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/tcp.h> 34 #include <net/smc.h> 35 #include <asm/ioctls.h> 36 37 #include <net/net_namespace.h> 38 #include <net/netns/generic.h> 39 #include "smc_netns.h" 40 41 #include "smc.h" 42 #include "smc_clc.h" 43 #include "smc_llc.h" 44 #include "smc_cdc.h" 45 #include "smc_core.h" 46 #include "smc_ib.h" 47 #include "smc_ism.h" 48 #include "smc_pnet.h" 49 #include "smc_netlink.h" 50 #include "smc_tx.h" 51 #include "smc_rx.h" 52 #include "smc_close.h" 53 #include "smc_stats.h" 54 #include "smc_tracepoint.h" 55 #include "smc_sysctl.h" 56 57 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 58 * creation on server 59 */ 60 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 61 * creation on client 62 */ 63 64 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 65 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 66 struct workqueue_struct *smc_close_wq; /* wq for close work */ 67 68 static void smc_tcp_listen_work(struct work_struct *); 69 static void smc_connect_work(struct work_struct *); 70 71 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 72 { 73 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 74 void *hdr; 75 76 if (cb_ctx->pos[0]) 77 goto out; 78 79 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 80 &smc_gen_nl_family, NLM_F_MULTI, 81 SMC_NETLINK_DUMP_HS_LIMITATION); 82 if (!hdr) 83 return -ENOMEM; 84 85 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 86 sock_net(skb->sk)->smc.limit_smc_hs)) 87 goto err; 88 89 genlmsg_end(skb, hdr); 90 cb_ctx->pos[0] = 1; 91 out: 92 return skb->len; 93 err: 94 genlmsg_cancel(skb, hdr); 95 return -EMSGSIZE; 96 } 97 98 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 99 { 100 sock_net(skb->sk)->smc.limit_smc_hs = true; 101 return 0; 102 } 103 104 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 105 { 106 sock_net(skb->sk)->smc.limit_smc_hs = false; 107 return 0; 108 } 109 110 static void smc_set_keepalive(struct sock *sk, int val) 111 { 112 struct smc_sock *smc = smc_sk(sk); 113 114 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 115 } 116 117 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 118 struct sk_buff *skb, 119 struct request_sock *req, 120 struct dst_entry *dst, 121 struct request_sock *req_unhash, 122 bool *own_req) 123 { 124 struct smc_sock *smc; 125 struct sock *child; 126 127 smc = smc_clcsock_user_data(sk); 128 129 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 130 sk->sk_max_ack_backlog) 131 goto drop; 132 133 if (sk_acceptq_is_full(&smc->sk)) { 134 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 135 goto drop; 136 } 137 138 /* passthrough to original syn recv sock fct */ 139 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 140 own_req); 141 /* child must not inherit smc or its ops */ 142 if (child) { 143 rcu_assign_sk_user_data(child, NULL); 144 145 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 146 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 147 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 148 } 149 return child; 150 151 drop: 152 dst_release(dst); 153 tcp_listendrop(sk); 154 return NULL; 155 } 156 157 static bool smc_hs_congested(const struct sock *sk) 158 { 159 const struct smc_sock *smc; 160 161 smc = smc_clcsock_user_data(sk); 162 163 if (!smc) 164 return true; 165 166 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 167 return true; 168 169 return false; 170 } 171 172 static struct smc_hashinfo smc_v4_hashinfo = { 173 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 174 }; 175 176 static struct smc_hashinfo smc_v6_hashinfo = { 177 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 178 }; 179 180 static int smc_hash_sk(struct sock *sk) 181 { 182 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 183 struct hlist_head *head; 184 185 head = &h->ht; 186 187 write_lock_bh(&h->lock); 188 sk_add_node(sk, head); 189 write_unlock_bh(&h->lock); 190 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 191 192 return 0; 193 } 194 195 static void smc_unhash_sk(struct sock *sk) 196 { 197 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 198 199 write_lock_bh(&h->lock); 200 if (sk_del_node_init(sk)) 201 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 202 write_unlock_bh(&h->lock); 203 } 204 205 /* This will be called before user really release sock_lock. So do the 206 * work which we didn't do because of user hold the sock_lock in the 207 * BH context 208 */ 209 static void smc_release_cb(struct sock *sk) 210 { 211 struct smc_sock *smc = smc_sk(sk); 212 213 if (smc->conn.tx_in_release_sock) { 214 smc_tx_pending(&smc->conn); 215 smc->conn.tx_in_release_sock = false; 216 } 217 } 218 219 struct proto smc_proto = { 220 .name = "SMC", 221 .owner = THIS_MODULE, 222 .keepalive = smc_set_keepalive, 223 .hash = smc_hash_sk, 224 .unhash = smc_unhash_sk, 225 .release_cb = smc_release_cb, 226 .obj_size = sizeof(struct smc_sock), 227 .h.smc_hash = &smc_v4_hashinfo, 228 .slab_flags = SLAB_TYPESAFE_BY_RCU, 229 }; 230 EXPORT_SYMBOL_GPL(smc_proto); 231 232 struct proto smc_proto6 = { 233 .name = "SMC6", 234 .owner = THIS_MODULE, 235 .keepalive = smc_set_keepalive, 236 .hash = smc_hash_sk, 237 .unhash = smc_unhash_sk, 238 .release_cb = smc_release_cb, 239 .obj_size = sizeof(struct smc_sock), 240 .h.smc_hash = &smc_v6_hashinfo, 241 .slab_flags = SLAB_TYPESAFE_BY_RCU, 242 }; 243 EXPORT_SYMBOL_GPL(smc_proto6); 244 245 static void smc_fback_restore_callbacks(struct smc_sock *smc) 246 { 247 struct sock *clcsk = smc->clcsock->sk; 248 249 write_lock_bh(&clcsk->sk_callback_lock); 250 clcsk->sk_user_data = NULL; 251 252 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 253 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 254 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 255 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 256 257 write_unlock_bh(&clcsk->sk_callback_lock); 258 } 259 260 static void smc_restore_fallback_changes(struct smc_sock *smc) 261 { 262 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 263 smc->clcsock->file->private_data = smc->sk.sk_socket; 264 smc->clcsock->file = NULL; 265 smc_fback_restore_callbacks(smc); 266 } 267 } 268 269 static int __smc_release(struct smc_sock *smc) 270 { 271 struct sock *sk = &smc->sk; 272 int rc = 0; 273 274 if (!smc->use_fallback) { 275 rc = smc_close_active(smc); 276 smc_sock_set_flag(sk, SOCK_DEAD); 277 sk->sk_shutdown |= SHUTDOWN_MASK; 278 } else { 279 if (sk->sk_state != SMC_CLOSED) { 280 if (sk->sk_state != SMC_LISTEN && 281 sk->sk_state != SMC_INIT) 282 sock_put(sk); /* passive closing */ 283 if (sk->sk_state == SMC_LISTEN) { 284 /* wake up clcsock accept */ 285 rc = kernel_sock_shutdown(smc->clcsock, 286 SHUT_RDWR); 287 } 288 sk->sk_state = SMC_CLOSED; 289 sk->sk_state_change(sk); 290 } 291 smc_restore_fallback_changes(smc); 292 } 293 294 sk->sk_prot->unhash(sk); 295 296 if (sk->sk_state == SMC_CLOSED) { 297 if (smc->clcsock) { 298 release_sock(sk); 299 smc_clcsock_release(smc); 300 lock_sock(sk); 301 } 302 if (!smc->use_fallback) 303 smc_conn_free(&smc->conn); 304 } 305 306 return rc; 307 } 308 309 static int smc_release(struct socket *sock) 310 { 311 struct sock *sk = sock->sk; 312 struct smc_sock *smc; 313 int old_state, rc = 0; 314 315 if (!sk) 316 goto out; 317 318 sock_hold(sk); /* sock_put below */ 319 smc = smc_sk(sk); 320 321 old_state = sk->sk_state; 322 323 /* cleanup for a dangling non-blocking connect */ 324 if (smc->connect_nonblock && old_state == SMC_INIT) 325 tcp_abort(smc->clcsock->sk, ECONNABORTED); 326 327 if (cancel_work_sync(&smc->connect_work)) 328 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 329 330 if (sk->sk_state == SMC_LISTEN) 331 /* smc_close_non_accepted() is called and acquires 332 * sock lock for child sockets again 333 */ 334 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 335 else 336 lock_sock(sk); 337 338 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 339 !smc->use_fallback) 340 smc_close_active_abort(smc); 341 342 rc = __smc_release(smc); 343 344 /* detach socket */ 345 sock_orphan(sk); 346 sock->sk = NULL; 347 release_sock(sk); 348 349 sock_put(sk); /* sock_hold above */ 350 sock_put(sk); /* final sock_put */ 351 out: 352 return rc; 353 } 354 355 static void smc_destruct(struct sock *sk) 356 { 357 if (sk->sk_state != SMC_CLOSED) 358 return; 359 if (!sock_flag(sk, SOCK_DEAD)) 360 return; 361 } 362 363 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 364 int protocol) 365 { 366 struct smc_sock *smc; 367 struct proto *prot; 368 struct sock *sk; 369 370 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 371 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 372 if (!sk) 373 return NULL; 374 375 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 376 sk->sk_state = SMC_INIT; 377 sk->sk_destruct = smc_destruct; 378 sk->sk_protocol = protocol; 379 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 380 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 381 smc = smc_sk(sk); 382 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 383 INIT_WORK(&smc->connect_work, smc_connect_work); 384 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 385 INIT_LIST_HEAD(&smc->accept_q); 386 spin_lock_init(&smc->accept_q_lock); 387 spin_lock_init(&smc->conn.send_lock); 388 sk->sk_prot->hash(sk); 389 mutex_init(&smc->clcsock_release_lock); 390 smc_init_saved_callbacks(smc); 391 392 return sk; 393 } 394 395 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 396 int addr_len) 397 { 398 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 399 struct sock *sk = sock->sk; 400 struct smc_sock *smc; 401 int rc; 402 403 smc = smc_sk(sk); 404 405 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 406 rc = -EINVAL; 407 if (addr_len < sizeof(struct sockaddr_in)) 408 goto out; 409 410 rc = -EAFNOSUPPORT; 411 if (addr->sin_family != AF_INET && 412 addr->sin_family != AF_INET6 && 413 addr->sin_family != AF_UNSPEC) 414 goto out; 415 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 416 if (addr->sin_family == AF_UNSPEC && 417 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 418 goto out; 419 420 lock_sock(sk); 421 422 /* Check if socket is already active */ 423 rc = -EINVAL; 424 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 425 goto out_rel; 426 427 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 428 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 429 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 430 431 out_rel: 432 release_sock(sk); 433 out: 434 return rc; 435 } 436 437 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 438 * clc socket (since smc is not called for these options from net/core) 439 */ 440 441 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 442 (1UL << SOCK_KEEPOPEN) | \ 443 (1UL << SOCK_LINGER) | \ 444 (1UL << SOCK_BROADCAST) | \ 445 (1UL << SOCK_TIMESTAMP) | \ 446 (1UL << SOCK_DBG) | \ 447 (1UL << SOCK_RCVTSTAMP) | \ 448 (1UL << SOCK_RCVTSTAMPNS) | \ 449 (1UL << SOCK_LOCALROUTE) | \ 450 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 451 (1UL << SOCK_RXQ_OVFL) | \ 452 (1UL << SOCK_WIFI_STATUS) | \ 453 (1UL << SOCK_NOFCS) | \ 454 (1UL << SOCK_FILTER_LOCKED) | \ 455 (1UL << SOCK_TSTAMP_NEW)) 456 457 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 458 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 459 unsigned long mask) 460 { 461 struct net *nnet = sock_net(nsk); 462 463 nsk->sk_userlocks = osk->sk_userlocks; 464 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) { 465 nsk->sk_sndbuf = osk->sk_sndbuf; 466 } else { 467 if (mask == SK_FLAGS_SMC_TO_CLC) 468 WRITE_ONCE(nsk->sk_sndbuf, 469 READ_ONCE(nnet->ipv4.sysctl_tcp_wmem[1])); 470 else 471 WRITE_ONCE(nsk->sk_sndbuf, 472 2 * READ_ONCE(nnet->smc.sysctl_wmem)); 473 } 474 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) { 475 nsk->sk_rcvbuf = osk->sk_rcvbuf; 476 } else { 477 if (mask == SK_FLAGS_SMC_TO_CLC) 478 WRITE_ONCE(nsk->sk_rcvbuf, 479 READ_ONCE(nnet->ipv4.sysctl_tcp_rmem[1])); 480 else 481 WRITE_ONCE(nsk->sk_rcvbuf, 482 2 * READ_ONCE(nnet->smc.sysctl_rmem)); 483 } 484 } 485 486 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 487 unsigned long mask) 488 { 489 /* options we don't get control via setsockopt for */ 490 nsk->sk_type = osk->sk_type; 491 nsk->sk_sndtimeo = osk->sk_sndtimeo; 492 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 493 nsk->sk_mark = READ_ONCE(osk->sk_mark); 494 nsk->sk_priority = READ_ONCE(osk->sk_priority); 495 nsk->sk_rcvlowat = osk->sk_rcvlowat; 496 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 497 nsk->sk_err = osk->sk_err; 498 499 nsk->sk_flags &= ~mask; 500 nsk->sk_flags |= osk->sk_flags & mask; 501 502 smc_adjust_sock_bufsizes(nsk, osk, mask); 503 } 504 505 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 506 { 507 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 508 } 509 510 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 511 (1UL << SOCK_KEEPOPEN) | \ 512 (1UL << SOCK_LINGER) | \ 513 (1UL << SOCK_DBG)) 514 /* copy only settings and flags relevant for smc from clc to smc socket */ 515 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 516 { 517 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 518 } 519 520 /* register the new vzalloced sndbuf on all links */ 521 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 522 struct smc_buf_desc *snd_desc) 523 { 524 struct smc_link_group *lgr = link->lgr; 525 int i, rc = 0; 526 527 if (!snd_desc->is_vm) 528 return -EINVAL; 529 530 /* protect against parallel smcr_link_reg_buf() */ 531 down_write(&lgr->llc_conf_mutex); 532 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 533 if (!smc_link_active(&lgr->lnk[i])) 534 continue; 535 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 536 if (rc) 537 break; 538 } 539 up_write(&lgr->llc_conf_mutex); 540 return rc; 541 } 542 543 /* register the new rmb on all links */ 544 static int smcr_lgr_reg_rmbs(struct smc_link *link, 545 struct smc_buf_desc *rmb_desc) 546 { 547 struct smc_link_group *lgr = link->lgr; 548 bool do_slow = false; 549 int i, rc = 0; 550 551 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 552 if (rc) 553 return rc; 554 555 down_read(&lgr->llc_conf_mutex); 556 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 557 if (!smc_link_active(&lgr->lnk[i])) 558 continue; 559 if (!rmb_desc->is_reg_mr[link->link_idx]) { 560 up_read(&lgr->llc_conf_mutex); 561 goto slow_path; 562 } 563 } 564 /* mr register already */ 565 goto fast_path; 566 slow_path: 567 do_slow = true; 568 /* protect against parallel smc_llc_cli_rkey_exchange() and 569 * parallel smcr_link_reg_buf() 570 */ 571 down_write(&lgr->llc_conf_mutex); 572 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 573 if (!smc_link_active(&lgr->lnk[i])) 574 continue; 575 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 576 if (rc) 577 goto out; 578 } 579 fast_path: 580 /* exchange confirm_rkey msg with peer */ 581 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 582 if (rc) { 583 rc = -EFAULT; 584 goto out; 585 } 586 rmb_desc->is_conf_rkey = true; 587 out: 588 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 589 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 590 return rc; 591 } 592 593 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 594 { 595 struct smc_link *link = smc->conn.lnk; 596 struct smc_llc_qentry *qentry; 597 int rc; 598 599 /* Receive CONFIRM LINK request from server over RoCE fabric. 600 * Increasing the client's timeout by twice as much as the server's 601 * timeout by default can temporarily avoid decline messages of 602 * both sides crossing or colliding 603 */ 604 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 605 SMC_LLC_CONFIRM_LINK); 606 if (!qentry) { 607 struct smc_clc_msg_decline dclc; 608 609 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 610 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 611 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 612 } 613 smc_llc_save_peer_uid(qentry); 614 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 615 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 616 if (rc) 617 return SMC_CLC_DECL_RMBE_EC; 618 619 rc = smc_ib_modify_qp_rts(link); 620 if (rc) 621 return SMC_CLC_DECL_ERR_RDYLNK; 622 623 smc_wr_remember_qp_attr(link); 624 625 /* reg the sndbuf if it was vzalloced */ 626 if (smc->conn.sndbuf_desc->is_vm) { 627 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 628 return SMC_CLC_DECL_ERR_REGBUF; 629 } 630 631 /* reg the rmb */ 632 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 633 return SMC_CLC_DECL_ERR_REGBUF; 634 635 /* confirm_rkey is implicit on 1st contact */ 636 smc->conn.rmb_desc->is_conf_rkey = true; 637 638 /* send CONFIRM LINK response over RoCE fabric */ 639 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 640 if (rc < 0) 641 return SMC_CLC_DECL_TIMEOUT_CL; 642 643 smc_llc_link_active(link); 644 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 645 646 if (link->lgr->max_links > 1) { 647 /* optional 2nd link, receive ADD LINK request from server */ 648 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 649 SMC_LLC_ADD_LINK); 650 if (!qentry) { 651 struct smc_clc_msg_decline dclc; 652 653 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 654 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 655 if (rc == -EAGAIN) 656 rc = 0; /* no DECLINE received, go with one link */ 657 return rc; 658 } 659 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 660 smc_llc_cli_add_link(link, qentry); 661 } 662 return 0; 663 } 664 665 static bool smc_isascii(char *hostname) 666 { 667 int i; 668 669 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 670 if (!isascii(hostname[i])) 671 return false; 672 return true; 673 } 674 675 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 676 struct smc_clc_msg_accept_confirm *clc) 677 { 678 struct smc_clc_first_contact_ext *fce; 679 int clc_v2_len; 680 681 if (clc->hdr.version == SMC_V1 || 682 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 683 return; 684 685 if (smc->conn.lgr->is_smcd) { 686 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 687 SMC_MAX_EID_LEN); 688 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 689 } else { 690 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 691 SMC_MAX_EID_LEN); 692 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 693 } 694 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 695 smc->conn.lgr->peer_os = fce->os_type; 696 smc->conn.lgr->peer_smc_release = fce->release; 697 if (smc_isascii(fce->hostname)) 698 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 699 SMC_MAX_HOSTNAME_LEN); 700 } 701 702 static void smcr_conn_save_peer_info(struct smc_sock *smc, 703 struct smc_clc_msg_accept_confirm *clc) 704 { 705 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 706 707 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 708 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 709 smc->conn.peer_rmbe_size = bufsize; 710 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 711 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 712 } 713 714 static void smcd_conn_save_peer_info(struct smc_sock *smc, 715 struct smc_clc_msg_accept_confirm *clc) 716 { 717 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 718 719 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 720 smc->conn.peer_token = ntohll(clc->d0.token); 721 /* msg header takes up space in the buffer */ 722 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 723 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 724 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 725 } 726 727 static void smc_conn_save_peer_info(struct smc_sock *smc, 728 struct smc_clc_msg_accept_confirm *clc) 729 { 730 if (smc->conn.lgr->is_smcd) 731 smcd_conn_save_peer_info(smc, clc); 732 else 733 smcr_conn_save_peer_info(smc, clc); 734 smc_conn_save_peer_info_fce(smc, clc); 735 } 736 737 static void smc_link_save_peer_info(struct smc_link *link, 738 struct smc_clc_msg_accept_confirm *clc, 739 struct smc_init_info *ini) 740 { 741 link->peer_qpn = ntoh24(clc->r0.qpn); 742 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 743 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 744 link->peer_psn = ntoh24(clc->r0.psn); 745 link->peer_mtu = clc->r0.qp_mtu; 746 } 747 748 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 749 struct smc_stats_fback *fback_arr) 750 { 751 int cnt; 752 753 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 754 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 755 fback_arr[cnt].count++; 756 break; 757 } 758 if (!fback_arr[cnt].fback_code) { 759 fback_arr[cnt].fback_code = smc->fallback_rsn; 760 fback_arr[cnt].count++; 761 break; 762 } 763 } 764 } 765 766 static void smc_stat_fallback(struct smc_sock *smc) 767 { 768 struct net *net = sock_net(&smc->sk); 769 770 mutex_lock(&net->smc.mutex_fback_rsn); 771 if (smc->listen_smc) { 772 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 773 net->smc.fback_rsn->srv_fback_cnt++; 774 } else { 775 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 776 net->smc.fback_rsn->clnt_fback_cnt++; 777 } 778 mutex_unlock(&net->smc.mutex_fback_rsn); 779 } 780 781 /* must be called under rcu read lock */ 782 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 783 { 784 struct socket_wq *wq; 785 __poll_t flags; 786 787 wq = rcu_dereference(smc->sk.sk_wq); 788 if (!skwq_has_sleeper(wq)) 789 return; 790 791 /* wake up smc sk->sk_wq */ 792 if (!key) { 793 /* sk_state_change */ 794 wake_up_interruptible_all(&wq->wait); 795 } else { 796 flags = key_to_poll(key); 797 if (flags & (EPOLLIN | EPOLLOUT)) 798 /* sk_data_ready or sk_write_space */ 799 wake_up_interruptible_sync_poll(&wq->wait, flags); 800 else if (flags & EPOLLERR) 801 /* sk_error_report */ 802 wake_up_interruptible_poll(&wq->wait, flags); 803 } 804 } 805 806 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 807 unsigned int mode, int sync, void *key) 808 { 809 struct smc_mark_woken *mark = 810 container_of(wait, struct smc_mark_woken, wait_entry); 811 812 mark->woken = true; 813 mark->key = key; 814 return 0; 815 } 816 817 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 818 void (*clcsock_callback)(struct sock *sk)) 819 { 820 struct smc_mark_woken mark = { .woken = false }; 821 struct socket_wq *wq; 822 823 init_waitqueue_func_entry(&mark.wait_entry, 824 smc_fback_mark_woken); 825 rcu_read_lock(); 826 wq = rcu_dereference(clcsk->sk_wq); 827 if (!wq) 828 goto out; 829 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 830 clcsock_callback(clcsk); 831 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 832 833 if (mark.woken) 834 smc_fback_wakeup_waitqueue(smc, mark.key); 835 out: 836 rcu_read_unlock(); 837 } 838 839 static void smc_fback_state_change(struct sock *clcsk) 840 { 841 struct smc_sock *smc; 842 843 read_lock_bh(&clcsk->sk_callback_lock); 844 smc = smc_clcsock_user_data(clcsk); 845 if (smc) 846 smc_fback_forward_wakeup(smc, clcsk, 847 smc->clcsk_state_change); 848 read_unlock_bh(&clcsk->sk_callback_lock); 849 } 850 851 static void smc_fback_data_ready(struct sock *clcsk) 852 { 853 struct smc_sock *smc; 854 855 read_lock_bh(&clcsk->sk_callback_lock); 856 smc = smc_clcsock_user_data(clcsk); 857 if (smc) 858 smc_fback_forward_wakeup(smc, clcsk, 859 smc->clcsk_data_ready); 860 read_unlock_bh(&clcsk->sk_callback_lock); 861 } 862 863 static void smc_fback_write_space(struct sock *clcsk) 864 { 865 struct smc_sock *smc; 866 867 read_lock_bh(&clcsk->sk_callback_lock); 868 smc = smc_clcsock_user_data(clcsk); 869 if (smc) 870 smc_fback_forward_wakeup(smc, clcsk, 871 smc->clcsk_write_space); 872 read_unlock_bh(&clcsk->sk_callback_lock); 873 } 874 875 static void smc_fback_error_report(struct sock *clcsk) 876 { 877 struct smc_sock *smc; 878 879 read_lock_bh(&clcsk->sk_callback_lock); 880 smc = smc_clcsock_user_data(clcsk); 881 if (smc) 882 smc_fback_forward_wakeup(smc, clcsk, 883 smc->clcsk_error_report); 884 read_unlock_bh(&clcsk->sk_callback_lock); 885 } 886 887 static void smc_fback_replace_callbacks(struct smc_sock *smc) 888 { 889 struct sock *clcsk = smc->clcsock->sk; 890 891 write_lock_bh(&clcsk->sk_callback_lock); 892 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 893 894 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 895 &smc->clcsk_state_change); 896 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 897 &smc->clcsk_data_ready); 898 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 899 &smc->clcsk_write_space); 900 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 901 &smc->clcsk_error_report); 902 903 write_unlock_bh(&clcsk->sk_callback_lock); 904 } 905 906 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 907 { 908 int rc = 0; 909 910 mutex_lock(&smc->clcsock_release_lock); 911 if (!smc->clcsock) { 912 rc = -EBADF; 913 goto out; 914 } 915 916 smc->use_fallback = true; 917 smc->fallback_rsn = reason_code; 918 smc_stat_fallback(smc); 919 trace_smc_switch_to_fallback(smc, reason_code); 920 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 921 smc->clcsock->file = smc->sk.sk_socket->file; 922 smc->clcsock->file->private_data = smc->clcsock; 923 smc->clcsock->wq.fasync_list = 924 smc->sk.sk_socket->wq.fasync_list; 925 smc->sk.sk_socket->wq.fasync_list = NULL; 926 927 /* There might be some wait entries remaining 928 * in smc sk->sk_wq and they should be woken up 929 * as clcsock's wait queue is woken up. 930 */ 931 smc_fback_replace_callbacks(smc); 932 } 933 out: 934 mutex_unlock(&smc->clcsock_release_lock); 935 return rc; 936 } 937 938 /* fall back during connect */ 939 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 940 { 941 struct net *net = sock_net(&smc->sk); 942 int rc = 0; 943 944 rc = smc_switch_to_fallback(smc, reason_code); 945 if (rc) { /* fallback fails */ 946 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 947 if (smc->sk.sk_state == SMC_INIT) 948 sock_put(&smc->sk); /* passive closing */ 949 return rc; 950 } 951 smc_copy_sock_settings_to_clc(smc); 952 smc->connect_nonblock = 0; 953 if (smc->sk.sk_state == SMC_INIT) 954 smc->sk.sk_state = SMC_ACTIVE; 955 return 0; 956 } 957 958 /* decline and fall back during connect */ 959 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 960 u8 version) 961 { 962 struct net *net = sock_net(&smc->sk); 963 int rc; 964 965 if (reason_code < 0) { /* error, fallback is not possible */ 966 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 967 if (smc->sk.sk_state == SMC_INIT) 968 sock_put(&smc->sk); /* passive closing */ 969 return reason_code; 970 } 971 if (reason_code != SMC_CLC_DECL_PEERDECL) { 972 rc = smc_clc_send_decline(smc, reason_code, version); 973 if (rc < 0) { 974 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 975 if (smc->sk.sk_state == SMC_INIT) 976 sock_put(&smc->sk); /* passive closing */ 977 return rc; 978 } 979 } 980 return smc_connect_fallback(smc, reason_code); 981 } 982 983 static void smc_conn_abort(struct smc_sock *smc, int local_first) 984 { 985 struct smc_connection *conn = &smc->conn; 986 struct smc_link_group *lgr = conn->lgr; 987 bool lgr_valid = false; 988 989 if (smc_conn_lgr_valid(conn)) 990 lgr_valid = true; 991 992 smc_conn_free(conn); 993 if (local_first && lgr_valid) 994 smc_lgr_cleanup_early(lgr); 995 } 996 997 /* check if there is a rdma device available for this connection. */ 998 /* called for connect and listen */ 999 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1000 { 1001 /* PNET table look up: search active ib_device and port 1002 * within same PNETID that also contains the ethernet device 1003 * used for the internal TCP socket 1004 */ 1005 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1006 if (!ini->check_smcrv2 && !ini->ib_dev) 1007 return SMC_CLC_DECL_NOSMCRDEV; 1008 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1009 return SMC_CLC_DECL_NOSMCRDEV; 1010 return 0; 1011 } 1012 1013 /* check if there is an ISM device available for this connection. */ 1014 /* called for connect and listen */ 1015 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1016 { 1017 /* Find ISM device with same PNETID as connecting interface */ 1018 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1019 if (!ini->ism_dev[0]) 1020 return SMC_CLC_DECL_NOSMCDDEV; 1021 else 1022 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1023 return 0; 1024 } 1025 1026 /* is chid unique for the ism devices that are already determined? */ 1027 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1028 int cnt) 1029 { 1030 int i = (!ini->ism_dev[0]) ? 1 : 0; 1031 1032 for (; i < cnt; i++) 1033 if (ini->ism_chid[i] == chid) 1034 return false; 1035 return true; 1036 } 1037 1038 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1039 * PNETID matching net_device) 1040 */ 1041 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1042 struct smc_init_info *ini) 1043 { 1044 int rc = SMC_CLC_DECL_NOSMCDDEV; 1045 struct smcd_dev *smcd; 1046 int i = 1, entry = 1; 1047 bool is_emulated; 1048 u16 chid; 1049 1050 if (smcd_indicated(ini->smc_type_v1)) 1051 rc = 0; /* already initialized for V1 */ 1052 mutex_lock(&smcd_dev_list.mutex); 1053 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1054 if (smcd->going_away || smcd == ini->ism_dev[0]) 1055 continue; 1056 chid = smc_ism_get_chid(smcd); 1057 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1058 continue; 1059 is_emulated = __smc_ism_is_emulated(chid); 1060 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1061 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1062 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1063 /* It's the last GID-CHID entry left in CLC 1064 * Proposal SMC-Dv2 extension, but an Emulated- 1065 * ISM device will take two entries. So give 1066 * up it and try the next potential ISM device. 1067 */ 1068 continue; 1069 ini->ism_dev[i] = smcd; 1070 ini->ism_chid[i] = chid; 1071 ini->is_smcd = true; 1072 rc = 0; 1073 i++; 1074 entry = is_emulated ? entry + 2 : entry + 1; 1075 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1076 break; 1077 } 1078 } 1079 mutex_unlock(&smcd_dev_list.mutex); 1080 ini->ism_offered_cnt = i - 1; 1081 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1082 ini->smcd_version = 0; 1083 1084 return rc; 1085 } 1086 1087 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1088 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1089 struct smc_init_info *ini) 1090 { 1091 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1092 return SMC_CLC_DECL_ISMVLANERR; 1093 return 0; 1094 } 1095 1096 static int smc_find_proposal_devices(struct smc_sock *smc, 1097 struct smc_init_info *ini) 1098 { 1099 int rc = 0; 1100 1101 /* check if there is an ism device available */ 1102 if (!(ini->smcd_version & SMC_V1) || 1103 smc_find_ism_device(smc, ini) || 1104 smc_connect_ism_vlan_setup(smc, ini)) 1105 ini->smcd_version &= ~SMC_V1; 1106 /* else ISM V1 is supported for this connection */ 1107 1108 /* check if there is an rdma device available */ 1109 if (!(ini->smcr_version & SMC_V1) || 1110 smc_find_rdma_device(smc, ini)) 1111 ini->smcr_version &= ~SMC_V1; 1112 /* else RDMA is supported for this connection */ 1113 1114 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1115 ini->smcr_version & SMC_V1); 1116 1117 /* check if there is an ism v2 device available */ 1118 if (!(ini->smcd_version & SMC_V2) || 1119 !smc_ism_is_v2_capable() || 1120 smc_find_ism_v2_device_clnt(smc, ini)) 1121 ini->smcd_version &= ~SMC_V2; 1122 1123 /* check if there is an rdma v2 device available */ 1124 ini->check_smcrv2 = true; 1125 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1126 if (!(ini->smcr_version & SMC_V2) || 1127 smc->clcsock->sk->sk_family != AF_INET || 1128 !smc_clc_ueid_count() || 1129 smc_find_rdma_device(smc, ini)) 1130 ini->smcr_version &= ~SMC_V2; 1131 ini->check_smcrv2 = false; 1132 1133 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1134 ini->smcr_version & SMC_V2); 1135 1136 /* if neither ISM nor RDMA are supported, fallback */ 1137 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1138 rc = SMC_CLC_DECL_NOSMCDEV; 1139 1140 return rc; 1141 } 1142 1143 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1144 * used, the VLAN ID will be registered again during the connection setup. 1145 */ 1146 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1147 struct smc_init_info *ini) 1148 { 1149 if (!smcd_indicated(ini->smc_type_v1)) 1150 return 0; 1151 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1152 return SMC_CLC_DECL_CNFERR; 1153 return 0; 1154 } 1155 1156 #define SMC_CLC_MAX_ACCEPT_LEN \ 1157 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1158 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1159 sizeof(struct smc_clc_msg_trail)) 1160 1161 /* CLC handshake during connect */ 1162 static int smc_connect_clc(struct smc_sock *smc, 1163 struct smc_clc_msg_accept_confirm *aclc, 1164 struct smc_init_info *ini) 1165 { 1166 int rc = 0; 1167 1168 /* do inband token exchange */ 1169 rc = smc_clc_send_proposal(smc, ini); 1170 if (rc) 1171 return rc; 1172 /* receive SMC Accept CLC message */ 1173 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1174 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1175 } 1176 1177 void smc_fill_gid_list(struct smc_link_group *lgr, 1178 struct smc_gidlist *gidlist, 1179 struct smc_ib_device *known_dev, u8 *known_gid) 1180 { 1181 struct smc_init_info *alt_ini = NULL; 1182 1183 memset(gidlist, 0, sizeof(*gidlist)); 1184 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1185 1186 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1187 if (!alt_ini) 1188 goto out; 1189 1190 alt_ini->vlan_id = lgr->vlan_id; 1191 alt_ini->check_smcrv2 = true; 1192 alt_ini->smcrv2.saddr = lgr->saddr; 1193 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1194 1195 if (!alt_ini->smcrv2.ib_dev_v2) 1196 goto out; 1197 1198 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1199 SMC_GID_SIZE); 1200 1201 out: 1202 kfree(alt_ini); 1203 } 1204 1205 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1206 struct smc_clc_msg_accept_confirm *aclc, 1207 struct smc_init_info *ini) 1208 { 1209 struct smc_clc_first_contact_ext *fce = 1210 smc_get_clc_first_contact_ext(aclc, false); 1211 struct net *net = sock_net(&smc->sk); 1212 int rc; 1213 1214 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1215 return 0; 1216 1217 if (fce->v2_direct) { 1218 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1219 ini->smcrv2.uses_gateway = false; 1220 } else { 1221 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1222 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1223 ini->smcrv2.nexthop_mac, 1224 &ini->smcrv2.uses_gateway)) 1225 return SMC_CLC_DECL_NOROUTE; 1226 if (!ini->smcrv2.uses_gateway) { 1227 /* mismatch: peer claims indirect, but its direct */ 1228 return SMC_CLC_DECL_NOINDIRECT; 1229 } 1230 } 1231 1232 ini->release_nr = fce->release; 1233 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1234 if (rc) 1235 return rc; 1236 1237 return 0; 1238 } 1239 1240 /* setup for RDMA connection of client */ 1241 static int smc_connect_rdma(struct smc_sock *smc, 1242 struct smc_clc_msg_accept_confirm *aclc, 1243 struct smc_init_info *ini) 1244 { 1245 int i, reason_code = 0; 1246 struct smc_link *link; 1247 u8 *eid = NULL; 1248 1249 ini->is_smcd = false; 1250 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1251 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1252 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1253 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1254 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1255 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1256 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1257 1258 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1259 if (reason_code) 1260 return reason_code; 1261 1262 mutex_lock(&smc_client_lgr_pending); 1263 reason_code = smc_conn_create(smc, ini); 1264 if (reason_code) { 1265 mutex_unlock(&smc_client_lgr_pending); 1266 return reason_code; 1267 } 1268 1269 smc_conn_save_peer_info(smc, aclc); 1270 1271 if (ini->first_contact_local) { 1272 link = smc->conn.lnk; 1273 } else { 1274 /* set link that was assigned by server */ 1275 link = NULL; 1276 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1277 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1278 1279 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1280 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1281 SMC_GID_SIZE) && 1282 (aclc->hdr.version > SMC_V1 || 1283 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1284 sizeof(l->peer_mac)))) { 1285 link = l; 1286 break; 1287 } 1288 } 1289 if (!link) { 1290 reason_code = SMC_CLC_DECL_NOSRVLINK; 1291 goto connect_abort; 1292 } 1293 smc_switch_link_and_count(&smc->conn, link); 1294 } 1295 1296 /* create send buffer and rmb */ 1297 if (smc_buf_create(smc, false)) { 1298 reason_code = SMC_CLC_DECL_MEM; 1299 goto connect_abort; 1300 } 1301 1302 if (ini->first_contact_local) 1303 smc_link_save_peer_info(link, aclc, ini); 1304 1305 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1306 reason_code = SMC_CLC_DECL_ERR_RTOK; 1307 goto connect_abort; 1308 } 1309 1310 smc_close_init(smc); 1311 smc_rx_init(smc); 1312 1313 if (ini->first_contact_local) { 1314 if (smc_ib_ready_link(link)) { 1315 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1316 goto connect_abort; 1317 } 1318 } else { 1319 /* reg sendbufs if they were vzalloced */ 1320 if (smc->conn.sndbuf_desc->is_vm) { 1321 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1322 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1323 goto connect_abort; 1324 } 1325 } 1326 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1327 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1328 goto connect_abort; 1329 } 1330 } 1331 1332 if (aclc->hdr.version > SMC_V1) { 1333 eid = aclc->r1.eid; 1334 if (ini->first_contact_local) 1335 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1336 link->smcibdev, link->gid); 1337 } 1338 1339 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1340 aclc->hdr.version, eid, ini); 1341 if (reason_code) 1342 goto connect_abort; 1343 1344 smc_tx_init(smc); 1345 1346 if (ini->first_contact_local) { 1347 /* QP confirmation over RoCE fabric */ 1348 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1349 reason_code = smcr_clnt_conf_first_link(smc); 1350 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1351 if (reason_code) 1352 goto connect_abort; 1353 } 1354 mutex_unlock(&smc_client_lgr_pending); 1355 1356 smc_copy_sock_settings_to_clc(smc); 1357 smc->connect_nonblock = 0; 1358 if (smc->sk.sk_state == SMC_INIT) 1359 smc->sk.sk_state = SMC_ACTIVE; 1360 1361 return 0; 1362 connect_abort: 1363 smc_conn_abort(smc, ini->first_contact_local); 1364 mutex_unlock(&smc_client_lgr_pending); 1365 smc->connect_nonblock = 0; 1366 1367 return reason_code; 1368 } 1369 1370 /* The server has chosen one of the proposed ISM devices for the communication. 1371 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1372 */ 1373 static int 1374 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1375 struct smc_init_info *ini) 1376 { 1377 int i; 1378 1379 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1380 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1381 ini->ism_selected = i; 1382 return 0; 1383 } 1384 } 1385 1386 return -EPROTO; 1387 } 1388 1389 /* setup for ISM connection of client */ 1390 static int smc_connect_ism(struct smc_sock *smc, 1391 struct smc_clc_msg_accept_confirm *aclc, 1392 struct smc_init_info *ini) 1393 { 1394 u8 *eid = NULL; 1395 int rc = 0; 1396 1397 ini->is_smcd = true; 1398 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1399 1400 if (aclc->hdr.version == SMC_V2) { 1401 if (ini->first_contact_peer) { 1402 struct smc_clc_first_contact_ext *fce = 1403 smc_get_clc_first_contact_ext(aclc, true); 1404 1405 ini->release_nr = fce->release; 1406 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1407 if (rc) 1408 return rc; 1409 } 1410 1411 rc = smc_v2_determine_accepted_chid(aclc, ini); 1412 if (rc) 1413 return rc; 1414 1415 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1416 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1417 ntohll(aclc->d1.gid_ext); 1418 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1419 } 1420 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1421 1422 /* there is only one lgr role for SMC-D; use server lock */ 1423 mutex_lock(&smc_server_lgr_pending); 1424 rc = smc_conn_create(smc, ini); 1425 if (rc) { 1426 mutex_unlock(&smc_server_lgr_pending); 1427 return rc; 1428 } 1429 1430 /* Create send and receive buffers */ 1431 rc = smc_buf_create(smc, true); 1432 if (rc) { 1433 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1434 goto connect_abort; 1435 } 1436 1437 smc_conn_save_peer_info(smc, aclc); 1438 smc_close_init(smc); 1439 smc_rx_init(smc); 1440 smc_tx_init(smc); 1441 1442 if (aclc->hdr.version > SMC_V1) 1443 eid = aclc->d1.eid; 1444 1445 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1446 aclc->hdr.version, eid, ini); 1447 if (rc) 1448 goto connect_abort; 1449 mutex_unlock(&smc_server_lgr_pending); 1450 1451 smc_copy_sock_settings_to_clc(smc); 1452 smc->connect_nonblock = 0; 1453 if (smc->sk.sk_state == SMC_INIT) 1454 smc->sk.sk_state = SMC_ACTIVE; 1455 1456 return 0; 1457 connect_abort: 1458 smc_conn_abort(smc, ini->first_contact_local); 1459 mutex_unlock(&smc_server_lgr_pending); 1460 smc->connect_nonblock = 0; 1461 1462 return rc; 1463 } 1464 1465 /* check if received accept type and version matches a proposed one */ 1466 static int smc_connect_check_aclc(struct smc_init_info *ini, 1467 struct smc_clc_msg_accept_confirm *aclc) 1468 { 1469 if (aclc->hdr.typev1 != SMC_TYPE_R && 1470 aclc->hdr.typev1 != SMC_TYPE_D) 1471 return SMC_CLC_DECL_MODEUNSUPP; 1472 1473 if (aclc->hdr.version >= SMC_V2) { 1474 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1475 !smcr_indicated(ini->smc_type_v2)) || 1476 (aclc->hdr.typev1 == SMC_TYPE_D && 1477 !smcd_indicated(ini->smc_type_v2))) 1478 return SMC_CLC_DECL_MODEUNSUPP; 1479 } else { 1480 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1481 !smcr_indicated(ini->smc_type_v1)) || 1482 (aclc->hdr.typev1 == SMC_TYPE_D && 1483 !smcd_indicated(ini->smc_type_v1))) 1484 return SMC_CLC_DECL_MODEUNSUPP; 1485 } 1486 1487 return 0; 1488 } 1489 1490 /* perform steps before actually connecting */ 1491 static int __smc_connect(struct smc_sock *smc) 1492 { 1493 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1494 struct smc_clc_msg_accept_confirm *aclc; 1495 struct smc_init_info *ini = NULL; 1496 u8 *buf = NULL; 1497 int rc = 0; 1498 1499 if (smc->use_fallback) 1500 return smc_connect_fallback(smc, smc->fallback_rsn); 1501 1502 /* if peer has not signalled SMC-capability, fall back */ 1503 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1504 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1505 1506 /* IPSec connections opt out of SMC optimizations */ 1507 if (using_ipsec(smc)) 1508 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1509 version); 1510 1511 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1512 if (!ini) 1513 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1514 version); 1515 1516 ini->smcd_version = SMC_V1 | SMC_V2; 1517 ini->smcr_version = SMC_V1 | SMC_V2; 1518 ini->smc_type_v1 = SMC_TYPE_B; 1519 ini->smc_type_v2 = SMC_TYPE_B; 1520 1521 /* get vlan id from IP device */ 1522 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1523 ini->smcd_version &= ~SMC_V1; 1524 ini->smcr_version = 0; 1525 ini->smc_type_v1 = SMC_TYPE_N; 1526 if (!ini->smcd_version) { 1527 rc = SMC_CLC_DECL_GETVLANERR; 1528 goto fallback; 1529 } 1530 } 1531 1532 rc = smc_find_proposal_devices(smc, ini); 1533 if (rc) 1534 goto fallback; 1535 1536 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1537 if (!buf) { 1538 rc = SMC_CLC_DECL_MEM; 1539 goto fallback; 1540 } 1541 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1542 1543 /* perform CLC handshake */ 1544 rc = smc_connect_clc(smc, aclc, ini); 1545 if (rc) { 1546 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1547 if (rc == -EAGAIN) { 1548 rc = -ETIMEDOUT; 1549 smc->sk.sk_err = ETIMEDOUT; 1550 } 1551 goto vlan_cleanup; 1552 } 1553 1554 /* check if smc modes and versions of CLC proposal and accept match */ 1555 rc = smc_connect_check_aclc(ini, aclc); 1556 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1557 if (rc) 1558 goto vlan_cleanup; 1559 1560 /* depending on previous steps, connect using rdma or ism */ 1561 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1562 ini->smcr_version = version; 1563 rc = smc_connect_rdma(smc, aclc, ini); 1564 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1565 ini->smcd_version = version; 1566 rc = smc_connect_ism(smc, aclc, ini); 1567 } 1568 if (rc) 1569 goto vlan_cleanup; 1570 1571 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1572 smc_connect_ism_vlan_cleanup(smc, ini); 1573 kfree(buf); 1574 kfree(ini); 1575 return 0; 1576 1577 vlan_cleanup: 1578 smc_connect_ism_vlan_cleanup(smc, ini); 1579 kfree(buf); 1580 fallback: 1581 kfree(ini); 1582 return smc_connect_decline_fallback(smc, rc, version); 1583 } 1584 1585 static void smc_connect_work(struct work_struct *work) 1586 { 1587 struct smc_sock *smc = container_of(work, struct smc_sock, 1588 connect_work); 1589 long timeo = smc->sk.sk_sndtimeo; 1590 int rc = 0; 1591 1592 if (!timeo) 1593 timeo = MAX_SCHEDULE_TIMEOUT; 1594 lock_sock(smc->clcsock->sk); 1595 if (smc->clcsock->sk->sk_err) { 1596 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1597 } else if ((1 << smc->clcsock->sk->sk_state) & 1598 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1599 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1600 if ((rc == -EPIPE) && 1601 ((1 << smc->clcsock->sk->sk_state) & 1602 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1603 rc = 0; 1604 } 1605 release_sock(smc->clcsock->sk); 1606 lock_sock(&smc->sk); 1607 if (rc != 0 || smc->sk.sk_err) { 1608 smc->sk.sk_state = SMC_CLOSED; 1609 if (rc == -EPIPE || rc == -EAGAIN) 1610 smc->sk.sk_err = EPIPE; 1611 else if (rc == -ECONNREFUSED) 1612 smc->sk.sk_err = ECONNREFUSED; 1613 else if (signal_pending(current)) 1614 smc->sk.sk_err = -sock_intr_errno(timeo); 1615 sock_put(&smc->sk); /* passive closing */ 1616 goto out; 1617 } 1618 1619 rc = __smc_connect(smc); 1620 if (rc < 0) 1621 smc->sk.sk_err = -rc; 1622 1623 out: 1624 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1625 if (smc->sk.sk_err) { 1626 smc->sk.sk_state_change(&smc->sk); 1627 } else { /* allow polling before and after fallback decision */ 1628 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1629 smc->sk.sk_write_space(&smc->sk); 1630 } 1631 } 1632 release_sock(&smc->sk); 1633 } 1634 1635 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1636 int alen, int flags) 1637 { 1638 struct sock *sk = sock->sk; 1639 struct smc_sock *smc; 1640 int rc = -EINVAL; 1641 1642 smc = smc_sk(sk); 1643 1644 /* separate smc parameter checking to be safe */ 1645 if (alen < sizeof(addr->sa_family)) 1646 goto out_err; 1647 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1648 goto out_err; 1649 1650 lock_sock(sk); 1651 switch (sock->state) { 1652 default: 1653 rc = -EINVAL; 1654 goto out; 1655 case SS_CONNECTED: 1656 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1657 goto out; 1658 case SS_CONNECTING: 1659 if (sk->sk_state == SMC_ACTIVE) 1660 goto connected; 1661 break; 1662 case SS_UNCONNECTED: 1663 sock->state = SS_CONNECTING; 1664 break; 1665 } 1666 1667 switch (sk->sk_state) { 1668 default: 1669 goto out; 1670 case SMC_CLOSED: 1671 rc = sock_error(sk) ? : -ECONNABORTED; 1672 sock->state = SS_UNCONNECTED; 1673 goto out; 1674 case SMC_ACTIVE: 1675 rc = -EISCONN; 1676 goto out; 1677 case SMC_INIT: 1678 break; 1679 } 1680 1681 smc_copy_sock_settings_to_clc(smc); 1682 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1683 if (smc->connect_nonblock) { 1684 rc = -EALREADY; 1685 goto out; 1686 } 1687 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1688 if (rc && rc != -EINPROGRESS) 1689 goto out; 1690 1691 if (smc->use_fallback) { 1692 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1693 goto out; 1694 } 1695 sock_hold(&smc->sk); /* sock put in passive closing */ 1696 if (flags & O_NONBLOCK) { 1697 if (queue_work(smc_hs_wq, &smc->connect_work)) 1698 smc->connect_nonblock = 1; 1699 rc = -EINPROGRESS; 1700 goto out; 1701 } else { 1702 rc = __smc_connect(smc); 1703 if (rc < 0) 1704 goto out; 1705 } 1706 1707 connected: 1708 rc = 0; 1709 sock->state = SS_CONNECTED; 1710 out: 1711 release_sock(sk); 1712 out_err: 1713 return rc; 1714 } 1715 1716 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1717 { 1718 struct socket *new_clcsock = NULL; 1719 struct sock *lsk = &lsmc->sk; 1720 struct sock *new_sk; 1721 int rc = -EINVAL; 1722 1723 release_sock(lsk); 1724 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1725 if (!new_sk) { 1726 rc = -ENOMEM; 1727 lsk->sk_err = ENOMEM; 1728 *new_smc = NULL; 1729 lock_sock(lsk); 1730 goto out; 1731 } 1732 *new_smc = smc_sk(new_sk); 1733 1734 mutex_lock(&lsmc->clcsock_release_lock); 1735 if (lsmc->clcsock) 1736 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1737 mutex_unlock(&lsmc->clcsock_release_lock); 1738 lock_sock(lsk); 1739 if (rc < 0 && rc != -EAGAIN) 1740 lsk->sk_err = -rc; 1741 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1742 new_sk->sk_prot->unhash(new_sk); 1743 if (new_clcsock) 1744 sock_release(new_clcsock); 1745 new_sk->sk_state = SMC_CLOSED; 1746 smc_sock_set_flag(new_sk, SOCK_DEAD); 1747 sock_put(new_sk); /* final */ 1748 *new_smc = NULL; 1749 goto out; 1750 } 1751 1752 /* new clcsock has inherited the smc listen-specific sk_data_ready 1753 * function; switch it back to the original sk_data_ready function 1754 */ 1755 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1756 1757 /* if new clcsock has also inherited the fallback-specific callback 1758 * functions, switch them back to the original ones. 1759 */ 1760 if (lsmc->use_fallback) { 1761 if (lsmc->clcsk_state_change) 1762 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1763 if (lsmc->clcsk_write_space) 1764 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1765 if (lsmc->clcsk_error_report) 1766 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1767 } 1768 1769 (*new_smc)->clcsock = new_clcsock; 1770 out: 1771 return rc; 1772 } 1773 1774 /* add a just created sock to the accept queue of the listen sock as 1775 * candidate for a following socket accept call from user space 1776 */ 1777 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1778 { 1779 struct smc_sock *par = smc_sk(parent); 1780 1781 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1782 spin_lock(&par->accept_q_lock); 1783 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1784 spin_unlock(&par->accept_q_lock); 1785 sk_acceptq_added(parent); 1786 } 1787 1788 /* remove a socket from the accept queue of its parental listening socket */ 1789 static void smc_accept_unlink(struct sock *sk) 1790 { 1791 struct smc_sock *par = smc_sk(sk)->listen_smc; 1792 1793 spin_lock(&par->accept_q_lock); 1794 list_del_init(&smc_sk(sk)->accept_q); 1795 spin_unlock(&par->accept_q_lock); 1796 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1797 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1798 } 1799 1800 /* remove a sock from the accept queue to bind it to a new socket created 1801 * for a socket accept call from user space 1802 */ 1803 struct sock *smc_accept_dequeue(struct sock *parent, 1804 struct socket *new_sock) 1805 { 1806 struct smc_sock *isk, *n; 1807 struct sock *new_sk; 1808 1809 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1810 new_sk = (struct sock *)isk; 1811 1812 smc_accept_unlink(new_sk); 1813 if (new_sk->sk_state == SMC_CLOSED) { 1814 new_sk->sk_prot->unhash(new_sk); 1815 if (isk->clcsock) { 1816 sock_release(isk->clcsock); 1817 isk->clcsock = NULL; 1818 } 1819 sock_put(new_sk); /* final */ 1820 continue; 1821 } 1822 if (new_sock) { 1823 sock_graft(new_sk, new_sock); 1824 new_sock->state = SS_CONNECTED; 1825 if (isk->use_fallback) { 1826 smc_sk(new_sk)->clcsock->file = new_sock->file; 1827 isk->clcsock->file->private_data = isk->clcsock; 1828 } 1829 } 1830 return new_sk; 1831 } 1832 return NULL; 1833 } 1834 1835 /* clean up for a created but never accepted sock */ 1836 void smc_close_non_accepted(struct sock *sk) 1837 { 1838 struct smc_sock *smc = smc_sk(sk); 1839 1840 sock_hold(sk); /* sock_put below */ 1841 lock_sock(sk); 1842 if (!sk->sk_lingertime) 1843 /* wait for peer closing */ 1844 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1845 __smc_release(smc); 1846 release_sock(sk); 1847 sock_put(sk); /* sock_hold above */ 1848 sock_put(sk); /* final sock_put */ 1849 } 1850 1851 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1852 { 1853 struct smc_link *link = smc->conn.lnk; 1854 struct smc_llc_qentry *qentry; 1855 int rc; 1856 1857 /* reg the sndbuf if it was vzalloced*/ 1858 if (smc->conn.sndbuf_desc->is_vm) { 1859 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1860 return SMC_CLC_DECL_ERR_REGBUF; 1861 } 1862 1863 /* reg the rmb */ 1864 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1865 return SMC_CLC_DECL_ERR_REGBUF; 1866 1867 /* send CONFIRM LINK request to client over the RoCE fabric */ 1868 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1869 if (rc < 0) 1870 return SMC_CLC_DECL_TIMEOUT_CL; 1871 1872 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1873 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1874 SMC_LLC_CONFIRM_LINK); 1875 if (!qentry) { 1876 struct smc_clc_msg_decline dclc; 1877 1878 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1879 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1880 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1881 } 1882 smc_llc_save_peer_uid(qentry); 1883 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1884 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1885 if (rc) 1886 return SMC_CLC_DECL_RMBE_EC; 1887 1888 /* confirm_rkey is implicit on 1st contact */ 1889 smc->conn.rmb_desc->is_conf_rkey = true; 1890 1891 smc_llc_link_active(link); 1892 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1893 1894 if (link->lgr->max_links > 1) { 1895 down_write(&link->lgr->llc_conf_mutex); 1896 /* initial contact - try to establish second link */ 1897 smc_llc_srv_add_link(link, NULL); 1898 up_write(&link->lgr->llc_conf_mutex); 1899 } 1900 return 0; 1901 } 1902 1903 /* listen worker: finish */ 1904 static void smc_listen_out(struct smc_sock *new_smc) 1905 { 1906 struct smc_sock *lsmc = new_smc->listen_smc; 1907 struct sock *newsmcsk = &new_smc->sk; 1908 1909 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1910 atomic_dec(&lsmc->queued_smc_hs); 1911 1912 if (lsmc->sk.sk_state == SMC_LISTEN) { 1913 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1914 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1915 release_sock(&lsmc->sk); 1916 } else { /* no longer listening */ 1917 smc_close_non_accepted(newsmcsk); 1918 } 1919 1920 /* Wake up accept */ 1921 lsmc->sk.sk_data_ready(&lsmc->sk); 1922 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1923 } 1924 1925 /* listen worker: finish in state connected */ 1926 static void smc_listen_out_connected(struct smc_sock *new_smc) 1927 { 1928 struct sock *newsmcsk = &new_smc->sk; 1929 1930 if (newsmcsk->sk_state == SMC_INIT) 1931 newsmcsk->sk_state = SMC_ACTIVE; 1932 1933 smc_listen_out(new_smc); 1934 } 1935 1936 /* listen worker: finish in error state */ 1937 static void smc_listen_out_err(struct smc_sock *new_smc) 1938 { 1939 struct sock *newsmcsk = &new_smc->sk; 1940 struct net *net = sock_net(newsmcsk); 1941 1942 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1943 if (newsmcsk->sk_state == SMC_INIT) 1944 sock_put(&new_smc->sk); /* passive closing */ 1945 newsmcsk->sk_state = SMC_CLOSED; 1946 1947 smc_listen_out(new_smc); 1948 } 1949 1950 /* listen worker: decline and fall back if possible */ 1951 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1952 int local_first, u8 version) 1953 { 1954 /* RDMA setup failed, switch back to TCP */ 1955 smc_conn_abort(new_smc, local_first); 1956 if (reason_code < 0 || 1957 smc_switch_to_fallback(new_smc, reason_code)) { 1958 /* error, no fallback possible */ 1959 smc_listen_out_err(new_smc); 1960 return; 1961 } 1962 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1963 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1964 smc_listen_out_err(new_smc); 1965 return; 1966 } 1967 } 1968 smc_listen_out_connected(new_smc); 1969 } 1970 1971 /* listen worker: version checking */ 1972 static int smc_listen_v2_check(struct smc_sock *new_smc, 1973 struct smc_clc_msg_proposal *pclc, 1974 struct smc_init_info *ini) 1975 { 1976 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1977 struct smc_clc_v2_extension *pclc_v2_ext; 1978 int rc = SMC_CLC_DECL_PEERNOSMC; 1979 1980 ini->smc_type_v1 = pclc->hdr.typev1; 1981 ini->smc_type_v2 = pclc->hdr.typev2; 1982 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1983 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1984 if (pclc->hdr.version > SMC_V1) { 1985 if (smcd_indicated(ini->smc_type_v2)) 1986 ini->smcd_version |= SMC_V2; 1987 if (smcr_indicated(ini->smc_type_v2)) 1988 ini->smcr_version |= SMC_V2; 1989 } 1990 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1991 rc = SMC_CLC_DECL_PEERNOSMC; 1992 goto out; 1993 } 1994 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1995 if (!pclc_v2_ext) { 1996 ini->smcd_version &= ~SMC_V2; 1997 ini->smcr_version &= ~SMC_V2; 1998 rc = SMC_CLC_DECL_NOV2EXT; 1999 goto out; 2000 } 2001 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2002 if (ini->smcd_version & SMC_V2) { 2003 if (!smc_ism_is_v2_capable()) { 2004 ini->smcd_version &= ~SMC_V2; 2005 rc = SMC_CLC_DECL_NOISM2SUPP; 2006 } else if (!pclc_smcd_v2_ext) { 2007 ini->smcd_version &= ~SMC_V2; 2008 rc = SMC_CLC_DECL_NOV2DEXT; 2009 } else if (!pclc_v2_ext->hdr.eid_cnt && 2010 !pclc_v2_ext->hdr.flag.seid) { 2011 ini->smcd_version &= ~SMC_V2; 2012 rc = SMC_CLC_DECL_NOUEID; 2013 } 2014 } 2015 if (ini->smcr_version & SMC_V2) { 2016 if (!pclc_v2_ext->hdr.eid_cnt) { 2017 ini->smcr_version &= ~SMC_V2; 2018 rc = SMC_CLC_DECL_NOUEID; 2019 } 2020 } 2021 2022 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2023 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2024 ini->release_nr = SMC_RELEASE; 2025 2026 out: 2027 if (!ini->smcd_version && !ini->smcr_version) 2028 return rc; 2029 2030 return 0; 2031 } 2032 2033 /* listen worker: check prefixes */ 2034 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2035 struct smc_clc_msg_proposal *pclc) 2036 { 2037 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2038 struct socket *newclcsock = new_smc->clcsock; 2039 2040 if (pclc->hdr.typev1 == SMC_TYPE_N) 2041 return 0; 2042 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2043 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2044 return SMC_CLC_DECL_DIFFPREFIX; 2045 2046 return 0; 2047 } 2048 2049 /* listen worker: initialize connection and buffers */ 2050 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2051 struct smc_init_info *ini) 2052 { 2053 int rc; 2054 2055 /* allocate connection / link group */ 2056 rc = smc_conn_create(new_smc, ini); 2057 if (rc) 2058 return rc; 2059 2060 /* create send buffer and rmb */ 2061 if (smc_buf_create(new_smc, false)) { 2062 smc_conn_abort(new_smc, ini->first_contact_local); 2063 return SMC_CLC_DECL_MEM; 2064 } 2065 2066 return 0; 2067 } 2068 2069 /* listen worker: initialize connection and buffers for SMC-D */ 2070 static int smc_listen_ism_init(struct smc_sock *new_smc, 2071 struct smc_init_info *ini) 2072 { 2073 int rc; 2074 2075 rc = smc_conn_create(new_smc, ini); 2076 if (rc) 2077 return rc; 2078 2079 /* Create send and receive buffers */ 2080 rc = smc_buf_create(new_smc, true); 2081 if (rc) { 2082 smc_conn_abort(new_smc, ini->first_contact_local); 2083 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2084 SMC_CLC_DECL_MEM; 2085 } 2086 2087 return 0; 2088 } 2089 2090 static bool smc_is_already_selected(struct smcd_dev *smcd, 2091 struct smc_init_info *ini, 2092 int matches) 2093 { 2094 int i; 2095 2096 for (i = 0; i < matches; i++) 2097 if (smcd == ini->ism_dev[i]) 2098 return true; 2099 2100 return false; 2101 } 2102 2103 /* check for ISM devices matching proposed ISM devices */ 2104 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2105 u16 proposed_chid, 2106 struct smcd_gid *proposed_gid, 2107 unsigned int *matches) 2108 { 2109 struct smcd_dev *smcd; 2110 2111 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2112 if (smcd->going_away) 2113 continue; 2114 if (smc_is_already_selected(smcd, ini, *matches)) 2115 continue; 2116 if (smc_ism_get_chid(smcd) == proposed_chid && 2117 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2118 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2119 if (__smc_ism_is_emulated(proposed_chid)) 2120 ini->ism_peer_gid[*matches].gid_ext = 2121 proposed_gid->gid_ext; 2122 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2123 ini->ism_dev[*matches] = smcd; 2124 (*matches)++; 2125 break; 2126 } 2127 } 2128 } 2129 2130 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2131 { 2132 if (!ini->rc) 2133 ini->rc = rc; 2134 } 2135 2136 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2137 struct smc_clc_msg_proposal *pclc, 2138 struct smc_init_info *ini) 2139 { 2140 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2141 struct smc_clc_v2_extension *smc_v2_ext; 2142 struct smc_clc_msg_smcd *pclc_smcd; 2143 unsigned int matches = 0; 2144 struct smcd_gid smcd_gid; 2145 u8 smcd_version; 2146 u8 *eid = NULL; 2147 int i, rc; 2148 u16 chid; 2149 2150 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2151 goto not_found; 2152 2153 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2154 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2155 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2156 2157 mutex_lock(&smcd_dev_list.mutex); 2158 if (pclc_smcd->ism.chid) { 2159 /* check for ISM device matching proposed native ISM device */ 2160 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2161 smcd_gid.gid_ext = 0; 2162 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2163 &smcd_gid, &matches); 2164 } 2165 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2166 /* check for ISM devices matching proposed non-native ISM 2167 * devices 2168 */ 2169 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2170 smcd_gid.gid_ext = 0; 2171 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2172 if (__smc_ism_is_emulated(chid)) { 2173 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2174 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2175 /* each Emulated-ISM device takes two GID-CHID 2176 * entries and CHID of the second entry repeats 2177 * that of the first entry. 2178 * 2179 * So check if the next GID-CHID entry exists 2180 * and both two entries' CHIDs are the same. 2181 */ 2182 continue; 2183 smcd_gid.gid_ext = 2184 ntohll(smcd_v2_ext->gidchid[++i].gid); 2185 } 2186 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2187 } 2188 mutex_unlock(&smcd_dev_list.mutex); 2189 2190 if (!ini->ism_dev[0]) { 2191 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2192 goto not_found; 2193 } 2194 2195 smc_ism_get_system_eid(&eid); 2196 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2197 smcd_v2_ext->system_eid, eid)) 2198 goto not_found; 2199 2200 /* separate - outside the smcd_dev_list.lock */ 2201 smcd_version = ini->smcd_version; 2202 for (i = 0; i < matches; i++) { 2203 ini->smcd_version = SMC_V2; 2204 ini->is_smcd = true; 2205 ini->ism_selected = i; 2206 rc = smc_listen_ism_init(new_smc, ini); 2207 if (rc) { 2208 smc_find_ism_store_rc(rc, ini); 2209 /* try next active ISM device */ 2210 continue; 2211 } 2212 return; /* matching and usable V2 ISM device found */ 2213 } 2214 /* no V2 ISM device could be initialized */ 2215 ini->smcd_version = smcd_version; /* restore original value */ 2216 ini->negotiated_eid[0] = 0; 2217 2218 not_found: 2219 ini->smcd_version &= ~SMC_V2; 2220 ini->ism_dev[0] = NULL; 2221 ini->is_smcd = false; 2222 } 2223 2224 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2225 struct smc_clc_msg_proposal *pclc, 2226 struct smc_init_info *ini) 2227 { 2228 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2229 int rc = 0; 2230 2231 /* check if ISM V1 is available */ 2232 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2233 goto not_found; 2234 ini->is_smcd = true; /* prepare ISM check */ 2235 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2236 ini->ism_peer_gid[0].gid_ext = 0; 2237 rc = smc_find_ism_device(new_smc, ini); 2238 if (rc) 2239 goto not_found; 2240 ini->ism_selected = 0; 2241 rc = smc_listen_ism_init(new_smc, ini); 2242 if (!rc) 2243 return; /* V1 ISM device found */ 2244 2245 not_found: 2246 smc_find_ism_store_rc(rc, ini); 2247 ini->smcd_version &= ~SMC_V1; 2248 ini->ism_dev[0] = NULL; 2249 ini->is_smcd = false; 2250 } 2251 2252 /* listen worker: register buffers */ 2253 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2254 { 2255 struct smc_connection *conn = &new_smc->conn; 2256 2257 if (!local_first) { 2258 /* reg sendbufs if they were vzalloced */ 2259 if (conn->sndbuf_desc->is_vm) { 2260 if (smcr_lgr_reg_sndbufs(conn->lnk, 2261 conn->sndbuf_desc)) 2262 return SMC_CLC_DECL_ERR_REGBUF; 2263 } 2264 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2265 return SMC_CLC_DECL_ERR_REGBUF; 2266 } 2267 2268 return 0; 2269 } 2270 2271 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2272 struct smc_clc_msg_proposal *pclc, 2273 struct smc_init_info *ini) 2274 { 2275 struct smc_clc_v2_extension *smc_v2_ext; 2276 u8 smcr_version; 2277 int rc; 2278 2279 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2280 goto not_found; 2281 2282 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2283 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2284 goto not_found; 2285 2286 /* prepare RDMA check */ 2287 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2288 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2289 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2290 ini->check_smcrv2 = true; 2291 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2292 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2293 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2294 rc = smc_find_rdma_device(new_smc, ini); 2295 if (rc) { 2296 smc_find_ism_store_rc(rc, ini); 2297 goto not_found; 2298 } 2299 if (!ini->smcrv2.uses_gateway) 2300 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2301 2302 smcr_version = ini->smcr_version; 2303 ini->smcr_version = SMC_V2; 2304 rc = smc_listen_rdma_init(new_smc, ini); 2305 if (!rc) { 2306 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2307 if (rc) 2308 smc_conn_abort(new_smc, ini->first_contact_local); 2309 } 2310 if (!rc) 2311 return; 2312 ini->smcr_version = smcr_version; 2313 smc_find_ism_store_rc(rc, ini); 2314 2315 not_found: 2316 ini->smcr_version &= ~SMC_V2; 2317 ini->smcrv2.ib_dev_v2 = NULL; 2318 ini->check_smcrv2 = false; 2319 } 2320 2321 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2322 struct smc_clc_msg_proposal *pclc, 2323 struct smc_init_info *ini) 2324 { 2325 int rc; 2326 2327 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2328 return SMC_CLC_DECL_NOSMCDEV; 2329 2330 /* prepare RDMA check */ 2331 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2332 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2333 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2334 rc = smc_find_rdma_device(new_smc, ini); 2335 if (rc) { 2336 /* no RDMA device found */ 2337 return SMC_CLC_DECL_NOSMCDEV; 2338 } 2339 rc = smc_listen_rdma_init(new_smc, ini); 2340 if (rc) 2341 return rc; 2342 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2343 } 2344 2345 /* determine the local device matching to proposal */ 2346 static int smc_listen_find_device(struct smc_sock *new_smc, 2347 struct smc_clc_msg_proposal *pclc, 2348 struct smc_init_info *ini) 2349 { 2350 int prfx_rc; 2351 2352 /* check for ISM device matching V2 proposed device */ 2353 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2354 if (ini->ism_dev[0]) 2355 return 0; 2356 2357 /* check for matching IP prefix and subnet length (V1) */ 2358 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2359 if (prfx_rc) 2360 smc_find_ism_store_rc(prfx_rc, ini); 2361 2362 /* get vlan id from IP device */ 2363 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2364 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2365 2366 /* check for ISM device matching V1 proposed device */ 2367 if (!prfx_rc) 2368 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2369 if (ini->ism_dev[0]) 2370 return 0; 2371 2372 if (!smcr_indicated(pclc->hdr.typev1) && 2373 !smcr_indicated(pclc->hdr.typev2)) 2374 /* skip RDMA and decline */ 2375 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2376 2377 /* check if RDMA V2 is available */ 2378 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2379 if (ini->smcrv2.ib_dev_v2) 2380 return 0; 2381 2382 /* check if RDMA V1 is available */ 2383 if (!prfx_rc) { 2384 int rc; 2385 2386 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2387 smc_find_ism_store_rc(rc, ini); 2388 return (!rc) ? 0 : ini->rc; 2389 } 2390 return prfx_rc; 2391 } 2392 2393 /* listen worker: finish RDMA setup */ 2394 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2395 struct smc_clc_msg_accept_confirm *cclc, 2396 bool local_first, 2397 struct smc_init_info *ini) 2398 { 2399 struct smc_link *link = new_smc->conn.lnk; 2400 int reason_code = 0; 2401 2402 if (local_first) 2403 smc_link_save_peer_info(link, cclc, ini); 2404 2405 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2406 return SMC_CLC_DECL_ERR_RTOK; 2407 2408 if (local_first) { 2409 if (smc_ib_ready_link(link)) 2410 return SMC_CLC_DECL_ERR_RDYLNK; 2411 /* QP confirmation over RoCE fabric */ 2412 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2413 reason_code = smcr_serv_conf_first_link(new_smc); 2414 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2415 } 2416 return reason_code; 2417 } 2418 2419 /* setup for connection of server */ 2420 static void smc_listen_work(struct work_struct *work) 2421 { 2422 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2423 smc_listen_work); 2424 struct socket *newclcsock = new_smc->clcsock; 2425 struct smc_clc_msg_accept_confirm *cclc; 2426 struct smc_clc_msg_proposal_area *buf; 2427 struct smc_clc_msg_proposal *pclc; 2428 struct smc_init_info *ini = NULL; 2429 u8 proposal_version = SMC_V1; 2430 u8 accept_version; 2431 int rc = 0; 2432 2433 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2434 return smc_listen_out_err(new_smc); 2435 2436 if (new_smc->use_fallback) { 2437 smc_listen_out_connected(new_smc); 2438 return; 2439 } 2440 2441 /* check if peer is smc capable */ 2442 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2443 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2444 if (rc) 2445 smc_listen_out_err(new_smc); 2446 else 2447 smc_listen_out_connected(new_smc); 2448 return; 2449 } 2450 2451 /* do inband token exchange - 2452 * wait for and receive SMC Proposal CLC message 2453 */ 2454 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2455 if (!buf) { 2456 rc = SMC_CLC_DECL_MEM; 2457 goto out_decl; 2458 } 2459 pclc = (struct smc_clc_msg_proposal *)buf; 2460 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2461 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2462 if (rc) 2463 goto out_decl; 2464 2465 if (pclc->hdr.version > SMC_V1) 2466 proposal_version = SMC_V2; 2467 2468 /* IPSec connections opt out of SMC optimizations */ 2469 if (using_ipsec(new_smc)) { 2470 rc = SMC_CLC_DECL_IPSEC; 2471 goto out_decl; 2472 } 2473 2474 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2475 if (!ini) { 2476 rc = SMC_CLC_DECL_MEM; 2477 goto out_decl; 2478 } 2479 2480 /* initial version checking */ 2481 rc = smc_listen_v2_check(new_smc, pclc, ini); 2482 if (rc) 2483 goto out_decl; 2484 2485 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2486 if (rc) 2487 goto out_decl; 2488 2489 mutex_lock(&smc_server_lgr_pending); 2490 smc_close_init(new_smc); 2491 smc_rx_init(new_smc); 2492 smc_tx_init(new_smc); 2493 2494 /* determine ISM or RoCE device used for connection */ 2495 rc = smc_listen_find_device(new_smc, pclc, ini); 2496 if (rc) 2497 goto out_unlock; 2498 2499 /* send SMC Accept CLC message */ 2500 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2501 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2502 accept_version, ini->negotiated_eid, ini); 2503 if (rc) 2504 goto out_unlock; 2505 2506 /* SMC-D does not need this lock any more */ 2507 if (ini->is_smcd) 2508 mutex_unlock(&smc_server_lgr_pending); 2509 2510 /* receive SMC Confirm CLC message */ 2511 memset(buf, 0, sizeof(*buf)); 2512 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2513 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2514 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2515 if (rc) { 2516 if (!ini->is_smcd) 2517 goto out_unlock; 2518 goto out_decl; 2519 } 2520 2521 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2522 if (rc) { 2523 if (!ini->is_smcd) 2524 goto out_unlock; 2525 goto out_decl; 2526 } 2527 2528 /* fce smc release version is needed in smc_listen_rdma_finish, 2529 * so save fce info here. 2530 */ 2531 smc_conn_save_peer_info_fce(new_smc, cclc); 2532 2533 /* finish worker */ 2534 if (!ini->is_smcd) { 2535 rc = smc_listen_rdma_finish(new_smc, cclc, 2536 ini->first_contact_local, ini); 2537 if (rc) 2538 goto out_unlock; 2539 mutex_unlock(&smc_server_lgr_pending); 2540 } 2541 smc_conn_save_peer_info(new_smc, cclc); 2542 smc_listen_out_connected(new_smc); 2543 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2544 goto out_free; 2545 2546 out_unlock: 2547 mutex_unlock(&smc_server_lgr_pending); 2548 out_decl: 2549 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2550 proposal_version); 2551 out_free: 2552 kfree(ini); 2553 kfree(buf); 2554 } 2555 2556 static void smc_tcp_listen_work(struct work_struct *work) 2557 { 2558 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2559 tcp_listen_work); 2560 struct sock *lsk = &lsmc->sk; 2561 struct smc_sock *new_smc; 2562 int rc = 0; 2563 2564 lock_sock(lsk); 2565 while (lsk->sk_state == SMC_LISTEN) { 2566 rc = smc_clcsock_accept(lsmc, &new_smc); 2567 if (rc) /* clcsock accept queue empty or error */ 2568 goto out; 2569 if (!new_smc) 2570 continue; 2571 2572 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2573 atomic_inc(&lsmc->queued_smc_hs); 2574 2575 new_smc->listen_smc = lsmc; 2576 new_smc->use_fallback = lsmc->use_fallback; 2577 new_smc->fallback_rsn = lsmc->fallback_rsn; 2578 sock_hold(lsk); /* sock_put in smc_listen_work */ 2579 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2580 smc_copy_sock_settings_to_smc(new_smc); 2581 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2582 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2583 sock_put(&new_smc->sk); 2584 } 2585 2586 out: 2587 release_sock(lsk); 2588 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2589 } 2590 2591 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2592 { 2593 struct smc_sock *lsmc; 2594 2595 read_lock_bh(&listen_clcsock->sk_callback_lock); 2596 lsmc = smc_clcsock_user_data(listen_clcsock); 2597 if (!lsmc) 2598 goto out; 2599 lsmc->clcsk_data_ready(listen_clcsock); 2600 if (lsmc->sk.sk_state == SMC_LISTEN) { 2601 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2602 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2603 sock_put(&lsmc->sk); 2604 } 2605 out: 2606 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2607 } 2608 2609 static int smc_listen(struct socket *sock, int backlog) 2610 { 2611 struct sock *sk = sock->sk; 2612 struct smc_sock *smc; 2613 int rc; 2614 2615 smc = smc_sk(sk); 2616 lock_sock(sk); 2617 2618 rc = -EINVAL; 2619 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2620 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2621 goto out; 2622 2623 rc = 0; 2624 if (sk->sk_state == SMC_LISTEN) { 2625 sk->sk_max_ack_backlog = backlog; 2626 goto out; 2627 } 2628 /* some socket options are handled in core, so we could not apply 2629 * them to the clc socket -- copy smc socket options to clc socket 2630 */ 2631 smc_copy_sock_settings_to_clc(smc); 2632 if (!smc->use_fallback) 2633 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2634 2635 /* save original sk_data_ready function and establish 2636 * smc-specific sk_data_ready function 2637 */ 2638 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2639 smc->clcsock->sk->sk_user_data = 2640 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2641 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2642 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2643 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2644 2645 /* save original ops */ 2646 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2647 2648 smc->af_ops = *smc->ori_af_ops; 2649 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2650 2651 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2652 2653 if (smc->limit_smc_hs) 2654 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2655 2656 rc = kernel_listen(smc->clcsock, backlog); 2657 if (rc) { 2658 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2659 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2660 &smc->clcsk_data_ready); 2661 smc->clcsock->sk->sk_user_data = NULL; 2662 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2663 goto out; 2664 } 2665 sk->sk_max_ack_backlog = backlog; 2666 sk->sk_ack_backlog = 0; 2667 sk->sk_state = SMC_LISTEN; 2668 2669 out: 2670 release_sock(sk); 2671 return rc; 2672 } 2673 2674 static int smc_accept(struct socket *sock, struct socket *new_sock, 2675 int flags, bool kern) 2676 { 2677 struct sock *sk = sock->sk, *nsk; 2678 DECLARE_WAITQUEUE(wait, current); 2679 struct smc_sock *lsmc; 2680 long timeo; 2681 int rc = 0; 2682 2683 lsmc = smc_sk(sk); 2684 sock_hold(sk); /* sock_put below */ 2685 lock_sock(sk); 2686 2687 if (lsmc->sk.sk_state != SMC_LISTEN) { 2688 rc = -EINVAL; 2689 release_sock(sk); 2690 goto out; 2691 } 2692 2693 /* Wait for an incoming connection */ 2694 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2695 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2696 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2697 set_current_state(TASK_INTERRUPTIBLE); 2698 if (!timeo) { 2699 rc = -EAGAIN; 2700 break; 2701 } 2702 release_sock(sk); 2703 timeo = schedule_timeout(timeo); 2704 /* wakeup by sk_data_ready in smc_listen_work() */ 2705 sched_annotate_sleep(); 2706 lock_sock(sk); 2707 if (signal_pending(current)) { 2708 rc = sock_intr_errno(timeo); 2709 break; 2710 } 2711 } 2712 set_current_state(TASK_RUNNING); 2713 remove_wait_queue(sk_sleep(sk), &wait); 2714 2715 if (!rc) 2716 rc = sock_error(nsk); 2717 release_sock(sk); 2718 if (rc) 2719 goto out; 2720 2721 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2722 /* wait till data arrives on the socket */ 2723 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2724 MSEC_PER_SEC); 2725 if (smc_sk(nsk)->use_fallback) { 2726 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2727 2728 lock_sock(clcsk); 2729 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2730 sk_wait_data(clcsk, &timeo, NULL); 2731 release_sock(clcsk); 2732 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2733 lock_sock(nsk); 2734 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2735 release_sock(nsk); 2736 } 2737 } 2738 2739 out: 2740 sock_put(sk); /* sock_hold above */ 2741 return rc; 2742 } 2743 2744 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2745 int peer) 2746 { 2747 struct smc_sock *smc; 2748 2749 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2750 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2751 return -ENOTCONN; 2752 2753 smc = smc_sk(sock->sk); 2754 2755 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2756 } 2757 2758 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2759 { 2760 struct sock *sk = sock->sk; 2761 struct smc_sock *smc; 2762 int rc; 2763 2764 smc = smc_sk(sk); 2765 lock_sock(sk); 2766 2767 /* SMC does not support connect with fastopen */ 2768 if (msg->msg_flags & MSG_FASTOPEN) { 2769 /* not connected yet, fallback */ 2770 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2771 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2772 if (rc) 2773 goto out; 2774 } else { 2775 rc = -EINVAL; 2776 goto out; 2777 } 2778 } else if ((sk->sk_state != SMC_ACTIVE) && 2779 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2780 (sk->sk_state != SMC_INIT)) { 2781 rc = -EPIPE; 2782 goto out; 2783 } 2784 2785 if (smc->use_fallback) { 2786 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2787 } else { 2788 rc = smc_tx_sendmsg(smc, msg, len); 2789 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2790 } 2791 out: 2792 release_sock(sk); 2793 return rc; 2794 } 2795 2796 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2797 int flags) 2798 { 2799 struct sock *sk = sock->sk; 2800 struct smc_sock *smc; 2801 int rc = -ENOTCONN; 2802 2803 smc = smc_sk(sk); 2804 lock_sock(sk); 2805 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2806 /* socket was connected before, no more data to read */ 2807 rc = 0; 2808 goto out; 2809 } 2810 if ((sk->sk_state == SMC_INIT) || 2811 (sk->sk_state == SMC_LISTEN) || 2812 (sk->sk_state == SMC_CLOSED)) 2813 goto out; 2814 2815 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2816 rc = 0; 2817 goto out; 2818 } 2819 2820 if (smc->use_fallback) { 2821 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2822 } else { 2823 msg->msg_namelen = 0; 2824 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2825 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2826 } 2827 2828 out: 2829 release_sock(sk); 2830 return rc; 2831 } 2832 2833 static __poll_t smc_accept_poll(struct sock *parent) 2834 { 2835 struct smc_sock *isk = smc_sk(parent); 2836 __poll_t mask = 0; 2837 2838 spin_lock(&isk->accept_q_lock); 2839 if (!list_empty(&isk->accept_q)) 2840 mask = EPOLLIN | EPOLLRDNORM; 2841 spin_unlock(&isk->accept_q_lock); 2842 2843 return mask; 2844 } 2845 2846 static __poll_t smc_poll(struct file *file, struct socket *sock, 2847 poll_table *wait) 2848 { 2849 struct sock *sk = sock->sk; 2850 struct smc_sock *smc; 2851 __poll_t mask = 0; 2852 2853 if (!sk) 2854 return EPOLLNVAL; 2855 2856 smc = smc_sk(sock->sk); 2857 if (smc->use_fallback) { 2858 /* delegate to CLC child sock */ 2859 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2860 sk->sk_err = smc->clcsock->sk->sk_err; 2861 } else { 2862 if (sk->sk_state != SMC_CLOSED) 2863 sock_poll_wait(file, sock, wait); 2864 if (sk->sk_err) 2865 mask |= EPOLLERR; 2866 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2867 (sk->sk_state == SMC_CLOSED)) 2868 mask |= EPOLLHUP; 2869 if (sk->sk_state == SMC_LISTEN) { 2870 /* woken up by sk_data_ready in smc_listen_work() */ 2871 mask |= smc_accept_poll(sk); 2872 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2873 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2874 wait); 2875 sk->sk_err = smc->clcsock->sk->sk_err; 2876 } else { 2877 if ((sk->sk_state != SMC_INIT && 2878 atomic_read(&smc->conn.sndbuf_space)) || 2879 sk->sk_shutdown & SEND_SHUTDOWN) { 2880 mask |= EPOLLOUT | EPOLLWRNORM; 2881 } else { 2882 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2883 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2884 } 2885 if (atomic_read(&smc->conn.bytes_to_rcv)) 2886 mask |= EPOLLIN | EPOLLRDNORM; 2887 if (sk->sk_shutdown & RCV_SHUTDOWN) 2888 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2889 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2890 mask |= EPOLLIN; 2891 if (smc->conn.urg_state == SMC_URG_VALID) 2892 mask |= EPOLLPRI; 2893 } 2894 } 2895 2896 return mask; 2897 } 2898 2899 static int smc_shutdown(struct socket *sock, int how) 2900 { 2901 struct sock *sk = sock->sk; 2902 bool do_shutdown = true; 2903 struct smc_sock *smc; 2904 int rc = -EINVAL; 2905 int old_state; 2906 int rc1 = 0; 2907 2908 smc = smc_sk(sk); 2909 2910 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2911 return rc; 2912 2913 lock_sock(sk); 2914 2915 if (sock->state == SS_CONNECTING) { 2916 if (sk->sk_state == SMC_ACTIVE) 2917 sock->state = SS_CONNECTED; 2918 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2919 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2920 sk->sk_state == SMC_APPCLOSEWAIT1 || 2921 sk->sk_state == SMC_APPCLOSEWAIT2 || 2922 sk->sk_state == SMC_APPFINCLOSEWAIT) 2923 sock->state = SS_DISCONNECTING; 2924 } 2925 2926 rc = -ENOTCONN; 2927 if ((sk->sk_state != SMC_ACTIVE) && 2928 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2929 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2930 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2931 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2932 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2933 goto out; 2934 if (smc->use_fallback) { 2935 rc = kernel_sock_shutdown(smc->clcsock, how); 2936 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2937 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2938 sk->sk_state = SMC_CLOSED; 2939 sk->sk_socket->state = SS_UNCONNECTED; 2940 sock_put(sk); 2941 } 2942 goto out; 2943 } 2944 switch (how) { 2945 case SHUT_RDWR: /* shutdown in both directions */ 2946 old_state = sk->sk_state; 2947 rc = smc_close_active(smc); 2948 if (old_state == SMC_ACTIVE && 2949 sk->sk_state == SMC_PEERCLOSEWAIT1) 2950 do_shutdown = false; 2951 break; 2952 case SHUT_WR: 2953 rc = smc_close_shutdown_write(smc); 2954 break; 2955 case SHUT_RD: 2956 rc = 0; 2957 /* nothing more to do because peer is not involved */ 2958 break; 2959 } 2960 if (do_shutdown && smc->clcsock) 2961 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2962 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2963 sk->sk_shutdown |= how + 1; 2964 2965 if (sk->sk_state == SMC_CLOSED) 2966 sock->state = SS_UNCONNECTED; 2967 else 2968 sock->state = SS_DISCONNECTING; 2969 out: 2970 release_sock(sk); 2971 return rc ? rc : rc1; 2972 } 2973 2974 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2975 char __user *optval, int __user *optlen) 2976 { 2977 struct smc_sock *smc; 2978 int val, len; 2979 2980 smc = smc_sk(sock->sk); 2981 2982 if (get_user(len, optlen)) 2983 return -EFAULT; 2984 2985 len = min_t(int, len, sizeof(int)); 2986 2987 if (len < 0) 2988 return -EINVAL; 2989 2990 switch (optname) { 2991 case SMC_LIMIT_HS: 2992 val = smc->limit_smc_hs; 2993 break; 2994 default: 2995 return -EOPNOTSUPP; 2996 } 2997 2998 if (put_user(len, optlen)) 2999 return -EFAULT; 3000 if (copy_to_user(optval, &val, len)) 3001 return -EFAULT; 3002 3003 return 0; 3004 } 3005 3006 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3007 sockptr_t optval, unsigned int optlen) 3008 { 3009 struct sock *sk = sock->sk; 3010 struct smc_sock *smc; 3011 int val, rc; 3012 3013 smc = smc_sk(sk); 3014 3015 lock_sock(sk); 3016 switch (optname) { 3017 case SMC_LIMIT_HS: 3018 if (optlen < sizeof(int)) { 3019 rc = -EINVAL; 3020 break; 3021 } 3022 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3023 rc = -EFAULT; 3024 break; 3025 } 3026 3027 smc->limit_smc_hs = !!val; 3028 rc = 0; 3029 break; 3030 default: 3031 rc = -EOPNOTSUPP; 3032 break; 3033 } 3034 release_sock(sk); 3035 3036 return rc; 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 if (level == SOL_TCP && optname == TCP_ULP) 3047 return -EOPNOTSUPP; 3048 else if (level == SOL_SMC) 3049 return __smc_setsockopt(sock, level, optname, optval, optlen); 3050 3051 smc = smc_sk(sk); 3052 3053 /* generic setsockopts reaching us here always apply to the 3054 * CLC socket 3055 */ 3056 mutex_lock(&smc->clcsock_release_lock); 3057 if (!smc->clcsock) { 3058 mutex_unlock(&smc->clcsock_release_lock); 3059 return -EBADF; 3060 } 3061 if (unlikely(!smc->clcsock->ops->setsockopt)) 3062 rc = -EOPNOTSUPP; 3063 else 3064 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3065 optval, optlen); 3066 if (smc->clcsock->sk->sk_err) { 3067 sk->sk_err = smc->clcsock->sk->sk_err; 3068 sk_error_report(sk); 3069 } 3070 mutex_unlock(&smc->clcsock_release_lock); 3071 3072 if (optlen < sizeof(int)) 3073 return -EINVAL; 3074 if (copy_from_sockptr(&val, optval, sizeof(int))) 3075 return -EFAULT; 3076 3077 lock_sock(sk); 3078 if (rc || smc->use_fallback) 3079 goto out; 3080 switch (optname) { 3081 case TCP_FASTOPEN: 3082 case TCP_FASTOPEN_CONNECT: 3083 case TCP_FASTOPEN_KEY: 3084 case TCP_FASTOPEN_NO_COOKIE: 3085 /* option not supported by SMC */ 3086 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3087 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3088 } else { 3089 rc = -EINVAL; 3090 } 3091 break; 3092 case TCP_NODELAY: 3093 if (sk->sk_state != SMC_INIT && 3094 sk->sk_state != SMC_LISTEN && 3095 sk->sk_state != SMC_CLOSED) { 3096 if (val) { 3097 SMC_STAT_INC(smc, ndly_cnt); 3098 smc_tx_pending(&smc->conn); 3099 cancel_delayed_work(&smc->conn.tx_work); 3100 } 3101 } 3102 break; 3103 case TCP_CORK: 3104 if (sk->sk_state != SMC_INIT && 3105 sk->sk_state != SMC_LISTEN && 3106 sk->sk_state != SMC_CLOSED) { 3107 if (!val) { 3108 SMC_STAT_INC(smc, cork_cnt); 3109 smc_tx_pending(&smc->conn); 3110 cancel_delayed_work(&smc->conn.tx_work); 3111 } 3112 } 3113 break; 3114 case TCP_DEFER_ACCEPT: 3115 smc->sockopt_defer_accept = val; 3116 break; 3117 default: 3118 break; 3119 } 3120 out: 3121 release_sock(sk); 3122 3123 return rc; 3124 } 3125 3126 static int smc_getsockopt(struct socket *sock, int level, int optname, 3127 char __user *optval, int __user *optlen) 3128 { 3129 struct smc_sock *smc; 3130 int rc; 3131 3132 if (level == SOL_SMC) 3133 return __smc_getsockopt(sock, level, optname, optval, optlen); 3134 3135 smc = smc_sk(sock->sk); 3136 mutex_lock(&smc->clcsock_release_lock); 3137 if (!smc->clcsock) { 3138 mutex_unlock(&smc->clcsock_release_lock); 3139 return -EBADF; 3140 } 3141 /* socket options apply to the CLC socket */ 3142 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3143 mutex_unlock(&smc->clcsock_release_lock); 3144 return -EOPNOTSUPP; 3145 } 3146 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3147 optval, optlen); 3148 mutex_unlock(&smc->clcsock_release_lock); 3149 return rc; 3150 } 3151 3152 static int smc_ioctl(struct socket *sock, unsigned int cmd, 3153 unsigned long arg) 3154 { 3155 union smc_host_cursor cons, urg; 3156 struct smc_connection *conn; 3157 struct smc_sock *smc; 3158 int answ; 3159 3160 smc = smc_sk(sock->sk); 3161 conn = &smc->conn; 3162 lock_sock(&smc->sk); 3163 if (smc->use_fallback) { 3164 if (!smc->clcsock) { 3165 release_sock(&smc->sk); 3166 return -EBADF; 3167 } 3168 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3169 release_sock(&smc->sk); 3170 return answ; 3171 } 3172 switch (cmd) { 3173 case SIOCINQ: /* same as FIONREAD */ 3174 if (smc->sk.sk_state == SMC_LISTEN) { 3175 release_sock(&smc->sk); 3176 return -EINVAL; 3177 } 3178 if (smc->sk.sk_state == SMC_INIT || 3179 smc->sk.sk_state == SMC_CLOSED) 3180 answ = 0; 3181 else 3182 answ = atomic_read(&smc->conn.bytes_to_rcv); 3183 break; 3184 case SIOCOUTQ: 3185 /* output queue size (not send + not acked) */ 3186 if (smc->sk.sk_state == SMC_LISTEN) { 3187 release_sock(&smc->sk); 3188 return -EINVAL; 3189 } 3190 if (smc->sk.sk_state == SMC_INIT || 3191 smc->sk.sk_state == SMC_CLOSED) 3192 answ = 0; 3193 else 3194 answ = smc->conn.sndbuf_desc->len - 3195 atomic_read(&smc->conn.sndbuf_space); 3196 break; 3197 case SIOCOUTQNSD: 3198 /* output queue size (not send only) */ 3199 if (smc->sk.sk_state == SMC_LISTEN) { 3200 release_sock(&smc->sk); 3201 return -EINVAL; 3202 } 3203 if (smc->sk.sk_state == SMC_INIT || 3204 smc->sk.sk_state == SMC_CLOSED) 3205 answ = 0; 3206 else 3207 answ = smc_tx_prepared_sends(&smc->conn); 3208 break; 3209 case SIOCATMARK: 3210 if (smc->sk.sk_state == SMC_LISTEN) { 3211 release_sock(&smc->sk); 3212 return -EINVAL; 3213 } 3214 if (smc->sk.sk_state == SMC_INIT || 3215 smc->sk.sk_state == SMC_CLOSED) { 3216 answ = 0; 3217 } else { 3218 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3219 smc_curs_copy(&urg, &conn->urg_curs, conn); 3220 answ = smc_curs_diff(conn->rmb_desc->len, 3221 &cons, &urg) == 1; 3222 } 3223 break; 3224 default: 3225 release_sock(&smc->sk); 3226 return -ENOIOCTLCMD; 3227 } 3228 release_sock(&smc->sk); 3229 3230 return put_user(answ, (int __user *)arg); 3231 } 3232 3233 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3234 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3235 * updates till whenever a respective page has been fully processed. 3236 * Note that subsequent recv() calls have to wait till all splice() processing 3237 * completed. 3238 */ 3239 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3240 struct pipe_inode_info *pipe, size_t len, 3241 unsigned int flags) 3242 { 3243 struct sock *sk = sock->sk; 3244 struct smc_sock *smc; 3245 int rc = -ENOTCONN; 3246 3247 smc = smc_sk(sk); 3248 lock_sock(sk); 3249 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3250 /* socket was connected before, no more data to read */ 3251 rc = 0; 3252 goto out; 3253 } 3254 if (sk->sk_state == SMC_INIT || 3255 sk->sk_state == SMC_LISTEN || 3256 sk->sk_state == SMC_CLOSED) 3257 goto out; 3258 3259 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3260 rc = 0; 3261 goto out; 3262 } 3263 3264 if (smc->use_fallback) { 3265 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3266 pipe, len, flags); 3267 } else { 3268 if (*ppos) { 3269 rc = -ESPIPE; 3270 goto out; 3271 } 3272 if (flags & SPLICE_F_NONBLOCK) 3273 flags = MSG_DONTWAIT; 3274 else 3275 flags = 0; 3276 SMC_STAT_INC(smc, splice_cnt); 3277 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3278 } 3279 out: 3280 release_sock(sk); 3281 3282 return rc; 3283 } 3284 3285 /* must look like tcp */ 3286 static const struct proto_ops smc_sock_ops = { 3287 .family = PF_SMC, 3288 .owner = THIS_MODULE, 3289 .release = smc_release, 3290 .bind = smc_bind, 3291 .connect = smc_connect, 3292 .socketpair = sock_no_socketpair, 3293 .accept = smc_accept, 3294 .getname = smc_getname, 3295 .poll = smc_poll, 3296 .ioctl = smc_ioctl, 3297 .listen = smc_listen, 3298 .shutdown = smc_shutdown, 3299 .setsockopt = smc_setsockopt, 3300 .getsockopt = smc_getsockopt, 3301 .sendmsg = smc_sendmsg, 3302 .recvmsg = smc_recvmsg, 3303 .mmap = sock_no_mmap, 3304 .splice_read = smc_splice_read, 3305 }; 3306 3307 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3308 int kern, struct socket *clcsock) 3309 { 3310 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3311 struct smc_sock *smc; 3312 struct sock *sk; 3313 int rc; 3314 3315 rc = -ESOCKTNOSUPPORT; 3316 if (sock->type != SOCK_STREAM) 3317 goto out; 3318 3319 rc = -EPROTONOSUPPORT; 3320 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3321 goto out; 3322 3323 rc = -ENOBUFS; 3324 sock->ops = &smc_sock_ops; 3325 sock->state = SS_UNCONNECTED; 3326 sk = smc_sock_alloc(net, sock, protocol); 3327 if (!sk) 3328 goto out; 3329 3330 /* create internal TCP socket for CLC handshake and fallback */ 3331 smc = smc_sk(sk); 3332 smc->use_fallback = false; /* assume rdma capability first */ 3333 smc->fallback_rsn = 0; 3334 3335 /* default behavior from limit_smc_hs in every net namespace */ 3336 smc->limit_smc_hs = net->smc.limit_smc_hs; 3337 3338 rc = 0; 3339 if (!clcsock) { 3340 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3341 &smc->clcsock); 3342 if (rc) { 3343 sk_common_release(sk); 3344 goto out; 3345 } 3346 3347 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3348 * destruction; its sk_state might not be TCP_CLOSE after 3349 * smc->sk is close()d, and TCP timers can be fired later, 3350 * which need net ref. 3351 */ 3352 sk = smc->clcsock->sk; 3353 __netns_tracker_free(net, &sk->ns_tracker, false); 3354 sk->sk_net_refcnt = 1; 3355 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3356 sock_inuse_add(net, 1); 3357 } else { 3358 smc->clcsock = clcsock; 3359 } 3360 3361 out: 3362 return rc; 3363 } 3364 3365 static int smc_create(struct net *net, struct socket *sock, int protocol, 3366 int kern) 3367 { 3368 return __smc_create(net, sock, protocol, kern, NULL); 3369 } 3370 3371 static const struct net_proto_family smc_sock_family_ops = { 3372 .family = PF_SMC, 3373 .owner = THIS_MODULE, 3374 .create = smc_create, 3375 }; 3376 3377 static int smc_ulp_init(struct sock *sk) 3378 { 3379 struct socket *tcp = sk->sk_socket; 3380 struct net *net = sock_net(sk); 3381 struct socket *smcsock; 3382 int protocol, ret; 3383 3384 /* only TCP can be replaced */ 3385 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3386 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3387 return -ESOCKTNOSUPPORT; 3388 /* don't handle wq now */ 3389 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3390 return -ENOTCONN; 3391 3392 if (sk->sk_family == AF_INET) 3393 protocol = SMCPROTO_SMC; 3394 else 3395 protocol = SMCPROTO_SMC6; 3396 3397 smcsock = sock_alloc(); 3398 if (!smcsock) 3399 return -ENFILE; 3400 3401 smcsock->type = SOCK_STREAM; 3402 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3403 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3404 if (ret) { 3405 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3406 return ret; 3407 } 3408 3409 /* replace tcp socket to smc */ 3410 smcsock->file = tcp->file; 3411 smcsock->file->private_data = smcsock; 3412 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3413 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3414 tcp->file = NULL; 3415 3416 return ret; 3417 } 3418 3419 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3420 const gfp_t priority) 3421 { 3422 struct inet_connection_sock *icsk = inet_csk(newsk); 3423 3424 /* don't inherit ulp ops to child when listen */ 3425 icsk->icsk_ulp_ops = NULL; 3426 } 3427 3428 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3429 .name = "smc", 3430 .owner = THIS_MODULE, 3431 .init = smc_ulp_init, 3432 .clone = smc_ulp_clone, 3433 }; 3434 3435 unsigned int smc_net_id; 3436 3437 static __net_init int smc_net_init(struct net *net) 3438 { 3439 int rc; 3440 3441 rc = smc_sysctl_net_init(net); 3442 if (rc) 3443 return rc; 3444 return smc_pnet_net_init(net); 3445 } 3446 3447 static void __net_exit smc_net_exit(struct net *net) 3448 { 3449 smc_sysctl_net_exit(net); 3450 smc_pnet_net_exit(net); 3451 } 3452 3453 static __net_init int smc_net_stat_init(struct net *net) 3454 { 3455 return smc_stats_init(net); 3456 } 3457 3458 static void __net_exit smc_net_stat_exit(struct net *net) 3459 { 3460 smc_stats_exit(net); 3461 } 3462 3463 static struct pernet_operations smc_net_ops = { 3464 .init = smc_net_init, 3465 .exit = smc_net_exit, 3466 .id = &smc_net_id, 3467 .size = sizeof(struct smc_net), 3468 }; 3469 3470 static struct pernet_operations smc_net_stat_ops = { 3471 .init = smc_net_stat_init, 3472 .exit = smc_net_stat_exit, 3473 }; 3474 3475 static int __init smc_init(void) 3476 { 3477 int rc; 3478 3479 rc = register_pernet_subsys(&smc_net_ops); 3480 if (rc) 3481 return rc; 3482 3483 rc = register_pernet_subsys(&smc_net_stat_ops); 3484 if (rc) 3485 goto out_pernet_subsys; 3486 3487 rc = smc_ism_init(); 3488 if (rc) 3489 goto out_pernet_subsys_stat; 3490 smc_clc_init(); 3491 3492 rc = smc_nl_init(); 3493 if (rc) 3494 goto out_ism; 3495 3496 rc = smc_pnet_init(); 3497 if (rc) 3498 goto out_nl; 3499 3500 rc = -ENOMEM; 3501 3502 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3503 if (!smc_tcp_ls_wq) 3504 goto out_pnet; 3505 3506 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3507 if (!smc_hs_wq) 3508 goto out_alloc_tcp_ls_wq; 3509 3510 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3511 if (!smc_close_wq) 3512 goto out_alloc_hs_wq; 3513 3514 rc = smc_core_init(); 3515 if (rc) { 3516 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3517 goto out_alloc_wqs; 3518 } 3519 3520 rc = smc_llc_init(); 3521 if (rc) { 3522 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3523 goto out_core; 3524 } 3525 3526 rc = smc_cdc_init(); 3527 if (rc) { 3528 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3529 goto out_core; 3530 } 3531 3532 rc = proto_register(&smc_proto, 1); 3533 if (rc) { 3534 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3535 goto out_core; 3536 } 3537 3538 rc = proto_register(&smc_proto6, 1); 3539 if (rc) { 3540 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3541 goto out_proto; 3542 } 3543 3544 rc = sock_register(&smc_sock_family_ops); 3545 if (rc) { 3546 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3547 goto out_proto6; 3548 } 3549 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3550 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3551 3552 rc = smc_ib_register_client(); 3553 if (rc) { 3554 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3555 goto out_sock; 3556 } 3557 3558 rc = tcp_register_ulp(&smc_ulp_ops); 3559 if (rc) { 3560 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3561 goto out_ib; 3562 } 3563 3564 static_branch_enable(&tcp_have_smc); 3565 return 0; 3566 3567 out_ib: 3568 smc_ib_unregister_client(); 3569 out_sock: 3570 sock_unregister(PF_SMC); 3571 out_proto6: 3572 proto_unregister(&smc_proto6); 3573 out_proto: 3574 proto_unregister(&smc_proto); 3575 out_core: 3576 smc_core_exit(); 3577 out_alloc_wqs: 3578 destroy_workqueue(smc_close_wq); 3579 out_alloc_hs_wq: 3580 destroy_workqueue(smc_hs_wq); 3581 out_alloc_tcp_ls_wq: 3582 destroy_workqueue(smc_tcp_ls_wq); 3583 out_pnet: 3584 smc_pnet_exit(); 3585 out_nl: 3586 smc_nl_exit(); 3587 out_ism: 3588 smc_clc_exit(); 3589 smc_ism_exit(); 3590 out_pernet_subsys_stat: 3591 unregister_pernet_subsys(&smc_net_stat_ops); 3592 out_pernet_subsys: 3593 unregister_pernet_subsys(&smc_net_ops); 3594 3595 return rc; 3596 } 3597 3598 static void __exit smc_exit(void) 3599 { 3600 static_branch_disable(&tcp_have_smc); 3601 tcp_unregister_ulp(&smc_ulp_ops); 3602 sock_unregister(PF_SMC); 3603 smc_core_exit(); 3604 smc_ib_unregister_client(); 3605 smc_ism_exit(); 3606 destroy_workqueue(smc_close_wq); 3607 destroy_workqueue(smc_tcp_ls_wq); 3608 destroy_workqueue(smc_hs_wq); 3609 proto_unregister(&smc_proto6); 3610 proto_unregister(&smc_proto); 3611 smc_pnet_exit(); 3612 smc_nl_exit(); 3613 smc_clc_exit(); 3614 unregister_pernet_subsys(&smc_net_stat_ops); 3615 unregister_pernet_subsys(&smc_net_ops); 3616 rcu_barrier(); 3617 } 3618 3619 module_init(smc_init); 3620 module_exit(smc_exit); 3621 3622 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3623 MODULE_DESCRIPTION("smc socket address family"); 3624 MODULE_LICENSE("GPL"); 3625 MODULE_ALIAS_NETPROTO(PF_SMC); 3626 MODULE_ALIAS_TCP_ULP("smc"); 3627 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3628