drivers/net/ethernet/chelsio/inline_crypto/chtls/chtls_cm.c

   1 // SPDX-License-Identifier: GPL-2.0-only
   2 /*
   3  * Copyright (c) 2018 Chelsio Communications, Inc.
   4  *
   5  * Written by: Atul Gupta (atul.gupta@chelsio.com)
   6  */
   7
   8 #include <linux/module.h>
   9 #include <linux/list.h>
  10 #include <linux/workqueue.h>
  11 #include <linux/skbuff.h>
  12 #include <linux/timer.h>
  13 #include <linux/notifier.h>
  14 #include <linux/inetdevice.h>
  15 #include <linux/ip.h>
  16 #include <linux/tcp.h>
  17 #include <linux/sched/signal.h>
  18 #include <linux/kallsyms.h>
  19 #include <linux/kprobes.h>
  20 #include <linux/if_vlan.h>
  21 #include <linux/ipv6.h>
  22 #include <net/ipv6.h>
  23 #include <net/transp_v6.h>
  24 #include <net/ip6_route.h>
  25 #include <net/inet_common.h>
  26 #include <net/tcp.h>
  27 #include <net/dst.h>
  28 #include <net/tls.h>
  29 #include <net/addrconf.h>
  30 #include <net/secure_seq.h>
  31
  32 #include "chtls.h"
  33 #include "chtls_cm.h"
  34 #include "clip_tbl.h"
  35
  36 /*
  37  * State transitions and actions for close.  Note that if we are in SYN_SENT
  38  * we remain in that state as we cannot control a connection while it's in
  39  * SYN_SENT; such connections are allowed to establish and are then aborted.
  40  */
  41 static unsigned char new_state[16] = {
  42         /* current state:     new state:      action: */
  43         /* (Invalid)       */ TCP_CLOSE,
  44         /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  45         /* TCP_SYN_SENT    */ TCP_SYN_SENT,
  46         /* TCP_SYN_RECV    */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  47         /* TCP_FIN_WAIT1   */ TCP_FIN_WAIT1,
  48         /* TCP_FIN_WAIT2   */ TCP_FIN_WAIT2,
  49         /* TCP_TIME_WAIT   */ TCP_CLOSE,
  50         /* TCP_CLOSE       */ TCP_CLOSE,
  51         /* TCP_CLOSE_WAIT  */ TCP_LAST_ACK | TCP_ACTION_FIN,
  52         /* TCP_LAST_ACK    */ TCP_LAST_ACK,
  53         /* TCP_LISTEN      */ TCP_CLOSE,
  54         /* TCP_CLOSING     */ TCP_CLOSING,
  55 };
  56
  57 static struct chtls_sock *chtls_sock_create(struct chtls_dev *cdev)
  58 {
  59         struct chtls_sock *csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
  60
  61         if (!csk)
  62                 return NULL;
  63
  64         csk->txdata_skb_cache = alloc_skb(TXDATA_SKB_LEN, GFP_ATOMIC);
  65         if (!csk->txdata_skb_cache) {
  66                 kfree(csk);
  67                 return NULL;
  68         }
  69
  70         kref_init(&csk->kref);
  71         csk->cdev = cdev;
  72         skb_queue_head_init(&csk->txq);
  73         csk->wr_skb_head = NULL;
  74         csk->wr_skb_tail = NULL;
  75         csk->mss = MAX_MSS;
  76         csk->tlshws.ofld = 1;
  77         csk->tlshws.txkey = -1;
  78         csk->tlshws.rxkey = -1;
  79         csk->tlshws.mfs = TLS_MFS;
  80         skb_queue_head_init(&csk->tlshws.sk_recv_queue);
  81         return csk;
  82 }
  83
  84 static void chtls_sock_release(struct kref *ref)
  85 {
  86         struct chtls_sock *csk =
  87                 container_of(ref, struct chtls_sock, kref);
  88
  89         kfree(csk);
  90 }
  91
  92 static struct net_device *chtls_find_netdev(struct chtls_dev *cdev,
  93                                             struct sock *sk)
  94 {
  95         struct adapter *adap = pci_get_drvdata(cdev->pdev);
  96         struct net_device *ndev = cdev->ports[0];
  97 #if IS_ENABLED(CONFIG_IPV6)
  98         struct net_device *temp;
  99         int addr_type;
 100 #endif
 101         int i;
 102
 103         switch (sk->sk_family) {
 104         case PF_INET:
 105                 if (likely(!inet_sk(sk)->inet_rcv_saddr))
 106                         return ndev;
 107                 ndev = __ip_dev_find(&init_net, inet_sk(sk)->inet_rcv_saddr, false);
 108                 break;
 109 #if IS_ENABLED(CONFIG_IPV6)
 110         case PF_INET6:
 111                 addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
 112                 if (likely(addr_type == IPV6_ADDR_ANY))
 113                         return ndev;
 114
 115                 for_each_netdev_rcu(&init_net, temp) {
 116                         if (ipv6_chk_addr(&init_net, (struct in6_addr *)
 117                                           &sk->sk_v6_rcv_saddr, temp, 1)) {
 118                                 ndev = temp;
 119                                 break;
 120                         }
 121                 }
 122         break;
 123 #endif
 124         default:
 125                 return NULL;
 126         }
 127
 128         if (!ndev)
 129                 return NULL;
 130
 131         if (is_vlan_dev(ndev))
 132                 ndev = vlan_dev_real_dev(ndev);
 133
 134         for_each_port(adap, i)
 135                 if (cdev->ports[i] == ndev)
 136                         return ndev;
 137         return NULL;
 138 }
 139
 140 static void assign_rxopt(struct sock *sk, unsigned int opt)
 141 {
 142         const struct chtls_dev *cdev;
 143         struct chtls_sock *csk;
 144         struct tcp_sock *tp;
 145
 146         csk = rcu_dereference_sk_user_data(sk);
 147         tp = tcp_sk(sk);
 148
 149         cdev = csk->cdev;
 150         tp->tcp_header_len           = sizeof(struct tcphdr);
 151         tp->rx_opt.mss_clamp         = cdev->mtus[TCPOPT_MSS_G(opt)] - 40;
 152         tp->mss_cache                = tp->rx_opt.mss_clamp;
 153         tp->rx_opt.tstamp_ok         = TCPOPT_TSTAMP_G(opt);
 154         tp->rx_opt.snd_wscale        = TCPOPT_SACK_G(opt);
 155         tp->rx_opt.wscale_ok         = TCPOPT_WSCALE_OK_G(opt);
 156         SND_WSCALE(tp)               = TCPOPT_SND_WSCALE_G(opt);
 157         if (!tp->rx_opt.wscale_ok)
 158                 tp->rx_opt.rcv_wscale = 0;
 159         if (tp->rx_opt.tstamp_ok) {
 160                 tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
 161                 tp->rx_opt.mss_clamp -= TCPOLEN_TSTAMP_ALIGNED;
 162         } else if (csk->opt2 & TSTAMPS_EN_F) {
 163                 csk->opt2 &= ~TSTAMPS_EN_F;
 164                 csk->mtu_idx = TCPOPT_MSS_G(opt);
 165         }
 166 }
 167
 168 static void chtls_purge_receive_queue(struct sock *sk)
 169 {
 170         struct sk_buff *skb;
 171
 172         while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 173                 skb_dst_set(skb, (void *)NULL);
 174                 kfree_skb(skb);
 175         }
 176 }
 177
 178 static void chtls_purge_write_queue(struct sock *sk)
 179 {
 180         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 181         struct sk_buff *skb;
 182
 183         while ((skb = __skb_dequeue(&csk->txq))) {
 184                 sk->sk_wmem_queued -= skb->truesize;
 185                 __kfree_skb(skb);
 186         }
 187 }
 188
 189 static void chtls_purge_recv_queue(struct sock *sk)
 190 {
 191         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 192         struct chtls_hws *tlsk = &csk->tlshws;
 193         struct sk_buff *skb;
 194
 195         while ((skb = __skb_dequeue(&tlsk->sk_recv_queue)) != NULL) {
 196                 skb_dst_set(skb, NULL);
 197                 kfree_skb(skb);
 198         }
 199 }
 200
 201 static void abort_arp_failure(void *handle, struct sk_buff *skb)
 202 {
 203         struct cpl_abort_req *req = cplhdr(skb);
 204         struct chtls_dev *cdev;
 205
 206         cdev = (struct chtls_dev *)handle;
 207         req->cmd = CPL_ABORT_NO_RST;
 208         cxgb4_ofld_send(cdev->lldi->ports[0], skb);
 209 }
 210
 211 static struct sk_buff *alloc_ctrl_skb(struct sk_buff *skb, int len)
 212 {
 213         if (likely(skb && !skb_shared(skb) && !skb_cloned(skb))) {
 214                 __skb_trim(skb, 0);
 215                 refcount_inc(&skb->users);
 216         } else {
 217                 skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
 218         }
 219         return skb;
 220 }
 221
 222 static void chtls_send_abort(struct sock *sk, int mode, struct sk_buff *skb)
 223 {
 224         struct cpl_abort_req *req;
 225         struct chtls_sock *csk;
 226         struct tcp_sock *tp;
 227
 228         csk = rcu_dereference_sk_user_data(sk);
 229         tp = tcp_sk(sk);
 230
 231         if (!skb)
 232                 skb = alloc_ctrl_skb(csk->txdata_skb_cache, sizeof(*req));
 233
 234         req = (struct cpl_abort_req *)skb_put(skb, sizeof(*req));
 235         INIT_TP_WR_CPL(req, CPL_ABORT_REQ, csk->tid);
 236         skb_set_queue_mapping(skb, (csk->txq_idx << 1) | CPL_PRIORITY_DATA);
 237         req->rsvd0 = htonl(tp->snd_nxt);
 238         req->rsvd1 = !csk_flag_nochk(csk, CSK_TX_DATA_SENT);
 239         req->cmd = mode;
 240         t4_set_arp_err_handler(skb, csk->cdev, abort_arp_failure);
 241         send_or_defer(sk, tp, skb, mode == CPL_ABORT_SEND_RST);
 242 }
 243
 244 static void chtls_send_reset(struct sock *sk, int mode, struct sk_buff *skb)
 245 {
 246         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 247
 248         if (unlikely(csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) ||
 249                      !csk->cdev)) {
 250                 if (sk->sk_state == TCP_SYN_RECV)
 251                         csk_set_flag(csk, CSK_RST_ABORTED);
 252                 goto out;
 253         }
 254
 255         if (!csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
 256                 struct tcp_sock *tp = tcp_sk(sk);
 257
 258                 if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
 259                         WARN_ONCE(1, "send tx flowc error");
 260                 csk_set_flag(csk, CSK_TX_DATA_SENT);
 261         }
 262
 263         csk_set_flag(csk, CSK_ABORT_RPL_PENDING);
 264         chtls_purge_write_queue(sk);
 265
 266         csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
 267         if (sk->sk_state != TCP_SYN_RECV)
 268                 chtls_send_abort(sk, mode, skb);
 269         else
 270                 goto out;
 271
 272         return;
 273 out:
 274         kfree_skb(skb);
 275 }
 276
 277 static void release_tcp_port(struct sock *sk)
 278 {
 279         if (inet_csk(sk)->icsk_bind_hash)
 280                 inet_put_port(sk);
 281 }
 282
 283 static void tcp_uncork(struct sock *sk)
 284 {
 285         struct tcp_sock *tp = tcp_sk(sk);
 286
 287         if (tp->nonagle & TCP_NAGLE_CORK) {
 288                 tp->nonagle &= ~TCP_NAGLE_CORK;
 289                 chtls_tcp_push(sk, 0);
 290         }
 291 }
 292
 293 static void chtls_close_conn(struct sock *sk)
 294 {
 295         struct cpl_close_con_req *req;
 296         struct chtls_sock *csk;
 297         struct sk_buff *skb;
 298         unsigned int tid;
 299         unsigned int len;
 300
 301         len = roundup(sizeof(struct cpl_close_con_req), 16);
 302         csk = rcu_dereference_sk_user_data(sk);
 303         tid = csk->tid;
 304
 305         skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
 306         req = (struct cpl_close_con_req *)__skb_put(skb, len);
 307         memset(req, 0, len);
 308         req->wr.wr_hi = htonl(FW_WR_OP_V(FW_TP_WR) |
 309                               FW_WR_IMMDLEN_V(sizeof(*req) -
 310                                               sizeof(req->wr)));
 311         req->wr.wr_mid = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)) |
 312                                FW_WR_FLOWID_V(tid));
 313
 314         OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
 315
 316         tcp_uncork(sk);
 317         skb_entail(sk, skb, ULPCB_FLAG_NO_HDR | ULPCB_FLAG_NO_APPEND);
 318         if (sk->sk_state != TCP_SYN_SENT)
 319                 chtls_push_frames(csk, 1);
 320 }
 321
 322 /*
 323  * Perform a state transition during close and return the actions indicated
 324  * for the transition.  Do not make this function inline, the main reason
 325  * it exists at all is to avoid multiple inlining of tcp_set_state.
 326  */
 327 static int make_close_transition(struct sock *sk)
 328 {
 329         int next = (int)new_state[sk->sk_state];
 330
 331         tcp_set_state(sk, next & TCP_STATE_MASK);
 332         return next & TCP_ACTION_FIN;
 333 }
 334
 335 void chtls_close(struct sock *sk, long timeout)
 336 {
 337         int data_lost, prev_state;
 338         struct chtls_sock *csk;
 339
 340         csk = rcu_dereference_sk_user_data(sk);
 341
 342         lock_sock(sk);
 343         sk->sk_shutdown |= SHUTDOWN_MASK;
 344
 345         data_lost = skb_queue_len(&sk->sk_receive_queue);
 346         data_lost |= skb_queue_len(&csk->tlshws.sk_recv_queue);
 347         chtls_purge_recv_queue(sk);
 348         chtls_purge_receive_queue(sk);
 349
 350         if (sk->sk_state == TCP_CLOSE) {
 351                 goto wait;
 352         } else if (data_lost || sk->sk_state == TCP_SYN_SENT) {
 353                 chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
 354                 release_tcp_port(sk);
 355                 goto unlock;
 356         } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
 357                 sk->sk_prot->disconnect(sk, 0);
 358         } else if (make_close_transition(sk)) {
 359                 chtls_close_conn(sk);
 360         }
 361 wait:
 362         if (timeout)
 363                 sk_stream_wait_close(sk, timeout);
 364
 365 unlock:
 366         prev_state = sk->sk_state;
 367         sock_hold(sk);
 368         sock_orphan(sk);
 369
 370         release_sock(sk);
 371
 372         local_bh_disable();
 373         bh_lock_sock(sk);
 374
 375         if (prev_state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
 376                 goto out;
 377
 378         if (sk->sk_state == TCP_FIN_WAIT2 && tcp_sk(sk)->linger2 < 0 &&
 379             !csk_flag(sk, CSK_ABORT_SHUTDOWN)) {
 380                 struct sk_buff *skb;
 381
 382                 skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
 383                 if (skb)
 384                         chtls_send_reset(sk, CPL_ABORT_SEND_RST, skb);
 385         }
 386
 387         if (sk->sk_state == TCP_CLOSE)
 388                 inet_csk_destroy_sock(sk);
 389
 390 out:
 391         bh_unlock_sock(sk);
 392         local_bh_enable();
 393         sock_put(sk);
 394 }
 395
 396 /*
 397  * Wait until a socket enters on of the given states.
 398  */
 399 static int wait_for_states(struct sock *sk, unsigned int states)
 400 {
 401         DECLARE_WAITQUEUE(wait, current);
 402         struct socket_wq _sk_wq;
 403         long current_timeo;
 404         int err = 0;
 405
 406         current_timeo = 200;
 407
 408         /*
 409          * We want this to work even when there's no associated struct socket.
 410          * In that case we provide a temporary wait_queue_head_t.
 411          */
 412         if (!sk->sk_wq) {
 413                 init_waitqueue_head(&_sk_wq.wait);
 414                 _sk_wq.fasync_list = NULL;
 415                 init_rcu_head_on_stack(&_sk_wq.rcu);
 416                 RCU_INIT_POINTER(sk->sk_wq, &_sk_wq);
 417         }
 418
 419         add_wait_queue(sk_sleep(sk), &wait);
 420         while (!sk_in_state(sk, states)) {
 421                 if (!current_timeo) {
 422                         err = -EBUSY;
 423                         break;
 424                 }
 425                 if (signal_pending(current)) {
 426                         err = sock_intr_errno(current_timeo);
 427                         break;
 428                 }
 429                 set_current_state(TASK_UNINTERRUPTIBLE);
 430                 release_sock(sk);
 431                 if (!sk_in_state(sk, states))
 432                         current_timeo = schedule_timeout(current_timeo);
 433                 __set_current_state(TASK_RUNNING);
 434                 lock_sock(sk);
 435         }
 436         remove_wait_queue(sk_sleep(sk), &wait);
 437
 438         if (rcu_dereference(sk->sk_wq) == &_sk_wq)
 439                 sk->sk_wq = NULL;
 440         return err;
 441 }
 442
 443 int chtls_disconnect(struct sock *sk, int flags)
 444 {
 445         struct tcp_sock *tp;
 446         int err;
 447
 448         tp = tcp_sk(sk);
 449         chtls_purge_recv_queue(sk);
 450         chtls_purge_receive_queue(sk);
 451         chtls_purge_write_queue(sk);
 452
 453         if (sk->sk_state != TCP_CLOSE) {
 454                 sk->sk_err = ECONNRESET;
 455                 chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
 456                 err = wait_for_states(sk, TCPF_CLOSE);
 457                 if (err)
 458                         return err;
 459         }
 460         chtls_purge_recv_queue(sk);
 461         chtls_purge_receive_queue(sk);
 462         tp->max_window = 0xFFFF << (tp->rx_opt.snd_wscale);
 463         return tcp_disconnect(sk, flags);
 464 }
 465
 466 #define SHUTDOWN_ELIGIBLE_STATE (TCPF_ESTABLISHED | \
 467                                  TCPF_SYN_RECV | TCPF_CLOSE_WAIT)
 468 void chtls_shutdown(struct sock *sk, int how)
 469 {
 470         if ((how & SEND_SHUTDOWN) &&
 471             sk_in_state(sk, SHUTDOWN_ELIGIBLE_STATE) &&
 472             make_close_transition(sk))
 473                 chtls_close_conn(sk);
 474 }
 475
 476 void chtls_destroy_sock(struct sock *sk)
 477 {
 478         struct chtls_sock *csk;
 479
 480         csk = rcu_dereference_sk_user_data(sk);
 481         chtls_purge_recv_queue(sk);
 482         csk->ulp_mode = ULP_MODE_NONE;
 483         chtls_purge_write_queue(sk);
 484         free_tls_keyid(sk);
 485         kref_put(&csk->kref, chtls_sock_release);
 486         if (sk->sk_family == AF_INET)
 487                 sk->sk_prot = &tcp_prot;
 488 #if IS_ENABLED(CONFIG_IPV6)
 489         else
 490                 sk->sk_prot = &tcpv6_prot;
 491 #endif
 492         sk->sk_prot->destroy(sk);
 493 }
 494
 495 static void reset_listen_child(struct sock *child)
 496 {
 497         struct chtls_sock *csk = rcu_dereference_sk_user_data(child);
 498         struct sk_buff *skb;
 499
 500         skb = alloc_ctrl_skb(csk->txdata_skb_cache,
 501                              sizeof(struct cpl_abort_req));
 502
 503         chtls_send_reset(child, CPL_ABORT_SEND_RST, skb);
 504         sock_orphan(child);
 505         INC_ORPHAN_COUNT(child);
 506         if (child->sk_state == TCP_CLOSE)
 507                 inet_csk_destroy_sock(child);
 508 }
 509
 510 static void chtls_disconnect_acceptq(struct sock *listen_sk)
 511 {
 512         struct request_sock **pprev;
 513
 514         pprev = ACCEPT_QUEUE(listen_sk);
 515         while (*pprev) {
 516                 struct request_sock *req = *pprev;
 517
 518                 if (req->rsk_ops == &chtls_rsk_ops ||
 519                     req->rsk_ops == &chtls_rsk_opsv6) {
 520                         struct sock *child = req->sk;
 521
 522                         *pprev = req->dl_next;
 523                         sk_acceptq_removed(listen_sk);
 524                         reqsk_put(req);
 525                         sock_hold(child);
 526                         local_bh_disable();
 527                         bh_lock_sock(child);
 528                         release_tcp_port(child);
 529                         reset_listen_child(child);
 530                         bh_unlock_sock(child);
 531                         local_bh_enable();
 532                         sock_put(child);
 533                 } else {
 534                         pprev = &req->dl_next;
 535                 }
 536         }
 537 }
 538
 539 static int listen_hashfn(const struct sock *sk)
 540 {
 541         return ((unsigned long)sk >> 10) & (LISTEN_INFO_HASH_SIZE - 1);
 542 }
 543
 544 static struct listen_info *listen_hash_add(struct chtls_dev *cdev,
 545                                            struct sock *sk,
 546                                            unsigned int stid)
 547 {
 548         struct listen_info *p = kmalloc(sizeof(*p), GFP_KERNEL);
 549
 550         if (p) {
 551                 int key = listen_hashfn(sk);
 552
 553                 p->sk = sk;
 554                 p->stid = stid;
 555                 spin_lock(&cdev->listen_lock);
 556                 p->next = cdev->listen_hash_tab[key];
 557                 cdev->listen_hash_tab[key] = p;
 558                 spin_unlock(&cdev->listen_lock);
 559         }
 560         return p;
 561 }
 562
 563 static int listen_hash_find(struct chtls_dev *cdev,
 564                             struct sock *sk)
 565 {
 566         struct listen_info *p;
 567         int stid = -1;
 568         int key;
 569
 570         key = listen_hashfn(sk);
 571
 572         spin_lock(&cdev->listen_lock);
 573         for (p = cdev->listen_hash_tab[key]; p; p = p->next)
 574                 if (p->sk == sk) {
 575                         stid = p->stid;
 576                         break;
 577                 }
 578         spin_unlock(&cdev->listen_lock);
 579         return stid;
 580 }
 581
 582 static int listen_hash_del(struct chtls_dev *cdev,
 583                            struct sock *sk)
 584 {
 585         struct listen_info *p, **prev;
 586         int stid = -1;
 587         int key;
 588
 589         key = listen_hashfn(sk);
 590         prev = &cdev->listen_hash_tab[key];
 591
 592         spin_lock(&cdev->listen_lock);
 593         for (p = *prev; p; prev = &p->next, p = p->next)
 594                 if (p->sk == sk) {
 595                         stid = p->stid;
 596                         *prev = p->next;
 597                         kfree(p);
 598                         break;
 599                 }
 600         spin_unlock(&cdev->listen_lock);
 601         return stid;
 602 }
 603
 604 static void cleanup_syn_rcv_conn(struct sock *child, struct sock *parent)
 605 {
 606         struct request_sock *req;
 607         struct chtls_sock *csk;
 608
 609         csk = rcu_dereference_sk_user_data(child);
 610         req = csk->passive_reap_next;
 611
 612         reqsk_queue_removed(&inet_csk(parent)->icsk_accept_queue, req);
 613         __skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
 614         chtls_reqsk_free(req);
 615         csk->passive_reap_next = NULL;
 616 }
 617
 618 static void chtls_reset_synq(struct listen_ctx *listen_ctx)
 619 {
 620         struct sock *listen_sk = listen_ctx->lsk;
 621
 622         while (!skb_queue_empty(&listen_ctx->synq)) {
 623                 struct chtls_sock *csk =
 624                         container_of((struct synq *)skb_peek
 625                                 (&listen_ctx->synq), struct chtls_sock, synq);
 626                 struct sock *child = csk->sk;
 627
 628                 cleanup_syn_rcv_conn(child, listen_sk);
 629                 sock_hold(child);
 630                 local_bh_disable();
 631                 bh_lock_sock(child);
 632                 release_tcp_port(child);
 633                 reset_listen_child(child);
 634                 bh_unlock_sock(child);
 635                 local_bh_enable();
 636                 sock_put(child);
 637         }
 638 }
 639
 640 int chtls_listen_start(struct chtls_dev *cdev, struct sock *sk)
 641 {
 642         struct net_device *ndev;
 643 #if IS_ENABLED(CONFIG_IPV6)
 644         bool clip_valid = false;
 645 #endif
 646         struct listen_ctx *ctx;
 647         struct adapter *adap;
 648         struct port_info *pi;
 649         int ret = 0;
 650         int stid;
 651
 652         rcu_read_lock();
 653         ndev = chtls_find_netdev(cdev, sk);
 654         rcu_read_unlock();
 655         if (!ndev)
 656                 return -EBADF;
 657
 658         pi = netdev_priv(ndev);
 659         adap = pi->adapter;
 660         if (!(adap->flags & CXGB4_FULL_INIT_DONE))
 661                 return -EBADF;
 662
 663         if (listen_hash_find(cdev, sk) >= 0)   /* already have it */
 664                 return -EADDRINUSE;
 665
 666         ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 667         if (!ctx)
 668                 return -ENOMEM;
 669
 670         __module_get(THIS_MODULE);
 671         ctx->lsk = sk;
 672         ctx->cdev = cdev;
 673         ctx->state = T4_LISTEN_START_PENDING;
 674         skb_queue_head_init(&ctx->synq);
 675
 676         stid = cxgb4_alloc_stid(cdev->tids, sk->sk_family, ctx);
 677         if (stid < 0)
 678                 goto free_ctx;
 679
 680         sock_hold(sk);
 681         if (!listen_hash_add(cdev, sk, stid))
 682                 goto free_stid;
 683
 684         if (sk->sk_family == PF_INET) {
 685                 ret = cxgb4_create_server(ndev, stid,
 686                                           inet_sk(sk)->inet_rcv_saddr,
 687                                           inet_sk(sk)->inet_sport, 0,
 688                                           cdev->lldi->rxq_ids[0]);
 689 #if IS_ENABLED(CONFIG_IPV6)
 690         } else {
 691                 int addr_type;
 692
 693                 addr_type = ipv6_addr_type(&sk->sk_v6_rcv_saddr);
 694                 if (addr_type != IPV6_ADDR_ANY) {
 695                         ret = cxgb4_clip_get(ndev, (const u32 *)
 696                                              &sk->sk_v6_rcv_saddr, 1);
 697                         if (ret)
 698                                 goto del_hash;
 699                         clip_valid = true;
 700                 }
 701                 ret = cxgb4_create_server6(ndev, stid,
 702                                            &sk->sk_v6_rcv_saddr,
 703                                            inet_sk(sk)->inet_sport,
 704                                            cdev->lldi->rxq_ids[0]);
 705 #endif
 706         }
 707         if (ret > 0)
 708                 ret = net_xmit_errno(ret);
 709         if (ret)
 710                 goto del_hash;
 711         return 0;
 712 del_hash:
 713 #if IS_ENABLED(CONFIG_IPV6)
 714         if (clip_valid)
 715                 cxgb4_clip_release(ndev, (const u32 *)&sk->sk_v6_rcv_saddr, 1);
 716 #endif
 717         listen_hash_del(cdev, sk);
 718 free_stid:
 719         cxgb4_free_stid(cdev->tids, stid, sk->sk_family);
 720         sock_put(sk);
 721 free_ctx:
 722         kfree(ctx);
 723         module_put(THIS_MODULE);
 724         return -EBADF;
 725 }
 726
 727 void chtls_listen_stop(struct chtls_dev *cdev, struct sock *sk)
 728 {
 729         struct listen_ctx *listen_ctx;
 730         int stid;
 731
 732         stid = listen_hash_del(cdev, sk);
 733         if (stid < 0)
 734                 return;
 735
 736         listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
 737         chtls_reset_synq(listen_ctx);
 738
 739         cxgb4_remove_server(cdev->lldi->ports[0], stid,
 740                             cdev->lldi->rxq_ids[0], sk->sk_family == PF_INET6);
 741
 742 #if IS_ENABLED(CONFIG_IPV6)
 743         if (sk->sk_family == PF_INET6) {
 744                 struct net_device *ndev = chtls_find_netdev(cdev, sk);
 745                 int addr_type = 0;
 746
 747                 addr_type = ipv6_addr_type((const struct in6_addr *)
 748                                           &sk->sk_v6_rcv_saddr);
 749                 if (addr_type != IPV6_ADDR_ANY)
 750                         cxgb4_clip_release(ndev, (const u32 *)
 751                                            &sk->sk_v6_rcv_saddr, 1);
 752         }
 753 #endif
 754         chtls_disconnect_acceptq(sk);
 755 }
 756
 757 static int chtls_pass_open_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
 758 {
 759         struct cpl_pass_open_rpl *rpl = cplhdr(skb) + RSS_HDR;
 760         unsigned int stid = GET_TID(rpl);
 761         struct listen_ctx *listen_ctx;
 762
 763         listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
 764         if (!listen_ctx)
 765                 return CPL_RET_BUF_DONE;
 766
 767         if (listen_ctx->state == T4_LISTEN_START_PENDING) {
 768                 listen_ctx->state = T4_LISTEN_STARTED;
 769                 return CPL_RET_BUF_DONE;
 770         }
 771
 772         if (rpl->status != CPL_ERR_NONE) {
 773                 pr_info("Unexpected PASS_OPEN_RPL status %u for STID %u\n",
 774                         rpl->status, stid);
 775         } else {
 776                 cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
 777                 sock_put(listen_ctx->lsk);
 778                 kfree(listen_ctx);
 779                 module_put(THIS_MODULE);
 780         }
 781         return CPL_RET_BUF_DONE;
 782 }
 783
 784 static int chtls_close_listsrv_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
 785 {
 786         struct cpl_close_listsvr_rpl *rpl = cplhdr(skb) + RSS_HDR;
 787         struct listen_ctx *listen_ctx;
 788         unsigned int stid;
 789         void *data;
 790
 791         stid = GET_TID(rpl);
 792         data = lookup_stid(cdev->tids, stid);
 793         listen_ctx = (struct listen_ctx *)data;
 794
 795         if (rpl->status != CPL_ERR_NONE) {
 796                 pr_info("Unexpected CLOSE_LISTSRV_RPL status %u for STID %u\n",
 797                         rpl->status, stid);
 798         } else {
 799                 cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
 800                 sock_put(listen_ctx->lsk);
 801                 kfree(listen_ctx);
 802                 module_put(THIS_MODULE);
 803         }
 804         return CPL_RET_BUF_DONE;
 805 }
 806
 807 static void chtls_purge_wr_queue(struct sock *sk)
 808 {
 809         struct sk_buff *skb;
 810
 811         while ((skb = dequeue_wr(sk)) != NULL)
 812                 kfree_skb(skb);
 813 }
 814
 815 static void chtls_release_resources(struct sock *sk)
 816 {
 817         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 818         struct chtls_dev *cdev = csk->cdev;
 819         unsigned int tid = csk->tid;
 820         struct tid_info *tids;
 821
 822         if (!cdev)
 823                 return;
 824
 825         tids = cdev->tids;
 826         kfree_skb(csk->txdata_skb_cache);
 827         csk->txdata_skb_cache = NULL;
 828
 829         if (csk->wr_credits != csk->wr_max_credits) {
 830                 chtls_purge_wr_queue(sk);
 831                 chtls_reset_wr_list(csk);
 832         }
 833
 834         if (csk->l2t_entry) {
 835                 cxgb4_l2t_release(csk->l2t_entry);
 836                 csk->l2t_entry = NULL;
 837         }
 838
 839         if (sk->sk_state != TCP_SYN_SENT) {
 840                 cxgb4_remove_tid(tids, csk->port_id, tid, sk->sk_family);
 841                 sock_put(sk);
 842         }
 843 }
 844
 845 static void chtls_conn_done(struct sock *sk)
 846 {
 847         if (sock_flag(sk, SOCK_DEAD))
 848                 chtls_purge_receive_queue(sk);
 849         sk_wakeup_sleepers(sk, 0);
 850         tcp_done(sk);
 851 }
 852
 853 static void do_abort_syn_rcv(struct sock *child, struct sock *parent)
 854 {
 855         /*
 856          * If the server is still open we clean up the child connection,
 857          * otherwise the server already did the clean up as it was purging
 858          * its SYN queue and the skb was just sitting in its backlog.
 859          */
 860         if (likely(parent->sk_state == TCP_LISTEN)) {
 861                 cleanup_syn_rcv_conn(child, parent);
 862                 /* Without the below call to sock_orphan,
 863                  * we leak the socket resource with syn_flood test
 864                  * as inet_csk_destroy_sock will not be called
 865                  * in tcp_done since SOCK_DEAD flag is not set.
 866                  * Kernel handles this differently where new socket is
 867                  * created only after 3 way handshake is done.
 868                  */
 869                 sock_orphan(child);
 870                 percpu_counter_inc((child)->sk_prot->orphan_count);
 871                 chtls_release_resources(child);
 872                 chtls_conn_done(child);
 873         } else {
 874                 if (csk_flag(child, CSK_RST_ABORTED)) {
 875                         chtls_release_resources(child);
 876                         chtls_conn_done(child);
 877                 }
 878         }
 879 }
 880
 881 static void pass_open_abort(struct sock *child, struct sock *parent,
 882                             struct sk_buff *skb)
 883 {
 884         do_abort_syn_rcv(child, parent);
 885         kfree_skb(skb);
 886 }
 887
 888 static void bl_pass_open_abort(struct sock *lsk, struct sk_buff *skb)
 889 {
 890         pass_open_abort(skb->sk, lsk, skb);
 891 }
 892
 893 static void chtls_pass_open_arp_failure(struct sock *sk,
 894                                         struct sk_buff *skb)
 895 {
 896         const struct request_sock *oreq;
 897         struct chtls_sock *csk;
 898         struct chtls_dev *cdev;
 899         struct sock *parent;
 900         void *data;
 901
 902         csk = rcu_dereference_sk_user_data(sk);
 903         cdev = csk->cdev;
 904
 905         /*
 906          * If the connection is being aborted due to the parent listening
 907          * socket going away there's nothing to do, the ABORT_REQ will close
 908          * the connection.
 909          */
 910         if (csk_flag(sk, CSK_ABORT_RPL_PENDING)) {
 911                 kfree_skb(skb);
 912                 return;
 913         }
 914
 915         oreq = csk->passive_reap_next;
 916         data = lookup_stid(cdev->tids, oreq->ts_recent);
 917         parent = ((struct listen_ctx *)data)->lsk;
 918
 919         bh_lock_sock(parent);
 920         if (!sock_owned_by_user(parent)) {
 921                 pass_open_abort(sk, parent, skb);
 922         } else {
 923                 BLOG_SKB_CB(skb)->backlog_rcv = bl_pass_open_abort;
 924                 __sk_add_backlog(parent, skb);
 925         }
 926         bh_unlock_sock(parent);
 927 }
 928
 929 static void chtls_accept_rpl_arp_failure(void *handle,
 930                                          struct sk_buff *skb)
 931 {
 932         struct sock *sk = (struct sock *)handle;
 933
 934         sock_hold(sk);
 935         process_cpl_msg(chtls_pass_open_arp_failure, sk, skb);
 936         sock_put(sk);
 937 }
 938
 939 static unsigned int chtls_select_mss(const struct chtls_sock *csk,
 940                                      unsigned int pmtu,
 941                                      struct cpl_pass_accept_req *req)
 942 {
 943         struct chtls_dev *cdev;
 944         struct dst_entry *dst;
 945         unsigned int tcpoptsz;
 946         unsigned int iphdrsz;
 947         unsigned int mtu_idx;
 948         struct tcp_sock *tp;
 949         unsigned int mss;
 950         struct sock *sk;
 951
 952         mss = ntohs(req->tcpopt.mss);
 953         sk = csk->sk;
 954         dst = __sk_dst_get(sk);
 955         cdev = csk->cdev;
 956         tp = tcp_sk(sk);
 957         tcpoptsz = 0;
 958
 959 #if IS_ENABLED(CONFIG_IPV6)
 960         if (sk->sk_family == AF_INET6)
 961                 iphdrsz = sizeof(struct ipv6hdr) + sizeof(struct tcphdr);
 962         else
 963 #endif
 964                 iphdrsz = sizeof(struct iphdr) + sizeof(struct tcphdr);
 965         if (req->tcpopt.tstamp)
 966                 tcpoptsz += round_up(TCPOLEN_TIMESTAMP, 4);
 967
 968         tp->advmss = dst_metric_advmss(dst);
 969         if (USER_MSS(tp) && tp->advmss > USER_MSS(tp))
 970                 tp->advmss = USER_MSS(tp);
 971         if (tp->advmss > pmtu - iphdrsz)
 972                 tp->advmss = pmtu - iphdrsz;
 973         if (mss && tp->advmss > mss)
 974                 tp->advmss = mss;
 975
 976         tp->advmss = cxgb4_best_aligned_mtu(cdev->lldi->mtus,
 977                                             iphdrsz + tcpoptsz,
 978                                             tp->advmss - tcpoptsz,
 979                                             8, &mtu_idx);
 980         tp->advmss -= iphdrsz;
 981
 982         inet_csk(sk)->icsk_pmtu_cookie = pmtu;
 983         return mtu_idx;
 984 }
 985
 986 static unsigned int select_rcv_wscale(int space, int wscale_ok, int win_clamp)
 987 {
 988         int wscale = 0;
 989
 990         if (space > MAX_RCV_WND)
 991                 space = MAX_RCV_WND;
 992         if (win_clamp && win_clamp < space)
 993                 space = win_clamp;
 994
 995         if (wscale_ok) {
 996                 while (wscale < 14 && (65535 << wscale) < space)
 997                         wscale++;
 998         }
 999         return wscale;
1000 }
1001
1002 static void chtls_pass_accept_rpl(struct sk_buff *skb,
1003                                   struct cpl_pass_accept_req *req,
1004                                   unsigned int tid)
1005
1006 {
1007         struct cpl_t5_pass_accept_rpl *rpl5;
1008         struct cxgb4_lld_info *lldi;
1009         const struct tcphdr *tcph;
1010         const struct tcp_sock *tp;
1011         struct chtls_sock *csk;
1012         unsigned int len;
1013         struct sock *sk;
1014         u32 opt2, hlen;
1015         u64 opt0;
1016
1017         sk = skb->sk;
1018         tp = tcp_sk(sk);
1019         csk = sk->sk_user_data;
1020         csk->tid = tid;
1021         lldi = csk->cdev->lldi;
1022         len = roundup(sizeof(*rpl5), 16);
1023
1024         rpl5 = __skb_put_zero(skb, len);
1025         INIT_TP_WR(rpl5, tid);
1026
1027         OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
1028                                                      csk->tid));
1029         csk->mtu_idx = chtls_select_mss(csk, dst_mtu(__sk_dst_get(sk)),
1030                                         req);
1031         opt0 = TCAM_BYPASS_F |
1032                WND_SCALE_V(RCV_WSCALE(tp)) |
1033                MSS_IDX_V(csk->mtu_idx) |
1034                L2T_IDX_V(csk->l2t_entry->idx) |
1035                NAGLE_V(!(tp->nonagle & TCP_NAGLE_OFF)) |
1036                TX_CHAN_V(csk->tx_chan) |
1037                SMAC_SEL_V(csk->smac_idx) |
1038                DSCP_V(csk->tos >> 2) |
1039                ULP_MODE_V(ULP_MODE_TLS) |
1040                RCV_BUFSIZ_V(min(tp->rcv_wnd >> 10, RCV_BUFSIZ_M));
1041
1042         opt2 = RX_CHANNEL_V(0) |
1043                 RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
1044
1045         if (!is_t5(lldi->adapter_type))
1046                 opt2 |= RX_FC_DISABLE_F;
1047         if (req->tcpopt.tstamp)
1048                 opt2 |= TSTAMPS_EN_F;
1049         if (req->tcpopt.sack)
1050                 opt2 |= SACK_EN_F;
1051         hlen = ntohl(req->hdr_len);
1052
1053         tcph = (struct tcphdr *)((u8 *)(req + 1) +
1054                         T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
1055         if (tcph->ece && tcph->cwr)
1056                 opt2 |= CCTRL_ECN_V(1);
1057         opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
1058         opt2 |= T5_ISS_F;
1059         opt2 |= T5_OPT_2_VALID_F;
1060         opt2 |= WND_SCALE_EN_V(WSCALE_OK(tp));
1061         rpl5->opt0 = cpu_to_be64(opt0);
1062         rpl5->opt2 = cpu_to_be32(opt2);
1063         rpl5->iss = cpu_to_be32((prandom_u32() & ~7UL) - 1);
1064         set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
1065         t4_set_arp_err_handler(skb, sk, chtls_accept_rpl_arp_failure);
1066         cxgb4_l2t_send(csk->egress_dev, skb, csk->l2t_entry);
1067 }
1068
1069 static void inet_inherit_port(struct inet_hashinfo *hash_info,
1070                               struct sock *lsk, struct sock *newsk)
1071 {
1072         local_bh_disable();
1073         __inet_inherit_port(lsk, newsk);
1074         local_bh_enable();
1075 }
1076
1077 static int chtls_backlog_rcv(struct sock *sk, struct sk_buff *skb)
1078 {
1079         if (skb->protocol) {
1080                 kfree_skb(skb);
1081                 return 0;
1082         }
1083         BLOG_SKB_CB(skb)->backlog_rcv(sk, skb);
1084         return 0;
1085 }
1086
1087 static void chtls_set_tcp_window(struct chtls_sock *csk)
1088 {
1089         struct net_device *ndev = csk->egress_dev;
1090         struct port_info *pi = netdev_priv(ndev);
1091         unsigned int linkspeed;
1092         u8 scale;
1093
1094         linkspeed = pi->link_cfg.speed;
1095         scale = linkspeed / SPEED_10000;
1096 #define CHTLS_10G_RCVWIN (256 * 1024)
1097         csk->rcv_win = CHTLS_10G_RCVWIN;
1098         if (scale)
1099                 csk->rcv_win *= scale;
1100 #define CHTLS_10G_SNDWIN (256 * 1024)
1101         csk->snd_win = CHTLS_10G_SNDWIN;
1102         if (scale)
1103                 csk->snd_win *= scale;
1104 }
1105
1106 static struct sock *chtls_recv_sock(struct sock *lsk,
1107                                     struct request_sock *oreq,
1108                                     void *network_hdr,
1109                                     const struct cpl_pass_accept_req *req,
1110                                     struct chtls_dev *cdev)
1111 {
1112         struct adapter *adap = pci_get_drvdata(cdev->pdev);
1113         struct neighbour *n = NULL;
1114         struct inet_sock *newinet;
1115         const struct iphdr *iph;
1116         struct tls_context *ctx;
1117         struct net_device *ndev;
1118         struct chtls_sock *csk;
1119         struct dst_entry *dst;
1120         struct tcp_sock *tp;
1121         struct sock *newsk;
1122         bool found = false;
1123         u16 port_id;
1124         int rxq_idx;
1125         int step, i;
1126
1127         iph = (const struct iphdr *)network_hdr;
1128         newsk = tcp_create_openreq_child(lsk, oreq, cdev->askb);
1129         if (!newsk)
1130                 goto free_oreq;
1131
1132         if (lsk->sk_family == AF_INET) {
1133                 dst = inet_csk_route_child_sock(lsk, newsk, oreq);
1134                 if (!dst)
1135                         goto free_sk;
1136
1137                 n = dst_neigh_lookup(dst, &iph->saddr);
1138 #if IS_ENABLED(CONFIG_IPV6)
1139         } else {
1140                 const struct ipv6hdr *ip6h;
1141                 struct flowi6 fl6;
1142
1143                 ip6h = (const struct ipv6hdr *)network_hdr;
1144                 memset(&fl6, 0, sizeof(fl6));
1145                 fl6.flowi6_proto = IPPROTO_TCP;
1146                 fl6.saddr = ip6h->daddr;
1147                 fl6.daddr = ip6h->saddr;
1148                 fl6.fl6_dport = inet_rsk(oreq)->ir_rmt_port;
1149                 fl6.fl6_sport = htons(inet_rsk(oreq)->ir_num);
1150                 security_req_classify_flow(oreq, flowi6_to_flowi_common(&fl6));
1151                 dst = ip6_dst_lookup_flow(sock_net(lsk), lsk, &fl6, NULL);
1152                 if (IS_ERR(dst))
1153                         goto free_sk;
1154                 n = dst_neigh_lookup(dst, &ip6h->saddr);
1155 #endif
1156         }
1157         if (!n || !n->dev)
1158                 goto free_sk;
1159
1160         ndev = n->dev;
1161         if (!ndev)
1162                 goto free_dst;
1163         if (is_vlan_dev(ndev))
1164                 ndev = vlan_dev_real_dev(ndev);
1165
1166         for_each_port(adap, i)
1167                 if (cdev->ports[i] == ndev)
1168                         found = true;
1169
1170         if (!found)
1171                 goto free_dst;
1172
1173         port_id = cxgb4_port_idx(ndev);
1174
1175         csk = chtls_sock_create(cdev);
1176         if (!csk)
1177                 goto free_dst;
1178
1179         csk->l2t_entry = cxgb4_l2t_get(cdev->lldi->l2t, n, ndev, 0);
1180         if (!csk->l2t_entry)
1181                 goto free_csk;
1182
1183         newsk->sk_user_data = csk;
1184         newsk->sk_backlog_rcv = chtls_backlog_rcv;
1185
1186         tp = tcp_sk(newsk);
1187         newinet = inet_sk(newsk);
1188
1189         if (iph->version == 0x4) {
1190                 newinet->inet_daddr = iph->saddr;
1191                 newinet->inet_rcv_saddr = iph->daddr;
1192                 newinet->inet_saddr = iph->daddr;
1193 #if IS_ENABLED(CONFIG_IPV6)
1194         } else {
1195                 struct tcp6_sock *newtcp6sk = (struct tcp6_sock *)newsk;
1196                 struct inet_request_sock *treq = inet_rsk(oreq);
1197                 struct ipv6_pinfo *newnp = inet6_sk(newsk);
1198                 struct ipv6_pinfo *np = inet6_sk(lsk);
1199
1200                 inet_sk(newsk)->pinet6 = &newtcp6sk->inet6;
1201                 memcpy(newnp, np, sizeof(struct ipv6_pinfo));
1202                 newsk->sk_v6_daddr = treq->ir_v6_rmt_addr;
1203                 newsk->sk_v6_rcv_saddr = treq->ir_v6_loc_addr;
1204                 inet6_sk(newsk)->saddr = treq->ir_v6_loc_addr;
1205                 newnp->ipv6_fl_list = NULL;
1206                 newnp->pktoptions = NULL;
1207                 newsk->sk_bound_dev_if = treq->ir_iif;
1208                 newinet->inet_opt = NULL;
1209                 newinet->inet_daddr = LOOPBACK4_IPV6;
1210                 newinet->inet_saddr = LOOPBACK4_IPV6;
1211 #endif
1212         }
1213
1214         oreq->ts_recent = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1215         sk_setup_caps(newsk, dst);
1216         ctx = tls_get_ctx(lsk);
1217         newsk->sk_destruct = ctx->sk_destruct;
1218         newsk->sk_prot_creator = lsk->sk_prot_creator;
1219         csk->sk = newsk;
1220         csk->passive_reap_next = oreq;
1221         csk->tx_chan = cxgb4_port_chan(ndev);
1222         csk->port_id = port_id;
1223         csk->egress_dev = ndev;
1224         csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1225         chtls_set_tcp_window(csk);
1226         tp->rcv_wnd = csk->rcv_win;
1227         csk->sndbuf = csk->snd_win;
1228         csk->ulp_mode = ULP_MODE_TLS;
1229         step = cdev->lldi->nrxq / cdev->lldi->nchan;
1230         rxq_idx = port_id * step;
1231         rxq_idx += cdev->round_robin_cnt++ % step;
1232         csk->rss_qid = cdev->lldi->rxq_ids[rxq_idx];
1233         csk->txq_idx = (rxq_idx < cdev->lldi->ntxq) ? rxq_idx :
1234                         port_id * step;
1235         csk->sndbuf = newsk->sk_sndbuf;
1236         csk->smac_idx = ((struct port_info *)netdev_priv(ndev))->smt_idx;
1237         RCV_WSCALE(tp) = select_rcv_wscale(tcp_full_space(newsk),
1238                                            sock_net(newsk)->
1239                                                 ipv4.sysctl_tcp_window_scaling,
1240                                            tp->window_clamp);
1241         neigh_release(n);
1242         inet_inherit_port(&tcp_hashinfo, lsk, newsk);
1243         csk_set_flag(csk, CSK_CONN_INLINE);
1244         bh_unlock_sock(newsk); /* tcp_create_openreq_child ->sk_clone_lock */
1245
1246         return newsk;
1247 free_csk:
1248         chtls_sock_release(&csk->kref);
1249 free_dst:
1250         neigh_release(n);
1251         dst_release(dst);
1252 free_sk:
1253         inet_csk_prepare_forced_close(newsk);
1254         tcp_done(newsk);
1255 free_oreq:
1256         chtls_reqsk_free(oreq);
1257         return NULL;
1258 }
1259
1260 /*
1261  * Populate a TID_RELEASE WR.  The skb must be already propely sized.
1262  */
1263 static  void mk_tid_release(struct sk_buff *skb,
1264                             unsigned int chan, unsigned int tid)
1265 {
1266         struct cpl_tid_release *req;
1267         unsigned int len;
1268
1269         len = roundup(sizeof(struct cpl_tid_release), 16);
1270         req = (struct cpl_tid_release *)__skb_put(skb, len);
1271         memset(req, 0, len);
1272         set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1273         INIT_TP_WR_CPL(req, CPL_TID_RELEASE, tid);
1274 }
1275
1276 static int chtls_get_module(struct sock *sk)
1277 {
1278         struct inet_connection_sock *icsk = inet_csk(sk);
1279
1280         if (!try_module_get(icsk->icsk_ulp_ops->owner))
1281                 return -1;
1282
1283         return 0;
1284 }
1285
1286 static void chtls_pass_accept_request(struct sock *sk,
1287                                       struct sk_buff *skb)
1288 {
1289         struct cpl_t5_pass_accept_rpl *rpl;
1290         struct cpl_pass_accept_req *req;
1291         struct listen_ctx *listen_ctx;
1292         struct vlan_ethhdr *vlan_eh;
1293         struct request_sock *oreq;
1294         struct sk_buff *reply_skb;
1295         struct chtls_sock *csk;
1296         struct chtls_dev *cdev;
1297         struct ipv6hdr *ip6h;
1298         struct tcphdr *tcph;
1299         struct sock *newsk;
1300         struct ethhdr *eh;
1301         struct iphdr *iph;
1302         void *network_hdr;
1303         unsigned int stid;
1304         unsigned int len;
1305         unsigned int tid;
1306         bool th_ecn, ect;
1307         __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
1308         u16 eth_hdr_len;
1309         bool ecn_ok;
1310
1311         req = cplhdr(skb) + RSS_HDR;
1312         tid = GET_TID(req);
1313         cdev = BLOG_SKB_CB(skb)->cdev;
1314         newsk = lookup_tid(cdev->tids, tid);
1315         stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1316         if (newsk) {
1317                 pr_info("tid (%d) already in use\n", tid);
1318                 return;
1319         }
1320
1321         len = roundup(sizeof(*rpl), 16);
1322         reply_skb = alloc_skb(len, GFP_ATOMIC);
1323         if (!reply_skb) {
1324                 cxgb4_remove_tid(cdev->tids, 0, tid, sk->sk_family);
1325                 kfree_skb(skb);
1326                 return;
1327         }
1328
1329         if (sk->sk_state != TCP_LISTEN)
1330                 goto reject;
1331
1332         if (inet_csk_reqsk_queue_is_full(sk))
1333                 goto reject;
1334
1335         if (sk_acceptq_is_full(sk))
1336                 goto reject;
1337
1338
1339         eth_hdr_len = T6_ETH_HDR_LEN_G(ntohl(req->hdr_len));
1340         if (eth_hdr_len == ETH_HLEN) {
1341                 eh = (struct ethhdr *)(req + 1);
1342                 iph = (struct iphdr *)(eh + 1);
1343                 ip6h = (struct ipv6hdr *)(eh + 1);
1344                 network_hdr = (void *)(eh + 1);
1345         } else {
1346                 vlan_eh = (struct vlan_ethhdr *)(req + 1);
1347                 iph = (struct iphdr *)(vlan_eh + 1);
1348                 ip6h = (struct ipv6hdr *)(vlan_eh + 1);
1349                 network_hdr = (void *)(vlan_eh + 1);
1350         }
1351
1352         if (iph->version == 0x4) {
1353                 tcph = (struct tcphdr *)(iph + 1);
1354                 skb_set_network_header(skb, (void *)iph - (void *)req);
1355                 oreq = inet_reqsk_alloc(&chtls_rsk_ops, sk, true);
1356         } else {
1357                 tcph = (struct tcphdr *)(ip6h + 1);
1358                 skb_set_network_header(skb, (void *)ip6h - (void *)req);
1359                 oreq = inet_reqsk_alloc(&chtls_rsk_opsv6, sk, false);
1360         }
1361
1362         if (!oreq)
1363                 goto reject;
1364
1365         oreq->rsk_rcv_wnd = 0;
1366         oreq->rsk_window_clamp = 0;
1367         oreq->syncookie = 0;
1368         oreq->mss = 0;
1369         oreq->ts_recent = 0;
1370
1371         tcp_rsk(oreq)->tfo_listener = false;
1372         tcp_rsk(oreq)->rcv_isn = ntohl(tcph->seq);
1373         chtls_set_req_port(oreq, tcph->source, tcph->dest);
1374         if (iph->version == 0x4) {
1375                 chtls_set_req_addr(oreq, iph->daddr, iph->saddr);
1376                 ip_dsfield = ipv4_get_dsfield(iph);
1377 #if IS_ENABLED(CONFIG_IPV6)
1378         } else {
1379                 inet_rsk(oreq)->ir_v6_rmt_addr = ipv6_hdr(skb)->saddr;
1380                 inet_rsk(oreq)->ir_v6_loc_addr = ipv6_hdr(skb)->daddr;
1381                 ip_dsfield = ipv6_get_dsfield(ipv6_hdr(skb));
1382 #endif
1383         }
1384         if (req->tcpopt.wsf <= 14 &&
1385             sock_net(sk)->ipv4.sysctl_tcp_window_scaling) {
1386                 inet_rsk(oreq)->wscale_ok = 1;
1387                 inet_rsk(oreq)->snd_wscale = req->tcpopt.wsf;
1388         }
1389         inet_rsk(oreq)->ir_iif = sk->sk_bound_dev_if;
1390         th_ecn = tcph->ece && tcph->cwr;
1391         if (th_ecn) {
1392                 ect = !INET_ECN_is_not_ect(ip_dsfield);
1393                 ecn_ok = sock_net(sk)->ipv4.sysctl_tcp_ecn;
1394                 if ((!ect && ecn_ok) || tcp_ca_needs_ecn(sk))
1395                         inet_rsk(oreq)->ecn_ok = 1;
1396         }
1397
1398         newsk = chtls_recv_sock(sk, oreq, network_hdr, req, cdev);
1399         if (!newsk)
1400                 goto reject;
1401
1402         if (chtls_get_module(newsk))
1403                 goto reject;
1404         inet_csk_reqsk_queue_added(sk);
1405         reply_skb->sk = newsk;
1406         chtls_install_cpl_ops(newsk);
1407         cxgb4_insert_tid(cdev->tids, newsk, tid, newsk->sk_family);
1408         csk = rcu_dereference_sk_user_data(newsk);
1409         listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
1410         csk->listen_ctx = listen_ctx;
1411         __skb_queue_tail(&listen_ctx->synq, (struct sk_buff *)&csk->synq);
1412         chtls_pass_accept_rpl(reply_skb, req, tid);
1413         kfree_skb(skb);
1414         return;
1415
1416 reject:
1417         mk_tid_release(reply_skb, 0, tid);
1418         cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1419         kfree_skb(skb);
1420 }
1421
1422 /*
1423  * Handle a CPL_PASS_ACCEPT_REQ message.
1424  */
1425 static int chtls_pass_accept_req(struct chtls_dev *cdev, struct sk_buff *skb)
1426 {
1427         struct cpl_pass_accept_req *req = cplhdr(skb) + RSS_HDR;
1428         struct listen_ctx *ctx;
1429         unsigned int stid;
1430         unsigned int tid;
1431         struct sock *lsk;
1432         void *data;
1433
1434         stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1435         tid = GET_TID(req);
1436
1437         data = lookup_stid(cdev->tids, stid);
1438         if (!data)
1439                 return 1;
1440
1441         ctx = (struct listen_ctx *)data;
1442         lsk = ctx->lsk;
1443
1444         if (unlikely(tid_out_of_range(cdev->tids, tid))) {
1445                 pr_info("passive open TID %u too large\n", tid);
1446                 return 1;
1447         }
1448
1449         BLOG_SKB_CB(skb)->cdev = cdev;
1450         process_cpl_msg(chtls_pass_accept_request, lsk, skb);
1451         return 0;
1452 }
1453
1454 /*
1455  * Completes some final bits of initialization for just established connections
1456  * and changes their state to TCP_ESTABLISHED.
1457  *
1458  * snd_isn here is the ISN after the SYN, i.e., the true ISN + 1.
1459  */
1460 static void make_established(struct sock *sk, u32 snd_isn, unsigned int opt)
1461 {
1462         struct tcp_sock *tp = tcp_sk(sk);
1463
1464         tp->pushed_seq = snd_isn;
1465         tp->write_seq = snd_isn;
1466         tp->snd_nxt = snd_isn;
1467         tp->snd_una = snd_isn;
1468         inet_sk(sk)->inet_id = prandom_u32();
1469         assign_rxopt(sk, opt);
1470
1471         if (tp->rcv_wnd > (RCV_BUFSIZ_M << 10))
1472                 tp->rcv_wup -= tp->rcv_wnd - (RCV_BUFSIZ_M << 10);
1473
1474         smp_mb();
1475         tcp_set_state(sk, TCP_ESTABLISHED);
1476 }
1477
1478 static void chtls_abort_conn(struct sock *sk, struct sk_buff *skb)
1479 {
1480         struct sk_buff *abort_skb;
1481
1482         abort_skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
1483         if (abort_skb)
1484                 chtls_send_reset(sk, CPL_ABORT_SEND_RST, abort_skb);
1485 }
1486
1487 static struct sock *reap_list;
1488 static DEFINE_SPINLOCK(reap_list_lock);
1489
1490 /*
1491  * Process the reap list.
1492  */
1493 DECLARE_TASK_FUNC(process_reap_list, task_param)
1494 {
1495         spin_lock_bh(&reap_list_lock);
1496         while (reap_list) {
1497                 struct sock *sk = reap_list;
1498                 struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1499
1500                 reap_list = csk->passive_reap_next;
1501                 csk->passive_reap_next = NULL;
1502                 spin_unlock(&reap_list_lock);
1503                 sock_hold(sk);
1504
1505                 bh_lock_sock(sk);
1506                 chtls_abort_conn(sk, NULL);
1507                 sock_orphan(sk);
1508                 if (sk->sk_state == TCP_CLOSE)
1509                         inet_csk_destroy_sock(sk);
1510                 bh_unlock_sock(sk);
1511                 sock_put(sk);
1512                 spin_lock(&reap_list_lock);
1513         }
1514         spin_unlock_bh(&reap_list_lock);
1515 }
1516
1517 static DECLARE_WORK(reap_task, process_reap_list);
1518
1519 static void add_to_reap_list(struct sock *sk)
1520 {
1521         struct chtls_sock *csk = sk->sk_user_data;
1522
1523         local_bh_disable();
1524         release_tcp_port(sk); /* release the port immediately */
1525
1526         spin_lock(&reap_list_lock);
1527         csk->passive_reap_next = reap_list;
1528         reap_list = sk;
1529         if (!csk->passive_reap_next)
1530                 schedule_work(&reap_task);
1531         spin_unlock(&reap_list_lock);
1532         local_bh_enable();
1533 }
1534
1535 static void add_pass_open_to_parent(struct sock *child, struct sock *lsk,
1536                                     struct chtls_dev *cdev)
1537 {
1538         struct request_sock *oreq;
1539         struct chtls_sock *csk;
1540
1541         if (lsk->sk_state != TCP_LISTEN)
1542                 return;
1543
1544         csk = child->sk_user_data;
1545         oreq = csk->passive_reap_next;
1546         csk->passive_reap_next = NULL;
1547
1548         reqsk_queue_removed(&inet_csk(lsk)->icsk_accept_queue, oreq);
1549         __skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
1550
1551         if (sk_acceptq_is_full(lsk)) {
1552                 chtls_reqsk_free(oreq);
1553                 add_to_reap_list(child);
1554         } else {
1555                 refcount_set(&oreq->rsk_refcnt, 1);
1556                 inet_csk_reqsk_queue_add(lsk, oreq, child);
1557                 lsk->sk_data_ready(lsk);
1558         }
1559 }
1560
1561 static void bl_add_pass_open_to_parent(struct sock *lsk, struct sk_buff *skb)
1562 {
1563         struct sock *child = skb->sk;
1564
1565         skb->sk = NULL;
1566         add_pass_open_to_parent(child, lsk, BLOG_SKB_CB(skb)->cdev);
1567         kfree_skb(skb);
1568 }
1569
1570 static int chtls_pass_establish(struct chtls_dev *cdev, struct sk_buff *skb)
1571 {
1572         struct cpl_pass_establish *req = cplhdr(skb) + RSS_HDR;
1573         struct chtls_sock *csk;
1574         struct sock *lsk, *sk;
1575         unsigned int hwtid;
1576
1577         hwtid = GET_TID(req);
1578         sk = lookup_tid(cdev->tids, hwtid);
1579         if (!sk)
1580                 return (CPL_RET_UNKNOWN_TID | CPL_RET_BUF_DONE);
1581
1582         bh_lock_sock(sk);
1583         if (unlikely(sock_owned_by_user(sk))) {
1584                 kfree_skb(skb);
1585         } else {
1586                 unsigned int stid;
1587                 void *data;
1588
1589                 csk = sk->sk_user_data;
1590                 csk->wr_max_credits = 64;
1591                 csk->wr_credits = 64;
1592                 csk->wr_unacked = 0;
1593                 make_established(sk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
1594                 stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1595                 sk->sk_state_change(sk);
1596                 if (unlikely(sk->sk_socket))
1597                         sk_wake_async(sk, 0, POLL_OUT);
1598
1599                 data = lookup_stid(cdev->tids, stid);
1600                 if (!data) {
1601                         /* listening server close */
1602                         kfree_skb(skb);
1603                         goto unlock;
1604                 }
1605                 lsk = ((struct listen_ctx *)data)->lsk;
1606
1607                 bh_lock_sock(lsk);
1608                 if (unlikely(skb_queue_empty(&csk->listen_ctx->synq))) {
1609                         /* removed from synq */
1610                         bh_unlock_sock(lsk);
1611                         kfree_skb(skb);
1612                         goto unlock;
1613                 }
1614
1615                 if (likely(!sock_owned_by_user(lsk))) {
1616                         kfree_skb(skb);
1617                         add_pass_open_to_parent(sk, lsk, cdev);
1618                 } else {
1619                         skb->sk = sk;
1620                         BLOG_SKB_CB(skb)->cdev = cdev;
1621                         BLOG_SKB_CB(skb)->backlog_rcv =
1622                                 bl_add_pass_open_to_parent;
1623                         __sk_add_backlog(lsk, skb);
1624                 }
1625                 bh_unlock_sock(lsk);
1626         }
1627 unlock:
1628         bh_unlock_sock(sk);
1629         return 0;
1630 }
1631
1632 /*
1633  * Handle receipt of an urgent pointer.
1634  */
1635 static void handle_urg_ptr(struct sock *sk, u32 urg_seq)
1636 {
1637         struct tcp_sock *tp = tcp_sk(sk);
1638
1639         urg_seq--;
1640         if (tp->urg_data && !after(urg_seq, tp->urg_seq))
1641                 return; /* duplicate pointer */
1642
1643         sk_send_sigurg(sk);
1644         if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
1645             !sock_flag(sk, SOCK_URGINLINE) &&
1646             tp->copied_seq != tp->rcv_nxt) {
1647                 struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1648
1649                 tp->copied_seq++;
1650                 if (skb && tp->copied_seq - ULP_SKB_CB(skb)->seq >= skb->len)
1651                         chtls_free_skb(sk, skb);
1652         }
1653
1654         tp->urg_data = TCP_URG_NOTYET;
1655         tp->urg_seq = urg_seq;
1656 }
1657
1658 static void check_sk_callbacks(struct chtls_sock *csk)
1659 {
1660         struct sock *sk = csk->sk;
1661
1662         if (unlikely(sk->sk_user_data &&
1663                      !csk_flag_nochk(csk, CSK_CALLBACKS_CHKD)))
1664                 csk_set_flag(csk, CSK_CALLBACKS_CHKD);
1665 }
1666
1667 /*
1668  * Handles Rx data that arrives in a state where the socket isn't accepting
1669  * new data.
1670  */
1671 static void handle_excess_rx(struct sock *sk, struct sk_buff *skb)
1672 {
1673         if (!csk_flag(sk, CSK_ABORT_SHUTDOWN))
1674                 chtls_abort_conn(sk, skb);
1675
1676         kfree_skb(skb);
1677 }
1678
1679 static void chtls_recv_data(struct sock *sk, struct sk_buff *skb)
1680 {
1681         struct cpl_rx_data *hdr = cplhdr(skb) + RSS_HDR;
1682         struct chtls_sock *csk;
1683         struct tcp_sock *tp;
1684
1685         csk = rcu_dereference_sk_user_data(sk);
1686         tp = tcp_sk(sk);
1687
1688         if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1689                 handle_excess_rx(sk, skb);
1690                 return;
1691         }
1692
1693         ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1694         ULP_SKB_CB(skb)->psh = hdr->psh;
1695         skb_ulp_mode(skb) = ULP_MODE_NONE;
1696
1697         skb_reset_transport_header(skb);
1698         __skb_pull(skb, sizeof(*hdr) + RSS_HDR);
1699         if (!skb->data_len)
1700                 __skb_trim(skb, ntohs(hdr->len));
1701
1702         if (unlikely(hdr->urg))
1703                 handle_urg_ptr(sk, tp->rcv_nxt + ntohs(hdr->urg));
1704         if (unlikely(tp->urg_data == TCP_URG_NOTYET &&
1705                      tp->urg_seq - tp->rcv_nxt < skb->len))
1706                 tp->urg_data = TCP_URG_VALID |
1707                                skb->data[tp->urg_seq - tp->rcv_nxt];
1708
1709         if (unlikely(hdr->dack_mode != csk->delack_mode)) {
1710                 csk->delack_mode = hdr->dack_mode;
1711                 csk->delack_seq = tp->rcv_nxt;
1712         }
1713
1714         tcp_hdr(skb)->fin = 0;
1715         tp->rcv_nxt += skb->len;
1716
1717         __skb_queue_tail(&sk->sk_receive_queue, skb);
1718
1719         if (!sock_flag(sk, SOCK_DEAD)) {
1720                 check_sk_callbacks(csk);
1721                 sk->sk_data_ready(sk);
1722         }
1723 }
1724
1725 static int chtls_rx_data(struct chtls_dev *cdev, struct sk_buff *skb)
1726 {
1727         struct cpl_rx_data *req = cplhdr(skb) + RSS_HDR;
1728         unsigned int hwtid = GET_TID(req);
1729         struct sock *sk;
1730
1731         sk = lookup_tid(cdev->tids, hwtid);
1732         if (unlikely(!sk)) {
1733                 pr_err("can't find conn. for hwtid %u.\n", hwtid);
1734                 return -EINVAL;
1735         }
1736         skb_dst_set(skb, NULL);
1737         process_cpl_msg(chtls_recv_data, sk, skb);
1738         return 0;
1739 }
1740
1741 static void chtls_recv_pdu(struct sock *sk, struct sk_buff *skb)
1742 {
1743         struct cpl_tls_data *hdr = cplhdr(skb);
1744         struct chtls_sock *csk;
1745         struct chtls_hws *tlsk;
1746         struct tcp_sock *tp;
1747
1748         csk = rcu_dereference_sk_user_data(sk);
1749         tlsk = &csk->tlshws;
1750         tp = tcp_sk(sk);
1751
1752         if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1753                 handle_excess_rx(sk, skb);
1754                 return;
1755         }
1756
1757         ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1758         ULP_SKB_CB(skb)->flags = 0;
1759         skb_ulp_mode(skb) = ULP_MODE_TLS;
1760
1761         skb_reset_transport_header(skb);
1762         __skb_pull(skb, sizeof(*hdr));
1763         if (!skb->data_len)
1764                 __skb_trim(skb,
1765                            CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd)));
1766
1767         if (unlikely(tp->urg_data == TCP_URG_NOTYET && tp->urg_seq -
1768                      tp->rcv_nxt < skb->len))
1769                 tp->urg_data = TCP_URG_VALID |
1770                                skb->data[tp->urg_seq - tp->rcv_nxt];
1771
1772         tcp_hdr(skb)->fin = 0;
1773         tlsk->pldlen = CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd));
1774         __skb_queue_tail(&tlsk->sk_recv_queue, skb);
1775 }
1776
1777 static int chtls_rx_pdu(struct chtls_dev *cdev, struct sk_buff *skb)
1778 {
1779         struct cpl_tls_data *req = cplhdr(skb);
1780         unsigned int hwtid = GET_TID(req);
1781         struct sock *sk;
1782
1783         sk = lookup_tid(cdev->tids, hwtid);
1784         if (unlikely(!sk)) {
1785                 pr_err("can't find conn. for hwtid %u.\n", hwtid);
1786                 return -EINVAL;
1787         }
1788         skb_dst_set(skb, NULL);
1789         process_cpl_msg(chtls_recv_pdu, sk, skb);
1790         return 0;
1791 }
1792
1793 static void chtls_set_hdrlen(struct sk_buff *skb, unsigned int nlen)
1794 {
1795         struct tlsrx_cmp_hdr *tls_cmp_hdr = cplhdr(skb);
1796
1797         skb->hdr_len = ntohs((__force __be16)tls_cmp_hdr->length);
1798         tls_cmp_hdr->length = ntohs((__force __be16)nlen);
1799 }
1800
1801 static void chtls_rx_hdr(struct sock *sk, struct sk_buff *skb)
1802 {
1803         struct tlsrx_cmp_hdr *tls_hdr_pkt;
1804         struct cpl_rx_tls_cmp *cmp_cpl;
1805         struct sk_buff *skb_rec;
1806         struct chtls_sock *csk;
1807         struct chtls_hws *tlsk;
1808         struct tcp_sock *tp;
1809
1810         cmp_cpl = cplhdr(skb);
1811         csk = rcu_dereference_sk_user_data(sk);
1812         tlsk = &csk->tlshws;
1813         tp = tcp_sk(sk);
1814
1815         ULP_SKB_CB(skb)->seq = ntohl(cmp_cpl->seq);
1816         ULP_SKB_CB(skb)->flags = 0;
1817
1818         skb_reset_transport_header(skb);
1819         __skb_pull(skb, sizeof(*cmp_cpl));
1820         tls_hdr_pkt = (struct tlsrx_cmp_hdr *)skb->data;
1821         if (tls_hdr_pkt->res_to_mac_error & TLSRX_HDR_PKT_ERROR_M)
1822                 tls_hdr_pkt->type = CONTENT_TYPE_ERROR;
1823         if (!skb->data_len)
1824                 __skb_trim(skb, TLS_HEADER_LENGTH);
1825
1826         tp->rcv_nxt +=
1827                 CPL_RX_TLS_CMP_PDULENGTH_G(ntohl(cmp_cpl->pdulength_length));
1828
1829         ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_TLS_HDR;
1830         skb_rec = __skb_dequeue(&tlsk->sk_recv_queue);
1831         if (!skb_rec) {
1832                 __skb_queue_tail(&sk->sk_receive_queue, skb);
1833         } else {
1834                 chtls_set_hdrlen(skb, tlsk->pldlen);
1835                 tlsk->pldlen = 0;
1836                 __skb_queue_tail(&sk->sk_receive_queue, skb);
1837                 __skb_queue_tail(&sk->sk_receive_queue, skb_rec);
1838         }
1839
1840         if (!sock_flag(sk, SOCK_DEAD)) {
1841                 check_sk_callbacks(csk);
1842                 sk->sk_data_ready(sk);
1843         }
1844 }
1845
1846 static int chtls_rx_cmp(struct chtls_dev *cdev, struct sk_buff *skb)
1847 {
1848         struct cpl_rx_tls_cmp *req = cplhdr(skb);
1849         unsigned int hwtid = GET_TID(req);
1850         struct sock *sk;
1851
1852         sk = lookup_tid(cdev->tids, hwtid);
1853         if (unlikely(!sk)) {
1854                 pr_err("can't find conn. for hwtid %u.\n", hwtid);
1855                 return -EINVAL;
1856         }
1857         skb_dst_set(skb, NULL);
1858         process_cpl_msg(chtls_rx_hdr, sk, skb);
1859
1860         return 0;
1861 }
1862
1863 static void chtls_timewait(struct sock *sk)
1864 {
1865         struct tcp_sock *tp = tcp_sk(sk);
1866
1867         tp->rcv_nxt++;
1868         tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
1869         tp->srtt_us = 0;
1870         tcp_time_wait(sk, TCP_TIME_WAIT, 0);
1871 }
1872
1873 static void chtls_peer_close(struct sock *sk, struct sk_buff *skb)
1874 {
1875         struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1876
1877         if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1878                 goto out;
1879
1880         sk->sk_shutdown |= RCV_SHUTDOWN;
1881         sock_set_flag(sk, SOCK_DONE);
1882
1883         switch (sk->sk_state) {
1884         case TCP_SYN_RECV:
1885         case TCP_ESTABLISHED:
1886                 tcp_set_state(sk, TCP_CLOSE_WAIT);
1887                 break;
1888         case TCP_FIN_WAIT1:
1889                 tcp_set_state(sk, TCP_CLOSING);
1890                 break;
1891         case TCP_FIN_WAIT2:
1892                 chtls_release_resources(sk);
1893                 if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1894                         chtls_conn_done(sk);
1895                 else
1896                         chtls_timewait(sk);
1897                 break;
1898         default:
1899                 pr_info("cpl_peer_close in bad state %d\n", sk->sk_state);
1900         }
1901
1902         if (!sock_flag(sk, SOCK_DEAD)) {
1903                 sk->sk_state_change(sk);
1904                 /* Do not send POLL_HUP for half duplex close. */
1905
1906                 if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1907                     sk->sk_state == TCP_CLOSE)
1908                         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
1909                 else
1910                         sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1911         }
1912 out:
1913         kfree_skb(skb);
1914 }
1915
1916 static void chtls_close_con_rpl(struct sock *sk, struct sk_buff *skb)
1917 {
1918         struct cpl_close_con_rpl *rpl = cplhdr(skb) + RSS_HDR;
1919         struct chtls_sock *csk;
1920         struct tcp_sock *tp;
1921
1922         csk = rcu_dereference_sk_user_data(sk);
1923
1924         if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1925                 goto out;
1926
1927         tp = tcp_sk(sk);
1928
1929         tp->snd_una = ntohl(rpl->snd_nxt) - 1;  /* exclude FIN */
1930
1931         switch (sk->sk_state) {
1932         case TCP_CLOSING:
1933                 chtls_release_resources(sk);
1934                 if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1935                         chtls_conn_done(sk);
1936                 else
1937                         chtls_timewait(sk);
1938                 break;
1939         case TCP_LAST_ACK:
1940                 chtls_release_resources(sk);
1941                 chtls_conn_done(sk);
1942                 break;
1943         case TCP_FIN_WAIT1:
1944                 tcp_set_state(sk, TCP_FIN_WAIT2);
1945                 sk->sk_shutdown |= SEND_SHUTDOWN;
1946
1947                 if (!sock_flag(sk, SOCK_DEAD))
1948                         sk->sk_state_change(sk);
1949                 else if (tcp_sk(sk)->linger2 < 0 &&
1950                          !csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN))
1951                         chtls_abort_conn(sk, skb);
1952                 break;
1953         default:
1954                 pr_info("close_con_rpl in bad state %d\n", sk->sk_state);
1955         }
1956 out:
1957         kfree_skb(skb);
1958 }
1959
1960 static struct sk_buff *get_cpl_skb(struct sk_buff *skb,
1961                                    size_t len, gfp_t gfp)
1962 {
1963         if (likely(!skb_is_nonlinear(skb) && !skb_cloned(skb))) {
1964                 WARN_ONCE(skb->len < len, "skb alloc error");
1965                 __skb_trim(skb, len);
1966                 skb_get(skb);
1967         } else {
1968                 skb = alloc_skb(len, gfp);
1969                 if (skb)
1970                         __skb_put(skb, len);
1971         }
1972         return skb;
1973 }
1974
1975 static void set_abort_rpl_wr(struct sk_buff *skb, unsigned int tid,
1976                              int cmd)
1977 {
1978         struct cpl_abort_rpl *rpl = cplhdr(skb);
1979
1980         INIT_TP_WR_CPL(rpl, CPL_ABORT_RPL, tid);
1981         rpl->cmd = cmd;
1982 }
1983
1984 static void send_defer_abort_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
1985 {
1986         struct cpl_abort_req_rss *req = cplhdr(skb);
1987         struct sk_buff *reply_skb;
1988
1989         reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1990                               GFP_KERNEL | __GFP_NOFAIL);
1991         __skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
1992         set_abort_rpl_wr(reply_skb, GET_TID(req),
1993                          (req->status & CPL_ABORT_NO_RST));
1994         set_wr_txq(reply_skb, CPL_PRIORITY_DATA, req->status >> 1);
1995         cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1996         kfree_skb(skb);
1997 }
1998
1999 /*
2000  * Add an skb to the deferred skb queue for processing from process context.
2001  */
2002 static void t4_defer_reply(struct sk_buff *skb, struct chtls_dev *cdev,
2003                            defer_handler_t handler)
2004 {
2005         DEFERRED_SKB_CB(skb)->handler = handler;
2006         spin_lock_bh(&cdev->deferq.lock);
2007         __skb_queue_tail(&cdev->deferq, skb);
2008         if (skb_queue_len(&cdev->deferq) == 1)
2009                 schedule_work(&cdev->deferq_task);
2010         spin_unlock_bh(&cdev->deferq.lock);
2011 }
2012
2013 static void chtls_send_abort_rpl(struct sock *sk, struct sk_buff *skb,
2014                                  struct chtls_dev *cdev,
2015                                  int status, int queue)
2016 {
2017         struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
2018         struct sk_buff *reply_skb;
2019         struct chtls_sock *csk;
2020         unsigned int tid;
2021
2022         csk = rcu_dereference_sk_user_data(sk);
2023         tid = GET_TID(req);
2024
2025         reply_skb = get_cpl_skb(skb, sizeof(struct cpl_abort_rpl), gfp_any());
2026         if (!reply_skb) {
2027                 req->status = (queue << 1) | status;
2028                 t4_defer_reply(skb, cdev, send_defer_abort_rpl);
2029                 return;
2030         }
2031
2032         set_abort_rpl_wr(reply_skb, tid, status);
2033         kfree_skb(skb);
2034         set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
2035         if (csk_conn_inline(csk)) {
2036                 struct l2t_entry *e = csk->l2t_entry;
2037
2038                 if (e && sk->sk_state != TCP_SYN_RECV) {
2039                         cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
2040                         return;
2041                 }
2042         }
2043         cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
2044 }
2045
2046 /*
2047  * This is run from a listener's backlog to abort a child connection in
2048  * SYN_RCV state (i.e., one on the listener's SYN queue).
2049  */
2050 static void bl_abort_syn_rcv(struct sock *lsk, struct sk_buff *skb)
2051 {
2052         struct chtls_sock *csk;
2053         struct sock *child;
2054         int queue;
2055
2056         child = skb->sk;
2057         csk = rcu_dereference_sk_user_data(child);
2058         queue = csk->txq_idx;
2059
2060         skb->sk = NULL;
2061         chtls_send_abort_rpl(child, skb, BLOG_SKB_CB(skb)->cdev,
2062                              CPL_ABORT_NO_RST, queue);
2063         do_abort_syn_rcv(child, lsk);
2064 }
2065
2066 static int abort_syn_rcv(struct sock *sk, struct sk_buff *skb)
2067 {
2068         const struct request_sock *oreq;
2069         struct listen_ctx *listen_ctx;
2070         struct chtls_sock *csk;
2071         struct chtls_dev *cdev;
2072         struct sock *psk;
2073         void *ctx;
2074
2075         csk = sk->sk_user_data;
2076         oreq = csk->passive_reap_next;
2077         cdev = csk->cdev;
2078
2079         if (!oreq)
2080                 return -1;
2081
2082         ctx = lookup_stid(cdev->tids, oreq->ts_recent);
2083         if (!ctx)
2084                 return -1;
2085
2086         listen_ctx = (struct listen_ctx *)ctx;
2087         psk = listen_ctx->lsk;
2088
2089         bh_lock_sock(psk);
2090         if (!sock_owned_by_user(psk)) {
2091                 int queue = csk->txq_idx;
2092
2093                 chtls_send_abort_rpl(sk, skb, cdev, CPL_ABORT_NO_RST, queue);
2094                 do_abort_syn_rcv(sk, psk);
2095         } else {
2096                 skb->sk = sk;
2097                 BLOG_SKB_CB(skb)->backlog_rcv = bl_abort_syn_rcv;
2098                 __sk_add_backlog(psk, skb);
2099         }
2100         bh_unlock_sock(psk);
2101         return 0;
2102 }
2103
2104 static void chtls_abort_req_rss(struct sock *sk, struct sk_buff *skb)
2105 {
2106         const struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
2107         struct chtls_sock *csk = sk->sk_user_data;
2108         int rst_status = CPL_ABORT_NO_RST;
2109         int queue = csk->txq_idx;
2110
2111         if (is_neg_adv(req->status)) {
2112                 kfree_skb(skb);
2113                 return;
2114         }
2115
2116         csk_reset_flag(csk, CSK_ABORT_REQ_RCVD);
2117
2118         if (!csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) &&
2119             !csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
2120                 struct tcp_sock *tp = tcp_sk(sk);
2121
2122                 if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
2123                         WARN_ONCE(1, "send_tx_flowc error");
2124                 csk_set_flag(csk, CSK_TX_DATA_SENT);
2125         }
2126
2127         csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
2128
2129         if (!csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2130                 sk->sk_err = ETIMEDOUT;
2131
2132                 if (!sock_flag(sk, SOCK_DEAD))
2133                         sk->sk_error_report(sk);
2134
2135                 if (sk->sk_state == TCP_SYN_RECV && !abort_syn_rcv(sk, skb))
2136                         return;
2137
2138         }
2139
2140         chtls_send_abort_rpl(sk, skb, BLOG_SKB_CB(skb)->cdev,
2141                              rst_status, queue);
2142         chtls_release_resources(sk);
2143         chtls_conn_done(sk);
2144 }
2145
2146 static void chtls_abort_rpl_rss(struct sock *sk, struct sk_buff *skb)
2147 {
2148         struct cpl_abort_rpl_rss *rpl = cplhdr(skb) + RSS_HDR;
2149         struct chtls_sock *csk;
2150         struct chtls_dev *cdev;
2151
2152         csk = rcu_dereference_sk_user_data(sk);
2153         cdev = csk->cdev;
2154
2155         if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2156                 csk_reset_flag(csk, CSK_ABORT_RPL_PENDING);
2157                 if (!csk_flag_nochk(csk, CSK_ABORT_REQ_RCVD)) {
2158                         if (sk->sk_state == TCP_SYN_SENT) {
2159                                 cxgb4_remove_tid(cdev->tids,
2160                                                  csk->port_id,
2161                                                  GET_TID(rpl),
2162                                                  sk->sk_family);
2163                                 sock_put(sk);
2164                         }
2165                         chtls_release_resources(sk);
2166                         chtls_conn_done(sk);
2167                 }
2168         }
2169         kfree_skb(skb);
2170 }
2171
2172 static int chtls_conn_cpl(struct chtls_dev *cdev, struct sk_buff *skb)
2173 {
2174         struct cpl_peer_close *req = cplhdr(skb) + RSS_HDR;
2175         void (*fn)(struct sock *sk, struct sk_buff *skb);
2176         unsigned int hwtid = GET_TID(req);
2177         struct chtls_sock *csk;
2178         struct sock *sk;
2179         u8 opcode;
2180
2181         opcode = ((const struct rss_header *)cplhdr(skb))->opcode;
2182
2183         sk = lookup_tid(cdev->tids, hwtid);
2184         if (!sk)
2185                 goto rel_skb;
2186
2187         csk = sk->sk_user_data;
2188
2189         switch (opcode) {
2190         case CPL_PEER_CLOSE:
2191                 fn = chtls_peer_close;
2192                 break;
2193         case CPL_CLOSE_CON_RPL:
2194                 fn = chtls_close_con_rpl;
2195                 break;
2196         case CPL_ABORT_REQ_RSS:
2197                 /*
2198                  * Save the offload device in the skb, we may process this
2199                  * message after the socket has closed.
2200                  */
2201                 BLOG_SKB_CB(skb)->cdev = csk->cdev;
2202                 fn = chtls_abort_req_rss;
2203                 break;
2204         case CPL_ABORT_RPL_RSS:
2205                 fn = chtls_abort_rpl_rss;
2206                 break;
2207         default:
2208                 goto rel_skb;
2209         }
2210
2211         process_cpl_msg(fn, sk, skb);
2212         return 0;
2213
2214 rel_skb:
2215         kfree_skb(skb);
2216         return 0;
2217 }
2218
2219 static void chtls_rx_ack(struct sock *sk, struct sk_buff *skb)
2220 {
2221         struct cpl_fw4_ack *hdr = cplhdr(skb) + RSS_HDR;
2222         struct chtls_sock *csk = sk->sk_user_data;
2223         struct tcp_sock *tp = tcp_sk(sk);
2224         u32 credits = hdr->credits;
2225         u32 snd_una;
2226
2227         snd_una = ntohl(hdr->snd_una);
2228         csk->wr_credits += credits;
2229
2230         if (csk->wr_unacked > csk->wr_max_credits - csk->wr_credits)
2231                 csk->wr_unacked = csk->wr_max_credits - csk->wr_credits;
2232
2233         while (credits) {
2234                 struct sk_buff *pskb = csk->wr_skb_head;
2235                 u32 csum;
2236
2237                 if (unlikely(!pskb)) {
2238                         if (csk->wr_nondata)
2239                                 csk->wr_nondata -= credits;
2240                         break;
2241                 }
2242                 csum = (__force u32)pskb->csum;
2243                 if (unlikely(credits < csum)) {
2244                         pskb->csum = (__force __wsum)(csum - credits);
2245                         break;
2246                 }
2247                 dequeue_wr(sk);
2248                 credits -= csum;
2249                 kfree_skb(pskb);
2250         }
2251         if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
2252                 if (unlikely(before(snd_una, tp->snd_una))) {
2253                         kfree_skb(skb);
2254                         return;
2255                 }
2256
2257                 if (tp->snd_una != snd_una) {
2258                         tp->snd_una = snd_una;
2259                         tp->rcv_tstamp = tcp_time_stamp(tp);
2260                         if (tp->snd_una == tp->snd_nxt &&
2261                             !csk_flag_nochk(csk, CSK_TX_FAILOVER))
2262                                 csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2263                 }
2264         }
2265
2266         if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_CH) {
2267                 unsigned int fclen16 = roundup(failover_flowc_wr_len, 16);
2268
2269                 csk->wr_credits -= fclen16;
2270                 csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2271                 csk_reset_flag(csk, CSK_TX_FAILOVER);
2272         }
2273         if (skb_queue_len(&csk->txq) && chtls_push_frames(csk, 0))
2274                 sk->sk_write_space(sk);
2275
2276         kfree_skb(skb);
2277 }
2278
2279 static int chtls_wr_ack(struct chtls_dev *cdev, struct sk_buff *skb)
2280 {
2281         struct cpl_fw4_ack *rpl = cplhdr(skb) + RSS_HDR;
2282         unsigned int hwtid = GET_TID(rpl);
2283         struct sock *sk;
2284
2285         sk = lookup_tid(cdev->tids, hwtid);
2286         if (unlikely(!sk)) {
2287                 pr_err("can't find conn. for hwtid %u.\n", hwtid);
2288                 return -EINVAL;
2289         }
2290         process_cpl_msg(chtls_rx_ack, sk, skb);
2291
2292         return 0;
2293 }
2294
2295 chtls_handler_func chtls_handlers[NUM_CPL_CMDS] = {
2296         [CPL_PASS_OPEN_RPL]     = chtls_pass_open_rpl,
2297         [CPL_CLOSE_LISTSRV_RPL] = chtls_close_listsrv_rpl,
2298         [CPL_PASS_ACCEPT_REQ]   = chtls_pass_accept_req,
2299         [CPL_PASS_ESTABLISH]    = chtls_pass_establish,
2300         [CPL_RX_DATA]           = chtls_rx_data,
2301         [CPL_TLS_DATA]          = chtls_rx_pdu,
2302         [CPL_RX_TLS_CMP]        = chtls_rx_cmp,
2303         [CPL_PEER_CLOSE]        = chtls_conn_cpl,
2304         [CPL_CLOSE_CON_RPL]     = chtls_conn_cpl,
2305         [CPL_ABORT_REQ_RSS]     = chtls_conn_cpl,
2306         [CPL_ABORT_RPL_RSS]     = chtls_conn_cpl,
2307         [CPL_FW4_ACK]           = chtls_wr_ack,
2308 };