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