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