1 /* 2 * Shared Memory Communications over RDMA (SMC-R) and RoCE 3 * 4 * AF_SMC protocol family socket handler keeping the AF_INET sock address type 5 * applies to SOCK_STREAM sockets only 6 * offers an alternative communication option for TCP-protocol sockets 7 * applicable with RoCE-cards only 8 * 9 * Initial restrictions: 10 * - support for alternate links postponed 11 * 12 * Copyright IBM Corp. 2016, 2018 13 * 14 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 15 * based on prototype from Frank Blaschka 16 */ 17 18 #define KMSG_COMPONENT "smc" 19 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 20 21 #include <linux/module.h> 22 #include <linux/socket.h> 23 #include <linux/workqueue.h> 24 #include <linux/in.h> 25 #include <linux/sched/signal.h> 26 #include <linux/if_vlan.h> 27 28 #include <net/sock.h> 29 #include <net/tcp.h> 30 #include <net/smc.h> 31 #include <asm/ioctls.h> 32 33 #include <net/net_namespace.h> 34 #include <net/netns/generic.h> 35 #include "smc_netns.h" 36 37 #include "smc.h" 38 #include "smc_clc.h" 39 #include "smc_llc.h" 40 #include "smc_cdc.h" 41 #include "smc_core.h" 42 #include "smc_ib.h" 43 #include "smc_ism.h" 44 #include "smc_pnet.h" 45 #include "smc_tx.h" 46 #include "smc_rx.h" 47 #include "smc_close.h" 48 49 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 50 * creation on server 51 */ 52 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 53 * creation on client 54 */ 55 56 static void smc_tcp_listen_work(struct work_struct *); 57 static void smc_connect_work(struct work_struct *); 58 59 static void smc_set_keepalive(struct sock *sk, int val) 60 { 61 struct smc_sock *smc = smc_sk(sk); 62 63 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 64 } 65 66 static struct smc_hashinfo smc_v4_hashinfo = { 67 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 68 }; 69 70 static struct smc_hashinfo smc_v6_hashinfo = { 71 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 72 }; 73 74 int smc_hash_sk(struct sock *sk) 75 { 76 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 77 struct hlist_head *head; 78 79 head = &h->ht; 80 81 write_lock_bh(&h->lock); 82 sk_add_node(sk, head); 83 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 84 write_unlock_bh(&h->lock); 85 86 return 0; 87 } 88 EXPORT_SYMBOL_GPL(smc_hash_sk); 89 90 void smc_unhash_sk(struct sock *sk) 91 { 92 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 93 94 write_lock_bh(&h->lock); 95 if (sk_del_node_init(sk)) 96 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 97 write_unlock_bh(&h->lock); 98 } 99 EXPORT_SYMBOL_GPL(smc_unhash_sk); 100 101 struct proto smc_proto = { 102 .name = "SMC", 103 .owner = THIS_MODULE, 104 .keepalive = smc_set_keepalive, 105 .hash = smc_hash_sk, 106 .unhash = smc_unhash_sk, 107 .obj_size = sizeof(struct smc_sock), 108 .h.smc_hash = &smc_v4_hashinfo, 109 .slab_flags = SLAB_TYPESAFE_BY_RCU, 110 }; 111 EXPORT_SYMBOL_GPL(smc_proto); 112 113 struct proto smc_proto6 = { 114 .name = "SMC6", 115 .owner = THIS_MODULE, 116 .keepalive = smc_set_keepalive, 117 .hash = smc_hash_sk, 118 .unhash = smc_unhash_sk, 119 .obj_size = sizeof(struct smc_sock), 120 .h.smc_hash = &smc_v6_hashinfo, 121 .slab_flags = SLAB_TYPESAFE_BY_RCU, 122 }; 123 EXPORT_SYMBOL_GPL(smc_proto6); 124 125 static int smc_release(struct socket *sock) 126 { 127 struct sock *sk = sock->sk; 128 struct smc_sock *smc; 129 int rc = 0; 130 131 if (!sk) 132 goto out; 133 134 smc = smc_sk(sk); 135 136 /* cleanup for a dangling non-blocking connect */ 137 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 138 tcp_abort(smc->clcsock->sk, ECONNABORTED); 139 flush_work(&smc->connect_work); 140 141 if (sk->sk_state == SMC_LISTEN) 142 /* smc_close_non_accepted() is called and acquires 143 * sock lock for child sockets again 144 */ 145 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 146 else 147 lock_sock(sk); 148 149 if (!smc->use_fallback) { 150 rc = smc_close_active(smc); 151 sock_set_flag(sk, SOCK_DEAD); 152 sk->sk_shutdown |= SHUTDOWN_MASK; 153 } else { 154 if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT) 155 sock_put(sk); /* passive closing */ 156 if (sk->sk_state == SMC_LISTEN) { 157 /* wake up clcsock accept */ 158 rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); 159 } 160 sk->sk_state = SMC_CLOSED; 161 sk->sk_state_change(sk); 162 } 163 164 sk->sk_prot->unhash(sk); 165 166 if (sk->sk_state == SMC_CLOSED) { 167 if (smc->clcsock) { 168 mutex_lock(&smc->clcsock_release_lock); 169 sock_release(smc->clcsock); 170 smc->clcsock = NULL; 171 mutex_unlock(&smc->clcsock_release_lock); 172 } 173 if (!smc->use_fallback) 174 smc_conn_free(&smc->conn); 175 } 176 177 /* detach socket */ 178 sock_orphan(sk); 179 sock->sk = NULL; 180 release_sock(sk); 181 182 sock_put(sk); /* final sock_put */ 183 out: 184 return rc; 185 } 186 187 static void smc_destruct(struct sock *sk) 188 { 189 if (sk->sk_state != SMC_CLOSED) 190 return; 191 if (!sock_flag(sk, SOCK_DEAD)) 192 return; 193 194 sk_refcnt_debug_dec(sk); 195 } 196 197 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 198 int protocol) 199 { 200 struct smc_sock *smc; 201 struct proto *prot; 202 struct sock *sk; 203 204 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 205 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 206 if (!sk) 207 return NULL; 208 209 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 210 sk->sk_state = SMC_INIT; 211 sk->sk_destruct = smc_destruct; 212 sk->sk_protocol = protocol; 213 smc = smc_sk(sk); 214 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 215 INIT_WORK(&smc->connect_work, smc_connect_work); 216 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 217 INIT_LIST_HEAD(&smc->accept_q); 218 spin_lock_init(&smc->accept_q_lock); 219 spin_lock_init(&smc->conn.send_lock); 220 sk->sk_prot->hash(sk); 221 sk_refcnt_debug_inc(sk); 222 mutex_init(&smc->clcsock_release_lock); 223 224 return sk; 225 } 226 227 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 228 int addr_len) 229 { 230 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 231 struct sock *sk = sock->sk; 232 struct smc_sock *smc; 233 int rc; 234 235 smc = smc_sk(sk); 236 237 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 238 rc = -EINVAL; 239 if (addr_len < sizeof(struct sockaddr_in)) 240 goto out; 241 242 rc = -EAFNOSUPPORT; 243 if (addr->sin_family != AF_INET && 244 addr->sin_family != AF_INET6 && 245 addr->sin_family != AF_UNSPEC) 246 goto out; 247 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 248 if (addr->sin_family == AF_UNSPEC && 249 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 250 goto out; 251 252 lock_sock(sk); 253 254 /* Check if socket is already active */ 255 rc = -EINVAL; 256 if (sk->sk_state != SMC_INIT) 257 goto out_rel; 258 259 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 260 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 261 262 out_rel: 263 release_sock(sk); 264 out: 265 return rc; 266 } 267 268 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 269 unsigned long mask) 270 { 271 /* options we don't get control via setsockopt for */ 272 nsk->sk_type = osk->sk_type; 273 nsk->sk_sndbuf = osk->sk_sndbuf; 274 nsk->sk_rcvbuf = osk->sk_rcvbuf; 275 nsk->sk_sndtimeo = osk->sk_sndtimeo; 276 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 277 nsk->sk_mark = osk->sk_mark; 278 nsk->sk_priority = osk->sk_priority; 279 nsk->sk_rcvlowat = osk->sk_rcvlowat; 280 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 281 nsk->sk_err = osk->sk_err; 282 283 nsk->sk_flags &= ~mask; 284 nsk->sk_flags |= osk->sk_flags & mask; 285 } 286 287 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 288 (1UL << SOCK_KEEPOPEN) | \ 289 (1UL << SOCK_LINGER) | \ 290 (1UL << SOCK_BROADCAST) | \ 291 (1UL << SOCK_TIMESTAMP) | \ 292 (1UL << SOCK_DBG) | \ 293 (1UL << SOCK_RCVTSTAMP) | \ 294 (1UL << SOCK_RCVTSTAMPNS) | \ 295 (1UL << SOCK_LOCALROUTE) | \ 296 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 297 (1UL << SOCK_RXQ_OVFL) | \ 298 (1UL << SOCK_WIFI_STATUS) | \ 299 (1UL << SOCK_NOFCS) | \ 300 (1UL << SOCK_FILTER_LOCKED) | \ 301 (1UL << SOCK_TSTAMP_NEW)) 302 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 303 * clc socket (since smc is not called for these options from net/core) 304 */ 305 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 306 { 307 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 308 } 309 310 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 311 (1UL << SOCK_KEEPOPEN) | \ 312 (1UL << SOCK_LINGER) | \ 313 (1UL << SOCK_DBG)) 314 /* copy only settings and flags relevant for smc from clc to smc socket */ 315 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 316 { 317 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 318 } 319 320 /* register a new rmb, send confirm_rkey msg to register with peer */ 321 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc, 322 bool conf_rkey) 323 { 324 if (!rmb_desc->wr_reg) { 325 /* register memory region for new rmb */ 326 if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) { 327 rmb_desc->regerr = 1; 328 return -EFAULT; 329 } 330 rmb_desc->wr_reg = 1; 331 } 332 if (!conf_rkey) 333 return 0; 334 /* exchange confirm_rkey msg with peer */ 335 if (smc_llc_do_confirm_rkey(link, rmb_desc)) { 336 rmb_desc->regerr = 1; 337 return -EFAULT; 338 } 339 return 0; 340 } 341 342 static int smc_clnt_conf_first_link(struct smc_sock *smc) 343 { 344 struct net *net = sock_net(smc->clcsock->sk); 345 struct smc_link_group *lgr = smc->conn.lgr; 346 struct smc_link *link; 347 int rest; 348 int rc; 349 350 link = &lgr->lnk[SMC_SINGLE_LINK]; 351 /* receive CONFIRM LINK request from server over RoCE fabric */ 352 rest = wait_for_completion_interruptible_timeout( 353 &link->llc_confirm, 354 SMC_LLC_WAIT_FIRST_TIME); 355 if (rest <= 0) { 356 struct smc_clc_msg_decline dclc; 357 358 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 359 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 360 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 361 } 362 363 if (link->llc_confirm_rc) 364 return SMC_CLC_DECL_RMBE_EC; 365 366 rc = smc_ib_modify_qp_rts(link); 367 if (rc) 368 return SMC_CLC_DECL_ERR_RDYLNK; 369 370 smc_wr_remember_qp_attr(link); 371 372 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 373 return SMC_CLC_DECL_ERR_REGRMB; 374 375 /* send CONFIRM LINK response over RoCE fabric */ 376 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 377 if (rc < 0) 378 return SMC_CLC_DECL_TIMEOUT_CL; 379 380 /* receive ADD LINK request from server over RoCE fabric */ 381 rest = wait_for_completion_interruptible_timeout(&link->llc_add, 382 SMC_LLC_WAIT_TIME); 383 if (rest <= 0) { 384 struct smc_clc_msg_decline dclc; 385 386 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 387 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 388 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 389 } 390 391 /* send add link reject message, only one link supported for now */ 392 rc = smc_llc_send_add_link(link, 393 link->smcibdev->mac[link->ibport - 1], 394 link->gid, SMC_LLC_RESP); 395 if (rc < 0) 396 return SMC_CLC_DECL_TIMEOUT_AL; 397 398 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 399 400 return 0; 401 } 402 403 static void smcr_conn_save_peer_info(struct smc_sock *smc, 404 struct smc_clc_msg_accept_confirm *clc) 405 { 406 int bufsize = smc_uncompress_bufsize(clc->rmbe_size); 407 408 smc->conn.peer_rmbe_idx = clc->rmbe_idx; 409 smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token); 410 smc->conn.peer_rmbe_size = bufsize; 411 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 412 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 413 } 414 415 static void smcd_conn_save_peer_info(struct smc_sock *smc, 416 struct smc_clc_msg_accept_confirm *clc) 417 { 418 int bufsize = smc_uncompress_bufsize(clc->dmbe_size); 419 420 smc->conn.peer_rmbe_idx = clc->dmbe_idx; 421 smc->conn.peer_token = clc->token; 422 /* msg header takes up space in the buffer */ 423 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 424 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 425 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 426 } 427 428 static void smc_conn_save_peer_info(struct smc_sock *smc, 429 struct smc_clc_msg_accept_confirm *clc) 430 { 431 if (smc->conn.lgr->is_smcd) 432 smcd_conn_save_peer_info(smc, clc); 433 else 434 smcr_conn_save_peer_info(smc, clc); 435 } 436 437 static void smc_link_save_peer_info(struct smc_link *link, 438 struct smc_clc_msg_accept_confirm *clc) 439 { 440 link->peer_qpn = ntoh24(clc->qpn); 441 memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE); 442 memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac)); 443 link->peer_psn = ntoh24(clc->psn); 444 link->peer_mtu = clc->qp_mtu; 445 } 446 447 /* fall back during connect */ 448 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 449 { 450 smc->use_fallback = true; 451 smc->fallback_rsn = reason_code; 452 smc_copy_sock_settings_to_clc(smc); 453 smc->connect_nonblock = 0; 454 if (smc->sk.sk_state == SMC_INIT) 455 smc->sk.sk_state = SMC_ACTIVE; 456 return 0; 457 } 458 459 /* decline and fall back during connect */ 460 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code) 461 { 462 int rc; 463 464 if (reason_code < 0) { /* error, fallback is not possible */ 465 if (smc->sk.sk_state == SMC_INIT) 466 sock_put(&smc->sk); /* passive closing */ 467 return reason_code; 468 } 469 if (reason_code != SMC_CLC_DECL_PEERDECL) { 470 rc = smc_clc_send_decline(smc, reason_code); 471 if (rc < 0) { 472 if (smc->sk.sk_state == SMC_INIT) 473 sock_put(&smc->sk); /* passive closing */ 474 return rc; 475 } 476 } 477 return smc_connect_fallback(smc, reason_code); 478 } 479 480 /* abort connecting */ 481 static int smc_connect_abort(struct smc_sock *smc, int reason_code, 482 int local_contact) 483 { 484 if (local_contact == SMC_FIRST_CONTACT) 485 smc_lgr_forget(smc->conn.lgr); 486 if (smc->conn.lgr->is_smcd) 487 /* there is only one lgr role for SMC-D; use server lock */ 488 mutex_unlock(&smc_server_lgr_pending); 489 else 490 mutex_unlock(&smc_client_lgr_pending); 491 492 smc_conn_free(&smc->conn); 493 smc->connect_nonblock = 0; 494 return reason_code; 495 } 496 497 /* check if there is a rdma device available for this connection. */ 498 /* called for connect and listen */ 499 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 500 { 501 /* PNET table look up: search active ib_device and port 502 * within same PNETID that also contains the ethernet device 503 * used for the internal TCP socket 504 */ 505 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 506 if (!ini->ib_dev) 507 return SMC_CLC_DECL_NOSMCRDEV; 508 return 0; 509 } 510 511 /* check if there is an ISM device available for this connection. */ 512 /* called for connect and listen */ 513 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 514 { 515 /* Find ISM device with same PNETID as connecting interface */ 516 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 517 if (!ini->ism_dev) 518 return SMC_CLC_DECL_NOSMCDDEV; 519 return 0; 520 } 521 522 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 523 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 524 struct smc_init_info *ini) 525 { 526 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 527 return SMC_CLC_DECL_ISMVLANERR; 528 return 0; 529 } 530 531 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 532 * used, the VLAN ID will be registered again during the connection setup. 533 */ 534 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 535 struct smc_init_info *ini) 536 { 537 if (!is_smcd) 538 return 0; 539 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 540 return SMC_CLC_DECL_CNFERR; 541 return 0; 542 } 543 544 /* CLC handshake during connect */ 545 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 546 struct smc_clc_msg_accept_confirm *aclc, 547 struct smc_init_info *ini) 548 { 549 int rc = 0; 550 551 /* do inband token exchange */ 552 rc = smc_clc_send_proposal(smc, smc_type, ini); 553 if (rc) 554 return rc; 555 /* receive SMC Accept CLC message */ 556 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 557 CLC_WAIT_TIME); 558 } 559 560 /* setup for RDMA connection of client */ 561 static int smc_connect_rdma(struct smc_sock *smc, 562 struct smc_clc_msg_accept_confirm *aclc, 563 struct smc_init_info *ini) 564 { 565 struct smc_link *link; 566 int reason_code = 0; 567 568 ini->is_smcd = false; 569 ini->ib_lcl = &aclc->lcl; 570 ini->ib_clcqpn = ntoh24(aclc->qpn); 571 ini->srv_first_contact = aclc->hdr.flag; 572 573 mutex_lock(&smc_client_lgr_pending); 574 reason_code = smc_conn_create(smc, ini); 575 if (reason_code) { 576 mutex_unlock(&smc_client_lgr_pending); 577 return reason_code; 578 } 579 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 580 581 smc_conn_save_peer_info(smc, aclc); 582 583 /* create send buffer and rmb */ 584 if (smc_buf_create(smc, false)) 585 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 586 ini->cln_first_contact); 587 588 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 589 smc_link_save_peer_info(link, aclc); 590 591 if (smc_rmb_rtoken_handling(&smc->conn, aclc)) 592 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 593 ini->cln_first_contact); 594 595 smc_close_init(smc); 596 smc_rx_init(smc); 597 598 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 599 if (smc_ib_ready_link(link)) 600 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 601 ini->cln_first_contact); 602 } else { 603 if (smc_reg_rmb(link, smc->conn.rmb_desc, true)) 604 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 605 ini->cln_first_contact); 606 } 607 smc_rmb_sync_sg_for_device(&smc->conn); 608 609 reason_code = smc_clc_send_confirm(smc); 610 if (reason_code) 611 return smc_connect_abort(smc, reason_code, 612 ini->cln_first_contact); 613 614 smc_tx_init(smc); 615 616 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 617 /* QP confirmation over RoCE fabric */ 618 reason_code = smc_clnt_conf_first_link(smc); 619 if (reason_code) 620 return smc_connect_abort(smc, reason_code, 621 ini->cln_first_contact); 622 } 623 mutex_unlock(&smc_client_lgr_pending); 624 625 smc_copy_sock_settings_to_clc(smc); 626 smc->connect_nonblock = 0; 627 if (smc->sk.sk_state == SMC_INIT) 628 smc->sk.sk_state = SMC_ACTIVE; 629 630 return 0; 631 } 632 633 /* setup for ISM connection of client */ 634 static int smc_connect_ism(struct smc_sock *smc, 635 struct smc_clc_msg_accept_confirm *aclc, 636 struct smc_init_info *ini) 637 { 638 int rc = 0; 639 640 ini->is_smcd = true; 641 ini->ism_gid = aclc->gid; 642 ini->srv_first_contact = aclc->hdr.flag; 643 644 /* there is only one lgr role for SMC-D; use server lock */ 645 mutex_lock(&smc_server_lgr_pending); 646 rc = smc_conn_create(smc, ini); 647 if (rc) { 648 mutex_unlock(&smc_server_lgr_pending); 649 return rc; 650 } 651 652 /* Create send and receive buffers */ 653 if (smc_buf_create(smc, true)) 654 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 655 ini->cln_first_contact); 656 657 smc_conn_save_peer_info(smc, aclc); 658 smc_close_init(smc); 659 smc_rx_init(smc); 660 smc_tx_init(smc); 661 662 rc = smc_clc_send_confirm(smc); 663 if (rc) 664 return smc_connect_abort(smc, rc, ini->cln_first_contact); 665 mutex_unlock(&smc_server_lgr_pending); 666 667 smc_copy_sock_settings_to_clc(smc); 668 smc->connect_nonblock = 0; 669 if (smc->sk.sk_state == SMC_INIT) 670 smc->sk.sk_state = SMC_ACTIVE; 671 672 return 0; 673 } 674 675 /* perform steps before actually connecting */ 676 static int __smc_connect(struct smc_sock *smc) 677 { 678 bool ism_supported = false, rdma_supported = false; 679 struct smc_clc_msg_accept_confirm aclc; 680 struct smc_init_info ini = {0}; 681 int smc_type; 682 int rc = 0; 683 684 sock_hold(&smc->sk); /* sock put in passive closing */ 685 686 if (smc->use_fallback) 687 return smc_connect_fallback(smc, smc->fallback_rsn); 688 689 /* if peer has not signalled SMC-capability, fall back */ 690 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 691 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 692 693 /* IPSec connections opt out of SMC-R optimizations */ 694 if (using_ipsec(smc)) 695 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 696 697 /* get vlan id from IP device */ 698 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 699 return smc_connect_decline_fallback(smc, 700 SMC_CLC_DECL_GETVLANERR); 701 702 /* check if there is an ism device available */ 703 if (!smc_find_ism_device(smc, &ini) && 704 !smc_connect_ism_vlan_setup(smc, &ini)) { 705 /* ISM is supported for this connection */ 706 ism_supported = true; 707 smc_type = SMC_TYPE_D; 708 } 709 710 /* check if there is a rdma device available */ 711 if (!smc_find_rdma_device(smc, &ini)) { 712 /* RDMA is supported for this connection */ 713 rdma_supported = true; 714 if (ism_supported) 715 smc_type = SMC_TYPE_B; /* both */ 716 else 717 smc_type = SMC_TYPE_R; /* only RDMA */ 718 } 719 720 /* if neither ISM nor RDMA are supported, fallback */ 721 if (!rdma_supported && !ism_supported) 722 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 723 724 /* perform CLC handshake */ 725 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 726 if (rc) { 727 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 728 return smc_connect_decline_fallback(smc, rc); 729 } 730 731 /* depending on previous steps, connect using rdma or ism */ 732 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 733 rc = smc_connect_rdma(smc, &aclc, &ini); 734 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 735 rc = smc_connect_ism(smc, &aclc, &ini); 736 else 737 rc = SMC_CLC_DECL_MODEUNSUPP; 738 if (rc) { 739 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 740 return smc_connect_decline_fallback(smc, rc); 741 } 742 743 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 744 return 0; 745 } 746 747 static void smc_connect_work(struct work_struct *work) 748 { 749 struct smc_sock *smc = container_of(work, struct smc_sock, 750 connect_work); 751 long timeo = smc->sk.sk_sndtimeo; 752 int rc = 0; 753 754 if (!timeo) 755 timeo = MAX_SCHEDULE_TIMEOUT; 756 lock_sock(smc->clcsock->sk); 757 if (smc->clcsock->sk->sk_err) { 758 smc->sk.sk_err = smc->clcsock->sk->sk_err; 759 } else if ((1 << smc->clcsock->sk->sk_state) & 760 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 761 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 762 if ((rc == -EPIPE) && 763 ((1 << smc->clcsock->sk->sk_state) & 764 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 765 rc = 0; 766 } 767 release_sock(smc->clcsock->sk); 768 lock_sock(&smc->sk); 769 if (rc != 0 || smc->sk.sk_err) { 770 smc->sk.sk_state = SMC_CLOSED; 771 if (rc == -EPIPE || rc == -EAGAIN) 772 smc->sk.sk_err = EPIPE; 773 else if (signal_pending(current)) 774 smc->sk.sk_err = -sock_intr_errno(timeo); 775 goto out; 776 } 777 778 rc = __smc_connect(smc); 779 if (rc < 0) 780 smc->sk.sk_err = -rc; 781 782 out: 783 if (smc->sk.sk_err) 784 smc->sk.sk_state_change(&smc->sk); 785 else 786 smc->sk.sk_write_space(&smc->sk); 787 release_sock(&smc->sk); 788 } 789 790 static int smc_connect(struct socket *sock, struct sockaddr *addr, 791 int alen, int flags) 792 { 793 struct sock *sk = sock->sk; 794 struct smc_sock *smc; 795 int rc = -EINVAL; 796 797 smc = smc_sk(sk); 798 799 /* separate smc parameter checking to be safe */ 800 if (alen < sizeof(addr->sa_family)) 801 goto out_err; 802 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 803 goto out_err; 804 805 lock_sock(sk); 806 switch (sk->sk_state) { 807 default: 808 goto out; 809 case SMC_ACTIVE: 810 rc = -EISCONN; 811 goto out; 812 case SMC_INIT: 813 rc = 0; 814 break; 815 } 816 817 smc_copy_sock_settings_to_clc(smc); 818 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 819 if (smc->connect_nonblock) { 820 rc = -EALREADY; 821 goto out; 822 } 823 rc = kernel_connect(smc->clcsock, addr, alen, flags); 824 if (rc && rc != -EINPROGRESS) 825 goto out; 826 if (flags & O_NONBLOCK) { 827 if (schedule_work(&smc->connect_work)) 828 smc->connect_nonblock = 1; 829 rc = -EINPROGRESS; 830 } else { 831 rc = __smc_connect(smc); 832 if (rc < 0) 833 goto out; 834 else 835 rc = 0; /* success cases including fallback */ 836 } 837 838 out: 839 release_sock(sk); 840 out_err: 841 return rc; 842 } 843 844 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 845 { 846 struct socket *new_clcsock = NULL; 847 struct sock *lsk = &lsmc->sk; 848 struct sock *new_sk; 849 int rc = -EINVAL; 850 851 release_sock(lsk); 852 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 853 if (!new_sk) { 854 rc = -ENOMEM; 855 lsk->sk_err = ENOMEM; 856 *new_smc = NULL; 857 lock_sock(lsk); 858 goto out; 859 } 860 *new_smc = smc_sk(new_sk); 861 862 mutex_lock(&lsmc->clcsock_release_lock); 863 if (lsmc->clcsock) 864 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 865 mutex_unlock(&lsmc->clcsock_release_lock); 866 lock_sock(lsk); 867 if (rc < 0) 868 lsk->sk_err = -rc; 869 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 870 if (new_clcsock) 871 sock_release(new_clcsock); 872 new_sk->sk_state = SMC_CLOSED; 873 sock_set_flag(new_sk, SOCK_DEAD); 874 new_sk->sk_prot->unhash(new_sk); 875 sock_put(new_sk); /* final */ 876 *new_smc = NULL; 877 goto out; 878 } 879 880 (*new_smc)->clcsock = new_clcsock; 881 out: 882 return rc; 883 } 884 885 /* add a just created sock to the accept queue of the listen sock as 886 * candidate for a following socket accept call from user space 887 */ 888 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 889 { 890 struct smc_sock *par = smc_sk(parent); 891 892 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 893 spin_lock(&par->accept_q_lock); 894 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 895 spin_unlock(&par->accept_q_lock); 896 sk_acceptq_added(parent); 897 } 898 899 /* remove a socket from the accept queue of its parental listening socket */ 900 static void smc_accept_unlink(struct sock *sk) 901 { 902 struct smc_sock *par = smc_sk(sk)->listen_smc; 903 904 spin_lock(&par->accept_q_lock); 905 list_del_init(&smc_sk(sk)->accept_q); 906 spin_unlock(&par->accept_q_lock); 907 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 908 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 909 } 910 911 /* remove a sock from the accept queue to bind it to a new socket created 912 * for a socket accept call from user space 913 */ 914 struct sock *smc_accept_dequeue(struct sock *parent, 915 struct socket *new_sock) 916 { 917 struct smc_sock *isk, *n; 918 struct sock *new_sk; 919 920 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 921 new_sk = (struct sock *)isk; 922 923 smc_accept_unlink(new_sk); 924 if (new_sk->sk_state == SMC_CLOSED) { 925 if (isk->clcsock) { 926 sock_release(isk->clcsock); 927 isk->clcsock = NULL; 928 } 929 new_sk->sk_prot->unhash(new_sk); 930 sock_put(new_sk); /* final */ 931 continue; 932 } 933 if (new_sock) 934 sock_graft(new_sk, new_sock); 935 return new_sk; 936 } 937 return NULL; 938 } 939 940 /* clean up for a created but never accepted sock */ 941 void smc_close_non_accepted(struct sock *sk) 942 { 943 struct smc_sock *smc = smc_sk(sk); 944 945 lock_sock(sk); 946 if (!sk->sk_lingertime) 947 /* wait for peer closing */ 948 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 949 if (!smc->use_fallback) { 950 smc_close_active(smc); 951 sock_set_flag(sk, SOCK_DEAD); 952 sk->sk_shutdown |= SHUTDOWN_MASK; 953 } 954 if (smc->clcsock) { 955 struct socket *tcp; 956 957 tcp = smc->clcsock; 958 smc->clcsock = NULL; 959 sock_release(tcp); 960 } 961 if (smc->use_fallback) { 962 sock_put(sk); /* passive closing */ 963 sk->sk_state = SMC_CLOSED; 964 } else { 965 if (sk->sk_state == SMC_CLOSED) 966 smc_conn_free(&smc->conn); 967 } 968 release_sock(sk); 969 sk->sk_prot->unhash(sk); 970 sock_put(sk); /* final sock_put */ 971 } 972 973 static int smc_serv_conf_first_link(struct smc_sock *smc) 974 { 975 struct net *net = sock_net(smc->clcsock->sk); 976 struct smc_link_group *lgr = smc->conn.lgr; 977 struct smc_link *link; 978 int rest; 979 int rc; 980 981 link = &lgr->lnk[SMC_SINGLE_LINK]; 982 983 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 984 return SMC_CLC_DECL_ERR_REGRMB; 985 986 /* send CONFIRM LINK request to client over the RoCE fabric */ 987 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 988 if (rc < 0) 989 return SMC_CLC_DECL_TIMEOUT_CL; 990 991 /* receive CONFIRM LINK response from client over the RoCE fabric */ 992 rest = wait_for_completion_interruptible_timeout( 993 &link->llc_confirm_resp, 994 SMC_LLC_WAIT_FIRST_TIME); 995 if (rest <= 0) { 996 struct smc_clc_msg_decline dclc; 997 998 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 999 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1000 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1001 } 1002 1003 if (link->llc_confirm_resp_rc) 1004 return SMC_CLC_DECL_RMBE_EC; 1005 1006 /* send ADD LINK request to client over the RoCE fabric */ 1007 rc = smc_llc_send_add_link(link, 1008 link->smcibdev->mac[link->ibport - 1], 1009 link->gid, SMC_LLC_REQ); 1010 if (rc < 0) 1011 return SMC_CLC_DECL_TIMEOUT_AL; 1012 1013 /* receive ADD LINK response from client over the RoCE fabric */ 1014 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1015 SMC_LLC_WAIT_TIME); 1016 if (rest <= 0) { 1017 struct smc_clc_msg_decline dclc; 1018 1019 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1020 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1021 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1022 } 1023 1024 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1025 1026 return 0; 1027 } 1028 1029 /* listen worker: finish */ 1030 static void smc_listen_out(struct smc_sock *new_smc) 1031 { 1032 struct smc_sock *lsmc = new_smc->listen_smc; 1033 struct sock *newsmcsk = &new_smc->sk; 1034 1035 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1036 if (lsmc->sk.sk_state == SMC_LISTEN) { 1037 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1038 } else { /* no longer listening */ 1039 smc_close_non_accepted(newsmcsk); 1040 } 1041 release_sock(&lsmc->sk); 1042 1043 /* Wake up accept */ 1044 lsmc->sk.sk_data_ready(&lsmc->sk); 1045 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1046 } 1047 1048 /* listen worker: finish in state connected */ 1049 static void smc_listen_out_connected(struct smc_sock *new_smc) 1050 { 1051 struct sock *newsmcsk = &new_smc->sk; 1052 1053 sk_refcnt_debug_inc(newsmcsk); 1054 if (newsmcsk->sk_state == SMC_INIT) 1055 newsmcsk->sk_state = SMC_ACTIVE; 1056 1057 smc_listen_out(new_smc); 1058 } 1059 1060 /* listen worker: finish in error state */ 1061 static void smc_listen_out_err(struct smc_sock *new_smc) 1062 { 1063 struct sock *newsmcsk = &new_smc->sk; 1064 1065 if (newsmcsk->sk_state == SMC_INIT) 1066 sock_put(&new_smc->sk); /* passive closing */ 1067 newsmcsk->sk_state = SMC_CLOSED; 1068 smc_conn_free(&new_smc->conn); 1069 1070 smc_listen_out(new_smc); 1071 } 1072 1073 /* listen worker: decline and fall back if possible */ 1074 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1075 int local_contact) 1076 { 1077 /* RDMA setup failed, switch back to TCP */ 1078 if (local_contact == SMC_FIRST_CONTACT) 1079 smc_lgr_forget(new_smc->conn.lgr); 1080 if (reason_code < 0) { /* error, no fallback possible */ 1081 smc_listen_out_err(new_smc); 1082 return; 1083 } 1084 smc_conn_free(&new_smc->conn); 1085 new_smc->use_fallback = true; 1086 new_smc->fallback_rsn = reason_code; 1087 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1088 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1089 smc_listen_out_err(new_smc); 1090 return; 1091 } 1092 } 1093 smc_listen_out_connected(new_smc); 1094 } 1095 1096 /* listen worker: check prefixes */ 1097 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1098 struct smc_clc_msg_proposal *pclc) 1099 { 1100 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1101 struct socket *newclcsock = new_smc->clcsock; 1102 1103 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1104 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1105 return SMC_CLC_DECL_DIFFPREFIX; 1106 1107 return 0; 1108 } 1109 1110 /* listen worker: initialize connection and buffers */ 1111 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1112 struct smc_init_info *ini) 1113 { 1114 int rc; 1115 1116 /* allocate connection / link group */ 1117 rc = smc_conn_create(new_smc, ini); 1118 if (rc) 1119 return rc; 1120 1121 /* create send buffer and rmb */ 1122 if (smc_buf_create(new_smc, false)) 1123 return SMC_CLC_DECL_MEM; 1124 1125 return 0; 1126 } 1127 1128 /* listen worker: initialize connection and buffers for SMC-D */ 1129 static int smc_listen_ism_init(struct smc_sock *new_smc, 1130 struct smc_clc_msg_proposal *pclc, 1131 struct smc_init_info *ini) 1132 { 1133 struct smc_clc_msg_smcd *pclc_smcd; 1134 int rc; 1135 1136 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1137 ini->ism_gid = pclc_smcd->gid; 1138 rc = smc_conn_create(new_smc, ini); 1139 if (rc) 1140 return rc; 1141 1142 /* Check if peer can be reached via ISM device */ 1143 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1144 new_smc->conn.lgr->vlan_id, 1145 new_smc->conn.lgr->smcd)) { 1146 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1147 smc_lgr_forget(new_smc->conn.lgr); 1148 smc_conn_free(&new_smc->conn); 1149 return SMC_CLC_DECL_SMCDNOTALK; 1150 } 1151 1152 /* Create send and receive buffers */ 1153 if (smc_buf_create(new_smc, true)) { 1154 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1155 smc_lgr_forget(new_smc->conn.lgr); 1156 smc_conn_free(&new_smc->conn); 1157 return SMC_CLC_DECL_MEM; 1158 } 1159 1160 return 0; 1161 } 1162 1163 /* listen worker: register buffers */ 1164 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1165 { 1166 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1167 1168 if (local_contact != SMC_FIRST_CONTACT) { 1169 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1170 return SMC_CLC_DECL_ERR_REGRMB; 1171 } 1172 smc_rmb_sync_sg_for_device(&new_smc->conn); 1173 1174 return 0; 1175 } 1176 1177 /* listen worker: finish RDMA setup */ 1178 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1179 struct smc_clc_msg_accept_confirm *cclc, 1180 int local_contact) 1181 { 1182 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1183 int reason_code = 0; 1184 1185 if (local_contact == SMC_FIRST_CONTACT) 1186 smc_link_save_peer_info(link, cclc); 1187 1188 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1189 reason_code = SMC_CLC_DECL_ERR_RTOK; 1190 goto decline; 1191 } 1192 1193 if (local_contact == SMC_FIRST_CONTACT) { 1194 if (smc_ib_ready_link(link)) { 1195 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1196 goto decline; 1197 } 1198 /* QP confirmation over RoCE fabric */ 1199 reason_code = smc_serv_conf_first_link(new_smc); 1200 if (reason_code) 1201 goto decline; 1202 } 1203 return 0; 1204 1205 decline: 1206 smc_listen_decline(new_smc, reason_code, local_contact); 1207 return reason_code; 1208 } 1209 1210 /* setup for RDMA connection of server */ 1211 static void smc_listen_work(struct work_struct *work) 1212 { 1213 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1214 smc_listen_work); 1215 struct socket *newclcsock = new_smc->clcsock; 1216 struct smc_clc_msg_accept_confirm cclc; 1217 struct smc_clc_msg_proposal *pclc; 1218 struct smc_init_info ini = {0}; 1219 bool ism_supported = false; 1220 u8 buf[SMC_CLC_MAX_LEN]; 1221 int rc = 0; 1222 1223 if (new_smc->use_fallback) { 1224 smc_listen_out_connected(new_smc); 1225 return; 1226 } 1227 1228 /* check if peer is smc capable */ 1229 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1230 new_smc->use_fallback = true; 1231 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1232 smc_listen_out_connected(new_smc); 1233 return; 1234 } 1235 1236 /* do inband token exchange - 1237 * wait for and receive SMC Proposal CLC message 1238 */ 1239 pclc = (struct smc_clc_msg_proposal *)&buf; 1240 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1241 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1242 if (rc) 1243 goto out_decl; 1244 1245 /* IPSec connections opt out of SMC-R optimizations */ 1246 if (using_ipsec(new_smc)) { 1247 rc = SMC_CLC_DECL_IPSEC; 1248 goto out_decl; 1249 } 1250 1251 /* check for matching IP prefix and subnet length */ 1252 rc = smc_listen_prfx_check(new_smc, pclc); 1253 if (rc) 1254 goto out_decl; 1255 1256 /* get vlan id from IP device */ 1257 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1258 rc = SMC_CLC_DECL_GETVLANERR; 1259 goto out_decl; 1260 } 1261 1262 mutex_lock(&smc_server_lgr_pending); 1263 smc_close_init(new_smc); 1264 smc_rx_init(new_smc); 1265 smc_tx_init(new_smc); 1266 1267 /* check if ISM is available */ 1268 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1269 ini.is_smcd = true; /* prepare ISM check */ 1270 rc = smc_find_ism_device(new_smc, &ini); 1271 if (!rc) 1272 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1273 if (!rc) 1274 ism_supported = true; 1275 else if (pclc->hdr.path == SMC_TYPE_D) 1276 goto out_unlock; /* skip RDMA and decline */ 1277 } 1278 1279 /* check if RDMA is available */ 1280 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1281 /* prepare RDMA check */ 1282 memset(&ini, 0, sizeof(ini)); 1283 ini.is_smcd = false; 1284 ini.ib_lcl = &pclc->lcl; 1285 rc = smc_find_rdma_device(new_smc, &ini); 1286 if (rc) { 1287 /* no RDMA device found */ 1288 if (pclc->hdr.path == SMC_TYPE_B) 1289 /* neither ISM nor RDMA device found */ 1290 rc = SMC_CLC_DECL_NOSMCDEV; 1291 goto out_unlock; 1292 } 1293 rc = smc_listen_rdma_init(new_smc, &ini); 1294 if (rc) 1295 goto out_unlock; 1296 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1297 if (rc) 1298 goto out_unlock; 1299 } 1300 1301 /* send SMC Accept CLC message */ 1302 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1303 if (rc) 1304 goto out_unlock; 1305 1306 /* SMC-D does not need this lock any more */ 1307 if (ism_supported) 1308 mutex_unlock(&smc_server_lgr_pending); 1309 1310 /* receive SMC Confirm CLC message */ 1311 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1312 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1313 if (rc) { 1314 if (!ism_supported) 1315 goto out_unlock; 1316 goto out_decl; 1317 } 1318 1319 /* finish worker */ 1320 if (!ism_supported) { 1321 rc = smc_listen_rdma_finish(new_smc, &cclc, 1322 ini.cln_first_contact); 1323 mutex_unlock(&smc_server_lgr_pending); 1324 if (rc) 1325 return; 1326 } 1327 smc_conn_save_peer_info(new_smc, &cclc); 1328 smc_listen_out_connected(new_smc); 1329 return; 1330 1331 out_unlock: 1332 mutex_unlock(&smc_server_lgr_pending); 1333 out_decl: 1334 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1335 } 1336 1337 static void smc_tcp_listen_work(struct work_struct *work) 1338 { 1339 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1340 tcp_listen_work); 1341 struct sock *lsk = &lsmc->sk; 1342 struct smc_sock *new_smc; 1343 int rc = 0; 1344 1345 lock_sock(lsk); 1346 while (lsk->sk_state == SMC_LISTEN) { 1347 rc = smc_clcsock_accept(lsmc, &new_smc); 1348 if (rc) 1349 goto out; 1350 if (!new_smc) 1351 continue; 1352 1353 new_smc->listen_smc = lsmc; 1354 new_smc->use_fallback = lsmc->use_fallback; 1355 new_smc->fallback_rsn = lsmc->fallback_rsn; 1356 sock_hold(lsk); /* sock_put in smc_listen_work */ 1357 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1358 smc_copy_sock_settings_to_smc(new_smc); 1359 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1360 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1361 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1362 if (!schedule_work(&new_smc->smc_listen_work)) 1363 sock_put(&new_smc->sk); 1364 } 1365 1366 out: 1367 release_sock(lsk); 1368 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1369 } 1370 1371 static int smc_listen(struct socket *sock, int backlog) 1372 { 1373 struct sock *sk = sock->sk; 1374 struct smc_sock *smc; 1375 int rc; 1376 1377 smc = smc_sk(sk); 1378 lock_sock(sk); 1379 1380 rc = -EINVAL; 1381 if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN)) 1382 goto out; 1383 1384 rc = 0; 1385 if (sk->sk_state == SMC_LISTEN) { 1386 sk->sk_max_ack_backlog = backlog; 1387 goto out; 1388 } 1389 /* some socket options are handled in core, so we could not apply 1390 * them to the clc socket -- copy smc socket options to clc socket 1391 */ 1392 smc_copy_sock_settings_to_clc(smc); 1393 if (!smc->use_fallback) 1394 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1395 1396 rc = kernel_listen(smc->clcsock, backlog); 1397 if (rc) 1398 goto out; 1399 sk->sk_max_ack_backlog = backlog; 1400 sk->sk_ack_backlog = 0; 1401 sk->sk_state = SMC_LISTEN; 1402 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1403 if (!schedule_work(&smc->tcp_listen_work)) 1404 sock_put(sk); 1405 1406 out: 1407 release_sock(sk); 1408 return rc; 1409 } 1410 1411 static int smc_accept(struct socket *sock, struct socket *new_sock, 1412 int flags, bool kern) 1413 { 1414 struct sock *sk = sock->sk, *nsk; 1415 DECLARE_WAITQUEUE(wait, current); 1416 struct smc_sock *lsmc; 1417 long timeo; 1418 int rc = 0; 1419 1420 lsmc = smc_sk(sk); 1421 sock_hold(sk); /* sock_put below */ 1422 lock_sock(sk); 1423 1424 if (lsmc->sk.sk_state != SMC_LISTEN) { 1425 rc = -EINVAL; 1426 release_sock(sk); 1427 goto out; 1428 } 1429 1430 /* Wait for an incoming connection */ 1431 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1432 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1433 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1434 set_current_state(TASK_INTERRUPTIBLE); 1435 if (!timeo) { 1436 rc = -EAGAIN; 1437 break; 1438 } 1439 release_sock(sk); 1440 timeo = schedule_timeout(timeo); 1441 /* wakeup by sk_data_ready in smc_listen_work() */ 1442 sched_annotate_sleep(); 1443 lock_sock(sk); 1444 if (signal_pending(current)) { 1445 rc = sock_intr_errno(timeo); 1446 break; 1447 } 1448 } 1449 set_current_state(TASK_RUNNING); 1450 remove_wait_queue(sk_sleep(sk), &wait); 1451 1452 if (!rc) 1453 rc = sock_error(nsk); 1454 release_sock(sk); 1455 if (rc) 1456 goto out; 1457 1458 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1459 /* wait till data arrives on the socket */ 1460 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1461 MSEC_PER_SEC); 1462 if (smc_sk(nsk)->use_fallback) { 1463 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1464 1465 lock_sock(clcsk); 1466 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1467 sk_wait_data(clcsk, &timeo, NULL); 1468 release_sock(clcsk); 1469 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1470 lock_sock(nsk); 1471 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1472 release_sock(nsk); 1473 } 1474 } 1475 1476 out: 1477 sock_put(sk); /* sock_hold above */ 1478 return rc; 1479 } 1480 1481 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1482 int peer) 1483 { 1484 struct smc_sock *smc; 1485 1486 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1487 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1488 return -ENOTCONN; 1489 1490 smc = smc_sk(sock->sk); 1491 1492 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1493 } 1494 1495 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1496 { 1497 struct sock *sk = sock->sk; 1498 struct smc_sock *smc; 1499 int rc = -EPIPE; 1500 1501 smc = smc_sk(sk); 1502 lock_sock(sk); 1503 if ((sk->sk_state != SMC_ACTIVE) && 1504 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1505 (sk->sk_state != SMC_INIT)) 1506 goto out; 1507 1508 if (msg->msg_flags & MSG_FASTOPEN) { 1509 if (sk->sk_state == SMC_INIT) { 1510 smc->use_fallback = true; 1511 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1512 } else { 1513 rc = -EINVAL; 1514 goto out; 1515 } 1516 } 1517 1518 if (smc->use_fallback) 1519 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1520 else 1521 rc = smc_tx_sendmsg(smc, msg, len); 1522 out: 1523 release_sock(sk); 1524 return rc; 1525 } 1526 1527 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1528 int flags) 1529 { 1530 struct sock *sk = sock->sk; 1531 struct smc_sock *smc; 1532 int rc = -ENOTCONN; 1533 1534 smc = smc_sk(sk); 1535 lock_sock(sk); 1536 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1537 /* socket was connected before, no more data to read */ 1538 rc = 0; 1539 goto out; 1540 } 1541 if ((sk->sk_state == SMC_INIT) || 1542 (sk->sk_state == SMC_LISTEN) || 1543 (sk->sk_state == SMC_CLOSED)) 1544 goto out; 1545 1546 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1547 rc = 0; 1548 goto out; 1549 } 1550 1551 if (smc->use_fallback) { 1552 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1553 } else { 1554 msg->msg_namelen = 0; 1555 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1556 } 1557 1558 out: 1559 release_sock(sk); 1560 return rc; 1561 } 1562 1563 static __poll_t smc_accept_poll(struct sock *parent) 1564 { 1565 struct smc_sock *isk = smc_sk(parent); 1566 __poll_t mask = 0; 1567 1568 spin_lock(&isk->accept_q_lock); 1569 if (!list_empty(&isk->accept_q)) 1570 mask = EPOLLIN | EPOLLRDNORM; 1571 spin_unlock(&isk->accept_q_lock); 1572 1573 return mask; 1574 } 1575 1576 static __poll_t smc_poll(struct file *file, struct socket *sock, 1577 poll_table *wait) 1578 { 1579 struct sock *sk = sock->sk; 1580 struct smc_sock *smc; 1581 __poll_t mask = 0; 1582 1583 if (!sk) 1584 return EPOLLNVAL; 1585 1586 smc = smc_sk(sock->sk); 1587 if (smc->use_fallback) { 1588 /* delegate to CLC child sock */ 1589 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1590 sk->sk_err = smc->clcsock->sk->sk_err; 1591 } else { 1592 if (sk->sk_state != SMC_CLOSED) 1593 sock_poll_wait(file, sock, wait); 1594 if (sk->sk_err) 1595 mask |= EPOLLERR; 1596 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1597 (sk->sk_state == SMC_CLOSED)) 1598 mask |= EPOLLHUP; 1599 if (sk->sk_state == SMC_LISTEN) { 1600 /* woken up by sk_data_ready in smc_listen_work() */ 1601 mask |= smc_accept_poll(sk); 1602 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1603 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1604 wait); 1605 sk->sk_err = smc->clcsock->sk->sk_err; 1606 } else { 1607 if ((sk->sk_state != SMC_INIT && 1608 atomic_read(&smc->conn.sndbuf_space)) || 1609 sk->sk_shutdown & SEND_SHUTDOWN) { 1610 mask |= EPOLLOUT | EPOLLWRNORM; 1611 } else { 1612 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1613 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1614 } 1615 if (atomic_read(&smc->conn.bytes_to_rcv)) 1616 mask |= EPOLLIN | EPOLLRDNORM; 1617 if (sk->sk_shutdown & RCV_SHUTDOWN) 1618 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1619 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1620 mask |= EPOLLIN; 1621 if (smc->conn.urg_state == SMC_URG_VALID) 1622 mask |= EPOLLPRI; 1623 } 1624 } 1625 1626 return mask; 1627 } 1628 1629 static int smc_shutdown(struct socket *sock, int how) 1630 { 1631 struct sock *sk = sock->sk; 1632 struct smc_sock *smc; 1633 int rc = -EINVAL; 1634 int rc1 = 0; 1635 1636 smc = smc_sk(sk); 1637 1638 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1639 return rc; 1640 1641 lock_sock(sk); 1642 1643 rc = -ENOTCONN; 1644 if ((sk->sk_state != SMC_ACTIVE) && 1645 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1646 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1647 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1648 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1649 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1650 goto out; 1651 if (smc->use_fallback) { 1652 rc = kernel_sock_shutdown(smc->clcsock, how); 1653 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1654 if (sk->sk_shutdown == SHUTDOWN_MASK) 1655 sk->sk_state = SMC_CLOSED; 1656 goto out; 1657 } 1658 switch (how) { 1659 case SHUT_RDWR: /* shutdown in both directions */ 1660 rc = smc_close_active(smc); 1661 break; 1662 case SHUT_WR: 1663 rc = smc_close_shutdown_write(smc); 1664 break; 1665 case SHUT_RD: 1666 rc = 0; 1667 /* nothing more to do because peer is not involved */ 1668 break; 1669 } 1670 if (smc->clcsock) 1671 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1672 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1673 sk->sk_shutdown |= how + 1; 1674 1675 out: 1676 release_sock(sk); 1677 return rc ? rc : rc1; 1678 } 1679 1680 static int smc_setsockopt(struct socket *sock, int level, int optname, 1681 char __user *optval, unsigned int optlen) 1682 { 1683 struct sock *sk = sock->sk; 1684 struct smc_sock *smc; 1685 int val, rc; 1686 1687 smc = smc_sk(sk); 1688 1689 /* generic setsockopts reaching us here always apply to the 1690 * CLC socket 1691 */ 1692 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1693 optval, optlen); 1694 if (smc->clcsock->sk->sk_err) { 1695 sk->sk_err = smc->clcsock->sk->sk_err; 1696 sk->sk_error_report(sk); 1697 } 1698 if (rc) 1699 return rc; 1700 1701 if (optlen < sizeof(int)) 1702 return -EINVAL; 1703 if (get_user(val, (int __user *)optval)) 1704 return -EFAULT; 1705 1706 lock_sock(sk); 1707 switch (optname) { 1708 case TCP_ULP: 1709 case TCP_FASTOPEN: 1710 case TCP_FASTOPEN_CONNECT: 1711 case TCP_FASTOPEN_KEY: 1712 case TCP_FASTOPEN_NO_COOKIE: 1713 /* option not supported by SMC */ 1714 if (sk->sk_state == SMC_INIT) { 1715 smc->use_fallback = true; 1716 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1717 } else { 1718 if (!smc->use_fallback) 1719 rc = -EINVAL; 1720 } 1721 break; 1722 case TCP_NODELAY: 1723 if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) { 1724 if (val && !smc->use_fallback) 1725 mod_delayed_work(system_wq, &smc->conn.tx_work, 1726 0); 1727 } 1728 break; 1729 case TCP_CORK: 1730 if (sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) { 1731 if (!val && !smc->use_fallback) 1732 mod_delayed_work(system_wq, &smc->conn.tx_work, 1733 0); 1734 } 1735 break; 1736 case TCP_DEFER_ACCEPT: 1737 smc->sockopt_defer_accept = val; 1738 break; 1739 default: 1740 break; 1741 } 1742 release_sock(sk); 1743 1744 return rc; 1745 } 1746 1747 static int smc_getsockopt(struct socket *sock, int level, int optname, 1748 char __user *optval, int __user *optlen) 1749 { 1750 struct smc_sock *smc; 1751 1752 smc = smc_sk(sock->sk); 1753 /* socket options apply to the CLC socket */ 1754 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1755 optval, optlen); 1756 } 1757 1758 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1759 unsigned long arg) 1760 { 1761 union smc_host_cursor cons, urg; 1762 struct smc_connection *conn; 1763 struct smc_sock *smc; 1764 int answ; 1765 1766 smc = smc_sk(sock->sk); 1767 conn = &smc->conn; 1768 lock_sock(&smc->sk); 1769 if (smc->use_fallback) { 1770 if (!smc->clcsock) { 1771 release_sock(&smc->sk); 1772 return -EBADF; 1773 } 1774 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1775 release_sock(&smc->sk); 1776 return answ; 1777 } 1778 switch (cmd) { 1779 case SIOCINQ: /* same as FIONREAD */ 1780 if (smc->sk.sk_state == SMC_LISTEN) { 1781 release_sock(&smc->sk); 1782 return -EINVAL; 1783 } 1784 if (smc->sk.sk_state == SMC_INIT || 1785 smc->sk.sk_state == SMC_CLOSED) 1786 answ = 0; 1787 else 1788 answ = atomic_read(&smc->conn.bytes_to_rcv); 1789 break; 1790 case SIOCOUTQ: 1791 /* output queue size (not send + not acked) */ 1792 if (smc->sk.sk_state == SMC_LISTEN) { 1793 release_sock(&smc->sk); 1794 return -EINVAL; 1795 } 1796 if (smc->sk.sk_state == SMC_INIT || 1797 smc->sk.sk_state == SMC_CLOSED) 1798 answ = 0; 1799 else 1800 answ = smc->conn.sndbuf_desc->len - 1801 atomic_read(&smc->conn.sndbuf_space); 1802 break; 1803 case SIOCOUTQNSD: 1804 /* output queue size (not send only) */ 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 = smc_tx_prepared_sends(&smc->conn); 1814 break; 1815 case SIOCATMARK: 1816 if (smc->sk.sk_state == SMC_LISTEN) { 1817 release_sock(&smc->sk); 1818 return -EINVAL; 1819 } 1820 if (smc->sk.sk_state == SMC_INIT || 1821 smc->sk.sk_state == SMC_CLOSED) { 1822 answ = 0; 1823 } else { 1824 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1825 smc_curs_copy(&urg, &conn->urg_curs, conn); 1826 answ = smc_curs_diff(conn->rmb_desc->len, 1827 &cons, &urg) == 1; 1828 } 1829 break; 1830 default: 1831 release_sock(&smc->sk); 1832 return -ENOIOCTLCMD; 1833 } 1834 release_sock(&smc->sk); 1835 1836 return put_user(answ, (int __user *)arg); 1837 } 1838 1839 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1840 int offset, size_t size, int flags) 1841 { 1842 struct sock *sk = sock->sk; 1843 struct smc_sock *smc; 1844 int rc = -EPIPE; 1845 1846 smc = smc_sk(sk); 1847 lock_sock(sk); 1848 if (sk->sk_state != SMC_ACTIVE) { 1849 release_sock(sk); 1850 goto out; 1851 } 1852 release_sock(sk); 1853 if (smc->use_fallback) 1854 rc = kernel_sendpage(smc->clcsock, page, offset, 1855 size, flags); 1856 else 1857 rc = sock_no_sendpage(sock, page, offset, size, flags); 1858 1859 out: 1860 return rc; 1861 } 1862 1863 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1864 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1865 * updates till whenever a respective page has been fully processed. 1866 * Note that subsequent recv() calls have to wait till all splice() processing 1867 * completed. 1868 */ 1869 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1870 struct pipe_inode_info *pipe, size_t len, 1871 unsigned int flags) 1872 { 1873 struct sock *sk = sock->sk; 1874 struct smc_sock *smc; 1875 int rc = -ENOTCONN; 1876 1877 smc = smc_sk(sk); 1878 lock_sock(sk); 1879 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1880 /* socket was connected before, no more data to read */ 1881 rc = 0; 1882 goto out; 1883 } 1884 if (sk->sk_state == SMC_INIT || 1885 sk->sk_state == SMC_LISTEN || 1886 sk->sk_state == SMC_CLOSED) 1887 goto out; 1888 1889 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1890 rc = 0; 1891 goto out; 1892 } 1893 1894 if (smc->use_fallback) { 1895 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1896 pipe, len, flags); 1897 } else { 1898 if (*ppos) { 1899 rc = -ESPIPE; 1900 goto out; 1901 } 1902 if (flags & SPLICE_F_NONBLOCK) 1903 flags = MSG_DONTWAIT; 1904 else 1905 flags = 0; 1906 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1907 } 1908 out: 1909 release_sock(sk); 1910 1911 return rc; 1912 } 1913 1914 /* must look like tcp */ 1915 static const struct proto_ops smc_sock_ops = { 1916 .family = PF_SMC, 1917 .owner = THIS_MODULE, 1918 .release = smc_release, 1919 .bind = smc_bind, 1920 .connect = smc_connect, 1921 .socketpair = sock_no_socketpair, 1922 .accept = smc_accept, 1923 .getname = smc_getname, 1924 .poll = smc_poll, 1925 .ioctl = smc_ioctl, 1926 .listen = smc_listen, 1927 .shutdown = smc_shutdown, 1928 .setsockopt = smc_setsockopt, 1929 .getsockopt = smc_getsockopt, 1930 .sendmsg = smc_sendmsg, 1931 .recvmsg = smc_recvmsg, 1932 .mmap = sock_no_mmap, 1933 .sendpage = smc_sendpage, 1934 .splice_read = smc_splice_read, 1935 }; 1936 1937 static int smc_create(struct net *net, struct socket *sock, int protocol, 1938 int kern) 1939 { 1940 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1941 struct smc_sock *smc; 1942 struct sock *sk; 1943 int rc; 1944 1945 rc = -ESOCKTNOSUPPORT; 1946 if (sock->type != SOCK_STREAM) 1947 goto out; 1948 1949 rc = -EPROTONOSUPPORT; 1950 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1951 goto out; 1952 1953 rc = -ENOBUFS; 1954 sock->ops = &smc_sock_ops; 1955 sk = smc_sock_alloc(net, sock, protocol); 1956 if (!sk) 1957 goto out; 1958 1959 /* create internal TCP socket for CLC handshake and fallback */ 1960 smc = smc_sk(sk); 1961 smc->use_fallback = false; /* assume rdma capability first */ 1962 smc->fallback_rsn = 0; 1963 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 1964 &smc->clcsock); 1965 if (rc) { 1966 sk_common_release(sk); 1967 goto out; 1968 } 1969 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 1970 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 1971 1972 out: 1973 return rc; 1974 } 1975 1976 static const struct net_proto_family smc_sock_family_ops = { 1977 .family = PF_SMC, 1978 .owner = THIS_MODULE, 1979 .create = smc_create, 1980 }; 1981 1982 unsigned int smc_net_id; 1983 1984 static __net_init int smc_net_init(struct net *net) 1985 { 1986 return smc_pnet_net_init(net); 1987 } 1988 1989 static void __net_exit smc_net_exit(struct net *net) 1990 { 1991 smc_pnet_net_exit(net); 1992 } 1993 1994 static struct pernet_operations smc_net_ops = { 1995 .init = smc_net_init, 1996 .exit = smc_net_exit, 1997 .id = &smc_net_id, 1998 .size = sizeof(struct smc_net), 1999 }; 2000 2001 static int __init smc_init(void) 2002 { 2003 int rc; 2004 2005 rc = register_pernet_subsys(&smc_net_ops); 2006 if (rc) 2007 return rc; 2008 2009 rc = smc_pnet_init(); 2010 if (rc) 2011 return rc; 2012 2013 rc = smc_llc_init(); 2014 if (rc) { 2015 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2016 goto out_pnet; 2017 } 2018 2019 rc = smc_cdc_init(); 2020 if (rc) { 2021 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2022 goto out_pnet; 2023 } 2024 2025 rc = proto_register(&smc_proto, 1); 2026 if (rc) { 2027 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2028 goto out_pnet; 2029 } 2030 2031 rc = proto_register(&smc_proto6, 1); 2032 if (rc) { 2033 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2034 goto out_proto; 2035 } 2036 2037 rc = sock_register(&smc_sock_family_ops); 2038 if (rc) { 2039 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2040 goto out_proto6; 2041 } 2042 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2043 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2044 2045 rc = smc_ib_register_client(); 2046 if (rc) { 2047 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2048 goto out_sock; 2049 } 2050 2051 static_branch_enable(&tcp_have_smc); 2052 return 0; 2053 2054 out_sock: 2055 sock_unregister(PF_SMC); 2056 out_proto6: 2057 proto_unregister(&smc_proto6); 2058 out_proto: 2059 proto_unregister(&smc_proto); 2060 out_pnet: 2061 smc_pnet_exit(); 2062 return rc; 2063 } 2064 2065 static void __exit smc_exit(void) 2066 { 2067 smc_core_exit(); 2068 static_branch_disable(&tcp_have_smc); 2069 smc_ib_unregister_client(); 2070 sock_unregister(PF_SMC); 2071 proto_unregister(&smc_proto6); 2072 proto_unregister(&smc_proto); 2073 smc_pnet_exit(); 2074 unregister_pernet_subsys(&smc_net_ops); 2075 } 2076 2077 module_init(smc_init); 2078 module_exit(smc_exit); 2079 2080 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2081 MODULE_DESCRIPTION("smc socket address family"); 2082 MODULE_LICENSE("GPL"); 2083 MODULE_ALIAS_NETPROTO(PF_SMC); 2084