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 #include "smc_loopback.h" 57 #include "smc_inet.h" 58 59 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 60 * creation on server 61 */ 62 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 63 * creation on client 64 */ 65 66 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 67 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 68 struct workqueue_struct *smc_close_wq; /* wq for close work */ 69 70 static void smc_tcp_listen_work(struct work_struct *); 71 static void smc_connect_work(struct work_struct *); 72 73 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 74 { 75 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 76 void *hdr; 77 78 if (cb_ctx->pos[0]) 79 goto out; 80 81 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 82 &smc_gen_nl_family, NLM_F_MULTI, 83 SMC_NETLINK_DUMP_HS_LIMITATION); 84 if (!hdr) 85 return -ENOMEM; 86 87 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 88 sock_net(skb->sk)->smc.limit_smc_hs)) 89 goto err; 90 91 genlmsg_end(skb, hdr); 92 cb_ctx->pos[0] = 1; 93 out: 94 return skb->len; 95 err: 96 genlmsg_cancel(skb, hdr); 97 return -EMSGSIZE; 98 } 99 100 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 101 { 102 sock_net(skb->sk)->smc.limit_smc_hs = true; 103 return 0; 104 } 105 106 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 107 { 108 sock_net(skb->sk)->smc.limit_smc_hs = false; 109 return 0; 110 } 111 112 static void smc_set_keepalive(struct sock *sk, int val) 113 { 114 struct smc_sock *smc = smc_sk(sk); 115 116 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 117 } 118 119 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 120 struct sk_buff *skb, 121 struct request_sock *req, 122 struct dst_entry *dst, 123 struct request_sock *req_unhash, 124 bool *own_req) 125 { 126 struct smc_sock *smc; 127 struct sock *child; 128 129 smc = smc_clcsock_user_data(sk); 130 131 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 132 sk->sk_max_ack_backlog) 133 goto drop; 134 135 if (sk_acceptq_is_full(&smc->sk)) { 136 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 137 goto drop; 138 } 139 140 /* passthrough to original syn recv sock fct */ 141 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 142 own_req); 143 /* child must not inherit smc or its ops */ 144 if (child) { 145 rcu_assign_sk_user_data(child, NULL); 146 147 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 148 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 149 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 150 } 151 return child; 152 153 drop: 154 dst_release(dst); 155 tcp_listendrop(sk); 156 return NULL; 157 } 158 159 static bool smc_hs_congested(const struct sock *sk) 160 { 161 const struct smc_sock *smc; 162 163 smc = smc_clcsock_user_data(sk); 164 165 if (!smc) 166 return true; 167 168 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 169 return true; 170 171 return false; 172 } 173 174 struct smc_hashinfo smc_v4_hashinfo = { 175 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 176 }; 177 178 struct smc_hashinfo smc_v6_hashinfo = { 179 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 180 }; 181 182 int smc_hash_sk(struct sock *sk) 183 { 184 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 185 struct hlist_head *head; 186 187 head = &h->ht; 188 189 write_lock_bh(&h->lock); 190 sk_add_node(sk, head); 191 write_unlock_bh(&h->lock); 192 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 193 194 return 0; 195 } 196 197 void smc_unhash_sk(struct sock *sk) 198 { 199 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 200 201 write_lock_bh(&h->lock); 202 if (sk_del_node_init(sk)) 203 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 204 write_unlock_bh(&h->lock); 205 } 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 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 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 void smc_sk_init(struct net *net, struct sock *sk, int protocol) 366 { 367 struct smc_sock *smc = smc_sk(sk); 368 369 sk->sk_state = SMC_INIT; 370 sk->sk_destruct = smc_destruct; 371 sk->sk_protocol = protocol; 372 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 373 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 374 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 375 INIT_WORK(&smc->connect_work, smc_connect_work); 376 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 377 INIT_LIST_HEAD(&smc->accept_q); 378 spin_lock_init(&smc->accept_q_lock); 379 spin_lock_init(&smc->conn.send_lock); 380 sk->sk_prot->hash(sk); 381 mutex_init(&smc->clcsock_release_lock); 382 smc_init_saved_callbacks(smc); 383 smc->limit_smc_hs = net->smc.limit_smc_hs; 384 smc->use_fallback = false; /* assume rdma capability first */ 385 smc->fallback_rsn = 0; 386 smc_close_init(smc); 387 } 388 389 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 390 int protocol) 391 { 392 struct proto *prot; 393 struct sock *sk; 394 395 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 396 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 397 if (!sk) 398 return NULL; 399 400 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 401 smc_sk_init(net, sk, protocol); 402 403 return sk; 404 } 405 406 int smc_bind(struct socket *sock, struct sockaddr *uaddr, 407 int addr_len) 408 { 409 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 410 struct sock *sk = sock->sk; 411 struct smc_sock *smc; 412 int rc; 413 414 smc = smc_sk(sk); 415 416 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 417 rc = -EINVAL; 418 if (addr_len < sizeof(struct sockaddr_in)) 419 goto out; 420 421 rc = -EAFNOSUPPORT; 422 if (addr->sin_family != AF_INET && 423 addr->sin_family != AF_INET6 && 424 addr->sin_family != AF_UNSPEC) 425 goto out; 426 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 427 if (addr->sin_family == AF_UNSPEC && 428 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 429 goto out; 430 431 lock_sock(sk); 432 433 /* Check if socket is already active */ 434 rc = -EINVAL; 435 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 436 goto out_rel; 437 438 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 439 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 440 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 441 442 out_rel: 443 release_sock(sk); 444 out: 445 return rc; 446 } 447 448 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 449 * clc socket (since smc is not called for these options from net/core) 450 */ 451 452 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 453 (1UL << SOCK_KEEPOPEN) | \ 454 (1UL << SOCK_LINGER) | \ 455 (1UL << SOCK_BROADCAST) | \ 456 (1UL << SOCK_TIMESTAMP) | \ 457 (1UL << SOCK_DBG) | \ 458 (1UL << SOCK_RCVTSTAMP) | \ 459 (1UL << SOCK_RCVTSTAMPNS) | \ 460 (1UL << SOCK_LOCALROUTE) | \ 461 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 462 (1UL << SOCK_RXQ_OVFL) | \ 463 (1UL << SOCK_WIFI_STATUS) | \ 464 (1UL << SOCK_NOFCS) | \ 465 (1UL << SOCK_FILTER_LOCKED) | \ 466 (1UL << SOCK_TSTAMP_NEW)) 467 468 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 469 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 470 unsigned long mask) 471 { 472 nsk->sk_userlocks = osk->sk_userlocks; 473 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) 474 nsk->sk_sndbuf = osk->sk_sndbuf; 475 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) 476 nsk->sk_rcvbuf = osk->sk_rcvbuf; 477 } 478 479 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 480 unsigned long mask) 481 { 482 /* options we don't get control via setsockopt for */ 483 nsk->sk_type = osk->sk_type; 484 nsk->sk_sndtimeo = osk->sk_sndtimeo; 485 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 486 nsk->sk_mark = READ_ONCE(osk->sk_mark); 487 nsk->sk_priority = READ_ONCE(osk->sk_priority); 488 nsk->sk_rcvlowat = osk->sk_rcvlowat; 489 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 490 nsk->sk_err = osk->sk_err; 491 492 nsk->sk_flags &= ~mask; 493 nsk->sk_flags |= osk->sk_flags & mask; 494 495 smc_adjust_sock_bufsizes(nsk, osk, mask); 496 } 497 498 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 499 { 500 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 501 } 502 503 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 504 (1UL << SOCK_KEEPOPEN) | \ 505 (1UL << SOCK_LINGER) | \ 506 (1UL << SOCK_DBG)) 507 /* copy only settings and flags relevant for smc from clc to smc socket */ 508 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 509 { 510 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 511 } 512 513 /* register the new vzalloced sndbuf on all links */ 514 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 515 struct smc_buf_desc *snd_desc) 516 { 517 struct smc_link_group *lgr = link->lgr; 518 int i, rc = 0; 519 520 if (!snd_desc->is_vm) 521 return -EINVAL; 522 523 /* protect against parallel smcr_link_reg_buf() */ 524 down_write(&lgr->llc_conf_mutex); 525 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 526 if (!smc_link_active(&lgr->lnk[i])) 527 continue; 528 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 529 if (rc) 530 break; 531 } 532 up_write(&lgr->llc_conf_mutex); 533 return rc; 534 } 535 536 /* register the new rmb on all links */ 537 static int smcr_lgr_reg_rmbs(struct smc_link *link, 538 struct smc_buf_desc *rmb_desc) 539 { 540 struct smc_link_group *lgr = link->lgr; 541 bool do_slow = false; 542 int i, rc = 0; 543 544 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 545 if (rc) 546 return rc; 547 548 down_read(&lgr->llc_conf_mutex); 549 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 550 if (!smc_link_active(&lgr->lnk[i])) 551 continue; 552 if (!rmb_desc->is_reg_mr[link->link_idx]) { 553 up_read(&lgr->llc_conf_mutex); 554 goto slow_path; 555 } 556 } 557 /* mr register already */ 558 goto fast_path; 559 slow_path: 560 do_slow = true; 561 /* protect against parallel smc_llc_cli_rkey_exchange() and 562 * parallel smcr_link_reg_buf() 563 */ 564 down_write(&lgr->llc_conf_mutex); 565 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 566 if (!smc_link_active(&lgr->lnk[i])) 567 continue; 568 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 569 if (rc) 570 goto out; 571 } 572 fast_path: 573 /* exchange confirm_rkey msg with peer */ 574 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 575 if (rc) { 576 rc = -EFAULT; 577 goto out; 578 } 579 rmb_desc->is_conf_rkey = true; 580 out: 581 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 582 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 583 return rc; 584 } 585 586 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 587 { 588 struct smc_link *link = smc->conn.lnk; 589 struct smc_llc_qentry *qentry; 590 int rc; 591 592 /* Receive CONFIRM LINK request from server over RoCE fabric. 593 * Increasing the client's timeout by twice as much as the server's 594 * timeout by default can temporarily avoid decline messages of 595 * both sides crossing or colliding 596 */ 597 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 598 SMC_LLC_CONFIRM_LINK); 599 if (!qentry) { 600 struct smc_clc_msg_decline dclc; 601 602 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 603 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 604 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 605 } 606 smc_llc_save_peer_uid(qentry); 607 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 608 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 609 if (rc) 610 return SMC_CLC_DECL_RMBE_EC; 611 612 rc = smc_ib_modify_qp_rts(link); 613 if (rc) 614 return SMC_CLC_DECL_ERR_RDYLNK; 615 616 smc_wr_remember_qp_attr(link); 617 618 /* reg the sndbuf if it was vzalloced */ 619 if (smc->conn.sndbuf_desc->is_vm) { 620 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 621 return SMC_CLC_DECL_ERR_REGBUF; 622 } 623 624 /* reg the rmb */ 625 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 626 return SMC_CLC_DECL_ERR_REGBUF; 627 628 /* confirm_rkey is implicit on 1st contact */ 629 smc->conn.rmb_desc->is_conf_rkey = true; 630 631 /* send CONFIRM LINK response over RoCE fabric */ 632 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 633 if (rc < 0) 634 return SMC_CLC_DECL_TIMEOUT_CL; 635 636 smc_llc_link_active(link); 637 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 638 639 if (link->lgr->max_links > 1) { 640 /* optional 2nd link, receive ADD LINK request from server */ 641 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 642 SMC_LLC_ADD_LINK); 643 if (!qentry) { 644 struct smc_clc_msg_decline dclc; 645 646 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 647 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 648 if (rc == -EAGAIN) 649 rc = 0; /* no DECLINE received, go with one link */ 650 return rc; 651 } 652 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 653 smc_llc_cli_add_link(link, qentry); 654 } 655 return 0; 656 } 657 658 static bool smc_isascii(char *hostname) 659 { 660 int i; 661 662 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 663 if (!isascii(hostname[i])) 664 return false; 665 return true; 666 } 667 668 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 669 struct smc_clc_msg_accept_confirm *clc) 670 { 671 struct smc_clc_first_contact_ext *fce; 672 int clc_v2_len; 673 674 if (clc->hdr.version == SMC_V1 || 675 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 676 return; 677 678 if (smc->conn.lgr->is_smcd) { 679 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid, 680 SMC_MAX_EID_LEN); 681 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1); 682 } else { 683 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid, 684 SMC_MAX_EID_LEN); 685 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1); 686 } 687 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len); 688 smc->conn.lgr->peer_os = fce->os_type; 689 smc->conn.lgr->peer_smc_release = fce->release; 690 if (smc_isascii(fce->hostname)) 691 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 692 SMC_MAX_HOSTNAME_LEN); 693 } 694 695 static void smcr_conn_save_peer_info(struct smc_sock *smc, 696 struct smc_clc_msg_accept_confirm *clc) 697 { 698 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 699 700 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 701 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 702 smc->conn.peer_rmbe_size = bufsize; 703 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 704 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 705 } 706 707 static void smcd_conn_save_peer_info(struct smc_sock *smc, 708 struct smc_clc_msg_accept_confirm *clc) 709 { 710 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 711 712 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 713 smc->conn.peer_token = ntohll(clc->d0.token); 714 /* msg header takes up space in the buffer */ 715 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 716 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 717 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 718 } 719 720 static void smc_conn_save_peer_info(struct smc_sock *smc, 721 struct smc_clc_msg_accept_confirm *clc) 722 { 723 if (smc->conn.lgr->is_smcd) 724 smcd_conn_save_peer_info(smc, clc); 725 else 726 smcr_conn_save_peer_info(smc, clc); 727 smc_conn_save_peer_info_fce(smc, clc); 728 } 729 730 static void smc_link_save_peer_info(struct smc_link *link, 731 struct smc_clc_msg_accept_confirm *clc, 732 struct smc_init_info *ini) 733 { 734 link->peer_qpn = ntoh24(clc->r0.qpn); 735 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 736 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 737 link->peer_psn = ntoh24(clc->r0.psn); 738 link->peer_mtu = clc->r0.qp_mtu; 739 } 740 741 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 742 struct smc_stats_fback *fback_arr) 743 { 744 int cnt; 745 746 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 747 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 748 fback_arr[cnt].count++; 749 break; 750 } 751 if (!fback_arr[cnt].fback_code) { 752 fback_arr[cnt].fback_code = smc->fallback_rsn; 753 fback_arr[cnt].count++; 754 break; 755 } 756 } 757 } 758 759 static void smc_stat_fallback(struct smc_sock *smc) 760 { 761 struct net *net = sock_net(&smc->sk); 762 763 mutex_lock(&net->smc.mutex_fback_rsn); 764 if (smc->listen_smc) { 765 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 766 net->smc.fback_rsn->srv_fback_cnt++; 767 } else { 768 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 769 net->smc.fback_rsn->clnt_fback_cnt++; 770 } 771 mutex_unlock(&net->smc.mutex_fback_rsn); 772 } 773 774 /* must be called under rcu read lock */ 775 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 776 { 777 struct socket_wq *wq; 778 __poll_t flags; 779 780 wq = rcu_dereference(smc->sk.sk_wq); 781 if (!skwq_has_sleeper(wq)) 782 return; 783 784 /* wake up smc sk->sk_wq */ 785 if (!key) { 786 /* sk_state_change */ 787 wake_up_interruptible_all(&wq->wait); 788 } else { 789 flags = key_to_poll(key); 790 if (flags & (EPOLLIN | EPOLLOUT)) 791 /* sk_data_ready or sk_write_space */ 792 wake_up_interruptible_sync_poll(&wq->wait, flags); 793 else if (flags & EPOLLERR) 794 /* sk_error_report */ 795 wake_up_interruptible_poll(&wq->wait, flags); 796 } 797 } 798 799 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 800 unsigned int mode, int sync, void *key) 801 { 802 struct smc_mark_woken *mark = 803 container_of(wait, struct smc_mark_woken, wait_entry); 804 805 mark->woken = true; 806 mark->key = key; 807 return 0; 808 } 809 810 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 811 void (*clcsock_callback)(struct sock *sk)) 812 { 813 struct smc_mark_woken mark = { .woken = false }; 814 struct socket_wq *wq; 815 816 init_waitqueue_func_entry(&mark.wait_entry, 817 smc_fback_mark_woken); 818 rcu_read_lock(); 819 wq = rcu_dereference(clcsk->sk_wq); 820 if (!wq) 821 goto out; 822 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 823 clcsock_callback(clcsk); 824 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 825 826 if (mark.woken) 827 smc_fback_wakeup_waitqueue(smc, mark.key); 828 out: 829 rcu_read_unlock(); 830 } 831 832 static void smc_fback_state_change(struct sock *clcsk) 833 { 834 struct smc_sock *smc; 835 836 read_lock_bh(&clcsk->sk_callback_lock); 837 smc = smc_clcsock_user_data(clcsk); 838 if (smc) 839 smc_fback_forward_wakeup(smc, clcsk, 840 smc->clcsk_state_change); 841 read_unlock_bh(&clcsk->sk_callback_lock); 842 } 843 844 static void smc_fback_data_ready(struct sock *clcsk) 845 { 846 struct smc_sock *smc; 847 848 read_lock_bh(&clcsk->sk_callback_lock); 849 smc = smc_clcsock_user_data(clcsk); 850 if (smc) 851 smc_fback_forward_wakeup(smc, clcsk, 852 smc->clcsk_data_ready); 853 read_unlock_bh(&clcsk->sk_callback_lock); 854 } 855 856 static void smc_fback_write_space(struct sock *clcsk) 857 { 858 struct smc_sock *smc; 859 860 read_lock_bh(&clcsk->sk_callback_lock); 861 smc = smc_clcsock_user_data(clcsk); 862 if (smc) 863 smc_fback_forward_wakeup(smc, clcsk, 864 smc->clcsk_write_space); 865 read_unlock_bh(&clcsk->sk_callback_lock); 866 } 867 868 static void smc_fback_error_report(struct sock *clcsk) 869 { 870 struct smc_sock *smc; 871 872 read_lock_bh(&clcsk->sk_callback_lock); 873 smc = smc_clcsock_user_data(clcsk); 874 if (smc) 875 smc_fback_forward_wakeup(smc, clcsk, 876 smc->clcsk_error_report); 877 read_unlock_bh(&clcsk->sk_callback_lock); 878 } 879 880 static void smc_fback_replace_callbacks(struct smc_sock *smc) 881 { 882 struct sock *clcsk = smc->clcsock->sk; 883 884 write_lock_bh(&clcsk->sk_callback_lock); 885 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 886 887 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 888 &smc->clcsk_state_change); 889 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 890 &smc->clcsk_data_ready); 891 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 892 &smc->clcsk_write_space); 893 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 894 &smc->clcsk_error_report); 895 896 write_unlock_bh(&clcsk->sk_callback_lock); 897 } 898 899 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 900 { 901 int rc = 0; 902 903 mutex_lock(&smc->clcsock_release_lock); 904 if (!smc->clcsock) { 905 rc = -EBADF; 906 goto out; 907 } 908 909 smc->use_fallback = true; 910 smc->fallback_rsn = reason_code; 911 smc_stat_fallback(smc); 912 trace_smc_switch_to_fallback(smc, reason_code); 913 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 914 smc->clcsock->file = smc->sk.sk_socket->file; 915 smc->clcsock->file->private_data = smc->clcsock; 916 smc->clcsock->wq.fasync_list = 917 smc->sk.sk_socket->wq.fasync_list; 918 smc->sk.sk_socket->wq.fasync_list = NULL; 919 920 /* There might be some wait entries remaining 921 * in smc sk->sk_wq and they should be woken up 922 * as clcsock's wait queue is woken up. 923 */ 924 smc_fback_replace_callbacks(smc); 925 } 926 out: 927 mutex_unlock(&smc->clcsock_release_lock); 928 return rc; 929 } 930 931 /* fall back during connect */ 932 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 933 { 934 struct net *net = sock_net(&smc->sk); 935 int rc = 0; 936 937 rc = smc_switch_to_fallback(smc, reason_code); 938 if (rc) { /* fallback fails */ 939 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 940 if (smc->sk.sk_state == SMC_INIT) 941 sock_put(&smc->sk); /* passive closing */ 942 return rc; 943 } 944 smc_copy_sock_settings_to_clc(smc); 945 smc->connect_nonblock = 0; 946 if (smc->sk.sk_state == SMC_INIT) 947 smc->sk.sk_state = SMC_ACTIVE; 948 return 0; 949 } 950 951 /* decline and fall back during connect */ 952 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 953 u8 version) 954 { 955 struct net *net = sock_net(&smc->sk); 956 int rc; 957 958 if (reason_code < 0) { /* error, fallback is not possible */ 959 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 960 if (smc->sk.sk_state == SMC_INIT) 961 sock_put(&smc->sk); /* passive closing */ 962 return reason_code; 963 } 964 if (reason_code != SMC_CLC_DECL_PEERDECL) { 965 rc = smc_clc_send_decline(smc, reason_code, version); 966 if (rc < 0) { 967 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 968 if (smc->sk.sk_state == SMC_INIT) 969 sock_put(&smc->sk); /* passive closing */ 970 return rc; 971 } 972 } 973 return smc_connect_fallback(smc, reason_code); 974 } 975 976 static void smc_conn_abort(struct smc_sock *smc, int local_first) 977 { 978 struct smc_connection *conn = &smc->conn; 979 struct smc_link_group *lgr = conn->lgr; 980 bool lgr_valid = false; 981 982 if (smc_conn_lgr_valid(conn)) 983 lgr_valid = true; 984 985 smc_conn_free(conn); 986 if (local_first && lgr_valid) 987 smc_lgr_cleanup_early(lgr); 988 } 989 990 /* check if there is a rdma device available for this connection. */ 991 /* called for connect and listen */ 992 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 993 { 994 /* PNET table look up: search active ib_device and port 995 * within same PNETID that also contains the ethernet device 996 * used for the internal TCP socket 997 */ 998 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 999 if (!ini->check_smcrv2 && !ini->ib_dev) 1000 return SMC_CLC_DECL_NOSMCRDEV; 1001 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 1002 return SMC_CLC_DECL_NOSMCRDEV; 1003 return 0; 1004 } 1005 1006 /* check if there is an ISM device available for this connection. */ 1007 /* called for connect and listen */ 1008 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1009 { 1010 /* Find ISM device with same PNETID as connecting interface */ 1011 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1012 if (!ini->ism_dev[0]) 1013 return SMC_CLC_DECL_NOSMCDDEV; 1014 else 1015 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1016 return 0; 1017 } 1018 1019 /* is chid unique for the ism devices that are already determined? */ 1020 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1021 int cnt) 1022 { 1023 int i = (!ini->ism_dev[0]) ? 1 : 0; 1024 1025 for (; i < cnt; i++) 1026 if (ini->ism_chid[i] == chid) 1027 return false; 1028 return true; 1029 } 1030 1031 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1032 * PNETID matching net_device) 1033 */ 1034 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1035 struct smc_init_info *ini) 1036 { 1037 int rc = SMC_CLC_DECL_NOSMCDDEV; 1038 struct smcd_dev *smcd; 1039 int i = 1, entry = 1; 1040 bool is_emulated; 1041 u16 chid; 1042 1043 if (smcd_indicated(ini->smc_type_v1)) 1044 rc = 0; /* already initialized for V1 */ 1045 mutex_lock(&smcd_dev_list.mutex); 1046 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1047 if (smcd->going_away || smcd == ini->ism_dev[0]) 1048 continue; 1049 chid = smc_ism_get_chid(smcd); 1050 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1051 continue; 1052 is_emulated = __smc_ism_is_emulated(chid); 1053 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1054 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1055 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES) 1056 /* It's the last GID-CHID entry left in CLC 1057 * Proposal SMC-Dv2 extension, but an Emulated- 1058 * ISM device will take two entries. So give 1059 * up it and try the next potential ISM device. 1060 */ 1061 continue; 1062 ini->ism_dev[i] = smcd; 1063 ini->ism_chid[i] = chid; 1064 ini->is_smcd = true; 1065 rc = 0; 1066 i++; 1067 entry = is_emulated ? entry + 2 : entry + 1; 1068 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES) 1069 break; 1070 } 1071 } 1072 mutex_unlock(&smcd_dev_list.mutex); 1073 ini->ism_offered_cnt = i - 1; 1074 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1075 ini->smcd_version = 0; 1076 1077 return rc; 1078 } 1079 1080 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1081 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1082 struct smc_init_info *ini) 1083 { 1084 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1085 return SMC_CLC_DECL_ISMVLANERR; 1086 return 0; 1087 } 1088 1089 static int smc_find_proposal_devices(struct smc_sock *smc, 1090 struct smc_init_info *ini) 1091 { 1092 int rc = 0; 1093 1094 /* check if there is an ism device available */ 1095 if (!(ini->smcd_version & SMC_V1) || 1096 smc_find_ism_device(smc, ini) || 1097 smc_connect_ism_vlan_setup(smc, ini)) 1098 ini->smcd_version &= ~SMC_V1; 1099 /* else ISM V1 is supported for this connection */ 1100 1101 /* check if there is an rdma device available */ 1102 if (!(ini->smcr_version & SMC_V1) || 1103 smc_find_rdma_device(smc, ini)) 1104 ini->smcr_version &= ~SMC_V1; 1105 /* else RDMA is supported for this connection */ 1106 1107 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1108 ini->smcr_version & SMC_V1); 1109 1110 /* check if there is an ism v2 device available */ 1111 if (!(ini->smcd_version & SMC_V2) || 1112 !smc_ism_is_v2_capable() || 1113 smc_find_ism_v2_device_clnt(smc, ini)) 1114 ini->smcd_version &= ~SMC_V2; 1115 1116 /* check if there is an rdma v2 device available */ 1117 ini->check_smcrv2 = true; 1118 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1119 if (!(ini->smcr_version & SMC_V2) || 1120 smc->clcsock->sk->sk_family != AF_INET || 1121 !smc_clc_ueid_count() || 1122 smc_find_rdma_device(smc, ini)) 1123 ini->smcr_version &= ~SMC_V2; 1124 ini->check_smcrv2 = false; 1125 1126 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1127 ini->smcr_version & SMC_V2); 1128 1129 /* if neither ISM nor RDMA are supported, fallback */ 1130 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1131 rc = SMC_CLC_DECL_NOSMCDEV; 1132 1133 return rc; 1134 } 1135 1136 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1137 * used, the VLAN ID will be registered again during the connection setup. 1138 */ 1139 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1140 struct smc_init_info *ini) 1141 { 1142 if (!smcd_indicated(ini->smc_type_v1)) 1143 return 0; 1144 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1145 return SMC_CLC_DECL_CNFERR; 1146 return 0; 1147 } 1148 1149 #define SMC_CLC_MAX_ACCEPT_LEN \ 1150 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1151 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1152 sizeof(struct smc_clc_msg_trail)) 1153 1154 /* CLC handshake during connect */ 1155 static int smc_connect_clc(struct smc_sock *smc, 1156 struct smc_clc_msg_accept_confirm *aclc, 1157 struct smc_init_info *ini) 1158 { 1159 int rc = 0; 1160 1161 /* do inband token exchange */ 1162 rc = smc_clc_send_proposal(smc, ini); 1163 if (rc) 1164 return rc; 1165 /* receive SMC Accept CLC message */ 1166 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1167 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1168 } 1169 1170 void smc_fill_gid_list(struct smc_link_group *lgr, 1171 struct smc_gidlist *gidlist, 1172 struct smc_ib_device *known_dev, u8 *known_gid) 1173 { 1174 struct smc_init_info *alt_ini = NULL; 1175 1176 memset(gidlist, 0, sizeof(*gidlist)); 1177 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1178 1179 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1180 if (!alt_ini) 1181 goto out; 1182 1183 alt_ini->vlan_id = lgr->vlan_id; 1184 alt_ini->check_smcrv2 = true; 1185 alt_ini->smcrv2.saddr = lgr->saddr; 1186 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1187 1188 if (!alt_ini->smcrv2.ib_dev_v2) 1189 goto out; 1190 1191 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1192 SMC_GID_SIZE); 1193 1194 out: 1195 kfree(alt_ini); 1196 } 1197 1198 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1199 struct smc_clc_msg_accept_confirm *aclc, 1200 struct smc_init_info *ini) 1201 { 1202 struct smc_clc_first_contact_ext *fce = 1203 smc_get_clc_first_contact_ext(aclc, false); 1204 struct net *net = sock_net(&smc->sk); 1205 int rc; 1206 1207 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1208 return 0; 1209 1210 if (fce->v2_direct) { 1211 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1212 ini->smcrv2.uses_gateway = false; 1213 } else { 1214 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1215 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1216 ini->smcrv2.nexthop_mac, 1217 &ini->smcrv2.uses_gateway)) 1218 return SMC_CLC_DECL_NOROUTE; 1219 if (!ini->smcrv2.uses_gateway) { 1220 /* mismatch: peer claims indirect, but its direct */ 1221 return SMC_CLC_DECL_NOINDIRECT; 1222 } 1223 } 1224 1225 ini->release_nr = fce->release; 1226 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1227 if (rc) 1228 return rc; 1229 1230 return 0; 1231 } 1232 1233 /* setup for RDMA connection of client */ 1234 static int smc_connect_rdma(struct smc_sock *smc, 1235 struct smc_clc_msg_accept_confirm *aclc, 1236 struct smc_init_info *ini) 1237 { 1238 int i, reason_code = 0; 1239 struct smc_link *link; 1240 u8 *eid = NULL; 1241 1242 ini->is_smcd = false; 1243 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1244 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1245 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1246 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1247 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1248 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1249 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1250 1251 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1252 if (reason_code) 1253 return reason_code; 1254 1255 mutex_lock(&smc_client_lgr_pending); 1256 reason_code = smc_conn_create(smc, ini); 1257 if (reason_code) { 1258 mutex_unlock(&smc_client_lgr_pending); 1259 return reason_code; 1260 } 1261 1262 smc_conn_save_peer_info(smc, aclc); 1263 1264 if (ini->first_contact_local) { 1265 link = smc->conn.lnk; 1266 } else { 1267 /* set link that was assigned by server */ 1268 link = NULL; 1269 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1270 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1271 1272 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1273 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1274 SMC_GID_SIZE) && 1275 (aclc->hdr.version > SMC_V1 || 1276 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1277 sizeof(l->peer_mac)))) { 1278 link = l; 1279 break; 1280 } 1281 } 1282 if (!link) { 1283 reason_code = SMC_CLC_DECL_NOSRVLINK; 1284 goto connect_abort; 1285 } 1286 smc_switch_link_and_count(&smc->conn, link); 1287 } 1288 1289 /* create send buffer and rmb */ 1290 if (smc_buf_create(smc, false)) { 1291 reason_code = SMC_CLC_DECL_MEM; 1292 goto connect_abort; 1293 } 1294 1295 if (ini->first_contact_local) 1296 smc_link_save_peer_info(link, aclc, ini); 1297 1298 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1299 reason_code = SMC_CLC_DECL_ERR_RTOK; 1300 goto connect_abort; 1301 } 1302 1303 smc_rx_init(smc); 1304 1305 if (ini->first_contact_local) { 1306 if (smc_ib_ready_link(link)) { 1307 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1308 goto connect_abort; 1309 } 1310 } else { 1311 /* reg sendbufs if they were vzalloced */ 1312 if (smc->conn.sndbuf_desc->is_vm) { 1313 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1314 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1315 goto connect_abort; 1316 } 1317 } 1318 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1319 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1320 goto connect_abort; 1321 } 1322 } 1323 1324 if (aclc->hdr.version > SMC_V1) { 1325 eid = aclc->r1.eid; 1326 if (ini->first_contact_local) 1327 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1328 link->smcibdev, link->gid); 1329 } 1330 1331 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1332 aclc->hdr.version, eid, ini); 1333 if (reason_code) 1334 goto connect_abort; 1335 1336 smc_tx_init(smc); 1337 1338 if (ini->first_contact_local) { 1339 /* QP confirmation over RoCE fabric */ 1340 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1341 reason_code = smcr_clnt_conf_first_link(smc); 1342 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1343 if (reason_code) 1344 goto connect_abort; 1345 } 1346 mutex_unlock(&smc_client_lgr_pending); 1347 1348 smc_copy_sock_settings_to_clc(smc); 1349 smc->connect_nonblock = 0; 1350 if (smc->sk.sk_state == SMC_INIT) 1351 smc->sk.sk_state = SMC_ACTIVE; 1352 1353 return 0; 1354 connect_abort: 1355 smc_conn_abort(smc, ini->first_contact_local); 1356 mutex_unlock(&smc_client_lgr_pending); 1357 smc->connect_nonblock = 0; 1358 1359 return reason_code; 1360 } 1361 1362 /* The server has chosen one of the proposed ISM devices for the communication. 1363 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1364 */ 1365 static int 1366 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1367 struct smc_init_info *ini) 1368 { 1369 int i; 1370 1371 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1372 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1373 ini->ism_selected = i; 1374 return 0; 1375 } 1376 } 1377 1378 return -EPROTO; 1379 } 1380 1381 /* setup for ISM connection of client */ 1382 static int smc_connect_ism(struct smc_sock *smc, 1383 struct smc_clc_msg_accept_confirm *aclc, 1384 struct smc_init_info *ini) 1385 { 1386 u8 *eid = NULL; 1387 int rc = 0; 1388 1389 ini->is_smcd = true; 1390 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1391 1392 if (aclc->hdr.version == SMC_V2) { 1393 if (ini->first_contact_peer) { 1394 struct smc_clc_first_contact_ext *fce = 1395 smc_get_clc_first_contact_ext(aclc, true); 1396 1397 ini->release_nr = fce->release; 1398 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1399 if (rc) 1400 return rc; 1401 } 1402 1403 rc = smc_v2_determine_accepted_chid(aclc, ini); 1404 if (rc) 1405 return rc; 1406 1407 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1408 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1409 ntohll(aclc->d1.gid_ext); 1410 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1411 } 1412 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1413 1414 /* there is only one lgr role for SMC-D; use server lock */ 1415 mutex_lock(&smc_server_lgr_pending); 1416 rc = smc_conn_create(smc, ini); 1417 if (rc) { 1418 mutex_unlock(&smc_server_lgr_pending); 1419 return rc; 1420 } 1421 1422 /* Create send and receive buffers */ 1423 rc = smc_buf_create(smc, true); 1424 if (rc) { 1425 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1426 goto connect_abort; 1427 } 1428 1429 smc_conn_save_peer_info(smc, aclc); 1430 1431 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) { 1432 rc = smcd_buf_attach(smc); 1433 if (rc) { 1434 rc = SMC_CLC_DECL_MEM; /* try to fallback */ 1435 goto connect_abort; 1436 } 1437 } 1438 smc_rx_init(smc); 1439 smc_tx_init(smc); 1440 1441 if (aclc->hdr.version > SMC_V1) 1442 eid = aclc->d1.eid; 1443 1444 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1445 aclc->hdr.version, eid, ini); 1446 if (rc) 1447 goto connect_abort; 1448 mutex_unlock(&smc_server_lgr_pending); 1449 1450 smc_copy_sock_settings_to_clc(smc); 1451 smc->connect_nonblock = 0; 1452 if (smc->sk.sk_state == SMC_INIT) 1453 smc->sk.sk_state = SMC_ACTIVE; 1454 1455 return 0; 1456 connect_abort: 1457 smc_conn_abort(smc, ini->first_contact_local); 1458 mutex_unlock(&smc_server_lgr_pending); 1459 smc->connect_nonblock = 0; 1460 1461 return rc; 1462 } 1463 1464 /* check if received accept type and version matches a proposed one */ 1465 static int smc_connect_check_aclc(struct smc_init_info *ini, 1466 struct smc_clc_msg_accept_confirm *aclc) 1467 { 1468 if (aclc->hdr.version >= SMC_V2) { 1469 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1470 !smcr_indicated(ini->smc_type_v2)) || 1471 (aclc->hdr.typev1 == SMC_TYPE_D && 1472 !smcd_indicated(ini->smc_type_v2))) 1473 return SMC_CLC_DECL_MODEUNSUPP; 1474 } else { 1475 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1476 !smcr_indicated(ini->smc_type_v1)) || 1477 (aclc->hdr.typev1 == SMC_TYPE_D && 1478 !smcd_indicated(ini->smc_type_v1))) 1479 return SMC_CLC_DECL_MODEUNSUPP; 1480 } 1481 1482 return 0; 1483 } 1484 1485 /* perform steps before actually connecting */ 1486 static int __smc_connect(struct smc_sock *smc) 1487 { 1488 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1489 struct smc_clc_msg_accept_confirm *aclc; 1490 struct smc_init_info *ini = NULL; 1491 u8 *buf = NULL; 1492 int rc = 0; 1493 1494 if (smc->use_fallback) 1495 return smc_connect_fallback(smc, smc->fallback_rsn); 1496 1497 /* if peer has not signalled SMC-capability, fall back */ 1498 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1499 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1500 1501 /* IPSec connections opt out of SMC optimizations */ 1502 if (using_ipsec(smc)) 1503 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1504 version); 1505 1506 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1507 if (!ini) 1508 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1509 version); 1510 1511 ini->smcd_version = SMC_V1 | SMC_V2; 1512 ini->smcr_version = SMC_V1 | SMC_V2; 1513 ini->smc_type_v1 = SMC_TYPE_B; 1514 ini->smc_type_v2 = SMC_TYPE_B; 1515 1516 /* get vlan id from IP device */ 1517 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1518 ini->smcd_version &= ~SMC_V1; 1519 ini->smcr_version = 0; 1520 ini->smc_type_v1 = SMC_TYPE_N; 1521 } 1522 1523 rc = smc_find_proposal_devices(smc, ini); 1524 if (rc) 1525 goto fallback; 1526 1527 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1528 if (!buf) { 1529 rc = SMC_CLC_DECL_MEM; 1530 goto fallback; 1531 } 1532 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1533 1534 /* perform CLC handshake */ 1535 rc = smc_connect_clc(smc, aclc, ini); 1536 if (rc) { 1537 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1538 if (rc == -EAGAIN) { 1539 rc = -ETIMEDOUT; 1540 smc->sk.sk_err = ETIMEDOUT; 1541 } 1542 goto vlan_cleanup; 1543 } 1544 1545 /* check if smc modes and versions of CLC proposal and accept match */ 1546 rc = smc_connect_check_aclc(ini, aclc); 1547 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1548 if (rc) 1549 goto vlan_cleanup; 1550 1551 /* depending on previous steps, connect using rdma or ism */ 1552 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1553 ini->smcr_version = version; 1554 rc = smc_connect_rdma(smc, aclc, ini); 1555 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1556 ini->smcd_version = version; 1557 rc = smc_connect_ism(smc, aclc, ini); 1558 } 1559 if (rc) 1560 goto vlan_cleanup; 1561 1562 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1563 smc_connect_ism_vlan_cleanup(smc, ini); 1564 kfree(buf); 1565 kfree(ini); 1566 return 0; 1567 1568 vlan_cleanup: 1569 smc_connect_ism_vlan_cleanup(smc, ini); 1570 kfree(buf); 1571 fallback: 1572 kfree(ini); 1573 return smc_connect_decline_fallback(smc, rc, version); 1574 } 1575 1576 static void smc_connect_work(struct work_struct *work) 1577 { 1578 struct smc_sock *smc = container_of(work, struct smc_sock, 1579 connect_work); 1580 long timeo = smc->sk.sk_sndtimeo; 1581 int rc = 0; 1582 1583 if (!timeo) 1584 timeo = MAX_SCHEDULE_TIMEOUT; 1585 lock_sock(smc->clcsock->sk); 1586 if (smc->clcsock->sk->sk_err) { 1587 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1588 } else if ((1 << smc->clcsock->sk->sk_state) & 1589 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1590 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1591 if ((rc == -EPIPE) && 1592 ((1 << smc->clcsock->sk->sk_state) & 1593 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1594 rc = 0; 1595 } 1596 release_sock(smc->clcsock->sk); 1597 lock_sock(&smc->sk); 1598 if (rc != 0 || smc->sk.sk_err) { 1599 smc->sk.sk_state = SMC_CLOSED; 1600 if (rc == -EPIPE || rc == -EAGAIN) 1601 smc->sk.sk_err = EPIPE; 1602 else if (rc == -ECONNREFUSED) 1603 smc->sk.sk_err = ECONNREFUSED; 1604 else if (signal_pending(current)) 1605 smc->sk.sk_err = -sock_intr_errno(timeo); 1606 sock_put(&smc->sk); /* passive closing */ 1607 goto out; 1608 } 1609 1610 rc = __smc_connect(smc); 1611 if (rc < 0) 1612 smc->sk.sk_err = -rc; 1613 1614 out: 1615 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1616 if (smc->sk.sk_err) { 1617 smc->sk.sk_state_change(&smc->sk); 1618 } else { /* allow polling before and after fallback decision */ 1619 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1620 smc->sk.sk_write_space(&smc->sk); 1621 } 1622 } 1623 release_sock(&smc->sk); 1624 } 1625 1626 int smc_connect(struct socket *sock, struct sockaddr *addr, 1627 int alen, int flags) 1628 { 1629 struct sock *sk = sock->sk; 1630 struct smc_sock *smc; 1631 int rc = -EINVAL; 1632 1633 smc = smc_sk(sk); 1634 1635 /* separate smc parameter checking to be safe */ 1636 if (alen < sizeof(addr->sa_family)) 1637 goto out_err; 1638 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1639 goto out_err; 1640 1641 lock_sock(sk); 1642 switch (sock->state) { 1643 default: 1644 rc = -EINVAL; 1645 goto out; 1646 case SS_CONNECTED: 1647 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1648 goto out; 1649 case SS_CONNECTING: 1650 if (sk->sk_state == SMC_ACTIVE) 1651 goto connected; 1652 break; 1653 case SS_UNCONNECTED: 1654 sock->state = SS_CONNECTING; 1655 break; 1656 } 1657 1658 switch (sk->sk_state) { 1659 default: 1660 goto out; 1661 case SMC_CLOSED: 1662 rc = sock_error(sk) ? : -ECONNABORTED; 1663 sock->state = SS_UNCONNECTED; 1664 goto out; 1665 case SMC_ACTIVE: 1666 rc = -EISCONN; 1667 goto out; 1668 case SMC_INIT: 1669 break; 1670 } 1671 1672 smc_copy_sock_settings_to_clc(smc); 1673 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1674 if (smc->connect_nonblock) { 1675 rc = -EALREADY; 1676 goto out; 1677 } 1678 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1679 if (rc && rc != -EINPROGRESS) 1680 goto out; 1681 1682 if (smc->use_fallback) { 1683 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1684 goto out; 1685 } 1686 sock_hold(&smc->sk); /* sock put in passive closing */ 1687 if (flags & O_NONBLOCK) { 1688 if (queue_work(smc_hs_wq, &smc->connect_work)) 1689 smc->connect_nonblock = 1; 1690 rc = -EINPROGRESS; 1691 goto out; 1692 } else { 1693 rc = __smc_connect(smc); 1694 if (rc < 0) 1695 goto out; 1696 } 1697 1698 connected: 1699 rc = 0; 1700 sock->state = SS_CONNECTED; 1701 out: 1702 release_sock(sk); 1703 out_err: 1704 return rc; 1705 } 1706 1707 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1708 { 1709 struct socket *new_clcsock = NULL; 1710 struct sock *lsk = &lsmc->sk; 1711 struct sock *new_sk; 1712 int rc = -EINVAL; 1713 1714 release_sock(lsk); 1715 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1716 if (!new_sk) { 1717 rc = -ENOMEM; 1718 lsk->sk_err = ENOMEM; 1719 *new_smc = NULL; 1720 lock_sock(lsk); 1721 goto out; 1722 } 1723 *new_smc = smc_sk(new_sk); 1724 1725 mutex_lock(&lsmc->clcsock_release_lock); 1726 if (lsmc->clcsock) 1727 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1728 mutex_unlock(&lsmc->clcsock_release_lock); 1729 lock_sock(lsk); 1730 if (rc < 0 && rc != -EAGAIN) 1731 lsk->sk_err = -rc; 1732 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1733 new_sk->sk_prot->unhash(new_sk); 1734 if (new_clcsock) 1735 sock_release(new_clcsock); 1736 new_sk->sk_state = SMC_CLOSED; 1737 smc_sock_set_flag(new_sk, SOCK_DEAD); 1738 sock_put(new_sk); /* final */ 1739 *new_smc = NULL; 1740 goto out; 1741 } 1742 1743 /* new clcsock has inherited the smc listen-specific sk_data_ready 1744 * function; switch it back to the original sk_data_ready function 1745 */ 1746 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1747 1748 /* if new clcsock has also inherited the fallback-specific callback 1749 * functions, switch them back to the original ones. 1750 */ 1751 if (lsmc->use_fallback) { 1752 if (lsmc->clcsk_state_change) 1753 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1754 if (lsmc->clcsk_write_space) 1755 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1756 if (lsmc->clcsk_error_report) 1757 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1758 } 1759 1760 (*new_smc)->clcsock = new_clcsock; 1761 out: 1762 return rc; 1763 } 1764 1765 /* add a just created sock to the accept queue of the listen sock as 1766 * candidate for a following socket accept call from user space 1767 */ 1768 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1769 { 1770 struct smc_sock *par = smc_sk(parent); 1771 1772 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1773 spin_lock(&par->accept_q_lock); 1774 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1775 spin_unlock(&par->accept_q_lock); 1776 sk_acceptq_added(parent); 1777 } 1778 1779 /* remove a socket from the accept queue of its parental listening socket */ 1780 static void smc_accept_unlink(struct sock *sk) 1781 { 1782 struct smc_sock *par = smc_sk(sk)->listen_smc; 1783 1784 spin_lock(&par->accept_q_lock); 1785 list_del_init(&smc_sk(sk)->accept_q); 1786 spin_unlock(&par->accept_q_lock); 1787 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1788 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1789 } 1790 1791 /* remove a sock from the accept queue to bind it to a new socket created 1792 * for a socket accept call from user space 1793 */ 1794 struct sock *smc_accept_dequeue(struct sock *parent, 1795 struct socket *new_sock) 1796 { 1797 struct smc_sock *isk, *n; 1798 struct sock *new_sk; 1799 1800 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1801 new_sk = (struct sock *)isk; 1802 1803 smc_accept_unlink(new_sk); 1804 if (new_sk->sk_state == SMC_CLOSED) { 1805 new_sk->sk_prot->unhash(new_sk); 1806 if (isk->clcsock) { 1807 sock_release(isk->clcsock); 1808 isk->clcsock = NULL; 1809 } 1810 sock_put(new_sk); /* final */ 1811 continue; 1812 } 1813 if (new_sock) { 1814 sock_graft(new_sk, new_sock); 1815 new_sock->state = SS_CONNECTED; 1816 if (isk->use_fallback) { 1817 smc_sk(new_sk)->clcsock->file = new_sock->file; 1818 isk->clcsock->file->private_data = isk->clcsock; 1819 } 1820 } 1821 return new_sk; 1822 } 1823 return NULL; 1824 } 1825 1826 /* clean up for a created but never accepted sock */ 1827 void smc_close_non_accepted(struct sock *sk) 1828 { 1829 struct smc_sock *smc = smc_sk(sk); 1830 1831 sock_hold(sk); /* sock_put below */ 1832 lock_sock(sk); 1833 if (!sk->sk_lingertime) 1834 /* wait for peer closing */ 1835 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1836 __smc_release(smc); 1837 release_sock(sk); 1838 sock_put(sk); /* sock_hold above */ 1839 sock_put(sk); /* final sock_put */ 1840 } 1841 1842 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1843 { 1844 struct smc_link *link = smc->conn.lnk; 1845 struct smc_llc_qentry *qentry; 1846 int rc; 1847 1848 /* reg the sndbuf if it was vzalloced*/ 1849 if (smc->conn.sndbuf_desc->is_vm) { 1850 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1851 return SMC_CLC_DECL_ERR_REGBUF; 1852 } 1853 1854 /* reg the rmb */ 1855 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1856 return SMC_CLC_DECL_ERR_REGBUF; 1857 1858 /* send CONFIRM LINK request to client over the RoCE fabric */ 1859 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1860 if (rc < 0) 1861 return SMC_CLC_DECL_TIMEOUT_CL; 1862 1863 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1864 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1865 SMC_LLC_CONFIRM_LINK); 1866 if (!qentry) { 1867 struct smc_clc_msg_decline dclc; 1868 1869 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1870 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1871 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1872 } 1873 smc_llc_save_peer_uid(qentry); 1874 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1875 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1876 if (rc) 1877 return SMC_CLC_DECL_RMBE_EC; 1878 1879 /* confirm_rkey is implicit on 1st contact */ 1880 smc->conn.rmb_desc->is_conf_rkey = true; 1881 1882 smc_llc_link_active(link); 1883 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1884 1885 if (link->lgr->max_links > 1) { 1886 down_write(&link->lgr->llc_conf_mutex); 1887 /* initial contact - try to establish second link */ 1888 smc_llc_srv_add_link(link, NULL); 1889 up_write(&link->lgr->llc_conf_mutex); 1890 } 1891 return 0; 1892 } 1893 1894 /* listen worker: finish */ 1895 static void smc_listen_out(struct smc_sock *new_smc) 1896 { 1897 struct smc_sock *lsmc = new_smc->listen_smc; 1898 struct sock *newsmcsk = &new_smc->sk; 1899 1900 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1901 atomic_dec(&lsmc->queued_smc_hs); 1902 1903 release_sock(newsmcsk); /* lock in smc_listen_work() */ 1904 if (lsmc->sk.sk_state == SMC_LISTEN) { 1905 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1906 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1907 release_sock(&lsmc->sk); 1908 } else { /* no longer listening */ 1909 smc_close_non_accepted(newsmcsk); 1910 } 1911 1912 /* Wake up accept */ 1913 lsmc->sk.sk_data_ready(&lsmc->sk); 1914 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1915 } 1916 1917 /* listen worker: finish in state connected */ 1918 static void smc_listen_out_connected(struct smc_sock *new_smc) 1919 { 1920 struct sock *newsmcsk = &new_smc->sk; 1921 1922 if (newsmcsk->sk_state == SMC_INIT) 1923 newsmcsk->sk_state = SMC_ACTIVE; 1924 1925 smc_listen_out(new_smc); 1926 } 1927 1928 /* listen worker: finish in error state */ 1929 static void smc_listen_out_err(struct smc_sock *new_smc) 1930 { 1931 struct sock *newsmcsk = &new_smc->sk; 1932 struct net *net = sock_net(newsmcsk); 1933 1934 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1935 if (newsmcsk->sk_state == SMC_INIT) 1936 sock_put(&new_smc->sk); /* passive closing */ 1937 newsmcsk->sk_state = SMC_CLOSED; 1938 1939 smc_listen_out(new_smc); 1940 } 1941 1942 /* listen worker: decline and fall back if possible */ 1943 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1944 int local_first, u8 version) 1945 { 1946 /* RDMA setup failed, switch back to TCP */ 1947 smc_conn_abort(new_smc, local_first); 1948 if (reason_code < 0 || 1949 smc_switch_to_fallback(new_smc, reason_code)) { 1950 /* error, no fallback possible */ 1951 smc_listen_out_err(new_smc); 1952 return; 1953 } 1954 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1955 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1956 smc_listen_out_err(new_smc); 1957 return; 1958 } 1959 } 1960 smc_listen_out_connected(new_smc); 1961 } 1962 1963 /* listen worker: version checking */ 1964 static int smc_listen_v2_check(struct smc_sock *new_smc, 1965 struct smc_clc_msg_proposal *pclc, 1966 struct smc_init_info *ini) 1967 { 1968 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1969 struct smc_clc_v2_extension *pclc_v2_ext; 1970 int rc = SMC_CLC_DECL_PEERNOSMC; 1971 1972 ini->smc_type_v1 = pclc->hdr.typev1; 1973 ini->smc_type_v2 = pclc->hdr.typev2; 1974 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1975 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1976 if (pclc->hdr.version > SMC_V1) { 1977 if (smcd_indicated(ini->smc_type_v2)) 1978 ini->smcd_version |= SMC_V2; 1979 if (smcr_indicated(ini->smc_type_v2)) 1980 ini->smcr_version |= SMC_V2; 1981 } 1982 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1983 rc = SMC_CLC_DECL_PEERNOSMC; 1984 goto out; 1985 } 1986 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1987 if (!pclc_v2_ext) { 1988 ini->smcd_version &= ~SMC_V2; 1989 ini->smcr_version &= ~SMC_V2; 1990 rc = SMC_CLC_DECL_NOV2EXT; 1991 goto out; 1992 } 1993 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1994 if (ini->smcd_version & SMC_V2) { 1995 if (!smc_ism_is_v2_capable()) { 1996 ini->smcd_version &= ~SMC_V2; 1997 rc = SMC_CLC_DECL_NOISM2SUPP; 1998 } else if (!pclc_smcd_v2_ext) { 1999 ini->smcd_version &= ~SMC_V2; 2000 rc = SMC_CLC_DECL_NOV2DEXT; 2001 } else if (!pclc_v2_ext->hdr.eid_cnt && 2002 !pclc_v2_ext->hdr.flag.seid) { 2003 ini->smcd_version &= ~SMC_V2; 2004 rc = SMC_CLC_DECL_NOUEID; 2005 } 2006 } 2007 if (ini->smcr_version & SMC_V2) { 2008 if (!pclc_v2_ext->hdr.eid_cnt) { 2009 ini->smcr_version &= ~SMC_V2; 2010 rc = SMC_CLC_DECL_NOUEID; 2011 } 2012 } 2013 2014 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2015 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2016 ini->release_nr = SMC_RELEASE; 2017 2018 out: 2019 if (!ini->smcd_version && !ini->smcr_version) 2020 return rc; 2021 2022 return 0; 2023 } 2024 2025 /* listen worker: check prefixes */ 2026 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2027 struct smc_clc_msg_proposal *pclc) 2028 { 2029 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2030 struct socket *newclcsock = new_smc->clcsock; 2031 2032 if (pclc->hdr.typev1 == SMC_TYPE_N) 2033 return 0; 2034 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2035 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2036 return SMC_CLC_DECL_DIFFPREFIX; 2037 2038 return 0; 2039 } 2040 2041 /* listen worker: initialize connection and buffers */ 2042 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2043 struct smc_init_info *ini) 2044 { 2045 int rc; 2046 2047 /* allocate connection / link group */ 2048 rc = smc_conn_create(new_smc, ini); 2049 if (rc) 2050 return rc; 2051 2052 /* create send buffer and rmb */ 2053 if (smc_buf_create(new_smc, false)) { 2054 smc_conn_abort(new_smc, ini->first_contact_local); 2055 return SMC_CLC_DECL_MEM; 2056 } 2057 2058 return 0; 2059 } 2060 2061 /* listen worker: initialize connection and buffers for SMC-D */ 2062 static int smc_listen_ism_init(struct smc_sock *new_smc, 2063 struct smc_init_info *ini) 2064 { 2065 int rc; 2066 2067 rc = smc_conn_create(new_smc, ini); 2068 if (rc) 2069 return rc; 2070 2071 /* Create send and receive buffers */ 2072 rc = smc_buf_create(new_smc, true); 2073 if (rc) { 2074 smc_conn_abort(new_smc, ini->first_contact_local); 2075 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2076 SMC_CLC_DECL_MEM; 2077 } 2078 2079 return 0; 2080 } 2081 2082 static bool smc_is_already_selected(struct smcd_dev *smcd, 2083 struct smc_init_info *ini, 2084 int matches) 2085 { 2086 int i; 2087 2088 for (i = 0; i < matches; i++) 2089 if (smcd == ini->ism_dev[i]) 2090 return true; 2091 2092 return false; 2093 } 2094 2095 /* check for ISM devices matching proposed ISM devices */ 2096 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2097 u16 proposed_chid, 2098 struct smcd_gid *proposed_gid, 2099 unsigned int *matches) 2100 { 2101 struct smcd_dev *smcd; 2102 2103 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2104 if (smcd->going_away) 2105 continue; 2106 if (smc_is_already_selected(smcd, ini, *matches)) 2107 continue; 2108 if (smc_ism_get_chid(smcd) == proposed_chid && 2109 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2110 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2111 if (__smc_ism_is_emulated(proposed_chid)) 2112 ini->ism_peer_gid[*matches].gid_ext = 2113 proposed_gid->gid_ext; 2114 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2115 ini->ism_dev[*matches] = smcd; 2116 (*matches)++; 2117 break; 2118 } 2119 } 2120 } 2121 2122 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2123 { 2124 if (!ini->rc) 2125 ini->rc = rc; 2126 } 2127 2128 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2129 struct smc_clc_msg_proposal *pclc, 2130 struct smc_init_info *ini) 2131 { 2132 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2133 struct smc_clc_v2_extension *smc_v2_ext; 2134 struct smc_clc_msg_smcd *pclc_smcd; 2135 unsigned int matches = 0; 2136 struct smcd_gid smcd_gid; 2137 u8 smcd_version; 2138 u8 *eid = NULL; 2139 int i, rc; 2140 u16 chid; 2141 2142 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2143 goto not_found; 2144 2145 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2146 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2147 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2148 2149 mutex_lock(&smcd_dev_list.mutex); 2150 if (pclc_smcd->ism.chid) { 2151 /* check for ISM device matching proposed native ISM device */ 2152 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2153 smcd_gid.gid_ext = 0; 2154 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2155 &smcd_gid, &matches); 2156 } 2157 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2158 /* check for ISM devices matching proposed non-native ISM 2159 * devices 2160 */ 2161 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2162 smcd_gid.gid_ext = 0; 2163 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2164 if (__smc_ism_is_emulated(chid)) { 2165 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2166 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2167 /* each Emulated-ISM device takes two GID-CHID 2168 * entries and CHID of the second entry repeats 2169 * that of the first entry. 2170 * 2171 * So check if the next GID-CHID entry exists 2172 * and both two entries' CHIDs are the same. 2173 */ 2174 continue; 2175 smcd_gid.gid_ext = 2176 ntohll(smcd_v2_ext->gidchid[++i].gid); 2177 } 2178 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2179 } 2180 mutex_unlock(&smcd_dev_list.mutex); 2181 2182 if (!ini->ism_dev[0]) { 2183 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2184 goto not_found; 2185 } 2186 2187 smc_ism_get_system_eid(&eid); 2188 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2189 smcd_v2_ext->system_eid, eid)) 2190 goto not_found; 2191 2192 /* separate - outside the smcd_dev_list.lock */ 2193 smcd_version = ini->smcd_version; 2194 for (i = 0; i < matches; i++) { 2195 ini->smcd_version = SMC_V2; 2196 ini->is_smcd = true; 2197 ini->ism_selected = i; 2198 rc = smc_listen_ism_init(new_smc, ini); 2199 if (rc) { 2200 smc_find_ism_store_rc(rc, ini); 2201 /* try next active ISM device */ 2202 continue; 2203 } 2204 return; /* matching and usable V2 ISM device found */ 2205 } 2206 /* no V2 ISM device could be initialized */ 2207 ini->smcd_version = smcd_version; /* restore original value */ 2208 ini->negotiated_eid[0] = 0; 2209 2210 not_found: 2211 ini->smcd_version &= ~SMC_V2; 2212 ini->ism_dev[0] = NULL; 2213 ini->is_smcd = false; 2214 } 2215 2216 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2217 struct smc_clc_msg_proposal *pclc, 2218 struct smc_init_info *ini) 2219 { 2220 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2221 int rc = 0; 2222 2223 /* check if ISM V1 is available */ 2224 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2225 goto not_found; 2226 ini->is_smcd = true; /* prepare ISM check */ 2227 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2228 ini->ism_peer_gid[0].gid_ext = 0; 2229 rc = smc_find_ism_device(new_smc, ini); 2230 if (rc) 2231 goto not_found; 2232 ini->ism_selected = 0; 2233 rc = smc_listen_ism_init(new_smc, ini); 2234 if (!rc) 2235 return; /* V1 ISM device found */ 2236 2237 not_found: 2238 smc_find_ism_store_rc(rc, ini); 2239 ini->smcd_version &= ~SMC_V1; 2240 ini->ism_dev[0] = NULL; 2241 ini->is_smcd = false; 2242 } 2243 2244 /* listen worker: register buffers */ 2245 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2246 { 2247 struct smc_connection *conn = &new_smc->conn; 2248 2249 if (!local_first) { 2250 /* reg sendbufs if they were vzalloced */ 2251 if (conn->sndbuf_desc->is_vm) { 2252 if (smcr_lgr_reg_sndbufs(conn->lnk, 2253 conn->sndbuf_desc)) 2254 return SMC_CLC_DECL_ERR_REGBUF; 2255 } 2256 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2257 return SMC_CLC_DECL_ERR_REGBUF; 2258 } 2259 2260 return 0; 2261 } 2262 2263 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2264 struct smc_clc_msg_proposal *pclc, 2265 struct smc_init_info *ini) 2266 { 2267 struct smc_clc_v2_extension *smc_v2_ext; 2268 u8 smcr_version; 2269 int rc; 2270 2271 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2272 goto not_found; 2273 2274 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2275 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2276 goto not_found; 2277 2278 /* prepare RDMA check */ 2279 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2280 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2281 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2282 ini->check_smcrv2 = true; 2283 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2284 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2285 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2286 rc = smc_find_rdma_device(new_smc, ini); 2287 if (rc) { 2288 smc_find_ism_store_rc(rc, ini); 2289 goto not_found; 2290 } 2291 if (!ini->smcrv2.uses_gateway) 2292 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2293 2294 smcr_version = ini->smcr_version; 2295 ini->smcr_version = SMC_V2; 2296 rc = smc_listen_rdma_init(new_smc, ini); 2297 if (!rc) { 2298 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2299 if (rc) 2300 smc_conn_abort(new_smc, ini->first_contact_local); 2301 } 2302 if (!rc) 2303 return; 2304 ini->smcr_version = smcr_version; 2305 smc_find_ism_store_rc(rc, ini); 2306 2307 not_found: 2308 ini->smcr_version &= ~SMC_V2; 2309 ini->smcrv2.ib_dev_v2 = NULL; 2310 ini->check_smcrv2 = false; 2311 } 2312 2313 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2314 struct smc_clc_msg_proposal *pclc, 2315 struct smc_init_info *ini) 2316 { 2317 int rc; 2318 2319 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2320 return SMC_CLC_DECL_NOSMCDEV; 2321 2322 /* prepare RDMA check */ 2323 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2324 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2325 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2326 rc = smc_find_rdma_device(new_smc, ini); 2327 if (rc) { 2328 /* no RDMA device found */ 2329 return SMC_CLC_DECL_NOSMCDEV; 2330 } 2331 rc = smc_listen_rdma_init(new_smc, ini); 2332 if (rc) 2333 return rc; 2334 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2335 } 2336 2337 /* determine the local device matching to proposal */ 2338 static int smc_listen_find_device(struct smc_sock *new_smc, 2339 struct smc_clc_msg_proposal *pclc, 2340 struct smc_init_info *ini) 2341 { 2342 int prfx_rc; 2343 2344 /* check for ISM device matching V2 proposed device */ 2345 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2346 if (ini->ism_dev[0]) 2347 return 0; 2348 2349 /* check for matching IP prefix and subnet length (V1) */ 2350 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2351 if (prfx_rc) 2352 smc_find_ism_store_rc(prfx_rc, ini); 2353 2354 /* get vlan id from IP device */ 2355 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2356 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2357 2358 /* check for ISM device matching V1 proposed device */ 2359 if (!prfx_rc) 2360 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2361 if (ini->ism_dev[0]) 2362 return 0; 2363 2364 if (!smcr_indicated(pclc->hdr.typev1) && 2365 !smcr_indicated(pclc->hdr.typev2)) 2366 /* skip RDMA and decline */ 2367 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2368 2369 /* check if RDMA V2 is available */ 2370 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2371 if (ini->smcrv2.ib_dev_v2) 2372 return 0; 2373 2374 /* check if RDMA V1 is available */ 2375 if (!prfx_rc) { 2376 int rc; 2377 2378 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2379 smc_find_ism_store_rc(rc, ini); 2380 return (!rc) ? 0 : ini->rc; 2381 } 2382 return prfx_rc; 2383 } 2384 2385 /* listen worker: finish RDMA setup */ 2386 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2387 struct smc_clc_msg_accept_confirm *cclc, 2388 bool local_first, 2389 struct smc_init_info *ini) 2390 { 2391 struct smc_link *link = new_smc->conn.lnk; 2392 int reason_code = 0; 2393 2394 if (local_first) 2395 smc_link_save_peer_info(link, cclc, ini); 2396 2397 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2398 return SMC_CLC_DECL_ERR_RTOK; 2399 2400 if (local_first) { 2401 if (smc_ib_ready_link(link)) 2402 return SMC_CLC_DECL_ERR_RDYLNK; 2403 /* QP confirmation over RoCE fabric */ 2404 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2405 reason_code = smcr_serv_conf_first_link(new_smc); 2406 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2407 } 2408 return reason_code; 2409 } 2410 2411 /* setup for connection of server */ 2412 static void smc_listen_work(struct work_struct *work) 2413 { 2414 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2415 smc_listen_work); 2416 struct socket *newclcsock = new_smc->clcsock; 2417 struct smc_clc_msg_accept_confirm *cclc; 2418 struct smc_clc_msg_proposal_area *buf; 2419 struct smc_clc_msg_proposal *pclc; 2420 struct smc_init_info *ini = NULL; 2421 u8 proposal_version = SMC_V1; 2422 u8 accept_version; 2423 int rc = 0; 2424 2425 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2426 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2427 return smc_listen_out_err(new_smc); 2428 2429 if (new_smc->use_fallback) { 2430 smc_listen_out_connected(new_smc); 2431 return; 2432 } 2433 2434 /* check if peer is smc capable */ 2435 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2436 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2437 if (rc) 2438 smc_listen_out_err(new_smc); 2439 else 2440 smc_listen_out_connected(new_smc); 2441 return; 2442 } 2443 2444 /* do inband token exchange - 2445 * wait for and receive SMC Proposal CLC message 2446 */ 2447 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2448 if (!buf) { 2449 rc = SMC_CLC_DECL_MEM; 2450 goto out_decl; 2451 } 2452 pclc = (struct smc_clc_msg_proposal *)buf; 2453 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2454 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2455 if (rc) 2456 goto out_decl; 2457 2458 if (pclc->hdr.version > SMC_V1) 2459 proposal_version = SMC_V2; 2460 2461 /* IPSec connections opt out of SMC optimizations */ 2462 if (using_ipsec(new_smc)) { 2463 rc = SMC_CLC_DECL_IPSEC; 2464 goto out_decl; 2465 } 2466 2467 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2468 if (!ini) { 2469 rc = SMC_CLC_DECL_MEM; 2470 goto out_decl; 2471 } 2472 2473 /* initial version checking */ 2474 rc = smc_listen_v2_check(new_smc, pclc, ini); 2475 if (rc) 2476 goto out_decl; 2477 2478 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2479 if (rc) 2480 goto out_decl; 2481 2482 mutex_lock(&smc_server_lgr_pending); 2483 smc_rx_init(new_smc); 2484 smc_tx_init(new_smc); 2485 2486 /* determine ISM or RoCE device used for connection */ 2487 rc = smc_listen_find_device(new_smc, pclc, ini); 2488 if (rc) 2489 goto out_unlock; 2490 2491 /* send SMC Accept CLC message */ 2492 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2493 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2494 accept_version, ini->negotiated_eid, ini); 2495 if (rc) 2496 goto out_unlock; 2497 2498 /* SMC-D does not need this lock any more */ 2499 if (ini->is_smcd) 2500 mutex_unlock(&smc_server_lgr_pending); 2501 2502 /* receive SMC Confirm CLC message */ 2503 memset(buf, 0, sizeof(*buf)); 2504 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2505 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2506 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2507 if (rc) { 2508 if (!ini->is_smcd) 2509 goto out_unlock; 2510 goto out_decl; 2511 } 2512 2513 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2514 if (rc) { 2515 if (!ini->is_smcd) 2516 goto out_unlock; 2517 goto out_decl; 2518 } 2519 2520 /* fce smc release version is needed in smc_listen_rdma_finish, 2521 * so save fce info here. 2522 */ 2523 smc_conn_save_peer_info_fce(new_smc, cclc); 2524 2525 /* finish worker */ 2526 if (!ini->is_smcd) { 2527 rc = smc_listen_rdma_finish(new_smc, cclc, 2528 ini->first_contact_local, ini); 2529 if (rc) 2530 goto out_unlock; 2531 mutex_unlock(&smc_server_lgr_pending); 2532 } 2533 smc_conn_save_peer_info(new_smc, cclc); 2534 2535 if (ini->is_smcd && 2536 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2537 rc = smcd_buf_attach(new_smc); 2538 if (rc) 2539 goto out_decl; 2540 } 2541 2542 smc_listen_out_connected(new_smc); 2543 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2544 goto out_free; 2545 2546 out_unlock: 2547 mutex_unlock(&smc_server_lgr_pending); 2548 out_decl: 2549 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2550 proposal_version); 2551 out_free: 2552 kfree(ini); 2553 kfree(buf); 2554 } 2555 2556 static void smc_tcp_listen_work(struct work_struct *work) 2557 { 2558 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2559 tcp_listen_work); 2560 struct sock *lsk = &lsmc->sk; 2561 struct smc_sock *new_smc; 2562 int rc = 0; 2563 2564 lock_sock(lsk); 2565 while (lsk->sk_state == SMC_LISTEN) { 2566 rc = smc_clcsock_accept(lsmc, &new_smc); 2567 if (rc) /* clcsock accept queue empty or error */ 2568 goto out; 2569 if (!new_smc) 2570 continue; 2571 2572 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2573 atomic_inc(&lsmc->queued_smc_hs); 2574 2575 new_smc->listen_smc = lsmc; 2576 new_smc->use_fallback = lsmc->use_fallback; 2577 new_smc->fallback_rsn = lsmc->fallback_rsn; 2578 sock_hold(lsk); /* sock_put in smc_listen_work */ 2579 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2580 smc_copy_sock_settings_to_smc(new_smc); 2581 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2582 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2583 sock_put(&new_smc->sk); 2584 } 2585 2586 out: 2587 release_sock(lsk); 2588 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2589 } 2590 2591 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2592 { 2593 struct smc_sock *lsmc; 2594 2595 read_lock_bh(&listen_clcsock->sk_callback_lock); 2596 lsmc = smc_clcsock_user_data(listen_clcsock); 2597 if (!lsmc) 2598 goto out; 2599 lsmc->clcsk_data_ready(listen_clcsock); 2600 if (lsmc->sk.sk_state == SMC_LISTEN) { 2601 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2602 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2603 sock_put(&lsmc->sk); 2604 } 2605 out: 2606 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2607 } 2608 2609 int smc_listen(struct socket *sock, int backlog) 2610 { 2611 struct sock *sk = sock->sk; 2612 struct smc_sock *smc; 2613 int rc; 2614 2615 smc = smc_sk(sk); 2616 lock_sock(sk); 2617 2618 rc = -EINVAL; 2619 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2620 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2621 goto out; 2622 2623 rc = 0; 2624 if (sk->sk_state == SMC_LISTEN) { 2625 sk->sk_max_ack_backlog = backlog; 2626 goto out; 2627 } 2628 /* some socket options are handled in core, so we could not apply 2629 * them to the clc socket -- copy smc socket options to clc socket 2630 */ 2631 smc_copy_sock_settings_to_clc(smc); 2632 if (!smc->use_fallback) 2633 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2634 2635 /* save original sk_data_ready function and establish 2636 * smc-specific sk_data_ready function 2637 */ 2638 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2639 smc->clcsock->sk->sk_user_data = 2640 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2641 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2642 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2643 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2644 2645 /* save original ops */ 2646 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2647 2648 smc->af_ops = *smc->ori_af_ops; 2649 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2650 2651 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2652 2653 if (smc->limit_smc_hs) 2654 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2655 2656 rc = kernel_listen(smc->clcsock, backlog); 2657 if (rc) { 2658 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2659 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2660 &smc->clcsk_data_ready); 2661 smc->clcsock->sk->sk_user_data = NULL; 2662 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2663 goto out; 2664 } 2665 sk->sk_max_ack_backlog = backlog; 2666 sk->sk_ack_backlog = 0; 2667 sk->sk_state = SMC_LISTEN; 2668 2669 out: 2670 release_sock(sk); 2671 return rc; 2672 } 2673 2674 int smc_accept(struct socket *sock, struct socket *new_sock, 2675 struct proto_accept_arg *arg) 2676 { 2677 struct sock *sk = sock->sk, *nsk; 2678 DECLARE_WAITQUEUE(wait, current); 2679 struct smc_sock *lsmc; 2680 long timeo; 2681 int rc = 0; 2682 2683 lsmc = smc_sk(sk); 2684 sock_hold(sk); /* sock_put below */ 2685 lock_sock(sk); 2686 2687 if (lsmc->sk.sk_state != SMC_LISTEN) { 2688 rc = -EINVAL; 2689 release_sock(sk); 2690 goto out; 2691 } 2692 2693 /* Wait for an incoming connection */ 2694 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2695 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2696 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2697 set_current_state(TASK_INTERRUPTIBLE); 2698 if (!timeo) { 2699 rc = -EAGAIN; 2700 break; 2701 } 2702 release_sock(sk); 2703 timeo = schedule_timeout(timeo); 2704 /* wakeup by sk_data_ready in smc_listen_work() */ 2705 sched_annotate_sleep(); 2706 lock_sock(sk); 2707 if (signal_pending(current)) { 2708 rc = sock_intr_errno(timeo); 2709 break; 2710 } 2711 } 2712 set_current_state(TASK_RUNNING); 2713 remove_wait_queue(sk_sleep(sk), &wait); 2714 2715 if (!rc) 2716 rc = sock_error(nsk); 2717 release_sock(sk); 2718 if (rc) 2719 goto out; 2720 2721 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2722 /* wait till data arrives on the socket */ 2723 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2724 MSEC_PER_SEC); 2725 if (smc_sk(nsk)->use_fallback) { 2726 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2727 2728 lock_sock(clcsk); 2729 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2730 sk_wait_data(clcsk, &timeo, NULL); 2731 release_sock(clcsk); 2732 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2733 lock_sock(nsk); 2734 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2735 release_sock(nsk); 2736 } 2737 } 2738 2739 out: 2740 sock_put(sk); /* sock_hold above */ 2741 return rc; 2742 } 2743 2744 int smc_getname(struct socket *sock, struct sockaddr *addr, 2745 int peer) 2746 { 2747 struct smc_sock *smc; 2748 2749 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2750 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2751 return -ENOTCONN; 2752 2753 smc = smc_sk(sock->sk); 2754 2755 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2756 } 2757 2758 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2759 { 2760 struct sock *sk = sock->sk; 2761 struct smc_sock *smc; 2762 int rc; 2763 2764 smc = smc_sk(sk); 2765 lock_sock(sk); 2766 2767 /* SMC does not support connect with fastopen */ 2768 if (msg->msg_flags & MSG_FASTOPEN) { 2769 /* not connected yet, fallback */ 2770 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2771 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2772 if (rc) 2773 goto out; 2774 } else { 2775 rc = -EINVAL; 2776 goto out; 2777 } 2778 } else if ((sk->sk_state != SMC_ACTIVE) && 2779 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2780 (sk->sk_state != SMC_INIT)) { 2781 rc = -EPIPE; 2782 goto out; 2783 } 2784 2785 if (smc->use_fallback) { 2786 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2787 } else { 2788 rc = smc_tx_sendmsg(smc, msg, len); 2789 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2790 } 2791 out: 2792 release_sock(sk); 2793 return rc; 2794 } 2795 2796 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2797 int flags) 2798 { 2799 struct sock *sk = sock->sk; 2800 struct smc_sock *smc; 2801 int rc = -ENOTCONN; 2802 2803 smc = smc_sk(sk); 2804 lock_sock(sk); 2805 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2806 /* socket was connected before, no more data to read */ 2807 rc = 0; 2808 goto out; 2809 } 2810 if ((sk->sk_state == SMC_INIT) || 2811 (sk->sk_state == SMC_LISTEN) || 2812 (sk->sk_state == SMC_CLOSED)) 2813 goto out; 2814 2815 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2816 rc = 0; 2817 goto out; 2818 } 2819 2820 if (smc->use_fallback) { 2821 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2822 } else { 2823 msg->msg_namelen = 0; 2824 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2825 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2826 } 2827 2828 out: 2829 release_sock(sk); 2830 return rc; 2831 } 2832 2833 static __poll_t smc_accept_poll(struct sock *parent) 2834 { 2835 struct smc_sock *isk = smc_sk(parent); 2836 __poll_t mask = 0; 2837 2838 spin_lock(&isk->accept_q_lock); 2839 if (!list_empty(&isk->accept_q)) 2840 mask = EPOLLIN | EPOLLRDNORM; 2841 spin_unlock(&isk->accept_q_lock); 2842 2843 return mask; 2844 } 2845 2846 __poll_t smc_poll(struct file *file, struct socket *sock, 2847 poll_table *wait) 2848 { 2849 struct sock *sk = sock->sk; 2850 struct smc_sock *smc; 2851 __poll_t mask = 0; 2852 2853 if (!sk) 2854 return EPOLLNVAL; 2855 2856 smc = smc_sk(sock->sk); 2857 if (smc->use_fallback) { 2858 /* delegate to CLC child sock */ 2859 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2860 sk->sk_err = smc->clcsock->sk->sk_err; 2861 } else { 2862 if (sk->sk_state != SMC_CLOSED) 2863 sock_poll_wait(file, sock, wait); 2864 if (sk->sk_err) 2865 mask |= EPOLLERR; 2866 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2867 (sk->sk_state == SMC_CLOSED)) 2868 mask |= EPOLLHUP; 2869 if (sk->sk_state == SMC_LISTEN) { 2870 /* woken up by sk_data_ready in smc_listen_work() */ 2871 mask |= smc_accept_poll(sk); 2872 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2873 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2874 wait); 2875 sk->sk_err = smc->clcsock->sk->sk_err; 2876 } else { 2877 if ((sk->sk_state != SMC_INIT && 2878 atomic_read(&smc->conn.sndbuf_space)) || 2879 sk->sk_shutdown & SEND_SHUTDOWN) { 2880 mask |= EPOLLOUT | EPOLLWRNORM; 2881 } else { 2882 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2883 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2884 } 2885 if (atomic_read(&smc->conn.bytes_to_rcv)) 2886 mask |= EPOLLIN | EPOLLRDNORM; 2887 if (sk->sk_shutdown & RCV_SHUTDOWN) 2888 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2889 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2890 mask |= EPOLLIN; 2891 if (smc->conn.urg_state == SMC_URG_VALID) 2892 mask |= EPOLLPRI; 2893 } 2894 } 2895 2896 return mask; 2897 } 2898 2899 int smc_shutdown(struct socket *sock, int how) 2900 { 2901 struct sock *sk = sock->sk; 2902 bool do_shutdown = true; 2903 struct smc_sock *smc; 2904 int rc = -EINVAL; 2905 int old_state; 2906 int rc1 = 0; 2907 2908 smc = smc_sk(sk); 2909 2910 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2911 return rc; 2912 2913 lock_sock(sk); 2914 2915 if (sock->state == SS_CONNECTING) { 2916 if (sk->sk_state == SMC_ACTIVE) 2917 sock->state = SS_CONNECTED; 2918 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2919 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2920 sk->sk_state == SMC_APPCLOSEWAIT1 || 2921 sk->sk_state == SMC_APPCLOSEWAIT2 || 2922 sk->sk_state == SMC_APPFINCLOSEWAIT) 2923 sock->state = SS_DISCONNECTING; 2924 } 2925 2926 rc = -ENOTCONN; 2927 if ((sk->sk_state != SMC_ACTIVE) && 2928 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2929 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2930 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2931 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2932 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2933 goto out; 2934 if (smc->use_fallback) { 2935 rc = kernel_sock_shutdown(smc->clcsock, how); 2936 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2937 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2938 sk->sk_state = SMC_CLOSED; 2939 sk->sk_socket->state = SS_UNCONNECTED; 2940 sock_put(sk); 2941 } 2942 goto out; 2943 } 2944 switch (how) { 2945 case SHUT_RDWR: /* shutdown in both directions */ 2946 old_state = sk->sk_state; 2947 rc = smc_close_active(smc); 2948 if (old_state == SMC_ACTIVE && 2949 sk->sk_state == SMC_PEERCLOSEWAIT1) 2950 do_shutdown = false; 2951 break; 2952 case SHUT_WR: 2953 rc = smc_close_shutdown_write(smc); 2954 break; 2955 case SHUT_RD: 2956 rc = 0; 2957 /* nothing more to do because peer is not involved */ 2958 break; 2959 } 2960 if (do_shutdown && smc->clcsock) 2961 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2962 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2963 sk->sk_shutdown |= how + 1; 2964 2965 if (sk->sk_state == SMC_CLOSED) 2966 sock->state = SS_UNCONNECTED; 2967 else 2968 sock->state = SS_DISCONNECTING; 2969 out: 2970 release_sock(sk); 2971 return rc ? rc : rc1; 2972 } 2973 2974 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2975 char __user *optval, int __user *optlen) 2976 { 2977 struct smc_sock *smc; 2978 int val, len; 2979 2980 smc = smc_sk(sock->sk); 2981 2982 if (get_user(len, optlen)) 2983 return -EFAULT; 2984 2985 len = min_t(int, len, sizeof(int)); 2986 2987 if (len < 0) 2988 return -EINVAL; 2989 2990 switch (optname) { 2991 case SMC_LIMIT_HS: 2992 val = smc->limit_smc_hs; 2993 break; 2994 default: 2995 return -EOPNOTSUPP; 2996 } 2997 2998 if (put_user(len, optlen)) 2999 return -EFAULT; 3000 if (copy_to_user(optval, &val, len)) 3001 return -EFAULT; 3002 3003 return 0; 3004 } 3005 3006 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3007 sockptr_t optval, unsigned int optlen) 3008 { 3009 struct sock *sk = sock->sk; 3010 struct smc_sock *smc; 3011 int val, rc; 3012 3013 smc = smc_sk(sk); 3014 3015 lock_sock(sk); 3016 switch (optname) { 3017 case SMC_LIMIT_HS: 3018 if (optlen < sizeof(int)) { 3019 rc = -EINVAL; 3020 break; 3021 } 3022 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3023 rc = -EFAULT; 3024 break; 3025 } 3026 3027 smc->limit_smc_hs = !!val; 3028 rc = 0; 3029 break; 3030 default: 3031 rc = -EOPNOTSUPP; 3032 break; 3033 } 3034 release_sock(sk); 3035 3036 return rc; 3037 } 3038 3039 int smc_setsockopt(struct socket *sock, int level, int optname, 3040 sockptr_t optval, unsigned int optlen) 3041 { 3042 struct sock *sk = sock->sk; 3043 struct smc_sock *smc; 3044 int val, rc; 3045 3046 if (level == SOL_TCP && optname == TCP_ULP) 3047 return -EOPNOTSUPP; 3048 else if (level == SOL_SMC) 3049 return __smc_setsockopt(sock, level, optname, optval, optlen); 3050 3051 smc = smc_sk(sk); 3052 3053 /* generic setsockopts reaching us here always apply to the 3054 * CLC socket 3055 */ 3056 mutex_lock(&smc->clcsock_release_lock); 3057 if (!smc->clcsock) { 3058 mutex_unlock(&smc->clcsock_release_lock); 3059 return -EBADF; 3060 } 3061 if (unlikely(!smc->clcsock->ops->setsockopt)) 3062 rc = -EOPNOTSUPP; 3063 else 3064 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3065 optval, optlen); 3066 if (smc->clcsock->sk->sk_err) { 3067 sk->sk_err = smc->clcsock->sk->sk_err; 3068 sk_error_report(sk); 3069 } 3070 mutex_unlock(&smc->clcsock_release_lock); 3071 3072 if (optlen < sizeof(int)) 3073 return -EINVAL; 3074 if (copy_from_sockptr(&val, optval, sizeof(int))) 3075 return -EFAULT; 3076 3077 lock_sock(sk); 3078 if (rc || smc->use_fallback) 3079 goto out; 3080 switch (optname) { 3081 case TCP_FASTOPEN: 3082 case TCP_FASTOPEN_CONNECT: 3083 case TCP_FASTOPEN_KEY: 3084 case TCP_FASTOPEN_NO_COOKIE: 3085 /* option not supported by SMC */ 3086 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3087 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3088 } else { 3089 rc = -EINVAL; 3090 } 3091 break; 3092 case TCP_NODELAY: 3093 if (sk->sk_state != SMC_INIT && 3094 sk->sk_state != SMC_LISTEN && 3095 sk->sk_state != SMC_CLOSED) { 3096 if (val) { 3097 SMC_STAT_INC(smc, ndly_cnt); 3098 smc_tx_pending(&smc->conn); 3099 cancel_delayed_work(&smc->conn.tx_work); 3100 } 3101 } 3102 break; 3103 case TCP_CORK: 3104 if (sk->sk_state != SMC_INIT && 3105 sk->sk_state != SMC_LISTEN && 3106 sk->sk_state != SMC_CLOSED) { 3107 if (!val) { 3108 SMC_STAT_INC(smc, cork_cnt); 3109 smc_tx_pending(&smc->conn); 3110 cancel_delayed_work(&smc->conn.tx_work); 3111 } 3112 } 3113 break; 3114 case TCP_DEFER_ACCEPT: 3115 smc->sockopt_defer_accept = val; 3116 break; 3117 default: 3118 break; 3119 } 3120 out: 3121 release_sock(sk); 3122 3123 return rc; 3124 } 3125 3126 int smc_getsockopt(struct socket *sock, int level, int optname, 3127 char __user *optval, int __user *optlen) 3128 { 3129 struct smc_sock *smc; 3130 int rc; 3131 3132 if (level == SOL_SMC) 3133 return __smc_getsockopt(sock, level, optname, optval, optlen); 3134 3135 smc = smc_sk(sock->sk); 3136 mutex_lock(&smc->clcsock_release_lock); 3137 if (!smc->clcsock) { 3138 mutex_unlock(&smc->clcsock_release_lock); 3139 return -EBADF; 3140 } 3141 /* socket options apply to the CLC socket */ 3142 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3143 mutex_unlock(&smc->clcsock_release_lock); 3144 return -EOPNOTSUPP; 3145 } 3146 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3147 optval, optlen); 3148 mutex_unlock(&smc->clcsock_release_lock); 3149 return rc; 3150 } 3151 3152 int smc_ioctl(struct socket *sock, unsigned int cmd, 3153 unsigned long arg) 3154 { 3155 union smc_host_cursor cons, urg; 3156 struct smc_connection *conn; 3157 struct smc_sock *smc; 3158 int answ; 3159 3160 smc = smc_sk(sock->sk); 3161 conn = &smc->conn; 3162 lock_sock(&smc->sk); 3163 if (smc->use_fallback) { 3164 if (!smc->clcsock) { 3165 release_sock(&smc->sk); 3166 return -EBADF; 3167 } 3168 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3169 release_sock(&smc->sk); 3170 return answ; 3171 } 3172 switch (cmd) { 3173 case SIOCINQ: /* same as FIONREAD */ 3174 if (smc->sk.sk_state == SMC_LISTEN) { 3175 release_sock(&smc->sk); 3176 return -EINVAL; 3177 } 3178 if (smc->sk.sk_state == SMC_INIT || 3179 smc->sk.sk_state == SMC_CLOSED) 3180 answ = 0; 3181 else 3182 answ = atomic_read(&smc->conn.bytes_to_rcv); 3183 break; 3184 case SIOCOUTQ: 3185 /* output queue size (not send + not acked) */ 3186 if (smc->sk.sk_state == SMC_LISTEN) { 3187 release_sock(&smc->sk); 3188 return -EINVAL; 3189 } 3190 if (smc->sk.sk_state == SMC_INIT || 3191 smc->sk.sk_state == SMC_CLOSED) 3192 answ = 0; 3193 else 3194 answ = smc->conn.sndbuf_desc->len - 3195 atomic_read(&smc->conn.sndbuf_space); 3196 break; 3197 case SIOCOUTQNSD: 3198 /* output queue size (not send only) */ 3199 if (smc->sk.sk_state == SMC_LISTEN) { 3200 release_sock(&smc->sk); 3201 return -EINVAL; 3202 } 3203 if (smc->sk.sk_state == SMC_INIT || 3204 smc->sk.sk_state == SMC_CLOSED) 3205 answ = 0; 3206 else 3207 answ = smc_tx_prepared_sends(&smc->conn); 3208 break; 3209 case SIOCATMARK: 3210 if (smc->sk.sk_state == SMC_LISTEN) { 3211 release_sock(&smc->sk); 3212 return -EINVAL; 3213 } 3214 if (smc->sk.sk_state == SMC_INIT || 3215 smc->sk.sk_state == SMC_CLOSED) { 3216 answ = 0; 3217 } else { 3218 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3219 smc_curs_copy(&urg, &conn->urg_curs, conn); 3220 answ = smc_curs_diff(conn->rmb_desc->len, 3221 &cons, &urg) == 1; 3222 } 3223 break; 3224 default: 3225 release_sock(&smc->sk); 3226 return -ENOIOCTLCMD; 3227 } 3228 release_sock(&smc->sk); 3229 3230 return put_user(answ, (int __user *)arg); 3231 } 3232 3233 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3234 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3235 * updates till whenever a respective page has been fully processed. 3236 * Note that subsequent recv() calls have to wait till all splice() processing 3237 * completed. 3238 */ 3239 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3240 struct pipe_inode_info *pipe, size_t len, 3241 unsigned int flags) 3242 { 3243 struct sock *sk = sock->sk; 3244 struct smc_sock *smc; 3245 int rc = -ENOTCONN; 3246 3247 smc = smc_sk(sk); 3248 lock_sock(sk); 3249 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3250 /* socket was connected before, no more data to read */ 3251 rc = 0; 3252 goto out; 3253 } 3254 if (sk->sk_state == SMC_INIT || 3255 sk->sk_state == SMC_LISTEN || 3256 sk->sk_state == SMC_CLOSED) 3257 goto out; 3258 3259 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3260 rc = 0; 3261 goto out; 3262 } 3263 3264 if (smc->use_fallback) { 3265 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3266 pipe, len, flags); 3267 } else { 3268 if (*ppos) { 3269 rc = -ESPIPE; 3270 goto out; 3271 } 3272 if (flags & SPLICE_F_NONBLOCK) 3273 flags = MSG_DONTWAIT; 3274 else 3275 flags = 0; 3276 SMC_STAT_INC(smc, splice_cnt); 3277 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3278 } 3279 out: 3280 release_sock(sk); 3281 3282 return rc; 3283 } 3284 3285 /* must look like tcp */ 3286 static const struct proto_ops smc_sock_ops = { 3287 .family = PF_SMC, 3288 .owner = THIS_MODULE, 3289 .release = smc_release, 3290 .bind = smc_bind, 3291 .connect = smc_connect, 3292 .socketpair = sock_no_socketpair, 3293 .accept = smc_accept, 3294 .getname = smc_getname, 3295 .poll = smc_poll, 3296 .ioctl = smc_ioctl, 3297 .listen = smc_listen, 3298 .shutdown = smc_shutdown, 3299 .setsockopt = smc_setsockopt, 3300 .getsockopt = smc_getsockopt, 3301 .sendmsg = smc_sendmsg, 3302 .recvmsg = smc_recvmsg, 3303 .mmap = sock_no_mmap, 3304 .splice_read = smc_splice_read, 3305 }; 3306 3307 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3308 { 3309 struct smc_sock *smc = smc_sk(sk); 3310 int rc; 3311 3312 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3313 &smc->clcsock); 3314 if (rc) 3315 return rc; 3316 3317 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3318 * destruction; its sk_state might not be TCP_CLOSE after 3319 * smc->sk is close()d, and TCP timers can be fired later, 3320 * which need net ref. 3321 */ 3322 sk = smc->clcsock->sk; 3323 __netns_tracker_free(net, &sk->ns_tracker, false); 3324 sk->sk_net_refcnt = 1; 3325 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3326 sock_inuse_add(net, 1); 3327 return 0; 3328 } 3329 3330 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3331 int kern, struct socket *clcsock) 3332 { 3333 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3334 struct smc_sock *smc; 3335 struct sock *sk; 3336 int rc; 3337 3338 rc = -ESOCKTNOSUPPORT; 3339 if (sock->type != SOCK_STREAM) 3340 goto out; 3341 3342 rc = -EPROTONOSUPPORT; 3343 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3344 goto out; 3345 3346 rc = -ENOBUFS; 3347 sock->ops = &smc_sock_ops; 3348 sock->state = SS_UNCONNECTED; 3349 sk = smc_sock_alloc(net, sock, protocol); 3350 if (!sk) 3351 goto out; 3352 3353 /* create internal TCP socket for CLC handshake and fallback */ 3354 smc = smc_sk(sk); 3355 3356 rc = 0; 3357 if (clcsock) 3358 smc->clcsock = clcsock; 3359 else 3360 rc = smc_create_clcsk(net, sk, family); 3361 3362 if (rc) { 3363 sk_common_release(sk); 3364 sock->sk = NULL; 3365 } 3366 out: 3367 return rc; 3368 } 3369 3370 static int smc_create(struct net *net, struct socket *sock, int protocol, 3371 int kern) 3372 { 3373 return __smc_create(net, sock, protocol, kern, NULL); 3374 } 3375 3376 static const struct net_proto_family smc_sock_family_ops = { 3377 .family = PF_SMC, 3378 .owner = THIS_MODULE, 3379 .create = smc_create, 3380 }; 3381 3382 static int smc_ulp_init(struct sock *sk) 3383 { 3384 struct socket *tcp = sk->sk_socket; 3385 struct net *net = sock_net(sk); 3386 struct socket *smcsock; 3387 int protocol, ret; 3388 3389 /* only TCP can be replaced */ 3390 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3391 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3392 return -ESOCKTNOSUPPORT; 3393 /* don't handle wq now */ 3394 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3395 return -ENOTCONN; 3396 3397 if (sk->sk_family == AF_INET) 3398 protocol = SMCPROTO_SMC; 3399 else 3400 protocol = SMCPROTO_SMC6; 3401 3402 smcsock = sock_alloc(); 3403 if (!smcsock) 3404 return -ENFILE; 3405 3406 smcsock->type = SOCK_STREAM; 3407 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3408 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3409 if (ret) { 3410 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3411 return ret; 3412 } 3413 3414 /* replace tcp socket to smc */ 3415 smcsock->file = tcp->file; 3416 smcsock->file->private_data = smcsock; 3417 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3418 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3419 tcp->file = NULL; 3420 3421 return ret; 3422 } 3423 3424 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3425 const gfp_t priority) 3426 { 3427 struct inet_connection_sock *icsk = inet_csk(newsk); 3428 3429 /* don't inherit ulp ops to child when listen */ 3430 icsk->icsk_ulp_ops = NULL; 3431 } 3432 3433 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3434 .name = "smc", 3435 .owner = THIS_MODULE, 3436 .init = smc_ulp_init, 3437 .clone = smc_ulp_clone, 3438 }; 3439 3440 unsigned int smc_net_id; 3441 3442 static __net_init int smc_net_init(struct net *net) 3443 { 3444 int rc; 3445 3446 rc = smc_sysctl_net_init(net); 3447 if (rc) 3448 return rc; 3449 return smc_pnet_net_init(net); 3450 } 3451 3452 static void __net_exit smc_net_exit(struct net *net) 3453 { 3454 smc_sysctl_net_exit(net); 3455 smc_pnet_net_exit(net); 3456 } 3457 3458 static __net_init int smc_net_stat_init(struct net *net) 3459 { 3460 return smc_stats_init(net); 3461 } 3462 3463 static void __net_exit smc_net_stat_exit(struct net *net) 3464 { 3465 smc_stats_exit(net); 3466 } 3467 3468 static struct pernet_operations smc_net_ops = { 3469 .init = smc_net_init, 3470 .exit = smc_net_exit, 3471 .id = &smc_net_id, 3472 .size = sizeof(struct smc_net), 3473 }; 3474 3475 static struct pernet_operations smc_net_stat_ops = { 3476 .init = smc_net_stat_init, 3477 .exit = smc_net_stat_exit, 3478 }; 3479 3480 static int __init smc_init(void) 3481 { 3482 int rc; 3483 3484 rc = register_pernet_subsys(&smc_net_ops); 3485 if (rc) 3486 return rc; 3487 3488 rc = register_pernet_subsys(&smc_net_stat_ops); 3489 if (rc) 3490 goto out_pernet_subsys; 3491 3492 rc = smc_ism_init(); 3493 if (rc) 3494 goto out_pernet_subsys_stat; 3495 smc_clc_init(); 3496 3497 rc = smc_nl_init(); 3498 if (rc) 3499 goto out_ism; 3500 3501 rc = smc_pnet_init(); 3502 if (rc) 3503 goto out_nl; 3504 3505 rc = -ENOMEM; 3506 3507 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3508 if (!smc_tcp_ls_wq) 3509 goto out_pnet; 3510 3511 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3512 if (!smc_hs_wq) 3513 goto out_alloc_tcp_ls_wq; 3514 3515 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3516 if (!smc_close_wq) 3517 goto out_alloc_hs_wq; 3518 3519 rc = smc_core_init(); 3520 if (rc) { 3521 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3522 goto out_alloc_wqs; 3523 } 3524 3525 rc = smc_llc_init(); 3526 if (rc) { 3527 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3528 goto out_core; 3529 } 3530 3531 rc = smc_cdc_init(); 3532 if (rc) { 3533 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3534 goto out_core; 3535 } 3536 3537 rc = proto_register(&smc_proto, 1); 3538 if (rc) { 3539 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3540 goto out_core; 3541 } 3542 3543 rc = proto_register(&smc_proto6, 1); 3544 if (rc) { 3545 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3546 goto out_proto; 3547 } 3548 3549 rc = sock_register(&smc_sock_family_ops); 3550 if (rc) { 3551 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3552 goto out_proto6; 3553 } 3554 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3555 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3556 3557 rc = smc_ib_register_client(); 3558 if (rc) { 3559 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3560 goto out_sock; 3561 } 3562 3563 rc = smc_loopback_init(); 3564 if (rc) { 3565 pr_err("%s: smc_loopback_init fails with %d\n", __func__, rc); 3566 goto out_ib; 3567 } 3568 3569 rc = tcp_register_ulp(&smc_ulp_ops); 3570 if (rc) { 3571 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3572 goto out_lo; 3573 } 3574 rc = smc_inet_init(); 3575 if (rc) { 3576 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3577 goto out_ulp; 3578 } 3579 static_branch_enable(&tcp_have_smc); 3580 return 0; 3581 out_ulp: 3582 tcp_unregister_ulp(&smc_ulp_ops); 3583 out_lo: 3584 smc_loopback_exit(); 3585 out_ib: 3586 smc_ib_unregister_client(); 3587 out_sock: 3588 sock_unregister(PF_SMC); 3589 out_proto6: 3590 proto_unregister(&smc_proto6); 3591 out_proto: 3592 proto_unregister(&smc_proto); 3593 out_core: 3594 smc_core_exit(); 3595 out_alloc_wqs: 3596 destroy_workqueue(smc_close_wq); 3597 out_alloc_hs_wq: 3598 destroy_workqueue(smc_hs_wq); 3599 out_alloc_tcp_ls_wq: 3600 destroy_workqueue(smc_tcp_ls_wq); 3601 out_pnet: 3602 smc_pnet_exit(); 3603 out_nl: 3604 smc_nl_exit(); 3605 out_ism: 3606 smc_clc_exit(); 3607 smc_ism_exit(); 3608 out_pernet_subsys_stat: 3609 unregister_pernet_subsys(&smc_net_stat_ops); 3610 out_pernet_subsys: 3611 unregister_pernet_subsys(&smc_net_ops); 3612 3613 return rc; 3614 } 3615 3616 static void __exit smc_exit(void) 3617 { 3618 static_branch_disable(&tcp_have_smc); 3619 smc_inet_exit(); 3620 tcp_unregister_ulp(&smc_ulp_ops); 3621 sock_unregister(PF_SMC); 3622 smc_core_exit(); 3623 smc_loopback_exit(); 3624 smc_ib_unregister_client(); 3625 smc_ism_exit(); 3626 destroy_workqueue(smc_close_wq); 3627 destroy_workqueue(smc_tcp_ls_wq); 3628 destroy_workqueue(smc_hs_wq); 3629 proto_unregister(&smc_proto6); 3630 proto_unregister(&smc_proto); 3631 smc_pnet_exit(); 3632 smc_nl_exit(); 3633 smc_clc_exit(); 3634 unregister_pernet_subsys(&smc_net_stat_ops); 3635 unregister_pernet_subsys(&smc_net_ops); 3636 rcu_barrier(); 3637 } 3638 3639 module_init(smc_init); 3640 module_exit(smc_exit); 3641 3642 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3643 MODULE_DESCRIPTION("smc socket address family"); 3644 MODULE_LICENSE("GPL"); 3645 MODULE_ALIAS_NETPROTO(PF_SMC); 3646 MODULE_ALIAS_TCP_ULP("smc"); 3647 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3648 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3649 #if IS_ENABLED(CONFIG_IPV6) 3650 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3651 #endif /* CONFIG_IPV6 */ 3652 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3653