1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type 6 * applies to SOCK_STREAM sockets only 7 * offers an alternative communication option for TCP-protocol sockets 8 * applicable with RoCE-cards only 9 * 10 * Initial restrictions: 11 * - support for alternate links postponed 12 * 13 * Copyright IBM Corp. 2016, 2018 14 * 15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 16 * based on prototype from Frank Blaschka 17 */ 18 19 #define KMSG_COMPONENT "smc" 20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 21 22 #include <linux/module.h> 23 #include <linux/socket.h> 24 #include <linux/workqueue.h> 25 #include <linux/in.h> 26 #include <linux/sched/signal.h> 27 #include <linux/if_vlan.h> 28 #include <linux/rcupdate_wait.h> 29 #include <linux/ctype.h> 30 #include <linux/splice.h> 31 32 #include <net/sock.h> 33 #include <net/tcp.h> 34 #include <net/smc.h> 35 #include <asm/ioctls.h> 36 37 #include <net/net_namespace.h> 38 #include <net/netns/generic.h> 39 #include "smc_netns.h" 40 41 #include "smc.h" 42 #include "smc_clc.h" 43 #include "smc_llc.h" 44 #include "smc_cdc.h" 45 #include "smc_core.h" 46 #include "smc_ib.h" 47 #include "smc_ism.h" 48 #include "smc_pnet.h" 49 #include "smc_netlink.h" 50 #include "smc_tx.h" 51 #include "smc_rx.h" 52 #include "smc_close.h" 53 #include "smc_stats.h" 54 #include "smc_tracepoint.h" 55 #include "smc_sysctl.h" 56 57 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 58 * creation on server 59 */ 60 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 61 * creation on client 62 */ 63 64 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 65 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 66 struct workqueue_struct *smc_close_wq; /* wq for close work */ 67 68 static void smc_tcp_listen_work(struct work_struct *); 69 static void smc_connect_work(struct work_struct *); 70 71 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 72 { 73 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 74 void *hdr; 75 76 if (cb_ctx->pos[0]) 77 goto out; 78 79 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 80 &smc_gen_nl_family, NLM_F_MULTI, 81 SMC_NETLINK_DUMP_HS_LIMITATION); 82 if (!hdr) 83 return -ENOMEM; 84 85 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 86 sock_net(skb->sk)->smc.limit_smc_hs)) 87 goto err; 88 89 genlmsg_end(skb, hdr); 90 cb_ctx->pos[0] = 1; 91 out: 92 return skb->len; 93 err: 94 genlmsg_cancel(skb, hdr); 95 return -EMSGSIZE; 96 } 97 98 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 99 { 100 sock_net(skb->sk)->smc.limit_smc_hs = true; 101 return 0; 102 } 103 104 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 105 { 106 sock_net(skb->sk)->smc.limit_smc_hs = false; 107 return 0; 108 } 109 110 static void smc_set_keepalive(struct sock *sk, int val) 111 { 112 struct smc_sock *smc = smc_sk(sk); 113 114 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 115 } 116 117 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 118 struct sk_buff *skb, 119 struct request_sock *req, 120 struct dst_entry *dst, 121 struct request_sock *req_unhash, 122 bool *own_req) 123 { 124 struct smc_sock *smc; 125 struct sock *child; 126 127 smc = smc_clcsock_user_data(sk); 128 129 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 130 sk->sk_max_ack_backlog) 131 goto drop; 132 133 if (sk_acceptq_is_full(&smc->sk)) { 134 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 135 goto drop; 136 } 137 138 /* passthrough to original syn recv sock fct */ 139 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 140 own_req); 141 /* child must not inherit smc or its ops */ 142 if (child) { 143 rcu_assign_sk_user_data(child, NULL); 144 145 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 146 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 147 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 148 } 149 return child; 150 151 drop: 152 dst_release(dst); 153 tcp_listendrop(sk); 154 return NULL; 155 } 156 157 static bool smc_hs_congested(const struct sock *sk) 158 { 159 const struct smc_sock *smc; 160 161 smc = smc_clcsock_user_data(sk); 162 163 if (!smc) 164 return true; 165 166 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 167 return true; 168 169 return false; 170 } 171 172 static struct smc_hashinfo smc_v4_hashinfo = { 173 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 174 }; 175 176 static struct smc_hashinfo smc_v6_hashinfo = { 177 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 178 }; 179 180 int smc_hash_sk(struct sock *sk) 181 { 182 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 183 struct hlist_head *head; 184 185 head = &h->ht; 186 187 write_lock_bh(&h->lock); 188 sk_add_node(sk, head); 189 write_unlock_bh(&h->lock); 190 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 191 192 return 0; 193 } 194 EXPORT_SYMBOL_GPL(smc_hash_sk); 195 196 void smc_unhash_sk(struct sock *sk) 197 { 198 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 199 200 write_lock_bh(&h->lock); 201 if (sk_del_node_init(sk)) 202 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 203 write_unlock_bh(&h->lock); 204 } 205 EXPORT_SYMBOL_GPL(smc_unhash_sk); 206 207 /* This will be called before user really release sock_lock. So do the 208 * work which we didn't do because of user hold the sock_lock in the 209 * BH context 210 */ 211 static void smc_release_cb(struct sock *sk) 212 { 213 struct smc_sock *smc = smc_sk(sk); 214 215 if (smc->conn.tx_in_release_sock) { 216 smc_tx_pending(&smc->conn); 217 smc->conn.tx_in_release_sock = false; 218 } 219 } 220 221 struct proto smc_proto = { 222 .name = "SMC", 223 .owner = THIS_MODULE, 224 .keepalive = smc_set_keepalive, 225 .hash = smc_hash_sk, 226 .unhash = smc_unhash_sk, 227 .release_cb = smc_release_cb, 228 .obj_size = sizeof(struct smc_sock), 229 .h.smc_hash = &smc_v4_hashinfo, 230 .slab_flags = SLAB_TYPESAFE_BY_RCU, 231 }; 232 EXPORT_SYMBOL_GPL(smc_proto); 233 234 struct proto smc_proto6 = { 235 .name = "SMC6", 236 .owner = THIS_MODULE, 237 .keepalive = smc_set_keepalive, 238 .hash = smc_hash_sk, 239 .unhash = smc_unhash_sk, 240 .release_cb = smc_release_cb, 241 .obj_size = sizeof(struct smc_sock), 242 .h.smc_hash = &smc_v6_hashinfo, 243 .slab_flags = SLAB_TYPESAFE_BY_RCU, 244 }; 245 EXPORT_SYMBOL_GPL(smc_proto6); 246 247 static void smc_fback_restore_callbacks(struct smc_sock *smc) 248 { 249 struct sock *clcsk = smc->clcsock->sk; 250 251 write_lock_bh(&clcsk->sk_callback_lock); 252 clcsk->sk_user_data = NULL; 253 254 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 255 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 256 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 257 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 258 259 write_unlock_bh(&clcsk->sk_callback_lock); 260 } 261 262 static void smc_restore_fallback_changes(struct smc_sock *smc) 263 { 264 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 265 smc->clcsock->file->private_data = smc->sk.sk_socket; 266 smc->clcsock->file = NULL; 267 smc_fback_restore_callbacks(smc); 268 } 269 } 270 271 static int __smc_release(struct smc_sock *smc) 272 { 273 struct sock *sk = &smc->sk; 274 int rc = 0; 275 276 if (!smc->use_fallback) { 277 rc = smc_close_active(smc); 278 smc_sock_set_flag(sk, SOCK_DEAD); 279 sk->sk_shutdown |= SHUTDOWN_MASK; 280 } else { 281 if (sk->sk_state != SMC_CLOSED) { 282 if (sk->sk_state != SMC_LISTEN && 283 sk->sk_state != SMC_INIT) 284 sock_put(sk); /* passive closing */ 285 if (sk->sk_state == SMC_LISTEN) { 286 /* wake up clcsock accept */ 287 rc = kernel_sock_shutdown(smc->clcsock, 288 SHUT_RDWR); 289 } 290 sk->sk_state = SMC_CLOSED; 291 sk->sk_state_change(sk); 292 } 293 smc_restore_fallback_changes(smc); 294 } 295 296 sk->sk_prot->unhash(sk); 297 298 if (sk->sk_state == SMC_CLOSED) { 299 if (smc->clcsock) { 300 release_sock(sk); 301 smc_clcsock_release(smc); 302 lock_sock(sk); 303 } 304 if (!smc->use_fallback) 305 smc_conn_free(&smc->conn); 306 } 307 308 return rc; 309 } 310 311 static int smc_release(struct socket *sock) 312 { 313 struct sock *sk = sock->sk; 314 struct smc_sock *smc; 315 int old_state, rc = 0; 316 317 if (!sk) 318 goto out; 319 320 sock_hold(sk); /* sock_put below */ 321 smc = smc_sk(sk); 322 323 old_state = sk->sk_state; 324 325 /* cleanup for a dangling non-blocking connect */ 326 if (smc->connect_nonblock && old_state == SMC_INIT) 327 tcp_abort(smc->clcsock->sk, ECONNABORTED); 328 329 if (cancel_work_sync(&smc->connect_work)) 330 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 331 332 if (sk->sk_state == SMC_LISTEN) 333 /* smc_close_non_accepted() is called and acquires 334 * sock lock for child sockets again 335 */ 336 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 337 else 338 lock_sock(sk); 339 340 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 341 !smc->use_fallback) 342 smc_close_active_abort(smc); 343 344 rc = __smc_release(smc); 345 346 /* detach socket */ 347 sock_orphan(sk); 348 sock->sk = NULL; 349 release_sock(sk); 350 351 sock_put(sk); /* sock_hold above */ 352 sock_put(sk); /* final sock_put */ 353 out: 354 return rc; 355 } 356 357 static void smc_destruct(struct sock *sk) 358 { 359 if (sk->sk_state != SMC_CLOSED) 360 return; 361 if (!sock_flag(sk, SOCK_DEAD)) 362 return; 363 } 364 365 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 366 int protocol) 367 { 368 struct smc_sock *smc; 369 struct proto *prot; 370 struct sock *sk; 371 372 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 373 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 374 if (!sk) 375 return NULL; 376 377 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 378 sk->sk_state = SMC_INIT; 379 sk->sk_destruct = smc_destruct; 380 sk->sk_protocol = protocol; 381 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem)); 382 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem)); 383 smc = smc_sk(sk); 384 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 385 INIT_WORK(&smc->connect_work, smc_connect_work); 386 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 387 INIT_LIST_HEAD(&smc->accept_q); 388 spin_lock_init(&smc->accept_q_lock); 389 spin_lock_init(&smc->conn.send_lock); 390 sk->sk_prot->hash(sk); 391 mutex_init(&smc->clcsock_release_lock); 392 smc_init_saved_callbacks(smc); 393 394 return sk; 395 } 396 397 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 398 int addr_len) 399 { 400 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 401 struct sock *sk = sock->sk; 402 struct smc_sock *smc; 403 int rc; 404 405 smc = smc_sk(sk); 406 407 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 408 rc = -EINVAL; 409 if (addr_len < sizeof(struct sockaddr_in)) 410 goto out; 411 412 rc = -EAFNOSUPPORT; 413 if (addr->sin_family != AF_INET && 414 addr->sin_family != AF_INET6 && 415 addr->sin_family != AF_UNSPEC) 416 goto out; 417 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 418 if (addr->sin_family == AF_UNSPEC && 419 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 420 goto out; 421 422 lock_sock(sk); 423 424 /* Check if socket is already active */ 425 rc = -EINVAL; 426 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 427 goto out_rel; 428 429 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 430 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport; 431 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 432 433 out_rel: 434 release_sock(sk); 435 out: 436 return rc; 437 } 438 439 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 440 * clc socket (since smc is not called for these options from net/core) 441 */ 442 443 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 444 (1UL << SOCK_KEEPOPEN) | \ 445 (1UL << SOCK_LINGER) | \ 446 (1UL << SOCK_BROADCAST) | \ 447 (1UL << SOCK_TIMESTAMP) | \ 448 (1UL << SOCK_DBG) | \ 449 (1UL << SOCK_RCVTSTAMP) | \ 450 (1UL << SOCK_RCVTSTAMPNS) | \ 451 (1UL << SOCK_LOCALROUTE) | \ 452 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 453 (1UL << SOCK_RXQ_OVFL) | \ 454 (1UL << SOCK_WIFI_STATUS) | \ 455 (1UL << SOCK_NOFCS) | \ 456 (1UL << SOCK_FILTER_LOCKED) | \ 457 (1UL << SOCK_TSTAMP_NEW)) 458 459 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */ 460 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk, 461 unsigned long mask) 462 { 463 nsk->sk_userlocks = osk->sk_userlocks; 464 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) 465 nsk->sk_sndbuf = osk->sk_sndbuf; 466 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) 467 nsk->sk_rcvbuf = osk->sk_rcvbuf; 468 } 469 470 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 471 unsigned long mask) 472 { 473 /* options we don't get control via setsockopt for */ 474 nsk->sk_type = osk->sk_type; 475 nsk->sk_sndtimeo = osk->sk_sndtimeo; 476 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 477 nsk->sk_mark = READ_ONCE(osk->sk_mark); 478 nsk->sk_priority = osk->sk_priority; 479 nsk->sk_rcvlowat = osk->sk_rcvlowat; 480 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 481 nsk->sk_err = osk->sk_err; 482 483 nsk->sk_flags &= ~mask; 484 nsk->sk_flags |= osk->sk_flags & mask; 485 486 smc_adjust_sock_bufsizes(nsk, osk, mask); 487 } 488 489 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 490 { 491 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 492 } 493 494 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 495 (1UL << SOCK_KEEPOPEN) | \ 496 (1UL << SOCK_LINGER) | \ 497 (1UL << SOCK_DBG)) 498 /* copy only settings and flags relevant for smc from clc to smc socket */ 499 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 500 { 501 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 502 } 503 504 /* register the new vzalloced sndbuf on all links */ 505 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 506 struct smc_buf_desc *snd_desc) 507 { 508 struct smc_link_group *lgr = link->lgr; 509 int i, rc = 0; 510 511 if (!snd_desc->is_vm) 512 return -EINVAL; 513 514 /* protect against parallel smcr_link_reg_buf() */ 515 down_write(&lgr->llc_conf_mutex); 516 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 517 if (!smc_link_active(&lgr->lnk[i])) 518 continue; 519 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 520 if (rc) 521 break; 522 } 523 up_write(&lgr->llc_conf_mutex); 524 return rc; 525 } 526 527 /* register the new rmb on all links */ 528 static int smcr_lgr_reg_rmbs(struct smc_link *link, 529 struct smc_buf_desc *rmb_desc) 530 { 531 struct smc_link_group *lgr = link->lgr; 532 bool do_slow = false; 533 int i, rc = 0; 534 535 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 536 if (rc) 537 return rc; 538 539 down_read(&lgr->llc_conf_mutex); 540 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 541 if (!smc_link_active(&lgr->lnk[i])) 542 continue; 543 if (!rmb_desc->is_reg_mr[link->link_idx]) { 544 up_read(&lgr->llc_conf_mutex); 545 goto slow_path; 546 } 547 } 548 /* mr register already */ 549 goto fast_path; 550 slow_path: 551 do_slow = true; 552 /* protect against parallel smc_llc_cli_rkey_exchange() and 553 * parallel smcr_link_reg_buf() 554 */ 555 down_write(&lgr->llc_conf_mutex); 556 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 557 if (!smc_link_active(&lgr->lnk[i])) 558 continue; 559 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 560 if (rc) 561 goto out; 562 } 563 fast_path: 564 /* exchange confirm_rkey msg with peer */ 565 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 566 if (rc) { 567 rc = -EFAULT; 568 goto out; 569 } 570 rmb_desc->is_conf_rkey = true; 571 out: 572 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex); 573 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 574 return rc; 575 } 576 577 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 578 { 579 struct smc_link *link = smc->conn.lnk; 580 struct smc_llc_qentry *qentry; 581 int rc; 582 583 /* Receive CONFIRM LINK request from server over RoCE fabric. 584 * Increasing the client's timeout by twice as much as the server's 585 * timeout by default can temporarily avoid decline messages of 586 * both sides crossing or colliding 587 */ 588 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME, 589 SMC_LLC_CONFIRM_LINK); 590 if (!qentry) { 591 struct smc_clc_msg_decline dclc; 592 593 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 594 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 595 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 596 } 597 smc_llc_save_peer_uid(qentry); 598 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 599 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 600 if (rc) 601 return SMC_CLC_DECL_RMBE_EC; 602 603 rc = smc_ib_modify_qp_rts(link); 604 if (rc) 605 return SMC_CLC_DECL_ERR_RDYLNK; 606 607 smc_wr_remember_qp_attr(link); 608 609 /* reg the sndbuf if it was vzalloced */ 610 if (smc->conn.sndbuf_desc->is_vm) { 611 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 612 return SMC_CLC_DECL_ERR_REGBUF; 613 } 614 615 /* reg the rmb */ 616 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 617 return SMC_CLC_DECL_ERR_REGBUF; 618 619 /* confirm_rkey is implicit on 1st contact */ 620 smc->conn.rmb_desc->is_conf_rkey = true; 621 622 /* send CONFIRM LINK response over RoCE fabric */ 623 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 624 if (rc < 0) 625 return SMC_CLC_DECL_TIMEOUT_CL; 626 627 smc_llc_link_active(link); 628 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 629 630 if (link->lgr->max_links > 1) { 631 /* optional 2nd link, receive ADD LINK request from server */ 632 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 633 SMC_LLC_ADD_LINK); 634 if (!qentry) { 635 struct smc_clc_msg_decline dclc; 636 637 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 638 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 639 if (rc == -EAGAIN) 640 rc = 0; /* no DECLINE received, go with one link */ 641 return rc; 642 } 643 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 644 smc_llc_cli_add_link(link, qentry); 645 } 646 return 0; 647 } 648 649 static bool smc_isascii(char *hostname) 650 { 651 int i; 652 653 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 654 if (!isascii(hostname[i])) 655 return false; 656 return true; 657 } 658 659 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 660 struct smc_clc_msg_accept_confirm *clc) 661 { 662 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 663 (struct smc_clc_msg_accept_confirm_v2 *)clc; 664 struct smc_clc_first_contact_ext *fce; 665 int clc_v2_len; 666 667 if (clc->hdr.version == SMC_V1 || 668 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 669 return; 670 671 if (smc->conn.lgr->is_smcd) { 672 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 673 SMC_MAX_EID_LEN); 674 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 675 d1); 676 } else { 677 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 678 SMC_MAX_EID_LEN); 679 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 680 r1); 681 } 682 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 683 smc->conn.lgr->peer_os = fce->os_type; 684 smc->conn.lgr->peer_smc_release = fce->release; 685 if (smc_isascii(fce->hostname)) 686 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 687 SMC_MAX_HOSTNAME_LEN); 688 } 689 690 static void smcr_conn_save_peer_info(struct smc_sock *smc, 691 struct smc_clc_msg_accept_confirm *clc) 692 { 693 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 694 695 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 696 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 697 smc->conn.peer_rmbe_size = bufsize; 698 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 699 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 700 } 701 702 static void smcd_conn_save_peer_info(struct smc_sock *smc, 703 struct smc_clc_msg_accept_confirm *clc) 704 { 705 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 706 707 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 708 smc->conn.peer_token = ntohll(clc->d0.token); 709 /* msg header takes up space in the buffer */ 710 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 711 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 712 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 713 } 714 715 static void smc_conn_save_peer_info(struct smc_sock *smc, 716 struct smc_clc_msg_accept_confirm *clc) 717 { 718 if (smc->conn.lgr->is_smcd) 719 smcd_conn_save_peer_info(smc, clc); 720 else 721 smcr_conn_save_peer_info(smc, clc); 722 smc_conn_save_peer_info_fce(smc, clc); 723 } 724 725 static void smc_link_save_peer_info(struct smc_link *link, 726 struct smc_clc_msg_accept_confirm *clc, 727 struct smc_init_info *ini) 728 { 729 link->peer_qpn = ntoh24(clc->r0.qpn); 730 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 731 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 732 link->peer_psn = ntoh24(clc->r0.psn); 733 link->peer_mtu = clc->r0.qp_mtu; 734 } 735 736 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 737 struct smc_stats_fback *fback_arr) 738 { 739 int cnt; 740 741 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 742 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 743 fback_arr[cnt].count++; 744 break; 745 } 746 if (!fback_arr[cnt].fback_code) { 747 fback_arr[cnt].fback_code = smc->fallback_rsn; 748 fback_arr[cnt].count++; 749 break; 750 } 751 } 752 } 753 754 static void smc_stat_fallback(struct smc_sock *smc) 755 { 756 struct net *net = sock_net(&smc->sk); 757 758 mutex_lock(&net->smc.mutex_fback_rsn); 759 if (smc->listen_smc) { 760 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 761 net->smc.fback_rsn->srv_fback_cnt++; 762 } else { 763 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 764 net->smc.fback_rsn->clnt_fback_cnt++; 765 } 766 mutex_unlock(&net->smc.mutex_fback_rsn); 767 } 768 769 /* must be called under rcu read lock */ 770 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 771 { 772 struct socket_wq *wq; 773 __poll_t flags; 774 775 wq = rcu_dereference(smc->sk.sk_wq); 776 if (!skwq_has_sleeper(wq)) 777 return; 778 779 /* wake up smc sk->sk_wq */ 780 if (!key) { 781 /* sk_state_change */ 782 wake_up_interruptible_all(&wq->wait); 783 } else { 784 flags = key_to_poll(key); 785 if (flags & (EPOLLIN | EPOLLOUT)) 786 /* sk_data_ready or sk_write_space */ 787 wake_up_interruptible_sync_poll(&wq->wait, flags); 788 else if (flags & EPOLLERR) 789 /* sk_error_report */ 790 wake_up_interruptible_poll(&wq->wait, flags); 791 } 792 } 793 794 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 795 unsigned int mode, int sync, void *key) 796 { 797 struct smc_mark_woken *mark = 798 container_of(wait, struct smc_mark_woken, wait_entry); 799 800 mark->woken = true; 801 mark->key = key; 802 return 0; 803 } 804 805 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 806 void (*clcsock_callback)(struct sock *sk)) 807 { 808 struct smc_mark_woken mark = { .woken = false }; 809 struct socket_wq *wq; 810 811 init_waitqueue_func_entry(&mark.wait_entry, 812 smc_fback_mark_woken); 813 rcu_read_lock(); 814 wq = rcu_dereference(clcsk->sk_wq); 815 if (!wq) 816 goto out; 817 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 818 clcsock_callback(clcsk); 819 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 820 821 if (mark.woken) 822 smc_fback_wakeup_waitqueue(smc, mark.key); 823 out: 824 rcu_read_unlock(); 825 } 826 827 static void smc_fback_state_change(struct sock *clcsk) 828 { 829 struct smc_sock *smc; 830 831 read_lock_bh(&clcsk->sk_callback_lock); 832 smc = smc_clcsock_user_data(clcsk); 833 if (smc) 834 smc_fback_forward_wakeup(smc, clcsk, 835 smc->clcsk_state_change); 836 read_unlock_bh(&clcsk->sk_callback_lock); 837 } 838 839 static void smc_fback_data_ready(struct sock *clcsk) 840 { 841 struct smc_sock *smc; 842 843 read_lock_bh(&clcsk->sk_callback_lock); 844 smc = smc_clcsock_user_data(clcsk); 845 if (smc) 846 smc_fback_forward_wakeup(smc, clcsk, 847 smc->clcsk_data_ready); 848 read_unlock_bh(&clcsk->sk_callback_lock); 849 } 850 851 static void smc_fback_write_space(struct sock *clcsk) 852 { 853 struct smc_sock *smc; 854 855 read_lock_bh(&clcsk->sk_callback_lock); 856 smc = smc_clcsock_user_data(clcsk); 857 if (smc) 858 smc_fback_forward_wakeup(smc, clcsk, 859 smc->clcsk_write_space); 860 read_unlock_bh(&clcsk->sk_callback_lock); 861 } 862 863 static void smc_fback_error_report(struct sock *clcsk) 864 { 865 struct smc_sock *smc; 866 867 read_lock_bh(&clcsk->sk_callback_lock); 868 smc = smc_clcsock_user_data(clcsk); 869 if (smc) 870 smc_fback_forward_wakeup(smc, clcsk, 871 smc->clcsk_error_report); 872 read_unlock_bh(&clcsk->sk_callback_lock); 873 } 874 875 static void smc_fback_replace_callbacks(struct smc_sock *smc) 876 { 877 struct sock *clcsk = smc->clcsock->sk; 878 879 write_lock_bh(&clcsk->sk_callback_lock); 880 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 881 882 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 883 &smc->clcsk_state_change); 884 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 885 &smc->clcsk_data_ready); 886 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 887 &smc->clcsk_write_space); 888 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 889 &smc->clcsk_error_report); 890 891 write_unlock_bh(&clcsk->sk_callback_lock); 892 } 893 894 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 895 { 896 int rc = 0; 897 898 mutex_lock(&smc->clcsock_release_lock); 899 if (!smc->clcsock) { 900 rc = -EBADF; 901 goto out; 902 } 903 904 smc->use_fallback = true; 905 smc->fallback_rsn = reason_code; 906 smc_stat_fallback(smc); 907 trace_smc_switch_to_fallback(smc, reason_code); 908 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 909 smc->clcsock->file = smc->sk.sk_socket->file; 910 smc->clcsock->file->private_data = smc->clcsock; 911 smc->clcsock->wq.fasync_list = 912 smc->sk.sk_socket->wq.fasync_list; 913 914 /* There might be some wait entries remaining 915 * in smc sk->sk_wq and they should be woken up 916 * as clcsock's wait queue is woken up. 917 */ 918 smc_fback_replace_callbacks(smc); 919 } 920 out: 921 mutex_unlock(&smc->clcsock_release_lock); 922 return rc; 923 } 924 925 /* fall back during connect */ 926 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 927 { 928 struct net *net = sock_net(&smc->sk); 929 int rc = 0; 930 931 rc = smc_switch_to_fallback(smc, reason_code); 932 if (rc) { /* fallback fails */ 933 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 934 if (smc->sk.sk_state == SMC_INIT) 935 sock_put(&smc->sk); /* passive closing */ 936 return rc; 937 } 938 smc_copy_sock_settings_to_clc(smc); 939 smc->connect_nonblock = 0; 940 if (smc->sk.sk_state == SMC_INIT) 941 smc->sk.sk_state = SMC_ACTIVE; 942 return 0; 943 } 944 945 /* decline and fall back during connect */ 946 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 947 u8 version) 948 { 949 struct net *net = sock_net(&smc->sk); 950 int rc; 951 952 if (reason_code < 0) { /* error, fallback is not possible */ 953 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 954 if (smc->sk.sk_state == SMC_INIT) 955 sock_put(&smc->sk); /* passive closing */ 956 return reason_code; 957 } 958 if (reason_code != SMC_CLC_DECL_PEERDECL) { 959 rc = smc_clc_send_decline(smc, reason_code, version); 960 if (rc < 0) { 961 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 962 if (smc->sk.sk_state == SMC_INIT) 963 sock_put(&smc->sk); /* passive closing */ 964 return rc; 965 } 966 } 967 return smc_connect_fallback(smc, reason_code); 968 } 969 970 static void smc_conn_abort(struct smc_sock *smc, int local_first) 971 { 972 struct smc_connection *conn = &smc->conn; 973 struct smc_link_group *lgr = conn->lgr; 974 bool lgr_valid = false; 975 976 if (smc_conn_lgr_valid(conn)) 977 lgr_valid = true; 978 979 smc_conn_free(conn); 980 if (local_first && lgr_valid) 981 smc_lgr_cleanup_early(lgr); 982 } 983 984 /* check if there is a rdma device available for this connection. */ 985 /* called for connect and listen */ 986 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 987 { 988 /* PNET table look up: search active ib_device and port 989 * within same PNETID that also contains the ethernet device 990 * used for the internal TCP socket 991 */ 992 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 993 if (!ini->check_smcrv2 && !ini->ib_dev) 994 return SMC_CLC_DECL_NOSMCRDEV; 995 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 996 return SMC_CLC_DECL_NOSMCRDEV; 997 return 0; 998 } 999 1000 /* check if there is an ISM device available for this connection. */ 1001 /* called for connect and listen */ 1002 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 1003 { 1004 /* Find ISM device with same PNETID as connecting interface */ 1005 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 1006 if (!ini->ism_dev[0]) 1007 return SMC_CLC_DECL_NOSMCDDEV; 1008 else 1009 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 1010 return 0; 1011 } 1012 1013 /* is chid unique for the ism devices that are already determined? */ 1014 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 1015 int cnt) 1016 { 1017 int i = (!ini->ism_dev[0]) ? 1 : 0; 1018 1019 for (; i < cnt; i++) 1020 if (ini->ism_chid[i] == chid) 1021 return false; 1022 return true; 1023 } 1024 1025 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 1026 * PNETID matching net_device) 1027 */ 1028 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 1029 struct smc_init_info *ini) 1030 { 1031 int rc = SMC_CLC_DECL_NOSMCDDEV; 1032 struct smcd_dev *smcd; 1033 int i = 1; 1034 u16 chid; 1035 1036 if (smcd_indicated(ini->smc_type_v1)) 1037 rc = 0; /* already initialized for V1 */ 1038 mutex_lock(&smcd_dev_list.mutex); 1039 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1040 if (smcd->going_away || smcd == ini->ism_dev[0]) 1041 continue; 1042 chid = smc_ism_get_chid(smcd); 1043 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1044 continue; 1045 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1046 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1047 ini->ism_dev[i] = smcd; 1048 ini->ism_chid[i] = chid; 1049 ini->is_smcd = true; 1050 rc = 0; 1051 i++; 1052 if (i > SMC_MAX_ISM_DEVS) 1053 break; 1054 } 1055 } 1056 mutex_unlock(&smcd_dev_list.mutex); 1057 ini->ism_offered_cnt = i - 1; 1058 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1059 ini->smcd_version = 0; 1060 1061 return rc; 1062 } 1063 1064 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1065 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1066 struct smc_init_info *ini) 1067 { 1068 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1069 return SMC_CLC_DECL_ISMVLANERR; 1070 return 0; 1071 } 1072 1073 static int smc_find_proposal_devices(struct smc_sock *smc, 1074 struct smc_init_info *ini) 1075 { 1076 int rc = 0; 1077 1078 /* check if there is an ism device available */ 1079 if (!(ini->smcd_version & SMC_V1) || 1080 smc_find_ism_device(smc, ini) || 1081 smc_connect_ism_vlan_setup(smc, ini)) 1082 ini->smcd_version &= ~SMC_V1; 1083 /* else ISM V1 is supported for this connection */ 1084 1085 /* check if there is an rdma device available */ 1086 if (!(ini->smcr_version & SMC_V1) || 1087 smc_find_rdma_device(smc, ini)) 1088 ini->smcr_version &= ~SMC_V1; 1089 /* else RDMA is supported for this connection */ 1090 1091 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1092 ini->smcr_version & SMC_V1); 1093 1094 /* check if there is an ism v2 device available */ 1095 if (!(ini->smcd_version & SMC_V2) || 1096 !smc_ism_is_v2_capable() || 1097 smc_find_ism_v2_device_clnt(smc, ini)) 1098 ini->smcd_version &= ~SMC_V2; 1099 1100 /* check if there is an rdma v2 device available */ 1101 ini->check_smcrv2 = true; 1102 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1103 if (!(ini->smcr_version & SMC_V2) || 1104 smc->clcsock->sk->sk_family != AF_INET || 1105 !smc_clc_ueid_count() || 1106 smc_find_rdma_device(smc, ini)) 1107 ini->smcr_version &= ~SMC_V2; 1108 ini->check_smcrv2 = false; 1109 1110 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1111 ini->smcr_version & SMC_V2); 1112 1113 /* if neither ISM nor RDMA are supported, fallback */ 1114 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1115 rc = SMC_CLC_DECL_NOSMCDEV; 1116 1117 return rc; 1118 } 1119 1120 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1121 * used, the VLAN ID will be registered again during the connection setup. 1122 */ 1123 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1124 struct smc_init_info *ini) 1125 { 1126 if (!smcd_indicated(ini->smc_type_v1)) 1127 return 0; 1128 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1129 return SMC_CLC_DECL_CNFERR; 1130 return 0; 1131 } 1132 1133 #define SMC_CLC_MAX_ACCEPT_LEN \ 1134 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 1135 sizeof(struct smc_clc_first_contact_ext_v2x) + \ 1136 sizeof(struct smc_clc_msg_trail)) 1137 1138 /* CLC handshake during connect */ 1139 static int smc_connect_clc(struct smc_sock *smc, 1140 struct smc_clc_msg_accept_confirm_v2 *aclc2, 1141 struct smc_init_info *ini) 1142 { 1143 int rc = 0; 1144 1145 /* do inband token exchange */ 1146 rc = smc_clc_send_proposal(smc, ini); 1147 if (rc) 1148 return rc; 1149 /* receive SMC Accept CLC message */ 1150 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 1151 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1152 } 1153 1154 void smc_fill_gid_list(struct smc_link_group *lgr, 1155 struct smc_gidlist *gidlist, 1156 struct smc_ib_device *known_dev, u8 *known_gid) 1157 { 1158 struct smc_init_info *alt_ini = NULL; 1159 1160 memset(gidlist, 0, sizeof(*gidlist)); 1161 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1162 1163 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1164 if (!alt_ini) 1165 goto out; 1166 1167 alt_ini->vlan_id = lgr->vlan_id; 1168 alt_ini->check_smcrv2 = true; 1169 alt_ini->smcrv2.saddr = lgr->saddr; 1170 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1171 1172 if (!alt_ini->smcrv2.ib_dev_v2) 1173 goto out; 1174 1175 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1176 SMC_GID_SIZE); 1177 1178 out: 1179 kfree(alt_ini); 1180 } 1181 1182 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1183 struct smc_clc_msg_accept_confirm *aclc, 1184 struct smc_init_info *ini) 1185 { 1186 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1187 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1188 struct smc_clc_first_contact_ext *fce = 1189 smc_get_clc_first_contact_ext(clc_v2, false); 1190 struct net *net = sock_net(&smc->sk); 1191 int rc; 1192 1193 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1194 return 0; 1195 1196 if (fce->v2_direct) { 1197 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1198 ini->smcrv2.uses_gateway = false; 1199 } else { 1200 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr, 1201 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1202 ini->smcrv2.nexthop_mac, 1203 &ini->smcrv2.uses_gateway)) 1204 return SMC_CLC_DECL_NOROUTE; 1205 if (!ini->smcrv2.uses_gateway) { 1206 /* mismatch: peer claims indirect, but its direct */ 1207 return SMC_CLC_DECL_NOINDIRECT; 1208 } 1209 } 1210 1211 ini->release_nr = fce->release; 1212 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1213 if (rc) 1214 return rc; 1215 1216 return 0; 1217 } 1218 1219 /* setup for RDMA connection of client */ 1220 static int smc_connect_rdma(struct smc_sock *smc, 1221 struct smc_clc_msg_accept_confirm *aclc, 1222 struct smc_init_info *ini) 1223 { 1224 int i, reason_code = 0; 1225 struct smc_link *link; 1226 u8 *eid = NULL; 1227 1228 ini->is_smcd = false; 1229 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1230 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1231 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1232 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1233 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1234 ini->max_conns = SMC_CONN_PER_LGR_MAX; 1235 ini->max_links = SMC_LINKS_ADD_LNK_MAX; 1236 1237 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1238 if (reason_code) 1239 return reason_code; 1240 1241 mutex_lock(&smc_client_lgr_pending); 1242 reason_code = smc_conn_create(smc, ini); 1243 if (reason_code) { 1244 mutex_unlock(&smc_client_lgr_pending); 1245 return reason_code; 1246 } 1247 1248 smc_conn_save_peer_info(smc, aclc); 1249 1250 if (ini->first_contact_local) { 1251 link = smc->conn.lnk; 1252 } else { 1253 /* set link that was assigned by server */ 1254 link = NULL; 1255 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1256 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1257 1258 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1259 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1260 SMC_GID_SIZE) && 1261 (aclc->hdr.version > SMC_V1 || 1262 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1263 sizeof(l->peer_mac)))) { 1264 link = l; 1265 break; 1266 } 1267 } 1268 if (!link) { 1269 reason_code = SMC_CLC_DECL_NOSRVLINK; 1270 goto connect_abort; 1271 } 1272 smc_switch_link_and_count(&smc->conn, link); 1273 } 1274 1275 /* create send buffer and rmb */ 1276 if (smc_buf_create(smc, false)) { 1277 reason_code = SMC_CLC_DECL_MEM; 1278 goto connect_abort; 1279 } 1280 1281 if (ini->first_contact_local) 1282 smc_link_save_peer_info(link, aclc, ini); 1283 1284 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1285 reason_code = SMC_CLC_DECL_ERR_RTOK; 1286 goto connect_abort; 1287 } 1288 1289 smc_close_init(smc); 1290 smc_rx_init(smc); 1291 1292 if (ini->first_contact_local) { 1293 if (smc_ib_ready_link(link)) { 1294 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1295 goto connect_abort; 1296 } 1297 } else { 1298 /* reg sendbufs if they were vzalloced */ 1299 if (smc->conn.sndbuf_desc->is_vm) { 1300 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1301 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1302 goto connect_abort; 1303 } 1304 } 1305 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1306 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1307 goto connect_abort; 1308 } 1309 } 1310 1311 if (aclc->hdr.version > SMC_V1) { 1312 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1313 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1314 1315 eid = clc_v2->r1.eid; 1316 if (ini->first_contact_local) 1317 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1318 link->smcibdev, link->gid); 1319 } 1320 1321 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1322 aclc->hdr.version, eid, ini); 1323 if (reason_code) 1324 goto connect_abort; 1325 1326 smc_tx_init(smc); 1327 1328 if (ini->first_contact_local) { 1329 /* QP confirmation over RoCE fabric */ 1330 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1331 reason_code = smcr_clnt_conf_first_link(smc); 1332 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1333 if (reason_code) 1334 goto connect_abort; 1335 } 1336 mutex_unlock(&smc_client_lgr_pending); 1337 1338 smc_copy_sock_settings_to_clc(smc); 1339 smc->connect_nonblock = 0; 1340 if (smc->sk.sk_state == SMC_INIT) 1341 smc->sk.sk_state = SMC_ACTIVE; 1342 1343 return 0; 1344 connect_abort: 1345 smc_conn_abort(smc, ini->first_contact_local); 1346 mutex_unlock(&smc_client_lgr_pending); 1347 smc->connect_nonblock = 0; 1348 1349 return reason_code; 1350 } 1351 1352 /* The server has chosen one of the proposed ISM devices for the communication. 1353 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1354 */ 1355 static int 1356 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 1357 struct smc_init_info *ini) 1358 { 1359 int i; 1360 1361 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1362 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1363 ini->ism_selected = i; 1364 return 0; 1365 } 1366 } 1367 1368 return -EPROTO; 1369 } 1370 1371 /* setup for ISM connection of client */ 1372 static int smc_connect_ism(struct smc_sock *smc, 1373 struct smc_clc_msg_accept_confirm *aclc, 1374 struct smc_init_info *ini) 1375 { 1376 u8 *eid = NULL; 1377 int rc = 0; 1378 1379 ini->is_smcd = true; 1380 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1381 1382 if (aclc->hdr.version == SMC_V2) { 1383 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1384 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1385 1386 if (ini->first_contact_peer) { 1387 struct smc_clc_first_contact_ext *fce = 1388 smc_get_clc_first_contact_ext(aclc_v2, true); 1389 1390 ini->release_nr = fce->release; 1391 rc = smc_clc_clnt_v2x_features_validate(fce, ini); 1392 if (rc) 1393 return rc; 1394 } 1395 1396 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1397 if (rc) 1398 return rc; 1399 } 1400 ini->ism_peer_gid[ini->ism_selected] = ntohll(aclc->d0.gid); 1401 1402 /* there is only one lgr role for SMC-D; use server lock */ 1403 mutex_lock(&smc_server_lgr_pending); 1404 rc = smc_conn_create(smc, ini); 1405 if (rc) { 1406 mutex_unlock(&smc_server_lgr_pending); 1407 return rc; 1408 } 1409 1410 /* Create send and receive buffers */ 1411 rc = smc_buf_create(smc, true); 1412 if (rc) { 1413 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1414 goto connect_abort; 1415 } 1416 1417 smc_conn_save_peer_info(smc, aclc); 1418 smc_close_init(smc); 1419 smc_rx_init(smc); 1420 smc_tx_init(smc); 1421 1422 if (aclc->hdr.version > SMC_V1) { 1423 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1424 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1425 1426 eid = clc_v2->d1.eid; 1427 } 1428 1429 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1430 aclc->hdr.version, eid, ini); 1431 if (rc) 1432 goto connect_abort; 1433 mutex_unlock(&smc_server_lgr_pending); 1434 1435 smc_copy_sock_settings_to_clc(smc); 1436 smc->connect_nonblock = 0; 1437 if (smc->sk.sk_state == SMC_INIT) 1438 smc->sk.sk_state = SMC_ACTIVE; 1439 1440 return 0; 1441 connect_abort: 1442 smc_conn_abort(smc, ini->first_contact_local); 1443 mutex_unlock(&smc_server_lgr_pending); 1444 smc->connect_nonblock = 0; 1445 1446 return rc; 1447 } 1448 1449 /* check if received accept type and version matches a proposed one */ 1450 static int smc_connect_check_aclc(struct smc_init_info *ini, 1451 struct smc_clc_msg_accept_confirm *aclc) 1452 { 1453 if (aclc->hdr.typev1 != SMC_TYPE_R && 1454 aclc->hdr.typev1 != SMC_TYPE_D) 1455 return SMC_CLC_DECL_MODEUNSUPP; 1456 1457 if (aclc->hdr.version >= SMC_V2) { 1458 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1459 !smcr_indicated(ini->smc_type_v2)) || 1460 (aclc->hdr.typev1 == SMC_TYPE_D && 1461 !smcd_indicated(ini->smc_type_v2))) 1462 return SMC_CLC_DECL_MODEUNSUPP; 1463 } else { 1464 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1465 !smcr_indicated(ini->smc_type_v1)) || 1466 (aclc->hdr.typev1 == SMC_TYPE_D && 1467 !smcd_indicated(ini->smc_type_v1))) 1468 return SMC_CLC_DECL_MODEUNSUPP; 1469 } 1470 1471 return 0; 1472 } 1473 1474 /* perform steps before actually connecting */ 1475 static int __smc_connect(struct smc_sock *smc) 1476 { 1477 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1478 struct smc_clc_msg_accept_confirm_v2 *aclc2; 1479 struct smc_clc_msg_accept_confirm *aclc; 1480 struct smc_init_info *ini = NULL; 1481 u8 *buf = NULL; 1482 int rc = 0; 1483 1484 if (smc->use_fallback) 1485 return smc_connect_fallback(smc, smc->fallback_rsn); 1486 1487 /* if peer has not signalled SMC-capability, fall back */ 1488 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1489 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1490 1491 /* IPSec connections opt out of SMC optimizations */ 1492 if (using_ipsec(smc)) 1493 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1494 version); 1495 1496 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1497 if (!ini) 1498 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1499 version); 1500 1501 ini->smcd_version = SMC_V1 | SMC_V2; 1502 ini->smcr_version = SMC_V1 | SMC_V2; 1503 ini->smc_type_v1 = SMC_TYPE_B; 1504 ini->smc_type_v2 = SMC_TYPE_B; 1505 1506 /* get vlan id from IP device */ 1507 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1508 ini->smcd_version &= ~SMC_V1; 1509 ini->smcr_version = 0; 1510 ini->smc_type_v1 = SMC_TYPE_N; 1511 if (!ini->smcd_version) { 1512 rc = SMC_CLC_DECL_GETVLANERR; 1513 goto fallback; 1514 } 1515 } 1516 1517 rc = smc_find_proposal_devices(smc, ini); 1518 if (rc) 1519 goto fallback; 1520 1521 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1522 if (!buf) { 1523 rc = SMC_CLC_DECL_MEM; 1524 goto fallback; 1525 } 1526 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1527 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1528 1529 /* perform CLC handshake */ 1530 rc = smc_connect_clc(smc, aclc2, ini); 1531 if (rc) { 1532 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1533 if (rc == -EAGAIN) { 1534 rc = -ETIMEDOUT; 1535 smc->sk.sk_err = ETIMEDOUT; 1536 } 1537 goto vlan_cleanup; 1538 } 1539 1540 /* check if smc modes and versions of CLC proposal and accept match */ 1541 rc = smc_connect_check_aclc(ini, aclc); 1542 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1543 if (rc) 1544 goto vlan_cleanup; 1545 1546 /* depending on previous steps, connect using rdma or ism */ 1547 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1548 ini->smcr_version = version; 1549 rc = smc_connect_rdma(smc, aclc, ini); 1550 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1551 ini->smcd_version = version; 1552 rc = smc_connect_ism(smc, aclc, ini); 1553 } 1554 if (rc) 1555 goto vlan_cleanup; 1556 1557 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1558 smc_connect_ism_vlan_cleanup(smc, ini); 1559 kfree(buf); 1560 kfree(ini); 1561 return 0; 1562 1563 vlan_cleanup: 1564 smc_connect_ism_vlan_cleanup(smc, ini); 1565 kfree(buf); 1566 fallback: 1567 kfree(ini); 1568 return smc_connect_decline_fallback(smc, rc, version); 1569 } 1570 1571 static void smc_connect_work(struct work_struct *work) 1572 { 1573 struct smc_sock *smc = container_of(work, struct smc_sock, 1574 connect_work); 1575 long timeo = smc->sk.sk_sndtimeo; 1576 int rc = 0; 1577 1578 if (!timeo) 1579 timeo = MAX_SCHEDULE_TIMEOUT; 1580 lock_sock(smc->clcsock->sk); 1581 if (smc->clcsock->sk->sk_err) { 1582 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1583 } else if ((1 << smc->clcsock->sk->sk_state) & 1584 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1585 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1586 if ((rc == -EPIPE) && 1587 ((1 << smc->clcsock->sk->sk_state) & 1588 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1589 rc = 0; 1590 } 1591 release_sock(smc->clcsock->sk); 1592 lock_sock(&smc->sk); 1593 if (rc != 0 || smc->sk.sk_err) { 1594 smc->sk.sk_state = SMC_CLOSED; 1595 if (rc == -EPIPE || rc == -EAGAIN) 1596 smc->sk.sk_err = EPIPE; 1597 else if (rc == -ECONNREFUSED) 1598 smc->sk.sk_err = ECONNREFUSED; 1599 else if (signal_pending(current)) 1600 smc->sk.sk_err = -sock_intr_errno(timeo); 1601 sock_put(&smc->sk); /* passive closing */ 1602 goto out; 1603 } 1604 1605 rc = __smc_connect(smc); 1606 if (rc < 0) 1607 smc->sk.sk_err = -rc; 1608 1609 out: 1610 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1611 if (smc->sk.sk_err) { 1612 smc->sk.sk_state_change(&smc->sk); 1613 } else { /* allow polling before and after fallback decision */ 1614 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1615 smc->sk.sk_write_space(&smc->sk); 1616 } 1617 } 1618 release_sock(&smc->sk); 1619 } 1620 1621 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1622 int alen, int flags) 1623 { 1624 struct sock *sk = sock->sk; 1625 struct smc_sock *smc; 1626 int rc = -EINVAL; 1627 1628 smc = smc_sk(sk); 1629 1630 /* separate smc parameter checking to be safe */ 1631 if (alen < sizeof(addr->sa_family)) 1632 goto out_err; 1633 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1634 goto out_err; 1635 1636 lock_sock(sk); 1637 switch (sock->state) { 1638 default: 1639 rc = -EINVAL; 1640 goto out; 1641 case SS_CONNECTED: 1642 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1643 goto out; 1644 case SS_CONNECTING: 1645 if (sk->sk_state == SMC_ACTIVE) 1646 goto connected; 1647 break; 1648 case SS_UNCONNECTED: 1649 sock->state = SS_CONNECTING; 1650 break; 1651 } 1652 1653 switch (sk->sk_state) { 1654 default: 1655 goto out; 1656 case SMC_CLOSED: 1657 rc = sock_error(sk) ? : -ECONNABORTED; 1658 sock->state = SS_UNCONNECTED; 1659 goto out; 1660 case SMC_ACTIVE: 1661 rc = -EISCONN; 1662 goto out; 1663 case SMC_INIT: 1664 break; 1665 } 1666 1667 smc_copy_sock_settings_to_clc(smc); 1668 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1669 if (smc->connect_nonblock) { 1670 rc = -EALREADY; 1671 goto out; 1672 } 1673 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1674 if (rc && rc != -EINPROGRESS) 1675 goto out; 1676 1677 if (smc->use_fallback) { 1678 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1679 goto out; 1680 } 1681 sock_hold(&smc->sk); /* sock put in passive closing */ 1682 if (flags & O_NONBLOCK) { 1683 if (queue_work(smc_hs_wq, &smc->connect_work)) 1684 smc->connect_nonblock = 1; 1685 rc = -EINPROGRESS; 1686 goto out; 1687 } else { 1688 rc = __smc_connect(smc); 1689 if (rc < 0) 1690 goto out; 1691 } 1692 1693 connected: 1694 rc = 0; 1695 sock->state = SS_CONNECTED; 1696 out: 1697 release_sock(sk); 1698 out_err: 1699 return rc; 1700 } 1701 1702 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1703 { 1704 struct socket *new_clcsock = NULL; 1705 struct sock *lsk = &lsmc->sk; 1706 struct sock *new_sk; 1707 int rc = -EINVAL; 1708 1709 release_sock(lsk); 1710 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1711 if (!new_sk) { 1712 rc = -ENOMEM; 1713 lsk->sk_err = ENOMEM; 1714 *new_smc = NULL; 1715 lock_sock(lsk); 1716 goto out; 1717 } 1718 *new_smc = smc_sk(new_sk); 1719 1720 mutex_lock(&lsmc->clcsock_release_lock); 1721 if (lsmc->clcsock) 1722 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1723 mutex_unlock(&lsmc->clcsock_release_lock); 1724 lock_sock(lsk); 1725 if (rc < 0 && rc != -EAGAIN) 1726 lsk->sk_err = -rc; 1727 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1728 new_sk->sk_prot->unhash(new_sk); 1729 if (new_clcsock) 1730 sock_release(new_clcsock); 1731 new_sk->sk_state = SMC_CLOSED; 1732 smc_sock_set_flag(new_sk, SOCK_DEAD); 1733 sock_put(new_sk); /* final */ 1734 *new_smc = NULL; 1735 goto out; 1736 } 1737 1738 /* new clcsock has inherited the smc listen-specific sk_data_ready 1739 * function; switch it back to the original sk_data_ready function 1740 */ 1741 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1742 1743 /* if new clcsock has also inherited the fallback-specific callback 1744 * functions, switch them back to the original ones. 1745 */ 1746 if (lsmc->use_fallback) { 1747 if (lsmc->clcsk_state_change) 1748 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1749 if (lsmc->clcsk_write_space) 1750 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1751 if (lsmc->clcsk_error_report) 1752 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1753 } 1754 1755 (*new_smc)->clcsock = new_clcsock; 1756 out: 1757 return rc; 1758 } 1759 1760 /* add a just created sock to the accept queue of the listen sock as 1761 * candidate for a following socket accept call from user space 1762 */ 1763 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1764 { 1765 struct smc_sock *par = smc_sk(parent); 1766 1767 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1768 spin_lock(&par->accept_q_lock); 1769 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1770 spin_unlock(&par->accept_q_lock); 1771 sk_acceptq_added(parent); 1772 } 1773 1774 /* remove a socket from the accept queue of its parental listening socket */ 1775 static void smc_accept_unlink(struct sock *sk) 1776 { 1777 struct smc_sock *par = smc_sk(sk)->listen_smc; 1778 1779 spin_lock(&par->accept_q_lock); 1780 list_del_init(&smc_sk(sk)->accept_q); 1781 spin_unlock(&par->accept_q_lock); 1782 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1783 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1784 } 1785 1786 /* remove a sock from the accept queue to bind it to a new socket created 1787 * for a socket accept call from user space 1788 */ 1789 struct sock *smc_accept_dequeue(struct sock *parent, 1790 struct socket *new_sock) 1791 { 1792 struct smc_sock *isk, *n; 1793 struct sock *new_sk; 1794 1795 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1796 new_sk = (struct sock *)isk; 1797 1798 smc_accept_unlink(new_sk); 1799 if (new_sk->sk_state == SMC_CLOSED) { 1800 new_sk->sk_prot->unhash(new_sk); 1801 if (isk->clcsock) { 1802 sock_release(isk->clcsock); 1803 isk->clcsock = NULL; 1804 } 1805 sock_put(new_sk); /* final */ 1806 continue; 1807 } 1808 if (new_sock) { 1809 sock_graft(new_sk, new_sock); 1810 new_sock->state = SS_CONNECTED; 1811 if (isk->use_fallback) { 1812 smc_sk(new_sk)->clcsock->file = new_sock->file; 1813 isk->clcsock->file->private_data = isk->clcsock; 1814 } 1815 } 1816 return new_sk; 1817 } 1818 return NULL; 1819 } 1820 1821 /* clean up for a created but never accepted sock */ 1822 void smc_close_non_accepted(struct sock *sk) 1823 { 1824 struct smc_sock *smc = smc_sk(sk); 1825 1826 sock_hold(sk); /* sock_put below */ 1827 lock_sock(sk); 1828 if (!sk->sk_lingertime) 1829 /* wait for peer closing */ 1830 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT); 1831 __smc_release(smc); 1832 release_sock(sk); 1833 sock_put(sk); /* sock_hold above */ 1834 sock_put(sk); /* final sock_put */ 1835 } 1836 1837 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1838 { 1839 struct smc_link *link = smc->conn.lnk; 1840 struct smc_llc_qentry *qentry; 1841 int rc; 1842 1843 /* reg the sndbuf if it was vzalloced*/ 1844 if (smc->conn.sndbuf_desc->is_vm) { 1845 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1846 return SMC_CLC_DECL_ERR_REGBUF; 1847 } 1848 1849 /* reg the rmb */ 1850 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1851 return SMC_CLC_DECL_ERR_REGBUF; 1852 1853 /* send CONFIRM LINK request to client over the RoCE fabric */ 1854 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1855 if (rc < 0) 1856 return SMC_CLC_DECL_TIMEOUT_CL; 1857 1858 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1859 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1860 SMC_LLC_CONFIRM_LINK); 1861 if (!qentry) { 1862 struct smc_clc_msg_decline dclc; 1863 1864 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1865 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1866 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1867 } 1868 smc_llc_save_peer_uid(qentry); 1869 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1870 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1871 if (rc) 1872 return SMC_CLC_DECL_RMBE_EC; 1873 1874 /* confirm_rkey is implicit on 1st contact */ 1875 smc->conn.rmb_desc->is_conf_rkey = true; 1876 1877 smc_llc_link_active(link); 1878 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1879 1880 if (link->lgr->max_links > 1) { 1881 down_write(&link->lgr->llc_conf_mutex); 1882 /* initial contact - try to establish second link */ 1883 smc_llc_srv_add_link(link, NULL); 1884 up_write(&link->lgr->llc_conf_mutex); 1885 } 1886 return 0; 1887 } 1888 1889 /* listen worker: finish */ 1890 static void smc_listen_out(struct smc_sock *new_smc) 1891 { 1892 struct smc_sock *lsmc = new_smc->listen_smc; 1893 struct sock *newsmcsk = &new_smc->sk; 1894 1895 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1896 atomic_dec(&lsmc->queued_smc_hs); 1897 1898 if (lsmc->sk.sk_state == SMC_LISTEN) { 1899 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1900 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1901 release_sock(&lsmc->sk); 1902 } else { /* no longer listening */ 1903 smc_close_non_accepted(newsmcsk); 1904 } 1905 1906 /* Wake up accept */ 1907 lsmc->sk.sk_data_ready(&lsmc->sk); 1908 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1909 } 1910 1911 /* listen worker: finish in state connected */ 1912 static void smc_listen_out_connected(struct smc_sock *new_smc) 1913 { 1914 struct sock *newsmcsk = &new_smc->sk; 1915 1916 if (newsmcsk->sk_state == SMC_INIT) 1917 newsmcsk->sk_state = SMC_ACTIVE; 1918 1919 smc_listen_out(new_smc); 1920 } 1921 1922 /* listen worker: finish in error state */ 1923 static void smc_listen_out_err(struct smc_sock *new_smc) 1924 { 1925 struct sock *newsmcsk = &new_smc->sk; 1926 struct net *net = sock_net(newsmcsk); 1927 1928 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1929 if (newsmcsk->sk_state == SMC_INIT) 1930 sock_put(&new_smc->sk); /* passive closing */ 1931 newsmcsk->sk_state = SMC_CLOSED; 1932 1933 smc_listen_out(new_smc); 1934 } 1935 1936 /* listen worker: decline and fall back if possible */ 1937 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1938 int local_first, u8 version) 1939 { 1940 /* RDMA setup failed, switch back to TCP */ 1941 smc_conn_abort(new_smc, local_first); 1942 if (reason_code < 0 || 1943 smc_switch_to_fallback(new_smc, reason_code)) { 1944 /* error, no fallback possible */ 1945 smc_listen_out_err(new_smc); 1946 return; 1947 } 1948 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1949 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1950 smc_listen_out_err(new_smc); 1951 return; 1952 } 1953 } 1954 smc_listen_out_connected(new_smc); 1955 } 1956 1957 /* listen worker: version checking */ 1958 static int smc_listen_v2_check(struct smc_sock *new_smc, 1959 struct smc_clc_msg_proposal *pclc, 1960 struct smc_init_info *ini) 1961 { 1962 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1963 struct smc_clc_v2_extension *pclc_v2_ext; 1964 int rc = SMC_CLC_DECL_PEERNOSMC; 1965 1966 ini->smc_type_v1 = pclc->hdr.typev1; 1967 ini->smc_type_v2 = pclc->hdr.typev2; 1968 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1969 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1970 if (pclc->hdr.version > SMC_V1) { 1971 if (smcd_indicated(ini->smc_type_v2)) 1972 ini->smcd_version |= SMC_V2; 1973 if (smcr_indicated(ini->smc_type_v2)) 1974 ini->smcr_version |= SMC_V2; 1975 } 1976 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1977 rc = SMC_CLC_DECL_PEERNOSMC; 1978 goto out; 1979 } 1980 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1981 if (!pclc_v2_ext) { 1982 ini->smcd_version &= ~SMC_V2; 1983 ini->smcr_version &= ~SMC_V2; 1984 rc = SMC_CLC_DECL_NOV2EXT; 1985 goto out; 1986 } 1987 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1988 if (ini->smcd_version & SMC_V2) { 1989 if (!smc_ism_is_v2_capable()) { 1990 ini->smcd_version &= ~SMC_V2; 1991 rc = SMC_CLC_DECL_NOISM2SUPP; 1992 } else if (!pclc_smcd_v2_ext) { 1993 ini->smcd_version &= ~SMC_V2; 1994 rc = SMC_CLC_DECL_NOV2DEXT; 1995 } else if (!pclc_v2_ext->hdr.eid_cnt && 1996 !pclc_v2_ext->hdr.flag.seid) { 1997 ini->smcd_version &= ~SMC_V2; 1998 rc = SMC_CLC_DECL_NOUEID; 1999 } 2000 } 2001 if (ini->smcr_version & SMC_V2) { 2002 if (!pclc_v2_ext->hdr.eid_cnt) { 2003 ini->smcr_version &= ~SMC_V2; 2004 rc = SMC_CLC_DECL_NOUEID; 2005 } 2006 } 2007 2008 ini->release_nr = pclc_v2_ext->hdr.flag.release; 2009 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE) 2010 ini->release_nr = SMC_RELEASE; 2011 2012 out: 2013 if (!ini->smcd_version && !ini->smcr_version) 2014 return rc; 2015 2016 return 0; 2017 } 2018 2019 /* listen worker: check prefixes */ 2020 static int smc_listen_prfx_check(struct smc_sock *new_smc, 2021 struct smc_clc_msg_proposal *pclc) 2022 { 2023 struct smc_clc_msg_proposal_prefix *pclc_prfx; 2024 struct socket *newclcsock = new_smc->clcsock; 2025 2026 if (pclc->hdr.typev1 == SMC_TYPE_N) 2027 return 0; 2028 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 2029 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 2030 return SMC_CLC_DECL_DIFFPREFIX; 2031 2032 return 0; 2033 } 2034 2035 /* listen worker: initialize connection and buffers */ 2036 static int smc_listen_rdma_init(struct smc_sock *new_smc, 2037 struct smc_init_info *ini) 2038 { 2039 int rc; 2040 2041 /* allocate connection / link group */ 2042 rc = smc_conn_create(new_smc, ini); 2043 if (rc) 2044 return rc; 2045 2046 /* create send buffer and rmb */ 2047 if (smc_buf_create(new_smc, false)) { 2048 smc_conn_abort(new_smc, ini->first_contact_local); 2049 return SMC_CLC_DECL_MEM; 2050 } 2051 2052 return 0; 2053 } 2054 2055 /* listen worker: initialize connection and buffers for SMC-D */ 2056 static int smc_listen_ism_init(struct smc_sock *new_smc, 2057 struct smc_init_info *ini) 2058 { 2059 int rc; 2060 2061 rc = smc_conn_create(new_smc, ini); 2062 if (rc) 2063 return rc; 2064 2065 /* Create send and receive buffers */ 2066 rc = smc_buf_create(new_smc, true); 2067 if (rc) { 2068 smc_conn_abort(new_smc, ini->first_contact_local); 2069 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2070 SMC_CLC_DECL_MEM; 2071 } 2072 2073 return 0; 2074 } 2075 2076 static bool smc_is_already_selected(struct smcd_dev *smcd, 2077 struct smc_init_info *ini, 2078 int matches) 2079 { 2080 int i; 2081 2082 for (i = 0; i < matches; i++) 2083 if (smcd == ini->ism_dev[i]) 2084 return true; 2085 2086 return false; 2087 } 2088 2089 /* check for ISM devices matching proposed ISM devices */ 2090 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2091 u16 proposed_chid, u64 proposed_gid, 2092 unsigned int *matches) 2093 { 2094 struct smcd_dev *smcd; 2095 2096 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2097 if (smcd->going_away) 2098 continue; 2099 if (smc_is_already_selected(smcd, ini, *matches)) 2100 continue; 2101 if (smc_ism_get_chid(smcd) == proposed_chid && 2102 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2103 ini->ism_peer_gid[*matches] = proposed_gid; 2104 ini->ism_dev[*matches] = smcd; 2105 (*matches)++; 2106 break; 2107 } 2108 } 2109 } 2110 2111 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2112 { 2113 if (!ini->rc) 2114 ini->rc = rc; 2115 } 2116 2117 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2118 struct smc_clc_msg_proposal *pclc, 2119 struct smc_init_info *ini) 2120 { 2121 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2122 struct smc_clc_v2_extension *smc_v2_ext; 2123 struct smc_clc_msg_smcd *pclc_smcd; 2124 unsigned int matches = 0; 2125 u8 smcd_version; 2126 u8 *eid = NULL; 2127 int i, rc; 2128 2129 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2130 goto not_found; 2131 2132 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2133 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2134 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2135 2136 mutex_lock(&smcd_dev_list.mutex); 2137 if (pclc_smcd->ism.chid) 2138 /* check for ISM device matching proposed native ISM device */ 2139 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2140 ntohll(pclc_smcd->ism.gid), &matches); 2141 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 2142 /* check for ISM devices matching proposed non-native ISM 2143 * devices 2144 */ 2145 smc_check_ism_v2_match(ini, 2146 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 2147 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 2148 &matches); 2149 } 2150 mutex_unlock(&smcd_dev_list.mutex); 2151 2152 if (!ini->ism_dev[0]) { 2153 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2154 goto not_found; 2155 } 2156 2157 smc_ism_get_system_eid(&eid); 2158 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2159 smcd_v2_ext->system_eid, eid)) 2160 goto not_found; 2161 2162 /* separate - outside the smcd_dev_list.lock */ 2163 smcd_version = ini->smcd_version; 2164 for (i = 0; i < matches; i++) { 2165 ini->smcd_version = SMC_V2; 2166 ini->is_smcd = true; 2167 ini->ism_selected = i; 2168 rc = smc_listen_ism_init(new_smc, ini); 2169 if (rc) { 2170 smc_find_ism_store_rc(rc, ini); 2171 /* try next active ISM device */ 2172 continue; 2173 } 2174 return; /* matching and usable V2 ISM device found */ 2175 } 2176 /* no V2 ISM device could be initialized */ 2177 ini->smcd_version = smcd_version; /* restore original value */ 2178 ini->negotiated_eid[0] = 0; 2179 2180 not_found: 2181 ini->smcd_version &= ~SMC_V2; 2182 ini->ism_dev[0] = NULL; 2183 ini->is_smcd = false; 2184 } 2185 2186 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2187 struct smc_clc_msg_proposal *pclc, 2188 struct smc_init_info *ini) 2189 { 2190 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2191 int rc = 0; 2192 2193 /* check if ISM V1 is available */ 2194 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2195 goto not_found; 2196 ini->is_smcd = true; /* prepare ISM check */ 2197 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 2198 rc = smc_find_ism_device(new_smc, ini); 2199 if (rc) 2200 goto not_found; 2201 ini->ism_selected = 0; 2202 rc = smc_listen_ism_init(new_smc, ini); 2203 if (!rc) 2204 return; /* V1 ISM device found */ 2205 2206 not_found: 2207 smc_find_ism_store_rc(rc, ini); 2208 ini->smcd_version &= ~SMC_V1; 2209 ini->ism_dev[0] = NULL; 2210 ini->is_smcd = false; 2211 } 2212 2213 /* listen worker: register buffers */ 2214 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2215 { 2216 struct smc_connection *conn = &new_smc->conn; 2217 2218 if (!local_first) { 2219 /* reg sendbufs if they were vzalloced */ 2220 if (conn->sndbuf_desc->is_vm) { 2221 if (smcr_lgr_reg_sndbufs(conn->lnk, 2222 conn->sndbuf_desc)) 2223 return SMC_CLC_DECL_ERR_REGBUF; 2224 } 2225 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2226 return SMC_CLC_DECL_ERR_REGBUF; 2227 } 2228 2229 return 0; 2230 } 2231 2232 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2233 struct smc_clc_msg_proposal *pclc, 2234 struct smc_init_info *ini) 2235 { 2236 struct smc_clc_v2_extension *smc_v2_ext; 2237 u8 smcr_version; 2238 int rc; 2239 2240 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2241 goto not_found; 2242 2243 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2244 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2245 goto not_found; 2246 2247 /* prepare RDMA check */ 2248 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2249 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2250 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2251 ini->check_smcrv2 = true; 2252 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2253 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2254 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2255 rc = smc_find_rdma_device(new_smc, ini); 2256 if (rc) { 2257 smc_find_ism_store_rc(rc, ini); 2258 goto not_found; 2259 } 2260 if (!ini->smcrv2.uses_gateway) 2261 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2262 2263 smcr_version = ini->smcr_version; 2264 ini->smcr_version = SMC_V2; 2265 rc = smc_listen_rdma_init(new_smc, ini); 2266 if (!rc) { 2267 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2268 if (rc) 2269 smc_conn_abort(new_smc, ini->first_contact_local); 2270 } 2271 if (!rc) 2272 return; 2273 ini->smcr_version = smcr_version; 2274 smc_find_ism_store_rc(rc, ini); 2275 2276 not_found: 2277 ini->smcr_version &= ~SMC_V2; 2278 ini->smcrv2.ib_dev_v2 = NULL; 2279 ini->check_smcrv2 = false; 2280 } 2281 2282 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2283 struct smc_clc_msg_proposal *pclc, 2284 struct smc_init_info *ini) 2285 { 2286 int rc; 2287 2288 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2289 return SMC_CLC_DECL_NOSMCDEV; 2290 2291 /* prepare RDMA check */ 2292 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2293 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2294 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2295 rc = smc_find_rdma_device(new_smc, ini); 2296 if (rc) { 2297 /* no RDMA device found */ 2298 return SMC_CLC_DECL_NOSMCDEV; 2299 } 2300 rc = smc_listen_rdma_init(new_smc, ini); 2301 if (rc) 2302 return rc; 2303 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2304 } 2305 2306 /* determine the local device matching to proposal */ 2307 static int smc_listen_find_device(struct smc_sock *new_smc, 2308 struct smc_clc_msg_proposal *pclc, 2309 struct smc_init_info *ini) 2310 { 2311 int prfx_rc; 2312 2313 /* check for ISM device matching V2 proposed device */ 2314 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2315 if (ini->ism_dev[0]) 2316 return 0; 2317 2318 /* check for matching IP prefix and subnet length (V1) */ 2319 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2320 if (prfx_rc) 2321 smc_find_ism_store_rc(prfx_rc, ini); 2322 2323 /* get vlan id from IP device */ 2324 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2325 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2326 2327 /* check for ISM device matching V1 proposed device */ 2328 if (!prfx_rc) 2329 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2330 if (ini->ism_dev[0]) 2331 return 0; 2332 2333 if (!smcr_indicated(pclc->hdr.typev1) && 2334 !smcr_indicated(pclc->hdr.typev2)) 2335 /* skip RDMA and decline */ 2336 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2337 2338 /* check if RDMA V2 is available */ 2339 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2340 if (ini->smcrv2.ib_dev_v2) 2341 return 0; 2342 2343 /* check if RDMA V1 is available */ 2344 if (!prfx_rc) { 2345 int rc; 2346 2347 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2348 smc_find_ism_store_rc(rc, ini); 2349 return (!rc) ? 0 : ini->rc; 2350 } 2351 return prfx_rc; 2352 } 2353 2354 /* listen worker: finish RDMA setup */ 2355 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2356 struct smc_clc_msg_accept_confirm *cclc, 2357 bool local_first, 2358 struct smc_init_info *ini) 2359 { 2360 struct smc_link *link = new_smc->conn.lnk; 2361 int reason_code = 0; 2362 2363 if (local_first) 2364 smc_link_save_peer_info(link, cclc, ini); 2365 2366 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2367 return SMC_CLC_DECL_ERR_RTOK; 2368 2369 if (local_first) { 2370 if (smc_ib_ready_link(link)) 2371 return SMC_CLC_DECL_ERR_RDYLNK; 2372 /* QP confirmation over RoCE fabric */ 2373 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2374 reason_code = smcr_serv_conf_first_link(new_smc); 2375 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2376 } 2377 return reason_code; 2378 } 2379 2380 /* setup for connection of server */ 2381 static void smc_listen_work(struct work_struct *work) 2382 { 2383 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2384 smc_listen_work); 2385 struct socket *newclcsock = new_smc->clcsock; 2386 struct smc_clc_msg_accept_confirm *cclc; 2387 struct smc_clc_msg_proposal_area *buf; 2388 struct smc_clc_msg_proposal *pclc; 2389 struct smc_init_info *ini = NULL; 2390 u8 proposal_version = SMC_V1; 2391 u8 accept_version; 2392 int rc = 0; 2393 2394 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2395 return smc_listen_out_err(new_smc); 2396 2397 if (new_smc->use_fallback) { 2398 smc_listen_out_connected(new_smc); 2399 return; 2400 } 2401 2402 /* check if peer is smc capable */ 2403 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2404 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2405 if (rc) 2406 smc_listen_out_err(new_smc); 2407 else 2408 smc_listen_out_connected(new_smc); 2409 return; 2410 } 2411 2412 /* do inband token exchange - 2413 * wait for and receive SMC Proposal CLC message 2414 */ 2415 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2416 if (!buf) { 2417 rc = SMC_CLC_DECL_MEM; 2418 goto out_decl; 2419 } 2420 pclc = (struct smc_clc_msg_proposal *)buf; 2421 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2422 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2423 if (rc) 2424 goto out_decl; 2425 2426 if (pclc->hdr.version > SMC_V1) 2427 proposal_version = SMC_V2; 2428 2429 /* IPSec connections opt out of SMC optimizations */ 2430 if (using_ipsec(new_smc)) { 2431 rc = SMC_CLC_DECL_IPSEC; 2432 goto out_decl; 2433 } 2434 2435 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2436 if (!ini) { 2437 rc = SMC_CLC_DECL_MEM; 2438 goto out_decl; 2439 } 2440 2441 /* initial version checking */ 2442 rc = smc_listen_v2_check(new_smc, pclc, ini); 2443 if (rc) 2444 goto out_decl; 2445 2446 rc = smc_clc_srv_v2x_features_validate(pclc, ini); 2447 if (rc) 2448 goto out_decl; 2449 2450 mutex_lock(&smc_server_lgr_pending); 2451 smc_close_init(new_smc); 2452 smc_rx_init(new_smc); 2453 smc_tx_init(new_smc); 2454 2455 /* determine ISM or RoCE device used for connection */ 2456 rc = smc_listen_find_device(new_smc, pclc, ini); 2457 if (rc) 2458 goto out_unlock; 2459 2460 /* send SMC Accept CLC message */ 2461 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2462 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2463 accept_version, ini->negotiated_eid, ini); 2464 if (rc) 2465 goto out_unlock; 2466 2467 /* SMC-D does not need this lock any more */ 2468 if (ini->is_smcd) 2469 mutex_unlock(&smc_server_lgr_pending); 2470 2471 /* receive SMC Confirm CLC message */ 2472 memset(buf, 0, sizeof(*buf)); 2473 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2474 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2475 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2476 if (rc) { 2477 if (!ini->is_smcd) 2478 goto out_unlock; 2479 goto out_decl; 2480 } 2481 2482 rc = smc_clc_v2x_features_confirm_check(cclc, ini); 2483 if (rc) { 2484 if (!ini->is_smcd) 2485 goto out_unlock; 2486 goto out_decl; 2487 } 2488 2489 /* fce smc release version is needed in smc_listen_rdma_finish, 2490 * so save fce info here. 2491 */ 2492 smc_conn_save_peer_info_fce(new_smc, cclc); 2493 2494 /* finish worker */ 2495 if (!ini->is_smcd) { 2496 rc = smc_listen_rdma_finish(new_smc, cclc, 2497 ini->first_contact_local, ini); 2498 if (rc) 2499 goto out_unlock; 2500 mutex_unlock(&smc_server_lgr_pending); 2501 } 2502 smc_conn_save_peer_info(new_smc, cclc); 2503 smc_listen_out_connected(new_smc); 2504 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2505 goto out_free; 2506 2507 out_unlock: 2508 mutex_unlock(&smc_server_lgr_pending); 2509 out_decl: 2510 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2511 proposal_version); 2512 out_free: 2513 kfree(ini); 2514 kfree(buf); 2515 } 2516 2517 static void smc_tcp_listen_work(struct work_struct *work) 2518 { 2519 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2520 tcp_listen_work); 2521 struct sock *lsk = &lsmc->sk; 2522 struct smc_sock *new_smc; 2523 int rc = 0; 2524 2525 lock_sock(lsk); 2526 while (lsk->sk_state == SMC_LISTEN) { 2527 rc = smc_clcsock_accept(lsmc, &new_smc); 2528 if (rc) /* clcsock accept queue empty or error */ 2529 goto out; 2530 if (!new_smc) 2531 continue; 2532 2533 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2534 atomic_inc(&lsmc->queued_smc_hs); 2535 2536 new_smc->listen_smc = lsmc; 2537 new_smc->use_fallback = lsmc->use_fallback; 2538 new_smc->fallback_rsn = lsmc->fallback_rsn; 2539 sock_hold(lsk); /* sock_put in smc_listen_work */ 2540 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2541 smc_copy_sock_settings_to_smc(new_smc); 2542 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2543 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2544 sock_put(&new_smc->sk); 2545 } 2546 2547 out: 2548 release_sock(lsk); 2549 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2550 } 2551 2552 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2553 { 2554 struct smc_sock *lsmc; 2555 2556 read_lock_bh(&listen_clcsock->sk_callback_lock); 2557 lsmc = smc_clcsock_user_data(listen_clcsock); 2558 if (!lsmc) 2559 goto out; 2560 lsmc->clcsk_data_ready(listen_clcsock); 2561 if (lsmc->sk.sk_state == SMC_LISTEN) { 2562 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2563 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2564 sock_put(&lsmc->sk); 2565 } 2566 out: 2567 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2568 } 2569 2570 static int smc_listen(struct socket *sock, int backlog) 2571 { 2572 struct sock *sk = sock->sk; 2573 struct smc_sock *smc; 2574 int rc; 2575 2576 smc = smc_sk(sk); 2577 lock_sock(sk); 2578 2579 rc = -EINVAL; 2580 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2581 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2582 goto out; 2583 2584 rc = 0; 2585 if (sk->sk_state == SMC_LISTEN) { 2586 sk->sk_max_ack_backlog = backlog; 2587 goto out; 2588 } 2589 /* some socket options are handled in core, so we could not apply 2590 * them to the clc socket -- copy smc socket options to clc socket 2591 */ 2592 smc_copy_sock_settings_to_clc(smc); 2593 if (!smc->use_fallback) 2594 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2595 2596 /* save original sk_data_ready function and establish 2597 * smc-specific sk_data_ready function 2598 */ 2599 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2600 smc->clcsock->sk->sk_user_data = 2601 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2602 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2603 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2604 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2605 2606 /* save original ops */ 2607 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2608 2609 smc->af_ops = *smc->ori_af_ops; 2610 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2611 2612 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2613 2614 if (smc->limit_smc_hs) 2615 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2616 2617 rc = kernel_listen(smc->clcsock, backlog); 2618 if (rc) { 2619 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2620 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2621 &smc->clcsk_data_ready); 2622 smc->clcsock->sk->sk_user_data = NULL; 2623 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2624 goto out; 2625 } 2626 sk->sk_max_ack_backlog = backlog; 2627 sk->sk_ack_backlog = 0; 2628 sk->sk_state = SMC_LISTEN; 2629 2630 out: 2631 release_sock(sk); 2632 return rc; 2633 } 2634 2635 static int smc_accept(struct socket *sock, struct socket *new_sock, 2636 int flags, bool kern) 2637 { 2638 struct sock *sk = sock->sk, *nsk; 2639 DECLARE_WAITQUEUE(wait, current); 2640 struct smc_sock *lsmc; 2641 long timeo; 2642 int rc = 0; 2643 2644 lsmc = smc_sk(sk); 2645 sock_hold(sk); /* sock_put below */ 2646 lock_sock(sk); 2647 2648 if (lsmc->sk.sk_state != SMC_LISTEN) { 2649 rc = -EINVAL; 2650 release_sock(sk); 2651 goto out; 2652 } 2653 2654 /* Wait for an incoming connection */ 2655 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2656 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2657 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2658 set_current_state(TASK_INTERRUPTIBLE); 2659 if (!timeo) { 2660 rc = -EAGAIN; 2661 break; 2662 } 2663 release_sock(sk); 2664 timeo = schedule_timeout(timeo); 2665 /* wakeup by sk_data_ready in smc_listen_work() */ 2666 sched_annotate_sleep(); 2667 lock_sock(sk); 2668 if (signal_pending(current)) { 2669 rc = sock_intr_errno(timeo); 2670 break; 2671 } 2672 } 2673 set_current_state(TASK_RUNNING); 2674 remove_wait_queue(sk_sleep(sk), &wait); 2675 2676 if (!rc) 2677 rc = sock_error(nsk); 2678 release_sock(sk); 2679 if (rc) 2680 goto out; 2681 2682 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2683 /* wait till data arrives on the socket */ 2684 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2685 MSEC_PER_SEC); 2686 if (smc_sk(nsk)->use_fallback) { 2687 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2688 2689 lock_sock(clcsk); 2690 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2691 sk_wait_data(clcsk, &timeo, NULL); 2692 release_sock(clcsk); 2693 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2694 lock_sock(nsk); 2695 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2696 release_sock(nsk); 2697 } 2698 } 2699 2700 out: 2701 sock_put(sk); /* sock_hold above */ 2702 return rc; 2703 } 2704 2705 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2706 int peer) 2707 { 2708 struct smc_sock *smc; 2709 2710 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2711 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2712 return -ENOTCONN; 2713 2714 smc = smc_sk(sock->sk); 2715 2716 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2717 } 2718 2719 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2720 { 2721 struct sock *sk = sock->sk; 2722 struct smc_sock *smc; 2723 int rc; 2724 2725 smc = smc_sk(sk); 2726 lock_sock(sk); 2727 2728 /* SMC does not support connect with fastopen */ 2729 if (msg->msg_flags & MSG_FASTOPEN) { 2730 /* not connected yet, fallback */ 2731 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2732 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2733 if (rc) 2734 goto out; 2735 } else { 2736 rc = -EINVAL; 2737 goto out; 2738 } 2739 } else if ((sk->sk_state != SMC_ACTIVE) && 2740 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2741 (sk->sk_state != SMC_INIT)) { 2742 rc = -EPIPE; 2743 goto out; 2744 } 2745 2746 if (smc->use_fallback) { 2747 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2748 } else { 2749 rc = smc_tx_sendmsg(smc, msg, len); 2750 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2751 } 2752 out: 2753 release_sock(sk); 2754 return rc; 2755 } 2756 2757 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2758 int flags) 2759 { 2760 struct sock *sk = sock->sk; 2761 struct smc_sock *smc; 2762 int rc = -ENOTCONN; 2763 2764 smc = smc_sk(sk); 2765 lock_sock(sk); 2766 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2767 /* socket was connected before, no more data to read */ 2768 rc = 0; 2769 goto out; 2770 } 2771 if ((sk->sk_state == SMC_INIT) || 2772 (sk->sk_state == SMC_LISTEN) || 2773 (sk->sk_state == SMC_CLOSED)) 2774 goto out; 2775 2776 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2777 rc = 0; 2778 goto out; 2779 } 2780 2781 if (smc->use_fallback) { 2782 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2783 } else { 2784 msg->msg_namelen = 0; 2785 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2786 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2787 } 2788 2789 out: 2790 release_sock(sk); 2791 return rc; 2792 } 2793 2794 static __poll_t smc_accept_poll(struct sock *parent) 2795 { 2796 struct smc_sock *isk = smc_sk(parent); 2797 __poll_t mask = 0; 2798 2799 spin_lock(&isk->accept_q_lock); 2800 if (!list_empty(&isk->accept_q)) 2801 mask = EPOLLIN | EPOLLRDNORM; 2802 spin_unlock(&isk->accept_q_lock); 2803 2804 return mask; 2805 } 2806 2807 static __poll_t smc_poll(struct file *file, struct socket *sock, 2808 poll_table *wait) 2809 { 2810 struct sock *sk = sock->sk; 2811 struct smc_sock *smc; 2812 __poll_t mask = 0; 2813 2814 if (!sk) 2815 return EPOLLNVAL; 2816 2817 smc = smc_sk(sock->sk); 2818 if (smc->use_fallback) { 2819 /* delegate to CLC child sock */ 2820 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2821 sk->sk_err = smc->clcsock->sk->sk_err; 2822 } else { 2823 if (sk->sk_state != SMC_CLOSED) 2824 sock_poll_wait(file, sock, wait); 2825 if (sk->sk_err) 2826 mask |= EPOLLERR; 2827 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2828 (sk->sk_state == SMC_CLOSED)) 2829 mask |= EPOLLHUP; 2830 if (sk->sk_state == SMC_LISTEN) { 2831 /* woken up by sk_data_ready in smc_listen_work() */ 2832 mask |= smc_accept_poll(sk); 2833 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2834 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2835 wait); 2836 sk->sk_err = smc->clcsock->sk->sk_err; 2837 } else { 2838 if ((sk->sk_state != SMC_INIT && 2839 atomic_read(&smc->conn.sndbuf_space)) || 2840 sk->sk_shutdown & SEND_SHUTDOWN) { 2841 mask |= EPOLLOUT | EPOLLWRNORM; 2842 } else { 2843 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2844 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2845 } 2846 if (atomic_read(&smc->conn.bytes_to_rcv)) 2847 mask |= EPOLLIN | EPOLLRDNORM; 2848 if (sk->sk_shutdown & RCV_SHUTDOWN) 2849 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2850 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2851 mask |= EPOLLIN; 2852 if (smc->conn.urg_state == SMC_URG_VALID) 2853 mask |= EPOLLPRI; 2854 } 2855 } 2856 2857 return mask; 2858 } 2859 2860 static int smc_shutdown(struct socket *sock, int how) 2861 { 2862 struct sock *sk = sock->sk; 2863 bool do_shutdown = true; 2864 struct smc_sock *smc; 2865 int rc = -EINVAL; 2866 int old_state; 2867 int rc1 = 0; 2868 2869 smc = smc_sk(sk); 2870 2871 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2872 return rc; 2873 2874 lock_sock(sk); 2875 2876 if (sock->state == SS_CONNECTING) { 2877 if (sk->sk_state == SMC_ACTIVE) 2878 sock->state = SS_CONNECTED; 2879 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2880 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2881 sk->sk_state == SMC_APPCLOSEWAIT1 || 2882 sk->sk_state == SMC_APPCLOSEWAIT2 || 2883 sk->sk_state == SMC_APPFINCLOSEWAIT) 2884 sock->state = SS_DISCONNECTING; 2885 } 2886 2887 rc = -ENOTCONN; 2888 if ((sk->sk_state != SMC_ACTIVE) && 2889 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2890 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2891 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2892 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2893 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2894 goto out; 2895 if (smc->use_fallback) { 2896 rc = kernel_sock_shutdown(smc->clcsock, how); 2897 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2898 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2899 sk->sk_state = SMC_CLOSED; 2900 sk->sk_socket->state = SS_UNCONNECTED; 2901 sock_put(sk); 2902 } 2903 goto out; 2904 } 2905 switch (how) { 2906 case SHUT_RDWR: /* shutdown in both directions */ 2907 old_state = sk->sk_state; 2908 rc = smc_close_active(smc); 2909 if (old_state == SMC_ACTIVE && 2910 sk->sk_state == SMC_PEERCLOSEWAIT1) 2911 do_shutdown = false; 2912 break; 2913 case SHUT_WR: 2914 rc = smc_close_shutdown_write(smc); 2915 break; 2916 case SHUT_RD: 2917 rc = 0; 2918 /* nothing more to do because peer is not involved */ 2919 break; 2920 } 2921 if (do_shutdown && smc->clcsock) 2922 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2923 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2924 sk->sk_shutdown |= how + 1; 2925 2926 if (sk->sk_state == SMC_CLOSED) 2927 sock->state = SS_UNCONNECTED; 2928 else 2929 sock->state = SS_DISCONNECTING; 2930 out: 2931 release_sock(sk); 2932 return rc ? rc : rc1; 2933 } 2934 2935 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2936 char __user *optval, int __user *optlen) 2937 { 2938 struct smc_sock *smc; 2939 int val, len; 2940 2941 smc = smc_sk(sock->sk); 2942 2943 if (get_user(len, optlen)) 2944 return -EFAULT; 2945 2946 len = min_t(int, len, sizeof(int)); 2947 2948 if (len < 0) 2949 return -EINVAL; 2950 2951 switch (optname) { 2952 case SMC_LIMIT_HS: 2953 val = smc->limit_smc_hs; 2954 break; 2955 default: 2956 return -EOPNOTSUPP; 2957 } 2958 2959 if (put_user(len, optlen)) 2960 return -EFAULT; 2961 if (copy_to_user(optval, &val, len)) 2962 return -EFAULT; 2963 2964 return 0; 2965 } 2966 2967 static int __smc_setsockopt(struct socket *sock, int level, int optname, 2968 sockptr_t optval, unsigned int optlen) 2969 { 2970 struct sock *sk = sock->sk; 2971 struct smc_sock *smc; 2972 int val, rc; 2973 2974 smc = smc_sk(sk); 2975 2976 lock_sock(sk); 2977 switch (optname) { 2978 case SMC_LIMIT_HS: 2979 if (optlen < sizeof(int)) { 2980 rc = -EINVAL; 2981 break; 2982 } 2983 if (copy_from_sockptr(&val, optval, sizeof(int))) { 2984 rc = -EFAULT; 2985 break; 2986 } 2987 2988 smc->limit_smc_hs = !!val; 2989 rc = 0; 2990 break; 2991 default: 2992 rc = -EOPNOTSUPP; 2993 break; 2994 } 2995 release_sock(sk); 2996 2997 return rc; 2998 } 2999 3000 static int smc_setsockopt(struct socket *sock, int level, int optname, 3001 sockptr_t optval, unsigned int optlen) 3002 { 3003 struct sock *sk = sock->sk; 3004 struct smc_sock *smc; 3005 int val, rc; 3006 3007 if (level == SOL_TCP && optname == TCP_ULP) 3008 return -EOPNOTSUPP; 3009 else if (level == SOL_SMC) 3010 return __smc_setsockopt(sock, level, optname, optval, optlen); 3011 3012 smc = smc_sk(sk); 3013 3014 /* generic setsockopts reaching us here always apply to the 3015 * CLC socket 3016 */ 3017 mutex_lock(&smc->clcsock_release_lock); 3018 if (!smc->clcsock) { 3019 mutex_unlock(&smc->clcsock_release_lock); 3020 return -EBADF; 3021 } 3022 if (unlikely(!smc->clcsock->ops->setsockopt)) 3023 rc = -EOPNOTSUPP; 3024 else 3025 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 3026 optval, optlen); 3027 if (smc->clcsock->sk->sk_err) { 3028 sk->sk_err = smc->clcsock->sk->sk_err; 3029 sk_error_report(sk); 3030 } 3031 mutex_unlock(&smc->clcsock_release_lock); 3032 3033 if (optlen < sizeof(int)) 3034 return -EINVAL; 3035 if (copy_from_sockptr(&val, optval, sizeof(int))) 3036 return -EFAULT; 3037 3038 lock_sock(sk); 3039 if (rc || smc->use_fallback) 3040 goto out; 3041 switch (optname) { 3042 case TCP_FASTOPEN: 3043 case TCP_FASTOPEN_CONNECT: 3044 case TCP_FASTOPEN_KEY: 3045 case TCP_FASTOPEN_NO_COOKIE: 3046 /* option not supported by SMC */ 3047 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 3048 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 3049 } else { 3050 rc = -EINVAL; 3051 } 3052 break; 3053 case TCP_NODELAY: 3054 if (sk->sk_state != SMC_INIT && 3055 sk->sk_state != SMC_LISTEN && 3056 sk->sk_state != SMC_CLOSED) { 3057 if (val) { 3058 SMC_STAT_INC(smc, ndly_cnt); 3059 smc_tx_pending(&smc->conn); 3060 cancel_delayed_work(&smc->conn.tx_work); 3061 } 3062 } 3063 break; 3064 case TCP_CORK: 3065 if (sk->sk_state != SMC_INIT && 3066 sk->sk_state != SMC_LISTEN && 3067 sk->sk_state != SMC_CLOSED) { 3068 if (!val) { 3069 SMC_STAT_INC(smc, cork_cnt); 3070 smc_tx_pending(&smc->conn); 3071 cancel_delayed_work(&smc->conn.tx_work); 3072 } 3073 } 3074 break; 3075 case TCP_DEFER_ACCEPT: 3076 smc->sockopt_defer_accept = val; 3077 break; 3078 default: 3079 break; 3080 } 3081 out: 3082 release_sock(sk); 3083 3084 return rc; 3085 } 3086 3087 static int smc_getsockopt(struct socket *sock, int level, int optname, 3088 char __user *optval, int __user *optlen) 3089 { 3090 struct smc_sock *smc; 3091 int rc; 3092 3093 if (level == SOL_SMC) 3094 return __smc_getsockopt(sock, level, optname, optval, optlen); 3095 3096 smc = smc_sk(sock->sk); 3097 mutex_lock(&smc->clcsock_release_lock); 3098 if (!smc->clcsock) { 3099 mutex_unlock(&smc->clcsock_release_lock); 3100 return -EBADF; 3101 } 3102 /* socket options apply to the CLC socket */ 3103 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3104 mutex_unlock(&smc->clcsock_release_lock); 3105 return -EOPNOTSUPP; 3106 } 3107 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3108 optval, optlen); 3109 mutex_unlock(&smc->clcsock_release_lock); 3110 return rc; 3111 } 3112 3113 static int smc_ioctl(struct socket *sock, unsigned int cmd, 3114 unsigned long arg) 3115 { 3116 union smc_host_cursor cons, urg; 3117 struct smc_connection *conn; 3118 struct smc_sock *smc; 3119 int answ; 3120 3121 smc = smc_sk(sock->sk); 3122 conn = &smc->conn; 3123 lock_sock(&smc->sk); 3124 if (smc->use_fallback) { 3125 if (!smc->clcsock) { 3126 release_sock(&smc->sk); 3127 return -EBADF; 3128 } 3129 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3130 release_sock(&smc->sk); 3131 return answ; 3132 } 3133 switch (cmd) { 3134 case SIOCINQ: /* same as FIONREAD */ 3135 if (smc->sk.sk_state == SMC_LISTEN) { 3136 release_sock(&smc->sk); 3137 return -EINVAL; 3138 } 3139 if (smc->sk.sk_state == SMC_INIT || 3140 smc->sk.sk_state == SMC_CLOSED) 3141 answ = 0; 3142 else 3143 answ = atomic_read(&smc->conn.bytes_to_rcv); 3144 break; 3145 case SIOCOUTQ: 3146 /* output queue size (not send + not acked) */ 3147 if (smc->sk.sk_state == SMC_LISTEN) { 3148 release_sock(&smc->sk); 3149 return -EINVAL; 3150 } 3151 if (smc->sk.sk_state == SMC_INIT || 3152 smc->sk.sk_state == SMC_CLOSED) 3153 answ = 0; 3154 else 3155 answ = smc->conn.sndbuf_desc->len - 3156 atomic_read(&smc->conn.sndbuf_space); 3157 break; 3158 case SIOCOUTQNSD: 3159 /* output queue size (not send only) */ 3160 if (smc->sk.sk_state == SMC_LISTEN) { 3161 release_sock(&smc->sk); 3162 return -EINVAL; 3163 } 3164 if (smc->sk.sk_state == SMC_INIT || 3165 smc->sk.sk_state == SMC_CLOSED) 3166 answ = 0; 3167 else 3168 answ = smc_tx_prepared_sends(&smc->conn); 3169 break; 3170 case SIOCATMARK: 3171 if (smc->sk.sk_state == SMC_LISTEN) { 3172 release_sock(&smc->sk); 3173 return -EINVAL; 3174 } 3175 if (smc->sk.sk_state == SMC_INIT || 3176 smc->sk.sk_state == SMC_CLOSED) { 3177 answ = 0; 3178 } else { 3179 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3180 smc_curs_copy(&urg, &conn->urg_curs, conn); 3181 answ = smc_curs_diff(conn->rmb_desc->len, 3182 &cons, &urg) == 1; 3183 } 3184 break; 3185 default: 3186 release_sock(&smc->sk); 3187 return -ENOIOCTLCMD; 3188 } 3189 release_sock(&smc->sk); 3190 3191 return put_user(answ, (int __user *)arg); 3192 } 3193 3194 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3195 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3196 * updates till whenever a respective page has been fully processed. 3197 * Note that subsequent recv() calls have to wait till all splice() processing 3198 * completed. 3199 */ 3200 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3201 struct pipe_inode_info *pipe, size_t len, 3202 unsigned int flags) 3203 { 3204 struct sock *sk = sock->sk; 3205 struct smc_sock *smc; 3206 int rc = -ENOTCONN; 3207 3208 smc = smc_sk(sk); 3209 lock_sock(sk); 3210 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3211 /* socket was connected before, no more data to read */ 3212 rc = 0; 3213 goto out; 3214 } 3215 if (sk->sk_state == SMC_INIT || 3216 sk->sk_state == SMC_LISTEN || 3217 sk->sk_state == SMC_CLOSED) 3218 goto out; 3219 3220 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3221 rc = 0; 3222 goto out; 3223 } 3224 3225 if (smc->use_fallback) { 3226 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3227 pipe, len, flags); 3228 } else { 3229 if (*ppos) { 3230 rc = -ESPIPE; 3231 goto out; 3232 } 3233 if (flags & SPLICE_F_NONBLOCK) 3234 flags = MSG_DONTWAIT; 3235 else 3236 flags = 0; 3237 SMC_STAT_INC(smc, splice_cnt); 3238 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3239 } 3240 out: 3241 release_sock(sk); 3242 3243 return rc; 3244 } 3245 3246 /* must look like tcp */ 3247 static const struct proto_ops smc_sock_ops = { 3248 .family = PF_SMC, 3249 .owner = THIS_MODULE, 3250 .release = smc_release, 3251 .bind = smc_bind, 3252 .connect = smc_connect, 3253 .socketpair = sock_no_socketpair, 3254 .accept = smc_accept, 3255 .getname = smc_getname, 3256 .poll = smc_poll, 3257 .ioctl = smc_ioctl, 3258 .listen = smc_listen, 3259 .shutdown = smc_shutdown, 3260 .setsockopt = smc_setsockopt, 3261 .getsockopt = smc_getsockopt, 3262 .sendmsg = smc_sendmsg, 3263 .recvmsg = smc_recvmsg, 3264 .mmap = sock_no_mmap, 3265 .splice_read = smc_splice_read, 3266 }; 3267 3268 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3269 int kern, struct socket *clcsock) 3270 { 3271 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3272 struct smc_sock *smc; 3273 struct sock *sk; 3274 int rc; 3275 3276 rc = -ESOCKTNOSUPPORT; 3277 if (sock->type != SOCK_STREAM) 3278 goto out; 3279 3280 rc = -EPROTONOSUPPORT; 3281 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3282 goto out; 3283 3284 rc = -ENOBUFS; 3285 sock->ops = &smc_sock_ops; 3286 sock->state = SS_UNCONNECTED; 3287 sk = smc_sock_alloc(net, sock, protocol); 3288 if (!sk) 3289 goto out; 3290 3291 /* create internal TCP socket for CLC handshake and fallback */ 3292 smc = smc_sk(sk); 3293 smc->use_fallback = false; /* assume rdma capability first */ 3294 smc->fallback_rsn = 0; 3295 3296 /* default behavior from limit_smc_hs in every net namespace */ 3297 smc->limit_smc_hs = net->smc.limit_smc_hs; 3298 3299 rc = 0; 3300 if (!clcsock) { 3301 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3302 &smc->clcsock); 3303 if (rc) { 3304 sk_common_release(sk); 3305 goto out; 3306 } 3307 3308 /* smc_clcsock_release() does not wait smc->clcsock->sk's 3309 * destruction; its sk_state might not be TCP_CLOSE after 3310 * smc->sk is close()d, and TCP timers can be fired later, 3311 * which need net ref. 3312 */ 3313 sk = smc->clcsock->sk; 3314 __netns_tracker_free(net, &sk->ns_tracker, false); 3315 sk->sk_net_refcnt = 1; 3316 get_net_track(net, &sk->ns_tracker, GFP_KERNEL); 3317 sock_inuse_add(net, 1); 3318 } else { 3319 smc->clcsock = clcsock; 3320 } 3321 3322 out: 3323 return rc; 3324 } 3325 3326 static int smc_create(struct net *net, struct socket *sock, int protocol, 3327 int kern) 3328 { 3329 return __smc_create(net, sock, protocol, kern, NULL); 3330 } 3331 3332 static const struct net_proto_family smc_sock_family_ops = { 3333 .family = PF_SMC, 3334 .owner = THIS_MODULE, 3335 .create = smc_create, 3336 }; 3337 3338 static int smc_ulp_init(struct sock *sk) 3339 { 3340 struct socket *tcp = sk->sk_socket; 3341 struct net *net = sock_net(sk); 3342 struct socket *smcsock; 3343 int protocol, ret; 3344 3345 /* only TCP can be replaced */ 3346 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3347 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3348 return -ESOCKTNOSUPPORT; 3349 /* don't handle wq now */ 3350 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3351 return -ENOTCONN; 3352 3353 if (sk->sk_family == AF_INET) 3354 protocol = SMCPROTO_SMC; 3355 else 3356 protocol = SMCPROTO_SMC6; 3357 3358 smcsock = sock_alloc(); 3359 if (!smcsock) 3360 return -ENFILE; 3361 3362 smcsock->type = SOCK_STREAM; 3363 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3364 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3365 if (ret) { 3366 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3367 return ret; 3368 } 3369 3370 /* replace tcp socket to smc */ 3371 smcsock->file = tcp->file; 3372 smcsock->file->private_data = smcsock; 3373 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3374 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3375 tcp->file = NULL; 3376 3377 return ret; 3378 } 3379 3380 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3381 const gfp_t priority) 3382 { 3383 struct inet_connection_sock *icsk = inet_csk(newsk); 3384 3385 /* don't inherit ulp ops to child when listen */ 3386 icsk->icsk_ulp_ops = NULL; 3387 } 3388 3389 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3390 .name = "smc", 3391 .owner = THIS_MODULE, 3392 .init = smc_ulp_init, 3393 .clone = smc_ulp_clone, 3394 }; 3395 3396 unsigned int smc_net_id; 3397 3398 static __net_init int smc_net_init(struct net *net) 3399 { 3400 int rc; 3401 3402 rc = smc_sysctl_net_init(net); 3403 if (rc) 3404 return rc; 3405 return smc_pnet_net_init(net); 3406 } 3407 3408 static void __net_exit smc_net_exit(struct net *net) 3409 { 3410 smc_sysctl_net_exit(net); 3411 smc_pnet_net_exit(net); 3412 } 3413 3414 static __net_init int smc_net_stat_init(struct net *net) 3415 { 3416 return smc_stats_init(net); 3417 } 3418 3419 static void __net_exit smc_net_stat_exit(struct net *net) 3420 { 3421 smc_stats_exit(net); 3422 } 3423 3424 static struct pernet_operations smc_net_ops = { 3425 .init = smc_net_init, 3426 .exit = smc_net_exit, 3427 .id = &smc_net_id, 3428 .size = sizeof(struct smc_net), 3429 }; 3430 3431 static struct pernet_operations smc_net_stat_ops = { 3432 .init = smc_net_stat_init, 3433 .exit = smc_net_stat_exit, 3434 }; 3435 3436 static int __init smc_init(void) 3437 { 3438 int rc; 3439 3440 rc = register_pernet_subsys(&smc_net_ops); 3441 if (rc) 3442 return rc; 3443 3444 rc = register_pernet_subsys(&smc_net_stat_ops); 3445 if (rc) 3446 goto out_pernet_subsys; 3447 3448 rc = smc_ism_init(); 3449 if (rc) 3450 goto out_pernet_subsys_stat; 3451 smc_clc_init(); 3452 3453 rc = smc_nl_init(); 3454 if (rc) 3455 goto out_ism; 3456 3457 rc = smc_pnet_init(); 3458 if (rc) 3459 goto out_nl; 3460 3461 rc = -ENOMEM; 3462 3463 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3464 if (!smc_tcp_ls_wq) 3465 goto out_pnet; 3466 3467 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3468 if (!smc_hs_wq) 3469 goto out_alloc_tcp_ls_wq; 3470 3471 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3472 if (!smc_close_wq) 3473 goto out_alloc_hs_wq; 3474 3475 rc = smc_core_init(); 3476 if (rc) { 3477 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3478 goto out_alloc_wqs; 3479 } 3480 3481 rc = smc_llc_init(); 3482 if (rc) { 3483 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3484 goto out_core; 3485 } 3486 3487 rc = smc_cdc_init(); 3488 if (rc) { 3489 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3490 goto out_core; 3491 } 3492 3493 rc = proto_register(&smc_proto, 1); 3494 if (rc) { 3495 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3496 goto out_core; 3497 } 3498 3499 rc = proto_register(&smc_proto6, 1); 3500 if (rc) { 3501 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3502 goto out_proto; 3503 } 3504 3505 rc = sock_register(&smc_sock_family_ops); 3506 if (rc) { 3507 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3508 goto out_proto6; 3509 } 3510 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3511 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3512 3513 rc = smc_ib_register_client(); 3514 if (rc) { 3515 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3516 goto out_sock; 3517 } 3518 3519 rc = tcp_register_ulp(&smc_ulp_ops); 3520 if (rc) { 3521 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3522 goto out_ib; 3523 } 3524 3525 static_branch_enable(&tcp_have_smc); 3526 return 0; 3527 3528 out_ib: 3529 smc_ib_unregister_client(); 3530 out_sock: 3531 sock_unregister(PF_SMC); 3532 out_proto6: 3533 proto_unregister(&smc_proto6); 3534 out_proto: 3535 proto_unregister(&smc_proto); 3536 out_core: 3537 smc_core_exit(); 3538 out_alloc_wqs: 3539 destroy_workqueue(smc_close_wq); 3540 out_alloc_hs_wq: 3541 destroy_workqueue(smc_hs_wq); 3542 out_alloc_tcp_ls_wq: 3543 destroy_workqueue(smc_tcp_ls_wq); 3544 out_pnet: 3545 smc_pnet_exit(); 3546 out_nl: 3547 smc_nl_exit(); 3548 out_ism: 3549 smc_clc_exit(); 3550 smc_ism_exit(); 3551 out_pernet_subsys_stat: 3552 unregister_pernet_subsys(&smc_net_stat_ops); 3553 out_pernet_subsys: 3554 unregister_pernet_subsys(&smc_net_ops); 3555 3556 return rc; 3557 } 3558 3559 static void __exit smc_exit(void) 3560 { 3561 static_branch_disable(&tcp_have_smc); 3562 tcp_unregister_ulp(&smc_ulp_ops); 3563 sock_unregister(PF_SMC); 3564 smc_core_exit(); 3565 smc_ib_unregister_client(); 3566 smc_ism_exit(); 3567 destroy_workqueue(smc_close_wq); 3568 destroy_workqueue(smc_tcp_ls_wq); 3569 destroy_workqueue(smc_hs_wq); 3570 proto_unregister(&smc_proto6); 3571 proto_unregister(&smc_proto); 3572 smc_pnet_exit(); 3573 smc_nl_exit(); 3574 smc_clc_exit(); 3575 unregister_pernet_subsys(&smc_net_stat_ops); 3576 unregister_pernet_subsys(&smc_net_ops); 3577 rcu_barrier(); 3578 } 3579 3580 module_init(smc_init); 3581 module_exit(smc_exit); 3582 3583 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3584 MODULE_DESCRIPTION("smc socket address family"); 3585 MODULE_LICENSE("GPL"); 3586 MODULE_ALIAS_NETPROTO(PF_SMC); 3587 MODULE_ALIAS_TCP_ULP("smc"); 3588 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3589