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 #if IS_ENABLED(CONFIG_IPV6) 1121 (smc->clcsock->sk->sk_family == AF_INET6 && 1122 !ipv6_addr_v4mapped(&smc->clcsock->sk->sk_v6_rcv_saddr)) || 1123 #endif 1124 !smc_clc_ueid_count() || 1125 smc_find_rdma_device(smc, ini)) 1126 ini->smcr_version &= ~SMC_V2; 1127 ini->check_smcrv2 = false; 1128 1129 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1130 ini->smcr_version & SMC_V2); 1131 1132 /* if neither ISM nor RDMA are supported, fallback */ 1133 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1134 rc = SMC_CLC_DECL_NOSMCDEV; 1135 1136 return rc; 1137 } 1138 1139 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1140 * used, the VLAN ID will be registered again during the connection setup. 1141 */ 1142 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1143 struct smc_init_info *ini) 1144 { 1145 if (!smcd_indicated(ini->smc_type_v1)) 1146 return 0; 1147 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1148 return SMC_CLC_DECL_CNFERR; 1149 return 0; 1150 } 1151 1152 #define SMC_CLC_MAX_ACCEPT_LEN \ 1153 (sizeof(struct smc_clc_msg_accept_confirm) + \ 1154 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1155 sizeof(struct smc_clc_msg_trail)) 1156 1157 /* CLC handshake during connect */ 1158 static int smc_connect_clc(struct smc_sock *smc, 1159 struct smc_clc_msg_accept_confirm *aclc, 1160 struct smc_init_info *ini) 1161 { 1162 int rc = 0; 1163 1164 /* do inband token exchange */ 1165 rc = smc_clc_send_proposal(smc, ini); 1166 if (rc) 1167 return rc; 1168 /* receive SMC Accept CLC message */ 1169 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN, 1170 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1171 } 1172 1173 void smc_fill_gid_list(struct smc_link_group *lgr, 1174 struct smc_gidlist *gidlist, 1175 struct smc_ib_device *known_dev, u8 *known_gid) 1176 { 1177 struct smc_init_info *alt_ini = NULL; 1178 1179 memset(gidlist, 0, sizeof(*gidlist)); 1180 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1181 1182 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1183 if (!alt_ini) 1184 goto out; 1185 1186 alt_ini->vlan_id = lgr->vlan_id; 1187 alt_ini->check_smcrv2 = true; 1188 alt_ini->smcrv2.saddr = lgr->saddr; 1189 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1190 1191 if (!alt_ini->smcrv2.ib_dev_v2) 1192 goto out; 1193 1194 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1195 SMC_GID_SIZE); 1196 1197 out: 1198 kfree(alt_ini); 1199 } 1200 1201 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1202 struct smc_clc_msg_accept_confirm *aclc, 1203 struct smc_init_info *ini) 1204 { 1205 struct smc_clc_first_contact_ext *fce = 1206 smc_get_clc_first_contact_ext(aclc, false); 1207 struct net *net = sock_net(&smc->sk); 1208 int rc; 1209 1210 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1211 return 0; 1212 1213 if (fce->v2_direct) { 1214 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1215 ini->smcrv2.uses_gateway = false; 1216 } else { 1217 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1218 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1219 ini->smcrv2.nexthop_mac, 1220 &ini->smcrv2.uses_gateway)) 1221 return SMC_CLC_DECL_NOROUTE; 1222 if (!ini->smcrv2.uses_gateway) { 1223 /* mismatch: peer claims indirect, but its direct */ 1224 return SMC_CLC_DECL_NOINDIRECT; 1225 } 1226 } 1227 1228 ini->release_nr = fce->release; 1229 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1230 if (rc) 1231 return rc; 1232 1233 return 0; 1234 } 1235 1236 /* setup for RDMA connection of client */ 1237 static int smc_connect_rdma(struct smc_sock *smc, 1238 struct smc_clc_msg_accept_confirm *aclc, 1239 struct smc_init_info *ini) 1240 { 1241 int i, reason_code = 0; 1242 struct smc_link *link; 1243 u8 *eid = NULL; 1244 1245 ini->is_smcd = false; 1246 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1247 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1248 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1249 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1250 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1251 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1252 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1253 1254 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1255 if (reason_code) 1256 return reason_code; 1257 1258 mutex_lock(&smc_client_lgr_pending); 1259 reason_code = smc_conn_create(smc, ini); 1260 if (reason_code) { 1261 mutex_unlock(&smc_client_lgr_pending); 1262 return reason_code; 1263 } 1264 1265 smc_conn_save_peer_info(smc, aclc); 1266 1267 if (ini->first_contact_local) { 1268 link = smc->conn.lnk; 1269 } else { 1270 /* set link that was assigned by server */ 1271 link = NULL; 1272 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1273 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1274 1275 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1276 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1277 SMC_GID_SIZE) && 1278 (aclc->hdr.version > SMC_V1 || 1279 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1280 sizeof(l->peer_mac)))) { 1281 link = l; 1282 break; 1283 } 1284 } 1285 if (!link) { 1286 reason_code = SMC_CLC_DECL_NOSRVLINK; 1287 goto connect_abort; 1288 } 1289 smc_switch_link_and_count(&smc->conn, link); 1290 } 1291 1292 /* create send buffer and rmb */ 1293 if (smc_buf_create(smc, false)) { 1294 reason_code = SMC_CLC_DECL_MEM; 1295 goto connect_abort; 1296 } 1297 1298 if (ini->first_contact_local) 1299 smc_link_save_peer_info(link, aclc, ini); 1300 1301 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1302 reason_code = SMC_CLC_DECL_ERR_RTOK; 1303 goto connect_abort; 1304 } 1305 1306 smc_rx_init(smc); 1307 1308 if (ini->first_contact_local) { 1309 if (smc_ib_ready_link(link)) { 1310 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1311 goto connect_abort; 1312 } 1313 } else { 1314 /* reg sendbufs if they were vzalloced */ 1315 if (smc->conn.sndbuf_desc->is_vm) { 1316 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1317 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1318 goto connect_abort; 1319 } 1320 } 1321 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1322 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1323 goto connect_abort; 1324 } 1325 } 1326 1327 if (aclc->hdr.version > SMC_V1) { 1328 eid = aclc->r1.eid; 1329 if (ini->first_contact_local) 1330 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1331 link->smcibdev, link->gid); 1332 } 1333 1334 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1335 aclc->hdr.version, eid, ini); 1336 if (reason_code) 1337 goto connect_abort; 1338 1339 smc_tx_init(smc); 1340 1341 if (ini->first_contact_local) { 1342 /* QP confirmation over RoCE fabric */ 1343 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1344 reason_code = smcr_clnt_conf_first_link(smc); 1345 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1346 if (reason_code) 1347 goto connect_abort; 1348 } 1349 mutex_unlock(&smc_client_lgr_pending); 1350 1351 smc_copy_sock_settings_to_clc(smc); 1352 smc->connect_nonblock = 0; 1353 if (smc->sk.sk_state == SMC_INIT) 1354 smc->sk.sk_state = SMC_ACTIVE; 1355 1356 return 0; 1357 connect_abort: 1358 smc_conn_abort(smc, ini->first_contact_local); 1359 mutex_unlock(&smc_client_lgr_pending); 1360 smc->connect_nonblock = 0; 1361 1362 return reason_code; 1363 } 1364 1365 /* The server has chosen one of the proposed ISM devices for the communication. 1366 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1367 */ 1368 static int 1369 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc, 1370 struct smc_init_info *ini) 1371 { 1372 int i; 1373 1374 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1375 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1376 ini->ism_selected = i; 1377 return 0; 1378 } 1379 } 1380 1381 return -EPROTO; 1382 } 1383 1384 /* setup for ISM connection of client */ 1385 static int smc_connect_ism(struct smc_sock *smc, 1386 struct smc_clc_msg_accept_confirm *aclc, 1387 struct smc_init_info *ini) 1388 { 1389 u8 *eid = NULL; 1390 int rc = 0; 1391 1392 ini->is_smcd = true; 1393 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1394 1395 if (aclc->hdr.version == SMC_V2) { 1396 if (ini->first_contact_peer) { 1397 struct smc_clc_first_contact_ext *fce = 1398 smc_get_clc_first_contact_ext(aclc, true); 1399 1400 ini->release_nr = fce->release; 1401 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1402 if (rc) 1403 return rc; 1404 } 1405 1406 rc = smc_v2_determine_accepted_chid(aclc, ini); 1407 if (rc) 1408 return rc; 1409 1410 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected])) 1411 ini->ism_peer_gid[ini->ism_selected].gid_ext = 1412 ntohll(aclc->d1.gid_ext); 1413 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */ 1414 } 1415 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid); 1416 1417 /* there is only one lgr role for SMC-D; use server lock */ 1418 mutex_lock(&smc_server_lgr_pending); 1419 rc = smc_conn_create(smc, ini); 1420 if (rc) { 1421 mutex_unlock(&smc_server_lgr_pending); 1422 return rc; 1423 } 1424 1425 /* Create send and receive buffers */ 1426 rc = smc_buf_create(smc, true); 1427 if (rc) { 1428 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1429 goto connect_abort; 1430 } 1431 1432 smc_conn_save_peer_info(smc, aclc); 1433 1434 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) { 1435 rc = smcd_buf_attach(smc); 1436 if (rc) { 1437 rc = SMC_CLC_DECL_MEM; /* try to fallback */ 1438 goto connect_abort; 1439 } 1440 } 1441 smc_rx_init(smc); 1442 smc_tx_init(smc); 1443 1444 if (aclc->hdr.version > SMC_V1) 1445 eid = aclc->d1.eid; 1446 1447 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1448 aclc->hdr.version, eid, ini); 1449 if (rc) 1450 goto connect_abort; 1451 mutex_unlock(&smc_server_lgr_pending); 1452 1453 smc_copy_sock_settings_to_clc(smc); 1454 smc->connect_nonblock = 0; 1455 if (smc->sk.sk_state == SMC_INIT) 1456 smc->sk.sk_state = SMC_ACTIVE; 1457 1458 return 0; 1459 connect_abort: 1460 smc_conn_abort(smc, ini->first_contact_local); 1461 mutex_unlock(&smc_server_lgr_pending); 1462 smc->connect_nonblock = 0; 1463 1464 return rc; 1465 } 1466 1467 /* check if received accept type and version matches a proposed one */ 1468 static int smc_connect_check_aclc(struct smc_init_info *ini, 1469 struct smc_clc_msg_accept_confirm *aclc) 1470 { 1471 if (aclc->hdr.version >= SMC_V2) { 1472 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1473 !smcr_indicated(ini->smc_type_v2)) || 1474 (aclc->hdr.typev1 == SMC_TYPE_D && 1475 !smcd_indicated(ini->smc_type_v2))) 1476 return SMC_CLC_DECL_MODEUNSUPP; 1477 } else { 1478 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1479 !smcr_indicated(ini->smc_type_v1)) || 1480 (aclc->hdr.typev1 == SMC_TYPE_D && 1481 !smcd_indicated(ini->smc_type_v1))) 1482 return SMC_CLC_DECL_MODEUNSUPP; 1483 } 1484 1485 return 0; 1486 } 1487 1488 /* perform steps before actually connecting */ 1489 static int __smc_connect(struct smc_sock *smc) 1490 { 1491 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1492 struct smc_clc_msg_accept_confirm *aclc; 1493 struct smc_init_info *ini = NULL; 1494 u8 *buf = NULL; 1495 int rc = 0; 1496 1497 if (smc->use_fallback) 1498 return smc_connect_fallback(smc, smc->fallback_rsn); 1499 1500 /* if peer has not signalled SMC-capability, fall back */ 1501 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1502 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1503 1504 /* IPSec connections opt out of SMC optimizations */ 1505 if (using_ipsec(smc)) 1506 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1507 version); 1508 1509 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1510 if (!ini) 1511 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1512 version); 1513 1514 ini->smcd_version = SMC_V1 | SMC_V2; 1515 ini->smcr_version = SMC_V1 | SMC_V2; 1516 ini->smc_type_v1 = SMC_TYPE_B; 1517 ini->smc_type_v2 = SMC_TYPE_B; 1518 1519 /* get vlan id from IP device */ 1520 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1521 ini->smcd_version &= ~SMC_V1; 1522 ini->smcr_version = 0; 1523 ini->smc_type_v1 = SMC_TYPE_N; 1524 } 1525 1526 rc = smc_find_proposal_devices(smc, ini); 1527 if (rc) 1528 goto fallback; 1529 1530 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1531 if (!buf) { 1532 rc = SMC_CLC_DECL_MEM; 1533 goto fallback; 1534 } 1535 aclc = (struct smc_clc_msg_accept_confirm *)buf; 1536 1537 /* perform CLC handshake */ 1538 rc = smc_connect_clc(smc, aclc, ini); 1539 if (rc) { 1540 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1541 if (rc == -EAGAIN) { 1542 rc = -ETIMEDOUT; 1543 smc->sk.sk_err = ETIMEDOUT; 1544 } 1545 goto vlan_cleanup; 1546 } 1547 1548 /* check if smc modes and versions of CLC proposal and accept match */ 1549 rc = smc_connect_check_aclc(ini, aclc); 1550 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1551 if (rc) 1552 goto vlan_cleanup; 1553 1554 /* depending on previous steps, connect using rdma or ism */ 1555 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1556 ini->smcr_version = version; 1557 rc = smc_connect_rdma(smc, aclc, ini); 1558 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1559 ini->smcd_version = version; 1560 rc = smc_connect_ism(smc, aclc, ini); 1561 } 1562 if (rc) 1563 goto vlan_cleanup; 1564 1565 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1566 smc_connect_ism_vlan_cleanup(smc, ini); 1567 kfree(buf); 1568 kfree(ini); 1569 return 0; 1570 1571 vlan_cleanup: 1572 smc_connect_ism_vlan_cleanup(smc, ini); 1573 kfree(buf); 1574 fallback: 1575 kfree(ini); 1576 return smc_connect_decline_fallback(smc, rc, version); 1577 } 1578 1579 static void smc_connect_work(struct work_struct *work) 1580 { 1581 struct smc_sock *smc = container_of(work, struct smc_sock, 1582 connect_work); 1583 long timeo = smc->sk.sk_sndtimeo; 1584 int rc = 0; 1585 1586 if (!timeo) 1587 timeo = MAX_SCHEDULE_TIMEOUT; 1588 lock_sock(smc->clcsock->sk); 1589 if (smc->clcsock->sk->sk_err) { 1590 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1591 } else if ((1 << smc->clcsock->sk->sk_state) & 1592 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1593 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1594 if ((rc == -EPIPE) && 1595 ((1 << smc->clcsock->sk->sk_state) & 1596 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1597 rc = 0; 1598 } 1599 release_sock(smc->clcsock->sk); 1600 lock_sock(&smc->sk); 1601 if (rc != 0 || smc->sk.sk_err) { 1602 smc->sk.sk_state = SMC_CLOSED; 1603 if (rc == -EPIPE || rc == -EAGAIN) 1604 smc->sk.sk_err = EPIPE; 1605 else if (rc == -ECONNREFUSED) 1606 smc->sk.sk_err = ECONNREFUSED; 1607 else if (signal_pending(current)) 1608 smc->sk.sk_err = -sock_intr_errno(timeo); 1609 sock_put(&smc->sk); /* passive closing */ 1610 goto out; 1611 } 1612 1613 rc = __smc_connect(smc); 1614 if (rc < 0) 1615 smc->sk.sk_err = -rc; 1616 1617 out: 1618 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1619 if (smc->sk.sk_err) { 1620 smc->sk.sk_state_change(&smc->sk); 1621 } else { /* allow polling before and after fallback decision */ 1622 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1623 smc->sk.sk_write_space(&smc->sk); 1624 } 1625 } 1626 release_sock(&smc->sk); 1627 } 1628 1629 int smc_connect(struct socket *sock, struct sockaddr *addr, 1630 int alen, int flags) 1631 { 1632 struct sock *sk = sock->sk; 1633 struct smc_sock *smc; 1634 int rc = -EINVAL; 1635 1636 smc = smc_sk(sk); 1637 1638 /* separate smc parameter checking to be safe */ 1639 if (alen < sizeof(addr->sa_family)) 1640 goto out_err; 1641 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1642 goto out_err; 1643 1644 lock_sock(sk); 1645 switch (sock->state) { 1646 default: 1647 rc = -EINVAL; 1648 goto out; 1649 case SS_CONNECTED: 1650 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1651 goto out; 1652 case SS_CONNECTING: 1653 if (sk->sk_state == SMC_ACTIVE) 1654 goto connected; 1655 break; 1656 case SS_UNCONNECTED: 1657 sock->state = SS_CONNECTING; 1658 break; 1659 } 1660 1661 switch (sk->sk_state) { 1662 default: 1663 goto out; 1664 case SMC_CLOSED: 1665 rc = sock_error(sk) ? : -ECONNABORTED; 1666 sock->state = SS_UNCONNECTED; 1667 goto out; 1668 case SMC_ACTIVE: 1669 rc = -EISCONN; 1670 goto out; 1671 case SMC_INIT: 1672 break; 1673 } 1674 1675 smc_copy_sock_settings_to_clc(smc); 1676 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1677 if (smc->connect_nonblock) { 1678 rc = -EALREADY; 1679 goto out; 1680 } 1681 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1682 if (rc && rc != -EINPROGRESS) 1683 goto out; 1684 1685 if (smc->use_fallback) { 1686 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1687 goto out; 1688 } 1689 sock_hold(&smc->sk); /* sock put in passive closing */ 1690 if (flags & O_NONBLOCK) { 1691 if (queue_work(smc_hs_wq, &smc->connect_work)) 1692 smc->connect_nonblock = 1; 1693 rc = -EINPROGRESS; 1694 goto out; 1695 } else { 1696 rc = __smc_connect(smc); 1697 if (rc < 0) 1698 goto out; 1699 } 1700 1701 connected: 1702 rc = 0; 1703 sock->state = SS_CONNECTED; 1704 out: 1705 release_sock(sk); 1706 out_err: 1707 return rc; 1708 } 1709 1710 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1711 { 1712 struct socket *new_clcsock = NULL; 1713 struct sock *lsk = &lsmc->sk; 1714 struct sock *new_sk; 1715 int rc = -EINVAL; 1716 1717 release_sock(lsk); 1718 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1719 if (!new_sk) { 1720 rc = -ENOMEM; 1721 lsk->sk_err = ENOMEM; 1722 *new_smc = NULL; 1723 lock_sock(lsk); 1724 goto out; 1725 } 1726 *new_smc = smc_sk(new_sk); 1727 1728 mutex_lock(&lsmc->clcsock_release_lock); 1729 if (lsmc->clcsock) 1730 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1731 mutex_unlock(&lsmc->clcsock_release_lock); 1732 lock_sock(lsk); 1733 if (rc < 0 && rc != -EAGAIN) 1734 lsk->sk_err = -rc; 1735 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1736 new_sk->sk_prot->unhash(new_sk); 1737 if (new_clcsock) 1738 sock_release(new_clcsock); 1739 new_sk->sk_state = SMC_CLOSED; 1740 smc_sock_set_flag(new_sk, SOCK_DEAD); 1741 sock_put(new_sk); /* final */ 1742 *new_smc = NULL; 1743 goto out; 1744 } 1745 1746 /* new clcsock has inherited the smc listen-specific sk_data_ready 1747 * function; switch it back to the original sk_data_ready function 1748 */ 1749 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1750 1751 /* if new clcsock has also inherited the fallback-specific callback 1752 * functions, switch them back to the original ones. 1753 */ 1754 if (lsmc->use_fallback) { 1755 if (lsmc->clcsk_state_change) 1756 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1757 if (lsmc->clcsk_write_space) 1758 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1759 if (lsmc->clcsk_error_report) 1760 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1761 } 1762 1763 (*new_smc)->clcsock = new_clcsock; 1764 out: 1765 return rc; 1766 } 1767 1768 /* add a just created sock to the accept queue of the listen sock as 1769 * candidate for a following socket accept call from user space 1770 */ 1771 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1772 { 1773 struct smc_sock *par = smc_sk(parent); 1774 1775 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1776 spin_lock(&par->accept_q_lock); 1777 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1778 spin_unlock(&par->accept_q_lock); 1779 sk_acceptq_added(parent); 1780 } 1781 1782 /* remove a socket from the accept queue of its parental listening socket */ 1783 static void smc_accept_unlink(struct sock *sk) 1784 { 1785 struct smc_sock *par = smc_sk(sk)->listen_smc; 1786 1787 spin_lock(&par->accept_q_lock); 1788 list_del_init(&smc_sk(sk)->accept_q); 1789 spin_unlock(&par->accept_q_lock); 1790 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1791 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1792 } 1793 1794 /* remove a sock from the accept queue to bind it to a new socket created 1795 * for a socket accept call from user space 1796 */ 1797 struct sock *smc_accept_dequeue(struct sock *parent, 1798 struct socket *new_sock) 1799 { 1800 struct smc_sock *isk, *n; 1801 struct sock *new_sk; 1802 1803 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1804 new_sk = (struct sock *)isk; 1805 1806 smc_accept_unlink(new_sk); 1807 if (new_sk->sk_state == SMC_CLOSED) { 1808 new_sk->sk_prot->unhash(new_sk); 1809 if (isk->clcsock) { 1810 sock_release(isk->clcsock); 1811 isk->clcsock = NULL; 1812 } 1813 sock_put(new_sk); /* final */ 1814 continue; 1815 } 1816 if (new_sock) { 1817 sock_graft(new_sk, new_sock); 1818 new_sock->state = SS_CONNECTED; 1819 if (isk->use_fallback) { 1820 smc_sk(new_sk)->clcsock->file = new_sock->file; 1821 isk->clcsock->file->private_data = isk->clcsock; 1822 } 1823 } 1824 return new_sk; 1825 } 1826 return NULL; 1827 } 1828 1829 /* clean up for a created but never accepted sock */ 1830 void smc_close_non_accepted(struct sock *sk) 1831 { 1832 struct smc_sock *smc = smc_sk(sk); 1833 1834 sock_hold(sk); /* sock_put below */ 1835 lock_sock(sk); 1836 if (!sk->sk_lingertime) 1837 /* wait for peer closing */ 1838 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1839 __smc_release(smc); 1840 release_sock(sk); 1841 sock_put(sk); /* sock_hold above */ 1842 sock_put(sk); /* final sock_put */ 1843 } 1844 1845 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1846 { 1847 struct smc_link *link = smc->conn.lnk; 1848 struct smc_llc_qentry *qentry; 1849 int rc; 1850 1851 /* reg the sndbuf if it was vzalloced*/ 1852 if (smc->conn.sndbuf_desc->is_vm) { 1853 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1854 return SMC_CLC_DECL_ERR_REGBUF; 1855 } 1856 1857 /* reg the rmb */ 1858 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1859 return SMC_CLC_DECL_ERR_REGBUF; 1860 1861 /* send CONFIRM LINK request to client over the RoCE fabric */ 1862 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1863 if (rc < 0) 1864 return SMC_CLC_DECL_TIMEOUT_CL; 1865 1866 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1867 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1868 SMC_LLC_CONFIRM_LINK); 1869 if (!qentry) { 1870 struct smc_clc_msg_decline dclc; 1871 1872 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1873 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1874 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1875 } 1876 smc_llc_save_peer_uid(qentry); 1877 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1878 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1879 if (rc) 1880 return SMC_CLC_DECL_RMBE_EC; 1881 1882 /* confirm_rkey is implicit on 1st contact */ 1883 smc->conn.rmb_desc->is_conf_rkey = true; 1884 1885 smc_llc_link_active(link); 1886 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1887 1888 if (link->lgr->max_links > 1) { 1889 down_write(&link->lgr->llc_conf_mutex); 1890 /* initial contact - try to establish second link */ 1891 smc_llc_srv_add_link(link, NULL); 1892 up_write(&link->lgr->llc_conf_mutex); 1893 } 1894 return 0; 1895 } 1896 1897 /* listen worker: finish */ 1898 static void smc_listen_out(struct smc_sock *new_smc) 1899 { 1900 struct smc_sock *lsmc = new_smc->listen_smc; 1901 struct sock *newsmcsk = &new_smc->sk; 1902 1903 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1904 atomic_dec(&lsmc->queued_smc_hs); 1905 1906 release_sock(newsmcsk); /* lock in smc_listen_work() */ 1907 if (lsmc->sk.sk_state == SMC_LISTEN) { 1908 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1909 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1910 release_sock(&lsmc->sk); 1911 } else { /* no longer listening */ 1912 smc_close_non_accepted(newsmcsk); 1913 } 1914 1915 /* Wake up accept */ 1916 lsmc->sk.sk_data_ready(&lsmc->sk); 1917 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1918 } 1919 1920 /* listen worker: finish in state connected */ 1921 static void smc_listen_out_connected(struct smc_sock *new_smc) 1922 { 1923 struct sock *newsmcsk = &new_smc->sk; 1924 1925 if (newsmcsk->sk_state == SMC_INIT) 1926 newsmcsk->sk_state = SMC_ACTIVE; 1927 1928 smc_listen_out(new_smc); 1929 } 1930 1931 /* listen worker: finish in error state */ 1932 static void smc_listen_out_err(struct smc_sock *new_smc) 1933 { 1934 struct sock *newsmcsk = &new_smc->sk; 1935 struct net *net = sock_net(newsmcsk); 1936 1937 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1938 if (newsmcsk->sk_state == SMC_INIT) 1939 sock_put(&new_smc->sk); /* passive closing */ 1940 newsmcsk->sk_state = SMC_CLOSED; 1941 1942 smc_listen_out(new_smc); 1943 } 1944 1945 /* listen worker: decline and fall back if possible */ 1946 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1947 int local_first, u8 version) 1948 { 1949 /* RDMA setup failed, switch back to TCP */ 1950 smc_conn_abort(new_smc, local_first); 1951 if (reason_code < 0 || 1952 smc_switch_to_fallback(new_smc, reason_code)) { 1953 /* error, no fallback possible */ 1954 smc_listen_out_err(new_smc); 1955 return; 1956 } 1957 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1958 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1959 smc_listen_out_err(new_smc); 1960 return; 1961 } 1962 } 1963 smc_listen_out_connected(new_smc); 1964 } 1965 1966 /* listen worker: version checking */ 1967 static int smc_listen_v2_check(struct smc_sock *new_smc, 1968 struct smc_clc_msg_proposal *pclc, 1969 struct smc_init_info *ini) 1970 { 1971 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1972 struct smc_clc_v2_extension *pclc_v2_ext; 1973 int rc = SMC_CLC_DECL_PEERNOSMC; 1974 1975 ini->smc_type_v1 = pclc->hdr.typev1; 1976 ini->smc_type_v2 = pclc->hdr.typev2; 1977 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1978 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1979 if (pclc->hdr.version > SMC_V1) { 1980 if (smcd_indicated(ini->smc_type_v2)) 1981 ini->smcd_version |= SMC_V2; 1982 if (smcr_indicated(ini->smc_type_v2)) 1983 ini->smcr_version |= SMC_V2; 1984 } 1985 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1986 rc = SMC_CLC_DECL_PEERNOSMC; 1987 goto out; 1988 } 1989 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1990 if (!pclc_v2_ext) { 1991 ini->smcd_version &= ~SMC_V2; 1992 ini->smcr_version &= ~SMC_V2; 1993 rc = SMC_CLC_DECL_NOV2EXT; 1994 goto out; 1995 } 1996 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1997 if (ini->smcd_version & SMC_V2) { 1998 if (!smc_ism_is_v2_capable()) { 1999 ini->smcd_version &= ~SMC_V2; 2000 rc = SMC_CLC_DECL_NOISM2SUPP; 2001 } else if (!pclc_smcd_v2_ext) { 2002 ini->smcd_version &= ~SMC_V2; 2003 rc = SMC_CLC_DECL_NOV2DEXT; 2004 } else if (!pclc_v2_ext->hdr.eid_cnt && 2005 !pclc_v2_ext->hdr.flag.seid) { 2006 ini->smcd_version &= ~SMC_V2; 2007 rc = SMC_CLC_DECL_NOUEID; 2008 } 2009 } 2010 if (ini->smcr_version & SMC_V2) { 2011 if (!pclc_v2_ext->hdr.eid_cnt) { 2012 ini->smcr_version &= ~SMC_V2; 2013 rc = SMC_CLC_DECL_NOUEID; 2014 } 2015 } 2016 2017 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2018 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2019 ini->release_nr = SMC_RELEASE; 2020 2021 out: 2022 if (!ini->smcd_version && !ini->smcr_version) 2023 return rc; 2024 2025 return 0; 2026 } 2027 2028 /* listen worker: check prefixes */ 2029 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2030 struct smc_clc_msg_proposal *pclc) 2031 { 2032 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2033 struct socket *newclcsock = new_smc->clcsock; 2034 2035 if (pclc->hdr.typev1 == SMC_TYPE_N) 2036 return 0; 2037 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2038 if (!pclc_prfx) 2039 return -EPROTO; 2040 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2041 return SMC_CLC_DECL_DIFFPREFIX; 2042 2043 return 0; 2044 } 2045 2046 /* listen worker: initialize connection and buffers */ 2047 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2048 struct smc_init_info *ini) 2049 { 2050 int rc; 2051 2052 /* allocate connection / link group */ 2053 rc = smc_conn_create(new_smc, ini); 2054 if (rc) 2055 return rc; 2056 2057 /* create send buffer and rmb */ 2058 if (smc_buf_create(new_smc, false)) { 2059 smc_conn_abort(new_smc, ini->first_contact_local); 2060 return SMC_CLC_DECL_MEM; 2061 } 2062 2063 return 0; 2064 } 2065 2066 /* listen worker: initialize connection and buffers for SMC-D */ 2067 static int smc_listen_ism_init(struct smc_sock *new_smc, 2068 struct smc_init_info *ini) 2069 { 2070 int rc; 2071 2072 rc = smc_conn_create(new_smc, ini); 2073 if (rc) 2074 return rc; 2075 2076 /* Create send and receive buffers */ 2077 rc = smc_buf_create(new_smc, true); 2078 if (rc) { 2079 smc_conn_abort(new_smc, ini->first_contact_local); 2080 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2081 SMC_CLC_DECL_MEM; 2082 } 2083 2084 return 0; 2085 } 2086 2087 static bool smc_is_already_selected(struct smcd_dev *smcd, 2088 struct smc_init_info *ini, 2089 int matches) 2090 { 2091 int i; 2092 2093 for (i = 0; i < matches; i++) 2094 if (smcd == ini->ism_dev[i]) 2095 return true; 2096 2097 return false; 2098 } 2099 2100 /* check for ISM devices matching proposed ISM devices */ 2101 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2102 u16 proposed_chid, 2103 struct smcd_gid *proposed_gid, 2104 unsigned int *matches) 2105 { 2106 struct smcd_dev *smcd; 2107 2108 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2109 if (smcd->going_away) 2110 continue; 2111 if (smc_is_already_selected(smcd, ini, *matches)) 2112 continue; 2113 if (smc_ism_get_chid(smcd) == proposed_chid && 2114 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2115 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2116 if (__smc_ism_is_emulated(proposed_chid)) 2117 ini->ism_peer_gid[*matches].gid_ext = 2118 proposed_gid->gid_ext; 2119 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2120 ini->ism_dev[*matches] = smcd; 2121 (*matches)++; 2122 break; 2123 } 2124 } 2125 } 2126 2127 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2128 { 2129 if (!ini->rc) 2130 ini->rc = rc; 2131 } 2132 2133 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2134 struct smc_clc_msg_proposal *pclc, 2135 struct smc_init_info *ini) 2136 { 2137 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2138 struct smc_clc_v2_extension *smc_v2_ext; 2139 struct smc_clc_msg_smcd *pclc_smcd; 2140 unsigned int matches = 0; 2141 struct smcd_gid smcd_gid; 2142 u8 smcd_version; 2143 u8 *eid = NULL; 2144 int i, rc; 2145 u16 chid; 2146 2147 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2148 goto not_found; 2149 2150 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2151 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2152 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2153 if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext) 2154 goto not_found; 2155 2156 mutex_lock(&smcd_dev_list.mutex); 2157 if (pclc_smcd->ism.chid) { 2158 /* check for ISM device matching proposed native ISM device */ 2159 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2160 smcd_gid.gid_ext = 0; 2161 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2162 &smcd_gid, &matches); 2163 } 2164 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2165 /* check for ISM devices matching proposed non-native ISM 2166 * devices 2167 */ 2168 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2169 smcd_gid.gid_ext = 0; 2170 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2171 if (__smc_ism_is_emulated(chid)) { 2172 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2173 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2174 /* each Emulated-ISM device takes two GID-CHID 2175 * entries and CHID of the second entry repeats 2176 * that of the first entry. 2177 * 2178 * So check if the next GID-CHID entry exists 2179 * and both two entries' CHIDs are the same. 2180 */ 2181 continue; 2182 smcd_gid.gid_ext = 2183 ntohll(smcd_v2_ext->gidchid[++i].gid); 2184 } 2185 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2186 } 2187 mutex_unlock(&smcd_dev_list.mutex); 2188 2189 if (!ini->ism_dev[0]) { 2190 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2191 goto not_found; 2192 } 2193 2194 smc_ism_get_system_eid(&eid); 2195 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2196 smcd_v2_ext->system_eid, eid)) 2197 goto not_found; 2198 2199 /* separate - outside the smcd_dev_list.lock */ 2200 smcd_version = ini->smcd_version; 2201 for (i = 0; i < matches; i++) { 2202 ini->smcd_version = SMC_V2; 2203 ini->is_smcd = true; 2204 ini->ism_selected = i; 2205 rc = smc_listen_ism_init(new_smc, ini); 2206 if (rc) { 2207 smc_find_ism_store_rc(rc, ini); 2208 /* try next active ISM device */ 2209 continue; 2210 } 2211 return; /* matching and usable V2 ISM device found */ 2212 } 2213 /* no V2 ISM device could be initialized */ 2214 ini->smcd_version = smcd_version; /* restore original value */ 2215 ini->negotiated_eid[0] = 0; 2216 2217 not_found: 2218 ini->smcd_version &= ~SMC_V2; 2219 ini->ism_dev[0] = NULL; 2220 ini->is_smcd = false; 2221 } 2222 2223 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2224 struct smc_clc_msg_proposal *pclc, 2225 struct smc_init_info *ini) 2226 { 2227 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2228 int rc = 0; 2229 2230 /* check if ISM V1 is available */ 2231 if (!(ini->smcd_version & SMC_V1) || 2232 !smcd_indicated(ini->smc_type_v1) || 2233 !pclc_smcd) 2234 goto not_found; 2235 ini->is_smcd = true; /* prepare ISM check */ 2236 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2237 ini->ism_peer_gid[0].gid_ext = 0; 2238 rc = smc_find_ism_device(new_smc, ini); 2239 if (rc) 2240 goto not_found; 2241 ini->ism_selected = 0; 2242 rc = smc_listen_ism_init(new_smc, ini); 2243 if (!rc) 2244 return; /* V1 ISM device found */ 2245 2246 not_found: 2247 smc_find_ism_store_rc(rc, ini); 2248 ini->smcd_version &= ~SMC_V1; 2249 ini->ism_dev[0] = NULL; 2250 ini->is_smcd = false; 2251 } 2252 2253 /* listen worker: register buffers */ 2254 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2255 { 2256 struct smc_connection *conn = &new_smc->conn; 2257 2258 if (!local_first) { 2259 /* reg sendbufs if they were vzalloced */ 2260 if (conn->sndbuf_desc->is_vm) { 2261 if (smcr_lgr_reg_sndbufs(conn->lnk, 2262 conn->sndbuf_desc)) 2263 return SMC_CLC_DECL_ERR_REGBUF; 2264 } 2265 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2266 return SMC_CLC_DECL_ERR_REGBUF; 2267 } 2268 2269 return 0; 2270 } 2271 2272 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2273 struct smc_clc_msg_proposal *pclc, 2274 struct smc_init_info *ini) 2275 { 2276 struct smc_clc_v2_extension *smc_v2_ext; 2277 u8 smcr_version; 2278 int rc; 2279 2280 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2281 goto not_found; 2282 2283 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2284 if (!smc_v2_ext || 2285 !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2286 goto not_found; 2287 2288 /* prepare RDMA check */ 2289 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2290 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2291 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2292 ini->check_smcrv2 = true; 2293 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2294 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2295 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2296 rc = smc_find_rdma_device(new_smc, ini); 2297 if (rc) { 2298 smc_find_ism_store_rc(rc, ini); 2299 goto not_found; 2300 } 2301 if (!ini->smcrv2.uses_gateway) 2302 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2303 2304 smcr_version = ini->smcr_version; 2305 ini->smcr_version = SMC_V2; 2306 rc = smc_listen_rdma_init(new_smc, ini); 2307 if (!rc) { 2308 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2309 if (rc) 2310 smc_conn_abort(new_smc, ini->first_contact_local); 2311 } 2312 if (!rc) 2313 return; 2314 ini->smcr_version = smcr_version; 2315 smc_find_ism_store_rc(rc, ini); 2316 2317 not_found: 2318 ini->smcr_version &= ~SMC_V2; 2319 ini->smcrv2.ib_dev_v2 = NULL; 2320 ini->check_smcrv2 = false; 2321 } 2322 2323 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2324 struct smc_clc_msg_proposal *pclc, 2325 struct smc_init_info *ini) 2326 { 2327 int rc; 2328 2329 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2330 return SMC_CLC_DECL_NOSMCDEV; 2331 2332 /* prepare RDMA check */ 2333 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2334 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2335 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2336 rc = smc_find_rdma_device(new_smc, ini); 2337 if (rc) { 2338 /* no RDMA device found */ 2339 return SMC_CLC_DECL_NOSMCDEV; 2340 } 2341 rc = smc_listen_rdma_init(new_smc, ini); 2342 if (rc) 2343 return rc; 2344 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2345 } 2346 2347 /* determine the local device matching to proposal */ 2348 static int smc_listen_find_device(struct smc_sock *new_smc, 2349 struct smc_clc_msg_proposal *pclc, 2350 struct smc_init_info *ini) 2351 { 2352 int prfx_rc; 2353 2354 /* check for ISM device matching V2 proposed device */ 2355 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2356 if (ini->ism_dev[0]) 2357 return 0; 2358 2359 /* check for matching IP prefix and subnet length (V1) */ 2360 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2361 if (prfx_rc) 2362 smc_find_ism_store_rc(prfx_rc, ini); 2363 2364 /* get vlan id from IP device */ 2365 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2366 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2367 2368 /* check for ISM device matching V1 proposed device */ 2369 if (!prfx_rc) 2370 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2371 if (ini->ism_dev[0]) 2372 return 0; 2373 2374 if (!smcr_indicated(pclc->hdr.typev1) && 2375 !smcr_indicated(pclc->hdr.typev2)) 2376 /* skip RDMA and decline */ 2377 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2378 2379 /* check if RDMA V2 is available */ 2380 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2381 if (ini->smcrv2.ib_dev_v2) 2382 return 0; 2383 2384 /* check if RDMA V1 is available */ 2385 if (!prfx_rc) { 2386 int rc; 2387 2388 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2389 smc_find_ism_store_rc(rc, ini); 2390 return (!rc) ? 0 : ini->rc; 2391 } 2392 return prfx_rc; 2393 } 2394 2395 /* listen worker: finish RDMA setup */ 2396 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2397 struct smc_clc_msg_accept_confirm *cclc, 2398 bool local_first, 2399 struct smc_init_info *ini) 2400 { 2401 struct smc_link *link = new_smc->conn.lnk; 2402 int reason_code = 0; 2403 2404 if (local_first) 2405 smc_link_save_peer_info(link, cclc, ini); 2406 2407 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2408 return SMC_CLC_DECL_ERR_RTOK; 2409 2410 if (local_first) { 2411 if (smc_ib_ready_link(link)) 2412 return SMC_CLC_DECL_ERR_RDYLNK; 2413 /* QP confirmation over RoCE fabric */ 2414 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2415 reason_code = smcr_serv_conf_first_link(new_smc); 2416 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2417 } 2418 return reason_code; 2419 } 2420 2421 /* setup for connection of server */ 2422 static void smc_listen_work(struct work_struct *work) 2423 { 2424 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2425 smc_listen_work); 2426 struct socket *newclcsock = new_smc->clcsock; 2427 struct smc_clc_msg_accept_confirm *cclc; 2428 struct smc_clc_msg_proposal_area *buf; 2429 struct smc_clc_msg_proposal *pclc; 2430 struct smc_init_info *ini = NULL; 2431 u8 proposal_version = SMC_V1; 2432 u8 accept_version; 2433 int rc = 0; 2434 2435 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2436 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2437 return smc_listen_out_err(new_smc); 2438 2439 if (new_smc->use_fallback) { 2440 smc_listen_out_connected(new_smc); 2441 return; 2442 } 2443 2444 /* check if peer is smc capable */ 2445 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2446 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2447 if (rc) 2448 smc_listen_out_err(new_smc); 2449 else 2450 smc_listen_out_connected(new_smc); 2451 return; 2452 } 2453 2454 /* do inband token exchange - 2455 * wait for and receive SMC Proposal CLC message 2456 */ 2457 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2458 if (!buf) { 2459 rc = SMC_CLC_DECL_MEM; 2460 goto out_decl; 2461 } 2462 pclc = (struct smc_clc_msg_proposal *)buf; 2463 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2464 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2465 if (rc) 2466 goto out_decl; 2467 2468 if (pclc->hdr.version > SMC_V1) 2469 proposal_version = SMC_V2; 2470 2471 /* IPSec connections opt out of SMC optimizations */ 2472 if (using_ipsec(new_smc)) { 2473 rc = SMC_CLC_DECL_IPSEC; 2474 goto out_decl; 2475 } 2476 2477 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2478 if (!ini) { 2479 rc = SMC_CLC_DECL_MEM; 2480 goto out_decl; 2481 } 2482 2483 /* initial version checking */ 2484 rc = smc_listen_v2_check(new_smc, pclc, ini); 2485 if (rc) 2486 goto out_decl; 2487 2488 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2489 if (rc) 2490 goto out_decl; 2491 2492 mutex_lock(&smc_server_lgr_pending); 2493 smc_rx_init(new_smc); 2494 smc_tx_init(new_smc); 2495 2496 /* determine ISM or RoCE device used for connection */ 2497 rc = smc_listen_find_device(new_smc, pclc, ini); 2498 if (rc) 2499 goto out_unlock; 2500 2501 /* send SMC Accept CLC message */ 2502 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2503 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2504 accept_version, ini->negotiated_eid, ini); 2505 if (rc) 2506 goto out_unlock; 2507 2508 /* SMC-D does not need this lock any more */ 2509 if (ini->is_smcd) 2510 mutex_unlock(&smc_server_lgr_pending); 2511 2512 /* receive SMC Confirm CLC message */ 2513 memset(buf, 0, sizeof(*buf)); 2514 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2515 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2516 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2517 if (rc) { 2518 if (!ini->is_smcd) 2519 goto out_unlock; 2520 goto out_decl; 2521 } 2522 2523 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2524 if (rc) { 2525 if (!ini->is_smcd) 2526 goto out_unlock; 2527 goto out_decl; 2528 } 2529 2530 /* fce smc release version is needed in smc_listen_rdma_finish, 2531 * so save fce info here. 2532 */ 2533 smc_conn_save_peer_info_fce(new_smc, cclc); 2534 2535 /* finish worker */ 2536 if (!ini->is_smcd) { 2537 rc = smc_listen_rdma_finish(new_smc, cclc, 2538 ini->first_contact_local, ini); 2539 if (rc) 2540 goto out_unlock; 2541 mutex_unlock(&smc_server_lgr_pending); 2542 } 2543 smc_conn_save_peer_info(new_smc, cclc); 2544 2545 if (ini->is_smcd && 2546 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2547 rc = smcd_buf_attach(new_smc); 2548 if (rc) 2549 goto out_decl; 2550 } 2551 2552 smc_listen_out_connected(new_smc); 2553 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2554 goto out_free; 2555 2556 out_unlock: 2557 mutex_unlock(&smc_server_lgr_pending); 2558 out_decl: 2559 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2560 proposal_version); 2561 out_free: 2562 kfree(ini); 2563 kfree(buf); 2564 } 2565 2566 static void smc_tcp_listen_work(struct work_struct *work) 2567 { 2568 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2569 tcp_listen_work); 2570 struct sock *lsk = &lsmc->sk; 2571 struct smc_sock *new_smc; 2572 int rc = 0; 2573 2574 lock_sock(lsk); 2575 while (lsk->sk_state == SMC_LISTEN) { 2576 rc = smc_clcsock_accept(lsmc, &new_smc); 2577 if (rc) /* clcsock accept queue empty or error */ 2578 goto out; 2579 if (!new_smc) 2580 continue; 2581 2582 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2583 atomic_inc(&lsmc->queued_smc_hs); 2584 2585 new_smc->listen_smc = lsmc; 2586 new_smc->use_fallback = lsmc->use_fallback; 2587 new_smc->fallback_rsn = lsmc->fallback_rsn; 2588 sock_hold(lsk); /* sock_put in smc_listen_work */ 2589 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2590 smc_copy_sock_settings_to_smc(new_smc); 2591 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2592 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2593 sock_put(&new_smc->sk); 2594 } 2595 2596 out: 2597 release_sock(lsk); 2598 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2599 } 2600 2601 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2602 { 2603 struct smc_sock *lsmc; 2604 2605 read_lock_bh(&listen_clcsock->sk_callback_lock); 2606 lsmc = smc_clcsock_user_data(listen_clcsock); 2607 if (!lsmc) 2608 goto out; 2609 lsmc->clcsk_data_ready(listen_clcsock); 2610 if (lsmc->sk.sk_state == SMC_LISTEN) { 2611 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2612 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2613 sock_put(&lsmc->sk); 2614 } 2615 out: 2616 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2617 } 2618 2619 int smc_listen(struct socket *sock, int backlog) 2620 { 2621 struct sock *sk = sock->sk; 2622 struct smc_sock *smc; 2623 int rc; 2624 2625 smc = smc_sk(sk); 2626 lock_sock(sk); 2627 2628 rc = -EINVAL; 2629 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2630 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2631 goto out; 2632 2633 rc = 0; 2634 if (sk->sk_state == SMC_LISTEN) { 2635 sk->sk_max_ack_backlog = backlog; 2636 goto out; 2637 } 2638 /* some socket options are handled in core, so we could not apply 2639 * them to the clc socket -- copy smc socket options to clc socket 2640 */ 2641 smc_copy_sock_settings_to_clc(smc); 2642 if (!smc->use_fallback) 2643 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2644 2645 /* save original sk_data_ready function and establish 2646 * smc-specific sk_data_ready function 2647 */ 2648 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2649 smc->clcsock->sk->sk_user_data = 2650 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2651 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2652 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2653 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2654 2655 /* save original ops */ 2656 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2657 2658 smc->af_ops = *smc->ori_af_ops; 2659 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2660 2661 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2662 2663 if (smc->limit_smc_hs) 2664 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2665 2666 rc = kernel_listen(smc->clcsock, backlog); 2667 if (rc) { 2668 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2669 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2670 &smc->clcsk_data_ready); 2671 smc->clcsock->sk->sk_user_data = NULL; 2672 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2673 goto out; 2674 } 2675 sk->sk_max_ack_backlog = backlog; 2676 sk->sk_ack_backlog = 0; 2677 sk->sk_state = SMC_LISTEN; 2678 2679 out: 2680 release_sock(sk); 2681 return rc; 2682 } 2683 2684 int smc_accept(struct socket *sock, struct socket *new_sock, 2685 struct proto_accept_arg *arg) 2686 { 2687 struct sock *sk = sock->sk, *nsk; 2688 DECLARE_WAITQUEUE(wait, current); 2689 struct smc_sock *lsmc; 2690 long timeo; 2691 int rc = 0; 2692 2693 lsmc = smc_sk(sk); 2694 sock_hold(sk); /* sock_put below */ 2695 lock_sock(sk); 2696 2697 if (lsmc->sk.sk_state != SMC_LISTEN) { 2698 rc = -EINVAL; 2699 release_sock(sk); 2700 goto out; 2701 } 2702 2703 /* Wait for an incoming connection */ 2704 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2705 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2706 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2707 set_current_state(TASK_INTERRUPTIBLE); 2708 if (!timeo) { 2709 rc = -EAGAIN; 2710 break; 2711 } 2712 release_sock(sk); 2713 timeo = schedule_timeout(timeo); 2714 /* wakeup by sk_data_ready in smc_listen_work() */ 2715 sched_annotate_sleep(); 2716 lock_sock(sk); 2717 if (signal_pending(current)) { 2718 rc = sock_intr_errno(timeo); 2719 break; 2720 } 2721 } 2722 set_current_state(TASK_RUNNING); 2723 remove_wait_queue(sk_sleep(sk), &wait); 2724 2725 if (!rc) 2726 rc = sock_error(nsk); 2727 release_sock(sk); 2728 if (rc) 2729 goto out; 2730 2731 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2732 /* wait till data arrives on the socket */ 2733 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2734 MSEC_PER_SEC); 2735 if (smc_sk(nsk)->use_fallback) { 2736 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2737 2738 lock_sock(clcsk); 2739 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2740 sk_wait_data(clcsk, &timeo, NULL); 2741 release_sock(clcsk); 2742 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2743 lock_sock(nsk); 2744 smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available); 2745 release_sock(nsk); 2746 } 2747 } 2748 2749 out: 2750 sock_put(sk); /* sock_hold above */ 2751 return rc; 2752 } 2753 2754 int smc_getname(struct socket *sock, struct sockaddr *addr, 2755 int peer) 2756 { 2757 struct smc_sock *smc; 2758 2759 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2760 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2761 return -ENOTCONN; 2762 2763 smc = smc_sk(sock->sk); 2764 2765 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2766 } 2767 2768 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2769 { 2770 struct sock *sk = sock->sk; 2771 struct smc_sock *smc; 2772 int rc; 2773 2774 smc = smc_sk(sk); 2775 lock_sock(sk); 2776 2777 /* SMC does not support connect with fastopen */ 2778 if (msg->msg_flags & MSG_FASTOPEN) { 2779 /* not connected yet, fallback */ 2780 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2781 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2782 if (rc) 2783 goto out; 2784 } else { 2785 rc = -EINVAL; 2786 goto out; 2787 } 2788 } else if ((sk->sk_state != SMC_ACTIVE) && 2789 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2790 (sk->sk_state != SMC_INIT)) { 2791 rc = -EPIPE; 2792 goto out; 2793 } 2794 2795 if (smc->use_fallback) { 2796 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2797 } else { 2798 rc = smc_tx_sendmsg(smc, msg, len); 2799 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2800 } 2801 out: 2802 release_sock(sk); 2803 return rc; 2804 } 2805 2806 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2807 int flags) 2808 { 2809 struct sock *sk = sock->sk; 2810 struct smc_sock *smc; 2811 int rc = -ENOTCONN; 2812 2813 smc = smc_sk(sk); 2814 lock_sock(sk); 2815 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2816 /* socket was connected before, no more data to read */ 2817 rc = 0; 2818 goto out; 2819 } 2820 if ((sk->sk_state == SMC_INIT) || 2821 (sk->sk_state == SMC_LISTEN) || 2822 (sk->sk_state == SMC_CLOSED)) 2823 goto out; 2824 2825 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2826 rc = 0; 2827 goto out; 2828 } 2829 2830 if (smc->use_fallback) { 2831 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2832 } else { 2833 msg->msg_namelen = 0; 2834 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2835 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2836 } 2837 2838 out: 2839 release_sock(sk); 2840 return rc; 2841 } 2842 2843 static __poll_t smc_accept_poll(struct sock *parent) 2844 { 2845 struct smc_sock *isk = smc_sk(parent); 2846 __poll_t mask = 0; 2847 2848 spin_lock(&isk->accept_q_lock); 2849 if (!list_empty(&isk->accept_q)) 2850 mask = EPOLLIN | EPOLLRDNORM; 2851 spin_unlock(&isk->accept_q_lock); 2852 2853 return mask; 2854 } 2855 2856 __poll_t smc_poll(struct file *file, struct socket *sock, 2857 poll_table *wait) 2858 { 2859 struct sock *sk = sock->sk; 2860 struct smc_sock *smc; 2861 __poll_t mask = 0; 2862 2863 if (!sk) 2864 return EPOLLNVAL; 2865 2866 smc = smc_sk(sock->sk); 2867 if (smc->use_fallback) { 2868 /* delegate to CLC child sock */ 2869 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2870 sk->sk_err = smc->clcsock->sk->sk_err; 2871 } else { 2872 if (sk->sk_state != SMC_CLOSED) 2873 sock_poll_wait(file, sock, wait); 2874 if (sk->sk_err) 2875 mask |= EPOLLERR; 2876 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2877 (sk->sk_state == SMC_CLOSED)) 2878 mask |= EPOLLHUP; 2879 if (sk->sk_state == SMC_LISTEN) { 2880 /* woken up by sk_data_ready in smc_listen_work() */ 2881 mask |= smc_accept_poll(sk); 2882 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2883 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2884 wait); 2885 sk->sk_err = smc->clcsock->sk->sk_err; 2886 } else { 2887 if ((sk->sk_state != SMC_INIT && 2888 atomic_read(&smc->conn.sndbuf_space)) || 2889 sk->sk_shutdown & SEND_SHUTDOWN) { 2890 mask |= EPOLLOUT | EPOLLWRNORM; 2891 } else { 2892 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2893 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2894 2895 if (sk->sk_state != SMC_INIT) { 2896 /* Race breaker the same way as tcp_poll(). */ 2897 smp_mb__after_atomic(); 2898 if (atomic_read(&smc->conn.sndbuf_space)) 2899 mask |= EPOLLOUT | EPOLLWRNORM; 2900 } 2901 } 2902 if (atomic_read(&smc->conn.bytes_to_rcv)) 2903 mask |= EPOLLIN | EPOLLRDNORM; 2904 if (sk->sk_shutdown & RCV_SHUTDOWN) 2905 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2906 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2907 mask |= EPOLLIN; 2908 if (smc->conn.urg_state == SMC_URG_VALID) 2909 mask |= EPOLLPRI; 2910 } 2911 } 2912 2913 return mask; 2914 } 2915 2916 int smc_shutdown(struct socket *sock, int how) 2917 { 2918 struct sock *sk = sock->sk; 2919 bool do_shutdown = true; 2920 struct smc_sock *smc; 2921 int rc = -EINVAL; 2922 int old_state; 2923 int rc1 = 0; 2924 2925 smc = smc_sk(sk); 2926 2927 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2928 return rc; 2929 2930 lock_sock(sk); 2931 2932 if (sock->state == SS_CONNECTING) { 2933 if (sk->sk_state == SMC_ACTIVE) 2934 sock->state = SS_CONNECTED; 2935 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2936 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2937 sk->sk_state == SMC_APPCLOSEWAIT1 || 2938 sk->sk_state == SMC_APPCLOSEWAIT2 || 2939 sk->sk_state == SMC_APPFINCLOSEWAIT) 2940 sock->state = SS_DISCONNECTING; 2941 } 2942 2943 rc = -ENOTCONN; 2944 if ((sk->sk_state != SMC_ACTIVE) && 2945 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2946 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2947 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2948 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2949 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2950 goto out; 2951 if (smc->use_fallback) { 2952 rc = kernel_sock_shutdown(smc->clcsock, how); 2953 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2954 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2955 sk->sk_state = SMC_CLOSED; 2956 sk->sk_socket->state = SS_UNCONNECTED; 2957 sock_put(sk); 2958 } 2959 goto out; 2960 } 2961 switch (how) { 2962 case SHUT_RDWR: /* shutdown in both directions */ 2963 old_state = sk->sk_state; 2964 rc = smc_close_active(smc); 2965 if (old_state == SMC_ACTIVE && 2966 sk->sk_state == SMC_PEERCLOSEWAIT1) 2967 do_shutdown = false; 2968 break; 2969 case SHUT_WR: 2970 rc = smc_close_shutdown_write(smc); 2971 break; 2972 case SHUT_RD: 2973 rc = 0; 2974 /* nothing more to do because peer is not involved */ 2975 break; 2976 } 2977 if (do_shutdown && smc->clcsock) 2978 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2979 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2980 sk->sk_shutdown |= how + 1; 2981 2982 if (sk->sk_state == SMC_CLOSED) 2983 sock->state = SS_UNCONNECTED; 2984 else 2985 sock->state = SS_DISCONNECTING; 2986 out: 2987 release_sock(sk); 2988 return rc ? rc : rc1; 2989 } 2990 2991 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2992 char __user *optval, int __user *optlen) 2993 { 2994 struct smc_sock *smc; 2995 int val, len; 2996 2997 smc = smc_sk(sock->sk); 2998 2999 if (get_user(len, optlen)) 3000 return -EFAULT; 3001 3002 len = min_t(int, len, sizeof(int)); 3003 3004 if (len < 0) 3005 return -EINVAL; 3006 3007 switch (optname) { 3008 case SMC_LIMIT_HS: 3009 val = smc->limit_smc_hs; 3010 break; 3011 default: 3012 return -EOPNOTSUPP; 3013 } 3014 3015 if (put_user(len, optlen)) 3016 return -EFAULT; 3017 if (copy_to_user(optval, &val, len)) 3018 return -EFAULT; 3019 3020 return 0; 3021 } 3022 3023 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3024 sockptr_t optval, unsigned int optlen) 3025 { 3026 struct sock *sk = sock->sk; 3027 struct smc_sock *smc; 3028 int val, rc; 3029 3030 smc = smc_sk(sk); 3031 3032 lock_sock(sk); 3033 switch (optname) { 3034 case SMC_LIMIT_HS: 3035 if (optlen < sizeof(int)) { 3036 rc = -EINVAL; 3037 break; 3038 } 3039 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3040 rc = -EFAULT; 3041 break; 3042 } 3043 3044 smc->limit_smc_hs = !!val; 3045 rc = 0; 3046 break; 3047 default: 3048 rc = -EOPNOTSUPP; 3049 break; 3050 } 3051 release_sock(sk); 3052 3053 return rc; 3054 } 3055 3056 int smc_setsockopt(struct socket *sock, int level, int optname, 3057 sockptr_t optval, unsigned int optlen) 3058 { 3059 struct sock *sk = sock->sk; 3060 struct smc_sock *smc; 3061 int val, rc; 3062 3063 if (level == SOL_TCP && optname == TCP_ULP) 3064 return -EOPNOTSUPP; 3065 else if (level == SOL_SMC) 3066 return __smc_setsockopt(sock, level, optname, optval, optlen); 3067 3068 smc = smc_sk(sk); 3069 3070 /* generic setsockopts reaching us here always apply to the 3071 * CLC socket 3072 */ 3073 mutex_lock(&smc->clcsock_release_lock); 3074 if (!smc->clcsock) { 3075 mutex_unlock(&smc->clcsock_release_lock); 3076 return -EBADF; 3077 } 3078 if (unlikely(!smc->clcsock->ops->setsockopt)) 3079 rc = -EOPNOTSUPP; 3080 else 3081 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3082 optval, optlen); 3083 if (smc->clcsock->sk->sk_err) { 3084 sk->sk_err = smc->clcsock->sk->sk_err; 3085 sk_error_report(sk); 3086 } 3087 mutex_unlock(&smc->clcsock_release_lock); 3088 3089 if (optlen < sizeof(int)) 3090 return -EINVAL; 3091 if (copy_from_sockptr(&val, optval, sizeof(int))) 3092 return -EFAULT; 3093 3094 lock_sock(sk); 3095 if (rc || smc->use_fallback) 3096 goto out; 3097 switch (optname) { 3098 case TCP_FASTOPEN: 3099 case TCP_FASTOPEN_CONNECT: 3100 case TCP_FASTOPEN_KEY: 3101 case TCP_FASTOPEN_NO_COOKIE: 3102 /* option not supported by SMC */ 3103 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3104 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3105 } else { 3106 rc = -EINVAL; 3107 } 3108 break; 3109 case TCP_NODELAY: 3110 if (sk->sk_state != SMC_INIT && 3111 sk->sk_state != SMC_LISTEN && 3112 sk->sk_state != SMC_CLOSED) { 3113 if (val) { 3114 SMC_STAT_INC(smc, ndly_cnt); 3115 smc_tx_pending(&smc->conn); 3116 cancel_delayed_work(&smc->conn.tx_work); 3117 } 3118 } 3119 break; 3120 case TCP_CORK: 3121 if (sk->sk_state != SMC_INIT && 3122 sk->sk_state != SMC_LISTEN && 3123 sk->sk_state != SMC_CLOSED) { 3124 if (!val) { 3125 SMC_STAT_INC(smc, cork_cnt); 3126 smc_tx_pending(&smc->conn); 3127 cancel_delayed_work(&smc->conn.tx_work); 3128 } 3129 } 3130 break; 3131 case TCP_DEFER_ACCEPT: 3132 smc->sockopt_defer_accept = val; 3133 break; 3134 default: 3135 break; 3136 } 3137 out: 3138 release_sock(sk); 3139 3140 return rc; 3141 } 3142 3143 int smc_getsockopt(struct socket *sock, int level, int optname, 3144 char __user *optval, int __user *optlen) 3145 { 3146 struct smc_sock *smc; 3147 int rc; 3148 3149 if (level == SOL_SMC) 3150 return __smc_getsockopt(sock, level, optname, optval, optlen); 3151 3152 smc = smc_sk(sock->sk); 3153 mutex_lock(&smc->clcsock_release_lock); 3154 if (!smc->clcsock) { 3155 mutex_unlock(&smc->clcsock_release_lock); 3156 return -EBADF; 3157 } 3158 /* socket options apply to the CLC socket */ 3159 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3160 mutex_unlock(&smc->clcsock_release_lock); 3161 return -EOPNOTSUPP; 3162 } 3163 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3164 optval, optlen); 3165 mutex_unlock(&smc->clcsock_release_lock); 3166 return rc; 3167 } 3168 3169 int smc_ioctl(struct socket *sock, unsigned int cmd, 3170 unsigned long arg) 3171 { 3172 union smc_host_cursor cons, urg; 3173 struct smc_connection *conn; 3174 struct smc_sock *smc; 3175 int answ; 3176 3177 smc = smc_sk(sock->sk); 3178 conn = &smc->conn; 3179 lock_sock(&smc->sk); 3180 if (smc->use_fallback) { 3181 if (!smc->clcsock) { 3182 release_sock(&smc->sk); 3183 return -EBADF; 3184 } 3185 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3186 release_sock(&smc->sk); 3187 return answ; 3188 } 3189 switch (cmd) { 3190 case SIOCINQ: /* same as FIONREAD */ 3191 if (smc->sk.sk_state == SMC_LISTEN) { 3192 release_sock(&smc->sk); 3193 return -EINVAL; 3194 } 3195 if (smc->sk.sk_state == SMC_INIT || 3196 smc->sk.sk_state == SMC_CLOSED) 3197 answ = 0; 3198 else 3199 answ = atomic_read(&smc->conn.bytes_to_rcv); 3200 break; 3201 case SIOCOUTQ: 3202 /* output queue size (not send + not acked) */ 3203 if (smc->sk.sk_state == SMC_LISTEN) { 3204 release_sock(&smc->sk); 3205 return -EINVAL; 3206 } 3207 if (smc->sk.sk_state == SMC_INIT || 3208 smc->sk.sk_state == SMC_CLOSED) 3209 answ = 0; 3210 else 3211 answ = smc->conn.sndbuf_desc->len - 3212 atomic_read(&smc->conn.sndbuf_space); 3213 break; 3214 case SIOCOUTQNSD: 3215 /* output queue size (not send only) */ 3216 if (smc->sk.sk_state == SMC_LISTEN) { 3217 release_sock(&smc->sk); 3218 return -EINVAL; 3219 } 3220 if (smc->sk.sk_state == SMC_INIT || 3221 smc->sk.sk_state == SMC_CLOSED) 3222 answ = 0; 3223 else 3224 answ = smc_tx_prepared_sends(&smc->conn); 3225 break; 3226 case SIOCATMARK: 3227 if (smc->sk.sk_state == SMC_LISTEN) { 3228 release_sock(&smc->sk); 3229 return -EINVAL; 3230 } 3231 if (smc->sk.sk_state == SMC_INIT || 3232 smc->sk.sk_state == SMC_CLOSED) { 3233 answ = 0; 3234 } else { 3235 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3236 smc_curs_copy(&urg, &conn->urg_curs, conn); 3237 answ = smc_curs_diff(conn->rmb_desc->len, 3238 &cons, &urg) == 1; 3239 } 3240 break; 3241 default: 3242 release_sock(&smc->sk); 3243 return -ENOIOCTLCMD; 3244 } 3245 release_sock(&smc->sk); 3246 3247 return put_user(answ, (int __user *)arg); 3248 } 3249 3250 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3251 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3252 * updates till whenever a respective page has been fully processed. 3253 * Note that subsequent recv() calls have to wait till all splice() processing 3254 * completed. 3255 */ 3256 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3257 struct pipe_inode_info *pipe, size_t len, 3258 unsigned int flags) 3259 { 3260 struct sock *sk = sock->sk; 3261 struct smc_sock *smc; 3262 int rc = -ENOTCONN; 3263 3264 smc = smc_sk(sk); 3265 lock_sock(sk); 3266 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3267 /* socket was connected before, no more data to read */ 3268 rc = 0; 3269 goto out; 3270 } 3271 if (sk->sk_state == SMC_INIT || 3272 sk->sk_state == SMC_LISTEN || 3273 sk->sk_state == SMC_CLOSED) 3274 goto out; 3275 3276 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3277 rc = 0; 3278 goto out; 3279 } 3280 3281 if (smc->use_fallback) { 3282 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3283 pipe, len, flags); 3284 } else { 3285 if (*ppos) { 3286 rc = -ESPIPE; 3287 goto out; 3288 } 3289 if (flags & SPLICE_F_NONBLOCK) 3290 flags = MSG_DONTWAIT; 3291 else 3292 flags = 0; 3293 SMC_STAT_INC(smc, splice_cnt); 3294 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3295 } 3296 out: 3297 release_sock(sk); 3298 3299 return rc; 3300 } 3301 3302 /* must look like tcp */ 3303 static const struct proto_ops smc_sock_ops = { 3304 .family = PF_SMC, 3305 .owner = THIS_MODULE, 3306 .release = smc_release, 3307 .bind = smc_bind, 3308 .connect = smc_connect, 3309 .socketpair = sock_no_socketpair, 3310 .accept = smc_accept, 3311 .getname = smc_getname, 3312 .poll = smc_poll, 3313 .ioctl = smc_ioctl, 3314 .listen = smc_listen, 3315 .shutdown = smc_shutdown, 3316 .setsockopt = smc_setsockopt, 3317 .getsockopt = smc_getsockopt, 3318 .sendmsg = smc_sendmsg, 3319 .recvmsg = smc_recvmsg, 3320 .mmap = sock_no_mmap, 3321 .splice_read = smc_splice_read, 3322 }; 3323 3324 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3325 { 3326 struct smc_sock *smc = smc_sk(sk); 3327 int rc; 3328 3329 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3330 &smc->clcsock); 3331 if (rc) 3332 return rc; 3333 3334 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3335 * destruction; its sk_state might not be TCP_CLOSE after 3336 * smc->sk is close()d, and TCP timers can be fired later, 3337 * which need net ref. 3338 */ 3339 sk = smc->clcsock->sk; 3340 __netns_tracker_free(net, &sk->ns_tracker, false); 3341 sk->sk_net_refcnt = 1; 3342 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3343 sock_inuse_add(net, 1); 3344 return 0; 3345 } 3346 3347 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3348 int kern, struct socket *clcsock) 3349 { 3350 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3351 struct smc_sock *smc; 3352 struct sock *sk; 3353 int rc; 3354 3355 rc = -ESOCKTNOSUPPORT; 3356 if (sock->type != SOCK_STREAM) 3357 goto out; 3358 3359 rc = -EPROTONOSUPPORT; 3360 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3361 goto out; 3362 3363 rc = -ENOBUFS; 3364 sock->ops = &smc_sock_ops; 3365 sock->state = SS_UNCONNECTED; 3366 sk = smc_sock_alloc(net, sock, protocol); 3367 if (!sk) 3368 goto out; 3369 3370 /* create internal TCP socket for CLC handshake and fallback */ 3371 smc = smc_sk(sk); 3372 3373 rc = 0; 3374 if (clcsock) 3375 smc->clcsock = clcsock; 3376 else 3377 rc = smc_create_clcsk(net, sk, family); 3378 3379 if (rc) { 3380 sk_common_release(sk); 3381 sock->sk = NULL; 3382 } 3383 out: 3384 return rc; 3385 } 3386 3387 static int smc_create(struct net *net, struct socket *sock, int protocol, 3388 int kern) 3389 { 3390 return __smc_create(net, sock, protocol, kern, NULL); 3391 } 3392 3393 static const struct net_proto_family smc_sock_family_ops = { 3394 .family = PF_SMC, 3395 .owner = THIS_MODULE, 3396 .create = smc_create, 3397 }; 3398 3399 static int smc_ulp_init(struct sock *sk) 3400 { 3401 struct socket *tcp = sk->sk_socket; 3402 struct net *net = sock_net(sk); 3403 struct socket *smcsock; 3404 int protocol, ret; 3405 3406 /* only TCP can be replaced */ 3407 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3408 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3409 return -ESOCKTNOSUPPORT; 3410 /* don't handle wq now */ 3411 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3412 return -ENOTCONN; 3413 3414 if (sk->sk_family == AF_INET) 3415 protocol = SMCPROTO_SMC; 3416 else 3417 protocol = SMCPROTO_SMC6; 3418 3419 smcsock = sock_alloc(); 3420 if (!smcsock) 3421 return -ENFILE; 3422 3423 smcsock->type = SOCK_STREAM; 3424 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3425 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3426 if (ret) { 3427 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3428 return ret; 3429 } 3430 3431 /* replace tcp socket to smc */ 3432 smcsock->file = tcp->file; 3433 smcsock->file->private_data = smcsock; 3434 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3435 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3436 tcp->file = NULL; 3437 3438 return ret; 3439 } 3440 3441 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3442 const gfp_t priority) 3443 { 3444 struct inet_connection_sock *icsk = inet_csk(newsk); 3445 3446 /* don't inherit ulp ops to child when listen */ 3447 icsk->icsk_ulp_ops = NULL; 3448 } 3449 3450 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3451 .name = "smc", 3452 .owner = THIS_MODULE, 3453 .init = smc_ulp_init, 3454 .clone = smc_ulp_clone, 3455 }; 3456 3457 unsigned int smc_net_id; 3458 3459 static __net_init int smc_net_init(struct net *net) 3460 { 3461 int rc; 3462 3463 rc = smc_sysctl_net_init(net); 3464 if (rc) 3465 return rc; 3466 return smc_pnet_net_init(net); 3467 } 3468 3469 static void __net_exit smc_net_exit(struct net *net) 3470 { 3471 smc_sysctl_net_exit(net); 3472 smc_pnet_net_exit(net); 3473 } 3474 3475 static __net_init int smc_net_stat_init(struct net *net) 3476 { 3477 return smc_stats_init(net); 3478 } 3479 3480 static void __net_exit smc_net_stat_exit(struct net *net) 3481 { 3482 smc_stats_exit(net); 3483 } 3484 3485 static struct pernet_operations smc_net_ops = { 3486 .init = smc_net_init, 3487 .exit = smc_net_exit, 3488 .id = &smc_net_id, 3489 .size = sizeof(struct smc_net), 3490 }; 3491 3492 static struct pernet_operations smc_net_stat_ops = { 3493 .init = smc_net_stat_init, 3494 .exit = smc_net_stat_exit, 3495 }; 3496 3497 static int __init smc_init(void) 3498 { 3499 int rc; 3500 3501 rc = register_pernet_subsys(&smc_net_ops); 3502 if (rc) 3503 return rc; 3504 3505 rc = register_pernet_subsys(&smc_net_stat_ops); 3506 if (rc) 3507 goto out_pernet_subsys; 3508 3509 rc = smc_ism_init(); 3510 if (rc) 3511 goto out_pernet_subsys_stat; 3512 smc_clc_init(); 3513 3514 rc = smc_nl_init(); 3515 if (rc) 3516 goto out_ism; 3517 3518 rc = smc_pnet_init(); 3519 if (rc) 3520 goto out_nl; 3521 3522 rc = -ENOMEM; 3523 3524 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3525 if (!smc_tcp_ls_wq) 3526 goto out_pnet; 3527 3528 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3529 if (!smc_hs_wq) 3530 goto out_alloc_tcp_ls_wq; 3531 3532 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3533 if (!smc_close_wq) 3534 goto out_alloc_hs_wq; 3535 3536 rc = smc_core_init(); 3537 if (rc) { 3538 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3539 goto out_alloc_wqs; 3540 } 3541 3542 rc = smc_llc_init(); 3543 if (rc) { 3544 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3545 goto out_core; 3546 } 3547 3548 rc = smc_cdc_init(); 3549 if (rc) { 3550 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3551 goto out_core; 3552 } 3553 3554 rc = proto_register(&smc_proto, 1); 3555 if (rc) { 3556 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3557 goto out_core; 3558 } 3559 3560 rc = proto_register(&smc_proto6, 1); 3561 if (rc) { 3562 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3563 goto out_proto; 3564 } 3565 3566 rc = sock_register(&smc_sock_family_ops); 3567 if (rc) { 3568 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3569 goto out_proto6; 3570 } 3571 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3572 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3573 3574 rc = smc_ib_register_client(); 3575 if (rc) { 3576 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3577 goto out_sock; 3578 } 3579 3580 rc = smc_loopback_init(); 3581 if (rc) { 3582 pr_err("%s: smc_loopback_init fails with %d\n", __func__, rc); 3583 goto out_ib; 3584 } 3585 3586 rc = tcp_register_ulp(&smc_ulp_ops); 3587 if (rc) { 3588 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3589 goto out_lo; 3590 } 3591 rc = smc_inet_init(); 3592 if (rc) { 3593 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3594 goto out_ulp; 3595 } 3596 static_branch_enable(&tcp_have_smc); 3597 return 0; 3598 out_ulp: 3599 tcp_unregister_ulp(&smc_ulp_ops); 3600 out_lo: 3601 smc_loopback_exit(); 3602 out_ib: 3603 smc_ib_unregister_client(); 3604 out_sock: 3605 sock_unregister(PF_SMC); 3606 out_proto6: 3607 proto_unregister(&smc_proto6); 3608 out_proto: 3609 proto_unregister(&smc_proto); 3610 out_core: 3611 smc_core_exit(); 3612 out_alloc_wqs: 3613 destroy_workqueue(smc_close_wq); 3614 out_alloc_hs_wq: 3615 destroy_workqueue(smc_hs_wq); 3616 out_alloc_tcp_ls_wq: 3617 destroy_workqueue(smc_tcp_ls_wq); 3618 out_pnet: 3619 smc_pnet_exit(); 3620 out_nl: 3621 smc_nl_exit(); 3622 out_ism: 3623 smc_clc_exit(); 3624 smc_ism_exit(); 3625 out_pernet_subsys_stat: 3626 unregister_pernet_subsys(&smc_net_stat_ops); 3627 out_pernet_subsys: 3628 unregister_pernet_subsys(&smc_net_ops); 3629 3630 return rc; 3631 } 3632 3633 static void __exit smc_exit(void) 3634 { 3635 static_branch_disable(&tcp_have_smc); 3636 smc_inet_exit(); 3637 tcp_unregister_ulp(&smc_ulp_ops); 3638 sock_unregister(PF_SMC); 3639 smc_core_exit(); 3640 smc_loopback_exit(); 3641 smc_ib_unregister_client(); 3642 smc_ism_exit(); 3643 destroy_workqueue(smc_close_wq); 3644 destroy_workqueue(smc_tcp_ls_wq); 3645 destroy_workqueue(smc_hs_wq); 3646 proto_unregister(&smc_proto6); 3647 proto_unregister(&smc_proto); 3648 smc_pnet_exit(); 3649 smc_nl_exit(); 3650 smc_clc_exit(); 3651 unregister_pernet_subsys(&smc_net_stat_ops); 3652 unregister_pernet_subsys(&smc_net_ops); 3653 rcu_barrier(); 3654 } 3655 3656 module_init(smc_init); 3657 module_exit(smc_exit); 3658 3659 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3660 MODULE_DESCRIPTION("smc socket address family"); 3661 MODULE_LICENSE("GPL"); 3662 MODULE_ALIAS_NETPROTO(PF_SMC); 3663 MODULE_ALIAS_TCP_ULP("smc"); 3664 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3665 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3666 #if IS_ENABLED(CONFIG_IPV6) 3667 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3668 #endif /* CONFIG_IPV6 */ 3669 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3670