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 (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2039 return SMC_CLC_DECL_DIFFPREFIX; 2040 2041 return 0; 2042 } 2043 2044 /* listen worker: initialize connection and buffers */ 2045 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2046 struct smc_init_info *ini) 2047 { 2048 int rc; 2049 2050 /* allocate connection / link group */ 2051 rc = smc_conn_create(new_smc, ini); 2052 if (rc) 2053 return rc; 2054 2055 /* create send buffer and rmb */ 2056 if (smc_buf_create(new_smc, false)) { 2057 smc_conn_abort(new_smc, ini->first_contact_local); 2058 return SMC_CLC_DECL_MEM; 2059 } 2060 2061 return 0; 2062 } 2063 2064 /* listen worker: initialize connection and buffers for SMC-D */ 2065 static int smc_listen_ism_init(struct smc_sock *new_smc, 2066 struct smc_init_info *ini) 2067 { 2068 int rc; 2069 2070 rc = smc_conn_create(new_smc, ini); 2071 if (rc) 2072 return rc; 2073 2074 /* Create send and receive buffers */ 2075 rc = smc_buf_create(new_smc, true); 2076 if (rc) { 2077 smc_conn_abort(new_smc, ini->first_contact_local); 2078 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2079 SMC_CLC_DECL_MEM; 2080 } 2081 2082 return 0; 2083 } 2084 2085 static bool smc_is_already_selected(struct smcd_dev *smcd, 2086 struct smc_init_info *ini, 2087 int matches) 2088 { 2089 int i; 2090 2091 for (i = 0; i < matches; i++) 2092 if (smcd == ini->ism_dev[i]) 2093 return true; 2094 2095 return false; 2096 } 2097 2098 /* check for ISM devices matching proposed ISM devices */ 2099 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2100 u16 proposed_chid, 2101 struct smcd_gid *proposed_gid, 2102 unsigned int *matches) 2103 { 2104 struct smcd_dev *smcd; 2105 2106 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2107 if (smcd->going_away) 2108 continue; 2109 if (smc_is_already_selected(smcd, ini, *matches)) 2110 continue; 2111 if (smc_ism_get_chid(smcd) == proposed_chid && 2112 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2113 ini->ism_peer_gid[*matches].gid = proposed_gid->gid; 2114 if (__smc_ism_is_emulated(proposed_chid)) 2115 ini->ism_peer_gid[*matches].gid_ext = 2116 proposed_gid->gid_ext; 2117 /* non-Emulated-ISM's peer gid_ext remains 0. */ 2118 ini->ism_dev[*matches] = smcd; 2119 (*matches)++; 2120 break; 2121 } 2122 } 2123 } 2124 2125 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2126 { 2127 if (!ini->rc) 2128 ini->rc = rc; 2129 } 2130 2131 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2132 struct smc_clc_msg_proposal *pclc, 2133 struct smc_init_info *ini) 2134 { 2135 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2136 struct smc_clc_v2_extension *smc_v2_ext; 2137 struct smc_clc_msg_smcd *pclc_smcd; 2138 unsigned int matches = 0; 2139 struct smcd_gid smcd_gid; 2140 u8 smcd_version; 2141 u8 *eid = NULL; 2142 int i, rc; 2143 u16 chid; 2144 2145 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2146 goto not_found; 2147 2148 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2149 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2150 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2151 2152 mutex_lock(&smcd_dev_list.mutex); 2153 if (pclc_smcd->ism.chid) { 2154 /* check for ISM device matching proposed native ISM device */ 2155 smcd_gid.gid = ntohll(pclc_smcd->ism.gid); 2156 smcd_gid.gid_ext = 0; 2157 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2158 &smcd_gid, &matches); 2159 } 2160 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) { 2161 /* check for ISM devices matching proposed non-native ISM 2162 * devices 2163 */ 2164 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid); 2165 smcd_gid.gid_ext = 0; 2166 chid = ntohs(smcd_v2_ext->gidchid[i].chid); 2167 if (__smc_ism_is_emulated(chid)) { 2168 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt || 2169 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid)) 2170 /* each Emulated-ISM device takes two GID-CHID 2171 * entries and CHID of the second entry repeats 2172 * that of the first entry. 2173 * 2174 * So check if the next GID-CHID entry exists 2175 * and both two entries' CHIDs are the same. 2176 */ 2177 continue; 2178 smcd_gid.gid_ext = 2179 ntohll(smcd_v2_ext->gidchid[++i].gid); 2180 } 2181 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches); 2182 } 2183 mutex_unlock(&smcd_dev_list.mutex); 2184 2185 if (!ini->ism_dev[0]) { 2186 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2187 goto not_found; 2188 } 2189 2190 smc_ism_get_system_eid(&eid); 2191 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2192 smcd_v2_ext->system_eid, eid)) 2193 goto not_found; 2194 2195 /* separate - outside the smcd_dev_list.lock */ 2196 smcd_version = ini->smcd_version; 2197 for (i = 0; i < matches; i++) { 2198 ini->smcd_version = SMC_V2; 2199 ini->is_smcd = true; 2200 ini->ism_selected = i; 2201 rc = smc_listen_ism_init(new_smc, ini); 2202 if (rc) { 2203 smc_find_ism_store_rc(rc, ini); 2204 /* try next active ISM device */ 2205 continue; 2206 } 2207 return; /* matching and usable V2 ISM device found */ 2208 } 2209 /* no V2 ISM device could be initialized */ 2210 ini->smcd_version = smcd_version; /* restore original value */ 2211 ini->negotiated_eid[0] = 0; 2212 2213 not_found: 2214 ini->smcd_version &= ~SMC_V2; 2215 ini->ism_dev[0] = NULL; 2216 ini->is_smcd = false; 2217 } 2218 2219 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2220 struct smc_clc_msg_proposal *pclc, 2221 struct smc_init_info *ini) 2222 { 2223 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2224 int rc = 0; 2225 2226 /* check if ISM V1 is available */ 2227 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2228 goto not_found; 2229 ini->is_smcd = true; /* prepare ISM check */ 2230 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid); 2231 ini->ism_peer_gid[0].gid_ext = 0; 2232 rc = smc_find_ism_device(new_smc, ini); 2233 if (rc) 2234 goto not_found; 2235 ini->ism_selected = 0; 2236 rc = smc_listen_ism_init(new_smc, ini); 2237 if (!rc) 2238 return; /* V1 ISM device found */ 2239 2240 not_found: 2241 smc_find_ism_store_rc(rc, ini); 2242 ini->smcd_version &= ~SMC_V1; 2243 ini->ism_dev[0] = NULL; 2244 ini->is_smcd = false; 2245 } 2246 2247 /* listen worker: register buffers */ 2248 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2249 { 2250 struct smc_connection *conn = &new_smc->conn; 2251 2252 if (!local_first) { 2253 /* reg sendbufs if they were vzalloced */ 2254 if (conn->sndbuf_desc->is_vm) { 2255 if (smcr_lgr_reg_sndbufs(conn->lnk, 2256 conn->sndbuf_desc)) 2257 return SMC_CLC_DECL_ERR_REGBUF; 2258 } 2259 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2260 return SMC_CLC_DECL_ERR_REGBUF; 2261 } 2262 2263 return 0; 2264 } 2265 2266 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2267 struct smc_clc_msg_proposal *pclc, 2268 struct smc_init_info *ini) 2269 { 2270 struct smc_clc_v2_extension *smc_v2_ext; 2271 u8 smcr_version; 2272 int rc; 2273 2274 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2275 goto not_found; 2276 2277 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2278 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2279 goto not_found; 2280 2281 /* prepare RDMA check */ 2282 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2283 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2284 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2285 ini->check_smcrv2 = true; 2286 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2287 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2288 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2289 rc = smc_find_rdma_device(new_smc, ini); 2290 if (rc) { 2291 smc_find_ism_store_rc(rc, ini); 2292 goto not_found; 2293 } 2294 if (!ini->smcrv2.uses_gateway) 2295 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2296 2297 smcr_version = ini->smcr_version; 2298 ini->smcr_version = SMC_V2; 2299 rc = smc_listen_rdma_init(new_smc, ini); 2300 if (!rc) { 2301 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2302 if (rc) 2303 smc_conn_abort(new_smc, ini->first_contact_local); 2304 } 2305 if (!rc) 2306 return; 2307 ini->smcr_version = smcr_version; 2308 smc_find_ism_store_rc(rc, ini); 2309 2310 not_found: 2311 ini->smcr_version &= ~SMC_V2; 2312 ini->smcrv2.ib_dev_v2 = NULL; 2313 ini->check_smcrv2 = false; 2314 } 2315 2316 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2317 struct smc_clc_msg_proposal *pclc, 2318 struct smc_init_info *ini) 2319 { 2320 int rc; 2321 2322 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2323 return SMC_CLC_DECL_NOSMCDEV; 2324 2325 /* prepare RDMA check */ 2326 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2327 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2328 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2329 rc = smc_find_rdma_device(new_smc, ini); 2330 if (rc) { 2331 /* no RDMA device found */ 2332 return SMC_CLC_DECL_NOSMCDEV; 2333 } 2334 rc = smc_listen_rdma_init(new_smc, ini); 2335 if (rc) 2336 return rc; 2337 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2338 } 2339 2340 /* determine the local device matching to proposal */ 2341 static int smc_listen_find_device(struct smc_sock *new_smc, 2342 struct smc_clc_msg_proposal *pclc, 2343 struct smc_init_info *ini) 2344 { 2345 int prfx_rc; 2346 2347 /* check for ISM device matching V2 proposed device */ 2348 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2349 if (ini->ism_dev[0]) 2350 return 0; 2351 2352 /* check for matching IP prefix and subnet length (V1) */ 2353 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2354 if (prfx_rc) 2355 smc_find_ism_store_rc(prfx_rc, ini); 2356 2357 /* get vlan id from IP device */ 2358 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2359 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2360 2361 /* check for ISM device matching V1 proposed device */ 2362 if (!prfx_rc) 2363 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2364 if (ini->ism_dev[0]) 2365 return 0; 2366 2367 if (!smcr_indicated(pclc->hdr.typev1) && 2368 !smcr_indicated(pclc->hdr.typev2)) 2369 /* skip RDMA and decline */ 2370 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2371 2372 /* check if RDMA V2 is available */ 2373 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2374 if (ini->smcrv2.ib_dev_v2) 2375 return 0; 2376 2377 /* check if RDMA V1 is available */ 2378 if (!prfx_rc) { 2379 int rc; 2380 2381 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2382 smc_find_ism_store_rc(rc, ini); 2383 return (!rc) ? 0 : ini->rc; 2384 } 2385 return prfx_rc; 2386 } 2387 2388 /* listen worker: finish RDMA setup */ 2389 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2390 struct smc_clc_msg_accept_confirm *cclc, 2391 bool local_first, 2392 struct smc_init_info *ini) 2393 { 2394 struct smc_link *link = new_smc->conn.lnk; 2395 int reason_code = 0; 2396 2397 if (local_first) 2398 smc_link_save_peer_info(link, cclc, ini); 2399 2400 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2401 return SMC_CLC_DECL_ERR_RTOK; 2402 2403 if (local_first) { 2404 if (smc_ib_ready_link(link)) 2405 return SMC_CLC_DECL_ERR_RDYLNK; 2406 /* QP confirmation over RoCE fabric */ 2407 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2408 reason_code = smcr_serv_conf_first_link(new_smc); 2409 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2410 } 2411 return reason_code; 2412 } 2413 2414 /* setup for connection of server */ 2415 static void smc_listen_work(struct work_struct *work) 2416 { 2417 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2418 smc_listen_work); 2419 struct socket *newclcsock = new_smc->clcsock; 2420 struct smc_clc_msg_accept_confirm *cclc; 2421 struct smc_clc_msg_proposal_area *buf; 2422 struct smc_clc_msg_proposal *pclc; 2423 struct smc_init_info *ini = NULL; 2424 u8 proposal_version = SMC_V1; 2425 u8 accept_version; 2426 int rc = 0; 2427 2428 lock_sock(&new_smc->sk); /* release in smc_listen_out() */ 2429 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2430 return smc_listen_out_err(new_smc); 2431 2432 if (new_smc->use_fallback) { 2433 smc_listen_out_connected(new_smc); 2434 return; 2435 } 2436 2437 /* check if peer is smc capable */ 2438 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2439 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2440 if (rc) 2441 smc_listen_out_err(new_smc); 2442 else 2443 smc_listen_out_connected(new_smc); 2444 return; 2445 } 2446 2447 /* do inband token exchange - 2448 * wait for and receive SMC Proposal CLC message 2449 */ 2450 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2451 if (!buf) { 2452 rc = SMC_CLC_DECL_MEM; 2453 goto out_decl; 2454 } 2455 pclc = (struct smc_clc_msg_proposal *)buf; 2456 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2457 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2458 if (rc) 2459 goto out_decl; 2460 2461 if (pclc->hdr.version > SMC_V1) 2462 proposal_version = SMC_V2; 2463 2464 /* IPSec connections opt out of SMC optimizations */ 2465 if (using_ipsec(new_smc)) { 2466 rc = SMC_CLC_DECL_IPSEC; 2467 goto out_decl; 2468 } 2469 2470 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2471 if (!ini) { 2472 rc = SMC_CLC_DECL_MEM; 2473 goto out_decl; 2474 } 2475 2476 /* initial version checking */ 2477 rc = smc_listen_v2_check(new_smc, pclc, ini); 2478 if (rc) 2479 goto out_decl; 2480 2481 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini); 2482 if (rc) 2483 goto out_decl; 2484 2485 mutex_lock(&smc_server_lgr_pending); 2486 smc_rx_init(new_smc); 2487 smc_tx_init(new_smc); 2488 2489 /* determine ISM or RoCE device used for connection */ 2490 rc = smc_listen_find_device(new_smc, pclc, ini); 2491 if (rc) 2492 goto out_unlock; 2493 2494 /* send SMC Accept CLC message */ 2495 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2496 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2497 accept_version, ini->negotiated_eid, ini); 2498 if (rc) 2499 goto out_unlock; 2500 2501 /* SMC-D does not need this lock any more */ 2502 if (ini->is_smcd) 2503 mutex_unlock(&smc_server_lgr_pending); 2504 2505 /* receive SMC Confirm CLC message */ 2506 memset(buf, 0, sizeof(*buf)); 2507 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2508 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2509 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2510 if (rc) { 2511 if (!ini->is_smcd) 2512 goto out_unlock; 2513 goto out_decl; 2514 } 2515 2516 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2517 if (rc) { 2518 if (!ini->is_smcd) 2519 goto out_unlock; 2520 goto out_decl; 2521 } 2522 2523 /* fce smc release version is needed in smc_listen_rdma_finish, 2524 * so save fce info here. 2525 */ 2526 smc_conn_save_peer_info_fce(new_smc, cclc); 2527 2528 /* finish worker */ 2529 if (!ini->is_smcd) { 2530 rc = smc_listen_rdma_finish(new_smc, cclc, 2531 ini->first_contact_local, ini); 2532 if (rc) 2533 goto out_unlock; 2534 mutex_unlock(&smc_server_lgr_pending); 2535 } 2536 smc_conn_save_peer_info(new_smc, cclc); 2537 2538 if (ini->is_smcd && 2539 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) { 2540 rc = smcd_buf_attach(new_smc); 2541 if (rc) 2542 goto out_decl; 2543 } 2544 2545 smc_listen_out_connected(new_smc); 2546 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2547 goto out_free; 2548 2549 out_unlock: 2550 mutex_unlock(&smc_server_lgr_pending); 2551 out_decl: 2552 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2553 proposal_version); 2554 out_free: 2555 kfree(ini); 2556 kfree(buf); 2557 } 2558 2559 static void smc_tcp_listen_work(struct work_struct *work) 2560 { 2561 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2562 tcp_listen_work); 2563 struct sock *lsk = &lsmc->sk; 2564 struct smc_sock *new_smc; 2565 int rc = 0; 2566 2567 lock_sock(lsk); 2568 while (lsk->sk_state == SMC_LISTEN) { 2569 rc = smc_clcsock_accept(lsmc, &new_smc); 2570 if (rc) /* clcsock accept queue empty or error */ 2571 goto out; 2572 if (!new_smc) 2573 continue; 2574 2575 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2576 atomic_inc(&lsmc->queued_smc_hs); 2577 2578 new_smc->listen_smc = lsmc; 2579 new_smc->use_fallback = lsmc->use_fallback; 2580 new_smc->fallback_rsn = lsmc->fallback_rsn; 2581 sock_hold(lsk); /* sock_put in smc_listen_work */ 2582 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2583 smc_copy_sock_settings_to_smc(new_smc); 2584 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2585 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2586 sock_put(&new_smc->sk); 2587 } 2588 2589 out: 2590 release_sock(lsk); 2591 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2592 } 2593 2594 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2595 { 2596 struct smc_sock *lsmc; 2597 2598 read_lock_bh(&listen_clcsock->sk_callback_lock); 2599 lsmc = smc_clcsock_user_data(listen_clcsock); 2600 if (!lsmc) 2601 goto out; 2602 lsmc->clcsk_data_ready(listen_clcsock); 2603 if (lsmc->sk.sk_state == SMC_LISTEN) { 2604 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2605 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2606 sock_put(&lsmc->sk); 2607 } 2608 out: 2609 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2610 } 2611 2612 int smc_listen(struct socket *sock, int backlog) 2613 { 2614 struct sock *sk = sock->sk; 2615 struct smc_sock *smc; 2616 int rc; 2617 2618 smc = smc_sk(sk); 2619 lock_sock(sk); 2620 2621 rc = -EINVAL; 2622 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2623 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2624 goto out; 2625 2626 rc = 0; 2627 if (sk->sk_state == SMC_LISTEN) { 2628 sk->sk_max_ack_backlog = backlog; 2629 goto out; 2630 } 2631 /* some socket options are handled in core, so we could not apply 2632 * them to the clc socket -- copy smc socket options to clc socket 2633 */ 2634 smc_copy_sock_settings_to_clc(smc); 2635 if (!smc->use_fallback) 2636 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2637 2638 /* save original sk_data_ready function and establish 2639 * smc-specific sk_data_ready function 2640 */ 2641 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2642 smc->clcsock->sk->sk_user_data = 2643 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2644 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2645 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2646 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2647 2648 /* save original ops */ 2649 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2650 2651 smc->af_ops = *smc->ori_af_ops; 2652 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2653 2654 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2655 2656 if (smc->limit_smc_hs) 2657 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2658 2659 rc = kernel_listen(smc->clcsock, backlog); 2660 if (rc) { 2661 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2662 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2663 &smc->clcsk_data_ready); 2664 smc->clcsock->sk->sk_user_data = NULL; 2665 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2666 goto out; 2667 } 2668 sk->sk_max_ack_backlog = backlog; 2669 sk->sk_ack_backlog = 0; 2670 sk->sk_state = SMC_LISTEN; 2671 2672 out: 2673 release_sock(sk); 2674 return rc; 2675 } 2676 2677 int smc_accept(struct socket *sock, struct socket *new_sock, 2678 struct proto_accept_arg *arg) 2679 { 2680 struct sock *sk = sock->sk, *nsk; 2681 DECLARE_WAITQUEUE(wait, current); 2682 struct smc_sock *lsmc; 2683 long timeo; 2684 int rc = 0; 2685 2686 lsmc = smc_sk(sk); 2687 sock_hold(sk); /* sock_put below */ 2688 lock_sock(sk); 2689 2690 if (lsmc->sk.sk_state != SMC_LISTEN) { 2691 rc = -EINVAL; 2692 release_sock(sk); 2693 goto out; 2694 } 2695 2696 /* Wait for an incoming connection */ 2697 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK); 2698 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2699 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2700 set_current_state(TASK_INTERRUPTIBLE); 2701 if (!timeo) { 2702 rc = -EAGAIN; 2703 break; 2704 } 2705 release_sock(sk); 2706 timeo = schedule_timeout(timeo); 2707 /* wakeup by sk_data_ready in smc_listen_work() */ 2708 sched_annotate_sleep(); 2709 lock_sock(sk); 2710 if (signal_pending(current)) { 2711 rc = sock_intr_errno(timeo); 2712 break; 2713 } 2714 } 2715 set_current_state(TASK_RUNNING); 2716 remove_wait_queue(sk_sleep(sk), &wait); 2717 2718 if (!rc) 2719 rc = sock_error(nsk); 2720 release_sock(sk); 2721 if (rc) 2722 goto out; 2723 2724 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) { 2725 /* wait till data arrives on the socket */ 2726 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2727 MSEC_PER_SEC); 2728 if (smc_sk(nsk)->use_fallback) { 2729 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2730 2731 lock_sock(clcsk); 2732 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2733 sk_wait_data(clcsk, &timeo, NULL); 2734 release_sock(clcsk); 2735 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2736 lock_sock(nsk); 2737 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2738 release_sock(nsk); 2739 } 2740 } 2741 2742 out: 2743 sock_put(sk); /* sock_hold above */ 2744 return rc; 2745 } 2746 2747 int smc_getname(struct socket *sock, struct sockaddr *addr, 2748 int peer) 2749 { 2750 struct smc_sock *smc; 2751 2752 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2753 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2754 return -ENOTCONN; 2755 2756 smc = smc_sk(sock->sk); 2757 2758 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2759 } 2760 2761 int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2762 { 2763 struct sock *sk = sock->sk; 2764 struct smc_sock *smc; 2765 int rc; 2766 2767 smc = smc_sk(sk); 2768 lock_sock(sk); 2769 2770 /* SMC does not support connect with fastopen */ 2771 if (msg->msg_flags & MSG_FASTOPEN) { 2772 /* not connected yet, fallback */ 2773 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2774 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2775 if (rc) 2776 goto out; 2777 } else { 2778 rc = -EINVAL; 2779 goto out; 2780 } 2781 } else if ((sk->sk_state != SMC_ACTIVE) && 2782 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2783 (sk->sk_state != SMC_INIT)) { 2784 rc = -EPIPE; 2785 goto out; 2786 } 2787 2788 if (smc->use_fallback) { 2789 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2790 } else { 2791 rc = smc_tx_sendmsg(smc, msg, len); 2792 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2793 } 2794 out: 2795 release_sock(sk); 2796 return rc; 2797 } 2798 2799 int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2800 int flags) 2801 { 2802 struct sock *sk = sock->sk; 2803 struct smc_sock *smc; 2804 int rc = -ENOTCONN; 2805 2806 smc = smc_sk(sk); 2807 lock_sock(sk); 2808 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2809 /* socket was connected before, no more data to read */ 2810 rc = 0; 2811 goto out; 2812 } 2813 if ((sk->sk_state == SMC_INIT) || 2814 (sk->sk_state == SMC_LISTEN) || 2815 (sk->sk_state == SMC_CLOSED)) 2816 goto out; 2817 2818 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2819 rc = 0; 2820 goto out; 2821 } 2822 2823 if (smc->use_fallback) { 2824 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2825 } else { 2826 msg->msg_namelen = 0; 2827 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2828 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2829 } 2830 2831 out: 2832 release_sock(sk); 2833 return rc; 2834 } 2835 2836 static __poll_t smc_accept_poll(struct sock *parent) 2837 { 2838 struct smc_sock *isk = smc_sk(parent); 2839 __poll_t mask = 0; 2840 2841 spin_lock(&isk->accept_q_lock); 2842 if (!list_empty(&isk->accept_q)) 2843 mask = EPOLLIN | EPOLLRDNORM; 2844 spin_unlock(&isk->accept_q_lock); 2845 2846 return mask; 2847 } 2848 2849 __poll_t smc_poll(struct file *file, struct socket *sock, 2850 poll_table *wait) 2851 { 2852 struct sock *sk = sock->sk; 2853 struct smc_sock *smc; 2854 __poll_t mask = 0; 2855 2856 if (!sk) 2857 return EPOLLNVAL; 2858 2859 smc = smc_sk(sock->sk); 2860 if (smc->use_fallback) { 2861 /* delegate to CLC child sock */ 2862 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2863 sk->sk_err = smc->clcsock->sk->sk_err; 2864 } else { 2865 if (sk->sk_state != SMC_CLOSED) 2866 sock_poll_wait(file, sock, wait); 2867 if (sk->sk_err) 2868 mask |= EPOLLERR; 2869 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2870 (sk->sk_state == SMC_CLOSED)) 2871 mask |= EPOLLHUP; 2872 if (sk->sk_state == SMC_LISTEN) { 2873 /* woken up by sk_data_ready in smc_listen_work() */ 2874 mask |= smc_accept_poll(sk); 2875 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2876 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2877 wait); 2878 sk->sk_err = smc->clcsock->sk->sk_err; 2879 } else { 2880 if ((sk->sk_state != SMC_INIT && 2881 atomic_read(&smc->conn.sndbuf_space)) || 2882 sk->sk_shutdown & SEND_SHUTDOWN) { 2883 mask |= EPOLLOUT | EPOLLWRNORM; 2884 } else { 2885 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2886 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2887 } 2888 if (atomic_read(&smc->conn.bytes_to_rcv)) 2889 mask |= EPOLLIN | EPOLLRDNORM; 2890 if (sk->sk_shutdown & RCV_SHUTDOWN) 2891 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2892 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2893 mask |= EPOLLIN; 2894 if (smc->conn.urg_state == SMC_URG_VALID) 2895 mask |= EPOLLPRI; 2896 } 2897 } 2898 2899 return mask; 2900 } 2901 2902 int smc_shutdown(struct socket *sock, int how) 2903 { 2904 struct sock *sk = sock->sk; 2905 bool do_shutdown = true; 2906 struct smc_sock *smc; 2907 int rc = -EINVAL; 2908 int old_state; 2909 int rc1 = 0; 2910 2911 smc = smc_sk(sk); 2912 2913 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2914 return rc; 2915 2916 lock_sock(sk); 2917 2918 if (sock->state == SS_CONNECTING) { 2919 if (sk->sk_state == SMC_ACTIVE) 2920 sock->state = SS_CONNECTED; 2921 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2922 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2923 sk->sk_state == SMC_APPCLOSEWAIT1 || 2924 sk->sk_state == SMC_APPCLOSEWAIT2 || 2925 sk->sk_state == SMC_APPFINCLOSEWAIT) 2926 sock->state = SS_DISCONNECTING; 2927 } 2928 2929 rc = -ENOTCONN; 2930 if ((sk->sk_state != SMC_ACTIVE) && 2931 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2932 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2933 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2934 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2935 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2936 goto out; 2937 if (smc->use_fallback) { 2938 rc = kernel_sock_shutdown(smc->clcsock, how); 2939 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2940 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2941 sk->sk_state = SMC_CLOSED; 2942 sk->sk_socket->state = SS_UNCONNECTED; 2943 sock_put(sk); 2944 } 2945 goto out; 2946 } 2947 switch (how) { 2948 case SHUT_RDWR: /* shutdown in both directions */ 2949 old_state = sk->sk_state; 2950 rc = smc_close_active(smc); 2951 if (old_state == SMC_ACTIVE && 2952 sk->sk_state == SMC_PEERCLOSEWAIT1) 2953 do_shutdown = false; 2954 break; 2955 case SHUT_WR: 2956 rc = smc_close_shutdown_write(smc); 2957 break; 2958 case SHUT_RD: 2959 rc = 0; 2960 /* nothing more to do because peer is not involved */ 2961 break; 2962 } 2963 if (do_shutdown && smc->clcsock) 2964 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2965 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2966 sk->sk_shutdown |= how + 1; 2967 2968 if (sk->sk_state == SMC_CLOSED) 2969 sock->state = SS_UNCONNECTED; 2970 else 2971 sock->state = SS_DISCONNECTING; 2972 out: 2973 release_sock(sk); 2974 return rc ? rc : rc1; 2975 } 2976 2977 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2978 char __user *optval, int __user *optlen) 2979 { 2980 struct smc_sock *smc; 2981 int val, len; 2982 2983 smc = smc_sk(sock->sk); 2984 2985 if (get_user(len, optlen)) 2986 return -EFAULT; 2987 2988 len = min_t(int, len, sizeof(int)); 2989 2990 if (len < 0) 2991 return -EINVAL; 2992 2993 switch (optname) { 2994 case SMC_LIMIT_HS: 2995 val = smc->limit_smc_hs; 2996 break; 2997 default: 2998 return -EOPNOTSUPP; 2999 } 3000 3001 if (put_user(len, optlen)) 3002 return -EFAULT; 3003 if (copy_to_user(optval, &val, len)) 3004 return -EFAULT; 3005 3006 return 0; 3007 } 3008 3009 static int __smc_setsockopt(struct socket *sock, int level, int optname, 3010 sockptr_t optval, unsigned int optlen) 3011 { 3012 struct sock *sk = sock->sk; 3013 struct smc_sock *smc; 3014 int val, rc; 3015 3016 smc = smc_sk(sk); 3017 3018 lock_sock(sk); 3019 switch (optname) { 3020 case SMC_LIMIT_HS: 3021 if (optlen < sizeof(int)) { 3022 rc = -EINVAL; 3023 break; 3024 } 3025 if (copy_from_sockptr(&val, optval, sizeof(int))) { 3026 rc = -EFAULT; 3027 break; 3028 } 3029 3030 smc->limit_smc_hs = !!val; 3031 rc = 0; 3032 break; 3033 default: 3034 rc = -EOPNOTSUPP; 3035 break; 3036 } 3037 release_sock(sk); 3038 3039 return rc; 3040 } 3041 3042 int smc_setsockopt(struct socket *sock, int level, int optname, 3043 sockptr_t optval, unsigned int optlen) 3044 { 3045 struct sock *sk = sock->sk; 3046 struct smc_sock *smc; 3047 int val, rc; 3048 3049 if (level == SOL_TCP && optname == TCP_ULP) 3050 return -EOPNOTSUPP; 3051 else if (level == SOL_SMC) 3052 return __smc_setsockopt(sock, level, optname, optval, optlen); 3053 3054 smc = smc_sk(sk); 3055 3056 /* generic setsockopts reaching us here always apply to the 3057 * CLC socket 3058 */ 3059 mutex_lock(&smc->clcsock_release_lock); 3060 if (!smc->clcsock) { 3061 mutex_unlock(&smc->clcsock_release_lock); 3062 return -EBADF; 3063 } 3064 if (unlikely(!smc->clcsock->ops->setsockopt)) 3065 rc = -EOPNOTSUPP; 3066 else 3067 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3068 optval, optlen); 3069 if (smc->clcsock->sk->sk_err) { 3070 sk->sk_err = smc->clcsock->sk->sk_err; 3071 sk_error_report(sk); 3072 } 3073 mutex_unlock(&smc->clcsock_release_lock); 3074 3075 if (optlen < sizeof(int)) 3076 return -EINVAL; 3077 if (copy_from_sockptr(&val, optval, sizeof(int))) 3078 return -EFAULT; 3079 3080 lock_sock(sk); 3081 if (rc || smc->use_fallback) 3082 goto out; 3083 switch (optname) { 3084 case TCP_FASTOPEN: 3085 case TCP_FASTOPEN_CONNECT: 3086 case TCP_FASTOPEN_KEY: 3087 case TCP_FASTOPEN_NO_COOKIE: 3088 /* option not supported by SMC */ 3089 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3090 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3091 } else { 3092 rc = -EINVAL; 3093 } 3094 break; 3095 case TCP_NODELAY: 3096 if (sk->sk_state != SMC_INIT && 3097 sk->sk_state != SMC_LISTEN && 3098 sk->sk_state != SMC_CLOSED) { 3099 if (val) { 3100 SMC_STAT_INC(smc, ndly_cnt); 3101 smc_tx_pending(&smc->conn); 3102 cancel_delayed_work(&smc->conn.tx_work); 3103 } 3104 } 3105 break; 3106 case TCP_CORK: 3107 if (sk->sk_state != SMC_INIT && 3108 sk->sk_state != SMC_LISTEN && 3109 sk->sk_state != SMC_CLOSED) { 3110 if (!val) { 3111 SMC_STAT_INC(smc, cork_cnt); 3112 smc_tx_pending(&smc->conn); 3113 cancel_delayed_work(&smc->conn.tx_work); 3114 } 3115 } 3116 break; 3117 case TCP_DEFER_ACCEPT: 3118 smc->sockopt_defer_accept = val; 3119 break; 3120 default: 3121 break; 3122 } 3123 out: 3124 release_sock(sk); 3125 3126 return rc; 3127 } 3128 3129 int smc_getsockopt(struct socket *sock, int level, int optname, 3130 char __user *optval, int __user *optlen) 3131 { 3132 struct smc_sock *smc; 3133 int rc; 3134 3135 if (level == SOL_SMC) 3136 return __smc_getsockopt(sock, level, optname, optval, optlen); 3137 3138 smc = smc_sk(sock->sk); 3139 mutex_lock(&smc->clcsock_release_lock); 3140 if (!smc->clcsock) { 3141 mutex_unlock(&smc->clcsock_release_lock); 3142 return -EBADF; 3143 } 3144 /* socket options apply to the CLC socket */ 3145 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3146 mutex_unlock(&smc->clcsock_release_lock); 3147 return -EOPNOTSUPP; 3148 } 3149 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3150 optval, optlen); 3151 mutex_unlock(&smc->clcsock_release_lock); 3152 return rc; 3153 } 3154 3155 int smc_ioctl(struct socket *sock, unsigned int cmd, 3156 unsigned long arg) 3157 { 3158 union smc_host_cursor cons, urg; 3159 struct smc_connection *conn; 3160 struct smc_sock *smc; 3161 int answ; 3162 3163 smc = smc_sk(sock->sk); 3164 conn = &smc->conn; 3165 lock_sock(&smc->sk); 3166 if (smc->use_fallback) { 3167 if (!smc->clcsock) { 3168 release_sock(&smc->sk); 3169 return -EBADF; 3170 } 3171 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3172 release_sock(&smc->sk); 3173 return answ; 3174 } 3175 switch (cmd) { 3176 case SIOCINQ: /* same as FIONREAD */ 3177 if (smc->sk.sk_state == SMC_LISTEN) { 3178 release_sock(&smc->sk); 3179 return -EINVAL; 3180 } 3181 if (smc->sk.sk_state == SMC_INIT || 3182 smc->sk.sk_state == SMC_CLOSED) 3183 answ = 0; 3184 else 3185 answ = atomic_read(&smc->conn.bytes_to_rcv); 3186 break; 3187 case SIOCOUTQ: 3188 /* output queue size (not send + not acked) */ 3189 if (smc->sk.sk_state == SMC_LISTEN) { 3190 release_sock(&smc->sk); 3191 return -EINVAL; 3192 } 3193 if (smc->sk.sk_state == SMC_INIT || 3194 smc->sk.sk_state == SMC_CLOSED) 3195 answ = 0; 3196 else 3197 answ = smc->conn.sndbuf_desc->len - 3198 atomic_read(&smc->conn.sndbuf_space); 3199 break; 3200 case SIOCOUTQNSD: 3201 /* output queue size (not send only) */ 3202 if (smc->sk.sk_state == SMC_LISTEN) { 3203 release_sock(&smc->sk); 3204 return -EINVAL; 3205 } 3206 if (smc->sk.sk_state == SMC_INIT || 3207 smc->sk.sk_state == SMC_CLOSED) 3208 answ = 0; 3209 else 3210 answ = smc_tx_prepared_sends(&smc->conn); 3211 break; 3212 case SIOCATMARK: 3213 if (smc->sk.sk_state == SMC_LISTEN) { 3214 release_sock(&smc->sk); 3215 return -EINVAL; 3216 } 3217 if (smc->sk.sk_state == SMC_INIT || 3218 smc->sk.sk_state == SMC_CLOSED) { 3219 answ = 0; 3220 } else { 3221 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3222 smc_curs_copy(&urg, &conn->urg_curs, conn); 3223 answ = smc_curs_diff(conn->rmb_desc->len, 3224 &cons, &urg) == 1; 3225 } 3226 break; 3227 default: 3228 release_sock(&smc->sk); 3229 return -ENOIOCTLCMD; 3230 } 3231 release_sock(&smc->sk); 3232 3233 return put_user(answ, (int __user *)arg); 3234 } 3235 3236 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3237 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3238 * updates till whenever a respective page has been fully processed. 3239 * Note that subsequent recv() calls have to wait till all splice() processing 3240 * completed. 3241 */ 3242 ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3243 struct pipe_inode_info *pipe, size_t len, 3244 unsigned int flags) 3245 { 3246 struct sock *sk = sock->sk; 3247 struct smc_sock *smc; 3248 int rc = -ENOTCONN; 3249 3250 smc = smc_sk(sk); 3251 lock_sock(sk); 3252 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3253 /* socket was connected before, no more data to read */ 3254 rc = 0; 3255 goto out; 3256 } 3257 if (sk->sk_state == SMC_INIT || 3258 sk->sk_state == SMC_LISTEN || 3259 sk->sk_state == SMC_CLOSED) 3260 goto out; 3261 3262 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3263 rc = 0; 3264 goto out; 3265 } 3266 3267 if (smc->use_fallback) { 3268 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3269 pipe, len, flags); 3270 } else { 3271 if (*ppos) { 3272 rc = -ESPIPE; 3273 goto out; 3274 } 3275 if (flags & SPLICE_F_NONBLOCK) 3276 flags = MSG_DONTWAIT; 3277 else 3278 flags = 0; 3279 SMC_STAT_INC(smc, splice_cnt); 3280 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3281 } 3282 out: 3283 release_sock(sk); 3284 3285 return rc; 3286 } 3287 3288 /* must look like tcp */ 3289 static const struct proto_ops smc_sock_ops = { 3290 .family = PF_SMC, 3291 .owner = THIS_MODULE, 3292 .release = smc_release, 3293 .bind = smc_bind, 3294 .connect = smc_connect, 3295 .socketpair = sock_no_socketpair, 3296 .accept = smc_accept, 3297 .getname = smc_getname, 3298 .poll = smc_poll, 3299 .ioctl = smc_ioctl, 3300 .listen = smc_listen, 3301 .shutdown = smc_shutdown, 3302 .setsockopt = smc_setsockopt, 3303 .getsockopt = smc_getsockopt, 3304 .sendmsg = smc_sendmsg, 3305 .recvmsg = smc_recvmsg, 3306 .mmap = sock_no_mmap, 3307 .splice_read = smc_splice_read, 3308 }; 3309 3310 int smc_create_clcsk(struct net *net, struct sock *sk, int family) 3311 { 3312 struct smc_sock *smc = smc_sk(sk); 3313 int rc; 3314 3315 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3316 &smc->clcsock); 3317 if (rc) 3318 return rc; 3319 3320 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3321 * destruction; its sk_state might not be TCP_CLOSE after 3322 * smc->sk is close()d, and TCP timers can be fired later, 3323 * which need net ref. 3324 */ 3325 sk = smc->clcsock->sk; 3326 __netns_tracker_free(net, &sk->ns_tracker, false); 3327 sk->sk_net_refcnt = 1; 3328 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3329 sock_inuse_add(net, 1); 3330 return 0; 3331 } 3332 3333 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3334 int kern, struct socket *clcsock) 3335 { 3336 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3337 struct smc_sock *smc; 3338 struct sock *sk; 3339 int rc; 3340 3341 rc = -ESOCKTNOSUPPORT; 3342 if (sock->type != SOCK_STREAM) 3343 goto out; 3344 3345 rc = -EPROTONOSUPPORT; 3346 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3347 goto out; 3348 3349 rc = -ENOBUFS; 3350 sock->ops = &smc_sock_ops; 3351 sock->state = SS_UNCONNECTED; 3352 sk = smc_sock_alloc(net, sock, protocol); 3353 if (!sk) 3354 goto out; 3355 3356 /* create internal TCP socket for CLC handshake and fallback */ 3357 smc = smc_sk(sk); 3358 3359 rc = 0; 3360 if (clcsock) 3361 smc->clcsock = clcsock; 3362 else 3363 rc = smc_create_clcsk(net, sk, family); 3364 3365 if (rc) { 3366 sk_common_release(sk); 3367 sock->sk = NULL; 3368 } 3369 out: 3370 return rc; 3371 } 3372 3373 static int smc_create(struct net *net, struct socket *sock, int protocol, 3374 int kern) 3375 { 3376 return __smc_create(net, sock, protocol, kern, NULL); 3377 } 3378 3379 static const struct net_proto_family smc_sock_family_ops = { 3380 .family = PF_SMC, 3381 .owner = THIS_MODULE, 3382 .create = smc_create, 3383 }; 3384 3385 static int smc_ulp_init(struct sock *sk) 3386 { 3387 struct socket *tcp = sk->sk_socket; 3388 struct net *net = sock_net(sk); 3389 struct socket *smcsock; 3390 int protocol, ret; 3391 3392 /* only TCP can be replaced */ 3393 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3394 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3395 return -ESOCKTNOSUPPORT; 3396 /* don't handle wq now */ 3397 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3398 return -ENOTCONN; 3399 3400 if (sk->sk_family == AF_INET) 3401 protocol = SMCPROTO_SMC; 3402 else 3403 protocol = SMCPROTO_SMC6; 3404 3405 smcsock = sock_alloc(); 3406 if (!smcsock) 3407 return -ENFILE; 3408 3409 smcsock->type = SOCK_STREAM; 3410 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3411 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3412 if (ret) { 3413 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3414 return ret; 3415 } 3416 3417 /* replace tcp socket to smc */ 3418 smcsock->file = tcp->file; 3419 smcsock->file->private_data = smcsock; 3420 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3421 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3422 tcp->file = NULL; 3423 3424 return ret; 3425 } 3426 3427 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3428 const gfp_t priority) 3429 { 3430 struct inet_connection_sock *icsk = inet_csk(newsk); 3431 3432 /* don't inherit ulp ops to child when listen */ 3433 icsk->icsk_ulp_ops = NULL; 3434 } 3435 3436 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3437 .name = "smc", 3438 .owner = THIS_MODULE, 3439 .init = smc_ulp_init, 3440 .clone = smc_ulp_clone, 3441 }; 3442 3443 unsigned int smc_net_id; 3444 3445 static __net_init int smc_net_init(struct net *net) 3446 { 3447 int rc; 3448 3449 rc = smc_sysctl_net_init(net); 3450 if (rc) 3451 return rc; 3452 return smc_pnet_net_init(net); 3453 } 3454 3455 static void __net_exit smc_net_exit(struct net *net) 3456 { 3457 smc_sysctl_net_exit(net); 3458 smc_pnet_net_exit(net); 3459 } 3460 3461 static __net_init int smc_net_stat_init(struct net *net) 3462 { 3463 return smc_stats_init(net); 3464 } 3465 3466 static void __net_exit smc_net_stat_exit(struct net *net) 3467 { 3468 smc_stats_exit(net); 3469 } 3470 3471 static struct pernet_operations smc_net_ops = { 3472 .init = smc_net_init, 3473 .exit = smc_net_exit, 3474 .id = &smc_net_id, 3475 .size = sizeof(struct smc_net), 3476 }; 3477 3478 static struct pernet_operations smc_net_stat_ops = { 3479 .init = smc_net_stat_init, 3480 .exit = smc_net_stat_exit, 3481 }; 3482 3483 static int __init smc_init(void) 3484 { 3485 int rc; 3486 3487 rc = register_pernet_subsys(&smc_net_ops); 3488 if (rc) 3489 return rc; 3490 3491 rc = register_pernet_subsys(&smc_net_stat_ops); 3492 if (rc) 3493 goto out_pernet_subsys; 3494 3495 rc = smc_ism_init(); 3496 if (rc) 3497 goto out_pernet_subsys_stat; 3498 smc_clc_init(); 3499 3500 rc = smc_nl_init(); 3501 if (rc) 3502 goto out_ism; 3503 3504 rc = smc_pnet_init(); 3505 if (rc) 3506 goto out_nl; 3507 3508 rc = -ENOMEM; 3509 3510 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3511 if (!smc_tcp_ls_wq) 3512 goto out_pnet; 3513 3514 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3515 if (!smc_hs_wq) 3516 goto out_alloc_tcp_ls_wq; 3517 3518 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3519 if (!smc_close_wq) 3520 goto out_alloc_hs_wq; 3521 3522 rc = smc_core_init(); 3523 if (rc) { 3524 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3525 goto out_alloc_wqs; 3526 } 3527 3528 rc = smc_llc_init(); 3529 if (rc) { 3530 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3531 goto out_core; 3532 } 3533 3534 rc = smc_cdc_init(); 3535 if (rc) { 3536 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3537 goto out_core; 3538 } 3539 3540 rc = proto_register(&smc_proto, 1); 3541 if (rc) { 3542 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3543 goto out_core; 3544 } 3545 3546 rc = proto_register(&smc_proto6, 1); 3547 if (rc) { 3548 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3549 goto out_proto; 3550 } 3551 3552 rc = sock_register(&smc_sock_family_ops); 3553 if (rc) { 3554 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3555 goto out_proto6; 3556 } 3557 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3558 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3559 3560 rc = smc_ib_register_client(); 3561 if (rc) { 3562 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3563 goto out_sock; 3564 } 3565 3566 rc = smc_loopback_init(); 3567 if (rc) { 3568 pr_err("%s: smc_loopback_init fails with %d\n", __func__, rc); 3569 goto out_ib; 3570 } 3571 3572 rc = tcp_register_ulp(&smc_ulp_ops); 3573 if (rc) { 3574 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3575 goto out_lo; 3576 } 3577 rc = smc_inet_init(); 3578 if (rc) { 3579 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc); 3580 goto out_ulp; 3581 } 3582 static_branch_enable(&tcp_have_smc); 3583 return 0; 3584 out_ulp: 3585 tcp_unregister_ulp(&smc_ulp_ops); 3586 out_lo: 3587 smc_loopback_exit(); 3588 out_ib: 3589 smc_ib_unregister_client(); 3590 out_sock: 3591 sock_unregister(PF_SMC); 3592 out_proto6: 3593 proto_unregister(&smc_proto6); 3594 out_proto: 3595 proto_unregister(&smc_proto); 3596 out_core: 3597 smc_core_exit(); 3598 out_alloc_wqs: 3599 destroy_workqueue(smc_close_wq); 3600 out_alloc_hs_wq: 3601 destroy_workqueue(smc_hs_wq); 3602 out_alloc_tcp_ls_wq: 3603 destroy_workqueue(smc_tcp_ls_wq); 3604 out_pnet: 3605 smc_pnet_exit(); 3606 out_nl: 3607 smc_nl_exit(); 3608 out_ism: 3609 smc_clc_exit(); 3610 smc_ism_exit(); 3611 out_pernet_subsys_stat: 3612 unregister_pernet_subsys(&smc_net_stat_ops); 3613 out_pernet_subsys: 3614 unregister_pernet_subsys(&smc_net_ops); 3615 3616 return rc; 3617 } 3618 3619 static void __exit smc_exit(void) 3620 { 3621 static_branch_disable(&tcp_have_smc); 3622 smc_inet_exit(); 3623 tcp_unregister_ulp(&smc_ulp_ops); 3624 sock_unregister(PF_SMC); 3625 smc_core_exit(); 3626 smc_loopback_exit(); 3627 smc_ib_unregister_client(); 3628 smc_ism_exit(); 3629 destroy_workqueue(smc_close_wq); 3630 destroy_workqueue(smc_tcp_ls_wq); 3631 destroy_workqueue(smc_hs_wq); 3632 proto_unregister(&smc_proto6); 3633 proto_unregister(&smc_proto); 3634 smc_pnet_exit(); 3635 smc_nl_exit(); 3636 smc_clc_exit(); 3637 unregister_pernet_subsys(&smc_net_stat_ops); 3638 unregister_pernet_subsys(&smc_net_ops); 3639 rcu_barrier(); 3640 } 3641 3642 module_init(smc_init); 3643 module_exit(smc_exit); 3644 3645 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3646 MODULE_DESCRIPTION("smc socket address family"); 3647 MODULE_LICENSE("GPL"); 3648 MODULE_ALIAS_NETPROTO(PF_SMC); 3649 MODULE_ALIAS_TCP_ULP("smc"); 3650 /* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */ 3651 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1); 3652 #if IS_ENABLED(CONFIG_IPV6) 3653 MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1); 3654 #endif /* CONFIG_IPV6 */ 3655 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3656