1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type 6 * applies to SOCK_STREAM sockets only 7 * offers an alternative communication option for TCP-protocol sockets 8 * applicable with RoCE-cards only 9 * 10 * Initial restrictions: 11 * - support for alternate links postponed 12 * 13 * Copyright IBM Corp. 2016, 2018 14 * 15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 16 * based on prototype from Frank Blaschka 17 */ 18 19 #define pr_fmt(fmt) "smc: " fmt 20 21 #include <linux/module.h> 22 #include <linux/socket.h> 23 #include <linux/workqueue.h> 24 #include <linux/in.h> 25 #include <linux/sched/signal.h> 26 #include <linux/if_vlan.h> 27 #include <linux/rcupdate_wait.h> 28 #include <linux/ctype.h> 29 #include <linux/splice.h> 30 31 #include <net/sock.h> 32 #include <net/inet_common.h> 33 #if IS_ENABLED(CONFIG_IPV6) 34 #include <net/ipv6.h> 35 #endif 36 #include <net/tcp.h> 37 #include <net/smc.h> 38 #include <asm/ioctls.h> 39 40 #include <net/net_namespace.h> 41 #include <net/netns/generic.h> 42 #include "smc_netns.h" 43 44 #include "smc.h" 45 #include "smc_clc.h" 46 #include "smc_llc.h" 47 #include "smc_cdc.h" 48 #include "smc_core.h" 49 #include "smc_ib.h" 50 #include "smc_ism.h" 51 #include "smc_pnet.h" 52 #include "smc_netlink.h" 53 #include "smc_tx.h" 54 #include "smc_rx.h" 55 #include "smc_close.h" 56 #include "smc_stats.h" 57 #include "smc_tracepoint.h" 58 #include "smc_sysctl.h" 59 #include "smc_inet.h" 60 #include "smc_hs_bpf.h" 61 62 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 63 * creation on server 64 */ 65 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 66 * creation on client 67 */ 68 69 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 70 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 71 struct workqueue_struct *smc_close_wq; /* wq for close work */ 72 73 static void smc_tcp_listen_work(struct work_struct *); 74 static void smc_connect_work(struct work_struct *); 75 76 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 77 { 78 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 79 void *hdr; 80 81 if (cb_ctx->pos[0]) 82 goto out; 83 84 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 85 &smc_gen_nl_family, NLM_F_MULTI, 86 SMC_NETLINK_DUMP_HS_LIMITATION); 87 if (!hdr) 88 return -ENOMEM; 89 90 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 91 sock_net(skb->sk)->smc.limit_smc_hs)) 92 goto err; 93 94 genlmsg_end(skb, hdr); 95 cb_ctx->pos[0] = 1; 96 out: 97 return skb->len; 98 err: 99 genlmsg_cancel(skb, hdr); 100 return -EMSGSIZE; 101 } 102 103 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 104 { 105 sock_net(skb->sk)->smc.limit_smc_hs = true; 106 return 0; 107 } 108 109 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 110 { 111 sock_net(skb->sk)->smc.limit_smc_hs = false; 112 return 0; 113 } 114 115 static void smc_set_keepalive(struct sock *sk, int val) 116 { 117 struct smc_sock *smc = smc_sk(sk); 118 119 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 120 } 121 122 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 123 struct sk_buff *skb, 124 struct request_sock *req, 125 struct dst_entry *dst, 126 struct request_sock *req_unhash, 127 bool *own_req, 128 void (*opt_child_init)(struct sock *newsk, 129 const struct sock *sk)) 130 { 131 struct smc_sock *smc; 132 struct sock *child; 133 134 rcu_read_lock(); 135 smc = smc_clcsock_user_data_rcu(sk); 136 if (!smc || !refcount_inc_not_zero(&smc->sk.sk_refcnt)) { 137 rcu_read_unlock(); 138 smc = NULL; 139 goto drop; 140 } 141 rcu_read_unlock(); 142 143 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 144 sk->sk_max_ack_backlog) 145 goto drop; 146 147 if (sk_acceptq_is_full(&smc->sk)) { 148 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 149 goto drop; 150 } 151 152 /* passthrough to original syn recv sock fct */ 153 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 154 own_req, opt_child_init); 155 /* child must not inherit smc or its ops */ 156 if (child) { 157 rcu_assign_sk_user_data(child, NULL); 158 159 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 160 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 161 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 162 } 163 sock_put(&smc->sk); 164 return child; 165 166 drop: 167 dst_release(dst); 168 tcp_listendrop(sk); 169 if (smc) 170 sock_put(&smc->sk); 171 return NULL; 172 } 173 174 static bool smc_hs_congested(const struct sock *sk) 175 { 176 const struct smc_sock *smc; 177 178 smc = smc_clcsock_user_data(sk); 179 180 if (!smc) 181 return true; 182 183 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 184 return true; 185 186 return false; 187 } 188 189 struct smc_hashinfo smc_v4_hashinfo = { 190 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 191 .ht = HLIST_HEAD_INIT, 192 }; 193 194 struct smc_hashinfo smc_v6_hashinfo = { 195 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 196 .ht = HLIST_HEAD_INIT, 197 }; 198 199 int smc_hash_sk(struct sock *sk) 200 { 201 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 202 struct hlist_head *head; 203 204 head = &h->ht; 205 206 write_lock_bh(&h->lock); 207 sk_add_node(sk, head); 208 write_unlock_bh(&h->lock); 209 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 210 211 return 0; 212 } 213 214 void smc_unhash_sk(struct sock *sk) 215 { 216 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 217 218 write_lock_bh(&h->lock); 219 if (sk_del_node_init(sk)) 220 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 221 write_unlock_bh(&h->lock); 222 } 223 224 /* This will be called before user really release sock_lock. So do the 225 * work which we didn't do because of user hold the sock_lock in the 226 * BH context 227 */ 228 void smc_release_cb(struct sock *sk) 229 { 230 struct smc_sock *smc = smc_sk(sk); 231 232 if (smc->conn.tx_in_release_sock) { 233 smc_tx_pending(&smc->conn); 234 smc->conn.tx_in_release_sock = false; 235 } 236 } 237 238 struct proto smc_proto = { 239 .name = "SMC", 240 .owner = THIS_MODULE, 241 .keepalive = smc_set_keepalive, 242 .hash = smc_hash_sk, 243 .unhash = smc_unhash_sk, 244 .release_cb = smc_release_cb, 245 .obj_size = sizeof(struct smc_sock), 246 .h.smc_hash = &smc_v4_hashinfo, 247 .slab_flags = SLAB_TYPESAFE_BY_RCU, 248 }; 249 EXPORT_SYMBOL_GPL(smc_proto); 250 251 struct proto smc_proto6 = { 252 .name = "SMC6", 253 .owner = THIS_MODULE, 254 .keepalive = smc_set_keepalive, 255 .hash = smc_hash_sk, 256 .unhash = smc_unhash_sk, 257 .release_cb = smc_release_cb, 258 .obj_size = sizeof(struct smc_sock), 259 .h.smc_hash = &smc_v6_hashinfo, 260 .slab_flags = SLAB_TYPESAFE_BY_RCU, 261 }; 262 EXPORT_SYMBOL_GPL(smc_proto6); 263 264 static void smc_fback_restore_callbacks(struct smc_sock *smc) 265 { 266 struct sock *clcsk = smc->clcsock->sk; 267 268 write_lock_bh(&clcsk->sk_callback_lock); 269 rcu_assign_sk_user_data(clcsk, NULL); 270 271 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 272 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 273 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 274 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 275 276 write_unlock_bh(&clcsk->sk_callback_lock); 277 } 278 279 static void smc_restore_fallback_changes(struct smc_sock *smc) 280 { 281 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 282 smc->clcsock->file->private_data = smc->sk.sk_socket; 283 smc->clcsock->file = NULL; 284 smc_fback_restore_callbacks(smc); 285 } 286 } 287 288 static int __smc_release(struct smc_sock *smc) 289 { 290 struct sock *sk = &smc->sk; 291 int rc = 0; 292 293 if (!smc->use_fallback) { 294 rc = smc_close_active(smc); 295 smc_sock_set_flag(sk, SOCK_DEAD); 296 sk->sk_shutdown |= SHUTDOWN_MASK; 297 } else { 298 if (sk->sk_state != SMC_CLOSED) { 299 if (sk->sk_state != SMC_LISTEN && 300 sk->sk_state != SMC_INIT) 301 sock_put(sk); /* passive closing */ 302 if (sk->sk_state == SMC_LISTEN) { 303 /* wake up clcsock accept */ 304 rc = kernel_sock_shutdown(smc->clcsock, 305 SHUT_RDWR); 306 } 307 sk->sk_state = SMC_CLOSED; 308 sk->sk_state_change(sk); 309 } 310 smc_restore_fallback_changes(smc); 311 } 312 313 sk->sk_prot->unhash(sk); 314 315 if (sk->sk_state == SMC_CLOSED) { 316 if (smc->clcsock) { 317 release_sock(sk); 318 smc_clcsock_release(smc); 319 lock_sock(sk); 320 } 321 if (!smc->use_fallback) 322 smc_conn_free(&smc->conn); 323 } 324 325 return rc; 326 } 327 328 int smc_release(struct socket *sock) 329 { 330 struct sock *sk = sock->sk; 331 struct smc_sock *smc; 332 int old_state, rc = 0; 333 334 if (!sk) 335 goto out; 336 337 sock_hold(sk); /* sock_put below */ 338 smc = smc_sk(sk); 339 340 old_state = sk->sk_state; 341 342 /* cleanup for a dangling non-blocking connect */ 343 if (smc->connect_nonblock && old_state == SMC_INIT) 344 tcp_abort(smc->clcsock->sk, ECONNABORTED); 345 346 if (cancel_work_sync(&smc->connect_work)) 347 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 348 349 if (sk->sk_state == SMC_LISTEN) 350 /* smc_close_non_accepted() is called and acquires 351 * sock lock for child sockets again 352 */ 353 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 354 else 355 lock_sock(sk); 356 357 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 358 !smc->use_fallback) 359 smc_close_active_abort(smc); 360 361 rc = __smc_release(smc); 362 363 /* detach socket */ 364 sock_orphan(sk); 365 sock->sk = NULL; 366 release_sock(sk); 367 368 sock_put(sk); /* sock_hold above */ 369 sock_put(sk); /* final sock_put */ 370 out: 371 return rc; 372 } 373 374 static void smc_destruct(struct sock *sk) 375 { 376 if (sk->sk_state != SMC_CLOSED) 377 return; 378 if (!sock_flag(sk, SOCK_DEAD)) 379 return; 380 switch (sk->sk_family) { 381 case AF_INET: 382 inet_sock_destruct(sk); 383 break; 384 #if IS_ENABLED(CONFIG_IPV6) 385 case AF_INET6: 386 inet6_sock_destruct(sk); 387 break; 388 #endif 389 } 390 } 391 392 static struct lock_class_key smc_key; 393 static struct lock_class_key smc_slock_key; 394 395 void smc_sk_init(struct net *net, struct sock *sk, int protocol) 396 { 397 struct smc_sock *smc = smc_sk(sk); 398 399 sk->sk_state = SMC_INIT; 400 sk->sk_destruct = smc_destruct; 401 sk->sk_protocol = protocol; 402 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 403 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 404 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 405 INIT_WORK(&smc->connect_work, smc_connect_work); 406 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 407 INIT_LIST_HEAD(&smc->accept_q); 408 sock_lock_init_class_and_name(sk, "slock-AF_SMC", &smc_slock_key, 409 "sk_lock-AF_SMC", &smc_key); 410 spin_lock_init(&smc->accept_q_lock); 411 spin_lock_init(&smc->conn.send_lock); 412 mutex_init(&smc->clcsock_release_lock); 413 smc_init_saved_callbacks(smc); 414 smc->limit_smc_hs = net->smc.limit_smc_hs; 415 smc->use_fallback = false; /* assume rdma capability first */ 416 smc->fallback_rsn = 0; 417 smc_close_init(smc); 418 sk->sk_prot->hash(sk); 419 } 420 421 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 422 int protocol) 423 { 424 struct proto *prot; 425 struct sock *sk; 426 427 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 428 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 429 if (!sk) 430 return NULL; 431 432 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 433 smc_sk_init(net, sk, protocol); 434 435 return sk; 436 } 437 438 int smc_bind(struct socket *sock, struct sockaddr_unsized *uaddr, 439 int addr_len) 440 { 441 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 442 struct sock *sk = sock->sk; 443 struct smc_sock *smc; 444 int rc; 445 446 smc = smc_sk(sk); 447 448 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 449 rc = -EINVAL; 450 if (addr_len < sizeof(struct sockaddr_in)) 451 goto out; 452 453 rc = -EAFNOSUPPORT; 454 if (addr->sin_family != AF_INET && 455 addr->sin_family != AF_INET6 && 456 addr->sin_family != AF_UNSPEC) 457 goto out; 458 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 459 if (addr->sin_family == AF_UNSPEC && 460 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 461 goto out; 462 463 lock_sock(sk); 464 465 /* Check if socket is already active */ 466 rc = -EINVAL; 467 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 468 goto out_rel; 469 470 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 471 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 472 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 473 474 out_rel: 475 release_sock(sk); 476 out: 477 return rc; 478 } 479 480 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 481 * clc socket (since smc is not called for these options from net/core) 482 */ 483 484 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 485 (1UL << SOCK_KEEPOPEN) | \ 486 (1UL << SOCK_LINGER) | \ 487 (1UL << SOCK_BROADCAST) | \ 488 (1UL << SOCK_TIMESTAMP) | \ 489 (1UL << SOCK_DBG) | \ 490 (1UL << SOCK_RCVTSTAMP) | \ 491 (1UL << SOCK_RCVTSTAMPNS) | \ 492 (1UL << SOCK_LOCALROUTE) | \ 493 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 494 (1UL << SOCK_RXQ_OVFL) | \ 495 (1UL << SOCK_WIFI_STATUS) | \ 496 (1UL << SOCK_NOFCS) | \ 497 (1UL << SOCK_FILTER_LOCKED) | \ 498 (1UL << SOCK_TSTAMP_NEW)) 499 500 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 501 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 502 unsigned long mask) 503 { 504 nsk->sk_userlocks = osk->sk_userlocks; 505 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) 506 nsk->sk_sndbuf = osk->sk_sndbuf; 507 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) 508 nsk->sk_rcvbuf = osk->sk_rcvbuf; 509 } 510 511 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 512 unsigned long mask) 513 { 514 /* options we don't get control via setsockopt for */ 515 nsk->sk_type = osk->sk_type; 516 nsk->sk_sndtimeo = READ_ONCE(osk->sk_sndtimeo); 517 nsk->sk_rcvtimeo = READ_ONCE(osk->sk_rcvtimeo); 518 nsk->sk_mark = READ_ONCE(osk->sk_mark); 519 nsk->sk_priority = READ_ONCE(osk->sk_priority); 520 nsk->sk_rcvlowat = osk->sk_rcvlowat; 521 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 522 nsk->sk_err = osk->sk_err; 523 524 nsk->sk_flags &= ~mask; 525 nsk->sk_flags |= osk->sk_flags & mask; 526 527 smc_adjust_sock_bufsizes(nsk, osk, mask); 528 } 529 530 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 531 { 532 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 533 } 534 535 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 536 (1UL << SOCK_KEEPOPEN) | \ 537 (1UL << SOCK_LINGER) | \ 538 (1UL << SOCK_DBG)) 539 /* copy only settings and flags relevant for smc from clc to smc socket */ 540 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 541 { 542 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 543 } 544 545 /* register the new vzalloced sndbuf on all links */ 546 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 547 struct smc_buf_desc *snd_desc) 548 { 549 struct smc_link_group *lgr = link->lgr; 550 int i, rc = 0; 551 552 if (!snd_desc->is_vm) 553 return -EINVAL; 554 555 /* protect against parallel smcr_link_reg_buf() */ 556 down_write(&lgr->llc_conf_mutex); 557 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 558 if (!smc_link_active(&lgr->lnk[i])) 559 continue; 560 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 561 if (rc) 562 break; 563 } 564 up_write(&lgr->llc_conf_mutex); 565 return rc; 566 } 567 568 /* register the new rmb on all links */ 569 static int smcr_lgr_reg_rmbs(struct smc_link *link, 570 struct smc_buf_desc *rmb_desc) 571 { 572 struct smc_link_group *lgr = link->lgr; 573 bool do_slow = false; 574 int i, rc = 0; 575 576 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 577 if (rc) 578 return rc; 579 580 down_read(&lgr->llc_conf_mutex); 581 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 582 if (!smc_link_active(&lgr->lnk[i])) 583 continue; 584 if (!rmb_desc->is_reg_mr[link->link_idx]) { 585 up_read(&lgr->llc_conf_mutex); 586 goto slow_path; 587 } 588 } 589 /* mr register already */ 590 goto fast_path; 591 slow_path: 592 do_slow = true; 593 /* protect against parallel smc_llc_cli_rkey_exchange() and 594 * parallel smcr_link_reg_buf() 595 */ 596 down_write(&lgr->llc_conf_mutex); 597 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 598 if (!smc_link_active(&lgr->lnk[i])) 599 continue; 600 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 601 if (rc) 602 goto out; 603 } 604 fast_path: 605 /* exchange confirm_rkey msg with peer */ 606 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 607 if (rc) { 608 rc = -EFAULT; 609 goto out; 610 } 611 rmb_desc->is_conf_rkey = true; 612 out: 613 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 614 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 615 return rc; 616 } 617 618 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 619 { 620 struct smc_link *link = smc->conn.lnk; 621 struct smc_llc_qentry *qentry; 622 int rc; 623 624 /* Receive CONFIRM LINK request from server over RoCE fabric. 625 * Increasing the client's timeout by twice as much as the server's 626 * timeout by default can temporarily avoid decline messages of 627 * both sides crossing or colliding 628 */ 629 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 630 SMC_LLC_CONFIRM_LINK); 631 if (!qentry) { 632 struct smc_clc_msg_decline dclc; 633 634 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 635 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 636 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 637 } 638 smc_llc_save_peer_uid(qentry); 639 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 640 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 641 if (rc) 642 return SMC_CLC_DECL_RMBE_EC; 643 644 rc = smc_ib_modify_qp_rts(link); 645 if (rc) 646 return SMC_CLC_DECL_ERR_RDYLNK; 647 648 smc_wr_remember_qp_attr(link); 649 650 /* reg the sndbuf if it was vzalloced */ 651 if (smc->conn.sndbuf_desc->is_vm) { 652 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 653 return SMC_CLC_DECL_ERR_REGBUF; 654 } 655 656 /* reg the rmb */ 657 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 658 return SMC_CLC_DECL_ERR_REGBUF; 659 660 /* confirm_rkey is implicit on 1st contact */ 661 smc->conn.rmb_desc->is_conf_rkey = true; 662 663 /* send CONFIRM LINK response over RoCE fabric */ 664 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 665 if (rc < 0) 666 return SMC_CLC_DECL_TIMEOUT_CL; 667 668 down_write(&link->lgr->llc_conf_mutex); 669 smc_llc_link_active(link); 670 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 671 up_write(&link->lgr->llc_conf_mutex); 672 673 if (link->lgr->max_links > 1) { 674 /* optional 2nd link, receive ADD LINK request from server */ 675 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 676 SMC_LLC_ADD_LINK); 677 if (!qentry) { 678 struct smc_clc_msg_decline dclc; 679 680 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 681 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 682 if (rc == -EAGAIN) 683 rc = 0; /* no DECLINE received, go with one link */ 684 return rc; 685 } 686 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 687 down_write(&link->lgr->llc_conf_mutex); 688 smc_llc_cli_add_link(link, qentry); 689 up_write(&link->lgr->llc_conf_mutex); 690 } 691 return 0; 692 } 693 694 static bool smc_isascii(char *hostname) 695 { 696 int i; 697 698 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 699 if (!isascii(hostname[i])) 700 return false; 701 return true; 702 } 703 704 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 705 struct smc_clc_msg_accept_confirm *clc) 706 { 707 struct smc_clc_first_contact_ext *fce; 708 int clc_v2_len; 709 710 if (clc->hdr.version == SMC_V1 || 711 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 712 return; 713 714 if (smc->conn.lgr->is_smcd) { 715 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 716 SMC_MAX_EID_LEN); 717 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 718 } else { 719 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 720 SMC_MAX_EID_LEN); 721 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 722 } 723 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 724 smc->conn.lgr->peer_os = fce->os_type; 725 smc->conn.lgr->peer_smc_release = fce->release; 726 if (smc_isascii(fce->hostname)) 727 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 728 SMC_MAX_HOSTNAME_LEN); 729 } 730 731 static void smcr_conn_save_peer_info(struct smc_sock *smc, 732 struct smc_clc_msg_accept_confirm *clc) 733 { 734 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 735 736 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 737 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 738 smc->conn.peer_rmbe_size = bufsize; 739 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 740 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 741 } 742 743 static void smcd_conn_save_peer_info(struct smc_sock *smc, 744 struct smc_clc_msg_accept_confirm *clc) 745 { 746 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 747 748 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 749 smc->conn.peer_token = ntohll(clc->d0.token); 750 /* msg header takes up space in the buffer */ 751 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 752 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 753 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 754 } 755 756 static void smc_conn_save_peer_info(struct smc_sock *smc, 757 struct smc_clc_msg_accept_confirm *clc) 758 { 759 if (smc->conn.lgr->is_smcd) 760 smcd_conn_save_peer_info(smc, clc); 761 else 762 smcr_conn_save_peer_info(smc, clc); 763 smc_conn_save_peer_info_fce(smc, clc); 764 } 765 766 static void smc_link_save_peer_info(struct smc_link *link, 767 struct smc_clc_msg_accept_confirm *clc, 768 struct smc_init_info *ini) 769 { 770 link->peer_qpn = ntoh24(clc->r0.qpn); 771 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 772 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 773 link->peer_psn = ntoh24(clc->r0.psn); 774 link->peer_mtu = clc->r0.qp_mtu; 775 } 776 777 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 778 struct smc_stats_fback *fback_arr) 779 { 780 int cnt; 781 782 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 783 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 784 fback_arr[cnt].count++; 785 break; 786 } 787 if (!fback_arr[cnt].fback_code) { 788 fback_arr[cnt].fback_code = smc->fallback_rsn; 789 fback_arr[cnt].count++; 790 break; 791 } 792 } 793 } 794 795 static void smc_stat_fallback(struct smc_sock *smc) 796 { 797 struct net *net = sock_net(&smc->sk); 798 799 mutex_lock(&net->smc.mutex_fback_rsn); 800 if (smc->listen_smc) { 801 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 802 net->smc.fback_rsn->srv_fback_cnt++; 803 } else { 804 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 805 net->smc.fback_rsn->clnt_fback_cnt++; 806 } 807 mutex_unlock(&net->smc.mutex_fback_rsn); 808 } 809 810 /* must be called under rcu read lock */ 811 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 812 { 813 struct socket_wq *wq; 814 __poll_t flags; 815 816 wq = rcu_dereference(smc->sk.sk_wq); 817 if (!skwq_has_sleeper(wq)) 818 return; 819 820 /* wake up smc sk->sk_wq */ 821 if (!key) { 822 /* sk_state_change */ 823 wake_up_interruptible_all(&wq->wait); 824 } else { 825 flags = key_to_poll(key); 826 if (flags & (EPOLLIN | EPOLLOUT)) 827 /* sk_data_ready or sk_write_space */ 828 wake_up_interruptible_sync_poll(&wq->wait, flags); 829 else if (flags & EPOLLERR) 830 /* sk_error_report */ 831 wake_up_interruptible_poll(&wq->wait, flags); 832 } 833 } 834 835 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 836 unsigned int mode, int sync, void *key) 837 { 838 struct smc_mark_woken *mark = 839 container_of(wait, struct smc_mark_woken, wait_entry); 840 841 mark->woken = true; 842 mark->key = key; 843 return 0; 844 } 845 846 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 847 void (*clcsock_callback)(struct sock *sk)) 848 { 849 struct smc_mark_woken mark = { .woken = false }; 850 struct socket_wq *wq; 851 852 init_waitqueue_func_entry(&mark.wait_entry, 853 smc_fback_mark_woken); 854 rcu_read_lock(); 855 wq = rcu_dereference(clcsk->sk_wq); 856 if (!wq) 857 goto out; 858 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 859 clcsock_callback(clcsk); 860 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 861 862 if (mark.woken) 863 smc_fback_wakeup_waitqueue(smc, mark.key); 864 out: 865 rcu_read_unlock(); 866 } 867 868 static void smc_fback_state_change(struct sock *clcsk) 869 { 870 struct smc_sock *smc; 871 872 read_lock_bh(&clcsk->sk_callback_lock); 873 smc = smc_clcsock_user_data(clcsk); 874 if (smc) 875 smc_fback_forward_wakeup(smc, clcsk, 876 smc->clcsk_state_change); 877 read_unlock_bh(&clcsk->sk_callback_lock); 878 } 879 880 static void smc_fback_data_ready(struct sock *clcsk) 881 { 882 struct smc_sock *smc; 883 884 read_lock_bh(&clcsk->sk_callback_lock); 885 smc = smc_clcsock_user_data(clcsk); 886 if (smc) 887 smc_fback_forward_wakeup(smc, clcsk, 888 smc->clcsk_data_ready); 889 read_unlock_bh(&clcsk->sk_callback_lock); 890 } 891 892 static void smc_fback_write_space(struct sock *clcsk) 893 { 894 struct smc_sock *smc; 895 896 read_lock_bh(&clcsk->sk_callback_lock); 897 smc = smc_clcsock_user_data(clcsk); 898 if (smc) 899 smc_fback_forward_wakeup(smc, clcsk, 900 smc->clcsk_write_space); 901 read_unlock_bh(&clcsk->sk_callback_lock); 902 } 903 904 static void smc_fback_error_report(struct sock *clcsk) 905 { 906 struct smc_sock *smc; 907 908 read_lock_bh(&clcsk->sk_callback_lock); 909 smc = smc_clcsock_user_data(clcsk); 910 if (smc) 911 smc_fback_forward_wakeup(smc, clcsk, 912 smc->clcsk_error_report); 913 read_unlock_bh(&clcsk->sk_callback_lock); 914 } 915 916 static void smc_fback_replace_callbacks(struct smc_sock *smc) 917 { 918 struct sock *clcsk = smc->clcsock->sk; 919 920 write_lock_bh(&clcsk->sk_callback_lock); 921 __rcu_assign_sk_user_data_with_flags(clcsk, smc, SK_USER_DATA_NOCOPY); 922 923 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 924 &smc->clcsk_state_change); 925 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 926 &smc->clcsk_data_ready); 927 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 928 &smc->clcsk_write_space); 929 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 930 &smc->clcsk_error_report); 931 932 write_unlock_bh(&clcsk->sk_callback_lock); 933 } 934 935 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 936 { 937 int rc = 0; 938 939 mutex_lock(&smc->clcsock_release_lock); 940 if (!smc->clcsock) { 941 rc = -EBADF; 942 goto out; 943 } 944 945 smc->use_fallback = true; 946 smc->fallback_rsn = reason_code; 947 smc_stat_fallback(smc); 948 trace_smc_switch_to_fallback(smc, reason_code); 949 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 950 smc->clcsock->file = smc->sk.sk_socket->file; 951 smc->clcsock->file->private_data = smc->clcsock; 952 smc->clcsock->wq.fasync_list = 953 smc->sk.sk_socket->wq.fasync_list; 954 smc->sk.sk_socket->wq.fasync_list = NULL; 955 956 /* There might be some wait entries remaining 957 * in smc sk->sk_wq and they should be woken up 958 * as clcsock's wait queue is woken up. 959 */ 960 smc_fback_replace_callbacks(smc); 961 } 962 out: 963 mutex_unlock(&smc->clcsock_release_lock); 964 return rc; 965 } 966 967 /* fall back during connect */ 968 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 969 { 970 struct net *net = sock_net(&smc->sk); 971 int rc = 0; 972 973 rc = smc_switch_to_fallback(smc, reason_code); 974 if (rc) { /* fallback fails */ 975 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 976 if (smc->sk.sk_state == SMC_INIT) 977 sock_put(&smc->sk); /* passive closing */ 978 return rc; 979 } 980 smc_copy_sock_settings_to_clc(smc); 981 smc->connect_nonblock = 0; 982 if (smc->sk.sk_state == SMC_INIT) 983 smc->sk.sk_state = SMC_ACTIVE; 984 return 0; 985 } 986 987 /* decline and fall back during connect */ 988 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 989 u8 version) 990 { 991 struct net *net = sock_net(&smc->sk); 992 int rc; 993 994 if (reason_code < 0) { /* error, fallback is not possible */ 995 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 996 if (smc->sk.sk_state == SMC_INIT) 997 sock_put(&smc->sk); /* passive closing */ 998 return reason_code; 999 } 1000 if (reason_code != SMC_CLC_DECL_PEERDECL) { 1001 rc = smc_clc_send_decline(smc, reason_code, version); 1002 if (rc < 0) { 1003 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 1004 if (smc->sk.sk_state == SMC_INIT) 1005 sock_put(&smc->sk); /* passive closing */ 1006 return rc; 1007 } 1008 } 1009 return smc_connect_fallback(smc, reason_code); 1010 } 1011 1012 static void smc_conn_abort(struct smc_sock *smc, int local_first) 1013 { 1014 struct smc_connection *conn = &smc->conn; 1015 struct smc_link_group *lgr = conn->lgr; 1016 bool lgr_valid = false; 1017 1018 if (smc_conn_lgr_valid(conn)) 1019 lgr_valid = true; 1020 1021 smc_conn_free(conn); 1022 if (local_first && lgr_valid) 1023 smc_lgr_cleanup_early(lgr); 1024 } 1025 1026 /* check if there is a rdma device available for this connection. */ 1027 /* called for connect and listen */ 1028 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1029 { 1030 /* PNET table look up: search active ib_device and port 1031 * within same PNETID that also contains the ethernet device 1032 * used for the internal TCP socket 1033 */ 1034 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1035 if (!ini->check_smcrv2 && !ini->ib_dev) 1036 return SMC_CLC_DECL_NOSMCRDEV; 1037 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1038 return SMC_CLC_DECL_NOSMCRDEV; 1039 return 0; 1040 } 1041 1042 /* check if there is an ISM device available for this connection. */ 1043 /* called for connect and listen */ 1044 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1045 { 1046 /* Find ISM device with same PNETID as connecting interface */ 1047 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1048 if (!ini->ism_dev[0]) 1049 return SMC_CLC_DECL_NOSMCDDEV; 1050 else 1051 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1052 return 0; 1053 } 1054 1055 /* is chid unique for the ism devices that are already determined? */ 1056 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1057 int cnt) 1058 { 1059 int i = (!ini->ism_dev[0]) ? 1 : 0; 1060 1061 for (; i < cnt; i++) 1062 if (ini->ism_chid[i] == chid) 1063 return false; 1064 return true; 1065 } 1066 1067 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1068 * PNETID matching net_device) 1069 */ 1070 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1071 struct smc_init_info *ini) 1072 { 1073 int rc = SMC_CLC_DECL_NOSMCDDEV; 1074 struct smcd_dev *smcd; 1075 int i = 1, entry = 1; 1076 bool is_emulated; 1077 u16 chid; 1078 1079 if (smcd_indicated(ini->smc_type_v1)) 1080 rc = 0; /* already initialized for V1 */ 1081 mutex_lock(&smcd_dev_list.mutex); 1082 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1083 if (smcd->going_away || smcd == ini->ism_dev[0]) 1084 continue; 1085 chid = smc_ism_get_chid(smcd); 1086 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1087 continue; 1088 is_emulated = __smc_ism_is_emulated(chid); 1089 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1090 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1091 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1092 /* It's the last GID-CHID entry left in CLC 1093 * Proposal SMC-Dv2 extension, but an Emulated- 1094 * ISM device will take two entries. So give 1095 * up it and try the next potential ISM device. 1096 */ 1097 continue; 1098 ini->ism_dev[i] = smcd; 1099 ini->ism_chid[i] = chid; 1100 ini->is_smcd = true; 1101 rc = 0; 1102 i++; 1103 entry = is_emulated ? entry + 2 : entry + 1; 1104 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1105 break; 1106 } 1107 } 1108 mutex_unlock(&smcd_dev_list.mutex); 1109 ini->ism_offered_cnt = i - 1; 1110 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1111 ini->smcd_version = 0; 1112 1113 return rc; 1114 } 1115 1116 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1117 static int smc_connect_ism_vlan_setup(struct smc_init_info *ini) 1118 { 1119 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1120 return SMC_CLC_DECL_ISMVLANERR; 1121 return 0; 1122 } 1123 1124 static int smc_find_proposal_devices(struct smc_sock *smc, 1125 struct smc_init_info *ini) 1126 { 1127 int rc = 0; 1128 1129 /* check if there is an ism device available */ 1130 if (!(ini->smcd_version & SMC_V1) || 1131 smc_find_ism_device(smc, ini) || 1132 smc_connect_ism_vlan_setup(ini)) 1133 ini->smcd_version &= ~SMC_V1; 1134 /* else ISM V1 is supported for this connection */ 1135 1136 /* check if there is an rdma device available */ 1137 if (!(ini->smcr_version & SMC_V1) || 1138 smc_find_rdma_device(smc, ini)) 1139 ini->smcr_version &= ~SMC_V1; 1140 /* else RDMA is supported for this connection */ 1141 1142 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1143 ini->smcr_version & SMC_V1); 1144 1145 /* check if there is an ism v2 device available */ 1146 if (!(ini->smcd_version & SMC_V2) || 1147 !smc_ism_is_v2_capable() || 1148 smc_find_ism_v2_device_clnt(smc, ini)) 1149 ini->smcd_version &= ~SMC_V2; 1150 1151 /* check if there is an rdma v2 device available */ 1152 ini->check_smcrv2 = true; 1153 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1154 if (!(ini->smcr_version & SMC_V2) || 1155 #if IS_ENABLED(CONFIG_IPV6) 1156 (smc->clcsock->sk->sk_family == AF_INET6 && 1157 !ipv6_addr_v4mapped(&smc->clcsock->sk->sk_v6_rcv_saddr)) || 1158 #endif 1159 !smc_clc_ueid_count() || 1160 smc_find_rdma_device(smc, ini)) 1161 ini->smcr_version &= ~SMC_V2; 1162 ini->check_smcrv2 = false; 1163 1164 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1165 ini->smcr_version & SMC_V2); 1166 1167 /* if neither ISM nor RDMA are supported, fallback */ 1168 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1169 rc = SMC_CLC_DECL_NOSMCDEV; 1170 1171 return rc; 1172 } 1173 1174 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1175 * used, the VLAN ID will be registered again during the connection setup. 1176 */ 1177 static int smc_connect_ism_vlan_cleanup(struct smc_init_info *ini) 1178 { 1179 if (!smcd_indicated(ini->smc_type_v1)) 1180 return 0; 1181 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1182 return SMC_CLC_DECL_CNFERR; 1183 return 0; 1184 } 1185 1186 #define SMC_CLC_MAX_ACCEPT_LEN \ 1187 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1188 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1189 sizeof(struct smc_clc_msg_trail)) 1190 1191 /* CLC handshake during connect */ 1192 static int smc_connect_clc(struct smc_sock *smc, 1193 struct smc_clc_msg_accept_confirm *aclc, 1194 struct smc_init_info *ini) 1195 { 1196 int rc = 0; 1197 1198 /* do inband token exchange */ 1199 rc = smc_clc_send_proposal(smc, ini); 1200 if (rc) 1201 return rc; 1202 /* receive SMC Accept CLC message */ 1203 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1204 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1205 } 1206 1207 void smc_fill_gid_list(struct smc_link_group *lgr, 1208 struct smc_gidlist *gidlist, 1209 struct smc_ib_device *known_dev, u8 *known_gid) 1210 { 1211 struct smc_init_info *alt_ini = NULL; 1212 1213 memset(gidlist, 0, sizeof(*gidlist)); 1214 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1215 1216 alt_ini = kzalloc_obj(*alt_ini); 1217 if (!alt_ini) 1218 goto out; 1219 1220 alt_ini->vlan_id = lgr->vlan_id; 1221 alt_ini->check_smcrv2 = true; 1222 alt_ini->smcrv2.saddr = lgr->saddr; 1223 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1224 1225 if (!alt_ini->smcrv2.ib_dev_v2) 1226 goto out; 1227 1228 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1229 SMC_GID_SIZE); 1230 1231 out: 1232 kfree(alt_ini); 1233 } 1234 1235 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1236 struct smc_clc_msg_accept_confirm *aclc, 1237 struct smc_init_info *ini) 1238 { 1239 struct smc_clc_first_contact_ext *fce = 1240 smc_get_clc_first_contact_ext(aclc, false); 1241 struct net *net = sock_net(&smc->sk); 1242 int rc; 1243 1244 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1245 return 0; 1246 1247 if (fce->v2_direct) { 1248 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1249 ini->smcrv2.uses_gateway = false; 1250 } else { 1251 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1252 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1253 ini->smcrv2.nexthop_mac, 1254 &ini->smcrv2.uses_gateway)) 1255 return SMC_CLC_DECL_NOROUTE; 1256 if (!ini->smcrv2.uses_gateway) { 1257 /* mismatch: peer claims indirect, but its direct */ 1258 return SMC_CLC_DECL_NOINDIRECT; 1259 } 1260 } 1261 1262 ini->release_nr = fce->release; 1263 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1264 if (rc) 1265 return rc; 1266 1267 return 0; 1268 } 1269 1270 /* setup for RDMA connection of client */ 1271 static int smc_connect_rdma(struct smc_sock *smc, 1272 struct smc_clc_msg_accept_confirm *aclc, 1273 struct smc_init_info *ini) 1274 { 1275 int i, reason_code = 0; 1276 struct smc_link *link; 1277 u8 *eid = NULL; 1278 1279 ini->is_smcd = false; 1280 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1281 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1282 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1283 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1284 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1285 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1286 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1287 1288 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1289 if (reason_code) 1290 return reason_code; 1291 1292 mutex_lock(&smc_client_lgr_pending); 1293 reason_code = smc_conn_create(smc, ini); 1294 if (reason_code) { 1295 mutex_unlock(&smc_client_lgr_pending); 1296 return reason_code; 1297 } 1298 1299 smc_conn_save_peer_info(smc, aclc); 1300 1301 if (ini->first_contact_local) { 1302 link = smc->conn.lnk; 1303 } else { 1304 /* set link that was assigned by server */ 1305 link = NULL; 1306 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1307 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1308 1309 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1310 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1311 SMC_GID_SIZE) && 1312 (aclc->hdr.version > SMC_V1 || 1313 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1314 sizeof(l->peer_mac)))) { 1315 link = l; 1316 break; 1317 } 1318 } 1319 if (!link) { 1320 reason_code = SMC_CLC_DECL_NOSRVLINK; 1321 goto connect_abort; 1322 } 1323 smc_switch_link_and_count(&smc->conn, link); 1324 } 1325 1326 /* create send buffer and rmb */ 1327 if (smc_buf_create(smc, false)) { 1328 reason_code = SMC_CLC_DECL_MEM; 1329 goto connect_abort; 1330 } 1331 1332 if (ini->first_contact_local) 1333 smc_link_save_peer_info(link, aclc, ini); 1334 1335 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1336 reason_code = SMC_CLC_DECL_ERR_RTOK; 1337 goto connect_abort; 1338 } 1339 1340 smc_rx_init(smc); 1341 1342 if (ini->first_contact_local) { 1343 if (smc_ib_ready_link(link)) { 1344 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1345 goto connect_abort; 1346 } 1347 } else { 1348 /* reg sendbufs if they were vzalloced */ 1349 if (smc->conn.sndbuf_desc->is_vm) { 1350 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1351 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1352 goto connect_abort; 1353 } 1354 } 1355 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1356 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1357 goto connect_abort; 1358 } 1359 } 1360 1361 if (aclc->hdr.version > SMC_V1) { 1362 eid = aclc->r1.eid; 1363 if (ini->first_contact_local) 1364 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1365 link->smcibdev, link->gid); 1366 } 1367 1368 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1369 aclc->hdr.version, eid, ini); 1370 if (reason_code) 1371 goto connect_abort; 1372 1373 smc_tx_init(smc); 1374 1375 if (ini->first_contact_local) { 1376 /* QP confirmation over RoCE fabric */ 1377 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1378 reason_code = smcr_clnt_conf_first_link(smc); 1379 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1380 if (reason_code) 1381 goto connect_abort; 1382 } 1383 mutex_unlock(&smc_client_lgr_pending); 1384 1385 smc_copy_sock_settings_to_clc(smc); 1386 smc->connect_nonblock = 0; 1387 if (smc->sk.sk_state == SMC_INIT) 1388 smc->sk.sk_state = SMC_ACTIVE; 1389 1390 return 0; 1391 connect_abort: 1392 smc_conn_abort(smc, ini->first_contact_local); 1393 mutex_unlock(&smc_client_lgr_pending); 1394 smc->connect_nonblock = 0; 1395 1396 return reason_code; 1397 } 1398 1399 /* The server has chosen one of the proposed ISM devices for the communication. 1400 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1401 */ 1402 static int 1403 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1404 struct smc_init_info *ini) 1405 { 1406 int i; 1407 1408 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1409 if (ini->ism_dev[i] && 1410 ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1411 ini->ism_selected = i; 1412 return 0; 1413 } 1414 } 1415 1416 return -EPROTO; 1417 } 1418 1419 /* setup for ISM connection of client */ 1420 static int smc_connect_ism(struct smc_sock *smc, 1421 struct smc_clc_msg_accept_confirm *aclc, 1422 struct smc_init_info *ini) 1423 { 1424 u8 *eid = NULL; 1425 int rc = 0; 1426 1427 ini->is_smcd = true; 1428 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1429 1430 if (aclc->hdr.version == SMC_V2) { 1431 if (ini->first_contact_peer) { 1432 struct smc_clc_first_contact_ext *fce = 1433 smc_get_clc_first_contact_ext(aclc, true); 1434 1435 ini->release_nr = fce->release; 1436 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1437 if (rc) 1438 return rc; 1439 } 1440 1441 rc = smc_v2_determine_accepted_chid(aclc, ini); 1442 if (rc) 1443 return rc; 1444 1445 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1446 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1447 ntohll(aclc->d1.gid_ext); 1448 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1449 } 1450 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1451 1452 /* there is only one lgr role for SMC-D; use server lock */ 1453 mutex_lock(&smc_server_lgr_pending); 1454 rc = smc_conn_create(smc, ini); 1455 if (rc) { 1456 mutex_unlock(&smc_server_lgr_pending); 1457 return rc; 1458 } 1459 1460 /* Create send and receive buffers */ 1461 rc = smc_buf_create(smc, true); 1462 if (rc) { 1463 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1464 goto connect_abort; 1465 } 1466 1467 smc_conn_save_peer_info(smc, aclc); 1468 1469 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) { 1470 rc = smcd_buf_attach(smc); 1471 if (rc) { 1472 rc = SMC_CLC_DECL_MEM; /* try to fallback */ 1473 goto connect_abort; 1474 } 1475 } 1476 smc_rx_init(smc); 1477 smc_tx_init(smc); 1478 1479 if (aclc->hdr.version > SMC_V1) 1480 eid = aclc->d1.eid; 1481 1482 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1483 aclc->hdr.version, eid, ini); 1484 if (rc) 1485 goto connect_abort; 1486 mutex_unlock(&smc_server_lgr_pending); 1487 1488 smc_copy_sock_settings_to_clc(smc); 1489 smc->connect_nonblock = 0; 1490 if (smc->sk.sk_state == SMC_INIT) 1491 smc->sk.sk_state = SMC_ACTIVE; 1492 1493 return 0; 1494 connect_abort: 1495 smc_conn_abort(smc, ini->first_contact_local); 1496 mutex_unlock(&smc_server_lgr_pending); 1497 smc->connect_nonblock = 0; 1498 1499 return rc; 1500 } 1501 1502 /* check if received accept type and version matches a proposed one */ 1503 static int smc_connect_check_aclc(struct smc_init_info *ini, 1504 struct smc_clc_msg_accept_confirm *aclc) 1505 { 1506 if (aclc->hdr.version >= SMC_V2) { 1507 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1508 !smcr_indicated(ini->smc_type_v2)) || 1509 (aclc->hdr.typev1 == SMC_TYPE_D && 1510 !smcd_indicated(ini->smc_type_v2))) 1511 return SMC_CLC_DECL_MODEUNSUPP; 1512 } else { 1513 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1514 !smcr_indicated(ini->smc_type_v1)) || 1515 (aclc->hdr.typev1 == SMC_TYPE_D && 1516 !smcd_indicated(ini->smc_type_v1))) 1517 return SMC_CLC_DECL_MODEUNSUPP; 1518 } 1519 1520 return 0; 1521 } 1522 1523 /* perform steps before actually connecting */ 1524 static int __smc_connect(struct smc_sock *smc) 1525 { 1526 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1527 struct smc_clc_msg_accept_confirm *aclc; 1528 struct smc_init_info *ini = NULL; 1529 u8 *buf = NULL; 1530 int rc = 0; 1531 1532 if (smc->use_fallback) 1533 return smc_connect_fallback(smc, smc->fallback_rsn); 1534 1535 /* if peer has not signalled SMC-capability, fall back */ 1536 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1537 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1538 1539 /* IPSec connections opt out of SMC optimizations */ 1540 if (using_ipsec(smc)) 1541 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1542 version); 1543 1544 ini = kzalloc_obj(*ini); 1545 if (!ini) 1546 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1547 version); 1548 1549 ini->smcd_version = SMC_V1 | SMC_V2; 1550 ini->smcr_version = SMC_V1 | SMC_V2; 1551 ini->smc_type_v1 = SMC_TYPE_B; 1552 ini->smc_type_v2 = SMC_TYPE_B; 1553 1554 /* get vlan id from IP device */ 1555 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1556 ini->smcd_version &= ~SMC_V1; 1557 ini->smcr_version = 0; 1558 ini->smc_type_v1 = SMC_TYPE_N; 1559 } 1560 1561 rc = smc_find_proposal_devices(smc, ini); 1562 if (rc) 1563 goto fallback; 1564 1565 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1566 if (!buf) { 1567 rc = SMC_CLC_DECL_MEM; 1568 goto fallback; 1569 } 1570 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1571 1572 /* perform CLC handshake */ 1573 rc = smc_connect_clc(smc, aclc, ini); 1574 if (rc) { 1575 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1576 if (rc == -EAGAIN) { 1577 rc = -ETIMEDOUT; 1578 smc->sk.sk_err = ETIMEDOUT; 1579 } 1580 goto vlan_cleanup; 1581 } 1582 1583 /* check if smc modes and versions of CLC proposal and accept match */ 1584 rc = smc_connect_check_aclc(ini, aclc); 1585 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1586 if (rc) 1587 goto vlan_cleanup; 1588 1589 /* depending on previous steps, connect using rdma or ism */ 1590 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1591 ini->smcr_version = version; 1592 rc = smc_connect_rdma(smc, aclc, ini); 1593 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1594 ini->smcd_version = version; 1595 rc = smc_connect_ism(smc, aclc, ini); 1596 } 1597 if (rc) 1598 goto vlan_cleanup; 1599 1600 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1601 smc_connect_ism_vlan_cleanup(ini); 1602 kfree(buf); 1603 kfree(ini); 1604 return 0; 1605 1606 vlan_cleanup: 1607 smc_connect_ism_vlan_cleanup(ini); 1608 kfree(buf); 1609 fallback: 1610 kfree(ini); 1611 return smc_connect_decline_fallback(smc, rc, version); 1612 } 1613 1614 static void smc_connect_work(struct work_struct *work) 1615 { 1616 struct smc_sock *smc = container_of(work, struct smc_sock, 1617 connect_work); 1618 long timeo = READ_ONCE(smc->sk.sk_sndtimeo); 1619 int rc = 0; 1620 1621 if (!timeo) 1622 timeo = MAX_SCHEDULE_TIMEOUT; 1623 lock_sock(smc->clcsock->sk); 1624 if (smc->clcsock->sk->sk_err) { 1625 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1626 } else if ((1 << smc->clcsock->sk->sk_state) & 1627 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1628 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1629 if ((rc == -EPIPE) && 1630 ((1 << smc->clcsock->sk->sk_state) & 1631 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1632 rc = 0; 1633 } 1634 release_sock(smc->clcsock->sk); 1635 lock_sock(&smc->sk); 1636 if (rc != 0 || smc->sk.sk_err) { 1637 smc->sk.sk_state = SMC_CLOSED; 1638 if (!smc->sk.sk_err) 1639 smc->sk.sk_err = (rc == -EAGAIN) ? EPIPE : -rc; 1640 sock_put(&smc->sk); /* passive closing */ 1641 goto out; 1642 } 1643 1644 rc = __smc_connect(smc); 1645 if (rc < 0) 1646 smc->sk.sk_err = -rc; 1647 1648 out: 1649 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1650 if (smc->sk.sk_err) { 1651 smc->sk.sk_state_change(&smc->sk); 1652 } else { /* allow polling before and after fallback decision */ 1653 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1654 smc->sk.sk_write_space(&smc->sk); 1655 } 1656 } 1657 release_sock(&smc->sk); 1658 } 1659 1660 int smc_connect(struct socket *sock, struct sockaddr_unsized *addr, 1661 int alen, int flags) 1662 { 1663 struct sock *sk = sock->sk; 1664 struct smc_sock *smc; 1665 int rc = -EINVAL; 1666 1667 smc = smc_sk(sk); 1668 1669 /* separate smc parameter checking to be safe */ 1670 if (alen < sizeof(addr->sa_family)) 1671 goto out_err; 1672 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1673 goto out_err; 1674 1675 lock_sock(sk); 1676 switch (sock->state) { 1677 default: 1678 rc = -EINVAL; 1679 goto out; 1680 case SS_CONNECTED: 1681 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1682 goto out; 1683 case SS_CONNECTING: 1684 if (sk->sk_state == SMC_ACTIVE) 1685 goto connected; 1686 break; 1687 case SS_UNCONNECTED: 1688 sock->state = SS_CONNECTING; 1689 break; 1690 } 1691 1692 switch (sk->sk_state) { 1693 default: 1694 goto out; 1695 case SMC_CLOSED: 1696 rc = sock_error(sk) ? : -ECONNABORTED; 1697 sock->state = SS_UNCONNECTED; 1698 goto out; 1699 case SMC_ACTIVE: 1700 rc = -EISCONN; 1701 goto out; 1702 case SMC_INIT: 1703 break; 1704 } 1705 1706 smc_copy_sock_settings_to_clc(smc); 1707 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1708 if (smc->connect_nonblock) { 1709 rc = -EALREADY; 1710 goto out; 1711 } 1712 rc = kernel_connect(smc->clcsock, (struct sockaddr_unsized *)addr, alen, flags); 1713 if (rc && rc != -EINPROGRESS) 1714 goto out; 1715 1716 if (smc->use_fallback) { 1717 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1718 goto out; 1719 } 1720 sock_hold(&smc->sk); /* sock put in passive closing */ 1721 if (flags & O_NONBLOCK) { 1722 if (queue_work(smc_hs_wq, &smc->connect_work)) 1723 smc->connect_nonblock = 1; 1724 rc = -EINPROGRESS; 1725 goto out; 1726 } else { 1727 rc = __smc_connect(smc); 1728 if (rc < 0) 1729 goto out; 1730 } 1731 1732 connected: 1733 rc = 0; 1734 sock->state = SS_CONNECTED; 1735 out: 1736 release_sock(sk); 1737 out_err: 1738 return rc; 1739 } 1740 1741 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1742 { 1743 struct socket *new_clcsock = NULL; 1744 struct sock *lsk = &lsmc->sk; 1745 struct sock *new_sk; 1746 int rc = -EINVAL; 1747 1748 release_sock(lsk); 1749 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1750 if (!new_sk) { 1751 rc = -ENOMEM; 1752 lsk->sk_err = ENOMEM; 1753 *new_smc = NULL; 1754 lock_sock(lsk); 1755 goto out; 1756 } 1757 *new_smc = smc_sk(new_sk); 1758 1759 mutex_lock(&lsmc->clcsock_release_lock); 1760 if (lsmc->clcsock) 1761 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1762 mutex_unlock(&lsmc->clcsock_release_lock); 1763 lock_sock(lsk); 1764 if (rc < 0 && rc != -EAGAIN) 1765 lsk->sk_err = -rc; 1766 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1767 new_sk->sk_prot->unhash(new_sk); 1768 if (new_clcsock) 1769 sock_release(new_clcsock); 1770 new_sk->sk_state = SMC_CLOSED; 1771 smc_sock_set_flag(new_sk, SOCK_DEAD); 1772 sock_put(new_sk); /* final */ 1773 *new_smc = NULL; 1774 goto out; 1775 } 1776 1777 /* new clcsock has inherited the smc listen-specific sk_data_ready 1778 * function; switch it back to the original sk_data_ready function 1779 */ 1780 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1781 1782 /* if new clcsock has also inherited the fallback-specific callback 1783 * functions, switch them back to the original ones. 1784 */ 1785 if (lsmc->use_fallback) { 1786 if (lsmc->clcsk_state_change) 1787 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1788 if (lsmc->clcsk_write_space) 1789 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1790 if (lsmc->clcsk_error_report) 1791 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1792 } 1793 1794 (*new_smc)->clcsock = new_clcsock; 1795 out: 1796 return rc; 1797 } 1798 1799 /* add a just created sock to the accept queue of the listen sock as 1800 * candidate for a following socket accept call from user space 1801 */ 1802 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1803 { 1804 struct smc_sock *par = smc_sk(parent); 1805 1806 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1807 spin_lock(&par->accept_q_lock); 1808 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1809 spin_unlock(&par->accept_q_lock); 1810 sk_acceptq_added(parent); 1811 } 1812 1813 /* remove a socket from the accept queue of its parental listening socket */ 1814 static void smc_accept_unlink(struct sock *sk) 1815 { 1816 struct smc_sock *par = smc_sk(sk)->listen_smc; 1817 1818 spin_lock(&par->accept_q_lock); 1819 list_del_init(&smc_sk(sk)->accept_q); 1820 spin_unlock(&par->accept_q_lock); 1821 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1822 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1823 } 1824 1825 /* remove a sock from the accept queue to bind it to a new socket created 1826 * for a socket accept call from user space 1827 */ 1828 struct sock *smc_accept_dequeue(struct sock *parent, 1829 struct socket *new_sock) 1830 { 1831 struct smc_sock *isk, *n; 1832 struct sock *new_sk; 1833 1834 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1835 new_sk = (struct sock *)isk; 1836 1837 smc_accept_unlink(new_sk); 1838 if (new_sk->sk_state == SMC_CLOSED) { 1839 new_sk->sk_prot->unhash(new_sk); 1840 if (isk->clcsock) { 1841 sock_release(isk->clcsock); 1842 isk->clcsock = NULL; 1843 } 1844 sock_put(new_sk); /* final */ 1845 continue; 1846 } 1847 if (new_sock) { 1848 sock_graft(new_sk, new_sock); 1849 new_sock->state = SS_CONNECTED; 1850 if (isk->use_fallback) { 1851 smc_sk(new_sk)->clcsock->file = new_sock->file; 1852 isk->clcsock->file->private_data = isk->clcsock; 1853 } 1854 } 1855 return new_sk; 1856 } 1857 return NULL; 1858 } 1859 1860 /* clean up for a created but never accepted sock */ 1861 void smc_close_non_accepted(struct sock *sk) 1862 { 1863 struct smc_sock *smc = smc_sk(sk); 1864 1865 sock_hold(sk); /* sock_put below */ 1866 lock_sock(sk); 1867 if (!sk->sk_lingertime) 1868 /* wait for peer closing */ 1869 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1870 __smc_release(smc); 1871 release_sock(sk); 1872 sock_put(sk); /* sock_hold above */ 1873 sock_put(sk); /* final sock_put */ 1874 } 1875 1876 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1877 { 1878 struct smc_link *link = smc->conn.lnk; 1879 struct smc_llc_qentry *qentry; 1880 int rc; 1881 1882 /* reg the sndbuf if it was vzalloced*/ 1883 if (smc->conn.sndbuf_desc->is_vm) { 1884 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1885 return SMC_CLC_DECL_ERR_REGBUF; 1886 } 1887 1888 /* reg the rmb */ 1889 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1890 return SMC_CLC_DECL_ERR_REGBUF; 1891 1892 /* send CONFIRM LINK request to client over the RoCE fabric */ 1893 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1894 if (rc < 0) 1895 return SMC_CLC_DECL_TIMEOUT_CL; 1896 1897 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1898 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1899 SMC_LLC_CONFIRM_LINK); 1900 if (!qentry) { 1901 struct smc_clc_msg_decline dclc; 1902 1903 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1904 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1905 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1906 } 1907 smc_llc_save_peer_uid(qentry); 1908 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1909 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1910 if (rc) 1911 return SMC_CLC_DECL_RMBE_EC; 1912 1913 /* confirm_rkey is implicit on 1st contact */ 1914 smc->conn.rmb_desc->is_conf_rkey = true; 1915 1916 down_write(&link->lgr->llc_conf_mutex); 1917 smc_llc_link_active(link); 1918 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1919 up_write(&link->lgr->llc_conf_mutex); 1920 1921 if (link->lgr->max_links > 1) { 1922 down_write(&link->lgr->llc_conf_mutex); 1923 /* initial contact - try to establish second link */ 1924 smc_llc_srv_add_link(link, NULL); 1925 up_write(&link->lgr->llc_conf_mutex); 1926 } 1927 return 0; 1928 } 1929 1930 /* listen worker: finish */ 1931 static void smc_listen_out(struct smc_sock *new_smc) 1932 { 1933 struct smc_sock *lsmc = new_smc->listen_smc; 1934 struct sock *newsmcsk = &new_smc->sk; 1935 1936 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1937 atomic_dec(&lsmc->queued_smc_hs); 1938 1939 release_sock(newsmcsk); /* lock in smc_listen_work() */ 1940 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1941 if (lsmc->sk.sk_state == SMC_LISTEN) { 1942 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1943 release_sock(&lsmc->sk); 1944 } else { /* no longer listening */ 1945 release_sock(&lsmc->sk); 1946 smc_close_non_accepted(newsmcsk); 1947 } 1948 1949 /* Wake up accept */ 1950 lsmc->sk.sk_data_ready(&lsmc->sk); 1951 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1952 } 1953 1954 /* listen worker: finish in state connected */ 1955 static void smc_listen_out_connected(struct smc_sock *new_smc) 1956 { 1957 struct sock *newsmcsk = &new_smc->sk; 1958 1959 if (newsmcsk->sk_state == SMC_INIT) 1960 newsmcsk->sk_state = SMC_ACTIVE; 1961 1962 smc_listen_out(new_smc); 1963 } 1964 1965 /* listen worker: finish in error state */ 1966 static void smc_listen_out_err(struct smc_sock *new_smc) 1967 { 1968 struct sock *newsmcsk = &new_smc->sk; 1969 struct net *net = sock_net(newsmcsk); 1970 1971 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1972 if (newsmcsk->sk_state == SMC_INIT) 1973 sock_put(&new_smc->sk); /* passive closing */ 1974 newsmcsk->sk_state = SMC_CLOSED; 1975 1976 smc_listen_out(new_smc); 1977 } 1978 1979 /* listen worker: decline and fall back if possible */ 1980 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1981 int local_first, u8 version) 1982 { 1983 /* RDMA setup failed, switch back to TCP */ 1984 smc_conn_abort(new_smc, local_first); 1985 if (reason_code < 0 || 1986 smc_switch_to_fallback(new_smc, reason_code)) { 1987 /* error, no fallback possible */ 1988 smc_listen_out_err(new_smc); 1989 return; 1990 } 1991 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1992 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1993 smc_listen_out_err(new_smc); 1994 return; 1995 } 1996 } 1997 smc_listen_out_connected(new_smc); 1998 } 1999 2000 /* listen worker: version checking */ 2001 static int smc_listen_v2_check(struct smc_sock *new_smc, 2002 struct smc_clc_msg_proposal *pclc, 2003 struct smc_init_info *ini) 2004 { 2005 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 2006 struct smc_clc_v2_extension *pclc_v2_ext; 2007 int rc = SMC_CLC_DECL_PEERNOSMC; 2008 2009 ini->smc_type_v1 = pclc->hdr.typev1; 2010 ini->smc_type_v2 = pclc->hdr.typev2; 2011 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 2012 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 2013 if (pclc->hdr.version > SMC_V1) { 2014 if (smcd_indicated(ini->smc_type_v2)) 2015 ini->smcd_version |= SMC_V2; 2016 if (smcr_indicated(ini->smc_type_v2)) 2017 ini->smcr_version |= SMC_V2; 2018 } 2019 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 2020 rc = SMC_CLC_DECL_PEERNOSMC; 2021 goto out; 2022 } 2023 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 2024 if (!pclc_v2_ext) { 2025 ini->smcd_version &= ~SMC_V2; 2026 ini->smcr_version &= ~SMC_V2; 2027 rc = SMC_CLC_DECL_NOV2EXT; 2028 goto out; 2029 } 2030 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2031 if (ini->smcd_version & SMC_V2) { 2032 if (!smc_ism_is_v2_capable()) { 2033 ini->smcd_version &= ~SMC_V2; 2034 rc = SMC_CLC_DECL_NOISM2SUPP; 2035 } else if (!pclc_smcd_v2_ext) { 2036 ini->smcd_version &= ~SMC_V2; 2037 rc = SMC_CLC_DECL_NOV2DEXT; 2038 } else if (!pclc_v2_ext->hdr.eid_cnt && 2039 !pclc_v2_ext->hdr.flag.seid) { 2040 ini->smcd_version &= ~SMC_V2; 2041 rc = SMC_CLC_DECL_NOUEID; 2042 } 2043 } 2044 if (ini->smcr_version & SMC_V2) { 2045 if (!pclc_v2_ext->hdr.eid_cnt) { 2046 ini->smcr_version &= ~SMC_V2; 2047 rc = SMC_CLC_DECL_NOUEID; 2048 } 2049 } 2050 2051 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2052 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2053 ini->release_nr = SMC_RELEASE; 2054 2055 out: 2056 if (!ini->smcd_version && !ini->smcr_version) 2057 return rc; 2058 2059 return 0; 2060 } 2061 2062 /* listen worker: check prefixes */ 2063 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2064 struct smc_clc_msg_proposal *pclc) 2065 { 2066 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2067 struct socket *newclcsock = new_smc->clcsock; 2068 2069 if (pclc->hdr.typev1 == SMC_TYPE_N) 2070 return 0; 2071 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2072 if (!pclc_prfx) 2073 return -EPROTO; 2074 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2075 return SMC_CLC_DECL_DIFFPREFIX; 2076 2077 return 0; 2078 } 2079 2080 /* listen worker: initialize connection and buffers */ 2081 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2082 struct smc_init_info *ini) 2083 { 2084 int rc; 2085 2086 /* allocate connection / link group */ 2087 rc = smc_conn_create(new_smc, ini); 2088 if (rc) 2089 return rc; 2090 2091 /* create send buffer and rmb */ 2092 if (smc_buf_create(new_smc, false)) { 2093 smc_conn_abort(new_smc, ini->first_contact_local); 2094 return SMC_CLC_DECL_MEM; 2095 } 2096 2097 return 0; 2098 } 2099 2100 /* listen worker: initialize connection and buffers for SMC-D */ 2101 static int smc_listen_ism_init(struct smc_sock *new_smc, 2102 struct smc_init_info *ini) 2103 { 2104 int rc; 2105 2106 rc = smc_conn_create(new_smc, ini); 2107 if (rc) 2108 return rc; 2109 2110 /* Create send and receive buffers */ 2111 rc = smc_buf_create(new_smc, true); 2112 if (rc) { 2113 smc_conn_abort(new_smc, ini->first_contact_local); 2114 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2115 SMC_CLC_DECL_MEM; 2116 } 2117 2118 return 0; 2119 } 2120 2121 static bool smc_is_already_selected(struct smcd_dev *smcd, 2122 struct smc_init_info *ini, 2123 int matches) 2124 { 2125 int i; 2126 2127 for (i = 0; i < matches; i++) 2128 if (smcd == ini->ism_dev[i]) 2129 return true; 2130 2131 return false; 2132 } 2133 2134 /* check for ISM devices matching proposed ISM devices */ 2135 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2136 u16 proposed_chid, 2137 struct smcd_gid *proposed_gid, 2138 unsigned int *matches) 2139 { 2140 struct smcd_dev *smcd; 2141 2142 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2143 if (smcd->going_away) 2144 continue; 2145 if (smc_is_already_selected(smcd, ini, *matches)) 2146 continue; 2147 if (smc_ism_get_chid(smcd) == proposed_chid && 2148 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2149 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2150 if (__smc_ism_is_emulated(proposed_chid)) 2151 ini->ism_peer_gid[*matches].gid_ext = 2152 proposed_gid->gid_ext; 2153 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2154 ini->ism_dev[*matches] = smcd; 2155 (*matches)++; 2156 break; 2157 } 2158 } 2159 } 2160 2161 static void smc_init_info_store_rc(u32 rc, struct smc_init_info *ini) 2162 { 2163 if (!ini->rc) 2164 ini->rc = rc; 2165 } 2166 2167 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2168 struct smc_clc_msg_proposal *pclc, 2169 struct smc_init_info *ini) 2170 { 2171 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2172 struct smc_clc_v2_extension *smc_v2_ext; 2173 struct smc_clc_msg_smcd *pclc_smcd; 2174 unsigned int matches = 0; 2175 struct smcd_gid smcd_gid; 2176 u8 smcd_version; 2177 u8 *eid = NULL; 2178 int i, rc; 2179 u16 chid; 2180 2181 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2182 goto not_found; 2183 2184 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2185 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2186 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2187 if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext) 2188 goto not_found; 2189 2190 mutex_lock(&smcd_dev_list.mutex); 2191 if (pclc_smcd->ism.chid) { 2192 /* check for ISM device matching proposed native ISM device */ 2193 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2194 smcd_gid.gid_ext = 0; 2195 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2196 &smcd_gid, &matches); 2197 } 2198 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2199 /* check for ISM devices matching proposed non-native ISM 2200 * devices 2201 */ 2202 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2203 smcd_gid.gid_ext = 0; 2204 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2205 if (__smc_ism_is_emulated(chid)) { 2206 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2207 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2208 /* each Emulated-ISM device takes two GID-CHID 2209 * entries and CHID of the second entry repeats 2210 * that of the first entry. 2211 * 2212 * So check if the next GID-CHID entry exists 2213 * and both two entries' CHIDs are the same. 2214 */ 2215 continue; 2216 smcd_gid.gid_ext = 2217 ntohll(smcd_v2_ext->gidchid[++i].gid); 2218 } 2219 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2220 } 2221 mutex_unlock(&smcd_dev_list.mutex); 2222 2223 if (!ini->ism_dev[0]) { 2224 smc_init_info_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2225 goto not_found; 2226 } 2227 2228 smc_ism_get_system_eid(&eid); 2229 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2230 smcd_v2_ext->system_eid, eid)) 2231 goto not_found; 2232 2233 /* separate - outside the smcd_dev_list.lock */ 2234 smcd_version = ini->smcd_version; 2235 for (i = 0; i < matches; i++) { 2236 ini->smcd_version = SMC_V2; 2237 ini->is_smcd = true; 2238 ini->ism_selected = i; 2239 rc = smc_listen_ism_init(new_smc, ini); 2240 if (rc) { 2241 smc_init_info_store_rc(rc, ini); 2242 /* try next active ISM device */ 2243 continue; 2244 } 2245 return; /* matching and usable V2 ISM device found */ 2246 } 2247 /* no V2 ISM device could be initialized */ 2248 ini->smcd_version = smcd_version; /* restore original value */ 2249 ini->negotiated_eid[0] = 0; 2250 2251 not_found: 2252 ini->smcd_version &= ~SMC_V2; 2253 ini->ism_dev[0] = NULL; 2254 ini->is_smcd = false; 2255 } 2256 2257 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2258 struct smc_clc_msg_proposal *pclc, 2259 struct smc_init_info *ini) 2260 { 2261 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2262 int rc = 0; 2263 2264 /* check if ISM V1 is available */ 2265 if (!(ini->smcd_version & SMC_V1) || 2266 !smcd_indicated(ini->smc_type_v1) || 2267 !pclc_smcd) 2268 goto not_found; 2269 ini->is_smcd = true; /* prepare ISM check */ 2270 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2271 ini->ism_peer_gid[0].gid_ext = 0; 2272 rc = smc_find_ism_device(new_smc, ini); 2273 if (rc) 2274 goto not_found; 2275 ini->ism_selected = 0; 2276 rc = smc_listen_ism_init(new_smc, ini); 2277 if (!rc) 2278 return; /* V1 ISM device found */ 2279 2280 not_found: 2281 smc_init_info_store_rc(rc, ini); 2282 ini->smcd_version &= ~SMC_V1; 2283 ini->ism_dev[0] = NULL; 2284 ini->is_smcd = false; 2285 } 2286 2287 /* listen worker: register buffers */ 2288 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2289 { 2290 struct smc_connection *conn = &new_smc->conn; 2291 2292 if (!local_first) { 2293 /* reg sendbufs if they were vzalloced */ 2294 if (conn->sndbuf_desc->is_vm) { 2295 if (smcr_lgr_reg_sndbufs(conn->lnk, 2296 conn->sndbuf_desc)) 2297 return SMC_CLC_DECL_ERR_REGBUF; 2298 } 2299 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2300 return SMC_CLC_DECL_ERR_REGBUF; 2301 } 2302 2303 return 0; 2304 } 2305 2306 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2307 struct smc_clc_msg_proposal *pclc, 2308 struct smc_init_info *ini) 2309 { 2310 struct smc_clc_v2_extension *smc_v2_ext; 2311 u8 smcr_version; 2312 int rc; 2313 2314 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2315 goto not_found; 2316 2317 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2318 if (!smc_v2_ext || 2319 !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2320 goto not_found; 2321 2322 /* prepare RDMA check */ 2323 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2324 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2325 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2326 ini->check_smcrv2 = true; 2327 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2328 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2329 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2330 rc = smc_find_rdma_device(new_smc, ini); 2331 if (rc) { 2332 smc_init_info_store_rc(rc, ini); 2333 goto not_found; 2334 } 2335 if (!ini->smcrv2.uses_gateway) 2336 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2337 2338 smcr_version = ini->smcr_version; 2339 ini->smcr_version = SMC_V2; 2340 rc = smc_listen_rdma_init(new_smc, ini); 2341 if (!rc) { 2342 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2343 if (rc) 2344 smc_conn_abort(new_smc, ini->first_contact_local); 2345 } 2346 if (!rc) 2347 return; 2348 ini->smcr_version = smcr_version; 2349 smc_init_info_store_rc(rc, ini); 2350 2351 not_found: 2352 ini->smcr_version &= ~SMC_V2; 2353 ini->smcrv2.ib_dev_v2 = NULL; 2354 ini->check_smcrv2 = false; 2355 } 2356 2357 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2358 struct smc_clc_msg_proposal *pclc, 2359 struct smc_init_info *ini) 2360 { 2361 int rc; 2362 2363 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2364 return SMC_CLC_DECL_NOSMCDEV; 2365 2366 /* prepare RDMA check */ 2367 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2368 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2369 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2370 rc = smc_find_rdma_device(new_smc, ini); 2371 if (rc) { 2372 /* no RDMA device found */ 2373 return SMC_CLC_DECL_NOSMCDEV; 2374 } 2375 rc = smc_listen_rdma_init(new_smc, ini); 2376 if (rc) 2377 return rc; 2378 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2379 } 2380 2381 /* determine the local device matching to proposal */ 2382 static int smc_listen_find_device(struct smc_sock *new_smc, 2383 struct smc_clc_msg_proposal *pclc, 2384 struct smc_init_info *ini) 2385 { 2386 int prfx_rc; 2387 2388 /* check for ISM device matching V2 proposed device */ 2389 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2390 if (ini->ism_dev[0]) 2391 return 0; 2392 2393 /* check for matching IP prefix and subnet length (V1) */ 2394 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2395 if (prfx_rc) 2396 smc_init_info_store_rc(prfx_rc, ini); 2397 2398 /* get vlan id from IP device */ 2399 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2400 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2401 2402 /* check for ISM device matching V1 proposed device */ 2403 if (!prfx_rc) 2404 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2405 if (ini->ism_dev[0]) 2406 return 0; 2407 2408 if (!smcr_indicated(pclc->hdr.typev1) && 2409 !smcr_indicated(pclc->hdr.typev2)) 2410 /* skip RDMA and decline */ 2411 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2412 2413 /* check if RDMA V2 is available */ 2414 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2415 if (ini->smcrv2.ib_dev_v2) 2416 return 0; 2417 2418 /* check if RDMA V1 is available */ 2419 if (!prfx_rc) { 2420 int rc; 2421 2422 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2423 smc_init_info_store_rc(rc, ini); 2424 return (!rc) ? 0 : ini->rc; 2425 } 2426 return prfx_rc; 2427 } 2428 2429 /* listen worker: finish RDMA setup */ 2430 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2431 struct smc_clc_msg_accept_confirm *cclc, 2432 bool local_first, 2433 struct smc_init_info *ini) 2434 { 2435 struct smc_link *link = new_smc->conn.lnk; 2436 int reason_code = 0; 2437 2438 if (local_first) 2439 smc_link_save_peer_info(link, cclc, ini); 2440 2441 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2442 return SMC_CLC_DECL_ERR_RTOK; 2443 2444 if (local_first) { 2445 if (smc_ib_ready_link(link)) 2446 return SMC_CLC_DECL_ERR_RDYLNK; 2447 /* QP confirmation over RoCE fabric */ 2448 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2449 reason_code = smcr_serv_conf_first_link(new_smc); 2450 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2451 } 2452 return reason_code; 2453 } 2454 2455 /* setup for connection of server */ 2456 static void smc_listen_work(struct work_struct *work) 2457 { 2458 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2459 smc_listen_work); 2460 struct socket *newclcsock = new_smc->clcsock; 2461 struct smc_clc_msg_accept_confirm *cclc; 2462 struct smc_clc_msg_proposal_area *buf; 2463 struct smc_clc_msg_proposal *pclc; 2464 struct smc_init_info *ini = NULL; 2465 u8 proposal_version = SMC_V1; 2466 u8 accept_version; 2467 int rc = 0; 2468 2469 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2470 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2471 return smc_listen_out_err(new_smc); 2472 2473 if (new_smc->use_fallback) { 2474 smc_listen_out_connected(new_smc); 2475 return; 2476 } 2477 2478 /* check if peer is smc capable */ 2479 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2480 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2481 if (rc) 2482 smc_listen_out_err(new_smc); 2483 else 2484 smc_listen_out_connected(new_smc); 2485 return; 2486 } 2487 2488 /* do inband token exchange - 2489 * wait for and receive SMC Proposal CLC message 2490 */ 2491 buf = kzalloc_obj(*buf); 2492 if (!buf) { 2493 rc = SMC_CLC_DECL_MEM; 2494 goto out_decl; 2495 } 2496 pclc = (struct smc_clc_msg_proposal *)buf; 2497 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2498 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2499 if (rc) 2500 goto out_decl; 2501 2502 if (pclc->hdr.version > SMC_V1) 2503 proposal_version = SMC_V2; 2504 2505 /* IPSec connections opt out of SMC optimizations */ 2506 if (using_ipsec(new_smc)) { 2507 rc = SMC_CLC_DECL_IPSEC; 2508 goto out_decl; 2509 } 2510 2511 ini = kzalloc_obj(*ini); 2512 if (!ini) { 2513 rc = SMC_CLC_DECL_MEM; 2514 goto out_decl; 2515 } 2516 2517 /* initial version checking */ 2518 rc = smc_listen_v2_check(new_smc, pclc, ini); 2519 if (rc) 2520 goto out_decl; 2521 2522 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2523 if (rc) 2524 goto out_decl; 2525 2526 mutex_lock(&smc_server_lgr_pending); 2527 smc_rx_init(new_smc); 2528 smc_tx_init(new_smc); 2529 2530 /* determine ISM or RoCE device used for connection */ 2531 rc = smc_listen_find_device(new_smc, pclc, ini); 2532 if (rc) 2533 goto out_unlock; 2534 2535 /* send SMC Accept CLC message */ 2536 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2537 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2538 accept_version, ini->negotiated_eid, ini); 2539 if (rc) 2540 goto out_unlock; 2541 2542 /* SMC-D does not need this lock any more */ 2543 if (ini->is_smcd) 2544 mutex_unlock(&smc_server_lgr_pending); 2545 2546 /* receive SMC Confirm CLC message */ 2547 memset(buf, 0, sizeof(*buf)); 2548 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2549 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2550 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2551 if (rc) { 2552 if (!ini->is_smcd) 2553 goto out_unlock; 2554 goto out_decl; 2555 } 2556 2557 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2558 if (rc) { 2559 if (!ini->is_smcd) 2560 goto out_unlock; 2561 goto out_decl; 2562 } 2563 2564 /* fce smc release version is needed in smc_listen_rdma_finish, 2565 * so save fce info here. 2566 */ 2567 smc_conn_save_peer_info_fce(new_smc, cclc); 2568 2569 /* finish worker */ 2570 if (!ini->is_smcd) { 2571 rc = smc_listen_rdma_finish(new_smc, cclc, 2572 ini->first_contact_local, ini); 2573 if (rc) 2574 goto out_unlock; 2575 mutex_unlock(&smc_server_lgr_pending); 2576 } 2577 smc_conn_save_peer_info(new_smc, cclc); 2578 2579 if (ini->is_smcd && 2580 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2581 rc = smcd_buf_attach(new_smc); 2582 if (rc) 2583 goto out_decl; 2584 } 2585 2586 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2587 /* smc_listen_out() will release smcsk */ 2588 smc_listen_out_connected(new_smc); 2589 goto out_free; 2590 2591 out_unlock: 2592 mutex_unlock(&smc_server_lgr_pending); 2593 out_decl: 2594 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2595 proposal_version); 2596 out_free: 2597 kfree(ini); 2598 kfree(buf); 2599 } 2600 2601 static void smc_tcp_listen_work(struct work_struct *work) 2602 { 2603 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2604 tcp_listen_work); 2605 struct sock *lsk = &lsmc->sk; 2606 struct smc_sock *new_smc; 2607 int rc = 0; 2608 2609 lock_sock(lsk); 2610 while (lsk->sk_state == SMC_LISTEN) { 2611 rc = smc_clcsock_accept(lsmc, &new_smc); 2612 if (rc) /* clcsock accept queue empty or error */ 2613 goto out; 2614 if (!new_smc) 2615 continue; 2616 2617 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2618 atomic_inc(&lsmc->queued_smc_hs); 2619 2620 new_smc->listen_smc = lsmc; 2621 new_smc->use_fallback = lsmc->use_fallback; 2622 new_smc->fallback_rsn = lsmc->fallback_rsn; 2623 sock_hold(lsk); /* sock_put in smc_listen_work */ 2624 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2625 smc_copy_sock_settings_to_smc(new_smc); 2626 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2627 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2628 sock_put(&new_smc->sk); 2629 } 2630 2631 out: 2632 release_sock(lsk); 2633 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2634 } 2635 2636 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2637 { 2638 struct smc_sock *lsmc; 2639 2640 read_lock_bh(&listen_clcsock->sk_callback_lock); 2641 lsmc = smc_clcsock_user_data(listen_clcsock); 2642 if (!lsmc) 2643 goto out; 2644 lsmc->clcsk_data_ready(listen_clcsock); 2645 if (lsmc->sk.sk_state == SMC_LISTEN) { 2646 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2647 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2648 sock_put(&lsmc->sk); 2649 } 2650 out: 2651 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2652 } 2653 2654 int smc_listen(struct socket *sock, int backlog) 2655 { 2656 struct sock *sk = sock->sk; 2657 struct smc_sock *smc; 2658 int rc; 2659 2660 smc = smc_sk(sk); 2661 lock_sock(sk); 2662 2663 rc = -EINVAL; 2664 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2665 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2666 goto out; 2667 2668 rc = 0; 2669 if (sk->sk_state == SMC_LISTEN) { 2670 sk->sk_max_ack_backlog = backlog; 2671 goto out; 2672 } 2673 /* some socket options are handled in core, so we could not apply 2674 * them to the clc socket -- copy smc socket options to clc socket 2675 */ 2676 smc_copy_sock_settings_to_clc(smc); 2677 if (!smc->use_fallback) 2678 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2679 2680 /* save original sk_data_ready function and establish 2681 * smc-specific sk_data_ready function 2682 */ 2683 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2684 __rcu_assign_sk_user_data_with_flags(smc->clcsock->sk, smc, 2685 SK_USER_DATA_NOCOPY); 2686 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2687 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2688 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2689 2690 /* save original ops */ 2691 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2692 2693 smc->af_ops = *smc->ori_af_ops; 2694 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2695 2696 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2697 2698 if (smc->limit_smc_hs) 2699 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2700 2701 rc = kernel_listen(smc->clcsock, backlog); 2702 if (rc) { 2703 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2704 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2705 &smc->clcsk_data_ready); 2706 rcu_assign_sk_user_data(smc->clcsock->sk, NULL); 2707 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2708 goto out; 2709 } 2710 sock_set_flag(sk, SOCK_RCU_FREE); 2711 sk->sk_max_ack_backlog = backlog; 2712 sk->sk_ack_backlog = 0; 2713 sk->sk_state = SMC_LISTEN; 2714 2715 out: 2716 release_sock(sk); 2717 return rc; 2718 } 2719 2720 int smc_accept(struct socket *sock, struct socket *new_sock, 2721 struct proto_accept_arg *arg) 2722 { 2723 struct sock *sk = sock->sk, *nsk; 2724 DECLARE_WAITQUEUE(wait, current); 2725 struct smc_sock *lsmc; 2726 long timeo; 2727 int rc = 0; 2728 2729 lsmc = smc_sk(sk); 2730 sock_hold(sk); /* sock_put below */ 2731 lock_sock(sk); 2732 2733 if (lsmc->sk.sk_state != SMC_LISTEN) { 2734 rc = -EINVAL; 2735 release_sock(sk); 2736 goto out; 2737 } 2738 2739 /* Wait for an incoming connection */ 2740 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2741 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2742 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2743 set_current_state(TASK_INTERRUPTIBLE); 2744 if (!timeo) { 2745 rc = -EAGAIN; 2746 break; 2747 } 2748 release_sock(sk); 2749 timeo = schedule_timeout(timeo); 2750 /* wakeup by sk_data_ready in smc_listen_work() */ 2751 sched_annotate_sleep(); 2752 lock_sock(sk); 2753 if (signal_pending(current)) { 2754 rc = sock_intr_errno(timeo); 2755 break; 2756 } 2757 } 2758 set_current_state(TASK_RUNNING); 2759 remove_wait_queue(sk_sleep(sk), &wait); 2760 2761 if (!rc) 2762 rc = sock_error(nsk); 2763 release_sock(sk); 2764 if (rc) 2765 goto out; 2766 2767 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2768 /* wait till data arrives on the socket */ 2769 timeo = secs_to_jiffies(lsmc->sockopt_defer_accept); 2770 if (smc_sk(nsk)->use_fallback) { 2771 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2772 2773 lock_sock(clcsk); 2774 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2775 sk_wait_data(clcsk, &timeo, NULL); 2776 release_sock(clcsk); 2777 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2778 lock_sock(nsk); 2779 smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available); 2780 release_sock(nsk); 2781 } 2782 } 2783 2784 out: 2785 sock_put(sk); /* sock_hold above */ 2786 return rc; 2787 } 2788 2789 int smc_getname(struct socket *sock, struct sockaddr *addr, 2790 int peer) 2791 { 2792 struct smc_sock *smc; 2793 int rc = -EBADF; 2794 2795 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2796 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2797 return -ENOTCONN; 2798 2799 smc = smc_sk(sock->sk); 2800 2801 mutex_lock(&smc->clcsock_release_lock); 2802 if (smc->clcsock) 2803 rc = smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2804 mutex_unlock(&smc->clcsock_release_lock); 2805 return rc; 2806 } 2807 2808 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2809 { 2810 struct sock *sk = sock->sk; 2811 struct smc_sock *smc; 2812 int rc; 2813 2814 smc = smc_sk(sk); 2815 lock_sock(sk); 2816 2817 /* SMC does not support connect with fastopen */ 2818 if (msg->msg_flags & MSG_FASTOPEN) { 2819 /* not connected yet, fallback */ 2820 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2821 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2822 if (rc) 2823 goto out; 2824 } else { 2825 rc = -EINVAL; 2826 goto out; 2827 } 2828 } else if ((sk->sk_state != SMC_ACTIVE) && 2829 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2830 (sk->sk_state != SMC_INIT)) { 2831 rc = -EPIPE; 2832 goto out; 2833 } 2834 2835 if (smc->use_fallback) { 2836 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2837 } else { 2838 rc = smc_tx_sendmsg(smc, msg, len); 2839 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2840 } 2841 out: 2842 release_sock(sk); 2843 return rc; 2844 } 2845 2846 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2847 int flags) 2848 { 2849 struct sock *sk = sock->sk; 2850 struct smc_sock *smc; 2851 int rc = -ENOTCONN; 2852 2853 smc = smc_sk(sk); 2854 lock_sock(sk); 2855 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2856 /* socket was connected before, no more data to read */ 2857 rc = 0; 2858 goto out; 2859 } 2860 if ((sk->sk_state == SMC_INIT) || 2861 (sk->sk_state == SMC_LISTEN) || 2862 (sk->sk_state == SMC_CLOSED)) 2863 goto out; 2864 2865 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2866 rc = 0; 2867 goto out; 2868 } 2869 2870 if (smc->use_fallback) { 2871 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2872 } else { 2873 msg->msg_namelen = 0; 2874 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2875 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2876 } 2877 2878 out: 2879 release_sock(sk); 2880 return rc; 2881 } 2882 2883 static __poll_t smc_accept_poll(struct sock *parent) 2884 { 2885 struct smc_sock *isk = smc_sk(parent); 2886 __poll_t mask = 0; 2887 2888 spin_lock(&isk->accept_q_lock); 2889 if (!list_empty(&isk->accept_q)) 2890 mask = EPOLLIN | EPOLLRDNORM; 2891 spin_unlock(&isk->accept_q_lock); 2892 2893 return mask; 2894 } 2895 2896 __poll_t smc_poll(struct file *file, struct socket *sock, 2897 poll_table *wait) 2898 { 2899 struct sock *sk = sock->sk; 2900 struct smc_sock *smc; 2901 __poll_t mask = 0; 2902 2903 if (!sk) 2904 return EPOLLNVAL; 2905 2906 smc = smc_sk(sock->sk); 2907 if (smc->use_fallback) { 2908 /* delegate to CLC child sock */ 2909 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2910 sk->sk_err = smc->clcsock->sk->sk_err; 2911 } else { 2912 if (sk->sk_state != SMC_CLOSED) 2913 sock_poll_wait(file, sock, wait); 2914 if (sk->sk_err) 2915 mask |= EPOLLERR; 2916 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2917 (sk->sk_state == SMC_CLOSED)) 2918 mask |= EPOLLHUP; 2919 if (sk->sk_state == SMC_LISTEN) { 2920 /* woken up by sk_data_ready in smc_listen_work() */ 2921 mask |= smc_accept_poll(sk); 2922 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2923 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2924 wait); 2925 sk->sk_err = smc->clcsock->sk->sk_err; 2926 } else { 2927 if ((sk->sk_state != SMC_INIT && 2928 atomic_read(&smc->conn.sndbuf_space)) || 2929 sk->sk_shutdown & SEND_SHUTDOWN) { 2930 mask |= EPOLLOUT | EPOLLWRNORM; 2931 } else { 2932 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2933 sk_set_nospace(sk); 2934 2935 if (sk->sk_state != SMC_INIT) { 2936 /* Race breaker the same way as tcp_poll(). */ 2937 smp_mb__after_atomic(); 2938 if (atomic_read(&smc->conn.sndbuf_space)) 2939 mask |= EPOLLOUT | EPOLLWRNORM; 2940 } 2941 } 2942 if (atomic_read(&smc->conn.bytes_to_rcv)) 2943 mask |= EPOLLIN | EPOLLRDNORM; 2944 if (sk->sk_shutdown & RCV_SHUTDOWN) 2945 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2946 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2947 mask |= EPOLLIN; 2948 if (smc->conn.urg_state == SMC_URG_VALID) 2949 mask |= EPOLLPRI; 2950 } 2951 } 2952 2953 return mask; 2954 } 2955 2956 int smc_shutdown(struct socket *sock, int how) 2957 { 2958 struct sock *sk = sock->sk; 2959 bool do_shutdown = true; 2960 struct smc_sock *smc; 2961 int rc = -EINVAL; 2962 int old_state; 2963 int rc1 = 0; 2964 2965 smc = smc_sk(sk); 2966 2967 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2968 return rc; 2969 2970 lock_sock(sk); 2971 2972 if (sock->state == SS_CONNECTING) { 2973 if (sk->sk_state == SMC_ACTIVE) 2974 sock->state = SS_CONNECTED; 2975 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2976 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2977 sk->sk_state == SMC_APPCLOSEWAIT1 || 2978 sk->sk_state == SMC_APPCLOSEWAIT2 || 2979 sk->sk_state == SMC_APPFINCLOSEWAIT) 2980 sock->state = SS_DISCONNECTING; 2981 } 2982 2983 rc = -ENOTCONN; 2984 if ((sk->sk_state != SMC_ACTIVE) && 2985 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2986 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2987 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2988 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2989 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2990 goto out; 2991 if (smc->use_fallback) { 2992 rc = kernel_sock_shutdown(smc->clcsock, how); 2993 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2994 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2995 sk->sk_state = SMC_CLOSED; 2996 sk->sk_socket->state = SS_UNCONNECTED; 2997 sock_put(sk); 2998 } 2999 goto out; 3000 } 3001 switch (how) { 3002 case SHUT_RDWR: /* shutdown in both directions */ 3003 old_state = sk->sk_state; 3004 rc = smc_close_active(smc); 3005 if (old_state == SMC_ACTIVE && 3006 sk->sk_state == SMC_PEERCLOSEWAIT1) 3007 do_shutdown = false; 3008 break; 3009 case SHUT_WR: 3010 rc = smc_close_shutdown_write(smc); 3011 break; 3012 case SHUT_RD: 3013 rc = 0; 3014 /* nothing more to do because peer is not involved */ 3015 break; 3016 } 3017 if (do_shutdown && smc->clcsock) 3018 rc1 = kernel_sock_shutdown(smc->clcsock, how); 3019 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 3020 sk->sk_shutdown |= how + 1; 3021 3022 if (sk->sk_state == SMC_CLOSED) 3023 sock->state = SS_UNCONNECTED; 3024 else 3025 sock->state = SS_DISCONNECTING; 3026 out: 3027 release_sock(sk); 3028 return rc ? rc : rc1; 3029 } 3030 3031 static int __smc_getsockopt(struct socket *sock, int level, int optname, 3032 char __user *optval, int __user *optlen) 3033 { 3034 struct smc_sock *smc; 3035 int val, len; 3036 3037 smc = smc_sk(sock->sk); 3038 3039 if (get_user(len, optlen)) 3040 return -EFAULT; 3041 3042 len = min_t(int, len, sizeof(int)); 3043 3044 if (len < 0) 3045 return -EINVAL; 3046 3047 switch (optname) { 3048 case SMC_LIMIT_HS: 3049 val = smc->limit_smc_hs; 3050 break; 3051 default: 3052 return -EOPNOTSUPP; 3053 } 3054 3055 if (put_user(len, optlen)) 3056 return -EFAULT; 3057 if (copy_to_user(optval, &val, len)) 3058 return -EFAULT; 3059 3060 return 0; 3061 } 3062 3063 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3064 sockptr_t optval, unsigned int optlen) 3065 { 3066 struct sock *sk = sock->sk; 3067 struct smc_sock *smc; 3068 int val, rc; 3069 3070 smc = smc_sk(sk); 3071 3072 /* pre-fetch user data outside the lock */ 3073 if (optname == SMC_LIMIT_HS) { 3074 if (optlen < sizeof(int)) 3075 return -EINVAL; 3076 if (copy_from_sockptr(&val, optval, sizeof(int))) 3077 return -EFAULT; 3078 } 3079 3080 lock_sock(sk); 3081 switch (optname) { 3082 case SMC_LIMIT_HS: 3083 smc->limit_smc_hs = !!val; 3084 rc = 0; 3085 break; 3086 default: 3087 rc = -EOPNOTSUPP; 3088 break; 3089 } 3090 release_sock(sk); 3091 3092 return rc; 3093 } 3094 3095 int smc_setsockopt(struct socket *sock, int level, int optname, 3096 sockptr_t optval, unsigned int optlen) 3097 { 3098 struct sock *sk = sock->sk; 3099 struct smc_sock *smc; 3100 int val, rc; 3101 3102 if (level == SOL_TCP && optname == TCP_ULP) 3103 return -EOPNOTSUPP; 3104 else if (level == SOL_SMC) 3105 return __smc_setsockopt(sock, level, optname, optval, optlen); 3106 3107 smc = smc_sk(sk); 3108 3109 /* generic setsockopts reaching us here always apply to the 3110 * CLC socket 3111 */ 3112 mutex_lock(&smc->clcsock_release_lock); 3113 if (!smc->clcsock) { 3114 mutex_unlock(&smc->clcsock_release_lock); 3115 return -EBADF; 3116 } 3117 if (unlikely(!smc->clcsock->ops->setsockopt)) 3118 rc = -EOPNOTSUPP; 3119 else 3120 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3121 optval, optlen); 3122 if (smc->clcsock->sk->sk_err) { 3123 sk->sk_err = smc->clcsock->sk->sk_err; 3124 sk_error_report(sk); 3125 } 3126 mutex_unlock(&smc->clcsock_release_lock); 3127 3128 if (optlen < sizeof(int)) 3129 return -EINVAL; 3130 if (copy_from_sockptr(&val, optval, sizeof(int))) 3131 return -EFAULT; 3132 3133 lock_sock(sk); 3134 if (rc || smc->use_fallback) 3135 goto out; 3136 switch (optname) { 3137 case TCP_FASTOPEN: 3138 case TCP_FASTOPEN_CONNECT: 3139 case TCP_FASTOPEN_KEY: 3140 case TCP_FASTOPEN_NO_COOKIE: 3141 /* option not supported by SMC */ 3142 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3143 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3144 } else { 3145 rc = -EINVAL; 3146 } 3147 break; 3148 case TCP_NODELAY: 3149 if (sk->sk_state != SMC_INIT && 3150 sk->sk_state != SMC_LISTEN && 3151 sk->sk_state != SMC_CLOSED) { 3152 if (val) { 3153 SMC_STAT_INC(smc, ndly_cnt); 3154 smc_tx_pending(&smc->conn); 3155 if (cancel_delayed_work(&smc->conn.tx_work)) 3156 sock_put(sk); 3157 } 3158 } 3159 break; 3160 case TCP_CORK: 3161 if (sk->sk_state != SMC_INIT && 3162 sk->sk_state != SMC_LISTEN && 3163 sk->sk_state != SMC_CLOSED) { 3164 if (!val) { 3165 SMC_STAT_INC(smc, cork_cnt); 3166 smc_tx_pending(&smc->conn); 3167 if (cancel_delayed_work(&smc->conn.tx_work)) 3168 sock_put(sk); 3169 } 3170 } 3171 break; 3172 case TCP_DEFER_ACCEPT: 3173 smc->sockopt_defer_accept = val; 3174 break; 3175 default: 3176 break; 3177 } 3178 out: 3179 release_sock(sk); 3180 3181 return rc; 3182 } 3183 3184 int smc_getsockopt(struct socket *sock, int level, int optname, 3185 char __user *optval, int __user *optlen) 3186 { 3187 struct smc_sock *smc; 3188 int rc; 3189 3190 if (level == SOL_SMC) 3191 return __smc_getsockopt(sock, level, optname, optval, optlen); 3192 3193 smc = smc_sk(sock->sk); 3194 mutex_lock(&smc->clcsock_release_lock); 3195 if (!smc->clcsock) { 3196 mutex_unlock(&smc->clcsock_release_lock); 3197 return -EBADF; 3198 } 3199 /* socket options apply to the CLC socket */ 3200 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3201 mutex_unlock(&smc->clcsock_release_lock); 3202 return -EOPNOTSUPP; 3203 } 3204 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3205 optval, optlen); 3206 mutex_unlock(&smc->clcsock_release_lock); 3207 return rc; 3208 } 3209 3210 int smc_ioctl(struct socket *sock, unsigned int cmd, 3211 unsigned long arg) 3212 { 3213 union smc_host_cursor cons, urg; 3214 struct smc_connection *conn; 3215 struct smc_sock *smc; 3216 int answ; 3217 3218 smc = smc_sk(sock->sk); 3219 conn = &smc->conn; 3220 lock_sock(&smc->sk); 3221 if (smc->use_fallback) { 3222 if (!smc->clcsock) { 3223 release_sock(&smc->sk); 3224 return -EBADF; 3225 } 3226 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3227 release_sock(&smc->sk); 3228 return answ; 3229 } 3230 switch (cmd) { 3231 case SIOCINQ: /* same as FIONREAD */ 3232 if (smc->sk.sk_state == SMC_LISTEN) { 3233 release_sock(&smc->sk); 3234 return -EINVAL; 3235 } 3236 if (smc->sk.sk_state == SMC_INIT || 3237 smc->sk.sk_state == SMC_CLOSED) 3238 answ = 0; 3239 else 3240 answ = atomic_read(&smc->conn.bytes_to_rcv); 3241 break; 3242 case SIOCOUTQ: 3243 /* output queue size (not send + not acked) */ 3244 if (smc->sk.sk_state == SMC_LISTEN) { 3245 release_sock(&smc->sk); 3246 return -EINVAL; 3247 } 3248 if (smc->sk.sk_state == SMC_INIT || 3249 smc->sk.sk_state == SMC_CLOSED || 3250 !READ_ONCE(smc->conn.sndbuf_desc)) 3251 answ = 0; 3252 else 3253 answ = smc->conn.sndbuf_desc->len - 3254 atomic_read(&smc->conn.sndbuf_space); 3255 break; 3256 case SIOCOUTQNSD: 3257 /* output queue size (not send only) */ 3258 if (smc->sk.sk_state == SMC_LISTEN) { 3259 release_sock(&smc->sk); 3260 return -EINVAL; 3261 } 3262 if (smc->sk.sk_state == SMC_INIT || 3263 smc->sk.sk_state == SMC_CLOSED) 3264 answ = 0; 3265 else 3266 answ = smc_tx_prepared_sends(&smc->conn); 3267 break; 3268 case SIOCATMARK: 3269 if (smc->sk.sk_state == SMC_LISTEN) { 3270 release_sock(&smc->sk); 3271 return -EINVAL; 3272 } 3273 if (smc->sk.sk_state == SMC_INIT || 3274 smc->sk.sk_state == SMC_CLOSED) { 3275 answ = 0; 3276 } else { 3277 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3278 smc_curs_copy(&urg, &conn->urg_curs, conn); 3279 answ = smc_curs_diff(conn->rmb_desc->len, 3280 &cons, &urg) == 1; 3281 } 3282 break; 3283 default: 3284 release_sock(&smc->sk); 3285 return -ENOIOCTLCMD; 3286 } 3287 release_sock(&smc->sk); 3288 3289 return put_user(answ, (int __user *)arg); 3290 } 3291 3292 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3293 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3294 * updates till whenever a respective page has been fully processed. 3295 * Note that subsequent recv() calls have to wait till all splice() processing 3296 * completed. 3297 */ 3298 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3299 struct pipe_inode_info *pipe, size_t len, 3300 unsigned int flags) 3301 { 3302 struct sock *sk = sock->sk; 3303 struct smc_sock *smc; 3304 int rc = -ENOTCONN; 3305 3306 smc = smc_sk(sk); 3307 lock_sock(sk); 3308 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3309 /* socket was connected before, no more data to read */ 3310 rc = 0; 3311 goto out; 3312 } 3313 if (sk->sk_state == SMC_INIT || 3314 sk->sk_state == SMC_LISTEN || 3315 sk->sk_state == SMC_CLOSED) 3316 goto out; 3317 3318 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3319 rc = 0; 3320 goto out; 3321 } 3322 3323 if (smc->use_fallback) { 3324 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3325 pipe, len, flags); 3326 } else { 3327 if (*ppos) { 3328 rc = -ESPIPE; 3329 goto out; 3330 } 3331 if (flags & SPLICE_F_NONBLOCK) 3332 flags = MSG_DONTWAIT; 3333 else 3334 flags = 0; 3335 SMC_STAT_INC(smc, splice_cnt); 3336 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3337 } 3338 out: 3339 release_sock(sk); 3340 3341 return rc; 3342 } 3343 3344 /* must look like tcp */ 3345 static const struct proto_ops smc_sock_ops = { 3346 .family = PF_SMC, 3347 .owner = THIS_MODULE, 3348 .release = smc_release, 3349 .bind = smc_bind, 3350 .connect = smc_connect, 3351 .socketpair = sock_no_socketpair, 3352 .accept = smc_accept, 3353 .getname = smc_getname, 3354 .poll = smc_poll, 3355 .ioctl = smc_ioctl, 3356 .listen = smc_listen, 3357 .shutdown = smc_shutdown, 3358 .setsockopt = smc_setsockopt, 3359 .getsockopt = smc_getsockopt, 3360 .sendmsg = smc_sendmsg, 3361 .recvmsg = smc_recvmsg, 3362 .mmap = sock_no_mmap, 3363 .splice_read = smc_splice_read, 3364 }; 3365 3366 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3367 { 3368 struct smc_sock *smc = smc_sk(sk); 3369 int rc; 3370 3371 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3372 &smc->clcsock); 3373 if (rc) 3374 return rc; 3375 3376 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3377 * destruction; its sk_state might not be TCP_CLOSE after 3378 * smc->sk is close()d, and TCP timers can be fired later, 3379 * which need net ref. 3380 */ 3381 sk = smc->clcsock->sk; 3382 sk_net_refcnt_upgrade(sk); 3383 return 0; 3384 } 3385 3386 static int smc_create(struct net *net, struct socket *sock, int protocol, 3387 int kern) 3388 { 3389 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3390 struct sock *sk; 3391 int rc; 3392 3393 rc = -ESOCKTNOSUPPORT; 3394 if (sock->type != SOCK_STREAM) 3395 goto out; 3396 3397 rc = -EPROTONOSUPPORT; 3398 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3399 goto out; 3400 3401 rc = -ENOBUFS; 3402 sock->ops = &smc_sock_ops; 3403 sock->state = SS_UNCONNECTED; 3404 sk = smc_sock_alloc(net, sock, protocol); 3405 if (!sk) 3406 goto out; 3407 3408 rc = smc_create_clcsk(net, sk, family); 3409 if (rc) { 3410 sk_common_release(sk); 3411 sock->sk = NULL; 3412 } 3413 out: 3414 return rc; 3415 } 3416 3417 static const struct net_proto_family smc_sock_family_ops = { 3418 .family = PF_SMC, 3419 .owner = THIS_MODULE, 3420 .create = smc_create, 3421 }; 3422 3423 unsigned int smc_net_id; 3424 3425 static __net_init int smc_net_init(struct net *net) 3426 { 3427 int rc; 3428 3429 rc = smc_sysctl_net_init(net); 3430 if (rc) 3431 return rc; 3432 return smc_pnet_net_init(net); 3433 } 3434 3435 static void __net_exit smc_net_exit(struct net *net) 3436 { 3437 smc_sysctl_net_exit(net); 3438 smc_pnet_net_exit(net); 3439 } 3440 3441 static __net_init int smc_net_stat_init(struct net *net) 3442 { 3443 return smc_stats_init(net); 3444 } 3445 3446 static void __net_exit smc_net_stat_exit(struct net *net) 3447 { 3448 smc_stats_exit(net); 3449 } 3450 3451 static struct pernet_operations smc_net_ops = { 3452 .init = smc_net_init, 3453 .exit = smc_net_exit, 3454 .id = &smc_net_id, 3455 .size = sizeof(struct smc_net), 3456 }; 3457 3458 static struct pernet_operations smc_net_stat_ops = { 3459 .init = smc_net_stat_init, 3460 .exit = smc_net_stat_exit, 3461 }; 3462 3463 static int __init smc_init(void) 3464 { 3465 int rc; 3466 3467 rc = register_pernet_subsys(&smc_net_ops); 3468 if (rc) 3469 return rc; 3470 3471 rc = register_pernet_subsys(&smc_net_stat_ops); 3472 if (rc) 3473 goto out_pernet_subsys; 3474 3475 rc = smc_ism_init(); 3476 if (rc) 3477 goto out_pernet_subsys_stat; 3478 smc_clc_init(); 3479 3480 rc = smc_nl_init(); 3481 if (rc) 3482 goto out_ism; 3483 3484 rc = smc_pnet_init(); 3485 if (rc) 3486 goto out_nl; 3487 3488 rc = -ENOMEM; 3489 3490 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", WQ_PERCPU, 0); 3491 if (!smc_tcp_ls_wq) 3492 goto out_pnet; 3493 3494 smc_hs_wq = alloc_workqueue("smc_hs_wq", WQ_PERCPU, 0); 3495 if (!smc_hs_wq) 3496 goto out_alloc_tcp_ls_wq; 3497 3498 smc_close_wq = alloc_workqueue("smc_close_wq", WQ_PERCPU, 0); 3499 if (!smc_close_wq) 3500 goto out_alloc_hs_wq; 3501 3502 rc = smc_core_init(); 3503 if (rc) { 3504 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3505 goto out_alloc_wqs; 3506 } 3507 3508 rc = smc_llc_init(); 3509 if (rc) { 3510 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3511 goto out_core; 3512 } 3513 3514 rc = smc_cdc_init(); 3515 if (rc) { 3516 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3517 goto out_core; 3518 } 3519 3520 rc = proto_register(&smc_proto, 1); 3521 if (rc) { 3522 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3523 goto out_core; 3524 } 3525 3526 rc = proto_register(&smc_proto6, 1); 3527 if (rc) { 3528 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3529 goto out_proto; 3530 } 3531 3532 rc = sock_register(&smc_sock_family_ops); 3533 if (rc) { 3534 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3535 goto out_proto6; 3536 } 3537 3538 rc = smc_ib_register_client(); 3539 if (rc) { 3540 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3541 goto out_sock; 3542 } 3543 rc = smc_inet_init(); 3544 if (rc) { 3545 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3546 goto out_ib; 3547 } 3548 rc = bpf_smc_hs_ctrl_init(); 3549 if (rc) { 3550 pr_err("%s: bpf_smc_hs_ctrl_init fails with %d\n", __func__, 3551 rc); 3552 goto out_inet; 3553 } 3554 static_branch_enable(&tcp_have_smc); 3555 return 0; 3556 out_inet: 3557 smc_inet_exit(); 3558 out_ib: 3559 smc_ib_unregister_client(); 3560 out_sock: 3561 sock_unregister(PF_SMC); 3562 out_proto6: 3563 proto_unregister(&smc_proto6); 3564 out_proto: 3565 proto_unregister(&smc_proto); 3566 out_core: 3567 smc_core_exit(); 3568 out_alloc_wqs: 3569 destroy_workqueue(smc_close_wq); 3570 out_alloc_hs_wq: 3571 destroy_workqueue(smc_hs_wq); 3572 out_alloc_tcp_ls_wq: 3573 destroy_workqueue(smc_tcp_ls_wq); 3574 out_pnet: 3575 smc_pnet_exit(); 3576 out_nl: 3577 smc_nl_exit(); 3578 out_ism: 3579 smc_clc_exit(); 3580 smc_ism_exit(); 3581 out_pernet_subsys_stat: 3582 unregister_pernet_subsys(&smc_net_stat_ops); 3583 out_pernet_subsys: 3584 unregister_pernet_subsys(&smc_net_ops); 3585 3586 return rc; 3587 } 3588 3589 static void __exit smc_exit(void) 3590 { 3591 static_branch_disable(&tcp_have_smc); 3592 smc_inet_exit(); 3593 sock_unregister(PF_SMC); 3594 smc_core_exit(); 3595 smc_ib_unregister_client(); 3596 smc_ism_exit(); 3597 destroy_workqueue(smc_close_wq); 3598 destroy_workqueue(smc_tcp_ls_wq); 3599 destroy_workqueue(smc_hs_wq); 3600 proto_unregister(&smc_proto6); 3601 proto_unregister(&smc_proto); 3602 smc_pnet_exit(); 3603 smc_nl_exit(); 3604 smc_clc_exit(); 3605 unregister_pernet_subsys(&smc_net_stat_ops); 3606 unregister_pernet_subsys(&smc_net_ops); 3607 rcu_barrier(); 3608 } 3609 3610 module_init(smc_init); 3611 module_exit(smc_exit); 3612 3613 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3614 MODULE_DESCRIPTION("smc socket address family"); 3615 MODULE_LICENSE("GPL"); 3616 MODULE_ALIAS_NETPROTO(PF_SMC); 3617 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3618 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3619 #if IS_ENABLED(CONFIG_IPV6) 3620 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3621 #endif /* CONFIG_IPV6 */ 3622 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3623