net/smc/af_smc.c

   1 /*
   2  *  Shared Memory Communications over RDMA (SMC-R) and RoCE
   3  *
   4  *  AF_SMC protocol family socket handler keeping the AF_INET sock address type
   5  *  applies to SOCK_STREAM sockets only
   6  *  offers an alternative communication option for TCP-protocol sockets
   7  *  applicable with RoCE-cards only
   8  *
   9  *  Initial restrictions:
  10  *    - IPv6 support postponed
  11  *    - support for alternate links postponed
  12  *    - partial support for non-blocking sockets only
  13  *    - support for urgent data postponed
  14  *
  15  *  Copyright IBM Corp. 2016
  16  *
  17  *  Author(s):  Ursula Braun <ubraun@linux.vnet.ibm.com>
  18  *              based on prototype from Frank Blaschka
  19  */
  20
  21 #define KMSG_COMPONENT "smc"
  22 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
  23
  24 #include <linux/module.h>
  25 #include <linux/socket.h>
  26 #include <linux/workqueue.h>
  27 #include <linux/in.h>
  28 #include <linux/sched/signal.h>
  29
  30 #include <net/sock.h>
  31 #include <net/tcp.h>
  32 #include <net/smc.h>
  33
  34 #include "smc.h"
  35 #include "smc_clc.h"
  36 #include "smc_llc.h"
  37 #include "smc_cdc.h"
  38 #include "smc_core.h"
  39 #include "smc_ib.h"
  40 #include "smc_pnet.h"
  41 #include "smc_tx.h"
  42 #include "smc_rx.h"
  43 #include "smc_close.h"
  44
  45 static DEFINE_MUTEX(smc_create_lgr_pending);    /* serialize link group
  46                                                  * creation
  47                                                  */
  48
  49 struct smc_lgr_list smc_lgr_list = {            /* established link groups */
  50         .lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
  51         .list = LIST_HEAD_INIT(smc_lgr_list.list),
  52 };
  53
  54 static void smc_tcp_listen_work(struct work_struct *);
  55
  56 static void smc_set_keepalive(struct sock *sk, int val)
  57 {
  58         struct smc_sock *smc = smc_sk(sk);
  59
  60         smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
  61 }
  62
  63 static struct smc_hashinfo smc_v4_hashinfo = {
  64         .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
  65 };
  66
  67 int smc_hash_sk(struct sock *sk)
  68 {
  69         struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
  70         struct hlist_head *head;
  71
  72         head = &h->ht;
  73
  74         write_lock_bh(&h->lock);
  75         sk_add_node(sk, head);
  76         sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
  77         write_unlock_bh(&h->lock);
  78
  79         return 0;
  80 }
  81 EXPORT_SYMBOL_GPL(smc_hash_sk);
  82
  83 void smc_unhash_sk(struct sock *sk)
  84 {
  85         struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
  86
  87         write_lock_bh(&h->lock);
  88         if (sk_del_node_init(sk))
  89                 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
  90         write_unlock_bh(&h->lock);
  91 }
  92 EXPORT_SYMBOL_GPL(smc_unhash_sk);
  93
  94 struct proto smc_proto = {
  95         .name           = "SMC",
  96         .owner          = THIS_MODULE,
  97         .keepalive      = smc_set_keepalive,
  98         .hash           = smc_hash_sk,
  99         .unhash         = smc_unhash_sk,
 100         .obj_size       = sizeof(struct smc_sock),
 101         .h.smc_hash     = &smc_v4_hashinfo,
 102         .slab_flags     = SLAB_TYPESAFE_BY_RCU,
 103 };
 104 EXPORT_SYMBOL_GPL(smc_proto);
 105
 106 static int smc_release(struct socket *sock)
 107 {
 108         struct sock *sk = sock->sk;
 109         struct smc_sock *smc;
 110         int rc = 0;
 111
 112         if (!sk)
 113                 goto out;
 114
 115         smc = smc_sk(sk);
 116         if (sk->sk_state == SMC_LISTEN)
 117                 /* smc_close_non_accepted() is called and acquires
 118                  * sock lock for child sockets again
 119                  */
 120                 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
 121         else
 122                 lock_sock(sk);
 123
 124         if (!smc->use_fallback) {
 125                 rc = smc_close_active(smc);
 126                 sock_set_flag(sk, SOCK_DEAD);
 127                 sk->sk_shutdown |= SHUTDOWN_MASK;
 128         }
 129         if (smc->clcsock) {
 130                 sock_release(smc->clcsock);
 131                 smc->clcsock = NULL;
 132         }
 133         if (smc->use_fallback) {
 134                 sock_put(sk); /* passive closing */
 135                 sk->sk_state = SMC_CLOSED;
 136                 sk->sk_state_change(sk);
 137         }
 138
 139         /* detach socket */
 140         sock_orphan(sk);
 141         sock->sk = NULL;
 142         if (!smc->use_fallback && sk->sk_state == SMC_CLOSED)
 143                 smc_conn_free(&smc->conn);
 144         release_sock(sk);
 145
 146         sk->sk_prot->unhash(sk);
 147         sock_put(sk); /* final sock_put */
 148 out:
 149         return rc;
 150 }
 151
 152 static void smc_destruct(struct sock *sk)
 153 {
 154         if (sk->sk_state != SMC_CLOSED)
 155                 return;
 156         if (!sock_flag(sk, SOCK_DEAD))
 157                 return;
 158
 159         sk_refcnt_debug_dec(sk);
 160 }
 161
 162 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock)
 163 {
 164         struct smc_sock *smc;
 165         struct sock *sk;
 166
 167         sk = sk_alloc(net, PF_SMC, GFP_KERNEL, &smc_proto, 0);
 168         if (!sk)
 169                 return NULL;
 170
 171         sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
 172         sk->sk_state = SMC_INIT;
 173         sk->sk_destruct = smc_destruct;
 174         sk->sk_protocol = SMCPROTO_SMC;
 175         smc = smc_sk(sk);
 176         INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
 177         INIT_LIST_HEAD(&smc->accept_q);
 178         spin_lock_init(&smc->accept_q_lock);
 179         sk->sk_prot->hash(sk);
 180         sk_refcnt_debug_inc(sk);
 181
 182         return sk;
 183 }
 184
 185 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
 186                     int addr_len)
 187 {
 188         struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
 189         struct sock *sk = sock->sk;
 190         struct smc_sock *smc;
 191         int rc;
 192
 193         smc = smc_sk(sk);
 194
 195         /* replicate tests from inet_bind(), to be safe wrt. future changes */
 196         rc = -EINVAL;
 197         if (addr_len < sizeof(struct sockaddr_in))
 198                 goto out;
 199
 200         rc = -EAFNOSUPPORT;
 201         /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
 202         if ((addr->sin_family != AF_INET) &&
 203             ((addr->sin_family != AF_UNSPEC) ||
 204              (addr->sin_addr.s_addr != htonl(INADDR_ANY))))
 205                 goto out;
 206
 207         lock_sock(sk);
 208
 209         /* Check if socket is already active */
 210         rc = -EINVAL;
 211         if (sk->sk_state != SMC_INIT)
 212                 goto out_rel;
 213
 214         smc->clcsock->sk->sk_reuse = sk->sk_reuse;
 215         rc = kernel_bind(smc->clcsock, uaddr, addr_len);
 216
 217 out_rel:
 218         release_sock(sk);
 219 out:
 220         return rc;
 221 }
 222
 223 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
 224                                    unsigned long mask)
 225 {
 226         /* options we don't get control via setsockopt for */
 227         nsk->sk_type = osk->sk_type;
 228         nsk->sk_sndbuf = osk->sk_sndbuf;
 229         nsk->sk_rcvbuf = osk->sk_rcvbuf;
 230         nsk->sk_sndtimeo = osk->sk_sndtimeo;
 231         nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
 232         nsk->sk_mark = osk->sk_mark;
 233         nsk->sk_priority = osk->sk_priority;
 234         nsk->sk_rcvlowat = osk->sk_rcvlowat;
 235         nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
 236         nsk->sk_err = osk->sk_err;
 237
 238         nsk->sk_flags &= ~mask;
 239         nsk->sk_flags |= osk->sk_flags & mask;
 240 }
 241
 242 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
 243                              (1UL << SOCK_KEEPOPEN) | \
 244                              (1UL << SOCK_LINGER) | \
 245                              (1UL << SOCK_BROADCAST) | \
 246                              (1UL << SOCK_TIMESTAMP) | \
 247                              (1UL << SOCK_DBG) | \
 248                              (1UL << SOCK_RCVTSTAMP) | \
 249                              (1UL << SOCK_RCVTSTAMPNS) | \
 250                              (1UL << SOCK_LOCALROUTE) | \
 251                              (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
 252                              (1UL << SOCK_RXQ_OVFL) | \
 253                              (1UL << SOCK_WIFI_STATUS) | \
 254                              (1UL << SOCK_NOFCS) | \
 255                              (1UL << SOCK_FILTER_LOCKED))
 256 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
 257  * clc socket (since smc is not called for these options from net/core)
 258  */
 259 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
 260 {
 261         smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
 262 }
 263
 264 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
 265                              (1UL << SOCK_KEEPOPEN) | \
 266                              (1UL << SOCK_LINGER) | \
 267                              (1UL << SOCK_DBG))
 268 /* copy only settings and flags relevant for smc from clc to smc socket */
 269 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
 270 {
 271         smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
 272 }
 273
 274 static int smc_clnt_conf_first_link(struct smc_sock *smc)
 275 {
 276         struct smc_link_group *lgr = smc->conn.lgr;
 277         struct smc_link *link;
 278         int rest;
 279         int rc;
 280
 281         link = &lgr->lnk[SMC_SINGLE_LINK];
 282         /* receive CONFIRM LINK request from server over RoCE fabric */
 283         rest = wait_for_completion_interruptible_timeout(
 284                 &link->llc_confirm,
 285                 SMC_LLC_WAIT_FIRST_TIME);
 286         if (rest <= 0) {
 287                 struct smc_clc_msg_decline dclc;
 288
 289                 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
 290                                       SMC_CLC_DECLINE);
 291                 return rc;
 292         }
 293
 294         if (link->llc_confirm_rc)
 295                 return SMC_CLC_DECL_RMBE_EC;
 296
 297         rc = smc_ib_modify_qp_rts(link);
 298         if (rc)
 299                 return SMC_CLC_DECL_INTERR;
 300
 301         smc_wr_remember_qp_attr(link);
 302
 303         rc = smc_wr_reg_send(link,
 304                              smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
 305         if (rc)
 306                 return SMC_CLC_DECL_INTERR;
 307
 308         /* send CONFIRM LINK response over RoCE fabric */
 309         rc = smc_llc_send_confirm_link(link,
 310                                        link->smcibdev->mac[link->ibport - 1],
 311                                        &link->smcibdev->gid[link->ibport - 1],
 312                                        SMC_LLC_RESP);
 313         if (rc < 0)
 314                 return SMC_CLC_DECL_TCL;
 315
 316         return 0;
 317 }
 318
 319 static void smc_conn_save_peer_info(struct smc_sock *smc,
 320                                     struct smc_clc_msg_accept_confirm *clc)
 321 {
 322         smc->conn.peer_conn_idx = clc->conn_idx;
 323         smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
 324         smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
 325         atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
 326 }
 327
 328 static void smc_link_save_peer_info(struct smc_link *link,
 329                                     struct smc_clc_msg_accept_confirm *clc)
 330 {
 331         link->peer_qpn = ntoh24(clc->qpn);
 332         memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
 333         memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
 334         link->peer_psn = ntoh24(clc->psn);
 335         link->peer_mtu = clc->qp_mtu;
 336 }
 337
 338 static void smc_lgr_forget(struct smc_link_group *lgr)
 339 {
 340         spin_lock_bh(&smc_lgr_list.lock);
 341         /* do not use this link group for new connections */
 342         if (!list_empty(&lgr->list))
 343                 list_del_init(&lgr->list);
 344         spin_unlock_bh(&smc_lgr_list.lock);
 345 }
 346
 347 /* setup for RDMA connection of client */
 348 static int smc_connect_rdma(struct smc_sock *smc)
 349 {
 350         struct smc_clc_msg_accept_confirm aclc;
 351         int local_contact = SMC_FIRST_CONTACT;
 352         struct smc_ib_device *smcibdev;
 353         struct smc_link *link;
 354         u8 srv_first_contact;
 355         int reason_code = 0;
 356         int rc = 0;
 357         u8 ibport;
 358
 359         sock_hold(&smc->sk); /* sock put in passive closing */
 360
 361         if (!tcp_sk(smc->clcsock->sk)->syn_smc) {
 362                 /* peer has not signalled SMC-capability */
 363                 smc->use_fallback = true;
 364                 goto out_connected;
 365         }
 366
 367         /* IPSec connections opt out of SMC-R optimizations */
 368         if (using_ipsec(smc)) {
 369                 reason_code = SMC_CLC_DECL_IPSEC;
 370                 goto decline_rdma;
 371         }
 372
 373         /* PNET table look up: search active ib_device and port
 374          * within same PNETID that also contains the ethernet device
 375          * used for the internal TCP socket
 376          */
 377         smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
 378         if (!smcibdev) {
 379                 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
 380                 goto decline_rdma;
 381         }
 382
 383         /* do inband token exchange */
 384         reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
 385         if (reason_code < 0) {
 386                 rc = reason_code;
 387                 goto out_err;
 388         }
 389         if (reason_code > 0) /* configuration error */
 390                 goto decline_rdma;
 391         /* receive SMC Accept CLC message */
 392         reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
 393                                        SMC_CLC_ACCEPT);
 394         if (reason_code < 0) {
 395                 rc = reason_code;
 396                 goto out_err;
 397         }
 398         if (reason_code > 0)
 399                 goto decline_rdma;
 400
 401         srv_first_contact = aclc.hdr.flag;
 402         mutex_lock(&smc_create_lgr_pending);
 403         local_contact = smc_conn_create(smc, smcibdev, ibport, &aclc.lcl,
 404                                         srv_first_contact);
 405         if (local_contact < 0) {
 406                 rc = local_contact;
 407                 if (rc == -ENOMEM)
 408                         reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
 409                 else if (rc == -ENOLINK)
 410                         reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
 411                 goto decline_rdma_unlock;
 412         }
 413         link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
 414
 415         smc_conn_save_peer_info(smc, &aclc);
 416
 417         /* create send buffer and rmb */
 418         rc = smc_buf_create(smc);
 419         if (rc) {
 420                 reason_code = SMC_CLC_DECL_MEM;
 421                 goto decline_rdma_unlock;
 422         }
 423
 424         if (local_contact == SMC_FIRST_CONTACT)
 425                 smc_link_save_peer_info(link, &aclc);
 426
 427         rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
 428         if (rc) {
 429                 reason_code = SMC_CLC_DECL_INTERR;
 430                 goto decline_rdma_unlock;
 431         }
 432
 433         smc_close_init(smc);
 434         smc_rx_init(smc);
 435
 436         if (local_contact == SMC_FIRST_CONTACT) {
 437                 rc = smc_ib_ready_link(link);
 438                 if (rc) {
 439                         reason_code = SMC_CLC_DECL_INTERR;
 440                         goto decline_rdma_unlock;
 441                 }
 442         } else {
 443                 struct smc_buf_desc *buf_desc = smc->conn.rmb_desc;
 444
 445                 if (!buf_desc->reused) {
 446                         /* register memory region for new rmb */
 447                         rc = smc_wr_reg_send(link,
 448                                              buf_desc->mr_rx[SMC_SINGLE_LINK]);
 449                         if (rc) {
 450                                 reason_code = SMC_CLC_DECL_INTERR;
 451                                 goto decline_rdma_unlock;
 452                         }
 453                 }
 454         }
 455         smc_rmb_sync_sg_for_device(&smc->conn);
 456
 457         rc = smc_clc_send_confirm(smc);
 458         if (rc)
 459                 goto out_err_unlock;
 460
 461         if (local_contact == SMC_FIRST_CONTACT) {
 462                 /* QP confirmation over RoCE fabric */
 463                 reason_code = smc_clnt_conf_first_link(smc);
 464                 if (reason_code < 0) {
 465                         rc = reason_code;
 466                         goto out_err_unlock;
 467                 }
 468                 if (reason_code > 0)
 469                         goto decline_rdma_unlock;
 470         }
 471
 472         mutex_unlock(&smc_create_lgr_pending);
 473         smc_tx_init(smc);
 474
 475 out_connected:
 476         smc_copy_sock_settings_to_clc(smc);
 477         if (smc->sk.sk_state == SMC_INIT)
 478                 smc->sk.sk_state = SMC_ACTIVE;
 479
 480         return rc ? rc : local_contact;
 481
 482 decline_rdma_unlock:
 483         if (local_contact == SMC_FIRST_CONTACT)
 484                 smc_lgr_forget(smc->conn.lgr);
 485         mutex_unlock(&smc_create_lgr_pending);
 486         smc_conn_free(&smc->conn);
 487 decline_rdma:
 488         /* RDMA setup failed, switch back to TCP */
 489         smc->use_fallback = true;
 490         if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
 491                 rc = smc_clc_send_decline(smc, reason_code);
 492                 if (rc < 0)
 493                         goto out_err;
 494         }
 495         goto out_connected;
 496
 497 out_err_unlock:
 498         if (local_contact == SMC_FIRST_CONTACT)
 499                 smc_lgr_forget(smc->conn.lgr);
 500         mutex_unlock(&smc_create_lgr_pending);
 501         smc_conn_free(&smc->conn);
 502 out_err:
 503         if (smc->sk.sk_state == SMC_INIT)
 504                 sock_put(&smc->sk); /* passive closing */
 505         return rc;
 506 }
 507
 508 static int smc_connect(struct socket *sock, struct sockaddr *addr,
 509                        int alen, int flags)
 510 {
 511         struct sock *sk = sock->sk;
 512         struct smc_sock *smc;
 513         int rc = -EINVAL;
 514
 515         smc = smc_sk(sk);
 516
 517         /* separate smc parameter checking to be safe */
 518         if (alen < sizeof(addr->sa_family))
 519                 goto out_err;
 520         if (addr->sa_family != AF_INET)
 521                 goto out_err;
 522
 523         lock_sock(sk);
 524         switch (sk->sk_state) {
 525         default:
 526                 goto out;
 527         case SMC_ACTIVE:
 528                 rc = -EISCONN;
 529                 goto out;
 530         case SMC_INIT:
 531                 rc = 0;
 532                 break;
 533         }
 534
 535         smc_copy_sock_settings_to_clc(smc);
 536         tcp_sk(smc->clcsock->sk)->syn_smc = 1;
 537         rc = kernel_connect(smc->clcsock, addr, alen, flags);
 538         if (rc)
 539                 goto out;
 540
 541         /* setup RDMA connection */
 542         rc = smc_connect_rdma(smc);
 543         if (rc < 0)
 544                 goto out;
 545         else
 546                 rc = 0; /* success cases including fallback */
 547
 548 out:
 549         release_sock(sk);
 550 out_err:
 551         return rc;
 552 }
 553
 554 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
 555 {
 556         struct socket *new_clcsock = NULL;
 557         struct sock *lsk = &lsmc->sk;
 558         struct sock *new_sk;
 559         int rc;
 560
 561         release_sock(lsk);
 562         new_sk = smc_sock_alloc(sock_net(lsk), NULL);
 563         if (!new_sk) {
 564                 rc = -ENOMEM;
 565                 lsk->sk_err = ENOMEM;
 566                 *new_smc = NULL;
 567                 lock_sock(lsk);
 568                 goto out;
 569         }
 570         *new_smc = smc_sk(new_sk);
 571
 572         rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
 573         lock_sock(lsk);
 574         if  (rc < 0)
 575                 lsk->sk_err = -rc;
 576         if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
 577                 if (new_clcsock)
 578                         sock_release(new_clcsock);
 579                 new_sk->sk_state = SMC_CLOSED;
 580                 sock_set_flag(new_sk, SOCK_DEAD);
 581                 new_sk->sk_prot->unhash(new_sk);
 582                 sock_put(new_sk); /* final */
 583                 *new_smc = NULL;
 584                 goto out;
 585         }
 586
 587         (*new_smc)->clcsock = new_clcsock;
 588 out:
 589         return rc;
 590 }
 591
 592 /* add a just created sock to the accept queue of the listen sock as
 593  * candidate for a following socket accept call from user space
 594  */
 595 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
 596 {
 597         struct smc_sock *par = smc_sk(parent);
 598
 599         sock_hold(sk); /* sock_put in smc_accept_unlink () */
 600         spin_lock(&par->accept_q_lock);
 601         list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
 602         spin_unlock(&par->accept_q_lock);
 603         sk_acceptq_added(parent);
 604 }
 605
 606 /* remove a socket from the accept queue of its parental listening socket */
 607 static void smc_accept_unlink(struct sock *sk)
 608 {
 609         struct smc_sock *par = smc_sk(sk)->listen_smc;
 610
 611         spin_lock(&par->accept_q_lock);
 612         list_del_init(&smc_sk(sk)->accept_q);
 613         spin_unlock(&par->accept_q_lock);
 614         sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
 615         sock_put(sk); /* sock_hold in smc_accept_enqueue */
 616 }
 617
 618 /* remove a sock from the accept queue to bind it to a new socket created
 619  * for a socket accept call from user space
 620  */
 621 struct sock *smc_accept_dequeue(struct sock *parent,
 622                                 struct socket *new_sock)
 623 {
 624         struct smc_sock *isk, *n;
 625         struct sock *new_sk;
 626
 627         list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
 628                 new_sk = (struct sock *)isk;
 629
 630                 smc_accept_unlink(new_sk);
 631                 if (new_sk->sk_state == SMC_CLOSED) {
 632                         if (isk->clcsock) {
 633                                 sock_release(isk->clcsock);
 634                                 isk->clcsock = NULL;
 635                         }
 636                         new_sk->sk_prot->unhash(new_sk);
 637                         sock_put(new_sk); /* final */
 638                         continue;
 639                 }
 640                 if (new_sock)
 641                         sock_graft(new_sk, new_sock);
 642                 return new_sk;
 643         }
 644         return NULL;
 645 }
 646
 647 /* clean up for a created but never accepted sock */
 648 void smc_close_non_accepted(struct sock *sk)
 649 {
 650         struct smc_sock *smc = smc_sk(sk);
 651
 652         lock_sock(sk);
 653         if (!sk->sk_lingertime)
 654                 /* wait for peer closing */
 655                 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
 656         if (!smc->use_fallback) {
 657                 smc_close_active(smc);
 658                 sock_set_flag(sk, SOCK_DEAD);
 659                 sk->sk_shutdown |= SHUTDOWN_MASK;
 660         }
 661         if (smc->clcsock) {
 662                 struct socket *tcp;
 663
 664                 tcp = smc->clcsock;
 665                 smc->clcsock = NULL;
 666                 sock_release(tcp);
 667         }
 668         if (smc->use_fallback) {
 669                 sock_put(sk); /* passive closing */
 670                 sk->sk_state = SMC_CLOSED;
 671         } else {
 672                 if (sk->sk_state == SMC_CLOSED)
 673                         smc_conn_free(&smc->conn);
 674         }
 675         release_sock(sk);
 676         sk->sk_prot->unhash(sk);
 677         sock_put(sk); /* final sock_put */
 678 }
 679
 680 static int smc_serv_conf_first_link(struct smc_sock *smc)
 681 {
 682         struct smc_link_group *lgr = smc->conn.lgr;
 683         struct smc_link *link;
 684         int rest;
 685         int rc;
 686
 687         link = &lgr->lnk[SMC_SINGLE_LINK];
 688
 689         rc = smc_wr_reg_send(link,
 690                              smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
 691         if (rc)
 692                 return SMC_CLC_DECL_INTERR;
 693
 694         /* send CONFIRM LINK request to client over the RoCE fabric */
 695         rc = smc_llc_send_confirm_link(link,
 696                                        link->smcibdev->mac[link->ibport - 1],
 697                                        &link->smcibdev->gid[link->ibport - 1],
 698                                        SMC_LLC_REQ);
 699         if (rc < 0)
 700                 return SMC_CLC_DECL_TCL;
 701
 702         /* receive CONFIRM LINK response from client over the RoCE fabric */
 703         rest = wait_for_completion_interruptible_timeout(
 704                 &link->llc_confirm_resp,
 705                 SMC_LLC_WAIT_FIRST_TIME);
 706         if (rest <= 0) {
 707                 struct smc_clc_msg_decline dclc;
 708
 709                 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
 710                                       SMC_CLC_DECLINE);
 711                 return rc;
 712         }
 713
 714         if (link->llc_confirm_resp_rc)
 715                 return SMC_CLC_DECL_RMBE_EC;
 716
 717         return 0;
 718 }
 719
 720 /* setup for RDMA connection of server */
 721 static void smc_listen_work(struct work_struct *work)
 722 {
 723         struct smc_sock *new_smc = container_of(work, struct smc_sock,
 724                                                 smc_listen_work);
 725         struct smc_clc_msg_proposal_prefix *pclc_prfx;
 726         struct socket *newclcsock = new_smc->clcsock;
 727         struct smc_sock *lsmc = new_smc->listen_smc;
 728         struct smc_clc_msg_accept_confirm cclc;
 729         int local_contact = SMC_REUSE_CONTACT;
 730         struct sock *newsmcsk = &new_smc->sk;
 731         struct smc_clc_msg_proposal *pclc;
 732         struct smc_ib_device *smcibdev;
 733         u8 buf[SMC_CLC_MAX_LEN];
 734         struct smc_link *link;
 735         int reason_code = 0;
 736         int rc = 0;
 737         __be32 subnet;
 738         u8 prefix_len;
 739         u8 ibport;
 740
 741         /* check if peer is smc capable */
 742         if (!tcp_sk(newclcsock->sk)->syn_smc) {
 743                 new_smc->use_fallback = true;
 744                 goto out_connected;
 745         }
 746
 747         /* do inband token exchange -
 748          *wait for and receive SMC Proposal CLC message
 749          */
 750         reason_code = smc_clc_wait_msg(new_smc, &buf, sizeof(buf),
 751                                        SMC_CLC_PROPOSAL);
 752         if (reason_code < 0)
 753                 goto out_err;
 754         if (reason_code > 0)
 755                 goto decline_rdma;
 756
 757         /* IPSec connections opt out of SMC-R optimizations */
 758         if (using_ipsec(new_smc)) {
 759                 reason_code = SMC_CLC_DECL_IPSEC;
 760                 goto decline_rdma;
 761         }
 762
 763         /* PNET table look up: search active ib_device and port
 764          * within same PNETID that also contains the ethernet device
 765          * used for the internal TCP socket
 766          */
 767         smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
 768         if (!smcibdev) {
 769                 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
 770                 goto decline_rdma;
 771         }
 772
 773         /* determine subnet and mask from internal TCP socket */
 774         rc = smc_clc_netinfo_by_tcpsk(newclcsock, &subnet, &prefix_len);
 775         if (rc) {
 776                 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
 777                 goto decline_rdma;
 778         }
 779
 780         pclc = (struct smc_clc_msg_proposal *)&buf;
 781         pclc_prfx = smc_clc_proposal_get_prefix(pclc);
 782         if (pclc_prfx->outgoing_subnet != subnet ||
 783             pclc_prfx->prefix_len != prefix_len) {
 784                 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
 785                 goto decline_rdma;
 786         }
 787
 788         /* allocate connection / link group */
 789         mutex_lock(&smc_create_lgr_pending);
 790         local_contact = smc_conn_create(new_smc, smcibdev, ibport, &pclc->lcl,
 791                                         0);
 792         if (local_contact < 0) {
 793                 rc = local_contact;
 794                 if (rc == -ENOMEM)
 795                         reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
 796                 goto decline_rdma_unlock;
 797         }
 798         link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
 799
 800         /* create send buffer and rmb */
 801         rc = smc_buf_create(new_smc);
 802         if (rc) {
 803                 reason_code = SMC_CLC_DECL_MEM;
 804                 goto decline_rdma_unlock;
 805         }
 806
 807         smc_close_init(new_smc);
 808         smc_rx_init(new_smc);
 809
 810         if (local_contact != SMC_FIRST_CONTACT) {
 811                 struct smc_buf_desc *buf_desc = new_smc->conn.rmb_desc;
 812
 813                 if (!buf_desc->reused) {
 814                         /* register memory region for new rmb */
 815                         rc = smc_wr_reg_send(link,
 816                                              buf_desc->mr_rx[SMC_SINGLE_LINK]);
 817                         if (rc) {
 818                                 reason_code = SMC_CLC_DECL_INTERR;
 819                                 goto decline_rdma_unlock;
 820                         }
 821                 }
 822         }
 823         smc_rmb_sync_sg_for_device(&new_smc->conn);
 824
 825         rc = smc_clc_send_accept(new_smc, local_contact);
 826         if (rc)
 827                 goto out_err_unlock;
 828
 829         /* receive SMC Confirm CLC message */
 830         reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
 831                                        SMC_CLC_CONFIRM);
 832         if (reason_code < 0)
 833                 goto out_err_unlock;
 834         if (reason_code > 0)
 835                 goto decline_rdma_unlock;
 836         smc_conn_save_peer_info(new_smc, &cclc);
 837         if (local_contact == SMC_FIRST_CONTACT)
 838                 smc_link_save_peer_info(link, &cclc);
 839
 840         rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
 841         if (rc) {
 842                 reason_code = SMC_CLC_DECL_INTERR;
 843                 goto decline_rdma_unlock;
 844         }
 845
 846         if (local_contact == SMC_FIRST_CONTACT) {
 847                 rc = smc_ib_ready_link(link);
 848                 if (rc) {
 849                         reason_code = SMC_CLC_DECL_INTERR;
 850                         goto decline_rdma_unlock;
 851                 }
 852                 /* QP confirmation over RoCE fabric */
 853                 reason_code = smc_serv_conf_first_link(new_smc);
 854                 if (reason_code < 0)
 855                         /* peer is not aware of a problem */
 856                         goto out_err_unlock;
 857                 if (reason_code > 0)
 858                         goto decline_rdma_unlock;
 859         }
 860
 861         smc_tx_init(new_smc);
 862         mutex_unlock(&smc_create_lgr_pending);
 863
 864 out_connected:
 865         sk_refcnt_debug_inc(newsmcsk);
 866         if (newsmcsk->sk_state == SMC_INIT)
 867                 newsmcsk->sk_state = SMC_ACTIVE;
 868 enqueue:
 869         lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
 870         if (lsmc->sk.sk_state == SMC_LISTEN) {
 871                 smc_accept_enqueue(&lsmc->sk, newsmcsk);
 872         } else { /* no longer listening */
 873                 smc_close_non_accepted(newsmcsk);
 874         }
 875         release_sock(&lsmc->sk);
 876
 877         /* Wake up accept */
 878         lsmc->sk.sk_data_ready(&lsmc->sk);
 879         sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
 880         return;
 881
 882 decline_rdma_unlock:
 883         if (local_contact == SMC_FIRST_CONTACT)
 884                 smc_lgr_forget(new_smc->conn.lgr);
 885         mutex_unlock(&smc_create_lgr_pending);
 886 decline_rdma:
 887         /* RDMA setup failed, switch back to TCP */
 888         smc_conn_free(&new_smc->conn);
 889         new_smc->use_fallback = true;
 890         if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
 891                 if (smc_clc_send_decline(new_smc, reason_code) < 0)
 892                         goto out_err;
 893         }
 894         goto out_connected;
 895
 896 out_err_unlock:
 897         if (local_contact == SMC_FIRST_CONTACT)
 898                 smc_lgr_forget(new_smc->conn.lgr);
 899         mutex_unlock(&smc_create_lgr_pending);
 900 out_err:
 901         if (newsmcsk->sk_state == SMC_INIT)
 902                 sock_put(&new_smc->sk); /* passive closing */
 903         newsmcsk->sk_state = SMC_CLOSED;
 904         smc_conn_free(&new_smc->conn);
 905         goto enqueue; /* queue new sock with sk_err set */
 906 }
 907
 908 static void smc_tcp_listen_work(struct work_struct *work)
 909 {
 910         struct smc_sock *lsmc = container_of(work, struct smc_sock,
 911                                              tcp_listen_work);
 912         struct sock *lsk = &lsmc->sk;
 913         struct smc_sock *new_smc;
 914         int rc = 0;
 915
 916         lock_sock(lsk);
 917         while (lsk->sk_state == SMC_LISTEN) {
 918                 rc = smc_clcsock_accept(lsmc, &new_smc);
 919                 if (rc)
 920                         goto out;
 921                 if (!new_smc)
 922                         continue;
 923
 924                 new_smc->listen_smc = lsmc;
 925                 new_smc->use_fallback = false; /* assume rdma capability first*/
 926                 sock_hold(lsk); /* sock_put in smc_listen_work */
 927                 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
 928                 smc_copy_sock_settings_to_smc(new_smc);
 929                 sock_hold(&new_smc->sk); /* sock_put in passive closing */
 930                 if (!schedule_work(&new_smc->smc_listen_work))
 931                         sock_put(&new_smc->sk);
 932         }
 933
 934 out:
 935         if (lsmc->clcsock) {
 936                 sock_release(lsmc->clcsock);
 937                 lsmc->clcsock = NULL;
 938         }
 939         release_sock(lsk);
 940         /* no more listening, wake up smc_close_wait_listen_clcsock and
 941          * accept
 942          */
 943         lsk->sk_state_change(lsk);
 944         sock_put(&lsmc->sk); /* sock_hold in smc_listen */
 945 }
 946
 947 static int smc_listen(struct socket *sock, int backlog)
 948 {
 949         struct sock *sk = sock->sk;
 950         struct smc_sock *smc;
 951         int rc;
 952
 953         smc = smc_sk(sk);
 954         lock_sock(sk);
 955
 956         rc = -EINVAL;
 957         if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
 958                 goto out;
 959
 960         rc = 0;
 961         if (sk->sk_state == SMC_LISTEN) {
 962                 sk->sk_max_ack_backlog = backlog;
 963                 goto out;
 964         }
 965         /* some socket options are handled in core, so we could not apply
 966          * them to the clc socket -- copy smc socket options to clc socket
 967          */
 968         smc_copy_sock_settings_to_clc(smc);
 969         tcp_sk(smc->clcsock->sk)->syn_smc = 1;
 970
 971         rc = kernel_listen(smc->clcsock, backlog);
 972         if (rc)
 973                 goto out;
 974         sk->sk_max_ack_backlog = backlog;
 975         sk->sk_ack_backlog = 0;
 976         sk->sk_state = SMC_LISTEN;
 977         INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
 978         sock_hold(sk); /* sock_hold in tcp_listen_worker */
 979         if (!schedule_work(&smc->tcp_listen_work))
 980                 sock_put(sk);
 981
 982 out:
 983         release_sock(sk);
 984         return rc;
 985 }
 986
 987 static int smc_accept(struct socket *sock, struct socket *new_sock,
 988                       int flags, bool kern)
 989 {
 990         struct sock *sk = sock->sk, *nsk;
 991         DECLARE_WAITQUEUE(wait, current);
 992         struct smc_sock *lsmc;
 993         long timeo;
 994         int rc = 0;
 995
 996         lsmc = smc_sk(sk);
 997         sock_hold(sk); /* sock_put below */
 998         lock_sock(sk);
 999
1000         if (lsmc->sk.sk_state != SMC_LISTEN) {
1001                 rc = -EINVAL;
1002                 goto out;
1003         }
1004
1005         /* Wait for an incoming connection */
1006         timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1007         add_wait_queue_exclusive(sk_sleep(sk), &wait);
1008         while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1009                 set_current_state(TASK_INTERRUPTIBLE);
1010                 if (!timeo) {
1011                         rc = -EAGAIN;
1012                         break;
1013                 }
1014                 release_sock(sk);
1015                 timeo = schedule_timeout(timeo);
1016                 /* wakeup by sk_data_ready in smc_listen_work() */
1017                 sched_annotate_sleep();
1018                 lock_sock(sk);
1019                 if (signal_pending(current)) {
1020                         rc = sock_intr_errno(timeo);
1021                         break;
1022                 }
1023         }
1024         set_current_state(TASK_RUNNING);
1025         remove_wait_queue(sk_sleep(sk), &wait);
1026
1027         if (!rc)
1028                 rc = sock_error(nsk);
1029
1030 out:
1031         release_sock(sk);
1032         sock_put(sk); /* sock_hold above */
1033         return rc;
1034 }
1035
1036 static int smc_getname(struct socket *sock, struct sockaddr *addr,
1037                        int peer)
1038 {
1039         struct smc_sock *smc;
1040
1041         if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1042             (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
1043                 return -ENOTCONN;
1044
1045         smc = smc_sk(sock->sk);
1046
1047         return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
1048 }
1049
1050 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1051 {
1052         struct sock *sk = sock->sk;
1053         struct smc_sock *smc;
1054         int rc = -EPIPE;
1055
1056         smc = smc_sk(sk);
1057         lock_sock(sk);
1058         if ((sk->sk_state != SMC_ACTIVE) &&
1059             (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1060             (sk->sk_state != SMC_INIT))
1061                 goto out;
1062         if (smc->use_fallback)
1063                 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1064         else
1065                 rc = smc_tx_sendmsg(smc, msg, len);
1066 out:
1067         release_sock(sk);
1068         return rc;
1069 }
1070
1071 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1072                        int flags)
1073 {
1074         struct sock *sk = sock->sk;
1075         struct smc_sock *smc;
1076         int rc = -ENOTCONN;
1077
1078         smc = smc_sk(sk);
1079         lock_sock(sk);
1080         if ((sk->sk_state == SMC_INIT) ||
1081             (sk->sk_state == SMC_LISTEN) ||
1082             (sk->sk_state == SMC_CLOSED))
1083                 goto out;
1084
1085         if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1086                 rc = 0;
1087                 goto out;
1088         }
1089
1090         if (smc->use_fallback)
1091                 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1092         else
1093                 rc = smc_rx_recvmsg(smc, msg, len, flags);
1094
1095 out:
1096         release_sock(sk);
1097         return rc;
1098 }
1099
1100 static __poll_t smc_accept_poll(struct sock *parent)
1101 {
1102         struct smc_sock *isk = smc_sk(parent);
1103         __poll_t mask = 0;
1104
1105         spin_lock(&isk->accept_q_lock);
1106         if (!list_empty(&isk->accept_q))
1107                 mask = EPOLLIN | EPOLLRDNORM;
1108         spin_unlock(&isk->accept_q_lock);
1109
1110         return mask;
1111 }
1112
1113 static __poll_t smc_poll(struct file *file, struct socket *sock,
1114                              poll_table *wait)
1115 {
1116         struct sock *sk = sock->sk;
1117         __poll_t mask = 0;
1118         struct smc_sock *smc;
1119         int rc;
1120
1121         if (!sk)
1122                 return EPOLLNVAL;
1123
1124         smc = smc_sk(sock->sk);
1125         sock_hold(sk);
1126         lock_sock(sk);
1127         if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1128                 /* delegate to CLC child sock */
1129                 release_sock(sk);
1130                 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1131                 /* if non-blocking connect finished ... */
1132                 lock_sock(sk);
1133                 if ((sk->sk_state == SMC_INIT) && (mask & EPOLLOUT)) {
1134                         sk->sk_err = smc->clcsock->sk->sk_err;
1135                         if (sk->sk_err) {
1136                                 mask |= EPOLLERR;
1137                         } else {
1138                                 rc = smc_connect_rdma(smc);
1139                                 if (rc < 0)
1140                                         mask |= EPOLLERR;
1141                                 /* success cases including fallback */
1142                                 mask |= EPOLLOUT | EPOLLWRNORM;
1143                         }
1144                 }
1145         } else {
1146                 if (sk->sk_state != SMC_CLOSED) {
1147                         release_sock(sk);
1148                         sock_poll_wait(file, sk_sleep(sk), wait);
1149                         lock_sock(sk);
1150                 }
1151                 if (sk->sk_err)
1152                         mask |= EPOLLERR;
1153                 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1154                     (sk->sk_state == SMC_CLOSED))
1155                         mask |= EPOLLHUP;
1156                 if (sk->sk_state == SMC_LISTEN) {
1157                         /* woken up by sk_data_ready in smc_listen_work() */
1158                         mask = smc_accept_poll(sk);
1159                 } else {
1160                         if (atomic_read(&smc->conn.sndbuf_space) ||
1161                             sk->sk_shutdown & SEND_SHUTDOWN) {
1162                                 mask |= EPOLLOUT | EPOLLWRNORM;
1163                         } else {
1164                                 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1165                                 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1166                         }
1167                         if (atomic_read(&smc->conn.bytes_to_rcv))
1168                                 mask |= EPOLLIN | EPOLLRDNORM;
1169                         if (sk->sk_shutdown & RCV_SHUTDOWN)
1170                                 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
1171                         if (sk->sk_state == SMC_APPCLOSEWAIT1)
1172                                 mask |= EPOLLIN;
1173                 }
1174
1175         }
1176         release_sock(sk);
1177         sock_put(sk);
1178
1179         return mask;
1180 }
1181
1182 static int smc_shutdown(struct socket *sock, int how)
1183 {
1184         struct sock *sk = sock->sk;
1185         struct smc_sock *smc;
1186         int rc = -EINVAL;
1187         int rc1 = 0;
1188
1189         smc = smc_sk(sk);
1190
1191         if ((how < SHUT_RD) || (how > SHUT_RDWR))
1192                 return rc;
1193
1194         lock_sock(sk);
1195
1196         rc = -ENOTCONN;
1197         if ((sk->sk_state != SMC_LISTEN) &&
1198             (sk->sk_state != SMC_ACTIVE) &&
1199             (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1200             (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1201             (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1202             (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1203             (sk->sk_state != SMC_APPFINCLOSEWAIT))
1204                 goto out;
1205         if (smc->use_fallback) {
1206                 rc = kernel_sock_shutdown(smc->clcsock, how);
1207                 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1208                 if (sk->sk_shutdown == SHUTDOWN_MASK)
1209                         sk->sk_state = SMC_CLOSED;
1210                 goto out;
1211         }
1212         switch (how) {
1213         case SHUT_RDWR:         /* shutdown in both directions */
1214                 rc = smc_close_active(smc);
1215                 break;
1216         case SHUT_WR:
1217                 rc = smc_close_shutdown_write(smc);
1218                 break;
1219         case SHUT_RD:
1220                 if (sk->sk_state == SMC_LISTEN)
1221                         rc = smc_close_active(smc);
1222                 else
1223                         rc = 0;
1224                         /* nothing more to do because peer is not involved */
1225                 break;
1226         }
1227         rc1 = kernel_sock_shutdown(smc->clcsock, how);
1228         /* map sock_shutdown_cmd constants to sk_shutdown value range */
1229         sk->sk_shutdown |= how + 1;
1230
1231 out:
1232         release_sock(sk);
1233         return rc ? rc : rc1;
1234 }
1235
1236 static int smc_setsockopt(struct socket *sock, int level, int optname,
1237                           char __user *optval, unsigned int optlen)
1238 {
1239         struct sock *sk = sock->sk;
1240         struct smc_sock *smc;
1241
1242         smc = smc_sk(sk);
1243
1244         /* generic setsockopts reaching us here always apply to the
1245          * CLC socket
1246          */
1247         return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1248                                              optval, optlen);
1249 }
1250
1251 static int smc_getsockopt(struct socket *sock, int level, int optname,
1252                           char __user *optval, int __user *optlen)
1253 {
1254         struct smc_sock *smc;
1255
1256         smc = smc_sk(sock->sk);
1257         /* socket options apply to the CLC socket */
1258         return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1259                                              optval, optlen);
1260 }
1261
1262 static int smc_ioctl(struct socket *sock, unsigned int cmd,
1263                      unsigned long arg)
1264 {
1265         struct smc_sock *smc;
1266
1267         smc = smc_sk(sock->sk);
1268         if (smc->use_fallback)
1269                 return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1270         else
1271                 return sock_no_ioctl(sock, cmd, arg);
1272 }
1273
1274 static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1275                             int offset, size_t size, int flags)
1276 {
1277         struct sock *sk = sock->sk;
1278         struct smc_sock *smc;
1279         int rc = -EPIPE;
1280
1281         smc = smc_sk(sk);
1282         lock_sock(sk);
1283         if (sk->sk_state != SMC_ACTIVE)
1284                 goto out;
1285         if (smc->use_fallback)
1286                 rc = kernel_sendpage(smc->clcsock, page, offset,
1287                                      size, flags);
1288         else
1289                 rc = sock_no_sendpage(sock, page, offset, size, flags);
1290
1291 out:
1292         release_sock(sk);
1293         return rc;
1294 }
1295
1296 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1297                                struct pipe_inode_info *pipe, size_t len,
1298                                     unsigned int flags)
1299 {
1300         struct sock *sk = sock->sk;
1301         struct smc_sock *smc;
1302         int rc = -ENOTCONN;
1303
1304         smc = smc_sk(sk);
1305         lock_sock(sk);
1306         if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1307                 goto out;
1308         if (smc->use_fallback) {
1309                 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1310                                                     pipe, len, flags);
1311         } else {
1312                 rc = -EOPNOTSUPP;
1313         }
1314 out:
1315         release_sock(sk);
1316         return rc;
1317 }
1318
1319 /* must look like tcp */
1320 static const struct proto_ops smc_sock_ops = {
1321         .family         = PF_SMC,
1322         .owner          = THIS_MODULE,
1323         .release        = smc_release,
1324         .bind           = smc_bind,
1325         .connect        = smc_connect,
1326         .socketpair     = sock_no_socketpair,
1327         .accept         = smc_accept,
1328         .getname        = smc_getname,
1329         .poll           = smc_poll,
1330         .ioctl          = smc_ioctl,
1331         .listen         = smc_listen,
1332         .shutdown       = smc_shutdown,
1333         .setsockopt     = smc_setsockopt,
1334         .getsockopt     = smc_getsockopt,
1335         .sendmsg        = smc_sendmsg,
1336         .recvmsg        = smc_recvmsg,
1337         .mmap           = sock_no_mmap,
1338         .sendpage       = smc_sendpage,
1339         .splice_read    = smc_splice_read,
1340 };
1341
1342 static int smc_create(struct net *net, struct socket *sock, int protocol,
1343                       int kern)
1344 {
1345         struct smc_sock *smc;
1346         struct sock *sk;
1347         int rc;
1348
1349         rc = -ESOCKTNOSUPPORT;
1350         if (sock->type != SOCK_STREAM)
1351                 goto out;
1352
1353         rc = -EPROTONOSUPPORT;
1354         if ((protocol != IPPROTO_IP) && (protocol != IPPROTO_TCP))
1355                 goto out;
1356
1357         rc = -ENOBUFS;
1358         sock->ops = &smc_sock_ops;
1359         sk = smc_sock_alloc(net, sock);
1360         if (!sk)
1361                 goto out;
1362
1363         /* create internal TCP socket for CLC handshake and fallback */
1364         smc = smc_sk(sk);
1365         smc->use_fallback = false; /* assume rdma capability first */
1366         rc = sock_create_kern(net, PF_INET, SOCK_STREAM,
1367                               IPPROTO_TCP, &smc->clcsock);
1368         if (rc)
1369                 sk_common_release(sk);
1370         smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1371         smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
1372
1373 out:
1374         return rc;
1375 }
1376
1377 static const struct net_proto_family smc_sock_family_ops = {
1378         .family = PF_SMC,
1379         .owner  = THIS_MODULE,
1380         .create = smc_create,
1381 };
1382
1383 static int __init smc_init(void)
1384 {
1385         int rc;
1386
1387         rc = smc_pnet_init();
1388         if (rc)
1389                 return rc;
1390
1391         rc = smc_llc_init();
1392         if (rc) {
1393                 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1394                 goto out_pnet;
1395         }
1396
1397         rc = smc_cdc_init();
1398         if (rc) {
1399                 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1400                 goto out_pnet;
1401         }
1402
1403         rc = proto_register(&smc_proto, 1);
1404         if (rc) {
1405                 pr_err("%s: proto_register fails with %d\n", __func__, rc);
1406                 goto out_pnet;
1407         }
1408
1409         rc = sock_register(&smc_sock_family_ops);
1410         if (rc) {
1411                 pr_err("%s: sock_register fails with %d\n", __func__, rc);
1412                 goto out_proto;
1413         }
1414         INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
1415
1416         rc = smc_ib_register_client();
1417         if (rc) {
1418                 pr_err("%s: ib_register fails with %d\n", __func__, rc);
1419                 goto out_sock;
1420         }
1421
1422         static_branch_enable(&tcp_have_smc);
1423         return 0;
1424
1425 out_sock:
1426         sock_unregister(PF_SMC);
1427 out_proto:
1428         proto_unregister(&smc_proto);
1429 out_pnet:
1430         smc_pnet_exit();
1431         return rc;
1432 }
1433
1434 static void __exit smc_exit(void)
1435 {
1436         struct smc_link_group *lgr, *lg;
1437         LIST_HEAD(lgr_freeing_list);
1438
1439         spin_lock_bh(&smc_lgr_list.lock);
1440         if (!list_empty(&smc_lgr_list.list))
1441                 list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1442         spin_unlock_bh(&smc_lgr_list.lock);
1443         list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1444                 list_del_init(&lgr->list);
1445                 smc_lgr_free(lgr); /* free link group */
1446         }
1447         static_branch_disable(&tcp_have_smc);
1448         smc_ib_unregister_client();
1449         sock_unregister(PF_SMC);
1450         proto_unregister(&smc_proto);
1451         smc_pnet_exit();
1452 }
1453
1454 module_init(smc_init);
1455 module_exit(smc_exit);
1456
1457 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1458 MODULE_DESCRIPTION("smc socket address family");
1459 MODULE_LICENSE("GPL");
1460 MODULE_ALIAS_NETPROTO(PF_SMC);