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 #include <linux/ctype.h> 30 31 #include <net/sock.h> 32 #include <net/tcp.h> 33 #include <net/smc.h> 34 #include <asm/ioctls.h> 35 36 #include <net/net_namespace.h> 37 #include <net/netns/generic.h> 38 #include "smc_netns.h" 39 40 #include "smc.h" 41 #include "smc_clc.h" 42 #include "smc_llc.h" 43 #include "smc_cdc.h" 44 #include "smc_core.h" 45 #include "smc_ib.h" 46 #include "smc_ism.h" 47 #include "smc_pnet.h" 48 #include "smc_tx.h" 49 #include "smc_rx.h" 50 #include "smc_close.h" 51 52 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 53 * creation on server 54 */ 55 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 56 * creation on client 57 */ 58 59 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 60 struct workqueue_struct *smc_close_wq; /* wq for close work */ 61 62 static void smc_tcp_listen_work(struct work_struct *); 63 static void smc_connect_work(struct work_struct *); 64 65 static void smc_set_keepalive(struct sock *sk, int val) 66 { 67 struct smc_sock *smc = smc_sk(sk); 68 69 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 70 } 71 72 static struct smc_hashinfo smc_v4_hashinfo = { 73 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 74 }; 75 76 static struct smc_hashinfo smc_v6_hashinfo = { 77 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 78 }; 79 80 int smc_hash_sk(struct sock *sk) 81 { 82 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 83 struct hlist_head *head; 84 85 head = &h->ht; 86 87 write_lock_bh(&h->lock); 88 sk_add_node(sk, head); 89 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 90 write_unlock_bh(&h->lock); 91 92 return 0; 93 } 94 EXPORT_SYMBOL_GPL(smc_hash_sk); 95 96 void smc_unhash_sk(struct sock *sk) 97 { 98 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 99 100 write_lock_bh(&h->lock); 101 if (sk_del_node_init(sk)) 102 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 103 write_unlock_bh(&h->lock); 104 } 105 EXPORT_SYMBOL_GPL(smc_unhash_sk); 106 107 struct proto smc_proto = { 108 .name = "SMC", 109 .owner = THIS_MODULE, 110 .keepalive = smc_set_keepalive, 111 .hash = smc_hash_sk, 112 .unhash = smc_unhash_sk, 113 .obj_size = sizeof(struct smc_sock), 114 .h.smc_hash = &smc_v4_hashinfo, 115 .slab_flags = SLAB_TYPESAFE_BY_RCU, 116 }; 117 EXPORT_SYMBOL_GPL(smc_proto); 118 119 struct proto smc_proto6 = { 120 .name = "SMC6", 121 .owner = THIS_MODULE, 122 .keepalive = smc_set_keepalive, 123 .hash = smc_hash_sk, 124 .unhash = smc_unhash_sk, 125 .obj_size = sizeof(struct smc_sock), 126 .h.smc_hash = &smc_v6_hashinfo, 127 .slab_flags = SLAB_TYPESAFE_BY_RCU, 128 }; 129 EXPORT_SYMBOL_GPL(smc_proto6); 130 131 static void smc_restore_fallback_changes(struct smc_sock *smc) 132 { 133 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 134 smc->clcsock->file->private_data = smc->sk.sk_socket; 135 smc->clcsock->file = NULL; 136 } 137 } 138 139 static int __smc_release(struct smc_sock *smc) 140 { 141 struct sock *sk = &smc->sk; 142 int rc = 0; 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 } else { 149 if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT) 150 sock_put(sk); /* passive closing */ 151 if (sk->sk_state == SMC_LISTEN) { 152 /* wake up clcsock accept */ 153 rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); 154 } 155 sk->sk_state = SMC_CLOSED; 156 sk->sk_state_change(sk); 157 smc_restore_fallback_changes(smc); 158 } 159 160 sk->sk_prot->unhash(sk); 161 162 if (sk->sk_state == SMC_CLOSED) { 163 if (smc->clcsock) { 164 release_sock(sk); 165 smc_clcsock_release(smc); 166 lock_sock(sk); 167 } 168 if (!smc->use_fallback) 169 smc_conn_free(&smc->conn); 170 } 171 172 return rc; 173 } 174 175 static int smc_release(struct socket *sock) 176 { 177 struct sock *sk = sock->sk; 178 struct smc_sock *smc; 179 int rc = 0; 180 181 if (!sk) 182 goto out; 183 184 sock_hold(sk); /* sock_put below */ 185 smc = smc_sk(sk); 186 187 /* cleanup for a dangling non-blocking connect */ 188 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 189 tcp_abort(smc->clcsock->sk, ECONNABORTED); 190 flush_work(&smc->connect_work); 191 192 if (sk->sk_state == SMC_LISTEN) 193 /* smc_close_non_accepted() is called and acquires 194 * sock lock for child sockets again 195 */ 196 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 197 else 198 lock_sock(sk); 199 200 rc = __smc_release(smc); 201 202 /* detach socket */ 203 sock_orphan(sk); 204 sock->sk = NULL; 205 release_sock(sk); 206 207 sock_put(sk); /* sock_hold above */ 208 sock_put(sk); /* final sock_put */ 209 out: 210 return rc; 211 } 212 213 static void smc_destruct(struct sock *sk) 214 { 215 if (sk->sk_state != SMC_CLOSED) 216 return; 217 if (!sock_flag(sk, SOCK_DEAD)) 218 return; 219 220 sk_refcnt_debug_dec(sk); 221 } 222 223 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 224 int protocol) 225 { 226 struct smc_sock *smc; 227 struct proto *prot; 228 struct sock *sk; 229 230 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 231 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 232 if (!sk) 233 return NULL; 234 235 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 236 sk->sk_state = SMC_INIT; 237 sk->sk_destruct = smc_destruct; 238 sk->sk_protocol = protocol; 239 smc = smc_sk(sk); 240 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 241 INIT_WORK(&smc->connect_work, smc_connect_work); 242 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 243 INIT_LIST_HEAD(&smc->accept_q); 244 spin_lock_init(&smc->accept_q_lock); 245 spin_lock_init(&smc->conn.send_lock); 246 sk->sk_prot->hash(sk); 247 sk_refcnt_debug_inc(sk); 248 mutex_init(&smc->clcsock_release_lock); 249 250 return sk; 251 } 252 253 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 254 int addr_len) 255 { 256 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 257 struct sock *sk = sock->sk; 258 struct smc_sock *smc; 259 int rc; 260 261 smc = smc_sk(sk); 262 263 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 264 rc = -EINVAL; 265 if (addr_len < sizeof(struct sockaddr_in)) 266 goto out; 267 268 rc = -EAFNOSUPPORT; 269 if (addr->sin_family != AF_INET && 270 addr->sin_family != AF_INET6 && 271 addr->sin_family != AF_UNSPEC) 272 goto out; 273 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 274 if (addr->sin_family == AF_UNSPEC && 275 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 276 goto out; 277 278 lock_sock(sk); 279 280 /* Check if socket is already active */ 281 rc = -EINVAL; 282 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 283 goto out_rel; 284 285 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 286 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 287 288 out_rel: 289 release_sock(sk); 290 out: 291 return rc; 292 } 293 294 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 295 unsigned long mask) 296 { 297 /* options we don't get control via setsockopt for */ 298 nsk->sk_type = osk->sk_type; 299 nsk->sk_sndbuf = osk->sk_sndbuf; 300 nsk->sk_rcvbuf = osk->sk_rcvbuf; 301 nsk->sk_sndtimeo = osk->sk_sndtimeo; 302 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 303 nsk->sk_mark = osk->sk_mark; 304 nsk->sk_priority = osk->sk_priority; 305 nsk->sk_rcvlowat = osk->sk_rcvlowat; 306 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 307 nsk->sk_err = osk->sk_err; 308 309 nsk->sk_flags &= ~mask; 310 nsk->sk_flags |= osk->sk_flags & mask; 311 } 312 313 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 314 (1UL << SOCK_KEEPOPEN) | \ 315 (1UL << SOCK_LINGER) | \ 316 (1UL << SOCK_BROADCAST) | \ 317 (1UL << SOCK_TIMESTAMP) | \ 318 (1UL << SOCK_DBG) | \ 319 (1UL << SOCK_RCVTSTAMP) | \ 320 (1UL << SOCK_RCVTSTAMPNS) | \ 321 (1UL << SOCK_LOCALROUTE) | \ 322 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 323 (1UL << SOCK_RXQ_OVFL) | \ 324 (1UL << SOCK_WIFI_STATUS) | \ 325 (1UL << SOCK_NOFCS) | \ 326 (1UL << SOCK_FILTER_LOCKED) | \ 327 (1UL << SOCK_TSTAMP_NEW)) 328 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 329 * clc socket (since smc is not called for these options from net/core) 330 */ 331 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 332 { 333 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 334 } 335 336 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 337 (1UL << SOCK_KEEPOPEN) | \ 338 (1UL << SOCK_LINGER) | \ 339 (1UL << SOCK_DBG)) 340 /* copy only settings and flags relevant for smc from clc to smc socket */ 341 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 342 { 343 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 344 } 345 346 /* register the new rmb on all links */ 347 static int smcr_lgr_reg_rmbs(struct smc_link *link, 348 struct smc_buf_desc *rmb_desc) 349 { 350 struct smc_link_group *lgr = link->lgr; 351 int i, rc = 0; 352 353 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 354 if (rc) 355 return rc; 356 /* protect against parallel smc_llc_cli_rkey_exchange() and 357 * parallel smcr_link_reg_rmb() 358 */ 359 mutex_lock(&lgr->llc_conf_mutex); 360 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 361 if (!smc_link_active(&lgr->lnk[i])) 362 continue; 363 rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc); 364 if (rc) 365 goto out; 366 } 367 368 /* exchange confirm_rkey msg with peer */ 369 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 370 if (rc) { 371 rc = -EFAULT; 372 goto out; 373 } 374 rmb_desc->is_conf_rkey = true; 375 out: 376 mutex_unlock(&lgr->llc_conf_mutex); 377 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 378 return rc; 379 } 380 381 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 382 { 383 struct smc_link *link = smc->conn.lnk; 384 struct smc_llc_qentry *qentry; 385 int rc; 386 387 /* receive CONFIRM LINK request from server over RoCE fabric */ 388 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 389 SMC_LLC_CONFIRM_LINK); 390 if (!qentry) { 391 struct smc_clc_msg_decline dclc; 392 393 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 394 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 395 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 396 } 397 smc_llc_save_peer_uid(qentry); 398 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 399 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 400 if (rc) 401 return SMC_CLC_DECL_RMBE_EC; 402 403 rc = smc_ib_modify_qp_rts(link); 404 if (rc) 405 return SMC_CLC_DECL_ERR_RDYLNK; 406 407 smc_wr_remember_qp_attr(link); 408 409 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 410 return SMC_CLC_DECL_ERR_REGRMB; 411 412 /* confirm_rkey is implicit on 1st contact */ 413 smc->conn.rmb_desc->is_conf_rkey = true; 414 415 /* send CONFIRM LINK response over RoCE fabric */ 416 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 417 if (rc < 0) 418 return SMC_CLC_DECL_TIMEOUT_CL; 419 420 smc_llc_link_active(link); 421 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 422 423 /* optional 2nd link, receive ADD LINK request from server */ 424 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 425 SMC_LLC_ADD_LINK); 426 if (!qentry) { 427 struct smc_clc_msg_decline dclc; 428 429 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 430 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 431 if (rc == -EAGAIN) 432 rc = 0; /* no DECLINE received, go with one link */ 433 return rc; 434 } 435 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 436 smc_llc_cli_add_link(link, qentry); 437 return 0; 438 } 439 440 static void smcr_conn_save_peer_info(struct smc_sock *smc, 441 struct smc_clc_msg_accept_confirm *clc) 442 { 443 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 444 445 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 446 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 447 smc->conn.peer_rmbe_size = bufsize; 448 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 449 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 450 } 451 452 static bool smc_isascii(char *hostname) 453 { 454 int i; 455 456 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 457 if (!isascii(hostname[i])) 458 return false; 459 return true; 460 } 461 462 static void smcd_conn_save_peer_info(struct smc_sock *smc, 463 struct smc_clc_msg_accept_confirm *clc) 464 { 465 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 466 467 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 468 smc->conn.peer_token = clc->d0.token; 469 /* msg header takes up space in the buffer */ 470 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 471 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 472 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 473 if (clc->hdr.version > SMC_V1 && 474 (clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) { 475 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 476 (struct smc_clc_msg_accept_confirm_v2 *)clc; 477 struct smc_clc_first_contact_ext *fce = 478 (struct smc_clc_first_contact_ext *) 479 (((u8 *)clc_v2) + sizeof(*clc_v2)); 480 481 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->eid, 482 SMC_MAX_EID_LEN); 483 smc->conn.lgr->peer_os = fce->os_type; 484 smc->conn.lgr->peer_smc_release = fce->release; 485 if (smc_isascii(fce->hostname)) 486 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 487 SMC_MAX_HOSTNAME_LEN); 488 } 489 } 490 491 static void smc_conn_save_peer_info(struct smc_sock *smc, 492 struct smc_clc_msg_accept_confirm *clc) 493 { 494 if (smc->conn.lgr->is_smcd) 495 smcd_conn_save_peer_info(smc, clc); 496 else 497 smcr_conn_save_peer_info(smc, clc); 498 } 499 500 static void smc_link_save_peer_info(struct smc_link *link, 501 struct smc_clc_msg_accept_confirm *clc) 502 { 503 link->peer_qpn = ntoh24(clc->r0.qpn); 504 memcpy(link->peer_gid, clc->r0.lcl.gid, SMC_GID_SIZE); 505 memcpy(link->peer_mac, clc->r0.lcl.mac, sizeof(link->peer_mac)); 506 link->peer_psn = ntoh24(clc->r0.psn); 507 link->peer_mtu = clc->r0.qp_mtu; 508 } 509 510 static void smc_switch_to_fallback(struct smc_sock *smc) 511 { 512 smc->use_fallback = true; 513 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 514 smc->clcsock->file = smc->sk.sk_socket->file; 515 smc->clcsock->file->private_data = smc->clcsock; 516 smc->clcsock->wq.fasync_list = 517 smc->sk.sk_socket->wq.fasync_list; 518 } 519 } 520 521 /* fall back during connect */ 522 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 523 { 524 smc_switch_to_fallback(smc); 525 smc->fallback_rsn = reason_code; 526 smc_copy_sock_settings_to_clc(smc); 527 smc->connect_nonblock = 0; 528 if (smc->sk.sk_state == SMC_INIT) 529 smc->sk.sk_state = SMC_ACTIVE; 530 return 0; 531 } 532 533 /* decline and fall back during connect */ 534 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 535 u8 version) 536 { 537 int rc; 538 539 if (reason_code < 0) { /* error, fallback is not possible */ 540 if (smc->sk.sk_state == SMC_INIT) 541 sock_put(&smc->sk); /* passive closing */ 542 return reason_code; 543 } 544 if (reason_code != SMC_CLC_DECL_PEERDECL) { 545 rc = smc_clc_send_decline(smc, reason_code, version); 546 if (rc < 0) { 547 if (smc->sk.sk_state == SMC_INIT) 548 sock_put(&smc->sk); /* passive closing */ 549 return rc; 550 } 551 } 552 return smc_connect_fallback(smc, reason_code); 553 } 554 555 /* abort connecting */ 556 static void smc_connect_abort(struct smc_sock *smc, int local_first) 557 { 558 if (local_first) 559 smc_lgr_cleanup_early(&smc->conn); 560 else 561 smc_conn_free(&smc->conn); 562 } 563 564 /* check if there is a rdma device available for this connection. */ 565 /* called for connect and listen */ 566 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 567 { 568 /* PNET table look up: search active ib_device and port 569 * within same PNETID that also contains the ethernet device 570 * used for the internal TCP socket 571 */ 572 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 573 if (!ini->ib_dev) 574 return SMC_CLC_DECL_NOSMCRDEV; 575 return 0; 576 } 577 578 /* check if there is an ISM device available for this connection. */ 579 /* called for connect and listen */ 580 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 581 { 582 /* Find ISM device with same PNETID as connecting interface */ 583 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 584 if (!ini->ism_dev[0]) 585 return SMC_CLC_DECL_NOSMCDDEV; 586 else 587 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 588 return 0; 589 } 590 591 /* is chid unique for the ism devices that are already determined? */ 592 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 593 int cnt) 594 { 595 int i = (!ini->ism_dev[0]) ? 1 : 0; 596 597 for (; i < cnt; i++) 598 if (ini->ism_chid[i] == chid) 599 return false; 600 return true; 601 } 602 603 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 604 * PNETID matching net_device) 605 */ 606 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 607 struct smc_init_info *ini) 608 { 609 int rc = SMC_CLC_DECL_NOSMCDDEV; 610 struct smcd_dev *smcd; 611 int i = 1; 612 u16 chid; 613 614 if (smcd_indicated(ini->smc_type_v1)) 615 rc = 0; /* already initialized for V1 */ 616 mutex_lock(&smcd_dev_list.mutex); 617 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 618 if (smcd->going_away || smcd == ini->ism_dev[0]) 619 continue; 620 chid = smc_ism_get_chid(smcd); 621 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 622 continue; 623 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 624 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 625 ini->ism_dev[i] = smcd; 626 ini->ism_chid[i] = chid; 627 ini->is_smcd = true; 628 rc = 0; 629 i++; 630 if (i > SMC_MAX_ISM_DEVS) 631 break; 632 } 633 } 634 mutex_unlock(&smcd_dev_list.mutex); 635 ini->ism_offered_cnt = i - 1; 636 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 637 ini->smcd_version = 0; 638 639 return rc; 640 } 641 642 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 643 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 644 struct smc_init_info *ini) 645 { 646 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 647 return SMC_CLC_DECL_ISMVLANERR; 648 return 0; 649 } 650 651 static int smc_find_proposal_devices(struct smc_sock *smc, 652 struct smc_init_info *ini) 653 { 654 int rc = 0; 655 656 /* check if there is an ism device available */ 657 if (ini->smcd_version & SMC_V1) { 658 if (smc_find_ism_device(smc, ini) || 659 smc_connect_ism_vlan_setup(smc, ini)) { 660 if (ini->smc_type_v1 == SMC_TYPE_B) 661 ini->smc_type_v1 = SMC_TYPE_R; 662 else 663 ini->smc_type_v1 = SMC_TYPE_N; 664 } /* else ISM V1 is supported for this connection */ 665 if (smc_find_rdma_device(smc, ini)) { 666 if (ini->smc_type_v1 == SMC_TYPE_B) 667 ini->smc_type_v1 = SMC_TYPE_D; 668 else 669 ini->smc_type_v1 = SMC_TYPE_N; 670 } /* else RDMA is supported for this connection */ 671 } 672 if (smc_ism_v2_capable && smc_find_ism_v2_device_clnt(smc, ini)) 673 ini->smc_type_v2 = SMC_TYPE_N; 674 675 /* if neither ISM nor RDMA are supported, fallback */ 676 if (!smcr_indicated(ini->smc_type_v1) && 677 ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 678 rc = SMC_CLC_DECL_NOSMCDEV; 679 680 return rc; 681 } 682 683 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 684 * used, the VLAN ID will be registered again during the connection setup. 685 */ 686 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 687 struct smc_init_info *ini) 688 { 689 if (!smcd_indicated(ini->smc_type_v1)) 690 return 0; 691 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 692 return SMC_CLC_DECL_CNFERR; 693 return 0; 694 } 695 696 #define SMC_CLC_MAX_ACCEPT_LEN \ 697 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 698 sizeof(struct smc_clc_first_contact_ext) + \ 699 sizeof(struct smc_clc_msg_trail)) 700 701 /* CLC handshake during connect */ 702 static int smc_connect_clc(struct smc_sock *smc, 703 struct smc_clc_msg_accept_confirm_v2 *aclc2, 704 struct smc_init_info *ini) 705 { 706 int rc = 0; 707 708 /* do inband token exchange */ 709 rc = smc_clc_send_proposal(smc, ini); 710 if (rc) 711 return rc; 712 /* receive SMC Accept CLC message */ 713 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 714 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 715 } 716 717 /* setup for RDMA connection of client */ 718 static int smc_connect_rdma(struct smc_sock *smc, 719 struct smc_clc_msg_accept_confirm *aclc, 720 struct smc_init_info *ini) 721 { 722 int i, reason_code = 0; 723 struct smc_link *link; 724 725 ini->is_smcd = false; 726 ini->ib_lcl = &aclc->r0.lcl; 727 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 728 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 729 730 mutex_lock(&smc_client_lgr_pending); 731 reason_code = smc_conn_create(smc, ini); 732 if (reason_code) { 733 mutex_unlock(&smc_client_lgr_pending); 734 return reason_code; 735 } 736 737 smc_conn_save_peer_info(smc, aclc); 738 739 if (ini->first_contact_local) { 740 link = smc->conn.lnk; 741 } else { 742 /* set link that was assigned by server */ 743 link = NULL; 744 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 745 struct smc_link *l = &smc->conn.lgr->lnk[i]; 746 747 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 748 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 749 SMC_GID_SIZE) && 750 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 751 sizeof(l->peer_mac))) { 752 link = l; 753 break; 754 } 755 } 756 if (!link) { 757 reason_code = SMC_CLC_DECL_NOSRVLINK; 758 goto connect_abort; 759 } 760 smc->conn.lnk = link; 761 } 762 763 /* create send buffer and rmb */ 764 if (smc_buf_create(smc, false)) { 765 reason_code = SMC_CLC_DECL_MEM; 766 goto connect_abort; 767 } 768 769 if (ini->first_contact_local) 770 smc_link_save_peer_info(link, aclc); 771 772 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 773 reason_code = SMC_CLC_DECL_ERR_RTOK; 774 goto connect_abort; 775 } 776 777 smc_close_init(smc); 778 smc_rx_init(smc); 779 780 if (ini->first_contact_local) { 781 if (smc_ib_ready_link(link)) { 782 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 783 goto connect_abort; 784 } 785 } else { 786 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 787 reason_code = SMC_CLC_DECL_ERR_REGRMB; 788 goto connect_abort; 789 } 790 } 791 smc_rmb_sync_sg_for_device(&smc->conn); 792 793 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 794 SMC_V1); 795 if (reason_code) 796 goto connect_abort; 797 798 smc_tx_init(smc); 799 800 if (ini->first_contact_local) { 801 /* QP confirmation over RoCE fabric */ 802 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 803 reason_code = smcr_clnt_conf_first_link(smc); 804 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 805 if (reason_code) 806 goto connect_abort; 807 } 808 mutex_unlock(&smc_client_lgr_pending); 809 810 smc_copy_sock_settings_to_clc(smc); 811 smc->connect_nonblock = 0; 812 if (smc->sk.sk_state == SMC_INIT) 813 smc->sk.sk_state = SMC_ACTIVE; 814 815 return 0; 816 connect_abort: 817 smc_connect_abort(smc, ini->first_contact_local); 818 mutex_unlock(&smc_client_lgr_pending); 819 smc->connect_nonblock = 0; 820 821 return reason_code; 822 } 823 824 /* The server has chosen one of the proposed ISM devices for the communication. 825 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 826 */ 827 static int 828 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 829 struct smc_init_info *ini) 830 { 831 int i; 832 833 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 834 if (ini->ism_chid[i] == ntohs(aclc->chid)) { 835 ini->ism_selected = i; 836 return 0; 837 } 838 } 839 840 return -EPROTO; 841 } 842 843 /* setup for ISM connection of client */ 844 static int smc_connect_ism(struct smc_sock *smc, 845 struct smc_clc_msg_accept_confirm *aclc, 846 struct smc_init_info *ini) 847 { 848 int rc = 0; 849 850 ini->is_smcd = true; 851 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 852 853 if (aclc->hdr.version == SMC_V2) { 854 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 855 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 856 857 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 858 if (rc) 859 return rc; 860 } 861 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 862 863 /* there is only one lgr role for SMC-D; use server lock */ 864 mutex_lock(&smc_server_lgr_pending); 865 rc = smc_conn_create(smc, ini); 866 if (rc) { 867 mutex_unlock(&smc_server_lgr_pending); 868 return rc; 869 } 870 871 /* Create send and receive buffers */ 872 rc = smc_buf_create(smc, true); 873 if (rc) { 874 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 875 goto connect_abort; 876 } 877 878 smc_conn_save_peer_info(smc, aclc); 879 smc_close_init(smc); 880 smc_rx_init(smc); 881 smc_tx_init(smc); 882 883 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 884 aclc->hdr.version); 885 if (rc) 886 goto connect_abort; 887 mutex_unlock(&smc_server_lgr_pending); 888 889 smc_copy_sock_settings_to_clc(smc); 890 smc->connect_nonblock = 0; 891 if (smc->sk.sk_state == SMC_INIT) 892 smc->sk.sk_state = SMC_ACTIVE; 893 894 return 0; 895 connect_abort: 896 smc_connect_abort(smc, ini->first_contact_local); 897 mutex_unlock(&smc_server_lgr_pending); 898 smc->connect_nonblock = 0; 899 900 return rc; 901 } 902 903 /* check if received accept type and version matches a proposed one */ 904 static int smc_connect_check_aclc(struct smc_init_info *ini, 905 struct smc_clc_msg_accept_confirm *aclc) 906 { 907 if ((aclc->hdr.typev1 == SMC_TYPE_R && 908 !smcr_indicated(ini->smc_type_v1)) || 909 (aclc->hdr.typev1 == SMC_TYPE_D && 910 ((!smcd_indicated(ini->smc_type_v1) && 911 !smcd_indicated(ini->smc_type_v2)) || 912 (aclc->hdr.version == SMC_V1 && 913 !smcd_indicated(ini->smc_type_v1)) || 914 (aclc->hdr.version == SMC_V2 && 915 !smcd_indicated(ini->smc_type_v2))))) 916 return SMC_CLC_DECL_MODEUNSUPP; 917 918 return 0; 919 } 920 921 /* perform steps before actually connecting */ 922 static int __smc_connect(struct smc_sock *smc) 923 { 924 u8 version = smc_ism_v2_capable ? SMC_V2 : SMC_V1; 925 struct smc_clc_msg_accept_confirm_v2 *aclc2; 926 struct smc_clc_msg_accept_confirm *aclc; 927 struct smc_init_info *ini = NULL; 928 u8 *buf = NULL; 929 int rc = 0; 930 931 if (smc->use_fallback) 932 return smc_connect_fallback(smc, smc->fallback_rsn); 933 934 /* if peer has not signalled SMC-capability, fall back */ 935 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 936 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 937 938 /* IPSec connections opt out of SMC optimizations */ 939 if (using_ipsec(smc)) 940 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 941 version); 942 943 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 944 if (!ini) 945 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 946 version); 947 948 ini->smcd_version = SMC_V1; 949 ini->smcd_version |= smc_ism_v2_capable ? SMC_V2 : 0; 950 ini->smc_type_v1 = SMC_TYPE_B; 951 ini->smc_type_v2 = smc_ism_v2_capable ? SMC_TYPE_D : SMC_TYPE_N; 952 953 /* get vlan id from IP device */ 954 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 955 ini->smcd_version &= ~SMC_V1; 956 ini->smc_type_v1 = SMC_TYPE_N; 957 if (!ini->smcd_version) { 958 rc = SMC_CLC_DECL_GETVLANERR; 959 goto fallback; 960 } 961 } 962 963 rc = smc_find_proposal_devices(smc, ini); 964 if (rc) 965 goto fallback; 966 967 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 968 if (!buf) { 969 rc = SMC_CLC_DECL_MEM; 970 goto fallback; 971 } 972 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 973 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 974 975 /* perform CLC handshake */ 976 rc = smc_connect_clc(smc, aclc2, ini); 977 if (rc) 978 goto vlan_cleanup; 979 980 /* check if smc modes and versions of CLC proposal and accept match */ 981 rc = smc_connect_check_aclc(ini, aclc); 982 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : version; 983 if (rc) 984 goto vlan_cleanup; 985 986 /* depending on previous steps, connect using rdma or ism */ 987 if (aclc->hdr.typev1 == SMC_TYPE_R) 988 rc = smc_connect_rdma(smc, aclc, ini); 989 else if (aclc->hdr.typev1 == SMC_TYPE_D) 990 rc = smc_connect_ism(smc, aclc, ini); 991 if (rc) 992 goto vlan_cleanup; 993 994 smc_connect_ism_vlan_cleanup(smc, ini); 995 kfree(buf); 996 kfree(ini); 997 return 0; 998 999 vlan_cleanup: 1000 smc_connect_ism_vlan_cleanup(smc, ini); 1001 kfree(buf); 1002 fallback: 1003 kfree(ini); 1004 return smc_connect_decline_fallback(smc, rc, version); 1005 } 1006 1007 static void smc_connect_work(struct work_struct *work) 1008 { 1009 struct smc_sock *smc = container_of(work, struct smc_sock, 1010 connect_work); 1011 long timeo = smc->sk.sk_sndtimeo; 1012 int rc = 0; 1013 1014 if (!timeo) 1015 timeo = MAX_SCHEDULE_TIMEOUT; 1016 lock_sock(smc->clcsock->sk); 1017 if (smc->clcsock->sk->sk_err) { 1018 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1019 } else if ((1 << smc->clcsock->sk->sk_state) & 1020 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 1021 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1022 if ((rc == -EPIPE) && 1023 ((1 << smc->clcsock->sk->sk_state) & 1024 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1025 rc = 0; 1026 } 1027 release_sock(smc->clcsock->sk); 1028 lock_sock(&smc->sk); 1029 if (rc != 0 || smc->sk.sk_err) { 1030 smc->sk.sk_state = SMC_CLOSED; 1031 if (rc == -EPIPE || rc == -EAGAIN) 1032 smc->sk.sk_err = EPIPE; 1033 else if (signal_pending(current)) 1034 smc->sk.sk_err = -sock_intr_errno(timeo); 1035 sock_put(&smc->sk); /* passive closing */ 1036 goto out; 1037 } 1038 1039 rc = __smc_connect(smc); 1040 if (rc < 0) 1041 smc->sk.sk_err = -rc; 1042 1043 out: 1044 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1045 if (smc->sk.sk_err) { 1046 smc->sk.sk_state_change(&smc->sk); 1047 } else { /* allow polling before and after fallback decision */ 1048 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1049 smc->sk.sk_write_space(&smc->sk); 1050 } 1051 } 1052 release_sock(&smc->sk); 1053 } 1054 1055 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1056 int alen, int flags) 1057 { 1058 struct sock *sk = sock->sk; 1059 struct smc_sock *smc; 1060 int rc = -EINVAL; 1061 1062 smc = smc_sk(sk); 1063 1064 /* separate smc parameter checking to be safe */ 1065 if (alen < sizeof(addr->sa_family)) 1066 goto out_err; 1067 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1068 goto out_err; 1069 1070 lock_sock(sk); 1071 switch (sk->sk_state) { 1072 default: 1073 goto out; 1074 case SMC_ACTIVE: 1075 rc = -EISCONN; 1076 goto out; 1077 case SMC_INIT: 1078 rc = 0; 1079 break; 1080 } 1081 1082 smc_copy_sock_settings_to_clc(smc); 1083 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1084 if (smc->connect_nonblock) { 1085 rc = -EALREADY; 1086 goto out; 1087 } 1088 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1089 if (rc && rc != -EINPROGRESS) 1090 goto out; 1091 1092 sock_hold(&smc->sk); /* sock put in passive closing */ 1093 if (smc->use_fallback) 1094 goto out; 1095 if (flags & O_NONBLOCK) { 1096 if (queue_work(smc_hs_wq, &smc->connect_work)) 1097 smc->connect_nonblock = 1; 1098 rc = -EINPROGRESS; 1099 } else { 1100 rc = __smc_connect(smc); 1101 if (rc < 0) 1102 goto out; 1103 else 1104 rc = 0; /* success cases including fallback */ 1105 } 1106 1107 out: 1108 release_sock(sk); 1109 out_err: 1110 return rc; 1111 } 1112 1113 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1114 { 1115 struct socket *new_clcsock = NULL; 1116 struct sock *lsk = &lsmc->sk; 1117 struct sock *new_sk; 1118 int rc = -EINVAL; 1119 1120 release_sock(lsk); 1121 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1122 if (!new_sk) { 1123 rc = -ENOMEM; 1124 lsk->sk_err = ENOMEM; 1125 *new_smc = NULL; 1126 lock_sock(lsk); 1127 goto out; 1128 } 1129 *new_smc = smc_sk(new_sk); 1130 1131 mutex_lock(&lsmc->clcsock_release_lock); 1132 if (lsmc->clcsock) 1133 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1134 mutex_unlock(&lsmc->clcsock_release_lock); 1135 lock_sock(lsk); 1136 if (rc < 0 && rc != -EAGAIN) 1137 lsk->sk_err = -rc; 1138 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1139 new_sk->sk_prot->unhash(new_sk); 1140 if (new_clcsock) 1141 sock_release(new_clcsock); 1142 new_sk->sk_state = SMC_CLOSED; 1143 sock_set_flag(new_sk, SOCK_DEAD); 1144 sock_put(new_sk); /* final */ 1145 *new_smc = NULL; 1146 goto out; 1147 } 1148 1149 /* new clcsock has inherited the smc listen-specific sk_data_ready 1150 * function; switch it back to the original sk_data_ready function 1151 */ 1152 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1153 (*new_smc)->clcsock = new_clcsock; 1154 out: 1155 return rc; 1156 } 1157 1158 /* add a just created sock to the accept queue of the listen sock as 1159 * candidate for a following socket accept call from user space 1160 */ 1161 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1162 { 1163 struct smc_sock *par = smc_sk(parent); 1164 1165 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1166 spin_lock(&par->accept_q_lock); 1167 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1168 spin_unlock(&par->accept_q_lock); 1169 sk_acceptq_added(parent); 1170 } 1171 1172 /* remove a socket from the accept queue of its parental listening socket */ 1173 static void smc_accept_unlink(struct sock *sk) 1174 { 1175 struct smc_sock *par = smc_sk(sk)->listen_smc; 1176 1177 spin_lock(&par->accept_q_lock); 1178 list_del_init(&smc_sk(sk)->accept_q); 1179 spin_unlock(&par->accept_q_lock); 1180 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1181 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1182 } 1183 1184 /* remove a sock from the accept queue to bind it to a new socket created 1185 * for a socket accept call from user space 1186 */ 1187 struct sock *smc_accept_dequeue(struct sock *parent, 1188 struct socket *new_sock) 1189 { 1190 struct smc_sock *isk, *n; 1191 struct sock *new_sk; 1192 1193 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1194 new_sk = (struct sock *)isk; 1195 1196 smc_accept_unlink(new_sk); 1197 if (new_sk->sk_state == SMC_CLOSED) { 1198 new_sk->sk_prot->unhash(new_sk); 1199 if (isk->clcsock) { 1200 sock_release(isk->clcsock); 1201 isk->clcsock = NULL; 1202 } 1203 sock_put(new_sk); /* final */ 1204 continue; 1205 } 1206 if (new_sock) { 1207 sock_graft(new_sk, new_sock); 1208 if (isk->use_fallback) { 1209 smc_sk(new_sk)->clcsock->file = new_sock->file; 1210 isk->clcsock->file->private_data = isk->clcsock; 1211 } 1212 } 1213 return new_sk; 1214 } 1215 return NULL; 1216 } 1217 1218 /* clean up for a created but never accepted sock */ 1219 void smc_close_non_accepted(struct sock *sk) 1220 { 1221 struct smc_sock *smc = smc_sk(sk); 1222 1223 sock_hold(sk); /* sock_put below */ 1224 lock_sock(sk); 1225 if (!sk->sk_lingertime) 1226 /* wait for peer closing */ 1227 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1228 __smc_release(smc); 1229 release_sock(sk); 1230 sock_put(sk); /* sock_hold above */ 1231 sock_put(sk); /* final sock_put */ 1232 } 1233 1234 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1235 { 1236 struct smc_link *link = smc->conn.lnk; 1237 struct smc_llc_qentry *qentry; 1238 int rc; 1239 1240 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 1241 return SMC_CLC_DECL_ERR_REGRMB; 1242 1243 /* send CONFIRM LINK request to client over the RoCE fabric */ 1244 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1245 if (rc < 0) 1246 return SMC_CLC_DECL_TIMEOUT_CL; 1247 1248 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1249 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1250 SMC_LLC_CONFIRM_LINK); 1251 if (!qentry) { 1252 struct smc_clc_msg_decline dclc; 1253 1254 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1255 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1256 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1257 } 1258 smc_llc_save_peer_uid(qentry); 1259 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1260 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1261 if (rc) 1262 return SMC_CLC_DECL_RMBE_EC; 1263 1264 /* confirm_rkey is implicit on 1st contact */ 1265 smc->conn.rmb_desc->is_conf_rkey = true; 1266 1267 smc_llc_link_active(link); 1268 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1269 1270 /* initial contact - try to establish second link */ 1271 smc_llc_srv_add_link(link); 1272 return 0; 1273 } 1274 1275 /* listen worker: finish */ 1276 static void smc_listen_out(struct smc_sock *new_smc) 1277 { 1278 struct smc_sock *lsmc = new_smc->listen_smc; 1279 struct sock *newsmcsk = &new_smc->sk; 1280 1281 if (lsmc->sk.sk_state == SMC_LISTEN) { 1282 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1283 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1284 release_sock(&lsmc->sk); 1285 } else { /* no longer listening */ 1286 smc_close_non_accepted(newsmcsk); 1287 } 1288 1289 /* Wake up accept */ 1290 lsmc->sk.sk_data_ready(&lsmc->sk); 1291 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1292 } 1293 1294 /* listen worker: finish in state connected */ 1295 static void smc_listen_out_connected(struct smc_sock *new_smc) 1296 { 1297 struct sock *newsmcsk = &new_smc->sk; 1298 1299 sk_refcnt_debug_inc(newsmcsk); 1300 if (newsmcsk->sk_state == SMC_INIT) 1301 newsmcsk->sk_state = SMC_ACTIVE; 1302 1303 smc_listen_out(new_smc); 1304 } 1305 1306 /* listen worker: finish in error state */ 1307 static void smc_listen_out_err(struct smc_sock *new_smc) 1308 { 1309 struct sock *newsmcsk = &new_smc->sk; 1310 1311 if (newsmcsk->sk_state == SMC_INIT) 1312 sock_put(&new_smc->sk); /* passive closing */ 1313 newsmcsk->sk_state = SMC_CLOSED; 1314 1315 smc_listen_out(new_smc); 1316 } 1317 1318 /* listen worker: decline and fall back if possible */ 1319 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1320 struct smc_init_info *ini, u8 version) 1321 { 1322 /* RDMA setup failed, switch back to TCP */ 1323 if (ini->first_contact_local) 1324 smc_lgr_cleanup_early(&new_smc->conn); 1325 else 1326 smc_conn_free(&new_smc->conn); 1327 if (reason_code < 0) { /* error, no fallback possible */ 1328 smc_listen_out_err(new_smc); 1329 return; 1330 } 1331 smc_switch_to_fallback(new_smc); 1332 new_smc->fallback_rsn = reason_code; 1333 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1334 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1335 smc_listen_out_err(new_smc); 1336 return; 1337 } 1338 } 1339 smc_listen_out_connected(new_smc); 1340 } 1341 1342 /* listen worker: version checking */ 1343 static int smc_listen_v2_check(struct smc_sock *new_smc, 1344 struct smc_clc_msg_proposal *pclc, 1345 struct smc_init_info *ini) 1346 { 1347 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1348 struct smc_clc_v2_extension *pclc_v2_ext; 1349 1350 ini->smc_type_v1 = pclc->hdr.typev1; 1351 ini->smc_type_v2 = pclc->hdr.typev2; 1352 ini->smcd_version = ini->smc_type_v1 != SMC_TYPE_N ? SMC_V1 : 0; 1353 if (pclc->hdr.version > SMC_V1) 1354 ini->smcd_version |= 1355 ini->smc_type_v2 != SMC_TYPE_N ? SMC_V2 : 0; 1356 if (!smc_ism_v2_capable) { 1357 ini->smcd_version &= ~SMC_V2; 1358 goto out; 1359 } 1360 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1361 if (!pclc_v2_ext) { 1362 ini->smcd_version &= ~SMC_V2; 1363 goto out; 1364 } 1365 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1366 if (!pclc_smcd_v2_ext) 1367 ini->smcd_version &= ~SMC_V2; 1368 1369 out: 1370 if (!ini->smcd_version) { 1371 if (pclc->hdr.typev1 == SMC_TYPE_B || 1372 pclc->hdr.typev2 == SMC_TYPE_B) 1373 return SMC_CLC_DECL_NOSMCDEV; 1374 if (pclc->hdr.typev1 == SMC_TYPE_D || 1375 pclc->hdr.typev2 == SMC_TYPE_D) 1376 return SMC_CLC_DECL_NOSMCDDEV; 1377 return SMC_CLC_DECL_NOSMCRDEV; 1378 } 1379 1380 return 0; 1381 } 1382 1383 /* listen worker: check prefixes */ 1384 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1385 struct smc_clc_msg_proposal *pclc) 1386 { 1387 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1388 struct socket *newclcsock = new_smc->clcsock; 1389 1390 if (pclc->hdr.typev1 == SMC_TYPE_N) 1391 return 0; 1392 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1393 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1394 return SMC_CLC_DECL_DIFFPREFIX; 1395 1396 return 0; 1397 } 1398 1399 /* listen worker: initialize connection and buffers */ 1400 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1401 struct smc_init_info *ini) 1402 { 1403 int rc; 1404 1405 /* allocate connection / link group */ 1406 rc = smc_conn_create(new_smc, ini); 1407 if (rc) 1408 return rc; 1409 1410 /* create send buffer and rmb */ 1411 if (smc_buf_create(new_smc, false)) 1412 return SMC_CLC_DECL_MEM; 1413 1414 return 0; 1415 } 1416 1417 /* listen worker: initialize connection and buffers for SMC-D */ 1418 static int smc_listen_ism_init(struct smc_sock *new_smc, 1419 struct smc_init_info *ini) 1420 { 1421 int rc; 1422 1423 rc = smc_conn_create(new_smc, ini); 1424 if (rc) 1425 return rc; 1426 1427 /* Create send and receive buffers */ 1428 rc = smc_buf_create(new_smc, true); 1429 if (rc) { 1430 if (ini->first_contact_local) 1431 smc_lgr_cleanup_early(&new_smc->conn); 1432 else 1433 smc_conn_free(&new_smc->conn); 1434 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 1435 SMC_CLC_DECL_MEM; 1436 } 1437 1438 return 0; 1439 } 1440 1441 static bool smc_is_already_selected(struct smcd_dev *smcd, 1442 struct smc_init_info *ini, 1443 int matches) 1444 { 1445 int i; 1446 1447 for (i = 0; i < matches; i++) 1448 if (smcd == ini->ism_dev[i]) 1449 return true; 1450 1451 return false; 1452 } 1453 1454 /* check for ISM devices matching proposed ISM devices */ 1455 static void smc_check_ism_v2_match(struct smc_init_info *ini, 1456 u16 proposed_chid, u64 proposed_gid, 1457 unsigned int *matches) 1458 { 1459 struct smcd_dev *smcd; 1460 1461 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1462 if (smcd->going_away) 1463 continue; 1464 if (smc_is_already_selected(smcd, ini, *matches)) 1465 continue; 1466 if (smc_ism_get_chid(smcd) == proposed_chid && 1467 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 1468 ini->ism_peer_gid[*matches] = proposed_gid; 1469 ini->ism_dev[*matches] = smcd; 1470 (*matches)++; 1471 break; 1472 } 1473 } 1474 } 1475 1476 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 1477 struct smc_clc_msg_proposal *pclc, 1478 struct smc_init_info *ini) 1479 { 1480 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 1481 struct smc_clc_v2_extension *smc_v2_ext; 1482 struct smc_clc_msg_smcd *pclc_smcd; 1483 unsigned int matches = 0; 1484 u8 smcd_version; 1485 u8 *eid = NULL; 1486 int i; 1487 1488 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 1489 goto not_found; 1490 1491 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1492 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1493 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 1494 if (!smcd_v2_ext || 1495 !smc_v2_ext->hdr.flag.seid) /* no system EID support for SMCD */ 1496 goto not_found; 1497 1498 mutex_lock(&smcd_dev_list.mutex); 1499 if (pclc_smcd->ism.chid) 1500 /* check for ISM device matching proposed native ISM device */ 1501 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 1502 ntohll(pclc_smcd->ism.gid), &matches); 1503 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 1504 /* check for ISM devices matching proposed non-native ISM 1505 * devices 1506 */ 1507 smc_check_ism_v2_match(ini, 1508 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 1509 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 1510 &matches); 1511 } 1512 mutex_unlock(&smcd_dev_list.mutex); 1513 1514 if (ini->ism_dev[0]) { 1515 smc_ism_get_system_eid(ini->ism_dev[0], &eid); 1516 if (memcmp(eid, smcd_v2_ext->system_eid, SMC_MAX_EID_LEN)) 1517 goto not_found; 1518 } else { 1519 goto not_found; 1520 } 1521 1522 /* separate - outside the smcd_dev_list.lock */ 1523 smcd_version = ini->smcd_version; 1524 for (i = 0; i < matches; i++) { 1525 ini->smcd_version = SMC_V2; 1526 ini->is_smcd = true; 1527 ini->ism_selected = i; 1528 if (smc_listen_ism_init(new_smc, ini)) 1529 /* try next active ISM device */ 1530 continue; 1531 return; /* matching and usable V2 ISM device found */ 1532 } 1533 /* no V2 ISM device could be initialized */ 1534 ini->smcd_version = smcd_version; /* restore original value */ 1535 1536 not_found: 1537 ini->smcd_version &= ~SMC_V2; 1538 ini->ism_dev[0] = NULL; 1539 ini->is_smcd = false; 1540 } 1541 1542 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 1543 struct smc_clc_msg_proposal *pclc, 1544 struct smc_init_info *ini) 1545 { 1546 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 1547 1548 /* check if ISM V1 is available */ 1549 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 1550 goto not_found; 1551 ini->is_smcd = true; /* prepare ISM check */ 1552 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 1553 if (smc_find_ism_device(new_smc, ini)) 1554 goto not_found; 1555 ini->ism_selected = 0; 1556 if (!smc_listen_ism_init(new_smc, ini)) 1557 return; /* V1 ISM device found */ 1558 1559 not_found: 1560 ini->ism_dev[0] = NULL; 1561 ini->is_smcd = false; 1562 } 1563 1564 /* listen worker: register buffers */ 1565 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 1566 { 1567 struct smc_connection *conn = &new_smc->conn; 1568 1569 if (!local_first) { 1570 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 1571 return SMC_CLC_DECL_ERR_REGRMB; 1572 } 1573 smc_rmb_sync_sg_for_device(&new_smc->conn); 1574 1575 return 0; 1576 } 1577 1578 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 1579 struct smc_clc_msg_proposal *pclc, 1580 struct smc_init_info *ini) 1581 { 1582 int rc; 1583 1584 if (!smcr_indicated(ini->smc_type_v1)) 1585 return SMC_CLC_DECL_NOSMCDEV; 1586 1587 /* prepare RDMA check */ 1588 ini->ib_lcl = &pclc->lcl; 1589 rc = smc_find_rdma_device(new_smc, ini); 1590 if (rc) { 1591 /* no RDMA device found */ 1592 if (ini->smc_type_v1 == SMC_TYPE_B) 1593 /* neither ISM nor RDMA device found */ 1594 rc = SMC_CLC_DECL_NOSMCDEV; 1595 return rc; 1596 } 1597 rc = smc_listen_rdma_init(new_smc, ini); 1598 if (rc) 1599 return rc; 1600 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 1601 } 1602 1603 /* determine the local device matching to proposal */ 1604 static int smc_listen_find_device(struct smc_sock *new_smc, 1605 struct smc_clc_msg_proposal *pclc, 1606 struct smc_init_info *ini) 1607 { 1608 int rc; 1609 1610 /* check for ISM device matching V2 proposed device */ 1611 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 1612 if (ini->ism_dev[0]) 1613 return 0; 1614 1615 if (!(ini->smcd_version & SMC_V1)) 1616 return SMC_CLC_DECL_NOSMCDEV; 1617 1618 /* check for matching IP prefix and subnet length */ 1619 rc = smc_listen_prfx_check(new_smc, pclc); 1620 if (rc) 1621 return rc; 1622 1623 /* get vlan id from IP device */ 1624 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 1625 return SMC_CLC_DECL_GETVLANERR; 1626 1627 /* check for ISM device matching V1 proposed device */ 1628 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 1629 if (ini->ism_dev[0]) 1630 return 0; 1631 1632 if (pclc->hdr.typev1 == SMC_TYPE_D) 1633 return SMC_CLC_DECL_NOSMCDDEV; /* skip RDMA and decline */ 1634 1635 /* check if RDMA is available */ 1636 return smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 1637 } 1638 1639 /* listen worker: finish RDMA setup */ 1640 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1641 struct smc_clc_msg_accept_confirm *cclc, 1642 bool local_first) 1643 { 1644 struct smc_link *link = new_smc->conn.lnk; 1645 int reason_code = 0; 1646 1647 if (local_first) 1648 smc_link_save_peer_info(link, cclc); 1649 1650 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 1651 return SMC_CLC_DECL_ERR_RTOK; 1652 1653 if (local_first) { 1654 if (smc_ib_ready_link(link)) 1655 return SMC_CLC_DECL_ERR_RDYLNK; 1656 /* QP confirmation over RoCE fabric */ 1657 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1658 reason_code = smcr_serv_conf_first_link(new_smc); 1659 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1660 } 1661 return reason_code; 1662 } 1663 1664 /* setup for connection of server */ 1665 static void smc_listen_work(struct work_struct *work) 1666 { 1667 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1668 smc_listen_work); 1669 u8 version = smc_ism_v2_capable ? SMC_V2 : SMC_V1; 1670 struct socket *newclcsock = new_smc->clcsock; 1671 struct smc_clc_msg_accept_confirm *cclc; 1672 struct smc_clc_msg_proposal_area *buf; 1673 struct smc_clc_msg_proposal *pclc; 1674 struct smc_init_info *ini = NULL; 1675 int rc = 0; 1676 1677 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1678 return smc_listen_out_err(new_smc); 1679 1680 if (new_smc->use_fallback) { 1681 smc_listen_out_connected(new_smc); 1682 return; 1683 } 1684 1685 /* check if peer is smc capable */ 1686 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1687 smc_switch_to_fallback(new_smc); 1688 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1689 smc_listen_out_connected(new_smc); 1690 return; 1691 } 1692 1693 /* do inband token exchange - 1694 * wait for and receive SMC Proposal CLC message 1695 */ 1696 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 1697 if (!buf) { 1698 rc = SMC_CLC_DECL_MEM; 1699 goto out_decl; 1700 } 1701 pclc = (struct smc_clc_msg_proposal *)buf; 1702 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 1703 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1704 if (rc) 1705 goto out_decl; 1706 version = pclc->hdr.version == SMC_V1 ? SMC_V1 : version; 1707 1708 /* IPSec connections opt out of SMC optimizations */ 1709 if (using_ipsec(new_smc)) { 1710 rc = SMC_CLC_DECL_IPSEC; 1711 goto out_decl; 1712 } 1713 1714 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1715 if (!ini) { 1716 rc = SMC_CLC_DECL_MEM; 1717 goto out_decl; 1718 } 1719 1720 /* initial version checking */ 1721 rc = smc_listen_v2_check(new_smc, pclc, ini); 1722 if (rc) 1723 goto out_decl; 1724 1725 mutex_lock(&smc_server_lgr_pending); 1726 smc_close_init(new_smc); 1727 smc_rx_init(new_smc); 1728 smc_tx_init(new_smc); 1729 1730 /* determine ISM or RoCE device used for connection */ 1731 rc = smc_listen_find_device(new_smc, pclc, ini); 1732 if (rc) 1733 goto out_unlock; 1734 1735 /* send SMC Accept CLC message */ 1736 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 1737 ini->smcd_version == SMC_V2 ? SMC_V2 : SMC_V1); 1738 if (rc) 1739 goto out_unlock; 1740 1741 /* SMC-D does not need this lock any more */ 1742 if (ini->is_smcd) 1743 mutex_unlock(&smc_server_lgr_pending); 1744 1745 /* receive SMC Confirm CLC message */ 1746 memset(buf, 0, sizeof(*buf)); 1747 cclc = (struct smc_clc_msg_accept_confirm *)buf; 1748 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 1749 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1750 if (rc) { 1751 if (!ini->is_smcd) 1752 goto out_unlock; 1753 goto out_decl; 1754 } 1755 1756 /* finish worker */ 1757 if (!ini->is_smcd) { 1758 rc = smc_listen_rdma_finish(new_smc, cclc, 1759 ini->first_contact_local); 1760 if (rc) 1761 goto out_unlock; 1762 mutex_unlock(&smc_server_lgr_pending); 1763 } 1764 smc_conn_save_peer_info(new_smc, cclc); 1765 smc_listen_out_connected(new_smc); 1766 goto out_free; 1767 1768 out_unlock: 1769 mutex_unlock(&smc_server_lgr_pending); 1770 out_decl: 1771 smc_listen_decline(new_smc, rc, ini, version); 1772 out_free: 1773 kfree(ini); 1774 kfree(buf); 1775 } 1776 1777 static void smc_tcp_listen_work(struct work_struct *work) 1778 { 1779 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1780 tcp_listen_work); 1781 struct sock *lsk = &lsmc->sk; 1782 struct smc_sock *new_smc; 1783 int rc = 0; 1784 1785 lock_sock(lsk); 1786 while (lsk->sk_state == SMC_LISTEN) { 1787 rc = smc_clcsock_accept(lsmc, &new_smc); 1788 if (rc) /* clcsock accept queue empty or error */ 1789 goto out; 1790 if (!new_smc) 1791 continue; 1792 1793 new_smc->listen_smc = lsmc; 1794 new_smc->use_fallback = lsmc->use_fallback; 1795 new_smc->fallback_rsn = lsmc->fallback_rsn; 1796 sock_hold(lsk); /* sock_put in smc_listen_work */ 1797 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1798 smc_copy_sock_settings_to_smc(new_smc); 1799 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1800 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1801 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1802 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 1803 sock_put(&new_smc->sk); 1804 } 1805 1806 out: 1807 release_sock(lsk); 1808 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 1809 } 1810 1811 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 1812 { 1813 struct smc_sock *lsmc; 1814 1815 lsmc = (struct smc_sock *) 1816 ((uintptr_t)listen_clcsock->sk_user_data & ~SK_USER_DATA_NOCOPY); 1817 if (!lsmc) 1818 return; 1819 lsmc->clcsk_data_ready(listen_clcsock); 1820 if (lsmc->sk.sk_state == SMC_LISTEN) { 1821 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 1822 if (!queue_work(smc_hs_wq, &lsmc->tcp_listen_work)) 1823 sock_put(&lsmc->sk); 1824 } 1825 } 1826 1827 static int smc_listen(struct socket *sock, int backlog) 1828 { 1829 struct sock *sk = sock->sk; 1830 struct smc_sock *smc; 1831 int rc; 1832 1833 smc = smc_sk(sk); 1834 lock_sock(sk); 1835 1836 rc = -EINVAL; 1837 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1838 smc->connect_nonblock) 1839 goto out; 1840 1841 rc = 0; 1842 if (sk->sk_state == SMC_LISTEN) { 1843 sk->sk_max_ack_backlog = backlog; 1844 goto out; 1845 } 1846 /* some socket options are handled in core, so we could not apply 1847 * them to the clc socket -- copy smc socket options to clc socket 1848 */ 1849 smc_copy_sock_settings_to_clc(smc); 1850 if (!smc->use_fallback) 1851 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1852 1853 /* save original sk_data_ready function and establish 1854 * smc-specific sk_data_ready function 1855 */ 1856 smc->clcsk_data_ready = smc->clcsock->sk->sk_data_ready; 1857 smc->clcsock->sk->sk_data_ready = smc_clcsock_data_ready; 1858 smc->clcsock->sk->sk_user_data = 1859 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 1860 rc = kernel_listen(smc->clcsock, backlog); 1861 if (rc) 1862 goto out; 1863 sk->sk_max_ack_backlog = backlog; 1864 sk->sk_ack_backlog = 0; 1865 sk->sk_state = SMC_LISTEN; 1866 1867 out: 1868 release_sock(sk); 1869 return rc; 1870 } 1871 1872 static int smc_accept(struct socket *sock, struct socket *new_sock, 1873 int flags, bool kern) 1874 { 1875 struct sock *sk = sock->sk, *nsk; 1876 DECLARE_WAITQUEUE(wait, current); 1877 struct smc_sock *lsmc; 1878 long timeo; 1879 int rc = 0; 1880 1881 lsmc = smc_sk(sk); 1882 sock_hold(sk); /* sock_put below */ 1883 lock_sock(sk); 1884 1885 if (lsmc->sk.sk_state != SMC_LISTEN) { 1886 rc = -EINVAL; 1887 release_sock(sk); 1888 goto out; 1889 } 1890 1891 /* Wait for an incoming connection */ 1892 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1893 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1894 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1895 set_current_state(TASK_INTERRUPTIBLE); 1896 if (!timeo) { 1897 rc = -EAGAIN; 1898 break; 1899 } 1900 release_sock(sk); 1901 timeo = schedule_timeout(timeo); 1902 /* wakeup by sk_data_ready in smc_listen_work() */ 1903 sched_annotate_sleep(); 1904 lock_sock(sk); 1905 if (signal_pending(current)) { 1906 rc = sock_intr_errno(timeo); 1907 break; 1908 } 1909 } 1910 set_current_state(TASK_RUNNING); 1911 remove_wait_queue(sk_sleep(sk), &wait); 1912 1913 if (!rc) 1914 rc = sock_error(nsk); 1915 release_sock(sk); 1916 if (rc) 1917 goto out; 1918 1919 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1920 /* wait till data arrives on the socket */ 1921 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1922 MSEC_PER_SEC); 1923 if (smc_sk(nsk)->use_fallback) { 1924 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1925 1926 lock_sock(clcsk); 1927 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1928 sk_wait_data(clcsk, &timeo, NULL); 1929 release_sock(clcsk); 1930 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1931 lock_sock(nsk); 1932 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1933 release_sock(nsk); 1934 } 1935 } 1936 1937 out: 1938 sock_put(sk); /* sock_hold above */ 1939 return rc; 1940 } 1941 1942 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1943 int peer) 1944 { 1945 struct smc_sock *smc; 1946 1947 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1948 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1949 return -ENOTCONN; 1950 1951 smc = smc_sk(sock->sk); 1952 1953 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1954 } 1955 1956 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1957 { 1958 struct sock *sk = sock->sk; 1959 struct smc_sock *smc; 1960 int rc = -EPIPE; 1961 1962 smc = smc_sk(sk); 1963 lock_sock(sk); 1964 if ((sk->sk_state != SMC_ACTIVE) && 1965 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1966 (sk->sk_state != SMC_INIT)) 1967 goto out; 1968 1969 if (msg->msg_flags & MSG_FASTOPEN) { 1970 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1971 smc_switch_to_fallback(smc); 1972 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1973 } else { 1974 rc = -EINVAL; 1975 goto out; 1976 } 1977 } 1978 1979 if (smc->use_fallback) 1980 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1981 else 1982 rc = smc_tx_sendmsg(smc, msg, len); 1983 out: 1984 release_sock(sk); 1985 return rc; 1986 } 1987 1988 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1989 int flags) 1990 { 1991 struct sock *sk = sock->sk; 1992 struct smc_sock *smc; 1993 int rc = -ENOTCONN; 1994 1995 smc = smc_sk(sk); 1996 lock_sock(sk); 1997 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1998 /* socket was connected before, no more data to read */ 1999 rc = 0; 2000 goto out; 2001 } 2002 if ((sk->sk_state == SMC_INIT) || 2003 (sk->sk_state == SMC_LISTEN) || 2004 (sk->sk_state == SMC_CLOSED)) 2005 goto out; 2006 2007 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2008 rc = 0; 2009 goto out; 2010 } 2011 2012 if (smc->use_fallback) { 2013 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2014 } else { 2015 msg->msg_namelen = 0; 2016 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2017 } 2018 2019 out: 2020 release_sock(sk); 2021 return rc; 2022 } 2023 2024 static __poll_t smc_accept_poll(struct sock *parent) 2025 { 2026 struct smc_sock *isk = smc_sk(parent); 2027 __poll_t mask = 0; 2028 2029 spin_lock(&isk->accept_q_lock); 2030 if (!list_empty(&isk->accept_q)) 2031 mask = EPOLLIN | EPOLLRDNORM; 2032 spin_unlock(&isk->accept_q_lock); 2033 2034 return mask; 2035 } 2036 2037 static __poll_t smc_poll(struct file *file, struct socket *sock, 2038 poll_table *wait) 2039 { 2040 struct sock *sk = sock->sk; 2041 struct smc_sock *smc; 2042 __poll_t mask = 0; 2043 2044 if (!sk) 2045 return EPOLLNVAL; 2046 2047 smc = smc_sk(sock->sk); 2048 if (smc->use_fallback) { 2049 /* delegate to CLC child sock */ 2050 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2051 sk->sk_err = smc->clcsock->sk->sk_err; 2052 } else { 2053 if (sk->sk_state != SMC_CLOSED) 2054 sock_poll_wait(file, sock, wait); 2055 if (sk->sk_err) 2056 mask |= EPOLLERR; 2057 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2058 (sk->sk_state == SMC_CLOSED)) 2059 mask |= EPOLLHUP; 2060 if (sk->sk_state == SMC_LISTEN) { 2061 /* woken up by sk_data_ready in smc_listen_work() */ 2062 mask |= smc_accept_poll(sk); 2063 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2064 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2065 wait); 2066 sk->sk_err = smc->clcsock->sk->sk_err; 2067 } else { 2068 if ((sk->sk_state != SMC_INIT && 2069 atomic_read(&smc->conn.sndbuf_space)) || 2070 sk->sk_shutdown & SEND_SHUTDOWN) { 2071 mask |= EPOLLOUT | EPOLLWRNORM; 2072 } else { 2073 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2074 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2075 } 2076 if (atomic_read(&smc->conn.bytes_to_rcv)) 2077 mask |= EPOLLIN | EPOLLRDNORM; 2078 if (sk->sk_shutdown & RCV_SHUTDOWN) 2079 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2080 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2081 mask |= EPOLLIN; 2082 if (smc->conn.urg_state == SMC_URG_VALID) 2083 mask |= EPOLLPRI; 2084 } 2085 } 2086 2087 return mask; 2088 } 2089 2090 static int smc_shutdown(struct socket *sock, int how) 2091 { 2092 struct sock *sk = sock->sk; 2093 struct smc_sock *smc; 2094 int rc = -EINVAL; 2095 int rc1 = 0; 2096 2097 smc = smc_sk(sk); 2098 2099 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2100 return rc; 2101 2102 lock_sock(sk); 2103 2104 rc = -ENOTCONN; 2105 if ((sk->sk_state != SMC_ACTIVE) && 2106 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2107 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2108 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2109 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2110 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2111 goto out; 2112 if (smc->use_fallback) { 2113 rc = kernel_sock_shutdown(smc->clcsock, how); 2114 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2115 if (sk->sk_shutdown == SHUTDOWN_MASK) 2116 sk->sk_state = SMC_CLOSED; 2117 goto out; 2118 } 2119 switch (how) { 2120 case SHUT_RDWR: /* shutdown in both directions */ 2121 rc = smc_close_active(smc); 2122 break; 2123 case SHUT_WR: 2124 rc = smc_close_shutdown_write(smc); 2125 break; 2126 case SHUT_RD: 2127 rc = 0; 2128 /* nothing more to do because peer is not involved */ 2129 break; 2130 } 2131 if (smc->clcsock) 2132 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2133 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2134 sk->sk_shutdown |= how + 1; 2135 2136 out: 2137 release_sock(sk); 2138 return rc ? rc : rc1; 2139 } 2140 2141 static int smc_setsockopt(struct socket *sock, int level, int optname, 2142 sockptr_t optval, unsigned int optlen) 2143 { 2144 struct sock *sk = sock->sk; 2145 struct smc_sock *smc; 2146 int val, rc; 2147 2148 smc = smc_sk(sk); 2149 2150 /* generic setsockopts reaching us here always apply to the 2151 * CLC socket 2152 */ 2153 if (unlikely(!smc->clcsock->ops->setsockopt)) 2154 rc = -EOPNOTSUPP; 2155 else 2156 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 2157 optval, optlen); 2158 if (smc->clcsock->sk->sk_err) { 2159 sk->sk_err = smc->clcsock->sk->sk_err; 2160 sk->sk_error_report(sk); 2161 } 2162 2163 if (optlen < sizeof(int)) 2164 return -EINVAL; 2165 if (copy_from_sockptr(&val, optval, sizeof(int))) 2166 return -EFAULT; 2167 2168 lock_sock(sk); 2169 if (rc || smc->use_fallback) 2170 goto out; 2171 switch (optname) { 2172 case TCP_ULP: 2173 case TCP_FASTOPEN: 2174 case TCP_FASTOPEN_CONNECT: 2175 case TCP_FASTOPEN_KEY: 2176 case TCP_FASTOPEN_NO_COOKIE: 2177 /* option not supported by SMC */ 2178 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2179 smc_switch_to_fallback(smc); 2180 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 2181 } else { 2182 rc = -EINVAL; 2183 } 2184 break; 2185 case TCP_NODELAY: 2186 if (sk->sk_state != SMC_INIT && 2187 sk->sk_state != SMC_LISTEN && 2188 sk->sk_state != SMC_CLOSED) { 2189 if (val) 2190 mod_delayed_work(smc->conn.lgr->tx_wq, 2191 &smc->conn.tx_work, 0); 2192 } 2193 break; 2194 case TCP_CORK: 2195 if (sk->sk_state != SMC_INIT && 2196 sk->sk_state != SMC_LISTEN && 2197 sk->sk_state != SMC_CLOSED) { 2198 if (!val) 2199 mod_delayed_work(smc->conn.lgr->tx_wq, 2200 &smc->conn.tx_work, 0); 2201 } 2202 break; 2203 case TCP_DEFER_ACCEPT: 2204 smc->sockopt_defer_accept = val; 2205 break; 2206 default: 2207 break; 2208 } 2209 out: 2210 release_sock(sk); 2211 2212 return rc; 2213 } 2214 2215 static int smc_getsockopt(struct socket *sock, int level, int optname, 2216 char __user *optval, int __user *optlen) 2217 { 2218 struct smc_sock *smc; 2219 2220 smc = smc_sk(sock->sk); 2221 /* socket options apply to the CLC socket */ 2222 if (unlikely(!smc->clcsock->ops->getsockopt)) 2223 return -EOPNOTSUPP; 2224 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 2225 optval, optlen); 2226 } 2227 2228 static int smc_ioctl(struct socket *sock, unsigned int cmd, 2229 unsigned long arg) 2230 { 2231 union smc_host_cursor cons, urg; 2232 struct smc_connection *conn; 2233 struct smc_sock *smc; 2234 int answ; 2235 2236 smc = smc_sk(sock->sk); 2237 conn = &smc->conn; 2238 lock_sock(&smc->sk); 2239 if (smc->use_fallback) { 2240 if (!smc->clcsock) { 2241 release_sock(&smc->sk); 2242 return -EBADF; 2243 } 2244 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 2245 release_sock(&smc->sk); 2246 return answ; 2247 } 2248 switch (cmd) { 2249 case SIOCINQ: /* same as FIONREAD */ 2250 if (smc->sk.sk_state == SMC_LISTEN) { 2251 release_sock(&smc->sk); 2252 return -EINVAL; 2253 } 2254 if (smc->sk.sk_state == SMC_INIT || 2255 smc->sk.sk_state == SMC_CLOSED) 2256 answ = 0; 2257 else 2258 answ = atomic_read(&smc->conn.bytes_to_rcv); 2259 break; 2260 case SIOCOUTQ: 2261 /* output queue size (not send + not acked) */ 2262 if (smc->sk.sk_state == SMC_LISTEN) { 2263 release_sock(&smc->sk); 2264 return -EINVAL; 2265 } 2266 if (smc->sk.sk_state == SMC_INIT || 2267 smc->sk.sk_state == SMC_CLOSED) 2268 answ = 0; 2269 else 2270 answ = smc->conn.sndbuf_desc->len - 2271 atomic_read(&smc->conn.sndbuf_space); 2272 break; 2273 case SIOCOUTQNSD: 2274 /* output queue size (not send only) */ 2275 if (smc->sk.sk_state == SMC_LISTEN) { 2276 release_sock(&smc->sk); 2277 return -EINVAL; 2278 } 2279 if (smc->sk.sk_state == SMC_INIT || 2280 smc->sk.sk_state == SMC_CLOSED) 2281 answ = 0; 2282 else 2283 answ = smc_tx_prepared_sends(&smc->conn); 2284 break; 2285 case SIOCATMARK: 2286 if (smc->sk.sk_state == SMC_LISTEN) { 2287 release_sock(&smc->sk); 2288 return -EINVAL; 2289 } 2290 if (smc->sk.sk_state == SMC_INIT || 2291 smc->sk.sk_state == SMC_CLOSED) { 2292 answ = 0; 2293 } else { 2294 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 2295 smc_curs_copy(&urg, &conn->urg_curs, conn); 2296 answ = smc_curs_diff(conn->rmb_desc->len, 2297 &cons, &urg) == 1; 2298 } 2299 break; 2300 default: 2301 release_sock(&smc->sk); 2302 return -ENOIOCTLCMD; 2303 } 2304 release_sock(&smc->sk); 2305 2306 return put_user(answ, (int __user *)arg); 2307 } 2308 2309 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 2310 int offset, size_t size, int flags) 2311 { 2312 struct sock *sk = sock->sk; 2313 struct smc_sock *smc; 2314 int rc = -EPIPE; 2315 2316 smc = smc_sk(sk); 2317 lock_sock(sk); 2318 if (sk->sk_state != SMC_ACTIVE) { 2319 release_sock(sk); 2320 goto out; 2321 } 2322 release_sock(sk); 2323 if (smc->use_fallback) 2324 rc = kernel_sendpage(smc->clcsock, page, offset, 2325 size, flags); 2326 else 2327 rc = sock_no_sendpage(sock, page, offset, size, flags); 2328 2329 out: 2330 return rc; 2331 } 2332 2333 /* Map the affected portions of the rmbe into an spd, note the number of bytes 2334 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 2335 * updates till whenever a respective page has been fully processed. 2336 * Note that subsequent recv() calls have to wait till all splice() processing 2337 * completed. 2338 */ 2339 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 2340 struct pipe_inode_info *pipe, size_t len, 2341 unsigned int flags) 2342 { 2343 struct sock *sk = sock->sk; 2344 struct smc_sock *smc; 2345 int rc = -ENOTCONN; 2346 2347 smc = smc_sk(sk); 2348 lock_sock(sk); 2349 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2350 /* socket was connected before, no more data to read */ 2351 rc = 0; 2352 goto out; 2353 } 2354 if (sk->sk_state == SMC_INIT || 2355 sk->sk_state == SMC_LISTEN || 2356 sk->sk_state == SMC_CLOSED) 2357 goto out; 2358 2359 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2360 rc = 0; 2361 goto out; 2362 } 2363 2364 if (smc->use_fallback) { 2365 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 2366 pipe, len, flags); 2367 } else { 2368 if (*ppos) { 2369 rc = -ESPIPE; 2370 goto out; 2371 } 2372 if (flags & SPLICE_F_NONBLOCK) 2373 flags = MSG_DONTWAIT; 2374 else 2375 flags = 0; 2376 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 2377 } 2378 out: 2379 release_sock(sk); 2380 2381 return rc; 2382 } 2383 2384 /* must look like tcp */ 2385 static const struct proto_ops smc_sock_ops = { 2386 .family = PF_SMC, 2387 .owner = THIS_MODULE, 2388 .release = smc_release, 2389 .bind = smc_bind, 2390 .connect = smc_connect, 2391 .socketpair = sock_no_socketpair, 2392 .accept = smc_accept, 2393 .getname = smc_getname, 2394 .poll = smc_poll, 2395 .ioctl = smc_ioctl, 2396 .listen = smc_listen, 2397 .shutdown = smc_shutdown, 2398 .setsockopt = smc_setsockopt, 2399 .getsockopt = smc_getsockopt, 2400 .sendmsg = smc_sendmsg, 2401 .recvmsg = smc_recvmsg, 2402 .mmap = sock_no_mmap, 2403 .sendpage = smc_sendpage, 2404 .splice_read = smc_splice_read, 2405 }; 2406 2407 static int smc_create(struct net *net, struct socket *sock, int protocol, 2408 int kern) 2409 { 2410 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 2411 struct smc_sock *smc; 2412 struct sock *sk; 2413 int rc; 2414 2415 rc = -ESOCKTNOSUPPORT; 2416 if (sock->type != SOCK_STREAM) 2417 goto out; 2418 2419 rc = -EPROTONOSUPPORT; 2420 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 2421 goto out; 2422 2423 rc = -ENOBUFS; 2424 sock->ops = &smc_sock_ops; 2425 sk = smc_sock_alloc(net, sock, protocol); 2426 if (!sk) 2427 goto out; 2428 2429 /* create internal TCP socket for CLC handshake and fallback */ 2430 smc = smc_sk(sk); 2431 smc->use_fallback = false; /* assume rdma capability first */ 2432 smc->fallback_rsn = 0; 2433 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 2434 &smc->clcsock); 2435 if (rc) { 2436 sk_common_release(sk); 2437 goto out; 2438 } 2439 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2440 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2441 2442 out: 2443 return rc; 2444 } 2445 2446 static const struct net_proto_family smc_sock_family_ops = { 2447 .family = PF_SMC, 2448 .owner = THIS_MODULE, 2449 .create = smc_create, 2450 }; 2451 2452 unsigned int smc_net_id; 2453 2454 static __net_init int smc_net_init(struct net *net) 2455 { 2456 return smc_pnet_net_init(net); 2457 } 2458 2459 static void __net_exit smc_net_exit(struct net *net) 2460 { 2461 smc_pnet_net_exit(net); 2462 } 2463 2464 static struct pernet_operations smc_net_ops = { 2465 .init = smc_net_init, 2466 .exit = smc_net_exit, 2467 .id = &smc_net_id, 2468 .size = sizeof(struct smc_net), 2469 }; 2470 2471 static int __init smc_init(void) 2472 { 2473 int rc; 2474 2475 rc = register_pernet_subsys(&smc_net_ops); 2476 if (rc) 2477 return rc; 2478 2479 smc_ism_init(); 2480 smc_clc_init(); 2481 2482 rc = smc_pnet_init(); 2483 if (rc) 2484 goto out_pernet_subsys; 2485 2486 rc = -ENOMEM; 2487 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 2488 if (!smc_hs_wq) 2489 goto out_pnet; 2490 2491 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 2492 if (!smc_close_wq) 2493 goto out_alloc_hs_wq; 2494 2495 rc = smc_core_init(); 2496 if (rc) { 2497 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2498 goto out_alloc_wqs; 2499 } 2500 2501 rc = smc_llc_init(); 2502 if (rc) { 2503 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2504 goto out_core; 2505 } 2506 2507 rc = smc_cdc_init(); 2508 if (rc) { 2509 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2510 goto out_core; 2511 } 2512 2513 rc = proto_register(&smc_proto, 1); 2514 if (rc) { 2515 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2516 goto out_core; 2517 } 2518 2519 rc = proto_register(&smc_proto6, 1); 2520 if (rc) { 2521 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2522 goto out_proto; 2523 } 2524 2525 rc = sock_register(&smc_sock_family_ops); 2526 if (rc) { 2527 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2528 goto out_proto6; 2529 } 2530 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2531 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2532 2533 rc = smc_ib_register_client(); 2534 if (rc) { 2535 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2536 goto out_sock; 2537 } 2538 2539 static_branch_enable(&tcp_have_smc); 2540 return 0; 2541 2542 out_sock: 2543 sock_unregister(PF_SMC); 2544 out_proto6: 2545 proto_unregister(&smc_proto6); 2546 out_proto: 2547 proto_unregister(&smc_proto); 2548 out_core: 2549 smc_core_exit(); 2550 out_alloc_wqs: 2551 destroy_workqueue(smc_close_wq); 2552 out_alloc_hs_wq: 2553 destroy_workqueue(smc_hs_wq); 2554 out_pnet: 2555 smc_pnet_exit(); 2556 out_pernet_subsys: 2557 unregister_pernet_subsys(&smc_net_ops); 2558 2559 return rc; 2560 } 2561 2562 static void __exit smc_exit(void) 2563 { 2564 static_branch_disable(&tcp_have_smc); 2565 sock_unregister(PF_SMC); 2566 smc_core_exit(); 2567 smc_ib_unregister_client(); 2568 destroy_workqueue(smc_close_wq); 2569 destroy_workqueue(smc_hs_wq); 2570 proto_unregister(&smc_proto6); 2571 proto_unregister(&smc_proto); 2572 smc_pnet_exit(); 2573 unregister_pernet_subsys(&smc_net_ops); 2574 rcu_barrier(); 2575 } 2576 2577 module_init(smc_init); 2578 module_exit(smc_exit); 2579 2580 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2581 MODULE_DESCRIPTION("smc socket address family"); 2582 MODULE_LICENSE("GPL"); 2583 MODULE_ALIAS_NETPROTO(PF_SMC); 2584