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_netlink.h" 49 #include "smc_tx.h" 50 #include "smc_rx.h" 51 #include "smc_close.h" 52 #include "smc_stats.h" 53 #include "smc_tracepoint.h" 54 55 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 56 * creation on server 57 */ 58 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 59 * creation on client 60 */ 61 62 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 63 struct workqueue_struct *smc_close_wq; /* wq for close work */ 64 65 static void smc_tcp_listen_work(struct work_struct *); 66 static void smc_connect_work(struct work_struct *); 67 68 static void smc_set_keepalive(struct sock *sk, int val) 69 { 70 struct smc_sock *smc = smc_sk(sk); 71 72 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 73 } 74 75 static struct smc_hashinfo smc_v4_hashinfo = { 76 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 77 }; 78 79 static struct smc_hashinfo smc_v6_hashinfo = { 80 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 81 }; 82 83 int smc_hash_sk(struct sock *sk) 84 { 85 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 86 struct hlist_head *head; 87 88 head = &h->ht; 89 90 write_lock_bh(&h->lock); 91 sk_add_node(sk, head); 92 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 93 write_unlock_bh(&h->lock); 94 95 return 0; 96 } 97 EXPORT_SYMBOL_GPL(smc_hash_sk); 98 99 void smc_unhash_sk(struct sock *sk) 100 { 101 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 102 103 write_lock_bh(&h->lock); 104 if (sk_del_node_init(sk)) 105 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 106 write_unlock_bh(&h->lock); 107 } 108 EXPORT_SYMBOL_GPL(smc_unhash_sk); 109 110 struct proto smc_proto = { 111 .name = "SMC", 112 .owner = THIS_MODULE, 113 .keepalive = smc_set_keepalive, 114 .hash = smc_hash_sk, 115 .unhash = smc_unhash_sk, 116 .obj_size = sizeof(struct smc_sock), 117 .h.smc_hash = &smc_v4_hashinfo, 118 .slab_flags = SLAB_TYPESAFE_BY_RCU, 119 }; 120 EXPORT_SYMBOL_GPL(smc_proto); 121 122 struct proto smc_proto6 = { 123 .name = "SMC6", 124 .owner = THIS_MODULE, 125 .keepalive = smc_set_keepalive, 126 .hash = smc_hash_sk, 127 .unhash = smc_unhash_sk, 128 .obj_size = sizeof(struct smc_sock), 129 .h.smc_hash = &smc_v6_hashinfo, 130 .slab_flags = SLAB_TYPESAFE_BY_RCU, 131 }; 132 EXPORT_SYMBOL_GPL(smc_proto6); 133 134 static void smc_restore_fallback_changes(struct smc_sock *smc) 135 { 136 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 137 smc->clcsock->file->private_data = smc->sk.sk_socket; 138 smc->clcsock->file = NULL; 139 } 140 } 141 142 static int __smc_release(struct smc_sock *smc) 143 { 144 struct sock *sk = &smc->sk; 145 int rc = 0; 146 147 if (!smc->use_fallback) { 148 rc = smc_close_active(smc); 149 sock_set_flag(sk, SOCK_DEAD); 150 sk->sk_shutdown |= SHUTDOWN_MASK; 151 } else { 152 if (sk->sk_state != SMC_CLOSED) { 153 if (sk->sk_state != SMC_LISTEN && 154 sk->sk_state != SMC_INIT) 155 sock_put(sk); /* passive closing */ 156 if (sk->sk_state == SMC_LISTEN) { 157 /* wake up clcsock accept */ 158 rc = kernel_sock_shutdown(smc->clcsock, 159 SHUT_RDWR); 160 } 161 sk->sk_state = SMC_CLOSED; 162 sk->sk_state_change(sk); 163 } 164 smc_restore_fallback_changes(smc); 165 } 166 167 sk->sk_prot->unhash(sk); 168 169 if (sk->sk_state == SMC_CLOSED) { 170 if (smc->clcsock) { 171 release_sock(sk); 172 smc_clcsock_release(smc); 173 lock_sock(sk); 174 } 175 if (!smc->use_fallback) 176 smc_conn_free(&smc->conn); 177 } 178 179 return rc; 180 } 181 182 static int smc_release(struct socket *sock) 183 { 184 struct sock *sk = sock->sk; 185 struct smc_sock *smc; 186 int rc = 0; 187 188 if (!sk) 189 goto out; 190 191 sock_hold(sk); /* sock_put below */ 192 smc = smc_sk(sk); 193 194 /* cleanup for a dangling non-blocking connect */ 195 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 196 tcp_abort(smc->clcsock->sk, ECONNABORTED); 197 flush_work(&smc->connect_work); 198 199 if (sk->sk_state == SMC_LISTEN) 200 /* smc_close_non_accepted() is called and acquires 201 * sock lock for child sockets again 202 */ 203 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 204 else 205 lock_sock(sk); 206 207 rc = __smc_release(smc); 208 209 /* detach socket */ 210 sock_orphan(sk); 211 sock->sk = NULL; 212 release_sock(sk); 213 214 sock_put(sk); /* sock_hold above */ 215 sock_put(sk); /* final sock_put */ 216 out: 217 return rc; 218 } 219 220 static void smc_destruct(struct sock *sk) 221 { 222 if (sk->sk_state != SMC_CLOSED) 223 return; 224 if (!sock_flag(sk, SOCK_DEAD)) 225 return; 226 227 sk_refcnt_debug_dec(sk); 228 } 229 230 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 231 int protocol) 232 { 233 struct smc_sock *smc; 234 struct proto *prot; 235 struct sock *sk; 236 237 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 238 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 239 if (!sk) 240 return NULL; 241 242 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 243 sk->sk_state = SMC_INIT; 244 sk->sk_destruct = smc_destruct; 245 sk->sk_protocol = protocol; 246 smc = smc_sk(sk); 247 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 248 INIT_WORK(&smc->connect_work, smc_connect_work); 249 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 250 INIT_LIST_HEAD(&smc->accept_q); 251 spin_lock_init(&smc->accept_q_lock); 252 spin_lock_init(&smc->conn.send_lock); 253 sk->sk_prot->hash(sk); 254 sk_refcnt_debug_inc(sk); 255 mutex_init(&smc->clcsock_release_lock); 256 257 return sk; 258 } 259 260 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 261 int addr_len) 262 { 263 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 264 struct sock *sk = sock->sk; 265 struct smc_sock *smc; 266 int rc; 267 268 smc = smc_sk(sk); 269 270 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 271 rc = -EINVAL; 272 if (addr_len < sizeof(struct sockaddr_in)) 273 goto out; 274 275 rc = -EAFNOSUPPORT; 276 if (addr->sin_family != AF_INET && 277 addr->sin_family != AF_INET6 && 278 addr->sin_family != AF_UNSPEC) 279 goto out; 280 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 281 if (addr->sin_family == AF_UNSPEC && 282 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 283 goto out; 284 285 lock_sock(sk); 286 287 /* Check if socket is already active */ 288 rc = -EINVAL; 289 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 290 goto out_rel; 291 292 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 293 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 294 295 out_rel: 296 release_sock(sk); 297 out: 298 return rc; 299 } 300 301 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 302 unsigned long mask) 303 { 304 /* options we don't get control via setsockopt for */ 305 nsk->sk_type = osk->sk_type; 306 nsk->sk_sndbuf = osk->sk_sndbuf; 307 nsk->sk_rcvbuf = osk->sk_rcvbuf; 308 nsk->sk_sndtimeo = osk->sk_sndtimeo; 309 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 310 nsk->sk_mark = osk->sk_mark; 311 nsk->sk_priority = osk->sk_priority; 312 nsk->sk_rcvlowat = osk->sk_rcvlowat; 313 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 314 nsk->sk_err = osk->sk_err; 315 316 nsk->sk_flags &= ~mask; 317 nsk->sk_flags |= osk->sk_flags & mask; 318 } 319 320 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 321 (1UL << SOCK_KEEPOPEN) | \ 322 (1UL << SOCK_LINGER) | \ 323 (1UL << SOCK_BROADCAST) | \ 324 (1UL << SOCK_TIMESTAMP) | \ 325 (1UL << SOCK_DBG) | \ 326 (1UL << SOCK_RCVTSTAMP) | \ 327 (1UL << SOCK_RCVTSTAMPNS) | \ 328 (1UL << SOCK_LOCALROUTE) | \ 329 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 330 (1UL << SOCK_RXQ_OVFL) | \ 331 (1UL << SOCK_WIFI_STATUS) | \ 332 (1UL << SOCK_NOFCS) | \ 333 (1UL << SOCK_FILTER_LOCKED) | \ 334 (1UL << SOCK_TSTAMP_NEW)) 335 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 336 * clc socket (since smc is not called for these options from net/core) 337 */ 338 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 339 { 340 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 341 } 342 343 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 344 (1UL << SOCK_KEEPOPEN) | \ 345 (1UL << SOCK_LINGER) | \ 346 (1UL << SOCK_DBG)) 347 /* copy only settings and flags relevant for smc from clc to smc socket */ 348 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 349 { 350 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 351 } 352 353 /* register the new rmb on all links */ 354 static int smcr_lgr_reg_rmbs(struct smc_link *link, 355 struct smc_buf_desc *rmb_desc) 356 { 357 struct smc_link_group *lgr = link->lgr; 358 int i, rc = 0; 359 360 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 361 if (rc) 362 return rc; 363 /* protect against parallel smc_llc_cli_rkey_exchange() and 364 * parallel smcr_link_reg_rmb() 365 */ 366 mutex_lock(&lgr->llc_conf_mutex); 367 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 368 if (!smc_link_active(&lgr->lnk[i])) 369 continue; 370 rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc); 371 if (rc) 372 goto out; 373 } 374 375 /* exchange confirm_rkey msg with peer */ 376 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 377 if (rc) { 378 rc = -EFAULT; 379 goto out; 380 } 381 rmb_desc->is_conf_rkey = true; 382 out: 383 mutex_unlock(&lgr->llc_conf_mutex); 384 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 385 return rc; 386 } 387 388 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 389 { 390 struct smc_link *link = smc->conn.lnk; 391 struct smc_llc_qentry *qentry; 392 int rc; 393 394 /* receive CONFIRM LINK request from server over RoCE fabric */ 395 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 396 SMC_LLC_CONFIRM_LINK); 397 if (!qentry) { 398 struct smc_clc_msg_decline dclc; 399 400 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 401 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 402 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 403 } 404 smc_llc_save_peer_uid(qentry); 405 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 406 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 407 if (rc) 408 return SMC_CLC_DECL_RMBE_EC; 409 410 rc = smc_ib_modify_qp_rts(link); 411 if (rc) 412 return SMC_CLC_DECL_ERR_RDYLNK; 413 414 smc_wr_remember_qp_attr(link); 415 416 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 417 return SMC_CLC_DECL_ERR_REGRMB; 418 419 /* confirm_rkey is implicit on 1st contact */ 420 smc->conn.rmb_desc->is_conf_rkey = true; 421 422 /* send CONFIRM LINK response over RoCE fabric */ 423 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 424 if (rc < 0) 425 return SMC_CLC_DECL_TIMEOUT_CL; 426 427 smc_llc_link_active(link); 428 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 429 430 /* optional 2nd link, receive ADD LINK request from server */ 431 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 432 SMC_LLC_ADD_LINK); 433 if (!qentry) { 434 struct smc_clc_msg_decline dclc; 435 436 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 437 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 438 if (rc == -EAGAIN) 439 rc = 0; /* no DECLINE received, go with one link */ 440 return rc; 441 } 442 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 443 smc_llc_cli_add_link(link, qentry); 444 return 0; 445 } 446 447 static bool smc_isascii(char *hostname) 448 { 449 int i; 450 451 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 452 if (!isascii(hostname[i])) 453 return false; 454 return true; 455 } 456 457 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 458 struct smc_clc_msg_accept_confirm *clc) 459 { 460 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 461 (struct smc_clc_msg_accept_confirm_v2 *)clc; 462 struct smc_clc_first_contact_ext *fce; 463 int clc_v2_len; 464 465 if (clc->hdr.version == SMC_V1 || 466 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 467 return; 468 469 if (smc->conn.lgr->is_smcd) { 470 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 471 SMC_MAX_EID_LEN); 472 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 473 d1); 474 } else { 475 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 476 SMC_MAX_EID_LEN); 477 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 478 r1); 479 } 480 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 481 smc->conn.lgr->peer_os = fce->os_type; 482 smc->conn.lgr->peer_smc_release = fce->release; 483 if (smc_isascii(fce->hostname)) 484 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 485 SMC_MAX_HOSTNAME_LEN); 486 } 487 488 static void smcr_conn_save_peer_info(struct smc_sock *smc, 489 struct smc_clc_msg_accept_confirm *clc) 490 { 491 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 492 493 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 494 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 495 smc->conn.peer_rmbe_size = bufsize; 496 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 497 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 498 } 499 500 static void smcd_conn_save_peer_info(struct smc_sock *smc, 501 struct smc_clc_msg_accept_confirm *clc) 502 { 503 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 504 505 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 506 smc->conn.peer_token = clc->d0.token; 507 /* msg header takes up space in the buffer */ 508 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 509 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 510 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 511 } 512 513 static void smc_conn_save_peer_info(struct smc_sock *smc, 514 struct smc_clc_msg_accept_confirm *clc) 515 { 516 if (smc->conn.lgr->is_smcd) 517 smcd_conn_save_peer_info(smc, clc); 518 else 519 smcr_conn_save_peer_info(smc, clc); 520 smc_conn_save_peer_info_fce(smc, clc); 521 } 522 523 static void smc_link_save_peer_info(struct smc_link *link, 524 struct smc_clc_msg_accept_confirm *clc, 525 struct smc_init_info *ini) 526 { 527 link->peer_qpn = ntoh24(clc->r0.qpn); 528 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 529 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 530 link->peer_psn = ntoh24(clc->r0.psn); 531 link->peer_mtu = clc->r0.qp_mtu; 532 } 533 534 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 535 struct smc_stats_fback *fback_arr) 536 { 537 int cnt; 538 539 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 540 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 541 fback_arr[cnt].count++; 542 break; 543 } 544 if (!fback_arr[cnt].fback_code) { 545 fback_arr[cnt].fback_code = smc->fallback_rsn; 546 fback_arr[cnt].count++; 547 break; 548 } 549 } 550 } 551 552 static void smc_stat_fallback(struct smc_sock *smc) 553 { 554 struct net *net = sock_net(&smc->sk); 555 556 mutex_lock(&net->smc.mutex_fback_rsn); 557 if (smc->listen_smc) { 558 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 559 net->smc.fback_rsn->srv_fback_cnt++; 560 } else { 561 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 562 net->smc.fback_rsn->clnt_fback_cnt++; 563 } 564 mutex_unlock(&net->smc.mutex_fback_rsn); 565 } 566 567 static void smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 568 { 569 wait_queue_head_t *smc_wait = sk_sleep(&smc->sk); 570 wait_queue_head_t *clc_wait = sk_sleep(smc->clcsock->sk); 571 unsigned long flags; 572 573 smc->use_fallback = true; 574 smc->fallback_rsn = reason_code; 575 smc_stat_fallback(smc); 576 trace_smc_switch_to_fallback(smc, reason_code); 577 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 578 smc->clcsock->file = smc->sk.sk_socket->file; 579 smc->clcsock->file->private_data = smc->clcsock; 580 smc->clcsock->wq.fasync_list = 581 smc->sk.sk_socket->wq.fasync_list; 582 583 /* There may be some entries remaining in 584 * smc socket->wq, which should be removed 585 * to clcsocket->wq during the fallback. 586 */ 587 spin_lock_irqsave(&smc_wait->lock, flags); 588 spin_lock_nested(&clc_wait->lock, SINGLE_DEPTH_NESTING); 589 list_splice_init(&smc_wait->head, &clc_wait->head); 590 spin_unlock(&clc_wait->lock); 591 spin_unlock_irqrestore(&smc_wait->lock, flags); 592 } 593 } 594 595 /* fall back during connect */ 596 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 597 { 598 smc_switch_to_fallback(smc, reason_code); 599 smc_copy_sock_settings_to_clc(smc); 600 smc->connect_nonblock = 0; 601 if (smc->sk.sk_state == SMC_INIT) 602 smc->sk.sk_state = SMC_ACTIVE; 603 return 0; 604 } 605 606 /* decline and fall back during connect */ 607 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 608 u8 version) 609 { 610 struct net *net = sock_net(&smc->sk); 611 int rc; 612 613 if (reason_code < 0) { /* error, fallback is not possible */ 614 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 615 if (smc->sk.sk_state == SMC_INIT) 616 sock_put(&smc->sk); /* passive closing */ 617 return reason_code; 618 } 619 if (reason_code != SMC_CLC_DECL_PEERDECL) { 620 rc = smc_clc_send_decline(smc, reason_code, version); 621 if (rc < 0) { 622 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 623 if (smc->sk.sk_state == SMC_INIT) 624 sock_put(&smc->sk); /* passive closing */ 625 return rc; 626 } 627 } 628 return smc_connect_fallback(smc, reason_code); 629 } 630 631 static void smc_conn_abort(struct smc_sock *smc, int local_first) 632 { 633 if (local_first) 634 smc_lgr_cleanup_early(&smc->conn); 635 else 636 smc_conn_free(&smc->conn); 637 } 638 639 /* check if there is a rdma device available for this connection. */ 640 /* called for connect and listen */ 641 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 642 { 643 /* PNET table look up: search active ib_device and port 644 * within same PNETID that also contains the ethernet device 645 * used for the internal TCP socket 646 */ 647 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 648 if (!ini->check_smcrv2 && !ini->ib_dev) 649 return SMC_CLC_DECL_NOSMCRDEV; 650 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 651 return SMC_CLC_DECL_NOSMCRDEV; 652 return 0; 653 } 654 655 /* check if there is an ISM device available for this connection. */ 656 /* called for connect and listen */ 657 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 658 { 659 /* Find ISM device with same PNETID as connecting interface */ 660 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 661 if (!ini->ism_dev[0]) 662 return SMC_CLC_DECL_NOSMCDDEV; 663 else 664 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 665 return 0; 666 } 667 668 /* is chid unique for the ism devices that are already determined? */ 669 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 670 int cnt) 671 { 672 int i = (!ini->ism_dev[0]) ? 1 : 0; 673 674 for (; i < cnt; i++) 675 if (ini->ism_chid[i] == chid) 676 return false; 677 return true; 678 } 679 680 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 681 * PNETID matching net_device) 682 */ 683 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 684 struct smc_init_info *ini) 685 { 686 int rc = SMC_CLC_DECL_NOSMCDDEV; 687 struct smcd_dev *smcd; 688 int i = 1; 689 u16 chid; 690 691 if (smcd_indicated(ini->smc_type_v1)) 692 rc = 0; /* already initialized for V1 */ 693 mutex_lock(&smcd_dev_list.mutex); 694 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 695 if (smcd->going_away || smcd == ini->ism_dev[0]) 696 continue; 697 chid = smc_ism_get_chid(smcd); 698 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 699 continue; 700 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 701 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 702 ini->ism_dev[i] = smcd; 703 ini->ism_chid[i] = chid; 704 ini->is_smcd = true; 705 rc = 0; 706 i++; 707 if (i > SMC_MAX_ISM_DEVS) 708 break; 709 } 710 } 711 mutex_unlock(&smcd_dev_list.mutex); 712 ini->ism_offered_cnt = i - 1; 713 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 714 ini->smcd_version = 0; 715 716 return rc; 717 } 718 719 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 720 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 721 struct smc_init_info *ini) 722 { 723 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 724 return SMC_CLC_DECL_ISMVLANERR; 725 return 0; 726 } 727 728 static int smc_find_proposal_devices(struct smc_sock *smc, 729 struct smc_init_info *ini) 730 { 731 int rc = 0; 732 733 /* check if there is an ism device available */ 734 if (!(ini->smcd_version & SMC_V1) || 735 smc_find_ism_device(smc, ini) || 736 smc_connect_ism_vlan_setup(smc, ini)) 737 ini->smcd_version &= ~SMC_V1; 738 /* else ISM V1 is supported for this connection */ 739 740 /* check if there is an rdma device available */ 741 if (!(ini->smcr_version & SMC_V1) || 742 smc_find_rdma_device(smc, ini)) 743 ini->smcr_version &= ~SMC_V1; 744 /* else RDMA is supported for this connection */ 745 746 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 747 ini->smcr_version & SMC_V1); 748 749 /* check if there is an ism v2 device available */ 750 if (!(ini->smcd_version & SMC_V2) || 751 !smc_ism_is_v2_capable() || 752 smc_find_ism_v2_device_clnt(smc, ini)) 753 ini->smcd_version &= ~SMC_V2; 754 755 /* check if there is an rdma v2 device available */ 756 ini->check_smcrv2 = true; 757 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 758 if (!(ini->smcr_version & SMC_V2) || 759 smc->clcsock->sk->sk_family != AF_INET || 760 !smc_clc_ueid_count() || 761 smc_find_rdma_device(smc, ini)) 762 ini->smcr_version &= ~SMC_V2; 763 ini->check_smcrv2 = false; 764 765 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 766 ini->smcr_version & SMC_V2); 767 768 /* if neither ISM nor RDMA are supported, fallback */ 769 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 770 rc = SMC_CLC_DECL_NOSMCDEV; 771 772 return rc; 773 } 774 775 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 776 * used, the VLAN ID will be registered again during the connection setup. 777 */ 778 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 779 struct smc_init_info *ini) 780 { 781 if (!smcd_indicated(ini->smc_type_v1)) 782 return 0; 783 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 784 return SMC_CLC_DECL_CNFERR; 785 return 0; 786 } 787 788 #define SMC_CLC_MAX_ACCEPT_LEN \ 789 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 790 sizeof(struct smc_clc_first_contact_ext) + \ 791 sizeof(struct smc_clc_msg_trail)) 792 793 /* CLC handshake during connect */ 794 static int smc_connect_clc(struct smc_sock *smc, 795 struct smc_clc_msg_accept_confirm_v2 *aclc2, 796 struct smc_init_info *ini) 797 { 798 int rc = 0; 799 800 /* do inband token exchange */ 801 rc = smc_clc_send_proposal(smc, ini); 802 if (rc) 803 return rc; 804 /* receive SMC Accept CLC message */ 805 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 806 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 807 } 808 809 void smc_fill_gid_list(struct smc_link_group *lgr, 810 struct smc_gidlist *gidlist, 811 struct smc_ib_device *known_dev, u8 *known_gid) 812 { 813 struct smc_init_info *alt_ini = NULL; 814 815 memset(gidlist, 0, sizeof(*gidlist)); 816 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 817 818 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 819 if (!alt_ini) 820 goto out; 821 822 alt_ini->vlan_id = lgr->vlan_id; 823 alt_ini->check_smcrv2 = true; 824 alt_ini->smcrv2.saddr = lgr->saddr; 825 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 826 827 if (!alt_ini->smcrv2.ib_dev_v2) 828 goto out; 829 830 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 831 SMC_GID_SIZE); 832 833 out: 834 kfree(alt_ini); 835 } 836 837 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 838 struct smc_clc_msg_accept_confirm *aclc, 839 struct smc_init_info *ini) 840 { 841 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 842 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 843 struct smc_clc_first_contact_ext *fce = 844 (struct smc_clc_first_contact_ext *) 845 (((u8 *)clc_v2) + sizeof(*clc_v2)); 846 847 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 848 return 0; 849 850 if (fce->v2_direct) { 851 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 852 ini->smcrv2.uses_gateway = false; 853 } else { 854 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr, 855 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 856 ini->smcrv2.nexthop_mac, 857 &ini->smcrv2.uses_gateway)) 858 return SMC_CLC_DECL_NOROUTE; 859 if (!ini->smcrv2.uses_gateway) { 860 /* mismatch: peer claims indirect, but its direct */ 861 return SMC_CLC_DECL_NOINDIRECT; 862 } 863 } 864 return 0; 865 } 866 867 /* setup for RDMA connection of client */ 868 static int smc_connect_rdma(struct smc_sock *smc, 869 struct smc_clc_msg_accept_confirm *aclc, 870 struct smc_init_info *ini) 871 { 872 int i, reason_code = 0; 873 struct smc_link *link; 874 u8 *eid = NULL; 875 876 ini->is_smcd = false; 877 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 878 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 879 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 880 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 881 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 882 883 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 884 if (reason_code) 885 return reason_code; 886 887 mutex_lock(&smc_client_lgr_pending); 888 reason_code = smc_conn_create(smc, ini); 889 if (reason_code) { 890 mutex_unlock(&smc_client_lgr_pending); 891 return reason_code; 892 } 893 894 smc_conn_save_peer_info(smc, aclc); 895 896 if (ini->first_contact_local) { 897 link = smc->conn.lnk; 898 } else { 899 /* set link that was assigned by server */ 900 link = NULL; 901 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 902 struct smc_link *l = &smc->conn.lgr->lnk[i]; 903 904 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 905 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 906 SMC_GID_SIZE) && 907 (aclc->hdr.version > SMC_V1 || 908 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 909 sizeof(l->peer_mac)))) { 910 link = l; 911 break; 912 } 913 } 914 if (!link) { 915 reason_code = SMC_CLC_DECL_NOSRVLINK; 916 goto connect_abort; 917 } 918 smc_switch_link_and_count(&smc->conn, link); 919 } 920 921 /* create send buffer and rmb */ 922 if (smc_buf_create(smc, false)) { 923 reason_code = SMC_CLC_DECL_MEM; 924 goto connect_abort; 925 } 926 927 if (ini->first_contact_local) 928 smc_link_save_peer_info(link, aclc, ini); 929 930 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 931 reason_code = SMC_CLC_DECL_ERR_RTOK; 932 goto connect_abort; 933 } 934 935 smc_close_init(smc); 936 smc_rx_init(smc); 937 938 if (ini->first_contact_local) { 939 if (smc_ib_ready_link(link)) { 940 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 941 goto connect_abort; 942 } 943 } else { 944 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 945 reason_code = SMC_CLC_DECL_ERR_REGRMB; 946 goto connect_abort; 947 } 948 } 949 smc_rmb_sync_sg_for_device(&smc->conn); 950 951 if (aclc->hdr.version > SMC_V1) { 952 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 953 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 954 955 eid = clc_v2->r1.eid; 956 if (ini->first_contact_local) 957 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 958 link->smcibdev, link->gid); 959 } 960 961 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 962 aclc->hdr.version, eid, ini); 963 if (reason_code) 964 goto connect_abort; 965 966 smc_tx_init(smc); 967 968 if (ini->first_contact_local) { 969 /* QP confirmation over RoCE fabric */ 970 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 971 reason_code = smcr_clnt_conf_first_link(smc); 972 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 973 if (reason_code) 974 goto connect_abort; 975 } 976 mutex_unlock(&smc_client_lgr_pending); 977 978 smc_copy_sock_settings_to_clc(smc); 979 smc->connect_nonblock = 0; 980 if (smc->sk.sk_state == SMC_INIT) 981 smc->sk.sk_state = SMC_ACTIVE; 982 983 return 0; 984 connect_abort: 985 smc_conn_abort(smc, ini->first_contact_local); 986 mutex_unlock(&smc_client_lgr_pending); 987 smc->connect_nonblock = 0; 988 989 return reason_code; 990 } 991 992 /* The server has chosen one of the proposed ISM devices for the communication. 993 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 994 */ 995 static int 996 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 997 struct smc_init_info *ini) 998 { 999 int i; 1000 1001 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1002 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1003 ini->ism_selected = i; 1004 return 0; 1005 } 1006 } 1007 1008 return -EPROTO; 1009 } 1010 1011 /* setup for ISM connection of client */ 1012 static int smc_connect_ism(struct smc_sock *smc, 1013 struct smc_clc_msg_accept_confirm *aclc, 1014 struct smc_init_info *ini) 1015 { 1016 u8 *eid = NULL; 1017 int rc = 0; 1018 1019 ini->is_smcd = true; 1020 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1021 1022 if (aclc->hdr.version == SMC_V2) { 1023 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1024 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1025 1026 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1027 if (rc) 1028 return rc; 1029 } 1030 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 1031 1032 /* there is only one lgr role for SMC-D; use server lock */ 1033 mutex_lock(&smc_server_lgr_pending); 1034 rc = smc_conn_create(smc, ini); 1035 if (rc) { 1036 mutex_unlock(&smc_server_lgr_pending); 1037 return rc; 1038 } 1039 1040 /* Create send and receive buffers */ 1041 rc = smc_buf_create(smc, true); 1042 if (rc) { 1043 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1044 goto connect_abort; 1045 } 1046 1047 smc_conn_save_peer_info(smc, aclc); 1048 smc_close_init(smc); 1049 smc_rx_init(smc); 1050 smc_tx_init(smc); 1051 1052 if (aclc->hdr.version > SMC_V1) { 1053 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1054 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1055 1056 eid = clc_v2->d1.eid; 1057 } 1058 1059 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1060 aclc->hdr.version, eid, NULL); 1061 if (rc) 1062 goto connect_abort; 1063 mutex_unlock(&smc_server_lgr_pending); 1064 1065 smc_copy_sock_settings_to_clc(smc); 1066 smc->connect_nonblock = 0; 1067 if (smc->sk.sk_state == SMC_INIT) 1068 smc->sk.sk_state = SMC_ACTIVE; 1069 1070 return 0; 1071 connect_abort: 1072 smc_conn_abort(smc, ini->first_contact_local); 1073 mutex_unlock(&smc_server_lgr_pending); 1074 smc->connect_nonblock = 0; 1075 1076 return rc; 1077 } 1078 1079 /* check if received accept type and version matches a proposed one */ 1080 static int smc_connect_check_aclc(struct smc_init_info *ini, 1081 struct smc_clc_msg_accept_confirm *aclc) 1082 { 1083 if (aclc->hdr.typev1 != SMC_TYPE_R && 1084 aclc->hdr.typev1 != SMC_TYPE_D) 1085 return SMC_CLC_DECL_MODEUNSUPP; 1086 1087 if (aclc->hdr.version >= SMC_V2) { 1088 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1089 !smcr_indicated(ini->smc_type_v2)) || 1090 (aclc->hdr.typev1 == SMC_TYPE_D && 1091 !smcd_indicated(ini->smc_type_v2))) 1092 return SMC_CLC_DECL_MODEUNSUPP; 1093 } else { 1094 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1095 !smcr_indicated(ini->smc_type_v1)) || 1096 (aclc->hdr.typev1 == SMC_TYPE_D && 1097 !smcd_indicated(ini->smc_type_v1))) 1098 return SMC_CLC_DECL_MODEUNSUPP; 1099 } 1100 1101 return 0; 1102 } 1103 1104 /* perform steps before actually connecting */ 1105 static int __smc_connect(struct smc_sock *smc) 1106 { 1107 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1108 struct smc_clc_msg_accept_confirm_v2 *aclc2; 1109 struct smc_clc_msg_accept_confirm *aclc; 1110 struct smc_init_info *ini = NULL; 1111 u8 *buf = NULL; 1112 int rc = 0; 1113 1114 if (smc->use_fallback) 1115 return smc_connect_fallback(smc, smc->fallback_rsn); 1116 1117 /* if peer has not signalled SMC-capability, fall back */ 1118 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1119 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1120 1121 /* IPSec connections opt out of SMC optimizations */ 1122 if (using_ipsec(smc)) 1123 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1124 version); 1125 1126 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1127 if (!ini) 1128 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1129 version); 1130 1131 ini->smcd_version = SMC_V1 | SMC_V2; 1132 ini->smcr_version = SMC_V1 | SMC_V2; 1133 ini->smc_type_v1 = SMC_TYPE_B; 1134 ini->smc_type_v2 = SMC_TYPE_B; 1135 1136 /* get vlan id from IP device */ 1137 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1138 ini->smcd_version &= ~SMC_V1; 1139 ini->smcr_version = 0; 1140 ini->smc_type_v1 = SMC_TYPE_N; 1141 if (!ini->smcd_version) { 1142 rc = SMC_CLC_DECL_GETVLANERR; 1143 goto fallback; 1144 } 1145 } 1146 1147 rc = smc_find_proposal_devices(smc, ini); 1148 if (rc) 1149 goto fallback; 1150 1151 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1152 if (!buf) { 1153 rc = SMC_CLC_DECL_MEM; 1154 goto fallback; 1155 } 1156 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1157 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1158 1159 /* perform CLC handshake */ 1160 rc = smc_connect_clc(smc, aclc2, ini); 1161 if (rc) 1162 goto vlan_cleanup; 1163 1164 /* check if smc modes and versions of CLC proposal and accept match */ 1165 rc = smc_connect_check_aclc(ini, aclc); 1166 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1167 if (rc) 1168 goto vlan_cleanup; 1169 1170 /* depending on previous steps, connect using rdma or ism */ 1171 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1172 ini->smcr_version = version; 1173 rc = smc_connect_rdma(smc, aclc, ini); 1174 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1175 ini->smcd_version = version; 1176 rc = smc_connect_ism(smc, aclc, ini); 1177 } 1178 if (rc) 1179 goto vlan_cleanup; 1180 1181 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1182 smc_connect_ism_vlan_cleanup(smc, ini); 1183 kfree(buf); 1184 kfree(ini); 1185 return 0; 1186 1187 vlan_cleanup: 1188 smc_connect_ism_vlan_cleanup(smc, ini); 1189 kfree(buf); 1190 fallback: 1191 kfree(ini); 1192 return smc_connect_decline_fallback(smc, rc, version); 1193 } 1194 1195 static void smc_connect_work(struct work_struct *work) 1196 { 1197 struct smc_sock *smc = container_of(work, struct smc_sock, 1198 connect_work); 1199 long timeo = smc->sk.sk_sndtimeo; 1200 int rc = 0; 1201 1202 if (!timeo) 1203 timeo = MAX_SCHEDULE_TIMEOUT; 1204 lock_sock(smc->clcsock->sk); 1205 if (smc->clcsock->sk->sk_err) { 1206 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1207 } else if ((1 << smc->clcsock->sk->sk_state) & 1208 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1209 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1210 if ((rc == -EPIPE) && 1211 ((1 << smc->clcsock->sk->sk_state) & 1212 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1213 rc = 0; 1214 } 1215 release_sock(smc->clcsock->sk); 1216 lock_sock(&smc->sk); 1217 if (rc != 0 || smc->sk.sk_err) { 1218 smc->sk.sk_state = SMC_CLOSED; 1219 if (rc == -EPIPE || rc == -EAGAIN) 1220 smc->sk.sk_err = EPIPE; 1221 else if (signal_pending(current)) 1222 smc->sk.sk_err = -sock_intr_errno(timeo); 1223 sock_put(&smc->sk); /* passive closing */ 1224 goto out; 1225 } 1226 1227 rc = __smc_connect(smc); 1228 if (rc < 0) 1229 smc->sk.sk_err = -rc; 1230 1231 out: 1232 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1233 if (smc->sk.sk_err) { 1234 smc->sk.sk_state_change(&smc->sk); 1235 } else { /* allow polling before and after fallback decision */ 1236 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1237 smc->sk.sk_write_space(&smc->sk); 1238 } 1239 } 1240 release_sock(&smc->sk); 1241 } 1242 1243 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1244 int alen, int flags) 1245 { 1246 struct sock *sk = sock->sk; 1247 struct smc_sock *smc; 1248 int rc = -EINVAL; 1249 1250 smc = smc_sk(sk); 1251 1252 /* separate smc parameter checking to be safe */ 1253 if (alen < sizeof(addr->sa_family)) 1254 goto out_err; 1255 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1256 goto out_err; 1257 1258 lock_sock(sk); 1259 switch (sk->sk_state) { 1260 default: 1261 goto out; 1262 case SMC_ACTIVE: 1263 rc = -EISCONN; 1264 goto out; 1265 case SMC_INIT: 1266 break; 1267 } 1268 1269 smc_copy_sock_settings_to_clc(smc); 1270 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1271 if (smc->connect_nonblock) { 1272 rc = -EALREADY; 1273 goto out; 1274 } 1275 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1276 if (rc && rc != -EINPROGRESS) 1277 goto out; 1278 1279 sock_hold(&smc->sk); /* sock put in passive closing */ 1280 if (smc->use_fallback) 1281 goto out; 1282 if (flags & O_NONBLOCK) { 1283 if (queue_work(smc_hs_wq, &smc->connect_work)) 1284 smc->connect_nonblock = 1; 1285 rc = -EINPROGRESS; 1286 } else { 1287 rc = __smc_connect(smc); 1288 if (rc < 0) 1289 goto out; 1290 else 1291 rc = 0; /* success cases including fallback */ 1292 } 1293 1294 out: 1295 release_sock(sk); 1296 out_err: 1297 return rc; 1298 } 1299 1300 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1301 { 1302 struct socket *new_clcsock = NULL; 1303 struct sock *lsk = &lsmc->sk; 1304 struct sock *new_sk; 1305 int rc = -EINVAL; 1306 1307 release_sock(lsk); 1308 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1309 if (!new_sk) { 1310 rc = -ENOMEM; 1311 lsk->sk_err = ENOMEM; 1312 *new_smc = NULL; 1313 lock_sock(lsk); 1314 goto out; 1315 } 1316 *new_smc = smc_sk(new_sk); 1317 1318 mutex_lock(&lsmc->clcsock_release_lock); 1319 if (lsmc->clcsock) 1320 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1321 mutex_unlock(&lsmc->clcsock_release_lock); 1322 lock_sock(lsk); 1323 if (rc < 0 && rc != -EAGAIN) 1324 lsk->sk_err = -rc; 1325 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1326 new_sk->sk_prot->unhash(new_sk); 1327 if (new_clcsock) 1328 sock_release(new_clcsock); 1329 new_sk->sk_state = SMC_CLOSED; 1330 sock_set_flag(new_sk, SOCK_DEAD); 1331 sock_put(new_sk); /* final */ 1332 *new_smc = NULL; 1333 goto out; 1334 } 1335 1336 /* new clcsock has inherited the smc listen-specific sk_data_ready 1337 * function; switch it back to the original sk_data_ready function 1338 */ 1339 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1340 (*new_smc)->clcsock = new_clcsock; 1341 out: 1342 return rc; 1343 } 1344 1345 /* add a just created sock to the accept queue of the listen sock as 1346 * candidate for a following socket accept call from user space 1347 */ 1348 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1349 { 1350 struct smc_sock *par = smc_sk(parent); 1351 1352 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1353 spin_lock(&par->accept_q_lock); 1354 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1355 spin_unlock(&par->accept_q_lock); 1356 sk_acceptq_added(parent); 1357 } 1358 1359 /* remove a socket from the accept queue of its parental listening socket */ 1360 static void smc_accept_unlink(struct sock *sk) 1361 { 1362 struct smc_sock *par = smc_sk(sk)->listen_smc; 1363 1364 spin_lock(&par->accept_q_lock); 1365 list_del_init(&smc_sk(sk)->accept_q); 1366 spin_unlock(&par->accept_q_lock); 1367 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1368 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1369 } 1370 1371 /* remove a sock from the accept queue to bind it to a new socket created 1372 * for a socket accept call from user space 1373 */ 1374 struct sock *smc_accept_dequeue(struct sock *parent, 1375 struct socket *new_sock) 1376 { 1377 struct smc_sock *isk, *n; 1378 struct sock *new_sk; 1379 1380 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1381 new_sk = (struct sock *)isk; 1382 1383 smc_accept_unlink(new_sk); 1384 if (new_sk->sk_state == SMC_CLOSED) { 1385 new_sk->sk_prot->unhash(new_sk); 1386 if (isk->clcsock) { 1387 sock_release(isk->clcsock); 1388 isk->clcsock = NULL; 1389 } 1390 sock_put(new_sk); /* final */ 1391 continue; 1392 } 1393 if (new_sock) { 1394 sock_graft(new_sk, new_sock); 1395 if (isk->use_fallback) { 1396 smc_sk(new_sk)->clcsock->file = new_sock->file; 1397 isk->clcsock->file->private_data = isk->clcsock; 1398 } 1399 } 1400 return new_sk; 1401 } 1402 return NULL; 1403 } 1404 1405 /* clean up for a created but never accepted sock */ 1406 void smc_close_non_accepted(struct sock *sk) 1407 { 1408 struct smc_sock *smc = smc_sk(sk); 1409 1410 sock_hold(sk); /* sock_put below */ 1411 lock_sock(sk); 1412 if (!sk->sk_lingertime) 1413 /* wait for peer closing */ 1414 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1415 __smc_release(smc); 1416 release_sock(sk); 1417 sock_put(sk); /* sock_hold above */ 1418 sock_put(sk); /* final sock_put */ 1419 } 1420 1421 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1422 { 1423 struct smc_link *link = smc->conn.lnk; 1424 struct smc_llc_qentry *qentry; 1425 int rc; 1426 1427 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 1428 return SMC_CLC_DECL_ERR_REGRMB; 1429 1430 /* send CONFIRM LINK request to client over the RoCE fabric */ 1431 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1432 if (rc < 0) 1433 return SMC_CLC_DECL_TIMEOUT_CL; 1434 1435 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1436 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1437 SMC_LLC_CONFIRM_LINK); 1438 if (!qentry) { 1439 struct smc_clc_msg_decline dclc; 1440 1441 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1442 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1443 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1444 } 1445 smc_llc_save_peer_uid(qentry); 1446 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1447 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1448 if (rc) 1449 return SMC_CLC_DECL_RMBE_EC; 1450 1451 /* confirm_rkey is implicit on 1st contact */ 1452 smc->conn.rmb_desc->is_conf_rkey = true; 1453 1454 smc_llc_link_active(link); 1455 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1456 1457 /* initial contact - try to establish second link */ 1458 smc_llc_srv_add_link(link, NULL); 1459 return 0; 1460 } 1461 1462 /* listen worker: finish */ 1463 static void smc_listen_out(struct smc_sock *new_smc) 1464 { 1465 struct smc_sock *lsmc = new_smc->listen_smc; 1466 struct sock *newsmcsk = &new_smc->sk; 1467 1468 if (lsmc->sk.sk_state == SMC_LISTEN) { 1469 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1470 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1471 release_sock(&lsmc->sk); 1472 } else { /* no longer listening */ 1473 smc_close_non_accepted(newsmcsk); 1474 } 1475 1476 /* Wake up accept */ 1477 lsmc->sk.sk_data_ready(&lsmc->sk); 1478 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1479 } 1480 1481 /* listen worker: finish in state connected */ 1482 static void smc_listen_out_connected(struct smc_sock *new_smc) 1483 { 1484 struct sock *newsmcsk = &new_smc->sk; 1485 1486 sk_refcnt_debug_inc(newsmcsk); 1487 if (newsmcsk->sk_state == SMC_INIT) 1488 newsmcsk->sk_state = SMC_ACTIVE; 1489 1490 smc_listen_out(new_smc); 1491 } 1492 1493 /* listen worker: finish in error state */ 1494 static void smc_listen_out_err(struct smc_sock *new_smc) 1495 { 1496 struct sock *newsmcsk = &new_smc->sk; 1497 struct net *net = sock_net(newsmcsk); 1498 1499 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1500 if (newsmcsk->sk_state == SMC_INIT) 1501 sock_put(&new_smc->sk); /* passive closing */ 1502 newsmcsk->sk_state = SMC_CLOSED; 1503 1504 smc_listen_out(new_smc); 1505 } 1506 1507 /* listen worker: decline and fall back if possible */ 1508 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1509 int local_first, u8 version) 1510 { 1511 /* RDMA setup failed, switch back to TCP */ 1512 smc_conn_abort(new_smc, local_first); 1513 if (reason_code < 0) { /* error, no fallback possible */ 1514 smc_listen_out_err(new_smc); 1515 return; 1516 } 1517 smc_switch_to_fallback(new_smc, reason_code); 1518 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1519 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1520 smc_listen_out_err(new_smc); 1521 return; 1522 } 1523 } 1524 smc_listen_out_connected(new_smc); 1525 } 1526 1527 /* listen worker: version checking */ 1528 static int smc_listen_v2_check(struct smc_sock *new_smc, 1529 struct smc_clc_msg_proposal *pclc, 1530 struct smc_init_info *ini) 1531 { 1532 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1533 struct smc_clc_v2_extension *pclc_v2_ext; 1534 int rc = SMC_CLC_DECL_PEERNOSMC; 1535 1536 ini->smc_type_v1 = pclc->hdr.typev1; 1537 ini->smc_type_v2 = pclc->hdr.typev2; 1538 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1539 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1540 if (pclc->hdr.version > SMC_V1) { 1541 if (smcd_indicated(ini->smc_type_v2)) 1542 ini->smcd_version |= SMC_V2; 1543 if (smcr_indicated(ini->smc_type_v2)) 1544 ini->smcr_version |= SMC_V2; 1545 } 1546 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1547 rc = SMC_CLC_DECL_PEERNOSMC; 1548 goto out; 1549 } 1550 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1551 if (!pclc_v2_ext) { 1552 ini->smcd_version &= ~SMC_V2; 1553 ini->smcr_version &= ~SMC_V2; 1554 rc = SMC_CLC_DECL_NOV2EXT; 1555 goto out; 1556 } 1557 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1558 if (ini->smcd_version & SMC_V2) { 1559 if (!smc_ism_is_v2_capable()) { 1560 ini->smcd_version &= ~SMC_V2; 1561 rc = SMC_CLC_DECL_NOISM2SUPP; 1562 } else if (!pclc_smcd_v2_ext) { 1563 ini->smcd_version &= ~SMC_V2; 1564 rc = SMC_CLC_DECL_NOV2DEXT; 1565 } else if (!pclc_v2_ext->hdr.eid_cnt && 1566 !pclc_v2_ext->hdr.flag.seid) { 1567 ini->smcd_version &= ~SMC_V2; 1568 rc = SMC_CLC_DECL_NOUEID; 1569 } 1570 } 1571 if (ini->smcr_version & SMC_V2) { 1572 if (!pclc_v2_ext->hdr.eid_cnt) { 1573 ini->smcr_version &= ~SMC_V2; 1574 rc = SMC_CLC_DECL_NOUEID; 1575 } 1576 } 1577 1578 out: 1579 if (!ini->smcd_version && !ini->smcr_version) 1580 return rc; 1581 1582 return 0; 1583 } 1584 1585 /* listen worker: check prefixes */ 1586 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1587 struct smc_clc_msg_proposal *pclc) 1588 { 1589 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1590 struct socket *newclcsock = new_smc->clcsock; 1591 1592 if (pclc->hdr.typev1 == SMC_TYPE_N) 1593 return 0; 1594 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1595 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1596 return SMC_CLC_DECL_DIFFPREFIX; 1597 1598 return 0; 1599 } 1600 1601 /* listen worker: initialize connection and buffers */ 1602 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1603 struct smc_init_info *ini) 1604 { 1605 int rc; 1606 1607 /* allocate connection / link group */ 1608 rc = smc_conn_create(new_smc, ini); 1609 if (rc) 1610 return rc; 1611 1612 /* create send buffer and rmb */ 1613 if (smc_buf_create(new_smc, false)) 1614 return SMC_CLC_DECL_MEM; 1615 1616 return 0; 1617 } 1618 1619 /* listen worker: initialize connection and buffers for SMC-D */ 1620 static int smc_listen_ism_init(struct smc_sock *new_smc, 1621 struct smc_init_info *ini) 1622 { 1623 int rc; 1624 1625 rc = smc_conn_create(new_smc, ini); 1626 if (rc) 1627 return rc; 1628 1629 /* Create send and receive buffers */ 1630 rc = smc_buf_create(new_smc, true); 1631 if (rc) { 1632 smc_conn_abort(new_smc, ini->first_contact_local); 1633 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 1634 SMC_CLC_DECL_MEM; 1635 } 1636 1637 return 0; 1638 } 1639 1640 static bool smc_is_already_selected(struct smcd_dev *smcd, 1641 struct smc_init_info *ini, 1642 int matches) 1643 { 1644 int i; 1645 1646 for (i = 0; i < matches; i++) 1647 if (smcd == ini->ism_dev[i]) 1648 return true; 1649 1650 return false; 1651 } 1652 1653 /* check for ISM devices matching proposed ISM devices */ 1654 static void smc_check_ism_v2_match(struct smc_init_info *ini, 1655 u16 proposed_chid, u64 proposed_gid, 1656 unsigned int *matches) 1657 { 1658 struct smcd_dev *smcd; 1659 1660 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1661 if (smcd->going_away) 1662 continue; 1663 if (smc_is_already_selected(smcd, ini, *matches)) 1664 continue; 1665 if (smc_ism_get_chid(smcd) == proposed_chid && 1666 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 1667 ini->ism_peer_gid[*matches] = proposed_gid; 1668 ini->ism_dev[*matches] = smcd; 1669 (*matches)++; 1670 break; 1671 } 1672 } 1673 } 1674 1675 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 1676 { 1677 if (!ini->rc) 1678 ini->rc = rc; 1679 } 1680 1681 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 1682 struct smc_clc_msg_proposal *pclc, 1683 struct smc_init_info *ini) 1684 { 1685 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 1686 struct smc_clc_v2_extension *smc_v2_ext; 1687 struct smc_clc_msg_smcd *pclc_smcd; 1688 unsigned int matches = 0; 1689 u8 smcd_version; 1690 u8 *eid = NULL; 1691 int i, rc; 1692 1693 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 1694 goto not_found; 1695 1696 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1697 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1698 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 1699 1700 mutex_lock(&smcd_dev_list.mutex); 1701 if (pclc_smcd->ism.chid) 1702 /* check for ISM device matching proposed native ISM device */ 1703 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 1704 ntohll(pclc_smcd->ism.gid), &matches); 1705 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 1706 /* check for ISM devices matching proposed non-native ISM 1707 * devices 1708 */ 1709 smc_check_ism_v2_match(ini, 1710 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 1711 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 1712 &matches); 1713 } 1714 mutex_unlock(&smcd_dev_list.mutex); 1715 1716 if (!ini->ism_dev[0]) { 1717 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 1718 goto not_found; 1719 } 1720 1721 smc_ism_get_system_eid(&eid); 1722 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 1723 smcd_v2_ext->system_eid, eid)) 1724 goto not_found; 1725 1726 /* separate - outside the smcd_dev_list.lock */ 1727 smcd_version = ini->smcd_version; 1728 for (i = 0; i < matches; i++) { 1729 ini->smcd_version = SMC_V2; 1730 ini->is_smcd = true; 1731 ini->ism_selected = i; 1732 rc = smc_listen_ism_init(new_smc, ini); 1733 if (rc) { 1734 smc_find_ism_store_rc(rc, ini); 1735 /* try next active ISM device */ 1736 continue; 1737 } 1738 return; /* matching and usable V2 ISM device found */ 1739 } 1740 /* no V2 ISM device could be initialized */ 1741 ini->smcd_version = smcd_version; /* restore original value */ 1742 ini->negotiated_eid[0] = 0; 1743 1744 not_found: 1745 ini->smcd_version &= ~SMC_V2; 1746 ini->ism_dev[0] = NULL; 1747 ini->is_smcd = false; 1748 } 1749 1750 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 1751 struct smc_clc_msg_proposal *pclc, 1752 struct smc_init_info *ini) 1753 { 1754 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 1755 int rc = 0; 1756 1757 /* check if ISM V1 is available */ 1758 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 1759 goto not_found; 1760 ini->is_smcd = true; /* prepare ISM check */ 1761 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 1762 rc = smc_find_ism_device(new_smc, ini); 1763 if (rc) 1764 goto not_found; 1765 ini->ism_selected = 0; 1766 rc = smc_listen_ism_init(new_smc, ini); 1767 if (!rc) 1768 return; /* V1 ISM device found */ 1769 1770 not_found: 1771 smc_find_ism_store_rc(rc, ini); 1772 ini->smcd_version &= ~SMC_V1; 1773 ini->ism_dev[0] = NULL; 1774 ini->is_smcd = false; 1775 } 1776 1777 /* listen worker: register buffers */ 1778 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 1779 { 1780 struct smc_connection *conn = &new_smc->conn; 1781 1782 if (!local_first) { 1783 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 1784 return SMC_CLC_DECL_ERR_REGRMB; 1785 } 1786 smc_rmb_sync_sg_for_device(&new_smc->conn); 1787 1788 return 0; 1789 } 1790 1791 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 1792 struct smc_clc_msg_proposal *pclc, 1793 struct smc_init_info *ini) 1794 { 1795 struct smc_clc_v2_extension *smc_v2_ext; 1796 u8 smcr_version; 1797 int rc; 1798 1799 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 1800 goto not_found; 1801 1802 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1803 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 1804 goto not_found; 1805 1806 /* prepare RDMA check */ 1807 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 1808 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 1809 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 1810 ini->check_smcrv2 = true; 1811 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 1812 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 1813 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 1814 rc = smc_find_rdma_device(new_smc, ini); 1815 if (rc) { 1816 smc_find_ism_store_rc(rc, ini); 1817 goto not_found; 1818 } 1819 if (!ini->smcrv2.uses_gateway) 1820 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 1821 1822 smcr_version = ini->smcr_version; 1823 ini->smcr_version = SMC_V2; 1824 rc = smc_listen_rdma_init(new_smc, ini); 1825 if (!rc) 1826 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 1827 if (!rc) 1828 return; 1829 ini->smcr_version = smcr_version; 1830 smc_find_ism_store_rc(rc, ini); 1831 1832 not_found: 1833 ini->smcr_version &= ~SMC_V2; 1834 ini->check_smcrv2 = false; 1835 } 1836 1837 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 1838 struct smc_clc_msg_proposal *pclc, 1839 struct smc_init_info *ini) 1840 { 1841 int rc; 1842 1843 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 1844 return SMC_CLC_DECL_NOSMCDEV; 1845 1846 /* prepare RDMA check */ 1847 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 1848 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 1849 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 1850 rc = smc_find_rdma_device(new_smc, ini); 1851 if (rc) { 1852 /* no RDMA device found */ 1853 return SMC_CLC_DECL_NOSMCDEV; 1854 } 1855 rc = smc_listen_rdma_init(new_smc, ini); 1856 if (rc) 1857 return rc; 1858 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 1859 } 1860 1861 /* determine the local device matching to proposal */ 1862 static int smc_listen_find_device(struct smc_sock *new_smc, 1863 struct smc_clc_msg_proposal *pclc, 1864 struct smc_init_info *ini) 1865 { 1866 int prfx_rc; 1867 1868 /* check for ISM device matching V2 proposed device */ 1869 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 1870 if (ini->ism_dev[0]) 1871 return 0; 1872 1873 /* check for matching IP prefix and subnet length (V1) */ 1874 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 1875 if (prfx_rc) 1876 smc_find_ism_store_rc(prfx_rc, ini); 1877 1878 /* get vlan id from IP device */ 1879 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 1880 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 1881 1882 /* check for ISM device matching V1 proposed device */ 1883 if (!prfx_rc) 1884 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 1885 if (ini->ism_dev[0]) 1886 return 0; 1887 1888 if (!smcr_indicated(pclc->hdr.typev1) && 1889 !smcr_indicated(pclc->hdr.typev2)) 1890 /* skip RDMA and decline */ 1891 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 1892 1893 /* check if RDMA V2 is available */ 1894 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 1895 if (ini->smcrv2.ib_dev_v2) 1896 return 0; 1897 1898 /* check if RDMA V1 is available */ 1899 if (!prfx_rc) { 1900 int rc; 1901 1902 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 1903 smc_find_ism_store_rc(rc, ini); 1904 return (!rc) ? 0 : ini->rc; 1905 } 1906 return SMC_CLC_DECL_NOSMCDEV; 1907 } 1908 1909 /* listen worker: finish RDMA setup */ 1910 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1911 struct smc_clc_msg_accept_confirm *cclc, 1912 bool local_first, 1913 struct smc_init_info *ini) 1914 { 1915 struct smc_link *link = new_smc->conn.lnk; 1916 int reason_code = 0; 1917 1918 if (local_first) 1919 smc_link_save_peer_info(link, cclc, ini); 1920 1921 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 1922 return SMC_CLC_DECL_ERR_RTOK; 1923 1924 if (local_first) { 1925 if (smc_ib_ready_link(link)) 1926 return SMC_CLC_DECL_ERR_RDYLNK; 1927 /* QP confirmation over RoCE fabric */ 1928 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1929 reason_code = smcr_serv_conf_first_link(new_smc); 1930 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1931 } 1932 return reason_code; 1933 } 1934 1935 /* setup for connection of server */ 1936 static void smc_listen_work(struct work_struct *work) 1937 { 1938 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1939 smc_listen_work); 1940 struct socket *newclcsock = new_smc->clcsock; 1941 struct smc_clc_msg_accept_confirm *cclc; 1942 struct smc_clc_msg_proposal_area *buf; 1943 struct smc_clc_msg_proposal *pclc; 1944 struct smc_init_info *ini = NULL; 1945 u8 proposal_version = SMC_V1; 1946 u8 accept_version; 1947 int rc = 0; 1948 1949 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1950 return smc_listen_out_err(new_smc); 1951 1952 if (new_smc->use_fallback) { 1953 smc_listen_out_connected(new_smc); 1954 return; 1955 } 1956 1957 /* check if peer is smc capable */ 1958 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1959 smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 1960 smc_listen_out_connected(new_smc); 1961 return; 1962 } 1963 1964 /* do inband token exchange - 1965 * wait for and receive SMC Proposal CLC message 1966 */ 1967 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 1968 if (!buf) { 1969 rc = SMC_CLC_DECL_MEM; 1970 goto out_decl; 1971 } 1972 pclc = (struct smc_clc_msg_proposal *)buf; 1973 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 1974 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1975 if (rc) 1976 goto out_decl; 1977 1978 if (pclc->hdr.version > SMC_V1) 1979 proposal_version = SMC_V2; 1980 1981 /* IPSec connections opt out of SMC optimizations */ 1982 if (using_ipsec(new_smc)) { 1983 rc = SMC_CLC_DECL_IPSEC; 1984 goto out_decl; 1985 } 1986 1987 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1988 if (!ini) { 1989 rc = SMC_CLC_DECL_MEM; 1990 goto out_decl; 1991 } 1992 1993 /* initial version checking */ 1994 rc = smc_listen_v2_check(new_smc, pclc, ini); 1995 if (rc) 1996 goto out_decl; 1997 1998 mutex_lock(&smc_server_lgr_pending); 1999 smc_close_init(new_smc); 2000 smc_rx_init(new_smc); 2001 smc_tx_init(new_smc); 2002 2003 /* determine ISM or RoCE device used for connection */ 2004 rc = smc_listen_find_device(new_smc, pclc, ini); 2005 if (rc) 2006 goto out_unlock; 2007 2008 /* send SMC Accept CLC message */ 2009 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2010 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2011 accept_version, ini->negotiated_eid); 2012 if (rc) 2013 goto out_unlock; 2014 2015 /* SMC-D does not need this lock any more */ 2016 if (ini->is_smcd) 2017 mutex_unlock(&smc_server_lgr_pending); 2018 2019 /* receive SMC Confirm CLC message */ 2020 memset(buf, 0, sizeof(*buf)); 2021 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2022 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2023 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2024 if (rc) { 2025 if (!ini->is_smcd) 2026 goto out_unlock; 2027 goto out_decl; 2028 } 2029 2030 /* finish worker */ 2031 if (!ini->is_smcd) { 2032 rc = smc_listen_rdma_finish(new_smc, cclc, 2033 ini->first_contact_local, ini); 2034 if (rc) 2035 goto out_unlock; 2036 mutex_unlock(&smc_server_lgr_pending); 2037 } 2038 smc_conn_save_peer_info(new_smc, cclc); 2039 smc_listen_out_connected(new_smc); 2040 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2041 goto out_free; 2042 2043 out_unlock: 2044 mutex_unlock(&smc_server_lgr_pending); 2045 out_decl: 2046 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2047 proposal_version); 2048 out_free: 2049 kfree(ini); 2050 kfree(buf); 2051 } 2052 2053 static void smc_tcp_listen_work(struct work_struct *work) 2054 { 2055 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2056 tcp_listen_work); 2057 struct sock *lsk = &lsmc->sk; 2058 struct smc_sock *new_smc; 2059 int rc = 0; 2060 2061 lock_sock(lsk); 2062 while (lsk->sk_state == SMC_LISTEN) { 2063 rc = smc_clcsock_accept(lsmc, &new_smc); 2064 if (rc) /* clcsock accept queue empty or error */ 2065 goto out; 2066 if (!new_smc) 2067 continue; 2068 2069 new_smc->listen_smc = lsmc; 2070 new_smc->use_fallback = lsmc->use_fallback; 2071 new_smc->fallback_rsn = lsmc->fallback_rsn; 2072 sock_hold(lsk); /* sock_put in smc_listen_work */ 2073 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2074 smc_copy_sock_settings_to_smc(new_smc); 2075 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 2076 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 2077 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2078 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2079 sock_put(&new_smc->sk); 2080 } 2081 2082 out: 2083 release_sock(lsk); 2084 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2085 } 2086 2087 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2088 { 2089 struct smc_sock *lsmc; 2090 2091 lsmc = (struct smc_sock *) 2092 ((uintptr_t)listen_clcsock->sk_user_data & ~SK_USER_DATA_NOCOPY); 2093 if (!lsmc) 2094 return; 2095 lsmc->clcsk_data_ready(listen_clcsock); 2096 if (lsmc->sk.sk_state == SMC_LISTEN) { 2097 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2098 if (!queue_work(smc_hs_wq, &lsmc->tcp_listen_work)) 2099 sock_put(&lsmc->sk); 2100 } 2101 } 2102 2103 static int smc_listen(struct socket *sock, int backlog) 2104 { 2105 struct sock *sk = sock->sk; 2106 struct smc_sock *smc; 2107 int rc; 2108 2109 smc = smc_sk(sk); 2110 lock_sock(sk); 2111 2112 rc = -EINVAL; 2113 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2114 smc->connect_nonblock) 2115 goto out; 2116 2117 rc = 0; 2118 if (sk->sk_state == SMC_LISTEN) { 2119 sk->sk_max_ack_backlog = backlog; 2120 goto out; 2121 } 2122 /* some socket options are handled in core, so we could not apply 2123 * them to the clc socket -- copy smc socket options to clc socket 2124 */ 2125 smc_copy_sock_settings_to_clc(smc); 2126 if (!smc->use_fallback) 2127 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2128 2129 /* save original sk_data_ready function and establish 2130 * smc-specific sk_data_ready function 2131 */ 2132 smc->clcsk_data_ready = smc->clcsock->sk->sk_data_ready; 2133 smc->clcsock->sk->sk_data_ready = smc_clcsock_data_ready; 2134 smc->clcsock->sk->sk_user_data = 2135 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2136 rc = kernel_listen(smc->clcsock, backlog); 2137 if (rc) { 2138 smc->clcsock->sk->sk_data_ready = smc->clcsk_data_ready; 2139 goto out; 2140 } 2141 sk->sk_max_ack_backlog = backlog; 2142 sk->sk_ack_backlog = 0; 2143 sk->sk_state = SMC_LISTEN; 2144 2145 out: 2146 release_sock(sk); 2147 return rc; 2148 } 2149 2150 static int smc_accept(struct socket *sock, struct socket *new_sock, 2151 int flags, bool kern) 2152 { 2153 struct sock *sk = sock->sk, *nsk; 2154 DECLARE_WAITQUEUE(wait, current); 2155 struct smc_sock *lsmc; 2156 long timeo; 2157 int rc = 0; 2158 2159 lsmc = smc_sk(sk); 2160 sock_hold(sk); /* sock_put below */ 2161 lock_sock(sk); 2162 2163 if (lsmc->sk.sk_state != SMC_LISTEN) { 2164 rc = -EINVAL; 2165 release_sock(sk); 2166 goto out; 2167 } 2168 2169 /* Wait for an incoming connection */ 2170 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2171 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2172 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2173 set_current_state(TASK_INTERRUPTIBLE); 2174 if (!timeo) { 2175 rc = -EAGAIN; 2176 break; 2177 } 2178 release_sock(sk); 2179 timeo = schedule_timeout(timeo); 2180 /* wakeup by sk_data_ready in smc_listen_work() */ 2181 sched_annotate_sleep(); 2182 lock_sock(sk); 2183 if (signal_pending(current)) { 2184 rc = sock_intr_errno(timeo); 2185 break; 2186 } 2187 } 2188 set_current_state(TASK_RUNNING); 2189 remove_wait_queue(sk_sleep(sk), &wait); 2190 2191 if (!rc) 2192 rc = sock_error(nsk); 2193 release_sock(sk); 2194 if (rc) 2195 goto out; 2196 2197 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2198 /* wait till data arrives on the socket */ 2199 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2200 MSEC_PER_SEC); 2201 if (smc_sk(nsk)->use_fallback) { 2202 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2203 2204 lock_sock(clcsk); 2205 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2206 sk_wait_data(clcsk, &timeo, NULL); 2207 release_sock(clcsk); 2208 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2209 lock_sock(nsk); 2210 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2211 release_sock(nsk); 2212 } 2213 } 2214 2215 out: 2216 sock_put(sk); /* sock_hold above */ 2217 return rc; 2218 } 2219 2220 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2221 int peer) 2222 { 2223 struct smc_sock *smc; 2224 2225 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2226 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2227 return -ENOTCONN; 2228 2229 smc = smc_sk(sock->sk); 2230 2231 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2232 } 2233 2234 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2235 { 2236 struct sock *sk = sock->sk; 2237 struct smc_sock *smc; 2238 int rc = -EPIPE; 2239 2240 smc = smc_sk(sk); 2241 lock_sock(sk); 2242 if ((sk->sk_state != SMC_ACTIVE) && 2243 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2244 (sk->sk_state != SMC_INIT)) 2245 goto out; 2246 2247 if (msg->msg_flags & MSG_FASTOPEN) { 2248 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2249 smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2250 } else { 2251 rc = -EINVAL; 2252 goto out; 2253 } 2254 } 2255 2256 if (smc->use_fallback) { 2257 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2258 } else { 2259 rc = smc_tx_sendmsg(smc, msg, len); 2260 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2261 } 2262 out: 2263 release_sock(sk); 2264 return rc; 2265 } 2266 2267 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2268 int flags) 2269 { 2270 struct sock *sk = sock->sk; 2271 struct smc_sock *smc; 2272 int rc = -ENOTCONN; 2273 2274 smc = smc_sk(sk); 2275 lock_sock(sk); 2276 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2277 /* socket was connected before, no more data to read */ 2278 rc = 0; 2279 goto out; 2280 } 2281 if ((sk->sk_state == SMC_INIT) || 2282 (sk->sk_state == SMC_LISTEN) || 2283 (sk->sk_state == SMC_CLOSED)) 2284 goto out; 2285 2286 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2287 rc = 0; 2288 goto out; 2289 } 2290 2291 if (smc->use_fallback) { 2292 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2293 } else { 2294 msg->msg_namelen = 0; 2295 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2296 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2297 } 2298 2299 out: 2300 release_sock(sk); 2301 return rc; 2302 } 2303 2304 static __poll_t smc_accept_poll(struct sock *parent) 2305 { 2306 struct smc_sock *isk = smc_sk(parent); 2307 __poll_t mask = 0; 2308 2309 spin_lock(&isk->accept_q_lock); 2310 if (!list_empty(&isk->accept_q)) 2311 mask = EPOLLIN | EPOLLRDNORM; 2312 spin_unlock(&isk->accept_q_lock); 2313 2314 return mask; 2315 } 2316 2317 static __poll_t smc_poll(struct file *file, struct socket *sock, 2318 poll_table *wait) 2319 { 2320 struct sock *sk = sock->sk; 2321 struct smc_sock *smc; 2322 __poll_t mask = 0; 2323 2324 if (!sk) 2325 return EPOLLNVAL; 2326 2327 smc = smc_sk(sock->sk); 2328 if (smc->use_fallback) { 2329 /* delegate to CLC child sock */ 2330 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2331 sk->sk_err = smc->clcsock->sk->sk_err; 2332 } else { 2333 if (sk->sk_state != SMC_CLOSED) 2334 sock_poll_wait(file, sock, wait); 2335 if (sk->sk_err) 2336 mask |= EPOLLERR; 2337 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2338 (sk->sk_state == SMC_CLOSED)) 2339 mask |= EPOLLHUP; 2340 if (sk->sk_state == SMC_LISTEN) { 2341 /* woken up by sk_data_ready in smc_listen_work() */ 2342 mask |= smc_accept_poll(sk); 2343 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2344 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2345 wait); 2346 sk->sk_err = smc->clcsock->sk->sk_err; 2347 } else { 2348 if ((sk->sk_state != SMC_INIT && 2349 atomic_read(&smc->conn.sndbuf_space)) || 2350 sk->sk_shutdown & SEND_SHUTDOWN) { 2351 mask |= EPOLLOUT | EPOLLWRNORM; 2352 } else { 2353 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2354 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2355 } 2356 if (atomic_read(&smc->conn.bytes_to_rcv)) 2357 mask |= EPOLLIN | EPOLLRDNORM; 2358 if (sk->sk_shutdown & RCV_SHUTDOWN) 2359 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2360 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2361 mask |= EPOLLIN; 2362 if (smc->conn.urg_state == SMC_URG_VALID) 2363 mask |= EPOLLPRI; 2364 } 2365 } 2366 2367 return mask; 2368 } 2369 2370 static int smc_shutdown(struct socket *sock, int how) 2371 { 2372 struct sock *sk = sock->sk; 2373 bool do_shutdown = true; 2374 struct smc_sock *smc; 2375 int rc = -EINVAL; 2376 int old_state; 2377 int rc1 = 0; 2378 2379 smc = smc_sk(sk); 2380 2381 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2382 return rc; 2383 2384 lock_sock(sk); 2385 2386 rc = -ENOTCONN; 2387 if ((sk->sk_state != SMC_ACTIVE) && 2388 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2389 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2390 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2391 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2392 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2393 goto out; 2394 if (smc->use_fallback) { 2395 rc = kernel_sock_shutdown(smc->clcsock, how); 2396 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2397 if (sk->sk_shutdown == SHUTDOWN_MASK) 2398 sk->sk_state = SMC_CLOSED; 2399 goto out; 2400 } 2401 switch (how) { 2402 case SHUT_RDWR: /* shutdown in both directions */ 2403 old_state = sk->sk_state; 2404 rc = smc_close_active(smc); 2405 if (old_state == SMC_ACTIVE && 2406 sk->sk_state == SMC_PEERCLOSEWAIT1) 2407 do_shutdown = false; 2408 break; 2409 case SHUT_WR: 2410 rc = smc_close_shutdown_write(smc); 2411 break; 2412 case SHUT_RD: 2413 rc = 0; 2414 /* nothing more to do because peer is not involved */ 2415 break; 2416 } 2417 if (do_shutdown && smc->clcsock) 2418 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2419 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2420 sk->sk_shutdown |= how + 1; 2421 2422 out: 2423 release_sock(sk); 2424 return rc ? rc : rc1; 2425 } 2426 2427 static int smc_setsockopt(struct socket *sock, int level, int optname, 2428 sockptr_t optval, unsigned int optlen) 2429 { 2430 struct sock *sk = sock->sk; 2431 struct smc_sock *smc; 2432 int val, rc; 2433 2434 if (level == SOL_TCP && optname == TCP_ULP) 2435 return -EOPNOTSUPP; 2436 2437 smc = smc_sk(sk); 2438 2439 /* generic setsockopts reaching us here always apply to the 2440 * CLC socket 2441 */ 2442 if (unlikely(!smc->clcsock->ops->setsockopt)) 2443 rc = -EOPNOTSUPP; 2444 else 2445 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 2446 optval, optlen); 2447 if (smc->clcsock->sk->sk_err) { 2448 sk->sk_err = smc->clcsock->sk->sk_err; 2449 sk_error_report(sk); 2450 } 2451 2452 if (optlen < sizeof(int)) 2453 return -EINVAL; 2454 if (copy_from_sockptr(&val, optval, sizeof(int))) 2455 return -EFAULT; 2456 2457 lock_sock(sk); 2458 if (rc || smc->use_fallback) 2459 goto out; 2460 switch (optname) { 2461 case TCP_FASTOPEN: 2462 case TCP_FASTOPEN_CONNECT: 2463 case TCP_FASTOPEN_KEY: 2464 case TCP_FASTOPEN_NO_COOKIE: 2465 /* option not supported by SMC */ 2466 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2467 smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2468 } else { 2469 rc = -EINVAL; 2470 } 2471 break; 2472 case TCP_NODELAY: 2473 if (sk->sk_state != SMC_INIT && 2474 sk->sk_state != SMC_LISTEN && 2475 sk->sk_state != SMC_CLOSED) { 2476 if (val) { 2477 SMC_STAT_INC(smc, ndly_cnt); 2478 mod_delayed_work(smc->conn.lgr->tx_wq, 2479 &smc->conn.tx_work, 0); 2480 } 2481 } 2482 break; 2483 case TCP_CORK: 2484 if (sk->sk_state != SMC_INIT && 2485 sk->sk_state != SMC_LISTEN && 2486 sk->sk_state != SMC_CLOSED) { 2487 if (!val) { 2488 SMC_STAT_INC(smc, cork_cnt); 2489 mod_delayed_work(smc->conn.lgr->tx_wq, 2490 &smc->conn.tx_work, 0); 2491 } 2492 } 2493 break; 2494 case TCP_DEFER_ACCEPT: 2495 smc->sockopt_defer_accept = val; 2496 break; 2497 default: 2498 break; 2499 } 2500 out: 2501 release_sock(sk); 2502 2503 return rc; 2504 } 2505 2506 static int smc_getsockopt(struct socket *sock, int level, int optname, 2507 char __user *optval, int __user *optlen) 2508 { 2509 struct smc_sock *smc; 2510 2511 smc = smc_sk(sock->sk); 2512 /* socket options apply to the CLC socket */ 2513 if (unlikely(!smc->clcsock->ops->getsockopt)) 2514 return -EOPNOTSUPP; 2515 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 2516 optval, optlen); 2517 } 2518 2519 static int smc_ioctl(struct socket *sock, unsigned int cmd, 2520 unsigned long arg) 2521 { 2522 union smc_host_cursor cons, urg; 2523 struct smc_connection *conn; 2524 struct smc_sock *smc; 2525 int answ; 2526 2527 smc = smc_sk(sock->sk); 2528 conn = &smc->conn; 2529 lock_sock(&smc->sk); 2530 if (smc->use_fallback) { 2531 if (!smc->clcsock) { 2532 release_sock(&smc->sk); 2533 return -EBADF; 2534 } 2535 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 2536 release_sock(&smc->sk); 2537 return answ; 2538 } 2539 switch (cmd) { 2540 case SIOCINQ: /* same as FIONREAD */ 2541 if (smc->sk.sk_state == SMC_LISTEN) { 2542 release_sock(&smc->sk); 2543 return -EINVAL; 2544 } 2545 if (smc->sk.sk_state == SMC_INIT || 2546 smc->sk.sk_state == SMC_CLOSED) 2547 answ = 0; 2548 else 2549 answ = atomic_read(&smc->conn.bytes_to_rcv); 2550 break; 2551 case SIOCOUTQ: 2552 /* output queue size (not send + not acked) */ 2553 if (smc->sk.sk_state == SMC_LISTEN) { 2554 release_sock(&smc->sk); 2555 return -EINVAL; 2556 } 2557 if (smc->sk.sk_state == SMC_INIT || 2558 smc->sk.sk_state == SMC_CLOSED) 2559 answ = 0; 2560 else 2561 answ = smc->conn.sndbuf_desc->len - 2562 atomic_read(&smc->conn.sndbuf_space); 2563 break; 2564 case SIOCOUTQNSD: 2565 /* output queue size (not send only) */ 2566 if (smc->sk.sk_state == SMC_LISTEN) { 2567 release_sock(&smc->sk); 2568 return -EINVAL; 2569 } 2570 if (smc->sk.sk_state == SMC_INIT || 2571 smc->sk.sk_state == SMC_CLOSED) 2572 answ = 0; 2573 else 2574 answ = smc_tx_prepared_sends(&smc->conn); 2575 break; 2576 case SIOCATMARK: 2577 if (smc->sk.sk_state == SMC_LISTEN) { 2578 release_sock(&smc->sk); 2579 return -EINVAL; 2580 } 2581 if (smc->sk.sk_state == SMC_INIT || 2582 smc->sk.sk_state == SMC_CLOSED) { 2583 answ = 0; 2584 } else { 2585 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 2586 smc_curs_copy(&urg, &conn->urg_curs, conn); 2587 answ = smc_curs_diff(conn->rmb_desc->len, 2588 &cons, &urg) == 1; 2589 } 2590 break; 2591 default: 2592 release_sock(&smc->sk); 2593 return -ENOIOCTLCMD; 2594 } 2595 release_sock(&smc->sk); 2596 2597 return put_user(answ, (int __user *)arg); 2598 } 2599 2600 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 2601 int offset, size_t size, int flags) 2602 { 2603 struct sock *sk = sock->sk; 2604 struct smc_sock *smc; 2605 int rc = -EPIPE; 2606 2607 smc = smc_sk(sk); 2608 lock_sock(sk); 2609 if (sk->sk_state != SMC_ACTIVE) { 2610 release_sock(sk); 2611 goto out; 2612 } 2613 release_sock(sk); 2614 if (smc->use_fallback) { 2615 rc = kernel_sendpage(smc->clcsock, page, offset, 2616 size, flags); 2617 } else { 2618 SMC_STAT_INC(smc, sendpage_cnt); 2619 rc = sock_no_sendpage(sock, page, offset, size, flags); 2620 } 2621 2622 out: 2623 return rc; 2624 } 2625 2626 /* Map the affected portions of the rmbe into an spd, note the number of bytes 2627 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 2628 * updates till whenever a respective page has been fully processed. 2629 * Note that subsequent recv() calls have to wait till all splice() processing 2630 * completed. 2631 */ 2632 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 2633 struct pipe_inode_info *pipe, size_t len, 2634 unsigned int flags) 2635 { 2636 struct sock *sk = sock->sk; 2637 struct smc_sock *smc; 2638 int rc = -ENOTCONN; 2639 2640 smc = smc_sk(sk); 2641 lock_sock(sk); 2642 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2643 /* socket was connected before, no more data to read */ 2644 rc = 0; 2645 goto out; 2646 } 2647 if (sk->sk_state == SMC_INIT || 2648 sk->sk_state == SMC_LISTEN || 2649 sk->sk_state == SMC_CLOSED) 2650 goto out; 2651 2652 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2653 rc = 0; 2654 goto out; 2655 } 2656 2657 if (smc->use_fallback) { 2658 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 2659 pipe, len, flags); 2660 } else { 2661 if (*ppos) { 2662 rc = -ESPIPE; 2663 goto out; 2664 } 2665 if (flags & SPLICE_F_NONBLOCK) 2666 flags = MSG_DONTWAIT; 2667 else 2668 flags = 0; 2669 SMC_STAT_INC(smc, splice_cnt); 2670 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 2671 } 2672 out: 2673 release_sock(sk); 2674 2675 return rc; 2676 } 2677 2678 /* must look like tcp */ 2679 static const struct proto_ops smc_sock_ops = { 2680 .family = PF_SMC, 2681 .owner = THIS_MODULE, 2682 .release = smc_release, 2683 .bind = smc_bind, 2684 .connect = smc_connect, 2685 .socketpair = sock_no_socketpair, 2686 .accept = smc_accept, 2687 .getname = smc_getname, 2688 .poll = smc_poll, 2689 .ioctl = smc_ioctl, 2690 .listen = smc_listen, 2691 .shutdown = smc_shutdown, 2692 .setsockopt = smc_setsockopt, 2693 .getsockopt = smc_getsockopt, 2694 .sendmsg = smc_sendmsg, 2695 .recvmsg = smc_recvmsg, 2696 .mmap = sock_no_mmap, 2697 .sendpage = smc_sendpage, 2698 .splice_read = smc_splice_read, 2699 }; 2700 2701 static int smc_create(struct net *net, struct socket *sock, int protocol, 2702 int kern) 2703 { 2704 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 2705 struct smc_sock *smc; 2706 struct sock *sk; 2707 int rc; 2708 2709 rc = -ESOCKTNOSUPPORT; 2710 if (sock->type != SOCK_STREAM) 2711 goto out; 2712 2713 rc = -EPROTONOSUPPORT; 2714 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 2715 goto out; 2716 2717 rc = -ENOBUFS; 2718 sock->ops = &smc_sock_ops; 2719 sk = smc_sock_alloc(net, sock, protocol); 2720 if (!sk) 2721 goto out; 2722 2723 /* create internal TCP socket for CLC handshake and fallback */ 2724 smc = smc_sk(sk); 2725 smc->use_fallback = false; /* assume rdma capability first */ 2726 smc->fallback_rsn = 0; 2727 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 2728 &smc->clcsock); 2729 if (rc) { 2730 sk_common_release(sk); 2731 goto out; 2732 } 2733 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2734 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2735 2736 out: 2737 return rc; 2738 } 2739 2740 static const struct net_proto_family smc_sock_family_ops = { 2741 .family = PF_SMC, 2742 .owner = THIS_MODULE, 2743 .create = smc_create, 2744 }; 2745 2746 unsigned int smc_net_id; 2747 2748 static __net_init int smc_net_init(struct net *net) 2749 { 2750 return smc_pnet_net_init(net); 2751 } 2752 2753 static void __net_exit smc_net_exit(struct net *net) 2754 { 2755 smc_pnet_net_exit(net); 2756 } 2757 2758 static __net_init int smc_net_stat_init(struct net *net) 2759 { 2760 return smc_stats_init(net); 2761 } 2762 2763 static void __net_exit smc_net_stat_exit(struct net *net) 2764 { 2765 smc_stats_exit(net); 2766 } 2767 2768 static struct pernet_operations smc_net_ops = { 2769 .init = smc_net_init, 2770 .exit = smc_net_exit, 2771 .id = &smc_net_id, 2772 .size = sizeof(struct smc_net), 2773 }; 2774 2775 static struct pernet_operations smc_net_stat_ops = { 2776 .init = smc_net_stat_init, 2777 .exit = smc_net_stat_exit, 2778 }; 2779 2780 static int __init smc_init(void) 2781 { 2782 int rc; 2783 2784 rc = register_pernet_subsys(&smc_net_ops); 2785 if (rc) 2786 return rc; 2787 2788 rc = register_pernet_subsys(&smc_net_stat_ops); 2789 if (rc) 2790 return rc; 2791 2792 smc_ism_init(); 2793 smc_clc_init(); 2794 2795 rc = smc_nl_init(); 2796 if (rc) 2797 goto out_pernet_subsys; 2798 2799 rc = smc_pnet_init(); 2800 if (rc) 2801 goto out_nl; 2802 2803 rc = -ENOMEM; 2804 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 2805 if (!smc_hs_wq) 2806 goto out_pnet; 2807 2808 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 2809 if (!smc_close_wq) 2810 goto out_alloc_hs_wq; 2811 2812 rc = smc_core_init(); 2813 if (rc) { 2814 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2815 goto out_alloc_wqs; 2816 } 2817 2818 rc = smc_llc_init(); 2819 if (rc) { 2820 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2821 goto out_core; 2822 } 2823 2824 rc = smc_cdc_init(); 2825 if (rc) { 2826 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2827 goto out_core; 2828 } 2829 2830 rc = proto_register(&smc_proto, 1); 2831 if (rc) { 2832 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2833 goto out_core; 2834 } 2835 2836 rc = proto_register(&smc_proto6, 1); 2837 if (rc) { 2838 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2839 goto out_proto; 2840 } 2841 2842 rc = sock_register(&smc_sock_family_ops); 2843 if (rc) { 2844 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2845 goto out_proto6; 2846 } 2847 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2848 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2849 2850 rc = smc_ib_register_client(); 2851 if (rc) { 2852 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2853 goto out_sock; 2854 } 2855 2856 static_branch_enable(&tcp_have_smc); 2857 return 0; 2858 2859 out_sock: 2860 sock_unregister(PF_SMC); 2861 out_proto6: 2862 proto_unregister(&smc_proto6); 2863 out_proto: 2864 proto_unregister(&smc_proto); 2865 out_core: 2866 smc_core_exit(); 2867 out_alloc_wqs: 2868 destroy_workqueue(smc_close_wq); 2869 out_alloc_hs_wq: 2870 destroy_workqueue(smc_hs_wq); 2871 out_pnet: 2872 smc_pnet_exit(); 2873 out_nl: 2874 smc_nl_exit(); 2875 out_pernet_subsys: 2876 unregister_pernet_subsys(&smc_net_ops); 2877 2878 return rc; 2879 } 2880 2881 static void __exit smc_exit(void) 2882 { 2883 static_branch_disable(&tcp_have_smc); 2884 sock_unregister(PF_SMC); 2885 smc_core_exit(); 2886 smc_ib_unregister_client(); 2887 destroy_workqueue(smc_close_wq); 2888 destroy_workqueue(smc_hs_wq); 2889 proto_unregister(&smc_proto6); 2890 proto_unregister(&smc_proto); 2891 smc_pnet_exit(); 2892 smc_nl_exit(); 2893 smc_clc_exit(); 2894 unregister_pernet_subsys(&smc_net_stat_ops); 2895 unregister_pernet_subsys(&smc_net_ops); 2896 rcu_barrier(); 2897 } 2898 2899 module_init(smc_init); 2900 module_exit(smc_exit); 2901 2902 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2903 MODULE_DESCRIPTION("smc socket address family"); 2904 MODULE_LICENSE("GPL"); 2905 MODULE_ALIAS_NETPROTO(PF_SMC); 2906