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 smc->clcsock->wq.fasync_list = 474 smc->sk.sk_socket->wq.fasync_list; 475 } 476 } 477 478 /* fall back during connect */ 479 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 480 { 481 smc_switch_to_fallback(smc); 482 smc->fallback_rsn = reason_code; 483 smc_copy_sock_settings_to_clc(smc); 484 smc->connect_nonblock = 0; 485 if (smc->sk.sk_state == SMC_INIT) 486 smc->sk.sk_state = SMC_ACTIVE; 487 return 0; 488 } 489 490 /* decline and fall back during connect */ 491 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code) 492 { 493 int rc; 494 495 if (reason_code < 0) { /* error, fallback is not possible */ 496 if (smc->sk.sk_state == SMC_INIT) 497 sock_put(&smc->sk); /* passive closing */ 498 return reason_code; 499 } 500 if (reason_code != SMC_CLC_DECL_PEERDECL) { 501 rc = smc_clc_send_decline(smc, reason_code); 502 if (rc < 0) { 503 if (smc->sk.sk_state == SMC_INIT) 504 sock_put(&smc->sk); /* passive closing */ 505 return rc; 506 } 507 } 508 return smc_connect_fallback(smc, reason_code); 509 } 510 511 /* abort connecting */ 512 static int smc_connect_abort(struct smc_sock *smc, int reason_code, 513 int local_contact) 514 { 515 if (local_contact == SMC_FIRST_CONTACT) 516 smc_lgr_forget(smc->conn.lgr); 517 if (smc->conn.lgr->is_smcd) 518 /* there is only one lgr role for SMC-D; use server lock */ 519 mutex_unlock(&smc_server_lgr_pending); 520 else 521 mutex_unlock(&smc_client_lgr_pending); 522 523 smc_conn_free(&smc->conn); 524 smc->connect_nonblock = 0; 525 return reason_code; 526 } 527 528 /* check if there is a rdma device available for this connection. */ 529 /* called for connect and listen */ 530 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 531 { 532 /* PNET table look up: search active ib_device and port 533 * within same PNETID that also contains the ethernet device 534 * used for the internal TCP socket 535 */ 536 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 537 if (!ini->ib_dev) 538 return SMC_CLC_DECL_NOSMCRDEV; 539 return 0; 540 } 541 542 /* check if there is an ISM device available for this connection. */ 543 /* called for connect and listen */ 544 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 545 { 546 /* Find ISM device with same PNETID as connecting interface */ 547 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 548 if (!ini->ism_dev) 549 return SMC_CLC_DECL_NOSMCDDEV; 550 return 0; 551 } 552 553 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 554 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 555 struct smc_init_info *ini) 556 { 557 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 558 return SMC_CLC_DECL_ISMVLANERR; 559 return 0; 560 } 561 562 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 563 * used, the VLAN ID will be registered again during the connection setup. 564 */ 565 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 566 struct smc_init_info *ini) 567 { 568 if (!is_smcd) 569 return 0; 570 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 571 return SMC_CLC_DECL_CNFERR; 572 return 0; 573 } 574 575 /* CLC handshake during connect */ 576 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 577 struct smc_clc_msg_accept_confirm *aclc, 578 struct smc_init_info *ini) 579 { 580 int rc = 0; 581 582 /* do inband token exchange */ 583 rc = smc_clc_send_proposal(smc, smc_type, ini); 584 if (rc) 585 return rc; 586 /* receive SMC Accept CLC message */ 587 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 588 CLC_WAIT_TIME); 589 } 590 591 /* setup for RDMA connection of client */ 592 static int smc_connect_rdma(struct smc_sock *smc, 593 struct smc_clc_msg_accept_confirm *aclc, 594 struct smc_init_info *ini) 595 { 596 struct smc_link *link; 597 int reason_code = 0; 598 599 ini->is_smcd = false; 600 ini->ib_lcl = &aclc->lcl; 601 ini->ib_clcqpn = ntoh24(aclc->qpn); 602 ini->srv_first_contact = aclc->hdr.flag; 603 604 mutex_lock(&smc_client_lgr_pending); 605 reason_code = smc_conn_create(smc, ini); 606 if (reason_code) { 607 mutex_unlock(&smc_client_lgr_pending); 608 return reason_code; 609 } 610 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 611 612 smc_conn_save_peer_info(smc, aclc); 613 614 /* create send buffer and rmb */ 615 if (smc_buf_create(smc, false)) 616 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 617 ini->cln_first_contact); 618 619 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 620 smc_link_save_peer_info(link, aclc); 621 622 if (smc_rmb_rtoken_handling(&smc->conn, aclc)) 623 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 624 ini->cln_first_contact); 625 626 smc_close_init(smc); 627 smc_rx_init(smc); 628 629 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 630 if (smc_ib_ready_link(link)) 631 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 632 ini->cln_first_contact); 633 } else { 634 if (smc_reg_rmb(link, smc->conn.rmb_desc, true)) 635 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 636 ini->cln_first_contact); 637 } 638 smc_rmb_sync_sg_for_device(&smc->conn); 639 640 reason_code = smc_clc_send_confirm(smc); 641 if (reason_code) 642 return smc_connect_abort(smc, reason_code, 643 ini->cln_first_contact); 644 645 smc_tx_init(smc); 646 647 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 648 /* QP confirmation over RoCE fabric */ 649 reason_code = smc_clnt_conf_first_link(smc); 650 if (reason_code) 651 return smc_connect_abort(smc, reason_code, 652 ini->cln_first_contact); 653 } 654 mutex_unlock(&smc_client_lgr_pending); 655 656 smc_copy_sock_settings_to_clc(smc); 657 smc->connect_nonblock = 0; 658 if (smc->sk.sk_state == SMC_INIT) 659 smc->sk.sk_state = SMC_ACTIVE; 660 661 return 0; 662 } 663 664 /* setup for ISM connection of client */ 665 static int smc_connect_ism(struct smc_sock *smc, 666 struct smc_clc_msg_accept_confirm *aclc, 667 struct smc_init_info *ini) 668 { 669 int rc = 0; 670 671 ini->is_smcd = true; 672 ini->ism_gid = aclc->gid; 673 ini->srv_first_contact = aclc->hdr.flag; 674 675 /* there is only one lgr role for SMC-D; use server lock */ 676 mutex_lock(&smc_server_lgr_pending); 677 rc = smc_conn_create(smc, ini); 678 if (rc) { 679 mutex_unlock(&smc_server_lgr_pending); 680 return rc; 681 } 682 683 /* Create send and receive buffers */ 684 if (smc_buf_create(smc, true)) 685 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 686 ini->cln_first_contact); 687 688 smc_conn_save_peer_info(smc, aclc); 689 smc_close_init(smc); 690 smc_rx_init(smc); 691 smc_tx_init(smc); 692 693 rc = smc_clc_send_confirm(smc); 694 if (rc) 695 return smc_connect_abort(smc, rc, ini->cln_first_contact); 696 mutex_unlock(&smc_server_lgr_pending); 697 698 smc_copy_sock_settings_to_clc(smc); 699 smc->connect_nonblock = 0; 700 if (smc->sk.sk_state == SMC_INIT) 701 smc->sk.sk_state = SMC_ACTIVE; 702 703 return 0; 704 } 705 706 /* perform steps before actually connecting */ 707 static int __smc_connect(struct smc_sock *smc) 708 { 709 bool ism_supported = false, rdma_supported = false; 710 struct smc_clc_msg_accept_confirm aclc; 711 struct smc_init_info ini = {0}; 712 int smc_type; 713 int rc = 0; 714 715 if (smc->use_fallback) 716 return smc_connect_fallback(smc, smc->fallback_rsn); 717 718 /* if peer has not signalled SMC-capability, fall back */ 719 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 720 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 721 722 /* IPSec connections opt out of SMC-R optimizations */ 723 if (using_ipsec(smc)) 724 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 725 726 /* get vlan id from IP device */ 727 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 728 return smc_connect_decline_fallback(smc, 729 SMC_CLC_DECL_GETVLANERR); 730 731 /* check if there is an ism device available */ 732 if (!smc_find_ism_device(smc, &ini) && 733 !smc_connect_ism_vlan_setup(smc, &ini)) { 734 /* ISM is supported for this connection */ 735 ism_supported = true; 736 smc_type = SMC_TYPE_D; 737 } 738 739 /* check if there is a rdma device available */ 740 if (!smc_find_rdma_device(smc, &ini)) { 741 /* RDMA is supported for this connection */ 742 rdma_supported = true; 743 if (ism_supported) 744 smc_type = SMC_TYPE_B; /* both */ 745 else 746 smc_type = SMC_TYPE_R; /* only RDMA */ 747 } 748 749 /* if neither ISM nor RDMA are supported, fallback */ 750 if (!rdma_supported && !ism_supported) 751 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 752 753 /* perform CLC handshake */ 754 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 755 if (rc) { 756 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 757 return smc_connect_decline_fallback(smc, rc); 758 } 759 760 /* depending on previous steps, connect using rdma or ism */ 761 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 762 rc = smc_connect_rdma(smc, &aclc, &ini); 763 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 764 rc = smc_connect_ism(smc, &aclc, &ini); 765 else 766 rc = SMC_CLC_DECL_MODEUNSUPP; 767 if (rc) { 768 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 769 return smc_connect_decline_fallback(smc, rc); 770 } 771 772 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 773 return 0; 774 } 775 776 static void smc_connect_work(struct work_struct *work) 777 { 778 struct smc_sock *smc = container_of(work, struct smc_sock, 779 connect_work); 780 long timeo = smc->sk.sk_sndtimeo; 781 int rc = 0; 782 783 if (!timeo) 784 timeo = MAX_SCHEDULE_TIMEOUT; 785 lock_sock(smc->clcsock->sk); 786 if (smc->clcsock->sk->sk_err) { 787 smc->sk.sk_err = smc->clcsock->sk->sk_err; 788 } else if ((1 << smc->clcsock->sk->sk_state) & 789 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 790 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 791 if ((rc == -EPIPE) && 792 ((1 << smc->clcsock->sk->sk_state) & 793 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 794 rc = 0; 795 } 796 release_sock(smc->clcsock->sk); 797 lock_sock(&smc->sk); 798 if (rc != 0 || smc->sk.sk_err) { 799 smc->sk.sk_state = SMC_CLOSED; 800 if (rc == -EPIPE || rc == -EAGAIN) 801 smc->sk.sk_err = EPIPE; 802 else if (signal_pending(current)) 803 smc->sk.sk_err = -sock_intr_errno(timeo); 804 sock_put(&smc->sk); /* passive closing */ 805 goto out; 806 } 807 808 rc = __smc_connect(smc); 809 if (rc < 0) 810 smc->sk.sk_err = -rc; 811 812 out: 813 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 814 if (smc->sk.sk_err) { 815 smc->sk.sk_state_change(&smc->sk); 816 } else { /* allow polling before and after fallback decision */ 817 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 818 smc->sk.sk_write_space(&smc->sk); 819 } 820 } 821 release_sock(&smc->sk); 822 } 823 824 static int smc_connect(struct socket *sock, struct sockaddr *addr, 825 int alen, int flags) 826 { 827 struct sock *sk = sock->sk; 828 struct smc_sock *smc; 829 int rc = -EINVAL; 830 831 smc = smc_sk(sk); 832 833 /* separate smc parameter checking to be safe */ 834 if (alen < sizeof(addr->sa_family)) 835 goto out_err; 836 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 837 goto out_err; 838 839 lock_sock(sk); 840 switch (sk->sk_state) { 841 default: 842 goto out; 843 case SMC_ACTIVE: 844 rc = -EISCONN; 845 goto out; 846 case SMC_INIT: 847 rc = 0; 848 break; 849 } 850 851 smc_copy_sock_settings_to_clc(smc); 852 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 853 if (smc->connect_nonblock) { 854 rc = -EALREADY; 855 goto out; 856 } 857 rc = kernel_connect(smc->clcsock, addr, alen, flags); 858 if (rc && rc != -EINPROGRESS) 859 goto out; 860 861 sock_hold(&smc->sk); /* sock put in passive closing */ 862 if (smc->use_fallback) 863 goto out; 864 if (flags & O_NONBLOCK) { 865 if (schedule_work(&smc->connect_work)) 866 smc->connect_nonblock = 1; 867 rc = -EINPROGRESS; 868 } else { 869 rc = __smc_connect(smc); 870 if (rc < 0) 871 goto out; 872 else 873 rc = 0; /* success cases including fallback */ 874 } 875 876 out: 877 release_sock(sk); 878 out_err: 879 return rc; 880 } 881 882 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 883 { 884 struct socket *new_clcsock = NULL; 885 struct sock *lsk = &lsmc->sk; 886 struct sock *new_sk; 887 int rc = -EINVAL; 888 889 release_sock(lsk); 890 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 891 if (!new_sk) { 892 rc = -ENOMEM; 893 lsk->sk_err = ENOMEM; 894 *new_smc = NULL; 895 lock_sock(lsk); 896 goto out; 897 } 898 *new_smc = smc_sk(new_sk); 899 900 mutex_lock(&lsmc->clcsock_release_lock); 901 if (lsmc->clcsock) 902 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 903 mutex_unlock(&lsmc->clcsock_release_lock); 904 lock_sock(lsk); 905 if (rc < 0) 906 lsk->sk_err = -rc; 907 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 908 new_sk->sk_prot->unhash(new_sk); 909 if (new_clcsock) 910 sock_release(new_clcsock); 911 new_sk->sk_state = SMC_CLOSED; 912 sock_set_flag(new_sk, SOCK_DEAD); 913 sock_put(new_sk); /* final */ 914 *new_smc = NULL; 915 goto out; 916 } 917 918 (*new_smc)->clcsock = new_clcsock; 919 out: 920 return rc; 921 } 922 923 /* add a just created sock to the accept queue of the listen sock as 924 * candidate for a following socket accept call from user space 925 */ 926 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 927 { 928 struct smc_sock *par = smc_sk(parent); 929 930 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 931 spin_lock(&par->accept_q_lock); 932 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 933 spin_unlock(&par->accept_q_lock); 934 sk_acceptq_added(parent); 935 } 936 937 /* remove a socket from the accept queue of its parental listening socket */ 938 static void smc_accept_unlink(struct sock *sk) 939 { 940 struct smc_sock *par = smc_sk(sk)->listen_smc; 941 942 spin_lock(&par->accept_q_lock); 943 list_del_init(&smc_sk(sk)->accept_q); 944 spin_unlock(&par->accept_q_lock); 945 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 946 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 947 } 948 949 /* remove a sock from the accept queue to bind it to a new socket created 950 * for a socket accept call from user space 951 */ 952 struct sock *smc_accept_dequeue(struct sock *parent, 953 struct socket *new_sock) 954 { 955 struct smc_sock *isk, *n; 956 struct sock *new_sk; 957 958 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 959 new_sk = (struct sock *)isk; 960 961 smc_accept_unlink(new_sk); 962 if (new_sk->sk_state == SMC_CLOSED) { 963 new_sk->sk_prot->unhash(new_sk); 964 if (isk->clcsock) { 965 sock_release(isk->clcsock); 966 isk->clcsock = NULL; 967 } 968 sock_put(new_sk); /* final */ 969 continue; 970 } 971 if (new_sock) { 972 sock_graft(new_sk, new_sock); 973 if (isk->use_fallback) { 974 smc_sk(new_sk)->clcsock->file = new_sock->file; 975 isk->clcsock->file->private_data = isk->clcsock; 976 } 977 } 978 return new_sk; 979 } 980 return NULL; 981 } 982 983 /* clean up for a created but never accepted sock */ 984 void smc_close_non_accepted(struct sock *sk) 985 { 986 struct smc_sock *smc = smc_sk(sk); 987 988 sock_hold(sk); /* sock_put below */ 989 lock_sock(sk); 990 if (!sk->sk_lingertime) 991 /* wait for peer closing */ 992 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 993 __smc_release(smc); 994 release_sock(sk); 995 sock_put(sk); /* sock_hold above */ 996 sock_put(sk); /* final sock_put */ 997 } 998 999 static int smc_serv_conf_first_link(struct smc_sock *smc) 1000 { 1001 struct net *net = sock_net(smc->clcsock->sk); 1002 struct smc_link_group *lgr = smc->conn.lgr; 1003 struct smc_link *link; 1004 int rest; 1005 int rc; 1006 1007 link = &lgr->lnk[SMC_SINGLE_LINK]; 1008 1009 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 1010 return SMC_CLC_DECL_ERR_REGRMB; 1011 1012 /* send CONFIRM LINK request to client over the RoCE fabric */ 1013 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1014 if (rc < 0) 1015 return SMC_CLC_DECL_TIMEOUT_CL; 1016 1017 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1018 rest = wait_for_completion_interruptible_timeout( 1019 &link->llc_confirm_resp, 1020 SMC_LLC_WAIT_FIRST_TIME); 1021 if (rest <= 0) { 1022 struct smc_clc_msg_decline dclc; 1023 1024 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1025 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1026 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1027 } 1028 1029 if (link->llc_confirm_resp_rc) 1030 return SMC_CLC_DECL_RMBE_EC; 1031 1032 /* send ADD LINK request to client over the RoCE fabric */ 1033 rc = smc_llc_send_add_link(link, 1034 link->smcibdev->mac[link->ibport - 1], 1035 link->gid, SMC_LLC_REQ); 1036 if (rc < 0) 1037 return SMC_CLC_DECL_TIMEOUT_AL; 1038 1039 /* receive ADD LINK response from client over the RoCE fabric */ 1040 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1041 SMC_LLC_WAIT_TIME); 1042 if (rest <= 0) { 1043 struct smc_clc_msg_decline dclc; 1044 1045 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1046 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1047 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1048 } 1049 1050 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1051 1052 return 0; 1053 } 1054 1055 /* listen worker: finish */ 1056 static void smc_listen_out(struct smc_sock *new_smc) 1057 { 1058 struct smc_sock *lsmc = new_smc->listen_smc; 1059 struct sock *newsmcsk = &new_smc->sk; 1060 1061 if (lsmc->sk.sk_state == SMC_LISTEN) { 1062 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1063 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1064 release_sock(&lsmc->sk); 1065 } else { /* no longer listening */ 1066 smc_close_non_accepted(newsmcsk); 1067 } 1068 1069 /* Wake up accept */ 1070 lsmc->sk.sk_data_ready(&lsmc->sk); 1071 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1072 } 1073 1074 /* listen worker: finish in state connected */ 1075 static void smc_listen_out_connected(struct smc_sock *new_smc) 1076 { 1077 struct sock *newsmcsk = &new_smc->sk; 1078 1079 sk_refcnt_debug_inc(newsmcsk); 1080 if (newsmcsk->sk_state == SMC_INIT) 1081 newsmcsk->sk_state = SMC_ACTIVE; 1082 1083 smc_listen_out(new_smc); 1084 } 1085 1086 /* listen worker: finish in error state */ 1087 static void smc_listen_out_err(struct smc_sock *new_smc) 1088 { 1089 struct sock *newsmcsk = &new_smc->sk; 1090 1091 if (newsmcsk->sk_state == SMC_INIT) 1092 sock_put(&new_smc->sk); /* passive closing */ 1093 newsmcsk->sk_state = SMC_CLOSED; 1094 smc_conn_free(&new_smc->conn); 1095 1096 smc_listen_out(new_smc); 1097 } 1098 1099 /* listen worker: decline and fall back if possible */ 1100 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1101 int local_contact) 1102 { 1103 /* RDMA setup failed, switch back to TCP */ 1104 if (local_contact == SMC_FIRST_CONTACT) 1105 smc_lgr_forget(new_smc->conn.lgr); 1106 if (reason_code < 0) { /* error, no fallback possible */ 1107 smc_listen_out_err(new_smc); 1108 return; 1109 } 1110 smc_conn_free(&new_smc->conn); 1111 smc_switch_to_fallback(new_smc); 1112 new_smc->fallback_rsn = reason_code; 1113 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1114 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1115 smc_listen_out_err(new_smc); 1116 return; 1117 } 1118 } 1119 smc_listen_out_connected(new_smc); 1120 } 1121 1122 /* listen worker: check prefixes */ 1123 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1124 struct smc_clc_msg_proposal *pclc) 1125 { 1126 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1127 struct socket *newclcsock = new_smc->clcsock; 1128 1129 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1130 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1131 return SMC_CLC_DECL_DIFFPREFIX; 1132 1133 return 0; 1134 } 1135 1136 /* listen worker: initialize connection and buffers */ 1137 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1138 struct smc_init_info *ini) 1139 { 1140 int rc; 1141 1142 /* allocate connection / link group */ 1143 rc = smc_conn_create(new_smc, ini); 1144 if (rc) 1145 return rc; 1146 1147 /* create send buffer and rmb */ 1148 if (smc_buf_create(new_smc, false)) 1149 return SMC_CLC_DECL_MEM; 1150 1151 return 0; 1152 } 1153 1154 /* listen worker: initialize connection and buffers for SMC-D */ 1155 static int smc_listen_ism_init(struct smc_sock *new_smc, 1156 struct smc_clc_msg_proposal *pclc, 1157 struct smc_init_info *ini) 1158 { 1159 struct smc_clc_msg_smcd *pclc_smcd; 1160 int rc; 1161 1162 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1163 ini->ism_gid = pclc_smcd->gid; 1164 rc = smc_conn_create(new_smc, ini); 1165 if (rc) 1166 return rc; 1167 1168 /* Check if peer can be reached via ISM device */ 1169 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1170 new_smc->conn.lgr->vlan_id, 1171 new_smc->conn.lgr->smcd)) { 1172 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1173 smc_lgr_forget(new_smc->conn.lgr); 1174 smc_conn_free(&new_smc->conn); 1175 return SMC_CLC_DECL_SMCDNOTALK; 1176 } 1177 1178 /* Create send and receive buffers */ 1179 if (smc_buf_create(new_smc, true)) { 1180 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1181 smc_lgr_forget(new_smc->conn.lgr); 1182 smc_conn_free(&new_smc->conn); 1183 return SMC_CLC_DECL_MEM; 1184 } 1185 1186 return 0; 1187 } 1188 1189 /* listen worker: register buffers */ 1190 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1191 { 1192 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1193 1194 if (local_contact != SMC_FIRST_CONTACT) { 1195 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1196 return SMC_CLC_DECL_ERR_REGRMB; 1197 } 1198 smc_rmb_sync_sg_for_device(&new_smc->conn); 1199 1200 return 0; 1201 } 1202 1203 /* listen worker: finish RDMA setup */ 1204 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1205 struct smc_clc_msg_accept_confirm *cclc, 1206 int local_contact) 1207 { 1208 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1209 int reason_code = 0; 1210 1211 if (local_contact == SMC_FIRST_CONTACT) 1212 smc_link_save_peer_info(link, cclc); 1213 1214 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1215 reason_code = SMC_CLC_DECL_ERR_RTOK; 1216 goto decline; 1217 } 1218 1219 if (local_contact == SMC_FIRST_CONTACT) { 1220 if (smc_ib_ready_link(link)) { 1221 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1222 goto decline; 1223 } 1224 /* QP confirmation over RoCE fabric */ 1225 reason_code = smc_serv_conf_first_link(new_smc); 1226 if (reason_code) 1227 goto decline; 1228 } 1229 return 0; 1230 1231 decline: 1232 smc_listen_decline(new_smc, reason_code, local_contact); 1233 return reason_code; 1234 } 1235 1236 /* setup for RDMA connection of server */ 1237 static void smc_listen_work(struct work_struct *work) 1238 { 1239 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1240 smc_listen_work); 1241 struct socket *newclcsock = new_smc->clcsock; 1242 struct smc_clc_msg_accept_confirm cclc; 1243 struct smc_clc_msg_proposal *pclc; 1244 struct smc_init_info ini = {0}; 1245 bool ism_supported = false; 1246 u8 buf[SMC_CLC_MAX_LEN]; 1247 int rc = 0; 1248 1249 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1250 return smc_listen_out_err(new_smc); 1251 1252 if (new_smc->use_fallback) { 1253 smc_listen_out_connected(new_smc); 1254 return; 1255 } 1256 1257 /* check if peer is smc capable */ 1258 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1259 smc_switch_to_fallback(new_smc); 1260 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1261 smc_listen_out_connected(new_smc); 1262 return; 1263 } 1264 1265 /* do inband token exchange - 1266 * wait for and receive SMC Proposal CLC message 1267 */ 1268 pclc = (struct smc_clc_msg_proposal *)&buf; 1269 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1270 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1271 if (rc) 1272 goto out_decl; 1273 1274 /* IPSec connections opt out of SMC-R optimizations */ 1275 if (using_ipsec(new_smc)) { 1276 rc = SMC_CLC_DECL_IPSEC; 1277 goto out_decl; 1278 } 1279 1280 /* check for matching IP prefix and subnet length */ 1281 rc = smc_listen_prfx_check(new_smc, pclc); 1282 if (rc) 1283 goto out_decl; 1284 1285 /* get vlan id from IP device */ 1286 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1287 rc = SMC_CLC_DECL_GETVLANERR; 1288 goto out_decl; 1289 } 1290 1291 mutex_lock(&smc_server_lgr_pending); 1292 smc_close_init(new_smc); 1293 smc_rx_init(new_smc); 1294 smc_tx_init(new_smc); 1295 1296 /* check if ISM is available */ 1297 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1298 ini.is_smcd = true; /* prepare ISM check */ 1299 rc = smc_find_ism_device(new_smc, &ini); 1300 if (!rc) 1301 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1302 if (!rc) 1303 ism_supported = true; 1304 else if (pclc->hdr.path == SMC_TYPE_D) 1305 goto out_unlock; /* skip RDMA and decline */ 1306 } 1307 1308 /* check if RDMA is available */ 1309 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1310 /* prepare RDMA check */ 1311 ini.is_smcd = false; 1312 ini.ism_dev = NULL; 1313 ini.ib_lcl = &pclc->lcl; 1314 rc = smc_find_rdma_device(new_smc, &ini); 1315 if (rc) { 1316 /* no RDMA device found */ 1317 if (pclc->hdr.path == SMC_TYPE_B) 1318 /* neither ISM nor RDMA device found */ 1319 rc = SMC_CLC_DECL_NOSMCDEV; 1320 goto out_unlock; 1321 } 1322 rc = smc_listen_rdma_init(new_smc, &ini); 1323 if (rc) 1324 goto out_unlock; 1325 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1326 if (rc) 1327 goto out_unlock; 1328 } 1329 1330 /* send SMC Accept CLC message */ 1331 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1332 if (rc) 1333 goto out_unlock; 1334 1335 /* SMC-D does not need this lock any more */ 1336 if (ism_supported) 1337 mutex_unlock(&smc_server_lgr_pending); 1338 1339 /* receive SMC Confirm CLC message */ 1340 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1341 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1342 if (rc) { 1343 if (!ism_supported) 1344 goto out_unlock; 1345 goto out_decl; 1346 } 1347 1348 /* finish worker */ 1349 if (!ism_supported) { 1350 rc = smc_listen_rdma_finish(new_smc, &cclc, 1351 ini.cln_first_contact); 1352 mutex_unlock(&smc_server_lgr_pending); 1353 if (rc) 1354 return; 1355 } 1356 smc_conn_save_peer_info(new_smc, &cclc); 1357 smc_listen_out_connected(new_smc); 1358 return; 1359 1360 out_unlock: 1361 mutex_unlock(&smc_server_lgr_pending); 1362 out_decl: 1363 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1364 } 1365 1366 static void smc_tcp_listen_work(struct work_struct *work) 1367 { 1368 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1369 tcp_listen_work); 1370 struct sock *lsk = &lsmc->sk; 1371 struct smc_sock *new_smc; 1372 int rc = 0; 1373 1374 lock_sock(lsk); 1375 while (lsk->sk_state == SMC_LISTEN) { 1376 rc = smc_clcsock_accept(lsmc, &new_smc); 1377 if (rc) 1378 goto out; 1379 if (!new_smc) 1380 continue; 1381 1382 new_smc->listen_smc = lsmc; 1383 new_smc->use_fallback = lsmc->use_fallback; 1384 new_smc->fallback_rsn = lsmc->fallback_rsn; 1385 sock_hold(lsk); /* sock_put in smc_listen_work */ 1386 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1387 smc_copy_sock_settings_to_smc(new_smc); 1388 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1389 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1390 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1391 if (!schedule_work(&new_smc->smc_listen_work)) 1392 sock_put(&new_smc->sk); 1393 } 1394 1395 out: 1396 release_sock(lsk); 1397 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1398 } 1399 1400 static int smc_listen(struct socket *sock, int backlog) 1401 { 1402 struct sock *sk = sock->sk; 1403 struct smc_sock *smc; 1404 int rc; 1405 1406 smc = smc_sk(sk); 1407 lock_sock(sk); 1408 1409 rc = -EINVAL; 1410 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1411 smc->connect_nonblock) 1412 goto out; 1413 1414 rc = 0; 1415 if (sk->sk_state == SMC_LISTEN) { 1416 sk->sk_max_ack_backlog = backlog; 1417 goto out; 1418 } 1419 /* some socket options are handled in core, so we could not apply 1420 * them to the clc socket -- copy smc socket options to clc socket 1421 */ 1422 smc_copy_sock_settings_to_clc(smc); 1423 if (!smc->use_fallback) 1424 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1425 1426 rc = kernel_listen(smc->clcsock, backlog); 1427 if (rc) 1428 goto out; 1429 sk->sk_max_ack_backlog = backlog; 1430 sk->sk_ack_backlog = 0; 1431 sk->sk_state = SMC_LISTEN; 1432 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1433 if (!schedule_work(&smc->tcp_listen_work)) 1434 sock_put(sk); 1435 1436 out: 1437 release_sock(sk); 1438 return rc; 1439 } 1440 1441 static int smc_accept(struct socket *sock, struct socket *new_sock, 1442 int flags, bool kern) 1443 { 1444 struct sock *sk = sock->sk, *nsk; 1445 DECLARE_WAITQUEUE(wait, current); 1446 struct smc_sock *lsmc; 1447 long timeo; 1448 int rc = 0; 1449 1450 lsmc = smc_sk(sk); 1451 sock_hold(sk); /* sock_put below */ 1452 lock_sock(sk); 1453 1454 if (lsmc->sk.sk_state != SMC_LISTEN) { 1455 rc = -EINVAL; 1456 release_sock(sk); 1457 goto out; 1458 } 1459 1460 /* Wait for an incoming connection */ 1461 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1462 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1463 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1464 set_current_state(TASK_INTERRUPTIBLE); 1465 if (!timeo) { 1466 rc = -EAGAIN; 1467 break; 1468 } 1469 release_sock(sk); 1470 timeo = schedule_timeout(timeo); 1471 /* wakeup by sk_data_ready in smc_listen_work() */ 1472 sched_annotate_sleep(); 1473 lock_sock(sk); 1474 if (signal_pending(current)) { 1475 rc = sock_intr_errno(timeo); 1476 break; 1477 } 1478 } 1479 set_current_state(TASK_RUNNING); 1480 remove_wait_queue(sk_sleep(sk), &wait); 1481 1482 if (!rc) 1483 rc = sock_error(nsk); 1484 release_sock(sk); 1485 if (rc) 1486 goto out; 1487 1488 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1489 /* wait till data arrives on the socket */ 1490 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1491 MSEC_PER_SEC); 1492 if (smc_sk(nsk)->use_fallback) { 1493 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1494 1495 lock_sock(clcsk); 1496 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1497 sk_wait_data(clcsk, &timeo, NULL); 1498 release_sock(clcsk); 1499 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1500 lock_sock(nsk); 1501 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1502 release_sock(nsk); 1503 } 1504 } 1505 1506 out: 1507 sock_put(sk); /* sock_hold above */ 1508 return rc; 1509 } 1510 1511 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1512 int peer) 1513 { 1514 struct smc_sock *smc; 1515 1516 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1517 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1518 return -ENOTCONN; 1519 1520 smc = smc_sk(sock->sk); 1521 1522 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1523 } 1524 1525 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1526 { 1527 struct sock *sk = sock->sk; 1528 struct smc_sock *smc; 1529 int rc = -EPIPE; 1530 1531 smc = smc_sk(sk); 1532 lock_sock(sk); 1533 if ((sk->sk_state != SMC_ACTIVE) && 1534 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1535 (sk->sk_state != SMC_INIT)) 1536 goto out; 1537 1538 if (msg->msg_flags & MSG_FASTOPEN) { 1539 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1540 smc_switch_to_fallback(smc); 1541 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1542 } else { 1543 rc = -EINVAL; 1544 goto out; 1545 } 1546 } 1547 1548 if (smc->use_fallback) 1549 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1550 else 1551 rc = smc_tx_sendmsg(smc, msg, len); 1552 out: 1553 release_sock(sk); 1554 return rc; 1555 } 1556 1557 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1558 int flags) 1559 { 1560 struct sock *sk = sock->sk; 1561 struct smc_sock *smc; 1562 int rc = -ENOTCONN; 1563 1564 smc = smc_sk(sk); 1565 lock_sock(sk); 1566 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1567 /* socket was connected before, no more data to read */ 1568 rc = 0; 1569 goto out; 1570 } 1571 if ((sk->sk_state == SMC_INIT) || 1572 (sk->sk_state == SMC_LISTEN) || 1573 (sk->sk_state == SMC_CLOSED)) 1574 goto out; 1575 1576 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1577 rc = 0; 1578 goto out; 1579 } 1580 1581 if (smc->use_fallback) { 1582 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1583 } else { 1584 msg->msg_namelen = 0; 1585 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1586 } 1587 1588 out: 1589 release_sock(sk); 1590 return rc; 1591 } 1592 1593 static __poll_t smc_accept_poll(struct sock *parent) 1594 { 1595 struct smc_sock *isk = smc_sk(parent); 1596 __poll_t mask = 0; 1597 1598 spin_lock(&isk->accept_q_lock); 1599 if (!list_empty(&isk->accept_q)) 1600 mask = EPOLLIN | EPOLLRDNORM; 1601 spin_unlock(&isk->accept_q_lock); 1602 1603 return mask; 1604 } 1605 1606 static __poll_t smc_poll(struct file *file, struct socket *sock, 1607 poll_table *wait) 1608 { 1609 struct sock *sk = sock->sk; 1610 struct smc_sock *smc; 1611 __poll_t mask = 0; 1612 1613 if (!sk) 1614 return EPOLLNVAL; 1615 1616 smc = smc_sk(sock->sk); 1617 if (smc->use_fallback) { 1618 /* delegate to CLC child sock */ 1619 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1620 sk->sk_err = smc->clcsock->sk->sk_err; 1621 } else { 1622 if (sk->sk_state != SMC_CLOSED) 1623 sock_poll_wait(file, sock, wait); 1624 if (sk->sk_err) 1625 mask |= EPOLLERR; 1626 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1627 (sk->sk_state == SMC_CLOSED)) 1628 mask |= EPOLLHUP; 1629 if (sk->sk_state == SMC_LISTEN) { 1630 /* woken up by sk_data_ready in smc_listen_work() */ 1631 mask |= smc_accept_poll(sk); 1632 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1633 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1634 wait); 1635 sk->sk_err = smc->clcsock->sk->sk_err; 1636 } else { 1637 if ((sk->sk_state != SMC_INIT && 1638 atomic_read(&smc->conn.sndbuf_space)) || 1639 sk->sk_shutdown & SEND_SHUTDOWN) { 1640 mask |= EPOLLOUT | EPOLLWRNORM; 1641 } else { 1642 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1643 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1644 } 1645 if (atomic_read(&smc->conn.bytes_to_rcv)) 1646 mask |= EPOLLIN | EPOLLRDNORM; 1647 if (sk->sk_shutdown & RCV_SHUTDOWN) 1648 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1649 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1650 mask |= EPOLLIN; 1651 if (smc->conn.urg_state == SMC_URG_VALID) 1652 mask |= EPOLLPRI; 1653 } 1654 } 1655 1656 return mask; 1657 } 1658 1659 static int smc_shutdown(struct socket *sock, int how) 1660 { 1661 struct sock *sk = sock->sk; 1662 struct smc_sock *smc; 1663 int rc = -EINVAL; 1664 int rc1 = 0; 1665 1666 smc = smc_sk(sk); 1667 1668 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1669 return rc; 1670 1671 lock_sock(sk); 1672 1673 rc = -ENOTCONN; 1674 if ((sk->sk_state != SMC_ACTIVE) && 1675 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1676 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1677 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1678 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1679 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1680 goto out; 1681 if (smc->use_fallback) { 1682 rc = kernel_sock_shutdown(smc->clcsock, how); 1683 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1684 if (sk->sk_shutdown == SHUTDOWN_MASK) 1685 sk->sk_state = SMC_CLOSED; 1686 goto out; 1687 } 1688 switch (how) { 1689 case SHUT_RDWR: /* shutdown in both directions */ 1690 rc = smc_close_active(smc); 1691 break; 1692 case SHUT_WR: 1693 rc = smc_close_shutdown_write(smc); 1694 break; 1695 case SHUT_RD: 1696 rc = 0; 1697 /* nothing more to do because peer is not involved */ 1698 break; 1699 } 1700 if (smc->clcsock) 1701 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1702 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1703 sk->sk_shutdown |= how + 1; 1704 1705 out: 1706 release_sock(sk); 1707 return rc ? rc : rc1; 1708 } 1709 1710 static int smc_setsockopt(struct socket *sock, int level, int optname, 1711 char __user *optval, unsigned int optlen) 1712 { 1713 struct sock *sk = sock->sk; 1714 struct smc_sock *smc; 1715 int val, rc; 1716 1717 smc = smc_sk(sk); 1718 1719 /* generic setsockopts reaching us here always apply to the 1720 * CLC socket 1721 */ 1722 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1723 optval, optlen); 1724 if (smc->clcsock->sk->sk_err) { 1725 sk->sk_err = smc->clcsock->sk->sk_err; 1726 sk->sk_error_report(sk); 1727 } 1728 1729 if (optlen < sizeof(int)) 1730 return -EINVAL; 1731 if (get_user(val, (int __user *)optval)) 1732 return -EFAULT; 1733 1734 lock_sock(sk); 1735 if (rc || smc->use_fallback) 1736 goto out; 1737 switch (optname) { 1738 case TCP_ULP: 1739 case TCP_FASTOPEN: 1740 case TCP_FASTOPEN_CONNECT: 1741 case TCP_FASTOPEN_KEY: 1742 case TCP_FASTOPEN_NO_COOKIE: 1743 /* option not supported by SMC */ 1744 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1745 smc_switch_to_fallback(smc); 1746 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1747 } else { 1748 rc = -EINVAL; 1749 } 1750 break; 1751 case TCP_NODELAY: 1752 if (sk->sk_state != SMC_INIT && 1753 sk->sk_state != SMC_LISTEN && 1754 sk->sk_state != SMC_CLOSED) { 1755 if (val) 1756 mod_delayed_work(system_wq, &smc->conn.tx_work, 1757 0); 1758 } 1759 break; 1760 case TCP_CORK: 1761 if (sk->sk_state != SMC_INIT && 1762 sk->sk_state != SMC_LISTEN && 1763 sk->sk_state != SMC_CLOSED) { 1764 if (!val) 1765 mod_delayed_work(system_wq, &smc->conn.tx_work, 1766 0); 1767 } 1768 break; 1769 case TCP_DEFER_ACCEPT: 1770 smc->sockopt_defer_accept = val; 1771 break; 1772 default: 1773 break; 1774 } 1775 out: 1776 release_sock(sk); 1777 1778 return rc; 1779 } 1780 1781 static int smc_getsockopt(struct socket *sock, int level, int optname, 1782 char __user *optval, int __user *optlen) 1783 { 1784 struct smc_sock *smc; 1785 1786 smc = smc_sk(sock->sk); 1787 /* socket options apply to the CLC socket */ 1788 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1789 optval, optlen); 1790 } 1791 1792 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1793 unsigned long arg) 1794 { 1795 union smc_host_cursor cons, urg; 1796 struct smc_connection *conn; 1797 struct smc_sock *smc; 1798 int answ; 1799 1800 smc = smc_sk(sock->sk); 1801 conn = &smc->conn; 1802 lock_sock(&smc->sk); 1803 if (smc->use_fallback) { 1804 if (!smc->clcsock) { 1805 release_sock(&smc->sk); 1806 return -EBADF; 1807 } 1808 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1809 release_sock(&smc->sk); 1810 return answ; 1811 } 1812 switch (cmd) { 1813 case SIOCINQ: /* same as FIONREAD */ 1814 if (smc->sk.sk_state == SMC_LISTEN) { 1815 release_sock(&smc->sk); 1816 return -EINVAL; 1817 } 1818 if (smc->sk.sk_state == SMC_INIT || 1819 smc->sk.sk_state == SMC_CLOSED) 1820 answ = 0; 1821 else 1822 answ = atomic_read(&smc->conn.bytes_to_rcv); 1823 break; 1824 case SIOCOUTQ: 1825 /* output queue size (not send + not acked) */ 1826 if (smc->sk.sk_state == SMC_LISTEN) { 1827 release_sock(&smc->sk); 1828 return -EINVAL; 1829 } 1830 if (smc->sk.sk_state == SMC_INIT || 1831 smc->sk.sk_state == SMC_CLOSED) 1832 answ = 0; 1833 else 1834 answ = smc->conn.sndbuf_desc->len - 1835 atomic_read(&smc->conn.sndbuf_space); 1836 break; 1837 case SIOCOUTQNSD: 1838 /* output queue size (not send only) */ 1839 if (smc->sk.sk_state == SMC_LISTEN) { 1840 release_sock(&smc->sk); 1841 return -EINVAL; 1842 } 1843 if (smc->sk.sk_state == SMC_INIT || 1844 smc->sk.sk_state == SMC_CLOSED) 1845 answ = 0; 1846 else 1847 answ = smc_tx_prepared_sends(&smc->conn); 1848 break; 1849 case SIOCATMARK: 1850 if (smc->sk.sk_state == SMC_LISTEN) { 1851 release_sock(&smc->sk); 1852 return -EINVAL; 1853 } 1854 if (smc->sk.sk_state == SMC_INIT || 1855 smc->sk.sk_state == SMC_CLOSED) { 1856 answ = 0; 1857 } else { 1858 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1859 smc_curs_copy(&urg, &conn->urg_curs, conn); 1860 answ = smc_curs_diff(conn->rmb_desc->len, 1861 &cons, &urg) == 1; 1862 } 1863 break; 1864 default: 1865 release_sock(&smc->sk); 1866 return -ENOIOCTLCMD; 1867 } 1868 release_sock(&smc->sk); 1869 1870 return put_user(answ, (int __user *)arg); 1871 } 1872 1873 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1874 int offset, size_t size, int flags) 1875 { 1876 struct sock *sk = sock->sk; 1877 struct smc_sock *smc; 1878 int rc = -EPIPE; 1879 1880 smc = smc_sk(sk); 1881 lock_sock(sk); 1882 if (sk->sk_state != SMC_ACTIVE) { 1883 release_sock(sk); 1884 goto out; 1885 } 1886 release_sock(sk); 1887 if (smc->use_fallback) 1888 rc = kernel_sendpage(smc->clcsock, page, offset, 1889 size, flags); 1890 else 1891 rc = sock_no_sendpage(sock, page, offset, size, flags); 1892 1893 out: 1894 return rc; 1895 } 1896 1897 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1898 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1899 * updates till whenever a respective page has been fully processed. 1900 * Note that subsequent recv() calls have to wait till all splice() processing 1901 * completed. 1902 */ 1903 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1904 struct pipe_inode_info *pipe, size_t len, 1905 unsigned int flags) 1906 { 1907 struct sock *sk = sock->sk; 1908 struct smc_sock *smc; 1909 int rc = -ENOTCONN; 1910 1911 smc = smc_sk(sk); 1912 lock_sock(sk); 1913 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1914 /* socket was connected before, no more data to read */ 1915 rc = 0; 1916 goto out; 1917 } 1918 if (sk->sk_state == SMC_INIT || 1919 sk->sk_state == SMC_LISTEN || 1920 sk->sk_state == SMC_CLOSED) 1921 goto out; 1922 1923 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1924 rc = 0; 1925 goto out; 1926 } 1927 1928 if (smc->use_fallback) { 1929 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1930 pipe, len, flags); 1931 } else { 1932 if (*ppos) { 1933 rc = -ESPIPE; 1934 goto out; 1935 } 1936 if (flags & SPLICE_F_NONBLOCK) 1937 flags = MSG_DONTWAIT; 1938 else 1939 flags = 0; 1940 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1941 } 1942 out: 1943 release_sock(sk); 1944 1945 return rc; 1946 } 1947 1948 /* must look like tcp */ 1949 static const struct proto_ops smc_sock_ops = { 1950 .family = PF_SMC, 1951 .owner = THIS_MODULE, 1952 .release = smc_release, 1953 .bind = smc_bind, 1954 .connect = smc_connect, 1955 .socketpair = sock_no_socketpair, 1956 .accept = smc_accept, 1957 .getname = smc_getname, 1958 .poll = smc_poll, 1959 .ioctl = smc_ioctl, 1960 .listen = smc_listen, 1961 .shutdown = smc_shutdown, 1962 .setsockopt = smc_setsockopt, 1963 .getsockopt = smc_getsockopt, 1964 .sendmsg = smc_sendmsg, 1965 .recvmsg = smc_recvmsg, 1966 .mmap = sock_no_mmap, 1967 .sendpage = smc_sendpage, 1968 .splice_read = smc_splice_read, 1969 }; 1970 1971 static int smc_create(struct net *net, struct socket *sock, int protocol, 1972 int kern) 1973 { 1974 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1975 struct smc_sock *smc; 1976 struct sock *sk; 1977 int rc; 1978 1979 rc = -ESOCKTNOSUPPORT; 1980 if (sock->type != SOCK_STREAM) 1981 goto out; 1982 1983 rc = -EPROTONOSUPPORT; 1984 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1985 goto out; 1986 1987 rc = -ENOBUFS; 1988 sock->ops = &smc_sock_ops; 1989 sk = smc_sock_alloc(net, sock, protocol); 1990 if (!sk) 1991 goto out; 1992 1993 /* create internal TCP socket for CLC handshake and fallback */ 1994 smc = smc_sk(sk); 1995 smc->use_fallback = false; /* assume rdma capability first */ 1996 smc->fallback_rsn = 0; 1997 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 1998 &smc->clcsock); 1999 if (rc) { 2000 sk_common_release(sk); 2001 goto out; 2002 } 2003 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2004 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2005 2006 out: 2007 return rc; 2008 } 2009 2010 static const struct net_proto_family smc_sock_family_ops = { 2011 .family = PF_SMC, 2012 .owner = THIS_MODULE, 2013 .create = smc_create, 2014 }; 2015 2016 unsigned int smc_net_id; 2017 2018 static __net_init int smc_net_init(struct net *net) 2019 { 2020 return smc_pnet_net_init(net); 2021 } 2022 2023 static void __net_exit smc_net_exit(struct net *net) 2024 { 2025 smc_pnet_net_exit(net); 2026 } 2027 2028 static struct pernet_operations smc_net_ops = { 2029 .init = smc_net_init, 2030 .exit = smc_net_exit, 2031 .id = &smc_net_id, 2032 .size = sizeof(struct smc_net), 2033 }; 2034 2035 static int __init smc_init(void) 2036 { 2037 int rc; 2038 2039 rc = register_pernet_subsys(&smc_net_ops); 2040 if (rc) 2041 return rc; 2042 2043 rc = smc_pnet_init(); 2044 if (rc) 2045 goto out_pernet_subsys; 2046 2047 rc = smc_core_init(); 2048 if (rc) { 2049 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2050 goto out_pnet; 2051 } 2052 2053 rc = smc_llc_init(); 2054 if (rc) { 2055 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2056 goto out_core; 2057 } 2058 2059 rc = smc_cdc_init(); 2060 if (rc) { 2061 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2062 goto out_core; 2063 } 2064 2065 rc = proto_register(&smc_proto, 1); 2066 if (rc) { 2067 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2068 goto out_core; 2069 } 2070 2071 rc = proto_register(&smc_proto6, 1); 2072 if (rc) { 2073 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2074 goto out_proto; 2075 } 2076 2077 rc = sock_register(&smc_sock_family_ops); 2078 if (rc) { 2079 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2080 goto out_proto6; 2081 } 2082 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2083 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2084 2085 rc = smc_ib_register_client(); 2086 if (rc) { 2087 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2088 goto out_sock; 2089 } 2090 2091 static_branch_enable(&tcp_have_smc); 2092 return 0; 2093 2094 out_sock: 2095 sock_unregister(PF_SMC); 2096 out_proto6: 2097 proto_unregister(&smc_proto6); 2098 out_proto: 2099 proto_unregister(&smc_proto); 2100 out_core: 2101 smc_core_exit(); 2102 out_pnet: 2103 smc_pnet_exit(); 2104 out_pernet_subsys: 2105 unregister_pernet_subsys(&smc_net_ops); 2106 2107 return rc; 2108 } 2109 2110 static void __exit smc_exit(void) 2111 { 2112 static_branch_disable(&tcp_have_smc); 2113 sock_unregister(PF_SMC); 2114 smc_core_exit(); 2115 smc_ib_unregister_client(); 2116 proto_unregister(&smc_proto6); 2117 proto_unregister(&smc_proto); 2118 smc_pnet_exit(); 2119 unregister_pernet_subsys(&smc_net_ops); 2120 rcu_barrier(); 2121 } 2122 2123 module_init(smc_init); 2124 module_exit(smc_exit); 2125 2126 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2127 MODULE_DESCRIPTION("smc socket address family"); 2128 MODULE_LICENSE("GPL"); 2129 MODULE_ALIAS_NETPROTO(PF_SMC); 2130