flag indicating disconnect all QPs 4109 */ 4110 static void irdma_teardown_list_prep(struct irdma_cm_core *cm_core, 4111 struct list_head *teardown_list, 4112 u32 *ipaddr, 4113 struct irdma_cm_info *nfo, 4114 bool disconnect_all) 4115 { 4116 struct irdma_cm_node *cm_node; 4117 int bkt; 4118 4119 hash_for_each_rcu(cm_core->cm_hash_tbl, bkt, cm_node, list) { 4120 if ((disconnect_all || 4121 (nfo->vlan_id == cm_node->vlan_id && 4122 !memcmp(cm_node->loc_addr, ipaddr, nfo->ipv4 ? 4 : 16))) && 4123 refcount_inc_not_zero(&cm_node->refcnt)) 4124 list_add(&cm_node->teardown_entry, teardown_list); 4125 } 4126 } 4127 4128 /** 4129 * irdma_cm_event_connected - handle connected active node 4130 * @event: the info for cm_node of connection 4131 */ 4132 static void irdma_cm_event_connected(struct irdma_cm_event *event) 4133 { 4134 struct irdma_qp *iwqp; 4135 struct irdma_device *iwdev; 4136 struct irdma_cm_node *cm_node; 4137 struct irdma_sc_dev *dev; 4138 struct ib_qp_attr attr = {}; 4139 struct iw_cm_id *cm_id; 4140 int status; 4141 bool read0; 4142 int wait_ret = 0; 4143 4144 cm_node = event->cm_node; 4145 cm_id = cm_node->cm_id; 4146 iwqp = cm_id->provider_data; 4147 iwdev = iwqp->iwdev; 4148 dev = &iwdev->rf->sc_dev; 4149 if (iwqp->sc_qp.qp_uk.destroy_pending) { 4150 status = -ETIMEDOUT; 4151 goto error; 4152 } 4153 4154 irdma_cm_init_tsa_conn(iwqp, cm_node); 4155 read0 = (cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO); 4156 if (iwqp->page) 4157 iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page); 4158 irdma_sc_send_rtt(&iwqp->sc_qp, read0); 4159 if (iwqp->page) 4160 kunmap_local(iwqp->sc_qp.qp_uk.sq_base); 4161 4162 attr.qp_state = IB_QPS_RTS; 4163 cm_node->qhash_set = false; 4164 irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL); 4165 if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) { 4166 wait_ret = wait_event_interruptible_timeout(iwqp->waitq, 4167 iwqp->rts_ae_rcvd, 4168 IRDMA_MAX_TIMEOUT); 4169 if (!wait_ret) 4170 ibdev_dbg(&iwdev->ibdev, 4171 "CM: Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n", 4172 cm_node, cm_node->loc_port, 4173 cm_node->rem_port, cm_node->cm_id); 4174 } 4175 4176 irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 0); 4177 cm_node->accelerated = true; 4178 complete(&cm_node->establish_comp); 4179 cm_node->cm_core->cm_free_ah(cm_node); 4180 return; 4181 4182 error: 4183 iwqp->cm_id = NULL; 4184 cm_id->provider_data = NULL; 4185 irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 4186 status); 4187 irdma_rem_ref_cm_node(event->cm_node); 4188 } 4189 4190 /** 4191 * irdma_cm_event_reset - handle reset 4192 * @event: the info for cm_node of connection 4193 */ 4194 static void irdma_cm_event_reset(struct irdma_cm_event *event) 4195 { 4196 struct irdma_cm_node *cm_node = event->cm_node; 4197 struct iw_cm_id *cm_id = cm_node->cm_id; 4198 struct irdma_qp *iwqp; 4199 4200 if (!cm_id) 4201 return; 4202 4203 iwqp = cm_id->provider_data; 4204 if (!iwqp) 4205 return; 4206 4207 ibdev_dbg(&cm_node->iwdev->ibdev, 4208 "CM: reset event %p - cm_id = %p\n", event->cm_node, cm_id); 4209 iwqp->cm_id = NULL; 4210 4211 irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_DISCONNECT, 4212 -ECONNRESET); 4213 irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_CLOSE, 0); 4214 } 4215 4216 /** 4217 * irdma_cm_event_handler - send event to cm upper layer 4218 * @work: pointer of cm event info. 4219 */ 4220 static void irdma_cm_event_handler(struct work_struct *work) 4221 { 4222 struct irdma_cm_event *event = container_of(work, struct irdma_cm_event, event_work); 4223 struct irdma_cm_node *cm_node; 4224 4225 if (!event || !event->cm_node || !event->cm_node->cm_core) 4226 return; 4227 4228 cm_node = event->cm_node; 4229 trace_irdma_cm_event_handler(cm_node, event->type, NULL); 4230 4231 switch (event->type) { 4232 case IRDMA_CM_EVENT_MPA_REQ: 4233 irdma_send_cm_event(cm_node, cm_node->cm_id, 4234 IW_CM_EVENT_CONNECT_REQUEST, 0); 4235 break; 4236 case IRDMA_CM_EVENT_RESET: 4237 irdma_cm_event_reset(event); 4238 break; 4239 case IRDMA_CM_EVENT_CONNECTED: 4240 if (!event->cm_node->cm_id || 4241 event->cm_node->state != IRDMA_CM_STATE_OFFLOADED) 4242 break; 4243 irdma_cm_event_connected(event); 4244 break; 4245 case IRDMA_CM_EVENT_MPA_REJECT: 4246 if (!event->cm_node->cm_id || 4247 cm_node->state == IRDMA_CM_STATE_OFFLOADED) 4248 break; 4249 irdma_send_cm_event(cm_node, cm_node->cm_id, 4250 IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED); 4251 break; 4252 case IRDMA_CM_EVENT_ABORTED: 4253 if (!event->cm_node->cm_id || 4254 event->cm_node->state == IRDMA_CM_STATE_OFFLOADED) 4255 break; 4256 irdma_event_connect_error(event); 4257 break; 4258 default: 4259 ibdev_dbg(&cm_node->iwdev->ibdev, 4260 "CM: bad event type = %d\n", event->type); 4261 break; 4262 } 4263 4264 irdma_rem_ref_cm_node(event->cm_node); 4265 kfree(event); 4266 } 4267 4268 /** 4269 * irdma_cm_post_event - queue event request for worker thread 4270 * @event: cm node's info for up event call 4271 */ 4272 static void irdma_cm_post_event(struct irdma_cm_event *event) 4273 { 4274 refcount_inc(&event->cm_node->refcnt); 4275 INIT_WORK(&event->event_work, irdma_cm_event_handler); 4276 queue_work(event->cm_node->cm_core->event_wq, &event->event_work); 4277 } 4278 4279 /** 4280 * irdma_cm_teardown_connections - teardown QPs 4281 * @iwdev: device pointer 4282 * @ipaddr: Pointer to IPv4 or IPv6 address 4283 * @nfo: Connection info 4284 * @disconnect_all: flag indicating disconnect all QPs 4285 * 4286 * teardown QPs where source or destination addr matches ip addr 4287 */ 4288 void irdma_cm_teardown_connections(struct irdma_device *iwdev, u32 *ipaddr, 4289 struct irdma_cm_info *nfo, 4290 bool disconnect_all) 4291 { 4292 struct irdma_cm_core *cm_core = &iwdev->cm_core; 4293 struct list_head *list_core_temp; 4294 struct list_head *list_node; 4295 struct irdma_cm_node *cm_node; 4296 struct list_head teardown_list; 4297 struct ib_qp_attr attr; 4298 4299 INIT_LIST_HEAD(&teardown_list); 4300 4301 rcu_read_lock(); 4302 irdma_teardown_list_prep(cm_core, &teardown_list, ipaddr, nfo, disconnect_all); 4303 rcu_read_unlock(); 4304 4305 list_for_each_safe (list_node, list_core_temp, &teardown_list) { 4306 cm_node = container_of(list_node, struct irdma_cm_node, 4307 teardown_entry); 4308 attr.qp_state = IB_QPS_ERR; 4309 irdma_modify_qp(&cm_node->iwqp->ibqp, &attr, IB_QP_STATE, NULL); 4310 if (iwdev->rf->reset) 4311 irdma_cm_disconn(cm_node->iwqp); 4312 irdma_rem_ref_cm_node(cm_node); 4313 } 4314 } 4315 4316 /** 4317 * irdma_qhash_ctrl - enable/disable qhash for list 4318 * @iwdev: device pointer 4319 * @parent_listen_node: parent listen node 4320 * @nfo: cm info node 4321 * @ipaddr: Pointer to IPv4 or IPv6 address 4322 * @ipv4: flag indicating IPv4 when true 4323 * @ifup: flag indicating interface up when true 4324 * 4325 * Enables or disables the qhash for the node in the child 4326 * listen list that matches ipaddr. If no matching IP was found 4327 * it will allocate and add a new child listen node to the 4328 * parent listen node. The listen_list_lock is assumed to be 4329 * held when called. 4330 */ 4331 static void irdma_qhash_ctrl(struct irdma_device *iwdev, 4332 struct irdma_cm_listener *parent_listen_node, 4333 struct irdma_cm_info *nfo, u32 *ipaddr, bool ipv4, 4334 bool ifup) 4335 { 4336 struct list_head *child_listen_list = &parent_listen_node->child_listen_list; 4337 struct irdma_cm_listener *child_listen_node; 4338 struct list_head *pos, *tpos; 4339 bool node_allocated = false; 4340 enum irdma_quad_hash_manage_type op = ifup ? 4341 IRDMA_QHASH_MANAGE_TYPE_ADD : 4342 IRDMA_QHASH_MANAGE_TYPE_DELETE; 4343 int err; 4344 4345 list_for_each_safe (pos, tpos, child_listen_list) { 4346 child_listen_node = list_entry(pos, struct irdma_cm_listener, 4347 child_listen_list); 4348 if (!memcmp(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16)) 4349 goto set_qhash; 4350 } 4351 4352 /* if not found then add a child listener if interface is going up */ 4353 if (!ifup) 4354 return; 4355 child_listen_node = kmemdup(parent_listen_node, 4356 sizeof(*child_listen_node), GFP_ATOMIC); 4357 if (!child_listen_node) 4358 return; 4359 4360 node_allocated = true; 4361 memcpy(child_listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16); 4362 4363 set_qhash: 4364 memcpy(nfo->loc_addr, child_listen_node->loc_addr, 4365 sizeof(nfo->loc_addr)); 4366 nfo->vlan_id = child_listen_node->vlan_id; 4367 err = irdma_manage_qhash(iwdev, nfo, IRDMA_QHASH_TYPE_TCP_SYN, op, NULL, 4368 false); 4369 if (!err) { 4370 child_listen_node->qhash_set = ifup; 4371 if (node_allocated) 4372 list_add(&child_listen_node->child_listen_list, 4373 &parent_listen_node->child_listen_list); 4374 } else if (node_allocated) { 4375 kfree(child_listen_node); 4376 } 4377 } 4378 4379 /** 4380 * irdma_if_notify - process an ifdown on an interface 4381 * @iwdev: device pointer 4382 * @netdev: network device structure 4383 * @ipaddr: Pointer to IPv4 or IPv6 address 4384 * @ipv4: flag indicating IPv4 when true 4385 * @ifup: flag indicating interface up when true 4386 */ 4387 void irdma_if_notify(struct irdma_device *iwdev, struct net_device *netdev, 4388 u32 *ipaddr, bool ipv4, bool ifup) 4389 { 4390 struct irdma_cm_core *cm_core = &iwdev->cm_core; 4391 unsigned long flags; 4392 struct irdma_cm_listener *listen_node; 4393 static const u32 ip_zero[4] = { 0, 0, 0, 0 }; 4394 struct irdma_cm_info nfo = {}; 4395 u16 vlan_id = rdma_vlan_dev_vlan_id(netdev); 4396 enum irdma_quad_hash_manage_type op = ifup ? 4397 IRDMA_QHASH_MANAGE_TYPE_ADD : 4398 IRDMA_QHASH_MANAGE_TYPE_DELETE; 4399 4400 nfo.vlan_id = vlan_id; 4401 nfo.ipv4 = ipv4; 4402 nfo.qh_qpid = 1; 4403 4404 /* Disable or enable qhash for listeners */ 4405 spin_lock_irqsave(&cm_core->listen_list_lock, flags); 4406 list_for_each_entry (listen_node, &cm_core->listen_list, list) { 4407 if (vlan_id != listen_node->vlan_id || 4408 (memcmp(listen_node->loc_addr, ipaddr, ipv4 ? 4 : 16) && 4409 memcmp(listen_node->loc_addr, ip_zero, ipv4 ? 4 : 16))) 4410 continue; 4411 4412 memcpy(nfo.loc_addr, listen_node->loc_addr, 4413 sizeof(nfo.loc_addr)); 4414 nfo.loc_port = listen_node->loc_port; 4415 nfo.user_pri = listen_node->user_pri; 4416 if (!list_empty(&listen_node->child_listen_list)) { 4417 irdma_qhash_ctrl(iwdev, listen_node, &nfo, ipaddr, ipv4, 4418 ifup); 4419 } else if (memcmp(listen_node->loc_addr, ip_zero, 4420 ipv4 ? 4 : 16)) { 4421 if (!irdma_manage_qhash(iwdev, &nfo, 4422 IRDMA_QHASH_TYPE_TCP_SYN, op, 4423 NULL, false)) 4424 listen_node->qhash_set = ifup; 4425 } 4426 } 4427 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags); 4428 4429 /* disconnect any connected qp's on ifdown */ 4430 if (!ifup) 4431 irdma_cm_teardown_connections(iwdev, ipaddr, &nfo, false); 4432 } 4433