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