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 write_unlock_bh(&h->lock); 93 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 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); 197 198 if (cancel_work_sync(&smc->connect_work)) 199 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 200 201 if (sk->sk_state == SMC_LISTEN) 202 /* smc_close_non_accepted() is called and acquires 203 * sock lock for child sockets again 204 */ 205 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 206 else 207 lock_sock(sk); 208 209 rc = __smc_release(smc); 210 211 /* detach socket */ 212 sock_orphan(sk); 213 sock->sk = NULL; 214 release_sock(sk); 215 216 sock_put(sk); /* sock_hold above */ 217 sock_put(sk); /* final sock_put */ 218 out: 219 return rc; 220 } 221 222 static void smc_destruct(struct sock *sk) 223 { 224 if (sk->sk_state != SMC_CLOSED) 225 return; 226 if (!sock_flag(sk, SOCK_DEAD)) 227 return; 228 229 sk_refcnt_debug_dec(sk); 230 } 231 232 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 233 int protocol) 234 { 235 struct smc_sock *smc; 236 struct proto *prot; 237 struct sock *sk; 238 239 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 240 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 241 if (!sk) 242 return NULL; 243 244 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 245 sk->sk_state = SMC_INIT; 246 sk->sk_destruct = smc_destruct; 247 sk->sk_protocol = protocol; 248 smc = smc_sk(sk); 249 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 250 INIT_WORK(&smc->connect_work, smc_connect_work); 251 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 252 INIT_LIST_HEAD(&smc->accept_q); 253 spin_lock_init(&smc->accept_q_lock); 254 spin_lock_init(&smc->conn.send_lock); 255 sk->sk_prot->hash(sk); 256 sk_refcnt_debug_inc(sk); 257 mutex_init(&smc->clcsock_release_lock); 258 259 return sk; 260 } 261 262 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 263 int addr_len) 264 { 265 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 266 struct sock *sk = sock->sk; 267 struct smc_sock *smc; 268 int rc; 269 270 smc = smc_sk(sk); 271 272 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 273 rc = -EINVAL; 274 if (addr_len < sizeof(struct sockaddr_in)) 275 goto out; 276 277 rc = -EAFNOSUPPORT; 278 if (addr->sin_family != AF_INET && 279 addr->sin_family != AF_INET6 && 280 addr->sin_family != AF_UNSPEC) 281 goto out; 282 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 283 if (addr->sin_family == AF_UNSPEC && 284 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 285 goto out; 286 287 lock_sock(sk); 288 289 /* Check if socket is already active */ 290 rc = -EINVAL; 291 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 292 goto out_rel; 293 294 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 295 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 296 297 out_rel: 298 release_sock(sk); 299 out: 300 return rc; 301 } 302 303 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 304 unsigned long mask) 305 { 306 /* options we don't get control via setsockopt for */ 307 nsk->sk_type = osk->sk_type; 308 nsk->sk_sndbuf = osk->sk_sndbuf; 309 nsk->sk_rcvbuf = osk->sk_rcvbuf; 310 nsk->sk_sndtimeo = osk->sk_sndtimeo; 311 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 312 nsk->sk_mark = osk->sk_mark; 313 nsk->sk_priority = osk->sk_priority; 314 nsk->sk_rcvlowat = osk->sk_rcvlowat; 315 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 316 nsk->sk_err = osk->sk_err; 317 318 nsk->sk_flags &= ~mask; 319 nsk->sk_flags |= osk->sk_flags & mask; 320 } 321 322 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 323 (1UL << SOCK_KEEPOPEN) | \ 324 (1UL << SOCK_LINGER) | \ 325 (1UL << SOCK_BROADCAST) | \ 326 (1UL << SOCK_TIMESTAMP) | \ 327 (1UL << SOCK_DBG) | \ 328 (1UL << SOCK_RCVTSTAMP) | \ 329 (1UL << SOCK_RCVTSTAMPNS) | \ 330 (1UL << SOCK_LOCALROUTE) | \ 331 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 332 (1UL << SOCK_RXQ_OVFL) | \ 333 (1UL << SOCK_WIFI_STATUS) | \ 334 (1UL << SOCK_NOFCS) | \ 335 (1UL << SOCK_FILTER_LOCKED) | \ 336 (1UL << SOCK_TSTAMP_NEW)) 337 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 338 * clc socket (since smc is not called for these options from net/core) 339 */ 340 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 341 { 342 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 343 } 344 345 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 346 (1UL << SOCK_KEEPOPEN) | \ 347 (1UL << SOCK_LINGER) | \ 348 (1UL << SOCK_DBG)) 349 /* copy only settings and flags relevant for smc from clc to smc socket */ 350 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 351 { 352 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 353 } 354 355 /* register the new rmb on all links */ 356 static int smcr_lgr_reg_rmbs(struct smc_link *link, 357 struct smc_buf_desc *rmb_desc) 358 { 359 struct smc_link_group *lgr = link->lgr; 360 int i, rc = 0; 361 362 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 363 if (rc) 364 return rc; 365 /* protect against parallel smc_llc_cli_rkey_exchange() and 366 * parallel smcr_link_reg_rmb() 367 */ 368 mutex_lock(&lgr->llc_conf_mutex); 369 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 370 if (!smc_link_active(&lgr->lnk[i])) 371 continue; 372 rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc); 373 if (rc) 374 goto out; 375 } 376 377 /* exchange confirm_rkey msg with peer */ 378 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 379 if (rc) { 380 rc = -EFAULT; 381 goto out; 382 } 383 rmb_desc->is_conf_rkey = true; 384 out: 385 mutex_unlock(&lgr->llc_conf_mutex); 386 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 387 return rc; 388 } 389 390 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 391 { 392 struct smc_link *link = smc->conn.lnk; 393 struct smc_llc_qentry *qentry; 394 int rc; 395 396 /* receive CONFIRM LINK request from server over RoCE fabric */ 397 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 398 SMC_LLC_CONFIRM_LINK); 399 if (!qentry) { 400 struct smc_clc_msg_decline dclc; 401 402 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 403 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 404 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 405 } 406 smc_llc_save_peer_uid(qentry); 407 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 408 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 409 if (rc) 410 return SMC_CLC_DECL_RMBE_EC; 411 412 rc = smc_ib_modify_qp_rts(link); 413 if (rc) 414 return SMC_CLC_DECL_ERR_RDYLNK; 415 416 smc_wr_remember_qp_attr(link); 417 418 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 419 return SMC_CLC_DECL_ERR_REGRMB; 420 421 /* confirm_rkey is implicit on 1st contact */ 422 smc->conn.rmb_desc->is_conf_rkey = true; 423 424 /* send CONFIRM LINK response over RoCE fabric */ 425 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 426 if (rc < 0) 427 return SMC_CLC_DECL_TIMEOUT_CL; 428 429 smc_llc_link_active(link); 430 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 431 432 /* optional 2nd link, receive ADD LINK request from server */ 433 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 434 SMC_LLC_ADD_LINK); 435 if (!qentry) { 436 struct smc_clc_msg_decline dclc; 437 438 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 439 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 440 if (rc == -EAGAIN) 441 rc = 0; /* no DECLINE received, go with one link */ 442 return rc; 443 } 444 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 445 smc_llc_cli_add_link(link, qentry); 446 return 0; 447 } 448 449 static bool smc_isascii(char *hostname) 450 { 451 int i; 452 453 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 454 if (!isascii(hostname[i])) 455 return false; 456 return true; 457 } 458 459 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 460 struct smc_clc_msg_accept_confirm *clc) 461 { 462 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 463 (struct smc_clc_msg_accept_confirm_v2 *)clc; 464 struct smc_clc_first_contact_ext *fce; 465 int clc_v2_len; 466 467 if (clc->hdr.version == SMC_V1 || 468 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 469 return; 470 471 if (smc->conn.lgr->is_smcd) { 472 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 473 SMC_MAX_EID_LEN); 474 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 475 d1); 476 } else { 477 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 478 SMC_MAX_EID_LEN); 479 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 480 r1); 481 } 482 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 483 smc->conn.lgr->peer_os = fce->os_type; 484 smc->conn.lgr->peer_smc_release = fce->release; 485 if (smc_isascii(fce->hostname)) 486 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 487 SMC_MAX_HOSTNAME_LEN); 488 } 489 490 static void smcr_conn_save_peer_info(struct smc_sock *smc, 491 struct smc_clc_msg_accept_confirm *clc) 492 { 493 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 494 495 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 496 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 497 smc->conn.peer_rmbe_size = bufsize; 498 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 499 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 500 } 501 502 static void smcd_conn_save_peer_info(struct smc_sock *smc, 503 struct smc_clc_msg_accept_confirm *clc) 504 { 505 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 506 507 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 508 smc->conn.peer_token = clc->d0.token; 509 /* msg header takes up space in the buffer */ 510 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 511 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 512 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 513 } 514 515 static void smc_conn_save_peer_info(struct smc_sock *smc, 516 struct smc_clc_msg_accept_confirm *clc) 517 { 518 if (smc->conn.lgr->is_smcd) 519 smcd_conn_save_peer_info(smc, clc); 520 else 521 smcr_conn_save_peer_info(smc, clc); 522 smc_conn_save_peer_info_fce(smc, clc); 523 } 524 525 static void smc_link_save_peer_info(struct smc_link *link, 526 struct smc_clc_msg_accept_confirm *clc, 527 struct smc_init_info *ini) 528 { 529 link->peer_qpn = ntoh24(clc->r0.qpn); 530 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 531 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 532 link->peer_psn = ntoh24(clc->r0.psn); 533 link->peer_mtu = clc->r0.qp_mtu; 534 } 535 536 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 537 struct smc_stats_fback *fback_arr) 538 { 539 int cnt; 540 541 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 542 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 543 fback_arr[cnt].count++; 544 break; 545 } 546 if (!fback_arr[cnt].fback_code) { 547 fback_arr[cnt].fback_code = smc->fallback_rsn; 548 fback_arr[cnt].count++; 549 break; 550 } 551 } 552 } 553 554 static void smc_stat_fallback(struct smc_sock *smc) 555 { 556 struct net *net = sock_net(&smc->sk); 557 558 mutex_lock(&net->smc.mutex_fback_rsn); 559 if (smc->listen_smc) { 560 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 561 net->smc.fback_rsn->srv_fback_cnt++; 562 } else { 563 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 564 net->smc.fback_rsn->clnt_fback_cnt++; 565 } 566 mutex_unlock(&net->smc.mutex_fback_rsn); 567 } 568 569 static void smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 570 { 571 wait_queue_head_t *smc_wait = sk_sleep(&smc->sk); 572 wait_queue_head_t *clc_wait = sk_sleep(smc->clcsock->sk); 573 unsigned long flags; 574 575 smc->use_fallback = true; 576 smc->fallback_rsn = reason_code; 577 smc_stat_fallback(smc); 578 trace_smc_switch_to_fallback(smc, reason_code); 579 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 580 smc->clcsock->file = smc->sk.sk_socket->file; 581 smc->clcsock->file->private_data = smc->clcsock; 582 smc->clcsock->wq.fasync_list = 583 smc->sk.sk_socket->wq.fasync_list; 584 585 /* There may be some entries remaining in 586 * smc socket->wq, which should be removed 587 * to clcsocket->wq during the fallback. 588 */ 589 spin_lock_irqsave(&smc_wait->lock, flags); 590 spin_lock_nested(&clc_wait->lock, SINGLE_DEPTH_NESTING); 591 list_splice_init(&smc_wait->head, &clc_wait->head); 592 spin_unlock(&clc_wait->lock); 593 spin_unlock_irqrestore(&smc_wait->lock, flags); 594 } 595 } 596 597 /* fall back during connect */ 598 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 599 { 600 smc_switch_to_fallback(smc, reason_code); 601 smc_copy_sock_settings_to_clc(smc); 602 smc->connect_nonblock = 0; 603 if (smc->sk.sk_state == SMC_INIT) 604 smc->sk.sk_state = SMC_ACTIVE; 605 return 0; 606 } 607 608 /* decline and fall back during connect */ 609 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 610 u8 version) 611 { 612 struct net *net = sock_net(&smc->sk); 613 int rc; 614 615 if (reason_code < 0) { /* error, fallback is not possible */ 616 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 617 if (smc->sk.sk_state == SMC_INIT) 618 sock_put(&smc->sk); /* passive closing */ 619 return reason_code; 620 } 621 if (reason_code != SMC_CLC_DECL_PEERDECL) { 622 rc = smc_clc_send_decline(smc, reason_code, version); 623 if (rc < 0) { 624 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 625 if (smc->sk.sk_state == SMC_INIT) 626 sock_put(&smc->sk); /* passive closing */ 627 return rc; 628 } 629 } 630 return smc_connect_fallback(smc, reason_code); 631 } 632 633 static void smc_conn_abort(struct smc_sock *smc, int local_first) 634 { 635 struct smc_connection *conn = &smc->conn; 636 struct smc_link_group *lgr = conn->lgr; 637 bool lgr_valid = false; 638 639 if (smc_conn_lgr_valid(conn)) 640 lgr_valid = true; 641 642 smc_conn_free(conn); 643 if (local_first && lgr_valid) 644 smc_lgr_cleanup_early(lgr); 645 } 646 647 /* check if there is a rdma device available for this connection. */ 648 /* called for connect and listen */ 649 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 650 { 651 /* PNET table look up: search active ib_device and port 652 * within same PNETID that also contains the ethernet device 653 * used for the internal TCP socket 654 */ 655 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 656 if (!ini->check_smcrv2 && !ini->ib_dev) 657 return SMC_CLC_DECL_NOSMCRDEV; 658 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 659 return SMC_CLC_DECL_NOSMCRDEV; 660 return 0; 661 } 662 663 /* check if there is an ISM device available for this connection. */ 664 /* called for connect and listen */ 665 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 666 { 667 /* Find ISM device with same PNETID as connecting interface */ 668 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 669 if (!ini->ism_dev[0]) 670 return SMC_CLC_DECL_NOSMCDDEV; 671 else 672 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 673 return 0; 674 } 675 676 /* is chid unique for the ism devices that are already determined? */ 677 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 678 int cnt) 679 { 680 int i = (!ini->ism_dev[0]) ? 1 : 0; 681 682 for (; i < cnt; i++) 683 if (ini->ism_chid[i] == chid) 684 return false; 685 return true; 686 } 687 688 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 689 * PNETID matching net_device) 690 */ 691 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 692 struct smc_init_info *ini) 693 { 694 int rc = SMC_CLC_DECL_NOSMCDDEV; 695 struct smcd_dev *smcd; 696 int i = 1; 697 u16 chid; 698 699 if (smcd_indicated(ini->smc_type_v1)) 700 rc = 0; /* already initialized for V1 */ 701 mutex_lock(&smcd_dev_list.mutex); 702 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 703 if (smcd->going_away || smcd == ini->ism_dev[0]) 704 continue; 705 chid = smc_ism_get_chid(smcd); 706 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 707 continue; 708 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 709 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 710 ini->ism_dev[i] = smcd; 711 ini->ism_chid[i] = chid; 712 ini->is_smcd = true; 713 rc = 0; 714 i++; 715 if (i > SMC_MAX_ISM_DEVS) 716 break; 717 } 718 } 719 mutex_unlock(&smcd_dev_list.mutex); 720 ini->ism_offered_cnt = i - 1; 721 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 722 ini->smcd_version = 0; 723 724 return rc; 725 } 726 727 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 728 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 729 struct smc_init_info *ini) 730 { 731 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 732 return SMC_CLC_DECL_ISMVLANERR; 733 return 0; 734 } 735 736 static int smc_find_proposal_devices(struct smc_sock *smc, 737 struct smc_init_info *ini) 738 { 739 int rc = 0; 740 741 /* check if there is an ism device available */ 742 if (!(ini->smcd_version & SMC_V1) || 743 smc_find_ism_device(smc, ini) || 744 smc_connect_ism_vlan_setup(smc, ini)) 745 ini->smcd_version &= ~SMC_V1; 746 /* else ISM V1 is supported for this connection */ 747 748 /* check if there is an rdma device available */ 749 if (!(ini->smcr_version & SMC_V1) || 750 smc_find_rdma_device(smc, ini)) 751 ini->smcr_version &= ~SMC_V1; 752 /* else RDMA is supported for this connection */ 753 754 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 755 ini->smcr_version & SMC_V1); 756 757 /* check if there is an ism v2 device available */ 758 if (!(ini->smcd_version & SMC_V2) || 759 !smc_ism_is_v2_capable() || 760 smc_find_ism_v2_device_clnt(smc, ini)) 761 ini->smcd_version &= ~SMC_V2; 762 763 /* check if there is an rdma v2 device available */ 764 ini->check_smcrv2 = true; 765 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 766 if (!(ini->smcr_version & SMC_V2) || 767 smc->clcsock->sk->sk_family != AF_INET || 768 !smc_clc_ueid_count() || 769 smc_find_rdma_device(smc, ini)) 770 ini->smcr_version &= ~SMC_V2; 771 ini->check_smcrv2 = false; 772 773 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 774 ini->smcr_version & SMC_V2); 775 776 /* if neither ISM nor RDMA are supported, fallback */ 777 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 778 rc = SMC_CLC_DECL_NOSMCDEV; 779 780 return rc; 781 } 782 783 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 784 * used, the VLAN ID will be registered again during the connection setup. 785 */ 786 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 787 struct smc_init_info *ini) 788 { 789 if (!smcd_indicated(ini->smc_type_v1)) 790 return 0; 791 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 792 return SMC_CLC_DECL_CNFERR; 793 return 0; 794 } 795 796 #define SMC_CLC_MAX_ACCEPT_LEN \ 797 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 798 sizeof(struct smc_clc_first_contact_ext) + \ 799 sizeof(struct smc_clc_msg_trail)) 800 801 /* CLC handshake during connect */ 802 static int smc_connect_clc(struct smc_sock *smc, 803 struct smc_clc_msg_accept_confirm_v2 *aclc2, 804 struct smc_init_info *ini) 805 { 806 int rc = 0; 807 808 /* do inband token exchange */ 809 rc = smc_clc_send_proposal(smc, ini); 810 if (rc) 811 return rc; 812 /* receive SMC Accept CLC message */ 813 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 814 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 815 } 816 817 void smc_fill_gid_list(struct smc_link_group *lgr, 818 struct smc_gidlist *gidlist, 819 struct smc_ib_device *known_dev, u8 *known_gid) 820 { 821 struct smc_init_info *alt_ini = NULL; 822 823 memset(gidlist, 0, sizeof(*gidlist)); 824 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 825 826 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 827 if (!alt_ini) 828 goto out; 829 830 alt_ini->vlan_id = lgr->vlan_id; 831 alt_ini->check_smcrv2 = true; 832 alt_ini->smcrv2.saddr = lgr->saddr; 833 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 834 835 if (!alt_ini->smcrv2.ib_dev_v2) 836 goto out; 837 838 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 839 SMC_GID_SIZE); 840 841 out: 842 kfree(alt_ini); 843 } 844 845 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 846 struct smc_clc_msg_accept_confirm *aclc, 847 struct smc_init_info *ini) 848 { 849 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 850 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 851 struct smc_clc_first_contact_ext *fce = 852 (struct smc_clc_first_contact_ext *) 853 (((u8 *)clc_v2) + sizeof(*clc_v2)); 854 855 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 856 return 0; 857 858 if (fce->v2_direct) { 859 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 860 ini->smcrv2.uses_gateway = false; 861 } else { 862 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr, 863 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 864 ini->smcrv2.nexthop_mac, 865 &ini->smcrv2.uses_gateway)) 866 return SMC_CLC_DECL_NOROUTE; 867 if (!ini->smcrv2.uses_gateway) { 868 /* mismatch: peer claims indirect, but its direct */ 869 return SMC_CLC_DECL_NOINDIRECT; 870 } 871 } 872 return 0; 873 } 874 875 /* setup for RDMA connection of client */ 876 static int smc_connect_rdma(struct smc_sock *smc, 877 struct smc_clc_msg_accept_confirm *aclc, 878 struct smc_init_info *ini) 879 { 880 int i, reason_code = 0; 881 struct smc_link *link; 882 u8 *eid = NULL; 883 884 ini->is_smcd = false; 885 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 886 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 887 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 888 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 889 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 890 891 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 892 if (reason_code) 893 return reason_code; 894 895 mutex_lock(&smc_client_lgr_pending); 896 reason_code = smc_conn_create(smc, ini); 897 if (reason_code) { 898 mutex_unlock(&smc_client_lgr_pending); 899 return reason_code; 900 } 901 902 smc_conn_save_peer_info(smc, aclc); 903 904 if (ini->first_contact_local) { 905 link = smc->conn.lnk; 906 } else { 907 /* set link that was assigned by server */ 908 link = NULL; 909 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 910 struct smc_link *l = &smc->conn.lgr->lnk[i]; 911 912 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 913 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 914 SMC_GID_SIZE) && 915 (aclc->hdr.version > SMC_V1 || 916 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 917 sizeof(l->peer_mac)))) { 918 link = l; 919 break; 920 } 921 } 922 if (!link) { 923 reason_code = SMC_CLC_DECL_NOSRVLINK; 924 goto connect_abort; 925 } 926 smc_switch_link_and_count(&smc->conn, link); 927 } 928 929 /* create send buffer and rmb */ 930 if (smc_buf_create(smc, false)) { 931 reason_code = SMC_CLC_DECL_MEM; 932 goto connect_abort; 933 } 934 935 if (ini->first_contact_local) 936 smc_link_save_peer_info(link, aclc, ini); 937 938 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 939 reason_code = SMC_CLC_DECL_ERR_RTOK; 940 goto connect_abort; 941 } 942 943 smc_close_init(smc); 944 smc_rx_init(smc); 945 946 if (ini->first_contact_local) { 947 if (smc_ib_ready_link(link)) { 948 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 949 goto connect_abort; 950 } 951 } else { 952 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 953 reason_code = SMC_CLC_DECL_ERR_REGRMB; 954 goto connect_abort; 955 } 956 } 957 smc_rmb_sync_sg_for_device(&smc->conn); 958 959 if (aclc->hdr.version > SMC_V1) { 960 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 961 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 962 963 eid = clc_v2->r1.eid; 964 if (ini->first_contact_local) 965 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 966 link->smcibdev, link->gid); 967 } 968 969 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 970 aclc->hdr.version, eid, ini); 971 if (reason_code) 972 goto connect_abort; 973 974 smc_tx_init(smc); 975 976 if (ini->first_contact_local) { 977 /* QP confirmation over RoCE fabric */ 978 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 979 reason_code = smcr_clnt_conf_first_link(smc); 980 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 981 if (reason_code) 982 goto connect_abort; 983 } 984 mutex_unlock(&smc_client_lgr_pending); 985 986 smc_copy_sock_settings_to_clc(smc); 987 smc->connect_nonblock = 0; 988 if (smc->sk.sk_state == SMC_INIT) 989 smc->sk.sk_state = SMC_ACTIVE; 990 991 return 0; 992 connect_abort: 993 smc_conn_abort(smc, ini->first_contact_local); 994 mutex_unlock(&smc_client_lgr_pending); 995 smc->connect_nonblock = 0; 996 997 return reason_code; 998 } 999 1000 /* The server has chosen one of the proposed ISM devices for the communication. 1001 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1002 */ 1003 static int 1004 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 1005 struct smc_init_info *ini) 1006 { 1007 int i; 1008 1009 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1010 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1011 ini->ism_selected = i; 1012 return 0; 1013 } 1014 } 1015 1016 return -EPROTO; 1017 } 1018 1019 /* setup for ISM connection of client */ 1020 static int smc_connect_ism(struct smc_sock *smc, 1021 struct smc_clc_msg_accept_confirm *aclc, 1022 struct smc_init_info *ini) 1023 { 1024 u8 *eid = NULL; 1025 int rc = 0; 1026 1027 ini->is_smcd = true; 1028 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1029 1030 if (aclc->hdr.version == SMC_V2) { 1031 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1032 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1033 1034 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1035 if (rc) 1036 return rc; 1037 } 1038 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 1039 1040 /* there is only one lgr role for SMC-D; use server lock */ 1041 mutex_lock(&smc_server_lgr_pending); 1042 rc = smc_conn_create(smc, ini); 1043 if (rc) { 1044 mutex_unlock(&smc_server_lgr_pending); 1045 return rc; 1046 } 1047 1048 /* Create send and receive buffers */ 1049 rc = smc_buf_create(smc, true); 1050 if (rc) { 1051 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1052 goto connect_abort; 1053 } 1054 1055 smc_conn_save_peer_info(smc, aclc); 1056 smc_close_init(smc); 1057 smc_rx_init(smc); 1058 smc_tx_init(smc); 1059 1060 if (aclc->hdr.version > SMC_V1) { 1061 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1062 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1063 1064 eid = clc_v2->d1.eid; 1065 } 1066 1067 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1068 aclc->hdr.version, eid, NULL); 1069 if (rc) 1070 goto connect_abort; 1071 mutex_unlock(&smc_server_lgr_pending); 1072 1073 smc_copy_sock_settings_to_clc(smc); 1074 smc->connect_nonblock = 0; 1075 if (smc->sk.sk_state == SMC_INIT) 1076 smc->sk.sk_state = SMC_ACTIVE; 1077 1078 return 0; 1079 connect_abort: 1080 smc_conn_abort(smc, ini->first_contact_local); 1081 mutex_unlock(&smc_server_lgr_pending); 1082 smc->connect_nonblock = 0; 1083 1084 return rc; 1085 } 1086 1087 /* check if received accept type and version matches a proposed one */ 1088 static int smc_connect_check_aclc(struct smc_init_info *ini, 1089 struct smc_clc_msg_accept_confirm *aclc) 1090 { 1091 if (aclc->hdr.typev1 != SMC_TYPE_R && 1092 aclc->hdr.typev1 != SMC_TYPE_D) 1093 return SMC_CLC_DECL_MODEUNSUPP; 1094 1095 if (aclc->hdr.version >= SMC_V2) { 1096 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1097 !smcr_indicated(ini->smc_type_v2)) || 1098 (aclc->hdr.typev1 == SMC_TYPE_D && 1099 !smcd_indicated(ini->smc_type_v2))) 1100 return SMC_CLC_DECL_MODEUNSUPP; 1101 } else { 1102 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1103 !smcr_indicated(ini->smc_type_v1)) || 1104 (aclc->hdr.typev1 == SMC_TYPE_D && 1105 !smcd_indicated(ini->smc_type_v1))) 1106 return SMC_CLC_DECL_MODEUNSUPP; 1107 } 1108 1109 return 0; 1110 } 1111 1112 /* perform steps before actually connecting */ 1113 static int __smc_connect(struct smc_sock *smc) 1114 { 1115 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1116 struct smc_clc_msg_accept_confirm_v2 *aclc2; 1117 struct smc_clc_msg_accept_confirm *aclc; 1118 struct smc_init_info *ini = NULL; 1119 u8 *buf = NULL; 1120 int rc = 0; 1121 1122 if (smc->use_fallback) 1123 return smc_connect_fallback(smc, smc->fallback_rsn); 1124 1125 /* if peer has not signalled SMC-capability, fall back */ 1126 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1127 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1128 1129 /* IPSec connections opt out of SMC optimizations */ 1130 if (using_ipsec(smc)) 1131 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1132 version); 1133 1134 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1135 if (!ini) 1136 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1137 version); 1138 1139 ini->smcd_version = SMC_V1 | SMC_V2; 1140 ini->smcr_version = SMC_V1 | SMC_V2; 1141 ini->smc_type_v1 = SMC_TYPE_B; 1142 ini->smc_type_v2 = SMC_TYPE_B; 1143 1144 /* get vlan id from IP device */ 1145 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1146 ini->smcd_version &= ~SMC_V1; 1147 ini->smcr_version = 0; 1148 ini->smc_type_v1 = SMC_TYPE_N; 1149 if (!ini->smcd_version) { 1150 rc = SMC_CLC_DECL_GETVLANERR; 1151 goto fallback; 1152 } 1153 } 1154 1155 rc = smc_find_proposal_devices(smc, ini); 1156 if (rc) 1157 goto fallback; 1158 1159 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1160 if (!buf) { 1161 rc = SMC_CLC_DECL_MEM; 1162 goto fallback; 1163 } 1164 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1165 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1166 1167 /* perform CLC handshake */ 1168 rc = smc_connect_clc(smc, aclc2, ini); 1169 if (rc) 1170 goto vlan_cleanup; 1171 1172 /* check if smc modes and versions of CLC proposal and accept match */ 1173 rc = smc_connect_check_aclc(ini, aclc); 1174 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1175 if (rc) 1176 goto vlan_cleanup; 1177 1178 /* depending on previous steps, connect using rdma or ism */ 1179 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1180 ini->smcr_version = version; 1181 rc = smc_connect_rdma(smc, aclc, ini); 1182 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1183 ini->smcd_version = version; 1184 rc = smc_connect_ism(smc, aclc, ini); 1185 } 1186 if (rc) 1187 goto vlan_cleanup; 1188 1189 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1190 smc_connect_ism_vlan_cleanup(smc, ini); 1191 kfree(buf); 1192 kfree(ini); 1193 return 0; 1194 1195 vlan_cleanup: 1196 smc_connect_ism_vlan_cleanup(smc, ini); 1197 kfree(buf); 1198 fallback: 1199 kfree(ini); 1200 return smc_connect_decline_fallback(smc, rc, version); 1201 } 1202 1203 static void smc_connect_work(struct work_struct *work) 1204 { 1205 struct smc_sock *smc = container_of(work, struct smc_sock, 1206 connect_work); 1207 long timeo = smc->sk.sk_sndtimeo; 1208 int rc = 0; 1209 1210 if (!timeo) 1211 timeo = MAX_SCHEDULE_TIMEOUT; 1212 lock_sock(smc->clcsock->sk); 1213 if (smc->clcsock->sk->sk_err) { 1214 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1215 } else if ((1 << smc->clcsock->sk->sk_state) & 1216 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1217 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1218 if ((rc == -EPIPE) && 1219 ((1 << smc->clcsock->sk->sk_state) & 1220 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1221 rc = 0; 1222 } 1223 release_sock(smc->clcsock->sk); 1224 lock_sock(&smc->sk); 1225 if (rc != 0 || smc->sk.sk_err) { 1226 smc->sk.sk_state = SMC_CLOSED; 1227 if (rc == -EPIPE || rc == -EAGAIN) 1228 smc->sk.sk_err = EPIPE; 1229 else if (signal_pending(current)) 1230 smc->sk.sk_err = -sock_intr_errno(timeo); 1231 sock_put(&smc->sk); /* passive closing */ 1232 goto out; 1233 } 1234 1235 rc = __smc_connect(smc); 1236 if (rc < 0) 1237 smc->sk.sk_err = -rc; 1238 1239 out: 1240 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1241 if (smc->sk.sk_err) { 1242 smc->sk.sk_state_change(&smc->sk); 1243 } else { /* allow polling before and after fallback decision */ 1244 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1245 smc->sk.sk_write_space(&smc->sk); 1246 } 1247 } 1248 release_sock(&smc->sk); 1249 } 1250 1251 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1252 int alen, int flags) 1253 { 1254 struct sock *sk = sock->sk; 1255 struct smc_sock *smc; 1256 int rc = -EINVAL; 1257 1258 smc = smc_sk(sk); 1259 1260 /* separate smc parameter checking to be safe */ 1261 if (alen < sizeof(addr->sa_family)) 1262 goto out_err; 1263 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1264 goto out_err; 1265 1266 lock_sock(sk); 1267 switch (sk->sk_state) { 1268 default: 1269 goto out; 1270 case SMC_ACTIVE: 1271 rc = -EISCONN; 1272 goto out; 1273 case SMC_INIT: 1274 break; 1275 } 1276 1277 smc_copy_sock_settings_to_clc(smc); 1278 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1279 if (smc->connect_nonblock) { 1280 rc = -EALREADY; 1281 goto out; 1282 } 1283 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1284 if (rc && rc != -EINPROGRESS) 1285 goto out; 1286 1287 sock_hold(&smc->sk); /* sock put in passive closing */ 1288 if (smc->use_fallback) 1289 goto out; 1290 if (flags & O_NONBLOCK) { 1291 if (queue_work(smc_hs_wq, &smc->connect_work)) 1292 smc->connect_nonblock = 1; 1293 rc = -EINPROGRESS; 1294 } else { 1295 rc = __smc_connect(smc); 1296 if (rc < 0) 1297 goto out; 1298 else 1299 rc = 0; /* success cases including fallback */ 1300 } 1301 1302 out: 1303 release_sock(sk); 1304 out_err: 1305 return rc; 1306 } 1307 1308 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1309 { 1310 struct socket *new_clcsock = NULL; 1311 struct sock *lsk = &lsmc->sk; 1312 struct sock *new_sk; 1313 int rc = -EINVAL; 1314 1315 release_sock(lsk); 1316 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1317 if (!new_sk) { 1318 rc = -ENOMEM; 1319 lsk->sk_err = ENOMEM; 1320 *new_smc = NULL; 1321 lock_sock(lsk); 1322 goto out; 1323 } 1324 *new_smc = smc_sk(new_sk); 1325 1326 mutex_lock(&lsmc->clcsock_release_lock); 1327 if (lsmc->clcsock) 1328 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1329 mutex_unlock(&lsmc->clcsock_release_lock); 1330 lock_sock(lsk); 1331 if (rc < 0 && rc != -EAGAIN) 1332 lsk->sk_err = -rc; 1333 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1334 new_sk->sk_prot->unhash(new_sk); 1335 if (new_clcsock) 1336 sock_release(new_clcsock); 1337 new_sk->sk_state = SMC_CLOSED; 1338 sock_set_flag(new_sk, SOCK_DEAD); 1339 sock_put(new_sk); /* final */ 1340 *new_smc = NULL; 1341 goto out; 1342 } 1343 1344 /* new clcsock has inherited the smc listen-specific sk_data_ready 1345 * function; switch it back to the original sk_data_ready function 1346 */ 1347 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1348 (*new_smc)->clcsock = new_clcsock; 1349 out: 1350 return rc; 1351 } 1352 1353 /* add a just created sock to the accept queue of the listen sock as 1354 * candidate for a following socket accept call from user space 1355 */ 1356 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1357 { 1358 struct smc_sock *par = smc_sk(parent); 1359 1360 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1361 spin_lock(&par->accept_q_lock); 1362 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1363 spin_unlock(&par->accept_q_lock); 1364 sk_acceptq_added(parent); 1365 } 1366 1367 /* remove a socket from the accept queue of its parental listening socket */ 1368 static void smc_accept_unlink(struct sock *sk) 1369 { 1370 struct smc_sock *par = smc_sk(sk)->listen_smc; 1371 1372 spin_lock(&par->accept_q_lock); 1373 list_del_init(&smc_sk(sk)->accept_q); 1374 spin_unlock(&par->accept_q_lock); 1375 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1376 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1377 } 1378 1379 /* remove a sock from the accept queue to bind it to a new socket created 1380 * for a socket accept call from user space 1381 */ 1382 struct sock *smc_accept_dequeue(struct sock *parent, 1383 struct socket *new_sock) 1384 { 1385 struct smc_sock *isk, *n; 1386 struct sock *new_sk; 1387 1388 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1389 new_sk = (struct sock *)isk; 1390 1391 smc_accept_unlink(new_sk); 1392 if (new_sk->sk_state == SMC_CLOSED) { 1393 new_sk->sk_prot->unhash(new_sk); 1394 if (isk->clcsock) { 1395 sock_release(isk->clcsock); 1396 isk->clcsock = NULL; 1397 } 1398 sock_put(new_sk); /* final */ 1399 continue; 1400 } 1401 if (new_sock) { 1402 sock_graft(new_sk, new_sock); 1403 if (isk->use_fallback) { 1404 smc_sk(new_sk)->clcsock->file = new_sock->file; 1405 isk->clcsock->file->private_data = isk->clcsock; 1406 } 1407 } 1408 return new_sk; 1409 } 1410 return NULL; 1411 } 1412 1413 /* clean up for a created but never accepted sock */ 1414 void smc_close_non_accepted(struct sock *sk) 1415 { 1416 struct smc_sock *smc = smc_sk(sk); 1417 1418 sock_hold(sk); /* sock_put below */ 1419 lock_sock(sk); 1420 if (!sk->sk_lingertime) 1421 /* wait for peer closing */ 1422 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1423 __smc_release(smc); 1424 release_sock(sk); 1425 sock_put(sk); /* sock_hold above */ 1426 sock_put(sk); /* final sock_put */ 1427 } 1428 1429 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1430 { 1431 struct smc_link *link = smc->conn.lnk; 1432 struct smc_llc_qentry *qentry; 1433 int rc; 1434 1435 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 1436 return SMC_CLC_DECL_ERR_REGRMB; 1437 1438 /* send CONFIRM LINK request to client over the RoCE fabric */ 1439 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1440 if (rc < 0) 1441 return SMC_CLC_DECL_TIMEOUT_CL; 1442 1443 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1444 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1445 SMC_LLC_CONFIRM_LINK); 1446 if (!qentry) { 1447 struct smc_clc_msg_decline dclc; 1448 1449 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1450 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1451 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1452 } 1453 smc_llc_save_peer_uid(qentry); 1454 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1455 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1456 if (rc) 1457 return SMC_CLC_DECL_RMBE_EC; 1458 1459 /* confirm_rkey is implicit on 1st contact */ 1460 smc->conn.rmb_desc->is_conf_rkey = true; 1461 1462 smc_llc_link_active(link); 1463 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1464 1465 /* initial contact - try to establish second link */ 1466 smc_llc_srv_add_link(link, NULL); 1467 return 0; 1468 } 1469 1470 /* listen worker: finish */ 1471 static void smc_listen_out(struct smc_sock *new_smc) 1472 { 1473 struct smc_sock *lsmc = new_smc->listen_smc; 1474 struct sock *newsmcsk = &new_smc->sk; 1475 1476 if (lsmc->sk.sk_state == SMC_LISTEN) { 1477 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1478 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1479 release_sock(&lsmc->sk); 1480 } else { /* no longer listening */ 1481 smc_close_non_accepted(newsmcsk); 1482 } 1483 1484 /* Wake up accept */ 1485 lsmc->sk.sk_data_ready(&lsmc->sk); 1486 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1487 } 1488 1489 /* listen worker: finish in state connected */ 1490 static void smc_listen_out_connected(struct smc_sock *new_smc) 1491 { 1492 struct sock *newsmcsk = &new_smc->sk; 1493 1494 sk_refcnt_debug_inc(newsmcsk); 1495 if (newsmcsk->sk_state == SMC_INIT) 1496 newsmcsk->sk_state = SMC_ACTIVE; 1497 1498 smc_listen_out(new_smc); 1499 } 1500 1501 /* listen worker: finish in error state */ 1502 static void smc_listen_out_err(struct smc_sock *new_smc) 1503 { 1504 struct sock *newsmcsk = &new_smc->sk; 1505 struct net *net = sock_net(newsmcsk); 1506 1507 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1508 if (newsmcsk->sk_state == SMC_INIT) 1509 sock_put(&new_smc->sk); /* passive closing */ 1510 newsmcsk->sk_state = SMC_CLOSED; 1511 1512 smc_listen_out(new_smc); 1513 } 1514 1515 /* listen worker: decline and fall back if possible */ 1516 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1517 int local_first, u8 version) 1518 { 1519 /* RDMA setup failed, switch back to TCP */ 1520 smc_conn_abort(new_smc, local_first); 1521 if (reason_code < 0) { /* error, no fallback possible */ 1522 smc_listen_out_err(new_smc); 1523 return; 1524 } 1525 smc_switch_to_fallback(new_smc, reason_code); 1526 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1527 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1528 smc_listen_out_err(new_smc); 1529 return; 1530 } 1531 } 1532 smc_listen_out_connected(new_smc); 1533 } 1534 1535 /* listen worker: version checking */ 1536 static int smc_listen_v2_check(struct smc_sock *new_smc, 1537 struct smc_clc_msg_proposal *pclc, 1538 struct smc_init_info *ini) 1539 { 1540 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1541 struct smc_clc_v2_extension *pclc_v2_ext; 1542 int rc = SMC_CLC_DECL_PEERNOSMC; 1543 1544 ini->smc_type_v1 = pclc->hdr.typev1; 1545 ini->smc_type_v2 = pclc->hdr.typev2; 1546 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1547 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1548 if (pclc->hdr.version > SMC_V1) { 1549 if (smcd_indicated(ini->smc_type_v2)) 1550 ini->smcd_version |= SMC_V2; 1551 if (smcr_indicated(ini->smc_type_v2)) 1552 ini->smcr_version |= SMC_V2; 1553 } 1554 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1555 rc = SMC_CLC_DECL_PEERNOSMC; 1556 goto out; 1557 } 1558 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1559 if (!pclc_v2_ext) { 1560 ini->smcd_version &= ~SMC_V2; 1561 ini->smcr_version &= ~SMC_V2; 1562 rc = SMC_CLC_DECL_NOV2EXT; 1563 goto out; 1564 } 1565 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1566 if (ini->smcd_version & SMC_V2) { 1567 if (!smc_ism_is_v2_capable()) { 1568 ini->smcd_version &= ~SMC_V2; 1569 rc = SMC_CLC_DECL_NOISM2SUPP; 1570 } else if (!pclc_smcd_v2_ext) { 1571 ini->smcd_version &= ~SMC_V2; 1572 rc = SMC_CLC_DECL_NOV2DEXT; 1573 } else if (!pclc_v2_ext->hdr.eid_cnt && 1574 !pclc_v2_ext->hdr.flag.seid) { 1575 ini->smcd_version &= ~SMC_V2; 1576 rc = SMC_CLC_DECL_NOUEID; 1577 } 1578 } 1579 if (ini->smcr_version & SMC_V2) { 1580 if (!pclc_v2_ext->hdr.eid_cnt) { 1581 ini->smcr_version &= ~SMC_V2; 1582 rc = SMC_CLC_DECL_NOUEID; 1583 } 1584 } 1585 1586 out: 1587 if (!ini->smcd_version && !ini->smcr_version) 1588 return rc; 1589 1590 return 0; 1591 } 1592 1593 /* listen worker: check prefixes */ 1594 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1595 struct smc_clc_msg_proposal *pclc) 1596 { 1597 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1598 struct socket *newclcsock = new_smc->clcsock; 1599 1600 if (pclc->hdr.typev1 == SMC_TYPE_N) 1601 return 0; 1602 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1603 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1604 return SMC_CLC_DECL_DIFFPREFIX; 1605 1606 return 0; 1607 } 1608 1609 /* listen worker: initialize connection and buffers */ 1610 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1611 struct smc_init_info *ini) 1612 { 1613 int rc; 1614 1615 /* allocate connection / link group */ 1616 rc = smc_conn_create(new_smc, ini); 1617 if (rc) 1618 return rc; 1619 1620 /* create send buffer and rmb */ 1621 if (smc_buf_create(new_smc, false)) 1622 return SMC_CLC_DECL_MEM; 1623 1624 return 0; 1625 } 1626 1627 /* listen worker: initialize connection and buffers for SMC-D */ 1628 static int smc_listen_ism_init(struct smc_sock *new_smc, 1629 struct smc_init_info *ini) 1630 { 1631 int rc; 1632 1633 rc = smc_conn_create(new_smc, ini); 1634 if (rc) 1635 return rc; 1636 1637 /* Create send and receive buffers */ 1638 rc = smc_buf_create(new_smc, true); 1639 if (rc) { 1640 smc_conn_abort(new_smc, ini->first_contact_local); 1641 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 1642 SMC_CLC_DECL_MEM; 1643 } 1644 1645 return 0; 1646 } 1647 1648 static bool smc_is_already_selected(struct smcd_dev *smcd, 1649 struct smc_init_info *ini, 1650 int matches) 1651 { 1652 int i; 1653 1654 for (i = 0; i < matches; i++) 1655 if (smcd == ini->ism_dev[i]) 1656 return true; 1657 1658 return false; 1659 } 1660 1661 /* check for ISM devices matching proposed ISM devices */ 1662 static void smc_check_ism_v2_match(struct smc_init_info *ini, 1663 u16 proposed_chid, u64 proposed_gid, 1664 unsigned int *matches) 1665 { 1666 struct smcd_dev *smcd; 1667 1668 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1669 if (smcd->going_away) 1670 continue; 1671 if (smc_is_already_selected(smcd, ini, *matches)) 1672 continue; 1673 if (smc_ism_get_chid(smcd) == proposed_chid && 1674 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 1675 ini->ism_peer_gid[*matches] = proposed_gid; 1676 ini->ism_dev[*matches] = smcd; 1677 (*matches)++; 1678 break; 1679 } 1680 } 1681 } 1682 1683 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 1684 { 1685 if (!ini->rc) 1686 ini->rc = rc; 1687 } 1688 1689 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 1690 struct smc_clc_msg_proposal *pclc, 1691 struct smc_init_info *ini) 1692 { 1693 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 1694 struct smc_clc_v2_extension *smc_v2_ext; 1695 struct smc_clc_msg_smcd *pclc_smcd; 1696 unsigned int matches = 0; 1697 u8 smcd_version; 1698 u8 *eid = NULL; 1699 int i, rc; 1700 1701 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 1702 goto not_found; 1703 1704 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1705 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1706 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 1707 1708 mutex_lock(&smcd_dev_list.mutex); 1709 if (pclc_smcd->ism.chid) 1710 /* check for ISM device matching proposed native ISM device */ 1711 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 1712 ntohll(pclc_smcd->ism.gid), &matches); 1713 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 1714 /* check for ISM devices matching proposed non-native ISM 1715 * devices 1716 */ 1717 smc_check_ism_v2_match(ini, 1718 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 1719 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 1720 &matches); 1721 } 1722 mutex_unlock(&smcd_dev_list.mutex); 1723 1724 if (!ini->ism_dev[0]) { 1725 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 1726 goto not_found; 1727 } 1728 1729 smc_ism_get_system_eid(&eid); 1730 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 1731 smcd_v2_ext->system_eid, eid)) 1732 goto not_found; 1733 1734 /* separate - outside the smcd_dev_list.lock */ 1735 smcd_version = ini->smcd_version; 1736 for (i = 0; i < matches; i++) { 1737 ini->smcd_version = SMC_V2; 1738 ini->is_smcd = true; 1739 ini->ism_selected = i; 1740 rc = smc_listen_ism_init(new_smc, ini); 1741 if (rc) { 1742 smc_find_ism_store_rc(rc, ini); 1743 /* try next active ISM device */ 1744 continue; 1745 } 1746 return; /* matching and usable V2 ISM device found */ 1747 } 1748 /* no V2 ISM device could be initialized */ 1749 ini->smcd_version = smcd_version; /* restore original value */ 1750 ini->negotiated_eid[0] = 0; 1751 1752 not_found: 1753 ini->smcd_version &= ~SMC_V2; 1754 ini->ism_dev[0] = NULL; 1755 ini->is_smcd = false; 1756 } 1757 1758 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 1759 struct smc_clc_msg_proposal *pclc, 1760 struct smc_init_info *ini) 1761 { 1762 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 1763 int rc = 0; 1764 1765 /* check if ISM V1 is available */ 1766 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 1767 goto not_found; 1768 ini->is_smcd = true; /* prepare ISM check */ 1769 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 1770 rc = smc_find_ism_device(new_smc, ini); 1771 if (rc) 1772 goto not_found; 1773 ini->ism_selected = 0; 1774 rc = smc_listen_ism_init(new_smc, ini); 1775 if (!rc) 1776 return; /* V1 ISM device found */ 1777 1778 not_found: 1779 smc_find_ism_store_rc(rc, ini); 1780 ini->smcd_version &= ~SMC_V1; 1781 ini->ism_dev[0] = NULL; 1782 ini->is_smcd = false; 1783 } 1784 1785 /* listen worker: register buffers */ 1786 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 1787 { 1788 struct smc_connection *conn = &new_smc->conn; 1789 1790 if (!local_first) { 1791 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 1792 return SMC_CLC_DECL_ERR_REGRMB; 1793 } 1794 smc_rmb_sync_sg_for_device(&new_smc->conn); 1795 1796 return 0; 1797 } 1798 1799 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 1800 struct smc_clc_msg_proposal *pclc, 1801 struct smc_init_info *ini) 1802 { 1803 struct smc_clc_v2_extension *smc_v2_ext; 1804 u8 smcr_version; 1805 int rc; 1806 1807 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 1808 goto not_found; 1809 1810 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1811 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 1812 goto not_found; 1813 1814 /* prepare RDMA check */ 1815 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 1816 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 1817 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 1818 ini->check_smcrv2 = true; 1819 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 1820 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 1821 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 1822 rc = smc_find_rdma_device(new_smc, ini); 1823 if (rc) { 1824 smc_find_ism_store_rc(rc, ini); 1825 goto not_found; 1826 } 1827 if (!ini->smcrv2.uses_gateway) 1828 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 1829 1830 smcr_version = ini->smcr_version; 1831 ini->smcr_version = SMC_V2; 1832 rc = smc_listen_rdma_init(new_smc, ini); 1833 if (!rc) 1834 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 1835 if (!rc) 1836 return; 1837 ini->smcr_version = smcr_version; 1838 smc_find_ism_store_rc(rc, ini); 1839 1840 not_found: 1841 ini->smcr_version &= ~SMC_V2; 1842 ini->check_smcrv2 = false; 1843 } 1844 1845 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 1846 struct smc_clc_msg_proposal *pclc, 1847 struct smc_init_info *ini) 1848 { 1849 int rc; 1850 1851 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 1852 return SMC_CLC_DECL_NOSMCDEV; 1853 1854 /* prepare RDMA check */ 1855 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 1856 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 1857 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 1858 rc = smc_find_rdma_device(new_smc, ini); 1859 if (rc) { 1860 /* no RDMA device found */ 1861 return SMC_CLC_DECL_NOSMCDEV; 1862 } 1863 rc = smc_listen_rdma_init(new_smc, ini); 1864 if (rc) 1865 return rc; 1866 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 1867 } 1868 1869 /* determine the local device matching to proposal */ 1870 static int smc_listen_find_device(struct smc_sock *new_smc, 1871 struct smc_clc_msg_proposal *pclc, 1872 struct smc_init_info *ini) 1873 { 1874 int prfx_rc; 1875 1876 /* check for ISM device matching V2 proposed device */ 1877 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 1878 if (ini->ism_dev[0]) 1879 return 0; 1880 1881 /* check for matching IP prefix and subnet length (V1) */ 1882 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 1883 if (prfx_rc) 1884 smc_find_ism_store_rc(prfx_rc, ini); 1885 1886 /* get vlan id from IP device */ 1887 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 1888 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 1889 1890 /* check for ISM device matching V1 proposed device */ 1891 if (!prfx_rc) 1892 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 1893 if (ini->ism_dev[0]) 1894 return 0; 1895 1896 if (!smcr_indicated(pclc->hdr.typev1) && 1897 !smcr_indicated(pclc->hdr.typev2)) 1898 /* skip RDMA and decline */ 1899 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 1900 1901 /* check if RDMA V2 is available */ 1902 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 1903 if (ini->smcrv2.ib_dev_v2) 1904 return 0; 1905 1906 /* check if RDMA V1 is available */ 1907 if (!prfx_rc) { 1908 int rc; 1909 1910 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 1911 smc_find_ism_store_rc(rc, ini); 1912 return (!rc) ? 0 : ini->rc; 1913 } 1914 return SMC_CLC_DECL_NOSMCDEV; 1915 } 1916 1917 /* listen worker: finish RDMA setup */ 1918 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1919 struct smc_clc_msg_accept_confirm *cclc, 1920 bool local_first, 1921 struct smc_init_info *ini) 1922 { 1923 struct smc_link *link = new_smc->conn.lnk; 1924 int reason_code = 0; 1925 1926 if (local_first) 1927 smc_link_save_peer_info(link, cclc, ini); 1928 1929 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 1930 return SMC_CLC_DECL_ERR_RTOK; 1931 1932 if (local_first) { 1933 if (smc_ib_ready_link(link)) 1934 return SMC_CLC_DECL_ERR_RDYLNK; 1935 /* QP confirmation over RoCE fabric */ 1936 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1937 reason_code = smcr_serv_conf_first_link(new_smc); 1938 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1939 } 1940 return reason_code; 1941 } 1942 1943 /* setup for connection of server */ 1944 static void smc_listen_work(struct work_struct *work) 1945 { 1946 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1947 smc_listen_work); 1948 struct socket *newclcsock = new_smc->clcsock; 1949 struct smc_clc_msg_accept_confirm *cclc; 1950 struct smc_clc_msg_proposal_area *buf; 1951 struct smc_clc_msg_proposal *pclc; 1952 struct smc_init_info *ini = NULL; 1953 u8 proposal_version = SMC_V1; 1954 u8 accept_version; 1955 int rc = 0; 1956 1957 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1958 return smc_listen_out_err(new_smc); 1959 1960 if (new_smc->use_fallback) { 1961 smc_listen_out_connected(new_smc); 1962 return; 1963 } 1964 1965 /* check if peer is smc capable */ 1966 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1967 smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 1968 smc_listen_out_connected(new_smc); 1969 return; 1970 } 1971 1972 /* do inband token exchange - 1973 * wait for and receive SMC Proposal CLC message 1974 */ 1975 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 1976 if (!buf) { 1977 rc = SMC_CLC_DECL_MEM; 1978 goto out_decl; 1979 } 1980 pclc = (struct smc_clc_msg_proposal *)buf; 1981 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 1982 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1983 if (rc) 1984 goto out_decl; 1985 1986 if (pclc->hdr.version > SMC_V1) 1987 proposal_version = SMC_V2; 1988 1989 /* IPSec connections opt out of SMC optimizations */ 1990 if (using_ipsec(new_smc)) { 1991 rc = SMC_CLC_DECL_IPSEC; 1992 goto out_decl; 1993 } 1994 1995 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1996 if (!ini) { 1997 rc = SMC_CLC_DECL_MEM; 1998 goto out_decl; 1999 } 2000 2001 /* initial version checking */ 2002 rc = smc_listen_v2_check(new_smc, pclc, ini); 2003 if (rc) 2004 goto out_decl; 2005 2006 mutex_lock(&smc_server_lgr_pending); 2007 smc_close_init(new_smc); 2008 smc_rx_init(new_smc); 2009 smc_tx_init(new_smc); 2010 2011 /* determine ISM or RoCE device used for connection */ 2012 rc = smc_listen_find_device(new_smc, pclc, ini); 2013 if (rc) 2014 goto out_unlock; 2015 2016 /* send SMC Accept CLC message */ 2017 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2018 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2019 accept_version, ini->negotiated_eid); 2020 if (rc) 2021 goto out_unlock; 2022 2023 /* SMC-D does not need this lock any more */ 2024 if (ini->is_smcd) 2025 mutex_unlock(&smc_server_lgr_pending); 2026 2027 /* receive SMC Confirm CLC message */ 2028 memset(buf, 0, sizeof(*buf)); 2029 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2030 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2031 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2032 if (rc) { 2033 if (!ini->is_smcd) 2034 goto out_unlock; 2035 goto out_decl; 2036 } 2037 2038 /* finish worker */ 2039 if (!ini->is_smcd) { 2040 rc = smc_listen_rdma_finish(new_smc, cclc, 2041 ini->first_contact_local, ini); 2042 if (rc) 2043 goto out_unlock; 2044 mutex_unlock(&smc_server_lgr_pending); 2045 } 2046 smc_conn_save_peer_info(new_smc, cclc); 2047 smc_listen_out_connected(new_smc); 2048 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2049 goto out_free; 2050 2051 out_unlock: 2052 mutex_unlock(&smc_server_lgr_pending); 2053 out_decl: 2054 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2055 proposal_version); 2056 out_free: 2057 kfree(ini); 2058 kfree(buf); 2059 } 2060 2061 static void smc_tcp_listen_work(struct work_struct *work) 2062 { 2063 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2064 tcp_listen_work); 2065 struct sock *lsk = &lsmc->sk; 2066 struct smc_sock *new_smc; 2067 int rc = 0; 2068 2069 lock_sock(lsk); 2070 while (lsk->sk_state == SMC_LISTEN) { 2071 rc = smc_clcsock_accept(lsmc, &new_smc); 2072 if (rc) /* clcsock accept queue empty or error */ 2073 goto out; 2074 if (!new_smc) 2075 continue; 2076 2077 new_smc->listen_smc = lsmc; 2078 new_smc->use_fallback = lsmc->use_fallback; 2079 new_smc->fallback_rsn = lsmc->fallback_rsn; 2080 sock_hold(lsk); /* sock_put in smc_listen_work */ 2081 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2082 smc_copy_sock_settings_to_smc(new_smc); 2083 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 2084 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 2085 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2086 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2087 sock_put(&new_smc->sk); 2088 } 2089 2090 out: 2091 release_sock(lsk); 2092 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2093 } 2094 2095 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2096 { 2097 struct smc_sock *lsmc; 2098 2099 lsmc = (struct smc_sock *) 2100 ((uintptr_t)listen_clcsock->sk_user_data & ~SK_USER_DATA_NOCOPY); 2101 if (!lsmc) 2102 return; 2103 lsmc->clcsk_data_ready(listen_clcsock); 2104 if (lsmc->sk.sk_state == SMC_LISTEN) { 2105 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2106 if (!queue_work(smc_hs_wq, &lsmc->tcp_listen_work)) 2107 sock_put(&lsmc->sk); 2108 } 2109 } 2110 2111 static int smc_listen(struct socket *sock, int backlog) 2112 { 2113 struct sock *sk = sock->sk; 2114 struct smc_sock *smc; 2115 int rc; 2116 2117 smc = smc_sk(sk); 2118 lock_sock(sk); 2119 2120 rc = -EINVAL; 2121 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2122 smc->connect_nonblock) 2123 goto out; 2124 2125 rc = 0; 2126 if (sk->sk_state == SMC_LISTEN) { 2127 sk->sk_max_ack_backlog = backlog; 2128 goto out; 2129 } 2130 /* some socket options are handled in core, so we could not apply 2131 * them to the clc socket -- copy smc socket options to clc socket 2132 */ 2133 smc_copy_sock_settings_to_clc(smc); 2134 if (!smc->use_fallback) 2135 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2136 2137 /* save original sk_data_ready function and establish 2138 * smc-specific sk_data_ready function 2139 */ 2140 smc->clcsk_data_ready = smc->clcsock->sk->sk_data_ready; 2141 smc->clcsock->sk->sk_data_ready = smc_clcsock_data_ready; 2142 smc->clcsock->sk->sk_user_data = 2143 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2144 rc = kernel_listen(smc->clcsock, backlog); 2145 if (rc) { 2146 smc->clcsock->sk->sk_data_ready = smc->clcsk_data_ready; 2147 goto out; 2148 } 2149 sk->sk_max_ack_backlog = backlog; 2150 sk->sk_ack_backlog = 0; 2151 sk->sk_state = SMC_LISTEN; 2152 2153 out: 2154 release_sock(sk); 2155 return rc; 2156 } 2157 2158 static int smc_accept(struct socket *sock, struct socket *new_sock, 2159 int flags, bool kern) 2160 { 2161 struct sock *sk = sock->sk, *nsk; 2162 DECLARE_WAITQUEUE(wait, current); 2163 struct smc_sock *lsmc; 2164 long timeo; 2165 int rc = 0; 2166 2167 lsmc = smc_sk(sk); 2168 sock_hold(sk); /* sock_put below */ 2169 lock_sock(sk); 2170 2171 if (lsmc->sk.sk_state != SMC_LISTEN) { 2172 rc = -EINVAL; 2173 release_sock(sk); 2174 goto out; 2175 } 2176 2177 /* Wait for an incoming connection */ 2178 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2179 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2180 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2181 set_current_state(TASK_INTERRUPTIBLE); 2182 if (!timeo) { 2183 rc = -EAGAIN; 2184 break; 2185 } 2186 release_sock(sk); 2187 timeo = schedule_timeout(timeo); 2188 /* wakeup by sk_data_ready in smc_listen_work() */ 2189 sched_annotate_sleep(); 2190 lock_sock(sk); 2191 if (signal_pending(current)) { 2192 rc = sock_intr_errno(timeo); 2193 break; 2194 } 2195 } 2196 set_current_state(TASK_RUNNING); 2197 remove_wait_queue(sk_sleep(sk), &wait); 2198 2199 if (!rc) 2200 rc = sock_error(nsk); 2201 release_sock(sk); 2202 if (rc) 2203 goto out; 2204 2205 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2206 /* wait till data arrives on the socket */ 2207 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2208 MSEC_PER_SEC); 2209 if (smc_sk(nsk)->use_fallback) { 2210 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2211 2212 lock_sock(clcsk); 2213 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2214 sk_wait_data(clcsk, &timeo, NULL); 2215 release_sock(clcsk); 2216 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2217 lock_sock(nsk); 2218 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2219 release_sock(nsk); 2220 } 2221 } 2222 2223 out: 2224 sock_put(sk); /* sock_hold above */ 2225 return rc; 2226 } 2227 2228 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2229 int peer) 2230 { 2231 struct smc_sock *smc; 2232 2233 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2234 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2235 return -ENOTCONN; 2236 2237 smc = smc_sk(sock->sk); 2238 2239 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2240 } 2241 2242 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2243 { 2244 struct sock *sk = sock->sk; 2245 struct smc_sock *smc; 2246 int rc = -EPIPE; 2247 2248 smc = smc_sk(sk); 2249 lock_sock(sk); 2250 if ((sk->sk_state != SMC_ACTIVE) && 2251 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2252 (sk->sk_state != SMC_INIT)) 2253 goto out; 2254 2255 if (msg->msg_flags & MSG_FASTOPEN) { 2256 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2257 smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2258 } else { 2259 rc = -EINVAL; 2260 goto out; 2261 } 2262 } 2263 2264 if (smc->use_fallback) { 2265 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2266 } else { 2267 rc = smc_tx_sendmsg(smc, msg, len); 2268 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2269 } 2270 out: 2271 release_sock(sk); 2272 return rc; 2273 } 2274 2275 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2276 int flags) 2277 { 2278 struct sock *sk = sock->sk; 2279 struct smc_sock *smc; 2280 int rc = -ENOTCONN; 2281 2282 smc = smc_sk(sk); 2283 lock_sock(sk); 2284 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2285 /* socket was connected before, no more data to read */ 2286 rc = 0; 2287 goto out; 2288 } 2289 if ((sk->sk_state == SMC_INIT) || 2290 (sk->sk_state == SMC_LISTEN) || 2291 (sk->sk_state == SMC_CLOSED)) 2292 goto out; 2293 2294 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2295 rc = 0; 2296 goto out; 2297 } 2298 2299 if (smc->use_fallback) { 2300 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2301 } else { 2302 msg->msg_namelen = 0; 2303 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2304 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2305 } 2306 2307 out: 2308 release_sock(sk); 2309 return rc; 2310 } 2311 2312 static __poll_t smc_accept_poll(struct sock *parent) 2313 { 2314 struct smc_sock *isk = smc_sk(parent); 2315 __poll_t mask = 0; 2316 2317 spin_lock(&isk->accept_q_lock); 2318 if (!list_empty(&isk->accept_q)) 2319 mask = EPOLLIN | EPOLLRDNORM; 2320 spin_unlock(&isk->accept_q_lock); 2321 2322 return mask; 2323 } 2324 2325 static __poll_t smc_poll(struct file *file, struct socket *sock, 2326 poll_table *wait) 2327 { 2328 struct sock *sk = sock->sk; 2329 struct smc_sock *smc; 2330 __poll_t mask = 0; 2331 2332 if (!sk) 2333 return EPOLLNVAL; 2334 2335 smc = smc_sk(sock->sk); 2336 if (smc->use_fallback) { 2337 /* delegate to CLC child sock */ 2338 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2339 sk->sk_err = smc->clcsock->sk->sk_err; 2340 } else { 2341 if (sk->sk_state != SMC_CLOSED) 2342 sock_poll_wait(file, sock, wait); 2343 if (sk->sk_err) 2344 mask |= EPOLLERR; 2345 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2346 (sk->sk_state == SMC_CLOSED)) 2347 mask |= EPOLLHUP; 2348 if (sk->sk_state == SMC_LISTEN) { 2349 /* woken up by sk_data_ready in smc_listen_work() */ 2350 mask |= smc_accept_poll(sk); 2351 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2352 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2353 wait); 2354 sk->sk_err = smc->clcsock->sk->sk_err; 2355 } else { 2356 if ((sk->sk_state != SMC_INIT && 2357 atomic_read(&smc->conn.sndbuf_space)) || 2358 sk->sk_shutdown & SEND_SHUTDOWN) { 2359 mask |= EPOLLOUT | EPOLLWRNORM; 2360 } else { 2361 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2362 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2363 } 2364 if (atomic_read(&smc->conn.bytes_to_rcv)) 2365 mask |= EPOLLIN | EPOLLRDNORM; 2366 if (sk->sk_shutdown & RCV_SHUTDOWN) 2367 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2368 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2369 mask |= EPOLLIN; 2370 if (smc->conn.urg_state == SMC_URG_VALID) 2371 mask |= EPOLLPRI; 2372 } 2373 } 2374 2375 return mask; 2376 } 2377 2378 static int smc_shutdown(struct socket *sock, int how) 2379 { 2380 struct sock *sk = sock->sk; 2381 bool do_shutdown = true; 2382 struct smc_sock *smc; 2383 int rc = -EINVAL; 2384 int old_state; 2385 int rc1 = 0; 2386 2387 smc = smc_sk(sk); 2388 2389 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2390 return rc; 2391 2392 lock_sock(sk); 2393 2394 rc = -ENOTCONN; 2395 if ((sk->sk_state != SMC_ACTIVE) && 2396 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2397 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2398 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2399 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2400 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2401 goto out; 2402 if (smc->use_fallback) { 2403 rc = kernel_sock_shutdown(smc->clcsock, how); 2404 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2405 if (sk->sk_shutdown == SHUTDOWN_MASK) 2406 sk->sk_state = SMC_CLOSED; 2407 goto out; 2408 } 2409 switch (how) { 2410 case SHUT_RDWR: /* shutdown in both directions */ 2411 old_state = sk->sk_state; 2412 rc = smc_close_active(smc); 2413 if (old_state == SMC_ACTIVE && 2414 sk->sk_state == SMC_PEERCLOSEWAIT1) 2415 do_shutdown = false; 2416 break; 2417 case SHUT_WR: 2418 rc = smc_close_shutdown_write(smc); 2419 break; 2420 case SHUT_RD: 2421 rc = 0; 2422 /* nothing more to do because peer is not involved */ 2423 break; 2424 } 2425 if (do_shutdown && smc->clcsock) 2426 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2427 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2428 sk->sk_shutdown |= how + 1; 2429 2430 out: 2431 release_sock(sk); 2432 return rc ? rc : rc1; 2433 } 2434 2435 static int smc_setsockopt(struct socket *sock, int level, int optname, 2436 sockptr_t optval, unsigned int optlen) 2437 { 2438 struct sock *sk = sock->sk; 2439 struct smc_sock *smc; 2440 int val, rc; 2441 2442 if (level == SOL_TCP && optname == TCP_ULP) 2443 return -EOPNOTSUPP; 2444 2445 smc = smc_sk(sk); 2446 2447 /* generic setsockopts reaching us here always apply to the 2448 * CLC socket 2449 */ 2450 if (unlikely(!smc->clcsock->ops->setsockopt)) 2451 rc = -EOPNOTSUPP; 2452 else 2453 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 2454 optval, optlen); 2455 if (smc->clcsock->sk->sk_err) { 2456 sk->sk_err = smc->clcsock->sk->sk_err; 2457 sk_error_report(sk); 2458 } 2459 2460 if (optlen < sizeof(int)) 2461 return -EINVAL; 2462 if (copy_from_sockptr(&val, optval, sizeof(int))) 2463 return -EFAULT; 2464 2465 lock_sock(sk); 2466 if (rc || smc->use_fallback) 2467 goto out; 2468 switch (optname) { 2469 case TCP_FASTOPEN: 2470 case TCP_FASTOPEN_CONNECT: 2471 case TCP_FASTOPEN_KEY: 2472 case TCP_FASTOPEN_NO_COOKIE: 2473 /* option not supported by SMC */ 2474 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2475 smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2476 } else { 2477 rc = -EINVAL; 2478 } 2479 break; 2480 case TCP_NODELAY: 2481 if (sk->sk_state != SMC_INIT && 2482 sk->sk_state != SMC_LISTEN && 2483 sk->sk_state != SMC_CLOSED) { 2484 if (val) { 2485 SMC_STAT_INC(smc, ndly_cnt); 2486 mod_delayed_work(smc->conn.lgr->tx_wq, 2487 &smc->conn.tx_work, 0); 2488 } 2489 } 2490 break; 2491 case TCP_CORK: 2492 if (sk->sk_state != SMC_INIT && 2493 sk->sk_state != SMC_LISTEN && 2494 sk->sk_state != SMC_CLOSED) { 2495 if (!val) { 2496 SMC_STAT_INC(smc, cork_cnt); 2497 mod_delayed_work(smc->conn.lgr->tx_wq, 2498 &smc->conn.tx_work, 0); 2499 } 2500 } 2501 break; 2502 case TCP_DEFER_ACCEPT: 2503 smc->sockopt_defer_accept = val; 2504 break; 2505 default: 2506 break; 2507 } 2508 out: 2509 release_sock(sk); 2510 2511 return rc; 2512 } 2513 2514 static int smc_getsockopt(struct socket *sock, int level, int optname, 2515 char __user *optval, int __user *optlen) 2516 { 2517 struct smc_sock *smc; 2518 2519 smc = smc_sk(sock->sk); 2520 /* socket options apply to the CLC socket */ 2521 if (unlikely(!smc->clcsock->ops->getsockopt)) 2522 return -EOPNOTSUPP; 2523 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 2524 optval, optlen); 2525 } 2526 2527 static int smc_ioctl(struct socket *sock, unsigned int cmd, 2528 unsigned long arg) 2529 { 2530 union smc_host_cursor cons, urg; 2531 struct smc_connection *conn; 2532 struct smc_sock *smc; 2533 int answ; 2534 2535 smc = smc_sk(sock->sk); 2536 conn = &smc->conn; 2537 lock_sock(&smc->sk); 2538 if (smc->use_fallback) { 2539 if (!smc->clcsock) { 2540 release_sock(&smc->sk); 2541 return -EBADF; 2542 } 2543 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 2544 release_sock(&smc->sk); 2545 return answ; 2546 } 2547 switch (cmd) { 2548 case SIOCINQ: /* same as FIONREAD */ 2549 if (smc->sk.sk_state == SMC_LISTEN) { 2550 release_sock(&smc->sk); 2551 return -EINVAL; 2552 } 2553 if (smc->sk.sk_state == SMC_INIT || 2554 smc->sk.sk_state == SMC_CLOSED) 2555 answ = 0; 2556 else 2557 answ = atomic_read(&smc->conn.bytes_to_rcv); 2558 break; 2559 case SIOCOUTQ: 2560 /* output queue size (not send + not acked) */ 2561 if (smc->sk.sk_state == SMC_LISTEN) { 2562 release_sock(&smc->sk); 2563 return -EINVAL; 2564 } 2565 if (smc->sk.sk_state == SMC_INIT || 2566 smc->sk.sk_state == SMC_CLOSED) 2567 answ = 0; 2568 else 2569 answ = smc->conn.sndbuf_desc->len - 2570 atomic_read(&smc->conn.sndbuf_space); 2571 break; 2572 case SIOCOUTQNSD: 2573 /* output queue size (not send only) */ 2574 if (smc->sk.sk_state == SMC_LISTEN) { 2575 release_sock(&smc->sk); 2576 return -EINVAL; 2577 } 2578 if (smc->sk.sk_state == SMC_INIT || 2579 smc->sk.sk_state == SMC_CLOSED) 2580 answ = 0; 2581 else 2582 answ = smc_tx_prepared_sends(&smc->conn); 2583 break; 2584 case SIOCATMARK: 2585 if (smc->sk.sk_state == SMC_LISTEN) { 2586 release_sock(&smc->sk); 2587 return -EINVAL; 2588 } 2589 if (smc->sk.sk_state == SMC_INIT || 2590 smc->sk.sk_state == SMC_CLOSED) { 2591 answ = 0; 2592 } else { 2593 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 2594 smc_curs_copy(&urg, &conn->urg_curs, conn); 2595 answ = smc_curs_diff(conn->rmb_desc->len, 2596 &cons, &urg) == 1; 2597 } 2598 break; 2599 default: 2600 release_sock(&smc->sk); 2601 return -ENOIOCTLCMD; 2602 } 2603 release_sock(&smc->sk); 2604 2605 return put_user(answ, (int __user *)arg); 2606 } 2607 2608 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 2609 int offset, size_t size, int flags) 2610 { 2611 struct sock *sk = sock->sk; 2612 struct smc_sock *smc; 2613 int rc = -EPIPE; 2614 2615 smc = smc_sk(sk); 2616 lock_sock(sk); 2617 if (sk->sk_state != SMC_ACTIVE) { 2618 release_sock(sk); 2619 goto out; 2620 } 2621 release_sock(sk); 2622 if (smc->use_fallback) { 2623 rc = kernel_sendpage(smc->clcsock, page, offset, 2624 size, flags); 2625 } else { 2626 SMC_STAT_INC(smc, sendpage_cnt); 2627 rc = sock_no_sendpage(sock, page, offset, size, flags); 2628 } 2629 2630 out: 2631 return rc; 2632 } 2633 2634 /* Map the affected portions of the rmbe into an spd, note the number of bytes 2635 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 2636 * updates till whenever a respective page has been fully processed. 2637 * Note that subsequent recv() calls have to wait till all splice() processing 2638 * completed. 2639 */ 2640 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 2641 struct pipe_inode_info *pipe, size_t len, 2642 unsigned int flags) 2643 { 2644 struct sock *sk = sock->sk; 2645 struct smc_sock *smc; 2646 int rc = -ENOTCONN; 2647 2648 smc = smc_sk(sk); 2649 lock_sock(sk); 2650 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2651 /* socket was connected before, no more data to read */ 2652 rc = 0; 2653 goto out; 2654 } 2655 if (sk->sk_state == SMC_INIT || 2656 sk->sk_state == SMC_LISTEN || 2657 sk->sk_state == SMC_CLOSED) 2658 goto out; 2659 2660 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2661 rc = 0; 2662 goto out; 2663 } 2664 2665 if (smc->use_fallback) { 2666 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 2667 pipe, len, flags); 2668 } else { 2669 if (*ppos) { 2670 rc = -ESPIPE; 2671 goto out; 2672 } 2673 if (flags & SPLICE_F_NONBLOCK) 2674 flags = MSG_DONTWAIT; 2675 else 2676 flags = 0; 2677 SMC_STAT_INC(smc, splice_cnt); 2678 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 2679 } 2680 out: 2681 release_sock(sk); 2682 2683 return rc; 2684 } 2685 2686 /* must look like tcp */ 2687 static const struct proto_ops smc_sock_ops = { 2688 .family = PF_SMC, 2689 .owner = THIS_MODULE, 2690 .release = smc_release, 2691 .bind = smc_bind, 2692 .connect = smc_connect, 2693 .socketpair = sock_no_socketpair, 2694 .accept = smc_accept, 2695 .getname = smc_getname, 2696 .poll = smc_poll, 2697 .ioctl = smc_ioctl, 2698 .listen = smc_listen, 2699 .shutdown = smc_shutdown, 2700 .setsockopt = smc_setsockopt, 2701 .getsockopt = smc_getsockopt, 2702 .sendmsg = smc_sendmsg, 2703 .recvmsg = smc_recvmsg, 2704 .mmap = sock_no_mmap, 2705 .sendpage = smc_sendpage, 2706 .splice_read = smc_splice_read, 2707 }; 2708 2709 static int __smc_create(struct net *net, struct socket *sock, int protocol, 2710 int kern, struct socket *clcsock) 2711 { 2712 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 2713 struct smc_sock *smc; 2714 struct sock *sk; 2715 int rc; 2716 2717 rc = -ESOCKTNOSUPPORT; 2718 if (sock->type != SOCK_STREAM) 2719 goto out; 2720 2721 rc = -EPROTONOSUPPORT; 2722 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 2723 goto out; 2724 2725 rc = -ENOBUFS; 2726 sock->ops = &smc_sock_ops; 2727 sk = smc_sock_alloc(net, sock, protocol); 2728 if (!sk) 2729 goto out; 2730 2731 /* create internal TCP socket for CLC handshake and fallback */ 2732 smc = smc_sk(sk); 2733 smc->use_fallback = false; /* assume rdma capability first */ 2734 smc->fallback_rsn = 0; 2735 2736 rc = 0; 2737 if (!clcsock) { 2738 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 2739 &smc->clcsock); 2740 if (rc) { 2741 sk_common_release(sk); 2742 goto out; 2743 } 2744 } else { 2745 smc->clcsock = clcsock; 2746 } 2747 2748 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2749 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2750 2751 out: 2752 return rc; 2753 } 2754 2755 static int smc_create(struct net *net, struct socket *sock, int protocol, 2756 int kern) 2757 { 2758 return __smc_create(net, sock, protocol, kern, NULL); 2759 } 2760 2761 static const struct net_proto_family smc_sock_family_ops = { 2762 .family = PF_SMC, 2763 .owner = THIS_MODULE, 2764 .create = smc_create, 2765 }; 2766 2767 static int smc_ulp_init(struct sock *sk) 2768 { 2769 struct socket *tcp = sk->sk_socket; 2770 struct net *net = sock_net(sk); 2771 struct socket *smcsock; 2772 int protocol, ret; 2773 2774 /* only TCP can be replaced */ 2775 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 2776 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 2777 return -ESOCKTNOSUPPORT; 2778 /* don't handle wq now */ 2779 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 2780 return -ENOTCONN; 2781 2782 if (sk->sk_family == AF_INET) 2783 protocol = SMCPROTO_SMC; 2784 else 2785 protocol = SMCPROTO_SMC6; 2786 2787 smcsock = sock_alloc(); 2788 if (!smcsock) 2789 return -ENFILE; 2790 2791 smcsock->type = SOCK_STREAM; 2792 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 2793 ret = __smc_create(net, smcsock, protocol, 1, tcp); 2794 if (ret) { 2795 sock_release(smcsock); /* module_put() which ops won't be NULL */ 2796 return ret; 2797 } 2798 2799 /* replace tcp socket to smc */ 2800 smcsock->file = tcp->file; 2801 smcsock->file->private_data = smcsock; 2802 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 2803 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 2804 tcp->file = NULL; 2805 2806 return ret; 2807 } 2808 2809 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 2810 const gfp_t priority) 2811 { 2812 struct inet_connection_sock *icsk = inet_csk(newsk); 2813 2814 /* don't inherit ulp ops to child when listen */ 2815 icsk->icsk_ulp_ops = NULL; 2816 } 2817 2818 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 2819 .name = "smc", 2820 .owner = THIS_MODULE, 2821 .init = smc_ulp_init, 2822 .clone = smc_ulp_clone, 2823 }; 2824 2825 unsigned int smc_net_id; 2826 2827 static __net_init int smc_net_init(struct net *net) 2828 { 2829 return smc_pnet_net_init(net); 2830 } 2831 2832 static void __net_exit smc_net_exit(struct net *net) 2833 { 2834 smc_pnet_net_exit(net); 2835 } 2836 2837 static __net_init int smc_net_stat_init(struct net *net) 2838 { 2839 return smc_stats_init(net); 2840 } 2841 2842 static void __net_exit smc_net_stat_exit(struct net *net) 2843 { 2844 smc_stats_exit(net); 2845 } 2846 2847 static struct pernet_operations smc_net_ops = { 2848 .init = smc_net_init, 2849 .exit = smc_net_exit, 2850 .id = &smc_net_id, 2851 .size = sizeof(struct smc_net), 2852 }; 2853 2854 static struct pernet_operations smc_net_stat_ops = { 2855 .init = smc_net_stat_init, 2856 .exit = smc_net_stat_exit, 2857 }; 2858 2859 static int __init smc_init(void) 2860 { 2861 int rc; 2862 2863 rc = register_pernet_subsys(&smc_net_ops); 2864 if (rc) 2865 return rc; 2866 2867 rc = register_pernet_subsys(&smc_net_stat_ops); 2868 if (rc) 2869 return rc; 2870 2871 smc_ism_init(); 2872 smc_clc_init(); 2873 2874 rc = smc_nl_init(); 2875 if (rc) 2876 goto out_pernet_subsys; 2877 2878 rc = smc_pnet_init(); 2879 if (rc) 2880 goto out_nl; 2881 2882 rc = -ENOMEM; 2883 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 2884 if (!smc_hs_wq) 2885 goto out_pnet; 2886 2887 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 2888 if (!smc_close_wq) 2889 goto out_alloc_hs_wq; 2890 2891 rc = smc_core_init(); 2892 if (rc) { 2893 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2894 goto out_alloc_wqs; 2895 } 2896 2897 rc = smc_llc_init(); 2898 if (rc) { 2899 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2900 goto out_core; 2901 } 2902 2903 rc = smc_cdc_init(); 2904 if (rc) { 2905 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2906 goto out_core; 2907 } 2908 2909 rc = proto_register(&smc_proto, 1); 2910 if (rc) { 2911 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2912 goto out_core; 2913 } 2914 2915 rc = proto_register(&smc_proto6, 1); 2916 if (rc) { 2917 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2918 goto out_proto; 2919 } 2920 2921 rc = sock_register(&smc_sock_family_ops); 2922 if (rc) { 2923 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2924 goto out_proto6; 2925 } 2926 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2927 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2928 2929 rc = smc_ib_register_client(); 2930 if (rc) { 2931 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2932 goto out_sock; 2933 } 2934 2935 rc = tcp_register_ulp(&smc_ulp_ops); 2936 if (rc) { 2937 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 2938 goto out_sock; 2939 } 2940 2941 static_branch_enable(&tcp_have_smc); 2942 return 0; 2943 2944 out_sock: 2945 sock_unregister(PF_SMC); 2946 out_proto6: 2947 proto_unregister(&smc_proto6); 2948 out_proto: 2949 proto_unregister(&smc_proto); 2950 out_core: 2951 smc_core_exit(); 2952 out_alloc_wqs: 2953 destroy_workqueue(smc_close_wq); 2954 out_alloc_hs_wq: 2955 destroy_workqueue(smc_hs_wq); 2956 out_pnet: 2957 smc_pnet_exit(); 2958 out_nl: 2959 smc_nl_exit(); 2960 out_pernet_subsys: 2961 unregister_pernet_subsys(&smc_net_ops); 2962 2963 return rc; 2964 } 2965 2966 static void __exit smc_exit(void) 2967 { 2968 static_branch_disable(&tcp_have_smc); 2969 tcp_unregister_ulp(&smc_ulp_ops); 2970 sock_unregister(PF_SMC); 2971 smc_core_exit(); 2972 smc_ib_unregister_client(); 2973 destroy_workqueue(smc_close_wq); 2974 destroy_workqueue(smc_hs_wq); 2975 proto_unregister(&smc_proto6); 2976 proto_unregister(&smc_proto); 2977 smc_pnet_exit(); 2978 smc_nl_exit(); 2979 smc_clc_exit(); 2980 unregister_pernet_subsys(&smc_net_stat_ops); 2981 unregister_pernet_subsys(&smc_net_ops); 2982 rcu_barrier(); 2983 } 2984 2985 module_init(smc_init); 2986 module_exit(smc_exit); 2987 2988 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2989 MODULE_DESCRIPTION("smc socket address family"); 2990 MODULE_LICENSE("GPL"); 2991 MODULE_ALIAS_NETPROTO(PF_SMC); 2992 MODULE_ALIAS_TCP_ULP("smc"); 2993