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