linux/drivers/crypto/chelsio/chtls/chtls_cm.c
<<
>>
Prefs
   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 <net/inet_common.h>
  22#include <net/tcp.h>
  23#include <net/dst.h>
  24#include <net/tls.h>
  25
  26#include "chtls.h"
  27#include "chtls_cm.h"
  28
  29/*
  30 * State transitions and actions for close.  Note that if we are in SYN_SENT
  31 * we remain in that state as we cannot control a connection while it's in
  32 * SYN_SENT; such connections are allowed to establish and are then aborted.
  33 */
  34static unsigned char new_state[16] = {
  35        /* current state:     new state:      action: */
  36        /* (Invalid)       */ TCP_CLOSE,
  37        /* TCP_ESTABLISHED */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  38        /* TCP_SYN_SENT    */ TCP_SYN_SENT,
  39        /* TCP_SYN_RECV    */ TCP_FIN_WAIT1 | TCP_ACTION_FIN,
  40        /* TCP_FIN_WAIT1   */ TCP_FIN_WAIT1,
  41        /* TCP_FIN_WAIT2   */ TCP_FIN_WAIT2,
  42        /* TCP_TIME_WAIT   */ TCP_CLOSE,
  43        /* TCP_CLOSE       */ TCP_CLOSE,
  44        /* TCP_CLOSE_WAIT  */ TCP_LAST_ACK | TCP_ACTION_FIN,
  45        /* TCP_LAST_ACK    */ TCP_LAST_ACK,
  46        /* TCP_LISTEN      */ TCP_CLOSE,
  47        /* TCP_CLOSING     */ TCP_CLOSING,
  48};
  49
  50static struct chtls_sock *chtls_sock_create(struct chtls_dev *cdev)
  51{
  52        struct chtls_sock *csk = kzalloc(sizeof(*csk), GFP_ATOMIC);
  53
  54        if (!csk)
  55                return NULL;
  56
  57        csk->txdata_skb_cache = alloc_skb(TXDATA_SKB_LEN, GFP_ATOMIC);
  58        if (!csk->txdata_skb_cache) {
  59                kfree(csk);
  60                return NULL;
  61        }
  62
  63        kref_init(&csk->kref);
  64        csk->cdev = cdev;
  65        skb_queue_head_init(&csk->txq);
  66        csk->wr_skb_head = NULL;
  67        csk->wr_skb_tail = NULL;
  68        csk->mss = MAX_MSS;
  69        csk->tlshws.ofld = 1;
  70        csk->tlshws.txkey = -1;
  71        csk->tlshws.rxkey = -1;
  72        csk->tlshws.mfs = TLS_MFS;
  73        skb_queue_head_init(&csk->tlshws.sk_recv_queue);
  74        return csk;
  75}
  76
  77static void chtls_sock_release(struct kref *ref)
  78{
  79        struct chtls_sock *csk =
  80                container_of(ref, struct chtls_sock, kref);
  81
  82        kfree(csk);
  83}
  84
  85static struct net_device *chtls_ipv4_netdev(struct chtls_dev *cdev,
  86                                            struct sock *sk)
  87{
  88        struct net_device *ndev = cdev->ports[0];
  89
  90        if (likely(!inet_sk(sk)->inet_rcv_saddr))
  91                return ndev;
  92
  93        ndev = ip_dev_find(&init_net, inet_sk(sk)->inet_rcv_saddr);
  94        if (!ndev)
  95                return NULL;
  96
  97        if (is_vlan_dev(ndev))
  98                return vlan_dev_real_dev(ndev);
  99        return ndev;
 100}
 101
 102static void assign_rxopt(struct sock *sk, unsigned int opt)
 103{
 104        const struct chtls_dev *cdev;
 105        struct chtls_sock *csk;
 106        struct tcp_sock *tp;
 107
 108        csk = rcu_dereference_sk_user_data(sk);
 109        tp = tcp_sk(sk);
 110
 111        cdev = csk->cdev;
 112        tp->tcp_header_len           = sizeof(struct tcphdr);
 113        tp->rx_opt.mss_clamp         = cdev->mtus[TCPOPT_MSS_G(opt)] - 40;
 114        tp->mss_cache                = tp->rx_opt.mss_clamp;
 115        tp->rx_opt.tstamp_ok         = TCPOPT_TSTAMP_G(opt);
 116        tp->rx_opt.snd_wscale        = TCPOPT_SACK_G(opt);
 117        tp->rx_opt.wscale_ok         = TCPOPT_WSCALE_OK_G(opt);
 118        SND_WSCALE(tp)               = TCPOPT_SND_WSCALE_G(opt);
 119        if (!tp->rx_opt.wscale_ok)
 120                tp->rx_opt.rcv_wscale = 0;
 121        if (tp->rx_opt.tstamp_ok) {
 122                tp->tcp_header_len += TCPOLEN_TSTAMP_ALIGNED;
 123                tp->rx_opt.mss_clamp -= TCPOLEN_TSTAMP_ALIGNED;
 124        } else if (csk->opt2 & TSTAMPS_EN_F) {
 125                csk->opt2 &= ~TSTAMPS_EN_F;
 126                csk->mtu_idx = TCPOPT_MSS_G(opt);
 127        }
 128}
 129
 130static void chtls_purge_receive_queue(struct sock *sk)
 131{
 132        struct sk_buff *skb;
 133
 134        while ((skb = __skb_dequeue(&sk->sk_receive_queue)) != NULL) {
 135                skb_dst_set(skb, (void *)NULL);
 136                kfree_skb(skb);
 137        }
 138}
 139
 140static void chtls_purge_write_queue(struct sock *sk)
 141{
 142        struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 143        struct sk_buff *skb;
 144
 145        while ((skb = __skb_dequeue(&csk->txq))) {
 146                sk->sk_wmem_queued -= skb->truesize;
 147                __kfree_skb(skb);
 148        }
 149}
 150
 151static void chtls_purge_recv_queue(struct sock *sk)
 152{
 153        struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 154        struct chtls_hws *tlsk = &csk->tlshws;
 155        struct sk_buff *skb;
 156
 157        while ((skb = __skb_dequeue(&tlsk->sk_recv_queue)) != NULL) {
 158                skb_dst_set(skb, NULL);
 159                kfree_skb(skb);
 160        }
 161}
 162
 163static void abort_arp_failure(void *handle, struct sk_buff *skb)
 164{
 165        struct cpl_abort_req *req = cplhdr(skb);
 166        struct chtls_dev *cdev;
 167
 168        cdev = (struct chtls_dev *)handle;
 169        req->cmd = CPL_ABORT_NO_RST;
 170        cxgb4_ofld_send(cdev->lldi->ports[0], skb);
 171}
 172
 173static struct sk_buff *alloc_ctrl_skb(struct sk_buff *skb, int len)
 174{
 175        if (likely(skb && !skb_shared(skb) && !skb_cloned(skb))) {
 176                __skb_trim(skb, 0);
 177                refcount_add(2, &skb->users);
 178        } else {
 179                skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
 180        }
 181        return skb;
 182}
 183
 184static void chtls_send_abort(struct sock *sk, int mode, struct sk_buff *skb)
 185{
 186        struct cpl_abort_req *req;
 187        struct chtls_sock *csk;
 188        struct tcp_sock *tp;
 189
 190        csk = rcu_dereference_sk_user_data(sk);
 191        tp = tcp_sk(sk);
 192
 193        if (!skb)
 194                skb = alloc_ctrl_skb(csk->txdata_skb_cache, sizeof(*req));
 195
 196        req = (struct cpl_abort_req *)skb_put(skb, sizeof(*req));
 197        INIT_TP_WR_CPL(req, CPL_ABORT_REQ, csk->tid);
 198        skb_set_queue_mapping(skb, (csk->txq_idx << 1) | CPL_PRIORITY_DATA);
 199        req->rsvd0 = htonl(tp->snd_nxt);
 200        req->rsvd1 = !csk_flag_nochk(csk, CSK_TX_DATA_SENT);
 201        req->cmd = mode;
 202        t4_set_arp_err_handler(skb, csk->cdev, abort_arp_failure);
 203        send_or_defer(sk, tp, skb, mode == CPL_ABORT_SEND_RST);
 204}
 205
 206static void chtls_send_reset(struct sock *sk, int mode, struct sk_buff *skb)
 207{
 208        struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 209
 210        if (unlikely(csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) ||
 211                     !csk->cdev)) {
 212                if (sk->sk_state == TCP_SYN_RECV)
 213                        csk_set_flag(csk, CSK_RST_ABORTED);
 214                goto out;
 215        }
 216
 217        if (!csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
 218                struct tcp_sock *tp = tcp_sk(sk);
 219
 220                if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
 221                        WARN_ONCE(1, "send tx flowc error");
 222                csk_set_flag(csk, CSK_TX_DATA_SENT);
 223        }
 224
 225        csk_set_flag(csk, CSK_ABORT_RPL_PENDING);
 226        chtls_purge_write_queue(sk);
 227
 228        csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
 229        if (sk->sk_state != TCP_SYN_RECV)
 230                chtls_send_abort(sk, mode, skb);
 231        else
 232                goto out;
 233
 234        return;
 235out:
 236        kfree_skb(skb);
 237}
 238
 239static void release_tcp_port(struct sock *sk)
 240{
 241        if (inet_csk(sk)->icsk_bind_hash)
 242                inet_put_port(sk);
 243}
 244
 245static void tcp_uncork(struct sock *sk)
 246{
 247        struct tcp_sock *tp = tcp_sk(sk);
 248
 249        if (tp->nonagle & TCP_NAGLE_CORK) {
 250                tp->nonagle &= ~TCP_NAGLE_CORK;
 251                chtls_tcp_push(sk, 0);
 252        }
 253}
 254
 255static void chtls_close_conn(struct sock *sk)
 256{
 257        struct cpl_close_con_req *req;
 258        struct chtls_sock *csk;
 259        struct sk_buff *skb;
 260        unsigned int tid;
 261        unsigned int len;
 262
 263        len = roundup(sizeof(struct cpl_close_con_req), 16);
 264        csk = rcu_dereference_sk_user_data(sk);
 265        tid = csk->tid;
 266
 267        skb = alloc_skb(len, GFP_KERNEL | __GFP_NOFAIL);
 268        req = (struct cpl_close_con_req *)__skb_put(skb, len);
 269        memset(req, 0, len);
 270        req->wr.wr_hi = htonl(FW_WR_OP_V(FW_TP_WR) |
 271                              FW_WR_IMMDLEN_V(sizeof(*req) -
 272                                              sizeof(req->wr)));
 273        req->wr.wr_mid = htonl(FW_WR_LEN16_V(DIV_ROUND_UP(sizeof(*req), 16)) |
 274                               FW_WR_FLOWID_V(tid));
 275
 276        OPCODE_TID(req) = htonl(MK_OPCODE_TID(CPL_CLOSE_CON_REQ, tid));
 277
 278        tcp_uncork(sk);
 279        skb_entail(sk, skb, ULPCB_FLAG_NO_HDR | ULPCB_FLAG_NO_APPEND);
 280        if (sk->sk_state != TCP_SYN_SENT)
 281                chtls_push_frames(csk, 1);
 282}
 283
 284/*
 285 * Perform a state transition during close and return the actions indicated
 286 * for the transition.  Do not make this function inline, the main reason
 287 * it exists at all is to avoid multiple inlining of tcp_set_state.
 288 */
 289static int make_close_transition(struct sock *sk)
 290{
 291        int next = (int)new_state[sk->sk_state];
 292
 293        tcp_set_state(sk, next & TCP_STATE_MASK);
 294        return next & TCP_ACTION_FIN;
 295}
 296
 297void chtls_close(struct sock *sk, long timeout)
 298{
 299        int data_lost, prev_state;
 300        struct chtls_sock *csk;
 301
 302        csk = rcu_dereference_sk_user_data(sk);
 303
 304        lock_sock(sk);
 305        sk->sk_shutdown |= SHUTDOWN_MASK;
 306
 307        data_lost = skb_queue_len(&sk->sk_receive_queue);
 308        data_lost |= skb_queue_len(&csk->tlshws.sk_recv_queue);
 309        chtls_purge_recv_queue(sk);
 310        chtls_purge_receive_queue(sk);
 311
 312        if (sk->sk_state == TCP_CLOSE) {
 313                goto wait;
 314        } else if (data_lost || sk->sk_state == TCP_SYN_SENT) {
 315                chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
 316                release_tcp_port(sk);
 317                goto unlock;
 318        } else if (sock_flag(sk, SOCK_LINGER) && !sk->sk_lingertime) {
 319                sk->sk_prot->disconnect(sk, 0);
 320        } else if (make_close_transition(sk)) {
 321                chtls_close_conn(sk);
 322        }
 323wait:
 324        if (timeout)
 325                sk_stream_wait_close(sk, timeout);
 326
 327unlock:
 328        prev_state = sk->sk_state;
 329        sock_hold(sk);
 330        sock_orphan(sk);
 331
 332        release_sock(sk);
 333
 334        local_bh_disable();
 335        bh_lock_sock(sk);
 336
 337        if (prev_state != TCP_CLOSE && sk->sk_state == TCP_CLOSE)
 338                goto out;
 339
 340        if (sk->sk_state == TCP_FIN_WAIT2 && tcp_sk(sk)->linger2 < 0 &&
 341            !csk_flag(sk, CSK_ABORT_SHUTDOWN)) {
 342                struct sk_buff *skb;
 343
 344                skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
 345                if (skb)
 346                        chtls_send_reset(sk, CPL_ABORT_SEND_RST, skb);
 347        }
 348
 349        if (sk->sk_state == TCP_CLOSE)
 350                inet_csk_destroy_sock(sk);
 351
 352out:
 353        bh_unlock_sock(sk);
 354        local_bh_enable();
 355        sock_put(sk);
 356}
 357
 358/*
 359 * Wait until a socket enters on of the given states.
 360 */
 361static int wait_for_states(struct sock *sk, unsigned int states)
 362{
 363        DECLARE_WAITQUEUE(wait, current);
 364        struct socket_wq _sk_wq;
 365        long current_timeo;
 366        int err = 0;
 367
 368        current_timeo = 200;
 369
 370        /*
 371         * We want this to work even when there's no associated struct socket.
 372         * In that case we provide a temporary wait_queue_head_t.
 373         */
 374        if (!sk->sk_wq) {
 375                init_waitqueue_head(&_sk_wq.wait);
 376                _sk_wq.fasync_list = NULL;
 377                init_rcu_head_on_stack(&_sk_wq.rcu);
 378                RCU_INIT_POINTER(sk->sk_wq, &_sk_wq);
 379        }
 380
 381        add_wait_queue(sk_sleep(sk), &wait);
 382        while (!sk_in_state(sk, states)) {
 383                if (!current_timeo) {
 384                        err = -EBUSY;
 385                        break;
 386                }
 387                if (signal_pending(current)) {
 388                        err = sock_intr_errno(current_timeo);
 389                        break;
 390                }
 391                set_current_state(TASK_UNINTERRUPTIBLE);
 392                release_sock(sk);
 393                if (!sk_in_state(sk, states))
 394                        current_timeo = schedule_timeout(current_timeo);
 395                __set_current_state(TASK_RUNNING);
 396                lock_sock(sk);
 397        }
 398        remove_wait_queue(sk_sleep(sk), &wait);
 399
 400        if (rcu_dereference(sk->sk_wq) == &_sk_wq)
 401                sk->sk_wq = NULL;
 402        return err;
 403}
 404
 405int chtls_disconnect(struct sock *sk, int flags)
 406{
 407        struct tcp_sock *tp;
 408        int err;
 409
 410        tp = tcp_sk(sk);
 411        chtls_purge_recv_queue(sk);
 412        chtls_purge_receive_queue(sk);
 413        chtls_purge_write_queue(sk);
 414
 415        if (sk->sk_state != TCP_CLOSE) {
 416                sk->sk_err = ECONNRESET;
 417                chtls_send_reset(sk, CPL_ABORT_SEND_RST, NULL);
 418                err = wait_for_states(sk, TCPF_CLOSE);
 419                if (err)
 420                        return err;
 421        }
 422        chtls_purge_recv_queue(sk);
 423        chtls_purge_receive_queue(sk);
 424        tp->max_window = 0xFFFF << (tp->rx_opt.snd_wscale);
 425        return tcp_disconnect(sk, flags);
 426}
 427
 428#define SHUTDOWN_ELIGIBLE_STATE (TCPF_ESTABLISHED | \
 429                                 TCPF_SYN_RECV | TCPF_CLOSE_WAIT)
 430void chtls_shutdown(struct sock *sk, int how)
 431{
 432        if ((how & SEND_SHUTDOWN) &&
 433            sk_in_state(sk, SHUTDOWN_ELIGIBLE_STATE) &&
 434            make_close_transition(sk))
 435                chtls_close_conn(sk);
 436}
 437
 438void chtls_destroy_sock(struct sock *sk)
 439{
 440        struct chtls_sock *csk;
 441
 442        csk = rcu_dereference_sk_user_data(sk);
 443        chtls_purge_recv_queue(sk);
 444        csk->ulp_mode = ULP_MODE_NONE;
 445        chtls_purge_write_queue(sk);
 446        free_tls_keyid(sk);
 447        kref_put(&csk->kref, chtls_sock_release);
 448        sk->sk_prot = &tcp_prot;
 449        sk->sk_prot->destroy(sk);
 450}
 451
 452static void reset_listen_child(struct sock *child)
 453{
 454        struct chtls_sock *csk = rcu_dereference_sk_user_data(child);
 455        struct sk_buff *skb;
 456
 457        skb = alloc_ctrl_skb(csk->txdata_skb_cache,
 458                             sizeof(struct cpl_abort_req));
 459
 460        chtls_send_reset(child, CPL_ABORT_SEND_RST, skb);
 461        sock_orphan(child);
 462        INC_ORPHAN_COUNT(child);
 463        if (child->sk_state == TCP_CLOSE)
 464                inet_csk_destroy_sock(child);
 465}
 466
 467static void chtls_disconnect_acceptq(struct sock *listen_sk)
 468{
 469        struct request_sock **pprev;
 470
 471        pprev = ACCEPT_QUEUE(listen_sk);
 472        while (*pprev) {
 473                struct request_sock *req = *pprev;
 474
 475                if (req->rsk_ops == &chtls_rsk_ops) {
 476                        struct sock *child = req->sk;
 477
 478                        *pprev = req->dl_next;
 479                        sk_acceptq_removed(listen_sk);
 480                        reqsk_put(req);
 481                        sock_hold(child);
 482                        local_bh_disable();
 483                        bh_lock_sock(child);
 484                        release_tcp_port(child);
 485                        reset_listen_child(child);
 486                        bh_unlock_sock(child);
 487                        local_bh_enable();
 488                        sock_put(child);
 489                } else {
 490                        pprev = &req->dl_next;
 491                }
 492        }
 493}
 494
 495static int listen_hashfn(const struct sock *sk)
 496{
 497        return ((unsigned long)sk >> 10) & (LISTEN_INFO_HASH_SIZE - 1);
 498}
 499
 500static struct listen_info *listen_hash_add(struct chtls_dev *cdev,
 501                                           struct sock *sk,
 502                                           unsigned int stid)
 503{
 504        struct listen_info *p = kmalloc(sizeof(*p), GFP_KERNEL);
 505
 506        if (p) {
 507                int key = listen_hashfn(sk);
 508
 509                p->sk = sk;
 510                p->stid = stid;
 511                spin_lock(&cdev->listen_lock);
 512                p->next = cdev->listen_hash_tab[key];
 513                cdev->listen_hash_tab[key] = p;
 514                spin_unlock(&cdev->listen_lock);
 515        }
 516        return p;
 517}
 518
 519static int listen_hash_find(struct chtls_dev *cdev,
 520                            struct sock *sk)
 521{
 522        struct listen_info *p;
 523        int stid = -1;
 524        int key;
 525
 526        key = listen_hashfn(sk);
 527
 528        spin_lock(&cdev->listen_lock);
 529        for (p = cdev->listen_hash_tab[key]; p; p = p->next)
 530                if (p->sk == sk) {
 531                        stid = p->stid;
 532                        break;
 533                }
 534        spin_unlock(&cdev->listen_lock);
 535        return stid;
 536}
 537
 538static int listen_hash_del(struct chtls_dev *cdev,
 539                           struct sock *sk)
 540{
 541        struct listen_info *p, **prev;
 542        int stid = -1;
 543        int key;
 544
 545        key = listen_hashfn(sk);
 546        prev = &cdev->listen_hash_tab[key];
 547
 548        spin_lock(&cdev->listen_lock);
 549        for (p = *prev; p; prev = &p->next, p = p->next)
 550                if (p->sk == sk) {
 551                        stid = p->stid;
 552                        *prev = p->next;
 553                        kfree(p);
 554                        break;
 555                }
 556        spin_unlock(&cdev->listen_lock);
 557        return stid;
 558}
 559
 560static void cleanup_syn_rcv_conn(struct sock *child, struct sock *parent)
 561{
 562        struct request_sock *req;
 563        struct chtls_sock *csk;
 564
 565        csk = rcu_dereference_sk_user_data(child);
 566        req = csk->passive_reap_next;
 567
 568        reqsk_queue_removed(&inet_csk(parent)->icsk_accept_queue, req);
 569        __skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
 570        chtls_reqsk_free(req);
 571        csk->passive_reap_next = NULL;
 572}
 573
 574static void chtls_reset_synq(struct listen_ctx *listen_ctx)
 575{
 576        struct sock *listen_sk = listen_ctx->lsk;
 577
 578        while (!skb_queue_empty(&listen_ctx->synq)) {
 579                struct chtls_sock *csk =
 580                        container_of((struct synq *)__skb_dequeue
 581                                (&listen_ctx->synq), struct chtls_sock, synq);
 582                struct sock *child = csk->sk;
 583
 584                cleanup_syn_rcv_conn(child, listen_sk);
 585                sock_hold(child);
 586                local_bh_disable();
 587                bh_lock_sock(child);
 588                release_tcp_port(child);
 589                reset_listen_child(child);
 590                bh_unlock_sock(child);
 591                local_bh_enable();
 592                sock_put(child);
 593        }
 594}
 595
 596int chtls_listen_start(struct chtls_dev *cdev, struct sock *sk)
 597{
 598        struct net_device *ndev;
 599        struct listen_ctx *ctx;
 600        struct adapter *adap;
 601        struct port_info *pi;
 602        int stid;
 603        int ret;
 604
 605        if (sk->sk_family != PF_INET)
 606                return -EAGAIN;
 607
 608        rcu_read_lock();
 609        ndev = chtls_ipv4_netdev(cdev, sk);
 610        rcu_read_unlock();
 611        if (!ndev)
 612                return -EBADF;
 613
 614        pi = netdev_priv(ndev);
 615        adap = pi->adapter;
 616        if (!(adap->flags & CXGB4_FULL_INIT_DONE))
 617                return -EBADF;
 618
 619        if (listen_hash_find(cdev, sk) >= 0)   /* already have it */
 620                return -EADDRINUSE;
 621
 622        ctx = kmalloc(sizeof(*ctx), GFP_KERNEL);
 623        if (!ctx)
 624                return -ENOMEM;
 625
 626        __module_get(THIS_MODULE);
 627        ctx->lsk = sk;
 628        ctx->cdev = cdev;
 629        ctx->state = T4_LISTEN_START_PENDING;
 630        skb_queue_head_init(&ctx->synq);
 631
 632        stid = cxgb4_alloc_stid(cdev->tids, sk->sk_family, ctx);
 633        if (stid < 0)
 634                goto free_ctx;
 635
 636        sock_hold(sk);
 637        if (!listen_hash_add(cdev, sk, stid))
 638                goto free_stid;
 639
 640        ret = cxgb4_create_server(ndev, stid,
 641                                  inet_sk(sk)->inet_rcv_saddr,
 642                                  inet_sk(sk)->inet_sport, 0,
 643                                  cdev->lldi->rxq_ids[0]);
 644        if (ret > 0)
 645                ret = net_xmit_errno(ret);
 646        if (ret)
 647                goto del_hash;
 648        return 0;
 649del_hash:
 650        listen_hash_del(cdev, sk);
 651free_stid:
 652        cxgb4_free_stid(cdev->tids, stid, sk->sk_family);
 653        sock_put(sk);
 654free_ctx:
 655        kfree(ctx);
 656        module_put(THIS_MODULE);
 657        return -EBADF;
 658}
 659
 660void chtls_listen_stop(struct chtls_dev *cdev, struct sock *sk)
 661{
 662        struct listen_ctx *listen_ctx;
 663        int stid;
 664
 665        stid = listen_hash_del(cdev, sk);
 666        if (stid < 0)
 667                return;
 668
 669        listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
 670        chtls_reset_synq(listen_ctx);
 671
 672        cxgb4_remove_server(cdev->lldi->ports[0], stid,
 673                            cdev->lldi->rxq_ids[0], 0);
 674        chtls_disconnect_acceptq(sk);
 675}
 676
 677static int chtls_pass_open_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
 678{
 679        struct cpl_pass_open_rpl *rpl = cplhdr(skb) + RSS_HDR;
 680        unsigned int stid = GET_TID(rpl);
 681        struct listen_ctx *listen_ctx;
 682
 683        listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
 684        if (!listen_ctx)
 685                return CPL_RET_BUF_DONE;
 686
 687        if (listen_ctx->state == T4_LISTEN_START_PENDING) {
 688                listen_ctx->state = T4_LISTEN_STARTED;
 689                return CPL_RET_BUF_DONE;
 690        }
 691
 692        if (rpl->status != CPL_ERR_NONE) {
 693                pr_info("Unexpected PASS_OPEN_RPL status %u for STID %u\n",
 694                        rpl->status, stid);
 695                return CPL_RET_BUF_DONE;
 696        }
 697        cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
 698        sock_put(listen_ctx->lsk);
 699        kfree(listen_ctx);
 700        module_put(THIS_MODULE);
 701
 702        return 0;
 703}
 704
 705static int chtls_close_listsrv_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
 706{
 707        struct cpl_close_listsvr_rpl *rpl = cplhdr(skb) + RSS_HDR;
 708        struct listen_ctx *listen_ctx;
 709        unsigned int stid;
 710        void *data;
 711
 712        stid = GET_TID(rpl);
 713        data = lookup_stid(cdev->tids, stid);
 714        listen_ctx = (struct listen_ctx *)data;
 715
 716        if (rpl->status != CPL_ERR_NONE) {
 717                pr_info("Unexpected CLOSE_LISTSRV_RPL status %u for STID %u\n",
 718                        rpl->status, stid);
 719                return CPL_RET_BUF_DONE;
 720        }
 721
 722        cxgb4_free_stid(cdev->tids, stid, listen_ctx->lsk->sk_family);
 723        sock_put(listen_ctx->lsk);
 724        kfree(listen_ctx);
 725        module_put(THIS_MODULE);
 726
 727        return 0;
 728}
 729
 730static void chtls_purge_wr_queue(struct sock *sk)
 731{
 732        struct sk_buff *skb;
 733
 734        while ((skb = dequeue_wr(sk)) != NULL)
 735                kfree_skb(skb);
 736}
 737
 738static void chtls_release_resources(struct sock *sk)
 739{
 740        struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
 741        struct chtls_dev *cdev = csk->cdev;
 742        unsigned int tid = csk->tid;
 743        struct tid_info *tids;
 744
 745        if (!cdev)
 746                return;
 747
 748        tids = cdev->tids;
 749        kfree_skb(csk->txdata_skb_cache);
 750        csk->txdata_skb_cache = NULL;
 751
 752        if (csk->wr_credits != csk->wr_max_credits) {
 753                chtls_purge_wr_queue(sk);
 754                chtls_reset_wr_list(csk);
 755        }
 756
 757        if (csk->l2t_entry) {
 758                cxgb4_l2t_release(csk->l2t_entry);
 759                csk->l2t_entry = NULL;
 760        }
 761
 762        cxgb4_remove_tid(tids, csk->port_id, tid, sk->sk_family);
 763        sock_put(sk);
 764}
 765
 766static void chtls_conn_done(struct sock *sk)
 767{
 768        if (sock_flag(sk, SOCK_DEAD))
 769                chtls_purge_receive_queue(sk);
 770        sk_wakeup_sleepers(sk, 0);
 771        tcp_done(sk);
 772}
 773
 774static void do_abort_syn_rcv(struct sock *child, struct sock *parent)
 775{
 776        /*
 777         * If the server is still open we clean up the child connection,
 778         * otherwise the server already did the clean up as it was purging
 779         * its SYN queue and the skb was just sitting in its backlog.
 780         */
 781        if (likely(parent->sk_state == TCP_LISTEN)) {
 782                cleanup_syn_rcv_conn(child, parent);
 783                /* Without the below call to sock_orphan,
 784                 * we leak the socket resource with syn_flood test
 785                 * as inet_csk_destroy_sock will not be called
 786                 * in tcp_done since SOCK_DEAD flag is not set.
 787                 * Kernel handles this differently where new socket is
 788                 * created only after 3 way handshake is done.
 789                 */
 790                sock_orphan(child);
 791                percpu_counter_inc((child)->sk_prot->orphan_count);
 792                chtls_release_resources(child);
 793                chtls_conn_done(child);
 794        } else {
 795                if (csk_flag(child, CSK_RST_ABORTED)) {
 796                        chtls_release_resources(child);
 797                        chtls_conn_done(child);
 798                }
 799        }
 800}
 801
 802static void pass_open_abort(struct sock *child, struct sock *parent,
 803                            struct sk_buff *skb)
 804{
 805        do_abort_syn_rcv(child, parent);
 806        kfree_skb(skb);
 807}
 808
 809static void bl_pass_open_abort(struct sock *lsk, struct sk_buff *skb)
 810{
 811        pass_open_abort(skb->sk, lsk, skb);
 812}
 813
 814static void chtls_pass_open_arp_failure(struct sock *sk,
 815                                        struct sk_buff *skb)
 816{
 817        const struct request_sock *oreq;
 818        struct chtls_sock *csk;
 819        struct chtls_dev *cdev;
 820        struct sock *parent;
 821        void *data;
 822
 823        csk = rcu_dereference_sk_user_data(sk);
 824        cdev = csk->cdev;
 825
 826        /*
 827         * If the connection is being aborted due to the parent listening
 828         * socket going away there's nothing to do, the ABORT_REQ will close
 829         * the connection.
 830         */
 831        if (csk_flag(sk, CSK_ABORT_RPL_PENDING)) {
 832                kfree_skb(skb);
 833                return;
 834        }
 835
 836        oreq = csk->passive_reap_next;
 837        data = lookup_stid(cdev->tids, oreq->ts_recent);
 838        parent = ((struct listen_ctx *)data)->lsk;
 839
 840        bh_lock_sock(parent);
 841        if (!sock_owned_by_user(parent)) {
 842                pass_open_abort(sk, parent, skb);
 843        } else {
 844                BLOG_SKB_CB(skb)->backlog_rcv = bl_pass_open_abort;
 845                __sk_add_backlog(parent, skb);
 846        }
 847        bh_unlock_sock(parent);
 848}
 849
 850static void chtls_accept_rpl_arp_failure(void *handle,
 851                                         struct sk_buff *skb)
 852{
 853        struct sock *sk = (struct sock *)handle;
 854
 855        sock_hold(sk);
 856        process_cpl_msg(chtls_pass_open_arp_failure, sk, skb);
 857        sock_put(sk);
 858}
 859
 860static unsigned int chtls_select_mss(const struct chtls_sock *csk,
 861                                     unsigned int pmtu,
 862                                     struct cpl_pass_accept_req *req)
 863{
 864        struct chtls_dev *cdev;
 865        struct dst_entry *dst;
 866        unsigned int tcpoptsz;
 867        unsigned int iphdrsz;
 868        unsigned int mtu_idx;
 869        struct tcp_sock *tp;
 870        unsigned int mss;
 871        struct sock *sk;
 872
 873        mss = ntohs(req->tcpopt.mss);
 874        sk = csk->sk;
 875        dst = __sk_dst_get(sk);
 876        cdev = csk->cdev;
 877        tp = tcp_sk(sk);
 878        tcpoptsz = 0;
 879
 880        iphdrsz = sizeof(struct iphdr) + sizeof(struct tcphdr);
 881        if (req->tcpopt.tstamp)
 882                tcpoptsz += round_up(TCPOLEN_TIMESTAMP, 4);
 883
 884        tp->advmss = dst_metric_advmss(dst);
 885        if (USER_MSS(tp) && tp->advmss > USER_MSS(tp))
 886                tp->advmss = USER_MSS(tp);
 887        if (tp->advmss > pmtu - iphdrsz)
 888                tp->advmss = pmtu - iphdrsz;
 889        if (mss && tp->advmss > mss)
 890                tp->advmss = mss;
 891
 892        tp->advmss = cxgb4_best_aligned_mtu(cdev->lldi->mtus,
 893                                            iphdrsz + tcpoptsz,
 894                                            tp->advmss - tcpoptsz,
 895                                            8, &mtu_idx);
 896        tp->advmss -= iphdrsz;
 897
 898        inet_csk(sk)->icsk_pmtu_cookie = pmtu;
 899        return mtu_idx;
 900}
 901
 902static unsigned int select_rcv_wscale(int space, int wscale_ok, int win_clamp)
 903{
 904        int wscale = 0;
 905
 906        if (space > MAX_RCV_WND)
 907                space = MAX_RCV_WND;
 908        if (win_clamp && win_clamp < space)
 909                space = win_clamp;
 910
 911        if (wscale_ok) {
 912                while (wscale < 14 && (65535 << wscale) < space)
 913                        wscale++;
 914        }
 915        return wscale;
 916}
 917
 918static void chtls_pass_accept_rpl(struct sk_buff *skb,
 919                                  struct cpl_pass_accept_req *req,
 920                                  unsigned int tid)
 921
 922{
 923        struct cpl_t5_pass_accept_rpl *rpl5;
 924        struct cxgb4_lld_info *lldi;
 925        const struct tcphdr *tcph;
 926        const struct tcp_sock *tp;
 927        struct chtls_sock *csk;
 928        unsigned int len;
 929        struct sock *sk;
 930        u32 opt2, hlen;
 931        u64 opt0;
 932
 933        sk = skb->sk;
 934        tp = tcp_sk(sk);
 935        csk = sk->sk_user_data;
 936        csk->tid = tid;
 937        lldi = csk->cdev->lldi;
 938        len = roundup(sizeof(*rpl5), 16);
 939
 940        rpl5 = __skb_put_zero(skb, len);
 941        INIT_TP_WR(rpl5, tid);
 942
 943        OPCODE_TID(rpl5) = cpu_to_be32(MK_OPCODE_TID(CPL_PASS_ACCEPT_RPL,
 944                                                     csk->tid));
 945        csk->mtu_idx = chtls_select_mss(csk, dst_mtu(__sk_dst_get(sk)),
 946                                        req);
 947        opt0 = TCAM_BYPASS_F |
 948               WND_SCALE_V(RCV_WSCALE(tp)) |
 949               MSS_IDX_V(csk->mtu_idx) |
 950               L2T_IDX_V(csk->l2t_entry->idx) |
 951               NAGLE_V(!(tp->nonagle & TCP_NAGLE_OFF)) |
 952               TX_CHAN_V(csk->tx_chan) |
 953               SMAC_SEL_V(csk->smac_idx) |
 954               DSCP_V(csk->tos >> 2) |
 955               ULP_MODE_V(ULP_MODE_TLS) |
 956               RCV_BUFSIZ_V(min(tp->rcv_wnd >> 10, RCV_BUFSIZ_M));
 957
 958        opt2 = RX_CHANNEL_V(0) |
 959                RSS_QUEUE_VALID_F | RSS_QUEUE_V(csk->rss_qid);
 960
 961        if (!is_t5(lldi->adapter_type))
 962                opt2 |= RX_FC_DISABLE_F;
 963        if (req->tcpopt.tstamp)
 964                opt2 |= TSTAMPS_EN_F;
 965        if (req->tcpopt.sack)
 966                opt2 |= SACK_EN_F;
 967        hlen = ntohl(req->hdr_len);
 968
 969        tcph = (struct tcphdr *)((u8 *)(req + 1) +
 970                        T6_ETH_HDR_LEN_G(hlen) + T6_IP_HDR_LEN_G(hlen));
 971        if (tcph->ece && tcph->cwr)
 972                opt2 |= CCTRL_ECN_V(1);
 973        opt2 |= CONG_CNTRL_V(CONG_ALG_NEWRENO);
 974        opt2 |= T5_ISS_F;
 975        opt2 |= T5_OPT_2_VALID_F;
 976        rpl5->opt0 = cpu_to_be64(opt0);
 977        rpl5->opt2 = cpu_to_be32(opt2);
 978        rpl5->iss = cpu_to_be32((prandom_u32() & ~7UL) - 1);
 979        set_wr_txq(skb, CPL_PRIORITY_SETUP, csk->port_id);
 980        t4_set_arp_err_handler(skb, sk, chtls_accept_rpl_arp_failure);
 981        cxgb4_l2t_send(csk->egress_dev, skb, csk->l2t_entry);
 982}
 983
 984static void inet_inherit_port(struct inet_hashinfo *hash_info,
 985                              struct sock *lsk, struct sock *newsk)
 986{
 987        local_bh_disable();
 988        __inet_inherit_port(lsk, newsk);
 989        local_bh_enable();
 990}
 991
 992static int chtls_backlog_rcv(struct sock *sk, struct sk_buff *skb)
 993{
 994        if (skb->protocol) {
 995                kfree_skb(skb);
 996                return 0;
 997        }
 998        BLOG_SKB_CB(skb)->backlog_rcv(sk, skb);
 999        return 0;
1000}
1001
1002static void chtls_set_tcp_window(struct chtls_sock *csk)
1003{
1004        struct net_device *ndev = csk->egress_dev;
1005        struct port_info *pi = netdev_priv(ndev);
1006        unsigned int linkspeed;
1007        u8 scale;
1008
1009        linkspeed = pi->link_cfg.speed;
1010        scale = linkspeed / SPEED_10000;
1011#define CHTLS_10G_RCVWIN (256 * 1024)
1012        csk->rcv_win = CHTLS_10G_RCVWIN;
1013        if (scale)
1014                csk->rcv_win *= scale;
1015#define CHTLS_10G_SNDWIN (256 * 1024)
1016        csk->snd_win = CHTLS_10G_SNDWIN;
1017        if (scale)
1018                csk->snd_win *= scale;
1019}
1020
1021static struct sock *chtls_recv_sock(struct sock *lsk,
1022                                    struct request_sock *oreq,
1023                                    void *network_hdr,
1024                                    const struct cpl_pass_accept_req *req,
1025                                    struct chtls_dev *cdev)
1026{
1027        struct inet_sock *newinet;
1028        const struct iphdr *iph;
1029        struct tls_context *ctx;
1030        struct net_device *ndev;
1031        struct chtls_sock *csk;
1032        struct dst_entry *dst;
1033        struct neighbour *n;
1034        struct tcp_sock *tp;
1035        struct sock *newsk;
1036        u16 port_id;
1037        int rxq_idx;
1038        int step;
1039
1040        iph = (const struct iphdr *)network_hdr;
1041        newsk = tcp_create_openreq_child(lsk, oreq, cdev->askb);
1042        if (!newsk)
1043                goto free_oreq;
1044
1045        dst = inet_csk_route_child_sock(lsk, newsk, oreq);
1046        if (!dst)
1047                goto free_sk;
1048
1049        n = dst_neigh_lookup(dst, &iph->saddr);
1050        if (!n)
1051                goto free_sk;
1052
1053        ndev = n->dev;
1054        if (!ndev)
1055                goto free_dst;
1056        port_id = cxgb4_port_idx(ndev);
1057
1058        csk = chtls_sock_create(cdev);
1059        if (!csk)
1060                goto free_dst;
1061
1062        csk->l2t_entry = cxgb4_l2t_get(cdev->lldi->l2t, n, ndev, 0);
1063        if (!csk->l2t_entry)
1064                goto free_csk;
1065
1066        newsk->sk_user_data = csk;
1067        newsk->sk_backlog_rcv = chtls_backlog_rcv;
1068
1069        tp = tcp_sk(newsk);
1070        newinet = inet_sk(newsk);
1071
1072        newinet->inet_daddr = iph->saddr;
1073        newinet->inet_rcv_saddr = iph->daddr;
1074        newinet->inet_saddr = iph->daddr;
1075
1076        oreq->ts_recent = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1077        sk_setup_caps(newsk, dst);
1078        ctx = tls_get_ctx(lsk);
1079        newsk->sk_destruct = ctx->sk_destruct;
1080        csk->sk = newsk;
1081        csk->passive_reap_next = oreq;
1082        csk->tx_chan = cxgb4_port_chan(ndev);
1083        csk->port_id = port_id;
1084        csk->egress_dev = ndev;
1085        csk->tos = PASS_OPEN_TOS_G(ntohl(req->tos_stid));
1086        chtls_set_tcp_window(csk);
1087        tp->rcv_wnd = csk->rcv_win;
1088        csk->sndbuf = csk->snd_win;
1089        csk->ulp_mode = ULP_MODE_TLS;
1090        step = cdev->lldi->nrxq / cdev->lldi->nchan;
1091        csk->rss_qid = cdev->lldi->rxq_ids[port_id * step];
1092        rxq_idx = port_id * step;
1093        csk->txq_idx = (rxq_idx < cdev->lldi->ntxq) ? rxq_idx :
1094                        port_id * step;
1095        csk->sndbuf = newsk->sk_sndbuf;
1096        csk->smac_idx = ((struct port_info *)netdev_priv(ndev))->smt_idx;
1097        RCV_WSCALE(tp) = select_rcv_wscale(tcp_full_space(newsk),
1098                                           sock_net(newsk)->
1099                                                ipv4.sysctl_tcp_window_scaling,
1100                                           tp->window_clamp);
1101        neigh_release(n);
1102        inet_inherit_port(&tcp_hashinfo, lsk, newsk);
1103        csk_set_flag(csk, CSK_CONN_INLINE);
1104        bh_unlock_sock(newsk); /* tcp_create_openreq_child ->sk_clone_lock */
1105
1106        return newsk;
1107free_csk:
1108        chtls_sock_release(&csk->kref);
1109free_dst:
1110        dst_release(dst);
1111free_sk:
1112        inet_csk_prepare_forced_close(newsk);
1113        tcp_done(newsk);
1114free_oreq:
1115        chtls_reqsk_free(oreq);
1116        return NULL;
1117}
1118
1119/*
1120 * Populate a TID_RELEASE WR.  The skb must be already propely sized.
1121 */
1122static  void mk_tid_release(struct sk_buff *skb,
1123                            unsigned int chan, unsigned int tid)
1124{
1125        struct cpl_tid_release *req;
1126        unsigned int len;
1127
1128        len = roundup(sizeof(struct cpl_tid_release), 16);
1129        req = (struct cpl_tid_release *)__skb_put(skb, len);
1130        memset(req, 0, len);
1131        set_wr_txq(skb, CPL_PRIORITY_SETUP, chan);
1132        INIT_TP_WR_CPL(req, CPL_TID_RELEASE, tid);
1133}
1134
1135static int chtls_get_module(struct sock *sk)
1136{
1137        struct inet_connection_sock *icsk = inet_csk(sk);
1138
1139        if (!try_module_get(icsk->icsk_ulp_ops->owner))
1140                return -1;
1141
1142        return 0;
1143}
1144
1145static void chtls_pass_accept_request(struct sock *sk,
1146                                      struct sk_buff *skb)
1147{
1148        struct cpl_t5_pass_accept_rpl *rpl;
1149        struct cpl_pass_accept_req *req;
1150        struct listen_ctx *listen_ctx;
1151        struct vlan_ethhdr *vlan_eh;
1152        struct request_sock *oreq;
1153        struct sk_buff *reply_skb;
1154        struct chtls_sock *csk;
1155        struct chtls_dev *cdev;
1156        struct tcphdr *tcph;
1157        struct sock *newsk;
1158        struct ethhdr *eh;
1159        struct iphdr *iph;
1160        void *network_hdr;
1161        unsigned int stid;
1162        unsigned int len;
1163        unsigned int tid;
1164        bool th_ecn, ect;
1165        __u8 ip_dsfield; /* IPv4 tos or IPv6 dsfield */
1166        u16 eth_hdr_len;
1167        bool ecn_ok;
1168
1169        req = cplhdr(skb) + RSS_HDR;
1170        tid = GET_TID(req);
1171        cdev = BLOG_SKB_CB(skb)->cdev;
1172        newsk = lookup_tid(cdev->tids, tid);
1173        stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1174        if (newsk) {
1175                pr_info("tid (%d) already in use\n", tid);
1176                return;
1177        }
1178
1179        len = roundup(sizeof(*rpl), 16);
1180        reply_skb = alloc_skb(len, GFP_ATOMIC);
1181        if (!reply_skb) {
1182                cxgb4_remove_tid(cdev->tids, 0, tid, sk->sk_family);
1183                kfree_skb(skb);
1184                return;
1185        }
1186
1187        if (sk->sk_state != TCP_LISTEN)
1188                goto reject;
1189
1190        if (inet_csk_reqsk_queue_is_full(sk))
1191                goto reject;
1192
1193        if (sk_acceptq_is_full(sk))
1194                goto reject;
1195
1196        oreq = inet_reqsk_alloc(&chtls_rsk_ops, sk, true);
1197        if (!oreq)
1198                goto reject;
1199
1200        oreq->rsk_rcv_wnd = 0;
1201        oreq->rsk_window_clamp = 0;
1202        oreq->cookie_ts = 0;
1203        oreq->mss = 0;
1204        oreq->ts_recent = 0;
1205
1206        eth_hdr_len = T6_ETH_HDR_LEN_G(ntohl(req->hdr_len));
1207        if (eth_hdr_len == ETH_HLEN) {
1208                eh = (struct ethhdr *)(req + 1);
1209                iph = (struct iphdr *)(eh + 1);
1210                network_hdr = (void *)(eh + 1);
1211        } else {
1212                vlan_eh = (struct vlan_ethhdr *)(req + 1);
1213                iph = (struct iphdr *)(vlan_eh + 1);
1214                network_hdr = (void *)(vlan_eh + 1);
1215        }
1216        if (iph->version != 0x4)
1217                goto free_oreq;
1218
1219        tcph = (struct tcphdr *)(iph + 1);
1220        skb_set_network_header(skb, (void *)iph - (void *)req);
1221
1222        tcp_rsk(oreq)->tfo_listener = false;
1223        tcp_rsk(oreq)->rcv_isn = ntohl(tcph->seq);
1224        chtls_set_req_port(oreq, tcph->source, tcph->dest);
1225        chtls_set_req_addr(oreq, iph->daddr, iph->saddr);
1226        ip_dsfield = ipv4_get_dsfield(iph);
1227        if (req->tcpopt.wsf <= 14 &&
1228            sock_net(sk)->ipv4.sysctl_tcp_window_scaling) {
1229                inet_rsk(oreq)->wscale_ok = 1;
1230                inet_rsk(oreq)->snd_wscale = req->tcpopt.wsf;
1231        }
1232        inet_rsk(oreq)->ir_iif = sk->sk_bound_dev_if;
1233        th_ecn = tcph->ece && tcph->cwr;
1234        if (th_ecn) {
1235                ect = !INET_ECN_is_not_ect(ip_dsfield);
1236                ecn_ok = sock_net(sk)->ipv4.sysctl_tcp_ecn;
1237                if ((!ect && ecn_ok) || tcp_ca_needs_ecn(sk))
1238                        inet_rsk(oreq)->ecn_ok = 1;
1239        }
1240
1241        newsk = chtls_recv_sock(sk, oreq, network_hdr, req, cdev);
1242        if (!newsk)
1243                goto reject;
1244
1245        if (chtls_get_module(newsk))
1246                goto reject;
1247        inet_csk_reqsk_queue_added(sk);
1248        reply_skb->sk = newsk;
1249        chtls_install_cpl_ops(newsk);
1250        cxgb4_insert_tid(cdev->tids, newsk, tid, newsk->sk_family);
1251        csk = rcu_dereference_sk_user_data(newsk);
1252        listen_ctx = (struct listen_ctx *)lookup_stid(cdev->tids, stid);
1253        csk->listen_ctx = listen_ctx;
1254        __skb_queue_tail(&listen_ctx->synq, (struct sk_buff *)&csk->synq);
1255        chtls_pass_accept_rpl(reply_skb, req, tid);
1256        kfree_skb(skb);
1257        return;
1258
1259free_oreq:
1260        chtls_reqsk_free(oreq);
1261reject:
1262        mk_tid_release(reply_skb, 0, tid);
1263        cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1264        kfree_skb(skb);
1265}
1266
1267/*
1268 * Handle a CPL_PASS_ACCEPT_REQ message.
1269 */
1270static int chtls_pass_accept_req(struct chtls_dev *cdev, struct sk_buff *skb)
1271{
1272        struct cpl_pass_accept_req *req = cplhdr(skb) + RSS_HDR;
1273        struct listen_ctx *ctx;
1274        unsigned int stid;
1275        unsigned int tid;
1276        struct sock *lsk;
1277        void *data;
1278
1279        stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1280        tid = GET_TID(req);
1281
1282        data = lookup_stid(cdev->tids, stid);
1283        if (!data)
1284                return 1;
1285
1286        ctx = (struct listen_ctx *)data;
1287        lsk = ctx->lsk;
1288
1289        if (unlikely(tid_out_of_range(cdev->tids, tid))) {
1290                pr_info("passive open TID %u too large\n", tid);
1291                return 1;
1292        }
1293
1294        BLOG_SKB_CB(skb)->cdev = cdev;
1295        process_cpl_msg(chtls_pass_accept_request, lsk, skb);
1296        return 0;
1297}
1298
1299/*
1300 * Completes some final bits of initialization for just established connections
1301 * and changes their state to TCP_ESTABLISHED.
1302 *
1303 * snd_isn here is the ISN after the SYN, i.e., the true ISN + 1.
1304 */
1305static void make_established(struct sock *sk, u32 snd_isn, unsigned int opt)
1306{
1307        struct tcp_sock *tp = tcp_sk(sk);
1308
1309        tp->pushed_seq = snd_isn;
1310        tp->write_seq = snd_isn;
1311        tp->snd_nxt = snd_isn;
1312        tp->snd_una = snd_isn;
1313        inet_sk(sk)->inet_id = prandom_u32();
1314        assign_rxopt(sk, opt);
1315
1316        if (tp->rcv_wnd > (RCV_BUFSIZ_M << 10))
1317                tp->rcv_wup -= tp->rcv_wnd - (RCV_BUFSIZ_M << 10);
1318
1319        smp_mb();
1320        tcp_set_state(sk, TCP_ESTABLISHED);
1321}
1322
1323static void chtls_abort_conn(struct sock *sk, struct sk_buff *skb)
1324{
1325        struct sk_buff *abort_skb;
1326
1327        abort_skb = alloc_skb(sizeof(struct cpl_abort_req), GFP_ATOMIC);
1328        if (abort_skb)
1329                chtls_send_reset(sk, CPL_ABORT_SEND_RST, abort_skb);
1330}
1331
1332static struct sock *reap_list;
1333static DEFINE_SPINLOCK(reap_list_lock);
1334
1335/*
1336 * Process the reap list.
1337 */
1338DECLARE_TASK_FUNC(process_reap_list, task_param)
1339{
1340        spin_lock_bh(&reap_list_lock);
1341        while (reap_list) {
1342                struct sock *sk = reap_list;
1343                struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1344
1345                reap_list = csk->passive_reap_next;
1346                csk->passive_reap_next = NULL;
1347                spin_unlock(&reap_list_lock);
1348                sock_hold(sk);
1349
1350                bh_lock_sock(sk);
1351                chtls_abort_conn(sk, NULL);
1352                sock_orphan(sk);
1353                if (sk->sk_state == TCP_CLOSE)
1354                        inet_csk_destroy_sock(sk);
1355                bh_unlock_sock(sk);
1356                sock_put(sk);
1357                spin_lock(&reap_list_lock);
1358        }
1359        spin_unlock_bh(&reap_list_lock);
1360}
1361
1362static DECLARE_WORK(reap_task, process_reap_list);
1363
1364static void add_to_reap_list(struct sock *sk)
1365{
1366        struct chtls_sock *csk = sk->sk_user_data;
1367
1368        local_bh_disable();
1369        bh_lock_sock(sk);
1370        release_tcp_port(sk); /* release the port immediately */
1371
1372        spin_lock(&reap_list_lock);
1373        csk->passive_reap_next = reap_list;
1374        reap_list = sk;
1375        if (!csk->passive_reap_next)
1376                schedule_work(&reap_task);
1377        spin_unlock(&reap_list_lock);
1378        bh_unlock_sock(sk);
1379        local_bh_enable();
1380}
1381
1382static void add_pass_open_to_parent(struct sock *child, struct sock *lsk,
1383                                    struct chtls_dev *cdev)
1384{
1385        struct request_sock *oreq;
1386        struct chtls_sock *csk;
1387
1388        if (lsk->sk_state != TCP_LISTEN)
1389                return;
1390
1391        csk = child->sk_user_data;
1392        oreq = csk->passive_reap_next;
1393        csk->passive_reap_next = NULL;
1394
1395        reqsk_queue_removed(&inet_csk(lsk)->icsk_accept_queue, oreq);
1396        __skb_unlink((struct sk_buff *)&csk->synq, &csk->listen_ctx->synq);
1397
1398        if (sk_acceptq_is_full(lsk)) {
1399                chtls_reqsk_free(oreq);
1400                add_to_reap_list(child);
1401        } else {
1402                refcount_set(&oreq->rsk_refcnt, 1);
1403                inet_csk_reqsk_queue_add(lsk, oreq, child);
1404                lsk->sk_data_ready(lsk);
1405        }
1406}
1407
1408static void bl_add_pass_open_to_parent(struct sock *lsk, struct sk_buff *skb)
1409{
1410        struct sock *child = skb->sk;
1411
1412        skb->sk = NULL;
1413        add_pass_open_to_parent(child, lsk, BLOG_SKB_CB(skb)->cdev);
1414        kfree_skb(skb);
1415}
1416
1417static int chtls_pass_establish(struct chtls_dev *cdev, struct sk_buff *skb)
1418{
1419        struct cpl_pass_establish *req = cplhdr(skb) + RSS_HDR;
1420        struct chtls_sock *csk;
1421        struct sock *lsk, *sk;
1422        unsigned int hwtid;
1423
1424        hwtid = GET_TID(req);
1425        sk = lookup_tid(cdev->tids, hwtid);
1426        if (!sk)
1427                return (CPL_RET_UNKNOWN_TID | CPL_RET_BUF_DONE);
1428
1429        bh_lock_sock(sk);
1430        if (unlikely(sock_owned_by_user(sk))) {
1431                kfree_skb(skb);
1432        } else {
1433                unsigned int stid;
1434                void *data;
1435
1436                csk = sk->sk_user_data;
1437                csk->wr_max_credits = 64;
1438                csk->wr_credits = 64;
1439                csk->wr_unacked = 0;
1440                make_established(sk, ntohl(req->snd_isn), ntohs(req->tcp_opt));
1441                stid = PASS_OPEN_TID_G(ntohl(req->tos_stid));
1442                sk->sk_state_change(sk);
1443                if (unlikely(sk->sk_socket))
1444                        sk_wake_async(sk, 0, POLL_OUT);
1445
1446                data = lookup_stid(cdev->tids, stid);
1447                lsk = ((struct listen_ctx *)data)->lsk;
1448
1449                bh_lock_sock(lsk);
1450                if (unlikely(skb_queue_empty(&csk->listen_ctx->synq))) {
1451                        /* removed from synq */
1452                        bh_unlock_sock(lsk);
1453                        kfree_skb(skb);
1454                        goto unlock;
1455                }
1456
1457                if (likely(!sock_owned_by_user(lsk))) {
1458                        kfree_skb(skb);
1459                        add_pass_open_to_parent(sk, lsk, cdev);
1460                } else {
1461                        skb->sk = sk;
1462                        BLOG_SKB_CB(skb)->cdev = cdev;
1463                        BLOG_SKB_CB(skb)->backlog_rcv =
1464                                bl_add_pass_open_to_parent;
1465                        __sk_add_backlog(lsk, skb);
1466                }
1467                bh_unlock_sock(lsk);
1468        }
1469unlock:
1470        bh_unlock_sock(sk);
1471        return 0;
1472}
1473
1474/*
1475 * Handle receipt of an urgent pointer.
1476 */
1477static void handle_urg_ptr(struct sock *sk, u32 urg_seq)
1478{
1479        struct tcp_sock *tp = tcp_sk(sk);
1480
1481        urg_seq--;
1482        if (tp->urg_data && !after(urg_seq, tp->urg_seq))
1483                return; /* duplicate pointer */
1484
1485        sk_send_sigurg(sk);
1486        if (tp->urg_seq == tp->copied_seq && tp->urg_data &&
1487            !sock_flag(sk, SOCK_URGINLINE) &&
1488            tp->copied_seq != tp->rcv_nxt) {
1489                struct sk_buff *skb = skb_peek(&sk->sk_receive_queue);
1490
1491                tp->copied_seq++;
1492                if (skb && tp->copied_seq - ULP_SKB_CB(skb)->seq >= skb->len)
1493                        chtls_free_skb(sk, skb);
1494        }
1495
1496        tp->urg_data = TCP_URG_NOTYET;
1497        tp->urg_seq = urg_seq;
1498}
1499
1500static void check_sk_callbacks(struct chtls_sock *csk)
1501{
1502        struct sock *sk = csk->sk;
1503
1504        if (unlikely(sk->sk_user_data &&
1505                     !csk_flag_nochk(csk, CSK_CALLBACKS_CHKD)))
1506                csk_set_flag(csk, CSK_CALLBACKS_CHKD);
1507}
1508
1509/*
1510 * Handles Rx data that arrives in a state where the socket isn't accepting
1511 * new data.
1512 */
1513static void handle_excess_rx(struct sock *sk, struct sk_buff *skb)
1514{
1515        if (!csk_flag(sk, CSK_ABORT_SHUTDOWN))
1516                chtls_abort_conn(sk, skb);
1517
1518        kfree_skb(skb);
1519}
1520
1521static void chtls_recv_data(struct sock *sk, struct sk_buff *skb)
1522{
1523        struct cpl_rx_data *hdr = cplhdr(skb) + RSS_HDR;
1524        struct chtls_sock *csk;
1525        struct tcp_sock *tp;
1526
1527        csk = rcu_dereference_sk_user_data(sk);
1528        tp = tcp_sk(sk);
1529
1530        if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1531                handle_excess_rx(sk, skb);
1532                return;
1533        }
1534
1535        ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1536        ULP_SKB_CB(skb)->psh = hdr->psh;
1537        skb_ulp_mode(skb) = ULP_MODE_NONE;
1538
1539        skb_reset_transport_header(skb);
1540        __skb_pull(skb, sizeof(*hdr) + RSS_HDR);
1541        if (!skb->data_len)
1542                __skb_trim(skb, ntohs(hdr->len));
1543
1544        if (unlikely(hdr->urg))
1545                handle_urg_ptr(sk, tp->rcv_nxt + ntohs(hdr->urg));
1546        if (unlikely(tp->urg_data == TCP_URG_NOTYET &&
1547                     tp->urg_seq - tp->rcv_nxt < skb->len))
1548                tp->urg_data = TCP_URG_VALID |
1549                               skb->data[tp->urg_seq - tp->rcv_nxt];
1550
1551        if (unlikely(hdr->dack_mode != csk->delack_mode)) {
1552                csk->delack_mode = hdr->dack_mode;
1553                csk->delack_seq = tp->rcv_nxt;
1554        }
1555
1556        tcp_hdr(skb)->fin = 0;
1557        tp->rcv_nxt += skb->len;
1558
1559        __skb_queue_tail(&sk->sk_receive_queue, skb);
1560
1561        if (!sock_flag(sk, SOCK_DEAD)) {
1562                check_sk_callbacks(csk);
1563                sk->sk_data_ready(sk);
1564        }
1565}
1566
1567static int chtls_rx_data(struct chtls_dev *cdev, struct sk_buff *skb)
1568{
1569        struct cpl_rx_data *req = cplhdr(skb) + RSS_HDR;
1570        unsigned int hwtid = GET_TID(req);
1571        struct sock *sk;
1572
1573        sk = lookup_tid(cdev->tids, hwtid);
1574        if (unlikely(!sk)) {
1575                pr_err("can't find conn. for hwtid %u.\n", hwtid);
1576                return -EINVAL;
1577        }
1578        skb_dst_set(skb, NULL);
1579        process_cpl_msg(chtls_recv_data, sk, skb);
1580        return 0;
1581}
1582
1583static void chtls_recv_pdu(struct sock *sk, struct sk_buff *skb)
1584{
1585        struct cpl_tls_data *hdr = cplhdr(skb);
1586        struct chtls_sock *csk;
1587        struct chtls_hws *tlsk;
1588        struct tcp_sock *tp;
1589
1590        csk = rcu_dereference_sk_user_data(sk);
1591        tlsk = &csk->tlshws;
1592        tp = tcp_sk(sk);
1593
1594        if (unlikely(sk->sk_shutdown & RCV_SHUTDOWN)) {
1595                handle_excess_rx(sk, skb);
1596                return;
1597        }
1598
1599        ULP_SKB_CB(skb)->seq = ntohl(hdr->seq);
1600        ULP_SKB_CB(skb)->flags = 0;
1601        skb_ulp_mode(skb) = ULP_MODE_TLS;
1602
1603        skb_reset_transport_header(skb);
1604        __skb_pull(skb, sizeof(*hdr));
1605        if (!skb->data_len)
1606                __skb_trim(skb,
1607                           CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd)));
1608
1609        if (unlikely(tp->urg_data == TCP_URG_NOTYET && tp->urg_seq -
1610                     tp->rcv_nxt < skb->len))
1611                tp->urg_data = TCP_URG_VALID |
1612                               skb->data[tp->urg_seq - tp->rcv_nxt];
1613
1614        tcp_hdr(skb)->fin = 0;
1615        tlsk->pldlen = CPL_TLS_DATA_LENGTH_G(ntohl(hdr->length_pkd));
1616        __skb_queue_tail(&tlsk->sk_recv_queue, skb);
1617}
1618
1619static int chtls_rx_pdu(struct chtls_dev *cdev, struct sk_buff *skb)
1620{
1621        struct cpl_tls_data *req = cplhdr(skb);
1622        unsigned int hwtid = GET_TID(req);
1623        struct sock *sk;
1624
1625        sk = lookup_tid(cdev->tids, hwtid);
1626        if (unlikely(!sk)) {
1627                pr_err("can't find conn. for hwtid %u.\n", hwtid);
1628                return -EINVAL;
1629        }
1630        skb_dst_set(skb, NULL);
1631        process_cpl_msg(chtls_recv_pdu, sk, skb);
1632        return 0;
1633}
1634
1635static void chtls_set_hdrlen(struct sk_buff *skb, unsigned int nlen)
1636{
1637        struct tlsrx_cmp_hdr *tls_cmp_hdr = cplhdr(skb);
1638
1639        skb->hdr_len = ntohs((__force __be16)tls_cmp_hdr->length);
1640        tls_cmp_hdr->length = ntohs((__force __be16)nlen);
1641}
1642
1643static void chtls_rx_hdr(struct sock *sk, struct sk_buff *skb)
1644{
1645        struct tlsrx_cmp_hdr *tls_hdr_pkt;
1646        struct cpl_rx_tls_cmp *cmp_cpl;
1647        struct sk_buff *skb_rec;
1648        struct chtls_sock *csk;
1649        struct chtls_hws *tlsk;
1650        struct tcp_sock *tp;
1651
1652        cmp_cpl = cplhdr(skb);
1653        csk = rcu_dereference_sk_user_data(sk);
1654        tlsk = &csk->tlshws;
1655        tp = tcp_sk(sk);
1656
1657        ULP_SKB_CB(skb)->seq = ntohl(cmp_cpl->seq);
1658        ULP_SKB_CB(skb)->flags = 0;
1659
1660        skb_reset_transport_header(skb);
1661        __skb_pull(skb, sizeof(*cmp_cpl));
1662        tls_hdr_pkt = (struct tlsrx_cmp_hdr *)skb->data;
1663        if (tls_hdr_pkt->res_to_mac_error & TLSRX_HDR_PKT_ERROR_M)
1664                tls_hdr_pkt->type = CONTENT_TYPE_ERROR;
1665        if (!skb->data_len)
1666                __skb_trim(skb, TLS_HEADER_LENGTH);
1667
1668        tp->rcv_nxt +=
1669                CPL_RX_TLS_CMP_PDULENGTH_G(ntohl(cmp_cpl->pdulength_length));
1670
1671        ULP_SKB_CB(skb)->flags |= ULPCB_FLAG_TLS_HDR;
1672        skb_rec = __skb_dequeue(&tlsk->sk_recv_queue);
1673        if (!skb_rec) {
1674                __skb_queue_tail(&sk->sk_receive_queue, skb);
1675        } else {
1676                chtls_set_hdrlen(skb, tlsk->pldlen);
1677                tlsk->pldlen = 0;
1678                __skb_queue_tail(&sk->sk_receive_queue, skb);
1679                __skb_queue_tail(&sk->sk_receive_queue, skb_rec);
1680        }
1681
1682        if (!sock_flag(sk, SOCK_DEAD)) {
1683                check_sk_callbacks(csk);
1684                sk->sk_data_ready(sk);
1685        }
1686}
1687
1688static int chtls_rx_cmp(struct chtls_dev *cdev, struct sk_buff *skb)
1689{
1690        struct cpl_rx_tls_cmp *req = cplhdr(skb);
1691        unsigned int hwtid = GET_TID(req);
1692        struct sock *sk;
1693
1694        sk = lookup_tid(cdev->tids, hwtid);
1695        if (unlikely(!sk)) {
1696                pr_err("can't find conn. for hwtid %u.\n", hwtid);
1697                return -EINVAL;
1698        }
1699        skb_dst_set(skb, NULL);
1700        process_cpl_msg(chtls_rx_hdr, sk, skb);
1701
1702        return 0;
1703}
1704
1705static void chtls_timewait(struct sock *sk)
1706{
1707        struct tcp_sock *tp = tcp_sk(sk);
1708
1709        tp->rcv_nxt++;
1710        tp->rx_opt.ts_recent_stamp = ktime_get_seconds();
1711        tp->srtt_us = 0;
1712        tcp_time_wait(sk, TCP_TIME_WAIT, 0);
1713}
1714
1715static void chtls_peer_close(struct sock *sk, struct sk_buff *skb)
1716{
1717        struct chtls_sock *csk = rcu_dereference_sk_user_data(sk);
1718
1719        sk->sk_shutdown |= RCV_SHUTDOWN;
1720        sock_set_flag(sk, SOCK_DONE);
1721
1722        switch (sk->sk_state) {
1723        case TCP_SYN_RECV:
1724        case TCP_ESTABLISHED:
1725                tcp_set_state(sk, TCP_CLOSE_WAIT);
1726                break;
1727        case TCP_FIN_WAIT1:
1728                tcp_set_state(sk, TCP_CLOSING);
1729                break;
1730        case TCP_FIN_WAIT2:
1731                chtls_release_resources(sk);
1732                if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1733                        chtls_conn_done(sk);
1734                else
1735                        chtls_timewait(sk);
1736                break;
1737        default:
1738                pr_info("cpl_peer_close in bad state %d\n", sk->sk_state);
1739        }
1740
1741        if (!sock_flag(sk, SOCK_DEAD)) {
1742                sk->sk_state_change(sk);
1743                /* Do not send POLL_HUP for half duplex close. */
1744
1745                if ((sk->sk_shutdown & SEND_SHUTDOWN) ||
1746                    sk->sk_state == TCP_CLOSE)
1747                        sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_HUP);
1748                else
1749                        sk_wake_async(sk, SOCK_WAKE_WAITD, POLL_IN);
1750        }
1751        kfree_skb(skb);
1752}
1753
1754static void chtls_close_con_rpl(struct sock *sk, struct sk_buff *skb)
1755{
1756        struct cpl_close_con_rpl *rpl = cplhdr(skb) + RSS_HDR;
1757        struct chtls_sock *csk;
1758        struct tcp_sock *tp;
1759
1760        csk = rcu_dereference_sk_user_data(sk);
1761        tp = tcp_sk(sk);
1762
1763        tp->snd_una = ntohl(rpl->snd_nxt) - 1;  /* exclude FIN */
1764
1765        switch (sk->sk_state) {
1766        case TCP_CLOSING:
1767                chtls_release_resources(sk);
1768                if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING))
1769                        chtls_conn_done(sk);
1770                else
1771                        chtls_timewait(sk);
1772                break;
1773        case TCP_LAST_ACK:
1774                chtls_release_resources(sk);
1775                chtls_conn_done(sk);
1776                break;
1777        case TCP_FIN_WAIT1:
1778                tcp_set_state(sk, TCP_FIN_WAIT2);
1779                sk->sk_shutdown |= SEND_SHUTDOWN;
1780
1781                if (!sock_flag(sk, SOCK_DEAD))
1782                        sk->sk_state_change(sk);
1783                else if (tcp_sk(sk)->linger2 < 0 &&
1784                         !csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN))
1785                        chtls_abort_conn(sk, skb);
1786                break;
1787        default:
1788                pr_info("close_con_rpl in bad state %d\n", sk->sk_state);
1789        }
1790        kfree_skb(skb);
1791}
1792
1793static struct sk_buff *get_cpl_skb(struct sk_buff *skb,
1794                                   size_t len, gfp_t gfp)
1795{
1796        if (likely(!skb_is_nonlinear(skb) && !skb_cloned(skb))) {
1797                WARN_ONCE(skb->len < len, "skb alloc error");
1798                __skb_trim(skb, len);
1799                skb_get(skb);
1800        } else {
1801                skb = alloc_skb(len, gfp);
1802                if (skb)
1803                        __skb_put(skb, len);
1804        }
1805        return skb;
1806}
1807
1808static void set_abort_rpl_wr(struct sk_buff *skb, unsigned int tid,
1809                             int cmd)
1810{
1811        struct cpl_abort_rpl *rpl = cplhdr(skb);
1812
1813        INIT_TP_WR_CPL(rpl, CPL_ABORT_RPL, tid);
1814        rpl->cmd = cmd;
1815}
1816
1817static void send_defer_abort_rpl(struct chtls_dev *cdev, struct sk_buff *skb)
1818{
1819        struct cpl_abort_req_rss *req = cplhdr(skb);
1820        struct sk_buff *reply_skb;
1821
1822        reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1823                              GFP_KERNEL | __GFP_NOFAIL);
1824        __skb_put(reply_skb, sizeof(struct cpl_abort_rpl));
1825        set_abort_rpl_wr(reply_skb, GET_TID(req),
1826                         (req->status & CPL_ABORT_NO_RST));
1827        set_wr_txq(reply_skb, CPL_PRIORITY_DATA, req->status >> 1);
1828        cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1829        kfree_skb(skb);
1830}
1831
1832/*
1833 * Add an skb to the deferred skb queue for processing from process context.
1834 */
1835static void t4_defer_reply(struct sk_buff *skb, struct chtls_dev *cdev,
1836                           defer_handler_t handler)
1837{
1838        DEFERRED_SKB_CB(skb)->handler = handler;
1839        spin_lock_bh(&cdev->deferq.lock);
1840        __skb_queue_tail(&cdev->deferq, skb);
1841        if (skb_queue_len(&cdev->deferq) == 1)
1842                schedule_work(&cdev->deferq_task);
1843        spin_unlock_bh(&cdev->deferq.lock);
1844}
1845
1846static void send_abort_rpl(struct sock *sk, struct sk_buff *skb,
1847                           struct chtls_dev *cdev, int status, int queue)
1848{
1849        struct cpl_abort_req_rss *req = cplhdr(skb);
1850        struct sk_buff *reply_skb;
1851        struct chtls_sock *csk;
1852
1853        csk = rcu_dereference_sk_user_data(sk);
1854
1855        reply_skb = alloc_skb(sizeof(struct cpl_abort_rpl),
1856                              GFP_KERNEL);
1857
1858        if (!reply_skb) {
1859                req->status = (queue << 1);
1860                t4_defer_reply(skb, cdev, send_defer_abort_rpl);
1861                return;
1862        }
1863
1864        set_abort_rpl_wr(reply_skb, GET_TID(req), status);
1865        kfree_skb(skb);
1866
1867        set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
1868        if (csk_conn_inline(csk)) {
1869                struct l2t_entry *e = csk->l2t_entry;
1870
1871                if (e && sk->sk_state != TCP_SYN_RECV) {
1872                        cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
1873                        return;
1874                }
1875        }
1876        cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1877}
1878
1879static void chtls_send_abort_rpl(struct sock *sk, struct sk_buff *skb,
1880                                 struct chtls_dev *cdev,
1881                                 int status, int queue)
1882{
1883        struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
1884        struct sk_buff *reply_skb;
1885        struct chtls_sock *csk;
1886        unsigned int tid;
1887
1888        csk = rcu_dereference_sk_user_data(sk);
1889        tid = GET_TID(req);
1890
1891        reply_skb = get_cpl_skb(skb, sizeof(struct cpl_abort_rpl), gfp_any());
1892        if (!reply_skb) {
1893                req->status = (queue << 1) | status;
1894                t4_defer_reply(skb, cdev, send_defer_abort_rpl);
1895                return;
1896        }
1897
1898        set_abort_rpl_wr(reply_skb, tid, status);
1899        set_wr_txq(reply_skb, CPL_PRIORITY_DATA, queue);
1900        if (csk_conn_inline(csk)) {
1901                struct l2t_entry *e = csk->l2t_entry;
1902
1903                if (e && sk->sk_state != TCP_SYN_RECV) {
1904                        cxgb4_l2t_send(csk->egress_dev, reply_skb, e);
1905                        return;
1906                }
1907        }
1908        cxgb4_ofld_send(cdev->lldi->ports[0], reply_skb);
1909        kfree_skb(skb);
1910}
1911
1912/*
1913 * This is run from a listener's backlog to abort a child connection in
1914 * SYN_RCV state (i.e., one on the listener's SYN queue).
1915 */
1916static void bl_abort_syn_rcv(struct sock *lsk, struct sk_buff *skb)
1917{
1918        struct chtls_sock *csk;
1919        struct sock *child;
1920        int queue;
1921
1922        child = skb->sk;
1923        csk = rcu_dereference_sk_user_data(child);
1924        queue = csk->txq_idx;
1925
1926        skb->sk = NULL;
1927        do_abort_syn_rcv(child, lsk);
1928        send_abort_rpl(child, skb, BLOG_SKB_CB(skb)->cdev,
1929                       CPL_ABORT_NO_RST, queue);
1930}
1931
1932static int abort_syn_rcv(struct sock *sk, struct sk_buff *skb)
1933{
1934        const struct request_sock *oreq;
1935        struct listen_ctx *listen_ctx;
1936        struct chtls_sock *csk;
1937        struct chtls_dev *cdev;
1938        struct sock *psk;
1939        void *ctx;
1940
1941        csk = sk->sk_user_data;
1942        oreq = csk->passive_reap_next;
1943        cdev = csk->cdev;
1944
1945        if (!oreq)
1946                return -1;
1947
1948        ctx = lookup_stid(cdev->tids, oreq->ts_recent);
1949        if (!ctx)
1950                return -1;
1951
1952        listen_ctx = (struct listen_ctx *)ctx;
1953        psk = listen_ctx->lsk;
1954
1955        bh_lock_sock(psk);
1956        if (!sock_owned_by_user(psk)) {
1957                int queue = csk->txq_idx;
1958
1959                do_abort_syn_rcv(sk, psk);
1960                send_abort_rpl(sk, skb, cdev, CPL_ABORT_NO_RST, queue);
1961        } else {
1962                skb->sk = sk;
1963                BLOG_SKB_CB(skb)->backlog_rcv = bl_abort_syn_rcv;
1964                __sk_add_backlog(psk, skb);
1965        }
1966        bh_unlock_sock(psk);
1967        return 0;
1968}
1969
1970static void chtls_abort_req_rss(struct sock *sk, struct sk_buff *skb)
1971{
1972        const struct cpl_abort_req_rss *req = cplhdr(skb) + RSS_HDR;
1973        struct chtls_sock *csk = sk->sk_user_data;
1974        int rst_status = CPL_ABORT_NO_RST;
1975        int queue = csk->txq_idx;
1976
1977        if (is_neg_adv(req->status)) {
1978                if (sk->sk_state == TCP_SYN_RECV)
1979                        chtls_set_tcb_tflag(sk, 0, 0);
1980
1981                kfree_skb(skb);
1982                return;
1983        }
1984
1985        csk_reset_flag(csk, CSK_ABORT_REQ_RCVD);
1986
1987        if (!csk_flag_nochk(csk, CSK_ABORT_SHUTDOWN) &&
1988            !csk_flag_nochk(csk, CSK_TX_DATA_SENT)) {
1989                struct tcp_sock *tp = tcp_sk(sk);
1990
1991                if (send_tx_flowc_wr(sk, 0, tp->snd_nxt, tp->rcv_nxt) < 0)
1992                        WARN_ONCE(1, "send_tx_flowc error");
1993                csk_set_flag(csk, CSK_TX_DATA_SENT);
1994        }
1995
1996        csk_set_flag(csk, CSK_ABORT_SHUTDOWN);
1997
1998        if (!csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
1999                sk->sk_err = ETIMEDOUT;
2000
2001                if (!sock_flag(sk, SOCK_DEAD))
2002                        sk->sk_error_report(sk);
2003
2004                if (sk->sk_state == TCP_SYN_RECV && !abort_syn_rcv(sk, skb))
2005                        return;
2006
2007                chtls_release_resources(sk);
2008                chtls_conn_done(sk);
2009        }
2010
2011        chtls_send_abort_rpl(sk, skb, csk->cdev, rst_status, queue);
2012}
2013
2014static void chtls_abort_rpl_rss(struct sock *sk, struct sk_buff *skb)
2015{
2016        struct cpl_abort_rpl_rss *rpl = cplhdr(skb) + RSS_HDR;
2017        struct chtls_sock *csk;
2018        struct chtls_dev *cdev;
2019
2020        csk = rcu_dereference_sk_user_data(sk);
2021        cdev = csk->cdev;
2022
2023        if (csk_flag_nochk(csk, CSK_ABORT_RPL_PENDING)) {
2024                csk_reset_flag(csk, CSK_ABORT_RPL_PENDING);
2025                if (!csk_flag_nochk(csk, CSK_ABORT_REQ_RCVD)) {
2026                        if (sk->sk_state == TCP_SYN_SENT) {
2027                                cxgb4_remove_tid(cdev->tids,
2028                                                 csk->port_id,
2029                                                 GET_TID(rpl),
2030                                                 sk->sk_family);
2031                                sock_put(sk);
2032                        }
2033                        chtls_release_resources(sk);
2034                        chtls_conn_done(sk);
2035                }
2036        }
2037        kfree_skb(skb);
2038}
2039
2040static int chtls_conn_cpl(struct chtls_dev *cdev, struct sk_buff *skb)
2041{
2042        struct cpl_peer_close *req = cplhdr(skb) + RSS_HDR;
2043        void (*fn)(struct sock *sk, struct sk_buff *skb);
2044        unsigned int hwtid = GET_TID(req);
2045        struct sock *sk;
2046        u8 opcode;
2047
2048        opcode = ((const struct rss_header *)cplhdr(skb))->opcode;
2049
2050        sk = lookup_tid(cdev->tids, hwtid);
2051        if (!sk)
2052                goto rel_skb;
2053
2054        switch (opcode) {
2055        case CPL_PEER_CLOSE:
2056                fn = chtls_peer_close;
2057                break;
2058        case CPL_CLOSE_CON_RPL:
2059                fn = chtls_close_con_rpl;
2060                break;
2061        case CPL_ABORT_REQ_RSS:
2062                fn = chtls_abort_req_rss;
2063                break;
2064        case CPL_ABORT_RPL_RSS:
2065                fn = chtls_abort_rpl_rss;
2066                break;
2067        default:
2068                goto rel_skb;
2069        }
2070
2071        process_cpl_msg(fn, sk, skb);
2072        return 0;
2073
2074rel_skb:
2075        kfree_skb(skb);
2076        return 0;
2077}
2078
2079static void chtls_rx_ack(struct sock *sk, struct sk_buff *skb)
2080{
2081        struct cpl_fw4_ack *hdr = cplhdr(skb) + RSS_HDR;
2082        struct chtls_sock *csk = sk->sk_user_data;
2083        struct tcp_sock *tp = tcp_sk(sk);
2084        u32 credits = hdr->credits;
2085        u32 snd_una;
2086
2087        snd_una = ntohl(hdr->snd_una);
2088        csk->wr_credits += credits;
2089
2090        if (csk->wr_unacked > csk->wr_max_credits - csk->wr_credits)
2091                csk->wr_unacked = csk->wr_max_credits - csk->wr_credits;
2092
2093        while (credits) {
2094                struct sk_buff *pskb = csk->wr_skb_head;
2095                u32 csum;
2096
2097                if (unlikely(!pskb)) {
2098                        if (csk->wr_nondata)
2099                                csk->wr_nondata -= credits;
2100                        break;
2101                }
2102                csum = (__force u32)pskb->csum;
2103                if (unlikely(credits < csum)) {
2104                        pskb->csum = (__force __wsum)(csum - credits);
2105                        break;
2106                }
2107                dequeue_wr(sk);
2108                credits -= csum;
2109                kfree_skb(pskb);
2110        }
2111        if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_SEQVAL) {
2112                if (unlikely(before(snd_una, tp->snd_una))) {
2113                        kfree_skb(skb);
2114                        return;
2115                }
2116
2117                if (tp->snd_una != snd_una) {
2118                        tp->snd_una = snd_una;
2119                        tp->rcv_tstamp = tcp_time_stamp(tp);
2120                        if (tp->snd_una == tp->snd_nxt &&
2121                            !csk_flag_nochk(csk, CSK_TX_FAILOVER))
2122                                csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2123                }
2124        }
2125
2126        if (hdr->seq_vld & CPL_FW4_ACK_FLAGS_CH) {
2127                unsigned int fclen16 = roundup(failover_flowc_wr_len, 16);
2128
2129                csk->wr_credits -= fclen16;
2130                csk_reset_flag(csk, CSK_TX_WAIT_IDLE);
2131                csk_reset_flag(csk, CSK_TX_FAILOVER);
2132        }
2133        if (skb_queue_len(&csk->txq) && chtls_push_frames(csk, 0))
2134                sk->sk_write_space(sk);
2135
2136        kfree_skb(skb);
2137}
2138
2139static int chtls_wr_ack(struct chtls_dev *cdev, struct sk_buff *skb)
2140{
2141        struct cpl_fw4_ack *rpl = cplhdr(skb) + RSS_HDR;
2142        unsigned int hwtid = GET_TID(rpl);
2143        struct sock *sk;
2144
2145        sk = lookup_tid(cdev->tids, hwtid);
2146        if (unlikely(!sk)) {
2147                pr_err("can't find conn. for hwtid %u.\n", hwtid);
2148                return -EINVAL;
2149        }
2150        process_cpl_msg(chtls_rx_ack, sk, skb);
2151
2152        return 0;
2153}
2154
2155chtls_handler_func chtls_handlers[NUM_CPL_CMDS] = {
2156        [CPL_PASS_OPEN_RPL]     = chtls_pass_open_rpl,
2157        [CPL_CLOSE_LISTSRV_RPL] = chtls_close_listsrv_rpl,
2158        [CPL_PASS_ACCEPT_REQ]   = chtls_pass_accept_req,
2159        [CPL_PASS_ESTABLISH]    = chtls_pass_establish,
2160        [CPL_RX_DATA]           = chtls_rx_data,
2161        [CPL_TLS_DATA]          = chtls_rx_pdu,
2162        [CPL_RX_TLS_CMP]        = chtls_rx_cmp,
2163        [CPL_PEER_CLOSE]        = chtls_conn_cpl,
2164        [CPL_CLOSE_CON_RPL]     = chtls_conn_cpl,
2165        [CPL_ABORT_REQ_RSS]     = chtls_conn_cpl,
2166        [CPL_ABORT_RPL_RSS]     = chtls_conn_cpl,
2167        [CPL_FW4_ACK]           = chtls_wr_ack,
2168};
2169