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 bool is_smcd = smc->conn.lgr->is_smcd; 516 517 if (local_contact == SMC_FIRST_CONTACT) 518 smc_lgr_cleanup_early(&smc->conn); 519 else 520 smc_conn_free(&smc->conn); 521 if (is_smcd) 522 /* there is only one lgr role for SMC-D; use server lock */ 523 mutex_unlock(&smc_server_lgr_pending); 524 else 525 mutex_unlock(&smc_client_lgr_pending); 526 527 smc->connect_nonblock = 0; 528 return reason_code; 529 } 530 531 /* check if there is a rdma device available for this connection. */ 532 /* called for connect and listen */ 533 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 534 { 535 /* PNET table look up: search active ib_device and port 536 * within same PNETID that also contains the ethernet device 537 * used for the internal TCP socket 538 */ 539 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 540 if (!ini->ib_dev) 541 return SMC_CLC_DECL_NOSMCRDEV; 542 return 0; 543 } 544 545 /* check if there is an ISM device available for this connection. */ 546 /* called for connect and listen */ 547 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 548 { 549 /* Find ISM device with same PNETID as connecting interface */ 550 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 551 if (!ini->ism_dev) 552 return SMC_CLC_DECL_NOSMCDDEV; 553 return 0; 554 } 555 556 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 557 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 558 struct smc_init_info *ini) 559 { 560 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 561 return SMC_CLC_DECL_ISMVLANERR; 562 return 0; 563 } 564 565 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 566 * used, the VLAN ID will be registered again during the connection setup. 567 */ 568 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 569 struct smc_init_info *ini) 570 { 571 if (!is_smcd) 572 return 0; 573 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 574 return SMC_CLC_DECL_CNFERR; 575 return 0; 576 } 577 578 /* CLC handshake during connect */ 579 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 580 struct smc_clc_msg_accept_confirm *aclc, 581 struct smc_init_info *ini) 582 { 583 int rc = 0; 584 585 /* do inband token exchange */ 586 rc = smc_clc_send_proposal(smc, smc_type, ini); 587 if (rc) 588 return rc; 589 /* receive SMC Accept CLC message */ 590 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 591 CLC_WAIT_TIME); 592 } 593 594 /* setup for RDMA connection of client */ 595 static int smc_connect_rdma(struct smc_sock *smc, 596 struct smc_clc_msg_accept_confirm *aclc, 597 struct smc_init_info *ini) 598 { 599 struct smc_link *link; 600 int reason_code = 0; 601 602 ini->is_smcd = false; 603 ini->ib_lcl = &aclc->lcl; 604 ini->ib_clcqpn = ntoh24(aclc->qpn); 605 ini->srv_first_contact = aclc->hdr.flag; 606 607 mutex_lock(&smc_client_lgr_pending); 608 reason_code = smc_conn_create(smc, ini); 609 if (reason_code) { 610 mutex_unlock(&smc_client_lgr_pending); 611 return reason_code; 612 } 613 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 614 615 smc_conn_save_peer_info(smc, aclc); 616 617 /* create send buffer and rmb */ 618 if (smc_buf_create(smc, false)) 619 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 620 ini->cln_first_contact); 621 622 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 623 smc_link_save_peer_info(link, aclc); 624 625 if (smc_rmb_rtoken_handling(&smc->conn, aclc)) 626 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 627 ini->cln_first_contact); 628 629 smc_close_init(smc); 630 smc_rx_init(smc); 631 632 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 633 if (smc_ib_ready_link(link)) 634 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 635 ini->cln_first_contact); 636 } else { 637 if (smc_reg_rmb(link, smc->conn.rmb_desc, true)) 638 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 639 ini->cln_first_contact); 640 } 641 smc_rmb_sync_sg_for_device(&smc->conn); 642 643 reason_code = smc_clc_send_confirm(smc); 644 if (reason_code) 645 return smc_connect_abort(smc, reason_code, 646 ini->cln_first_contact); 647 648 smc_tx_init(smc); 649 650 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 651 /* QP confirmation over RoCE fabric */ 652 reason_code = smc_clnt_conf_first_link(smc); 653 if (reason_code) 654 return smc_connect_abort(smc, reason_code, 655 ini->cln_first_contact); 656 } 657 mutex_unlock(&smc_client_lgr_pending); 658 659 smc_copy_sock_settings_to_clc(smc); 660 smc->connect_nonblock = 0; 661 if (smc->sk.sk_state == SMC_INIT) 662 smc->sk.sk_state = SMC_ACTIVE; 663 664 return 0; 665 } 666 667 /* setup for ISM connection of client */ 668 static int smc_connect_ism(struct smc_sock *smc, 669 struct smc_clc_msg_accept_confirm *aclc, 670 struct smc_init_info *ini) 671 { 672 int rc = 0; 673 674 ini->is_smcd = true; 675 ini->ism_gid = aclc->gid; 676 ini->srv_first_contact = aclc->hdr.flag; 677 678 /* there is only one lgr role for SMC-D; use server lock */ 679 mutex_lock(&smc_server_lgr_pending); 680 rc = smc_conn_create(smc, ini); 681 if (rc) { 682 mutex_unlock(&smc_server_lgr_pending); 683 return rc; 684 } 685 686 /* Create send and receive buffers */ 687 if (smc_buf_create(smc, true)) 688 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 689 ini->cln_first_contact); 690 691 smc_conn_save_peer_info(smc, aclc); 692 smc_close_init(smc); 693 smc_rx_init(smc); 694 smc_tx_init(smc); 695 696 rc = smc_clc_send_confirm(smc); 697 if (rc) 698 return smc_connect_abort(smc, rc, ini->cln_first_contact); 699 mutex_unlock(&smc_server_lgr_pending); 700 701 smc_copy_sock_settings_to_clc(smc); 702 smc->connect_nonblock = 0; 703 if (smc->sk.sk_state == SMC_INIT) 704 smc->sk.sk_state = SMC_ACTIVE; 705 706 return 0; 707 } 708 709 /* perform steps before actually connecting */ 710 static int __smc_connect(struct smc_sock *smc) 711 { 712 bool ism_supported = false, rdma_supported = false; 713 struct smc_clc_msg_accept_confirm aclc; 714 struct smc_init_info ini = {0}; 715 int smc_type; 716 int rc = 0; 717 718 if (smc->use_fallback) 719 return smc_connect_fallback(smc, smc->fallback_rsn); 720 721 /* if peer has not signalled SMC-capability, fall back */ 722 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 723 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 724 725 /* IPSec connections opt out of SMC-R optimizations */ 726 if (using_ipsec(smc)) 727 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 728 729 /* get vlan id from IP device */ 730 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 731 return smc_connect_decline_fallback(smc, 732 SMC_CLC_DECL_GETVLANERR); 733 734 /* check if there is an ism device available */ 735 if (!smc_find_ism_device(smc, &ini) && 736 !smc_connect_ism_vlan_setup(smc, &ini)) { 737 /* ISM is supported for this connection */ 738 ism_supported = true; 739 smc_type = SMC_TYPE_D; 740 } 741 742 /* check if there is a rdma device available */ 743 if (!smc_find_rdma_device(smc, &ini)) { 744 /* RDMA is supported for this connection */ 745 rdma_supported = true; 746 if (ism_supported) 747 smc_type = SMC_TYPE_B; /* both */ 748 else 749 smc_type = SMC_TYPE_R; /* only RDMA */ 750 } 751 752 /* if neither ISM nor RDMA are supported, fallback */ 753 if (!rdma_supported && !ism_supported) 754 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 755 756 /* perform CLC handshake */ 757 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 758 if (rc) { 759 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 760 return smc_connect_decline_fallback(smc, rc); 761 } 762 763 /* depending on previous steps, connect using rdma or ism */ 764 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 765 rc = smc_connect_rdma(smc, &aclc, &ini); 766 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 767 rc = smc_connect_ism(smc, &aclc, &ini); 768 else 769 rc = SMC_CLC_DECL_MODEUNSUPP; 770 if (rc) { 771 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 772 return smc_connect_decline_fallback(smc, rc); 773 } 774 775 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 776 return 0; 777 } 778 779 static void smc_connect_work(struct work_struct *work) 780 { 781 struct smc_sock *smc = container_of(work, struct smc_sock, 782 connect_work); 783 long timeo = smc->sk.sk_sndtimeo; 784 int rc = 0; 785 786 if (!timeo) 787 timeo = MAX_SCHEDULE_TIMEOUT; 788 lock_sock(smc->clcsock->sk); 789 if (smc->clcsock->sk->sk_err) { 790 smc->sk.sk_err = smc->clcsock->sk->sk_err; 791 } else if ((1 << smc->clcsock->sk->sk_state) & 792 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 793 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 794 if ((rc == -EPIPE) && 795 ((1 << smc->clcsock->sk->sk_state) & 796 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 797 rc = 0; 798 } 799 release_sock(smc->clcsock->sk); 800 lock_sock(&smc->sk); 801 if (rc != 0 || smc->sk.sk_err) { 802 smc->sk.sk_state = SMC_CLOSED; 803 if (rc == -EPIPE || rc == -EAGAIN) 804 smc->sk.sk_err = EPIPE; 805 else if (signal_pending(current)) 806 smc->sk.sk_err = -sock_intr_errno(timeo); 807 sock_put(&smc->sk); /* passive closing */ 808 goto out; 809 } 810 811 rc = __smc_connect(smc); 812 if (rc < 0) 813 smc->sk.sk_err = -rc; 814 815 out: 816 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 817 if (smc->sk.sk_err) { 818 smc->sk.sk_state_change(&smc->sk); 819 } else { /* allow polling before and after fallback decision */ 820 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 821 smc->sk.sk_write_space(&smc->sk); 822 } 823 } 824 release_sock(&smc->sk); 825 } 826 827 static int smc_connect(struct socket *sock, struct sockaddr *addr, 828 int alen, int flags) 829 { 830 struct sock *sk = sock->sk; 831 struct smc_sock *smc; 832 int rc = -EINVAL; 833 834 smc = smc_sk(sk); 835 836 /* separate smc parameter checking to be safe */ 837 if (alen < sizeof(addr->sa_family)) 838 goto out_err; 839 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 840 goto out_err; 841 842 lock_sock(sk); 843 switch (sk->sk_state) { 844 default: 845 goto out; 846 case SMC_ACTIVE: 847 rc = -EISCONN; 848 goto out; 849 case SMC_INIT: 850 rc = 0; 851 break; 852 } 853 854 smc_copy_sock_settings_to_clc(smc); 855 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 856 if (smc->connect_nonblock) { 857 rc = -EALREADY; 858 goto out; 859 } 860 rc = kernel_connect(smc->clcsock, addr, alen, flags); 861 if (rc && rc != -EINPROGRESS) 862 goto out; 863 864 sock_hold(&smc->sk); /* sock put in passive closing */ 865 if (smc->use_fallback) 866 goto out; 867 if (flags & O_NONBLOCK) { 868 if (schedule_work(&smc->connect_work)) 869 smc->connect_nonblock = 1; 870 rc = -EINPROGRESS; 871 } else { 872 rc = __smc_connect(smc); 873 if (rc < 0) 874 goto out; 875 else 876 rc = 0; /* success cases including fallback */ 877 } 878 879 out: 880 release_sock(sk); 881 out_err: 882 return rc; 883 } 884 885 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 886 { 887 struct socket *new_clcsock = NULL; 888 struct sock *lsk = &lsmc->sk; 889 struct sock *new_sk; 890 int rc = -EINVAL; 891 892 release_sock(lsk); 893 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 894 if (!new_sk) { 895 rc = -ENOMEM; 896 lsk->sk_err = ENOMEM; 897 *new_smc = NULL; 898 lock_sock(lsk); 899 goto out; 900 } 901 *new_smc = smc_sk(new_sk); 902 903 mutex_lock(&lsmc->clcsock_release_lock); 904 if (lsmc->clcsock) 905 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 906 mutex_unlock(&lsmc->clcsock_release_lock); 907 lock_sock(lsk); 908 if (rc < 0) 909 lsk->sk_err = -rc; 910 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 911 new_sk->sk_prot->unhash(new_sk); 912 if (new_clcsock) 913 sock_release(new_clcsock); 914 new_sk->sk_state = SMC_CLOSED; 915 sock_set_flag(new_sk, SOCK_DEAD); 916 sock_put(new_sk); /* final */ 917 *new_smc = NULL; 918 goto out; 919 } 920 921 (*new_smc)->clcsock = new_clcsock; 922 out: 923 return rc; 924 } 925 926 /* add a just created sock to the accept queue of the listen sock as 927 * candidate for a following socket accept call from user space 928 */ 929 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 930 { 931 struct smc_sock *par = smc_sk(parent); 932 933 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 934 spin_lock(&par->accept_q_lock); 935 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 936 spin_unlock(&par->accept_q_lock); 937 sk_acceptq_added(parent); 938 } 939 940 /* remove a socket from the accept queue of its parental listening socket */ 941 static void smc_accept_unlink(struct sock *sk) 942 { 943 struct smc_sock *par = smc_sk(sk)->listen_smc; 944 945 spin_lock(&par->accept_q_lock); 946 list_del_init(&smc_sk(sk)->accept_q); 947 spin_unlock(&par->accept_q_lock); 948 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 949 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 950 } 951 952 /* remove a sock from the accept queue to bind it to a new socket created 953 * for a socket accept call from user space 954 */ 955 struct sock *smc_accept_dequeue(struct sock *parent, 956 struct socket *new_sock) 957 { 958 struct smc_sock *isk, *n; 959 struct sock *new_sk; 960 961 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 962 new_sk = (struct sock *)isk; 963 964 smc_accept_unlink(new_sk); 965 if (new_sk->sk_state == SMC_CLOSED) { 966 new_sk->sk_prot->unhash(new_sk); 967 if (isk->clcsock) { 968 sock_release(isk->clcsock); 969 isk->clcsock = NULL; 970 } 971 sock_put(new_sk); /* final */ 972 continue; 973 } 974 if (new_sock) { 975 sock_graft(new_sk, new_sock); 976 if (isk->use_fallback) { 977 smc_sk(new_sk)->clcsock->file = new_sock->file; 978 isk->clcsock->file->private_data = isk->clcsock; 979 } 980 } 981 return new_sk; 982 } 983 return NULL; 984 } 985 986 /* clean up for a created but never accepted sock */ 987 void smc_close_non_accepted(struct sock *sk) 988 { 989 struct smc_sock *smc = smc_sk(sk); 990 991 sock_hold(sk); /* sock_put below */ 992 lock_sock(sk); 993 if (!sk->sk_lingertime) 994 /* wait for peer closing */ 995 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 996 __smc_release(smc); 997 release_sock(sk); 998 sock_put(sk); /* sock_hold above */ 999 sock_put(sk); /* final sock_put */ 1000 } 1001 1002 static int smc_serv_conf_first_link(struct smc_sock *smc) 1003 { 1004 struct net *net = sock_net(smc->clcsock->sk); 1005 struct smc_link_group *lgr = smc->conn.lgr; 1006 struct smc_link *link; 1007 int rest; 1008 int rc; 1009 1010 link = &lgr->lnk[SMC_SINGLE_LINK]; 1011 1012 if (smc_reg_rmb(link, smc->conn.rmb_desc, false)) 1013 return SMC_CLC_DECL_ERR_REGRMB; 1014 1015 /* send CONFIRM LINK request to client over the RoCE fabric */ 1016 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1017 if (rc < 0) 1018 return SMC_CLC_DECL_TIMEOUT_CL; 1019 1020 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1021 rest = wait_for_completion_interruptible_timeout( 1022 &link->llc_confirm_resp, 1023 SMC_LLC_WAIT_FIRST_TIME); 1024 if (rest <= 0) { 1025 struct smc_clc_msg_decline dclc; 1026 1027 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1028 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1029 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1030 } 1031 1032 if (link->llc_confirm_resp_rc) 1033 return SMC_CLC_DECL_RMBE_EC; 1034 1035 /* send ADD LINK request to client over the RoCE fabric */ 1036 rc = smc_llc_send_add_link(link, 1037 link->smcibdev->mac[link->ibport - 1], 1038 link->gid, SMC_LLC_REQ); 1039 if (rc < 0) 1040 return SMC_CLC_DECL_TIMEOUT_AL; 1041 1042 /* receive ADD LINK response from client over the RoCE fabric */ 1043 rest = wait_for_completion_interruptible_timeout(&link->llc_add_resp, 1044 SMC_LLC_WAIT_TIME); 1045 if (rest <= 0) { 1046 struct smc_clc_msg_decline dclc; 1047 1048 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1049 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1050 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_AL : rc; 1051 } 1052 1053 smc_llc_link_active(link, net->ipv4.sysctl_tcp_keepalive_time); 1054 1055 return 0; 1056 } 1057 1058 /* listen worker: finish */ 1059 static void smc_listen_out(struct smc_sock *new_smc) 1060 { 1061 struct smc_sock *lsmc = new_smc->listen_smc; 1062 struct sock *newsmcsk = &new_smc->sk; 1063 1064 if (lsmc->sk.sk_state == SMC_LISTEN) { 1065 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1066 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1067 release_sock(&lsmc->sk); 1068 } else { /* no longer listening */ 1069 smc_close_non_accepted(newsmcsk); 1070 } 1071 1072 /* Wake up accept */ 1073 lsmc->sk.sk_data_ready(&lsmc->sk); 1074 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1075 } 1076 1077 /* listen worker: finish in state connected */ 1078 static void smc_listen_out_connected(struct smc_sock *new_smc) 1079 { 1080 struct sock *newsmcsk = &new_smc->sk; 1081 1082 sk_refcnt_debug_inc(newsmcsk); 1083 if (newsmcsk->sk_state == SMC_INIT) 1084 newsmcsk->sk_state = SMC_ACTIVE; 1085 1086 smc_listen_out(new_smc); 1087 } 1088 1089 /* listen worker: finish in error state */ 1090 static void smc_listen_out_err(struct smc_sock *new_smc) 1091 { 1092 struct sock *newsmcsk = &new_smc->sk; 1093 1094 if (newsmcsk->sk_state == SMC_INIT) 1095 sock_put(&new_smc->sk); /* passive closing */ 1096 newsmcsk->sk_state = SMC_CLOSED; 1097 1098 smc_listen_out(new_smc); 1099 } 1100 1101 /* listen worker: decline and fall back if possible */ 1102 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1103 int local_contact) 1104 { 1105 /* RDMA setup failed, switch back to TCP */ 1106 if (local_contact == SMC_FIRST_CONTACT) 1107 smc_lgr_cleanup_early(&new_smc->conn); 1108 else 1109 smc_conn_free(&new_smc->conn); 1110 if (reason_code < 0) { /* error, no fallback possible */ 1111 smc_listen_out_err(new_smc); 1112 return; 1113 } 1114 smc_switch_to_fallback(new_smc); 1115 new_smc->fallback_rsn = reason_code; 1116 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1117 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1118 smc_listen_out_err(new_smc); 1119 return; 1120 } 1121 } 1122 smc_listen_out_connected(new_smc); 1123 } 1124 1125 /* listen worker: check prefixes */ 1126 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1127 struct smc_clc_msg_proposal *pclc) 1128 { 1129 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1130 struct socket *newclcsock = new_smc->clcsock; 1131 1132 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1133 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1134 return SMC_CLC_DECL_DIFFPREFIX; 1135 1136 return 0; 1137 } 1138 1139 /* listen worker: initialize connection and buffers */ 1140 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1141 struct smc_init_info *ini) 1142 { 1143 int rc; 1144 1145 /* allocate connection / link group */ 1146 rc = smc_conn_create(new_smc, ini); 1147 if (rc) 1148 return rc; 1149 1150 /* create send buffer and rmb */ 1151 if (smc_buf_create(new_smc, false)) 1152 return SMC_CLC_DECL_MEM; 1153 1154 return 0; 1155 } 1156 1157 /* listen worker: initialize connection and buffers for SMC-D */ 1158 static int smc_listen_ism_init(struct smc_sock *new_smc, 1159 struct smc_clc_msg_proposal *pclc, 1160 struct smc_init_info *ini) 1161 { 1162 struct smc_clc_msg_smcd *pclc_smcd; 1163 int rc; 1164 1165 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1166 ini->ism_gid = pclc_smcd->gid; 1167 rc = smc_conn_create(new_smc, ini); 1168 if (rc) 1169 return rc; 1170 1171 /* Check if peer can be reached via ISM device */ 1172 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1173 new_smc->conn.lgr->vlan_id, 1174 new_smc->conn.lgr->smcd)) { 1175 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1176 smc_lgr_cleanup_early(&new_smc->conn); 1177 else 1178 smc_conn_free(&new_smc->conn); 1179 return SMC_CLC_DECL_SMCDNOTALK; 1180 } 1181 1182 /* Create send and receive buffers */ 1183 if (smc_buf_create(new_smc, true)) { 1184 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1185 smc_lgr_cleanup_early(&new_smc->conn); 1186 else 1187 smc_conn_free(&new_smc->conn); 1188 return SMC_CLC_DECL_MEM; 1189 } 1190 1191 return 0; 1192 } 1193 1194 /* listen worker: register buffers */ 1195 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1196 { 1197 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1198 1199 if (local_contact != SMC_FIRST_CONTACT) { 1200 if (smc_reg_rmb(link, new_smc->conn.rmb_desc, true)) 1201 return SMC_CLC_DECL_ERR_REGRMB; 1202 } 1203 smc_rmb_sync_sg_for_device(&new_smc->conn); 1204 1205 return 0; 1206 } 1207 1208 /* listen worker: finish RDMA setup */ 1209 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1210 struct smc_clc_msg_accept_confirm *cclc, 1211 int local_contact) 1212 { 1213 struct smc_link *link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK]; 1214 int reason_code = 0; 1215 1216 if (local_contact == SMC_FIRST_CONTACT) 1217 smc_link_save_peer_info(link, cclc); 1218 1219 if (smc_rmb_rtoken_handling(&new_smc->conn, cclc)) { 1220 reason_code = SMC_CLC_DECL_ERR_RTOK; 1221 goto decline; 1222 } 1223 1224 if (local_contact == SMC_FIRST_CONTACT) { 1225 if (smc_ib_ready_link(link)) { 1226 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1227 goto decline; 1228 } 1229 /* QP confirmation over RoCE fabric */ 1230 reason_code = smc_serv_conf_first_link(new_smc); 1231 if (reason_code) 1232 goto decline; 1233 } 1234 return 0; 1235 1236 decline: 1237 smc_listen_decline(new_smc, reason_code, local_contact); 1238 return reason_code; 1239 } 1240 1241 /* setup for RDMA connection of server */ 1242 static void smc_listen_work(struct work_struct *work) 1243 { 1244 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1245 smc_listen_work); 1246 struct socket *newclcsock = new_smc->clcsock; 1247 struct smc_clc_msg_accept_confirm cclc; 1248 struct smc_clc_msg_proposal *pclc; 1249 struct smc_init_info ini = {0}; 1250 bool ism_supported = false; 1251 u8 buf[SMC_CLC_MAX_LEN]; 1252 int rc = 0; 1253 1254 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1255 return smc_listen_out_err(new_smc); 1256 1257 if (new_smc->use_fallback) { 1258 smc_listen_out_connected(new_smc); 1259 return; 1260 } 1261 1262 /* check if peer is smc capable */ 1263 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1264 smc_switch_to_fallback(new_smc); 1265 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1266 smc_listen_out_connected(new_smc); 1267 return; 1268 } 1269 1270 /* do inband token exchange - 1271 * wait for and receive SMC Proposal CLC message 1272 */ 1273 pclc = (struct smc_clc_msg_proposal *)&buf; 1274 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1275 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1276 if (rc) 1277 goto out_decl; 1278 1279 /* IPSec connections opt out of SMC-R optimizations */ 1280 if (using_ipsec(new_smc)) { 1281 rc = SMC_CLC_DECL_IPSEC; 1282 goto out_decl; 1283 } 1284 1285 /* check for matching IP prefix and subnet length */ 1286 rc = smc_listen_prfx_check(new_smc, pclc); 1287 if (rc) 1288 goto out_decl; 1289 1290 /* get vlan id from IP device */ 1291 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1292 rc = SMC_CLC_DECL_GETVLANERR; 1293 goto out_decl; 1294 } 1295 1296 mutex_lock(&smc_server_lgr_pending); 1297 smc_close_init(new_smc); 1298 smc_rx_init(new_smc); 1299 smc_tx_init(new_smc); 1300 1301 /* check if ISM is available */ 1302 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1303 ini.is_smcd = true; /* prepare ISM check */ 1304 rc = smc_find_ism_device(new_smc, &ini); 1305 if (!rc) 1306 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1307 if (!rc) 1308 ism_supported = true; 1309 else if (pclc->hdr.path == SMC_TYPE_D) 1310 goto out_unlock; /* skip RDMA and decline */ 1311 } 1312 1313 /* check if RDMA is available */ 1314 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1315 /* prepare RDMA check */ 1316 ini.is_smcd = false; 1317 ini.ism_dev = NULL; 1318 ini.ib_lcl = &pclc->lcl; 1319 rc = smc_find_rdma_device(new_smc, &ini); 1320 if (rc) { 1321 /* no RDMA device found */ 1322 if (pclc->hdr.path == SMC_TYPE_B) 1323 /* neither ISM nor RDMA device found */ 1324 rc = SMC_CLC_DECL_NOSMCDEV; 1325 goto out_unlock; 1326 } 1327 rc = smc_listen_rdma_init(new_smc, &ini); 1328 if (rc) 1329 goto out_unlock; 1330 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1331 if (rc) 1332 goto out_unlock; 1333 } 1334 1335 /* send SMC Accept CLC message */ 1336 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1337 if (rc) 1338 goto out_unlock; 1339 1340 /* SMC-D does not need this lock any more */ 1341 if (ism_supported) 1342 mutex_unlock(&smc_server_lgr_pending); 1343 1344 /* receive SMC Confirm CLC message */ 1345 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1346 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1347 if (rc) { 1348 if (!ism_supported) 1349 goto out_unlock; 1350 goto out_decl; 1351 } 1352 1353 /* finish worker */ 1354 if (!ism_supported) { 1355 rc = smc_listen_rdma_finish(new_smc, &cclc, 1356 ini.cln_first_contact); 1357 mutex_unlock(&smc_server_lgr_pending); 1358 if (rc) 1359 return; 1360 } 1361 smc_conn_save_peer_info(new_smc, &cclc); 1362 smc_listen_out_connected(new_smc); 1363 return; 1364 1365 out_unlock: 1366 mutex_unlock(&smc_server_lgr_pending); 1367 out_decl: 1368 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1369 } 1370 1371 static void smc_tcp_listen_work(struct work_struct *work) 1372 { 1373 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1374 tcp_listen_work); 1375 struct sock *lsk = &lsmc->sk; 1376 struct smc_sock *new_smc; 1377 int rc = 0; 1378 1379 lock_sock(lsk); 1380 while (lsk->sk_state == SMC_LISTEN) { 1381 rc = smc_clcsock_accept(lsmc, &new_smc); 1382 if (rc) 1383 goto out; 1384 if (!new_smc) 1385 continue; 1386 1387 new_smc->listen_smc = lsmc; 1388 new_smc->use_fallback = lsmc->use_fallback; 1389 new_smc->fallback_rsn = lsmc->fallback_rsn; 1390 sock_hold(lsk); /* sock_put in smc_listen_work */ 1391 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1392 smc_copy_sock_settings_to_smc(new_smc); 1393 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1394 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1395 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1396 if (!schedule_work(&new_smc->smc_listen_work)) 1397 sock_put(&new_smc->sk); 1398 } 1399 1400 out: 1401 release_sock(lsk); 1402 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1403 } 1404 1405 static int smc_listen(struct socket *sock, int backlog) 1406 { 1407 struct sock *sk = sock->sk; 1408 struct smc_sock *smc; 1409 int rc; 1410 1411 smc = smc_sk(sk); 1412 lock_sock(sk); 1413 1414 rc = -EINVAL; 1415 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1416 smc->connect_nonblock) 1417 goto out; 1418 1419 rc = 0; 1420 if (sk->sk_state == SMC_LISTEN) { 1421 sk->sk_max_ack_backlog = backlog; 1422 goto out; 1423 } 1424 /* some socket options are handled in core, so we could not apply 1425 * them to the clc socket -- copy smc socket options to clc socket 1426 */ 1427 smc_copy_sock_settings_to_clc(smc); 1428 if (!smc->use_fallback) 1429 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1430 1431 rc = kernel_listen(smc->clcsock, backlog); 1432 if (rc) 1433 goto out; 1434 sk->sk_max_ack_backlog = backlog; 1435 sk->sk_ack_backlog = 0; 1436 sk->sk_state = SMC_LISTEN; 1437 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1438 if (!schedule_work(&smc->tcp_listen_work)) 1439 sock_put(sk); 1440 1441 out: 1442 release_sock(sk); 1443 return rc; 1444 } 1445 1446 static int smc_accept(struct socket *sock, struct socket *new_sock, 1447 int flags, bool kern) 1448 { 1449 struct sock *sk = sock->sk, *nsk; 1450 DECLARE_WAITQUEUE(wait, current); 1451 struct smc_sock *lsmc; 1452 long timeo; 1453 int rc = 0; 1454 1455 lsmc = smc_sk(sk); 1456 sock_hold(sk); /* sock_put below */ 1457 lock_sock(sk); 1458 1459 if (lsmc->sk.sk_state != SMC_LISTEN) { 1460 rc = -EINVAL; 1461 release_sock(sk); 1462 goto out; 1463 } 1464 1465 /* Wait for an incoming connection */ 1466 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1467 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1468 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1469 set_current_state(TASK_INTERRUPTIBLE); 1470 if (!timeo) { 1471 rc = -EAGAIN; 1472 break; 1473 } 1474 release_sock(sk); 1475 timeo = schedule_timeout(timeo); 1476 /* wakeup by sk_data_ready in smc_listen_work() */ 1477 sched_annotate_sleep(); 1478 lock_sock(sk); 1479 if (signal_pending(current)) { 1480 rc = sock_intr_errno(timeo); 1481 break; 1482 } 1483 } 1484 set_current_state(TASK_RUNNING); 1485 remove_wait_queue(sk_sleep(sk), &wait); 1486 1487 if (!rc) 1488 rc = sock_error(nsk); 1489 release_sock(sk); 1490 if (rc) 1491 goto out; 1492 1493 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1494 /* wait till data arrives on the socket */ 1495 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1496 MSEC_PER_SEC); 1497 if (smc_sk(nsk)->use_fallback) { 1498 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1499 1500 lock_sock(clcsk); 1501 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1502 sk_wait_data(clcsk, &timeo, NULL); 1503 release_sock(clcsk); 1504 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1505 lock_sock(nsk); 1506 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1507 release_sock(nsk); 1508 } 1509 } 1510 1511 out: 1512 sock_put(sk); /* sock_hold above */ 1513 return rc; 1514 } 1515 1516 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1517 int peer) 1518 { 1519 struct smc_sock *smc; 1520 1521 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1522 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1523 return -ENOTCONN; 1524 1525 smc = smc_sk(sock->sk); 1526 1527 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1528 } 1529 1530 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1531 { 1532 struct sock *sk = sock->sk; 1533 struct smc_sock *smc; 1534 int rc = -EPIPE; 1535 1536 smc = smc_sk(sk); 1537 lock_sock(sk); 1538 if ((sk->sk_state != SMC_ACTIVE) && 1539 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1540 (sk->sk_state != SMC_INIT)) 1541 goto out; 1542 1543 if (msg->msg_flags & MSG_FASTOPEN) { 1544 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1545 smc_switch_to_fallback(smc); 1546 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1547 } else { 1548 rc = -EINVAL; 1549 goto out; 1550 } 1551 } 1552 1553 if (smc->use_fallback) 1554 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1555 else 1556 rc = smc_tx_sendmsg(smc, msg, len); 1557 out: 1558 release_sock(sk); 1559 return rc; 1560 } 1561 1562 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1563 int flags) 1564 { 1565 struct sock *sk = sock->sk; 1566 struct smc_sock *smc; 1567 int rc = -ENOTCONN; 1568 1569 smc = smc_sk(sk); 1570 lock_sock(sk); 1571 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1572 /* socket was connected before, no more data to read */ 1573 rc = 0; 1574 goto out; 1575 } 1576 if ((sk->sk_state == SMC_INIT) || 1577 (sk->sk_state == SMC_LISTEN) || 1578 (sk->sk_state == SMC_CLOSED)) 1579 goto out; 1580 1581 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1582 rc = 0; 1583 goto out; 1584 } 1585 1586 if (smc->use_fallback) { 1587 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1588 } else { 1589 msg->msg_namelen = 0; 1590 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1591 } 1592 1593 out: 1594 release_sock(sk); 1595 return rc; 1596 } 1597 1598 static __poll_t smc_accept_poll(struct sock *parent) 1599 { 1600 struct smc_sock *isk = smc_sk(parent); 1601 __poll_t mask = 0; 1602 1603 spin_lock(&isk->accept_q_lock); 1604 if (!list_empty(&isk->accept_q)) 1605 mask = EPOLLIN | EPOLLRDNORM; 1606 spin_unlock(&isk->accept_q_lock); 1607 1608 return mask; 1609 } 1610 1611 static __poll_t smc_poll(struct file *file, struct socket *sock, 1612 poll_table *wait) 1613 { 1614 struct sock *sk = sock->sk; 1615 struct smc_sock *smc; 1616 __poll_t mask = 0; 1617 1618 if (!sk) 1619 return EPOLLNVAL; 1620 1621 smc = smc_sk(sock->sk); 1622 if (smc->use_fallback) { 1623 /* delegate to CLC child sock */ 1624 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1625 sk->sk_err = smc->clcsock->sk->sk_err; 1626 } else { 1627 if (sk->sk_state != SMC_CLOSED) 1628 sock_poll_wait(file, sock, wait); 1629 if (sk->sk_err) 1630 mask |= EPOLLERR; 1631 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1632 (sk->sk_state == SMC_CLOSED)) 1633 mask |= EPOLLHUP; 1634 if (sk->sk_state == SMC_LISTEN) { 1635 /* woken up by sk_data_ready in smc_listen_work() */ 1636 mask |= smc_accept_poll(sk); 1637 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1638 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1639 wait); 1640 sk->sk_err = smc->clcsock->sk->sk_err; 1641 } else { 1642 if ((sk->sk_state != SMC_INIT && 1643 atomic_read(&smc->conn.sndbuf_space)) || 1644 sk->sk_shutdown & SEND_SHUTDOWN) { 1645 mask |= EPOLLOUT | EPOLLWRNORM; 1646 } else { 1647 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1648 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1649 } 1650 if (atomic_read(&smc->conn.bytes_to_rcv)) 1651 mask |= EPOLLIN | EPOLLRDNORM; 1652 if (sk->sk_shutdown & RCV_SHUTDOWN) 1653 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1654 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1655 mask |= EPOLLIN; 1656 if (smc->conn.urg_state == SMC_URG_VALID) 1657 mask |= EPOLLPRI; 1658 } 1659 } 1660 1661 return mask; 1662 } 1663 1664 static int smc_shutdown(struct socket *sock, int how) 1665 { 1666 struct sock *sk = sock->sk; 1667 struct smc_sock *smc; 1668 int rc = -EINVAL; 1669 int rc1 = 0; 1670 1671 smc = smc_sk(sk); 1672 1673 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1674 return rc; 1675 1676 lock_sock(sk); 1677 1678 rc = -ENOTCONN; 1679 if ((sk->sk_state != SMC_ACTIVE) && 1680 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1681 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1682 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1683 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1684 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1685 goto out; 1686 if (smc->use_fallback) { 1687 rc = kernel_sock_shutdown(smc->clcsock, how); 1688 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1689 if (sk->sk_shutdown == SHUTDOWN_MASK) 1690 sk->sk_state = SMC_CLOSED; 1691 goto out; 1692 } 1693 switch (how) { 1694 case SHUT_RDWR: /* shutdown in both directions */ 1695 rc = smc_close_active(smc); 1696 break; 1697 case SHUT_WR: 1698 rc = smc_close_shutdown_write(smc); 1699 break; 1700 case SHUT_RD: 1701 rc = 0; 1702 /* nothing more to do because peer is not involved */ 1703 break; 1704 } 1705 if (smc->clcsock) 1706 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1707 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1708 sk->sk_shutdown |= how + 1; 1709 1710 out: 1711 release_sock(sk); 1712 return rc ? rc : rc1; 1713 } 1714 1715 static int smc_setsockopt(struct socket *sock, int level, int optname, 1716 char __user *optval, unsigned int optlen) 1717 { 1718 struct sock *sk = sock->sk; 1719 struct smc_sock *smc; 1720 int val, rc; 1721 1722 smc = smc_sk(sk); 1723 1724 /* generic setsockopts reaching us here always apply to the 1725 * CLC socket 1726 */ 1727 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1728 optval, optlen); 1729 if (smc->clcsock->sk->sk_err) { 1730 sk->sk_err = smc->clcsock->sk->sk_err; 1731 sk->sk_error_report(sk); 1732 } 1733 1734 if (optlen < sizeof(int)) 1735 return -EINVAL; 1736 if (get_user(val, (int __user *)optval)) 1737 return -EFAULT; 1738 1739 lock_sock(sk); 1740 if (rc || smc->use_fallback) 1741 goto out; 1742 switch (optname) { 1743 case TCP_ULP: 1744 case TCP_FASTOPEN: 1745 case TCP_FASTOPEN_CONNECT: 1746 case TCP_FASTOPEN_KEY: 1747 case TCP_FASTOPEN_NO_COOKIE: 1748 /* option not supported by SMC */ 1749 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1750 smc_switch_to_fallback(smc); 1751 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1752 } else { 1753 rc = -EINVAL; 1754 } 1755 break; 1756 case TCP_NODELAY: 1757 if (sk->sk_state != SMC_INIT && 1758 sk->sk_state != SMC_LISTEN && 1759 sk->sk_state != SMC_CLOSED) { 1760 if (val) 1761 mod_delayed_work(system_wq, &smc->conn.tx_work, 1762 0); 1763 } 1764 break; 1765 case TCP_CORK: 1766 if (sk->sk_state != SMC_INIT && 1767 sk->sk_state != SMC_LISTEN && 1768 sk->sk_state != SMC_CLOSED) { 1769 if (!val) 1770 mod_delayed_work(system_wq, &smc->conn.tx_work, 1771 0); 1772 } 1773 break; 1774 case TCP_DEFER_ACCEPT: 1775 smc->sockopt_defer_accept = val; 1776 break; 1777 default: 1778 break; 1779 } 1780 out: 1781 release_sock(sk); 1782 1783 return rc; 1784 } 1785 1786 static int smc_getsockopt(struct socket *sock, int level, int optname, 1787 char __user *optval, int __user *optlen) 1788 { 1789 struct smc_sock *smc; 1790 1791 smc = smc_sk(sock->sk); 1792 /* socket options apply to the CLC socket */ 1793 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1794 optval, optlen); 1795 } 1796 1797 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1798 unsigned long arg) 1799 { 1800 union smc_host_cursor cons, urg; 1801 struct smc_connection *conn; 1802 struct smc_sock *smc; 1803 int answ; 1804 1805 smc = smc_sk(sock->sk); 1806 conn = &smc->conn; 1807 lock_sock(&smc->sk); 1808 if (smc->use_fallback) { 1809 if (!smc->clcsock) { 1810 release_sock(&smc->sk); 1811 return -EBADF; 1812 } 1813 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1814 release_sock(&smc->sk); 1815 return answ; 1816 } 1817 switch (cmd) { 1818 case SIOCINQ: /* same as FIONREAD */ 1819 if (smc->sk.sk_state == SMC_LISTEN) { 1820 release_sock(&smc->sk); 1821 return -EINVAL; 1822 } 1823 if (smc->sk.sk_state == SMC_INIT || 1824 smc->sk.sk_state == SMC_CLOSED) 1825 answ = 0; 1826 else 1827 answ = atomic_read(&smc->conn.bytes_to_rcv); 1828 break; 1829 case SIOCOUTQ: 1830 /* output queue size (not send + not acked) */ 1831 if (smc->sk.sk_state == SMC_LISTEN) { 1832 release_sock(&smc->sk); 1833 return -EINVAL; 1834 } 1835 if (smc->sk.sk_state == SMC_INIT || 1836 smc->sk.sk_state == SMC_CLOSED) 1837 answ = 0; 1838 else 1839 answ = smc->conn.sndbuf_desc->len - 1840 atomic_read(&smc->conn.sndbuf_space); 1841 break; 1842 case SIOCOUTQNSD: 1843 /* output queue size (not send only) */ 1844 if (smc->sk.sk_state == SMC_LISTEN) { 1845 release_sock(&smc->sk); 1846 return -EINVAL; 1847 } 1848 if (smc->sk.sk_state == SMC_INIT || 1849 smc->sk.sk_state == SMC_CLOSED) 1850 answ = 0; 1851 else 1852 answ = smc_tx_prepared_sends(&smc->conn); 1853 break; 1854 case SIOCATMARK: 1855 if (smc->sk.sk_state == SMC_LISTEN) { 1856 release_sock(&smc->sk); 1857 return -EINVAL; 1858 } 1859 if (smc->sk.sk_state == SMC_INIT || 1860 smc->sk.sk_state == SMC_CLOSED) { 1861 answ = 0; 1862 } else { 1863 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1864 smc_curs_copy(&urg, &conn->urg_curs, conn); 1865 answ = smc_curs_diff(conn->rmb_desc->len, 1866 &cons, &urg) == 1; 1867 } 1868 break; 1869 default: 1870 release_sock(&smc->sk); 1871 return -ENOIOCTLCMD; 1872 } 1873 release_sock(&smc->sk); 1874 1875 return put_user(answ, (int __user *)arg); 1876 } 1877 1878 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1879 int offset, size_t size, int flags) 1880 { 1881 struct sock *sk = sock->sk; 1882 struct smc_sock *smc; 1883 int rc = -EPIPE; 1884 1885 smc = smc_sk(sk); 1886 lock_sock(sk); 1887 if (sk->sk_state != SMC_ACTIVE) { 1888 release_sock(sk); 1889 goto out; 1890 } 1891 release_sock(sk); 1892 if (smc->use_fallback) 1893 rc = kernel_sendpage(smc->clcsock, page, offset, 1894 size, flags); 1895 else 1896 rc = sock_no_sendpage(sock, page, offset, size, flags); 1897 1898 out: 1899 return rc; 1900 } 1901 1902 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1903 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1904 * updates till whenever a respective page has been fully processed. 1905 * Note that subsequent recv() calls have to wait till all splice() processing 1906 * completed. 1907 */ 1908 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1909 struct pipe_inode_info *pipe, size_t len, 1910 unsigned int flags) 1911 { 1912 struct sock *sk = sock->sk; 1913 struct smc_sock *smc; 1914 int rc = -ENOTCONN; 1915 1916 smc = smc_sk(sk); 1917 lock_sock(sk); 1918 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1919 /* socket was connected before, no more data to read */ 1920 rc = 0; 1921 goto out; 1922 } 1923 if (sk->sk_state == SMC_INIT || 1924 sk->sk_state == SMC_LISTEN || 1925 sk->sk_state == SMC_CLOSED) 1926 goto out; 1927 1928 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1929 rc = 0; 1930 goto out; 1931 } 1932 1933 if (smc->use_fallback) { 1934 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1935 pipe, len, flags); 1936 } else { 1937 if (*ppos) { 1938 rc = -ESPIPE; 1939 goto out; 1940 } 1941 if (flags & SPLICE_F_NONBLOCK) 1942 flags = MSG_DONTWAIT; 1943 else 1944 flags = 0; 1945 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1946 } 1947 out: 1948 release_sock(sk); 1949 1950 return rc; 1951 } 1952 1953 /* must look like tcp */ 1954 static const struct proto_ops smc_sock_ops = { 1955 .family = PF_SMC, 1956 .owner = THIS_MODULE, 1957 .release = smc_release, 1958 .bind = smc_bind, 1959 .connect = smc_connect, 1960 .socketpair = sock_no_socketpair, 1961 .accept = smc_accept, 1962 .getname = smc_getname, 1963 .poll = smc_poll, 1964 .ioctl = smc_ioctl, 1965 .listen = smc_listen, 1966 .shutdown = smc_shutdown, 1967 .setsockopt = smc_setsockopt, 1968 .getsockopt = smc_getsockopt, 1969 .sendmsg = smc_sendmsg, 1970 .recvmsg = smc_recvmsg, 1971 .mmap = sock_no_mmap, 1972 .sendpage = smc_sendpage, 1973 .splice_read = smc_splice_read, 1974 }; 1975 1976 static int smc_create(struct net *net, struct socket *sock, int protocol, 1977 int kern) 1978 { 1979 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 1980 struct smc_sock *smc; 1981 struct sock *sk; 1982 int rc; 1983 1984 rc = -ESOCKTNOSUPPORT; 1985 if (sock->type != SOCK_STREAM) 1986 goto out; 1987 1988 rc = -EPROTONOSUPPORT; 1989 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 1990 goto out; 1991 1992 rc = -ENOBUFS; 1993 sock->ops = &smc_sock_ops; 1994 sk = smc_sock_alloc(net, sock, protocol); 1995 if (!sk) 1996 goto out; 1997 1998 /* create internal TCP socket for CLC handshake and fallback */ 1999 smc = smc_sk(sk); 2000 smc->use_fallback = false; /* assume rdma capability first */ 2001 smc->fallback_rsn = 0; 2002 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 2003 &smc->clcsock); 2004 if (rc) { 2005 sk_common_release(sk); 2006 goto out; 2007 } 2008 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2009 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2010 2011 out: 2012 return rc; 2013 } 2014 2015 static const struct net_proto_family smc_sock_family_ops = { 2016 .family = PF_SMC, 2017 .owner = THIS_MODULE, 2018 .create = smc_create, 2019 }; 2020 2021 unsigned int smc_net_id; 2022 2023 static __net_init int smc_net_init(struct net *net) 2024 { 2025 return smc_pnet_net_init(net); 2026 } 2027 2028 static void __net_exit smc_net_exit(struct net *net) 2029 { 2030 smc_pnet_net_exit(net); 2031 } 2032 2033 static struct pernet_operations smc_net_ops = { 2034 .init = smc_net_init, 2035 .exit = smc_net_exit, 2036 .id = &smc_net_id, 2037 .size = sizeof(struct smc_net), 2038 }; 2039 2040 static int __init smc_init(void) 2041 { 2042 int rc; 2043 2044 rc = register_pernet_subsys(&smc_net_ops); 2045 if (rc) 2046 return rc; 2047 2048 rc = smc_pnet_init(); 2049 if (rc) 2050 goto out_pernet_subsys; 2051 2052 rc = smc_core_init(); 2053 if (rc) { 2054 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2055 goto out_pnet; 2056 } 2057 2058 rc = smc_llc_init(); 2059 if (rc) { 2060 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2061 goto out_core; 2062 } 2063 2064 rc = smc_cdc_init(); 2065 if (rc) { 2066 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2067 goto out_core; 2068 } 2069 2070 rc = proto_register(&smc_proto, 1); 2071 if (rc) { 2072 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2073 goto out_core; 2074 } 2075 2076 rc = proto_register(&smc_proto6, 1); 2077 if (rc) { 2078 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2079 goto out_proto; 2080 } 2081 2082 rc = sock_register(&smc_sock_family_ops); 2083 if (rc) { 2084 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2085 goto out_proto6; 2086 } 2087 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2088 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2089 2090 rc = smc_ib_register_client(); 2091 if (rc) { 2092 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2093 goto out_sock; 2094 } 2095 2096 static_branch_enable(&tcp_have_smc); 2097 return 0; 2098 2099 out_sock: 2100 sock_unregister(PF_SMC); 2101 out_proto6: 2102 proto_unregister(&smc_proto6); 2103 out_proto: 2104 proto_unregister(&smc_proto); 2105 out_core: 2106 smc_core_exit(); 2107 out_pnet: 2108 smc_pnet_exit(); 2109 out_pernet_subsys: 2110 unregister_pernet_subsys(&smc_net_ops); 2111 2112 return rc; 2113 } 2114 2115 static void __exit smc_exit(void) 2116 { 2117 static_branch_disable(&tcp_have_smc); 2118 sock_unregister(PF_SMC); 2119 smc_core_exit(); 2120 smc_ib_unregister_client(); 2121 proto_unregister(&smc_proto6); 2122 proto_unregister(&smc_proto); 2123 smc_pnet_exit(); 2124 unregister_pernet_subsys(&smc_net_ops); 2125 rcu_barrier(); 2126 } 2127 2128 module_init(smc_init); 2129 module_exit(smc_exit); 2130 2131 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2132 MODULE_DESCRIPTION("smc socket address family"); 2133 MODULE_LICENSE("GPL"); 2134 MODULE_ALIAS_NETPROTO(PF_SMC); 2135