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 (flags & O_NONBLOCK) { 861 if (schedule_work(&smc->connect_work)) 862 smc->connect_nonblock = 1; 863 rc = -EINPROGRESS; 864 } else { 865 rc = __smc_connect(smc); 866 if (rc < 0) 867 goto out; 868 else 869 rc = 0; /* success cases including fallback */ 870 } 871 872 out: 873 release_sock(sk); 874 out_err: 875 return rc; 876 } 877 878 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 879 { 880 struct socket *new_clcsock = NULL; 881 struct sock *lsk = &lsmc->sk; 882 struct sock *new_sk; 883 int rc = -EINVAL; 884 885 release_sock(lsk); 886 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 887 if (!new_sk) { 888 rc = -ENOMEM; 889 lsk->sk_err = ENOMEM; 890 *new_smc = NULL; 891 lock_sock(lsk); 892 goto out; 893 } 894 *new_smc = smc_sk(new_sk); 895 896 mutex_lock(&lsmc->clcsock_release_lock); 897 if (lsmc->clcsock) 898 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 899 mutex_unlock(&lsmc->clcsock_release_lock); 900 lock_sock(lsk); 901 if (rc < 0) 902 lsk->sk_err = -rc; 903 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 904 new_sk->sk_prot->unhash(new_sk); 905 if (new_clcsock) 906 sock_release(new_clcsock); 907 new_sk->sk_state = SMC_CLOSED; 908 sock_set_flag(new_sk, SOCK_DEAD); 909 sock_put(new_sk); /* final */ 910 *new_smc = NULL; 911 goto out; 912 } 913 914 (*new_smc)->clcsock = new_clcsock; 915 out: 916 return rc; 917 } 918 919 /* add a just created sock to the accept queue of the listen sock as 920 * candidate for a following socket accept call from user space 921 */ 922 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 923 { 924 struct smc_sock *par = smc_sk(parent); 925 926 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 927 spin_lock(&par->accept_q_lock); 928 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 929 spin_unlock(&par->accept_q_lock); 930 sk_acceptq_added(parent); 931 } 932 933 /* remove a socket from the accept queue of its parental listening socket */ 934 static void smc_accept_unlink(struct sock *sk) 935 { 936 struct smc_sock *par = smc_sk(sk)->listen_smc; 937 938 spin_lock(&par->accept_q_lock); 939 list_del_init(&smc_sk(sk)->accept_q); 940 spin_unlock(&par->accept_q_lock); 941 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 942 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 943 } 944 945 /* remove a sock from the accept queue to bind it to a new socket created 946 * for a socket accept call from user space 947 */ 948 struct sock *smc_accept_dequeue(struct sock *parent, 949 struct socket *new_sock) 950 { 951 struct smc_sock *isk, *n; 952 struct sock *new_sk; 953 954 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 955 new_sk = (struct sock *)isk; 956 957 smc_accept_unlink(new_sk); 958 if (new_sk->sk_state == SMC_CLOSED) { 959 new_sk->sk_prot->unhash(new_sk); 960 if (isk->clcsock) { 961 sock_release(isk->clcsock); 962 isk->clcsock = NULL; 963 } 964 sock_put(new_sk); /* final */ 965 continue; 966 } 967 if (new_sock) { 968 sock_graft(new_sk, new_sock); 969 if (isk->use_fallback) { 970 smc_sk(new_sk)->clcsock->file = new_sock->file; 971 isk->clcsock->file->private_data = isk->clcsock; 972 } 973 } 974 return new_sk; 975 } 976 return NULL; 977 } 978 979 /* clean up for a created but never accepted sock */ 980 void smc_close_non_accepted(struct sock *sk) 981 { 982 struct smc_sock *smc = smc_sk(sk); 983 984 sock_hold(sk); /* sock_put below */ 985 lock_sock(sk); 986 if (!sk->sk_lingertime) 987 /* wait for peer closing */ 988 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 989 __smc_release(smc); 990 release_sock(sk); 991 sock_put(sk); /* sock_hold above */ 992 sock_put(sk); /* final sock_put */ 993 } 994 995 static int smc_serv_conf_first_link(struct smc_sock *smc) 996 { 997 struct net *net = sock_net(smc->clcsock->sk); 998 struct smc_link_group *lgr = smc->conn.lgr; 999 struct smc_link *link; 1000 int rest; 1001 int rc; 1002 1003 link = &lgr->lnk[SMC_SINGLE_LINK]; 1004 1005 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 1006 return SMC_CLC_DECL_ERR_REGRMB; 1007 1008 /* send CONFIRM LINK request to client over the RoCE fabric */ 1009 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1010 if (rc < 0) 1011 return SMC_CLC_DECL_TIMEOUT_CL; 1012 1013 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1014 rest = wait_for_completion_interruptible_timeout( 1015 &link->llc_confirm_resp, 1016 SMC_LLC_WAIT_FIRST_TIME); 1017 if (rest <= 0) { 1018 struct smc_clc_msg_decline dclc; 1019 1020 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1021 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1022 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1023 } 1024 1025 if (link->llc_confirm_resp_rc) 1026 return SMC_CLC_DECL_RMBE_EC; 1027 1028 /* send ADD LINK request to client over the RoCE fabric */ 1029 rc = smc_llc_send_add_link(link, 1030 link->smcibdev->mac[link->ibport - 1], 1031 link->gid, SMC_LLC_REQ); 1032 if (rc < 0) 1033 return SMC_CLC_DECL_TIMEOUT_AL; 1034 1035 /* receive ADD LINK response from client over the RoCE fabric */ 1036 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1037 SMC_LLC_WAIT_TIME); 1038 if (rest <= 0) { 1039 struct smc_clc_msg_decline dclc; 1040 1041 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1042 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1043 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1044 } 1045 1046 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1047 1048 return 0; 1049 } 1050 1051 /* listen worker: finish */ 1052 static void smc_listen_out(struct smc_sock *new_smc) 1053 { 1054 struct smc_sock *lsmc = new_smc->listen_smc; 1055 struct sock *newsmcsk = &new_smc->sk; 1056 1057 if (lsmc->sk.sk_state == SMC_LISTEN) { 1058 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1059 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1060 release_sock(&lsmc->sk); 1061 } else { /* no longer listening */ 1062 smc_close_non_accepted(newsmcsk); 1063 } 1064 1065 /* Wake up accept */ 1066 lsmc->sk.sk_data_ready(&lsmc->sk); 1067 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1068 } 1069 1070 /* listen worker: finish in state connected */ 1071 static void smc_listen_out_connected(struct smc_sock *new_smc) 1072 { 1073 struct sock *newsmcsk = &new_smc->sk; 1074 1075 sk_refcnt_debug_inc(newsmcsk); 1076 if (newsmcsk->sk_state == SMC_INIT) 1077 newsmcsk->sk_state = SMC_ACTIVE; 1078 1079 smc_listen_out(new_smc); 1080 } 1081 1082 /* listen worker: finish in error state */ 1083 static void smc_listen_out_err(struct smc_sock *new_smc) 1084 { 1085 struct sock *newsmcsk = &new_smc->sk; 1086 1087 if (newsmcsk->sk_state == SMC_INIT) 1088 sock_put(&new_smc->sk); /* passive closing */ 1089 newsmcsk->sk_state = SMC_CLOSED; 1090 smc_conn_free(&new_smc->conn); 1091 1092 smc_listen_out(new_smc); 1093 } 1094 1095 /* listen worker: decline and fall back if possible */ 1096 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1097 int local_contact) 1098 { 1099 /* RDMA setup failed, switch back to TCP */ 1100 if (local_contact == SMC_FIRST_CONTACT) 1101 smc_lgr_forget(new_smc->conn.lgr); 1102 if (reason_code < 0) { /* error, no fallback possible */ 1103 smc_listen_out_err(new_smc); 1104 return; 1105 } 1106 smc_conn_free(&new_smc->conn); 1107 smc_switch_to_fallback(new_smc); 1108 new_smc->fallback_rsn = reason_code; 1109 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1110 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1111 smc_listen_out_err(new_smc); 1112 return; 1113 } 1114 } 1115 smc_listen_out_connected(new_smc); 1116 } 1117 1118 /* listen worker: check prefixes */ 1119 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1120 struct smc_clc_msg_proposal *pclc) 1121 { 1122 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1123 struct socket *newclcsock = new_smc->clcsock; 1124 1125 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1126 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1127 return SMC_CLC_DECL_DIFFPREFIX; 1128 1129 return 0; 1130 } 1131 1132 /* listen worker: initialize connection and buffers */ 1133 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1134 struct smc_init_info *ini) 1135 { 1136 int rc; 1137 1138 /* allocate connection / link group */ 1139 rc = smc_conn_create(new_smc, ini); 1140 if (rc) 1141 return rc; 1142 1143 /* create send buffer and rmb */ 1144 if (smc_buf_create(new_smc, false)) 1145 return SMC_CLC_DECL_MEM; 1146 1147 return 0; 1148 } 1149 1150 /* listen worker: initialize connection and buffers for SMC-D */ 1151 static int smc_listen_ism_init(struct smc_sock *new_smc, 1152 struct smc_clc_msg_proposal *pclc, 1153 struct smc_init_info *ini) 1154 { 1155 struct smc_clc_msg_smcd *pclc_smcd; 1156 int rc; 1157 1158 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1159 ini->ism_gid = pclc_smcd->gid; 1160 rc = smc_conn_create(new_smc, ini); 1161 if (rc) 1162 return rc; 1163 1164 /* Check if peer can be reached via ISM device */ 1165 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1166 new_smc->conn.lgr->vlan_id, 1167 new_smc->conn.lgr->smcd)) { 1168 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1169 smc_lgr_forget(new_smc->conn.lgr); 1170 smc_conn_free(&new_smc->conn); 1171 return SMC_CLC_DECL_SMCDNOTALK; 1172 } 1173 1174 /* Create send and receive buffers */ 1175 if (smc_buf_create(new_smc, true)) { 1176 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1177 smc_lgr_forget(new_smc->conn.lgr); 1178 smc_conn_free(&new_smc->conn); 1179 return SMC_CLC_DECL_MEM; 1180 } 1181 1182 return 0; 1183 } 1184 1185 /* listen worker: register buffers */ 1186 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1187 { 1188 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1189 1190 if (local_contact != SMC_FIRST_CONTACT) { 1191 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1192 return SMC_CLC_DECL_ERR_REGRMB; 1193 } 1194 smc_rmb_sync_sg_for_device(&new_smc->conn); 1195 1196 return 0; 1197 } 1198 1199 /* listen worker: finish RDMA setup */ 1200 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1201 struct smc_clc_msg_accept_confirm *cclc, 1202 int local_contact) 1203 { 1204 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1205 int reason_code = 0; 1206 1207 if (local_contact == SMC_FIRST_CONTACT) 1208 smc_link_save_peer_info(link, cclc); 1209 1210 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1211 reason_code = SMC_CLC_DECL_ERR_RTOK; 1212 goto decline; 1213 } 1214 1215 if (local_contact == SMC_FIRST_CONTACT) { 1216 if (smc_ib_ready_link(link)) { 1217 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1218 goto decline; 1219 } 1220 /* QP confirmation over RoCE fabric */ 1221 reason_code = smc_serv_conf_first_link(new_smc); 1222 if (reason_code) 1223 goto decline; 1224 } 1225 return 0; 1226 1227 decline: 1228 smc_listen_decline(new_smc, reason_code, local_contact); 1229 return reason_code; 1230 } 1231 1232 /* setup for RDMA connection of server */ 1233 static void smc_listen_work(struct work_struct *work) 1234 { 1235 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1236 smc_listen_work); 1237 struct socket *newclcsock = new_smc->clcsock; 1238 struct smc_clc_msg_accept_confirm cclc; 1239 struct smc_clc_msg_proposal *pclc; 1240 struct smc_init_info ini = {0}; 1241 bool ism_supported = false; 1242 u8 buf[SMC_CLC_MAX_LEN]; 1243 int rc = 0; 1244 1245 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1246 return smc_listen_out_err(new_smc); 1247 1248 if (new_smc->use_fallback) { 1249 smc_listen_out_connected(new_smc); 1250 return; 1251 } 1252 1253 /* check if peer is smc capable */ 1254 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1255 smc_switch_to_fallback(new_smc); 1256 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1257 smc_listen_out_connected(new_smc); 1258 return; 1259 } 1260 1261 /* do inband token exchange - 1262 * wait for and receive SMC Proposal CLC message 1263 */ 1264 pclc = (struct smc_clc_msg_proposal *)&buf; 1265 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1266 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1267 if (rc) 1268 goto out_decl; 1269 1270 /* IPSec connections opt out of SMC-R optimizations */ 1271 if (using_ipsec(new_smc)) { 1272 rc = SMC_CLC_DECL_IPSEC; 1273 goto out_decl; 1274 } 1275 1276 /* check for matching IP prefix and subnet length */ 1277 rc = smc_listen_prfx_check(new_smc, pclc); 1278 if (rc) 1279 goto out_decl; 1280 1281 /* get vlan id from IP device */ 1282 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1283 rc = SMC_CLC_DECL_GETVLANERR; 1284 goto out_decl; 1285 } 1286 1287 mutex_lock(&smc_server_lgr_pending); 1288 smc_close_init(new_smc); 1289 smc_rx_init(new_smc); 1290 smc_tx_init(new_smc); 1291 1292 /* check if ISM is available */ 1293 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1294 ini.is_smcd = true; /* prepare ISM check */ 1295 rc = smc_find_ism_device(new_smc, &ini); 1296 if (!rc) 1297 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1298 if (!rc) 1299 ism_supported = true; 1300 else if (pclc->hdr.path == SMC_TYPE_D) 1301 goto out_unlock; /* skip RDMA and decline */ 1302 } 1303 1304 /* check if RDMA is available */ 1305 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1306 /* prepare RDMA check */ 1307 ini.is_smcd = false; 1308 ini.ism_dev = NULL; 1309 ini.ib_lcl = &pclc->lcl; 1310 rc = smc_find_rdma_device(new_smc, &ini); 1311 if (rc) { 1312 /* no RDMA device found */ 1313 if (pclc->hdr.path == SMC_TYPE_B) 1314 /* neither ISM nor RDMA device found */ 1315 rc = SMC_CLC_DECL_NOSMCDEV; 1316 goto out_unlock; 1317 } 1318 rc = smc_listen_rdma_init(new_smc, &ini); 1319 if (rc) 1320 goto out_unlock; 1321 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1322 if (rc) 1323 goto out_unlock; 1324 } 1325 1326 /* send SMC Accept CLC message */ 1327 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1328 if (rc) 1329 goto out_unlock; 1330 1331 /* SMC-D does not need this lock any more */ 1332 if (ism_supported) 1333 mutex_unlock(&smc_server_lgr_pending); 1334 1335 /* receive SMC Confirm CLC message */ 1336 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1337 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1338 if (rc) { 1339 if (!ism_supported) 1340 goto out_unlock; 1341 goto out_decl; 1342 } 1343 1344 /* finish worker */ 1345 if (!ism_supported) { 1346 rc = smc_listen_rdma_finish(new_smc, &cclc, 1347 ini.cln_first_contact); 1348 mutex_unlock(&smc_server_lgr_pending); 1349 if (rc) 1350 return; 1351 } 1352 smc_conn_save_peer_info(new_smc, &cclc); 1353 smc_listen_out_connected(new_smc); 1354 return; 1355 1356 out_unlock: 1357 mutex_unlock(&smc_server_lgr_pending); 1358 out_decl: 1359 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1360 } 1361 1362 static void smc_tcp_listen_work(struct work_struct *work) 1363 { 1364 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1365 tcp_listen_work); 1366 struct sock *lsk = &lsmc->sk; 1367 struct smc_sock *new_smc; 1368 int rc = 0; 1369 1370 lock_sock(lsk); 1371 while (lsk->sk_state == SMC_LISTEN) { 1372 rc = smc_clcsock_accept(lsmc, &new_smc); 1373 if (rc) 1374 goto out; 1375 if (!new_smc) 1376 continue; 1377 1378 new_smc->listen_smc = lsmc; 1379 new_smc->use_fallback = lsmc->use_fallback; 1380 new_smc->fallback_rsn = lsmc->fallback_rsn; 1381 sock_hold(lsk); /* sock_put in smc_listen_work */ 1382 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1383 smc_copy_sock_settings_to_smc(new_smc); 1384 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1385 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1386 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1387 if (!schedule_work(&new_smc->smc_listen_work)) 1388 sock_put(&new_smc->sk); 1389 } 1390 1391 out: 1392 release_sock(lsk); 1393 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1394 } 1395 1396 static int smc_listen(struct socket *sock, int backlog) 1397 { 1398 struct sock *sk = sock->sk; 1399 struct smc_sock *smc; 1400 int rc; 1401 1402 smc = smc_sk(sk); 1403 lock_sock(sk); 1404 1405 rc = -EINVAL; 1406 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1407 smc->connect_nonblock) 1408 goto out; 1409 1410 rc = 0; 1411 if (sk->sk_state == SMC_LISTEN) { 1412 sk->sk_max_ack_backlog = backlog; 1413 goto out; 1414 } 1415 /* some socket options are handled in core, so we could not apply 1416 * them to the clc socket -- copy smc socket options to clc socket 1417 */ 1418 smc_copy_sock_settings_to_clc(smc); 1419 if (!smc->use_fallback) 1420 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1421 1422 rc = kernel_listen(smc->clcsock, backlog); 1423 if (rc) 1424 goto out; 1425 sk->sk_max_ack_backlog = backlog; 1426 sk->sk_ack_backlog = 0; 1427 sk->sk_state = SMC_LISTEN; 1428 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1429 if (!schedule_work(&smc->tcp_listen_work)) 1430 sock_put(sk); 1431 1432 out: 1433 release_sock(sk); 1434 return rc; 1435 } 1436 1437 static int smc_accept(struct socket *sock, struct socket *new_sock, 1438 int flags, bool kern) 1439 { 1440 struct sock *sk = sock->sk, *nsk; 1441 DECLARE_WAITQUEUE(wait, current); 1442 struct smc_sock *lsmc; 1443 long timeo; 1444 int rc = 0; 1445 1446 lsmc = smc_sk(sk); 1447 sock_hold(sk); /* sock_put below */ 1448 lock_sock(sk); 1449 1450 if (lsmc->sk.sk_state != SMC_LISTEN) { 1451 rc = -EINVAL; 1452 release_sock(sk); 1453 goto out; 1454 } 1455 1456 /* Wait for an incoming connection */ 1457 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1458 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1459 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1460 set_current_state(TASK_INTERRUPTIBLE); 1461 if (!timeo) { 1462 rc = -EAGAIN; 1463 break; 1464 } 1465 release_sock(sk); 1466 timeo = schedule_timeout(timeo); 1467 /* wakeup by sk_data_ready in smc_listen_work() */ 1468 sched_annotate_sleep(); 1469 lock_sock(sk); 1470 if (signal_pending(current)) { 1471 rc = sock_intr_errno(timeo); 1472 break; 1473 } 1474 } 1475 set_current_state(TASK_RUNNING); 1476 remove_wait_queue(sk_sleep(sk), &wait); 1477 1478 if (!rc) 1479 rc = sock_error(nsk); 1480 release_sock(sk); 1481 if (rc) 1482 goto out; 1483 1484 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1485 /* wait till data arrives on the socket */ 1486 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1487 MSEC_PER_SEC); 1488 if (smc_sk(nsk)->use_fallback) { 1489 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1490 1491 lock_sock(clcsk); 1492 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1493 sk_wait_data(clcsk, &timeo, NULL); 1494 release_sock(clcsk); 1495 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1496 lock_sock(nsk); 1497 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1498 release_sock(nsk); 1499 } 1500 } 1501 1502 out: 1503 sock_put(sk); /* sock_hold above */ 1504 return rc; 1505 } 1506 1507 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1508 int peer) 1509 { 1510 struct smc_sock *smc; 1511 1512 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1513 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1514 return -ENOTCONN; 1515 1516 smc = smc_sk(sock->sk); 1517 1518 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1519 } 1520 1521 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1522 { 1523 struct sock *sk = sock->sk; 1524 struct smc_sock *smc; 1525 int rc = -EPIPE; 1526 1527 smc = smc_sk(sk); 1528 lock_sock(sk); 1529 if ((sk->sk_state != SMC_ACTIVE) && 1530 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1531 (sk->sk_state != SMC_INIT)) 1532 goto out; 1533 1534 if (msg->msg_flags & MSG_FASTOPEN) { 1535 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1536 smc_switch_to_fallback(smc); 1537 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1538 } else { 1539 rc = -EINVAL; 1540 goto out; 1541 } 1542 } 1543 1544 if (smc->use_fallback) 1545 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1546 else 1547 rc = smc_tx_sendmsg(smc, msg, len); 1548 out: 1549 release_sock(sk); 1550 return rc; 1551 } 1552 1553 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1554 int flags) 1555 { 1556 struct sock *sk = sock->sk; 1557 struct smc_sock *smc; 1558 int rc = -ENOTCONN; 1559 1560 smc = smc_sk(sk); 1561 lock_sock(sk); 1562 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1563 /* socket was connected before, no more data to read */ 1564 rc = 0; 1565 goto out; 1566 } 1567 if ((sk->sk_state == SMC_INIT) || 1568 (sk->sk_state == SMC_LISTEN) || 1569 (sk->sk_state == SMC_CLOSED)) 1570 goto out; 1571 1572 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1573 rc = 0; 1574 goto out; 1575 } 1576 1577 if (smc->use_fallback) { 1578 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1579 } else { 1580 msg->msg_namelen = 0; 1581 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1582 } 1583 1584 out: 1585 release_sock(sk); 1586 return rc; 1587 } 1588 1589 static __poll_t smc_accept_poll(struct sock *parent) 1590 { 1591 struct smc_sock *isk = smc_sk(parent); 1592 __poll_t mask = 0; 1593 1594 spin_lock(&isk->accept_q_lock); 1595 if (!list_empty(&isk->accept_q)) 1596 mask = EPOLLIN | EPOLLRDNORM; 1597 spin_unlock(&isk->accept_q_lock); 1598 1599 return mask; 1600 } 1601 1602 static __poll_t smc_poll(struct file *file, struct socket *sock, 1603 poll_table *wait) 1604 { 1605 struct sock *sk = sock->sk; 1606 struct smc_sock *smc; 1607 __poll_t mask = 0; 1608 1609 if (!sk) 1610 return EPOLLNVAL; 1611 1612 smc = smc_sk(sock->sk); 1613 if (smc->use_fallback) { 1614 /* delegate to CLC child sock */ 1615 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1616 sk->sk_err = smc->clcsock->sk->sk_err; 1617 } else { 1618 if (sk->sk_state != SMC_CLOSED) 1619 sock_poll_wait(file, sock, wait); 1620 if (sk->sk_err) 1621 mask |= EPOLLERR; 1622 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1623 (sk->sk_state == SMC_CLOSED)) 1624 mask |= EPOLLHUP; 1625 if (sk->sk_state == SMC_LISTEN) { 1626 /* woken up by sk_data_ready in smc_listen_work() */ 1627 mask |= smc_accept_poll(sk); 1628 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1629 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1630 wait); 1631 sk->sk_err = smc->clcsock->sk->sk_err; 1632 } else { 1633 if ((sk->sk_state != SMC_INIT && 1634 atomic_read(&smc->conn.sndbuf_space)) || 1635 sk->sk_shutdown & SEND_SHUTDOWN) { 1636 mask |= EPOLLOUT | EPOLLWRNORM; 1637 } else { 1638 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1639 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1640 } 1641 if (atomic_read(&smc->conn.bytes_to_rcv)) 1642 mask |= EPOLLIN | EPOLLRDNORM; 1643 if (sk->sk_shutdown & RCV_SHUTDOWN) 1644 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1645 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1646 mask |= EPOLLIN; 1647 if (smc->conn.urg_state == SMC_URG_VALID) 1648 mask |= EPOLLPRI; 1649 } 1650 } 1651 1652 return mask; 1653 } 1654 1655 static int smc_shutdown(struct socket *sock, int how) 1656 { 1657 struct sock *sk = sock->sk; 1658 struct smc_sock *smc; 1659 int rc = -EINVAL; 1660 int rc1 = 0; 1661 1662 smc = smc_sk(sk); 1663 1664 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1665 return rc; 1666 1667 lock_sock(sk); 1668 1669 rc = -ENOTCONN; 1670 if ((sk->sk_state != SMC_ACTIVE) && 1671 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1672 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1673 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1674 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1675 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1676 goto out; 1677 if (smc->use_fallback) { 1678 rc = kernel_sock_shutdown(smc->clcsock, how); 1679 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1680 if (sk->sk_shutdown == SHUTDOWN_MASK) 1681 sk->sk_state = SMC_CLOSED; 1682 goto out; 1683 } 1684 switch (how) { 1685 case SHUT_RDWR: /* shutdown in both directions */ 1686 rc = smc_close_active(smc); 1687 break; 1688 case SHUT_WR: 1689 rc = smc_close_shutdown_write(smc); 1690 break; 1691 case SHUT_RD: 1692 rc = 0; 1693 /* nothing more to do because peer is not involved */ 1694 break; 1695 } 1696 if (smc->clcsock) 1697 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1698 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1699 sk->sk_shutdown |= how + 1; 1700 1701 out: 1702 release_sock(sk); 1703 return rc ? rc : rc1; 1704 } 1705 1706 static int smc_setsockopt(struct socket *sock, int level, int optname, 1707 char __user *optval, unsigned int optlen) 1708 { 1709 struct sock *sk = sock->sk; 1710 struct smc_sock *smc; 1711 int val, rc; 1712 1713 smc = smc_sk(sk); 1714 1715 /* generic setsockopts reaching us here always apply to the 1716 * CLC socket 1717 */ 1718 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1719 optval, optlen); 1720 if (smc->clcsock->sk->sk_err) { 1721 sk->sk_err = smc->clcsock->sk->sk_err; 1722 sk->sk_error_report(sk); 1723 } 1724 if (rc) 1725 return rc; 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 switch (optname) { 1734 case TCP_ULP: 1735 case TCP_FASTOPEN: 1736 case TCP_FASTOPEN_CONNECT: 1737 case TCP_FASTOPEN_KEY: 1738 case TCP_FASTOPEN_NO_COOKIE: 1739 /* option not supported by SMC */ 1740 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1741 smc_switch_to_fallback(smc); 1742 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1743 } else { 1744 if (!smc->use_fallback) 1745 rc = -EINVAL; 1746 } 1747 break; 1748 case TCP_NODELAY: 1749 if (sk->sk_state != SMC_INIT && 1750 sk->sk_state != SMC_LISTEN && 1751 sk->sk_state != SMC_CLOSED) { 1752 if (val && !smc->use_fallback) 1753 mod_delayed_work(system_wq, &smc->conn.tx_work, 1754 0); 1755 } 1756 break; 1757 case TCP_CORK: 1758 if (sk->sk_state != SMC_INIT && 1759 sk->sk_state != SMC_LISTEN && 1760 sk->sk_state != SMC_CLOSED) { 1761 if (!val && !smc->use_fallback) 1762 mod_delayed_work(system_wq, &smc->conn.tx_work, 1763 0); 1764 } 1765 break; 1766 case TCP_DEFER_ACCEPT: 1767 smc->sockopt_defer_accept = val; 1768 break; 1769 default: 1770 break; 1771 } 1772 release_sock(sk); 1773 1774 return rc; 1775 } 1776 1777 static int smc_getsockopt(struct socket *sock, int level, int optname, 1778 char __user *optval, int __user *optlen) 1779 { 1780 struct smc_sock *smc; 1781 1782 smc = smc_sk(sock->sk); 1783 /* socket options apply to the CLC socket */ 1784 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1785 optval, optlen); 1786 } 1787 1788 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1789 unsigned long arg) 1790 { 1791 union smc_host_cursor cons, urg; 1792 struct smc_connection *conn; 1793 struct smc_sock *smc; 1794 int answ; 1795 1796 smc = smc_sk(sock->sk); 1797 conn = &smc->conn; 1798 lock_sock(&smc->sk); 1799 if (smc->use_fallback) { 1800 if (!smc->clcsock) { 1801 release_sock(&smc->sk); 1802 return -EBADF; 1803 } 1804 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1805 release_sock(&smc->sk); 1806 return answ; 1807 } 1808 switch (cmd) { 1809 case SIOCINQ: /* same as FIONREAD */ 1810 if (smc->sk.sk_state == SMC_LISTEN) { 1811 release_sock(&smc->sk); 1812 return -EINVAL; 1813 } 1814 if (smc->sk.sk_state == SMC_INIT || 1815 smc->sk.sk_state == SMC_CLOSED) 1816 answ = 0; 1817 else 1818 answ = atomic_read(&smc->conn.bytes_to_rcv); 1819 break; 1820 case SIOCOUTQ: 1821 /* output queue size (not send + not acked) */ 1822 if (smc->sk.sk_state == SMC_LISTEN) { 1823 release_sock(&smc->sk); 1824 return -EINVAL; 1825 } 1826 if (smc->sk.sk_state == SMC_INIT || 1827 smc->sk.sk_state == SMC_CLOSED) 1828 answ = 0; 1829 else 1830 answ = smc->conn.sndbuf_desc->len - 1831 atomic_read(&smc->conn.sndbuf_space); 1832 break; 1833 case SIOCOUTQNSD: 1834 /* output queue size (not send only) */ 1835 if (smc->sk.sk_state == SMC_LISTEN) { 1836 release_sock(&smc->sk); 1837 return -EINVAL; 1838 } 1839 if (smc->sk.sk_state == SMC_INIT || 1840 smc->sk.sk_state == SMC_CLOSED) 1841 answ = 0; 1842 else 1843 answ = smc_tx_prepared_sends(&smc->conn); 1844 break; 1845 case SIOCATMARK: 1846 if (smc->sk.sk_state == SMC_LISTEN) { 1847 release_sock(&smc->sk); 1848 return -EINVAL; 1849 } 1850 if (smc->sk.sk_state == SMC_INIT || 1851 smc->sk.sk_state == SMC_CLOSED) { 1852 answ = 0; 1853 } else { 1854 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1855 smc_curs_copy(&urg, &conn->urg_curs, conn); 1856 answ = smc_curs_diff(conn->rmb_desc->len, 1857 &cons, &urg) == 1; 1858 } 1859 break; 1860 default: 1861 release_sock(&smc->sk); 1862 return -ENOIOCTLCMD; 1863 } 1864 release_sock(&smc->sk); 1865 1866 return put_user(answ, (int __user *)arg); 1867 } 1868 1869 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1870 int offset, size_t size, int flags) 1871 { 1872 struct sock *sk = sock->sk; 1873 struct smc_sock *smc; 1874 int rc = -EPIPE; 1875 1876 smc = smc_sk(sk); 1877 lock_sock(sk); 1878 if (sk->sk_state != SMC_ACTIVE) { 1879 release_sock(sk); 1880 goto out; 1881 } 1882 release_sock(sk); 1883 if (smc->use_fallback) 1884 rc = kernel_sendpage(smc->clcsock, page, offset, 1885 size, flags); 1886 else 1887 rc = sock_no_sendpage(sock, page, offset, size, flags); 1888 1889 out: 1890 return rc; 1891 } 1892 1893 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1894 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1895 * updates till whenever a respective page has been fully processed. 1896 * Note that subsequent recv() calls have to wait till all splice() processing 1897 * completed. 1898 */ 1899 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1900 struct pipe_inode_info *pipe, size_t len, 1901 unsigned int flags) 1902 { 1903 struct sock *sk = sock->sk; 1904 struct smc_sock *smc; 1905 int rc = -ENOTCONN; 1906 1907 smc = smc_sk(sk); 1908 lock_sock(sk); 1909 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1910 /* socket was connected before, no more data to read */ 1911 rc = 0; 1912 goto out; 1913 } 1914 if (sk->sk_state == SMC_INIT || 1915 sk->sk_state == SMC_LISTEN || 1916 sk->sk_state == SMC_CLOSED) 1917 goto out; 1918 1919 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1920 rc = 0; 1921 goto out; 1922 } 1923 1924 if (smc->use_fallback) { 1925 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1926 pipe, len, flags); 1927 } else { 1928 if (*ppos) { 1929 rc = -ESPIPE; 1930 goto out; 1931 } 1932 if (flags & SPLICE_F_NONBLOCK) 1933 flags = MSG_DONTWAIT; 1934 else 1935 flags = 0; 1936 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1937 } 1938 out: 1939 release_sock(sk); 1940 1941 return rc; 1942 } 1943 1944 /* must look like tcp */ 1945 static const struct proto_ops smc_sock_ops = { 1946 .family = PF_SMC, 1947 .owner = THIS_MODULE, 1948 .release = smc_release, 1949 .bind = smc_bind, 1950 .connect = smc_connect, 1951 .socketpair = sock_no_socketpair, 1952 .accept = smc_accept, 1953 .getname = smc_getname, 1954 .poll = smc_poll, 1955 .ioctl = smc_ioctl, 1956 .listen = smc_listen, 1957 .shutdown = smc_shutdown, 1958 .setsockopt = smc_setsockopt, 1959 .getsockopt = smc_getsockopt, 1960 .sendmsg = smc_sendmsg, 1961 .recvmsg = smc_recvmsg, 1962 .mmap = sock_no_mmap, 1963 .sendpage = smc_sendpage, 1964 .splice_read = smc_splice_read, 1965 }; 1966 1967 static int smc_create(struct net *net, struct socket *sock, int protocol, 1968 int kern) 1969 { 1970 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1971 struct smc_sock *smc; 1972 struct sock *sk; 1973 int rc; 1974 1975 rc = -ESOCKTNOSUPPORT; 1976 if (sock->type != SOCK_STREAM) 1977 goto out; 1978 1979 rc = -EPROTONOSUPPORT; 1980 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1981 goto out; 1982 1983 rc = -ENOBUFS; 1984 sock->ops = &smc_sock_ops; 1985 sk = smc_sock_alloc(net, sock, protocol); 1986 if (!sk) 1987 goto out; 1988 1989 /* create internal TCP socket for CLC handshake and fallback */ 1990 smc = smc_sk(sk); 1991 smc->use_fallback = false; /* assume rdma capability first */ 1992 smc->fallback_rsn = 0; 1993 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 1994 &smc->clcsock); 1995 if (rc) { 1996 sk_common_release(sk); 1997 goto out; 1998 } 1999 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2000 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2001 2002 out: 2003 return rc; 2004 } 2005 2006 static const struct net_proto_family smc_sock_family_ops = { 2007 .family = PF_SMC, 2008 .owner = THIS_MODULE, 2009 .create = smc_create, 2010 }; 2011 2012 unsigned int smc_net_id; 2013 2014 static __net_init int smc_net_init(struct net *net) 2015 { 2016 return smc_pnet_net_init(net); 2017 } 2018 2019 static void __net_exit smc_net_exit(struct net *net) 2020 { 2021 smc_pnet_net_exit(net); 2022 } 2023 2024 static struct pernet_operations smc_net_ops = { 2025 .init = smc_net_init, 2026 .exit = smc_net_exit, 2027 .id = &smc_net_id, 2028 .size = sizeof(struct smc_net), 2029 }; 2030 2031 static int __init smc_init(void) 2032 { 2033 int rc; 2034 2035 rc = register_pernet_subsys(&smc_net_ops); 2036 if (rc) 2037 return rc; 2038 2039 rc = smc_pnet_init(); 2040 if (rc) 2041 goto out_pernet_subsys; 2042 2043 rc = smc_core_init(); 2044 if (rc) { 2045 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2046 goto out_pnet; 2047 } 2048 2049 rc = smc_llc_init(); 2050 if (rc) { 2051 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2052 goto out_core; 2053 } 2054 2055 rc = smc_cdc_init(); 2056 if (rc) { 2057 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2058 goto out_core; 2059 } 2060 2061 rc = proto_register(&smc_proto, 1); 2062 if (rc) { 2063 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2064 goto out_core; 2065 } 2066 2067 rc = proto_register(&smc_proto6, 1); 2068 if (rc) { 2069 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2070 goto out_proto; 2071 } 2072 2073 rc = sock_register(&smc_sock_family_ops); 2074 if (rc) { 2075 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2076 goto out_proto6; 2077 } 2078 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2079 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2080 2081 rc = smc_ib_register_client(); 2082 if (rc) { 2083 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2084 goto out_sock; 2085 } 2086 2087 static_branch_enable(&tcp_have_smc); 2088 return 0; 2089 2090 out_sock: 2091 sock_unregister(PF_SMC); 2092 out_proto6: 2093 proto_unregister(&smc_proto6); 2094 out_proto: 2095 proto_unregister(&smc_proto); 2096 out_core: 2097 smc_core_exit(); 2098 out_pnet: 2099 smc_pnet_exit(); 2100 out_pernet_subsys: 2101 unregister_pernet_subsys(&smc_net_ops); 2102 2103 return rc; 2104 } 2105 2106 static void __exit smc_exit(void) 2107 { 2108 static_branch_disable(&tcp_have_smc); 2109 sock_unregister(PF_SMC); 2110 smc_core_exit(); 2111 smc_ib_unregister_client(); 2112 proto_unregister(&smc_proto6); 2113 proto_unregister(&smc_proto); 2114 smc_pnet_exit(); 2115 unregister_pernet_subsys(&smc_net_ops); 2116 rcu_barrier(); 2117 } 2118 2119 module_init(smc_init); 2120 module_exit(smc_exit); 2121 2122 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2123 MODULE_DESCRIPTION("smc socket address family"); 2124 MODULE_LICENSE("GPL"); 2125 MODULE_ALIAS_NETPROTO(PF_SMC); 2126