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 = 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_msg_accept_confirm_v2 *clc_v2 = 681 (struct smc_clc_msg_accept_confirm_v2 *)clc; 682 struct smc_clc_first_contact_ext *fce; 683 int clc_v2_len; 684 685 if (clc->hdr.version == SMC_V1 || 686 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 687 return; 688 689 if (smc->conn.lgr->is_smcd) { 690 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 691 SMC_MAX_EID_LEN); 692 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 693 d1); 694 } else { 695 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 696 SMC_MAX_EID_LEN); 697 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 698 r1); 699 } 700 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 701 smc->conn.lgr->peer_os = fce->os_type; 702 smc->conn.lgr->peer_smc_release = fce->release; 703 if (smc_isascii(fce->hostname)) 704 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 705 SMC_MAX_HOSTNAME_LEN); 706 } 707 708 static void smcr_conn_save_peer_info(struct smc_sock *smc, 709 struct smc_clc_msg_accept_confirm *clc) 710 { 711 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 712 713 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 714 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 715 smc->conn.peer_rmbe_size = bufsize; 716 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 717 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 718 } 719 720 static void smcd_conn_save_peer_info(struct smc_sock *smc, 721 struct smc_clc_msg_accept_confirm *clc) 722 { 723 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 724 725 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 726 smc->conn.peer_token = clc->d0.token; 727 /* msg header takes up space in the buffer */ 728 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 729 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 730 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 731 } 732 733 static void smc_conn_save_peer_info(struct smc_sock *smc, 734 struct smc_clc_msg_accept_confirm *clc) 735 { 736 if (smc->conn.lgr->is_smcd) 737 smcd_conn_save_peer_info(smc, clc); 738 else 739 smcr_conn_save_peer_info(smc, clc); 740 smc_conn_save_peer_info_fce(smc, clc); 741 } 742 743 static void smc_link_save_peer_info(struct smc_link *link, 744 struct smc_clc_msg_accept_confirm *clc, 745 struct smc_init_info *ini) 746 { 747 link->peer_qpn = ntoh24(clc->r0.qpn); 748 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 749 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 750 link->peer_psn = ntoh24(clc->r0.psn); 751 link->peer_mtu = clc->r0.qp_mtu; 752 } 753 754 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 755 struct smc_stats_fback *fback_arr) 756 { 757 int cnt; 758 759 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 760 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 761 fback_arr[cnt].count++; 762 break; 763 } 764 if (!fback_arr[cnt].fback_code) { 765 fback_arr[cnt].fback_code = smc->fallback_rsn; 766 fback_arr[cnt].count++; 767 break; 768 } 769 } 770 } 771 772 static void smc_stat_fallback(struct smc_sock *smc) 773 { 774 struct net *net = sock_net(&smc->sk); 775 776 mutex_lock(&net->smc.mutex_fback_rsn); 777 if (smc->listen_smc) { 778 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 779 net->smc.fback_rsn->srv_fback_cnt++; 780 } else { 781 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 782 net->smc.fback_rsn->clnt_fback_cnt++; 783 } 784 mutex_unlock(&net->smc.mutex_fback_rsn); 785 } 786 787 /* must be called under rcu read lock */ 788 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 789 { 790 struct socket_wq *wq; 791 __poll_t flags; 792 793 wq = rcu_dereference(smc->sk.sk_wq); 794 if (!skwq_has_sleeper(wq)) 795 return; 796 797 /* wake up smc sk->sk_wq */ 798 if (!key) { 799 /* sk_state_change */ 800 wake_up_interruptible_all(&wq->wait); 801 } else { 802 flags = key_to_poll(key); 803 if (flags & (EPOLLIN | EPOLLOUT)) 804 /* sk_data_ready or sk_write_space */ 805 wake_up_interruptible_sync_poll(&wq->wait, flags); 806 else if (flags & EPOLLERR) 807 /* sk_error_report */ 808 wake_up_interruptible_poll(&wq->wait, flags); 809 } 810 } 811 812 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 813 unsigned int mode, int sync, void *key) 814 { 815 struct smc_mark_woken *mark = 816 container_of(wait, struct smc_mark_woken, wait_entry); 817 818 mark->woken = true; 819 mark->key = key; 820 return 0; 821 } 822 823 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 824 void (*clcsock_callback)(struct sock *sk)) 825 { 826 struct smc_mark_woken mark = { .woken = false }; 827 struct socket_wq *wq; 828 829 init_waitqueue_func_entry(&mark.wait_entry, 830 smc_fback_mark_woken); 831 rcu_read_lock(); 832 wq = rcu_dereference(clcsk->sk_wq); 833 if (!wq) 834 goto out; 835 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 836 clcsock_callback(clcsk); 837 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 838 839 if (mark.woken) 840 smc_fback_wakeup_waitqueue(smc, mark.key); 841 out: 842 rcu_read_unlock(); 843 } 844 845 static void smc_fback_state_change(struct sock *clcsk) 846 { 847 struct smc_sock *smc; 848 849 read_lock_bh(&clcsk->sk_callback_lock); 850 smc = smc_clcsock_user_data(clcsk); 851 if (smc) 852 smc_fback_forward_wakeup(smc, clcsk, 853 smc->clcsk_state_change); 854 read_unlock_bh(&clcsk->sk_callback_lock); 855 } 856 857 static void smc_fback_data_ready(struct sock *clcsk) 858 { 859 struct smc_sock *smc; 860 861 read_lock_bh(&clcsk->sk_callback_lock); 862 smc = smc_clcsock_user_data(clcsk); 863 if (smc) 864 smc_fback_forward_wakeup(smc, clcsk, 865 smc->clcsk_data_ready); 866 read_unlock_bh(&clcsk->sk_callback_lock); 867 } 868 869 static void smc_fback_write_space(struct sock *clcsk) 870 { 871 struct smc_sock *smc; 872 873 read_lock_bh(&clcsk->sk_callback_lock); 874 smc = smc_clcsock_user_data(clcsk); 875 if (smc) 876 smc_fback_forward_wakeup(smc, clcsk, 877 smc->clcsk_write_space); 878 read_unlock_bh(&clcsk->sk_callback_lock); 879 } 880 881 static void smc_fback_error_report(struct sock *clcsk) 882 { 883 struct smc_sock *smc; 884 885 read_lock_bh(&clcsk->sk_callback_lock); 886 smc = smc_clcsock_user_data(clcsk); 887 if (smc) 888 smc_fback_forward_wakeup(smc, clcsk, 889 smc->clcsk_error_report); 890 read_unlock_bh(&clcsk->sk_callback_lock); 891 } 892 893 static void smc_fback_replace_callbacks(struct smc_sock *smc) 894 { 895 struct sock *clcsk = smc->clcsock->sk; 896 897 write_lock_bh(&clcsk->sk_callback_lock); 898 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 899 900 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 901 &smc->clcsk_state_change); 902 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 903 &smc->clcsk_data_ready); 904 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 905 &smc->clcsk_write_space); 906 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 907 &smc->clcsk_error_report); 908 909 write_unlock_bh(&clcsk->sk_callback_lock); 910 } 911 912 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 913 { 914 int rc = 0; 915 916 mutex_lock(&smc->clcsock_release_lock); 917 if (!smc->clcsock) { 918 rc = -EBADF; 919 goto out; 920 } 921 922 smc->use_fallback = true; 923 smc->fallback_rsn = reason_code; 924 smc_stat_fallback(smc); 925 trace_smc_switch_to_fallback(smc, reason_code); 926 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 927 smc->clcsock->file = smc->sk.sk_socket->file; 928 smc->clcsock->file->private_data = smc->clcsock; 929 smc->clcsock->wq.fasync_list = 930 smc->sk.sk_socket->wq.fasync_list; 931 932 /* There might be some wait entries remaining 933 * in smc sk->sk_wq and they should be woken up 934 * as clcsock's wait queue is woken up. 935 */ 936 smc_fback_replace_callbacks(smc); 937 } 938 out: 939 mutex_unlock(&smc->clcsock_release_lock); 940 return rc; 941 } 942 943 /* fall back during connect */ 944 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 945 { 946 struct net *net = sock_net(&smc->sk); 947 int rc = 0; 948 949 rc = smc_switch_to_fallback(smc, reason_code); 950 if (rc) { /* fallback fails */ 951 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 952 if (smc->sk.sk_state == SMC_INIT) 953 sock_put(&smc->sk); /* passive closing */ 954 return rc; 955 } 956 smc_copy_sock_settings_to_clc(smc); 957 smc->connect_nonblock = 0; 958 if (smc->sk.sk_state == SMC_INIT) 959 smc->sk.sk_state = SMC_ACTIVE; 960 return 0; 961 } 962 963 /* decline and fall back during connect */ 964 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 965 u8 version) 966 { 967 struct net *net = sock_net(&smc->sk); 968 int rc; 969 970 if (reason_code < 0) { /* error, fallback is not possible */ 971 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 972 if (smc->sk.sk_state == SMC_INIT) 973 sock_put(&smc->sk); /* passive closing */ 974 return reason_code; 975 } 976 if (reason_code != SMC_CLC_DECL_PEERDECL) { 977 rc = smc_clc_send_decline(smc, reason_code, version); 978 if (rc < 0) { 979 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 980 if (smc->sk.sk_state == SMC_INIT) 981 sock_put(&smc->sk); /* passive closing */ 982 return rc; 983 } 984 } 985 return smc_connect_fallback(smc, reason_code); 986 } 987 988 static void smc_conn_abort(struct smc_sock *smc, int local_first) 989 { 990 struct smc_connection *conn = &smc->conn; 991 struct smc_link_group *lgr = conn->lgr; 992 bool lgr_valid = false; 993 994 if (smc_conn_lgr_valid(conn)) 995 lgr_valid = true; 996 997 smc_conn_free(conn); 998 if (local_first && lgr_valid) 999 smc_lgr_cleanup_early(lgr); 1000 } 1001 1002 /* check if there is a rdma device available for this connection. */ 1003 /* called for connect and listen */ 1004 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 1005 { 1006 /* PNET table look up: search active ib_device and port 1007 * within same PNETID that also contains the ethernet device 1008 * used for the internal TCP socket 1009 */ 1010 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 1011 if (!ini->check_smcrv2 && !ini->ib_dev) 1012 return SMC_CLC_DECL_NOSMCRDEV; 1013 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1014 return SMC_CLC_DECL_NOSMCRDEV; 1015 return 0; 1016 } 1017 1018 /* check if there is an ISM device available for this connection. */ 1019 /* called for connect and listen */ 1020 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1021 { 1022 /* Find ISM device with same PNETID as connecting interface */ 1023 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1024 if (!ini->ism_dev[0]) 1025 return SMC_CLC_DECL_NOSMCDDEV; 1026 else 1027 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1028 return 0; 1029 } 1030 1031 /* is chid unique for the ism devices that are already determined? */ 1032 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1033 int cnt) 1034 { 1035 int i = (!ini->ism_dev[0]) ? 1 : 0; 1036 1037 for (; i < cnt; i++) 1038 if (ini->ism_chid[i] == chid) 1039 return false; 1040 return true; 1041 } 1042 1043 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1044 * PNETID matching net_device) 1045 */ 1046 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1047 struct smc_init_info *ini) 1048 { 1049 int rc = SMC_CLC_DECL_NOSMCDDEV; 1050 struct smcd_dev *smcd; 1051 int i = 1; 1052 u16 chid; 1053 1054 if (smcd_indicated(ini->smc_type_v1)) 1055 rc = 0; /* already initialized for V1 */ 1056 mutex_lock(&smcd_dev_list.mutex); 1057 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1058 if (smcd->going_away || smcd == ini->ism_dev[0]) 1059 continue; 1060 chid = smc_ism_get_chid(smcd); 1061 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1062 continue; 1063 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1064 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1065 ini->ism_dev[i] = smcd; 1066 ini->ism_chid[i] = chid; 1067 ini->is_smcd = true; 1068 rc = 0; 1069 i++; 1070 if (i > SMC_MAX_ISM_DEVS) 1071 break; 1072 } 1073 } 1074 mutex_unlock(&smcd_dev_list.mutex); 1075 ini->ism_offered_cnt = i - 1; 1076 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1077 ini->smcd_version = 0; 1078 1079 return rc; 1080 } 1081 1082 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1083 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1084 struct smc_init_info *ini) 1085 { 1086 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1087 return SMC_CLC_DECL_ISMVLANERR; 1088 return 0; 1089 } 1090 1091 static int smc_find_proposal_devices(struct smc_sock *smc, 1092 struct smc_init_info *ini) 1093 { 1094 int rc = 0; 1095 1096 /* check if there is an ism device available */ 1097 if (!(ini->smcd_version & SMC_V1) || 1098 smc_find_ism_device(smc, ini) || 1099 smc_connect_ism_vlan_setup(smc, ini)) 1100 ini->smcd_version &= ~SMC_V1; 1101 /* else ISM V1 is supported for this connection */ 1102 1103 /* check if there is an rdma device available */ 1104 if (!(ini->smcr_version & SMC_V1) || 1105 smc_find_rdma_device(smc, ini)) 1106 ini->smcr_version &= ~SMC_V1; 1107 /* else RDMA is supported for this connection */ 1108 1109 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1110 ini->smcr_version & SMC_V1); 1111 1112 /* check if there is an ism v2 device available */ 1113 if (!(ini->smcd_version & SMC_V2) || 1114 !smc_ism_is_v2_capable() || 1115 smc_find_ism_v2_device_clnt(smc, ini)) 1116 ini->smcd_version &= ~SMC_V2; 1117 1118 /* check if there is an rdma v2 device available */ 1119 ini->check_smcrv2 = true; 1120 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1121 if (!(ini->smcr_version & SMC_V2) || 1122 smc->clcsock->sk->sk_family != AF_INET || 1123 !smc_clc_ueid_count() || 1124 smc_find_rdma_device(smc, ini)) 1125 ini->smcr_version &= ~SMC_V2; 1126 ini->check_smcrv2 = false; 1127 1128 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1129 ini->smcr_version & SMC_V2); 1130 1131 /* if neither ISM nor RDMA are supported, fallback */ 1132 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1133 rc = SMC_CLC_DECL_NOSMCDEV; 1134 1135 return rc; 1136 } 1137 1138 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1139 * used, the VLAN ID will be registered again during the connection setup. 1140 */ 1141 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1142 struct smc_init_info *ini) 1143 { 1144 if (!smcd_indicated(ini->smc_type_v1)) 1145 return 0; 1146 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1147 return SMC_CLC_DECL_CNFERR; 1148 return 0; 1149 } 1150 1151 #define SMC_CLC_MAX_ACCEPT_LEN \ 1152 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 1153 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1154 sizeof(struct smc_clc_msg_trail)) 1155 1156 /* CLC handshake during connect */ 1157 static int smc_connect_clc(struct smc_sock *smc, 1158 struct smc_clc_msg_accept_confirm_v2 *aclc2, 1159 struct smc_init_info *ini) 1160 { 1161 int rc = 0; 1162 1163 /* do inband token exchange */ 1164 rc = smc_clc_send_proposal(smc, ini); 1165 if (rc) 1166 return rc; 1167 /* receive SMC Accept CLC message */ 1168 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 1169 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1170 } 1171 1172 void smc_fill_gid_list(struct smc_link_group *lgr, 1173 struct smc_gidlist *gidlist, 1174 struct smc_ib_device *known_dev, u8 *known_gid) 1175 { 1176 struct smc_init_info *alt_ini = NULL; 1177 1178 memset(gidlist, 0, sizeof(*gidlist)); 1179 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1180 1181 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1182 if (!alt_ini) 1183 goto out; 1184 1185 alt_ini->vlan_id = lgr->vlan_id; 1186 alt_ini->check_smcrv2 = true; 1187 alt_ini->smcrv2.saddr = lgr->saddr; 1188 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1189 1190 if (!alt_ini->smcrv2.ib_dev_v2) 1191 goto out; 1192 1193 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1194 SMC_GID_SIZE); 1195 1196 out: 1197 kfree(alt_ini); 1198 } 1199 1200 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1201 struct smc_clc_msg_accept_confirm *aclc, 1202 struct smc_init_info *ini) 1203 { 1204 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1205 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1206 struct smc_clc_first_contact_ext *fce = 1207 smc_get_clc_first_contact_ext(clc_v2, false); 1208 struct net *net = sock_net(&smc->sk); 1209 int rc; 1210 1211 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1212 return 0; 1213 1214 if (fce->v2_direct) { 1215 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1216 ini->smcrv2.uses_gateway = false; 1217 } else { 1218 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1219 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1220 ini->smcrv2.nexthop_mac, 1221 &ini->smcrv2.uses_gateway)) 1222 return SMC_CLC_DECL_NOROUTE; 1223 if (!ini->smcrv2.uses_gateway) { 1224 /* mismatch: peer claims indirect, but its direct */ 1225 return SMC_CLC_DECL_NOINDIRECT; 1226 } 1227 } 1228 1229 ini->release_nr = fce->release; 1230 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1231 if (rc) 1232 return rc; 1233 1234 return 0; 1235 } 1236 1237 /* setup for RDMA connection of client */ 1238 static int smc_connect_rdma(struct smc_sock *smc, 1239 struct smc_clc_msg_accept_confirm *aclc, 1240 struct smc_init_info *ini) 1241 { 1242 int i, reason_code = 0; 1243 struct smc_link *link; 1244 u8 *eid = NULL; 1245 1246 ini->is_smcd = false; 1247 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1248 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1249 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1250 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1251 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1252 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1253 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1254 1255 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1256 if (reason_code) 1257 return reason_code; 1258 1259 mutex_lock(&smc_client_lgr_pending); 1260 reason_code = smc_conn_create(smc, ini); 1261 if (reason_code) { 1262 mutex_unlock(&smc_client_lgr_pending); 1263 return reason_code; 1264 } 1265 1266 smc_conn_save_peer_info(smc, aclc); 1267 1268 if (ini->first_contact_local) { 1269 link = smc->conn.lnk; 1270 } else { 1271 /* set link that was assigned by server */ 1272 link = NULL; 1273 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1274 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1275 1276 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1277 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1278 SMC_GID_SIZE) && 1279 (aclc->hdr.version > SMC_V1 || 1280 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1281 sizeof(l->peer_mac)))) { 1282 link = l; 1283 break; 1284 } 1285 } 1286 if (!link) { 1287 reason_code = SMC_CLC_DECL_NOSRVLINK; 1288 goto connect_abort; 1289 } 1290 smc_switch_link_and_count(&smc->conn, link); 1291 } 1292 1293 /* create send buffer and rmb */ 1294 if (smc_buf_create(smc, false)) { 1295 reason_code = SMC_CLC_DECL_MEM; 1296 goto connect_abort; 1297 } 1298 1299 if (ini->first_contact_local) 1300 smc_link_save_peer_info(link, aclc, ini); 1301 1302 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1303 reason_code = SMC_CLC_DECL_ERR_RTOK; 1304 goto connect_abort; 1305 } 1306 1307 smc_close_init(smc); 1308 smc_rx_init(smc); 1309 1310 if (ini->first_contact_local) { 1311 if (smc_ib_ready_link(link)) { 1312 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1313 goto connect_abort; 1314 } 1315 } else { 1316 /* reg sendbufs if they were vzalloced */ 1317 if (smc->conn.sndbuf_desc->is_vm) { 1318 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1319 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1320 goto connect_abort; 1321 } 1322 } 1323 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1324 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1325 goto connect_abort; 1326 } 1327 } 1328 1329 if (aclc->hdr.version > SMC_V1) { 1330 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1331 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1332 1333 eid = clc_v2->r1.eid; 1334 if (ini->first_contact_local) 1335 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1336 link->smcibdev, link->gid); 1337 } 1338 1339 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1340 aclc->hdr.version, eid, ini); 1341 if (reason_code) 1342 goto connect_abort; 1343 1344 smc_tx_init(smc); 1345 1346 if (ini->first_contact_local) { 1347 /* QP confirmation over RoCE fabric */ 1348 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1349 reason_code = smcr_clnt_conf_first_link(smc); 1350 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1351 if (reason_code) 1352 goto connect_abort; 1353 } 1354 mutex_unlock(&smc_client_lgr_pending); 1355 1356 smc_copy_sock_settings_to_clc(smc); 1357 smc->connect_nonblock = 0; 1358 if (smc->sk.sk_state == SMC_INIT) 1359 smc->sk.sk_state = SMC_ACTIVE; 1360 1361 return 0; 1362 connect_abort: 1363 smc_conn_abort(smc, ini->first_contact_local); 1364 mutex_unlock(&smc_client_lgr_pending); 1365 smc->connect_nonblock = 0; 1366 1367 return reason_code; 1368 } 1369 1370 /* The server has chosen one of the proposed ISM devices for the communication. 1371 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1372 */ 1373 static int 1374 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 1375 struct smc_init_info *ini) 1376 { 1377 int i; 1378 1379 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1380 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1381 ini->ism_selected = i; 1382 return 0; 1383 } 1384 } 1385 1386 return -EPROTO; 1387 } 1388 1389 /* setup for ISM connection of client */ 1390 static int smc_connect_ism(struct smc_sock *smc, 1391 struct smc_clc_msg_accept_confirm *aclc, 1392 struct smc_init_info *ini) 1393 { 1394 u8 *eid = NULL; 1395 int rc = 0; 1396 1397 ini->is_smcd = true; 1398 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1399 1400 if (aclc->hdr.version == SMC_V2) { 1401 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1402 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1403 1404 if (ini->first_contact_peer) { 1405 struct smc_clc_first_contact_ext *fce = 1406 smc_get_clc_first_contact_ext(aclc_v2, true); 1407 1408 ini->release_nr = fce->release; 1409 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1410 if (rc) 1411 return rc; 1412 } 1413 1414 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1415 if (rc) 1416 return rc; 1417 } 1418 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 1419 1420 /* there is only one lgr role for SMC-D; use server lock */ 1421 mutex_lock(&smc_server_lgr_pending); 1422 rc = smc_conn_create(smc, ini); 1423 if (rc) { 1424 mutex_unlock(&smc_server_lgr_pending); 1425 return rc; 1426 } 1427 1428 /* Create send and receive buffers */ 1429 rc = smc_buf_create(smc, true); 1430 if (rc) { 1431 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1432 goto connect_abort; 1433 } 1434 1435 smc_conn_save_peer_info(smc, aclc); 1436 smc_close_init(smc); 1437 smc_rx_init(smc); 1438 smc_tx_init(smc); 1439 1440 if (aclc->hdr.version > SMC_V1) { 1441 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1442 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1443 1444 eid = clc_v2->d1.eid; 1445 } 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_v2 *aclc2; 1497 struct smc_clc_msg_accept_confirm *aclc; 1498 struct smc_init_info *ini = NULL; 1499 u8 *buf = NULL; 1500 int rc = 0; 1501 1502 if (smc->use_fallback) 1503 return smc_connect_fallback(smc, smc->fallback_rsn); 1504 1505 /* if peer has not signalled SMC-capability, fall back */ 1506 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1507 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1508 1509 /* IPSec connections opt out of SMC optimizations */ 1510 if (using_ipsec(smc)) 1511 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1512 version); 1513 1514 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1515 if (!ini) 1516 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1517 version); 1518 1519 ini->smcd_version = SMC_V1 | SMC_V2; 1520 ini->smcr_version = SMC_V1 | SMC_V2; 1521 ini->smc_type_v1 = SMC_TYPE_B; 1522 ini->smc_type_v2 = SMC_TYPE_B; 1523 1524 /* get vlan id from IP device */ 1525 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1526 ini->smcd_version &= ~SMC_V1; 1527 ini->smcr_version = 0; 1528 ini->smc_type_v1 = SMC_TYPE_N; 1529 if (!ini->smcd_version) { 1530 rc = SMC_CLC_DECL_GETVLANERR; 1531 goto fallback; 1532 } 1533 } 1534 1535 rc = smc_find_proposal_devices(smc, ini); 1536 if (rc) 1537 goto fallback; 1538 1539 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1540 if (!buf) { 1541 rc = SMC_CLC_DECL_MEM; 1542 goto fallback; 1543 } 1544 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1545 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1546 1547 /* perform CLC handshake */ 1548 rc = smc_connect_clc(smc, aclc2, ini); 1549 if (rc) { 1550 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1551 if (rc == -EAGAIN) { 1552 rc = -ETIMEDOUT; 1553 smc->sk.sk_err = ETIMEDOUT; 1554 } 1555 goto vlan_cleanup; 1556 } 1557 1558 /* check if smc modes and versions of CLC proposal and accept match */ 1559 rc = smc_connect_check_aclc(ini, aclc); 1560 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1561 if (rc) 1562 goto vlan_cleanup; 1563 1564 /* depending on previous steps, connect using rdma or ism */ 1565 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1566 ini->smcr_version = version; 1567 rc = smc_connect_rdma(smc, aclc, ini); 1568 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1569 ini->smcd_version = version; 1570 rc = smc_connect_ism(smc, aclc, ini); 1571 } 1572 if (rc) 1573 goto vlan_cleanup; 1574 1575 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1576 smc_connect_ism_vlan_cleanup(smc, ini); 1577 kfree(buf); 1578 kfree(ini); 1579 return 0; 1580 1581 vlan_cleanup: 1582 smc_connect_ism_vlan_cleanup(smc, ini); 1583 kfree(buf); 1584 fallback: 1585 kfree(ini); 1586 return smc_connect_decline_fallback(smc, rc, version); 1587 } 1588 1589 static void smc_connect_work(struct work_struct *work) 1590 { 1591 struct smc_sock *smc = container_of(work, struct smc_sock, 1592 connect_work); 1593 long timeo = smc->sk.sk_sndtimeo; 1594 int rc = 0; 1595 1596 if (!timeo) 1597 timeo = MAX_SCHEDULE_TIMEOUT; 1598 lock_sock(smc->clcsock->sk); 1599 if (smc->clcsock->sk->sk_err) { 1600 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1601 } else if ((1 << smc->clcsock->sk->sk_state) & 1602 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1603 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1604 if ((rc == -EPIPE) && 1605 ((1 << smc->clcsock->sk->sk_state) & 1606 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1607 rc = 0; 1608 } 1609 release_sock(smc->clcsock->sk); 1610 lock_sock(&smc->sk); 1611 if (rc != 0 || smc->sk.sk_err) { 1612 smc->sk.sk_state = SMC_CLOSED; 1613 if (rc == -EPIPE || rc == -EAGAIN) 1614 smc->sk.sk_err = EPIPE; 1615 else if (rc == -ECONNREFUSED) 1616 smc->sk.sk_err = ECONNREFUSED; 1617 else if (signal_pending(current)) 1618 smc->sk.sk_err = -sock_intr_errno(timeo); 1619 sock_put(&smc->sk); /* passive closing */ 1620 goto out; 1621 } 1622 1623 rc = __smc_connect(smc); 1624 if (rc < 0) 1625 smc->sk.sk_err = -rc; 1626 1627 out: 1628 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1629 if (smc->sk.sk_err) { 1630 smc->sk.sk_state_change(&smc->sk); 1631 } else { /* allow polling before and after fallback decision */ 1632 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1633 smc->sk.sk_write_space(&smc->sk); 1634 } 1635 } 1636 release_sock(&smc->sk); 1637 } 1638 1639 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1640 int alen, int flags) 1641 { 1642 struct sock *sk = sock->sk; 1643 struct smc_sock *smc; 1644 int rc = -EINVAL; 1645 1646 smc = smc_sk(sk); 1647 1648 /* separate smc parameter checking to be safe */ 1649 if (alen < sizeof(addr->sa_family)) 1650 goto out_err; 1651 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1652 goto out_err; 1653 1654 lock_sock(sk); 1655 switch (sock->state) { 1656 default: 1657 rc = -EINVAL; 1658 goto out; 1659 case SS_CONNECTED: 1660 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1661 goto out; 1662 case SS_CONNECTING: 1663 if (sk->sk_state == SMC_ACTIVE) 1664 goto connected; 1665 break; 1666 case SS_UNCONNECTED: 1667 sock->state = SS_CONNECTING; 1668 break; 1669 } 1670 1671 switch (sk->sk_state) { 1672 default: 1673 goto out; 1674 case SMC_CLOSED: 1675 rc = sock_error(sk) ? : -ECONNABORTED; 1676 sock->state = SS_UNCONNECTED; 1677 goto out; 1678 case SMC_ACTIVE: 1679 rc = -EISCONN; 1680 goto out; 1681 case SMC_INIT: 1682 break; 1683 } 1684 1685 smc_copy_sock_settings_to_clc(smc); 1686 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1687 if (smc->connect_nonblock) { 1688 rc = -EALREADY; 1689 goto out; 1690 } 1691 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1692 if (rc && rc != -EINPROGRESS) 1693 goto out; 1694 1695 if (smc->use_fallback) { 1696 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1697 goto out; 1698 } 1699 sock_hold(&smc->sk); /* sock put in passive closing */ 1700 if (flags & O_NONBLOCK) { 1701 if (queue_work(smc_hs_wq, &smc->connect_work)) 1702 smc->connect_nonblock = 1; 1703 rc = -EINPROGRESS; 1704 goto out; 1705 } else { 1706 rc = __smc_connect(smc); 1707 if (rc < 0) 1708 goto out; 1709 } 1710 1711 connected: 1712 rc = 0; 1713 sock->state = SS_CONNECTED; 1714 out: 1715 release_sock(sk); 1716 out_err: 1717 return rc; 1718 } 1719 1720 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1721 { 1722 struct socket *new_clcsock = NULL; 1723 struct sock *lsk = &lsmc->sk; 1724 struct sock *new_sk; 1725 int rc = -EINVAL; 1726 1727 release_sock(lsk); 1728 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1729 if (!new_sk) { 1730 rc = -ENOMEM; 1731 lsk->sk_err = ENOMEM; 1732 *new_smc = NULL; 1733 lock_sock(lsk); 1734 goto out; 1735 } 1736 *new_smc = smc_sk(new_sk); 1737 1738 mutex_lock(&lsmc->clcsock_release_lock); 1739 if (lsmc->clcsock) 1740 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1741 mutex_unlock(&lsmc->clcsock_release_lock); 1742 lock_sock(lsk); 1743 if (rc < 0 && rc != -EAGAIN) 1744 lsk->sk_err = -rc; 1745 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1746 new_sk->sk_prot->unhash(new_sk); 1747 if (new_clcsock) 1748 sock_release(new_clcsock); 1749 new_sk->sk_state = SMC_CLOSED; 1750 smc_sock_set_flag(new_sk, SOCK_DEAD); 1751 sock_put(new_sk); /* final */ 1752 *new_smc = NULL; 1753 goto out; 1754 } 1755 1756 /* new clcsock has inherited the smc listen-specific sk_data_ready 1757 * function; switch it back to the original sk_data_ready function 1758 */ 1759 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1760 1761 /* if new clcsock has also inherited the fallback-specific callback 1762 * functions, switch them back to the original ones. 1763 */ 1764 if (lsmc->use_fallback) { 1765 if (lsmc->clcsk_state_change) 1766 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1767 if (lsmc->clcsk_write_space) 1768 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1769 if (lsmc->clcsk_error_report) 1770 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1771 } 1772 1773 (*new_smc)->clcsock = new_clcsock; 1774 out: 1775 return rc; 1776 } 1777 1778 /* add a just created sock to the accept queue of the listen sock as 1779 * candidate for a following socket accept call from user space 1780 */ 1781 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1782 { 1783 struct smc_sock *par = smc_sk(parent); 1784 1785 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1786 spin_lock(&par->accept_q_lock); 1787 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1788 spin_unlock(&par->accept_q_lock); 1789 sk_acceptq_added(parent); 1790 } 1791 1792 /* remove a socket from the accept queue of its parental listening socket */ 1793 static void smc_accept_unlink(struct sock *sk) 1794 { 1795 struct smc_sock *par = smc_sk(sk)->listen_smc; 1796 1797 spin_lock(&par->accept_q_lock); 1798 list_del_init(&smc_sk(sk)->accept_q); 1799 spin_unlock(&par->accept_q_lock); 1800 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1801 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1802 } 1803 1804 /* remove a sock from the accept queue to bind it to a new socket created 1805 * for a socket accept call from user space 1806 */ 1807 struct sock *smc_accept_dequeue(struct sock *parent, 1808 struct socket *new_sock) 1809 { 1810 struct smc_sock *isk, *n; 1811 struct sock *new_sk; 1812 1813 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1814 new_sk = (struct sock *)isk; 1815 1816 smc_accept_unlink(new_sk); 1817 if (new_sk->sk_state == SMC_CLOSED) { 1818 new_sk->sk_prot->unhash(new_sk); 1819 if (isk->clcsock) { 1820 sock_release(isk->clcsock); 1821 isk->clcsock = NULL; 1822 } 1823 sock_put(new_sk); /* final */ 1824 continue; 1825 } 1826 if (new_sock) { 1827 sock_graft(new_sk, new_sock); 1828 new_sock->state = SS_CONNECTED; 1829 if (isk->use_fallback) { 1830 smc_sk(new_sk)->clcsock->file = new_sock->file; 1831 isk->clcsock->file->private_data = isk->clcsock; 1832 } 1833 } 1834 return new_sk; 1835 } 1836 return NULL; 1837 } 1838 1839 /* clean up for a created but never accepted sock */ 1840 void smc_close_non_accepted(struct sock *sk) 1841 { 1842 struct smc_sock *smc = smc_sk(sk); 1843 1844 sock_hold(sk); /* sock_put below */ 1845 lock_sock(sk); 1846 if (!sk->sk_lingertime) 1847 /* wait for peer closing */ 1848 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1849 __smc_release(smc); 1850 release_sock(sk); 1851 sock_put(sk); /* sock_hold above */ 1852 sock_put(sk); /* final sock_put */ 1853 } 1854 1855 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1856 { 1857 struct smc_link *link = smc->conn.lnk; 1858 struct smc_llc_qentry *qentry; 1859 int rc; 1860 1861 /* reg the sndbuf if it was vzalloced*/ 1862 if (smc->conn.sndbuf_desc->is_vm) { 1863 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1864 return SMC_CLC_DECL_ERR_REGBUF; 1865 } 1866 1867 /* reg the rmb */ 1868 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1869 return SMC_CLC_DECL_ERR_REGBUF; 1870 1871 /* send CONFIRM LINK request to client over the RoCE fabric */ 1872 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1873 if (rc < 0) 1874 return SMC_CLC_DECL_TIMEOUT_CL; 1875 1876 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1877 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1878 SMC_LLC_CONFIRM_LINK); 1879 if (!qentry) { 1880 struct smc_clc_msg_decline dclc; 1881 1882 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1883 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1884 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1885 } 1886 smc_llc_save_peer_uid(qentry); 1887 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1888 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1889 if (rc) 1890 return SMC_CLC_DECL_RMBE_EC; 1891 1892 /* confirm_rkey is implicit on 1st contact */ 1893 smc->conn.rmb_desc->is_conf_rkey = true; 1894 1895 smc_llc_link_active(link); 1896 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1897 1898 if (link->lgr->max_links > 1) { 1899 down_write(&link->lgr->llc_conf_mutex); 1900 /* initial contact - try to establish second link */ 1901 smc_llc_srv_add_link(link, NULL); 1902 up_write(&link->lgr->llc_conf_mutex); 1903 } 1904 return 0; 1905 } 1906 1907 /* listen worker: finish */ 1908 static void smc_listen_out(struct smc_sock *new_smc) 1909 { 1910 struct smc_sock *lsmc = new_smc->listen_smc; 1911 struct sock *newsmcsk = &new_smc->sk; 1912 1913 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1914 atomic_dec(&lsmc->queued_smc_hs); 1915 1916 if (lsmc->sk.sk_state == SMC_LISTEN) { 1917 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1918 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1919 release_sock(&lsmc->sk); 1920 } else { /* no longer listening */ 1921 smc_close_non_accepted(newsmcsk); 1922 } 1923 1924 /* Wake up accept */ 1925 lsmc->sk.sk_data_ready(&lsmc->sk); 1926 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1927 } 1928 1929 /* listen worker: finish in state connected */ 1930 static void smc_listen_out_connected(struct smc_sock *new_smc) 1931 { 1932 struct sock *newsmcsk = &new_smc->sk; 1933 1934 if (newsmcsk->sk_state == SMC_INIT) 1935 newsmcsk->sk_state = SMC_ACTIVE; 1936 1937 smc_listen_out(new_smc); 1938 } 1939 1940 /* listen worker: finish in error state */ 1941 static void smc_listen_out_err(struct smc_sock *new_smc) 1942 { 1943 struct sock *newsmcsk = &new_smc->sk; 1944 struct net *net = sock_net(newsmcsk); 1945 1946 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1947 if (newsmcsk->sk_state == SMC_INIT) 1948 sock_put(&new_smc->sk); /* passive closing */ 1949 newsmcsk->sk_state = SMC_CLOSED; 1950 1951 smc_listen_out(new_smc); 1952 } 1953 1954 /* listen worker: decline and fall back if possible */ 1955 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1956 int local_first, u8 version) 1957 { 1958 /* RDMA setup failed, switch back to TCP */ 1959 smc_conn_abort(new_smc, local_first); 1960 if (reason_code < 0 || 1961 smc_switch_to_fallback(new_smc, reason_code)) { 1962 /* error, no fallback possible */ 1963 smc_listen_out_err(new_smc); 1964 return; 1965 } 1966 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1967 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1968 smc_listen_out_err(new_smc); 1969 return; 1970 } 1971 } 1972 smc_listen_out_connected(new_smc); 1973 } 1974 1975 /* listen worker: version checking */ 1976 static int smc_listen_v2_check(struct smc_sock *new_smc, 1977 struct smc_clc_msg_proposal *pclc, 1978 struct smc_init_info *ini) 1979 { 1980 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1981 struct smc_clc_v2_extension *pclc_v2_ext; 1982 int rc = SMC_CLC_DECL_PEERNOSMC; 1983 1984 ini->smc_type_v1 = pclc->hdr.typev1; 1985 ini->smc_type_v2 = pclc->hdr.typev2; 1986 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1987 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1988 if (pclc->hdr.version > SMC_V1) { 1989 if (smcd_indicated(ini->smc_type_v2)) 1990 ini->smcd_version |= SMC_V2; 1991 if (smcr_indicated(ini->smc_type_v2)) 1992 ini->smcr_version |= SMC_V2; 1993 } 1994 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1995 rc = SMC_CLC_DECL_PEERNOSMC; 1996 goto out; 1997 } 1998 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1999 if (!pclc_v2_ext) { 2000 ini->smcd_version &= ~SMC_V2; 2001 ini->smcr_version &= ~SMC_V2; 2002 rc = SMC_CLC_DECL_NOV2EXT; 2003 goto out; 2004 } 2005 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 2006 if (ini->smcd_version & SMC_V2) { 2007 if (!smc_ism_is_v2_capable()) { 2008 ini->smcd_version &= ~SMC_V2; 2009 rc = SMC_CLC_DECL_NOISM2SUPP; 2010 } else if (!pclc_smcd_v2_ext) { 2011 ini->smcd_version &= ~SMC_V2; 2012 rc = SMC_CLC_DECL_NOV2DEXT; 2013 } else if (!pclc_v2_ext->hdr.eid_cnt && 2014 !pclc_v2_ext->hdr.flag.seid) { 2015 ini->smcd_version &= ~SMC_V2; 2016 rc = SMC_CLC_DECL_NOUEID; 2017 } 2018 } 2019 if (ini->smcr_version & SMC_V2) { 2020 if (!pclc_v2_ext->hdr.eid_cnt) { 2021 ini->smcr_version &= ~SMC_V2; 2022 rc = SMC_CLC_DECL_NOUEID; 2023 } 2024 } 2025 2026 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2027 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2028 ini->release_nr = SMC_RELEASE; 2029 2030 out: 2031 if (!ini->smcd_version && !ini->smcr_version) 2032 return rc; 2033 2034 return 0; 2035 } 2036 2037 /* listen worker: check prefixes */ 2038 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2039 struct smc_clc_msg_proposal *pclc) 2040 { 2041 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2042 struct socket *newclcsock = new_smc->clcsock; 2043 2044 if (pclc->hdr.typev1 == SMC_TYPE_N) 2045 return 0; 2046 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2047 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2048 return SMC_CLC_DECL_DIFFPREFIX; 2049 2050 return 0; 2051 } 2052 2053 /* listen worker: initialize connection and buffers */ 2054 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2055 struct smc_init_info *ini) 2056 { 2057 int rc; 2058 2059 /* allocate connection / link group */ 2060 rc = smc_conn_create(new_smc, ini); 2061 if (rc) 2062 return rc; 2063 2064 /* create send buffer and rmb */ 2065 if (smc_buf_create(new_smc, false)) { 2066 smc_conn_abort(new_smc, ini->first_contact_local); 2067 return SMC_CLC_DECL_MEM; 2068 } 2069 2070 return 0; 2071 } 2072 2073 /* listen worker: initialize connection and buffers for SMC-D */ 2074 static int smc_listen_ism_init(struct smc_sock *new_smc, 2075 struct smc_init_info *ini) 2076 { 2077 int rc; 2078 2079 rc = smc_conn_create(new_smc, ini); 2080 if (rc) 2081 return rc; 2082 2083 /* Create send and receive buffers */ 2084 rc = smc_buf_create(new_smc, true); 2085 if (rc) { 2086 smc_conn_abort(new_smc, ini->first_contact_local); 2087 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2088 SMC_CLC_DECL_MEM; 2089 } 2090 2091 return 0; 2092 } 2093 2094 static bool smc_is_already_selected(struct smcd_dev *smcd, 2095 struct smc_init_info *ini, 2096 int matches) 2097 { 2098 int i; 2099 2100 for (i = 0; i < matches; i++) 2101 if (smcd == ini->ism_dev[i]) 2102 return true; 2103 2104 return false; 2105 } 2106 2107 /* check for ISM devices matching proposed ISM devices */ 2108 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2109 u16 proposed_chid, u64 proposed_gid, 2110 unsigned int *matches) 2111 { 2112 struct smcd_dev *smcd; 2113 2114 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2115 if (smcd->going_away) 2116 continue; 2117 if (smc_is_already_selected(smcd, ini, *matches)) 2118 continue; 2119 if (smc_ism_get_chid(smcd) == proposed_chid && 2120 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2121 ini->ism_peer_gid[*matches] = proposed_gid; 2122 ini->ism_dev[*matches] = smcd; 2123 (*matches)++; 2124 break; 2125 } 2126 } 2127 } 2128 2129 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2130 { 2131 if (!ini->rc) 2132 ini->rc = rc; 2133 } 2134 2135 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2136 struct smc_clc_msg_proposal *pclc, 2137 struct smc_init_info *ini) 2138 { 2139 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2140 struct smc_clc_v2_extension *smc_v2_ext; 2141 struct smc_clc_msg_smcd *pclc_smcd; 2142 unsigned int matches = 0; 2143 u8 smcd_version; 2144 u8 *eid = NULL; 2145 int i, rc; 2146 2147 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2148 goto not_found; 2149 2150 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2151 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2152 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2153 2154 mutex_lock(&smcd_dev_list.mutex); 2155 if (pclc_smcd->ism.chid) 2156 /* check for ISM device matching proposed native ISM device */ 2157 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2158 ntohll(pclc_smcd->ism.gid), &matches); 2159 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 2160 /* check for ISM devices matching proposed non-native ISM 2161 * devices 2162 */ 2163 smc_check_ism_v2_match(ini, 2164 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 2165 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 2166 &matches); 2167 } 2168 mutex_unlock(&smcd_dev_list.mutex); 2169 2170 if (!ini->ism_dev[0]) { 2171 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2172 goto not_found; 2173 } 2174 2175 smc_ism_get_system_eid(&eid); 2176 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2177 smcd_v2_ext->system_eid, eid)) 2178 goto not_found; 2179 2180 /* separate - outside the smcd_dev_list.lock */ 2181 smcd_version = ini->smcd_version; 2182 for (i = 0; i < matches; i++) { 2183 ini->smcd_version = SMC_V2; 2184 ini->is_smcd = true; 2185 ini->ism_selected = i; 2186 rc = smc_listen_ism_init(new_smc, ini); 2187 if (rc) { 2188 smc_find_ism_store_rc(rc, ini); 2189 /* try next active ISM device */ 2190 continue; 2191 } 2192 return; /* matching and usable V2 ISM device found */ 2193 } 2194 /* no V2 ISM device could be initialized */ 2195 ini->smcd_version = smcd_version; /* restore original value */ 2196 ini->negotiated_eid[0] = 0; 2197 2198 not_found: 2199 ini->smcd_version &= ~SMC_V2; 2200 ini->ism_dev[0] = NULL; 2201 ini->is_smcd = false; 2202 } 2203 2204 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2205 struct smc_clc_msg_proposal *pclc, 2206 struct smc_init_info *ini) 2207 { 2208 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2209 int rc = 0; 2210 2211 /* check if ISM V1 is available */ 2212 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2213 goto not_found; 2214 ini->is_smcd = true; /* prepare ISM check */ 2215 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 2216 rc = smc_find_ism_device(new_smc, ini); 2217 if (rc) 2218 goto not_found; 2219 ini->ism_selected = 0; 2220 rc = smc_listen_ism_init(new_smc, ini); 2221 if (!rc) 2222 return; /* V1 ISM device found */ 2223 2224 not_found: 2225 smc_find_ism_store_rc(rc, ini); 2226 ini->smcd_version &= ~SMC_V1; 2227 ini->ism_dev[0] = NULL; 2228 ini->is_smcd = false; 2229 } 2230 2231 /* listen worker: register buffers */ 2232 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2233 { 2234 struct smc_connection *conn = &new_smc->conn; 2235 2236 if (!local_first) { 2237 /* reg sendbufs if they were vzalloced */ 2238 if (conn->sndbuf_desc->is_vm) { 2239 if (smcr_lgr_reg_sndbufs(conn->lnk, 2240 conn->sndbuf_desc)) 2241 return SMC_CLC_DECL_ERR_REGBUF; 2242 } 2243 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2244 return SMC_CLC_DECL_ERR_REGBUF; 2245 } 2246 2247 return 0; 2248 } 2249 2250 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2251 struct smc_clc_msg_proposal *pclc, 2252 struct smc_init_info *ini) 2253 { 2254 struct smc_clc_v2_extension *smc_v2_ext; 2255 u8 smcr_version; 2256 int rc; 2257 2258 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2259 goto not_found; 2260 2261 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2262 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2263 goto not_found; 2264 2265 /* prepare RDMA check */ 2266 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2267 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2268 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2269 ini->check_smcrv2 = true; 2270 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2271 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2272 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2273 rc = smc_find_rdma_device(new_smc, ini); 2274 if (rc) { 2275 smc_find_ism_store_rc(rc, ini); 2276 goto not_found; 2277 } 2278 if (!ini->smcrv2.uses_gateway) 2279 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2280 2281 smcr_version = ini->smcr_version; 2282 ini->smcr_version = SMC_V2; 2283 rc = smc_listen_rdma_init(new_smc, ini); 2284 if (!rc) { 2285 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2286 if (rc) 2287 smc_conn_abort(new_smc, ini->first_contact_local); 2288 } 2289 if (!rc) 2290 return; 2291 ini->smcr_version = smcr_version; 2292 smc_find_ism_store_rc(rc, ini); 2293 2294 not_found: 2295 ini->smcr_version &= ~SMC_V2; 2296 ini->smcrv2.ib_dev_v2 = NULL; 2297 ini->check_smcrv2 = false; 2298 } 2299 2300 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2301 struct smc_clc_msg_proposal *pclc, 2302 struct smc_init_info *ini) 2303 { 2304 int rc; 2305 2306 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2307 return SMC_CLC_DECL_NOSMCDEV; 2308 2309 /* prepare RDMA check */ 2310 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2311 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2312 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2313 rc = smc_find_rdma_device(new_smc, ini); 2314 if (rc) { 2315 /* no RDMA device found */ 2316 return SMC_CLC_DECL_NOSMCDEV; 2317 } 2318 rc = smc_listen_rdma_init(new_smc, ini); 2319 if (rc) 2320 return rc; 2321 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2322 } 2323 2324 /* determine the local device matching to proposal */ 2325 static int smc_listen_find_device(struct smc_sock *new_smc, 2326 struct smc_clc_msg_proposal *pclc, 2327 struct smc_init_info *ini) 2328 { 2329 int prfx_rc; 2330 2331 /* check for ISM device matching V2 proposed device */ 2332 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2333 if (ini->ism_dev[0]) 2334 return 0; 2335 2336 /* check for matching IP prefix and subnet length (V1) */ 2337 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2338 if (prfx_rc) 2339 smc_find_ism_store_rc(prfx_rc, ini); 2340 2341 /* get vlan id from IP device */ 2342 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2343 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2344 2345 /* check for ISM device matching V1 proposed device */ 2346 if (!prfx_rc) 2347 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2348 if (ini->ism_dev[0]) 2349 return 0; 2350 2351 if (!smcr_indicated(pclc->hdr.typev1) && 2352 !smcr_indicated(pclc->hdr.typev2)) 2353 /* skip RDMA and decline */ 2354 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2355 2356 /* check if RDMA V2 is available */ 2357 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2358 if (ini->smcrv2.ib_dev_v2) 2359 return 0; 2360 2361 /* check if RDMA V1 is available */ 2362 if (!prfx_rc) { 2363 int rc; 2364 2365 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2366 smc_find_ism_store_rc(rc, ini); 2367 return (!rc) ? 0 : ini->rc; 2368 } 2369 return prfx_rc; 2370 } 2371 2372 /* listen worker: finish RDMA setup */ 2373 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2374 struct smc_clc_msg_accept_confirm *cclc, 2375 bool local_first, 2376 struct smc_init_info *ini) 2377 { 2378 struct smc_link *link = new_smc->conn.lnk; 2379 int reason_code = 0; 2380 2381 if (local_first) 2382 smc_link_save_peer_info(link, cclc, ini); 2383 2384 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2385 return SMC_CLC_DECL_ERR_RTOK; 2386 2387 if (local_first) { 2388 if (smc_ib_ready_link(link)) 2389 return SMC_CLC_DECL_ERR_RDYLNK; 2390 /* QP confirmation over RoCE fabric */ 2391 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2392 reason_code = smcr_serv_conf_first_link(new_smc); 2393 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2394 } 2395 return reason_code; 2396 } 2397 2398 /* setup for connection of server */ 2399 static void smc_listen_work(struct work_struct *work) 2400 { 2401 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2402 smc_listen_work); 2403 struct socket *newclcsock = new_smc->clcsock; 2404 struct smc_clc_msg_accept_confirm *cclc; 2405 struct smc_clc_msg_proposal_area *buf; 2406 struct smc_clc_msg_proposal *pclc; 2407 struct smc_init_info *ini = NULL; 2408 u8 proposal_version = SMC_V1; 2409 u8 accept_version; 2410 int rc = 0; 2411 2412 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2413 return smc_listen_out_err(new_smc); 2414 2415 if (new_smc->use_fallback) { 2416 smc_listen_out_connected(new_smc); 2417 return; 2418 } 2419 2420 /* check if peer is smc capable */ 2421 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2422 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2423 if (rc) 2424 smc_listen_out_err(new_smc); 2425 else 2426 smc_listen_out_connected(new_smc); 2427 return; 2428 } 2429 2430 /* do inband token exchange - 2431 * wait for and receive SMC Proposal CLC message 2432 */ 2433 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2434 if (!buf) { 2435 rc = SMC_CLC_DECL_MEM; 2436 goto out_decl; 2437 } 2438 pclc = (struct smc_clc_msg_proposal *)buf; 2439 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2440 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2441 if (rc) 2442 goto out_decl; 2443 2444 if (pclc->hdr.version > SMC_V1) 2445 proposal_version = SMC_V2; 2446 2447 /* IPSec connections opt out of SMC optimizations */ 2448 if (using_ipsec(new_smc)) { 2449 rc = SMC_CLC_DECL_IPSEC; 2450 goto out_decl; 2451 } 2452 2453 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2454 if (!ini) { 2455 rc = SMC_CLC_DECL_MEM; 2456 goto out_decl; 2457 } 2458 2459 /* initial version checking */ 2460 rc = smc_listen_v2_check(new_smc, pclc, ini); 2461 if (rc) 2462 goto out_decl; 2463 2464 rc = smc_clc_srv_v2x_features_validate(pclc, ini); 2465 if (rc) 2466 goto out_decl; 2467 2468 mutex_lock(&smc_server_lgr_pending); 2469 smc_close_init(new_smc); 2470 smc_rx_init(new_smc); 2471 smc_tx_init(new_smc); 2472 2473 /* determine ISM or RoCE device used for connection */ 2474 rc = smc_listen_find_device(new_smc, pclc, ini); 2475 if (rc) 2476 goto out_unlock; 2477 2478 /* send SMC Accept CLC message */ 2479 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2480 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2481 accept_version, ini->negotiated_eid, ini); 2482 if (rc) 2483 goto out_unlock; 2484 2485 /* SMC-D does not need this lock any more */ 2486 if (ini->is_smcd) 2487 mutex_unlock(&smc_server_lgr_pending); 2488 2489 /* receive SMC Confirm CLC message */ 2490 memset(buf, 0, sizeof(*buf)); 2491 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2492 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2493 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2494 if (rc) { 2495 if (!ini->is_smcd) 2496 goto out_unlock; 2497 goto out_decl; 2498 } 2499 2500 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2501 if (rc) { 2502 if (!ini->is_smcd) 2503 goto out_unlock; 2504 goto out_decl; 2505 } 2506 2507 /* fce smc release version is needed in smc_listen_rdma_finish, 2508 * so save fce info here. 2509 */ 2510 smc_conn_save_peer_info_fce(new_smc, cclc); 2511 2512 /* finish worker */ 2513 if (!ini->is_smcd) { 2514 rc = smc_listen_rdma_finish(new_smc, cclc, 2515 ini->first_contact_local, ini); 2516 if (rc) 2517 goto out_unlock; 2518 mutex_unlock(&smc_server_lgr_pending); 2519 } 2520 smc_conn_save_peer_info(new_smc, cclc); 2521 smc_listen_out_connected(new_smc); 2522 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2523 goto out_free; 2524 2525 out_unlock: 2526 mutex_unlock(&smc_server_lgr_pending); 2527 out_decl: 2528 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2529 proposal_version); 2530 out_free: 2531 kfree(ini); 2532 kfree(buf); 2533 } 2534 2535 static void smc_tcp_listen_work(struct work_struct *work) 2536 { 2537 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2538 tcp_listen_work); 2539 struct sock *lsk = &lsmc->sk; 2540 struct smc_sock *new_smc; 2541 int rc = 0; 2542 2543 lock_sock(lsk); 2544 while (lsk->sk_state == SMC_LISTEN) { 2545 rc = smc_clcsock_accept(lsmc, &new_smc); 2546 if (rc) /* clcsock accept queue empty or error */ 2547 goto out; 2548 if (!new_smc) 2549 continue; 2550 2551 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2552 atomic_inc(&lsmc->queued_smc_hs); 2553 2554 new_smc->listen_smc = lsmc; 2555 new_smc->use_fallback = lsmc->use_fallback; 2556 new_smc->fallback_rsn = lsmc->fallback_rsn; 2557 sock_hold(lsk); /* sock_put in smc_listen_work */ 2558 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2559 smc_copy_sock_settings_to_smc(new_smc); 2560 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2561 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2562 sock_put(&new_smc->sk); 2563 } 2564 2565 out: 2566 release_sock(lsk); 2567 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2568 } 2569 2570 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2571 { 2572 struct smc_sock *lsmc; 2573 2574 read_lock_bh(&listen_clcsock->sk_callback_lock); 2575 lsmc = smc_clcsock_user_data(listen_clcsock); 2576 if (!lsmc) 2577 goto out; 2578 lsmc->clcsk_data_ready(listen_clcsock); 2579 if (lsmc->sk.sk_state == SMC_LISTEN) { 2580 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2581 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2582 sock_put(&lsmc->sk); 2583 } 2584 out: 2585 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2586 } 2587 2588 static int smc_listen(struct socket *sock, int backlog) 2589 { 2590 struct sock *sk = sock->sk; 2591 struct smc_sock *smc; 2592 int rc; 2593 2594 smc = smc_sk(sk); 2595 lock_sock(sk); 2596 2597 rc = -EINVAL; 2598 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2599 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2600 goto out; 2601 2602 rc = 0; 2603 if (sk->sk_state == SMC_LISTEN) { 2604 sk->sk_max_ack_backlog = backlog; 2605 goto out; 2606 } 2607 /* some socket options are handled in core, so we could not apply 2608 * them to the clc socket -- copy smc socket options to clc socket 2609 */ 2610 smc_copy_sock_settings_to_clc(smc); 2611 if (!smc->use_fallback) 2612 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2613 2614 /* save original sk_data_ready function and establish 2615 * smc-specific sk_data_ready function 2616 */ 2617 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2618 smc->clcsock->sk->sk_user_data = 2619 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2620 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2621 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2622 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2623 2624 /* save original ops */ 2625 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2626 2627 smc->af_ops = *smc->ori_af_ops; 2628 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2629 2630 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2631 2632 if (smc->limit_smc_hs) 2633 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2634 2635 rc = kernel_listen(smc->clcsock, backlog); 2636 if (rc) { 2637 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2638 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2639 &smc->clcsk_data_ready); 2640 smc->clcsock->sk->sk_user_data = NULL; 2641 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2642 goto out; 2643 } 2644 sk->sk_max_ack_backlog = backlog; 2645 sk->sk_ack_backlog = 0; 2646 sk->sk_state = SMC_LISTEN; 2647 2648 out: 2649 release_sock(sk); 2650 return rc; 2651 } 2652 2653 static int smc_accept(struct socket *sock, struct socket *new_sock, 2654 int flags, bool kern) 2655 { 2656 struct sock *sk = sock->sk, *nsk; 2657 DECLARE_WAITQUEUE(wait, current); 2658 struct smc_sock *lsmc; 2659 long timeo; 2660 int rc = 0; 2661 2662 lsmc = smc_sk(sk); 2663 sock_hold(sk); /* sock_put below */ 2664 lock_sock(sk); 2665 2666 if (lsmc->sk.sk_state != SMC_LISTEN) { 2667 rc = -EINVAL; 2668 release_sock(sk); 2669 goto out; 2670 } 2671 2672 /* Wait for an incoming connection */ 2673 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2674 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2675 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2676 set_current_state(TASK_INTERRUPTIBLE); 2677 if (!timeo) { 2678 rc = -EAGAIN; 2679 break; 2680 } 2681 release_sock(sk); 2682 timeo = schedule_timeout(timeo); 2683 /* wakeup by sk_data_ready in smc_listen_work() */ 2684 sched_annotate_sleep(); 2685 lock_sock(sk); 2686 if (signal_pending(current)) { 2687 rc = sock_intr_errno(timeo); 2688 break; 2689 } 2690 } 2691 set_current_state(TASK_RUNNING); 2692 remove_wait_queue(sk_sleep(sk), &wait); 2693 2694 if (!rc) 2695 rc = sock_error(nsk); 2696 release_sock(sk); 2697 if (rc) 2698 goto out; 2699 2700 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2701 /* wait till data arrives on the socket */ 2702 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2703 MSEC_PER_SEC); 2704 if (smc_sk(nsk)->use_fallback) { 2705 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2706 2707 lock_sock(clcsk); 2708 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2709 sk_wait_data(clcsk, &timeo, NULL); 2710 release_sock(clcsk); 2711 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2712 lock_sock(nsk); 2713 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2714 release_sock(nsk); 2715 } 2716 } 2717 2718 out: 2719 sock_put(sk); /* sock_hold above */ 2720 return rc; 2721 } 2722 2723 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2724 int peer) 2725 { 2726 struct smc_sock *smc; 2727 2728 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2729 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2730 return -ENOTCONN; 2731 2732 smc = smc_sk(sock->sk); 2733 2734 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2735 } 2736 2737 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2738 { 2739 struct sock *sk = sock->sk; 2740 struct smc_sock *smc; 2741 int rc; 2742 2743 smc = smc_sk(sk); 2744 lock_sock(sk); 2745 2746 /* SMC does not support connect with fastopen */ 2747 if (msg->msg_flags & MSG_FASTOPEN) { 2748 /* not connected yet, fallback */ 2749 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2750 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2751 if (rc) 2752 goto out; 2753 } else { 2754 rc = -EINVAL; 2755 goto out; 2756 } 2757 } else if ((sk->sk_state != SMC_ACTIVE) && 2758 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2759 (sk->sk_state != SMC_INIT)) { 2760 rc = -EPIPE; 2761 goto out; 2762 } 2763 2764 if (smc->use_fallback) { 2765 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2766 } else { 2767 rc = smc_tx_sendmsg(smc, msg, len); 2768 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2769 } 2770 out: 2771 release_sock(sk); 2772 return rc; 2773 } 2774 2775 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2776 int flags) 2777 { 2778 struct sock *sk = sock->sk; 2779 struct smc_sock *smc; 2780 int rc = -ENOTCONN; 2781 2782 smc = smc_sk(sk); 2783 lock_sock(sk); 2784 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2785 /* socket was connected before, no more data to read */ 2786 rc = 0; 2787 goto out; 2788 } 2789 if ((sk->sk_state == SMC_INIT) || 2790 (sk->sk_state == SMC_LISTEN) || 2791 (sk->sk_state == SMC_CLOSED)) 2792 goto out; 2793 2794 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2795 rc = 0; 2796 goto out; 2797 } 2798 2799 if (smc->use_fallback) { 2800 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2801 } else { 2802 msg->msg_namelen = 0; 2803 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2804 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2805 } 2806 2807 out: 2808 release_sock(sk); 2809 return rc; 2810 } 2811 2812 static __poll_t smc_accept_poll(struct sock *parent) 2813 { 2814 struct smc_sock *isk = smc_sk(parent); 2815 __poll_t mask = 0; 2816 2817 spin_lock(&isk->accept_q_lock); 2818 if (!list_empty(&isk->accept_q)) 2819 mask = EPOLLIN | EPOLLRDNORM; 2820 spin_unlock(&isk->accept_q_lock); 2821 2822 return mask; 2823 } 2824 2825 static __poll_t smc_poll(struct file *file, struct socket *sock, 2826 poll_table *wait) 2827 { 2828 struct sock *sk = sock->sk; 2829 struct smc_sock *smc; 2830 __poll_t mask = 0; 2831 2832 if (!sk) 2833 return EPOLLNVAL; 2834 2835 smc = smc_sk(sock->sk); 2836 if (smc->use_fallback) { 2837 /* delegate to CLC child sock */ 2838 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2839 sk->sk_err = smc->clcsock->sk->sk_err; 2840 } else { 2841 if (sk->sk_state != SMC_CLOSED) 2842 sock_poll_wait(file, sock, wait); 2843 if (sk->sk_err) 2844 mask |= EPOLLERR; 2845 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2846 (sk->sk_state == SMC_CLOSED)) 2847 mask |= EPOLLHUP; 2848 if (sk->sk_state == SMC_LISTEN) { 2849 /* woken up by sk_data_ready in smc_listen_work() */ 2850 mask |= smc_accept_poll(sk); 2851 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2852 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2853 wait); 2854 sk->sk_err = smc->clcsock->sk->sk_err; 2855 } else { 2856 if ((sk->sk_state != SMC_INIT && 2857 atomic_read(&smc->conn.sndbuf_space)) || 2858 sk->sk_shutdown & SEND_SHUTDOWN) { 2859 mask |= EPOLLOUT | EPOLLWRNORM; 2860 } else { 2861 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2862 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2863 } 2864 if (atomic_read(&smc->conn.bytes_to_rcv)) 2865 mask |= EPOLLIN | EPOLLRDNORM; 2866 if (sk->sk_shutdown & RCV_SHUTDOWN) 2867 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2868 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2869 mask |= EPOLLIN; 2870 if (smc->conn.urg_state == SMC_URG_VALID) 2871 mask |= EPOLLPRI; 2872 } 2873 } 2874 2875 return mask; 2876 } 2877 2878 static int smc_shutdown(struct socket *sock, int how) 2879 { 2880 struct sock *sk = sock->sk; 2881 bool do_shutdown = true; 2882 struct smc_sock *smc; 2883 int rc = -EINVAL; 2884 int old_state; 2885 int rc1 = 0; 2886 2887 smc = smc_sk(sk); 2888 2889 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2890 return rc; 2891 2892 lock_sock(sk); 2893 2894 if (sock->state == SS_CONNECTING) { 2895 if (sk->sk_state == SMC_ACTIVE) 2896 sock->state = SS_CONNECTED; 2897 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2898 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2899 sk->sk_state == SMC_APPCLOSEWAIT1 || 2900 sk->sk_state == SMC_APPCLOSEWAIT2 || 2901 sk->sk_state == SMC_APPFINCLOSEWAIT) 2902 sock->state = SS_DISCONNECTING; 2903 } 2904 2905 rc = -ENOTCONN; 2906 if ((sk->sk_state != SMC_ACTIVE) && 2907 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2908 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2909 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2910 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2911 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2912 goto out; 2913 if (smc->use_fallback) { 2914 rc = kernel_sock_shutdown(smc->clcsock, how); 2915 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2916 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2917 sk->sk_state = SMC_CLOSED; 2918 sk->sk_socket->state = SS_UNCONNECTED; 2919 sock_put(sk); 2920 } 2921 goto out; 2922 } 2923 switch (how) { 2924 case SHUT_RDWR: /* shutdown in both directions */ 2925 old_state = sk->sk_state; 2926 rc = smc_close_active(smc); 2927 if (old_state == SMC_ACTIVE && 2928 sk->sk_state == SMC_PEERCLOSEWAIT1) 2929 do_shutdown = false; 2930 break; 2931 case SHUT_WR: 2932 rc = smc_close_shutdown_write(smc); 2933 break; 2934 case SHUT_RD: 2935 rc = 0; 2936 /* nothing more to do because peer is not involved */ 2937 break; 2938 } 2939 if (do_shutdown && smc->clcsock) 2940 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2941 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2942 sk->sk_shutdown |= how + 1; 2943 2944 if (sk->sk_state == SMC_CLOSED) 2945 sock->state = SS_UNCONNECTED; 2946 else 2947 sock->state = SS_DISCONNECTING; 2948 out: 2949 release_sock(sk); 2950 return rc ? rc : rc1; 2951 } 2952 2953 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2954 char __user *optval, int __user *optlen) 2955 { 2956 struct smc_sock *smc; 2957 int val, len; 2958 2959 smc = smc_sk(sock->sk); 2960 2961 if (get_user(len, optlen)) 2962 return -EFAULT; 2963 2964 len = min_t(int, len, sizeof(int)); 2965 2966 if (len < 0) 2967 return -EINVAL; 2968 2969 switch (optname) { 2970 case SMC_LIMIT_HS: 2971 val = smc->limit_smc_hs; 2972 break; 2973 default: 2974 return -EOPNOTSUPP; 2975 } 2976 2977 if (put_user(len, optlen)) 2978 return -EFAULT; 2979 if (copy_to_user(optval, &val, len)) 2980 return -EFAULT; 2981 2982 return 0; 2983 } 2984 2985 static int __smc_setsockopt(struct socket *sock, int level, int optname, 2986 sockptr_t optval, unsigned int optlen) 2987 { 2988 struct sock *sk = sock->sk; 2989 struct smc_sock *smc; 2990 int val, rc; 2991 2992 smc = smc_sk(sk); 2993 2994 lock_sock(sk); 2995 switch (optname) { 2996 case SMC_LIMIT_HS: 2997 if (optlen < sizeof(int)) { 2998 rc = -EINVAL; 2999 break; 3000 } 3001 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3002 rc = -EFAULT; 3003 break; 3004 } 3005 3006 smc->limit_smc_hs = !!val; 3007 rc = 0; 3008 break; 3009 default: 3010 rc = -EOPNOTSUPP; 3011 break; 3012 } 3013 release_sock(sk); 3014 3015 return rc; 3016 } 3017 3018 static int smc_setsockopt(struct socket *sock, int level, int optname, 3019 sockptr_t optval, unsigned int optlen) 3020 { 3021 struct sock *sk = sock->sk; 3022 struct smc_sock *smc; 3023 int val, rc; 3024 3025 if (level == SOL_TCP && optname == TCP_ULP) 3026 return -EOPNOTSUPP; 3027 else if (level == SOL_SMC) 3028 return __smc_setsockopt(sock, level, optname, optval, optlen); 3029 3030 smc = smc_sk(sk); 3031 3032 /* generic setsockopts reaching us here always apply to the 3033 * CLC socket 3034 */ 3035 mutex_lock(&smc->clcsock_release_lock); 3036 if (!smc->clcsock) { 3037 mutex_unlock(&smc->clcsock_release_lock); 3038 return -EBADF; 3039 } 3040 if (unlikely(!smc->clcsock->ops->setsockopt)) 3041 rc = -EOPNOTSUPP; 3042 else 3043 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3044 optval, optlen); 3045 if (smc->clcsock->sk->sk_err) { 3046 sk->sk_err = smc->clcsock->sk->sk_err; 3047 sk_error_report(sk); 3048 } 3049 mutex_unlock(&smc->clcsock_release_lock); 3050 3051 if (optlen < sizeof(int)) 3052 return -EINVAL; 3053 if (copy_from_sockptr(&val, optval, sizeof(int))) 3054 return -EFAULT; 3055 3056 lock_sock(sk); 3057 if (rc || smc->use_fallback) 3058 goto out; 3059 switch (optname) { 3060 case TCP_FASTOPEN: 3061 case TCP_FASTOPEN_CONNECT: 3062 case TCP_FASTOPEN_KEY: 3063 case TCP_FASTOPEN_NO_COOKIE: 3064 /* option not supported by SMC */ 3065 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3066 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3067 } else { 3068 rc = -EINVAL; 3069 } 3070 break; 3071 case TCP_NODELAY: 3072 if (sk->sk_state != SMC_INIT && 3073 sk->sk_state != SMC_LISTEN && 3074 sk->sk_state != SMC_CLOSED) { 3075 if (val) { 3076 SMC_STAT_INC(smc, ndly_cnt); 3077 smc_tx_pending(&smc->conn); 3078 cancel_delayed_work(&smc->conn.tx_work); 3079 } 3080 } 3081 break; 3082 case TCP_CORK: 3083 if (sk->sk_state != SMC_INIT && 3084 sk->sk_state != SMC_LISTEN && 3085 sk->sk_state != SMC_CLOSED) { 3086 if (!val) { 3087 SMC_STAT_INC(smc, cork_cnt); 3088 smc_tx_pending(&smc->conn); 3089 cancel_delayed_work(&smc->conn.tx_work); 3090 } 3091 } 3092 break; 3093 case TCP_DEFER_ACCEPT: 3094 smc->sockopt_defer_accept = val; 3095 break; 3096 default: 3097 break; 3098 } 3099 out: 3100 release_sock(sk); 3101 3102 return rc; 3103 } 3104 3105 static int smc_getsockopt(struct socket *sock, int level, int optname, 3106 char __user *optval, int __user *optlen) 3107 { 3108 struct smc_sock *smc; 3109 int rc; 3110 3111 if (level == SOL_SMC) 3112 return __smc_getsockopt(sock, level, optname, optval, optlen); 3113 3114 smc = smc_sk(sock->sk); 3115 mutex_lock(&smc->clcsock_release_lock); 3116 if (!smc->clcsock) { 3117 mutex_unlock(&smc->clcsock_release_lock); 3118 return -EBADF; 3119 } 3120 /* socket options apply to the CLC socket */ 3121 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3122 mutex_unlock(&smc->clcsock_release_lock); 3123 return -EOPNOTSUPP; 3124 } 3125 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3126 optval, optlen); 3127 mutex_unlock(&smc->clcsock_release_lock); 3128 return rc; 3129 } 3130 3131 static int smc_ioctl(struct socket *sock, unsigned int cmd, 3132 unsigned long arg) 3133 { 3134 union smc_host_cursor cons, urg; 3135 struct smc_connection *conn; 3136 struct smc_sock *smc; 3137 int answ; 3138 3139 smc = smc_sk(sock->sk); 3140 conn = &smc->conn; 3141 lock_sock(&smc->sk); 3142 if (smc->use_fallback) { 3143 if (!smc->clcsock) { 3144 release_sock(&smc->sk); 3145 return -EBADF; 3146 } 3147 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3148 release_sock(&smc->sk); 3149 return answ; 3150 } 3151 switch (cmd) { 3152 case SIOCINQ: /* same as FIONREAD */ 3153 if (smc->sk.sk_state == SMC_LISTEN) { 3154 release_sock(&smc->sk); 3155 return -EINVAL; 3156 } 3157 if (smc->sk.sk_state == SMC_INIT || 3158 smc->sk.sk_state == SMC_CLOSED) 3159 answ = 0; 3160 else 3161 answ = atomic_read(&smc->conn.bytes_to_rcv); 3162 break; 3163 case SIOCOUTQ: 3164 /* output queue size (not send + not acked) */ 3165 if (smc->sk.sk_state == SMC_LISTEN) { 3166 release_sock(&smc->sk); 3167 return -EINVAL; 3168 } 3169 if (smc->sk.sk_state == SMC_INIT || 3170 smc->sk.sk_state == SMC_CLOSED) 3171 answ = 0; 3172 else 3173 answ = smc->conn.sndbuf_desc->len - 3174 atomic_read(&smc->conn.sndbuf_space); 3175 break; 3176 case SIOCOUTQNSD: 3177 /* output queue size (not send only) */ 3178 if (smc->sk.sk_state == SMC_LISTEN) { 3179 release_sock(&smc->sk); 3180 return -EINVAL; 3181 } 3182 if (smc->sk.sk_state == SMC_INIT || 3183 smc->sk.sk_state == SMC_CLOSED) 3184 answ = 0; 3185 else 3186 answ = smc_tx_prepared_sends(&smc->conn); 3187 break; 3188 case SIOCATMARK: 3189 if (smc->sk.sk_state == SMC_LISTEN) { 3190 release_sock(&smc->sk); 3191 return -EINVAL; 3192 } 3193 if (smc->sk.sk_state == SMC_INIT || 3194 smc->sk.sk_state == SMC_CLOSED) { 3195 answ = 0; 3196 } else { 3197 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3198 smc_curs_copy(&urg, &conn->urg_curs, conn); 3199 answ = smc_curs_diff(conn->rmb_desc->len, 3200 &cons, &urg) == 1; 3201 } 3202 break; 3203 default: 3204 release_sock(&smc->sk); 3205 return -ENOIOCTLCMD; 3206 } 3207 release_sock(&smc->sk); 3208 3209 return put_user(answ, (int __user *)arg); 3210 } 3211 3212 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3213 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3214 * updates till whenever a respective page has been fully processed. 3215 * Note that subsequent recv() calls have to wait till all splice() processing 3216 * completed. 3217 */ 3218 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3219 struct pipe_inode_info *pipe, size_t len, 3220 unsigned int flags) 3221 { 3222 struct sock *sk = sock->sk; 3223 struct smc_sock *smc; 3224 int rc = -ENOTCONN; 3225 3226 smc = smc_sk(sk); 3227 lock_sock(sk); 3228 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3229 /* socket was connected before, no more data to read */ 3230 rc = 0; 3231 goto out; 3232 } 3233 if (sk->sk_state == SMC_INIT || 3234 sk->sk_state == SMC_LISTEN || 3235 sk->sk_state == SMC_CLOSED) 3236 goto out; 3237 3238 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3239 rc = 0; 3240 goto out; 3241 } 3242 3243 if (smc->use_fallback) { 3244 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3245 pipe, len, flags); 3246 } else { 3247 if (*ppos) { 3248 rc = -ESPIPE; 3249 goto out; 3250 } 3251 if (flags & SPLICE_F_NONBLOCK) 3252 flags = MSG_DONTWAIT; 3253 else 3254 flags = 0; 3255 SMC_STAT_INC(smc, splice_cnt); 3256 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3257 } 3258 out: 3259 release_sock(sk); 3260 3261 return rc; 3262 } 3263 3264 /* must look like tcp */ 3265 static const struct proto_ops smc_sock_ops = { 3266 .family = PF_SMC, 3267 .owner = THIS_MODULE, 3268 .release = smc_release, 3269 .bind = smc_bind, 3270 .connect = smc_connect, 3271 .socketpair = sock_no_socketpair, 3272 .accept = smc_accept, 3273 .getname = smc_getname, 3274 .poll = smc_poll, 3275 .ioctl = smc_ioctl, 3276 .listen = smc_listen, 3277 .shutdown = smc_shutdown, 3278 .setsockopt = smc_setsockopt, 3279 .getsockopt = smc_getsockopt, 3280 .sendmsg = smc_sendmsg, 3281 .recvmsg = smc_recvmsg, 3282 .mmap = sock_no_mmap, 3283 .splice_read = smc_splice_read, 3284 }; 3285 3286 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3287 int kern, struct socket *clcsock) 3288 { 3289 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3290 struct smc_sock *smc; 3291 struct sock *sk; 3292 int rc; 3293 3294 rc = -ESOCKTNOSUPPORT; 3295 if (sock->type != SOCK_STREAM) 3296 goto out; 3297 3298 rc = -EPROTONOSUPPORT; 3299 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3300 goto out; 3301 3302 rc = -ENOBUFS; 3303 sock->ops = &smc_sock_ops; 3304 sock->state = SS_UNCONNECTED; 3305 sk = smc_sock_alloc(net, sock, protocol); 3306 if (!sk) 3307 goto out; 3308 3309 /* create internal TCP socket for CLC handshake and fallback */ 3310 smc = smc_sk(sk); 3311 smc->use_fallback = false; /* assume rdma capability first */ 3312 smc->fallback_rsn = 0; 3313 3314 /* default behavior from limit_smc_hs in every net namespace */ 3315 smc->limit_smc_hs = net->smc.limit_smc_hs; 3316 3317 rc = 0; 3318 if (!clcsock) { 3319 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3320 &smc->clcsock); 3321 if (rc) { 3322 sk_common_release(sk); 3323 goto out; 3324 } 3325 3326 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3327 * destruction; its sk_state might not be TCP_CLOSE after 3328 * smc->sk is close()d, and TCP timers can be fired later, 3329 * which need net ref. 3330 */ 3331 sk = smc->clcsock->sk; 3332 __netns_tracker_free(net, &sk->ns_tracker, false); 3333 sk->sk_net_refcnt = 1; 3334 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3335 sock_inuse_add(net, 1); 3336 } else { 3337 smc->clcsock = clcsock; 3338 } 3339 3340 out: 3341 return rc; 3342 } 3343 3344 static int smc_create(struct net *net, struct socket *sock, int protocol, 3345 int kern) 3346 { 3347 return __smc_create(net, sock, protocol, kern, NULL); 3348 } 3349 3350 static const struct net_proto_family smc_sock_family_ops = { 3351 .family = PF_SMC, 3352 .owner = THIS_MODULE, 3353 .create = smc_create, 3354 }; 3355 3356 static int smc_ulp_init(struct sock *sk) 3357 { 3358 struct socket *tcp = sk->sk_socket; 3359 struct net *net = sock_net(sk); 3360 struct socket *smcsock; 3361 int protocol, ret; 3362 3363 /* only TCP can be replaced */ 3364 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3365 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3366 return -ESOCKTNOSUPPORT; 3367 /* don't handle wq now */ 3368 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3369 return -ENOTCONN; 3370 3371 if (sk->sk_family == AF_INET) 3372 protocol = SMCPROTO_SMC; 3373 else 3374 protocol = SMCPROTO_SMC6; 3375 3376 smcsock = sock_alloc(); 3377 if (!smcsock) 3378 return -ENFILE; 3379 3380 smcsock->type = SOCK_STREAM; 3381 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3382 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3383 if (ret) { 3384 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3385 return ret; 3386 } 3387 3388 /* replace tcp socket to smc */ 3389 smcsock->file = tcp->file; 3390 smcsock->file->private_data = smcsock; 3391 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3392 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3393 tcp->file = NULL; 3394 3395 return ret; 3396 } 3397 3398 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3399 const gfp_t priority) 3400 { 3401 struct inet_connection_sock *icsk = inet_csk(newsk); 3402 3403 /* don't inherit ulp ops to child when listen */ 3404 icsk->icsk_ulp_ops = NULL; 3405 } 3406 3407 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3408 .name = "smc", 3409 .owner = THIS_MODULE, 3410 .init = smc_ulp_init, 3411 .clone = smc_ulp_clone, 3412 }; 3413 3414 unsigned int smc_net_id; 3415 3416 static __net_init int smc_net_init(struct net *net) 3417 { 3418 int rc; 3419 3420 rc = smc_sysctl_net_init(net); 3421 if (rc) 3422 return rc; 3423 return smc_pnet_net_init(net); 3424 } 3425 3426 static void __net_exit smc_net_exit(struct net *net) 3427 { 3428 smc_sysctl_net_exit(net); 3429 smc_pnet_net_exit(net); 3430 } 3431 3432 static __net_init int smc_net_stat_init(struct net *net) 3433 { 3434 return smc_stats_init(net); 3435 } 3436 3437 static void __net_exit smc_net_stat_exit(struct net *net) 3438 { 3439 smc_stats_exit(net); 3440 } 3441 3442 static struct pernet_operations smc_net_ops = { 3443 .init = smc_net_init, 3444 .exit = smc_net_exit, 3445 .id = &smc_net_id, 3446 .size = sizeof(struct smc_net), 3447 }; 3448 3449 static struct pernet_operations smc_net_stat_ops = { 3450 .init = smc_net_stat_init, 3451 .exit = smc_net_stat_exit, 3452 }; 3453 3454 static int __init smc_init(void) 3455 { 3456 int rc; 3457 3458 rc = register_pernet_subsys(&smc_net_ops); 3459 if (rc) 3460 return rc; 3461 3462 rc = register_pernet_subsys(&smc_net_stat_ops); 3463 if (rc) 3464 goto out_pernet_subsys; 3465 3466 rc = smc_ism_init(); 3467 if (rc) 3468 goto out_pernet_subsys_stat; 3469 smc_clc_init(); 3470 3471 rc = smc_nl_init(); 3472 if (rc) 3473 goto out_ism; 3474 3475 rc = smc_pnet_init(); 3476 if (rc) 3477 goto out_nl; 3478 3479 rc = -ENOMEM; 3480 3481 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3482 if (!smc_tcp_ls_wq) 3483 goto out_pnet; 3484 3485 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3486 if (!smc_hs_wq) 3487 goto out_alloc_tcp_ls_wq; 3488 3489 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3490 if (!smc_close_wq) 3491 goto out_alloc_hs_wq; 3492 3493 rc = smc_core_init(); 3494 if (rc) { 3495 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3496 goto out_alloc_wqs; 3497 } 3498 3499 rc = smc_llc_init(); 3500 if (rc) { 3501 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3502 goto out_core; 3503 } 3504 3505 rc = smc_cdc_init(); 3506 if (rc) { 3507 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3508 goto out_core; 3509 } 3510 3511 rc = proto_register(&smc_proto, 1); 3512 if (rc) { 3513 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3514 goto out_core; 3515 } 3516 3517 rc = proto_register(&smc_proto6, 1); 3518 if (rc) { 3519 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3520 goto out_proto; 3521 } 3522 3523 rc = sock_register(&smc_sock_family_ops); 3524 if (rc) { 3525 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3526 goto out_proto6; 3527 } 3528 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3529 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3530 3531 rc = smc_ib_register_client(); 3532 if (rc) { 3533 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3534 goto out_sock; 3535 } 3536 3537 rc = tcp_register_ulp(&smc_ulp_ops); 3538 if (rc) { 3539 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3540 goto out_ib; 3541 } 3542 3543 static_branch_enable(&tcp_have_smc); 3544 return 0; 3545 3546 out_ib: 3547 smc_ib_unregister_client(); 3548 out_sock: 3549 sock_unregister(PF_SMC); 3550 out_proto6: 3551 proto_unregister(&smc_proto6); 3552 out_proto: 3553 proto_unregister(&smc_proto); 3554 out_core: 3555 smc_core_exit(); 3556 out_alloc_wqs: 3557 destroy_workqueue(smc_close_wq); 3558 out_alloc_hs_wq: 3559 destroy_workqueue(smc_hs_wq); 3560 out_alloc_tcp_ls_wq: 3561 destroy_workqueue(smc_tcp_ls_wq); 3562 out_pnet: 3563 smc_pnet_exit(); 3564 out_nl: 3565 smc_nl_exit(); 3566 out_ism: 3567 smc_clc_exit(); 3568 smc_ism_exit(); 3569 out_pernet_subsys_stat: 3570 unregister_pernet_subsys(&smc_net_stat_ops); 3571 out_pernet_subsys: 3572 unregister_pernet_subsys(&smc_net_ops); 3573 3574 return rc; 3575 } 3576 3577 static void __exit smc_exit(void) 3578 { 3579 static_branch_disable(&tcp_have_smc); 3580 tcp_unregister_ulp(&smc_ulp_ops); 3581 sock_unregister(PF_SMC); 3582 smc_core_exit(); 3583 smc_ib_unregister_client(); 3584 smc_ism_exit(); 3585 destroy_workqueue(smc_close_wq); 3586 destroy_workqueue(smc_tcp_ls_wq); 3587 destroy_workqueue(smc_hs_wq); 3588 proto_unregister(&smc_proto6); 3589 proto_unregister(&smc_proto); 3590 smc_pnet_exit(); 3591 smc_nl_exit(); 3592 smc_clc_exit(); 3593 unregister_pernet_subsys(&smc_net_stat_ops); 3594 unregister_pernet_subsys(&smc_net_ops); 3595 rcu_barrier(); 3596 } 3597 3598 module_init(smc_init); 3599 module_exit(smc_exit); 3600 3601 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3602 MODULE_DESCRIPTION("smc socket address family"); 3603 MODULE_LICENSE("GPL"); 3604 MODULE_ALIAS_NETPROTO(PF_SMC); 3605 MODULE_ALIAS_TCP_ULP("smc"); 3606 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3607