irdma_cm.c

Version: - FREEBSD - FREEBSD-13-STABLE - FREEBSD-13-0 - FREEBSD-12-STABLE - FREEBSD-12-0 - FREEBSD-11-STABLE - FREEBSD-11-0 - FREEBSD-10-STABLE - FREEBSD-10-0 - FREEBSD-9-STABLE - FREEBSD-9-0 - FREEBSD-8-STABLE - FREEBSD-8-0 - FREEBSD-7-STABLE - FREEBSD-7-0 - FREEBSD-6-STABLE - FREEBSD-6-0 - FREEBSD-5-STABLE - FREEBSD-5-0 - FREEBSD-4-STABLE - FREEBSD-3-STABLE - FREEBSD22 - l41 - OPENBSD - linux-2.6 - MK84 - PLAN9 - xnu-8792
SearchContext: - none - 3 - 10
    1 /*-
    2  * SPDX-License-Identifier: GPL-2.0 or Linux-OpenIB
    3  *
    4  * Copyright (c) 2015 - 2022 Intel Corporation
    5  *
    6  * This software is available to you under a choice of one of two
    7  * licenses.  You may choose to be licensed under the terms of the GNU
    8  * General Public License (GPL) Version 2, available from the file
    9  * COPYING in the main directory of this source tree, or the
   10  * OpenFabrics.org BSD license below:
   11  *
   12  *   Redistribution and use in source and binary forms, with or
   13  *   without modification, are permitted provided that the following
   14  *   conditions are met:
   15  *
   16  *    - Redistributions of source code must retain the above
   17  *      copyright notice, this list of conditions and the following
   18  *      disclaimer.
   19  *
   20  *    - Redistributions in binary form must reproduce the above
   21  *      copyright notice, this list of conditions and the following
   22  *      disclaimer in the documentation and/or other materials
   23  *      provided with the distribution.
   24  *
   25  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
   26  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
   27  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
   28  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
   29  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
   30  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
   31  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
   32  * SOFTWARE.
   33  */
   34 /*$FreeBSD$*/
   35 
   36 #include "irdma_main.h"
   37 
   38 static void irdma_cm_post_event(struct irdma_cm_event *event);
   39 static void irdma_disconnect_worker(struct work_struct *work);
   40 
   41 /**
   42  * irdma_free_sqbuf - put back puda buffer if refcount is 0
   43  * @vsi: The VSI structure of the device
   44  * @bufp: puda buffer to free
   45  */
   46 void
   47 irdma_free_sqbuf(struct irdma_sc_vsi *vsi, void *bufp)
   48 {
   49         struct irdma_puda_buf *buf = bufp;
   50         struct irdma_puda_rsrc *ilq = vsi->ilq;
   51 
   52         if (atomic_dec_and_test(&buf->refcount))
   53                 irdma_puda_ret_bufpool(ilq, buf);
   54 }
   55 
   56 /**
   57  * irdma_record_ird_ord - Record IRD/ORD passed in
   58  * @cm_node: connection's node
   59  * @conn_ird: connection IRD
   60  * @conn_ord: connection ORD
   61  */
   62 static void
   63 irdma_record_ird_ord(struct irdma_cm_node *cm_node, u32 conn_ird,
   64                      u32 conn_ord)
   65 {
   66         if (conn_ird > cm_node->dev->hw_attrs.max_hw_ird)
   67                 conn_ird = cm_node->dev->hw_attrs.max_hw_ird;
   68 
   69         if (conn_ord > cm_node->dev->hw_attrs.max_hw_ord)
   70                 conn_ord = cm_node->dev->hw_attrs.max_hw_ord;
   71         else if (!conn_ord && cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO)
   72                 conn_ord = 1;
   73         cm_node->ird_size = conn_ird;
   74         cm_node->ord_size = conn_ord;
   75 }
   76 
   77 /**
   78  * irdma_copy_ip_ntohl - copy IP address from  network to host
   79  * @dst: IP address in host order
   80  * @src: IP address in network order (big endian)
   81  */
   82 void
   83 irdma_copy_ip_ntohl(u32 *dst, __be32 *src)
   84 {
   85         *dst++ = ntohl(*src++);
   86         *dst++ = ntohl(*src++);
   87         *dst++ = ntohl(*src++);
   88         *dst = ntohl(*src);
   89 }
   90 
   91 /**
   92  * irdma_copy_ip_htonl - copy IP address from host to network order
   93  * @dst: IP address in network order (big endian)
   94  * @src: IP address in host order
   95  */
   96 void
   97 irdma_copy_ip_htonl(__be32 *dst, u32 *src)
   98 {
   99         *dst++ = htonl(*src++);
  100         *dst++ = htonl(*src++);
  101         *dst++ = htonl(*src++);
  102         *dst = htonl(*src);
  103 }
  104 
  105 /**
  106  * irdma_get_addr_info
  107  * @cm_node: contains ip/tcp info
  108  * @cm_info: to get a copy of the cm_node ip/tcp info
  109  */
  110 static void
  111 irdma_get_addr_info(struct irdma_cm_node *cm_node,
  112                     struct irdma_cm_info *cm_info)
  113 {
  114         memset(cm_info, 0, sizeof(*cm_info));
  115         cm_info->ipv4 = cm_node->ipv4;
  116         cm_info->vlan_id = cm_node->vlan_id;
  117         memcpy(cm_info->loc_addr, cm_node->loc_addr, sizeof(cm_info->loc_addr));
  118         memcpy(cm_info->rem_addr, cm_node->rem_addr, sizeof(cm_info->rem_addr));
  119         cm_info->loc_port = cm_node->loc_port;
  120         cm_info->rem_port = cm_node->rem_port;
  121 }
  122 
  123 /**
  124  * irdma_fill_sockaddr4 - fill in addr info for IPv4 connection
  125  * @cm_node: connection's node
  126  * @event: upper layer's cm event
  127  */
  128 static inline void
  129 irdma_fill_sockaddr4(struct irdma_cm_node *cm_node,
  130                      struct iw_cm_event *event)
  131 {
  132         struct sockaddr_in *laddr = (struct sockaddr_in *)&event->local_addr;
  133         struct sockaddr_in *raddr = (struct sockaddr_in *)&event->remote_addr;
  134 
  135         laddr->sin_family = AF_INET;
  136         raddr->sin_family = AF_INET;
  137 
  138         laddr->sin_port = htons(cm_node->loc_port);
  139         raddr->sin_port = htons(cm_node->rem_port);
  140 
  141         laddr->sin_addr.s_addr = htonl(cm_node->loc_addr[0]);
  142         raddr->sin_addr.s_addr = htonl(cm_node->rem_addr[0]);
  143 }
  144 
  145 /**
  146  * irdma_fill_sockaddr6 - fill in addr info for IPv6 connection
  147  * @cm_node: connection's node
  148  * @event: upper layer's cm event
  149  */
  150 static inline void
  151 irdma_fill_sockaddr6(struct irdma_cm_node *cm_node,
  152                      struct iw_cm_event *event)
  153 {
  154         struct sockaddr_in6 *laddr6 = (struct sockaddr_in6 *)&event->local_addr;
  155         struct sockaddr_in6 *raddr6 = (struct sockaddr_in6 *)&event->remote_addr;
  156 
  157         laddr6->sin6_family = AF_INET6;
  158         raddr6->sin6_family = AF_INET6;
  159 
  160         laddr6->sin6_port = htons(cm_node->loc_port);
  161         raddr6->sin6_port = htons(cm_node->rem_port);
  162 
  163         irdma_copy_ip_htonl(laddr6->sin6_addr.__u6_addr.__u6_addr32,
  164                             cm_node->loc_addr);
  165         irdma_copy_ip_htonl(raddr6->sin6_addr.__u6_addr.__u6_addr32,
  166                             cm_node->rem_addr);
  167 }
  168 
  169 /**
  170  * irdma_get_cmevent_info - for cm event upcall
  171  * @cm_node: connection's node
  172  * @cm_id: upper layers cm struct for the event
  173  * @event: upper layer's cm event
  174  */
  175 static inline void
  176 irdma_get_cmevent_info(struct irdma_cm_node *cm_node,
  177                        struct iw_cm_id *cm_id,
  178                        struct iw_cm_event *event)
  179 {
  180         memcpy(&event->local_addr, &cm_id->m_local_addr,
  181                sizeof(event->local_addr));
  182         memcpy(&event->remote_addr, &cm_id->m_remote_addr,
  183                sizeof(event->remote_addr));
  184         if (cm_node) {
  185                 event->private_data = cm_node->pdata_buf;
  186                 event->private_data_len = (u8)cm_node->pdata.size;
  187                 event->ird = cm_node->ird_size;
  188                 event->ord = cm_node->ord_size;
  189         }
  190 }
  191 
  192 /**
  193  * irdma_send_cm_event - upcall cm's event handler
  194  * @cm_node: connection's node
  195  * @cm_id: upper layer's cm info struct
  196  * @type: Event type to indicate
  197  * @status: status for the event type
  198  */
  199 static int
  200 irdma_send_cm_event(struct irdma_cm_node *cm_node,
  201                     struct iw_cm_id *cm_id,
  202                     enum iw_cm_event_type type, int status)
  203 {
  204         struct iw_cm_event event = {0};
  205 
  206         event.event = type;
  207         event.status = status;
  208 
  209         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  210                     "cm_node %p cm_id=%p state=%d accel=%d event_type=%d status=%d\n",
  211                     cm_node, cm_id, cm_node->accelerated, cm_node->state, type, status);
  212 
  213         switch (type) {
  214         case IW_CM_EVENT_CONNECT_REQUEST:
  215                 if (cm_node->ipv4)
  216                         irdma_fill_sockaddr4(cm_node, &event);
  217                 else
  218                         irdma_fill_sockaddr6(cm_node, &event);
  219                 event.provider_data = cm_node;
  220                 event.private_data = cm_node->pdata_buf;
  221                 event.private_data_len = (u8)cm_node->pdata.size;
  222                 event.ird = cm_node->ird_size;
  223                 break;
  224         case IW_CM_EVENT_CONNECT_REPLY:
  225                 irdma_get_cmevent_info(cm_node, cm_id, &event);
  226                 break;
  227         case IW_CM_EVENT_ESTABLISHED:
  228                 event.ird = cm_node->ird_size;
  229                 event.ord = cm_node->ord_size;
  230                 break;
  231         case IW_CM_EVENT_DISCONNECT:
  232         case IW_CM_EVENT_CLOSE:
  233                 /* Wait if we are in RTS but havent issued the iwcm event upcall */
  234                 if (!cm_node->accelerated)
  235                         wait_for_completion(&cm_node->establish_comp);
  236                 break;
  237         default:
  238                 return -EINVAL;
  239         }
  240 
  241         return cm_id->event_handler(cm_id, &event);
  242 }
  243 
  244 /**
  245  * irdma_timer_list_prep - add connection nodes to a list to perform timer tasks
  246  * @cm_core: cm's core
  247  * @timer_list: a timer list to which cm_node will be selected
  248  */
  249 static void
  250 irdma_timer_list_prep(struct irdma_cm_core *cm_core,
  251                       struct list_head *timer_list)
  252 {
  253         struct irdma_cm_node *cm_node;
  254         int bkt;
  255 
  256         HASH_FOR_EACH_RCU(cm_core->cm_hash_tbl, bkt, cm_node, list) {
  257                 if ((cm_node->close_entry || cm_node->send_entry) &&
  258                     atomic_inc_not_zero(&cm_node->refcnt))
  259                         list_add(&cm_node->timer_entry, timer_list);
  260         }
  261 }
  262 
  263 /**
  264  * irdma_create_event - create cm event
  265  * @cm_node: connection's node
  266  * @type: Event type to generate
  267  */
  268 static struct irdma_cm_event *
  269 irdma_create_event(struct irdma_cm_node *cm_node,
  270                    enum irdma_cm_event_type type)
  271 {
  272         struct irdma_cm_event *event;
  273 
  274         if (!cm_node->cm_id)
  275                 return NULL;
  276 
  277         event = kzalloc(sizeof(*event), GFP_ATOMIC);
  278 
  279         if (!event)
  280                 return NULL;
  281 
  282         event->type = type;
  283         event->cm_node = cm_node;
  284         memcpy(event->cm_info.rem_addr, cm_node->rem_addr,
  285                sizeof(event->cm_info.rem_addr));
  286         memcpy(event->cm_info.loc_addr, cm_node->loc_addr,
  287                sizeof(event->cm_info.loc_addr));
  288         event->cm_info.rem_port = cm_node->rem_port;
  289         event->cm_info.loc_port = cm_node->loc_port;
  290         event->cm_info.cm_id = cm_node->cm_id;
  291         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  292                     "node=%p event=%p type=%u dst=%pI4 src=%pI4\n", cm_node,
  293                     event, type, event->cm_info.loc_addr,
  294                     event->cm_info.rem_addr);
  295         irdma_cm_post_event(event);
  296 
  297         return event;
  298 }
  299 
  300 /**
  301  * irdma_free_retrans_entry - free send entry
  302  * @cm_node: connection's node
  303  */
  304 static void
  305 irdma_free_retrans_entry(struct irdma_cm_node *cm_node)
  306 {
  307         struct irdma_device *iwdev = cm_node->iwdev;
  308         struct irdma_timer_entry *send_entry;
  309 
  310         send_entry = cm_node->send_entry;
  311         if (!send_entry)
  312                 return;
  313 
  314         cm_node->send_entry = NULL;
  315         irdma_free_sqbuf(&iwdev->vsi, send_entry->sqbuf);
  316         kfree(send_entry);
  317         atomic_dec(&cm_node->refcnt);
  318 }
  319 
  320 /**
  321  * irdma_cleanup_retrans_entry - free send entry with lock
  322  * @cm_node: connection's node
  323  */
  324 static void
  325 irdma_cleanup_retrans_entry(struct irdma_cm_node *cm_node)
  326 {
  327         unsigned long flags;
  328 
  329         spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
  330         irdma_free_retrans_entry(cm_node);
  331         spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
  332 }
  333 
  334 /**
  335  * irdma_form_ah_cm_frame - get a free packet and build frame with address handle
  336  * @cm_node: connection's node ionfo to use in frame
  337  * @options: pointer to options info
  338  * @hdr: pointer mpa header
  339  * @pdata: pointer to private data
  340  * @flags:  indicates FIN or ACK
  341  */
  342 static struct irdma_puda_buf *
  343 irdma_form_ah_cm_frame(struct irdma_cm_node *cm_node,
  344                        struct irdma_kmem_info *options,
  345                        struct irdma_kmem_info *hdr,
  346                        struct irdma_mpa_priv_info *pdata,
  347                        u8 flags)
  348 {
  349         struct irdma_puda_buf *sqbuf;
  350         struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
  351         u8 *buf;
  352         struct tcphdr *tcph;
  353         u16 pktsize;
  354         u32 opts_len = 0;
  355         u32 pd_len = 0;
  356         u32 hdr_len = 0;
  357 
  358         if (!cm_node->ah || !cm_node->ah->ah_info.ah_valid) {
  359                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  360                             "AH invalid\n");
  361                 return NULL;
  362         }
  363 
  364         sqbuf = irdma_puda_get_bufpool(vsi->ilq);
  365         if (!sqbuf) {
  366                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  367                             "SQ buf NULL\n");
  368                 return NULL;
  369         }
  370 
  371         sqbuf->ah_id = cm_node->ah->ah_info.ah_idx;
  372         buf = sqbuf->mem.va;
  373         if (options)
  374                 opts_len = (u32)options->size;
  375 
  376         if (hdr)
  377                 hdr_len = hdr->size;
  378 
  379         if (pdata)
  380                 pd_len = pdata->size;
  381 
  382         pktsize = sizeof(*tcph) + opts_len + hdr_len + pd_len;
  383 
  384         memset(buf, 0, pktsize);
  385 
  386         sqbuf->totallen = pktsize;
  387         sqbuf->tcphlen = sizeof(*tcph) + opts_len;
  388         sqbuf->scratch = cm_node;
  389 
  390         tcph = (struct tcphdr *)buf;
  391         buf += sizeof(*tcph);
  392 
  393         tcph->th_sport = htons(cm_node->loc_port);
  394         tcph->th_dport = htons(cm_node->rem_port);
  395         tcph->th_seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
  396 
  397         if (flags & SET_ACK) {
  398                 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
  399                 tcph->th_ack = htonl(cm_node->tcp_cntxt.loc_ack_num);
  400                 tcph->th_flags |= TH_ACK;
  401         } else {
  402                 tcph->th_ack = 0;
  403         }
  404 
  405         if (flags & SET_SYN) {
  406                 cm_node->tcp_cntxt.loc_seq_num++;
  407                 tcph->th_flags |= TH_SYN;
  408         } else {
  409                 cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
  410         }
  411 
  412         if (flags & SET_FIN) {
  413                 cm_node->tcp_cntxt.loc_seq_num++;
  414                 tcph->th_flags |= TH_FIN;
  415         }
  416 
  417         if (flags & SET_RST)
  418                 tcph->th_flags |= TH_RST;
  419 
  420         tcph->th_off = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
  421         sqbuf->tcphlen = tcph->th_off << 2;
  422         tcph->th_win = htons(cm_node->tcp_cntxt.rcv_wnd);
  423         tcph->th_urp = 0;
  424 
  425         if (opts_len) {
  426                 memcpy(buf, options->addr, opts_len);
  427                 buf += opts_len;
  428         }
  429 
  430         if (hdr_len) {
  431                 memcpy(buf, hdr->addr, hdr_len);
  432                 buf += hdr_len;
  433         }
  434 
  435         if (pdata && pdata->addr)
  436                 memcpy(buf, pdata->addr, pdata->size);
  437 
  438         atomic_set(&sqbuf->refcount, 1);
  439 
  440         irdma_debug_buf(vsi->dev, IRDMA_DEBUG_ILQ, "TRANSMIT ILQ BUFFER",
  441                         sqbuf->mem.va, sqbuf->totallen);
  442 
  443         return sqbuf;
  444 }
  445 
  446 /**
  447  * irdma_form_uda_cm_frame - get a free packet and build frame full tcpip packet
  448  * @cm_node: connection's node ionfo to use in frame
  449  * @options: pointer to options info
  450  * @hdr: pointer mpa header
  451  * @pdata: pointer to private data
  452  * @flags:  indicates FIN or ACK
  453  */
  454 static struct irdma_puda_buf *
  455 irdma_form_uda_cm_frame(struct irdma_cm_node *cm_node,
  456                         struct irdma_kmem_info *options,
  457                         struct irdma_kmem_info *hdr,
  458                         struct irdma_mpa_priv_info *pdata,
  459                         u8 flags)
  460 {
  461         struct irdma_puda_buf *sqbuf;
  462         struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
  463         u8 *buf;
  464 
  465         struct tcphdr *tcph;
  466         struct ip *iph;
  467         struct ip6_hdr *ip6h;
  468         struct ether_header *ethh;
  469         u16 pktsize;
  470         u16 eth_hlen = ETH_HLEN;
  471         u32 opts_len = 0;
  472         u32 pd_len = 0;
  473         u32 hdr_len = 0;
  474 
  475         u16 vtag;
  476 
  477         sqbuf = irdma_puda_get_bufpool(vsi->ilq);
  478         if (!sqbuf)
  479                 return NULL;
  480 
  481         buf = sqbuf->mem.va;
  482 
  483         if (options)
  484                 opts_len = (u32)options->size;
  485 
  486         if (hdr)
  487                 hdr_len = hdr->size;
  488 
  489         if (pdata)
  490                 pd_len = pdata->size;
  491 
  492         if (cm_node->vlan_id < VLAN_N_VID)
  493                 eth_hlen += 4;
  494 
  495         if (cm_node->ipv4)
  496                 pktsize = sizeof(*iph) + sizeof(*tcph);
  497         else
  498                 pktsize = sizeof(*ip6h) + sizeof(*tcph);
  499         pktsize += opts_len + hdr_len + pd_len;
  500 
  501         memset(buf, 0, eth_hlen + pktsize);
  502 
  503         sqbuf->totallen = pktsize + eth_hlen;
  504         sqbuf->maclen = eth_hlen;
  505         sqbuf->tcphlen = sizeof(*tcph) + opts_len;
  506         sqbuf->scratch = cm_node;
  507 
  508         ethh = (struct ether_header *)buf;
  509         buf += eth_hlen;
  510 
  511         if (cm_node->do_lpb)
  512                 sqbuf->do_lpb = true;
  513 
  514         if (cm_node->ipv4) {
  515                 sqbuf->ipv4 = true;
  516 
  517                 iph = (struct ip *)buf;
  518                 buf += sizeof(*iph);
  519                 tcph = (struct tcphdr *)buf;
  520                 buf += sizeof(*tcph);
  521 
  522                 ether_addr_copy(ethh->ether_dhost, cm_node->rem_mac);
  523                 ether_addr_copy(ethh->ether_shost, cm_node->loc_mac);
  524                 if (cm_node->vlan_id < VLAN_N_VID) {
  525                         ((struct ether_vlan_header *)ethh)->evl_proto =
  526                             htons(ETH_P_8021Q);
  527                         vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
  528                             cm_node->vlan_id;
  529                         ((struct ether_vlan_header *)ethh)->evl_tag = htons(vtag);
  530 
  531                         ((struct ether_vlan_header *)ethh)->evl_encap_proto =
  532                             htons(ETH_P_IP);
  533                 } else {
  534                         ethh->ether_type = htons(ETH_P_IP);
  535                 }
  536 
  537                 iph->ip_v = IPVERSION;
  538                 iph->ip_hl = 5; /* 5 * 4Byte words, IP headr len */
  539                 iph->ip_tos = cm_node->tos;
  540                 iph->ip_len = htons(pktsize);
  541                 iph->ip_id = htons(++cm_node->tcp_cntxt.loc_id);
  542 
  543                 iph->ip_off = htons(0x4000);
  544                 iph->ip_ttl = 0x40;
  545                 iph->ip_p = IPPROTO_TCP;
  546                 iph->ip_src.s_addr = htonl(cm_node->loc_addr[0]);
  547                 iph->ip_dst.s_addr = htonl(cm_node->rem_addr[0]);
  548         } else {
  549                 sqbuf->ipv4 = false;
  550                 ip6h = (struct ip6_hdr *)buf;
  551                 buf += sizeof(*ip6h);
  552                 tcph = (struct tcphdr *)buf;
  553                 buf += sizeof(*tcph);
  554 
  555                 ether_addr_copy(ethh->ether_dhost, cm_node->rem_mac);
  556                 ether_addr_copy(ethh->ether_shost, cm_node->loc_mac);
  557                 if (cm_node->vlan_id < VLAN_N_VID) {
  558                         ((struct ether_vlan_header *)ethh)->evl_proto =
  559                             htons(ETH_P_8021Q);
  560                         vtag = (cm_node->user_pri << VLAN_PRIO_SHIFT) |
  561                             cm_node->vlan_id;
  562                         ((struct ether_vlan_header *)ethh)->evl_tag = htons(vtag);
  563                         ((struct ether_vlan_header *)ethh)->evl_encap_proto =
  564                             htons(ETH_P_IPV6);
  565                 } else {
  566                         ethh->ether_type = htons(ETH_P_IPV6);
  567                 }
  568                 ip6h->ip6_vfc = 6 << 4;
  569                 ip6h->ip6_vfc |= cm_node->tos >> 4;
  570                 ip6h->ip6_flow = cm_node->tos << 20;
  571                 ip6h->ip6_plen = htons(pktsize - sizeof(*ip6h));
  572                 ip6h->ip6_nxt = 6;
  573                 ip6h->ip6_hops = 128;
  574                 irdma_copy_ip_htonl(ip6h->ip6_src.__u6_addr.__u6_addr32,
  575                                     cm_node->loc_addr);
  576                 irdma_copy_ip_htonl(ip6h->ip6_dst.__u6_addr.__u6_addr32,
  577                                     cm_node->rem_addr);
  578         }
  579 
  580         tcph->th_sport = htons(cm_node->loc_port);
  581         tcph->th_dport = htons(cm_node->rem_port);
  582         tcph->th_seq = htonl(cm_node->tcp_cntxt.loc_seq_num);
  583 
  584         if (flags & SET_ACK) {
  585                 cm_node->tcp_cntxt.loc_ack_num = cm_node->tcp_cntxt.rcv_nxt;
  586                 tcph->th_ack = htonl(cm_node->tcp_cntxt.loc_ack_num);
  587                 tcph->th_flags |= TH_ACK;
  588         } else {
  589                 tcph->th_ack = 0;
  590         }
  591 
  592         if (flags & SET_SYN) {
  593                 cm_node->tcp_cntxt.loc_seq_num++;
  594                 tcph->th_flags |= TH_SYN;
  595         } else {
  596                 cm_node->tcp_cntxt.loc_seq_num += hdr_len + pd_len;
  597         }
  598 
  599         if (flags & SET_FIN) {
  600                 cm_node->tcp_cntxt.loc_seq_num++;
  601                 tcph->th_flags |= TH_FIN;
  602         }
  603 
  604         if (flags & SET_RST)
  605                 tcph->th_flags |= TH_RST;
  606 
  607         tcph->th_off = (u16)((sizeof(*tcph) + opts_len + 3) >> 2);
  608         sqbuf->tcphlen = tcph->th_off << 2;
  609         tcph->th_win = htons(cm_node->tcp_cntxt.rcv_wnd);
  610         tcph->th_urp = 0;
  611 
  612         if (opts_len) {
  613                 memcpy(buf, options->addr, opts_len);
  614                 buf += opts_len;
  615         }
  616 
  617         if (hdr_len) {
  618                 memcpy(buf, hdr->addr, hdr_len);
  619                 buf += hdr_len;
  620         }
  621 
  622         if (pdata && pdata->addr)
  623                 memcpy(buf, pdata->addr, pdata->size);
  624 
  625         atomic_set(&sqbuf->refcount, 1);
  626 
  627         irdma_debug_buf(vsi->dev, IRDMA_DEBUG_ILQ, "TRANSMIT ILQ BUFFER",
  628                         sqbuf->mem.va, sqbuf->totallen);
  629 
  630         return sqbuf;
  631 }
  632 
  633 /**
  634  * irdma_send_reset - Send RST packet
  635  * @cm_node: connection's node
  636  */
  637 int
  638 irdma_send_reset(struct irdma_cm_node *cm_node)
  639 {
  640         struct irdma_puda_buf *sqbuf;
  641         int flags = SET_RST | SET_ACK;
  642 
  643         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
  644                                                 flags);
  645         if (!sqbuf)
  646                 return -ENOMEM;
  647 
  648         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  649                     "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",
  650                     __builtin_return_address(0), cm_node, cm_node->cm_id,
  651                     cm_node->accelerated, cm_node->state, cm_node->rem_port,
  652                     cm_node->loc_port, cm_node->rem_addr, cm_node->loc_addr);
  653 
  654         return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 0,
  655                                        1);
  656 }
  657 
  658 /**
  659  * irdma_active_open_err - send event for active side cm error
  660  * @cm_node: connection's node
  661  * @reset: Flag to send reset or not
  662  */
  663 static void
  664 irdma_active_open_err(struct irdma_cm_node *cm_node, bool reset)
  665 {
  666         irdma_cleanup_retrans_entry(cm_node);
  667         cm_node->cm_core->stats_connect_errs++;
  668         if (reset) {
  669                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  670                             "cm_node=%p state=%d\n", cm_node, cm_node->state);
  671                 atomic_inc(&cm_node->refcnt);
  672                 irdma_send_reset(cm_node);
  673         }
  674 
  675         cm_node->state = IRDMA_CM_STATE_CLOSED;
  676         irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
  677 }
  678 
  679 /**
  680  * irdma_passive_open_err - handle passive side cm error
  681  * @cm_node: connection's node
  682  * @reset: send reset or just free cm_node
  683  */
  684 static void
  685 irdma_passive_open_err(struct irdma_cm_node *cm_node, bool reset)
  686 {
  687         irdma_cleanup_retrans_entry(cm_node);
  688         cm_node->cm_core->stats_passive_errs++;
  689         cm_node->state = IRDMA_CM_STATE_CLOSED;
  690         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  691                     "cm_node=%p state =%d\n", cm_node, cm_node->state);
  692         if (reset)
  693                 irdma_send_reset(cm_node);
  694         else
  695                 irdma_rem_ref_cm_node(cm_node);
  696 }
  697 
  698 /**
  699  * irdma_event_connect_error - to create connect error event
  700  * @event: cm information for connect event
  701  */
  702 static void
  703 irdma_event_connect_error(struct irdma_cm_event *event)
  704 {
  705         struct irdma_qp *iwqp;
  706         struct iw_cm_id *cm_id;
  707 
  708         cm_id = event->cm_node->cm_id;
  709         if (!cm_id)
  710                 return;
  711 
  712         iwqp = cm_id->provider_data;
  713 
  714         if (!iwqp || !iwqp->iwdev)
  715                 return;
  716 
  717         iwqp->cm_id = NULL;
  718         cm_id->provider_data = NULL;
  719         irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
  720                             -ECONNRESET);
  721         irdma_rem_ref_cm_node(event->cm_node);
  722 }
  723 
  724 /**
  725  * irdma_process_options - process options from TCP header
  726  * @cm_node: connection's node
  727  * @optionsloc: point to start of options
  728  * @optionsize: size of all options
  729  * @syn_pkt: flag if syn packet
  730  */
  731 static int
  732 irdma_process_options(struct irdma_cm_node *cm_node, u8 *optionsloc,
  733                       u32 optionsize, u32 syn_pkt)
  734 {
  735         u32 tmp;
  736         u32 offset = 0;
  737         union all_known_options *all_options;
  738         char got_mss_option = 0;
  739 
  740         while (offset < optionsize) {
  741                 all_options = (union all_known_options *)(optionsloc + offset);
  742                 switch (all_options->base.optionnum) {
  743                 case OPTION_NUM_EOL:
  744                         offset = optionsize;
  745                         break;
  746                 case OPTION_NUM_NONE:
  747                         offset += 1;
  748                         continue;
  749                 case OPTION_NUM_MSS:
  750                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
  751                                     IRDMA_DEBUG_CM,
  752                                     "MSS Length: %d Offset: %d Size: %d\n",
  753                                     all_options->mss.len, offset, optionsize);
  754                         got_mss_option = 1;
  755                         if (all_options->mss.len != 4)
  756                                 return -EINVAL;
  757                         tmp = ntohs(all_options->mss.mss);
  758                         if ((cm_node->ipv4 &&
  759                              (tmp + IRDMA_MTU_TO_MSS_IPV4) < IRDMA_MIN_MTU_IPV4) ||
  760                             (!cm_node->ipv4 &&
  761                              (tmp + IRDMA_MTU_TO_MSS_IPV6) < IRDMA_MIN_MTU_IPV6))
  762                                 return -EINVAL;
  763                         if (tmp < cm_node->tcp_cntxt.mss)
  764                                 cm_node->tcp_cntxt.mss = tmp;
  765                         break;
  766                 case OPTION_NUM_WINDOW_SCALE:
  767                         cm_node->tcp_cntxt.snd_wscale =
  768                             all_options->windowscale.shiftcount;
  769                         break;
  770                 default:
  771                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
  772                                     IRDMA_DEBUG_CM,
  773                                     "Unsupported TCP Option: %x\n",
  774                                     all_options->base.optionnum);
  775                         break;
  776                 }
  777                 offset += all_options->base.len;
  778         }
  779         if (!got_mss_option && syn_pkt)
  780                 cm_node->tcp_cntxt.mss = IRDMA_CM_DEFAULT_MSS;
  781 
  782         return 0;
  783 }
  784 
  785 /**
  786  * irdma_handle_tcp_options - setup TCP context info after parsing TCP options
  787  * @cm_node: connection's node
  788  * @tcph: pointer tcp header
  789  * @optionsize: size of options rcvd
  790  * @passive: active or passive flag
  791  */
  792 static int
  793 irdma_handle_tcp_options(struct irdma_cm_node *cm_node,
  794                          struct tcphdr *tcph, int optionsize,
  795                          int passive)
  796 {
  797         u8 *optionsloc = (u8 *)&tcph[1];
  798         int ret;
  799 
  800         if (optionsize) {
  801                 ret = irdma_process_options(cm_node, optionsloc, optionsize,
  802                                             (u32)tcph->th_flags & TH_SYN);
  803                 if (ret) {
  804                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
  805                                     IRDMA_DEBUG_CM,
  806                                     "Node %p, Sending Reset\n", cm_node);
  807                         if (passive)
  808                                 irdma_passive_open_err(cm_node, true);
  809                         else
  810                                 irdma_active_open_err(cm_node, true);
  811                         return ret;
  812                 }
  813         }
  814 
  815         cm_node->tcp_cntxt.snd_wnd = ntohs(tcph->th_win)
  816             << cm_node->tcp_cntxt.snd_wscale;
  817 
  818         if (cm_node->tcp_cntxt.snd_wnd > cm_node->tcp_cntxt.max_snd_wnd)
  819                 cm_node->tcp_cntxt.max_snd_wnd = cm_node->tcp_cntxt.snd_wnd;
  820 
  821         return 0;
  822 }
  823 
  824 /**
  825  * irdma_build_mpa_v1 - build a MPA V1 frame
  826  * @cm_node: connection's node
  827  * @start_addr: address where to build frame
  828  * @mpa_key: to do read0 or write0
  829  */
  830 static void
  831 irdma_build_mpa_v1(struct irdma_cm_node *cm_node, void *start_addr,
  832                    u8 mpa_key)
  833 {
  834         struct ietf_mpa_v1 *mpa_frame = start_addr;
  835 
  836         switch (mpa_key) {
  837         case MPA_KEY_REQUEST:
  838                 memcpy(mpa_frame->key, IEFT_MPA_KEY_REQ, IETF_MPA_KEY_SIZE);
  839                 break;
  840         case MPA_KEY_REPLY:
  841                 memcpy(mpa_frame->key, IEFT_MPA_KEY_REP, IETF_MPA_KEY_SIZE);
  842                 break;
  843         default:
  844                 break;
  845         }
  846         mpa_frame->flags = IETF_MPA_FLAGS_CRC;
  847         mpa_frame->rev = cm_node->mpa_frame_rev;
  848         mpa_frame->priv_data_len = htons(cm_node->pdata.size);
  849 }
  850 
  851 /**
  852  * irdma_build_mpa_v2 - build a MPA V2 frame
  853  * @cm_node: connection's node
  854  * @start_addr: buffer start address
  855  * @mpa_key: to do read0 or write0
  856  */
  857 static void
  858 irdma_build_mpa_v2(struct irdma_cm_node *cm_node, void *start_addr,
  859                    u8 mpa_key)
  860 {
  861         struct ietf_mpa_v2 *mpa_frame = start_addr;
  862         struct ietf_rtr_msg *rtr_msg = &mpa_frame->rtr_msg;
  863         u16 ctrl_ird, ctrl_ord;
  864 
  865         /* initialize the upper 5 bytes of the frame */
  866         irdma_build_mpa_v1(cm_node, start_addr, mpa_key);
  867         mpa_frame->flags |= IETF_MPA_V2_FLAG;
  868         if (cm_node->iwdev->iw_ooo) {
  869                 mpa_frame->flags |= IETF_MPA_FLAGS_MARKERS;
  870                 cm_node->rcv_mark_en = true;
  871         }
  872         mpa_frame->priv_data_len = cpu_to_be16(be16_to_cpu(mpa_frame->priv_data_len) +
  873                                                IETF_RTR_MSG_SIZE);
  874 
  875         /* initialize RTR msg */
  876         if (cm_node->mpav2_ird_ord == IETF_NO_IRD_ORD) {
  877                 ctrl_ird = IETF_NO_IRD_ORD;
  878                 ctrl_ord = IETF_NO_IRD_ORD;
  879         } else {
  880                 ctrl_ird = (cm_node->ird_size > IETF_NO_IRD_ORD) ?
  881                     IETF_NO_IRD_ORD :
  882                     cm_node->ird_size;
  883                 ctrl_ord = (cm_node->ord_size > IETF_NO_IRD_ORD) ?
  884                     IETF_NO_IRD_ORD :
  885                     cm_node->ord_size;
  886         }
  887         ctrl_ird |= IETF_PEER_TO_PEER;
  888 
  889         switch (mpa_key) {
  890         case MPA_KEY_REQUEST:
  891                 ctrl_ord |= IETF_RDMA0_WRITE;
  892                 ctrl_ord |= IETF_RDMA0_READ;
  893                 break;
  894         case MPA_KEY_REPLY:
  895                 switch (cm_node->send_rdma0_op) {
  896                 case SEND_RDMA_WRITE_ZERO:
  897                         ctrl_ord |= IETF_RDMA0_WRITE;
  898                         break;
  899                 case SEND_RDMA_READ_ZERO:
  900                         ctrl_ord |= IETF_RDMA0_READ;
  901                         break;
  902                 }
  903                 break;
  904         default:
  905                 break;
  906         }
  907         rtr_msg->ctrl_ird = htons(ctrl_ird);
  908         rtr_msg->ctrl_ord = htons(ctrl_ord);
  909 }
  910 
  911 /**
  912  * irdma_cm_build_mpa_frame - build mpa frame for mpa version 1 or version 2
  913  * @cm_node: connection's node
  914  * @mpa: mpa: data buffer
  915  * @mpa_key: to do read0 or write0
  916  */
  917 static int
  918 irdma_cm_build_mpa_frame(struct irdma_cm_node *cm_node,
  919                          struct irdma_kmem_info *mpa, u8 mpa_key)
  920 {
  921         int hdr_len = 0;
  922 
  923         switch (cm_node->mpa_frame_rev) {
  924         case IETF_MPA_V1:
  925                 hdr_len = sizeof(struct ietf_mpa_v1);
  926                 irdma_build_mpa_v1(cm_node, mpa->addr, mpa_key);
  927                 break;
  928         case IETF_MPA_V2:
  929                 hdr_len = sizeof(struct ietf_mpa_v2);
  930                 irdma_build_mpa_v2(cm_node, mpa->addr, mpa_key);
  931                 break;
  932         default:
  933                 break;
  934         }
  935 
  936         return hdr_len;
  937 }
  938 
  939 /**
  940  * irdma_send_mpa_request - active node send mpa request to passive node
  941  * @cm_node: connection's node
  942  */
  943 static int
  944 irdma_send_mpa_request(struct irdma_cm_node *cm_node)
  945 {
  946         struct irdma_puda_buf *sqbuf;
  947 
  948         cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
  949         cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
  950                                                          &cm_node->mpa_hdr,
  951                                                          MPA_KEY_REQUEST);
  952         if (!cm_node->mpa_hdr.size) {
  953                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
  954                             "mpa size = %d\n", cm_node->mpa_hdr.size);
  955                 return -EINVAL;
  956         }
  957 
  958         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
  959                                                 &cm_node->mpa_hdr,
  960                                                 &cm_node->pdata, SET_ACK);
  961         if (!sqbuf)
  962                 return -ENOMEM;
  963 
  964         return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
  965                                        0);
  966 }
  967 
  968 /**
  969  * irdma_send_mpa_reject -
  970  * @cm_node: connection's node
  971  * @pdata: reject data for connection
  972  * @plen: length of reject data
  973  */
  974 static int
  975 irdma_send_mpa_reject(struct irdma_cm_node *cm_node,
  976                       const void *pdata, u8 plen)
  977 {
  978         struct irdma_puda_buf *sqbuf;
  979         struct irdma_mpa_priv_info priv_info;
  980 
  981         cm_node->mpa_hdr.addr = &cm_node->mpa_v2_frame;
  982         cm_node->mpa_hdr.size = irdma_cm_build_mpa_frame(cm_node,
  983                                                          &cm_node->mpa_hdr,
  984                                                          MPA_KEY_REPLY);
  985 
  986         cm_node->mpa_v2_frame.flags |= IETF_MPA_FLAGS_REJECT;
  987         priv_info.addr = pdata;
  988         priv_info.size = plen;
  989 
  990         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL,
  991                                                 &cm_node->mpa_hdr, &priv_info,
  992                                                 SET_ACK | SET_FIN);
  993         if (!sqbuf)
  994                 return -ENOMEM;
  995 
  996         cm_node->state = IRDMA_CM_STATE_FIN_WAIT1;
  997 
  998         return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
  999                                        0);
 1000 }
 1001 
 1002 /**
 1003  * irdma_negotiate_mpa_v2_ird_ord - negotiate MPAv2 IRD/ORD
 1004  * @cm_node: connection's node
 1005  * @buf: Data pointer
 1006  */
 1007 static int
 1008 irdma_negotiate_mpa_v2_ird_ord(struct irdma_cm_node *cm_node,
 1009                                u8 *buf)
 1010 {
 1011         struct ietf_mpa_v2 *mpa_v2_frame;
 1012         struct ietf_rtr_msg *rtr_msg;
 1013         u16 ird_size;
 1014         u16 ord_size;
 1015         u16 ctrl_ord;
 1016         u16 ctrl_ird;
 1017 
 1018         mpa_v2_frame = (struct ietf_mpa_v2 *)buf;
 1019         rtr_msg = &mpa_v2_frame->rtr_msg;
 1020 
 1021         /* parse rtr message */
 1022         ctrl_ord = ntohs(rtr_msg->ctrl_ord);
 1023         ctrl_ird = ntohs(rtr_msg->ctrl_ird);
 1024         ird_size = ctrl_ird & IETF_NO_IRD_ORD;
 1025         ord_size = ctrl_ord & IETF_NO_IRD_ORD;
 1026 
 1027         if (!(ctrl_ird & IETF_PEER_TO_PEER))
 1028                 return -EOPNOTSUPP;
 1029 
 1030         if (ird_size == IETF_NO_IRD_ORD || ord_size == IETF_NO_IRD_ORD) {
 1031                 cm_node->mpav2_ird_ord = IETF_NO_IRD_ORD;
 1032                 goto negotiate_done;
 1033         }
 1034 
 1035         if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
 1036                 /* responder */
 1037                 if (!ord_size && (ctrl_ord & IETF_RDMA0_READ))
 1038                         cm_node->ird_size = 1;
 1039                 if (cm_node->ord_size > ird_size)
 1040                         cm_node->ord_size = ird_size;
 1041         } else {
 1042                 /* initiator */
 1043                 if (!ird_size && (ctrl_ord & IETF_RDMA0_READ))
 1044                         /* Remote peer doesn't support RDMA0_READ */
 1045                         return -EOPNOTSUPP;
 1046 
 1047                 if (cm_node->ord_size > ird_size)
 1048                         cm_node->ord_size = ird_size;
 1049 
 1050                 if (cm_node->ird_size < ord_size)
 1051                         /* no resources available */
 1052                         return -EINVAL;
 1053         }
 1054 
 1055 negotiate_done:
 1056         if (ctrl_ord & IETF_RDMA0_READ)
 1057                 cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
 1058         else if (ctrl_ord & IETF_RDMA0_WRITE)
 1059                 cm_node->send_rdma0_op = SEND_RDMA_WRITE_ZERO;
 1060         else
 1061                 /* Not supported RDMA0 operation */
 1062                 return -EOPNOTSUPP;
 1063 
 1064         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1065                     "MPAV2 Negotiated ORD: %d, IRD: %d\n", cm_node->ord_size,
 1066                     cm_node->ird_size);
 1067         return 0;
 1068 }
 1069 
 1070 /**
 1071  * irdma_parse_mpa - process an IETF MPA frame
 1072  * @cm_node: connection's node
 1073  * @buf: Data pointer
 1074  * @type: to return accept or reject
 1075  * @len: Len of mpa buffer
 1076  */
 1077 static int
 1078 irdma_parse_mpa(struct irdma_cm_node *cm_node, u8 *buf, u32 *type,
 1079                 u32 len)
 1080 {
 1081         struct ietf_mpa_v1 *mpa_frame;
 1082         int mpa_hdr_len, priv_data_len, ret;
 1083 
 1084         *type = IRDMA_MPA_REQUEST_ACCEPT;
 1085 
 1086         if (len < sizeof(struct ietf_mpa_v1)) {
 1087                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1088                             "ietf buffer small (%x)\n", len);
 1089                 return -EINVAL;
 1090         }
 1091 
 1092         mpa_frame = (struct ietf_mpa_v1 *)buf;
 1093         mpa_hdr_len = sizeof(struct ietf_mpa_v1);
 1094         priv_data_len = ntohs(mpa_frame->priv_data_len);
 1095 
 1096         if (priv_data_len > IETF_MAX_PRIV_DATA_LEN) {
 1097                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1098                             "private_data too big %d\n", priv_data_len);
 1099                 return -EOVERFLOW;
 1100         }
 1101 
 1102         if (mpa_frame->rev != IETF_MPA_V1 && mpa_frame->rev != IETF_MPA_V2) {
 1103                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1104                             "unsupported mpa rev = %d\n", mpa_frame->rev);
 1105                 return -EINVAL;
 1106         }
 1107 
 1108         if (mpa_frame->rev > cm_node->mpa_frame_rev) {
 1109                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1110                             "rev %d\n", mpa_frame->rev);
 1111                 return -EINVAL;
 1112         }
 1113 
 1114         cm_node->mpa_frame_rev = mpa_frame->rev;
 1115         if (cm_node->state != IRDMA_CM_STATE_MPAREQ_SENT) {
 1116                 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REQ,
 1117                            IETF_MPA_KEY_SIZE)) {
 1118                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 1119                                     IRDMA_DEBUG_CM,
 1120                                     "Unexpected MPA Key received\n");
 1121                         return -EINVAL;
 1122                 }
 1123         } else {
 1124                 if (memcmp(mpa_frame->key, IEFT_MPA_KEY_REP,
 1125                            IETF_MPA_KEY_SIZE)) {
 1126                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 1127                                     IRDMA_DEBUG_CM,
 1128                                     "Unexpected MPA Key received\n");
 1129                         return -EINVAL;
 1130                 }
 1131         }
 1132 
 1133         if (priv_data_len + mpa_hdr_len > len) {
 1134                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1135                             "ietf buffer len(%x + %x != %x)\n", priv_data_len,
 1136                             mpa_hdr_len, len);
 1137                 return -EOVERFLOW;
 1138         }
 1139 
 1140         if (len > IRDMA_MAX_CM_BUF) {
 1141                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 1142                             "ietf buffer large len = %d\n", len);
 1143                 return -EOVERFLOW;
 1144         }
 1145 
 1146         switch (mpa_frame->rev) {
 1147         case IETF_MPA_V2:
 1148                 mpa_hdr_len += IETF_RTR_MSG_SIZE;
 1149                 ret = irdma_negotiate_mpa_v2_ird_ord(cm_node, buf);
 1150                 if (ret)
 1151                         return ret;
 1152                 break;
 1153         case IETF_MPA_V1:
 1154         default:
 1155                 break;
 1156         }
 1157 
 1158         memcpy(cm_node->pdata_buf, buf + mpa_hdr_len, priv_data_len);
 1159         cm_node->pdata.size = priv_data_len;
 1160 
 1161         if (mpa_frame->flags & IETF_MPA_FLAGS_REJECT)
 1162                 *type = IRDMA_MPA_REQUEST_REJECT;
 1163 
 1164         if (mpa_frame->flags & IETF_MPA_FLAGS_MARKERS)
 1165                 cm_node->snd_mark_en = true;
 1166 
 1167         return 0;
 1168 }
 1169 
 1170 /**
 1171  * irdma_schedule_cm_timer
 1172  * @cm_node: connection's node
 1173  * @sqbuf: buffer to send
 1174  * @type: if it is send or close
 1175  * @send_retrans: if rexmits to be done
 1176  * @close_when_complete: is cm_node to be removed
 1177  *
 1178  * note - cm_node needs to be protected before calling this. Encase in:
 1179  *              irdma_rem_ref_cm_node(cm_core, cm_node);
 1180  *              irdma_schedule_cm_timer(...)
 1181  *              atomic_inc(&cm_node->refcnt);
 1182  */
 1183 int
 1184 irdma_schedule_cm_timer(struct irdma_cm_node *cm_node,
 1185                         struct irdma_puda_buf *sqbuf,
 1186                         enum irdma_timer_type type, int send_retrans,
 1187                         int close_when_complete)
 1188 {
 1189         struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
 1190         struct irdma_cm_core *cm_core = cm_node->cm_core;
 1191         struct irdma_timer_entry *new_send;
 1192         u32 was_timer_set;
 1193         unsigned long flags;
 1194 
 1195         new_send = kzalloc(sizeof(*new_send), GFP_ATOMIC);
 1196         if (!new_send) {
 1197                 if (type != IRDMA_TIMER_TYPE_CLOSE)
 1198                         irdma_free_sqbuf(vsi, sqbuf);
 1199                 return -ENOMEM;
 1200         }
 1201 
 1202         new_send->retrycount = IRDMA_DEFAULT_RETRYS;
 1203         new_send->retranscount = IRDMA_DEFAULT_RETRANS;
 1204         new_send->sqbuf = sqbuf;
 1205         new_send->timetosend = jiffies;
 1206         new_send->type = type;
 1207         new_send->send_retrans = send_retrans;
 1208         new_send->close_when_complete = close_when_complete;
 1209 
 1210         if (type == IRDMA_TIMER_TYPE_CLOSE) {
 1211                 new_send->timetosend += (HZ / 10);
 1212                 if (cm_node->close_entry) {
 1213                         kfree(new_send);
 1214                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 1215                                     IRDMA_DEBUG_CM, "already close entry\n");
 1216                         return -EINVAL;
 1217                 }
 1218 
 1219                 cm_node->close_entry = new_send;
 1220         } else {                /* type == IRDMA_TIMER_TYPE_SEND */
 1221                 spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 1222                 cm_node->send_entry = new_send;
 1223                 atomic_inc(&cm_node->refcnt);
 1224                 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 1225                 new_send->timetosend = jiffies + IRDMA_RETRY_TIMEOUT;
 1226 
 1227                 atomic_inc(&sqbuf->refcount);
 1228                 irdma_puda_send_buf(vsi->ilq, sqbuf);
 1229                 if (!send_retrans) {
 1230                         irdma_cleanup_retrans_entry(cm_node);
 1231                         if (close_when_complete)
 1232                                 irdma_rem_ref_cm_node(cm_node);
 1233                         return 0;
 1234                 }
 1235         }
 1236 
 1237         spin_lock_irqsave(&cm_core->ht_lock, flags);
 1238         was_timer_set = timer_pending(&cm_core->tcp_timer);
 1239 
 1240         if (!was_timer_set) {
 1241                 cm_core->tcp_timer.expires = new_send->timetosend;
 1242                 add_timer(&cm_core->tcp_timer);
 1243         }
 1244         spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 1245 
 1246         return 0;
 1247 }
 1248 
 1249 /**
 1250  * irdma_retrans_expired - Could not rexmit the packet
 1251  * @cm_node: connection's node
 1252  */
 1253 static void
 1254 irdma_retrans_expired(struct irdma_cm_node *cm_node)
 1255 {
 1256         enum irdma_cm_node_state state = cm_node->state;
 1257 
 1258         cm_node->state = IRDMA_CM_STATE_CLOSED;
 1259         switch (state) {
 1260         case IRDMA_CM_STATE_SYN_RCVD:
 1261         case IRDMA_CM_STATE_CLOSING:
 1262                 irdma_rem_ref_cm_node(cm_node);
 1263                 break;
 1264         case IRDMA_CM_STATE_FIN_WAIT1:
 1265         case IRDMA_CM_STATE_LAST_ACK:
 1266                 irdma_send_reset(cm_node);
 1267                 break;
 1268         default:
 1269                 atomic_inc(&cm_node->refcnt);
 1270                 irdma_send_reset(cm_node);
 1271                 irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
 1272                 break;
 1273         }
 1274 }
 1275 
 1276 /**
 1277  * irdma_handle_close_entry - for handling retry/timeouts
 1278  * @cm_node: connection's node
 1279  * @rem_node: flag for remove cm_node
 1280  */
 1281 static void
 1282 irdma_handle_close_entry(struct irdma_cm_node *cm_node,
 1283                          u32 rem_node)
 1284 {
 1285         struct irdma_timer_entry *close_entry = cm_node->close_entry;
 1286         struct irdma_qp *iwqp;
 1287         unsigned long flags;
 1288 
 1289         if (!close_entry)
 1290                 return;
 1291         iwqp = (struct irdma_qp *)close_entry->sqbuf;
 1292         if (iwqp) {
 1293                 spin_lock_irqsave(&iwqp->lock, flags);
 1294                 if (iwqp->cm_id) {
 1295                         iwqp->hw_tcp_state = IRDMA_TCP_STATE_CLOSED;
 1296                         iwqp->hw_iwarp_state = IRDMA_QP_STATE_ERROR;
 1297                         iwqp->last_aeq = IRDMA_AE_RESET_SENT;
 1298                         iwqp->ibqp_state = IB_QPS_ERR;
 1299                         spin_unlock_irqrestore(&iwqp->lock, flags);
 1300                         irdma_cm_disconn(iwqp);
 1301                 } else {
 1302                         spin_unlock_irqrestore(&iwqp->lock, flags);
 1303                 }
 1304         } else if (rem_node) {
 1305                 /* TIME_WAIT state */
 1306                 irdma_rem_ref_cm_node(cm_node);
 1307         }
 1308 
 1309         kfree(close_entry);
 1310         cm_node->close_entry = NULL;
 1311 }
 1312 
 1313 /**
 1314  * irdma_cm_timer_tick - system's timer expired callback
 1315  * @t: Pointer to timer_list
 1316  */
 1317 static void
 1318 irdma_cm_timer_tick(struct timer_list *t)
 1319 {
 1320         unsigned long nexttimeout = jiffies + IRDMA_LONG_TIME;
 1321         struct irdma_cm_node *cm_node;
 1322         struct irdma_timer_entry *send_entry, *close_entry;
 1323         struct list_head *list_core_temp;
 1324         struct list_head *list_node;
 1325         struct irdma_cm_core *cm_core = from_timer(cm_core, t, tcp_timer);
 1326         struct irdma_sc_vsi *vsi;
 1327         u32 settimer = 0;
 1328         unsigned long timetosend;
 1329         unsigned long flags;
 1330         struct list_head timer_list;
 1331 
 1332         INIT_LIST_HEAD(&timer_list);
 1333 
 1334         rcu_read_lock();
 1335         irdma_timer_list_prep(cm_core, &timer_list);
 1336         rcu_read_unlock();
 1337 
 1338         list_for_each_safe(list_node, list_core_temp, &timer_list) {
 1339                 cm_node = container_of(list_node, struct irdma_cm_node,
 1340                                        timer_entry);
 1341                 close_entry = cm_node->close_entry;
 1342 
 1343                 if (close_entry) {
 1344                         if (time_after(close_entry->timetosend, jiffies)) {
 1345                                 if (nexttimeout > close_entry->timetosend ||
 1346                                     !settimer) {
 1347                                         nexttimeout = close_entry->timetosend;
 1348                                         settimer = 1;
 1349                                 }
 1350                         } else {
 1351                                 irdma_handle_close_entry(cm_node, 1);
 1352                         }
 1353                 }
 1354 
 1355                 spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 1356 
 1357                 send_entry = cm_node->send_entry;
 1358                 if (!send_entry)
 1359                         goto done;
 1360                 if (time_after(send_entry->timetosend, jiffies)) {
 1361                         if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
 1362                                 if (nexttimeout > send_entry->timetosend ||
 1363                                     !settimer) {
 1364                                         nexttimeout = send_entry->timetosend;
 1365                                         settimer = 1;
 1366                                 }
 1367                         } else {
 1368                                 irdma_free_retrans_entry(cm_node);
 1369                         }
 1370                         goto done;
 1371                 }
 1372 
 1373                 if (cm_node->state == IRDMA_CM_STATE_OFFLOADED ||
 1374                     cm_node->state == IRDMA_CM_STATE_CLOSED) {
 1375                         irdma_free_retrans_entry(cm_node);
 1376                         goto done;
 1377                 }
 1378 
 1379                 if (!send_entry->retranscount || !send_entry->retrycount) {
 1380                         irdma_free_retrans_entry(cm_node);
 1381 
 1382                         spin_unlock_irqrestore(&cm_node->retrans_list_lock,
 1383                                                flags);
 1384                         irdma_retrans_expired(cm_node);
 1385                         cm_node->state = IRDMA_CM_STATE_CLOSED;
 1386                         spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 1387                         goto done;
 1388                 }
 1389                 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 1390 
 1391                 vsi = &cm_node->iwdev->vsi;
 1392                 if (!cm_node->ack_rcvd) {
 1393                         atomic_inc(&send_entry->sqbuf->refcount);
 1394                         irdma_puda_send_buf(vsi->ilq, send_entry->sqbuf);
 1395                         cm_node->cm_core->stats_pkt_retrans++;
 1396                 }
 1397 
 1398                 spin_lock_irqsave(&cm_node->retrans_list_lock, flags);
 1399                 if (send_entry->send_retrans) {
 1400                         send_entry->retranscount--;
 1401                         timetosend = (IRDMA_RETRY_TIMEOUT <<
 1402                                       (IRDMA_DEFAULT_RETRANS -
 1403                                        send_entry->retranscount));
 1404 
 1405                         send_entry->timetosend = jiffies +
 1406                             min(timetosend, IRDMA_MAX_TIMEOUT);
 1407                         if (nexttimeout > send_entry->timetosend || !settimer) {
 1408                                 nexttimeout = send_entry->timetosend;
 1409                                 settimer = 1;
 1410                         }
 1411                 } else {
 1412                         int close_when_complete;
 1413 
 1414                         close_when_complete = send_entry->close_when_complete;
 1415                         irdma_free_retrans_entry(cm_node);
 1416                         if (close_when_complete)
 1417                                 irdma_rem_ref_cm_node(cm_node);
 1418                 }
 1419 done:
 1420                 spin_unlock_irqrestore(&cm_node->retrans_list_lock, flags);
 1421                 irdma_rem_ref_cm_node(cm_node);
 1422         }
 1423 
 1424         if (settimer) {
 1425                 spin_lock_irqsave(&cm_core->ht_lock, flags);
 1426                 if (!timer_pending(&cm_core->tcp_timer)) {
 1427                         cm_core->tcp_timer.expires = nexttimeout;
 1428                         add_timer(&cm_core->tcp_timer);
 1429                 }
 1430                 spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 1431         }
 1432 }
 1433 
 1434 /**
 1435  * irdma_send_syn - send SYN packet
 1436  * @cm_node: connection's node
 1437  * @sendack: flag to set ACK bit or not
 1438  */
 1439 int
 1440 irdma_send_syn(struct irdma_cm_node *cm_node, u32 sendack)
 1441 {
 1442         struct irdma_puda_buf *sqbuf;
 1443         int flags = SET_SYN;
 1444         char optionsbuf[sizeof(struct option_mss) +
 1445                         sizeof(struct option_windowscale) +
 1446                         sizeof(struct option_base) + TCP_OPTIONS_PADDING];
 1447         struct irdma_kmem_info opts;
 1448         int optionssize = 0;
 1449         /* Sending MSS option */
 1450         union all_known_options *options;
 1451 
 1452         opts.addr = optionsbuf;
 1453         if (!cm_node)
 1454                 return -EINVAL;
 1455 
 1456         options = (union all_known_options *)&optionsbuf[optionssize];
 1457         options->mss.optionnum = OPTION_NUM_MSS;
 1458         options->mss.len = sizeof(struct option_mss);
 1459         options->mss.mss = htons(cm_node->tcp_cntxt.mss);
 1460         optionssize += sizeof(struct option_mss);
 1461 
 1462         options = (union all_known_options *)&optionsbuf[optionssize];
 1463         options->windowscale.optionnum = OPTION_NUM_WINDOW_SCALE;
 1464         options->windowscale.len = sizeof(struct option_windowscale);
 1465         options->windowscale.shiftcount = cm_node->tcp_cntxt.rcv_wscale;
 1466         optionssize += sizeof(struct option_windowscale);
 1467         options = (union all_known_options *)&optionsbuf[optionssize];
 1468         options->eol = OPTION_NUM_EOL;
 1469         optionssize += 1;
 1470 
 1471         if (sendack)
 1472                 flags |= SET_ACK;
 1473 
 1474         opts.size = optionssize;
 1475 
 1476         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, &opts, NULL, NULL,
 1477                                                 flags);
 1478         if (!sqbuf)
 1479                 return -ENOMEM;
 1480 
 1481         return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
 1482                                        0);
 1483 }
 1484 
 1485 /**
 1486  * irdma_send_ack - Send ACK packet
 1487  * @cm_node: connection's node
 1488  */
 1489 void
 1490 irdma_send_ack(struct irdma_cm_node *cm_node)
 1491 {
 1492         struct irdma_puda_buf *sqbuf;
 1493         struct irdma_sc_vsi *vsi = &cm_node->iwdev->vsi;
 1494 
 1495         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
 1496                                                 SET_ACK);
 1497         if (sqbuf)
 1498                 irdma_puda_send_buf(vsi->ilq, sqbuf);
 1499 }
 1500 
 1501 /**
 1502  * irdma_send_fin - Send FIN pkt
 1503  * @cm_node: connection's node
 1504  */
 1505 static int
 1506 irdma_send_fin(struct irdma_cm_node *cm_node)
 1507 {
 1508         struct irdma_puda_buf *sqbuf;
 1509 
 1510         sqbuf = cm_node->cm_core->form_cm_frame(cm_node, NULL, NULL, NULL,
 1511                                                 SET_ACK | SET_FIN);
 1512         if (!sqbuf)
 1513                 return -ENOMEM;
 1514 
 1515         return irdma_schedule_cm_timer(cm_node, sqbuf, IRDMA_TIMER_TYPE_SEND, 1,
 1516                                        0);
 1517 }
 1518 
 1519 /**
 1520  * irdma_find_listener - find a cm node listening on this addr-port pair
 1521  * @cm_core: cm's core
 1522  * @dst_addr: listener ip addr
 1523  * @dst_port: listener tcp port num
 1524  * @vlan_id: virtual LAN ID
 1525  * @listener_state: state to match with listen node's
 1526  */
 1527 static struct irdma_cm_listener *
 1528 irdma_find_listener(struct irdma_cm_core *cm_core, u32 *dst_addr, u16 dst_port,
 1529                     u16 vlan_id, enum irdma_cm_listener_state listener_state)
 1530 {
 1531         struct irdma_cm_listener *listen_node;
 1532         static const u32 ip_zero[4] = {0, 0, 0, 0};
 1533         u32 listen_addr[4];
 1534         u16 listen_port;
 1535         unsigned long flags;
 1536 
 1537         /* walk list and find cm_node associated with this session ID */
 1538         spin_lock_irqsave(&cm_core->listen_list_lock, flags);
 1539         list_for_each_entry(listen_node, &cm_core->listen_list, list) {
 1540                 memcpy(listen_addr, listen_node->loc_addr, sizeof(listen_addr));
 1541                 listen_port = listen_node->loc_port;
 1542                 if (listen_port != dst_port ||
 1543                     !(listener_state & listen_node->listener_state))
 1544                         continue;
 1545                 /* compare node pair, return node handle if a match */
 1546                 if (!memcmp(listen_addr, ip_zero, sizeof(listen_addr)) ||
 1547                     (!memcmp(listen_addr, dst_addr, sizeof(listen_addr)) &&
 1548                      vlan_id == listen_node->vlan_id)) {
 1549                         atomic_inc(&listen_node->refcnt);
 1550                         spin_unlock_irqrestore(&cm_core->listen_list_lock,
 1551                                                flags);
 1552                         return listen_node;
 1553                 }
 1554         }
 1555         spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 1556 
 1557         return NULL;
 1558 }
 1559 
 1560 /**
 1561  * irdma_del_multiple_qhash - Remove qhash and child listens
 1562  * @iwdev: iWarp device
 1563  * @cm_info: CM info for parent listen node
 1564  * @cm_parent_listen_node: The parent listen node
 1565  */
 1566 static int
 1567 irdma_del_multiple_qhash(struct irdma_device *iwdev,
 1568                          struct irdma_cm_info *cm_info,
 1569                          struct irdma_cm_listener *cm_parent_listen_node)
 1570 {
 1571         struct irdma_cm_listener *child_listen_node;
 1572         struct list_head *pos, *tpos;
 1573         unsigned long flags;
 1574         int ret = -EINVAL;
 1575 
 1576         spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
 1577         list_for_each_safe(pos, tpos,
 1578                            &cm_parent_listen_node->child_listen_list) {
 1579                 child_listen_node = list_entry(pos, struct irdma_cm_listener,
 1580                                                child_listen_list);
 1581                 if (child_listen_node->ipv4)
 1582                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1583                                     "removing child listen for IP=%pI4, port=%d, vlan=%d\n",
 1584                                     child_listen_node->loc_addr,
 1585                                     child_listen_node->loc_port,
 1586                                     child_listen_node->vlan_id);
 1587                 else
 1588                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1589                                     "removing child listen for IP=%pI6, port=%d, vlan=%d\n",
 1590                                     child_listen_node->loc_addr,
 1591                                     child_listen_node->loc_port,
 1592                                     child_listen_node->vlan_id);
 1593                 list_del(pos);
 1594                 memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
 1595                        sizeof(cm_info->loc_addr));
 1596                 cm_info->vlan_id = child_listen_node->vlan_id;
 1597                 if (child_listen_node->qhash_set) {
 1598                         ret = irdma_manage_qhash(iwdev, cm_info,
 1599                                                  IRDMA_QHASH_TYPE_TCP_SYN,
 1600                                                  IRDMA_QHASH_MANAGE_TYPE_DELETE,
 1601                                                  NULL, false);
 1602                         child_listen_node->qhash_set = false;
 1603                 } else {
 1604                         ret = 0;
 1605                 }
 1606                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1607                             "Child listen node freed = %p\n",
 1608                             child_listen_node);
 1609                 kfree(child_listen_node);
 1610                 cm_parent_listen_node->cm_core->stats_listen_nodes_destroyed++;
 1611         }
 1612         spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
 1613 
 1614         return ret;
 1615 }
 1616 
 1617 /**
 1618  * irdma_netdev_vlan_ipv6 - Gets the netdev and mac
 1619  * @addr: local IPv6 address
 1620  * @vlan_id: vlan id for the given IPv6 address
 1621  * @mac: mac address for the given IPv6 address
 1622  *
 1623  * Returns the net_device of the IPv6 address and also sets the
 1624  * vlan id and mac for that address.
 1625  */
 1626 struct ifnet *
 1627 irdma_netdev_vlan_ipv6(u32 *addr, u16 *vlan_id, u8 *mac)
 1628 {
 1629         struct ifnet *ip_dev = NULL;
 1630         struct in6_addr laddr6;
 1631         u16 scope_id = 0;
 1632 
 1633         irdma_copy_ip_htonl(laddr6.__u6_addr.__u6_addr32, addr);
 1634         if (vlan_id)
 1635                 *vlan_id = 0xFFFF;      /* Match rdma_vlan_dev_vlan_id() */
 1636         if (mac)
 1637                 eth_zero_addr(mac);
 1638 
 1639         if (IN6_IS_SCOPE_LINKLOCAL(&laddr6) ||
 1640             IN6_IS_ADDR_MC_INTFACELOCAL(&laddr6))
 1641                 scope_id = ntohs(laddr6.__u6_addr.__u6_addr16[1]);
 1642 
 1643         ip_dev = ip6_ifp_find(&init_net, laddr6, scope_id);
 1644         if (ip_dev) {
 1645                 if (vlan_id)
 1646                         *vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
 1647                 if (ip_dev->if_addr && ip_dev->if_addr->ifa_addr && mac)
 1648                         ether_addr_copy(mac, IF_LLADDR(ip_dev));
 1649         }
 1650 
 1651         return ip_dev;
 1652 }
 1653 
 1654 /**
 1655  * irdma_get_vlan_ipv4 - Returns the vlan_id for IPv4 address
 1656  * @addr: local IPv4 address
 1657  */
 1658 u16
 1659 irdma_get_vlan_ipv4(u32 *addr)
 1660 {
 1661         struct ifnet *netdev;
 1662         u16 vlan_id = 0xFFFF;
 1663 
 1664         netdev = ip_ifp_find(&init_net, htonl(addr[0]));
 1665         if (netdev) {
 1666                 vlan_id = rdma_vlan_dev_vlan_id(netdev);
 1667                 dev_put(netdev);
 1668         }
 1669 
 1670         return vlan_id;
 1671 }
 1672 
 1673 /**
 1674  * irdma_add_mqh_6 - Adds multiple qhashes for IPv6
 1675  * @iwdev: iWarp device
 1676  * @cm_info: CM info for parent listen node
 1677  * @cm_parent_listen_node: The parent listen node
 1678  *
 1679  * Adds a qhash and a child listen node for every IPv6 address
 1680  * on the adapter and adds the associated qhash filter
 1681  */
 1682 static int
 1683 irdma_add_mqh_6(struct irdma_device *iwdev,
 1684                 struct irdma_cm_info *cm_info,
 1685                 struct irdma_cm_listener *cm_parent_listen_node)
 1686 {
 1687         struct ifnet *ip_dev;
 1688         struct ifaddr *ifp;
 1689         struct irdma_cm_listener *child_listen_node;
 1690         unsigned long flags;
 1691         int ret = 0;
 1692 
 1693         IFNET_RLOCK();
 1694         IRDMA_TAILQ_FOREACH((ip_dev), &V_ifnet, if_link) {
 1695                 if (!(ip_dev->if_flags & IFF_UP))
 1696                         continue;
 1697 
 1698                 if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
 1699                      (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
 1700                     ip_dev != iwdev->netdev)
 1701                         continue;
 1702 
 1703                 if_addr_rlock(ip_dev);
 1704                 IRDMA_TAILQ_FOREACH(ifp, &ip_dev->if_addrhead, ifa_link) {
 1705                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1706                                     "IP=%pI6, vlan_id=%d, MAC=%pM\n",
 1707                                     &((struct sockaddr_in6 *)ifp->ifa_addr)->sin6_addr, rdma_vlan_dev_vlan_id(ip_dev),
 1708                                     IF_LLADDR(ip_dev));
 1709                         if (((struct sockaddr_in6 *)ifp->ifa_addr)->sin6_family != AF_INET6)
 1710                                 continue;
 1711                         child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
 1712                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1713                                     "Allocating child listener %p\n",
 1714                                     child_listen_node);
 1715                         if (!child_listen_node) {
 1716                                 irdma_debug(iwdev_to_idev(iwdev),
 1717                                             IRDMA_DEBUG_CM,
 1718                                             "listener memory allocation\n");
 1719                                 ret = -ENOMEM;
 1720                                 if_addr_runlock(ip_dev);
 1721                                 goto exit;
 1722                         }
 1723 
 1724                         memcpy(child_listen_node, cm_parent_listen_node,
 1725                                sizeof(*child_listen_node));
 1726                         cm_info->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
 1727                         child_listen_node->vlan_id = cm_info->vlan_id;
 1728                         irdma_copy_ip_ntohl(child_listen_node->loc_addr,
 1729                                             ((struct sockaddr_in6 *)ifp->ifa_addr)->sin6_addr.__u6_addr.__u6_addr32);
 1730                         memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
 1731                                sizeof(cm_info->loc_addr));
 1732                         ret = irdma_manage_qhash(iwdev, cm_info,
 1733                                                  IRDMA_QHASH_TYPE_TCP_SYN,
 1734                                                  IRDMA_QHASH_MANAGE_TYPE_ADD,
 1735                                                  NULL, true);
 1736                         if (ret) {
 1737                                 kfree(child_listen_node);
 1738                                 continue;
 1739                         }
 1740 
 1741                         child_listen_node->qhash_set = true;
 1742                         spin_lock_irqsave(&iwdev->cm_core.listen_list_lock, flags);
 1743                         list_add(&child_listen_node->child_listen_list,
 1744                                  &cm_parent_listen_node->child_listen_list);
 1745                         spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
 1746                         cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
 1747                 }
 1748                 if_addr_runlock(ip_dev);
 1749         }
 1750 exit:
 1751         IFNET_RUNLOCK();
 1752 
 1753         return ret;
 1754 }
 1755 
 1756 /**
 1757  * irdma_add_mqh_4 - Adds multiple qhashes for IPv4
 1758  * @iwdev: iWarp device
 1759  * @cm_info: CM info for parent listen node
 1760  * @cm_parent_listen_node: The parent listen node
 1761  *
 1762  * Adds a qhash and a child listen node for every IPv4 address
 1763  * on the adapter and adds the associated qhash filter
 1764  */
 1765 static int
 1766 irdma_add_mqh_4(struct irdma_device *iwdev,
 1767                 struct irdma_cm_info *cm_info,
 1768                 struct irdma_cm_listener *cm_parent_listen_node)
 1769 {
 1770         struct ifnet *ip_dev;
 1771         struct irdma_cm_listener *child_listen_node;
 1772         unsigned long flags;
 1773         struct ifaddr *ifa;
 1774         int ret = 0;
 1775 
 1776         IFNET_RLOCK();
 1777         IRDMA_TAILQ_FOREACH((ip_dev), &V_ifnet, if_link) {
 1778                 if (!(ip_dev->if_flags & IFF_UP))
 1779                         continue;
 1780 
 1781                 if (((rdma_vlan_dev_vlan_id(ip_dev) >= VLAN_N_VID) ||
 1782                      (rdma_vlan_dev_real_dev(ip_dev) != iwdev->netdev)) &&
 1783                     ip_dev != iwdev->netdev)
 1784                         continue;
 1785 
 1786                 if_addr_rlock(ip_dev);
 1787                 IRDMA_TAILQ_FOREACH(ifa, &ip_dev->if_addrhead, ifa_link) {
 1788                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1789                                     "Allocating child CM Listener forIP=%pI4, vlan_id=%d, MAC=%pM\n",
 1790                                     &ifa->ifa_addr,
 1791                                     rdma_vlan_dev_vlan_id(ip_dev), IF_LLADDR(ip_dev));
 1792                         if (((struct sockaddr_in *)ifa->ifa_addr)->sin_family != AF_INET)
 1793                                 continue;
 1794                         child_listen_node = kzalloc(sizeof(*child_listen_node), GFP_KERNEL);
 1795                         cm_parent_listen_node->cm_core->stats_listen_nodes_created++;
 1796                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1797                                     "Allocating child listener %p\n",
 1798                                     child_listen_node);
 1799                         if (!child_listen_node) {
 1800                                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 1801                                             "listener memory allocation\n");
 1802                                 if_addr_runlock(ip_dev);
 1803                                 ret = -ENOMEM;
 1804                                 goto exit;
 1805                         }
 1806 
 1807                         memcpy(child_listen_node, cm_parent_listen_node,
 1808                                sizeof(*child_listen_node));
 1809                         child_listen_node->vlan_id = rdma_vlan_dev_vlan_id(ip_dev);
 1810                         cm_info->vlan_id = child_listen_node->vlan_id;
 1811                         child_listen_node->loc_addr[0] =
 1812                             ntohl(((struct sockaddr_in *)ifa->ifa_addr)->sin_addr.s_addr);
 1813                         memcpy(cm_info->loc_addr, child_listen_node->loc_addr,
 1814                                sizeof(cm_info->loc_addr));
 1815                         ret = irdma_manage_qhash(iwdev, cm_info,
 1816                                                  IRDMA_QHASH_TYPE_TCP_SYN,
 1817                                                  IRDMA_QHASH_MANAGE_TYPE_ADD,
 1818                                                  NULL, true);
 1819                         if (ret) {
 1820                                 kfree(child_listen_node);
 1821                                 cm_parent_listen_node->cm_core
 1822                                     ->stats_listen_nodes_created--;
 1823                                 continue;
 1824                         }
 1825 
 1826                         child_listen_node->qhash_set = true;
 1827                         spin_lock_irqsave(&iwdev->cm_core.listen_list_lock,
 1828                                           flags);
 1829                         list_add(&child_listen_node->child_listen_list,
 1830                                  &cm_parent_listen_node->child_listen_list);
 1831                         spin_unlock_irqrestore(&iwdev->cm_core.listen_list_lock, flags);
 1832                 }
 1833                 if_addr_runlock(ip_dev);
 1834         }
 1835 exit:
 1836         IFNET_RUNLOCK();
 1837 
 1838         return ret;
 1839 }
 1840 
 1841 /**
 1842  * irdma_add_mqh - Adds multiple qhashes
 1843  * @iwdev: iWarp device
 1844  * @cm_info: CM info for parent listen node
 1845  * @cm_listen_node: The parent listen node
 1846  */
 1847 static int
 1848 irdma_add_mqh(struct irdma_device *iwdev,
 1849               struct irdma_cm_info *cm_info,
 1850               struct irdma_cm_listener *cm_listen_node)
 1851 {
 1852         int err;
 1853         VNET_ITERATOR_DECL(vnet_iter);
 1854 
 1855         VNET_LIST_RLOCK();
 1856         VNET_FOREACH(vnet_iter) {
 1857                 IFNET_RLOCK();
 1858                 CURVNET_SET_QUIET(vnet_iter);
 1859 
 1860                 if (cm_info->ipv4)
 1861                         err = irdma_add_mqh_4(iwdev, cm_info, cm_listen_node);
 1862                 else
 1863                         err = irdma_add_mqh_6(iwdev, cm_info, cm_listen_node);
 1864                 CURVNET_RESTORE();
 1865                 IFNET_RUNLOCK();
 1866         }
 1867         VNET_LIST_RUNLOCK();
 1868 
 1869         return err;
 1870 }
 1871 
 1872 /**
 1873  * irdma_reset_list_prep - add connection nodes slated for reset to list
 1874  * @cm_core: cm's core
 1875  * @listener: pointer to listener node
 1876  * @reset_list: a list to which cm_node will be selected
 1877  */
 1878 static void
 1879 irdma_reset_list_prep(struct irdma_cm_core *cm_core,
 1880                       struct irdma_cm_listener *listener,
 1881                       struct list_head *reset_list)
 1882 {
 1883         struct irdma_cm_node *cm_node;
 1884         int bkt;
 1885 
 1886         HASH_FOR_EACH_RCU(cm_core->cm_hash_tbl, bkt, cm_node, list) {
 1887                 if (cm_node->listener == listener &&
 1888                     !cm_node->accelerated &&
 1889                     atomic_inc_not_zero(&cm_node->refcnt))
 1890                         list_add(&cm_node->reset_entry, reset_list);
 1891         }
 1892 }
 1893 
 1894 /**
 1895  * irdma_dec_refcnt_listen - delete listener and associated cm nodes
 1896  * @cm_core: cm's core
 1897  * @listener: pointer to listener node
 1898  * @free_hanging_nodes: to free associated cm_nodes
 1899  * @apbvt_del: flag to delete the apbvt
 1900  */
 1901 static int
 1902 irdma_dec_refcnt_listen(struct irdma_cm_core *cm_core,
 1903                         struct irdma_cm_listener *listener,
 1904                         int free_hanging_nodes, bool apbvt_del)
 1905 {
 1906         struct list_head *list_pos;
 1907         struct list_head *list_temp;
 1908         struct irdma_cm_node *cm_node;
 1909         struct list_head reset_list;
 1910         struct irdma_cm_info nfo;
 1911         enum irdma_cm_node_state old_state;
 1912         unsigned long flags;
 1913         int err;
 1914 
 1915         /* free non-accelerated child nodes for this listener */
 1916         INIT_LIST_HEAD(&reset_list);
 1917         if (free_hanging_nodes) {
 1918                 rcu_read_lock();
 1919                 irdma_reset_list_prep(cm_core, listener, &reset_list);
 1920                 rcu_read_unlock();
 1921         }
 1922 
 1923         list_for_each_safe(list_pos, list_temp, &reset_list) {
 1924                 cm_node = container_of(list_pos, struct irdma_cm_node,
 1925                                        reset_entry);
 1926                 if (cm_node->state >= IRDMA_CM_STATE_FIN_WAIT1) {
 1927                         irdma_rem_ref_cm_node(cm_node);
 1928                         continue;
 1929                 }
 1930 
 1931                 irdma_cleanup_retrans_entry(cm_node);
 1932                 err = irdma_send_reset(cm_node);
 1933                 if (err) {
 1934                         cm_node->state = IRDMA_CM_STATE_CLOSED;
 1935                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 1936                                     IRDMA_DEBUG_CM, "send reset failed\n");
 1937                 } else {
 1938                         old_state = cm_node->state;
 1939                         cm_node->state = IRDMA_CM_STATE_LISTENER_DESTROYED;
 1940                         if (old_state != IRDMA_CM_STATE_MPAREQ_RCVD)
 1941                                 irdma_rem_ref_cm_node(cm_node);
 1942                 }
 1943         }
 1944 
 1945         if (atomic_dec_and_test(&listener->refcnt)) {
 1946                 spin_lock_irqsave(&cm_core->listen_list_lock, flags);
 1947                 list_del(&listener->list);
 1948                 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 1949 
 1950                 if (apbvt_del)
 1951                         irdma_del_apbvt(listener->iwdev,
 1952                                         listener->apbvt_entry);
 1953                 memcpy(nfo.loc_addr, listener->loc_addr, sizeof(nfo.loc_addr));
 1954                 nfo.loc_port = listener->loc_port;
 1955                 nfo.ipv4 = listener->ipv4;
 1956                 nfo.vlan_id = listener->vlan_id;
 1957                 nfo.user_pri = listener->user_pri;
 1958                 nfo.qh_qpid = listener->iwdev->vsi.ilq->qp_id;
 1959 
 1960                 if (!list_empty(&listener->child_listen_list)) {
 1961                         irdma_del_multiple_qhash(listener->iwdev, &nfo,
 1962                                                  listener);
 1963                 } else {
 1964                         if (listener->qhash_set)
 1965                                 irdma_manage_qhash(listener->iwdev,
 1966                                                    &nfo,
 1967                                                    IRDMA_QHASH_TYPE_TCP_SYN,
 1968                                                    IRDMA_QHASH_MANAGE_TYPE_DELETE,
 1969                                                    NULL, false);
 1970                 }
 1971 
 1972                 cm_core->stats_listen_destroyed++;
 1973                 cm_core->stats_listen_nodes_destroyed++;
 1974                 irdma_debug(iwdev_to_idev(listener->iwdev), IRDMA_DEBUG_CM,
 1975                     "loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d apbvt_del=%d\n",
 1976                             listener->loc_port, listener->loc_addr, listener, listener->cm_id, listener->qhash_set,
 1977                             listener->vlan_id, apbvt_del);
 1978                 kfree(listener);
 1979                 listener = NULL;
 1980                 return 0;
 1981         }
 1982 
 1983         return -EINVAL;
 1984 }
 1985 
 1986 /**
 1987  * irdma_cm_del_listen - delete a listener
 1988  * @cm_core: cm's core
 1989  * @listener: passive connection's listener
 1990  * @apbvt_del: flag to delete apbvt
 1991  */
 1992 static int
 1993 irdma_cm_del_listen(struct irdma_cm_core *cm_core,
 1994                     struct irdma_cm_listener *listener,
 1995                     bool apbvt_del)
 1996 {
 1997         listener->listener_state = IRDMA_CM_LISTENER_PASSIVE_STATE;
 1998         listener->cm_id = NULL;
 1999 
 2000         return irdma_dec_refcnt_listen(cm_core, listener, 1, apbvt_del);
 2001 }
 2002 
 2003 /**
 2004  * irdma_find_node - find a cm node that matches the reference cm node
 2005  * @cm_core: cm's core
 2006  * @rem_port: remote tcp port num
 2007  * @rem_addr: remote ip addr
 2008  * @loc_port: local tcp port num
 2009  * @loc_addr: local ip addr
 2010  * @vlan_id: local VLAN ID
 2011  */
 2012 struct irdma_cm_node *
 2013 irdma_find_node(struct irdma_cm_core *cm_core,
 2014                 u16 rem_port, u32 *rem_addr, u16 loc_port,
 2015                 u32 *loc_addr, u16 vlan_id)
 2016 {
 2017         struct irdma_cm_node *cm_node;
 2018         u32 key = (rem_port << 16) | loc_port;
 2019 
 2020         rcu_read_lock();
 2021         HASH_FOR_EACH_POSSIBLE_RCU(cm_core->cm_hash_tbl, cm_node, list, key) {
 2022                 if (cm_node->vlan_id == vlan_id &&
 2023                     cm_node->loc_port == loc_port && cm_node->rem_port == rem_port &&
 2024                     !memcmp(cm_node->loc_addr, loc_addr, sizeof(cm_node->loc_addr)) &&
 2025                     !memcmp(cm_node->rem_addr, rem_addr, sizeof(cm_node->rem_addr))) {
 2026                         if (!atomic_inc_not_zero(&cm_node->refcnt))
 2027                                 goto exit;
 2028                         rcu_read_unlock();
 2029                         return cm_node;
 2030                 }
 2031         }
 2032 
 2033 exit:
 2034         rcu_read_unlock();
 2035 
 2036         /* no owner node */
 2037         return NULL;
 2038 }
 2039 
 2040 /**
 2041  * irdma_add_hte_node - add a cm node to the hash table
 2042  * @cm_core: cm's core
 2043  * @cm_node: connection's node
 2044  */
 2045 static void
 2046 irdma_add_hte_node(struct irdma_cm_core *cm_core,
 2047                    struct irdma_cm_node *cm_node)
 2048 {
 2049         unsigned long flags;
 2050         u32 key = (cm_node->rem_port << 16) | cm_node->loc_port;
 2051 
 2052         spin_lock_irqsave(&cm_core->ht_lock, flags);
 2053         HASH_ADD_RCU(cm_core->cm_hash_tbl, &cm_node->list, key);
 2054         spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 2055 }
 2056 
 2057 /**
 2058  * irdma_ipv4_is_lpb - check if loopback
 2059  * @loc_addr: local addr to compare
 2060  * @rem_addr: remote address
 2061  */
 2062 bool
 2063 irdma_ipv4_is_lpb(u32 loc_addr, u32 rem_addr)
 2064 {
 2065         return ipv4_is_loopback(htonl(rem_addr)) || (loc_addr == rem_addr);
 2066 }
 2067 
 2068 /**
 2069  * irdma_ipv6_is_lpb - check if loopback
 2070  * @loc_addr: local addr to compare
 2071  * @rem_addr: remote address
 2072  */
 2073 bool
 2074 irdma_ipv6_is_lpb(u32 *loc_addr, u32 *rem_addr)
 2075 {
 2076         struct in6_addr raddr6;
 2077 
 2078         irdma_copy_ip_htonl(raddr6.__u6_addr.__u6_addr32, rem_addr);
 2079 
 2080         return !memcmp(loc_addr, rem_addr, 16) || ipv6_addr_loopback(&raddr6);
 2081 }
 2082 
 2083 /**
 2084  * irdma_cm_create_ah - create a cm address handle
 2085  * @cm_node: The connection manager node to create AH for
 2086  * @wait: Provides option to wait for ah creation or not
 2087  */
 2088 static int
 2089 irdma_cm_create_ah(struct irdma_cm_node *cm_node, bool wait)
 2090 {
 2091         struct irdma_ah_info ah_info = {0};
 2092         struct irdma_device *iwdev = cm_node->iwdev;
 2093 #ifdef VIMAGE
 2094         struct rdma_cm_id *rdma_id = (struct rdma_cm_id *)cm_node->cm_id->context;
 2095         struct vnet *vnet = rdma_id->route.addr.dev_addr.net;
 2096 #endif
 2097 
 2098         ether_addr_copy(ah_info.mac_addr, IF_LLADDR(iwdev->netdev));
 2099 
 2100         ah_info.hop_ttl = 0x40;
 2101         ah_info.tc_tos = cm_node->tos;
 2102         ah_info.vsi = &iwdev->vsi;
 2103 
 2104         if (cm_node->ipv4) {
 2105                 ah_info.ipv4_valid = true;
 2106                 ah_info.dest_ip_addr[0] = cm_node->rem_addr[0];
 2107                 ah_info.src_ip_addr[0] = cm_node->loc_addr[0];
 2108 #ifdef VIMAGE
 2109                 CURVNET_SET_QUIET(vnet);
 2110                 ah_info.do_lpbk = irdma_ipv4_is_lpb(ah_info.src_ip_addr[0],
 2111                                                     ah_info.dest_ip_addr[0]);
 2112                 CURVNET_RESTORE();
 2113 #endif
 2114         } else {
 2115                 memcpy(ah_info.dest_ip_addr, cm_node->rem_addr,
 2116                        sizeof(ah_info.dest_ip_addr));
 2117                 memcpy(ah_info.src_ip_addr, cm_node->loc_addr,
 2118                        sizeof(ah_info.src_ip_addr));
 2119                 ah_info.do_lpbk = irdma_ipv6_is_lpb(ah_info.src_ip_addr,
 2120                                                     ah_info.dest_ip_addr);
 2121         }
 2122 
 2123         ah_info.vlan_tag = cm_node->vlan_id;
 2124         if (cm_node->vlan_id < VLAN_N_VID) {
 2125                 ah_info.insert_vlan_tag = 1;
 2126                 ah_info.vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
 2127         }
 2128 
 2129         ah_info.dst_arpindex =
 2130             irdma_arp_table(iwdev->rf, ah_info.dest_ip_addr,
 2131                             NULL, IRDMA_ARP_RESOLVE);
 2132 
 2133         if (irdma_puda_create_ah(&iwdev->rf->sc_dev, &ah_info, wait,
 2134                                  IRDMA_PUDA_RSRC_TYPE_ILQ, cm_node,
 2135                                  &cm_node->ah))
 2136                 return -ENOMEM;
 2137 
 2138         return 0;
 2139 }
 2140 
 2141 /**
 2142  * irdma_cm_free_ah - free a cm address handle
 2143  * @cm_node: The connection manager node to create AH for
 2144  */
 2145 static void
 2146 irdma_cm_free_ah(struct irdma_cm_node *cm_node)
 2147 {
 2148         struct irdma_device *iwdev = cm_node->iwdev;
 2149 
 2150         irdma_puda_free_ah(&iwdev->rf->sc_dev, cm_node->ah);
 2151         cm_node->ah = NULL;
 2152 }
 2153 
 2154 /**
 2155  * irdma_make_cm_node - create a new instance of a cm node
 2156  * @cm_core: cm's core
 2157  * @iwdev: iwarp device structure
 2158  * @cm_info: quad info for connection
 2159  * @listener: passive connection's listener
 2160  */
 2161 static struct irdma_cm_node *
 2162 irdma_make_cm_node(struct irdma_cm_core *cm_core, struct irdma_device *iwdev,
 2163                    struct irdma_cm_info *cm_info,
 2164                    struct irdma_cm_listener *listener)
 2165 {
 2166         struct irdma_cm_node *cm_node;
 2167         int arpindex;
 2168         struct ifnet *netdev = iwdev->netdev;
 2169 
 2170         /* create an hte and cm_node for this instance */
 2171         cm_node = kzalloc(sizeof(*cm_node), GFP_ATOMIC);
 2172         if (!cm_node)
 2173                 return NULL;
 2174 
 2175         /* set our node specific transport info */
 2176         cm_node->ipv4 = cm_info->ipv4;
 2177         cm_node->vlan_id = cm_info->vlan_id;
 2178         if (cm_node->vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
 2179                 cm_node->vlan_id = 0;
 2180         cm_node->tos = cm_info->tos;
 2181         cm_node->user_pri = cm_info->user_pri;
 2182         if (listener) {
 2183                 if (listener->tos != cm_info->tos)
 2184                         irdma_dev_warn(
 2185                                        &iwdev->rf->sc_dev,
 2186                                        "application TOS[%d] and remote client TOS[%d] mismatch\n",
 2187                                        listener->tos, cm_info->tos);
 2188                 if (iwdev->vsi.dscp_mode) {
 2189                         cm_node->user_pri = listener->user_pri;
 2190                 } else {
 2191                         cm_node->tos = max(listener->tos, cm_info->tos);
 2192                         cm_node->user_pri = rt_tos2priority(cm_node->tos);
 2193                 }
 2194                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_DCB,
 2195                             "listener: TOS:[%d] UP:[%d]\n", cm_node->tos,
 2196                             cm_node->user_pri);
 2197         }
 2198         memcpy(cm_node->loc_addr, cm_info->loc_addr, sizeof(cm_node->loc_addr));
 2199         memcpy(cm_node->rem_addr, cm_info->rem_addr, sizeof(cm_node->rem_addr));
 2200         cm_node->loc_port = cm_info->loc_port;
 2201         cm_node->rem_port = cm_info->rem_port;
 2202 
 2203         cm_node->mpa_frame_rev = IRDMA_CM_DEFAULT_MPA_VER;
 2204         cm_node->send_rdma0_op = SEND_RDMA_READ_ZERO;
 2205         cm_node->iwdev = iwdev;
 2206         cm_node->dev = &iwdev->rf->sc_dev;
 2207 
 2208         cm_node->ird_size = cm_node->dev->hw_attrs.max_hw_ird;
 2209         cm_node->ord_size = cm_node->dev->hw_attrs.max_hw_ord;
 2210 
 2211         cm_node->listener = listener;
 2212         cm_node->cm_id = cm_info->cm_id;
 2213         ether_addr_copy(cm_node->loc_mac, IF_LLADDR(netdev));
 2214         spin_lock_init(&cm_node->retrans_list_lock);
 2215         cm_node->ack_rcvd = false;
 2216 
 2217         init_completion(&cm_node->establish_comp);
 2218         atomic_set(&cm_node->refcnt, 1);
 2219         /* associate our parent CM core */
 2220         cm_node->cm_core = cm_core;
 2221         cm_node->tcp_cntxt.loc_id = IRDMA_CM_DEFAULT_LOCAL_ID;
 2222         cm_node->tcp_cntxt.rcv_wscale = iwdev->rcv_wscale;
 2223         cm_node->tcp_cntxt.rcv_wnd = iwdev->rcv_wnd >> cm_node->tcp_cntxt.rcv_wscale;
 2224         kc_set_loc_seq_num_mss(cm_node);
 2225 
 2226         arpindex = irdma_resolve_neigh_lpb_chk(iwdev, cm_node, cm_info);
 2227         if (arpindex < 0)
 2228                 goto err;
 2229 
 2230         ether_addr_copy(cm_node->rem_mac, iwdev->rf->arp_table[arpindex].mac_addr);
 2231         irdma_add_hte_node(cm_core, cm_node);
 2232         cm_core->stats_nodes_created++;
 2233         return cm_node;
 2234 
 2235 err:
 2236         kfree(cm_node);
 2237 
 2238         return NULL;
 2239 }
 2240 
 2241 static void
 2242 irdma_destroy_connection(struct irdma_cm_node *cm_node)
 2243 {
 2244         struct irdma_cm_core *cm_core = cm_node->cm_core;
 2245         struct irdma_qp *iwqp;
 2246         struct irdma_cm_info nfo;
 2247 
 2248         /* if the node is destroyed before connection was accelerated */
 2249         if (!cm_node->accelerated && cm_node->accept_pend) {
 2250                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2251                             "node destroyed before established\n");
 2252                 atomic_dec(&cm_node->listener->pend_accepts_cnt);
 2253         }
 2254         if (cm_node->close_entry)
 2255                 irdma_handle_close_entry(cm_node, 0);
 2256         if (cm_node->listener) {
 2257                 irdma_dec_refcnt_listen(cm_core, cm_node->listener, 0, true);
 2258         } else {
 2259                 if (cm_node->apbvt_set) {
 2260                         irdma_del_apbvt(cm_node->iwdev, cm_node->apbvt_entry);
 2261                         cm_node->apbvt_set = 0;
 2262                 }
 2263                 irdma_get_addr_info(cm_node, &nfo);
 2264                 if (cm_node->qhash_set) {
 2265                         nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
 2266                         irdma_manage_qhash(cm_node->iwdev, &nfo,
 2267                                            IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
 2268                                            IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL,
 2269                                            false);
 2270                         cm_node->qhash_set = 0;
 2271                 }
 2272         }
 2273 
 2274         iwqp = cm_node->iwqp;
 2275         if (iwqp) {
 2276                 cm_node->cm_id->rem_ref(cm_node->cm_id);
 2277                 cm_node->cm_id = NULL;
 2278                 iwqp->cm_id = NULL;
 2279                 irdma_qp_rem_ref(&iwqp->ibqp);
 2280                 cm_node->iwqp = NULL;
 2281         } else if (cm_node->qhash_set) {
 2282                 irdma_get_addr_info(cm_node, &nfo);
 2283                 nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
 2284                 irdma_manage_qhash(cm_node->iwdev, &nfo,
 2285                                    IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
 2286                                    IRDMA_QHASH_MANAGE_TYPE_DELETE, NULL, false);
 2287                 cm_node->qhash_set = 0;
 2288         }
 2289 
 2290         cm_core->cm_free_ah(cm_node);
 2291 }
 2292 
 2293 /**
 2294  * irdma_rem_ref_cm_node - destroy an instance of a cm node
 2295  * @cm_node: connection's node
 2296  */
 2297 void
 2298 irdma_rem_ref_cm_node(struct irdma_cm_node *cm_node)
 2299 {
 2300         struct irdma_cm_core *cm_core = cm_node->cm_core;
 2301         unsigned long flags;
 2302 
 2303         spin_lock_irqsave(&cm_core->ht_lock, flags);
 2304 
 2305         if (!atomic_dec_and_test(&cm_node->refcnt)) {
 2306                 spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 2307                 return;
 2308         }
 2309         if (cm_node->iwqp) {
 2310                 cm_node->iwqp->cm_node = NULL;
 2311                 cm_node->iwqp->cm_id = NULL;
 2312         }
 2313         HASH_DEL_RCU(cm_core->cm_hash_tbl, &cm_node->list);
 2314         cm_node->cm_core->stats_nodes_destroyed++;
 2315 
 2316         spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 2317 
 2318         irdma_destroy_connection(cm_node);
 2319 
 2320         kfree_rcu(cm_node, rcu_head);
 2321 }
 2322 
 2323 /**
 2324  * irdma_handle_fin_pkt - FIN packet received
 2325  * @cm_node: connection's node
 2326  */
 2327 static void
 2328 irdma_handle_fin_pkt(struct irdma_cm_node *cm_node)
 2329 {
 2330         switch (cm_node->state) {
 2331         case IRDMA_CM_STATE_SYN_RCVD:
 2332         case IRDMA_CM_STATE_SYN_SENT:
 2333         case IRDMA_CM_STATE_ESTABLISHED:
 2334         case IRDMA_CM_STATE_MPAREJ_RCVD:
 2335                 cm_node->tcp_cntxt.rcv_nxt++;
 2336                 irdma_cleanup_retrans_entry(cm_node);
 2337                 cm_node->state = IRDMA_CM_STATE_LAST_ACK;
 2338                 irdma_send_fin(cm_node);
 2339                 break;
 2340         case IRDMA_CM_STATE_MPAREQ_SENT:
 2341                 irdma_create_event(cm_node, IRDMA_CM_EVENT_ABORTED);
 2342                 cm_node->tcp_cntxt.rcv_nxt++;
 2343                 irdma_cleanup_retrans_entry(cm_node);
 2344                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2345                 atomic_inc(&cm_node->refcnt);
 2346                 irdma_send_reset(cm_node);
 2347                 break;
 2348         case IRDMA_CM_STATE_FIN_WAIT1:
 2349                 cm_node->tcp_cntxt.rcv_nxt++;
 2350                 irdma_cleanup_retrans_entry(cm_node);
 2351                 cm_node->state = IRDMA_CM_STATE_CLOSING;
 2352                 irdma_send_ack(cm_node);
 2353                 /*
 2354                  * Wait for ACK as this is simultaneous close. After we receive ACK, do not send anything. Just rm the
 2355                  * node.
 2356                  */
 2357                 break;
 2358         case IRDMA_CM_STATE_FIN_WAIT2:
 2359                 cm_node->tcp_cntxt.rcv_nxt++;
 2360                 irdma_cleanup_retrans_entry(cm_node);
 2361                 cm_node->state = IRDMA_CM_STATE_TIME_WAIT;
 2362                 irdma_send_ack(cm_node);
 2363                 irdma_schedule_cm_timer(cm_node, NULL, IRDMA_TIMER_TYPE_CLOSE,
 2364                                         1, 0);
 2365                 break;
 2366         case IRDMA_CM_STATE_TIME_WAIT:
 2367                 cm_node->tcp_cntxt.rcv_nxt++;
 2368                 irdma_cleanup_retrans_entry(cm_node);
 2369                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2370                 irdma_rem_ref_cm_node(cm_node);
 2371                 break;
 2372         case IRDMA_CM_STATE_OFFLOADED:
 2373         default:
 2374                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2375                             "bad state node state = %d\n", cm_node->state);
 2376                 break;
 2377         }
 2378 }
 2379 
 2380 /**
 2381  * irdma_handle_rst_pkt - process received RST packet
 2382  * @cm_node: connection's node
 2383  * @rbuf: receive buffer
 2384  */
 2385 static void
 2386 irdma_handle_rst_pkt(struct irdma_cm_node *cm_node,
 2387                      struct irdma_puda_buf *rbuf)
 2388 {
 2389         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2390                     "caller: %pS cm_node=%p state=%d rem_port=0x%04x loc_port=0x%04x rem_addr=%pI4 loc_addr=%pI4\n",
 2391                     __builtin_return_address(0), cm_node, cm_node->state,
 2392                     cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr, cm_node->loc_addr);
 2393 
 2394         irdma_cleanup_retrans_entry(cm_node);
 2395         switch (cm_node->state) {
 2396         case IRDMA_CM_STATE_SYN_SENT:
 2397         case IRDMA_CM_STATE_MPAREQ_SENT:
 2398                 switch (cm_node->mpa_frame_rev) {
 2399                 case IETF_MPA_V2:
 2400                         /* Drop down to MPA_V1 */
 2401                         cm_node->mpa_frame_rev = IETF_MPA_V1;
 2402                         /* send a syn and goto syn sent state */
 2403                         cm_node->state = IRDMA_CM_STATE_SYN_SENT;
 2404                         if (irdma_send_syn(cm_node, 0))
 2405                                 irdma_active_open_err(cm_node, false);
 2406                         break;
 2407                 case IETF_MPA_V1:
 2408                 default:
 2409                         irdma_active_open_err(cm_node, false);
 2410                         break;
 2411                 }
 2412                 break;
 2413         case IRDMA_CM_STATE_MPAREQ_RCVD:
 2414                 atomic_inc(&cm_node->passive_state);
 2415                 break;
 2416         case IRDMA_CM_STATE_ESTABLISHED:
 2417         case IRDMA_CM_STATE_SYN_RCVD:
 2418         case IRDMA_CM_STATE_LISTENING:
 2419                 irdma_passive_open_err(cm_node, false);
 2420                 break;
 2421         case IRDMA_CM_STATE_OFFLOADED:
 2422                 irdma_active_open_err(cm_node, false);
 2423                 break;
 2424         case IRDMA_CM_STATE_CLOSED:
 2425                 break;
 2426         case IRDMA_CM_STATE_FIN_WAIT2:
 2427         case IRDMA_CM_STATE_FIN_WAIT1:
 2428         case IRDMA_CM_STATE_LAST_ACK:
 2429         case IRDMA_CM_STATE_TIME_WAIT:
 2430                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2431                 irdma_rem_ref_cm_node(cm_node);
 2432                 break;
 2433         default:
 2434                 break;
 2435         }
 2436 }
 2437 
 2438 /**
 2439  * irdma_handle_rcv_mpa - Process a recv'd mpa buffer
 2440  * @cm_node: connection's node
 2441  * @rbuf: receive buffer
 2442  */
 2443 static void
 2444 irdma_handle_rcv_mpa(struct irdma_cm_node *cm_node,
 2445                      struct irdma_puda_buf *rbuf)
 2446 {
 2447         int err;
 2448         int datasize = rbuf->datalen;
 2449         u8 *dataloc = rbuf->data;
 2450 
 2451         enum irdma_cm_event_type type = IRDMA_CM_EVENT_UNKNOWN;
 2452         u32 res_type;
 2453 
 2454         err = irdma_parse_mpa(cm_node, dataloc, &res_type, datasize);
 2455         if (err) {
 2456                 if (cm_node->state == IRDMA_CM_STATE_MPAREQ_SENT)
 2457                         irdma_active_open_err(cm_node, true);
 2458                 else
 2459                         irdma_passive_open_err(cm_node, true);
 2460                 return;
 2461         }
 2462 
 2463         switch (cm_node->state) {
 2464         case IRDMA_CM_STATE_ESTABLISHED:
 2465                 if (res_type == IRDMA_MPA_REQUEST_REJECT)
 2466                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 2467                                     IRDMA_DEBUG_CM, "state for reject\n");
 2468                 cm_node->state = IRDMA_CM_STATE_MPAREQ_RCVD;
 2469                 type = IRDMA_CM_EVENT_MPA_REQ;
 2470                 irdma_send_ack(cm_node);        /* ACK received MPA request */
 2471                 atomic_set(&cm_node->passive_state,
 2472                            IRDMA_PASSIVE_STATE_INDICATED);
 2473                 break;
 2474         case IRDMA_CM_STATE_MPAREQ_SENT:
 2475                 irdma_cleanup_retrans_entry(cm_node);
 2476                 if (res_type == IRDMA_MPA_REQUEST_REJECT) {
 2477                         type = IRDMA_CM_EVENT_MPA_REJECT;
 2478                         cm_node->state = IRDMA_CM_STATE_MPAREJ_RCVD;
 2479                 } else {
 2480                         type = IRDMA_CM_EVENT_CONNECTED;
 2481                         cm_node->state = IRDMA_CM_STATE_OFFLOADED;
 2482                 }
 2483                 irdma_send_ack(cm_node);
 2484                 break;
 2485         default:
 2486                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2487                             "wrong cm_node state =%d\n", cm_node->state);
 2488                 break;
 2489         }
 2490         irdma_create_event(cm_node, type);
 2491 }
 2492 
 2493 /**
 2494  * irdma_check_syn - Check for error on received syn ack
 2495  * @cm_node: connection's node
 2496  * @tcph: pointer tcp header
 2497  */
 2498 static int
 2499 irdma_check_syn(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
 2500 {
 2501         if (ntohl(tcph->th_ack) != cm_node->tcp_cntxt.loc_seq_num) {
 2502                 irdma_active_open_err(cm_node, true);
 2503                 return 1;
 2504         }
 2505 
 2506         return 0;
 2507 }
 2508 
 2509 /**
 2510  * irdma_check_seq - check seq numbers if OK
 2511  * @cm_node: connection's node
 2512  * @tcph: pointer tcp header
 2513  */
 2514 static int
 2515 irdma_check_seq(struct irdma_cm_node *cm_node, struct tcphdr *tcph)
 2516 {
 2517         u32 seq;
 2518         u32 ack_seq;
 2519         u32 loc_seq_num = cm_node->tcp_cntxt.loc_seq_num;
 2520         u32 rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
 2521         u32 rcv_wnd;
 2522         int err = 0;
 2523 
 2524         seq = ntohl(tcph->th_seq);
 2525         ack_seq = ntohl(tcph->th_ack);
 2526         rcv_wnd = cm_node->tcp_cntxt.rcv_wnd;
 2527         if (ack_seq != loc_seq_num ||
 2528             !between(seq, rcv_nxt, (rcv_nxt + rcv_wnd)))
 2529                 err = -1;
 2530         if (err)
 2531                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2532                             "seq number err\n");
 2533 
 2534         return err;
 2535 }
 2536 
 2537 void
 2538 irdma_add_conn_est_qh(struct irdma_cm_node *cm_node)
 2539 {
 2540         struct irdma_cm_info nfo;
 2541 
 2542         irdma_get_addr_info(cm_node, &nfo);
 2543         nfo.qh_qpid = cm_node->iwdev->vsi.ilq->qp_id;
 2544         irdma_manage_qhash(cm_node->iwdev, &nfo,
 2545                            IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
 2546                            IRDMA_QHASH_MANAGE_TYPE_ADD,
 2547                            cm_node, false);
 2548         cm_node->qhash_set = true;
 2549 }
 2550 
 2551 /**
 2552  * irdma_handle_syn_pkt - is for Passive node
 2553  * @cm_node: connection's node
 2554  * @rbuf: receive buffer
 2555  */
 2556 static void
 2557 irdma_handle_syn_pkt(struct irdma_cm_node *cm_node,
 2558                      struct irdma_puda_buf *rbuf)
 2559 {
 2560         struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
 2561         int err;
 2562         u32 inc_sequence;
 2563         int optionsize;
 2564 
 2565         optionsize = (tcph->th_off << 2) - sizeof(struct tcphdr);
 2566         inc_sequence = ntohl(tcph->th_seq);
 2567 
 2568         switch (cm_node->state) {
 2569         case IRDMA_CM_STATE_SYN_SENT:
 2570         case IRDMA_CM_STATE_MPAREQ_SENT:
 2571                 /* Rcvd syn on active open connection */
 2572                 irdma_active_open_err(cm_node, 1);
 2573                 break;
 2574         case IRDMA_CM_STATE_LISTENING:
 2575                 /* Passive OPEN */
 2576                 if (atomic_read(&cm_node->listener->pend_accepts_cnt) >
 2577                     cm_node->listener->backlog) {
 2578                         cm_node->cm_core->stats_backlog_drops++;
 2579                         irdma_passive_open_err(cm_node, false);
 2580                         break;
 2581                 }
 2582                 err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
 2583                 if (err) {
 2584                         irdma_passive_open_err(cm_node, false);
 2585                         /* drop pkt */
 2586                         break;
 2587                 }
 2588                 err = cm_node->cm_core->cm_create_ah(cm_node, false);
 2589                 if (err) {
 2590                         irdma_passive_open_err(cm_node, false);
 2591                         /* drop pkt */
 2592                         break;
 2593                 }
 2594                 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
 2595                 cm_node->accept_pend = 1;
 2596                 atomic_inc(&cm_node->listener->pend_accepts_cnt);
 2597 
 2598                 cm_node->state = IRDMA_CM_STATE_SYN_RCVD;
 2599                 break;
 2600         case IRDMA_CM_STATE_CLOSED:
 2601                 irdma_cleanup_retrans_entry(cm_node);
 2602                 atomic_inc(&cm_node->refcnt);
 2603                 irdma_send_reset(cm_node);
 2604                 break;
 2605         case IRDMA_CM_STATE_OFFLOADED:
 2606         case IRDMA_CM_STATE_ESTABLISHED:
 2607         case IRDMA_CM_STATE_FIN_WAIT1:
 2608         case IRDMA_CM_STATE_FIN_WAIT2:
 2609         case IRDMA_CM_STATE_MPAREQ_RCVD:
 2610         case IRDMA_CM_STATE_LAST_ACK:
 2611         case IRDMA_CM_STATE_CLOSING:
 2612         case IRDMA_CM_STATE_UNKNOWN:
 2613         default:
 2614                 break;
 2615         }
 2616 }
 2617 
 2618 /**
 2619  * irdma_handle_synack_pkt - Process SYN+ACK packet (active side)
 2620  * @cm_node: connection's node
 2621  * @rbuf: receive buffer
 2622  */
 2623 static void
 2624 irdma_handle_synack_pkt(struct irdma_cm_node *cm_node,
 2625                         struct irdma_puda_buf *rbuf)
 2626 {
 2627         struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
 2628         int err;
 2629         u32 inc_sequence;
 2630         int optionsize;
 2631 
 2632         optionsize = (tcph->th_off << 2) - sizeof(struct tcphdr);
 2633         inc_sequence = ntohl(tcph->th_seq);
 2634         switch (cm_node->state) {
 2635         case IRDMA_CM_STATE_SYN_SENT:
 2636                 irdma_cleanup_retrans_entry(cm_node);
 2637                 /* active open */
 2638                 if (irdma_check_syn(cm_node, tcph)) {
 2639                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 2640                                     IRDMA_DEBUG_CM, "check syn fail\n");
 2641                         return;
 2642                 }
 2643                 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->th_ack);
 2644                 /* setup options */
 2645                 err = irdma_handle_tcp_options(cm_node, tcph, optionsize, 0);
 2646                 if (err) {
 2647                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 2648                                     IRDMA_DEBUG_CM,
 2649                                     "cm_node=%p tcp_options failed\n",
 2650                                     cm_node);
 2651                         break;
 2652                 }
 2653                 irdma_cleanup_retrans_entry(cm_node);
 2654                 cm_node->tcp_cntxt.rcv_nxt = inc_sequence + 1;
 2655                 irdma_send_ack(cm_node);        /* ACK  for the syn_ack */
 2656                 err = irdma_send_mpa_request(cm_node);
 2657                 if (err) {
 2658                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 2659                                     IRDMA_DEBUG_CM,
 2660                                     "cm_node=%p irdma_send_mpa_request failed\n",
 2661                                     cm_node);
 2662                         break;
 2663                 }
 2664                 cm_node->state = IRDMA_CM_STATE_MPAREQ_SENT;
 2665                 break;
 2666         case IRDMA_CM_STATE_MPAREQ_RCVD:
 2667                 irdma_passive_open_err(cm_node, true);
 2668                 break;
 2669         case IRDMA_CM_STATE_LISTENING:
 2670                 cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->th_ack);
 2671                 irdma_cleanup_retrans_entry(cm_node);
 2672                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2673                 irdma_send_reset(cm_node);
 2674                 break;
 2675         case IRDMA_CM_STATE_CLOSED:
 2676                 cm_node->tcp_cntxt.loc_seq_num = ntohl(tcph->th_ack);
 2677                 irdma_cleanup_retrans_entry(cm_node);
 2678                 atomic_inc(&cm_node->refcnt);
 2679                 irdma_send_reset(cm_node);
 2680                 break;
 2681         case IRDMA_CM_STATE_ESTABLISHED:
 2682         case IRDMA_CM_STATE_FIN_WAIT1:
 2683         case IRDMA_CM_STATE_FIN_WAIT2:
 2684         case IRDMA_CM_STATE_LAST_ACK:
 2685         case IRDMA_CM_STATE_OFFLOADED:
 2686         case IRDMA_CM_STATE_CLOSING:
 2687         case IRDMA_CM_STATE_UNKNOWN:
 2688         case IRDMA_CM_STATE_MPAREQ_SENT:
 2689         default:
 2690                 break;
 2691         }
 2692 }
 2693 
 2694 /**
 2695  * irdma_handle_ack_pkt - process packet with ACK
 2696  * @cm_node: connection's node
 2697  * @rbuf: receive buffer
 2698  */
 2699 static int
 2700 irdma_handle_ack_pkt(struct irdma_cm_node *cm_node,
 2701                      struct irdma_puda_buf *rbuf)
 2702 {
 2703         struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
 2704         u32 inc_sequence;
 2705         int ret;
 2706         int optionsize;
 2707         u32 datasize = rbuf->datalen;
 2708 
 2709         optionsize = (tcph->th_off << 2) - sizeof(struct tcphdr);
 2710 
 2711         if (irdma_check_seq(cm_node, tcph))
 2712                 return -EINVAL;
 2713 
 2714         inc_sequence = ntohl(tcph->th_seq);
 2715         switch (cm_node->state) {
 2716         case IRDMA_CM_STATE_SYN_RCVD:
 2717                 irdma_cleanup_retrans_entry(cm_node);
 2718                 ret = irdma_handle_tcp_options(cm_node, tcph, optionsize, 1);
 2719                 if (ret)
 2720                         return ret;
 2721                 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->th_ack);
 2722                 cm_node->state = IRDMA_CM_STATE_ESTABLISHED;
 2723                 if (datasize) {
 2724                         cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
 2725                         irdma_handle_rcv_mpa(cm_node, rbuf);
 2726                 }
 2727                 break;
 2728         case IRDMA_CM_STATE_ESTABLISHED:
 2729                 irdma_cleanup_retrans_entry(cm_node);
 2730                 if (datasize) {
 2731                         cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
 2732                         irdma_handle_rcv_mpa(cm_node, rbuf);
 2733                 }
 2734                 break;
 2735         case IRDMA_CM_STATE_MPAREQ_SENT:
 2736                 cm_node->tcp_cntxt.rem_ack_num = ntohl(tcph->th_ack);
 2737                 if (datasize) {
 2738                         cm_node->tcp_cntxt.rcv_nxt = inc_sequence + datasize;
 2739                         cm_node->ack_rcvd = false;
 2740                         irdma_handle_rcv_mpa(cm_node, rbuf);
 2741                 } else {
 2742                         cm_node->ack_rcvd = true;
 2743                 }
 2744                 break;
 2745         case IRDMA_CM_STATE_LISTENING:
 2746                 irdma_cleanup_retrans_entry(cm_node);
 2747                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2748                 irdma_send_reset(cm_node);
 2749                 break;
 2750         case IRDMA_CM_STATE_CLOSED:
 2751                 irdma_cleanup_retrans_entry(cm_node);
 2752                 atomic_inc(&cm_node->refcnt);
 2753                 irdma_send_reset(cm_node);
 2754                 break;
 2755         case IRDMA_CM_STATE_LAST_ACK:
 2756         case IRDMA_CM_STATE_CLOSING:
 2757                 irdma_cleanup_retrans_entry(cm_node);
 2758                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2759                 irdma_rem_ref_cm_node(cm_node);
 2760                 break;
 2761         case IRDMA_CM_STATE_FIN_WAIT1:
 2762                 irdma_cleanup_retrans_entry(cm_node);
 2763                 cm_node->state = IRDMA_CM_STATE_FIN_WAIT2;
 2764                 break;
 2765         case IRDMA_CM_STATE_SYN_SENT:
 2766         case IRDMA_CM_STATE_FIN_WAIT2:
 2767         case IRDMA_CM_STATE_OFFLOADED:
 2768         case IRDMA_CM_STATE_MPAREQ_RCVD:
 2769         case IRDMA_CM_STATE_UNKNOWN:
 2770         default:
 2771                 irdma_cleanup_retrans_entry(cm_node);
 2772                 break;
 2773         }
 2774 
 2775         return 0;
 2776 }
 2777 
 2778 /**
 2779  * irdma_process_pkt - process cm packet
 2780  * @cm_node: connection's node
 2781  * @rbuf: receive buffer
 2782  */
 2783 static void
 2784 irdma_process_pkt(struct irdma_cm_node *cm_node,
 2785                   struct irdma_puda_buf *rbuf)
 2786 {
 2787         enum irdma_tcpip_pkt_type pkt_type = IRDMA_PKT_TYPE_UNKNOWN;
 2788         struct tcphdr *tcph = (struct tcphdr *)rbuf->tcph;
 2789         u32 fin_set = 0;
 2790         int err;
 2791 
 2792         if (tcph->th_flags & TH_RST) {
 2793                 pkt_type = IRDMA_PKT_TYPE_RST;
 2794         } else if (tcph->th_flags & TH_SYN) {
 2795                 pkt_type = IRDMA_PKT_TYPE_SYN;
 2796                 if (tcph->th_flags & TH_ACK)
 2797                         pkt_type = IRDMA_PKT_TYPE_SYNACK;
 2798         } else if (tcph->th_flags & TH_ACK) {
 2799                 pkt_type = IRDMA_PKT_TYPE_ACK;
 2800         }
 2801         if (tcph->th_flags & TH_FIN)
 2802                 fin_set = 1;
 2803 
 2804         switch (pkt_type) {
 2805         case IRDMA_PKT_TYPE_SYN:
 2806                 irdma_handle_syn_pkt(cm_node, rbuf);
 2807                 break;
 2808         case IRDMA_PKT_TYPE_SYNACK:
 2809                 irdma_handle_synack_pkt(cm_node, rbuf);
 2810                 break;
 2811         case IRDMA_PKT_TYPE_ACK:
 2812                 err = irdma_handle_ack_pkt(cm_node, rbuf);
 2813                 if (fin_set && !err)
 2814                         irdma_handle_fin_pkt(cm_node);
 2815                 break;
 2816         case IRDMA_PKT_TYPE_RST:
 2817                 irdma_handle_rst_pkt(cm_node, rbuf);
 2818                 break;
 2819         default:
 2820                 if (fin_set &&
 2821                     (!irdma_check_seq(cm_node, (struct tcphdr *)rbuf->tcph)))
 2822                         irdma_handle_fin_pkt(cm_node);
 2823                 break;
 2824         }
 2825 }
 2826 
 2827 /**
 2828  * irdma_make_listen_node - create a listen node with params
 2829  * @cm_core: cm's core
 2830  * @iwdev: iwarp device structure
 2831  * @cm_info: quad info for connection
 2832  */
 2833 static struct irdma_cm_listener *
 2834 irdma_make_listen_node(struct irdma_cm_core *cm_core,
 2835                        struct irdma_device *iwdev,
 2836                        struct irdma_cm_info *cm_info)
 2837 {
 2838         struct irdma_cm_listener *listener;
 2839         unsigned long flags;
 2840 
 2841         /* cannot have multiple matching listeners */
 2842         listener = irdma_find_listener(cm_core, cm_info->loc_addr,
 2843                                        cm_info->loc_port, cm_info->vlan_id,
 2844                                        IRDMA_CM_LISTENER_EITHER_STATE);
 2845         if (listener &&
 2846             listener->listener_state == IRDMA_CM_LISTENER_ACTIVE_STATE) {
 2847                 atomic_dec(&listener->refcnt);
 2848                 return NULL;
 2849         }
 2850 
 2851         if (!listener) {
 2852                 /*
 2853                  * create a CM listen node 1/2 node to compare incoming traffic to
 2854                  */
 2855                 listener = kzalloc(sizeof(*listener), GFP_KERNEL);
 2856                 if (!listener)
 2857                         return NULL;
 2858                 cm_core->stats_listen_nodes_created++;
 2859                 memcpy(listener->loc_addr, cm_info->loc_addr,
 2860                        sizeof(listener->loc_addr));
 2861                 listener->loc_port = cm_info->loc_port;
 2862 
 2863                 INIT_LIST_HEAD(&listener->child_listen_list);
 2864 
 2865                 atomic_set(&listener->refcnt, 1);
 2866         } else {
 2867                 listener->reused_node = 1;
 2868         }
 2869 
 2870         listener->cm_id = cm_info->cm_id;
 2871         listener->ipv4 = cm_info->ipv4;
 2872         listener->vlan_id = cm_info->vlan_id;
 2873         atomic_set(&listener->pend_accepts_cnt, 0);
 2874         listener->cm_core = cm_core;
 2875         listener->iwdev = iwdev;
 2876 
 2877         listener->backlog = cm_info->backlog;
 2878         listener->listener_state = IRDMA_CM_LISTENER_ACTIVE_STATE;
 2879 
 2880         if (!listener->reused_node) {
 2881                 spin_lock_irqsave(&cm_core->listen_list_lock, flags);
 2882                 list_add(&listener->list, &cm_core->listen_list);
 2883                 spin_unlock_irqrestore(&cm_core->listen_list_lock, flags);
 2884         }
 2885 
 2886         return listener;
 2887 }
 2888 
 2889 /**
 2890  * irdma_create_cm_node - make a connection node with params
 2891  * @cm_core: cm's core
 2892  * @iwdev: iwarp device structure
 2893  * @conn_param: connection parameters
 2894  * @cm_info: quad info for connection
 2895  * @caller_cm_node: pointer to cm_node structure to return
 2896  */
 2897 static int
 2898 irdma_create_cm_node(struct irdma_cm_core *cm_core,
 2899                      struct irdma_device *iwdev,
 2900                      struct iw_cm_conn_param *conn_param,
 2901                      struct irdma_cm_info *cm_info,
 2902                      struct irdma_cm_node **caller_cm_node)
 2903 {
 2904         struct irdma_cm_node *cm_node;
 2905         u16 private_data_len = conn_param->private_data_len;
 2906         const void *private_data = conn_param->private_data;
 2907 
 2908         /* create a CM connection node */
 2909         cm_node = irdma_make_cm_node(cm_core, iwdev, cm_info, NULL);
 2910         if (!cm_node)
 2911                 return -ENOMEM;
 2912 
 2913         /* set our node side to client (active) side */
 2914         cm_node->tcp_cntxt.client = 1;
 2915         cm_node->tcp_cntxt.rcv_wscale = IRDMA_CM_DEFAULT_RCV_WND_SCALE;
 2916 
 2917         irdma_record_ird_ord(cm_node, conn_param->ird, conn_param->ord);
 2918 
 2919         cm_node->pdata.size = private_data_len;
 2920         cm_node->pdata.addr = cm_node->pdata_buf;
 2921 
 2922         memcpy(cm_node->pdata_buf, private_data, private_data_len);
 2923         *caller_cm_node = cm_node;
 2924 
 2925         return 0;
 2926 }
 2927 
 2928 /**
 2929  * irdma_cm_reject - reject and teardown a connection
 2930  * @cm_node: connection's node
 2931  * @pdata: ptr to private data for reject
 2932  * @plen: size of private data
 2933  */
 2934 static int
 2935 irdma_cm_reject(struct irdma_cm_node *cm_node, const void *pdata,
 2936                 u8 plen)
 2937 {
 2938         int ret;
 2939         int passive_state;
 2940 
 2941         if (cm_node->tcp_cntxt.client)
 2942                 return 0;
 2943 
 2944         irdma_cleanup_retrans_entry(cm_node);
 2945 
 2946         passive_state = atomic_add_return(1, &cm_node->passive_state);
 2947         if (passive_state == IRDMA_SEND_RESET_EVENT) {
 2948                 cm_node->state = IRDMA_CM_STATE_CLOSED;
 2949                 irdma_rem_ref_cm_node(cm_node);
 2950                 return 0;
 2951         }
 2952 
 2953         if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
 2954                 irdma_rem_ref_cm_node(cm_node);
 2955                 return 0;
 2956         }
 2957 
 2958         ret = irdma_send_mpa_reject(cm_node, pdata, plen);
 2959         if (!ret)
 2960                 return 0;
 2961 
 2962         cm_node->state = IRDMA_CM_STATE_CLOSED;
 2963         if (irdma_send_reset(cm_node))
 2964                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 2965                             "send reset failed\n");
 2966 
 2967         return ret;
 2968 }
 2969 
 2970 /**
 2971  * irdma_cm_close - close of cm connection
 2972  * @cm_node: connection's node
 2973  */
 2974 static int
 2975 irdma_cm_close(struct irdma_cm_node *cm_node)
 2976 {
 2977         switch (cm_node->state) {
 2978         case IRDMA_CM_STATE_SYN_RCVD:
 2979         case IRDMA_CM_STATE_SYN_SENT:
 2980         case IRDMA_CM_STATE_ONE_SIDE_ESTABLISHED:
 2981         case IRDMA_CM_STATE_ESTABLISHED:
 2982         case IRDMA_CM_STATE_ACCEPTING:
 2983         case IRDMA_CM_STATE_MPAREQ_SENT:
 2984         case IRDMA_CM_STATE_MPAREQ_RCVD:
 2985                 irdma_cleanup_retrans_entry(cm_node);
 2986                 irdma_send_reset(cm_node);
 2987                 break;
 2988         case IRDMA_CM_STATE_CLOSE_WAIT:
 2989                 cm_node->state = IRDMA_CM_STATE_LAST_ACK;
 2990                 irdma_send_fin(cm_node);
 2991                 break;
 2992         case IRDMA_CM_STATE_FIN_WAIT1:
 2993         case IRDMA_CM_STATE_FIN_WAIT2:
 2994         case IRDMA_CM_STATE_LAST_ACK:
 2995         case IRDMA_CM_STATE_TIME_WAIT:
 2996         case IRDMA_CM_STATE_CLOSING:
 2997                 return -EINVAL;
 2998         case IRDMA_CM_STATE_LISTENING:
 2999                 irdma_cleanup_retrans_entry(cm_node);
 3000                 irdma_send_reset(cm_node);
 3001                 break;
 3002         case IRDMA_CM_STATE_MPAREJ_RCVD:
 3003         case IRDMA_CM_STATE_UNKNOWN:
 3004         case IRDMA_CM_STATE_INITED:
 3005         case IRDMA_CM_STATE_CLOSED:
 3006         case IRDMA_CM_STATE_LISTENER_DESTROYED:
 3007                 irdma_rem_ref_cm_node(cm_node);
 3008                 break;
 3009         case IRDMA_CM_STATE_OFFLOADED:
 3010                 if (cm_node->send_entry)
 3011                         irdma_debug(iwdev_to_idev(cm_node->iwdev),
 3012                                     IRDMA_DEBUG_CM,
 3013                                     "CM send_entry in OFFLOADED state\n");
 3014                 irdma_rem_ref_cm_node(cm_node);
 3015                 break;
 3016         }
 3017 
 3018         return 0;
 3019 }
 3020 
 3021 /**
 3022  * irdma_receive_ilq - recv an ETHERNET packet, and process it
 3023  * through CM
 3024  * @vsi: VSI structure of dev
 3025  * @rbuf: receive buffer
 3026  */
 3027 void
 3028 irdma_receive_ilq(struct irdma_sc_vsi *vsi, struct irdma_puda_buf *rbuf)
 3029 {
 3030         struct irdma_cm_node *cm_node;
 3031         struct irdma_cm_listener *listener;
 3032         struct ip *iph;
 3033         struct ip6_hdr *ip6h;
 3034         struct tcphdr *tcph;
 3035         struct irdma_cm_info cm_info = {0};
 3036         struct irdma_device *iwdev = vsi->back_vsi;
 3037         struct irdma_cm_core *cm_core = &iwdev->cm_core;
 3038         struct ether_vlan_header *ethh;
 3039         u16 vtag;
 3040 
 3041         /* if vlan, then maclen = 18 else 14 */
 3042         iph = (struct ip *)rbuf->iph;
 3043         irdma_debug_buf(vsi->dev, IRDMA_DEBUG_ILQ, "RECEIVE ILQ BUFFER",
 3044                         rbuf->mem.va, rbuf->totallen);
 3045         if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev >= IRDMA_GEN_2) {
 3046                 if (rbuf->vlan_valid) {
 3047                         vtag = rbuf->vlan_id;
 3048                         cm_info.user_pri = (vtag & EVL_PRI_MASK) >>
 3049                             VLAN_PRIO_SHIFT;
 3050                         cm_info.vlan_id = vtag & EVL_VLID_MASK;
 3051                 } else {
 3052                         cm_info.vlan_id = 0xFFFF;
 3053                 }
 3054         } else {
 3055                 ethh = rbuf->mem.va;
 3056 
 3057                 if (ethh->evl_proto == htons(ETH_P_8021Q)) {
 3058                         vtag = ntohs(ethh->evl_tag);
 3059                         cm_info.user_pri = (vtag & EVL_PRI_MASK) >>
 3060                             VLAN_PRIO_SHIFT;
 3061                         cm_info.vlan_id = vtag & EVL_VLID_MASK;
 3062                         irdma_debug(iwdev_to_idev(cm_core->iwdev),
 3063                                     IRDMA_DEBUG_CM, "vlan_id=%d\n",
 3064                                     cm_info.vlan_id);
 3065                 } else {
 3066                         cm_info.vlan_id = 0xFFFF;
 3067                 }
 3068         }
 3069         tcph = (struct tcphdr *)rbuf->tcph;
 3070 
 3071         if (rbuf->ipv4) {
 3072                 cm_info.loc_addr[0] = ntohl(iph->ip_dst.s_addr);
 3073                 cm_info.rem_addr[0] = ntohl(iph->ip_src.s_addr);
 3074                 cm_info.ipv4 = true;
 3075                 cm_info.tos = iph->ip_tos;
 3076         } else {
 3077                 ip6h = (struct ip6_hdr *)rbuf->iph;
 3078                 irdma_copy_ip_ntohl(cm_info.loc_addr,
 3079                                     ip6h->ip6_dst.__u6_addr.__u6_addr32);
 3080                 irdma_copy_ip_ntohl(cm_info.rem_addr,
 3081                                     ip6h->ip6_src.__u6_addr.__u6_addr32);
 3082                 cm_info.ipv4 = false;
 3083                 cm_info.tos = (ip6h->ip6_vfc << 4) | ip6h->ip6_flow;
 3084         }
 3085         cm_info.loc_port = ntohs(tcph->th_dport);
 3086         cm_info.rem_port = ntohs(tcph->th_sport);
 3087         cm_node = irdma_find_node(cm_core, cm_info.rem_port, cm_info.rem_addr,
 3088                                   cm_info.loc_port, cm_info.loc_addr, cm_info.vlan_id);
 3089 
 3090         if (!cm_node) {
 3091                 /*
 3092                  * Only type of packet accepted are for the PASSIVE open (syn only)
 3093                  */
 3094                 if (!(tcph->th_flags & TH_SYN) || tcph->th_flags & TH_ACK)
 3095                         return;
 3096 
 3097                 listener = irdma_find_listener(cm_core,
 3098                                                cm_info.loc_addr,
 3099                                                cm_info.loc_port,
 3100                                                cm_info.vlan_id,
 3101                                                IRDMA_CM_LISTENER_ACTIVE_STATE);
 3102                 if (!listener) {
 3103                         cm_info.cm_id = NULL;
 3104                         irdma_debug(iwdev_to_idev(cm_core->iwdev),
 3105                                     IRDMA_DEBUG_CM, "no listener found\n");
 3106                         return;
 3107                 }
 3108 
 3109                 cm_info.cm_id = listener->cm_id;
 3110                 cm_node = irdma_make_cm_node(cm_core, iwdev, &cm_info,
 3111                                              listener);
 3112                 if (!cm_node) {
 3113                         irdma_debug(iwdev_to_idev(cm_core->iwdev),
 3114                                     IRDMA_DEBUG_CM, "allocate node failed\n");
 3115                         atomic_dec(&listener->refcnt);
 3116                         return;
 3117                 }
 3118 
 3119                 if (!(tcph->th_flags & (TH_RST | TH_FIN))) {
 3120                         cm_node->state = IRDMA_CM_STATE_LISTENING;
 3121                 } else {
 3122                         irdma_rem_ref_cm_node(cm_node);
 3123                         return;
 3124                 }
 3125 
 3126                 atomic_inc(&cm_node->refcnt);
 3127         } else if (cm_node->state == IRDMA_CM_STATE_OFFLOADED) {
 3128                 irdma_rem_ref_cm_node(cm_node);
 3129                 return;
 3130         }
 3131 
 3132         irdma_process_pkt(cm_node, rbuf);
 3133         irdma_rem_ref_cm_node(cm_node);
 3134 }
 3135 
 3136 static int
 3137 irdma_add_qh(struct irdma_cm_node *cm_node, bool active)
 3138 {
 3139         if (!active)
 3140                 irdma_add_conn_est_qh(cm_node);
 3141         return 0;
 3142 }
 3143 
 3144 static void
 3145 irdma_cm_free_ah_nop(struct irdma_cm_node *cm_node)
 3146 {
 3147 }
 3148 
 3149 /**
 3150  * irdma_setup_cm_core - setup top level instance of a cm core
 3151  * @iwdev: iwarp device structure
 3152  * @rdma_ver: HW version
 3153  */
 3154 int
 3155 irdma_setup_cm_core(struct irdma_device *iwdev, u8 rdma_ver)
 3156 {
 3157         struct irdma_cm_core *cm_core = &iwdev->cm_core;
 3158 
 3159         cm_core->iwdev = iwdev;
 3160         cm_core->dev = &iwdev->rf->sc_dev;
 3161 
 3162         /* Handles CM event work items send to Iwarp core */
 3163         cm_core->event_wq = alloc_ordered_workqueue("iwarp-event-wq", 0);
 3164         if (!cm_core->event_wq)
 3165                 return -ENOMEM;
 3166 
 3167         INIT_LIST_HEAD(&cm_core->listen_list);
 3168 
 3169         timer_setup(&cm_core->tcp_timer, irdma_cm_timer_tick, 0);
 3170 
 3171         spin_lock_init(&cm_core->ht_lock);
 3172         spin_lock_init(&cm_core->listen_list_lock);
 3173         spin_lock_init(&cm_core->apbvt_lock);
 3174         switch (rdma_ver) {
 3175         case IRDMA_GEN_1:
 3176                 cm_core->form_cm_frame = irdma_form_uda_cm_frame;
 3177                 cm_core->cm_create_ah = irdma_add_qh;
 3178                 cm_core->cm_free_ah = irdma_cm_free_ah_nop;
 3179                 break;
 3180         case IRDMA_GEN_2:
 3181         default:
 3182                 cm_core->form_cm_frame = irdma_form_ah_cm_frame;
 3183                 cm_core->cm_create_ah = irdma_cm_create_ah;
 3184                 cm_core->cm_free_ah = irdma_cm_free_ah;
 3185         }
 3186 
 3187         return 0;
 3188 }
 3189 
 3190 /**
 3191  * irdma_cleanup_cm_core - deallocate a top level instance of a
 3192  * cm core
 3193  * @cm_core: cm's core
 3194  */
 3195 void
 3196 irdma_cleanup_cm_core(struct irdma_cm_core *cm_core)
 3197 {
 3198         unsigned long flags;
 3199 
 3200         if (!cm_core)
 3201                 return;
 3202 
 3203         spin_lock_irqsave(&cm_core->ht_lock, flags);
 3204         if (timer_pending(&cm_core->tcp_timer))
 3205                 del_timer_sync(&cm_core->tcp_timer);
 3206         spin_unlock_irqrestore(&cm_core->ht_lock, flags);
 3207 
 3208         destroy_workqueue(cm_core->event_wq);
 3209         cm_core->dev->ws_reset(&cm_core->iwdev->vsi);
 3210 }
 3211 
 3212 /**
 3213  * irdma_init_tcp_ctx - setup qp context
 3214  * @cm_node: connection's node
 3215  * @tcp_info: offload info for tcp
 3216  * @iwqp: associate qp for the connection
 3217  */
 3218 static void
 3219 irdma_init_tcp_ctx(struct irdma_cm_node *cm_node,
 3220                    struct irdma_tcp_offload_info *tcp_info,
 3221                    struct irdma_qp *iwqp)
 3222 {
 3223         tcp_info->ipv4 = cm_node->ipv4;
 3224         tcp_info->drop_ooo_seg = !iwqp->iwdev->iw_ooo;
 3225         tcp_info->wscale = true;
 3226         tcp_info->ignore_tcp_opt = true;
 3227         tcp_info->ignore_tcp_uns_opt = true;
 3228         tcp_info->no_nagle = false;
 3229 
 3230         tcp_info->ttl = IRDMA_DEFAULT_TTL;
 3231         tcp_info->rtt_var = IRDMA_DEFAULT_RTT_VAR;
 3232         tcp_info->ss_thresh = IRDMA_DEFAULT_SS_THRESH;
 3233         tcp_info->rexmit_thresh = IRDMA_DEFAULT_REXMIT_THRESH;
 3234 
 3235         tcp_info->tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
 3236         tcp_info->snd_wscale = cm_node->tcp_cntxt.snd_wscale;
 3237         tcp_info->rcv_wscale = cm_node->tcp_cntxt.rcv_wscale;
 3238 
 3239         tcp_info->snd_nxt = cm_node->tcp_cntxt.loc_seq_num;
 3240         tcp_info->snd_wnd = cm_node->tcp_cntxt.snd_wnd;
 3241         tcp_info->rcv_nxt = cm_node->tcp_cntxt.rcv_nxt;
 3242         tcp_info->snd_max = cm_node->tcp_cntxt.loc_seq_num;
 3243 
 3244         tcp_info->snd_una = cm_node->tcp_cntxt.loc_seq_num;
 3245         tcp_info->cwnd = 2 * cm_node->tcp_cntxt.mss;
 3246         tcp_info->snd_wl1 = cm_node->tcp_cntxt.rcv_nxt;
 3247         tcp_info->snd_wl2 = cm_node->tcp_cntxt.loc_seq_num;
 3248         tcp_info->max_snd_window = cm_node->tcp_cntxt.max_snd_wnd;
 3249         tcp_info->rcv_wnd = cm_node->tcp_cntxt.rcv_wnd
 3250             << cm_node->tcp_cntxt.rcv_wscale;
 3251 
 3252         tcp_info->flow_label = 0;
 3253         tcp_info->snd_mss = (u32)cm_node->tcp_cntxt.mss;
 3254         tcp_info->tos = cm_node->tos;
 3255         if (cm_node->vlan_id < VLAN_N_VID) {
 3256                 tcp_info->insert_vlan_tag = true;
 3257                 tcp_info->vlan_tag = cm_node->vlan_id;
 3258                 tcp_info->vlan_tag |= cm_node->user_pri << VLAN_PRIO_SHIFT;
 3259         }
 3260         tcp_info->src_port = cm_node->loc_port;
 3261         tcp_info->dst_port = cm_node->rem_port;
 3262         tcp_info->arp_idx = (u16)irdma_arp_table(iwqp->iwdev->rf,
 3263                                                   cm_node->rem_addr, NULL,
 3264                                                   IRDMA_ARP_RESOLVE);
 3265         if (cm_node->ipv4) {
 3266                 tcp_info->dest_ip_addr[3] = cm_node->rem_addr[0];
 3267                 tcp_info->local_ipaddr[3] = cm_node->loc_addr[0];
 3268         } else {
 3269                 memcpy(tcp_info->dest_ip_addr, cm_node->rem_addr,
 3270                        sizeof(tcp_info->dest_ip_addr));
 3271                 memcpy(tcp_info->local_ipaddr, cm_node->loc_addr,
 3272                        sizeof(tcp_info->local_ipaddr));
 3273         }
 3274 }
 3275 
 3276 /**
 3277  * irdma_cm_init_tsa_conn - setup qp for RTS
 3278  * @iwqp: associate qp for the connection
 3279  * @cm_node: connection's node
 3280  */
 3281 static void
 3282 irdma_cm_init_tsa_conn(struct irdma_qp *iwqp,
 3283                        struct irdma_cm_node *cm_node)
 3284 {
 3285         struct irdma_iwarp_offload_info *iwarp_info;
 3286         struct irdma_qp_host_ctx_info *ctx_info;
 3287 
 3288         iwarp_info = &iwqp->iwarp_info;
 3289         ctx_info = &iwqp->ctx_info;
 3290 
 3291         ctx_info->tcp_info = &iwqp->tcp_info;
 3292         ctx_info->send_cq_num = iwqp->iwscq->sc_cq.cq_uk.cq_id;
 3293         ctx_info->rcv_cq_num = iwqp->iwrcq->sc_cq.cq_uk.cq_id;
 3294 
 3295         iwarp_info->ord_size = cm_node->ord_size;
 3296         iwarp_info->ird_size = cm_node->ird_size;
 3297         iwarp_info->rd_en = true;
 3298         iwarp_info->rdmap_ver = 1;
 3299         iwarp_info->ddp_ver = 1;
 3300         iwarp_info->pd_id = iwqp->iwpd->sc_pd.pd_id;
 3301 
 3302         ctx_info->tcp_info_valid = true;
 3303         ctx_info->iwarp_info_valid = true;
 3304         ctx_info->user_pri = cm_node->user_pri;
 3305 
 3306         irdma_init_tcp_ctx(cm_node, &iwqp->tcp_info, iwqp);
 3307         if (cm_node->snd_mark_en) {
 3308                 iwarp_info->snd_mark_en = true;
 3309                 iwarp_info->snd_mark_offset = (iwqp->tcp_info.snd_nxt & SNDMARKER_SEQNMASK) +
 3310                     cm_node->lsmm_size;
 3311         }
 3312 
 3313         cm_node->state = IRDMA_CM_STATE_OFFLOADED;
 3314         iwqp->tcp_info.tcp_state = IRDMA_TCP_STATE_ESTABLISHED;
 3315         iwqp->tcp_info.src_mac_addr_idx = iwqp->iwdev->mac_ip_table_idx;
 3316 
 3317         if (cm_node->rcv_mark_en) {
 3318                 iwarp_info->rcv_mark_en = true;
 3319                 iwarp_info->align_hdrs = true;
 3320         }
 3321 
 3322         irdma_sc_qp_setctx(&iwqp->sc_qp, iwqp->host_ctx.va, ctx_info);
 3323 
 3324         /* once tcp_info is set, no need to do it again */
 3325         ctx_info->tcp_info_valid = false;
 3326         ctx_info->iwarp_info_valid = false;
 3327 }
 3328 
 3329 /**
 3330  * irdma_cm_disconn - when a connection is being closed
 3331  * @iwqp: associated qp for the connection
 3332  */
 3333 void
 3334 irdma_cm_disconn(struct irdma_qp *iwqp)
 3335 {
 3336         struct irdma_device *iwdev = iwqp->iwdev;
 3337         struct disconn_work *work;
 3338         unsigned long flags;
 3339 
 3340         work = kzalloc(sizeof(*work), GFP_ATOMIC);
 3341         if (!work)
 3342                 return;
 3343 
 3344         spin_lock_irqsave(&iwdev->rf->qptable_lock, flags);
 3345         if (!iwdev->rf->qp_table[iwqp->ibqp.qp_num]) {
 3346                 spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
 3347                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3348                             "qp_id %d is already freed\n", iwqp->ibqp.qp_num);
 3349                 kfree(work);
 3350                 return;
 3351         }
 3352         irdma_qp_add_ref(&iwqp->ibqp);
 3353         spin_unlock_irqrestore(&iwdev->rf->qptable_lock, flags);
 3354 
 3355         work->iwqp = iwqp;
 3356         INIT_WORK(&work->work, irdma_disconnect_worker);
 3357         queue_work(iwdev->cleanup_wq, &work->work);
 3358 }
 3359 
 3360 /**
 3361  * irdma_qp_disconnect - free qp and close cm
 3362  * @iwqp: associate qp for the connection
 3363  */
 3364 static void
 3365 irdma_qp_disconnect(struct irdma_qp *iwqp)
 3366 {
 3367         struct irdma_device *iwdev = iwqp->iwdev;
 3368 
 3369         iwqp->active_conn = 0;
 3370         /* close the CM node down if it is still active */
 3371         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3372                     "Call close API\n");
 3373         irdma_cm_close(iwqp->cm_node);
 3374 }
 3375 
 3376 /**
 3377  * irdma_cm_disconn_true - called by worker thread to disconnect qp
 3378  * @iwqp: associate qp for the connection
 3379  */
 3380 static void
 3381 irdma_cm_disconn_true(struct irdma_qp *iwqp)
 3382 {
 3383         struct iw_cm_id *cm_id;
 3384         struct irdma_device *iwdev;
 3385         struct irdma_sc_qp *qp = &iwqp->sc_qp;
 3386         u16 last_ae;
 3387         u8 original_hw_tcp_state;
 3388         u8 original_ibqp_state;
 3389         int disconn_status = 0;
 3390         int issue_disconn = 0;
 3391         int issue_close = 0;
 3392         int issue_flush = 0;
 3393         unsigned long flags;
 3394         int err;
 3395 
 3396         iwdev = iwqp->iwdev;
 3397         spin_lock_irqsave(&iwqp->lock, flags);
 3398         if (rdma_protocol_roce(&iwdev->ibdev, 1)) {
 3399                 struct ib_qp_attr attr;
 3400 
 3401                 if (iwqp->flush_issued || iwqp->sc_qp.qp_uk.destroy_pending) {
 3402                         spin_unlock_irqrestore(&iwqp->lock, flags);
 3403                         return;
 3404                 }
 3405 
 3406                 spin_unlock_irqrestore(&iwqp->lock, flags);
 3407 
 3408                 attr.qp_state = IB_QPS_ERR;
 3409                 irdma_modify_qp_roce(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
 3410                 irdma_ib_qp_event(iwqp, qp->event_type);
 3411                 return;
 3412         }
 3413 
 3414         cm_id = iwqp->cm_id;
 3415         original_hw_tcp_state = iwqp->hw_tcp_state;
 3416         original_ibqp_state = iwqp->ibqp_state;
 3417         last_ae = iwqp->last_aeq;
 3418 
 3419         if (qp->term_flags) {
 3420                 issue_disconn = 1;
 3421                 issue_close = 1;
 3422                 iwqp->cm_id = NULL;
 3423                 irdma_terminate_del_timer(qp);
 3424                 if (!iwqp->flush_issued) {
 3425                         iwqp->flush_issued = 1;
 3426                         issue_flush = 1;
 3427                 }
 3428         } else if ((original_hw_tcp_state == IRDMA_TCP_STATE_CLOSE_WAIT) ||
 3429                    ((original_ibqp_state == IB_QPS_RTS) &&
 3430                     (last_ae == IRDMA_AE_LLP_CONNECTION_RESET))) {
 3431                 issue_disconn = 1;
 3432                 if (last_ae == IRDMA_AE_LLP_CONNECTION_RESET)
 3433                         disconn_status = -ECONNRESET;
 3434         }
 3435 
 3436         if (original_hw_tcp_state == IRDMA_TCP_STATE_CLOSED ||
 3437             original_hw_tcp_state == IRDMA_TCP_STATE_TIME_WAIT ||
 3438             last_ae == IRDMA_AE_RDMAP_ROE_BAD_LLP_CLOSE ||
 3439             last_ae == IRDMA_AE_BAD_CLOSE ||
 3440             last_ae == IRDMA_AE_LLP_CONNECTION_RESET || iwdev->rf->reset || !cm_id) {
 3441                 issue_close = 1;
 3442                 iwqp->cm_id = NULL;
 3443                 qp->term_flags = 0;
 3444                 if (!iwqp->flush_issued) {
 3445                         iwqp->flush_issued = 1;
 3446                         issue_flush = 1;
 3447                 }
 3448         }
 3449 
 3450         spin_unlock_irqrestore(&iwqp->lock, flags);
 3451         if (issue_flush && !iwqp->sc_qp.qp_uk.destroy_pending) {
 3452                 irdma_flush_wqes(iwqp, IRDMA_FLUSH_SQ | IRDMA_FLUSH_RQ |
 3453                                  IRDMA_FLUSH_WAIT);
 3454 
 3455                 if (qp->term_flags)
 3456                         irdma_ib_qp_event(iwqp, qp->event_type);
 3457         }
 3458 
 3459         if (!cm_id || !cm_id->event_handler)
 3460                 return;
 3461 
 3462         spin_lock_irqsave(&iwdev->cm_core.ht_lock, flags);
 3463         if (!iwqp->cm_node) {
 3464                 spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
 3465                 return;
 3466         }
 3467         atomic_inc(&iwqp->cm_node->refcnt);
 3468 
 3469         spin_unlock_irqrestore(&iwdev->cm_core.ht_lock, flags);
 3470 
 3471         if (issue_disconn) {
 3472                 err = irdma_send_cm_event(iwqp->cm_node, cm_id,
 3473                                           IW_CM_EVENT_DISCONNECT,
 3474                                           disconn_status);
 3475                 if (err)
 3476                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3477                                     "disconnect event failed: - cm_id = %p\n",
 3478                                     cm_id);
 3479         }
 3480         if (issue_close) {
 3481                 cm_id->provider_data = iwqp;
 3482                 err = irdma_send_cm_event(iwqp->cm_node, cm_id,
 3483                                           IW_CM_EVENT_CLOSE, 0);
 3484                 if (err)
 3485                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3486                                     "close event failed: - cm_id = %p\n",
 3487                                     cm_id);
 3488                 irdma_qp_disconnect(iwqp);
 3489         }
 3490         irdma_rem_ref_cm_node(iwqp->cm_node);
 3491 }
 3492 
 3493 /**
 3494  * irdma_disconnect_worker - worker for connection close
 3495  * @work: points or disconn structure
 3496  */
 3497 static void
 3498 irdma_disconnect_worker(struct work_struct *work)
 3499 {
 3500         struct disconn_work *dwork = container_of(work, struct disconn_work, work);
 3501         struct irdma_qp *iwqp = dwork->iwqp;
 3502 
 3503         kfree(dwork);
 3504         irdma_cm_disconn_true(iwqp);
 3505         irdma_qp_rem_ref(&iwqp->ibqp);
 3506 }
 3507 
 3508 /**
 3509  * irdma_free_lsmm_rsrc - free lsmm memory and deregister
 3510  * @iwqp: associate qp for the connection
 3511  */
 3512 void
 3513 irdma_free_lsmm_rsrc(struct irdma_qp *iwqp)
 3514 {
 3515         struct irdma_device *iwdev;
 3516 
 3517         iwdev = iwqp->iwdev;
 3518 
 3519         if (iwqp->ietf_mem.va) {
 3520                 if (iwqp->lsmm_mr)
 3521                         kc_free_lsmm_dereg_mr(iwdev, iwqp);
 3522                 irdma_free_dma_mem(iwdev->rf->sc_dev.hw,
 3523                                    &iwqp->ietf_mem);
 3524                 iwqp->ietf_mem.va = NULL;
 3525         }
 3526 }
 3527 
 3528 /**
 3529  * irdma_accept - registered call for connection to be accepted
 3530  * @cm_id: cm information for passive connection
 3531  * @conn_param: accpet parameters
 3532  */
 3533 int
 3534 irdma_accept(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 3535 {
 3536         struct ib_qp *ibqp;
 3537         struct irdma_qp *iwqp;
 3538         struct irdma_device *iwdev;
 3539         struct irdma_sc_dev *dev;
 3540         struct irdma_cm_node *cm_node;
 3541         struct ib_qp_attr attr = {0};
 3542         int passive_state;
 3543         struct ib_mr *ibmr;
 3544         struct irdma_pd *iwpd;
 3545         u16 buf_len = 0;
 3546         struct irdma_kmem_info accept;
 3547         u64 tagged_offset;
 3548         int wait_ret;
 3549         int ret = 0;
 3550 
 3551         ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
 3552         if (!ibqp)
 3553                 return -EINVAL;
 3554 
 3555         iwqp = to_iwqp(ibqp);
 3556         iwdev = iwqp->iwdev;
 3557         dev = &iwdev->rf->sc_dev;
 3558         cm_node = cm_id->provider_data;
 3559 
 3560         if (((struct sockaddr_in *)&cm_id->local_addr)->sin_family == AF_INET) {
 3561                 cm_node->ipv4 = true;
 3562                 cm_node->vlan_id = irdma_get_vlan_ipv4(cm_node->loc_addr);
 3563         } else {
 3564                 cm_node->ipv4 = false;
 3565                 irdma_netdev_vlan_ipv6(cm_node->loc_addr, &cm_node->vlan_id,
 3566                                        NULL);
 3567         }
 3568         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3569                     "Accept vlan_id=%d\n", cm_node->vlan_id);
 3570 
 3571         if (cm_node->state == IRDMA_CM_STATE_LISTENER_DESTROYED) {
 3572                 ret = -EINVAL;
 3573                 goto error;
 3574         }
 3575 
 3576         passive_state = atomic_add_return(1, &cm_node->passive_state);
 3577         if (passive_state == IRDMA_SEND_RESET_EVENT) {
 3578                 ret = -ECONNRESET;
 3579                 goto error;
 3580         }
 3581 
 3582         buf_len = conn_param->private_data_len + IRDMA_MAX_IETF_SIZE;
 3583         iwqp->ietf_mem.size = buf_len;
 3584         iwqp->ietf_mem.va = irdma_allocate_dma_mem(dev->hw, &iwqp->ietf_mem,
 3585                                                    iwqp->ietf_mem.size, 1);
 3586         if (!iwqp->ietf_mem.va) {
 3587                 ret = -ENOMEM;
 3588                 goto error;
 3589         }
 3590 
 3591         cm_node->pdata.size = conn_param->private_data_len;
 3592         accept.addr = iwqp->ietf_mem.va;
 3593         accept.size = irdma_cm_build_mpa_frame(cm_node, &accept, MPA_KEY_REPLY);
 3594         memcpy((u8 *)accept.addr + accept.size, conn_param->private_data,
 3595                conn_param->private_data_len);
 3596 
 3597         if (cm_node->dev->ws_add(iwqp->sc_qp.vsi, cm_node->user_pri)) {
 3598                 ret = -ENOMEM;
 3599                 goto error;
 3600         }
 3601         iwqp->sc_qp.user_pri = cm_node->user_pri;
 3602         irdma_qp_add_qos(&iwqp->sc_qp);
 3603         if (cm_node->dev->hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_2)
 3604                 iwdev->rf->check_fc(&iwdev->vsi, &iwqp->sc_qp);
 3605         /* setup our first outgoing iWarp send WQE (the IETF frame response) */
 3606         iwpd = iwqp->iwpd;
 3607         tagged_offset = (uintptr_t)iwqp->ietf_mem.va;
 3608         ibmr = irdma_reg_phys_mr(&iwpd->ibpd, iwqp->ietf_mem.pa, buf_len,
 3609                                  IB_ACCESS_LOCAL_WRITE, &tagged_offset);
 3610         if (IS_ERR(ibmr)) {
 3611                 ret = -ENOMEM;
 3612                 goto error;
 3613         }
 3614 
 3615         ibmr->pd = &iwpd->ibpd;
 3616         ibmr->device = iwpd->ibpd.device;
 3617         iwqp->lsmm_mr = ibmr;
 3618         if (iwqp->page)
 3619                 iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
 3620 
 3621         cm_node->lsmm_size = accept.size + conn_param->private_data_len;
 3622         irdma_sc_send_lsmm(&iwqp->sc_qp, iwqp->ietf_mem.va, cm_node->lsmm_size,
 3623                            ibmr->lkey);
 3624 
 3625         if (iwqp->page)
 3626                 kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
 3627 
 3628         iwqp->cm_id = cm_id;
 3629         cm_node->cm_id = cm_id;
 3630 
 3631         cm_id->provider_data = iwqp;
 3632         iwqp->active_conn = 0;
 3633         iwqp->cm_node = cm_node;
 3634         cm_node->iwqp = iwqp;
 3635         irdma_cm_init_tsa_conn(iwqp, cm_node);
 3636         irdma_qp_add_ref(&iwqp->ibqp);
 3637         cm_id->add_ref(cm_id);
 3638 
 3639         attr.qp_state = IB_QPS_RTS;
 3640         cm_node->qhash_set = false;
 3641         cm_node->cm_core->cm_free_ah(cm_node);
 3642 
 3643         irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
 3644         if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
 3645                 wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
 3646                                                             iwqp->rts_ae_rcvd,
 3647                                                             IRDMA_MAX_TIMEOUT);
 3648                 if (!wait_ret) {
 3649                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3650                                     "Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
 3651                                     cm_node, cm_node->loc_port,
 3652                                     cm_node->rem_port, cm_node->cm_id);
 3653                         ret = -ECONNRESET;
 3654                         goto error;
 3655                 }
 3656         }
 3657 
 3658         irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_ESTABLISHED, 0);
 3659         cm_node->accelerated = true;
 3660         complete(&cm_node->establish_comp);
 3661 
 3662         if (cm_node->accept_pend) {
 3663                 atomic_dec(&cm_node->listener->pend_accepts_cnt);
 3664                 cm_node->accept_pend = 0;
 3665         }
 3666 
 3667         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3668                     "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",
 3669                     cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
 3670                     cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
 3671         cm_node->cm_core->stats_accepts++;
 3672 
 3673         return 0;
 3674 error:
 3675         irdma_free_lsmm_rsrc(iwqp);
 3676         irdma_rem_ref_cm_node(cm_node);
 3677 
 3678         return ret;
 3679 }
 3680 
 3681 /**
 3682  * irdma_reject - registered call for connection to be rejected
 3683  * @cm_id: cm information for passive connection
 3684  * @pdata: private data to be sent
 3685  * @pdata_len: private data length
 3686  */
 3687 int
 3688 irdma_reject(struct iw_cm_id *cm_id, const void *pdata, u8 pdata_len)
 3689 {
 3690         struct irdma_device *iwdev;
 3691         struct irdma_cm_node *cm_node;
 3692 
 3693         cm_node = cm_id->provider_data;
 3694         cm_node->pdata.size = pdata_len;
 3695 
 3696         iwdev = to_iwdev(cm_id->device);
 3697         if (!iwdev)
 3698                 return -EINVAL;
 3699 
 3700         cm_node->cm_core->stats_rejects++;
 3701 
 3702         if (pdata_len + sizeof(struct ietf_mpa_v2) > IRDMA_MAX_CM_BUF)
 3703                 return -EINVAL;
 3704 
 3705         return irdma_cm_reject(cm_node, pdata, pdata_len);
 3706 }
 3707 
 3708 /**
 3709  * irdma_connect - registered call for connection to be established
 3710  * @cm_id: cm information for passive connection
 3711  * @conn_param: Information about the connection
 3712  */
 3713 int
 3714 irdma_connect(struct iw_cm_id *cm_id, struct iw_cm_conn_param *conn_param)
 3715 {
 3716         struct ib_qp *ibqp;
 3717         struct irdma_qp *iwqp;
 3718         struct irdma_device *iwdev;
 3719         struct irdma_cm_node *cm_node;
 3720         struct irdma_cm_info cm_info;
 3721         struct sockaddr_in *laddr;
 3722         struct sockaddr_in *raddr;
 3723         struct sockaddr_in6 *laddr6;
 3724         struct sockaddr_in6 *raddr6;
 3725         int ret = 0;
 3726 
 3727         ibqp = irdma_get_qp(cm_id->device, conn_param->qpn);
 3728         if (!ibqp)
 3729                 return -EINVAL;
 3730         iwqp = to_iwqp(ibqp);
 3731         if (!iwqp)
 3732                 return -EINVAL;
 3733         iwdev = iwqp->iwdev;
 3734         if (!iwdev)
 3735                 return -EINVAL;
 3736 
 3737         laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
 3738         raddr = (struct sockaddr_in *)&cm_id->m_remote_addr;
 3739         laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
 3740         raddr6 = (struct sockaddr_in6 *)&cm_id->m_remote_addr;
 3741 
 3742         if (!(laddr->sin_port) || !(raddr->sin_port))
 3743                 return -EINVAL;
 3744 
 3745         iwqp->active_conn = 1;
 3746         iwqp->cm_id = NULL;
 3747         cm_id->provider_data = iwqp;
 3748 
 3749         /* set up the connection params for the node */
 3750         if (cm_id->remote_addr.ss_family == AF_INET) {
 3751                 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
 3752                         return -EINVAL;
 3753 
 3754                 cm_info.ipv4 = true;
 3755                 memset(cm_info.loc_addr, 0, sizeof(cm_info.loc_addr));
 3756                 memset(cm_info.rem_addr, 0, sizeof(cm_info.rem_addr));
 3757                 cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
 3758                 cm_info.rem_addr[0] = ntohl(raddr->sin_addr.s_addr);
 3759                 cm_info.loc_port = ntohs(laddr->sin_port);
 3760                 cm_info.rem_port = ntohs(raddr->sin_port);
 3761                 cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
 3762         } else {
 3763                 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
 3764                         return -EINVAL;
 3765 
 3766                 cm_info.ipv4 = false;
 3767                 irdma_copy_ip_ntohl(cm_info.loc_addr,
 3768                                     laddr6->sin6_addr.__u6_addr.__u6_addr32);
 3769                 irdma_copy_ip_ntohl(cm_info.rem_addr,
 3770                                     raddr6->sin6_addr.__u6_addr.__u6_addr32);
 3771                 cm_info.loc_port = ntohs(laddr6->sin6_port);
 3772                 cm_info.rem_port = ntohs(raddr6->sin6_port);
 3773                 irdma_netdev_vlan_ipv6(cm_info.loc_addr, &cm_info.vlan_id,
 3774                                        NULL);
 3775         }
 3776         cm_info.cm_id = cm_id;
 3777         cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
 3778         cm_info.tos = cm_id->tos;
 3779         if (iwdev->vsi.dscp_mode)
 3780                 cm_info.user_pri =
 3781                     iwqp->sc_qp.vsi->dscp_map[irdma_tos2dscp(cm_info.tos)];
 3782         else
 3783                 cm_info.user_pri = rt_tos2priority(cm_id->tos);
 3784 
 3785         if (iwqp->sc_qp.dev->ws_add(iwqp->sc_qp.vsi, cm_info.user_pri))
 3786                 return -ENOMEM;
 3787         iwqp->sc_qp.user_pri = cm_info.user_pri;
 3788         irdma_qp_add_qos(&iwqp->sc_qp);
 3789         if (iwdev->rf->sc_dev.hw_attrs.uk_attrs.hw_rev == IRDMA_GEN_2)
 3790                 iwdev->rf->check_fc(&iwdev->vsi, &iwqp->sc_qp);
 3791         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_DCB,
 3792                     "TOS:[%d] UP:[%d]\n", cm_id->tos, cm_info.user_pri);
 3793 
 3794         ret = irdma_create_cm_node(&iwdev->cm_core, iwdev, conn_param, &cm_info,
 3795                                    &cm_node);
 3796         if (ret)
 3797                 return ret;
 3798         ret = cm_node->cm_core->cm_create_ah(cm_node, true);
 3799         if (ret)
 3800                 goto err;
 3801         if (irdma_manage_qhash(iwdev, &cm_info,
 3802                                IRDMA_QHASH_TYPE_TCP_ESTABLISHED,
 3803                                IRDMA_QHASH_MANAGE_TYPE_ADD, NULL, true)) {
 3804                 ret = -EINVAL;
 3805                 goto err;
 3806         }
 3807         cm_node->qhash_set = true;
 3808 
 3809         cm_node->apbvt_entry = irdma_add_apbvt(iwdev, cm_info.loc_port);
 3810         if (!cm_node->apbvt_entry) {
 3811                 ret = -EINVAL;
 3812                 goto err;
 3813         }
 3814 
 3815         cm_node->apbvt_set = true;
 3816         iwqp->cm_node = cm_node;
 3817         cm_node->iwqp = iwqp;
 3818         iwqp->cm_id = cm_id;
 3819         irdma_qp_add_ref(&iwqp->ibqp);
 3820         cm_id->add_ref(cm_id);
 3821 
 3822         if (cm_node->state != IRDMA_CM_STATE_OFFLOADED) {
 3823                 cm_node->state = IRDMA_CM_STATE_SYN_SENT;
 3824                 ret = irdma_send_syn(cm_node, 0);
 3825                 if (ret)
 3826                         goto err;
 3827         }
 3828 
 3829         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3830                     "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",
 3831                     cm_node->rem_port, cm_node->loc_port, cm_node->rem_addr,
 3832                     cm_node->loc_addr, cm_node, cm_id, ibqp->qp_num);
 3833 
 3834         return 0;
 3835 
 3836 err:
 3837         if (cm_info.ipv4)
 3838                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3839                             "connect() FAILED: dest addr=%pI4",
 3840                             cm_info.rem_addr);
 3841         else
 3842                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3843                             "connect() FAILED: dest addr=%pI6",
 3844                             cm_info.rem_addr);
 3845         irdma_rem_ref_cm_node(cm_node);
 3846         iwdev->cm_core.stats_connect_errs++;
 3847 
 3848         return ret;
 3849 }
 3850 
 3851 /**
 3852  * irdma_create_listen - registered call creating listener
 3853  * @cm_id: cm information for passive connection
 3854  * @backlog: to max accept pending count
 3855  */
 3856 int
 3857 irdma_create_listen(struct iw_cm_id *cm_id, int backlog)
 3858 {
 3859         struct irdma_device *iwdev;
 3860         struct irdma_cm_listener *cm_listen_node;
 3861         struct irdma_cm_info cm_info = {0};
 3862         struct sockaddr_in *laddr;
 3863         struct sockaddr_in6 *laddr6;
 3864         bool wildcard = false;
 3865         int err;
 3866 
 3867         iwdev = to_iwdev(cm_id->device);
 3868         if (!iwdev)
 3869                 return -EINVAL;
 3870 
 3871         laddr = (struct sockaddr_in *)&cm_id->m_local_addr;
 3872         laddr6 = (struct sockaddr_in6 *)&cm_id->m_local_addr;
 3873         cm_info.qh_qpid = iwdev->vsi.ilq->qp_id;
 3874 
 3875         if (laddr->sin_family == AF_INET) {
 3876                 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV4)
 3877                         return -EINVAL;
 3878 
 3879                 cm_info.ipv4 = true;
 3880                 cm_info.loc_addr[0] = ntohl(laddr->sin_addr.s_addr);
 3881                 cm_info.loc_port = ntohs(laddr->sin_port);
 3882 
 3883                 if (laddr->sin_addr.s_addr != htonl(INADDR_ANY)) {
 3884                         cm_info.vlan_id = irdma_get_vlan_ipv4(cm_info.loc_addr);
 3885                 } else {
 3886                         cm_info.vlan_id = 0xFFFF;
 3887                         wildcard = true;
 3888                 }
 3889         } else {
 3890                 if (iwdev->vsi.mtu < IRDMA_MIN_MTU_IPV6)
 3891                         return -EINVAL;
 3892 
 3893                 cm_info.ipv4 = false;
 3894                 irdma_copy_ip_ntohl(cm_info.loc_addr,
 3895                                     laddr6->sin6_addr.__u6_addr.__u6_addr32);
 3896                 cm_info.loc_port = ntohs(laddr6->sin6_port);
 3897                 if (!IN6_IS_ADDR_UNSPECIFIED(&laddr6->sin6_addr)) {
 3898                         irdma_netdev_vlan_ipv6(cm_info.loc_addr,
 3899                                                &cm_info.vlan_id, NULL);
 3900                 } else {
 3901                         cm_info.vlan_id = 0xFFFF;
 3902                         wildcard = true;
 3903                 }
 3904         }
 3905 
 3906         if (cm_info.vlan_id >= VLAN_N_VID && iwdev->dcb_vlan_mode)
 3907                 cm_info.vlan_id = 0;
 3908         cm_info.backlog = backlog;
 3909         cm_info.cm_id = cm_id;
 3910 
 3911         cm_listen_node = irdma_make_listen_node(&iwdev->cm_core, iwdev,
 3912                                                 &cm_info);
 3913         if (!cm_listen_node) {
 3914                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3915                             "cm_listen_node == NULL\n");
 3916                 return -ENOMEM;
 3917         }
 3918 
 3919         cm_id->provider_data = cm_listen_node;
 3920 
 3921         cm_listen_node->tos = cm_id->tos;
 3922         if (iwdev->vsi.dscp_mode)
 3923                 cm_listen_node->user_pri =
 3924                     iwdev->vsi.dscp_map[irdma_tos2dscp(cm_id->tos)];
 3925         else
 3926                 cm_listen_node->user_pri = rt_tos2priority(cm_id->tos);
 3927         cm_info.user_pri = cm_listen_node->user_pri;
 3928         if (!cm_listen_node->reused_node) {
 3929                 if (wildcard) {
 3930                         err = irdma_add_mqh(iwdev, &cm_info, cm_listen_node);
 3931                         if (err)
 3932                                 goto error;
 3933                 } else {
 3934                         err = irdma_manage_qhash(iwdev, &cm_info,
 3935                                                  IRDMA_QHASH_TYPE_TCP_SYN,
 3936                                                  IRDMA_QHASH_MANAGE_TYPE_ADD,
 3937                                                  NULL, true);
 3938                         if (err)
 3939                                 goto error;
 3940 
 3941                         cm_listen_node->qhash_set = true;
 3942                 }
 3943 
 3944                 cm_listen_node->apbvt_entry = irdma_add_apbvt(iwdev,
 3945                                                               cm_info.loc_port);
 3946                 if (!cm_listen_node->apbvt_entry)
 3947                         goto error;
 3948         }
 3949         cm_id->add_ref(cm_id);
 3950         cm_listen_node->cm_core->stats_listen_created++;
 3951         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3952                     "loc_port=0x%04x loc_addr=%pI4 cm_listen_node=%p cm_id=%p qhash_set=%d vlan_id=%d\n",
 3953                     cm_listen_node->loc_port, cm_listen_node->loc_addr, cm_listen_node, cm_listen_node->cm_id,
 3954                     cm_listen_node->qhash_set, cm_listen_node->vlan_id);
 3955 
 3956         return 0;
 3957 
 3958 error:
 3959 
 3960         irdma_cm_del_listen(&iwdev->cm_core, cm_listen_node, false);
 3961 
 3962         return -EINVAL;
 3963 }
 3964 
 3965 /**
 3966  * irdma_destroy_listen - registered call to destroy listener
 3967  * @cm_id: cm information for passive connection
 3968  */
 3969 int
 3970 irdma_destroy_listen(struct iw_cm_id *cm_id)
 3971 {
 3972         struct irdma_device *iwdev;
 3973 
 3974         iwdev = to_iwdev(cm_id->device);
 3975         if (cm_id->provider_data)
 3976                 irdma_cm_del_listen(&iwdev->cm_core, cm_id->provider_data,
 3977                                     true);
 3978         else
 3979                 irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 3980                             "cm_id->provider_data was NULL\n");
 3981 
 3982         cm_id->rem_ref(cm_id);
 3983 
 3984         return 0;
 3985 }
 3986 
 3987 /**
 3988  * irdma_teardown_list_prep - add conn nodes slated for tear down to list
 3989  * @cm_core: cm's core
 3990  * @teardown_list: a list to which cm_node will be selected
 3991  * @ipaddr: pointer to ip address
 3992  * @nfo: pointer to cm_info structure instance
 3993  * @disconnect_all: flag indicating disconnect all QPs
 3994  */
 3995 static void
 3996 irdma_teardown_list_prep(struct irdma_cm_core *cm_core,
 3997                          struct list_head *teardown_list,
 3998                          u32 *ipaddr,
 3999                          struct irdma_cm_info *nfo,
 4000                          bool disconnect_all)
 4001 {
 4002         struct irdma_cm_node *cm_node;
 4003         int bkt;
 4004 
 4005         HASH_FOR_EACH_RCU(cm_core->cm_hash_tbl, bkt, cm_node, list) {
 4006                 if ((disconnect_all ||
 4007                      (nfo->vlan_id == cm_node->vlan_id &&
 4008                       !memcmp(cm_node->loc_addr, ipaddr, nfo->ipv4 ? 4 : 16))) &&
 4009                     atomic_inc_not_zero(&cm_node->refcnt))
 4010                         list_add(&cm_node->teardown_entry, teardown_list);
 4011         }
 4012 }
 4013 
 4014 /**
 4015  * irdma_cm_event_connected - handle connected active node
 4016  * @event: the info for cm_node of connection
 4017  */
 4018 static void
 4019 irdma_cm_event_connected(struct irdma_cm_event *event)
 4020 {
 4021         struct irdma_qp *iwqp;
 4022         struct irdma_device *iwdev;
 4023         struct irdma_cm_node *cm_node;
 4024         struct irdma_sc_dev *dev;
 4025         struct ib_qp_attr attr = {0};
 4026         struct iw_cm_id *cm_id;
 4027         int status;
 4028         bool read0;
 4029         int wait_ret = 0;
 4030 
 4031         cm_node = event->cm_node;
 4032         cm_id = cm_node->cm_id;
 4033         iwqp = cm_id->provider_data;
 4034         iwdev = iwqp->iwdev;
 4035         dev = &iwdev->rf->sc_dev;
 4036         if (iwqp->sc_qp.qp_uk.destroy_pending) {
 4037                 status = -ETIMEDOUT;
 4038                 goto error;
 4039         }
 4040 
 4041         irdma_cm_init_tsa_conn(iwqp, cm_node);
 4042         read0 = (cm_node->send_rdma0_op == SEND_RDMA_READ_ZERO);
 4043         if (iwqp->page)
 4044                 iwqp->sc_qp.qp_uk.sq_base = kmap_local_page(iwqp->page);
 4045         irdma_sc_send_rtt(&iwqp->sc_qp, read0);
 4046         if (iwqp->page)
 4047                 kunmap_local(iwqp->sc_qp.qp_uk.sq_base);
 4048 
 4049         attr.qp_state = IB_QPS_RTS;
 4050         cm_node->qhash_set = false;
 4051         irdma_modify_qp(&iwqp->ibqp, &attr, IB_QP_STATE, NULL);
 4052         if (dev->hw_attrs.uk_attrs.feature_flags & IRDMA_FEATURE_RTS_AE) {
 4053                 wait_ret = wait_event_interruptible_timeout(iwqp->waitq,
 4054                                                             iwqp->rts_ae_rcvd,
 4055                                                             IRDMA_MAX_TIMEOUT);
 4056                 if (!wait_ret)
 4057                         irdma_debug(iwdev_to_idev(iwdev), IRDMA_DEBUG_CM,
 4058                                     "Slow Connection: cm_node=%p, loc_port=%d, rem_port=%d, cm_id=%p\n",
 4059                                     cm_node, cm_node->loc_port,
 4060                                     cm_node->rem_port, cm_node->cm_id);
 4061         }
 4062 
 4063         irdma_send_cm_event(cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY, 0);
 4064         cm_node->accelerated = true;
 4065         complete(&cm_node->establish_comp);
 4066         cm_node->cm_core->cm_free_ah(cm_node);
 4067         return;
 4068 
 4069 error:
 4070         iwqp->cm_id = NULL;
 4071         cm_id->provider_data = NULL;
 4072         irdma_send_cm_event(event->cm_node, cm_id, IW_CM_EVENT_CONNECT_REPLY,
 4073                             status);
 4074         irdma_rem_ref_cm_node(event->cm_node);
 4075 }
 4076 
 4077 /**
 4078  * irdma_cm_event_reset - handle reset
 4079  * @event: the info for cm_node of connection
 4080  */
 4081 static void
 4082 irdma_cm_event_reset(struct irdma_cm_event *event)
 4083 {
 4084         struct irdma_cm_node *cm_node = event->cm_node;
 4085         struct iw_cm_id *cm_id = cm_node->cm_id;
 4086         struct irdma_qp *iwqp;
 4087 
 4088         if (!cm_id)
 4089                 return;
 4090 
 4091         iwqp = cm_id->provider_data;
 4092         if (!iwqp)
 4093                 return;
 4094 
 4095         irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 4096                     "reset event %p - cm_id = %p\n", event->cm_node, cm_id);
 4097         iwqp->cm_id = NULL;
 4098 
 4099         irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_DISCONNECT,
 4100                             -ECONNRESET);
 4101         irdma_send_cm_event(cm_node, cm_node->cm_id, IW_CM_EVENT_CLOSE, 0);
 4102 }
 4103 
 4104 /**
 4105  * irdma_cm_event_handler - send event to cm upper layer
 4106  * @work: pointer of cm event info.
 4107  */
 4108 static void
 4109 irdma_cm_event_handler(struct work_struct *work)
 4110 {
 4111         struct irdma_cm_event *event = container_of(work, struct irdma_cm_event, event_work);
 4112         struct irdma_cm_node *cm_node;
 4113 
 4114         if (!event || !event->cm_node || !event->cm_node->cm_core)
 4115                 return;
 4116 
 4117         cm_node = event->cm_node;
 4118 
 4119         switch (event->type) {
 4120         case IRDMA_CM_EVENT_MPA_REQ:
 4121                 irdma_send_cm_event(cm_node, cm_node->cm_id,
 4122                                     IW_CM_EVENT_CONNECT_REQUEST, 0);
 4123                 break;
 4124         case IRDMA_CM_EVENT_RESET:
 4125                 irdma_cm_event_reset(event);
 4126                 break;
 4127         case IRDMA_CM_EVENT_CONNECTED:
 4128                 if (!event->cm_node->cm_id ||
 4129                     event->cm_node->state != IRDMA_CM_STATE_OFFLOADED)
 4130                         break;
 4131                 irdma_cm_event_connected(event);
 4132                 break;
 4133         case IRDMA_CM_EVENT_MPA_REJECT:
 4134                 if (!event->cm_node->cm_id ||
 4135                     cm_node->state == IRDMA_CM_STATE_OFFLOADED)
 4136                         break;
 4137                 irdma_send_cm_event(cm_node, cm_node->cm_id,
 4138                                     IW_CM_EVENT_CONNECT_REPLY, -ECONNREFUSED);
 4139                 break;
 4140         case IRDMA_CM_EVENT_ABORTED:
 4141                 if (!event->cm_node->cm_id ||
 4142                     event->cm_node->state == IRDMA_CM_STATE_OFFLOADED)
 4143                         break;
 4144                 irdma_event_connect_error(event);
 4145                 break;
 4146         default:
 4147                 irdma_debug(iwdev_to_idev(cm_node->iwdev), IRDMA_DEBUG_CM,
 4148                             "bad event type = %d\n", event->type);
 4149                 break;
 4150         }
 4151 
 4152         irdma_rem_ref_cm_node(event->cm_node);
 4153         kfree(event);
 4154 }
 4155 
 4156 /**
 4157  * irdma_cm_post_event - queue event request for worker thread
 4158  * @event: cm node's info for up event call
 4159  */
 4160 static void
 4161 irdma_cm_post_event(struct irdma_cm_event *event)
 4162 {
 4163         atomic_inc(&event->cm_node->refcnt);
 4164         INIT_WORK(&event->event_work, irdma_cm_event_handler);
 4165         queue_work(event->cm_node->cm_core->event_wq, &event->event_work);
 4166 }
 4167 
 4168 /**
 4169  * irdma_cm_teardown_connections - teardown QPs
 4170  * @iwdev: device pointer
 4171  * @ipaddr: Pointer to IPv4 or IPv6 address
 4172  * @nfo: Connection info
 4173  * @disconnect_all: flag indicating disconnect all QPs
 4174  *
 4175  * teardown QPs where source or destination addr matches ip addr
 4176  */
 4177 void
 4178 irdma_cm_teardown_connections(struct irdma_device *iwdev, u32 *ipaddr,
 4179                               struct irdma_cm_info *nfo,
 4180                               bool disconnect_all)
 4181 {
 4182         struct irdma_cm_core *cm_core = &iwdev->cm_core;
 4183         struct list_head *list_core_temp;
 4184         struct list_head *list_node;
 4185         struct irdma_cm_node *cm_node;
 4186         struct list_head teardown_list;
 4187         struct ib_qp_attr attr;
 4188 
 4189         INIT_LIST_HEAD(&teardown_list);
 4190 
 4191         rcu_read_lock();
 4192         irdma_teardown_list_prep(cm_core, &teardown_list, ipaddr, nfo, disconnect_all);
 4193         rcu_read_unlock();
 4194 
 4195         list_for_each_safe(list_node, list_core_temp, &teardown_list) {
 4196                 cm_node = container_of(list_node, struct irdma_cm_node,
 4197                                        teardown_entry);
 4198                 attr.qp_state = IB_QPS_ERR;
 4199                 irdma_modify_qp(&cm_node->iwqp->ibqp, &attr, IB_QP_STATE, NULL);
 4200                 if (iwdev->rf->reset)
 4201                         irdma_cm_disconn(cm_node->iwqp);
 4202                 irdma_rem_ref_cm_node(cm_node);
 4203         }
 4204 }
Cache object: f7b93f80ee58c47dd3acdad7643f454f
[ source navigation ] [ diff markup ] [ identifier search ] [ freetext search ] [ file search ] [ list types ] [ track identifier ]
This page is part of the FreeBSD/Linux Linux Kernel Cross-Reference, and was automatically generated using a modified version of the LXR engine.
FreeBSD/Linux Kernel Cross Reference sys/dev/irdma/irdma_cm.c

FreeBSD/Linux Kernel Cross Reference
sys/dev/irdma/irdma_cm.c
