1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type 6 * applies to SOCK_STREAM sockets only 7 * offers an alternative communication option for TCP-protocol sockets 8 * applicable with RoCE-cards only 9 * 10 * Initial restrictions: 11 * - support for alternate links postponed 12 * 13 * Copyright IBM Corp. 2016, 2018 14 * 15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 16 * based on prototype from Frank Blaschka 17 */ 18 19 #define KMSG_COMPONENT "smc" 20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt 21 22 #include <linux/module.h> 23 #include <linux/socket.h> 24 #include <linux/workqueue.h> 25 #include <linux/in.h> 26 #include <linux/sched/signal.h> 27 #include <linux/if_vlan.h> 28 #include <linux/rcupdate_wait.h> 29 30 #include <net/sock.h> 31 #include <net/tcp.h> 32 #include <net/smc.h> 33 #include <asm/ioctls.h> 34 35 #include <net/net_namespace.h> 36 #include <net/netns/generic.h> 37 #include "smc_netns.h" 38 39 #include "smc.h" 40 #include "smc_clc.h" 41 #include "smc_llc.h" 42 #include "smc_cdc.h" 43 #include "smc_core.h" 44 #include "smc_ib.h" 45 #include "smc_ism.h" 46 #include "smc_pnet.h" 47 #include "smc_tx.h" 48 #include "smc_rx.h" 49 #include "smc_close.h" 50 51 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 52 * creation on server 53 */ 54 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 55 * creation on client 56 */ 57 58 static void smc_tcp_listen_work(struct work_struct *); 59 static void smc_connect_work(struct work_struct *); 60 61 static void smc_set_keepalive(struct sock *sk, int val) 62 { 63 struct smc_sock *smc = smc_sk(sk); 64 65 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 66 } 67 68 static struct smc_hashinfo smc_v4_hashinfo = { 69 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 70 }; 71 72 static struct smc_hashinfo smc_v6_hashinfo = { 73 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 74 }; 75 76 int smc_hash_sk(struct sock *sk) 77 { 78 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 79 struct hlist_head *head; 80 81 head = &h->ht; 82 83 write_lock_bh(&h->lock); 84 sk_add_node(sk, head); 85 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 86 write_unlock_bh(&h->lock); 87 88 return 0; 89 } 90 EXPORT_SYMBOL_GPL(smc_hash_sk); 91 92 void smc_unhash_sk(struct sock *sk) 93 { 94 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 95 96 write_lock_bh(&h->lock); 97 if (sk_del_node_init(sk)) 98 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 99 write_unlock_bh(&h->lock); 100 } 101 EXPORT_SYMBOL_GPL(smc_unhash_sk); 102 103 struct proto smc_proto = { 104 .name = "SMC", 105 .owner = THIS_MODULE, 106 .keepalive = smc_set_keepalive, 107 .hash = smc_hash_sk, 108 .unhash = smc_unhash_sk, 109 .obj_size = sizeof(struct smc_sock), 110 .h.smc_hash = &smc_v4_hashinfo, 111 .slab_flags = SLAB_TYPESAFE_BY_RCU, 112 }; 113 EXPORT_SYMBOL_GPL(smc_proto); 114 115 struct proto smc_proto6 = { 116 .name = "SMC6", 117 .owner = THIS_MODULE, 118 .keepalive = smc_set_keepalive, 119 .hash = smc_hash_sk, 120 .unhash = smc_unhash_sk, 121 .obj_size = sizeof(struct smc_sock), 122 .h.smc_hash = &smc_v6_hashinfo, 123 .slab_flags = SLAB_TYPESAFE_BY_RCU, 124 }; 125 EXPORT_SYMBOL_GPL(smc_proto6); 126 127 static void smc_restore_fallback_changes(struct smc_sock *smc) 128 { 129 smc->clcsock->file->private_data = smc->sk.sk_socket; 130 smc->clcsock->file = NULL; 131 } 132 133 static int __smc_release(struct smc_sock *smc) 134 { 135 struct sock *sk = &smc->sk; 136 int rc = 0; 137 138 if (!smc->use_fallback) { 139 rc = smc_close_active(smc); 140 sock_set_flag(sk, SOCK_DEAD); 141 sk->sk_shutdown |= SHUTDOWN_MASK; 142 } else { 143 if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT) 144 sock_put(sk); /* passive closing */ 145 if (sk->sk_state == SMC_LISTEN) { 146 /* wake up clcsock accept */ 147 rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); 148 } 149 sk->sk_state = SMC_CLOSED; 150 sk->sk_state_change(sk); 151 smc_restore_fallback_changes(smc); 152 } 153 154 sk->sk_prot->unhash(sk); 155 156 if (sk->sk_state == SMC_CLOSED) { 157 if (smc->clcsock) { 158 release_sock(sk); 159 smc_clcsock_release(smc); 160 lock_sock(sk); 161 } 162 if (!smc->use_fallback) 163 smc_conn_free(&smc->conn); 164 } 165 166 return rc; 167 } 168 169 static int smc_release(struct socket *sock) 170 { 171 struct sock *sk = sock->sk; 172 struct smc_sock *smc; 173 int rc = 0; 174 175 if (!sk) 176 goto out; 177 178 sock_hold(sk); /* sock_put below */ 179 smc = smc_sk(sk); 180 181 /* cleanup for a dangling non-blocking connect */ 182 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 183 tcp_abort(smc->clcsock->sk, ECONNABORTED); 184 flush_work(&smc->connect_work); 185 186 if (sk->sk_state == SMC_LISTEN) 187 /* smc_close_non_accepted() is called and acquires 188 * sock lock for child sockets again 189 */ 190 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 191 else 192 lock_sock(sk); 193 194 rc = __smc_release(smc); 195 196 /* detach socket */ 197 sock_orphan(sk); 198 sock->sk = NULL; 199 release_sock(sk); 200 201 sock_put(sk); /* sock_hold above */ 202 sock_put(sk); /* final sock_put */ 203 out: 204 return rc; 205 } 206 207 static void smc_destruct(struct sock *sk) 208 { 209 if (sk->sk_state != SMC_CLOSED) 210 return; 211 if (!sock_flag(sk, SOCK_DEAD)) 212 return; 213 214 sk_refcnt_debug_dec(sk); 215 } 216 217 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 218 int protocol) 219 { 220 struct smc_sock *smc; 221 struct proto *prot; 222 struct sock *sk; 223 224 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 225 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 226 if (!sk) 227 return NULL; 228 229 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 230 sk->sk_state = SMC_INIT; 231 sk->sk_destruct = smc_destruct; 232 sk->sk_protocol = protocol; 233 smc = smc_sk(sk); 234 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 235 INIT_WORK(&smc->connect_work, smc_connect_work); 236 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 237 INIT_LIST_HEAD(&smc->accept_q); 238 spin_lock_init(&smc->accept_q_lock); 239 spin_lock_init(&smc->conn.send_lock); 240 sk->sk_prot->hash(sk); 241 sk_refcnt_debug_inc(sk); 242 mutex_init(&smc->clcsock_release_lock); 243 244 return sk; 245 } 246 247 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 248 int addr_len) 249 { 250 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 251 struct sock *sk = sock->sk; 252 struct smc_sock *smc; 253 int rc; 254 255 smc = smc_sk(sk); 256 257 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 258 rc = -EINVAL; 259 if (addr_len < sizeof(struct sockaddr_in)) 260 goto out; 261 262 rc = -EAFNOSUPPORT; 263 if (addr->sin_family != AF_INET && 264 addr->sin_family != AF_INET6 && 265 addr->sin_family != AF_UNSPEC) 266 goto out; 267 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 268 if (addr->sin_family == AF_UNSPEC && 269 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 270 goto out; 271 272 lock_sock(sk); 273 274 /* Check if socket is already active */ 275 rc = -EINVAL; 276 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 277 goto out_rel; 278 279 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 280 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 281 282 out_rel: 283 release_sock(sk); 284 out: 285 return rc; 286 } 287 288 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 289 unsigned long mask) 290 { 291 /* options we don't get control via setsockopt for */ 292 nsk->sk_type = osk->sk_type; 293 nsk->sk_sndbuf = osk->sk_sndbuf; 294 nsk->sk_rcvbuf = osk->sk_rcvbuf; 295 nsk->sk_sndtimeo = osk->sk_sndtimeo; 296 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 297 nsk->sk_mark = osk->sk_mark; 298 nsk->sk_priority = osk->sk_priority; 299 nsk->sk_rcvlowat = osk->sk_rcvlowat; 300 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 301 nsk->sk_err = osk->sk_err; 302 303 nsk->sk_flags &= ~mask; 304 nsk->sk_flags |= osk->sk_flags & mask; 305 } 306 307 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 308 (1UL << SOCK_KEEPOPEN) | \ 309 (1UL << SOCK_LINGER) | \ 310 (1UL << SOCK_BROADCAST) | \ 311 (1UL << SOCK_TIMESTAMP) | \ 312 (1UL << SOCK_DBG) | \ 313 (1UL << SOCK_RCVTSTAMP) | \ 314 (1UL << SOCK_RCVTSTAMPNS) | \ 315 (1UL << SOCK_LOCALROUTE) | \ 316 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 317 (1UL << SOCK_RXQ_OVFL) | \ 318 (1UL << SOCK_WIFI_STATUS) | \ 319 (1UL << SOCK_NOFCS) | \ 320 (1UL << SOCK_FILTER_LOCKED) | \ 321 (1UL << SOCK_TSTAMP_NEW)) 322 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 323 * clc socket (since smc is not called for these options from net/core) 324 */ 325 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 326 { 327 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 328 } 329 330 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 331 (1UL << SOCK_KEEPOPEN) | \ 332 (1UL << SOCK_LINGER) | \ 333 (1UL << SOCK_DBG)) 334 /* copy only settings and flags relevant for smc from clc to smc socket */ 335 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 336 { 337 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 338 } 339 340 /* register a new rmb, send confirm_rkey msg to register with peer */ 341 static int smc_reg_rmb(struct smc_link *link, struct smc_buf_desc *rmb_desc, 342 bool conf_rkey) 343 { 344 if (!rmb_desc->wr_reg) { 345 /* register memory region for new rmb */ 346 if (smc_wr_reg_send(link, rmb_desc->mr_rx[SMC_SINGLE_LINK])) { 347 rmb_desc->regerr = 1; 348 return -EFAULT; 349 } 350 rmb_desc->wr_reg = 1; 351 } 352 if (!conf_rkey) 353 return 0; 354 /* exchange confirm_rkey msg with peer */ 355 if (smc_llc_do_confirm_rkey(link, rmb_desc)) { 356 rmb_desc->regerr = 1; 357 return -EFAULT; 358 } 359 return 0; 360 } 361 362 static int smc_clnt_conf_first_link(struct smc_sock *smc) 363 { 364 struct net *net = sock_net(smc->clcsock->sk); 365 struct smc_link_group *lgr = smc->conn.lgr; 366 struct smc_link *link; 367 int rest; 368 int rc; 369 370 link = &lgr->lnk[SMC_SINGLE_LINK]; 371 /* receive CONFIRM LINK request from server over RoCE fabric */ 372 rest = wait_for_completion_interruptible_timeout( 373 &link->llc_confirm, 374 SMC_LLC_WAIT_FIRST_TIME); 375 if (rest <= 0) { 376 struct smc_clc_msg_decline dclc; 377 378 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 379 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 380 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 381 } 382 383 if (link->llc_confirm_rc) 384 return SMC_CLC_DECL_RMBE_EC; 385 386 rc = smc_ib_modify_qp_rts(link); 387 if (rc) 388 return SMC_CLC_DECL_ERR_RDYLNK; 389 390 smc_wr_remember_qp_attr(link); 391 392 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 393 return SMC_CLC_DECL_ERR_REGRMB; 394 395 /* send CONFIRM LINK response over RoCE fabric */ 396 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 397 if (rc < 0) 398 return SMC_CLC_DECL_TIMEOUT_CL; 399 400 /* receive ADD LINK request from server over RoCE fabric */ 401 rest = wait_for_completion_interruptible_timeout(&link->llc_add, 402 SMC_LLC_WAIT_TIME); 403 if (rest <= 0) { 404 struct smc_clc_msg_decline dclc; 405 406 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 407 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 408 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 409 } 410 411 /* send add link reject message, only one link supported for now */ 412 rc = smc_llc_send_add_link(link, 413 link->smcibdev->mac[link->ibport - 1], 414 link->gid, SMC_LLC_RESP); 415 if (rc < 0) 416 return SMC_CLC_DECL_TIMEOUT_AL; 417 418 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 419 420 return 0; 421 } 422 423 static void smcr_conn_save_peer_info(struct smc_sock *smc, 424 struct smc_clc_msg_accept_confirm *clc) 425 { 426 int bufsize = smc_uncompress_bufsize(clc->rmbe_size); 427 428 smc->conn.peer_rmbe_idx = clc->rmbe_idx; 429 smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token); 430 smc->conn.peer_rmbe_size = bufsize; 431 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 432 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 433 } 434 435 static void smcd_conn_save_peer_info(struct smc_sock *smc, 436 struct smc_clc_msg_accept_confirm *clc) 437 { 438 int bufsize = smc_uncompress_bufsize(clc->dmbe_size); 439 440 smc->conn.peer_rmbe_idx = clc->dmbe_idx; 441 smc->conn.peer_token = clc->token; 442 /* msg header takes up space in the buffer */ 443 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 444 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 445 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 446 } 447 448 static void smc_conn_save_peer_info(struct smc_sock *smc, 449 struct smc_clc_msg_accept_confirm *clc) 450 { 451 if (smc->conn.lgr->is_smcd) 452 smcd_conn_save_peer_info(smc, clc); 453 else 454 smcr_conn_save_peer_info(smc, clc); 455 } 456 457 static void smc_link_save_peer_info(struct smc_link *link, 458 struct smc_clc_msg_accept_confirm *clc) 459 { 460 link->peer_qpn = ntoh24(clc->qpn); 461 memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE); 462 memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac)); 463 link->peer_psn = ntoh24(clc->psn); 464 link->peer_mtu = clc->qp_mtu; 465 } 466 467 static void smc_switch_to_fallback(struct smc_sock *smc) 468 { 469 smc->use_fallback = true; 470 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 471 smc->clcsock->file = smc->sk.sk_socket->file; 472 smc->clcsock->file->private_data = smc->clcsock; 473 } 474 } 475 476 /* fall back during connect */ 477 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 478 { 479 smc_switch_to_fallback(smc); 480 smc->fallback_rsn = reason_code; 481 smc_copy_sock_settings_to_clc(smc); 482 smc->connect_nonblock = 0; 483 if (smc->sk.sk_state == SMC_INIT) 484 smc->sk.sk_state = SMC_ACTIVE; 485 return 0; 486 } 487 488 /* decline and fall back during connect */ 489 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code) 490 { 491 int rc; 492 493 if (reason_code < 0) { /* error, fallback is not possible */ 494 if (smc->sk.sk_state == SMC_INIT) 495 sock_put(&smc->sk); /* passive closing */ 496 return reason_code; 497 } 498 if (reason_code != SMC_CLC_DECL_PEERDECL) { 499 rc = smc_clc_send_decline(smc, reason_code); 500 if (rc < 0) { 501 if (smc->sk.sk_state == SMC_INIT) 502 sock_put(&smc->sk); /* passive closing */ 503 return rc; 504 } 505 } 506 return smc_connect_fallback(smc, reason_code); 507 } 508 509 /* abort connecting */ 510 static int smc_connect_abort(struct smc_sock *smc, int reason_code, 511 int local_contact) 512 { 513 if (local_contact == SMC_FIRST_CONTACT) 514 smc_lgr_forget(smc->conn.lgr); 515 if (smc->conn.lgr->is_smcd) 516 /* there is only one lgr role for SMC-D; use server lock */ 517 mutex_unlock(&smc_server_lgr_pending); 518 else 519 mutex_unlock(&smc_client_lgr_pending); 520 521 smc_conn_free(&smc->conn); 522 smc->connect_nonblock = 0; 523 return reason_code; 524 } 525 526 /* check if there is a rdma device available for this connection. */ 527 /* called for connect and listen */ 528 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 529 { 530 /* PNET table look up: search active ib_device and port 531 * within same PNETID that also contains the ethernet device 532 * used for the internal TCP socket 533 */ 534 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 535 if (!ini->ib_dev) 536 return SMC_CLC_DECL_NOSMCRDEV; 537 return 0; 538 } 539 540 /* check if there is an ISM device available for this connection. */ 541 /* called for connect and listen */ 542 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 543 { 544 /* Find ISM device with same PNETID as connecting interface */ 545 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 546 if (!ini->ism_dev) 547 return SMC_CLC_DECL_NOSMCDDEV; 548 return 0; 549 } 550 551 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 552 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 553 struct smc_init_info *ini) 554 { 555 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 556 return SMC_CLC_DECL_ISMVLANERR; 557 return 0; 558 } 559 560 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 561 * used, the VLAN ID will be registered again during the connection setup. 562 */ 563 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 564 struct smc_init_info *ini) 565 { 566 if (!is_smcd) 567 return 0; 568 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 569 return SMC_CLC_DECL_CNFERR; 570 return 0; 571 } 572 573 /* CLC handshake during connect */ 574 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 575 struct smc_clc_msg_accept_confirm *aclc, 576 struct smc_init_info *ini) 577 { 578 int rc = 0; 579 580 /* do inband token exchange */ 581 rc = smc_clc_send_proposal(smc, smc_type, ini); 582 if (rc) 583 return rc; 584 /* receive SMC Accept CLC message */ 585 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 586 CLC_WAIT_TIME); 587 } 588 589 /* setup for RDMA connection of client */ 590 static int smc_connect_rdma(struct smc_sock *smc, 591 struct smc_clc_msg_accept_confirm *aclc, 592 struct smc_init_info *ini) 593 { 594 struct smc_link *link; 595 int reason_code = 0; 596 597 ini->is_smcd = false; 598 ini->ib_lcl = &aclc->lcl; 599 ini->ib_clcqpn = ntoh24(aclc->qpn); 600 ini->srv_first_contact = aclc->hdr.flag; 601 602 mutex_lock(&smc_client_lgr_pending); 603 reason_code = smc_conn_create(smc, ini); 604 if (reason_code) { 605 mutex_unlock(&smc_client_lgr_pending); 606 return reason_code; 607 } 608 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 609 610 smc_conn_save_peer_info(smc, aclc); 611 612 /* create send buffer and rmb */ 613 if (smc_buf_create(smc, false)) 614 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 615 ini->cln_first_contact); 616 617 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 618 smc_link_save_peer_info(link, aclc); 619 620 if (smc_rmb_rtoken_handling(&smc->conn, aclc)) 621 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 622 ini->cln_first_contact); 623 624 smc_close_init(smc); 625 smc_rx_init(smc); 626 627 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 628 if (smc_ib_ready_link(link)) 629 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 630 ini->cln_first_contact); 631 } else { 632 if (smc_reg_rmb(link, smc->conn.rmb_desc, true)) 633 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 634 ini->cln_first_contact); 635 } 636 smc_rmb_sync_sg_for_device(&smc->conn); 637 638 reason_code = smc_clc_send_confirm(smc); 639 if (reason_code) 640 return smc_connect_abort(smc, reason_code, 641 ini->cln_first_contact); 642 643 smc_tx_init(smc); 644 645 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 646 /* QP confirmation over RoCE fabric */ 647 reason_code = smc_clnt_conf_first_link(smc); 648 if (reason_code) 649 return smc_connect_abort(smc, reason_code, 650 ini->cln_first_contact); 651 } 652 mutex_unlock(&smc_client_lgr_pending); 653 654 smc_copy_sock_settings_to_clc(smc); 655 smc->connect_nonblock = 0; 656 if (smc->sk.sk_state == SMC_INIT) 657 smc->sk.sk_state = SMC_ACTIVE; 658 659 return 0; 660 } 661 662 /* setup for ISM connection of client */ 663 static int smc_connect_ism(struct smc_sock *smc, 664 struct smc_clc_msg_accept_confirm *aclc, 665 struct smc_init_info *ini) 666 { 667 int rc = 0; 668 669 ini->is_smcd = true; 670 ini->ism_gid = aclc->gid; 671 ini->srv_first_contact = aclc->hdr.flag; 672 673 /* there is only one lgr role for SMC-D; use server lock */ 674 mutex_lock(&smc_server_lgr_pending); 675 rc = smc_conn_create(smc, ini); 676 if (rc) { 677 mutex_unlock(&smc_server_lgr_pending); 678 return rc; 679 } 680 681 /* Create send and receive buffers */ 682 if (smc_buf_create(smc, true)) 683 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 684 ini->cln_first_contact); 685 686 smc_conn_save_peer_info(smc, aclc); 687 smc_close_init(smc); 688 smc_rx_init(smc); 689 smc_tx_init(smc); 690 691 rc = smc_clc_send_confirm(smc); 692 if (rc) 693 return smc_connect_abort(smc, rc, ini->cln_first_contact); 694 mutex_unlock(&smc_server_lgr_pending); 695 696 smc_copy_sock_settings_to_clc(smc); 697 smc->connect_nonblock = 0; 698 if (smc->sk.sk_state == SMC_INIT) 699 smc->sk.sk_state = SMC_ACTIVE; 700 701 return 0; 702 } 703 704 /* perform steps before actually connecting */ 705 static int __smc_connect(struct smc_sock *smc) 706 { 707 bool ism_supported = false, rdma_supported = false; 708 struct smc_clc_msg_accept_confirm aclc; 709 struct smc_init_info ini = {0}; 710 int smc_type; 711 int rc = 0; 712 713 if (smc->use_fallback) 714 return smc_connect_fallback(smc, smc->fallback_rsn); 715 716 /* if peer has not signalled SMC-capability, fall back */ 717 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 718 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 719 720 /* IPSec connections opt out of SMC-R optimizations */ 721 if (using_ipsec(smc)) 722 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 723 724 /* get vlan id from IP device */ 725 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 726 return smc_connect_decline_fallback(smc, 727 SMC_CLC_DECL_GETVLANERR); 728 729 /* check if there is an ism device available */ 730 if (!smc_find_ism_device(smc, &ini) && 731 !smc_connect_ism_vlan_setup(smc, &ini)) { 732 /* ISM is supported for this connection */ 733 ism_supported = true; 734 smc_type = SMC_TYPE_D; 735 } 736 737 /* check if there is a rdma device available */ 738 if (!smc_find_rdma_device(smc, &ini)) { 739 /* RDMA is supported for this connection */ 740 rdma_supported = true; 741 if (ism_supported) 742 smc_type = SMC_TYPE_B; /* both */ 743 else 744 smc_type = SMC_TYPE_R; /* only RDMA */ 745 } 746 747 /* if neither ISM nor RDMA are supported, fallback */ 748 if (!rdma_supported && !ism_supported) 749 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 750 751 /* perform CLC handshake */ 752 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 753 if (rc) { 754 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 755 return smc_connect_decline_fallback(smc, rc); 756 } 757 758 /* depending on previous steps, connect using rdma or ism */ 759 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 760 rc = smc_connect_rdma(smc, &aclc, &ini); 761 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 762 rc = smc_connect_ism(smc, &aclc, &ini); 763 else 764 rc = SMC_CLC_DECL_MODEUNSUPP; 765 if (rc) { 766 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 767 return smc_connect_decline_fallback(smc, rc); 768 } 769 770 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 771 return 0; 772 } 773 774 static void smc_connect_work(struct work_struct *work) 775 { 776 struct smc_sock *smc = container_of(work, struct smc_sock, 777 connect_work); 778 long timeo = smc->sk.sk_sndtimeo; 779 int rc = 0; 780 781 if (!timeo) 782 timeo = MAX_SCHEDULE_TIMEOUT; 783 lock_sock(smc->clcsock->sk); 784 if (smc->clcsock->sk->sk_err) { 785 smc->sk.sk_err = smc->clcsock->sk->sk_err; 786 } else if ((1 << smc->clcsock->sk->sk_state) & 787 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 788 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 789 if ((rc == -EPIPE) && 790 ((1 << smc->clcsock->sk->sk_state) & 791 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 792 rc = 0; 793 } 794 release_sock(smc->clcsock->sk); 795 lock_sock(&smc->sk); 796 if (rc != 0 || smc->sk.sk_err) { 797 smc->sk.sk_state = SMC_CLOSED; 798 if (rc == -EPIPE || rc == -EAGAIN) 799 smc->sk.sk_err = EPIPE; 800 else if (signal_pending(current)) 801 smc->sk.sk_err = -sock_intr_errno(timeo); 802 sock_put(&smc->sk); /* passive closing */ 803 goto out; 804 } 805 806 rc = __smc_connect(smc); 807 if (rc < 0) 808 smc->sk.sk_err = -rc; 809 810 out: 811 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 812 if (smc->sk.sk_err) { 813 smc->sk.sk_state_change(&smc->sk); 814 } else { /* allow polling before and after fallback decision */ 815 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 816 smc->sk.sk_write_space(&smc->sk); 817 } 818 } 819 release_sock(&smc->sk); 820 } 821 822 static int smc_connect(struct socket *sock, struct sockaddr *addr, 823 int alen, int flags) 824 { 825 struct sock *sk = sock->sk; 826 struct smc_sock *smc; 827 int rc = -EINVAL; 828 829 smc = smc_sk(sk); 830 831 /* separate smc parameter checking to be safe */ 832 if (alen < sizeof(addr->sa_family)) 833 goto out_err; 834 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 835 goto out_err; 836 837 lock_sock(sk); 838 switch (sk->sk_state) { 839 default: 840 goto out; 841 case SMC_ACTIVE: 842 rc = -EISCONN; 843 goto out; 844 case SMC_INIT: 845 rc = 0; 846 break; 847 } 848 849 smc_copy_sock_settings_to_clc(smc); 850 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 851 if (smc->connect_nonblock) { 852 rc = -EALREADY; 853 goto out; 854 } 855 rc = kernel_connect(smc->clcsock, addr, alen, flags); 856 if (rc && rc != -EINPROGRESS) 857 goto out; 858 859 sock_hold(&smc->sk); /* sock put in passive closing */ 860 if (smc->use_fallback) 861 goto out; 862 if (flags & O_NONBLOCK) { 863 if (schedule_work(&smc->connect_work)) 864 smc->connect_nonblock = 1; 865 rc = -EINPROGRESS; 866 } else { 867 rc = __smc_connect(smc); 868 if (rc < 0) 869 goto out; 870 else 871 rc = 0; /* success cases including fallback */ 872 } 873 874 out: 875 release_sock(sk); 876 out_err: 877 return rc; 878 } 879 880 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 881 { 882 struct socket *new_clcsock = NULL; 883 struct sock *lsk = &lsmc->sk; 884 struct sock *new_sk; 885 int rc = -EINVAL; 886 887 release_sock(lsk); 888 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 889 if (!new_sk) { 890 rc = -ENOMEM; 891 lsk->sk_err = ENOMEM; 892 *new_smc = NULL; 893 lock_sock(lsk); 894 goto out; 895 } 896 *new_smc = smc_sk(new_sk); 897 898 mutex_lock(&lsmc->clcsock_release_lock); 899 if (lsmc->clcsock) 900 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 901 mutex_unlock(&lsmc->clcsock_release_lock); 902 lock_sock(lsk); 903 if (rc < 0) 904 lsk->sk_err = -rc; 905 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 906 new_sk->sk_prot->unhash(new_sk); 907 if (new_clcsock) 908 sock_release(new_clcsock); 909 new_sk->sk_state = SMC_CLOSED; 910 sock_set_flag(new_sk, SOCK_DEAD); 911 sock_put(new_sk); /* final */ 912 *new_smc = NULL; 913 goto out; 914 } 915 916 (*new_smc)->clcsock = new_clcsock; 917 out: 918 return rc; 919 } 920 921 /* add a just created sock to the accept queue of the listen sock as 922 * candidate for a following socket accept call from user space 923 */ 924 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 925 { 926 struct smc_sock *par = smc_sk(parent); 927 928 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 929 spin_lock(&par->accept_q_lock); 930 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 931 spin_unlock(&par->accept_q_lock); 932 sk_acceptq_added(parent); 933 } 934 935 /* remove a socket from the accept queue of its parental listening socket */ 936 static void smc_accept_unlink(struct sock *sk) 937 { 938 struct smc_sock *par = smc_sk(sk)->listen_smc; 939 940 spin_lock(&par->accept_q_lock); 941 list_del_init(&smc_sk(sk)->accept_q); 942 spin_unlock(&par->accept_q_lock); 943 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 944 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 945 } 946 947 /* remove a sock from the accept queue to bind it to a new socket created 948 * for a socket accept call from user space 949 */ 950 struct sock *smc_accept_dequeue(struct sock *parent, 951 struct socket *new_sock) 952 { 953 struct smc_sock *isk, *n; 954 struct sock *new_sk; 955 956 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 957 new_sk = (struct sock *)isk; 958 959 smc_accept_unlink(new_sk); 960 if (new_sk->sk_state == SMC_CLOSED) { 961 new_sk->sk_prot->unhash(new_sk); 962 if (isk->clcsock) { 963 sock_release(isk->clcsock); 964 isk->clcsock = NULL; 965 } 966 sock_put(new_sk); /* final */ 967 continue; 968 } 969 if (new_sock) { 970 sock_graft(new_sk, new_sock); 971 if (isk->use_fallback) { 972 smc_sk(new_sk)->clcsock->file = new_sock->file; 973 isk->clcsock->file->private_data = isk->clcsock; 974 } 975 } 976 return new_sk; 977 } 978 return NULL; 979 } 980 981 /* clean up for a created but never accepted sock */ 982 void smc_close_non_accepted(struct sock *sk) 983 { 984 struct smc_sock *smc = smc_sk(sk); 985 986 sock_hold(sk); /* sock_put below */ 987 lock_sock(sk); 988 if (!sk->sk_lingertime) 989 /* wait for peer closing */ 990 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 991 __smc_release(smc); 992 release_sock(sk); 993 sock_put(sk); /* sock_hold above */ 994 sock_put(sk); /* final sock_put */ 995 } 996 997 static int smc_serv_conf_first_link(struct smc_sock *smc) 998 { 999 struct net *net = sock_net(smc->clcsock->sk); 1000 struct smc_link_group *lgr = smc->conn.lgr; 1001 struct smc_link *link; 1002 int rest; 1003 int rc; 1004 1005 link = &lgr->lnk[SMC_SINGLE_LINK]; 1006 1007 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 1008 return SMC_CLC_DECL_ERR_REGRMB; 1009 1010 /* send CONFIRM LINK request to client over the RoCE fabric */ 1011 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1012 if (rc < 0) 1013 return SMC_CLC_DECL_TIMEOUT_CL; 1014 1015 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1016 rest = wait_for_completion_interruptible_timeout( 1017 &link->llc_confirm_resp, 1018 SMC_LLC_WAIT_FIRST_TIME); 1019 if (rest <= 0) { 1020 struct smc_clc_msg_decline dclc; 1021 1022 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1023 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1024 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1025 } 1026 1027 if (link->llc_confirm_resp_rc) 1028 return SMC_CLC_DECL_RMBE_EC; 1029 1030 /* send ADD LINK request to client over the RoCE fabric */ 1031 rc = smc_llc_send_add_link(link, 1032 link->smcibdev->mac[link->ibport - 1], 1033 link->gid, SMC_LLC_REQ); 1034 if (rc < 0) 1035 return SMC_CLC_DECL_TIMEOUT_AL; 1036 1037 /* receive ADD LINK response from client over the RoCE fabric */ 1038 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1039 SMC_LLC_WAIT_TIME); 1040 if (rest <= 0) { 1041 struct smc_clc_msg_decline dclc; 1042 1043 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1044 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1045 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1046 } 1047 1048 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1049 1050 return 0; 1051 } 1052 1053 /* listen worker: finish */ 1054 static void smc_listen_out(struct smc_sock *new_smc) 1055 { 1056 struct smc_sock *lsmc = new_smc->listen_smc; 1057 struct sock *newsmcsk = &new_smc->sk; 1058 1059 if (lsmc->sk.sk_state == SMC_LISTEN) { 1060 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1061 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1062 release_sock(&lsmc->sk); 1063 } else { /* no longer listening */ 1064 smc_close_non_accepted(newsmcsk); 1065 } 1066 1067 /* Wake up accept */ 1068 lsmc->sk.sk_data_ready(&lsmc->sk); 1069 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1070 } 1071 1072 /* listen worker: finish in state connected */ 1073 static void smc_listen_out_connected(struct smc_sock *new_smc) 1074 { 1075 struct sock *newsmcsk = &new_smc->sk; 1076 1077 sk_refcnt_debug_inc(newsmcsk); 1078 if (newsmcsk->sk_state == SMC_INIT) 1079 newsmcsk->sk_state = SMC_ACTIVE; 1080 1081 smc_listen_out(new_smc); 1082 } 1083 1084 /* listen worker: finish in error state */ 1085 static void smc_listen_out_err(struct smc_sock *new_smc) 1086 { 1087 struct sock *newsmcsk = &new_smc->sk; 1088 1089 if (newsmcsk->sk_state == SMC_INIT) 1090 sock_put(&new_smc->sk); /* passive closing */ 1091 newsmcsk->sk_state = SMC_CLOSED; 1092 smc_conn_free(&new_smc->conn); 1093 1094 smc_listen_out(new_smc); 1095 } 1096 1097 /* listen worker: decline and fall back if possible */ 1098 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1099 int local_contact) 1100 { 1101 /* RDMA setup failed, switch back to TCP */ 1102 if (local_contact == SMC_FIRST_CONTACT) 1103 smc_lgr_forget(new_smc->conn.lgr); 1104 if (reason_code < 0) { /* error, no fallback possible */ 1105 smc_listen_out_err(new_smc); 1106 return; 1107 } 1108 smc_conn_free(&new_smc->conn); 1109 smc_switch_to_fallback(new_smc); 1110 new_smc->fallback_rsn = reason_code; 1111 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1112 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1113 smc_listen_out_err(new_smc); 1114 return; 1115 } 1116 } 1117 smc_listen_out_connected(new_smc); 1118 } 1119 1120 /* listen worker: check prefixes */ 1121 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1122 struct smc_clc_msg_proposal *pclc) 1123 { 1124 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1125 struct socket *newclcsock = new_smc->clcsock; 1126 1127 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1128 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1129 return SMC_CLC_DECL_DIFFPREFIX; 1130 1131 return 0; 1132 } 1133 1134 /* listen worker: initialize connection and buffers */ 1135 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1136 struct smc_init_info *ini) 1137 { 1138 int rc; 1139 1140 /* allocate connection / link group */ 1141 rc = smc_conn_create(new_smc, ini); 1142 if (rc) 1143 return rc; 1144 1145 /* create send buffer and rmb */ 1146 if (smc_buf_create(new_smc, false)) 1147 return SMC_CLC_DECL_MEM; 1148 1149 return 0; 1150 } 1151 1152 /* listen worker: initialize connection and buffers for SMC-D */ 1153 static int smc_listen_ism_init(struct smc_sock *new_smc, 1154 struct smc_clc_msg_proposal *pclc, 1155 struct smc_init_info *ini) 1156 { 1157 struct smc_clc_msg_smcd *pclc_smcd; 1158 int rc; 1159 1160 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1161 ini->ism_gid = pclc_smcd->gid; 1162 rc = smc_conn_create(new_smc, ini); 1163 if (rc) 1164 return rc; 1165 1166 /* Check if peer can be reached via ISM device */ 1167 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1168 new_smc->conn.lgr->vlan_id, 1169 new_smc->conn.lgr->smcd)) { 1170 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1171 smc_lgr_forget(new_smc->conn.lgr); 1172 smc_conn_free(&new_smc->conn); 1173 return SMC_CLC_DECL_SMCDNOTALK; 1174 } 1175 1176 /* Create send and receive buffers */ 1177 if (smc_buf_create(new_smc, true)) { 1178 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1179 smc_lgr_forget(new_smc->conn.lgr); 1180 smc_conn_free(&new_smc->conn); 1181 return SMC_CLC_DECL_MEM; 1182 } 1183 1184 return 0; 1185 } 1186 1187 /* listen worker: register buffers */ 1188 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1189 { 1190 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1191 1192 if (local_contact != SMC_FIRST_CONTACT) { 1193 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1194 return SMC_CLC_DECL_ERR_REGRMB; 1195 } 1196 smc_rmb_sync_sg_for_device(&new_smc->conn); 1197 1198 return 0; 1199 } 1200 1201 /* listen worker: finish RDMA setup */ 1202 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1203 struct smc_clc_msg_accept_confirm *cclc, 1204 int local_contact) 1205 { 1206 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1207 int reason_code = 0; 1208 1209 if (local_contact == SMC_FIRST_CONTACT) 1210 smc_link_save_peer_info(link, cclc); 1211 1212 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1213 reason_code = SMC_CLC_DECL_ERR_RTOK; 1214 goto decline; 1215 } 1216 1217 if (local_contact == SMC_FIRST_CONTACT) { 1218 if (smc_ib_ready_link(link)) { 1219 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1220 goto decline; 1221 } 1222 /* QP confirmation over RoCE fabric */ 1223 reason_code = smc_serv_conf_first_link(new_smc); 1224 if (reason_code) 1225 goto decline; 1226 } 1227 return 0; 1228 1229 decline: 1230 smc_listen_decline(new_smc, reason_code, local_contact); 1231 return reason_code; 1232 } 1233 1234 /* setup for RDMA connection of server */ 1235 static void smc_listen_work(struct work_struct *work) 1236 { 1237 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1238 smc_listen_work); 1239 struct socket *newclcsock = new_smc->clcsock; 1240 struct smc_clc_msg_accept_confirm cclc; 1241 struct smc_clc_msg_proposal *pclc; 1242 struct smc_init_info ini = {0}; 1243 bool ism_supported = false; 1244 u8 buf[SMC_CLC_MAX_LEN]; 1245 int rc = 0; 1246 1247 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1248 return smc_listen_out_err(new_smc); 1249 1250 if (new_smc->use_fallback) { 1251 smc_listen_out_connected(new_smc); 1252 return; 1253 } 1254 1255 /* check if peer is smc capable */ 1256 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1257 smc_switch_to_fallback(new_smc); 1258 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1259 smc_listen_out_connected(new_smc); 1260 return; 1261 } 1262 1263 /* do inband token exchange - 1264 * wait for and receive SMC Proposal CLC message 1265 */ 1266 pclc = (struct smc_clc_msg_proposal *)&buf; 1267 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1268 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1269 if (rc) 1270 goto out_decl; 1271 1272 /* IPSec connections opt out of SMC-R optimizations */ 1273 if (using_ipsec(new_smc)) { 1274 rc = SMC_CLC_DECL_IPSEC; 1275 goto out_decl; 1276 } 1277 1278 /* check for matching IP prefix and subnet length */ 1279 rc = smc_listen_prfx_check(new_smc, pclc); 1280 if (rc) 1281 goto out_decl; 1282 1283 /* get vlan id from IP device */ 1284 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1285 rc = SMC_CLC_DECL_GETVLANERR; 1286 goto out_decl; 1287 } 1288 1289 mutex_lock(&smc_server_lgr_pending); 1290 smc_close_init(new_smc); 1291 smc_rx_init(new_smc); 1292 smc_tx_init(new_smc); 1293 1294 /* check if ISM is available */ 1295 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1296 ini.is_smcd = true; /* prepare ISM check */ 1297 rc = smc_find_ism_device(new_smc, &ini); 1298 if (!rc) 1299 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1300 if (!rc) 1301 ism_supported = true; 1302 else if (pclc->hdr.path == SMC_TYPE_D) 1303 goto out_unlock; /* skip RDMA and decline */ 1304 } 1305 1306 /* check if RDMA is available */ 1307 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1308 /* prepare RDMA check */ 1309 ini.is_smcd = false; 1310 ini.ism_dev = NULL; 1311 ini.ib_lcl = &pclc->lcl; 1312 rc = smc_find_rdma_device(new_smc, &ini); 1313 if (rc) { 1314 /* no RDMA device found */ 1315 if (pclc->hdr.path == SMC_TYPE_B) 1316 /* neither ISM nor RDMA device found */ 1317 rc = SMC_CLC_DECL_NOSMCDEV; 1318 goto out_unlock; 1319 } 1320 rc = smc_listen_rdma_init(new_smc, &ini); 1321 if (rc) 1322 goto out_unlock; 1323 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1324 if (rc) 1325 goto out_unlock; 1326 } 1327 1328 /* send SMC Accept CLC message */ 1329 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1330 if (rc) 1331 goto out_unlock; 1332 1333 /* SMC-D does not need this lock any more */ 1334 if (ism_supported) 1335 mutex_unlock(&smc_server_lgr_pending); 1336 1337 /* receive SMC Confirm CLC message */ 1338 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1339 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1340 if (rc) { 1341 if (!ism_supported) 1342 goto out_unlock; 1343 goto out_decl; 1344 } 1345 1346 /* finish worker */ 1347 if (!ism_supported) { 1348 rc = smc_listen_rdma_finish(new_smc, &cclc, 1349 ini.cln_first_contact); 1350 mutex_unlock(&smc_server_lgr_pending); 1351 if (rc) 1352 return; 1353 } 1354 smc_conn_save_peer_info(new_smc, &cclc); 1355 smc_listen_out_connected(new_smc); 1356 return; 1357 1358 out_unlock: 1359 mutex_unlock(&smc_server_lgr_pending); 1360 out_decl: 1361 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1362 } 1363 1364 static void smc_tcp_listen_work(struct work_struct *work) 1365 { 1366 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1367 tcp_listen_work); 1368 struct sock *lsk = &lsmc->sk; 1369 struct smc_sock *new_smc; 1370 int rc = 0; 1371 1372 lock_sock(lsk); 1373 while (lsk->sk_state == SMC_LISTEN) { 1374 rc = smc_clcsock_accept(lsmc, &new_smc); 1375 if (rc) 1376 goto out; 1377 if (!new_smc) 1378 continue; 1379 1380 new_smc->listen_smc = lsmc; 1381 new_smc->use_fallback = lsmc->use_fallback; 1382 new_smc->fallback_rsn = lsmc->fallback_rsn; 1383 sock_hold(lsk); /* sock_put in smc_listen_work */ 1384 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1385 smc_copy_sock_settings_to_smc(new_smc); 1386 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1387 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1388 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1389 if (!schedule_work(&new_smc->smc_listen_work)) 1390 sock_put(&new_smc->sk); 1391 } 1392 1393 out: 1394 release_sock(lsk); 1395 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1396 } 1397 1398 static int smc_listen(struct socket *sock, int backlog) 1399 { 1400 struct sock *sk = sock->sk; 1401 struct smc_sock *smc; 1402 int rc; 1403 1404 smc = smc_sk(sk); 1405 lock_sock(sk); 1406 1407 rc = -EINVAL; 1408 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1409 smc->connect_nonblock) 1410 goto out; 1411 1412 rc = 0; 1413 if (sk->sk_state == SMC_LISTEN) { 1414 sk->sk_max_ack_backlog = backlog; 1415 goto out; 1416 } 1417 /* some socket options are handled in core, so we could not apply 1418 * them to the clc socket -- copy smc socket options to clc socket 1419 */ 1420 smc_copy_sock_settings_to_clc(smc); 1421 if (!smc->use_fallback) 1422 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1423 1424 rc = kernel_listen(smc->clcsock, backlog); 1425 if (rc) 1426 goto out; 1427 sk->sk_max_ack_backlog = backlog; 1428 sk->sk_ack_backlog = 0; 1429 sk->sk_state = SMC_LISTEN; 1430 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1431 if (!schedule_work(&smc->tcp_listen_work)) 1432 sock_put(sk); 1433 1434 out: 1435 release_sock(sk); 1436 return rc; 1437 } 1438 1439 static int smc_accept(struct socket *sock, struct socket *new_sock, 1440 int flags, bool kern) 1441 { 1442 struct sock *sk = sock->sk, *nsk; 1443 DECLARE_WAITQUEUE(wait, current); 1444 struct smc_sock *lsmc; 1445 long timeo; 1446 int rc = 0; 1447 1448 lsmc = smc_sk(sk); 1449 sock_hold(sk); /* sock_put below */ 1450 lock_sock(sk); 1451 1452 if (lsmc->sk.sk_state != SMC_LISTEN) { 1453 rc = -EINVAL; 1454 release_sock(sk); 1455 goto out; 1456 } 1457 1458 /* Wait for an incoming connection */ 1459 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1460 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1461 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1462 set_current_state(TASK_INTERRUPTIBLE); 1463 if (!timeo) { 1464 rc = -EAGAIN; 1465 break; 1466 } 1467 release_sock(sk); 1468 timeo = schedule_timeout(timeo); 1469 /* wakeup by sk_data_ready in smc_listen_work() */ 1470 sched_annotate_sleep(); 1471 lock_sock(sk); 1472 if (signal_pending(current)) { 1473 rc = sock_intr_errno(timeo); 1474 break; 1475 } 1476 } 1477 set_current_state(TASK_RUNNING); 1478 remove_wait_queue(sk_sleep(sk), &wait); 1479 1480 if (!rc) 1481 rc = sock_error(nsk); 1482 release_sock(sk); 1483 if (rc) 1484 goto out; 1485 1486 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1487 /* wait till data arrives on the socket */ 1488 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1489 MSEC_PER_SEC); 1490 if (smc_sk(nsk)->use_fallback) { 1491 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1492 1493 lock_sock(clcsk); 1494 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1495 sk_wait_data(clcsk, &timeo, NULL); 1496 release_sock(clcsk); 1497 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1498 lock_sock(nsk); 1499 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1500 release_sock(nsk); 1501 } 1502 } 1503 1504 out: 1505 sock_put(sk); /* sock_hold above */ 1506 return rc; 1507 } 1508 1509 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1510 int peer) 1511 { 1512 struct smc_sock *smc; 1513 1514 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1515 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1516 return -ENOTCONN; 1517 1518 smc = smc_sk(sock->sk); 1519 1520 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1521 } 1522 1523 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1524 { 1525 struct sock *sk = sock->sk; 1526 struct smc_sock *smc; 1527 int rc = -EPIPE; 1528 1529 smc = smc_sk(sk); 1530 lock_sock(sk); 1531 if ((sk->sk_state != SMC_ACTIVE) && 1532 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1533 (sk->sk_state != SMC_INIT)) 1534 goto out; 1535 1536 if (msg->msg_flags & MSG_FASTOPEN) { 1537 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1538 smc_switch_to_fallback(smc); 1539 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1540 } else { 1541 rc = -EINVAL; 1542 goto out; 1543 } 1544 } 1545 1546 if (smc->use_fallback) 1547 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1548 else 1549 rc = smc_tx_sendmsg(smc, msg, len); 1550 out: 1551 release_sock(sk); 1552 return rc; 1553 } 1554 1555 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1556 int flags) 1557 { 1558 struct sock *sk = sock->sk; 1559 struct smc_sock *smc; 1560 int rc = -ENOTCONN; 1561 1562 smc = smc_sk(sk); 1563 lock_sock(sk); 1564 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1565 /* socket was connected before, no more data to read */ 1566 rc = 0; 1567 goto out; 1568 } 1569 if ((sk->sk_state == SMC_INIT) || 1570 (sk->sk_state == SMC_LISTEN) || 1571 (sk->sk_state == SMC_CLOSED)) 1572 goto out; 1573 1574 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1575 rc = 0; 1576 goto out; 1577 } 1578 1579 if (smc->use_fallback) { 1580 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1581 } else { 1582 msg->msg_namelen = 0; 1583 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1584 } 1585 1586 out: 1587 release_sock(sk); 1588 return rc; 1589 } 1590 1591 static __poll_t smc_accept_poll(struct sock *parent) 1592 { 1593 struct smc_sock *isk = smc_sk(parent); 1594 __poll_t mask = 0; 1595 1596 spin_lock(&isk->accept_q_lock); 1597 if (!list_empty(&isk->accept_q)) 1598 mask = EPOLLIN | EPOLLRDNORM; 1599 spin_unlock(&isk->accept_q_lock); 1600 1601 return mask; 1602 } 1603 1604 static __poll_t smc_poll(struct file *file, struct socket *sock, 1605 poll_table *wait) 1606 { 1607 struct sock *sk = sock->sk; 1608 struct smc_sock *smc; 1609 __poll_t mask = 0; 1610 1611 if (!sk) 1612 return EPOLLNVAL; 1613 1614 smc = smc_sk(sock->sk); 1615 if (smc->use_fallback) { 1616 /* delegate to CLC child sock */ 1617 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1618 sk->sk_err = smc->clcsock->sk->sk_err; 1619 } else { 1620 if (sk->sk_state != SMC_CLOSED) 1621 sock_poll_wait(file, sock, wait); 1622 if (sk->sk_err) 1623 mask |= EPOLLERR; 1624 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1625 (sk->sk_state == SMC_CLOSED)) 1626 mask |= EPOLLHUP; 1627 if (sk->sk_state == SMC_LISTEN) { 1628 /* woken up by sk_data_ready in smc_listen_work() */ 1629 mask |= smc_accept_poll(sk); 1630 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1631 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1632 wait); 1633 sk->sk_err = smc->clcsock->sk->sk_err; 1634 } else { 1635 if ((sk->sk_state != SMC_INIT && 1636 atomic_read(&smc->conn.sndbuf_space)) || 1637 sk->sk_shutdown & SEND_SHUTDOWN) { 1638 mask |= EPOLLOUT | EPOLLWRNORM; 1639 } else { 1640 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1641 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1642 } 1643 if (atomic_read(&smc->conn.bytes_to_rcv)) 1644 mask |= EPOLLIN | EPOLLRDNORM; 1645 if (sk->sk_shutdown & RCV_SHUTDOWN) 1646 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1647 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1648 mask |= EPOLLIN; 1649 if (smc->conn.urg_state == SMC_URG_VALID) 1650 mask |= EPOLLPRI; 1651 } 1652 } 1653 1654 return mask; 1655 } 1656 1657 static int smc_shutdown(struct socket *sock, int how) 1658 { 1659 struct sock *sk = sock->sk; 1660 struct smc_sock *smc; 1661 int rc = -EINVAL; 1662 int rc1 = 0; 1663 1664 smc = smc_sk(sk); 1665 1666 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1667 return rc; 1668 1669 lock_sock(sk); 1670 1671 rc = -ENOTCONN; 1672 if ((sk->sk_state != SMC_ACTIVE) && 1673 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1674 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1675 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1676 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1677 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1678 goto out; 1679 if (smc->use_fallback) { 1680 rc = kernel_sock_shutdown(smc->clcsock, how); 1681 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1682 if (sk->sk_shutdown == SHUTDOWN_MASK) 1683 sk->sk_state = SMC_CLOSED; 1684 goto out; 1685 } 1686 switch (how) { 1687 case SHUT_RDWR: /* shutdown in both directions */ 1688 rc = smc_close_active(smc); 1689 break; 1690 case SHUT_WR: 1691 rc = smc_close_shutdown_write(smc); 1692 break; 1693 case SHUT_RD: 1694 rc = 0; 1695 /* nothing more to do because peer is not involved */ 1696 break; 1697 } 1698 if (smc->clcsock) 1699 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1700 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1701 sk->sk_shutdown |= how + 1; 1702 1703 out: 1704 release_sock(sk); 1705 return rc ? rc : rc1; 1706 } 1707 1708 static int smc_setsockopt(struct socket *sock, int level, int optname, 1709 char __user *optval, unsigned int optlen) 1710 { 1711 struct sock *sk = sock->sk; 1712 struct smc_sock *smc; 1713 int val, rc; 1714 1715 smc = smc_sk(sk); 1716 1717 /* generic setsockopts reaching us here always apply to the 1718 * CLC socket 1719 */ 1720 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1721 optval, optlen); 1722 if (smc->clcsock->sk->sk_err) { 1723 sk->sk_err = smc->clcsock->sk->sk_err; 1724 sk->sk_error_report(sk); 1725 } 1726 1727 if (optlen < sizeof(int)) 1728 return -EINVAL; 1729 if (get_user(val, (int __user *)optval)) 1730 return -EFAULT; 1731 1732 lock_sock(sk); 1733 if (rc || smc->use_fallback) 1734 goto out; 1735 switch (optname) { 1736 case TCP_ULP: 1737 case TCP_FASTOPEN: 1738 case TCP_FASTOPEN_CONNECT: 1739 case TCP_FASTOPEN_KEY: 1740 case TCP_FASTOPEN_NO_COOKIE: 1741 /* option not supported by SMC */ 1742 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1743 smc_switch_to_fallback(smc); 1744 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1745 } else { 1746 rc = -EINVAL; 1747 } 1748 break; 1749 case TCP_NODELAY: 1750 if (sk->sk_state != SMC_INIT && 1751 sk->sk_state != SMC_LISTEN && 1752 sk->sk_state != SMC_CLOSED) { 1753 if (val) 1754 mod_delayed_work(system_wq, &smc->conn.tx_work, 1755 0); 1756 } 1757 break; 1758 case TCP_CORK: 1759 if (sk->sk_state != SMC_INIT && 1760 sk->sk_state != SMC_LISTEN && 1761 sk->sk_state != SMC_CLOSED) { 1762 if (!val) 1763 mod_delayed_work(system_wq, &smc->conn.tx_work, 1764 0); 1765 } 1766 break; 1767 case TCP_DEFER_ACCEPT: 1768 smc->sockopt_defer_accept = val; 1769 break; 1770 default: 1771 break; 1772 } 1773 out: 1774 release_sock(sk); 1775 1776 return rc; 1777 } 1778 1779 static int smc_getsockopt(struct socket *sock, int level, int optname, 1780 char __user *optval, int __user *optlen) 1781 { 1782 struct smc_sock *smc; 1783 1784 smc = smc_sk(sock->sk); 1785 /* socket options apply to the CLC socket */ 1786 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1787 optval, optlen); 1788 } 1789 1790 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1791 unsigned long arg) 1792 { 1793 union smc_host_cursor cons, urg; 1794 struct smc_connection *conn; 1795 struct smc_sock *smc; 1796 int answ; 1797 1798 smc = smc_sk(sock->sk); 1799 conn = &smc->conn; 1800 lock_sock(&smc->sk); 1801 if (smc->use_fallback) { 1802 if (!smc->clcsock) { 1803 release_sock(&smc->sk); 1804 return -EBADF; 1805 } 1806 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1807 release_sock(&smc->sk); 1808 return answ; 1809 } 1810 switch (cmd) { 1811 case SIOCINQ: /* same as FIONREAD */ 1812 if (smc->sk.sk_state == SMC_LISTEN) { 1813 release_sock(&smc->sk); 1814 return -EINVAL; 1815 } 1816 if (smc->sk.sk_state == SMC_INIT || 1817 smc->sk.sk_state == SMC_CLOSED) 1818 answ = 0; 1819 else 1820 answ = atomic_read(&smc->conn.bytes_to_rcv); 1821 break; 1822 case SIOCOUTQ: 1823 /* output queue size (not send + not acked) */ 1824 if (smc->sk.sk_state == SMC_LISTEN) { 1825 release_sock(&smc->sk); 1826 return -EINVAL; 1827 } 1828 if (smc->sk.sk_state == SMC_INIT || 1829 smc->sk.sk_state == SMC_CLOSED) 1830 answ = 0; 1831 else 1832 answ = smc->conn.sndbuf_desc->len - 1833 atomic_read(&smc->conn.sndbuf_space); 1834 break; 1835 case SIOCOUTQNSD: 1836 /* output queue size (not send only) */ 1837 if (smc->sk.sk_state == SMC_LISTEN) { 1838 release_sock(&smc->sk); 1839 return -EINVAL; 1840 } 1841 if (smc->sk.sk_state == SMC_INIT || 1842 smc->sk.sk_state == SMC_CLOSED) 1843 answ = 0; 1844 else 1845 answ = smc_tx_prepared_sends(&smc->conn); 1846 break; 1847 case SIOCATMARK: 1848 if (smc->sk.sk_state == SMC_LISTEN) { 1849 release_sock(&smc->sk); 1850 return -EINVAL; 1851 } 1852 if (smc->sk.sk_state == SMC_INIT || 1853 smc->sk.sk_state == SMC_CLOSED) { 1854 answ = 0; 1855 } else { 1856 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1857 smc_curs_copy(&urg, &conn->urg_curs, conn); 1858 answ = smc_curs_diff(conn->rmb_desc->len, 1859 &cons, &urg) == 1; 1860 } 1861 break; 1862 default: 1863 release_sock(&smc->sk); 1864 return -ENOIOCTLCMD; 1865 } 1866 release_sock(&smc->sk); 1867 1868 return put_user(answ, (int __user *)arg); 1869 } 1870 1871 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1872 int offset, size_t size, int flags) 1873 { 1874 struct sock *sk = sock->sk; 1875 struct smc_sock *smc; 1876 int rc = -EPIPE; 1877 1878 smc = smc_sk(sk); 1879 lock_sock(sk); 1880 if (sk->sk_state != SMC_ACTIVE) { 1881 release_sock(sk); 1882 goto out; 1883 } 1884 release_sock(sk); 1885 if (smc->use_fallback) 1886 rc = kernel_sendpage(smc->clcsock, page, offset, 1887 size, flags); 1888 else 1889 rc = sock_no_sendpage(sock, page, offset, size, flags); 1890 1891 out: 1892 return rc; 1893 } 1894 1895 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1896 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1897 * updates till whenever a respective page has been fully processed. 1898 * Note that subsequent recv() calls have to wait till all splice() processing 1899 * completed. 1900 */ 1901 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1902 struct pipe_inode_info *pipe, size_t len, 1903 unsigned int flags) 1904 { 1905 struct sock *sk = sock->sk; 1906 struct smc_sock *smc; 1907 int rc = -ENOTCONN; 1908 1909 smc = smc_sk(sk); 1910 lock_sock(sk); 1911 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1912 /* socket was connected before, no more data to read */ 1913 rc = 0; 1914 goto out; 1915 } 1916 if (sk->sk_state == SMC_INIT || 1917 sk->sk_state == SMC_LISTEN || 1918 sk->sk_state == SMC_CLOSED) 1919 goto out; 1920 1921 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1922 rc = 0; 1923 goto out; 1924 } 1925 1926 if (smc->use_fallback) { 1927 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1928 pipe, len, flags); 1929 } else { 1930 if (*ppos) { 1931 rc = -ESPIPE; 1932 goto out; 1933 } 1934 if (flags & SPLICE_F_NONBLOCK) 1935 flags = MSG_DONTWAIT; 1936 else 1937 flags = 0; 1938 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1939 } 1940 out: 1941 release_sock(sk); 1942 1943 return rc; 1944 } 1945 1946 /* must look like tcp */ 1947 static const struct proto_ops smc_sock_ops = { 1948 .family = PF_SMC, 1949 .owner = THIS_MODULE, 1950 .release = smc_release, 1951 .bind = smc_bind, 1952 .connect = smc_connect, 1953 .socketpair = sock_no_socketpair, 1954 .accept = smc_accept, 1955 .getname = smc_getname, 1956 .poll = smc_poll, 1957 .ioctl = smc_ioctl, 1958 .listen = smc_listen, 1959 .shutdown = smc_shutdown, 1960 .setsockopt = smc_setsockopt, 1961 .getsockopt = smc_getsockopt, 1962 .sendmsg = smc_sendmsg, 1963 .recvmsg = smc_recvmsg, 1964 .mmap = sock_no_mmap, 1965 .sendpage = smc_sendpage, 1966 .splice_read = smc_splice_read, 1967 }; 1968 1969 static int smc_create(struct net *net, struct socket *sock, int protocol, 1970 int kern) 1971 { 1972 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1973 struct smc_sock *smc; 1974 struct sock *sk; 1975 int rc; 1976 1977 rc = -ESOCKTNOSUPPORT; 1978 if (sock->type != SOCK_STREAM) 1979 goto out; 1980 1981 rc = -EPROTONOSUPPORT; 1982 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1983 goto out; 1984 1985 rc = -ENOBUFS; 1986 sock->ops = &smc_sock_ops; 1987 sk = smc_sock_alloc(net, sock, protocol); 1988 if (!sk) 1989 goto out; 1990 1991 /* create internal TCP socket for CLC handshake and fallback */ 1992 smc = smc_sk(sk); 1993 smc->use_fallback = false; /* assume rdma capability first */ 1994 smc->fallback_rsn = 0; 1995 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 1996 &smc->clcsock); 1997 if (rc) { 1998 sk_common_release(sk); 1999 goto out; 2000 } 2001 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2002 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2003 2004 out: 2005 return rc; 2006 } 2007 2008 static const struct net_proto_family smc_sock_family_ops = { 2009 .family = PF_SMC, 2010 .owner = THIS_MODULE, 2011 .create = smc_create, 2012 }; 2013 2014 unsigned int smc_net_id; 2015 2016 static __net_init int smc_net_init(struct net *net) 2017 { 2018 return smc_pnet_net_init(net); 2019 } 2020 2021 static void __net_exit smc_net_exit(struct net *net) 2022 { 2023 smc_pnet_net_exit(net); 2024 } 2025 2026 static struct pernet_operations smc_net_ops = { 2027 .init = smc_net_init, 2028 .exit = smc_net_exit, 2029 .id = &smc_net_id, 2030 .size = sizeof(struct smc_net), 2031 }; 2032 2033 static int __init smc_init(void) 2034 { 2035 int rc; 2036 2037 rc = register_pernet_subsys(&smc_net_ops); 2038 if (rc) 2039 return rc; 2040 2041 rc = smc_pnet_init(); 2042 if (rc) 2043 goto out_pernet_subsys; 2044 2045 rc = smc_core_init(); 2046 if (rc) { 2047 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2048 goto out_pnet; 2049 } 2050 2051 rc = smc_llc_init(); 2052 if (rc) { 2053 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2054 goto out_core; 2055 } 2056 2057 rc = smc_cdc_init(); 2058 if (rc) { 2059 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2060 goto out_core; 2061 } 2062 2063 rc = proto_register(&smc_proto, 1); 2064 if (rc) { 2065 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2066 goto out_core; 2067 } 2068 2069 rc = proto_register(&smc_proto6, 1); 2070 if (rc) { 2071 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2072 goto out_proto; 2073 } 2074 2075 rc = sock_register(&smc_sock_family_ops); 2076 if (rc) { 2077 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2078 goto out_proto6; 2079 } 2080 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2081 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2082 2083 rc = smc_ib_register_client(); 2084 if (rc) { 2085 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2086 goto out_sock; 2087 } 2088 2089 static_branch_enable(&tcp_have_smc); 2090 return 0; 2091 2092 out_sock: 2093 sock_unregister(PF_SMC); 2094 out_proto6: 2095 proto_unregister(&smc_proto6); 2096 out_proto: 2097 proto_unregister(&smc_proto); 2098 out_core: 2099 smc_core_exit(); 2100 out_pnet: 2101 smc_pnet_exit(); 2102 out_pernet_subsys: 2103 unregister_pernet_subsys(&smc_net_ops); 2104 2105 return rc; 2106 } 2107 2108 static void __exit smc_exit(void) 2109 { 2110 static_branch_disable(&tcp_have_smc); 2111 sock_unregister(PF_SMC); 2112 smc_core_exit(); 2113 smc_ib_unregister_client(); 2114 proto_unregister(&smc_proto6); 2115 proto_unregister(&smc_proto); 2116 smc_pnet_exit(); 2117 unregister_pernet_subsys(&smc_net_ops); 2118 rcu_barrier(); 2119 } 2120 2121 module_init(smc_init); 2122 module_exit(smc_exit); 2123 2124 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2125 MODULE_DESCRIPTION("smc socket address family"); 2126 MODULE_LICENSE("GPL"); 2127 MODULE_ALIAS_NETPROTO(PF_SMC); 2128