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 KMSG_COMPONENT "smc" 20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 21 22 #include <linux/module.h> 23 #include <linux/socket.h> 24 #include <linux/workqueue.h> 25 #include <linux/in.h> 26 #include <linux/sched/signal.h> 27 #include <linux/if_vlan.h> 28 #include <linux/rcupdate_wait.h> 29 #include <linux/ctype.h> 30 #include <linux/splice.h> 31 32 #include <net/sock.h> 33 #include <net/inet_common.h> 34 #if IS_ENABLED(CONFIG_IPV6) 35 #include <net/ipv6.h> 36 #endif 37 #include <net/tcp.h> 38 #include <net/smc.h> 39 #include <asm/ioctls.h> 40 41 #include <net/net_namespace.h> 42 #include <net/netns/generic.h> 43 #include "smc_netns.h" 44 45 #include "smc.h" 46 #include "smc_clc.h" 47 #include "smc_llc.h" 48 #include "smc_cdc.h" 49 #include "smc_core.h" 50 #include "smc_ib.h" 51 #include "smc_ism.h" 52 #include "smc_pnet.h" 53 #include "smc_netlink.h" 54 #include "smc_tx.h" 55 #include "smc_rx.h" 56 #include "smc_close.h" 57 #include "smc_stats.h" 58 #include "smc_tracepoint.h" 59 #include "smc_sysctl.h" 60 #include "smc_inet.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 { 129 struct smc_sock *smc; 130 struct sock *child; 131 132 smc = smc_clcsock_user_data(sk); 133 134 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 135 sk->sk_max_ack_backlog) 136 goto drop; 137 138 if (sk_acceptq_is_full(&smc->sk)) { 139 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 140 goto drop; 141 } 142 143 /* passthrough to original syn recv sock fct */ 144 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 145 own_req); 146 /* child must not inherit smc or its ops */ 147 if (child) { 148 rcu_assign_sk_user_data(child, NULL); 149 150 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 151 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 152 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 153 } 154 return child; 155 156 drop: 157 dst_release(dst); 158 tcp_listendrop(sk); 159 return NULL; 160 } 161 162 static bool smc_hs_congested(const struct sock *sk) 163 { 164 const struct smc_sock *smc; 165 166 smc = smc_clcsock_user_data(sk); 167 168 if (!smc) 169 return true; 170 171 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 172 return true; 173 174 return false; 175 } 176 177 struct smc_hashinfo smc_v4_hashinfo = { 178 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 179 }; 180 181 struct smc_hashinfo smc_v6_hashinfo = { 182 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 183 }; 184 185 int smc_hash_sk(struct sock *sk) 186 { 187 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 188 struct hlist_head *head; 189 190 head = &h->ht; 191 192 write_lock_bh(&h->lock); 193 sk_add_node(sk, head); 194 write_unlock_bh(&h->lock); 195 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 196 197 return 0; 198 } 199 200 void smc_unhash_sk(struct sock *sk) 201 { 202 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 203 204 write_lock_bh(&h->lock); 205 if (sk_del_node_init(sk)) 206 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 207 write_unlock_bh(&h->lock); 208 } 209 210 /* This will be called before user really release sock_lock. So do the 211 * work which we didn't do because of user hold the sock_lock in the 212 * BH context 213 */ 214 void smc_release_cb(struct sock *sk) 215 { 216 struct smc_sock *smc = smc_sk(sk); 217 218 if (smc->conn.tx_in_release_sock) { 219 smc_tx_pending(&smc->conn); 220 smc->conn.tx_in_release_sock = false; 221 } 222 } 223 224 struct proto smc_proto = { 225 .name = "SMC", 226 .owner = THIS_MODULE, 227 .keepalive = smc_set_keepalive, 228 .hash = smc_hash_sk, 229 .unhash = smc_unhash_sk, 230 .release_cb = smc_release_cb, 231 .obj_size = sizeof(struct smc_sock), 232 .h.smc_hash = &smc_v4_hashinfo, 233 .slab_flags = SLAB_TYPESAFE_BY_RCU, 234 }; 235 EXPORT_SYMBOL_GPL(smc_proto); 236 237 struct proto smc_proto6 = { 238 .name = "SMC6", 239 .owner = THIS_MODULE, 240 .keepalive = smc_set_keepalive, 241 .hash = smc_hash_sk, 242 .unhash = smc_unhash_sk, 243 .release_cb = smc_release_cb, 244 .obj_size = sizeof(struct smc_sock), 245 .h.smc_hash = &smc_v6_hashinfo, 246 .slab_flags = SLAB_TYPESAFE_BY_RCU, 247 }; 248 EXPORT_SYMBOL_GPL(smc_proto6); 249 250 static void smc_fback_restore_callbacks(struct smc_sock *smc) 251 { 252 struct sock *clcsk = smc->clcsock->sk; 253 254 write_lock_bh(&clcsk->sk_callback_lock); 255 clcsk->sk_user_data = NULL; 256 257 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 258 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 259 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 260 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 261 262 write_unlock_bh(&clcsk->sk_callback_lock); 263 } 264 265 static void smc_restore_fallback_changes(struct smc_sock *smc) 266 { 267 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 268 smc->clcsock->file->private_data = smc->sk.sk_socket; 269 smc->clcsock->file = NULL; 270 smc_fback_restore_callbacks(smc); 271 } 272 } 273 274 static int __smc_release(struct smc_sock *smc) 275 { 276 struct sock *sk = &smc->sk; 277 int rc = 0; 278 279 if (!smc->use_fallback) { 280 rc = smc_close_active(smc); 281 smc_sock_set_flag(sk, SOCK_DEAD); 282 sk->sk_shutdown |= SHUTDOWN_MASK; 283 } else { 284 if (sk->sk_state != SMC_CLOSED) { 285 if (sk->sk_state != SMC_LISTEN && 286 sk->sk_state != SMC_INIT) 287 sock_put(sk); /* passive closing */ 288 if (sk->sk_state == SMC_LISTEN) { 289 /* wake up clcsock accept */ 290 rc = kernel_sock_shutdown(smc->clcsock, 291 SHUT_RDWR); 292 } 293 sk->sk_state = SMC_CLOSED; 294 sk->sk_state_change(sk); 295 } 296 smc_restore_fallback_changes(smc); 297 } 298 299 sk->sk_prot->unhash(sk); 300 301 if (sk->sk_state == SMC_CLOSED) { 302 if (smc->clcsock) { 303 release_sock(sk); 304 smc_clcsock_release(smc); 305 lock_sock(sk); 306 } 307 if (!smc->use_fallback) 308 smc_conn_free(&smc->conn); 309 } 310 311 return rc; 312 } 313 314 int smc_release(struct socket *sock) 315 { 316 struct sock *sk = sock->sk; 317 struct smc_sock *smc; 318 int old_state, rc = 0; 319 320 if (!sk) 321 goto out; 322 323 sock_hold(sk); /* sock_put below */ 324 smc = smc_sk(sk); 325 326 old_state = sk->sk_state; 327 328 /* cleanup for a dangling non-blocking connect */ 329 if (smc->connect_nonblock && old_state == SMC_INIT) 330 tcp_abort(smc->clcsock->sk, ECONNABORTED); 331 332 if (cancel_work_sync(&smc->connect_work)) 333 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 334 335 if (sk->sk_state == SMC_LISTEN) 336 /* smc_close_non_accepted() is called and acquires 337 * sock lock for child sockets again 338 */ 339 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 340 else 341 lock_sock(sk); 342 343 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 344 !smc->use_fallback) 345 smc_close_active_abort(smc); 346 347 rc = __smc_release(smc); 348 349 /* detach socket */ 350 sock_orphan(sk); 351 sock->sk = NULL; 352 release_sock(sk); 353 354 sock_put(sk); /* sock_hold above */ 355 sock_put(sk); /* final sock_put */ 356 out: 357 return rc; 358 } 359 360 static void smc_destruct(struct sock *sk) 361 { 362 if (sk->sk_state != SMC_CLOSED) 363 return; 364 if (!sock_flag(sk, SOCK_DEAD)) 365 return; 366 switch (sk->sk_family) { 367 case AF_INET: 368 inet_sock_destruct(sk); 369 break; 370 #if IS_ENABLED(CONFIG_IPV6) 371 case AF_INET6: 372 inet6_sock_destruct(sk); 373 break; 374 #endif 375 } 376 } 377 378 static struct lock_class_key smc_key; 379 static struct lock_class_key smc_slock_key; 380 381 void smc_sk_init(struct net *net, struct sock *sk, int protocol) 382 { 383 struct smc_sock *smc = smc_sk(sk); 384 385 sk->sk_state = SMC_INIT; 386 sk->sk_destruct = smc_destruct; 387 sk->sk_protocol = protocol; 388 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 389 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 390 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 391 INIT_WORK(&smc->connect_work, smc_connect_work); 392 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 393 INIT_LIST_HEAD(&smc->accept_q); 394 sock_lock_init_class_and_name(sk, "slock-AF_SMC", &smc_slock_key, 395 "sk_lock-AF_SMC", &smc_key); 396 spin_lock_init(&smc->accept_q_lock); 397 spin_lock_init(&smc->conn.send_lock); 398 sk->sk_prot->hash(sk); 399 mutex_init(&smc->clcsock_release_lock); 400 smc_init_saved_callbacks(smc); 401 smc->limit_smc_hs = net->smc.limit_smc_hs; 402 smc->use_fallback = false; /* assume rdma capability first */ 403 smc->fallback_rsn = 0; 404 smc_close_init(smc); 405 } 406 407 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 408 int protocol) 409 { 410 struct proto *prot; 411 struct sock *sk; 412 413 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 414 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 415 if (!sk) 416 return NULL; 417 418 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 419 smc_sk_init(net, sk, protocol); 420 421 return sk; 422 } 423 424 int smc_bind(struct socket *sock, struct sockaddr *uaddr, 425 int addr_len) 426 { 427 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 428 struct sock *sk = sock->sk; 429 struct smc_sock *smc; 430 int rc; 431 432 smc = smc_sk(sk); 433 434 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 435 rc = -EINVAL; 436 if (addr_len < sizeof(struct sockaddr_in)) 437 goto out; 438 439 rc = -EAFNOSUPPORT; 440 if (addr->sin_family != AF_INET && 441 addr->sin_family != AF_INET6 && 442 addr->sin_family != AF_UNSPEC) 443 goto out; 444 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 445 if (addr->sin_family == AF_UNSPEC && 446 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 447 goto out; 448 449 lock_sock(sk); 450 451 /* Check if socket is already active */ 452 rc = -EINVAL; 453 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 454 goto out_rel; 455 456 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 457 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 458 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 459 460 out_rel: 461 release_sock(sk); 462 out: 463 return rc; 464 } 465 466 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 467 * clc socket (since smc is not called for these options from net/core) 468 */ 469 470 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 471 (1UL << SOCK_KEEPOPEN) | \ 472 (1UL << SOCK_LINGER) | \ 473 (1UL << SOCK_BROADCAST) | \ 474 (1UL << SOCK_TIMESTAMP) | \ 475 (1UL << SOCK_DBG) | \ 476 (1UL << SOCK_RCVTSTAMP) | \ 477 (1UL << SOCK_RCVTSTAMPNS) | \ 478 (1UL << SOCK_LOCALROUTE) | \ 479 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 480 (1UL << SOCK_RXQ_OVFL) | \ 481 (1UL << SOCK_WIFI_STATUS) | \ 482 (1UL << SOCK_NOFCS) | \ 483 (1UL << SOCK_FILTER_LOCKED) | \ 484 (1UL << SOCK_TSTAMP_NEW)) 485 486 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 487 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 488 unsigned long mask) 489 { 490 nsk->sk_userlocks = osk->sk_userlocks; 491 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) 492 nsk->sk_sndbuf = osk->sk_sndbuf; 493 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) 494 nsk->sk_rcvbuf = osk->sk_rcvbuf; 495 } 496 497 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 498 unsigned long mask) 499 { 500 /* options we don't get control via setsockopt for */ 501 nsk->sk_type = osk->sk_type; 502 nsk->sk_sndtimeo = READ_ONCE(osk->sk_sndtimeo); 503 nsk->sk_rcvtimeo = READ_ONCE(osk->sk_rcvtimeo); 504 nsk->sk_mark = READ_ONCE(osk->sk_mark); 505 nsk->sk_priority = READ_ONCE(osk->sk_priority); 506 nsk->sk_rcvlowat = osk->sk_rcvlowat; 507 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 508 nsk->sk_err = osk->sk_err; 509 510 nsk->sk_flags &= ~mask; 511 nsk->sk_flags |= osk->sk_flags & mask; 512 513 smc_adjust_sock_bufsizes(nsk, osk, mask); 514 } 515 516 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 517 { 518 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 519 } 520 521 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 522 (1UL << SOCK_KEEPOPEN) | \ 523 (1UL << SOCK_LINGER) | \ 524 (1UL << SOCK_DBG)) 525 /* copy only settings and flags relevant for smc from clc to smc socket */ 526 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 527 { 528 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 529 } 530 531 /* register the new vzalloced sndbuf on all links */ 532 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 533 struct smc_buf_desc *snd_desc) 534 { 535 struct smc_link_group *lgr = link->lgr; 536 int i, rc = 0; 537 538 if (!snd_desc->is_vm) 539 return -EINVAL; 540 541 /* protect against parallel smcr_link_reg_buf() */ 542 down_write(&lgr->llc_conf_mutex); 543 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 544 if (!smc_link_active(&lgr->lnk[i])) 545 continue; 546 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 547 if (rc) 548 break; 549 } 550 up_write(&lgr->llc_conf_mutex); 551 return rc; 552 } 553 554 /* register the new rmb on all links */ 555 static int smcr_lgr_reg_rmbs(struct smc_link *link, 556 struct smc_buf_desc *rmb_desc) 557 { 558 struct smc_link_group *lgr = link->lgr; 559 bool do_slow = false; 560 int i, rc = 0; 561 562 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 563 if (rc) 564 return rc; 565 566 down_read(&lgr->llc_conf_mutex); 567 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 568 if (!smc_link_active(&lgr->lnk[i])) 569 continue; 570 if (!rmb_desc->is_reg_mr[link->link_idx]) { 571 up_read(&lgr->llc_conf_mutex); 572 goto slow_path; 573 } 574 } 575 /* mr register already */ 576 goto fast_path; 577 slow_path: 578 do_slow = true; 579 /* protect against parallel smc_llc_cli_rkey_exchange() and 580 * parallel smcr_link_reg_buf() 581 */ 582 down_write(&lgr->llc_conf_mutex); 583 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 584 if (!smc_link_active(&lgr->lnk[i])) 585 continue; 586 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 587 if (rc) 588 goto out; 589 } 590 fast_path: 591 /* exchange confirm_rkey msg with peer */ 592 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 593 if (rc) { 594 rc = -EFAULT; 595 goto out; 596 } 597 rmb_desc->is_conf_rkey = true; 598 out: 599 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 600 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 601 return rc; 602 } 603 604 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 605 { 606 struct smc_link *link = smc->conn.lnk; 607 struct smc_llc_qentry *qentry; 608 int rc; 609 610 /* Receive CONFIRM LINK request from server over RoCE fabric. 611 * Increasing the client's timeout by twice as much as the server's 612 * timeout by default can temporarily avoid decline messages of 613 * both sides crossing or colliding 614 */ 615 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 616 SMC_LLC_CONFIRM_LINK); 617 if (!qentry) { 618 struct smc_clc_msg_decline dclc; 619 620 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 621 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 622 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 623 } 624 smc_llc_save_peer_uid(qentry); 625 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 626 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 627 if (rc) 628 return SMC_CLC_DECL_RMBE_EC; 629 630 rc = smc_ib_modify_qp_rts(link); 631 if (rc) 632 return SMC_CLC_DECL_ERR_RDYLNK; 633 634 smc_wr_remember_qp_attr(link); 635 636 /* reg the sndbuf if it was vzalloced */ 637 if (smc->conn.sndbuf_desc->is_vm) { 638 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 639 return SMC_CLC_DECL_ERR_REGBUF; 640 } 641 642 /* reg the rmb */ 643 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 644 return SMC_CLC_DECL_ERR_REGBUF; 645 646 /* confirm_rkey is implicit on 1st contact */ 647 smc->conn.rmb_desc->is_conf_rkey = true; 648 649 /* send CONFIRM LINK response over RoCE fabric */ 650 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 651 if (rc < 0) 652 return SMC_CLC_DECL_TIMEOUT_CL; 653 654 smc_llc_link_active(link); 655 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 656 657 if (link->lgr->max_links > 1) { 658 /* optional 2nd link, receive ADD LINK request from server */ 659 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 660 SMC_LLC_ADD_LINK); 661 if (!qentry) { 662 struct smc_clc_msg_decline dclc; 663 664 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 665 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 666 if (rc == -EAGAIN) 667 rc = 0; /* no DECLINE received, go with one link */ 668 return rc; 669 } 670 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 671 smc_llc_cli_add_link(link, qentry); 672 } 673 return 0; 674 } 675 676 static bool smc_isascii(char *hostname) 677 { 678 int i; 679 680 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 681 if (!isascii(hostname[i])) 682 return false; 683 return true; 684 } 685 686 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 687 struct smc_clc_msg_accept_confirm *clc) 688 { 689 struct smc_clc_first_contact_ext *fce; 690 int clc_v2_len; 691 692 if (clc->hdr.version == SMC_V1 || 693 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 694 return; 695 696 if (smc->conn.lgr->is_smcd) { 697 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 698 SMC_MAX_EID_LEN); 699 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 700 } else { 701 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 702 SMC_MAX_EID_LEN); 703 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 704 } 705 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 706 smc->conn.lgr->peer_os = fce->os_type; 707 smc->conn.lgr->peer_smc_release = fce->release; 708 if (smc_isascii(fce->hostname)) 709 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 710 SMC_MAX_HOSTNAME_LEN); 711 } 712 713 static void smcr_conn_save_peer_info(struct smc_sock *smc, 714 struct smc_clc_msg_accept_confirm *clc) 715 { 716 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 717 718 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 719 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 720 smc->conn.peer_rmbe_size = bufsize; 721 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 722 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 723 } 724 725 static void smcd_conn_save_peer_info(struct smc_sock *smc, 726 struct smc_clc_msg_accept_confirm *clc) 727 { 728 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 729 730 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 731 smc->conn.peer_token = ntohll(clc->d0.token); 732 /* msg header takes up space in the buffer */ 733 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 734 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 735 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 736 } 737 738 static void smc_conn_save_peer_info(struct smc_sock *smc, 739 struct smc_clc_msg_accept_confirm *clc) 740 { 741 if (smc->conn.lgr->is_smcd) 742 smcd_conn_save_peer_info(smc, clc); 743 else 744 smcr_conn_save_peer_info(smc, clc); 745 smc_conn_save_peer_info_fce(smc, clc); 746 } 747 748 static void smc_link_save_peer_info(struct smc_link *link, 749 struct smc_clc_msg_accept_confirm *clc, 750 struct smc_init_info *ini) 751 { 752 link->peer_qpn = ntoh24(clc->r0.qpn); 753 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 754 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 755 link->peer_psn = ntoh24(clc->r0.psn); 756 link->peer_mtu = clc->r0.qp_mtu; 757 } 758 759 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 760 struct smc_stats_fback *fback_arr) 761 { 762 int cnt; 763 764 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 765 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 766 fback_arr[cnt].count++; 767 break; 768 } 769 if (!fback_arr[cnt].fback_code) { 770 fback_arr[cnt].fback_code = smc->fallback_rsn; 771 fback_arr[cnt].count++; 772 break; 773 } 774 } 775 } 776 777 static void smc_stat_fallback(struct smc_sock *smc) 778 { 779 struct net *net = sock_net(&smc->sk); 780 781 mutex_lock(&net->smc.mutex_fback_rsn); 782 if (smc->listen_smc) { 783 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 784 net->smc.fback_rsn->srv_fback_cnt++; 785 } else { 786 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 787 net->smc.fback_rsn->clnt_fback_cnt++; 788 } 789 mutex_unlock(&net->smc.mutex_fback_rsn); 790 } 791 792 /* must be called under rcu read lock */ 793 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 794 { 795 struct socket_wq *wq; 796 __poll_t flags; 797 798 wq = rcu_dereference(smc->sk.sk_wq); 799 if (!skwq_has_sleeper(wq)) 800 return; 801 802 /* wake up smc sk->sk_wq */ 803 if (!key) { 804 /* sk_state_change */ 805 wake_up_interruptible_all(&wq->wait); 806 } else { 807 flags = key_to_poll(key); 808 if (flags & (EPOLLIN | EPOLLOUT)) 809 /* sk_data_ready or sk_write_space */ 810 wake_up_interruptible_sync_poll(&wq->wait, flags); 811 else if (flags & EPOLLERR) 812 /* sk_error_report */ 813 wake_up_interruptible_poll(&wq->wait, flags); 814 } 815 } 816 817 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 818 unsigned int mode, int sync, void *key) 819 { 820 struct smc_mark_woken *mark = 821 container_of(wait, struct smc_mark_woken, wait_entry); 822 823 mark->woken = true; 824 mark->key = key; 825 return 0; 826 } 827 828 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 829 void (*clcsock_callback)(struct sock *sk)) 830 { 831 struct smc_mark_woken mark = { .woken = false }; 832 struct socket_wq *wq; 833 834 init_waitqueue_func_entry(&mark.wait_entry, 835 smc_fback_mark_woken); 836 rcu_read_lock(); 837 wq = rcu_dereference(clcsk->sk_wq); 838 if (!wq) 839 goto out; 840 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 841 clcsock_callback(clcsk); 842 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 843 844 if (mark.woken) 845 smc_fback_wakeup_waitqueue(smc, mark.key); 846 out: 847 rcu_read_unlock(); 848 } 849 850 static void smc_fback_state_change(struct sock *clcsk) 851 { 852 struct smc_sock *smc; 853 854 read_lock_bh(&clcsk->sk_callback_lock); 855 smc = smc_clcsock_user_data(clcsk); 856 if (smc) 857 smc_fback_forward_wakeup(smc, clcsk, 858 smc->clcsk_state_change); 859 read_unlock_bh(&clcsk->sk_callback_lock); 860 } 861 862 static void smc_fback_data_ready(struct sock *clcsk) 863 { 864 struct smc_sock *smc; 865 866 read_lock_bh(&clcsk->sk_callback_lock); 867 smc = smc_clcsock_user_data(clcsk); 868 if (smc) 869 smc_fback_forward_wakeup(smc, clcsk, 870 smc->clcsk_data_ready); 871 read_unlock_bh(&clcsk->sk_callback_lock); 872 } 873 874 static void smc_fback_write_space(struct sock *clcsk) 875 { 876 struct smc_sock *smc; 877 878 read_lock_bh(&clcsk->sk_callback_lock); 879 smc = smc_clcsock_user_data(clcsk); 880 if (smc) 881 smc_fback_forward_wakeup(smc, clcsk, 882 smc->clcsk_write_space); 883 read_unlock_bh(&clcsk->sk_callback_lock); 884 } 885 886 static void smc_fback_error_report(struct sock *clcsk) 887 { 888 struct smc_sock *smc; 889 890 read_lock_bh(&clcsk->sk_callback_lock); 891 smc = smc_clcsock_user_data(clcsk); 892 if (smc) 893 smc_fback_forward_wakeup(smc, clcsk, 894 smc->clcsk_error_report); 895 read_unlock_bh(&clcsk->sk_callback_lock); 896 } 897 898 static void smc_fback_replace_callbacks(struct smc_sock *smc) 899 { 900 struct sock *clcsk = smc->clcsock->sk; 901 902 write_lock_bh(&clcsk->sk_callback_lock); 903 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 904 905 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 906 &smc->clcsk_state_change); 907 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 908 &smc->clcsk_data_ready); 909 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 910 &smc->clcsk_write_space); 911 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 912 &smc->clcsk_error_report); 913 914 write_unlock_bh(&clcsk->sk_callback_lock); 915 } 916 917 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 918 { 919 int rc = 0; 920 921 mutex_lock(&smc->clcsock_release_lock); 922 if (!smc->clcsock) { 923 rc = -EBADF; 924 goto out; 925 } 926 927 smc->use_fallback = true; 928 smc->fallback_rsn = reason_code; 929 smc_stat_fallback(smc); 930 trace_smc_switch_to_fallback(smc, reason_code); 931 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 932 smc->clcsock->file = smc->sk.sk_socket->file; 933 smc->clcsock->file->private_data = smc->clcsock; 934 smc->clcsock->wq.fasync_list = 935 smc->sk.sk_socket->wq.fasync_list; 936 smc->sk.sk_socket->wq.fasync_list = NULL; 937 938 /* There might be some wait entries remaining 939 * in smc sk->sk_wq and they should be woken up 940 * as clcsock's wait queue is woken up. 941 */ 942 smc_fback_replace_callbacks(smc); 943 } 944 out: 945 mutex_unlock(&smc->clcsock_release_lock); 946 return rc; 947 } 948 949 /* fall back during connect */ 950 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 951 { 952 struct net *net = sock_net(&smc->sk); 953 int rc = 0; 954 955 rc = smc_switch_to_fallback(smc, reason_code); 956 if (rc) { /* fallback fails */ 957 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 958 if (smc->sk.sk_state == SMC_INIT) 959 sock_put(&smc->sk); /* passive closing */ 960 return rc; 961 } 962 smc_copy_sock_settings_to_clc(smc); 963 smc->connect_nonblock = 0; 964 if (smc->sk.sk_state == SMC_INIT) 965 smc->sk.sk_state = SMC_ACTIVE; 966 return 0; 967 } 968 969 /* decline and fall back during connect */ 970 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 971 u8 version) 972 { 973 struct net *net = sock_net(&smc->sk); 974 int rc; 975 976 if (reason_code < 0) { /* error, fallback is not possible */ 977 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 978 if (smc->sk.sk_state == SMC_INIT) 979 sock_put(&smc->sk); /* passive closing */ 980 return reason_code; 981 } 982 if (reason_code != SMC_CLC_DECL_PEERDECL) { 983 rc = smc_clc_send_decline(smc, reason_code, version); 984 if (rc < 0) { 985 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 986 if (smc->sk.sk_state == SMC_INIT) 987 sock_put(&smc->sk); /* passive closing */ 988 return rc; 989 } 990 } 991 return smc_connect_fallback(smc, reason_code); 992 } 993 994 static void smc_conn_abort(struct smc_sock *smc, int local_first) 995 { 996 struct smc_connection *conn = &smc->conn; 997 struct smc_link_group *lgr = conn->lgr; 998 bool lgr_valid = false; 999 1000 if (smc_conn_lgr_valid(conn)) 1001 lgr_valid = true; 1002 1003 smc_conn_free(conn); 1004 if (local_first && lgr_valid) 1005 smc_lgr_cleanup_early(lgr); 1006 } 1007 1008 /* check if there is a rdma device available for this connection. */ 1009 /* called for connect and listen */ 1010 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1011 { 1012 /* PNET table look up: search active ib_device and port 1013 * within same PNETID that also contains the ethernet device 1014 * used for the internal TCP socket 1015 */ 1016 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1017 if (!ini->check_smcrv2 && !ini->ib_dev) 1018 return SMC_CLC_DECL_NOSMCRDEV; 1019 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1020 return SMC_CLC_DECL_NOSMCRDEV; 1021 return 0; 1022 } 1023 1024 /* check if there is an ISM device available for this connection. */ 1025 /* called for connect and listen */ 1026 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1027 { 1028 /* Find ISM device with same PNETID as connecting interface */ 1029 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1030 if (!ini->ism_dev[0]) 1031 return SMC_CLC_DECL_NOSMCDDEV; 1032 else 1033 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1034 return 0; 1035 } 1036 1037 /* is chid unique for the ism devices that are already determined? */ 1038 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1039 int cnt) 1040 { 1041 int i = (!ini->ism_dev[0]) ? 1 : 0; 1042 1043 for (; i < cnt; i++) 1044 if (ini->ism_chid[i] == chid) 1045 return false; 1046 return true; 1047 } 1048 1049 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1050 * PNETID matching net_device) 1051 */ 1052 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1053 struct smc_init_info *ini) 1054 { 1055 int rc = SMC_CLC_DECL_NOSMCDDEV; 1056 struct smcd_dev *smcd; 1057 int i = 1, entry = 1; 1058 bool is_emulated; 1059 u16 chid; 1060 1061 if (smcd_indicated(ini->smc_type_v1)) 1062 rc = 0; /* already initialized for V1 */ 1063 mutex_lock(&smcd_dev_list.mutex); 1064 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1065 if (smcd->going_away || smcd == ini->ism_dev[0]) 1066 continue; 1067 chid = smc_ism_get_chid(smcd); 1068 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1069 continue; 1070 is_emulated = __smc_ism_is_emulated(chid); 1071 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1072 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1073 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1074 /* It's the last GID-CHID entry left in CLC 1075 * Proposal SMC-Dv2 extension, but an Emulated- 1076 * ISM device will take two entries. So give 1077 * up it and try the next potential ISM device. 1078 */ 1079 continue; 1080 ini->ism_dev[i] = smcd; 1081 ini->ism_chid[i] = chid; 1082 ini->is_smcd = true; 1083 rc = 0; 1084 i++; 1085 entry = is_emulated ? entry + 2 : entry + 1; 1086 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1087 break; 1088 } 1089 } 1090 mutex_unlock(&smcd_dev_list.mutex); 1091 ini->ism_offered_cnt = i - 1; 1092 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1093 ini->smcd_version = 0; 1094 1095 return rc; 1096 } 1097 1098 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1099 static int smc_connect_ism_vlan_setup(struct smc_init_info *ini) 1100 { 1101 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1102 return SMC_CLC_DECL_ISMVLANERR; 1103 return 0; 1104 } 1105 1106 static int smc_find_proposal_devices(struct smc_sock *smc, 1107 struct smc_init_info *ini) 1108 { 1109 int rc = 0; 1110 1111 /* check if there is an ism device available */ 1112 if (!(ini->smcd_version & SMC_V1) || 1113 smc_find_ism_device(smc, ini) || 1114 smc_connect_ism_vlan_setup(ini)) 1115 ini->smcd_version &= ~SMC_V1; 1116 /* else ISM V1 is supported for this connection */ 1117 1118 /* check if there is an rdma device available */ 1119 if (!(ini->smcr_version & SMC_V1) || 1120 smc_find_rdma_device(smc, ini)) 1121 ini->smcr_version &= ~SMC_V1; 1122 /* else RDMA is supported for this connection */ 1123 1124 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1125 ini->smcr_version & SMC_V1); 1126 1127 /* check if there is an ism v2 device available */ 1128 if (!(ini->smcd_version & SMC_V2) || 1129 !smc_ism_is_v2_capable() || 1130 smc_find_ism_v2_device_clnt(smc, ini)) 1131 ini->smcd_version &= ~SMC_V2; 1132 1133 /* check if there is an rdma v2 device available */ 1134 ini->check_smcrv2 = true; 1135 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1136 if (!(ini->smcr_version & SMC_V2) || 1137 #if IS_ENABLED(CONFIG_IPV6) 1138 (smc->clcsock->sk->sk_family == AF_INET6 && 1139 !ipv6_addr_v4mapped(&smc->clcsock->sk->sk_v6_rcv_saddr)) || 1140 #endif 1141 !smc_clc_ueid_count() || 1142 smc_find_rdma_device(smc, ini)) 1143 ini->smcr_version &= ~SMC_V2; 1144 ini->check_smcrv2 = false; 1145 1146 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1147 ini->smcr_version & SMC_V2); 1148 1149 /* if neither ISM nor RDMA are supported, fallback */ 1150 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1151 rc = SMC_CLC_DECL_NOSMCDEV; 1152 1153 return rc; 1154 } 1155 1156 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1157 * used, the VLAN ID will be registered again during the connection setup. 1158 */ 1159 static int smc_connect_ism_vlan_cleanup(struct smc_init_info *ini) 1160 { 1161 if (!smcd_indicated(ini->smc_type_v1)) 1162 return 0; 1163 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1164 return SMC_CLC_DECL_CNFERR; 1165 return 0; 1166 } 1167 1168 #define SMC_CLC_MAX_ACCEPT_LEN \ 1169 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1170 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1171 sizeof(struct smc_clc_msg_trail)) 1172 1173 /* CLC handshake during connect */ 1174 static int smc_connect_clc(struct smc_sock *smc, 1175 struct smc_clc_msg_accept_confirm *aclc, 1176 struct smc_init_info *ini) 1177 { 1178 int rc = 0; 1179 1180 /* do inband token exchange */ 1181 rc = smc_clc_send_proposal(smc, ini); 1182 if (rc) 1183 return rc; 1184 /* receive SMC Accept CLC message */ 1185 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1186 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1187 } 1188 1189 void smc_fill_gid_list(struct smc_link_group *lgr, 1190 struct smc_gidlist *gidlist, 1191 struct smc_ib_device *known_dev, u8 *known_gid) 1192 { 1193 struct smc_init_info *alt_ini = NULL; 1194 1195 memset(gidlist, 0, sizeof(*gidlist)); 1196 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1197 1198 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1199 if (!alt_ini) 1200 goto out; 1201 1202 alt_ini->vlan_id = lgr->vlan_id; 1203 alt_ini->check_smcrv2 = true; 1204 alt_ini->smcrv2.saddr = lgr->saddr; 1205 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1206 1207 if (!alt_ini->smcrv2.ib_dev_v2) 1208 goto out; 1209 1210 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1211 SMC_GID_SIZE); 1212 1213 out: 1214 kfree(alt_ini); 1215 } 1216 1217 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1218 struct smc_clc_msg_accept_confirm *aclc, 1219 struct smc_init_info *ini) 1220 { 1221 struct smc_clc_first_contact_ext *fce = 1222 smc_get_clc_first_contact_ext(aclc, false); 1223 struct net *net = sock_net(&smc->sk); 1224 int rc; 1225 1226 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1227 return 0; 1228 1229 if (fce->v2_direct) { 1230 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1231 ini->smcrv2.uses_gateway = false; 1232 } else { 1233 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1234 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1235 ini->smcrv2.nexthop_mac, 1236 &ini->smcrv2.uses_gateway)) 1237 return SMC_CLC_DECL_NOROUTE; 1238 if (!ini->smcrv2.uses_gateway) { 1239 /* mismatch: peer claims indirect, but its direct */ 1240 return SMC_CLC_DECL_NOINDIRECT; 1241 } 1242 } 1243 1244 ini->release_nr = fce->release; 1245 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1246 if (rc) 1247 return rc; 1248 1249 return 0; 1250 } 1251 1252 /* setup for RDMA connection of client */ 1253 static int smc_connect_rdma(struct smc_sock *smc, 1254 struct smc_clc_msg_accept_confirm *aclc, 1255 struct smc_init_info *ini) 1256 { 1257 int i, reason_code = 0; 1258 struct smc_link *link; 1259 u8 *eid = NULL; 1260 1261 ini->is_smcd = false; 1262 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1263 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1264 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1265 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1266 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1267 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1268 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1269 1270 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1271 if (reason_code) 1272 return reason_code; 1273 1274 mutex_lock(&smc_client_lgr_pending); 1275 reason_code = smc_conn_create(smc, ini); 1276 if (reason_code) { 1277 mutex_unlock(&smc_client_lgr_pending); 1278 return reason_code; 1279 } 1280 1281 smc_conn_save_peer_info(smc, aclc); 1282 1283 if (ini->first_contact_local) { 1284 link = smc->conn.lnk; 1285 } else { 1286 /* set link that was assigned by server */ 1287 link = NULL; 1288 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1289 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1290 1291 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1292 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1293 SMC_GID_SIZE) && 1294 (aclc->hdr.version > SMC_V1 || 1295 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1296 sizeof(l->peer_mac)))) { 1297 link = l; 1298 break; 1299 } 1300 } 1301 if (!link) { 1302 reason_code = SMC_CLC_DECL_NOSRVLINK; 1303 goto connect_abort; 1304 } 1305 smc_switch_link_and_count(&smc->conn, link); 1306 } 1307 1308 /* create send buffer and rmb */ 1309 if (smc_buf_create(smc, false)) { 1310 reason_code = SMC_CLC_DECL_MEM; 1311 goto connect_abort; 1312 } 1313 1314 if (ini->first_contact_local) 1315 smc_link_save_peer_info(link, aclc, ini); 1316 1317 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1318 reason_code = SMC_CLC_DECL_ERR_RTOK; 1319 goto connect_abort; 1320 } 1321 1322 smc_rx_init(smc); 1323 1324 if (ini->first_contact_local) { 1325 if (smc_ib_ready_link(link)) { 1326 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1327 goto connect_abort; 1328 } 1329 } else { 1330 /* reg sendbufs if they were vzalloced */ 1331 if (smc->conn.sndbuf_desc->is_vm) { 1332 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1333 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1334 goto connect_abort; 1335 } 1336 } 1337 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1338 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1339 goto connect_abort; 1340 } 1341 } 1342 1343 if (aclc->hdr.version > SMC_V1) { 1344 eid = aclc->r1.eid; 1345 if (ini->first_contact_local) 1346 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1347 link->smcibdev, link->gid); 1348 } 1349 1350 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1351 aclc->hdr.version, eid, ini); 1352 if (reason_code) 1353 goto connect_abort; 1354 1355 smc_tx_init(smc); 1356 1357 if (ini->first_contact_local) { 1358 /* QP confirmation over RoCE fabric */ 1359 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1360 reason_code = smcr_clnt_conf_first_link(smc); 1361 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1362 if (reason_code) 1363 goto connect_abort; 1364 } 1365 mutex_unlock(&smc_client_lgr_pending); 1366 1367 smc_copy_sock_settings_to_clc(smc); 1368 smc->connect_nonblock = 0; 1369 if (smc->sk.sk_state == SMC_INIT) 1370 smc->sk.sk_state = SMC_ACTIVE; 1371 1372 return 0; 1373 connect_abort: 1374 smc_conn_abort(smc, ini->first_contact_local); 1375 mutex_unlock(&smc_client_lgr_pending); 1376 smc->connect_nonblock = 0; 1377 1378 return reason_code; 1379 } 1380 1381 /* The server has chosen one of the proposed ISM devices for the communication. 1382 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1383 */ 1384 static int 1385 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1386 struct smc_init_info *ini) 1387 { 1388 int i; 1389 1390 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1391 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1392 ini->ism_selected = i; 1393 return 0; 1394 } 1395 } 1396 1397 return -EPROTO; 1398 } 1399 1400 /* setup for ISM connection of client */ 1401 static int smc_connect_ism(struct smc_sock *smc, 1402 struct smc_clc_msg_accept_confirm *aclc, 1403 struct smc_init_info *ini) 1404 { 1405 u8 *eid = NULL; 1406 int rc = 0; 1407 1408 ini->is_smcd = true; 1409 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1410 1411 if (aclc->hdr.version == SMC_V2) { 1412 if (ini->first_contact_peer) { 1413 struct smc_clc_first_contact_ext *fce = 1414 smc_get_clc_first_contact_ext(aclc, true); 1415 1416 ini->release_nr = fce->release; 1417 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1418 if (rc) 1419 return rc; 1420 } 1421 1422 rc = smc_v2_determine_accepted_chid(aclc, ini); 1423 if (rc) 1424 return rc; 1425 1426 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1427 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1428 ntohll(aclc->d1.gid_ext); 1429 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1430 } 1431 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1432 1433 /* there is only one lgr role for SMC-D; use server lock */ 1434 mutex_lock(&smc_server_lgr_pending); 1435 rc = smc_conn_create(smc, ini); 1436 if (rc) { 1437 mutex_unlock(&smc_server_lgr_pending); 1438 return rc; 1439 } 1440 1441 /* Create send and receive buffers */ 1442 rc = smc_buf_create(smc, true); 1443 if (rc) { 1444 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1445 goto connect_abort; 1446 } 1447 1448 smc_conn_save_peer_info(smc, aclc); 1449 1450 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) { 1451 rc = smcd_buf_attach(smc); 1452 if (rc) { 1453 rc = SMC_CLC_DECL_MEM; /* try to fallback */ 1454 goto connect_abort; 1455 } 1456 } 1457 smc_rx_init(smc); 1458 smc_tx_init(smc); 1459 1460 if (aclc->hdr.version > SMC_V1) 1461 eid = aclc->d1.eid; 1462 1463 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1464 aclc->hdr.version, eid, ini); 1465 if (rc) 1466 goto connect_abort; 1467 mutex_unlock(&smc_server_lgr_pending); 1468 1469 smc_copy_sock_settings_to_clc(smc); 1470 smc->connect_nonblock = 0; 1471 if (smc->sk.sk_state == SMC_INIT) 1472 smc->sk.sk_state = SMC_ACTIVE; 1473 1474 return 0; 1475 connect_abort: 1476 smc_conn_abort(smc, ini->first_contact_local); 1477 mutex_unlock(&smc_server_lgr_pending); 1478 smc->connect_nonblock = 0; 1479 1480 return rc; 1481 } 1482 1483 /* check if received accept type and version matches a proposed one */ 1484 static int smc_connect_check_aclc(struct smc_init_info *ini, 1485 struct smc_clc_msg_accept_confirm *aclc) 1486 { 1487 if (aclc->hdr.version >= SMC_V2) { 1488 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1489 !smcr_indicated(ini->smc_type_v2)) || 1490 (aclc->hdr.typev1 == SMC_TYPE_D && 1491 !smcd_indicated(ini->smc_type_v2))) 1492 return SMC_CLC_DECL_MODEUNSUPP; 1493 } else { 1494 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1495 !smcr_indicated(ini->smc_type_v1)) || 1496 (aclc->hdr.typev1 == SMC_TYPE_D && 1497 !smcd_indicated(ini->smc_type_v1))) 1498 return SMC_CLC_DECL_MODEUNSUPP; 1499 } 1500 1501 return 0; 1502 } 1503 1504 /* perform steps before actually connecting */ 1505 static int __smc_connect(struct smc_sock *smc) 1506 { 1507 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1508 struct smc_clc_msg_accept_confirm *aclc; 1509 struct smc_init_info *ini = NULL; 1510 u8 *buf = NULL; 1511 int rc = 0; 1512 1513 if (smc->use_fallback) 1514 return smc_connect_fallback(smc, smc->fallback_rsn); 1515 1516 /* if peer has not signalled SMC-capability, fall back */ 1517 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1518 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1519 1520 /* IPSec connections opt out of SMC optimizations */ 1521 if (using_ipsec(smc)) 1522 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1523 version); 1524 1525 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1526 if (!ini) 1527 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1528 version); 1529 1530 ini->smcd_version = SMC_V1 | SMC_V2; 1531 ini->smcr_version = SMC_V1 | SMC_V2; 1532 ini->smc_type_v1 = SMC_TYPE_B; 1533 ini->smc_type_v2 = SMC_TYPE_B; 1534 1535 /* get vlan id from IP device */ 1536 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1537 ini->smcd_version &= ~SMC_V1; 1538 ini->smcr_version = 0; 1539 ini->smc_type_v1 = SMC_TYPE_N; 1540 } 1541 1542 rc = smc_find_proposal_devices(smc, ini); 1543 if (rc) 1544 goto fallback; 1545 1546 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1547 if (!buf) { 1548 rc = SMC_CLC_DECL_MEM; 1549 goto fallback; 1550 } 1551 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1552 1553 /* perform CLC handshake */ 1554 rc = smc_connect_clc(smc, aclc, ini); 1555 if (rc) { 1556 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1557 if (rc == -EAGAIN) { 1558 rc = -ETIMEDOUT; 1559 smc->sk.sk_err = ETIMEDOUT; 1560 } 1561 goto vlan_cleanup; 1562 } 1563 1564 /* check if smc modes and versions of CLC proposal and accept match */ 1565 rc = smc_connect_check_aclc(ini, aclc); 1566 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1567 if (rc) 1568 goto vlan_cleanup; 1569 1570 /* depending on previous steps, connect using rdma or ism */ 1571 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1572 ini->smcr_version = version; 1573 rc = smc_connect_rdma(smc, aclc, ini); 1574 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1575 ini->smcd_version = version; 1576 rc = smc_connect_ism(smc, aclc, ini); 1577 } 1578 if (rc) 1579 goto vlan_cleanup; 1580 1581 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1582 smc_connect_ism_vlan_cleanup(ini); 1583 kfree(buf); 1584 kfree(ini); 1585 return 0; 1586 1587 vlan_cleanup: 1588 smc_connect_ism_vlan_cleanup(ini); 1589 kfree(buf); 1590 fallback: 1591 kfree(ini); 1592 return smc_connect_decline_fallback(smc, rc, version); 1593 } 1594 1595 static void smc_connect_work(struct work_struct *work) 1596 { 1597 struct smc_sock *smc = container_of(work, struct smc_sock, 1598 connect_work); 1599 long timeo = READ_ONCE(smc->sk.sk_sndtimeo); 1600 int rc = 0; 1601 1602 if (!timeo) 1603 timeo = MAX_SCHEDULE_TIMEOUT; 1604 lock_sock(smc->clcsock->sk); 1605 if (smc->clcsock->sk->sk_err) { 1606 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1607 } else if ((1 << smc->clcsock->sk->sk_state) & 1608 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1609 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1610 if ((rc == -EPIPE) && 1611 ((1 << smc->clcsock->sk->sk_state) & 1612 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1613 rc = 0; 1614 } 1615 release_sock(smc->clcsock->sk); 1616 lock_sock(&smc->sk); 1617 if (rc != 0 || smc->sk.sk_err) { 1618 smc->sk.sk_state = SMC_CLOSED; 1619 if (rc == -EPIPE || rc == -EAGAIN) 1620 smc->sk.sk_err = EPIPE; 1621 else if (rc == -ECONNREFUSED) 1622 smc->sk.sk_err = ECONNREFUSED; 1623 else if (signal_pending(current)) 1624 smc->sk.sk_err = -sock_intr_errno(timeo); 1625 sock_put(&smc->sk); /* passive closing */ 1626 goto out; 1627 } 1628 1629 rc = __smc_connect(smc); 1630 if (rc < 0) 1631 smc->sk.sk_err = -rc; 1632 1633 out: 1634 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1635 if (smc->sk.sk_err) { 1636 smc->sk.sk_state_change(&smc->sk); 1637 } else { /* allow polling before and after fallback decision */ 1638 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1639 smc->sk.sk_write_space(&smc->sk); 1640 } 1641 } 1642 release_sock(&smc->sk); 1643 } 1644 1645 int smc_connect(struct socket *sock, struct sockaddr *addr, 1646 int alen, int flags) 1647 { 1648 struct sock *sk = sock->sk; 1649 struct smc_sock *smc; 1650 int rc = -EINVAL; 1651 1652 smc = smc_sk(sk); 1653 1654 /* separate smc parameter checking to be safe */ 1655 if (alen < sizeof(addr->sa_family)) 1656 goto out_err; 1657 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1658 goto out_err; 1659 1660 lock_sock(sk); 1661 switch (sock->state) { 1662 default: 1663 rc = -EINVAL; 1664 goto out; 1665 case SS_CONNECTED: 1666 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1667 goto out; 1668 case SS_CONNECTING: 1669 if (sk->sk_state == SMC_ACTIVE) 1670 goto connected; 1671 break; 1672 case SS_UNCONNECTED: 1673 sock->state = SS_CONNECTING; 1674 break; 1675 } 1676 1677 switch (sk->sk_state) { 1678 default: 1679 goto out; 1680 case SMC_CLOSED: 1681 rc = sock_error(sk) ? : -ECONNABORTED; 1682 sock->state = SS_UNCONNECTED; 1683 goto out; 1684 case SMC_ACTIVE: 1685 rc = -EISCONN; 1686 goto out; 1687 case SMC_INIT: 1688 break; 1689 } 1690 1691 smc_copy_sock_settings_to_clc(smc); 1692 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1693 if (smc->connect_nonblock) { 1694 rc = -EALREADY; 1695 goto out; 1696 } 1697 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1698 if (rc && rc != -EINPROGRESS) 1699 goto out; 1700 1701 if (smc->use_fallback) { 1702 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1703 goto out; 1704 } 1705 sock_hold(&smc->sk); /* sock put in passive closing */ 1706 if (flags & O_NONBLOCK) { 1707 if (queue_work(smc_hs_wq, &smc->connect_work)) 1708 smc->connect_nonblock = 1; 1709 rc = -EINPROGRESS; 1710 goto out; 1711 } else { 1712 rc = __smc_connect(smc); 1713 if (rc < 0) 1714 goto out; 1715 } 1716 1717 connected: 1718 rc = 0; 1719 sock->state = SS_CONNECTED; 1720 out: 1721 release_sock(sk); 1722 out_err: 1723 return rc; 1724 } 1725 1726 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1727 { 1728 struct socket *new_clcsock = NULL; 1729 struct sock *lsk = &lsmc->sk; 1730 struct sock *new_sk; 1731 int rc = -EINVAL; 1732 1733 release_sock(lsk); 1734 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1735 if (!new_sk) { 1736 rc = -ENOMEM; 1737 lsk->sk_err = ENOMEM; 1738 *new_smc = NULL; 1739 lock_sock(lsk); 1740 goto out; 1741 } 1742 *new_smc = smc_sk(new_sk); 1743 1744 mutex_lock(&lsmc->clcsock_release_lock); 1745 if (lsmc->clcsock) 1746 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1747 mutex_unlock(&lsmc->clcsock_release_lock); 1748 lock_sock(lsk); 1749 if (rc < 0 && rc != -EAGAIN) 1750 lsk->sk_err = -rc; 1751 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1752 new_sk->sk_prot->unhash(new_sk); 1753 if (new_clcsock) 1754 sock_release(new_clcsock); 1755 new_sk->sk_state = SMC_CLOSED; 1756 smc_sock_set_flag(new_sk, SOCK_DEAD); 1757 sock_put(new_sk); /* final */ 1758 *new_smc = NULL; 1759 goto out; 1760 } 1761 1762 /* new clcsock has inherited the smc listen-specific sk_data_ready 1763 * function; switch it back to the original sk_data_ready function 1764 */ 1765 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1766 1767 /* if new clcsock has also inherited the fallback-specific callback 1768 * functions, switch them back to the original ones. 1769 */ 1770 if (lsmc->use_fallback) { 1771 if (lsmc->clcsk_state_change) 1772 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1773 if (lsmc->clcsk_write_space) 1774 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1775 if (lsmc->clcsk_error_report) 1776 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1777 } 1778 1779 (*new_smc)->clcsock = new_clcsock; 1780 out: 1781 return rc; 1782 } 1783 1784 /* add a just created sock to the accept queue of the listen sock as 1785 * candidate for a following socket accept call from user space 1786 */ 1787 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1788 { 1789 struct smc_sock *par = smc_sk(parent); 1790 1791 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1792 spin_lock(&par->accept_q_lock); 1793 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1794 spin_unlock(&par->accept_q_lock); 1795 sk_acceptq_added(parent); 1796 } 1797 1798 /* remove a socket from the accept queue of its parental listening socket */ 1799 static void smc_accept_unlink(struct sock *sk) 1800 { 1801 struct smc_sock *par = smc_sk(sk)->listen_smc; 1802 1803 spin_lock(&par->accept_q_lock); 1804 list_del_init(&smc_sk(sk)->accept_q); 1805 spin_unlock(&par->accept_q_lock); 1806 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1807 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1808 } 1809 1810 /* remove a sock from the accept queue to bind it to a new socket created 1811 * for a socket accept call from user space 1812 */ 1813 struct sock *smc_accept_dequeue(struct sock *parent, 1814 struct socket *new_sock) 1815 { 1816 struct smc_sock *isk, *n; 1817 struct sock *new_sk; 1818 1819 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1820 new_sk = (struct sock *)isk; 1821 1822 smc_accept_unlink(new_sk); 1823 if (new_sk->sk_state == SMC_CLOSED) { 1824 new_sk->sk_prot->unhash(new_sk); 1825 if (isk->clcsock) { 1826 sock_release(isk->clcsock); 1827 isk->clcsock = NULL; 1828 } 1829 sock_put(new_sk); /* final */ 1830 continue; 1831 } 1832 if (new_sock) { 1833 sock_graft(new_sk, new_sock); 1834 new_sock->state = SS_CONNECTED; 1835 if (isk->use_fallback) { 1836 smc_sk(new_sk)->clcsock->file = new_sock->file; 1837 isk->clcsock->file->private_data = isk->clcsock; 1838 } 1839 } 1840 return new_sk; 1841 } 1842 return NULL; 1843 } 1844 1845 /* clean up for a created but never accepted sock */ 1846 void smc_close_non_accepted(struct sock *sk) 1847 { 1848 struct smc_sock *smc = smc_sk(sk); 1849 1850 sock_hold(sk); /* sock_put below */ 1851 lock_sock(sk); 1852 if (!sk->sk_lingertime) 1853 /* wait for peer closing */ 1854 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1855 __smc_release(smc); 1856 release_sock(sk); 1857 sock_put(sk); /* sock_hold above */ 1858 sock_put(sk); /* final sock_put */ 1859 } 1860 1861 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1862 { 1863 struct smc_link *link = smc->conn.lnk; 1864 struct smc_llc_qentry *qentry; 1865 int rc; 1866 1867 /* reg the sndbuf if it was vzalloced*/ 1868 if (smc->conn.sndbuf_desc->is_vm) { 1869 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1870 return SMC_CLC_DECL_ERR_REGBUF; 1871 } 1872 1873 /* reg the rmb */ 1874 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1875 return SMC_CLC_DECL_ERR_REGBUF; 1876 1877 /* send CONFIRM LINK request to client over the RoCE fabric */ 1878 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1879 if (rc < 0) 1880 return SMC_CLC_DECL_TIMEOUT_CL; 1881 1882 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1883 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1884 SMC_LLC_CONFIRM_LINK); 1885 if (!qentry) { 1886 struct smc_clc_msg_decline dclc; 1887 1888 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1889 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1890 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1891 } 1892 smc_llc_save_peer_uid(qentry); 1893 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1894 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1895 if (rc) 1896 return SMC_CLC_DECL_RMBE_EC; 1897 1898 /* confirm_rkey is implicit on 1st contact */ 1899 smc->conn.rmb_desc->is_conf_rkey = true; 1900 1901 smc_llc_link_active(link); 1902 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1903 1904 if (link->lgr->max_links > 1) { 1905 down_write(&link->lgr->llc_conf_mutex); 1906 /* initial contact - try to establish second link */ 1907 smc_llc_srv_add_link(link, NULL); 1908 up_write(&link->lgr->llc_conf_mutex); 1909 } 1910 return 0; 1911 } 1912 1913 /* listen worker: finish */ 1914 static void smc_listen_out(struct smc_sock *new_smc) 1915 { 1916 struct smc_sock *lsmc = new_smc->listen_smc; 1917 struct sock *newsmcsk = &new_smc->sk; 1918 1919 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1920 atomic_dec(&lsmc->queued_smc_hs); 1921 1922 release_sock(newsmcsk); /* lock in smc_listen_work() */ 1923 if (lsmc->sk.sk_state == SMC_LISTEN) { 1924 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1925 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1926 release_sock(&lsmc->sk); 1927 } else { /* no longer listening */ 1928 smc_close_non_accepted(newsmcsk); 1929 } 1930 1931 /* Wake up accept */ 1932 lsmc->sk.sk_data_ready(&lsmc->sk); 1933 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1934 } 1935 1936 /* listen worker: finish in state connected */ 1937 static void smc_listen_out_connected(struct smc_sock *new_smc) 1938 { 1939 struct sock *newsmcsk = &new_smc->sk; 1940 1941 if (newsmcsk->sk_state == SMC_INIT) 1942 newsmcsk->sk_state = SMC_ACTIVE; 1943 1944 smc_listen_out(new_smc); 1945 } 1946 1947 /* listen worker: finish in error state */ 1948 static void smc_listen_out_err(struct smc_sock *new_smc) 1949 { 1950 struct sock *newsmcsk = &new_smc->sk; 1951 struct net *net = sock_net(newsmcsk); 1952 1953 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1954 if (newsmcsk->sk_state == SMC_INIT) 1955 sock_put(&new_smc->sk); /* passive closing */ 1956 newsmcsk->sk_state = SMC_CLOSED; 1957 1958 smc_listen_out(new_smc); 1959 } 1960 1961 /* listen worker: decline and fall back if possible */ 1962 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1963 int local_first, u8 version) 1964 { 1965 /* RDMA setup failed, switch back to TCP */ 1966 smc_conn_abort(new_smc, local_first); 1967 if (reason_code < 0 || 1968 smc_switch_to_fallback(new_smc, reason_code)) { 1969 /* error, no fallback possible */ 1970 smc_listen_out_err(new_smc); 1971 return; 1972 } 1973 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1974 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1975 smc_listen_out_err(new_smc); 1976 return; 1977 } 1978 } 1979 smc_listen_out_connected(new_smc); 1980 } 1981 1982 /* listen worker: version checking */ 1983 static int smc_listen_v2_check(struct smc_sock *new_smc, 1984 struct smc_clc_msg_proposal *pclc, 1985 struct smc_init_info *ini) 1986 { 1987 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1988 struct smc_clc_v2_extension *pclc_v2_ext; 1989 int rc = SMC_CLC_DECL_PEERNOSMC; 1990 1991 ini->smc_type_v1 = pclc->hdr.typev1; 1992 ini->smc_type_v2 = pclc->hdr.typev2; 1993 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1994 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1995 if (pclc->hdr.version > SMC_V1) { 1996 if (smcd_indicated(ini->smc_type_v2)) 1997 ini->smcd_version |= SMC_V2; 1998 if (smcr_indicated(ini->smc_type_v2)) 1999 ini->smcr_version |= SMC_V2; 2000 } 2001 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 2002 rc = SMC_CLC_DECL_PEERNOSMC; 2003 goto out; 2004 } 2005 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 2006 if (!pclc_v2_ext) { 2007 ini->smcd_version &= ~SMC_V2; 2008 ini->smcr_version &= ~SMC_V2; 2009 rc = SMC_CLC_DECL_NOV2EXT; 2010 goto out; 2011 } 2012 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2013 if (ini->smcd_version & SMC_V2) { 2014 if (!smc_ism_is_v2_capable()) { 2015 ini->smcd_version &= ~SMC_V2; 2016 rc = SMC_CLC_DECL_NOISM2SUPP; 2017 } else if (!pclc_smcd_v2_ext) { 2018 ini->smcd_version &= ~SMC_V2; 2019 rc = SMC_CLC_DECL_NOV2DEXT; 2020 } else if (!pclc_v2_ext->hdr.eid_cnt && 2021 !pclc_v2_ext->hdr.flag.seid) { 2022 ini->smcd_version &= ~SMC_V2; 2023 rc = SMC_CLC_DECL_NOUEID; 2024 } 2025 } 2026 if (ini->smcr_version & SMC_V2) { 2027 if (!pclc_v2_ext->hdr.eid_cnt) { 2028 ini->smcr_version &= ~SMC_V2; 2029 rc = SMC_CLC_DECL_NOUEID; 2030 } 2031 } 2032 2033 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2034 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2035 ini->release_nr = SMC_RELEASE; 2036 2037 out: 2038 if (!ini->smcd_version && !ini->smcr_version) 2039 return rc; 2040 2041 return 0; 2042 } 2043 2044 /* listen worker: check prefixes */ 2045 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2046 struct smc_clc_msg_proposal *pclc) 2047 { 2048 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2049 struct socket *newclcsock = new_smc->clcsock; 2050 2051 if (pclc->hdr.typev1 == SMC_TYPE_N) 2052 return 0; 2053 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2054 if (!pclc_prfx) 2055 return -EPROTO; 2056 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2057 return SMC_CLC_DECL_DIFFPREFIX; 2058 2059 return 0; 2060 } 2061 2062 /* listen worker: initialize connection and buffers */ 2063 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2064 struct smc_init_info *ini) 2065 { 2066 int rc; 2067 2068 /* allocate connection / link group */ 2069 rc = smc_conn_create(new_smc, ini); 2070 if (rc) 2071 return rc; 2072 2073 /* create send buffer and rmb */ 2074 if (smc_buf_create(new_smc, false)) { 2075 smc_conn_abort(new_smc, ini->first_contact_local); 2076 return SMC_CLC_DECL_MEM; 2077 } 2078 2079 return 0; 2080 } 2081 2082 /* listen worker: initialize connection and buffers for SMC-D */ 2083 static int smc_listen_ism_init(struct smc_sock *new_smc, 2084 struct smc_init_info *ini) 2085 { 2086 int rc; 2087 2088 rc = smc_conn_create(new_smc, ini); 2089 if (rc) 2090 return rc; 2091 2092 /* Create send and receive buffers */ 2093 rc = smc_buf_create(new_smc, true); 2094 if (rc) { 2095 smc_conn_abort(new_smc, ini->first_contact_local); 2096 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2097 SMC_CLC_DECL_MEM; 2098 } 2099 2100 return 0; 2101 } 2102 2103 static bool smc_is_already_selected(struct smcd_dev *smcd, 2104 struct smc_init_info *ini, 2105 int matches) 2106 { 2107 int i; 2108 2109 for (i = 0; i < matches; i++) 2110 if (smcd == ini->ism_dev[i]) 2111 return true; 2112 2113 return false; 2114 } 2115 2116 /* check for ISM devices matching proposed ISM devices */ 2117 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2118 u16 proposed_chid, 2119 struct smcd_gid *proposed_gid, 2120 unsigned int *matches) 2121 { 2122 struct smcd_dev *smcd; 2123 2124 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2125 if (smcd->going_away) 2126 continue; 2127 if (smc_is_already_selected(smcd, ini, *matches)) 2128 continue; 2129 if (smc_ism_get_chid(smcd) == proposed_chid && 2130 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2131 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2132 if (__smc_ism_is_emulated(proposed_chid)) 2133 ini->ism_peer_gid[*matches].gid_ext = 2134 proposed_gid->gid_ext; 2135 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2136 ini->ism_dev[*matches] = smcd; 2137 (*matches)++; 2138 break; 2139 } 2140 } 2141 } 2142 2143 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2144 { 2145 if (!ini->rc) 2146 ini->rc = rc; 2147 } 2148 2149 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2150 struct smc_clc_msg_proposal *pclc, 2151 struct smc_init_info *ini) 2152 { 2153 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2154 struct smc_clc_v2_extension *smc_v2_ext; 2155 struct smc_clc_msg_smcd *pclc_smcd; 2156 unsigned int matches = 0; 2157 struct smcd_gid smcd_gid; 2158 u8 smcd_version; 2159 u8 *eid = NULL; 2160 int i, rc; 2161 u16 chid; 2162 2163 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2164 goto not_found; 2165 2166 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2167 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2168 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2169 if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext) 2170 goto not_found; 2171 2172 mutex_lock(&smcd_dev_list.mutex); 2173 if (pclc_smcd->ism.chid) { 2174 /* check for ISM device matching proposed native ISM device */ 2175 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2176 smcd_gid.gid_ext = 0; 2177 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2178 &smcd_gid, &matches); 2179 } 2180 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2181 /* check for ISM devices matching proposed non-native ISM 2182 * devices 2183 */ 2184 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2185 smcd_gid.gid_ext = 0; 2186 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2187 if (__smc_ism_is_emulated(chid)) { 2188 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2189 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2190 /* each Emulated-ISM device takes two GID-CHID 2191 * entries and CHID of the second entry repeats 2192 * that of the first entry. 2193 * 2194 * So check if the next GID-CHID entry exists 2195 * and both two entries' CHIDs are the same. 2196 */ 2197 continue; 2198 smcd_gid.gid_ext = 2199 ntohll(smcd_v2_ext->gidchid[++i].gid); 2200 } 2201 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2202 } 2203 mutex_unlock(&smcd_dev_list.mutex); 2204 2205 if (!ini->ism_dev[0]) { 2206 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2207 goto not_found; 2208 } 2209 2210 smc_ism_get_system_eid(&eid); 2211 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2212 smcd_v2_ext->system_eid, eid)) 2213 goto not_found; 2214 2215 /* separate - outside the smcd_dev_list.lock */ 2216 smcd_version = ini->smcd_version; 2217 for (i = 0; i < matches; i++) { 2218 ini->smcd_version = SMC_V2; 2219 ini->is_smcd = true; 2220 ini->ism_selected = i; 2221 rc = smc_listen_ism_init(new_smc, ini); 2222 if (rc) { 2223 smc_find_ism_store_rc(rc, ini); 2224 /* try next active ISM device */ 2225 continue; 2226 } 2227 return; /* matching and usable V2 ISM device found */ 2228 } 2229 /* no V2 ISM device could be initialized */ 2230 ini->smcd_version = smcd_version; /* restore original value */ 2231 ini->negotiated_eid[0] = 0; 2232 2233 not_found: 2234 ini->smcd_version &= ~SMC_V2; 2235 ini->ism_dev[0] = NULL; 2236 ini->is_smcd = false; 2237 } 2238 2239 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2240 struct smc_clc_msg_proposal *pclc, 2241 struct smc_init_info *ini) 2242 { 2243 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2244 int rc = 0; 2245 2246 /* check if ISM V1 is available */ 2247 if (!(ini->smcd_version & SMC_V1) || 2248 !smcd_indicated(ini->smc_type_v1) || 2249 !pclc_smcd) 2250 goto not_found; 2251 ini->is_smcd = true; /* prepare ISM check */ 2252 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2253 ini->ism_peer_gid[0].gid_ext = 0; 2254 rc = smc_find_ism_device(new_smc, ini); 2255 if (rc) 2256 goto not_found; 2257 ini->ism_selected = 0; 2258 rc = smc_listen_ism_init(new_smc, ini); 2259 if (!rc) 2260 return; /* V1 ISM device found */ 2261 2262 not_found: 2263 smc_find_ism_store_rc(rc, ini); 2264 ini->smcd_version &= ~SMC_V1; 2265 ini->ism_dev[0] = NULL; 2266 ini->is_smcd = false; 2267 } 2268 2269 /* listen worker: register buffers */ 2270 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2271 { 2272 struct smc_connection *conn = &new_smc->conn; 2273 2274 if (!local_first) { 2275 /* reg sendbufs if they were vzalloced */ 2276 if (conn->sndbuf_desc->is_vm) { 2277 if (smcr_lgr_reg_sndbufs(conn->lnk, 2278 conn->sndbuf_desc)) 2279 return SMC_CLC_DECL_ERR_REGBUF; 2280 } 2281 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2282 return SMC_CLC_DECL_ERR_REGBUF; 2283 } 2284 2285 return 0; 2286 } 2287 2288 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2289 struct smc_clc_msg_proposal *pclc, 2290 struct smc_init_info *ini) 2291 { 2292 struct smc_clc_v2_extension *smc_v2_ext; 2293 u8 smcr_version; 2294 int rc; 2295 2296 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2297 goto not_found; 2298 2299 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2300 if (!smc_v2_ext || 2301 !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2302 goto not_found; 2303 2304 /* prepare RDMA check */ 2305 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2306 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2307 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2308 ini->check_smcrv2 = true; 2309 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2310 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2311 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2312 rc = smc_find_rdma_device(new_smc, ini); 2313 if (rc) { 2314 smc_find_ism_store_rc(rc, ini); 2315 goto not_found; 2316 } 2317 if (!ini->smcrv2.uses_gateway) 2318 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2319 2320 smcr_version = ini->smcr_version; 2321 ini->smcr_version = SMC_V2; 2322 rc = smc_listen_rdma_init(new_smc, ini); 2323 if (!rc) { 2324 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2325 if (rc) 2326 smc_conn_abort(new_smc, ini->first_contact_local); 2327 } 2328 if (!rc) 2329 return; 2330 ini->smcr_version = smcr_version; 2331 smc_find_ism_store_rc(rc, ini); 2332 2333 not_found: 2334 ini->smcr_version &= ~SMC_V2; 2335 ini->smcrv2.ib_dev_v2 = NULL; 2336 ini->check_smcrv2 = false; 2337 } 2338 2339 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2340 struct smc_clc_msg_proposal *pclc, 2341 struct smc_init_info *ini) 2342 { 2343 int rc; 2344 2345 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2346 return SMC_CLC_DECL_NOSMCDEV; 2347 2348 /* prepare RDMA check */ 2349 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2350 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2351 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2352 rc = smc_find_rdma_device(new_smc, ini); 2353 if (rc) { 2354 /* no RDMA device found */ 2355 return SMC_CLC_DECL_NOSMCDEV; 2356 } 2357 rc = smc_listen_rdma_init(new_smc, ini); 2358 if (rc) 2359 return rc; 2360 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2361 } 2362 2363 /* determine the local device matching to proposal */ 2364 static int smc_listen_find_device(struct smc_sock *new_smc, 2365 struct smc_clc_msg_proposal *pclc, 2366 struct smc_init_info *ini) 2367 { 2368 int prfx_rc; 2369 2370 /* check for ISM device matching V2 proposed device */ 2371 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2372 if (ini->ism_dev[0]) 2373 return 0; 2374 2375 /* check for matching IP prefix and subnet length (V1) */ 2376 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2377 if (prfx_rc) 2378 smc_find_ism_store_rc(prfx_rc, ini); 2379 2380 /* get vlan id from IP device */ 2381 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2382 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2383 2384 /* check for ISM device matching V1 proposed device */ 2385 if (!prfx_rc) 2386 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2387 if (ini->ism_dev[0]) 2388 return 0; 2389 2390 if (!smcr_indicated(pclc->hdr.typev1) && 2391 !smcr_indicated(pclc->hdr.typev2)) 2392 /* skip RDMA and decline */ 2393 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2394 2395 /* check if RDMA V2 is available */ 2396 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2397 if (ini->smcrv2.ib_dev_v2) 2398 return 0; 2399 2400 /* check if RDMA V1 is available */ 2401 if (!prfx_rc) { 2402 int rc; 2403 2404 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2405 smc_find_ism_store_rc(rc, ini); 2406 return (!rc) ? 0 : ini->rc; 2407 } 2408 return prfx_rc; 2409 } 2410 2411 /* listen worker: finish RDMA setup */ 2412 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2413 struct smc_clc_msg_accept_confirm *cclc, 2414 bool local_first, 2415 struct smc_init_info *ini) 2416 { 2417 struct smc_link *link = new_smc->conn.lnk; 2418 int reason_code = 0; 2419 2420 if (local_first) 2421 smc_link_save_peer_info(link, cclc, ini); 2422 2423 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2424 return SMC_CLC_DECL_ERR_RTOK; 2425 2426 if (local_first) { 2427 if (smc_ib_ready_link(link)) 2428 return SMC_CLC_DECL_ERR_RDYLNK; 2429 /* QP confirmation over RoCE fabric */ 2430 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2431 reason_code = smcr_serv_conf_first_link(new_smc); 2432 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2433 } 2434 return reason_code; 2435 } 2436 2437 /* setup for connection of server */ 2438 static void smc_listen_work(struct work_struct *work) 2439 { 2440 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2441 smc_listen_work); 2442 struct socket *newclcsock = new_smc->clcsock; 2443 struct smc_clc_msg_accept_confirm *cclc; 2444 struct smc_clc_msg_proposal_area *buf; 2445 struct smc_clc_msg_proposal *pclc; 2446 struct smc_init_info *ini = NULL; 2447 u8 proposal_version = SMC_V1; 2448 u8 accept_version; 2449 int rc = 0; 2450 2451 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2452 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2453 return smc_listen_out_err(new_smc); 2454 2455 if (new_smc->use_fallback) { 2456 smc_listen_out_connected(new_smc); 2457 return; 2458 } 2459 2460 /* check if peer is smc capable */ 2461 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2462 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2463 if (rc) 2464 smc_listen_out_err(new_smc); 2465 else 2466 smc_listen_out_connected(new_smc); 2467 return; 2468 } 2469 2470 /* do inband token exchange - 2471 * wait for and receive SMC Proposal CLC message 2472 */ 2473 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2474 if (!buf) { 2475 rc = SMC_CLC_DECL_MEM; 2476 goto out_decl; 2477 } 2478 pclc = (struct smc_clc_msg_proposal *)buf; 2479 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2480 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2481 if (rc) 2482 goto out_decl; 2483 2484 if (pclc->hdr.version > SMC_V1) 2485 proposal_version = SMC_V2; 2486 2487 /* IPSec connections opt out of SMC optimizations */ 2488 if (using_ipsec(new_smc)) { 2489 rc = SMC_CLC_DECL_IPSEC; 2490 goto out_decl; 2491 } 2492 2493 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2494 if (!ini) { 2495 rc = SMC_CLC_DECL_MEM; 2496 goto out_decl; 2497 } 2498 2499 /* initial version checking */ 2500 rc = smc_listen_v2_check(new_smc, pclc, ini); 2501 if (rc) 2502 goto out_decl; 2503 2504 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2505 if (rc) 2506 goto out_decl; 2507 2508 mutex_lock(&smc_server_lgr_pending); 2509 smc_rx_init(new_smc); 2510 smc_tx_init(new_smc); 2511 2512 /* determine ISM or RoCE device used for connection */ 2513 rc = smc_listen_find_device(new_smc, pclc, ini); 2514 if (rc) 2515 goto out_unlock; 2516 2517 /* send SMC Accept CLC message */ 2518 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2519 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2520 accept_version, ini->negotiated_eid, ini); 2521 if (rc) 2522 goto out_unlock; 2523 2524 /* SMC-D does not need this lock any more */ 2525 if (ini->is_smcd) 2526 mutex_unlock(&smc_server_lgr_pending); 2527 2528 /* receive SMC Confirm CLC message */ 2529 memset(buf, 0, sizeof(*buf)); 2530 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2531 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2532 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2533 if (rc) { 2534 if (!ini->is_smcd) 2535 goto out_unlock; 2536 goto out_decl; 2537 } 2538 2539 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2540 if (rc) { 2541 if (!ini->is_smcd) 2542 goto out_unlock; 2543 goto out_decl; 2544 } 2545 2546 /* fce smc release version is needed in smc_listen_rdma_finish, 2547 * so save fce info here. 2548 */ 2549 smc_conn_save_peer_info_fce(new_smc, cclc); 2550 2551 /* finish worker */ 2552 if (!ini->is_smcd) { 2553 rc = smc_listen_rdma_finish(new_smc, cclc, 2554 ini->first_contact_local, ini); 2555 if (rc) 2556 goto out_unlock; 2557 mutex_unlock(&smc_server_lgr_pending); 2558 } 2559 smc_conn_save_peer_info(new_smc, cclc); 2560 2561 if (ini->is_smcd && 2562 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2563 rc = smcd_buf_attach(new_smc); 2564 if (rc) 2565 goto out_decl; 2566 } 2567 2568 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2569 /* smc_listen_out() will release smcsk */ 2570 smc_listen_out_connected(new_smc); 2571 goto out_free; 2572 2573 out_unlock: 2574 mutex_unlock(&smc_server_lgr_pending); 2575 out_decl: 2576 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2577 proposal_version); 2578 out_free: 2579 kfree(ini); 2580 kfree(buf); 2581 } 2582 2583 static void smc_tcp_listen_work(struct work_struct *work) 2584 { 2585 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2586 tcp_listen_work); 2587 struct sock *lsk = &lsmc->sk; 2588 struct smc_sock *new_smc; 2589 int rc = 0; 2590 2591 lock_sock(lsk); 2592 while (lsk->sk_state == SMC_LISTEN) { 2593 rc = smc_clcsock_accept(lsmc, &new_smc); 2594 if (rc) /* clcsock accept queue empty or error */ 2595 goto out; 2596 if (!new_smc) 2597 continue; 2598 2599 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2600 atomic_inc(&lsmc->queued_smc_hs); 2601 2602 new_smc->listen_smc = lsmc; 2603 new_smc->use_fallback = lsmc->use_fallback; 2604 new_smc->fallback_rsn = lsmc->fallback_rsn; 2605 sock_hold(lsk); /* sock_put in smc_listen_work */ 2606 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2607 smc_copy_sock_settings_to_smc(new_smc); 2608 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2609 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2610 sock_put(&new_smc->sk); 2611 } 2612 2613 out: 2614 release_sock(lsk); 2615 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2616 } 2617 2618 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2619 { 2620 struct smc_sock *lsmc; 2621 2622 read_lock_bh(&listen_clcsock->sk_callback_lock); 2623 lsmc = smc_clcsock_user_data(listen_clcsock); 2624 if (!lsmc) 2625 goto out; 2626 lsmc->clcsk_data_ready(listen_clcsock); 2627 if (lsmc->sk.sk_state == SMC_LISTEN) { 2628 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2629 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2630 sock_put(&lsmc->sk); 2631 } 2632 out: 2633 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2634 } 2635 2636 int smc_listen(struct socket *sock, int backlog) 2637 { 2638 struct sock *sk = sock->sk; 2639 struct smc_sock *smc; 2640 int rc; 2641 2642 smc = smc_sk(sk); 2643 lock_sock(sk); 2644 2645 rc = -EINVAL; 2646 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2647 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2648 goto out; 2649 2650 rc = 0; 2651 if (sk->sk_state == SMC_LISTEN) { 2652 sk->sk_max_ack_backlog = backlog; 2653 goto out; 2654 } 2655 /* some socket options are handled in core, so we could not apply 2656 * them to the clc socket -- copy smc socket options to clc socket 2657 */ 2658 smc_copy_sock_settings_to_clc(smc); 2659 if (!smc->use_fallback) 2660 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2661 2662 /* save original sk_data_ready function and establish 2663 * smc-specific sk_data_ready function 2664 */ 2665 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2666 smc->clcsock->sk->sk_user_data = 2667 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2668 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2669 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2670 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2671 2672 /* save original ops */ 2673 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2674 2675 smc->af_ops = *smc->ori_af_ops; 2676 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2677 2678 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2679 2680 if (smc->limit_smc_hs) 2681 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2682 2683 rc = kernel_listen(smc->clcsock, backlog); 2684 if (rc) { 2685 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2686 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2687 &smc->clcsk_data_ready); 2688 smc->clcsock->sk->sk_user_data = NULL; 2689 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2690 goto out; 2691 } 2692 sk->sk_max_ack_backlog = backlog; 2693 sk->sk_ack_backlog = 0; 2694 sk->sk_state = SMC_LISTEN; 2695 2696 out: 2697 release_sock(sk); 2698 return rc; 2699 } 2700 2701 int smc_accept(struct socket *sock, struct socket *new_sock, 2702 struct proto_accept_arg *arg) 2703 { 2704 struct sock *sk = sock->sk, *nsk; 2705 DECLARE_WAITQUEUE(wait, current); 2706 struct smc_sock *lsmc; 2707 long timeo; 2708 int rc = 0; 2709 2710 lsmc = smc_sk(sk); 2711 sock_hold(sk); /* sock_put below */ 2712 lock_sock(sk); 2713 2714 if (lsmc->sk.sk_state != SMC_LISTEN) { 2715 rc = -EINVAL; 2716 release_sock(sk); 2717 goto out; 2718 } 2719 2720 /* Wait for an incoming connection */ 2721 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2722 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2723 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2724 set_current_state(TASK_INTERRUPTIBLE); 2725 if (!timeo) { 2726 rc = -EAGAIN; 2727 break; 2728 } 2729 release_sock(sk); 2730 timeo = schedule_timeout(timeo); 2731 /* wakeup by sk_data_ready in smc_listen_work() */ 2732 sched_annotate_sleep(); 2733 lock_sock(sk); 2734 if (signal_pending(current)) { 2735 rc = sock_intr_errno(timeo); 2736 break; 2737 } 2738 } 2739 set_current_state(TASK_RUNNING); 2740 remove_wait_queue(sk_sleep(sk), &wait); 2741 2742 if (!rc) 2743 rc = sock_error(nsk); 2744 release_sock(sk); 2745 if (rc) 2746 goto out; 2747 2748 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2749 /* wait till data arrives on the socket */ 2750 timeo = secs_to_jiffies(lsmc->sockopt_defer_accept); 2751 if (smc_sk(nsk)->use_fallback) { 2752 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2753 2754 lock_sock(clcsk); 2755 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2756 sk_wait_data(clcsk, &timeo, NULL); 2757 release_sock(clcsk); 2758 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2759 lock_sock(nsk); 2760 smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available); 2761 release_sock(nsk); 2762 } 2763 } 2764 2765 out: 2766 sock_put(sk); /* sock_hold above */ 2767 return rc; 2768 } 2769 2770 int smc_getname(struct socket *sock, struct sockaddr *addr, 2771 int peer) 2772 { 2773 struct smc_sock *smc; 2774 2775 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2776 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2777 return -ENOTCONN; 2778 2779 smc = smc_sk(sock->sk); 2780 2781 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2782 } 2783 2784 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2785 { 2786 struct sock *sk = sock->sk; 2787 struct smc_sock *smc; 2788 int rc; 2789 2790 smc = smc_sk(sk); 2791 lock_sock(sk); 2792 2793 /* SMC does not support connect with fastopen */ 2794 if (msg->msg_flags & MSG_FASTOPEN) { 2795 /* not connected yet, fallback */ 2796 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2797 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2798 if (rc) 2799 goto out; 2800 } else { 2801 rc = -EINVAL; 2802 goto out; 2803 } 2804 } else if ((sk->sk_state != SMC_ACTIVE) && 2805 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2806 (sk->sk_state != SMC_INIT)) { 2807 rc = -EPIPE; 2808 goto out; 2809 } 2810 2811 if (smc->use_fallback) { 2812 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2813 } else { 2814 rc = smc_tx_sendmsg(smc, msg, len); 2815 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2816 } 2817 out: 2818 release_sock(sk); 2819 return rc; 2820 } 2821 2822 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2823 int flags) 2824 { 2825 struct sock *sk = sock->sk; 2826 struct smc_sock *smc; 2827 int rc = -ENOTCONN; 2828 2829 smc = smc_sk(sk); 2830 lock_sock(sk); 2831 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2832 /* socket was connected before, no more data to read */ 2833 rc = 0; 2834 goto out; 2835 } 2836 if ((sk->sk_state == SMC_INIT) || 2837 (sk->sk_state == SMC_LISTEN) || 2838 (sk->sk_state == SMC_CLOSED)) 2839 goto out; 2840 2841 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2842 rc = 0; 2843 goto out; 2844 } 2845 2846 if (smc->use_fallback) { 2847 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2848 } else { 2849 msg->msg_namelen = 0; 2850 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2851 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2852 } 2853 2854 out: 2855 release_sock(sk); 2856 return rc; 2857 } 2858 2859 static __poll_t smc_accept_poll(struct sock *parent) 2860 { 2861 struct smc_sock *isk = smc_sk(parent); 2862 __poll_t mask = 0; 2863 2864 spin_lock(&isk->accept_q_lock); 2865 if (!list_empty(&isk->accept_q)) 2866 mask = EPOLLIN | EPOLLRDNORM; 2867 spin_unlock(&isk->accept_q_lock); 2868 2869 return mask; 2870 } 2871 2872 __poll_t smc_poll(struct file *file, struct socket *sock, 2873 poll_table *wait) 2874 { 2875 struct sock *sk = sock->sk; 2876 struct smc_sock *smc; 2877 __poll_t mask = 0; 2878 2879 if (!sk) 2880 return EPOLLNVAL; 2881 2882 smc = smc_sk(sock->sk); 2883 if (smc->use_fallback) { 2884 /* delegate to CLC child sock */ 2885 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2886 sk->sk_err = smc->clcsock->sk->sk_err; 2887 } else { 2888 if (sk->sk_state != SMC_CLOSED) 2889 sock_poll_wait(file, sock, wait); 2890 if (sk->sk_err) 2891 mask |= EPOLLERR; 2892 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2893 (sk->sk_state == SMC_CLOSED)) 2894 mask |= EPOLLHUP; 2895 if (sk->sk_state == SMC_LISTEN) { 2896 /* woken up by sk_data_ready in smc_listen_work() */ 2897 mask |= smc_accept_poll(sk); 2898 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2899 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2900 wait); 2901 sk->sk_err = smc->clcsock->sk->sk_err; 2902 } else { 2903 if ((sk->sk_state != SMC_INIT && 2904 atomic_read(&smc->conn.sndbuf_space)) || 2905 sk->sk_shutdown & SEND_SHUTDOWN) { 2906 mask |= EPOLLOUT | EPOLLWRNORM; 2907 } else { 2908 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2909 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2910 2911 if (sk->sk_state != SMC_INIT) { 2912 /* Race breaker the same way as tcp_poll(). */ 2913 smp_mb__after_atomic(); 2914 if (atomic_read(&smc->conn.sndbuf_space)) 2915 mask |= EPOLLOUT | EPOLLWRNORM; 2916 } 2917 } 2918 if (atomic_read(&smc->conn.bytes_to_rcv)) 2919 mask |= EPOLLIN | EPOLLRDNORM; 2920 if (sk->sk_shutdown & RCV_SHUTDOWN) 2921 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2922 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2923 mask |= EPOLLIN; 2924 if (smc->conn.urg_state == SMC_URG_VALID) 2925 mask |= EPOLLPRI; 2926 } 2927 } 2928 2929 return mask; 2930 } 2931 2932 int smc_shutdown(struct socket *sock, int how) 2933 { 2934 struct sock *sk = sock->sk; 2935 bool do_shutdown = true; 2936 struct smc_sock *smc; 2937 int rc = -EINVAL; 2938 int old_state; 2939 int rc1 = 0; 2940 2941 smc = smc_sk(sk); 2942 2943 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2944 return rc; 2945 2946 lock_sock(sk); 2947 2948 if (sock->state == SS_CONNECTING) { 2949 if (sk->sk_state == SMC_ACTIVE) 2950 sock->state = SS_CONNECTED; 2951 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2952 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2953 sk->sk_state == SMC_APPCLOSEWAIT1 || 2954 sk->sk_state == SMC_APPCLOSEWAIT2 || 2955 sk->sk_state == SMC_APPFINCLOSEWAIT) 2956 sock->state = SS_DISCONNECTING; 2957 } 2958 2959 rc = -ENOTCONN; 2960 if ((sk->sk_state != SMC_ACTIVE) && 2961 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2962 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2963 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2964 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2965 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2966 goto out; 2967 if (smc->use_fallback) { 2968 rc = kernel_sock_shutdown(smc->clcsock, how); 2969 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2970 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2971 sk->sk_state = SMC_CLOSED; 2972 sk->sk_socket->state = SS_UNCONNECTED; 2973 sock_put(sk); 2974 } 2975 goto out; 2976 } 2977 switch (how) { 2978 case SHUT_RDWR: /* shutdown in both directions */ 2979 old_state = sk->sk_state; 2980 rc = smc_close_active(smc); 2981 if (old_state == SMC_ACTIVE && 2982 sk->sk_state == SMC_PEERCLOSEWAIT1) 2983 do_shutdown = false; 2984 break; 2985 case SHUT_WR: 2986 rc = smc_close_shutdown_write(smc); 2987 break; 2988 case SHUT_RD: 2989 rc = 0; 2990 /* nothing more to do because peer is not involved */ 2991 break; 2992 } 2993 if (do_shutdown && smc->clcsock) 2994 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2995 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2996 sk->sk_shutdown |= how + 1; 2997 2998 if (sk->sk_state == SMC_CLOSED) 2999 sock->state = SS_UNCONNECTED; 3000 else 3001 sock->state = SS_DISCONNECTING; 3002 out: 3003 release_sock(sk); 3004 return rc ? rc : rc1; 3005 } 3006 3007 static int __smc_getsockopt(struct socket *sock, int level, int optname, 3008 char __user *optval, int __user *optlen) 3009 { 3010 struct smc_sock *smc; 3011 int val, len; 3012 3013 smc = smc_sk(sock->sk); 3014 3015 if (get_user(len, optlen)) 3016 return -EFAULT; 3017 3018 len = min_t(int, len, sizeof(int)); 3019 3020 if (len < 0) 3021 return -EINVAL; 3022 3023 switch (optname) { 3024 case SMC_LIMIT_HS: 3025 val = smc->limit_smc_hs; 3026 break; 3027 default: 3028 return -EOPNOTSUPP; 3029 } 3030 3031 if (put_user(len, optlen)) 3032 return -EFAULT; 3033 if (copy_to_user(optval, &val, len)) 3034 return -EFAULT; 3035 3036 return 0; 3037 } 3038 3039 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3040 sockptr_t optval, unsigned int optlen) 3041 { 3042 struct sock *sk = sock->sk; 3043 struct smc_sock *smc; 3044 int val, rc; 3045 3046 smc = smc_sk(sk); 3047 3048 lock_sock(sk); 3049 switch (optname) { 3050 case SMC_LIMIT_HS: 3051 if (optlen < sizeof(int)) { 3052 rc = -EINVAL; 3053 break; 3054 } 3055 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3056 rc = -EFAULT; 3057 break; 3058 } 3059 3060 smc->limit_smc_hs = !!val; 3061 rc = 0; 3062 break; 3063 default: 3064 rc = -EOPNOTSUPP; 3065 break; 3066 } 3067 release_sock(sk); 3068 3069 return rc; 3070 } 3071 3072 int smc_setsockopt(struct socket *sock, int level, int optname, 3073 sockptr_t optval, unsigned int optlen) 3074 { 3075 struct sock *sk = sock->sk; 3076 struct smc_sock *smc; 3077 int val, rc; 3078 3079 if (level == SOL_TCP && optname == TCP_ULP) 3080 return -EOPNOTSUPP; 3081 else if (level == SOL_SMC) 3082 return __smc_setsockopt(sock, level, optname, optval, optlen); 3083 3084 smc = smc_sk(sk); 3085 3086 /* generic setsockopts reaching us here always apply to the 3087 * CLC socket 3088 */ 3089 mutex_lock(&smc->clcsock_release_lock); 3090 if (!smc->clcsock) { 3091 mutex_unlock(&smc->clcsock_release_lock); 3092 return -EBADF; 3093 } 3094 if (unlikely(!smc->clcsock->ops->setsockopt)) 3095 rc = -EOPNOTSUPP; 3096 else 3097 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3098 optval, optlen); 3099 if (smc->clcsock->sk->sk_err) { 3100 sk->sk_err = smc->clcsock->sk->sk_err; 3101 sk_error_report(sk); 3102 } 3103 mutex_unlock(&smc->clcsock_release_lock); 3104 3105 if (optlen < sizeof(int)) 3106 return -EINVAL; 3107 if (copy_from_sockptr(&val, optval, sizeof(int))) 3108 return -EFAULT; 3109 3110 lock_sock(sk); 3111 if (rc || smc->use_fallback) 3112 goto out; 3113 switch (optname) { 3114 case TCP_FASTOPEN: 3115 case TCP_FASTOPEN_CONNECT: 3116 case TCP_FASTOPEN_KEY: 3117 case TCP_FASTOPEN_NO_COOKIE: 3118 /* option not supported by SMC */ 3119 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3120 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3121 } else { 3122 rc = -EINVAL; 3123 } 3124 break; 3125 case TCP_NODELAY: 3126 if (sk->sk_state != SMC_INIT && 3127 sk->sk_state != SMC_LISTEN && 3128 sk->sk_state != SMC_CLOSED) { 3129 if (val) { 3130 SMC_STAT_INC(smc, ndly_cnt); 3131 smc_tx_pending(&smc->conn); 3132 cancel_delayed_work(&smc->conn.tx_work); 3133 } 3134 } 3135 break; 3136 case TCP_CORK: 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, cork_cnt); 3142 smc_tx_pending(&smc->conn); 3143 cancel_delayed_work(&smc->conn.tx_work); 3144 } 3145 } 3146 break; 3147 case TCP_DEFER_ACCEPT: 3148 smc->sockopt_defer_accept = val; 3149 break; 3150 default: 3151 break; 3152 } 3153 out: 3154 release_sock(sk); 3155 3156 return rc; 3157 } 3158 3159 int smc_getsockopt(struct socket *sock, int level, int optname, 3160 char __user *optval, int __user *optlen) 3161 { 3162 struct smc_sock *smc; 3163 int rc; 3164 3165 if (level == SOL_SMC) 3166 return __smc_getsockopt(sock, level, optname, optval, optlen); 3167 3168 smc = smc_sk(sock->sk); 3169 mutex_lock(&smc->clcsock_release_lock); 3170 if (!smc->clcsock) { 3171 mutex_unlock(&smc->clcsock_release_lock); 3172 return -EBADF; 3173 } 3174 /* socket options apply to the CLC socket */ 3175 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3176 mutex_unlock(&smc->clcsock_release_lock); 3177 return -EOPNOTSUPP; 3178 } 3179 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3180 optval, optlen); 3181 mutex_unlock(&smc->clcsock_release_lock); 3182 return rc; 3183 } 3184 3185 int smc_ioctl(struct socket *sock, unsigned int cmd, 3186 unsigned long arg) 3187 { 3188 union smc_host_cursor cons, urg; 3189 struct smc_connection *conn; 3190 struct smc_sock *smc; 3191 int answ; 3192 3193 smc = smc_sk(sock->sk); 3194 conn = &smc->conn; 3195 lock_sock(&smc->sk); 3196 if (smc->use_fallback) { 3197 if (!smc->clcsock) { 3198 release_sock(&smc->sk); 3199 return -EBADF; 3200 } 3201 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3202 release_sock(&smc->sk); 3203 return answ; 3204 } 3205 switch (cmd) { 3206 case SIOCINQ: /* same as FIONREAD */ 3207 if (smc->sk.sk_state == SMC_LISTEN) { 3208 release_sock(&smc->sk); 3209 return -EINVAL; 3210 } 3211 if (smc->sk.sk_state == SMC_INIT || 3212 smc->sk.sk_state == SMC_CLOSED) 3213 answ = 0; 3214 else 3215 answ = atomic_read(&smc->conn.bytes_to_rcv); 3216 break; 3217 case SIOCOUTQ: 3218 /* output queue size (not send + not acked) */ 3219 if (smc->sk.sk_state == SMC_LISTEN) { 3220 release_sock(&smc->sk); 3221 return -EINVAL; 3222 } 3223 if (smc->sk.sk_state == SMC_INIT || 3224 smc->sk.sk_state == SMC_CLOSED) 3225 answ = 0; 3226 else 3227 answ = smc->conn.sndbuf_desc->len - 3228 atomic_read(&smc->conn.sndbuf_space); 3229 break; 3230 case SIOCOUTQNSD: 3231 /* output queue size (not send only) */ 3232 if (smc->sk.sk_state == SMC_LISTEN) { 3233 release_sock(&smc->sk); 3234 return -EINVAL; 3235 } 3236 if (smc->sk.sk_state == SMC_INIT || 3237 smc->sk.sk_state == SMC_CLOSED) 3238 answ = 0; 3239 else 3240 answ = smc_tx_prepared_sends(&smc->conn); 3241 break; 3242 case SIOCATMARK: 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 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3252 smc_curs_copy(&urg, &conn->urg_curs, conn); 3253 answ = smc_curs_diff(conn->rmb_desc->len, 3254 &cons, &urg) == 1; 3255 } 3256 break; 3257 default: 3258 release_sock(&smc->sk); 3259 return -ENOIOCTLCMD; 3260 } 3261 release_sock(&smc->sk); 3262 3263 return put_user(answ, (int __user *)arg); 3264 } 3265 3266 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3267 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3268 * updates till whenever a respective page has been fully processed. 3269 * Note that subsequent recv() calls have to wait till all splice() processing 3270 * completed. 3271 */ 3272 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3273 struct pipe_inode_info *pipe, size_t len, 3274 unsigned int flags) 3275 { 3276 struct sock *sk = sock->sk; 3277 struct smc_sock *smc; 3278 int rc = -ENOTCONN; 3279 3280 smc = smc_sk(sk); 3281 lock_sock(sk); 3282 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3283 /* socket was connected before, no more data to read */ 3284 rc = 0; 3285 goto out; 3286 } 3287 if (sk->sk_state == SMC_INIT || 3288 sk->sk_state == SMC_LISTEN || 3289 sk->sk_state == SMC_CLOSED) 3290 goto out; 3291 3292 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3293 rc = 0; 3294 goto out; 3295 } 3296 3297 if (smc->use_fallback) { 3298 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3299 pipe, len, flags); 3300 } else { 3301 if (*ppos) { 3302 rc = -ESPIPE; 3303 goto out; 3304 } 3305 if (flags & SPLICE_F_NONBLOCK) 3306 flags = MSG_DONTWAIT; 3307 else 3308 flags = 0; 3309 SMC_STAT_INC(smc, splice_cnt); 3310 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3311 } 3312 out: 3313 release_sock(sk); 3314 3315 return rc; 3316 } 3317 3318 /* must look like tcp */ 3319 static const struct proto_ops smc_sock_ops = { 3320 .family = PF_SMC, 3321 .owner = THIS_MODULE, 3322 .release = smc_release, 3323 .bind = smc_bind, 3324 .connect = smc_connect, 3325 .socketpair = sock_no_socketpair, 3326 .accept = smc_accept, 3327 .getname = smc_getname, 3328 .poll = smc_poll, 3329 .ioctl = smc_ioctl, 3330 .listen = smc_listen, 3331 .shutdown = smc_shutdown, 3332 .setsockopt = smc_setsockopt, 3333 .getsockopt = smc_getsockopt, 3334 .sendmsg = smc_sendmsg, 3335 .recvmsg = smc_recvmsg, 3336 .mmap = sock_no_mmap, 3337 .splice_read = smc_splice_read, 3338 }; 3339 3340 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3341 { 3342 struct smc_sock *smc = smc_sk(sk); 3343 int rc; 3344 3345 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3346 &smc->clcsock); 3347 if (rc) 3348 return rc; 3349 3350 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3351 * destruction; its sk_state might not be TCP_CLOSE after 3352 * smc->sk is close()d, and TCP timers can be fired later, 3353 * which need net ref. 3354 */ 3355 sk = smc->clcsock->sk; 3356 sk_net_refcnt_upgrade(sk); 3357 return 0; 3358 } 3359 3360 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3361 int kern, struct socket *clcsock) 3362 { 3363 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3364 struct smc_sock *smc; 3365 struct sock *sk; 3366 int rc; 3367 3368 rc = -ESOCKTNOSUPPORT; 3369 if (sock->type != SOCK_STREAM) 3370 goto out; 3371 3372 rc = -EPROTONOSUPPORT; 3373 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3374 goto out; 3375 3376 rc = -ENOBUFS; 3377 sock->ops = &smc_sock_ops; 3378 sock->state = SS_UNCONNECTED; 3379 sk = smc_sock_alloc(net, sock, protocol); 3380 if (!sk) 3381 goto out; 3382 3383 /* create internal TCP socket for CLC handshake and fallback */ 3384 smc = smc_sk(sk); 3385 3386 rc = 0; 3387 if (clcsock) 3388 smc->clcsock = clcsock; 3389 else 3390 rc = smc_create_clcsk(net, sk, family); 3391 3392 if (rc) { 3393 sk_common_release(sk); 3394 sock->sk = NULL; 3395 } 3396 out: 3397 return rc; 3398 } 3399 3400 static int smc_create(struct net *net, struct socket *sock, int protocol, 3401 int kern) 3402 { 3403 return __smc_create(net, sock, protocol, kern, NULL); 3404 } 3405 3406 static const struct net_proto_family smc_sock_family_ops = { 3407 .family = PF_SMC, 3408 .owner = THIS_MODULE, 3409 .create = smc_create, 3410 }; 3411 3412 static int smc_ulp_init(struct sock *sk) 3413 { 3414 struct socket *tcp = sk->sk_socket; 3415 struct net *net = sock_net(sk); 3416 struct socket *smcsock; 3417 int protocol, ret; 3418 3419 /* only TCP can be replaced */ 3420 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3421 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3422 return -ESOCKTNOSUPPORT; 3423 /* don't handle wq now */ 3424 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3425 return -ENOTCONN; 3426 3427 if (sk->sk_family == AF_INET) 3428 protocol = SMCPROTO_SMC; 3429 else 3430 protocol = SMCPROTO_SMC6; 3431 3432 smcsock = sock_alloc(); 3433 if (!smcsock) 3434 return -ENFILE; 3435 3436 smcsock->type = SOCK_STREAM; 3437 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3438 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3439 if (ret) { 3440 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3441 return ret; 3442 } 3443 3444 /* replace tcp socket to smc */ 3445 smcsock->file = tcp->file; 3446 smcsock->file->private_data = smcsock; 3447 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3448 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3449 tcp->file = NULL; 3450 3451 return ret; 3452 } 3453 3454 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3455 const gfp_t priority) 3456 { 3457 struct inet_connection_sock *icsk = inet_csk(newsk); 3458 3459 /* don't inherit ulp ops to child when listen */ 3460 icsk->icsk_ulp_ops = NULL; 3461 } 3462 3463 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3464 .name = "smc", 3465 .owner = THIS_MODULE, 3466 .init = smc_ulp_init, 3467 .clone = smc_ulp_clone, 3468 }; 3469 3470 unsigned int smc_net_id; 3471 3472 static __net_init int smc_net_init(struct net *net) 3473 { 3474 int rc; 3475 3476 rc = smc_sysctl_net_init(net); 3477 if (rc) 3478 return rc; 3479 return smc_pnet_net_init(net); 3480 } 3481 3482 static void __net_exit smc_net_exit(struct net *net) 3483 { 3484 smc_sysctl_net_exit(net); 3485 smc_pnet_net_exit(net); 3486 } 3487 3488 static __net_init int smc_net_stat_init(struct net *net) 3489 { 3490 return smc_stats_init(net); 3491 } 3492 3493 static void __net_exit smc_net_stat_exit(struct net *net) 3494 { 3495 smc_stats_exit(net); 3496 } 3497 3498 static struct pernet_operations smc_net_ops = { 3499 .init = smc_net_init, 3500 .exit = smc_net_exit, 3501 .id = &smc_net_id, 3502 .size = sizeof(struct smc_net), 3503 }; 3504 3505 static struct pernet_operations smc_net_stat_ops = { 3506 .init = smc_net_stat_init, 3507 .exit = smc_net_stat_exit, 3508 }; 3509 3510 static int __init smc_init(void) 3511 { 3512 int rc; 3513 3514 rc = register_pernet_subsys(&smc_net_ops); 3515 if (rc) 3516 return rc; 3517 3518 rc = register_pernet_subsys(&smc_net_stat_ops); 3519 if (rc) 3520 goto out_pernet_subsys; 3521 3522 rc = smc_ism_init(); 3523 if (rc) 3524 goto out_pernet_subsys_stat; 3525 smc_clc_init(); 3526 3527 rc = smc_nl_init(); 3528 if (rc) 3529 goto out_ism; 3530 3531 rc = smc_pnet_init(); 3532 if (rc) 3533 goto out_nl; 3534 3535 rc = -ENOMEM; 3536 3537 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", WQ_PERCPU, 0); 3538 if (!smc_tcp_ls_wq) 3539 goto out_pnet; 3540 3541 smc_hs_wq = alloc_workqueue("smc_hs_wq", WQ_PERCPU, 0); 3542 if (!smc_hs_wq) 3543 goto out_alloc_tcp_ls_wq; 3544 3545 smc_close_wq = alloc_workqueue("smc_close_wq", WQ_PERCPU, 0); 3546 if (!smc_close_wq) 3547 goto out_alloc_hs_wq; 3548 3549 rc = smc_core_init(); 3550 if (rc) { 3551 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3552 goto out_alloc_wqs; 3553 } 3554 3555 rc = smc_llc_init(); 3556 if (rc) { 3557 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3558 goto out_core; 3559 } 3560 3561 rc = smc_cdc_init(); 3562 if (rc) { 3563 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3564 goto out_core; 3565 } 3566 3567 rc = proto_register(&smc_proto, 1); 3568 if (rc) { 3569 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3570 goto out_core; 3571 } 3572 3573 rc = proto_register(&smc_proto6, 1); 3574 if (rc) { 3575 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3576 goto out_proto; 3577 } 3578 3579 rc = sock_register(&smc_sock_family_ops); 3580 if (rc) { 3581 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3582 goto out_proto6; 3583 } 3584 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3585 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3586 3587 rc = smc_ib_register_client(); 3588 if (rc) { 3589 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3590 goto out_sock; 3591 } 3592 3593 rc = tcp_register_ulp(&smc_ulp_ops); 3594 if (rc) { 3595 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3596 goto out_ib; 3597 } 3598 rc = smc_inet_init(); 3599 if (rc) { 3600 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3601 goto out_ulp; 3602 } 3603 static_branch_enable(&tcp_have_smc); 3604 return 0; 3605 out_ulp: 3606 tcp_unregister_ulp(&smc_ulp_ops); 3607 out_ib: 3608 smc_ib_unregister_client(); 3609 out_sock: 3610 sock_unregister(PF_SMC); 3611 out_proto6: 3612 proto_unregister(&smc_proto6); 3613 out_proto: 3614 proto_unregister(&smc_proto); 3615 out_core: 3616 smc_core_exit(); 3617 out_alloc_wqs: 3618 destroy_workqueue(smc_close_wq); 3619 out_alloc_hs_wq: 3620 destroy_workqueue(smc_hs_wq); 3621 out_alloc_tcp_ls_wq: 3622 destroy_workqueue(smc_tcp_ls_wq); 3623 out_pnet: 3624 smc_pnet_exit(); 3625 out_nl: 3626 smc_nl_exit(); 3627 out_ism: 3628 smc_clc_exit(); 3629 smc_ism_exit(); 3630 out_pernet_subsys_stat: 3631 unregister_pernet_subsys(&smc_net_stat_ops); 3632 out_pernet_subsys: 3633 unregister_pernet_subsys(&smc_net_ops); 3634 3635 return rc; 3636 } 3637 3638 static void __exit smc_exit(void) 3639 { 3640 static_branch_disable(&tcp_have_smc); 3641 smc_inet_exit(); 3642 tcp_unregister_ulp(&smc_ulp_ops); 3643 sock_unregister(PF_SMC); 3644 smc_core_exit(); 3645 smc_ib_unregister_client(); 3646 smc_ism_exit(); 3647 destroy_workqueue(smc_close_wq); 3648 destroy_workqueue(smc_tcp_ls_wq); 3649 destroy_workqueue(smc_hs_wq); 3650 proto_unregister(&smc_proto6); 3651 proto_unregister(&smc_proto); 3652 smc_pnet_exit(); 3653 smc_nl_exit(); 3654 smc_clc_exit(); 3655 unregister_pernet_subsys(&smc_net_stat_ops); 3656 unregister_pernet_subsys(&smc_net_ops); 3657 rcu_barrier(); 3658 } 3659 3660 module_init(smc_init); 3661 module_exit(smc_exit); 3662 3663 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3664 MODULE_DESCRIPTION("smc socket address family"); 3665 MODULE_LICENSE("GPL"); 3666 MODULE_ALIAS_NETPROTO(PF_SMC); 3667 MODULE_ALIAS_TCP_ULP("smc"); 3668 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3669 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3670 #if IS_ENABLED(CONFIG_IPV6) 3671 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3672 #endif /* CONFIG_IPV6 */ 3673 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3674