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 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 EXPORT_SYMBOL_GPL(smc_hash_sk); 195 196 void smc_unhash_sk(struct sock *sk) 197 { 198 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 199 200 write_lock_bh(&h->lock); 201 if (sk_del_node_init(sk)) 202 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 203 write_unlock_bh(&h->lock); 204 } 205 EXPORT_SYMBOL_GPL(smc_unhash_sk); 206 207 /* This will be called before user really release sock_lock. So do the 208 * work which we didn't do because of user hold the sock_lock in the 209 * BH context 210 */ 211 static void smc_release_cb(struct sock *sk) 212 { 213 struct smc_sock *smc = smc_sk(sk); 214 215 if (smc->conn.tx_in_release_sock) { 216 smc_tx_pending(&smc->conn); 217 smc->conn.tx_in_release_sock = false; 218 } 219 } 220 221 struct proto smc_proto = { 222 .name = "SMC", 223 .owner = THIS_MODULE, 224 .keepalive = smc_set_keepalive, 225 .hash = smc_hash_sk, 226 .unhash = smc_unhash_sk, 227 .release_cb = smc_release_cb, 228 .obj_size = sizeof(struct smc_sock), 229 .h.smc_hash = &smc_v4_hashinfo, 230 .slab_flags = SLAB_TYPESAFE_BY_RCU, 231 }; 232 EXPORT_SYMBOL_GPL(smc_proto); 233 234 struct proto smc_proto6 = { 235 .name = "SMC6", 236 .owner = THIS_MODULE, 237 .keepalive = smc_set_keepalive, 238 .hash = smc_hash_sk, 239 .unhash = smc_unhash_sk, 240 .release_cb = smc_release_cb, 241 .obj_size = sizeof(struct smc_sock), 242 .h.smc_hash = &smc_v6_hashinfo, 243 .slab_flags = SLAB_TYPESAFE_BY_RCU, 244 }; 245 EXPORT_SYMBOL_GPL(smc_proto6); 246 247 static void smc_fback_restore_callbacks(struct smc_sock *smc) 248 { 249 struct sock *clcsk = smc->clcsock->sk; 250 251 write_lock_bh(&clcsk->sk_callback_lock); 252 clcsk->sk_user_data = NULL; 253 254 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 255 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 256 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 257 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 258 259 write_unlock_bh(&clcsk->sk_callback_lock); 260 } 261 262 static void smc_restore_fallback_changes(struct smc_sock *smc) 263 { 264 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 265 smc->clcsock->file->private_data = smc->sk.sk_socket; 266 smc->clcsock->file = NULL; 267 smc_fback_restore_callbacks(smc); 268 } 269 } 270 271 static int __smc_release(struct smc_sock *smc) 272 { 273 struct sock *sk = &smc->sk; 274 int rc = 0; 275 276 if (!smc->use_fallback) { 277 rc = smc_close_active(smc); 278 smc_sock_set_flag(sk, SOCK_DEAD); 279 sk->sk_shutdown |= SHUTDOWN_MASK; 280 } else { 281 if (sk->sk_state != SMC_CLOSED) { 282 if (sk->sk_state != SMC_LISTEN && 283 sk->sk_state != SMC_INIT) 284 sock_put(sk); /* passive closing */ 285 if (sk->sk_state == SMC_LISTEN) { 286 /* wake up clcsock accept */ 287 rc = kernel_sock_shutdown(smc->clcsock, 288 SHUT_RDWR); 289 } 290 sk->sk_state = SMC_CLOSED; 291 sk->sk_state_change(sk); 292 } 293 smc_restore_fallback_changes(smc); 294 } 295 296 sk->sk_prot->unhash(sk); 297 298 if (sk->sk_state == SMC_CLOSED) { 299 if (smc->clcsock) { 300 release_sock(sk); 301 smc_clcsock_release(smc); 302 lock_sock(sk); 303 } 304 if (!smc->use_fallback) 305 smc_conn_free(&smc->conn); 306 } 307 308 return rc; 309 } 310 311 static int smc_release(struct socket *sock) 312 { 313 struct sock *sk = sock->sk; 314 struct smc_sock *smc; 315 int old_state, rc = 0; 316 317 if (!sk) 318 goto out; 319 320 sock_hold(sk); /* sock_put below */ 321 smc = smc_sk(sk); 322 323 old_state = sk->sk_state; 324 325 /* cleanup for a dangling non-blocking connect */ 326 if (smc->connect_nonblock && old_state == SMC_INIT) 327 tcp_abort(smc->clcsock->sk, ECONNABORTED); 328 329 if (cancel_work_sync(&smc->connect_work)) 330 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 331 332 if (sk->sk_state == SMC_LISTEN) 333 /* smc_close_non_accepted() is called and acquires 334 * sock lock for child sockets again 335 */ 336 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 337 else 338 lock_sock(sk); 339 340 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 341 !smc->use_fallback) 342 smc_close_active_abort(smc); 343 344 rc = __smc_release(smc); 345 346 /* detach socket */ 347 sock_orphan(sk); 348 sock->sk = NULL; 349 release_sock(sk); 350 351 sock_put(sk); /* sock_hold above */ 352 sock_put(sk); /* final sock_put */ 353 out: 354 return rc; 355 } 356 357 static void smc_destruct(struct sock *sk) 358 { 359 if (sk->sk_state != SMC_CLOSED) 360 return; 361 if (!sock_flag(sk, SOCK_DEAD)) 362 return; 363 } 364 365 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 366 int protocol) 367 { 368 struct smc_sock *smc; 369 struct proto *prot; 370 struct sock *sk; 371 372 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 373 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 374 if (!sk) 375 return NULL; 376 377 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 378 sk->sk_state = SMC_INIT; 379 sk->sk_destruct = smc_destruct; 380 sk->sk_protocol = protocol; 381 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 382 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 383 smc = smc_sk(sk); 384 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 385 INIT_WORK(&smc->connect_work, smc_connect_work); 386 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 387 INIT_LIST_HEAD(&smc->accept_q); 388 spin_lock_init(&smc->accept_q_lock); 389 spin_lock_init(&smc->conn.send_lock); 390 sk->sk_prot->hash(sk); 391 mutex_init(&smc->clcsock_release_lock); 392 smc_init_saved_callbacks(smc); 393 394 return sk; 395 } 396 397 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 398 int addr_len) 399 { 400 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 401 struct sock *sk = sock->sk; 402 struct smc_sock *smc; 403 int rc; 404 405 smc = smc_sk(sk); 406 407 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 408 rc = -EINVAL; 409 if (addr_len < sizeof(struct sockaddr_in)) 410 goto out; 411 412 rc = -EAFNOSUPPORT; 413 if (addr->sin_family != AF_INET && 414 addr->sin_family != AF_INET6 && 415 addr->sin_family != AF_UNSPEC) 416 goto out; 417 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 418 if (addr->sin_family == AF_UNSPEC && 419 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 420 goto out; 421 422 lock_sock(sk); 423 424 /* Check if socket is already active */ 425 rc = -EINVAL; 426 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 427 goto out_rel; 428 429 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 430 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 431 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 432 433 out_rel: 434 release_sock(sk); 435 out: 436 return rc; 437 } 438 439 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 440 * clc socket (since smc is not called for these options from net/core) 441 */ 442 443 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 444 (1UL << SOCK_KEEPOPEN) | \ 445 (1UL << SOCK_LINGER) | \ 446 (1UL << SOCK_BROADCAST) | \ 447 (1UL << SOCK_TIMESTAMP) | \ 448 (1UL << SOCK_DBG) | \ 449 (1UL << SOCK_RCVTSTAMP) | \ 450 (1UL << SOCK_RCVTSTAMPNS) | \ 451 (1UL << SOCK_LOCALROUTE) | \ 452 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 453 (1UL << SOCK_RXQ_OVFL) | \ 454 (1UL << SOCK_WIFI_STATUS) | \ 455 (1UL << SOCK_NOFCS) | \ 456 (1UL << SOCK_FILTER_LOCKED) | \ 457 (1UL << SOCK_TSTAMP_NEW)) 458 459 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 460 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 461 unsigned long mask) 462 { 463 struct net *nnet = sock_net(nsk); 464 465 nsk->sk_userlocks = osk->sk_userlocks; 466 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) { 467 nsk->sk_sndbuf = osk->sk_sndbuf; 468 } else { 469 if (mask == SK_FLAGS_SMC_TO_CLC) 470 WRITE_ONCE(nsk->sk_sndbuf, 471 READ_ONCE(nnet->ipv4.sysctl_tcp_wmem[1])); 472 else 473 WRITE_ONCE(nsk->sk_sndbuf, 474 2 * READ_ONCE(nnet->smc.sysctl_wmem)); 475 } 476 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) { 477 nsk->sk_rcvbuf = osk->sk_rcvbuf; 478 } else { 479 if (mask == SK_FLAGS_SMC_TO_CLC) 480 WRITE_ONCE(nsk->sk_rcvbuf, 481 READ_ONCE(nnet->ipv4.sysctl_tcp_rmem[1])); 482 else 483 WRITE_ONCE(nsk->sk_rcvbuf, 484 2 * READ_ONCE(nnet->smc.sysctl_rmem)); 485 } 486 } 487 488 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 489 unsigned long mask) 490 { 491 /* options we don't get control via setsockopt for */ 492 nsk->sk_type = osk->sk_type; 493 nsk->sk_sndtimeo = osk->sk_sndtimeo; 494 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 495 nsk->sk_mark = READ_ONCE(osk->sk_mark); 496 nsk->sk_priority = READ_ONCE(osk->sk_priority); 497 nsk->sk_rcvlowat = osk->sk_rcvlowat; 498 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 499 nsk->sk_err = osk->sk_err; 500 501 nsk->sk_flags &= ~mask; 502 nsk->sk_flags |= osk->sk_flags & mask; 503 504 smc_adjust_sock_bufsizes(nsk, osk, mask); 505 } 506 507 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 508 { 509 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 510 } 511 512 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 513 (1UL << SOCK_KEEPOPEN) | \ 514 (1UL << SOCK_LINGER) | \ 515 (1UL << SOCK_DBG)) 516 /* copy only settings and flags relevant for smc from clc to smc socket */ 517 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 518 { 519 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 520 } 521 522 /* register the new vzalloced sndbuf on all links */ 523 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 524 struct smc_buf_desc *snd_desc) 525 { 526 struct smc_link_group *lgr = link->lgr; 527 int i, rc = 0; 528 529 if (!snd_desc->is_vm) 530 return -EINVAL; 531 532 /* protect against parallel smcr_link_reg_buf() */ 533 down_write(&lgr->llc_conf_mutex); 534 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 535 if (!smc_link_active(&lgr->lnk[i])) 536 continue; 537 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 538 if (rc) 539 break; 540 } 541 up_write(&lgr->llc_conf_mutex); 542 return rc; 543 } 544 545 /* register the new rmb on all links */ 546 static int smcr_lgr_reg_rmbs(struct smc_link *link, 547 struct smc_buf_desc *rmb_desc) 548 { 549 struct smc_link_group *lgr = link->lgr; 550 bool do_slow = false; 551 int i, rc = 0; 552 553 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 554 if (rc) 555 return rc; 556 557 down_read(&lgr->llc_conf_mutex); 558 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 559 if (!smc_link_active(&lgr->lnk[i])) 560 continue; 561 if (!rmb_desc->is_reg_mr[link->link_idx]) { 562 up_read(&lgr->llc_conf_mutex); 563 goto slow_path; 564 } 565 } 566 /* mr register already */ 567 goto fast_path; 568 slow_path: 569 do_slow = true; 570 /* protect against parallel smc_llc_cli_rkey_exchange() and 571 * parallel smcr_link_reg_buf() 572 */ 573 down_write(&lgr->llc_conf_mutex); 574 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 575 if (!smc_link_active(&lgr->lnk[i])) 576 continue; 577 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 578 if (rc) 579 goto out; 580 } 581 fast_path: 582 /* exchange confirm_rkey msg with peer */ 583 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 584 if (rc) { 585 rc = -EFAULT; 586 goto out; 587 } 588 rmb_desc->is_conf_rkey = true; 589 out: 590 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 591 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 592 return rc; 593 } 594 595 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 596 { 597 struct smc_link *link = smc->conn.lnk; 598 struct smc_llc_qentry *qentry; 599 int rc; 600 601 /* Receive CONFIRM LINK request from server over RoCE fabric. 602 * Increasing the client's timeout by twice as much as the server's 603 * timeout by default can temporarily avoid decline messages of 604 * both sides crossing or colliding 605 */ 606 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 607 SMC_LLC_CONFIRM_LINK); 608 if (!qentry) { 609 struct smc_clc_msg_decline dclc; 610 611 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 612 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 613 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 614 } 615 smc_llc_save_peer_uid(qentry); 616 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 617 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 618 if (rc) 619 return SMC_CLC_DECL_RMBE_EC; 620 621 rc = smc_ib_modify_qp_rts(link); 622 if (rc) 623 return SMC_CLC_DECL_ERR_RDYLNK; 624 625 smc_wr_remember_qp_attr(link); 626 627 /* reg the sndbuf if it was vzalloced */ 628 if (smc->conn.sndbuf_desc->is_vm) { 629 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 630 return SMC_CLC_DECL_ERR_REGBUF; 631 } 632 633 /* reg the rmb */ 634 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 635 return SMC_CLC_DECL_ERR_REGBUF; 636 637 /* confirm_rkey is implicit on 1st contact */ 638 smc->conn.rmb_desc->is_conf_rkey = true; 639 640 /* send CONFIRM LINK response over RoCE fabric */ 641 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 642 if (rc < 0) 643 return SMC_CLC_DECL_TIMEOUT_CL; 644 645 smc_llc_link_active(link); 646 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 647 648 if (link->lgr->max_links > 1) { 649 /* optional 2nd link, receive ADD LINK request from server */ 650 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 651 SMC_LLC_ADD_LINK); 652 if (!qentry) { 653 struct smc_clc_msg_decline dclc; 654 655 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 656 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 657 if (rc == -EAGAIN) 658 rc = 0; /* no DECLINE received, go with one link */ 659 return rc; 660 } 661 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 662 smc_llc_cli_add_link(link, qentry); 663 } 664 return 0; 665 } 666 667 static bool smc_isascii(char *hostname) 668 { 669 int i; 670 671 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 672 if (!isascii(hostname[i])) 673 return false; 674 return true; 675 } 676 677 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 678 struct smc_clc_msg_accept_confirm *clc) 679 { 680 struct smc_clc_first_contact_ext *fce; 681 int clc_v2_len; 682 683 if (clc->hdr.version == SMC_V1 || 684 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 685 return; 686 687 if (smc->conn.lgr->is_smcd) { 688 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 689 SMC_MAX_EID_LEN); 690 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 691 } else { 692 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 693 SMC_MAX_EID_LEN); 694 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 695 } 696 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 697 smc->conn.lgr->peer_os = fce->os_type; 698 smc->conn.lgr->peer_smc_release = fce->release; 699 if (smc_isascii(fce->hostname)) 700 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 701 SMC_MAX_HOSTNAME_LEN); 702 } 703 704 static void smcr_conn_save_peer_info(struct smc_sock *smc, 705 struct smc_clc_msg_accept_confirm *clc) 706 { 707 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 708 709 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 710 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 711 smc->conn.peer_rmbe_size = bufsize; 712 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 713 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 714 } 715 716 static void smcd_conn_save_peer_info(struct smc_sock *smc, 717 struct smc_clc_msg_accept_confirm *clc) 718 { 719 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 720 721 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 722 smc->conn.peer_token = ntohll(clc->d0.token); 723 /* msg header takes up space in the buffer */ 724 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 725 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 726 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 727 } 728 729 static void smc_conn_save_peer_info(struct smc_sock *smc, 730 struct smc_clc_msg_accept_confirm *clc) 731 { 732 if (smc->conn.lgr->is_smcd) 733 smcd_conn_save_peer_info(smc, clc); 734 else 735 smcr_conn_save_peer_info(smc, clc); 736 smc_conn_save_peer_info_fce(smc, clc); 737 } 738 739 static void smc_link_save_peer_info(struct smc_link *link, 740 struct smc_clc_msg_accept_confirm *clc, 741 struct smc_init_info *ini) 742 { 743 link->peer_qpn = ntoh24(clc->r0.qpn); 744 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 745 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 746 link->peer_psn = ntoh24(clc->r0.psn); 747 link->peer_mtu = clc->r0.qp_mtu; 748 } 749 750 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 751 struct smc_stats_fback *fback_arr) 752 { 753 int cnt; 754 755 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 756 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 757 fback_arr[cnt].count++; 758 break; 759 } 760 if (!fback_arr[cnt].fback_code) { 761 fback_arr[cnt].fback_code = smc->fallback_rsn; 762 fback_arr[cnt].count++; 763 break; 764 } 765 } 766 } 767 768 static void smc_stat_fallback(struct smc_sock *smc) 769 { 770 struct net *net = sock_net(&smc->sk); 771 772 mutex_lock(&net->smc.mutex_fback_rsn); 773 if (smc->listen_smc) { 774 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 775 net->smc.fback_rsn->srv_fback_cnt++; 776 } else { 777 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 778 net->smc.fback_rsn->clnt_fback_cnt++; 779 } 780 mutex_unlock(&net->smc.mutex_fback_rsn); 781 } 782 783 /* must be called under rcu read lock */ 784 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 785 { 786 struct socket_wq *wq; 787 __poll_t flags; 788 789 wq = rcu_dereference(smc->sk.sk_wq); 790 if (!skwq_has_sleeper(wq)) 791 return; 792 793 /* wake up smc sk->sk_wq */ 794 if (!key) { 795 /* sk_state_change */ 796 wake_up_interruptible_all(&wq->wait); 797 } else { 798 flags = key_to_poll(key); 799 if (flags & (EPOLLIN | EPOLLOUT)) 800 /* sk_data_ready or sk_write_space */ 801 wake_up_interruptible_sync_poll(&wq->wait, flags); 802 else if (flags & EPOLLERR) 803 /* sk_error_report */ 804 wake_up_interruptible_poll(&wq->wait, flags); 805 } 806 } 807 808 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 809 unsigned int mode, int sync, void *key) 810 { 811 struct smc_mark_woken *mark = 812 container_of(wait, struct smc_mark_woken, wait_entry); 813 814 mark->woken = true; 815 mark->key = key; 816 return 0; 817 } 818 819 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 820 void (*clcsock_callback)(struct sock *sk)) 821 { 822 struct smc_mark_woken mark = { .woken = false }; 823 struct socket_wq *wq; 824 825 init_waitqueue_func_entry(&mark.wait_entry, 826 smc_fback_mark_woken); 827 rcu_read_lock(); 828 wq = rcu_dereference(clcsk->sk_wq); 829 if (!wq) 830 goto out; 831 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 832 clcsock_callback(clcsk); 833 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 834 835 if (mark.woken) 836 smc_fback_wakeup_waitqueue(smc, mark.key); 837 out: 838 rcu_read_unlock(); 839 } 840 841 static void smc_fback_state_change(struct sock *clcsk) 842 { 843 struct smc_sock *smc; 844 845 read_lock_bh(&clcsk->sk_callback_lock); 846 smc = smc_clcsock_user_data(clcsk); 847 if (smc) 848 smc_fback_forward_wakeup(smc, clcsk, 849 smc->clcsk_state_change); 850 read_unlock_bh(&clcsk->sk_callback_lock); 851 } 852 853 static void smc_fback_data_ready(struct sock *clcsk) 854 { 855 struct smc_sock *smc; 856 857 read_lock_bh(&clcsk->sk_callback_lock); 858 smc = smc_clcsock_user_data(clcsk); 859 if (smc) 860 smc_fback_forward_wakeup(smc, clcsk, 861 smc->clcsk_data_ready); 862 read_unlock_bh(&clcsk->sk_callback_lock); 863 } 864 865 static void smc_fback_write_space(struct sock *clcsk) 866 { 867 struct smc_sock *smc; 868 869 read_lock_bh(&clcsk->sk_callback_lock); 870 smc = smc_clcsock_user_data(clcsk); 871 if (smc) 872 smc_fback_forward_wakeup(smc, clcsk, 873 smc->clcsk_write_space); 874 read_unlock_bh(&clcsk->sk_callback_lock); 875 } 876 877 static void smc_fback_error_report(struct sock *clcsk) 878 { 879 struct smc_sock *smc; 880 881 read_lock_bh(&clcsk->sk_callback_lock); 882 smc = smc_clcsock_user_data(clcsk); 883 if (smc) 884 smc_fback_forward_wakeup(smc, clcsk, 885 smc->clcsk_error_report); 886 read_unlock_bh(&clcsk->sk_callback_lock); 887 } 888 889 static void smc_fback_replace_callbacks(struct smc_sock *smc) 890 { 891 struct sock *clcsk = smc->clcsock->sk; 892 893 write_lock_bh(&clcsk->sk_callback_lock); 894 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 895 896 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 897 &smc->clcsk_state_change); 898 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 899 &smc->clcsk_data_ready); 900 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 901 &smc->clcsk_write_space); 902 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 903 &smc->clcsk_error_report); 904 905 write_unlock_bh(&clcsk->sk_callback_lock); 906 } 907 908 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 909 { 910 int rc = 0; 911 912 mutex_lock(&smc->clcsock_release_lock); 913 if (!smc->clcsock) { 914 rc = -EBADF; 915 goto out; 916 } 917 918 smc->use_fallback = true; 919 smc->fallback_rsn = reason_code; 920 smc_stat_fallback(smc); 921 trace_smc_switch_to_fallback(smc, reason_code); 922 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 923 smc->clcsock->file = smc->sk.sk_socket->file; 924 smc->clcsock->file->private_data = smc->clcsock; 925 smc->clcsock->wq.fasync_list = 926 smc->sk.sk_socket->wq.fasync_list; 927 smc->sk.sk_socket->wq.fasync_list = NULL; 928 929 /* There might be some wait entries remaining 930 * in smc sk->sk_wq and they should be woken up 931 * as clcsock's wait queue is woken up. 932 */ 933 smc_fback_replace_callbacks(smc); 934 } 935 out: 936 mutex_unlock(&smc->clcsock_release_lock); 937 return rc; 938 } 939 940 /* fall back during connect */ 941 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 942 { 943 struct net *net = sock_net(&smc->sk); 944 int rc = 0; 945 946 rc = smc_switch_to_fallback(smc, reason_code); 947 if (rc) { /* fallback fails */ 948 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 949 if (smc->sk.sk_state == SMC_INIT) 950 sock_put(&smc->sk); /* passive closing */ 951 return rc; 952 } 953 smc_copy_sock_settings_to_clc(smc); 954 smc->connect_nonblock = 0; 955 if (smc->sk.sk_state == SMC_INIT) 956 smc->sk.sk_state = SMC_ACTIVE; 957 return 0; 958 } 959 960 /* decline and fall back during connect */ 961 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 962 u8 version) 963 { 964 struct net *net = sock_net(&smc->sk); 965 int rc; 966 967 if (reason_code < 0) { /* error, fallback is not possible */ 968 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 969 if (smc->sk.sk_state == SMC_INIT) 970 sock_put(&smc->sk); /* passive closing */ 971 return reason_code; 972 } 973 if (reason_code != SMC_CLC_DECL_PEERDECL) { 974 rc = smc_clc_send_decline(smc, reason_code, version); 975 if (rc < 0) { 976 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 977 if (smc->sk.sk_state == SMC_INIT) 978 sock_put(&smc->sk); /* passive closing */ 979 return rc; 980 } 981 } 982 return smc_connect_fallback(smc, reason_code); 983 } 984 985 static void smc_conn_abort(struct smc_sock *smc, int local_first) 986 { 987 struct smc_connection *conn = &smc->conn; 988 struct smc_link_group *lgr = conn->lgr; 989 bool lgr_valid = false; 990 991 if (smc_conn_lgr_valid(conn)) 992 lgr_valid = true; 993 994 smc_conn_free(conn); 995 if (local_first && lgr_valid) 996 smc_lgr_cleanup_early(lgr); 997 } 998 999 /* check if there is a rdma device available for this connection. */ 1000 /* called for connect and listen */ 1001 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1002 { 1003 /* PNET table look up: search active ib_device and port 1004 * within same PNETID that also contains the ethernet device 1005 * used for the internal TCP socket 1006 */ 1007 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1008 if (!ini->check_smcrv2 && !ini->ib_dev) 1009 return SMC_CLC_DECL_NOSMCRDEV; 1010 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1011 return SMC_CLC_DECL_NOSMCRDEV; 1012 return 0; 1013 } 1014 1015 /* check if there is an ISM device available for this connection. */ 1016 /* called for connect and listen */ 1017 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1018 { 1019 /* Find ISM device with same PNETID as connecting interface */ 1020 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1021 if (!ini->ism_dev[0]) 1022 return SMC_CLC_DECL_NOSMCDDEV; 1023 else 1024 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1025 return 0; 1026 } 1027 1028 /* is chid unique for the ism devices that are already determined? */ 1029 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1030 int cnt) 1031 { 1032 int i = (!ini->ism_dev[0]) ? 1 : 0; 1033 1034 for (; i < cnt; i++) 1035 if (ini->ism_chid[i] == chid) 1036 return false; 1037 return true; 1038 } 1039 1040 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1041 * PNETID matching net_device) 1042 */ 1043 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1044 struct smc_init_info *ini) 1045 { 1046 int rc = SMC_CLC_DECL_NOSMCDDEV; 1047 struct smcd_dev *smcd; 1048 int i = 1, entry = 1; 1049 bool is_virtual; 1050 u16 chid; 1051 1052 if (smcd_indicated(ini->smc_type_v1)) 1053 rc = 0; /* already initialized for V1 */ 1054 mutex_lock(&smcd_dev_list.mutex); 1055 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1056 if (smcd->going_away || smcd == ini->ism_dev[0]) 1057 continue; 1058 chid = smc_ism_get_chid(smcd); 1059 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1060 continue; 1061 is_virtual = __smc_ism_is_virtual(chid); 1062 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1063 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1064 if (is_virtual && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1065 /* It's the last GID-CHID entry left in CLC 1066 * Proposal SMC-Dv2 extension, but a virtual 1067 * ISM device will take two entries. So give 1068 * up it and try the next potential ISM device. 1069 */ 1070 continue; 1071 ini->ism_dev[i] = smcd; 1072 ini->ism_chid[i] = chid; 1073 ini->is_smcd = true; 1074 rc = 0; 1075 i++; 1076 entry = is_virtual ? entry + 2 : entry + 1; 1077 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1078 break; 1079 } 1080 } 1081 mutex_unlock(&smcd_dev_list.mutex); 1082 ini->ism_offered_cnt = i - 1; 1083 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1084 ini->smcd_version = 0; 1085 1086 return rc; 1087 } 1088 1089 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1090 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1091 struct smc_init_info *ini) 1092 { 1093 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1094 return SMC_CLC_DECL_ISMVLANERR; 1095 return 0; 1096 } 1097 1098 static int smc_find_proposal_devices(struct smc_sock *smc, 1099 struct smc_init_info *ini) 1100 { 1101 int rc = 0; 1102 1103 /* check if there is an ism device available */ 1104 if (!(ini->smcd_version & SMC_V1) || 1105 smc_find_ism_device(smc, ini) || 1106 smc_connect_ism_vlan_setup(smc, ini)) 1107 ini->smcd_version &= ~SMC_V1; 1108 /* else ISM V1 is supported for this connection */ 1109 1110 /* check if there is an rdma device available */ 1111 if (!(ini->smcr_version & SMC_V1) || 1112 smc_find_rdma_device(smc, ini)) 1113 ini->smcr_version &= ~SMC_V1; 1114 /* else RDMA is supported for this connection */ 1115 1116 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1117 ini->smcr_version & SMC_V1); 1118 1119 /* check if there is an ism v2 device available */ 1120 if (!(ini->smcd_version & SMC_V2) || 1121 !smc_ism_is_v2_capable() || 1122 smc_find_ism_v2_device_clnt(smc, ini)) 1123 ini->smcd_version &= ~SMC_V2; 1124 1125 /* check if there is an rdma v2 device available */ 1126 ini->check_smcrv2 = true; 1127 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1128 if (!(ini->smcr_version & SMC_V2) || 1129 smc->clcsock->sk->sk_family != AF_INET || 1130 !smc_clc_ueid_count() || 1131 smc_find_rdma_device(smc, ini)) 1132 ini->smcr_version &= ~SMC_V2; 1133 ini->check_smcrv2 = false; 1134 1135 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1136 ini->smcr_version & SMC_V2); 1137 1138 /* if neither ISM nor RDMA are supported, fallback */ 1139 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1140 rc = SMC_CLC_DECL_NOSMCDEV; 1141 1142 return rc; 1143 } 1144 1145 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1146 * used, the VLAN ID will be registered again during the connection setup. 1147 */ 1148 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1149 struct smc_init_info *ini) 1150 { 1151 if (!smcd_indicated(ini->smc_type_v1)) 1152 return 0; 1153 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1154 return SMC_CLC_DECL_CNFERR; 1155 return 0; 1156 } 1157 1158 #define SMC_CLC_MAX_ACCEPT_LEN \ 1159 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1160 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1161 sizeof(struct smc_clc_msg_trail)) 1162 1163 /* CLC handshake during connect */ 1164 static int smc_connect_clc(struct smc_sock *smc, 1165 struct smc_clc_msg_accept_confirm *aclc, 1166 struct smc_init_info *ini) 1167 { 1168 int rc = 0; 1169 1170 /* do inband token exchange */ 1171 rc = smc_clc_send_proposal(smc, ini); 1172 if (rc) 1173 return rc; 1174 /* receive SMC Accept CLC message */ 1175 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1176 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1177 } 1178 1179 void smc_fill_gid_list(struct smc_link_group *lgr, 1180 struct smc_gidlist *gidlist, 1181 struct smc_ib_device *known_dev, u8 *known_gid) 1182 { 1183 struct smc_init_info *alt_ini = NULL; 1184 1185 memset(gidlist, 0, sizeof(*gidlist)); 1186 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1187 1188 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1189 if (!alt_ini) 1190 goto out; 1191 1192 alt_ini->vlan_id = lgr->vlan_id; 1193 alt_ini->check_smcrv2 = true; 1194 alt_ini->smcrv2.saddr = lgr->saddr; 1195 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1196 1197 if (!alt_ini->smcrv2.ib_dev_v2) 1198 goto out; 1199 1200 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1201 SMC_GID_SIZE); 1202 1203 out: 1204 kfree(alt_ini); 1205 } 1206 1207 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1208 struct smc_clc_msg_accept_confirm *aclc, 1209 struct smc_init_info *ini) 1210 { 1211 struct smc_clc_first_contact_ext *fce = 1212 smc_get_clc_first_contact_ext(aclc, false); 1213 struct net *net = sock_net(&smc->sk); 1214 int rc; 1215 1216 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1217 return 0; 1218 1219 if (fce->v2_direct) { 1220 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1221 ini->smcrv2.uses_gateway = false; 1222 } else { 1223 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1224 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1225 ini->smcrv2.nexthop_mac, 1226 &ini->smcrv2.uses_gateway)) 1227 return SMC_CLC_DECL_NOROUTE; 1228 if (!ini->smcrv2.uses_gateway) { 1229 /* mismatch: peer claims indirect, but its direct */ 1230 return SMC_CLC_DECL_NOINDIRECT; 1231 } 1232 } 1233 1234 ini->release_nr = fce->release; 1235 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1236 if (rc) 1237 return rc; 1238 1239 return 0; 1240 } 1241 1242 /* setup for RDMA connection of client */ 1243 static int smc_connect_rdma(struct smc_sock *smc, 1244 struct smc_clc_msg_accept_confirm *aclc, 1245 struct smc_init_info *ini) 1246 { 1247 int i, reason_code = 0; 1248 struct smc_link *link; 1249 u8 *eid = NULL; 1250 1251 ini->is_smcd = false; 1252 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1253 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1254 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1255 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1256 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1257 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1258 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1259 1260 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1261 if (reason_code) 1262 return reason_code; 1263 1264 mutex_lock(&smc_client_lgr_pending); 1265 reason_code = smc_conn_create(smc, ini); 1266 if (reason_code) { 1267 mutex_unlock(&smc_client_lgr_pending); 1268 return reason_code; 1269 } 1270 1271 smc_conn_save_peer_info(smc, aclc); 1272 1273 if (ini->first_contact_local) { 1274 link = smc->conn.lnk; 1275 } else { 1276 /* set link that was assigned by server */ 1277 link = NULL; 1278 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1279 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1280 1281 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1282 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1283 SMC_GID_SIZE) && 1284 (aclc->hdr.version > SMC_V1 || 1285 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1286 sizeof(l->peer_mac)))) { 1287 link = l; 1288 break; 1289 } 1290 } 1291 if (!link) { 1292 reason_code = SMC_CLC_DECL_NOSRVLINK; 1293 goto connect_abort; 1294 } 1295 smc_switch_link_and_count(&smc->conn, link); 1296 } 1297 1298 /* create send buffer and rmb */ 1299 if (smc_buf_create(smc, false)) { 1300 reason_code = SMC_CLC_DECL_MEM; 1301 goto connect_abort; 1302 } 1303 1304 if (ini->first_contact_local) 1305 smc_link_save_peer_info(link, aclc, ini); 1306 1307 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1308 reason_code = SMC_CLC_DECL_ERR_RTOK; 1309 goto connect_abort; 1310 } 1311 1312 smc_close_init(smc); 1313 smc_rx_init(smc); 1314 1315 if (ini->first_contact_local) { 1316 if (smc_ib_ready_link(link)) { 1317 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1318 goto connect_abort; 1319 } 1320 } else { 1321 /* reg sendbufs if they were vzalloced */ 1322 if (smc->conn.sndbuf_desc->is_vm) { 1323 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1324 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1325 goto connect_abort; 1326 } 1327 } 1328 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1329 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1330 goto connect_abort; 1331 } 1332 } 1333 1334 if (aclc->hdr.version > SMC_V1) { 1335 eid = aclc->r1.eid; 1336 if (ini->first_contact_local) 1337 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1338 link->smcibdev, link->gid); 1339 } 1340 1341 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1342 aclc->hdr.version, eid, ini); 1343 if (reason_code) 1344 goto connect_abort; 1345 1346 smc_tx_init(smc); 1347 1348 if (ini->first_contact_local) { 1349 /* QP confirmation over RoCE fabric */ 1350 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1351 reason_code = smcr_clnt_conf_first_link(smc); 1352 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1353 if (reason_code) 1354 goto connect_abort; 1355 } 1356 mutex_unlock(&smc_client_lgr_pending); 1357 1358 smc_copy_sock_settings_to_clc(smc); 1359 smc->connect_nonblock = 0; 1360 if (smc->sk.sk_state == SMC_INIT) 1361 smc->sk.sk_state = SMC_ACTIVE; 1362 1363 return 0; 1364 connect_abort: 1365 smc_conn_abort(smc, ini->first_contact_local); 1366 mutex_unlock(&smc_client_lgr_pending); 1367 smc->connect_nonblock = 0; 1368 1369 return reason_code; 1370 } 1371 1372 /* The server has chosen one of the proposed ISM devices for the communication. 1373 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1374 */ 1375 static int 1376 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1377 struct smc_init_info *ini) 1378 { 1379 int i; 1380 1381 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1382 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1383 ini->ism_selected = i; 1384 return 0; 1385 } 1386 } 1387 1388 return -EPROTO; 1389 } 1390 1391 /* setup for ISM connection of client */ 1392 static int smc_connect_ism(struct smc_sock *smc, 1393 struct smc_clc_msg_accept_confirm *aclc, 1394 struct smc_init_info *ini) 1395 { 1396 u8 *eid = NULL; 1397 int rc = 0; 1398 1399 ini->is_smcd = true; 1400 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1401 1402 if (aclc->hdr.version == SMC_V2) { 1403 if (ini->first_contact_peer) { 1404 struct smc_clc_first_contact_ext *fce = 1405 smc_get_clc_first_contact_ext(aclc, true); 1406 1407 ini->release_nr = fce->release; 1408 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1409 if (rc) 1410 return rc; 1411 } 1412 1413 rc = smc_v2_determine_accepted_chid(aclc, ini); 1414 if (rc) 1415 return rc; 1416 1417 if (__smc_ism_is_virtual(ini->ism_chid[ini->ism_selected])) 1418 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1419 ntohll(aclc->d1.gid_ext); 1420 /* for non-virtual ISM devices, peer gid_ext remains 0. */ 1421 } 1422 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1423 1424 /* there is only one lgr role for SMC-D; use server lock */ 1425 mutex_lock(&smc_server_lgr_pending); 1426 rc = smc_conn_create(smc, ini); 1427 if (rc) { 1428 mutex_unlock(&smc_server_lgr_pending); 1429 return rc; 1430 } 1431 1432 /* Create send and receive buffers */ 1433 rc = smc_buf_create(smc, true); 1434 if (rc) { 1435 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1436 goto connect_abort; 1437 } 1438 1439 smc_conn_save_peer_info(smc, aclc); 1440 smc_close_init(smc); 1441 smc_rx_init(smc); 1442 smc_tx_init(smc); 1443 1444 if (aclc->hdr.version > SMC_V1) 1445 eid = aclc->d1.eid; 1446 1447 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1448 aclc->hdr.version, eid, ini); 1449 if (rc) 1450 goto connect_abort; 1451 mutex_unlock(&smc_server_lgr_pending); 1452 1453 smc_copy_sock_settings_to_clc(smc); 1454 smc->connect_nonblock = 0; 1455 if (smc->sk.sk_state == SMC_INIT) 1456 smc->sk.sk_state = SMC_ACTIVE; 1457 1458 return 0; 1459 connect_abort: 1460 smc_conn_abort(smc, ini->first_contact_local); 1461 mutex_unlock(&smc_server_lgr_pending); 1462 smc->connect_nonblock = 0; 1463 1464 return rc; 1465 } 1466 1467 /* check if received accept type and version matches a proposed one */ 1468 static int smc_connect_check_aclc(struct smc_init_info *ini, 1469 struct smc_clc_msg_accept_confirm *aclc) 1470 { 1471 if (aclc->hdr.typev1 != SMC_TYPE_R && 1472 aclc->hdr.typev1 != SMC_TYPE_D) 1473 return SMC_CLC_DECL_MODEUNSUPP; 1474 1475 if (aclc->hdr.version >= SMC_V2) { 1476 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1477 !smcr_indicated(ini->smc_type_v2)) || 1478 (aclc->hdr.typev1 == SMC_TYPE_D && 1479 !smcd_indicated(ini->smc_type_v2))) 1480 return SMC_CLC_DECL_MODEUNSUPP; 1481 } else { 1482 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1483 !smcr_indicated(ini->smc_type_v1)) || 1484 (aclc->hdr.typev1 == SMC_TYPE_D && 1485 !smcd_indicated(ini->smc_type_v1))) 1486 return SMC_CLC_DECL_MODEUNSUPP; 1487 } 1488 1489 return 0; 1490 } 1491 1492 /* perform steps before actually connecting */ 1493 static int __smc_connect(struct smc_sock *smc) 1494 { 1495 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1496 struct smc_clc_msg_accept_confirm *aclc; 1497 struct smc_init_info *ini = NULL; 1498 u8 *buf = NULL; 1499 int rc = 0; 1500 1501 if (smc->use_fallback) 1502 return smc_connect_fallback(smc, smc->fallback_rsn); 1503 1504 /* if peer has not signalled SMC-capability, fall back */ 1505 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1506 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1507 1508 /* IPSec connections opt out of SMC optimizations */ 1509 if (using_ipsec(smc)) 1510 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1511 version); 1512 1513 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1514 if (!ini) 1515 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1516 version); 1517 1518 ini->smcd_version = SMC_V1 | SMC_V2; 1519 ini->smcr_version = SMC_V1 | SMC_V2; 1520 ini->smc_type_v1 = SMC_TYPE_B; 1521 ini->smc_type_v2 = SMC_TYPE_B; 1522 1523 /* get vlan id from IP device */ 1524 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1525 ini->smcd_version &= ~SMC_V1; 1526 ini->smcr_version = 0; 1527 ini->smc_type_v1 = SMC_TYPE_N; 1528 if (!ini->smcd_version) { 1529 rc = SMC_CLC_DECL_GETVLANERR; 1530 goto fallback; 1531 } 1532 } 1533 1534 rc = smc_find_proposal_devices(smc, ini); 1535 if (rc) 1536 goto fallback; 1537 1538 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1539 if (!buf) { 1540 rc = SMC_CLC_DECL_MEM; 1541 goto fallback; 1542 } 1543 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1544 1545 /* perform CLC handshake */ 1546 rc = smc_connect_clc(smc, aclc, ini); 1547 if (rc) { 1548 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1549 if (rc == -EAGAIN) { 1550 rc = -ETIMEDOUT; 1551 smc->sk.sk_err = ETIMEDOUT; 1552 } 1553 goto vlan_cleanup; 1554 } 1555 1556 /* check if smc modes and versions of CLC proposal and accept match */ 1557 rc = smc_connect_check_aclc(ini, aclc); 1558 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1559 if (rc) 1560 goto vlan_cleanup; 1561 1562 /* depending on previous steps, connect using rdma or ism */ 1563 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1564 ini->smcr_version = version; 1565 rc = smc_connect_rdma(smc, aclc, ini); 1566 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1567 ini->smcd_version = version; 1568 rc = smc_connect_ism(smc, aclc, ini); 1569 } 1570 if (rc) 1571 goto vlan_cleanup; 1572 1573 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1574 smc_connect_ism_vlan_cleanup(smc, ini); 1575 kfree(buf); 1576 kfree(ini); 1577 return 0; 1578 1579 vlan_cleanup: 1580 smc_connect_ism_vlan_cleanup(smc, ini); 1581 kfree(buf); 1582 fallback: 1583 kfree(ini); 1584 return smc_connect_decline_fallback(smc, rc, version); 1585 } 1586 1587 static void smc_connect_work(struct work_struct *work) 1588 { 1589 struct smc_sock *smc = container_of(work, struct smc_sock, 1590 connect_work); 1591 long timeo = smc->sk.sk_sndtimeo; 1592 int rc = 0; 1593 1594 if (!timeo) 1595 timeo = MAX_SCHEDULE_TIMEOUT; 1596 lock_sock(smc->clcsock->sk); 1597 if (smc->clcsock->sk->sk_err) { 1598 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1599 } else if ((1 << smc->clcsock->sk->sk_state) & 1600 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1601 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1602 if ((rc == -EPIPE) && 1603 ((1 << smc->clcsock->sk->sk_state) & 1604 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1605 rc = 0; 1606 } 1607 release_sock(smc->clcsock->sk); 1608 lock_sock(&smc->sk); 1609 if (rc != 0 || smc->sk.sk_err) { 1610 smc->sk.sk_state = SMC_CLOSED; 1611 if (rc == -EPIPE || rc == -EAGAIN) 1612 smc->sk.sk_err = EPIPE; 1613 else if (rc == -ECONNREFUSED) 1614 smc->sk.sk_err = ECONNREFUSED; 1615 else if (signal_pending(current)) 1616 smc->sk.sk_err = -sock_intr_errno(timeo); 1617 sock_put(&smc->sk); /* passive closing */ 1618 goto out; 1619 } 1620 1621 rc = __smc_connect(smc); 1622 if (rc < 0) 1623 smc->sk.sk_err = -rc; 1624 1625 out: 1626 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1627 if (smc->sk.sk_err) { 1628 smc->sk.sk_state_change(&smc->sk); 1629 } else { /* allow polling before and after fallback decision */ 1630 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1631 smc->sk.sk_write_space(&smc->sk); 1632 } 1633 } 1634 release_sock(&smc->sk); 1635 } 1636 1637 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1638 int alen, int flags) 1639 { 1640 struct sock *sk = sock->sk; 1641 struct smc_sock *smc; 1642 int rc = -EINVAL; 1643 1644 smc = smc_sk(sk); 1645 1646 /* separate smc parameter checking to be safe */ 1647 if (alen < sizeof(addr->sa_family)) 1648 goto out_err; 1649 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1650 goto out_err; 1651 1652 lock_sock(sk); 1653 switch (sock->state) { 1654 default: 1655 rc = -EINVAL; 1656 goto out; 1657 case SS_CONNECTED: 1658 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1659 goto out; 1660 case SS_CONNECTING: 1661 if (sk->sk_state == SMC_ACTIVE) 1662 goto connected; 1663 break; 1664 case SS_UNCONNECTED: 1665 sock->state = SS_CONNECTING; 1666 break; 1667 } 1668 1669 switch (sk->sk_state) { 1670 default: 1671 goto out; 1672 case SMC_CLOSED: 1673 rc = sock_error(sk) ? : -ECONNABORTED; 1674 sock->state = SS_UNCONNECTED; 1675 goto out; 1676 case SMC_ACTIVE: 1677 rc = -EISCONN; 1678 goto out; 1679 case SMC_INIT: 1680 break; 1681 } 1682 1683 smc_copy_sock_settings_to_clc(smc); 1684 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1685 if (smc->connect_nonblock) { 1686 rc = -EALREADY; 1687 goto out; 1688 } 1689 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1690 if (rc && rc != -EINPROGRESS) 1691 goto out; 1692 1693 if (smc->use_fallback) { 1694 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1695 goto out; 1696 } 1697 sock_hold(&smc->sk); /* sock put in passive closing */ 1698 if (flags & O_NONBLOCK) { 1699 if (queue_work(smc_hs_wq, &smc->connect_work)) 1700 smc->connect_nonblock = 1; 1701 rc = -EINPROGRESS; 1702 goto out; 1703 } else { 1704 rc = __smc_connect(smc); 1705 if (rc < 0) 1706 goto out; 1707 } 1708 1709 connected: 1710 rc = 0; 1711 sock->state = SS_CONNECTED; 1712 out: 1713 release_sock(sk); 1714 out_err: 1715 return rc; 1716 } 1717 1718 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1719 { 1720 struct socket *new_clcsock = NULL; 1721 struct sock *lsk = &lsmc->sk; 1722 struct sock *new_sk; 1723 int rc = -EINVAL; 1724 1725 release_sock(lsk); 1726 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1727 if (!new_sk) { 1728 rc = -ENOMEM; 1729 lsk->sk_err = ENOMEM; 1730 *new_smc = NULL; 1731 lock_sock(lsk); 1732 goto out; 1733 } 1734 *new_smc = smc_sk(new_sk); 1735 1736 mutex_lock(&lsmc->clcsock_release_lock); 1737 if (lsmc->clcsock) 1738 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1739 mutex_unlock(&lsmc->clcsock_release_lock); 1740 lock_sock(lsk); 1741 if (rc < 0 && rc != -EAGAIN) 1742 lsk->sk_err = -rc; 1743 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1744 new_sk->sk_prot->unhash(new_sk); 1745 if (new_clcsock) 1746 sock_release(new_clcsock); 1747 new_sk->sk_state = SMC_CLOSED; 1748 smc_sock_set_flag(new_sk, SOCK_DEAD); 1749 sock_put(new_sk); /* final */ 1750 *new_smc = NULL; 1751 goto out; 1752 } 1753 1754 /* new clcsock has inherited the smc listen-specific sk_data_ready 1755 * function; switch it back to the original sk_data_ready function 1756 */ 1757 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1758 1759 /* if new clcsock has also inherited the fallback-specific callback 1760 * functions, switch them back to the original ones. 1761 */ 1762 if (lsmc->use_fallback) { 1763 if (lsmc->clcsk_state_change) 1764 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1765 if (lsmc->clcsk_write_space) 1766 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1767 if (lsmc->clcsk_error_report) 1768 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1769 } 1770 1771 (*new_smc)->clcsock = new_clcsock; 1772 out: 1773 return rc; 1774 } 1775 1776 /* add a just created sock to the accept queue of the listen sock as 1777 * candidate for a following socket accept call from user space 1778 */ 1779 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1780 { 1781 struct smc_sock *par = smc_sk(parent); 1782 1783 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1784 spin_lock(&par->accept_q_lock); 1785 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1786 spin_unlock(&par->accept_q_lock); 1787 sk_acceptq_added(parent); 1788 } 1789 1790 /* remove a socket from the accept queue of its parental listening socket */ 1791 static void smc_accept_unlink(struct sock *sk) 1792 { 1793 struct smc_sock *par = smc_sk(sk)->listen_smc; 1794 1795 spin_lock(&par->accept_q_lock); 1796 list_del_init(&smc_sk(sk)->accept_q); 1797 spin_unlock(&par->accept_q_lock); 1798 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1799 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1800 } 1801 1802 /* remove a sock from the accept queue to bind it to a new socket created 1803 * for a socket accept call from user space 1804 */ 1805 struct sock *smc_accept_dequeue(struct sock *parent, 1806 struct socket *new_sock) 1807 { 1808 struct smc_sock *isk, *n; 1809 struct sock *new_sk; 1810 1811 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1812 new_sk = (struct sock *)isk; 1813 1814 smc_accept_unlink(new_sk); 1815 if (new_sk->sk_state == SMC_CLOSED) { 1816 new_sk->sk_prot->unhash(new_sk); 1817 if (isk->clcsock) { 1818 sock_release(isk->clcsock); 1819 isk->clcsock = NULL; 1820 } 1821 sock_put(new_sk); /* final */ 1822 continue; 1823 } 1824 if (new_sock) { 1825 sock_graft(new_sk, new_sock); 1826 new_sock->state = SS_CONNECTED; 1827 if (isk->use_fallback) { 1828 smc_sk(new_sk)->clcsock->file = new_sock->file; 1829 isk->clcsock->file->private_data = isk->clcsock; 1830 } 1831 } 1832 return new_sk; 1833 } 1834 return NULL; 1835 } 1836 1837 /* clean up for a created but never accepted sock */ 1838 void smc_close_non_accepted(struct sock *sk) 1839 { 1840 struct smc_sock *smc = smc_sk(sk); 1841 1842 sock_hold(sk); /* sock_put below */ 1843 lock_sock(sk); 1844 if (!sk->sk_lingertime) 1845 /* wait for peer closing */ 1846 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1847 __smc_release(smc); 1848 release_sock(sk); 1849 sock_put(sk); /* sock_hold above */ 1850 sock_put(sk); /* final sock_put */ 1851 } 1852 1853 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1854 { 1855 struct smc_link *link = smc->conn.lnk; 1856 struct smc_llc_qentry *qentry; 1857 int rc; 1858 1859 /* reg the sndbuf if it was vzalloced*/ 1860 if (smc->conn.sndbuf_desc->is_vm) { 1861 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1862 return SMC_CLC_DECL_ERR_REGBUF; 1863 } 1864 1865 /* reg the rmb */ 1866 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1867 return SMC_CLC_DECL_ERR_REGBUF; 1868 1869 /* send CONFIRM LINK request to client over the RoCE fabric */ 1870 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1871 if (rc < 0) 1872 return SMC_CLC_DECL_TIMEOUT_CL; 1873 1874 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1875 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1876 SMC_LLC_CONFIRM_LINK); 1877 if (!qentry) { 1878 struct smc_clc_msg_decline dclc; 1879 1880 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1881 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1882 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1883 } 1884 smc_llc_save_peer_uid(qentry); 1885 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1886 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1887 if (rc) 1888 return SMC_CLC_DECL_RMBE_EC; 1889 1890 /* confirm_rkey is implicit on 1st contact */ 1891 smc->conn.rmb_desc->is_conf_rkey = true; 1892 1893 smc_llc_link_active(link); 1894 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1895 1896 if (link->lgr->max_links > 1) { 1897 down_write(&link->lgr->llc_conf_mutex); 1898 /* initial contact - try to establish second link */ 1899 smc_llc_srv_add_link(link, NULL); 1900 up_write(&link->lgr->llc_conf_mutex); 1901 } 1902 return 0; 1903 } 1904 1905 /* listen worker: finish */ 1906 static void smc_listen_out(struct smc_sock *new_smc) 1907 { 1908 struct smc_sock *lsmc = new_smc->listen_smc; 1909 struct sock *newsmcsk = &new_smc->sk; 1910 1911 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1912 atomic_dec(&lsmc->queued_smc_hs); 1913 1914 if (lsmc->sk.sk_state == SMC_LISTEN) { 1915 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1916 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1917 release_sock(&lsmc->sk); 1918 } else { /* no longer listening */ 1919 smc_close_non_accepted(newsmcsk); 1920 } 1921 1922 /* Wake up accept */ 1923 lsmc->sk.sk_data_ready(&lsmc->sk); 1924 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1925 } 1926 1927 /* listen worker: finish in state connected */ 1928 static void smc_listen_out_connected(struct smc_sock *new_smc) 1929 { 1930 struct sock *newsmcsk = &new_smc->sk; 1931 1932 if (newsmcsk->sk_state == SMC_INIT) 1933 newsmcsk->sk_state = SMC_ACTIVE; 1934 1935 smc_listen_out(new_smc); 1936 } 1937 1938 /* listen worker: finish in error state */ 1939 static void smc_listen_out_err(struct smc_sock *new_smc) 1940 { 1941 struct sock *newsmcsk = &new_smc->sk; 1942 struct net *net = sock_net(newsmcsk); 1943 1944 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1945 if (newsmcsk->sk_state == SMC_INIT) 1946 sock_put(&new_smc->sk); /* passive closing */ 1947 newsmcsk->sk_state = SMC_CLOSED; 1948 1949 smc_listen_out(new_smc); 1950 } 1951 1952 /* listen worker: decline and fall back if possible */ 1953 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1954 int local_first, u8 version) 1955 { 1956 /* RDMA setup failed, switch back to TCP */ 1957 smc_conn_abort(new_smc, local_first); 1958 if (reason_code < 0 || 1959 smc_switch_to_fallback(new_smc, reason_code)) { 1960 /* error, no fallback possible */ 1961 smc_listen_out_err(new_smc); 1962 return; 1963 } 1964 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1965 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1966 smc_listen_out_err(new_smc); 1967 return; 1968 } 1969 } 1970 smc_listen_out_connected(new_smc); 1971 } 1972 1973 /* listen worker: version checking */ 1974 static int smc_listen_v2_check(struct smc_sock *new_smc, 1975 struct smc_clc_msg_proposal *pclc, 1976 struct smc_init_info *ini) 1977 { 1978 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1979 struct smc_clc_v2_extension *pclc_v2_ext; 1980 int rc = SMC_CLC_DECL_PEERNOSMC; 1981 1982 ini->smc_type_v1 = pclc->hdr.typev1; 1983 ini->smc_type_v2 = pclc->hdr.typev2; 1984 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1985 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1986 if (pclc->hdr.version > SMC_V1) { 1987 if (smcd_indicated(ini->smc_type_v2)) 1988 ini->smcd_version |= SMC_V2; 1989 if (smcr_indicated(ini->smc_type_v2)) 1990 ini->smcr_version |= SMC_V2; 1991 } 1992 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1993 rc = SMC_CLC_DECL_PEERNOSMC; 1994 goto out; 1995 } 1996 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1997 if (!pclc_v2_ext) { 1998 ini->smcd_version &= ~SMC_V2; 1999 ini->smcr_version &= ~SMC_V2; 2000 rc = SMC_CLC_DECL_NOV2EXT; 2001 goto out; 2002 } 2003 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2004 if (ini->smcd_version & SMC_V2) { 2005 if (!smc_ism_is_v2_capable()) { 2006 ini->smcd_version &= ~SMC_V2; 2007 rc = SMC_CLC_DECL_NOISM2SUPP; 2008 } else if (!pclc_smcd_v2_ext) { 2009 ini->smcd_version &= ~SMC_V2; 2010 rc = SMC_CLC_DECL_NOV2DEXT; 2011 } else if (!pclc_v2_ext->hdr.eid_cnt && 2012 !pclc_v2_ext->hdr.flag.seid) { 2013 ini->smcd_version &= ~SMC_V2; 2014 rc = SMC_CLC_DECL_NOUEID; 2015 } 2016 } 2017 if (ini->smcr_version & SMC_V2) { 2018 if (!pclc_v2_ext->hdr.eid_cnt) { 2019 ini->smcr_version &= ~SMC_V2; 2020 rc = SMC_CLC_DECL_NOUEID; 2021 } 2022 } 2023 2024 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2025 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2026 ini->release_nr = SMC_RELEASE; 2027 2028 out: 2029 if (!ini->smcd_version && !ini->smcr_version) 2030 return rc; 2031 2032 return 0; 2033 } 2034 2035 /* listen worker: check prefixes */ 2036 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2037 struct smc_clc_msg_proposal *pclc) 2038 { 2039 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2040 struct socket *newclcsock = new_smc->clcsock; 2041 2042 if (pclc->hdr.typev1 == SMC_TYPE_N) 2043 return 0; 2044 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2045 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2046 return SMC_CLC_DECL_DIFFPREFIX; 2047 2048 return 0; 2049 } 2050 2051 /* listen worker: initialize connection and buffers */ 2052 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2053 struct smc_init_info *ini) 2054 { 2055 int rc; 2056 2057 /* allocate connection / link group */ 2058 rc = smc_conn_create(new_smc, ini); 2059 if (rc) 2060 return rc; 2061 2062 /* create send buffer and rmb */ 2063 if (smc_buf_create(new_smc, false)) { 2064 smc_conn_abort(new_smc, ini->first_contact_local); 2065 return SMC_CLC_DECL_MEM; 2066 } 2067 2068 return 0; 2069 } 2070 2071 /* listen worker: initialize connection and buffers for SMC-D */ 2072 static int smc_listen_ism_init(struct smc_sock *new_smc, 2073 struct smc_init_info *ini) 2074 { 2075 int rc; 2076 2077 rc = smc_conn_create(new_smc, ini); 2078 if (rc) 2079 return rc; 2080 2081 /* Create send and receive buffers */ 2082 rc = smc_buf_create(new_smc, true); 2083 if (rc) { 2084 smc_conn_abort(new_smc, ini->first_contact_local); 2085 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2086 SMC_CLC_DECL_MEM; 2087 } 2088 2089 return 0; 2090 } 2091 2092 static bool smc_is_already_selected(struct smcd_dev *smcd, 2093 struct smc_init_info *ini, 2094 int matches) 2095 { 2096 int i; 2097 2098 for (i = 0; i < matches; i++) 2099 if (smcd == ini->ism_dev[i]) 2100 return true; 2101 2102 return false; 2103 } 2104 2105 /* check for ISM devices matching proposed ISM devices */ 2106 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2107 u16 proposed_chid, 2108 struct smcd_gid *proposed_gid, 2109 unsigned int *matches) 2110 { 2111 struct smcd_dev *smcd; 2112 2113 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2114 if (smcd->going_away) 2115 continue; 2116 if (smc_is_already_selected(smcd, ini, *matches)) 2117 continue; 2118 if (smc_ism_get_chid(smcd) == proposed_chid && 2119 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2120 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2121 if (__smc_ism_is_virtual(proposed_chid)) 2122 ini->ism_peer_gid[*matches].gid_ext = 2123 proposed_gid->gid_ext; 2124 /* non-virtual ISM's peer gid_ext remains 0. */ 2125 ini->ism_dev[*matches] = smcd; 2126 (*matches)++; 2127 break; 2128 } 2129 } 2130 } 2131 2132 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2133 { 2134 if (!ini->rc) 2135 ini->rc = rc; 2136 } 2137 2138 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2139 struct smc_clc_msg_proposal *pclc, 2140 struct smc_init_info *ini) 2141 { 2142 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2143 struct smc_clc_v2_extension *smc_v2_ext; 2144 struct smc_clc_msg_smcd *pclc_smcd; 2145 unsigned int matches = 0; 2146 struct smcd_gid smcd_gid; 2147 u8 smcd_version; 2148 u8 *eid = NULL; 2149 int i, rc; 2150 u16 chid; 2151 2152 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2153 goto not_found; 2154 2155 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2156 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2157 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2158 2159 mutex_lock(&smcd_dev_list.mutex); 2160 if (pclc_smcd->ism.chid) { 2161 /* check for ISM device matching proposed native ISM device */ 2162 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2163 smcd_gid.gid_ext = 0; 2164 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2165 &smcd_gid, &matches); 2166 } 2167 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2168 /* check for ISM devices matching proposed non-native ISM 2169 * devices 2170 */ 2171 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2172 smcd_gid.gid_ext = 0; 2173 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2174 if (__smc_ism_is_virtual(chid)) { 2175 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2176 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2177 /* each virtual ISM device takes two GID-CHID 2178 * entries and CHID of the second entry repeats 2179 * that of the first entry. 2180 * 2181 * So check if the next GID-CHID entry exists 2182 * and both two entries' CHIDs are the same. 2183 */ 2184 continue; 2185 smcd_gid.gid_ext = 2186 ntohll(smcd_v2_ext->gidchid[++i].gid); 2187 } 2188 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2189 } 2190 mutex_unlock(&smcd_dev_list.mutex); 2191 2192 if (!ini->ism_dev[0]) { 2193 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2194 goto not_found; 2195 } 2196 2197 smc_ism_get_system_eid(&eid); 2198 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2199 smcd_v2_ext->system_eid, eid)) 2200 goto not_found; 2201 2202 /* separate - outside the smcd_dev_list.lock */ 2203 smcd_version = ini->smcd_version; 2204 for (i = 0; i < matches; i++) { 2205 ini->smcd_version = SMC_V2; 2206 ini->is_smcd = true; 2207 ini->ism_selected = i; 2208 rc = smc_listen_ism_init(new_smc, ini); 2209 if (rc) { 2210 smc_find_ism_store_rc(rc, ini); 2211 /* try next active ISM device */ 2212 continue; 2213 } 2214 return; /* matching and usable V2 ISM device found */ 2215 } 2216 /* no V2 ISM device could be initialized */ 2217 ini->smcd_version = smcd_version; /* restore original value */ 2218 ini->negotiated_eid[0] = 0; 2219 2220 not_found: 2221 ini->smcd_version &= ~SMC_V2; 2222 ini->ism_dev[0] = NULL; 2223 ini->is_smcd = false; 2224 } 2225 2226 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2227 struct smc_clc_msg_proposal *pclc, 2228 struct smc_init_info *ini) 2229 { 2230 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2231 int rc = 0; 2232 2233 /* check if ISM V1 is available */ 2234 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2235 goto not_found; 2236 ini->is_smcd = true; /* prepare ISM check */ 2237 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2238 ini->ism_peer_gid[0].gid_ext = 0; 2239 rc = smc_find_ism_device(new_smc, ini); 2240 if (rc) 2241 goto not_found; 2242 ini->ism_selected = 0; 2243 rc = smc_listen_ism_init(new_smc, ini); 2244 if (!rc) 2245 return; /* V1 ISM device found */ 2246 2247 not_found: 2248 smc_find_ism_store_rc(rc, ini); 2249 ini->smcd_version &= ~SMC_V1; 2250 ini->ism_dev[0] = NULL; 2251 ini->is_smcd = false; 2252 } 2253 2254 /* listen worker: register buffers */ 2255 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2256 { 2257 struct smc_connection *conn = &new_smc->conn; 2258 2259 if (!local_first) { 2260 /* reg sendbufs if they were vzalloced */ 2261 if (conn->sndbuf_desc->is_vm) { 2262 if (smcr_lgr_reg_sndbufs(conn->lnk, 2263 conn->sndbuf_desc)) 2264 return SMC_CLC_DECL_ERR_REGBUF; 2265 } 2266 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2267 return SMC_CLC_DECL_ERR_REGBUF; 2268 } 2269 2270 return 0; 2271 } 2272 2273 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2274 struct smc_clc_msg_proposal *pclc, 2275 struct smc_init_info *ini) 2276 { 2277 struct smc_clc_v2_extension *smc_v2_ext; 2278 u8 smcr_version; 2279 int rc; 2280 2281 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2282 goto not_found; 2283 2284 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2285 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2286 goto not_found; 2287 2288 /* prepare RDMA check */ 2289 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2290 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2291 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2292 ini->check_smcrv2 = true; 2293 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2294 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2295 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2296 rc = smc_find_rdma_device(new_smc, ini); 2297 if (rc) { 2298 smc_find_ism_store_rc(rc, ini); 2299 goto not_found; 2300 } 2301 if (!ini->smcrv2.uses_gateway) 2302 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2303 2304 smcr_version = ini->smcr_version; 2305 ini->smcr_version = SMC_V2; 2306 rc = smc_listen_rdma_init(new_smc, ini); 2307 if (!rc) { 2308 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2309 if (rc) 2310 smc_conn_abort(new_smc, ini->first_contact_local); 2311 } 2312 if (!rc) 2313 return; 2314 ini->smcr_version = smcr_version; 2315 smc_find_ism_store_rc(rc, ini); 2316 2317 not_found: 2318 ini->smcr_version &= ~SMC_V2; 2319 ini->smcrv2.ib_dev_v2 = NULL; 2320 ini->check_smcrv2 = false; 2321 } 2322 2323 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2324 struct smc_clc_msg_proposal *pclc, 2325 struct smc_init_info *ini) 2326 { 2327 int rc; 2328 2329 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2330 return SMC_CLC_DECL_NOSMCDEV; 2331 2332 /* prepare RDMA check */ 2333 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2334 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2335 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2336 rc = smc_find_rdma_device(new_smc, ini); 2337 if (rc) { 2338 /* no RDMA device found */ 2339 return SMC_CLC_DECL_NOSMCDEV; 2340 } 2341 rc = smc_listen_rdma_init(new_smc, ini); 2342 if (rc) 2343 return rc; 2344 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2345 } 2346 2347 /* determine the local device matching to proposal */ 2348 static int smc_listen_find_device(struct smc_sock *new_smc, 2349 struct smc_clc_msg_proposal *pclc, 2350 struct smc_init_info *ini) 2351 { 2352 int prfx_rc; 2353 2354 /* check for ISM device matching V2 proposed device */ 2355 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2356 if (ini->ism_dev[0]) 2357 return 0; 2358 2359 /* check for matching IP prefix and subnet length (V1) */ 2360 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2361 if (prfx_rc) 2362 smc_find_ism_store_rc(prfx_rc, ini); 2363 2364 /* get vlan id from IP device */ 2365 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2366 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2367 2368 /* check for ISM device matching V1 proposed device */ 2369 if (!prfx_rc) 2370 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2371 if (ini->ism_dev[0]) 2372 return 0; 2373 2374 if (!smcr_indicated(pclc->hdr.typev1) && 2375 !smcr_indicated(pclc->hdr.typev2)) 2376 /* skip RDMA and decline */ 2377 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2378 2379 /* check if RDMA V2 is available */ 2380 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2381 if (ini->smcrv2.ib_dev_v2) 2382 return 0; 2383 2384 /* check if RDMA V1 is available */ 2385 if (!prfx_rc) { 2386 int rc; 2387 2388 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2389 smc_find_ism_store_rc(rc, ini); 2390 return (!rc) ? 0 : ini->rc; 2391 } 2392 return prfx_rc; 2393 } 2394 2395 /* listen worker: finish RDMA setup */ 2396 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2397 struct smc_clc_msg_accept_confirm *cclc, 2398 bool local_first, 2399 struct smc_init_info *ini) 2400 { 2401 struct smc_link *link = new_smc->conn.lnk; 2402 int reason_code = 0; 2403 2404 if (local_first) 2405 smc_link_save_peer_info(link, cclc, ini); 2406 2407 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2408 return SMC_CLC_DECL_ERR_RTOK; 2409 2410 if (local_first) { 2411 if (smc_ib_ready_link(link)) 2412 return SMC_CLC_DECL_ERR_RDYLNK; 2413 /* QP confirmation over RoCE fabric */ 2414 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2415 reason_code = smcr_serv_conf_first_link(new_smc); 2416 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2417 } 2418 return reason_code; 2419 } 2420 2421 /* setup for connection of server */ 2422 static void smc_listen_work(struct work_struct *work) 2423 { 2424 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2425 smc_listen_work); 2426 struct socket *newclcsock = new_smc->clcsock; 2427 struct smc_clc_msg_accept_confirm *cclc; 2428 struct smc_clc_msg_proposal_area *buf; 2429 struct smc_clc_msg_proposal *pclc; 2430 struct smc_init_info *ini = NULL; 2431 u8 proposal_version = SMC_V1; 2432 u8 accept_version; 2433 int rc = 0; 2434 2435 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2436 return smc_listen_out_err(new_smc); 2437 2438 if (new_smc->use_fallback) { 2439 smc_listen_out_connected(new_smc); 2440 return; 2441 } 2442 2443 /* check if peer is smc capable */ 2444 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2445 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2446 if (rc) 2447 smc_listen_out_err(new_smc); 2448 else 2449 smc_listen_out_connected(new_smc); 2450 return; 2451 } 2452 2453 /* do inband token exchange - 2454 * wait for and receive SMC Proposal CLC message 2455 */ 2456 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2457 if (!buf) { 2458 rc = SMC_CLC_DECL_MEM; 2459 goto out_decl; 2460 } 2461 pclc = (struct smc_clc_msg_proposal *)buf; 2462 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2463 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2464 if (rc) 2465 goto out_decl; 2466 2467 if (pclc->hdr.version > SMC_V1) 2468 proposal_version = SMC_V2; 2469 2470 /* IPSec connections opt out of SMC optimizations */ 2471 if (using_ipsec(new_smc)) { 2472 rc = SMC_CLC_DECL_IPSEC; 2473 goto out_decl; 2474 } 2475 2476 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2477 if (!ini) { 2478 rc = SMC_CLC_DECL_MEM; 2479 goto out_decl; 2480 } 2481 2482 /* initial version checking */ 2483 rc = smc_listen_v2_check(new_smc, pclc, ini); 2484 if (rc) 2485 goto out_decl; 2486 2487 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2488 if (rc) 2489 goto out_decl; 2490 2491 mutex_lock(&smc_server_lgr_pending); 2492 smc_close_init(new_smc); 2493 smc_rx_init(new_smc); 2494 smc_tx_init(new_smc); 2495 2496 /* determine ISM or RoCE device used for connection */ 2497 rc = smc_listen_find_device(new_smc, pclc, ini); 2498 if (rc) 2499 goto out_unlock; 2500 2501 /* send SMC Accept CLC message */ 2502 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2503 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2504 accept_version, ini->negotiated_eid, ini); 2505 if (rc) 2506 goto out_unlock; 2507 2508 /* SMC-D does not need this lock any more */ 2509 if (ini->is_smcd) 2510 mutex_unlock(&smc_server_lgr_pending); 2511 2512 /* receive SMC Confirm CLC message */ 2513 memset(buf, 0, sizeof(*buf)); 2514 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2515 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2516 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2517 if (rc) { 2518 if (!ini->is_smcd) 2519 goto out_unlock; 2520 goto out_decl; 2521 } 2522 2523 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2524 if (rc) { 2525 if (!ini->is_smcd) 2526 goto out_unlock; 2527 goto out_decl; 2528 } 2529 2530 /* fce smc release version is needed in smc_listen_rdma_finish, 2531 * so save fce info here. 2532 */ 2533 smc_conn_save_peer_info_fce(new_smc, cclc); 2534 2535 /* finish worker */ 2536 if (!ini->is_smcd) { 2537 rc = smc_listen_rdma_finish(new_smc, cclc, 2538 ini->first_contact_local, ini); 2539 if (rc) 2540 goto out_unlock; 2541 mutex_unlock(&smc_server_lgr_pending); 2542 } 2543 smc_conn_save_peer_info(new_smc, cclc); 2544 smc_listen_out_connected(new_smc); 2545 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2546 goto out_free; 2547 2548 out_unlock: 2549 mutex_unlock(&smc_server_lgr_pending); 2550 out_decl: 2551 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2552 proposal_version); 2553 out_free: 2554 kfree(ini); 2555 kfree(buf); 2556 } 2557 2558 static void smc_tcp_listen_work(struct work_struct *work) 2559 { 2560 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2561 tcp_listen_work); 2562 struct sock *lsk = &lsmc->sk; 2563 struct smc_sock *new_smc; 2564 int rc = 0; 2565 2566 lock_sock(lsk); 2567 while (lsk->sk_state == SMC_LISTEN) { 2568 rc = smc_clcsock_accept(lsmc, &new_smc); 2569 if (rc) /* clcsock accept queue empty or error */ 2570 goto out; 2571 if (!new_smc) 2572 continue; 2573 2574 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2575 atomic_inc(&lsmc->queued_smc_hs); 2576 2577 new_smc->listen_smc = lsmc; 2578 new_smc->use_fallback = lsmc->use_fallback; 2579 new_smc->fallback_rsn = lsmc->fallback_rsn; 2580 sock_hold(lsk); /* sock_put in smc_listen_work */ 2581 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2582 smc_copy_sock_settings_to_smc(new_smc); 2583 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2584 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2585 sock_put(&new_smc->sk); 2586 } 2587 2588 out: 2589 release_sock(lsk); 2590 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2591 } 2592 2593 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2594 { 2595 struct smc_sock *lsmc; 2596 2597 read_lock_bh(&listen_clcsock->sk_callback_lock); 2598 lsmc = smc_clcsock_user_data(listen_clcsock); 2599 if (!lsmc) 2600 goto out; 2601 lsmc->clcsk_data_ready(listen_clcsock); 2602 if (lsmc->sk.sk_state == SMC_LISTEN) { 2603 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2604 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2605 sock_put(&lsmc->sk); 2606 } 2607 out: 2608 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2609 } 2610 2611 static int smc_listen(struct socket *sock, int backlog) 2612 { 2613 struct sock *sk = sock->sk; 2614 struct smc_sock *smc; 2615 int rc; 2616 2617 smc = smc_sk(sk); 2618 lock_sock(sk); 2619 2620 rc = -EINVAL; 2621 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2622 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2623 goto out; 2624 2625 rc = 0; 2626 if (sk->sk_state == SMC_LISTEN) { 2627 sk->sk_max_ack_backlog = backlog; 2628 goto out; 2629 } 2630 /* some socket options are handled in core, so we could not apply 2631 * them to the clc socket -- copy smc socket options to clc socket 2632 */ 2633 smc_copy_sock_settings_to_clc(smc); 2634 if (!smc->use_fallback) 2635 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2636 2637 /* save original sk_data_ready function and establish 2638 * smc-specific sk_data_ready function 2639 */ 2640 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2641 smc->clcsock->sk->sk_user_data = 2642 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2643 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2644 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2645 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2646 2647 /* save original ops */ 2648 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2649 2650 smc->af_ops = *smc->ori_af_ops; 2651 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2652 2653 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2654 2655 if (smc->limit_smc_hs) 2656 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2657 2658 rc = kernel_listen(smc->clcsock, backlog); 2659 if (rc) { 2660 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2661 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2662 &smc->clcsk_data_ready); 2663 smc->clcsock->sk->sk_user_data = NULL; 2664 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2665 goto out; 2666 } 2667 sk->sk_max_ack_backlog = backlog; 2668 sk->sk_ack_backlog = 0; 2669 sk->sk_state = SMC_LISTEN; 2670 2671 out: 2672 release_sock(sk); 2673 return rc; 2674 } 2675 2676 static int smc_accept(struct socket *sock, struct socket *new_sock, 2677 int flags, bool kern) 2678 { 2679 struct sock *sk = sock->sk, *nsk; 2680 DECLARE_WAITQUEUE(wait, current); 2681 struct smc_sock *lsmc; 2682 long timeo; 2683 int rc = 0; 2684 2685 lsmc = smc_sk(sk); 2686 sock_hold(sk); /* sock_put below */ 2687 lock_sock(sk); 2688 2689 if (lsmc->sk.sk_state != SMC_LISTEN) { 2690 rc = -EINVAL; 2691 release_sock(sk); 2692 goto out; 2693 } 2694 2695 /* Wait for an incoming connection */ 2696 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2697 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2698 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2699 set_current_state(TASK_INTERRUPTIBLE); 2700 if (!timeo) { 2701 rc = -EAGAIN; 2702 break; 2703 } 2704 release_sock(sk); 2705 timeo = schedule_timeout(timeo); 2706 /* wakeup by sk_data_ready in smc_listen_work() */ 2707 sched_annotate_sleep(); 2708 lock_sock(sk); 2709 if (signal_pending(current)) { 2710 rc = sock_intr_errno(timeo); 2711 break; 2712 } 2713 } 2714 set_current_state(TASK_RUNNING); 2715 remove_wait_queue(sk_sleep(sk), &wait); 2716 2717 if (!rc) 2718 rc = sock_error(nsk); 2719 release_sock(sk); 2720 if (rc) 2721 goto out; 2722 2723 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2724 /* wait till data arrives on the socket */ 2725 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2726 MSEC_PER_SEC); 2727 if (smc_sk(nsk)->use_fallback) { 2728 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2729 2730 lock_sock(clcsk); 2731 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2732 sk_wait_data(clcsk, &timeo, NULL); 2733 release_sock(clcsk); 2734 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2735 lock_sock(nsk); 2736 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2737 release_sock(nsk); 2738 } 2739 } 2740 2741 out: 2742 sock_put(sk); /* sock_hold above */ 2743 return rc; 2744 } 2745 2746 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2747 int peer) 2748 { 2749 struct smc_sock *smc; 2750 2751 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2752 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2753 return -ENOTCONN; 2754 2755 smc = smc_sk(sock->sk); 2756 2757 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2758 } 2759 2760 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2761 { 2762 struct sock *sk = sock->sk; 2763 struct smc_sock *smc; 2764 int rc; 2765 2766 smc = smc_sk(sk); 2767 lock_sock(sk); 2768 2769 /* SMC does not support connect with fastopen */ 2770 if (msg->msg_flags & MSG_FASTOPEN) { 2771 /* not connected yet, fallback */ 2772 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2773 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2774 if (rc) 2775 goto out; 2776 } else { 2777 rc = -EINVAL; 2778 goto out; 2779 } 2780 } else if ((sk->sk_state != SMC_ACTIVE) && 2781 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2782 (sk->sk_state != SMC_INIT)) { 2783 rc = -EPIPE; 2784 goto out; 2785 } 2786 2787 if (smc->use_fallback) { 2788 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2789 } else { 2790 rc = smc_tx_sendmsg(smc, msg, len); 2791 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2792 } 2793 out: 2794 release_sock(sk); 2795 return rc; 2796 } 2797 2798 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2799 int flags) 2800 { 2801 struct sock *sk = sock->sk; 2802 struct smc_sock *smc; 2803 int rc = -ENOTCONN; 2804 2805 smc = smc_sk(sk); 2806 lock_sock(sk); 2807 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2808 /* socket was connected before, no more data to read */ 2809 rc = 0; 2810 goto out; 2811 } 2812 if ((sk->sk_state == SMC_INIT) || 2813 (sk->sk_state == SMC_LISTEN) || 2814 (sk->sk_state == SMC_CLOSED)) 2815 goto out; 2816 2817 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2818 rc = 0; 2819 goto out; 2820 } 2821 2822 if (smc->use_fallback) { 2823 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2824 } else { 2825 msg->msg_namelen = 0; 2826 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2827 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2828 } 2829 2830 out: 2831 release_sock(sk); 2832 return rc; 2833 } 2834 2835 static __poll_t smc_accept_poll(struct sock *parent) 2836 { 2837 struct smc_sock *isk = smc_sk(parent); 2838 __poll_t mask = 0; 2839 2840 spin_lock(&isk->accept_q_lock); 2841 if (!list_empty(&isk->accept_q)) 2842 mask = EPOLLIN | EPOLLRDNORM; 2843 spin_unlock(&isk->accept_q_lock); 2844 2845 return mask; 2846 } 2847 2848 static __poll_t smc_poll(struct file *file, struct socket *sock, 2849 poll_table *wait) 2850 { 2851 struct sock *sk = sock->sk; 2852 struct smc_sock *smc; 2853 __poll_t mask = 0; 2854 2855 if (!sk) 2856 return EPOLLNVAL; 2857 2858 smc = smc_sk(sock->sk); 2859 if (smc->use_fallback) { 2860 /* delegate to CLC child sock */ 2861 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2862 sk->sk_err = smc->clcsock->sk->sk_err; 2863 } else { 2864 if (sk->sk_state != SMC_CLOSED) 2865 sock_poll_wait(file, sock, wait); 2866 if (sk->sk_err) 2867 mask |= EPOLLERR; 2868 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2869 (sk->sk_state == SMC_CLOSED)) 2870 mask |= EPOLLHUP; 2871 if (sk->sk_state == SMC_LISTEN) { 2872 /* woken up by sk_data_ready in smc_listen_work() */ 2873 mask |= smc_accept_poll(sk); 2874 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2875 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2876 wait); 2877 sk->sk_err = smc->clcsock->sk->sk_err; 2878 } else { 2879 if ((sk->sk_state != SMC_INIT && 2880 atomic_read(&smc->conn.sndbuf_space)) || 2881 sk->sk_shutdown & SEND_SHUTDOWN) { 2882 mask |= EPOLLOUT | EPOLLWRNORM; 2883 } else { 2884 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2885 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2886 } 2887 if (atomic_read(&smc->conn.bytes_to_rcv)) 2888 mask |= EPOLLIN | EPOLLRDNORM; 2889 if (sk->sk_shutdown & RCV_SHUTDOWN) 2890 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2891 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2892 mask |= EPOLLIN; 2893 if (smc->conn.urg_state == SMC_URG_VALID) 2894 mask |= EPOLLPRI; 2895 } 2896 } 2897 2898 return mask; 2899 } 2900 2901 static int smc_shutdown(struct socket *sock, int how) 2902 { 2903 struct sock *sk = sock->sk; 2904 bool do_shutdown = true; 2905 struct smc_sock *smc; 2906 int rc = -EINVAL; 2907 int old_state; 2908 int rc1 = 0; 2909 2910 smc = smc_sk(sk); 2911 2912 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2913 return rc; 2914 2915 lock_sock(sk); 2916 2917 if (sock->state == SS_CONNECTING) { 2918 if (sk->sk_state == SMC_ACTIVE) 2919 sock->state = SS_CONNECTED; 2920 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2921 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2922 sk->sk_state == SMC_APPCLOSEWAIT1 || 2923 sk->sk_state == SMC_APPCLOSEWAIT2 || 2924 sk->sk_state == SMC_APPFINCLOSEWAIT) 2925 sock->state = SS_DISCONNECTING; 2926 } 2927 2928 rc = -ENOTCONN; 2929 if ((sk->sk_state != SMC_ACTIVE) && 2930 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2931 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2932 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2933 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2934 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2935 goto out; 2936 if (smc->use_fallback) { 2937 rc = kernel_sock_shutdown(smc->clcsock, how); 2938 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2939 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2940 sk->sk_state = SMC_CLOSED; 2941 sk->sk_socket->state = SS_UNCONNECTED; 2942 sock_put(sk); 2943 } 2944 goto out; 2945 } 2946 switch (how) { 2947 case SHUT_RDWR: /* shutdown in both directions */ 2948 old_state = sk->sk_state; 2949 rc = smc_close_active(smc); 2950 if (old_state == SMC_ACTIVE && 2951 sk->sk_state == SMC_PEERCLOSEWAIT1) 2952 do_shutdown = false; 2953 break; 2954 case SHUT_WR: 2955 rc = smc_close_shutdown_write(smc); 2956 break; 2957 case SHUT_RD: 2958 rc = 0; 2959 /* nothing more to do because peer is not involved */ 2960 break; 2961 } 2962 if (do_shutdown && smc->clcsock) 2963 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2964 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2965 sk->sk_shutdown |= how + 1; 2966 2967 if (sk->sk_state == SMC_CLOSED) 2968 sock->state = SS_UNCONNECTED; 2969 else 2970 sock->state = SS_DISCONNECTING; 2971 out: 2972 release_sock(sk); 2973 return rc ? rc : rc1; 2974 } 2975 2976 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2977 char __user *optval, int __user *optlen) 2978 { 2979 struct smc_sock *smc; 2980 int val, len; 2981 2982 smc = smc_sk(sock->sk); 2983 2984 if (get_user(len, optlen)) 2985 return -EFAULT; 2986 2987 len = min_t(int, len, sizeof(int)); 2988 2989 if (len < 0) 2990 return -EINVAL; 2991 2992 switch (optname) { 2993 case SMC_LIMIT_HS: 2994 val = smc->limit_smc_hs; 2995 break; 2996 default: 2997 return -EOPNOTSUPP; 2998 } 2999 3000 if (put_user(len, optlen)) 3001 return -EFAULT; 3002 if (copy_to_user(optval, &val, len)) 3003 return -EFAULT; 3004 3005 return 0; 3006 } 3007 3008 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3009 sockptr_t optval, unsigned int optlen) 3010 { 3011 struct sock *sk = sock->sk; 3012 struct smc_sock *smc; 3013 int val, rc; 3014 3015 smc = smc_sk(sk); 3016 3017 lock_sock(sk); 3018 switch (optname) { 3019 case SMC_LIMIT_HS: 3020 if (optlen < sizeof(int)) { 3021 rc = -EINVAL; 3022 break; 3023 } 3024 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3025 rc = -EFAULT; 3026 break; 3027 } 3028 3029 smc->limit_smc_hs = !!val; 3030 rc = 0; 3031 break; 3032 default: 3033 rc = -EOPNOTSUPP; 3034 break; 3035 } 3036 release_sock(sk); 3037 3038 return rc; 3039 } 3040 3041 static int smc_setsockopt(struct socket *sock, int level, int optname, 3042 sockptr_t optval, unsigned int optlen) 3043 { 3044 struct sock *sk = sock->sk; 3045 struct smc_sock *smc; 3046 int val, rc; 3047 3048 if (level == SOL_TCP && optname == TCP_ULP) 3049 return -EOPNOTSUPP; 3050 else if (level == SOL_SMC) 3051 return __smc_setsockopt(sock, level, optname, optval, optlen); 3052 3053 smc = smc_sk(sk); 3054 3055 /* generic setsockopts reaching us here always apply to the 3056 * CLC socket 3057 */ 3058 mutex_lock(&smc->clcsock_release_lock); 3059 if (!smc->clcsock) { 3060 mutex_unlock(&smc->clcsock_release_lock); 3061 return -EBADF; 3062 } 3063 if (unlikely(!smc->clcsock->ops->setsockopt)) 3064 rc = -EOPNOTSUPP; 3065 else 3066 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3067 optval, optlen); 3068 if (smc->clcsock->sk->sk_err) { 3069 sk->sk_err = smc->clcsock->sk->sk_err; 3070 sk_error_report(sk); 3071 } 3072 mutex_unlock(&smc->clcsock_release_lock); 3073 3074 if (optlen < sizeof(int)) 3075 return -EINVAL; 3076 if (copy_from_sockptr(&val, optval, sizeof(int))) 3077 return -EFAULT; 3078 3079 lock_sock(sk); 3080 if (rc || smc->use_fallback) 3081 goto out; 3082 switch (optname) { 3083 case TCP_FASTOPEN: 3084 case TCP_FASTOPEN_CONNECT: 3085 case TCP_FASTOPEN_KEY: 3086 case TCP_FASTOPEN_NO_COOKIE: 3087 /* option not supported by SMC */ 3088 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3089 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3090 } else { 3091 rc = -EINVAL; 3092 } 3093 break; 3094 case TCP_NODELAY: 3095 if (sk->sk_state != SMC_INIT && 3096 sk->sk_state != SMC_LISTEN && 3097 sk->sk_state != SMC_CLOSED) { 3098 if (val) { 3099 SMC_STAT_INC(smc, ndly_cnt); 3100 smc_tx_pending(&smc->conn); 3101 cancel_delayed_work(&smc->conn.tx_work); 3102 } 3103 } 3104 break; 3105 case TCP_CORK: 3106 if (sk->sk_state != SMC_INIT && 3107 sk->sk_state != SMC_LISTEN && 3108 sk->sk_state != SMC_CLOSED) { 3109 if (!val) { 3110 SMC_STAT_INC(smc, cork_cnt); 3111 smc_tx_pending(&smc->conn); 3112 cancel_delayed_work(&smc->conn.tx_work); 3113 } 3114 } 3115 break; 3116 case TCP_DEFER_ACCEPT: 3117 smc->sockopt_defer_accept = val; 3118 break; 3119 default: 3120 break; 3121 } 3122 out: 3123 release_sock(sk); 3124 3125 return rc; 3126 } 3127 3128 static int smc_getsockopt(struct socket *sock, int level, int optname, 3129 char __user *optval, int __user *optlen) 3130 { 3131 struct smc_sock *smc; 3132 int rc; 3133 3134 if (level == SOL_SMC) 3135 return __smc_getsockopt(sock, level, optname, optval, optlen); 3136 3137 smc = smc_sk(sock->sk); 3138 mutex_lock(&smc->clcsock_release_lock); 3139 if (!smc->clcsock) { 3140 mutex_unlock(&smc->clcsock_release_lock); 3141 return -EBADF; 3142 } 3143 /* socket options apply to the CLC socket */ 3144 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3145 mutex_unlock(&smc->clcsock_release_lock); 3146 return -EOPNOTSUPP; 3147 } 3148 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3149 optval, optlen); 3150 mutex_unlock(&smc->clcsock_release_lock); 3151 return rc; 3152 } 3153 3154 static int smc_ioctl(struct socket *sock, unsigned int cmd, 3155 unsigned long arg) 3156 { 3157 union smc_host_cursor cons, urg; 3158 struct smc_connection *conn; 3159 struct smc_sock *smc; 3160 int answ; 3161 3162 smc = smc_sk(sock->sk); 3163 conn = &smc->conn; 3164 lock_sock(&smc->sk); 3165 if (smc->use_fallback) { 3166 if (!smc->clcsock) { 3167 release_sock(&smc->sk); 3168 return -EBADF; 3169 } 3170 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3171 release_sock(&smc->sk); 3172 return answ; 3173 } 3174 switch (cmd) { 3175 case SIOCINQ: /* same as FIONREAD */ 3176 if (smc->sk.sk_state == SMC_LISTEN) { 3177 release_sock(&smc->sk); 3178 return -EINVAL; 3179 } 3180 if (smc->sk.sk_state == SMC_INIT || 3181 smc->sk.sk_state == SMC_CLOSED) 3182 answ = 0; 3183 else 3184 answ = atomic_read(&smc->conn.bytes_to_rcv); 3185 break; 3186 case SIOCOUTQ: 3187 /* output queue size (not send + not acked) */ 3188 if (smc->sk.sk_state == SMC_LISTEN) { 3189 release_sock(&smc->sk); 3190 return -EINVAL; 3191 } 3192 if (smc->sk.sk_state == SMC_INIT || 3193 smc->sk.sk_state == SMC_CLOSED) 3194 answ = 0; 3195 else 3196 answ = smc->conn.sndbuf_desc->len - 3197 atomic_read(&smc->conn.sndbuf_space); 3198 break; 3199 case SIOCOUTQNSD: 3200 /* output queue size (not send only) */ 3201 if (smc->sk.sk_state == SMC_LISTEN) { 3202 release_sock(&smc->sk); 3203 return -EINVAL; 3204 } 3205 if (smc->sk.sk_state == SMC_INIT || 3206 smc->sk.sk_state == SMC_CLOSED) 3207 answ = 0; 3208 else 3209 answ = smc_tx_prepared_sends(&smc->conn); 3210 break; 3211 case SIOCATMARK: 3212 if (smc->sk.sk_state == SMC_LISTEN) { 3213 release_sock(&smc->sk); 3214 return -EINVAL; 3215 } 3216 if (smc->sk.sk_state == SMC_INIT || 3217 smc->sk.sk_state == SMC_CLOSED) { 3218 answ = 0; 3219 } else { 3220 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3221 smc_curs_copy(&urg, &conn->urg_curs, conn); 3222 answ = smc_curs_diff(conn->rmb_desc->len, 3223 &cons, &urg) == 1; 3224 } 3225 break; 3226 default: 3227 release_sock(&smc->sk); 3228 return -ENOIOCTLCMD; 3229 } 3230 release_sock(&smc->sk); 3231 3232 return put_user(answ, (int __user *)arg); 3233 } 3234 3235 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3236 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3237 * updates till whenever a respective page has been fully processed. 3238 * Note that subsequent recv() calls have to wait till all splice() processing 3239 * completed. 3240 */ 3241 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3242 struct pipe_inode_info *pipe, size_t len, 3243 unsigned int flags) 3244 { 3245 struct sock *sk = sock->sk; 3246 struct smc_sock *smc; 3247 int rc = -ENOTCONN; 3248 3249 smc = smc_sk(sk); 3250 lock_sock(sk); 3251 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3252 /* socket was connected before, no more data to read */ 3253 rc = 0; 3254 goto out; 3255 } 3256 if (sk->sk_state == SMC_INIT || 3257 sk->sk_state == SMC_LISTEN || 3258 sk->sk_state == SMC_CLOSED) 3259 goto out; 3260 3261 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3262 rc = 0; 3263 goto out; 3264 } 3265 3266 if (smc->use_fallback) { 3267 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3268 pipe, len, flags); 3269 } else { 3270 if (*ppos) { 3271 rc = -ESPIPE; 3272 goto out; 3273 } 3274 if (flags & SPLICE_F_NONBLOCK) 3275 flags = MSG_DONTWAIT; 3276 else 3277 flags = 0; 3278 SMC_STAT_INC(smc, splice_cnt); 3279 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3280 } 3281 out: 3282 release_sock(sk); 3283 3284 return rc; 3285 } 3286 3287 /* must look like tcp */ 3288 static const struct proto_ops smc_sock_ops = { 3289 .family = PF_SMC, 3290 .owner = THIS_MODULE, 3291 .release = smc_release, 3292 .bind = smc_bind, 3293 .connect = smc_connect, 3294 .socketpair = sock_no_socketpair, 3295 .accept = smc_accept, 3296 .getname = smc_getname, 3297 .poll = smc_poll, 3298 .ioctl = smc_ioctl, 3299 .listen = smc_listen, 3300 .shutdown = smc_shutdown, 3301 .setsockopt = smc_setsockopt, 3302 .getsockopt = smc_getsockopt, 3303 .sendmsg = smc_sendmsg, 3304 .recvmsg = smc_recvmsg, 3305 .mmap = sock_no_mmap, 3306 .splice_read = smc_splice_read, 3307 }; 3308 3309 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3310 int kern, struct socket *clcsock) 3311 { 3312 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3313 struct smc_sock *smc; 3314 struct sock *sk; 3315 int rc; 3316 3317 rc = -ESOCKTNOSUPPORT; 3318 if (sock->type != SOCK_STREAM) 3319 goto out; 3320 3321 rc = -EPROTONOSUPPORT; 3322 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3323 goto out; 3324 3325 rc = -ENOBUFS; 3326 sock->ops = &smc_sock_ops; 3327 sock->state = SS_UNCONNECTED; 3328 sk = smc_sock_alloc(net, sock, protocol); 3329 if (!sk) 3330 goto out; 3331 3332 /* create internal TCP socket for CLC handshake and fallback */ 3333 smc = smc_sk(sk); 3334 smc->use_fallback = false; /* assume rdma capability first */ 3335 smc->fallback_rsn = 0; 3336 3337 /* default behavior from limit_smc_hs in every net namespace */ 3338 smc->limit_smc_hs = net->smc.limit_smc_hs; 3339 3340 rc = 0; 3341 if (!clcsock) { 3342 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3343 &smc->clcsock); 3344 if (rc) { 3345 sk_common_release(sk); 3346 goto out; 3347 } 3348 3349 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3350 * destruction; its sk_state might not be TCP_CLOSE after 3351 * smc->sk is close()d, and TCP timers can be fired later, 3352 * which need net ref. 3353 */ 3354 sk = smc->clcsock->sk; 3355 __netns_tracker_free(net, &sk->ns_tracker, false); 3356 sk->sk_net_refcnt = 1; 3357 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3358 sock_inuse_add(net, 1); 3359 } else { 3360 smc->clcsock = clcsock; 3361 } 3362 3363 out: 3364 return rc; 3365 } 3366 3367 static int smc_create(struct net *net, struct socket *sock, int protocol, 3368 int kern) 3369 { 3370 return __smc_create(net, sock, protocol, kern, NULL); 3371 } 3372 3373 static const struct net_proto_family smc_sock_family_ops = { 3374 .family = PF_SMC, 3375 .owner = THIS_MODULE, 3376 .create = smc_create, 3377 }; 3378 3379 static int smc_ulp_init(struct sock *sk) 3380 { 3381 struct socket *tcp = sk->sk_socket; 3382 struct net *net = sock_net(sk); 3383 struct socket *smcsock; 3384 int protocol, ret; 3385 3386 /* only TCP can be replaced */ 3387 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3388 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3389 return -ESOCKTNOSUPPORT; 3390 /* don't handle wq now */ 3391 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3392 return -ENOTCONN; 3393 3394 if (sk->sk_family == AF_INET) 3395 protocol = SMCPROTO_SMC; 3396 else 3397 protocol = SMCPROTO_SMC6; 3398 3399 smcsock = sock_alloc(); 3400 if (!smcsock) 3401 return -ENFILE; 3402 3403 smcsock->type = SOCK_STREAM; 3404 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3405 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3406 if (ret) { 3407 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3408 return ret; 3409 } 3410 3411 /* replace tcp socket to smc */ 3412 smcsock->file = tcp->file; 3413 smcsock->file->private_data = smcsock; 3414 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3415 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3416 tcp->file = NULL; 3417 3418 return ret; 3419 } 3420 3421 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3422 const gfp_t priority) 3423 { 3424 struct inet_connection_sock *icsk = inet_csk(newsk); 3425 3426 /* don't inherit ulp ops to child when listen */ 3427 icsk->icsk_ulp_ops = NULL; 3428 } 3429 3430 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3431 .name = "smc", 3432 .owner = THIS_MODULE, 3433 .init = smc_ulp_init, 3434 .clone = smc_ulp_clone, 3435 }; 3436 3437 unsigned int smc_net_id; 3438 3439 static __net_init int smc_net_init(struct net *net) 3440 { 3441 int rc; 3442 3443 rc = smc_sysctl_net_init(net); 3444 if (rc) 3445 return rc; 3446 return smc_pnet_net_init(net); 3447 } 3448 3449 static void __net_exit smc_net_exit(struct net *net) 3450 { 3451 smc_sysctl_net_exit(net); 3452 smc_pnet_net_exit(net); 3453 } 3454 3455 static __net_init int smc_net_stat_init(struct net *net) 3456 { 3457 return smc_stats_init(net); 3458 } 3459 3460 static void __net_exit smc_net_stat_exit(struct net *net) 3461 { 3462 smc_stats_exit(net); 3463 } 3464 3465 static struct pernet_operations smc_net_ops = { 3466 .init = smc_net_init, 3467 .exit = smc_net_exit, 3468 .id = &smc_net_id, 3469 .size = sizeof(struct smc_net), 3470 }; 3471 3472 static struct pernet_operations smc_net_stat_ops = { 3473 .init = smc_net_stat_init, 3474 .exit = smc_net_stat_exit, 3475 }; 3476 3477 static int __init smc_init(void) 3478 { 3479 int rc; 3480 3481 rc = register_pernet_subsys(&smc_net_ops); 3482 if (rc) 3483 return rc; 3484 3485 rc = register_pernet_subsys(&smc_net_stat_ops); 3486 if (rc) 3487 goto out_pernet_subsys; 3488 3489 rc = smc_ism_init(); 3490 if (rc) 3491 goto out_pernet_subsys_stat; 3492 smc_clc_init(); 3493 3494 rc = smc_nl_init(); 3495 if (rc) 3496 goto out_ism; 3497 3498 rc = smc_pnet_init(); 3499 if (rc) 3500 goto out_nl; 3501 3502 rc = -ENOMEM; 3503 3504 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3505 if (!smc_tcp_ls_wq) 3506 goto out_pnet; 3507 3508 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3509 if (!smc_hs_wq) 3510 goto out_alloc_tcp_ls_wq; 3511 3512 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3513 if (!smc_close_wq) 3514 goto out_alloc_hs_wq; 3515 3516 rc = smc_core_init(); 3517 if (rc) { 3518 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3519 goto out_alloc_wqs; 3520 } 3521 3522 rc = smc_llc_init(); 3523 if (rc) { 3524 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3525 goto out_core; 3526 } 3527 3528 rc = smc_cdc_init(); 3529 if (rc) { 3530 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3531 goto out_core; 3532 } 3533 3534 rc = proto_register(&smc_proto, 1); 3535 if (rc) { 3536 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3537 goto out_core; 3538 } 3539 3540 rc = proto_register(&smc_proto6, 1); 3541 if (rc) { 3542 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3543 goto out_proto; 3544 } 3545 3546 rc = sock_register(&smc_sock_family_ops); 3547 if (rc) { 3548 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3549 goto out_proto6; 3550 } 3551 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3552 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3553 3554 rc = smc_ib_register_client(); 3555 if (rc) { 3556 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3557 goto out_sock; 3558 } 3559 3560 rc = tcp_register_ulp(&smc_ulp_ops); 3561 if (rc) { 3562 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3563 goto out_ib; 3564 } 3565 3566 static_branch_enable(&tcp_have_smc); 3567 return 0; 3568 3569 out_ib: 3570 smc_ib_unregister_client(); 3571 out_sock: 3572 sock_unregister(PF_SMC); 3573 out_proto6: 3574 proto_unregister(&smc_proto6); 3575 out_proto: 3576 proto_unregister(&smc_proto); 3577 out_core: 3578 smc_core_exit(); 3579 out_alloc_wqs: 3580 destroy_workqueue(smc_close_wq); 3581 out_alloc_hs_wq: 3582 destroy_workqueue(smc_hs_wq); 3583 out_alloc_tcp_ls_wq: 3584 destroy_workqueue(smc_tcp_ls_wq); 3585 out_pnet: 3586 smc_pnet_exit(); 3587 out_nl: 3588 smc_nl_exit(); 3589 out_ism: 3590 smc_clc_exit(); 3591 smc_ism_exit(); 3592 out_pernet_subsys_stat: 3593 unregister_pernet_subsys(&smc_net_stat_ops); 3594 out_pernet_subsys: 3595 unregister_pernet_subsys(&smc_net_ops); 3596 3597 return rc; 3598 } 3599 3600 static void __exit smc_exit(void) 3601 { 3602 static_branch_disable(&tcp_have_smc); 3603 tcp_unregister_ulp(&smc_ulp_ops); 3604 sock_unregister(PF_SMC); 3605 smc_core_exit(); 3606 smc_ib_unregister_client(); 3607 smc_ism_exit(); 3608 destroy_workqueue(smc_close_wq); 3609 destroy_workqueue(smc_tcp_ls_wq); 3610 destroy_workqueue(smc_hs_wq); 3611 proto_unregister(&smc_proto6); 3612 proto_unregister(&smc_proto); 3613 smc_pnet_exit(); 3614 smc_nl_exit(); 3615 smc_clc_exit(); 3616 unregister_pernet_subsys(&smc_net_stat_ops); 3617 unregister_pernet_subsys(&smc_net_ops); 3618 rcu_barrier(); 3619 } 3620 3621 module_init(smc_init); 3622 module_exit(smc_exit); 3623 3624 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3625 MODULE_DESCRIPTION("smc socket address family"); 3626 MODULE_LICENSE("GPL"); 3627 MODULE_ALIAS_NETPROTO(PF_SMC); 3628 MODULE_ALIAS_TCP_ULP("smc"); 3629 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3630