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 29 #include <net/sock.h> 30 #include <net/tcp.h> 31 #include <net/smc.h> 32 #include <asm/ioctls.h> 33 34 #include <net/net_namespace.h> 35 #include <net/netns/generic.h> 36 #include "smc_netns.h" 37 38 #include "smc.h" 39 #include "smc_clc.h" 40 #include "smc_llc.h" 41 #include "smc_cdc.h" 42 #include "smc_core.h" 43 #include "smc_ib.h" 44 #include "smc_ism.h" 45 #include "smc_pnet.h" 46 #include "smc_tx.h" 47 #include "smc_rx.h" 48 #include "smc_close.h" 49 50 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 51 * creation on server 52 */ 53 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 54 * creation on client 55 */ 56 57 static void smc_tcp_listen_work(struct work_struct *); 58 static void smc_connect_work(struct work_struct *); 59 60 static void smc_set_keepalive(struct sock *sk, int val) 61 { 62 struct smc_sock *smc = smc_sk(sk); 63 64 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 65 } 66 67 static struct smc_hashinfo smc_v4_hashinfo = { 68 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 69 }; 70 71 static struct smc_hashinfo smc_v6_hashinfo = { 72 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 73 }; 74 75 int smc_hash_sk(struct sock *sk) 76 { 77 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 78 struct hlist_head *head; 79 80 head = &h->ht; 81 82 write_lock_bh(&h->lock); 83 sk_add_node(sk, head); 84 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 85 write_unlock_bh(&h->lock); 86 87 return 0; 88 } 89 EXPORT_SYMBOL_GPL(smc_hash_sk); 90 91 void smc_unhash_sk(struct sock *sk) 92 { 93 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 94 95 write_lock_bh(&h->lock); 96 if (sk_del_node_init(sk)) 97 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 98 write_unlock_bh(&h->lock); 99 } 100 EXPORT_SYMBOL_GPL(smc_unhash_sk); 101 102 struct proto smc_proto = { 103 .name = "SMC", 104 .owner = THIS_MODULE, 105 .keepalive = smc_set_keepalive, 106 .hash = smc_hash_sk, 107 .unhash = smc_unhash_sk, 108 .obj_size = sizeof(struct smc_sock), 109 .h.smc_hash = &smc_v4_hashinfo, 110 .slab_flags = SLAB_TYPESAFE_BY_RCU, 111 }; 112 EXPORT_SYMBOL_GPL(smc_proto); 113 114 struct proto smc_proto6 = { 115 .name = "SMC6", 116 .owner = THIS_MODULE, 117 .keepalive = smc_set_keepalive, 118 .hash = smc_hash_sk, 119 .unhash = smc_unhash_sk, 120 .obj_size = sizeof(struct smc_sock), 121 .h.smc_hash = &smc_v6_hashinfo, 122 .slab_flags = SLAB_TYPESAFE_BY_RCU, 123 }; 124 EXPORT_SYMBOL_GPL(smc_proto6); 125 126 static void smc_restore_fallback_changes(struct smc_sock *smc) 127 { 128 smc->clcsock->file->private_data = smc->sk.sk_socket; 129 smc->clcsock->file = NULL; 130 } 131 132 static int __smc_release(struct smc_sock *smc) 133 { 134 struct sock *sk = &smc->sk; 135 int rc = 0; 136 137 if (!smc->use_fallback) { 138 rc = smc_close_active(smc); 139 sock_set_flag(sk, SOCK_DEAD); 140 sk->sk_shutdown |= SHUTDOWN_MASK; 141 } else { 142 if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT) 143 sock_put(sk); /* passive closing */ 144 if (sk->sk_state == SMC_LISTEN) { 145 /* wake up clcsock accept */ 146 rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); 147 } 148 sk->sk_state = SMC_CLOSED; 149 sk->sk_state_change(sk); 150 smc_restore_fallback_changes(smc); 151 } 152 153 sk->sk_prot->unhash(sk); 154 155 if (sk->sk_state == SMC_CLOSED) { 156 if (smc->clcsock) { 157 release_sock(sk); 158 smc_clcsock_release(smc); 159 lock_sock(sk); 160 } 161 if (!smc->use_fallback) 162 smc_conn_free(&smc->conn); 163 } 164 165 return rc; 166 } 167 168 static int smc_release(struct socket *sock) 169 { 170 struct sock *sk = sock->sk; 171 struct smc_sock *smc; 172 int rc = 0; 173 174 if (!sk) 175 goto out; 176 177 smc = smc_sk(sk); 178 179 /* cleanup for a dangling non-blocking connect */ 180 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 181 tcp_abort(smc->clcsock->sk, ECONNABORTED); 182 flush_work(&smc->connect_work); 183 184 if (sk->sk_state == SMC_LISTEN) 185 /* smc_close_non_accepted() is called and acquires 186 * sock lock for child sockets again 187 */ 188 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 189 else 190 lock_sock(sk); 191 192 rc = __smc_release(smc); 193 194 /* detach socket */ 195 sock_orphan(sk); 196 sock->sk = NULL; 197 release_sock(sk); 198 199 sock_put(sk); /* final sock_put */ 200 out: 201 return rc; 202 } 203 204 static void smc_destruct(struct sock *sk) 205 { 206 if (sk->sk_state != SMC_CLOSED) 207 return; 208 if (!sock_flag(sk, SOCK_DEAD)) 209 return; 210 211 sk_refcnt_debug_dec(sk); 212 } 213 214 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 215 int protocol) 216 { 217 struct smc_sock *smc; 218 struct proto *prot; 219 struct sock *sk; 220 221 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 222 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 223 if (!sk) 224 return NULL; 225 226 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 227 sk->sk_state = SMC_INIT; 228 sk->sk_destruct = smc_destruct; 229 sk->sk_protocol = protocol; 230 smc = smc_sk(sk); 231 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 232 INIT_WORK(&smc->connect_work, smc_connect_work); 233 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 234 INIT_LIST_HEAD(&smc->accept_q); 235 spin_lock_init(&smc->accept_q_lock); 236 spin_lock_init(&smc->conn.send_lock); 237 sk->sk_prot->hash(sk); 238 sk_refcnt_debug_inc(sk); 239 mutex_init(&smc->clcsock_release_lock); 240 241 return sk; 242 } 243 244 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 245 int addr_len) 246 { 247 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 248 struct sock *sk = sock->sk; 249 struct smc_sock *smc; 250 int rc; 251 252 smc = smc_sk(sk); 253 254 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 255 rc = -EINVAL; 256 if (addr_len < sizeof(struct sockaddr_in)) 257 goto out; 258 259 rc = -EAFNOSUPPORT; 260 if (addr->sin_family != AF_INET && 261 addr->sin_family != AF_INET6 && 262 addr->sin_family != AF_UNSPEC) 263 goto out; 264 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 265 if (addr->sin_family == AF_UNSPEC && 266 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 267 goto out; 268 269 lock_sock(sk); 270 271 /* Check if socket is already active */ 272 rc = -EINVAL; 273 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 274 goto out_rel; 275 276 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 277 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 278 279 out_rel: 280 release_sock(sk); 281 out: 282 return rc; 283 } 284 285 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 286 unsigned long mask) 287 { 288 /* options we don't get control via setsockopt for */ 289 nsk->sk_type = osk->sk_type; 290 nsk->sk_sndbuf = osk->sk_sndbuf; 291 nsk->sk_rcvbuf = osk->sk_rcvbuf; 292 nsk->sk_sndtimeo = osk->sk_sndtimeo; 293 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 294 nsk->sk_mark = osk->sk_mark; 295 nsk->sk_priority = osk->sk_priority; 296 nsk->sk_rcvlowat = osk->sk_rcvlowat; 297 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 298 nsk->sk_err = osk->sk_err; 299 300 nsk->sk_flags &= ~mask; 301 nsk->sk_flags |= osk->sk_flags & mask; 302 } 303 304 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 305 (1UL << SOCK_KEEPOPEN) | \ 306 (1UL << SOCK_LINGER) | \ 307 (1UL << SOCK_BROADCAST) | \ 308 (1UL << SOCK_TIMESTAMP) | \ 309 (1UL << SOCK_DBG) | \ 310 (1UL << SOCK_RCVTSTAMP) | \ 311 (1UL << SOCK_RCVTSTAMPNS) | \ 312 (1UL << SOCK_LOCALROUTE) | \ 313 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 314 (1UL << SOCK_RXQ_OVFL) | \ 315 (1UL << SOCK_WIFI_STATUS) | \ 316 (1UL << SOCK_NOFCS) | \ 317 (1UL << SOCK_FILTER_LOCKED) | \ 318 (1UL << SOCK_TSTAMP_NEW)) 319 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 320 * clc socket (since smc is not called for these options from net/core) 321 */ 322 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 323 { 324 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 325 } 326 327 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 328 (1UL << SOCK_KEEPOPEN) | \ 329 (1UL << SOCK_LINGER) | \ 330 (1UL << SOCK_DBG)) 331 /* copy only settings and flags relevant for smc from clc to smc socket */ 332 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 333 { 334 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 335 } 336 337 /* register a new rmb, send confirm_rkey msg to register with peer */ 338 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc, 339 bool conf_rkey) 340 { 341 if (!rmb_desc->wr_reg) { 342 /* register memory region for new rmb */ 343 if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) { 344 rmb_desc->regerr = 1; 345 return -EFAULT; 346 } 347 rmb_desc->wr_reg = 1; 348 } 349 if (!conf_rkey) 350 return 0; 351 /* exchange confirm_rkey msg with peer */ 352 if (smc_llc_do_confirm_rkey(link, rmb_desc)) { 353 rmb_desc->regerr = 1; 354 return -EFAULT; 355 } 356 return 0; 357 } 358 359 static int smc_clnt_conf_first_link(struct smc_sock *smc) 360 { 361 struct net *net = sock_net(smc->clcsock->sk); 362 struct smc_link_group *lgr = smc->conn.lgr; 363 struct smc_link *link; 364 int rest; 365 int rc; 366 367 link = &lgr->lnk[SMC_SINGLE_LINK]; 368 /* receive CONFIRM LINK request from server over RoCE fabric */ 369 rest = wait_for_completion_interruptible_timeout( 370 &link->llc_confirm, 371 SMC_LLC_WAIT_FIRST_TIME); 372 if (rest <= 0) { 373 struct smc_clc_msg_decline dclc; 374 375 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 376 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 377 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 378 } 379 380 if (link->llc_confirm_rc) 381 return SMC_CLC_DECL_RMBE_EC; 382 383 rc = smc_ib_modify_qp_rts(link); 384 if (rc) 385 return SMC_CLC_DECL_ERR_RDYLNK; 386 387 smc_wr_remember_qp_attr(link); 388 389 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 390 return SMC_CLC_DECL_ERR_REGRMB; 391 392 /* send CONFIRM LINK response over RoCE fabric */ 393 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 394 if (rc < 0) 395 return SMC_CLC_DECL_TIMEOUT_CL; 396 397 /* receive ADD LINK request from server over RoCE fabric */ 398 rest = wait_for_completion_interruptible_timeout(&link->llc_add, 399 SMC_LLC_WAIT_TIME); 400 if (rest <= 0) { 401 struct smc_clc_msg_decline dclc; 402 403 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 404 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 405 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 406 } 407 408 /* send add link reject message, only one link supported for now */ 409 rc = smc_llc_send_add_link(link, 410 link->smcibdev->mac[link->ibport - 1], 411 link->gid, SMC_LLC_RESP); 412 if (rc < 0) 413 return SMC_CLC_DECL_TIMEOUT_AL; 414 415 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 416 417 return 0; 418 } 419 420 static void smcr_conn_save_peer_info(struct smc_sock *smc, 421 struct smc_clc_msg_accept_confirm *clc) 422 { 423 int bufsize = smc_uncompress_bufsize(clc->rmbe_size); 424 425 smc->conn.peer_rmbe_idx = clc->rmbe_idx; 426 smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token); 427 smc->conn.peer_rmbe_size = bufsize; 428 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 429 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 430 } 431 432 static void smcd_conn_save_peer_info(struct smc_sock *smc, 433 struct smc_clc_msg_accept_confirm *clc) 434 { 435 int bufsize = smc_uncompress_bufsize(clc->dmbe_size); 436 437 smc->conn.peer_rmbe_idx = clc->dmbe_idx; 438 smc->conn.peer_token = clc->token; 439 /* msg header takes up space in the buffer */ 440 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 441 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 442 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 443 } 444 445 static void smc_conn_save_peer_info(struct smc_sock *smc, 446 struct smc_clc_msg_accept_confirm *clc) 447 { 448 if (smc->conn.lgr->is_smcd) 449 smcd_conn_save_peer_info(smc, clc); 450 else 451 smcr_conn_save_peer_info(smc, clc); 452 } 453 454 static void smc_link_save_peer_info(struct smc_link *link, 455 struct smc_clc_msg_accept_confirm *clc) 456 { 457 link->peer_qpn = ntoh24(clc->qpn); 458 memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE); 459 memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac)); 460 link->peer_psn = ntoh24(clc->psn); 461 link->peer_mtu = clc->qp_mtu; 462 } 463 464 static void smc_switch_to_fallback(struct smc_sock *smc) 465 { 466 smc->use_fallback = true; 467 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 468 smc->clcsock->file = smc->sk.sk_socket->file; 469 smc->clcsock->file->private_data = smc->clcsock; 470 } 471 } 472 473 /* fall back during connect */ 474 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 475 { 476 smc_switch_to_fallback(smc); 477 smc->fallback_rsn = reason_code; 478 smc_copy_sock_settings_to_clc(smc); 479 smc->connect_nonblock = 0; 480 if (smc->sk.sk_state == SMC_INIT) 481 smc->sk.sk_state = SMC_ACTIVE; 482 return 0; 483 } 484 485 /* decline and fall back during connect */ 486 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code) 487 { 488 int rc; 489 490 if (reason_code < 0) { /* error, fallback is not possible */ 491 if (smc->sk.sk_state == SMC_INIT) 492 sock_put(&smc->sk); /* passive closing */ 493 return reason_code; 494 } 495 if (reason_code != SMC_CLC_DECL_PEERDECL) { 496 rc = smc_clc_send_decline(smc, reason_code); 497 if (rc < 0) { 498 if (smc->sk.sk_state == SMC_INIT) 499 sock_put(&smc->sk); /* passive closing */ 500 return rc; 501 } 502 } 503 return smc_connect_fallback(smc, reason_code); 504 } 505 506 /* abort connecting */ 507 static int smc_connect_abort(struct smc_sock *smc, int reason_code, 508 int local_contact) 509 { 510 if (local_contact == SMC_FIRST_CONTACT) 511 smc_lgr_forget(smc->conn.lgr); 512 if (smc->conn.lgr->is_smcd) 513 /* there is only one lgr role for SMC-D; use server lock */ 514 mutex_unlock(&smc_server_lgr_pending); 515 else 516 mutex_unlock(&smc_client_lgr_pending); 517 518 smc_conn_free(&smc->conn); 519 smc->connect_nonblock = 0; 520 return reason_code; 521 } 522 523 /* check if there is a rdma device available for this connection. */ 524 /* called for connect and listen */ 525 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 526 { 527 /* PNET table look up: search active ib_device and port 528 * within same PNETID that also contains the ethernet device 529 * used for the internal TCP socket 530 */ 531 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 532 if (!ini->ib_dev) 533 return SMC_CLC_DECL_NOSMCRDEV; 534 return 0; 535 } 536 537 /* check if there is an ISM device available for this connection. */ 538 /* called for connect and listen */ 539 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 540 { 541 /* Find ISM device with same PNETID as connecting interface */ 542 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 543 if (!ini->ism_dev) 544 return SMC_CLC_DECL_NOSMCDDEV; 545 return 0; 546 } 547 548 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 549 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 550 struct smc_init_info *ini) 551 { 552 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 553 return SMC_CLC_DECL_ISMVLANERR; 554 return 0; 555 } 556 557 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 558 * used, the VLAN ID will be registered again during the connection setup. 559 */ 560 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 561 struct smc_init_info *ini) 562 { 563 if (!is_smcd) 564 return 0; 565 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 566 return SMC_CLC_DECL_CNFERR; 567 return 0; 568 } 569 570 /* CLC handshake during connect */ 571 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 572 struct smc_clc_msg_accept_confirm *aclc, 573 struct smc_init_info *ini) 574 { 575 int rc = 0; 576 577 /* do inband token exchange */ 578 rc = smc_clc_send_proposal(smc, smc_type, ini); 579 if (rc) 580 return rc; 581 /* receive SMC Accept CLC message */ 582 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 583 CLC_WAIT_TIME); 584 } 585 586 /* setup for RDMA connection of client */ 587 static int smc_connect_rdma(struct smc_sock *smc, 588 struct smc_clc_msg_accept_confirm *aclc, 589 struct smc_init_info *ini) 590 { 591 struct smc_link *link; 592 int reason_code = 0; 593 594 ini->is_smcd = false; 595 ini->ib_lcl = &aclc->lcl; 596 ini->ib_clcqpn = ntoh24(aclc->qpn); 597 ini->srv_first_contact = aclc->hdr.flag; 598 599 mutex_lock(&smc_client_lgr_pending); 600 reason_code = smc_conn_create(smc, ini); 601 if (reason_code) { 602 mutex_unlock(&smc_client_lgr_pending); 603 return reason_code; 604 } 605 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 606 607 smc_conn_save_peer_info(smc, aclc); 608 609 /* create send buffer and rmb */ 610 if (smc_buf_create(smc, false)) 611 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 612 ini->cln_first_contact); 613 614 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 615 smc_link_save_peer_info(link, aclc); 616 617 if (smc_rmb_rtoken_handling(&smc->conn, aclc)) 618 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 619 ini->cln_first_contact); 620 621 smc_close_init(smc); 622 smc_rx_init(smc); 623 624 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 625 if (smc_ib_ready_link(link)) 626 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 627 ini->cln_first_contact); 628 } else { 629 if (smc_reg_rmb(link, smc->conn.rmb_desc, true)) 630 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 631 ini->cln_first_contact); 632 } 633 smc_rmb_sync_sg_for_device(&smc->conn); 634 635 reason_code = smc_clc_send_confirm(smc); 636 if (reason_code) 637 return smc_connect_abort(smc, reason_code, 638 ini->cln_first_contact); 639 640 smc_tx_init(smc); 641 642 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 643 /* QP confirmation over RoCE fabric */ 644 reason_code = smc_clnt_conf_first_link(smc); 645 if (reason_code) 646 return smc_connect_abort(smc, reason_code, 647 ini->cln_first_contact); 648 } 649 mutex_unlock(&smc_client_lgr_pending); 650 651 smc_copy_sock_settings_to_clc(smc); 652 smc->connect_nonblock = 0; 653 if (smc->sk.sk_state == SMC_INIT) 654 smc->sk.sk_state = SMC_ACTIVE; 655 656 return 0; 657 } 658 659 /* setup for ISM connection of client */ 660 static int smc_connect_ism(struct smc_sock *smc, 661 struct smc_clc_msg_accept_confirm *aclc, 662 struct smc_init_info *ini) 663 { 664 int rc = 0; 665 666 ini->is_smcd = true; 667 ini->ism_gid = aclc->gid; 668 ini->srv_first_contact = aclc->hdr.flag; 669 670 /* there is only one lgr role for SMC-D; use server lock */ 671 mutex_lock(&smc_server_lgr_pending); 672 rc = smc_conn_create(smc, ini); 673 if (rc) { 674 mutex_unlock(&smc_server_lgr_pending); 675 return rc; 676 } 677 678 /* Create send and receive buffers */ 679 if (smc_buf_create(smc, true)) 680 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 681 ini->cln_first_contact); 682 683 smc_conn_save_peer_info(smc, aclc); 684 smc_close_init(smc); 685 smc_rx_init(smc); 686 smc_tx_init(smc); 687 688 rc = smc_clc_send_confirm(smc); 689 if (rc) 690 return smc_connect_abort(smc, rc, ini->cln_first_contact); 691 mutex_unlock(&smc_server_lgr_pending); 692 693 smc_copy_sock_settings_to_clc(smc); 694 smc->connect_nonblock = 0; 695 if (smc->sk.sk_state == SMC_INIT) 696 smc->sk.sk_state = SMC_ACTIVE; 697 698 return 0; 699 } 700 701 /* perform steps before actually connecting */ 702 static int __smc_connect(struct smc_sock *smc) 703 { 704 bool ism_supported = false, rdma_supported = false; 705 struct smc_clc_msg_accept_confirm aclc; 706 struct smc_init_info ini = {0}; 707 int smc_type; 708 int rc = 0; 709 710 if (smc->use_fallback) 711 return smc_connect_fallback(smc, smc->fallback_rsn); 712 713 /* if peer has not signalled SMC-capability, fall back */ 714 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 715 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 716 717 /* IPSec connections opt out of SMC-R optimizations */ 718 if (using_ipsec(smc)) 719 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 720 721 /* get vlan id from IP device */ 722 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 723 return smc_connect_decline_fallback(smc, 724 SMC_CLC_DECL_GETVLANERR); 725 726 /* check if there is an ism device available */ 727 if (!smc_find_ism_device(smc, &ini) && 728 !smc_connect_ism_vlan_setup(smc, &ini)) { 729 /* ISM is supported for this connection */ 730 ism_supported = true; 731 smc_type = SMC_TYPE_D; 732 } 733 734 /* check if there is a rdma device available */ 735 if (!smc_find_rdma_device(smc, &ini)) { 736 /* RDMA is supported for this connection */ 737 rdma_supported = true; 738 if (ism_supported) 739 smc_type = SMC_TYPE_B; /* both */ 740 else 741 smc_type = SMC_TYPE_R; /* only RDMA */ 742 } 743 744 /* if neither ISM nor RDMA are supported, fallback */ 745 if (!rdma_supported && !ism_supported) 746 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 747 748 /* perform CLC handshake */ 749 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 750 if (rc) { 751 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 752 return smc_connect_decline_fallback(smc, rc); 753 } 754 755 /* depending on previous steps, connect using rdma or ism */ 756 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 757 rc = smc_connect_rdma(smc, &aclc, &ini); 758 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 759 rc = smc_connect_ism(smc, &aclc, &ini); 760 else 761 rc = SMC_CLC_DECL_MODEUNSUPP; 762 if (rc) { 763 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 764 return smc_connect_decline_fallback(smc, rc); 765 } 766 767 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 768 return 0; 769 } 770 771 static void smc_connect_work(struct work_struct *work) 772 { 773 struct smc_sock *smc = container_of(work, struct smc_sock, 774 connect_work); 775 long timeo = smc->sk.sk_sndtimeo; 776 int rc = 0; 777 778 if (!timeo) 779 timeo = MAX_SCHEDULE_TIMEOUT; 780 lock_sock(smc->clcsock->sk); 781 if (smc->clcsock->sk->sk_err) { 782 smc->sk.sk_err = smc->clcsock->sk->sk_err; 783 } else if ((1 << smc->clcsock->sk->sk_state) & 784 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 785 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 786 if ((rc == -EPIPE) && 787 ((1 << smc->clcsock->sk->sk_state) & 788 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 789 rc = 0; 790 } 791 release_sock(smc->clcsock->sk); 792 lock_sock(&smc->sk); 793 if (rc != 0 || smc->sk.sk_err) { 794 smc->sk.sk_state = SMC_CLOSED; 795 if (rc == -EPIPE || rc == -EAGAIN) 796 smc->sk.sk_err = EPIPE; 797 else if (signal_pending(current)) 798 smc->sk.sk_err = -sock_intr_errno(timeo); 799 sock_put(&smc->sk); /* passive closing */ 800 goto out; 801 } 802 803 rc = __smc_connect(smc); 804 if (rc < 0) 805 smc->sk.sk_err = -rc; 806 807 out: 808 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 809 if (smc->sk.sk_err) { 810 smc->sk.sk_state_change(&smc->sk); 811 } else { /* allow polling before and after fallback decision */ 812 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 813 smc->sk.sk_write_space(&smc->sk); 814 } 815 } 816 release_sock(&smc->sk); 817 } 818 819 static int smc_connect(struct socket *sock, struct sockaddr *addr, 820 int alen, int flags) 821 { 822 struct sock *sk = sock->sk; 823 struct smc_sock *smc; 824 int rc = -EINVAL; 825 826 smc = smc_sk(sk); 827 828 /* separate smc parameter checking to be safe */ 829 if (alen < sizeof(addr->sa_family)) 830 goto out_err; 831 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 832 goto out_err; 833 834 lock_sock(sk); 835 switch (sk->sk_state) { 836 default: 837 goto out; 838 case SMC_ACTIVE: 839 rc = -EISCONN; 840 goto out; 841 case SMC_INIT: 842 rc = 0; 843 break; 844 } 845 846 smc_copy_sock_settings_to_clc(smc); 847 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 848 if (smc->connect_nonblock) { 849 rc = -EALREADY; 850 goto out; 851 } 852 rc = kernel_connect(smc->clcsock, addr, alen, flags); 853 if (rc && rc != -EINPROGRESS) 854 goto out; 855 856 sock_hold(&smc->sk); /* sock put in passive closing */ 857 if (flags & O_NONBLOCK) { 858 if (schedule_work(&smc->connect_work)) 859 smc->connect_nonblock = 1; 860 rc = -EINPROGRESS; 861 } else { 862 rc = __smc_connect(smc); 863 if (rc < 0) 864 goto out; 865 else 866 rc = 0; /* success cases including fallback */ 867 } 868 869 out: 870 release_sock(sk); 871 out_err: 872 return rc; 873 } 874 875 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 876 { 877 struct socket *new_clcsock = NULL; 878 struct sock *lsk = &lsmc->sk; 879 struct sock *new_sk; 880 int rc = -EINVAL; 881 882 release_sock(lsk); 883 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 884 if (!new_sk) { 885 rc = -ENOMEM; 886 lsk->sk_err = ENOMEM; 887 *new_smc = NULL; 888 lock_sock(lsk); 889 goto out; 890 } 891 *new_smc = smc_sk(new_sk); 892 893 mutex_lock(&lsmc->clcsock_release_lock); 894 if (lsmc->clcsock) 895 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 896 mutex_unlock(&lsmc->clcsock_release_lock); 897 lock_sock(lsk); 898 if (rc < 0) 899 lsk->sk_err = -rc; 900 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 901 new_sk->sk_prot->unhash(new_sk); 902 if (new_clcsock) 903 sock_release(new_clcsock); 904 new_sk->sk_state = SMC_CLOSED; 905 sock_set_flag(new_sk, SOCK_DEAD); 906 sock_put(new_sk); /* final */ 907 *new_smc = NULL; 908 goto out; 909 } 910 911 (*new_smc)->clcsock = new_clcsock; 912 out: 913 return rc; 914 } 915 916 /* add a just created sock to the accept queue of the listen sock as 917 * candidate for a following socket accept call from user space 918 */ 919 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 920 { 921 struct smc_sock *par = smc_sk(parent); 922 923 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 924 spin_lock(&par->accept_q_lock); 925 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 926 spin_unlock(&par->accept_q_lock); 927 sk_acceptq_added(parent); 928 } 929 930 /* remove a socket from the accept queue of its parental listening socket */ 931 static void smc_accept_unlink(struct sock *sk) 932 { 933 struct smc_sock *par = smc_sk(sk)->listen_smc; 934 935 spin_lock(&par->accept_q_lock); 936 list_del_init(&smc_sk(sk)->accept_q); 937 spin_unlock(&par->accept_q_lock); 938 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 939 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 940 } 941 942 /* remove a sock from the accept queue to bind it to a new socket created 943 * for a socket accept call from user space 944 */ 945 struct sock *smc_accept_dequeue(struct sock *parent, 946 struct socket *new_sock) 947 { 948 struct smc_sock *isk, *n; 949 struct sock *new_sk; 950 951 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 952 new_sk = (struct sock *)isk; 953 954 smc_accept_unlink(new_sk); 955 if (new_sk->sk_state == SMC_CLOSED) { 956 new_sk->sk_prot->unhash(new_sk); 957 if (isk->clcsock) { 958 sock_release(isk->clcsock); 959 isk->clcsock = NULL; 960 } 961 sock_put(new_sk); /* final */ 962 continue; 963 } 964 if (new_sock) { 965 sock_graft(new_sk, new_sock); 966 if (isk->use_fallback) { 967 smc_sk(new_sk)->clcsock->file = new_sock->file; 968 isk->clcsock->file->private_data = isk->clcsock; 969 } 970 } 971 return new_sk; 972 } 973 return NULL; 974 } 975 976 /* clean up for a created but never accepted sock */ 977 void smc_close_non_accepted(struct sock *sk) 978 { 979 struct smc_sock *smc = smc_sk(sk); 980 981 lock_sock(sk); 982 if (!sk->sk_lingertime) 983 /* wait for peer closing */ 984 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 985 __smc_release(smc); 986 release_sock(sk); 987 sock_put(sk); /* final sock_put */ 988 } 989 990 static int smc_serv_conf_first_link(struct smc_sock *smc) 991 { 992 struct net *net = sock_net(smc->clcsock->sk); 993 struct smc_link_group *lgr = smc->conn.lgr; 994 struct smc_link *link; 995 int rest; 996 int rc; 997 998 link = &lgr->lnk[SMC_SINGLE_LINK]; 999 1000 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 1001 return SMC_CLC_DECL_ERR_REGRMB; 1002 1003 /* send CONFIRM LINK request to client over the RoCE fabric */ 1004 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1005 if (rc < 0) 1006 return SMC_CLC_DECL_TIMEOUT_CL; 1007 1008 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1009 rest = wait_for_completion_interruptible_timeout( 1010 &link->llc_confirm_resp, 1011 SMC_LLC_WAIT_FIRST_TIME); 1012 if (rest <= 0) { 1013 struct smc_clc_msg_decline dclc; 1014 1015 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1016 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1017 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1018 } 1019 1020 if (link->llc_confirm_resp_rc) 1021 return SMC_CLC_DECL_RMBE_EC; 1022 1023 /* send ADD LINK request to client over the RoCE fabric */ 1024 rc = smc_llc_send_add_link(link, 1025 link->smcibdev->mac[link->ibport - 1], 1026 link->gid, SMC_LLC_REQ); 1027 if (rc < 0) 1028 return SMC_CLC_DECL_TIMEOUT_AL; 1029 1030 /* receive ADD LINK response from client over the RoCE fabric */ 1031 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1032 SMC_LLC_WAIT_TIME); 1033 if (rest <= 0) { 1034 struct smc_clc_msg_decline dclc; 1035 1036 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1037 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1038 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1039 } 1040 1041 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1042 1043 return 0; 1044 } 1045 1046 /* listen worker: finish */ 1047 static void smc_listen_out(struct smc_sock *new_smc) 1048 { 1049 struct smc_sock *lsmc = new_smc->listen_smc; 1050 struct sock *newsmcsk = &new_smc->sk; 1051 1052 if (lsmc->sk.sk_state == SMC_LISTEN) { 1053 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1054 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1055 release_sock(&lsmc->sk); 1056 } else { /* no longer listening */ 1057 smc_close_non_accepted(newsmcsk); 1058 } 1059 1060 /* Wake up accept */ 1061 lsmc->sk.sk_data_ready(&lsmc->sk); 1062 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1063 } 1064 1065 /* listen worker: finish in state connected */ 1066 static void smc_listen_out_connected(struct smc_sock *new_smc) 1067 { 1068 struct sock *newsmcsk = &new_smc->sk; 1069 1070 sk_refcnt_debug_inc(newsmcsk); 1071 if (newsmcsk->sk_state == SMC_INIT) 1072 newsmcsk->sk_state = SMC_ACTIVE; 1073 1074 smc_listen_out(new_smc); 1075 } 1076 1077 /* listen worker: finish in error state */ 1078 static void smc_listen_out_err(struct smc_sock *new_smc) 1079 { 1080 struct sock *newsmcsk = &new_smc->sk; 1081 1082 if (newsmcsk->sk_state == SMC_INIT) 1083 sock_put(&new_smc->sk); /* passive closing */ 1084 newsmcsk->sk_state = SMC_CLOSED; 1085 smc_conn_free(&new_smc->conn); 1086 1087 smc_listen_out(new_smc); 1088 } 1089 1090 /* listen worker: decline and fall back if possible */ 1091 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1092 int local_contact) 1093 { 1094 /* RDMA setup failed, switch back to TCP */ 1095 if (local_contact == SMC_FIRST_CONTACT) 1096 smc_lgr_forget(new_smc->conn.lgr); 1097 if (reason_code < 0) { /* error, no fallback possible */ 1098 smc_listen_out_err(new_smc); 1099 return; 1100 } 1101 smc_conn_free(&new_smc->conn); 1102 smc_switch_to_fallback(new_smc); 1103 new_smc->fallback_rsn = reason_code; 1104 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1105 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1106 smc_listen_out_err(new_smc); 1107 return; 1108 } 1109 } 1110 smc_listen_out_connected(new_smc); 1111 } 1112 1113 /* listen worker: check prefixes */ 1114 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1115 struct smc_clc_msg_proposal *pclc) 1116 { 1117 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1118 struct socket *newclcsock = new_smc->clcsock; 1119 1120 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1121 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1122 return SMC_CLC_DECL_DIFFPREFIX; 1123 1124 return 0; 1125 } 1126 1127 /* listen worker: initialize connection and buffers */ 1128 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1129 struct smc_init_info *ini) 1130 { 1131 int rc; 1132 1133 /* allocate connection / link group */ 1134 rc = smc_conn_create(new_smc, ini); 1135 if (rc) 1136 return rc; 1137 1138 /* create send buffer and rmb */ 1139 if (smc_buf_create(new_smc, false)) 1140 return SMC_CLC_DECL_MEM; 1141 1142 return 0; 1143 } 1144 1145 /* listen worker: initialize connection and buffers for SMC-D */ 1146 static int smc_listen_ism_init(struct smc_sock *new_smc, 1147 struct smc_clc_msg_proposal *pclc, 1148 struct smc_init_info *ini) 1149 { 1150 struct smc_clc_msg_smcd *pclc_smcd; 1151 int rc; 1152 1153 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1154 ini->ism_gid = pclc_smcd->gid; 1155 rc = smc_conn_create(new_smc, ini); 1156 if (rc) 1157 return rc; 1158 1159 /* Check if peer can be reached via ISM device */ 1160 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1161 new_smc->conn.lgr->vlan_id, 1162 new_smc->conn.lgr->smcd)) { 1163 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1164 smc_lgr_forget(new_smc->conn.lgr); 1165 smc_conn_free(&new_smc->conn); 1166 return SMC_CLC_DECL_SMCDNOTALK; 1167 } 1168 1169 /* Create send and receive buffers */ 1170 if (smc_buf_create(new_smc, true)) { 1171 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1172 smc_lgr_forget(new_smc->conn.lgr); 1173 smc_conn_free(&new_smc->conn); 1174 return SMC_CLC_DECL_MEM; 1175 } 1176 1177 return 0; 1178 } 1179 1180 /* listen worker: register buffers */ 1181 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1182 { 1183 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1184 1185 if (local_contact != SMC_FIRST_CONTACT) { 1186 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1187 return SMC_CLC_DECL_ERR_REGRMB; 1188 } 1189 smc_rmb_sync_sg_for_device(&new_smc->conn); 1190 1191 return 0; 1192 } 1193 1194 /* listen worker: finish RDMA setup */ 1195 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1196 struct smc_clc_msg_accept_confirm *cclc, 1197 int local_contact) 1198 { 1199 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1200 int reason_code = 0; 1201 1202 if (local_contact == SMC_FIRST_CONTACT) 1203 smc_link_save_peer_info(link, cclc); 1204 1205 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1206 reason_code = SMC_CLC_DECL_ERR_RTOK; 1207 goto decline; 1208 } 1209 1210 if (local_contact == SMC_FIRST_CONTACT) { 1211 if (smc_ib_ready_link(link)) { 1212 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1213 goto decline; 1214 } 1215 /* QP confirmation over RoCE fabric */ 1216 reason_code = smc_serv_conf_first_link(new_smc); 1217 if (reason_code) 1218 goto decline; 1219 } 1220 return 0; 1221 1222 decline: 1223 smc_listen_decline(new_smc, reason_code, local_contact); 1224 return reason_code; 1225 } 1226 1227 /* setup for RDMA connection of server */ 1228 static void smc_listen_work(struct work_struct *work) 1229 { 1230 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1231 smc_listen_work); 1232 struct socket *newclcsock = new_smc->clcsock; 1233 struct smc_clc_msg_accept_confirm cclc; 1234 struct smc_clc_msg_proposal *pclc; 1235 struct smc_init_info ini = {0}; 1236 bool ism_supported = false; 1237 u8 buf[SMC_CLC_MAX_LEN]; 1238 int rc = 0; 1239 1240 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1241 return smc_listen_out_err(new_smc); 1242 1243 if (new_smc->use_fallback) { 1244 smc_listen_out_connected(new_smc); 1245 return; 1246 } 1247 1248 /* check if peer is smc capable */ 1249 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1250 smc_switch_to_fallback(new_smc); 1251 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1252 smc_listen_out_connected(new_smc); 1253 return; 1254 } 1255 1256 /* do inband token exchange - 1257 * wait for and receive SMC Proposal CLC message 1258 */ 1259 pclc = (struct smc_clc_msg_proposal *)&buf; 1260 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1261 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1262 if (rc) 1263 goto out_decl; 1264 1265 /* IPSec connections opt out of SMC-R optimizations */ 1266 if (using_ipsec(new_smc)) { 1267 rc = SMC_CLC_DECL_IPSEC; 1268 goto out_decl; 1269 } 1270 1271 /* check for matching IP prefix and subnet length */ 1272 rc = smc_listen_prfx_check(new_smc, pclc); 1273 if (rc) 1274 goto out_decl; 1275 1276 /* get vlan id from IP device */ 1277 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1278 rc = SMC_CLC_DECL_GETVLANERR; 1279 goto out_decl; 1280 } 1281 1282 mutex_lock(&smc_server_lgr_pending); 1283 smc_close_init(new_smc); 1284 smc_rx_init(new_smc); 1285 smc_tx_init(new_smc); 1286 1287 /* check if ISM is available */ 1288 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1289 ini.is_smcd = true; /* prepare ISM check */ 1290 rc = smc_find_ism_device(new_smc, &ini); 1291 if (!rc) 1292 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1293 if (!rc) 1294 ism_supported = true; 1295 else if (pclc->hdr.path == SMC_TYPE_D) 1296 goto out_unlock; /* skip RDMA and decline */ 1297 } 1298 1299 /* check if RDMA is available */ 1300 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1301 /* prepare RDMA check */ 1302 ini.is_smcd = false; 1303 ini.ism_dev = NULL; 1304 ini.ib_lcl = &pclc->lcl; 1305 rc = smc_find_rdma_device(new_smc, &ini); 1306 if (rc) { 1307 /* no RDMA device found */ 1308 if (pclc->hdr.path == SMC_TYPE_B) 1309 /* neither ISM nor RDMA device found */ 1310 rc = SMC_CLC_DECL_NOSMCDEV; 1311 goto out_unlock; 1312 } 1313 rc = smc_listen_rdma_init(new_smc, &ini); 1314 if (rc) 1315 goto out_unlock; 1316 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1317 if (rc) 1318 goto out_unlock; 1319 } 1320 1321 /* send SMC Accept CLC message */ 1322 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1323 if (rc) 1324 goto out_unlock; 1325 1326 /* SMC-D does not need this lock any more */ 1327 if (ism_supported) 1328 mutex_unlock(&smc_server_lgr_pending); 1329 1330 /* receive SMC Confirm CLC message */ 1331 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1332 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1333 if (rc) { 1334 if (!ism_supported) 1335 goto out_unlock; 1336 goto out_decl; 1337 } 1338 1339 /* finish worker */ 1340 if (!ism_supported) { 1341 rc = smc_listen_rdma_finish(new_smc, &cclc, 1342 ini.cln_first_contact); 1343 mutex_unlock(&smc_server_lgr_pending); 1344 if (rc) 1345 return; 1346 } 1347 smc_conn_save_peer_info(new_smc, &cclc); 1348 smc_listen_out_connected(new_smc); 1349 return; 1350 1351 out_unlock: 1352 mutex_unlock(&smc_server_lgr_pending); 1353 out_decl: 1354 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1355 } 1356 1357 static void smc_tcp_listen_work(struct work_struct *work) 1358 { 1359 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1360 tcp_listen_work); 1361 struct sock *lsk = &lsmc->sk; 1362 struct smc_sock *new_smc; 1363 int rc = 0; 1364 1365 lock_sock(lsk); 1366 while (lsk->sk_state == SMC_LISTEN) { 1367 rc = smc_clcsock_accept(lsmc, &new_smc); 1368 if (rc) 1369 goto out; 1370 if (!new_smc) 1371 continue; 1372 1373 new_smc->listen_smc = lsmc; 1374 new_smc->use_fallback = lsmc->use_fallback; 1375 new_smc->fallback_rsn = lsmc->fallback_rsn; 1376 sock_hold(lsk); /* sock_put in smc_listen_work */ 1377 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1378 smc_copy_sock_settings_to_smc(new_smc); 1379 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1380 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1381 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1382 if (!schedule_work(&new_smc->smc_listen_work)) 1383 sock_put(&new_smc->sk); 1384 } 1385 1386 out: 1387 release_sock(lsk); 1388 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1389 } 1390 1391 static int smc_listen(struct socket *sock, int backlog) 1392 { 1393 struct sock *sk = sock->sk; 1394 struct smc_sock *smc; 1395 int rc; 1396 1397 smc = smc_sk(sk); 1398 lock_sock(sk); 1399 1400 rc = -EINVAL; 1401 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1402 smc->connect_nonblock) 1403 goto out; 1404 1405 rc = 0; 1406 if (sk->sk_state == SMC_LISTEN) { 1407 sk->sk_max_ack_backlog = backlog; 1408 goto out; 1409 } 1410 /* some socket options are handled in core, so we could not apply 1411 * them to the clc socket -- copy smc socket options to clc socket 1412 */ 1413 smc_copy_sock_settings_to_clc(smc); 1414 if (!smc->use_fallback) 1415 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1416 1417 rc = kernel_listen(smc->clcsock, backlog); 1418 if (rc) 1419 goto out; 1420 sk->sk_max_ack_backlog = backlog; 1421 sk->sk_ack_backlog = 0; 1422 sk->sk_state = SMC_LISTEN; 1423 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1424 if (!schedule_work(&smc->tcp_listen_work)) 1425 sock_put(sk); 1426 1427 out: 1428 release_sock(sk); 1429 return rc; 1430 } 1431 1432 static int smc_accept(struct socket *sock, struct socket *new_sock, 1433 int flags, bool kern) 1434 { 1435 struct sock *sk = sock->sk, *nsk; 1436 DECLARE_WAITQUEUE(wait, current); 1437 struct smc_sock *lsmc; 1438 long timeo; 1439 int rc = 0; 1440 1441 lsmc = smc_sk(sk); 1442 sock_hold(sk); /* sock_put below */ 1443 lock_sock(sk); 1444 1445 if (lsmc->sk.sk_state != SMC_LISTEN) { 1446 rc = -EINVAL; 1447 release_sock(sk); 1448 goto out; 1449 } 1450 1451 /* Wait for an incoming connection */ 1452 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1453 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1454 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1455 set_current_state(TASK_INTERRUPTIBLE); 1456 if (!timeo) { 1457 rc = -EAGAIN; 1458 break; 1459 } 1460 release_sock(sk); 1461 timeo = schedule_timeout(timeo); 1462 /* wakeup by sk_data_ready in smc_listen_work() */ 1463 sched_annotate_sleep(); 1464 lock_sock(sk); 1465 if (signal_pending(current)) { 1466 rc = sock_intr_errno(timeo); 1467 break; 1468 } 1469 } 1470 set_current_state(TASK_RUNNING); 1471 remove_wait_queue(sk_sleep(sk), &wait); 1472 1473 if (!rc) 1474 rc = sock_error(nsk); 1475 release_sock(sk); 1476 if (rc) 1477 goto out; 1478 1479 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1480 /* wait till data arrives on the socket */ 1481 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1482 MSEC_PER_SEC); 1483 if (smc_sk(nsk)->use_fallback) { 1484 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1485 1486 lock_sock(clcsk); 1487 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1488 sk_wait_data(clcsk, &timeo, NULL); 1489 release_sock(clcsk); 1490 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1491 lock_sock(nsk); 1492 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1493 release_sock(nsk); 1494 } 1495 } 1496 1497 out: 1498 sock_put(sk); /* sock_hold above */ 1499 return rc; 1500 } 1501 1502 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1503 int peer) 1504 { 1505 struct smc_sock *smc; 1506 1507 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1508 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1509 return -ENOTCONN; 1510 1511 smc = smc_sk(sock->sk); 1512 1513 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1514 } 1515 1516 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1517 { 1518 struct sock *sk = sock->sk; 1519 struct smc_sock *smc; 1520 int rc = -EPIPE; 1521 1522 smc = smc_sk(sk); 1523 lock_sock(sk); 1524 if ((sk->sk_state != SMC_ACTIVE) && 1525 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1526 (sk->sk_state != SMC_INIT)) 1527 goto out; 1528 1529 if (msg->msg_flags & MSG_FASTOPEN) { 1530 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1531 smc_switch_to_fallback(smc); 1532 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1533 } else { 1534 rc = -EINVAL; 1535 goto out; 1536 } 1537 } 1538 1539 if (smc->use_fallback) 1540 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1541 else 1542 rc = smc_tx_sendmsg(smc, msg, len); 1543 out: 1544 release_sock(sk); 1545 return rc; 1546 } 1547 1548 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1549 int flags) 1550 { 1551 struct sock *sk = sock->sk; 1552 struct smc_sock *smc; 1553 int rc = -ENOTCONN; 1554 1555 smc = smc_sk(sk); 1556 lock_sock(sk); 1557 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1558 /* socket was connected before, no more data to read */ 1559 rc = 0; 1560 goto out; 1561 } 1562 if ((sk->sk_state == SMC_INIT) || 1563 (sk->sk_state == SMC_LISTEN) || 1564 (sk->sk_state == SMC_CLOSED)) 1565 goto out; 1566 1567 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1568 rc = 0; 1569 goto out; 1570 } 1571 1572 if (smc->use_fallback) { 1573 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1574 } else { 1575 msg->msg_namelen = 0; 1576 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1577 } 1578 1579 out: 1580 release_sock(sk); 1581 return rc; 1582 } 1583 1584 static __poll_t smc_accept_poll(struct sock *parent) 1585 { 1586 struct smc_sock *isk = smc_sk(parent); 1587 __poll_t mask = 0; 1588 1589 spin_lock(&isk->accept_q_lock); 1590 if (!list_empty(&isk->accept_q)) 1591 mask = EPOLLIN | EPOLLRDNORM; 1592 spin_unlock(&isk->accept_q_lock); 1593 1594 return mask; 1595 } 1596 1597 static __poll_t smc_poll(struct file *file, struct socket *sock, 1598 poll_table *wait) 1599 { 1600 struct sock *sk = sock->sk; 1601 struct smc_sock *smc; 1602 __poll_t mask = 0; 1603 1604 if (!sk) 1605 return EPOLLNVAL; 1606 1607 smc = smc_sk(sock->sk); 1608 if (smc->use_fallback) { 1609 /* delegate to CLC child sock */ 1610 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1611 sk->sk_err = smc->clcsock->sk->sk_err; 1612 } else { 1613 if (sk->sk_state != SMC_CLOSED) 1614 sock_poll_wait(file, sock, wait); 1615 if (sk->sk_err) 1616 mask |= EPOLLERR; 1617 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1618 (sk->sk_state == SMC_CLOSED)) 1619 mask |= EPOLLHUP; 1620 if (sk->sk_state == SMC_LISTEN) { 1621 /* woken up by sk_data_ready in smc_listen_work() */ 1622 mask |= smc_accept_poll(sk); 1623 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1624 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1625 wait); 1626 sk->sk_err = smc->clcsock->sk->sk_err; 1627 } else { 1628 if ((sk->sk_state != SMC_INIT && 1629 atomic_read(&smc->conn.sndbuf_space)) || 1630 sk->sk_shutdown & SEND_SHUTDOWN) { 1631 mask |= EPOLLOUT | EPOLLWRNORM; 1632 } else { 1633 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1634 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1635 } 1636 if (atomic_read(&smc->conn.bytes_to_rcv)) 1637 mask |= EPOLLIN | EPOLLRDNORM; 1638 if (sk->sk_shutdown & RCV_SHUTDOWN) 1639 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1640 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1641 mask |= EPOLLIN; 1642 if (smc->conn.urg_state == SMC_URG_VALID) 1643 mask |= EPOLLPRI; 1644 } 1645 } 1646 1647 return mask; 1648 } 1649 1650 static int smc_shutdown(struct socket *sock, int how) 1651 { 1652 struct sock *sk = sock->sk; 1653 struct smc_sock *smc; 1654 int rc = -EINVAL; 1655 int rc1 = 0; 1656 1657 smc = smc_sk(sk); 1658 1659 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1660 return rc; 1661 1662 lock_sock(sk); 1663 1664 rc = -ENOTCONN; 1665 if ((sk->sk_state != SMC_ACTIVE) && 1666 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1667 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1668 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1669 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1670 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1671 goto out; 1672 if (smc->use_fallback) { 1673 rc = kernel_sock_shutdown(smc->clcsock, how); 1674 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1675 if (sk->sk_shutdown == SHUTDOWN_MASK) 1676 sk->sk_state = SMC_CLOSED; 1677 goto out; 1678 } 1679 switch (how) { 1680 case SHUT_RDWR: /* shutdown in both directions */ 1681 rc = smc_close_active(smc); 1682 break; 1683 case SHUT_WR: 1684 rc = smc_close_shutdown_write(smc); 1685 break; 1686 case SHUT_RD: 1687 rc = 0; 1688 /* nothing more to do because peer is not involved */ 1689 break; 1690 } 1691 if (smc->clcsock) 1692 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1693 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1694 sk->sk_shutdown |= how + 1; 1695 1696 out: 1697 release_sock(sk); 1698 return rc ? rc : rc1; 1699 } 1700 1701 static int smc_setsockopt(struct socket *sock, int level, int optname, 1702 char __user *optval, unsigned int optlen) 1703 { 1704 struct sock *sk = sock->sk; 1705 struct smc_sock *smc; 1706 int val, rc; 1707 1708 smc = smc_sk(sk); 1709 1710 /* generic setsockopts reaching us here always apply to the 1711 * CLC socket 1712 */ 1713 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1714 optval, optlen); 1715 if (smc->clcsock->sk->sk_err) { 1716 sk->sk_err = smc->clcsock->sk->sk_err; 1717 sk->sk_error_report(sk); 1718 } 1719 if (rc) 1720 return rc; 1721 1722 if (optlen < sizeof(int)) 1723 return -EINVAL; 1724 if (get_user(val, (int __user *)optval)) 1725 return -EFAULT; 1726 1727 lock_sock(sk); 1728 switch (optname) { 1729 case TCP_ULP: 1730 case TCP_FASTOPEN: 1731 case TCP_FASTOPEN_CONNECT: 1732 case TCP_FASTOPEN_KEY: 1733 case TCP_FASTOPEN_NO_COOKIE: 1734 /* option not supported by SMC */ 1735 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1736 smc_switch_to_fallback(smc); 1737 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1738 } else { 1739 if (!smc->use_fallback) 1740 rc = -EINVAL; 1741 } 1742 break; 1743 case TCP_NODELAY: 1744 if (sk->sk_state != SMC_INIT && 1745 sk->sk_state != SMC_LISTEN && 1746 sk->sk_state != SMC_CLOSED) { 1747 if (val && !smc->use_fallback) 1748 mod_delayed_work(system_wq, &smc->conn.tx_work, 1749 0); 1750 } 1751 break; 1752 case TCP_CORK: 1753 if (sk->sk_state != SMC_INIT && 1754 sk->sk_state != SMC_LISTEN && 1755 sk->sk_state != SMC_CLOSED) { 1756 if (!val && !smc->use_fallback) 1757 mod_delayed_work(system_wq, &smc->conn.tx_work, 1758 0); 1759 } 1760 break; 1761 case TCP_DEFER_ACCEPT: 1762 smc->sockopt_defer_accept = val; 1763 break; 1764 default: 1765 break; 1766 } 1767 release_sock(sk); 1768 1769 return rc; 1770 } 1771 1772 static int smc_getsockopt(struct socket *sock, int level, int optname, 1773 char __user *optval, int __user *optlen) 1774 { 1775 struct smc_sock *smc; 1776 1777 smc = smc_sk(sock->sk); 1778 /* socket options apply to the CLC socket */ 1779 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1780 optval, optlen); 1781 } 1782 1783 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1784 unsigned long arg) 1785 { 1786 union smc_host_cursor cons, urg; 1787 struct smc_connection *conn; 1788 struct smc_sock *smc; 1789 int answ; 1790 1791 smc = smc_sk(sock->sk); 1792 conn = &smc->conn; 1793 lock_sock(&smc->sk); 1794 if (smc->use_fallback) { 1795 if (!smc->clcsock) { 1796 release_sock(&smc->sk); 1797 return -EBADF; 1798 } 1799 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1800 release_sock(&smc->sk); 1801 return answ; 1802 } 1803 switch (cmd) { 1804 case SIOCINQ: /* same as FIONREAD */ 1805 if (smc->sk.sk_state == SMC_LISTEN) { 1806 release_sock(&smc->sk); 1807 return -EINVAL; 1808 } 1809 if (smc->sk.sk_state == SMC_INIT || 1810 smc->sk.sk_state == SMC_CLOSED) 1811 answ = 0; 1812 else 1813 answ = atomic_read(&smc->conn.bytes_to_rcv); 1814 break; 1815 case SIOCOUTQ: 1816 /* output queue size (not send + not acked) */ 1817 if (smc->sk.sk_state == SMC_LISTEN) { 1818 release_sock(&smc->sk); 1819 return -EINVAL; 1820 } 1821 if (smc->sk.sk_state == SMC_INIT || 1822 smc->sk.sk_state == SMC_CLOSED) 1823 answ = 0; 1824 else 1825 answ = smc->conn.sndbuf_desc->len - 1826 atomic_read(&smc->conn.sndbuf_space); 1827 break; 1828 case SIOCOUTQNSD: 1829 /* output queue size (not send only) */ 1830 if (smc->sk.sk_state == SMC_LISTEN) { 1831 release_sock(&smc->sk); 1832 return -EINVAL; 1833 } 1834 if (smc->sk.sk_state == SMC_INIT || 1835 smc->sk.sk_state == SMC_CLOSED) 1836 answ = 0; 1837 else 1838 answ = smc_tx_prepared_sends(&smc->conn); 1839 break; 1840 case SIOCATMARK: 1841 if (smc->sk.sk_state == SMC_LISTEN) { 1842 release_sock(&smc->sk); 1843 return -EINVAL; 1844 } 1845 if (smc->sk.sk_state == SMC_INIT || 1846 smc->sk.sk_state == SMC_CLOSED) { 1847 answ = 0; 1848 } else { 1849 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1850 smc_curs_copy(&urg, &conn->urg_curs, conn); 1851 answ = smc_curs_diff(conn->rmb_desc->len, 1852 &cons, &urg) == 1; 1853 } 1854 break; 1855 default: 1856 release_sock(&smc->sk); 1857 return -ENOIOCTLCMD; 1858 } 1859 release_sock(&smc->sk); 1860 1861 return put_user(answ, (int __user *)arg); 1862 } 1863 1864 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1865 int offset, size_t size, int flags) 1866 { 1867 struct sock *sk = sock->sk; 1868 struct smc_sock *smc; 1869 int rc = -EPIPE; 1870 1871 smc = smc_sk(sk); 1872 lock_sock(sk); 1873 if (sk->sk_state != SMC_ACTIVE) { 1874 release_sock(sk); 1875 goto out; 1876 } 1877 release_sock(sk); 1878 if (smc->use_fallback) 1879 rc = kernel_sendpage(smc->clcsock, page, offset, 1880 size, flags); 1881 else 1882 rc = sock_no_sendpage(sock, page, offset, size, flags); 1883 1884 out: 1885 return rc; 1886 } 1887 1888 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1889 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1890 * updates till whenever a respective page has been fully processed. 1891 * Note that subsequent recv() calls have to wait till all splice() processing 1892 * completed. 1893 */ 1894 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1895 struct pipe_inode_info *pipe, size_t len, 1896 unsigned int flags) 1897 { 1898 struct sock *sk = sock->sk; 1899 struct smc_sock *smc; 1900 int rc = -ENOTCONN; 1901 1902 smc = smc_sk(sk); 1903 lock_sock(sk); 1904 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1905 /* socket was connected before, no more data to read */ 1906 rc = 0; 1907 goto out; 1908 } 1909 if (sk->sk_state == SMC_INIT || 1910 sk->sk_state == SMC_LISTEN || 1911 sk->sk_state == SMC_CLOSED) 1912 goto out; 1913 1914 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1915 rc = 0; 1916 goto out; 1917 } 1918 1919 if (smc->use_fallback) { 1920 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1921 pipe, len, flags); 1922 } else { 1923 if (*ppos) { 1924 rc = -ESPIPE; 1925 goto out; 1926 } 1927 if (flags & SPLICE_F_NONBLOCK) 1928 flags = MSG_DONTWAIT; 1929 else 1930 flags = 0; 1931 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1932 } 1933 out: 1934 release_sock(sk); 1935 1936 return rc; 1937 } 1938 1939 /* must look like tcp */ 1940 static const struct proto_ops smc_sock_ops = { 1941 .family = PF_SMC, 1942 .owner = THIS_MODULE, 1943 .release = smc_release, 1944 .bind = smc_bind, 1945 .connect = smc_connect, 1946 .socketpair = sock_no_socketpair, 1947 .accept = smc_accept, 1948 .getname = smc_getname, 1949 .poll = smc_poll, 1950 .ioctl = smc_ioctl, 1951 .listen = smc_listen, 1952 .shutdown = smc_shutdown, 1953 .setsockopt = smc_setsockopt, 1954 .getsockopt = smc_getsockopt, 1955 .sendmsg = smc_sendmsg, 1956 .recvmsg = smc_recvmsg, 1957 .mmap = sock_no_mmap, 1958 .sendpage = smc_sendpage, 1959 .splice_read = smc_splice_read, 1960 }; 1961 1962 static int smc_create(struct net *net, struct socket *sock, int protocol, 1963 int kern) 1964 { 1965 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1966 struct smc_sock *smc; 1967 struct sock *sk; 1968 int rc; 1969 1970 rc = -ESOCKTNOSUPPORT; 1971 if (sock->type != SOCK_STREAM) 1972 goto out; 1973 1974 rc = -EPROTONOSUPPORT; 1975 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1976 goto out; 1977 1978 rc = -ENOBUFS; 1979 sock->ops = &smc_sock_ops; 1980 sk = smc_sock_alloc(net, sock, protocol); 1981 if (!sk) 1982 goto out; 1983 1984 /* create internal TCP socket for CLC handshake and fallback */ 1985 smc = smc_sk(sk); 1986 smc->use_fallback = false; /* assume rdma capability first */ 1987 smc->fallback_rsn = 0; 1988 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 1989 &smc->clcsock); 1990 if (rc) { 1991 sk_common_release(sk); 1992 goto out; 1993 } 1994 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 1995 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 1996 1997 out: 1998 return rc; 1999 } 2000 2001 static const struct net_proto_family smc_sock_family_ops = { 2002 .family = PF_SMC, 2003 .owner = THIS_MODULE, 2004 .create = smc_create, 2005 }; 2006 2007 unsigned int smc_net_id; 2008 2009 static __net_init int smc_net_init(struct net *net) 2010 { 2011 return smc_pnet_net_init(net); 2012 } 2013 2014 static void __net_exit smc_net_exit(struct net *net) 2015 { 2016 smc_pnet_net_exit(net); 2017 } 2018 2019 static struct pernet_operations smc_net_ops = { 2020 .init = smc_net_init, 2021 .exit = smc_net_exit, 2022 .id = &smc_net_id, 2023 .size = sizeof(struct smc_net), 2024 }; 2025 2026 static int __init smc_init(void) 2027 { 2028 int rc; 2029 2030 rc = register_pernet_subsys(&smc_net_ops); 2031 if (rc) 2032 return rc; 2033 2034 rc = smc_pnet_init(); 2035 if (rc) 2036 goto out_pernet_subsys; 2037 2038 rc = smc_llc_init(); 2039 if (rc) { 2040 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2041 goto out_pnet; 2042 } 2043 2044 rc = smc_cdc_init(); 2045 if (rc) { 2046 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2047 goto out_pnet; 2048 } 2049 2050 rc = proto_register(&smc_proto, 1); 2051 if (rc) { 2052 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2053 goto out_pnet; 2054 } 2055 2056 rc = proto_register(&smc_proto6, 1); 2057 if (rc) { 2058 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2059 goto out_proto; 2060 } 2061 2062 rc = sock_register(&smc_sock_family_ops); 2063 if (rc) { 2064 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2065 goto out_proto6; 2066 } 2067 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2068 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2069 2070 rc = smc_ib_register_client(); 2071 if (rc) { 2072 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2073 goto out_sock; 2074 } 2075 2076 static_branch_enable(&tcp_have_smc); 2077 return 0; 2078 2079 out_sock: 2080 sock_unregister(PF_SMC); 2081 out_proto6: 2082 proto_unregister(&smc_proto6); 2083 out_proto: 2084 proto_unregister(&smc_proto); 2085 out_pnet: 2086 smc_pnet_exit(); 2087 out_pernet_subsys: 2088 unregister_pernet_subsys(&smc_net_ops); 2089 2090 return rc; 2091 } 2092 2093 static void __exit smc_exit(void) 2094 { 2095 smc_core_exit(); 2096 static_branch_disable(&tcp_have_smc); 2097 smc_ib_unregister_client(); 2098 sock_unregister(PF_SMC); 2099 proto_unregister(&smc_proto6); 2100 proto_unregister(&smc_proto); 2101 smc_pnet_exit(); 2102 unregister_pernet_subsys(&smc_net_ops); 2103 } 2104 2105 module_init(smc_init); 2106 module_exit(smc_exit); 2107 2108 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2109 MODULE_DESCRIPTION("smc socket address family"); 2110 MODULE_LICENSE("GPL"); 2111 MODULE_ALIAS_NETPROTO(PF_SMC); 2112