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 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 63 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 64 struct workqueue_struct *smc_close_wq; /* wq for close work */ 65 66 static void smc_tcp_listen_work(struct work_struct *); 67 static void smc_connect_work(struct work_struct *); 68 69 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 70 { 71 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 72 void *hdr; 73 74 if (cb_ctx->pos[0]) 75 goto out; 76 77 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 78 &smc_gen_nl_family, NLM_F_MULTI, 79 SMC_NETLINK_DUMP_HS_LIMITATION); 80 if (!hdr) 81 return -ENOMEM; 82 83 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 84 sock_net(skb->sk)->smc.limit_smc_hs)) 85 goto err; 86 87 genlmsg_end(skb, hdr); 88 cb_ctx->pos[0] = 1; 89 out: 90 return skb->len; 91 err: 92 genlmsg_cancel(skb, hdr); 93 return -EMSGSIZE; 94 } 95 96 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 97 { 98 sock_net(skb->sk)->smc.limit_smc_hs = true; 99 return 0; 100 } 101 102 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 103 { 104 sock_net(skb->sk)->smc.limit_smc_hs = false; 105 return 0; 106 } 107 108 static void smc_set_keepalive(struct sock *sk, int val) 109 { 110 struct smc_sock *smc = smc_sk(sk); 111 112 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 113 } 114 115 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 116 struct sk_buff *skb, 117 struct request_sock *req, 118 struct dst_entry *dst, 119 struct request_sock *req_unhash, 120 bool *own_req) 121 { 122 struct smc_sock *smc; 123 124 smc = smc_clcsock_user_data(sk); 125 126 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 127 sk->sk_max_ack_backlog) 128 goto drop; 129 130 if (sk_acceptq_is_full(&smc->sk)) { 131 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 132 goto drop; 133 } 134 135 /* passthrough to original syn recv sock fct */ 136 return smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 137 own_req); 138 139 drop: 140 dst_release(dst); 141 tcp_listendrop(sk); 142 return NULL; 143 } 144 145 static bool smc_hs_congested(const struct sock *sk) 146 { 147 const struct smc_sock *smc; 148 149 smc = smc_clcsock_user_data(sk); 150 151 if (!smc) 152 return true; 153 154 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 155 return true; 156 157 return false; 158 } 159 160 static struct smc_hashinfo smc_v4_hashinfo = { 161 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 162 }; 163 164 static struct smc_hashinfo smc_v6_hashinfo = { 165 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 166 }; 167 168 int smc_hash_sk(struct sock *sk) 169 { 170 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 171 struct hlist_head *head; 172 173 head = &h->ht; 174 175 write_lock_bh(&h->lock); 176 sk_add_node(sk, head); 177 write_unlock_bh(&h->lock); 178 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 179 180 return 0; 181 } 182 EXPORT_SYMBOL_GPL(smc_hash_sk); 183 184 void smc_unhash_sk(struct sock *sk) 185 { 186 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 187 188 write_lock_bh(&h->lock); 189 if (sk_del_node_init(sk)) 190 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 191 write_unlock_bh(&h->lock); 192 } 193 EXPORT_SYMBOL_GPL(smc_unhash_sk); 194 195 struct proto smc_proto = { 196 .name = "SMC", 197 .owner = THIS_MODULE, 198 .keepalive = smc_set_keepalive, 199 .hash = smc_hash_sk, 200 .unhash = smc_unhash_sk, 201 .obj_size = sizeof(struct smc_sock), 202 .h.smc_hash = &smc_v4_hashinfo, 203 .slab_flags = SLAB_TYPESAFE_BY_RCU, 204 }; 205 EXPORT_SYMBOL_GPL(smc_proto); 206 207 struct proto smc_proto6 = { 208 .name = "SMC6", 209 .owner = THIS_MODULE, 210 .keepalive = smc_set_keepalive, 211 .hash = smc_hash_sk, 212 .unhash = smc_unhash_sk, 213 .obj_size = sizeof(struct smc_sock), 214 .h.smc_hash = &smc_v6_hashinfo, 215 .slab_flags = SLAB_TYPESAFE_BY_RCU, 216 }; 217 EXPORT_SYMBOL_GPL(smc_proto6); 218 219 static void smc_restore_fallback_changes(struct smc_sock *smc) 220 { 221 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 222 smc->clcsock->file->private_data = smc->sk.sk_socket; 223 smc->clcsock->file = NULL; 224 } 225 } 226 227 static int __smc_release(struct smc_sock *smc) 228 { 229 struct sock *sk = &smc->sk; 230 int rc = 0; 231 232 if (!smc->use_fallback) { 233 rc = smc_close_active(smc); 234 sock_set_flag(sk, SOCK_DEAD); 235 sk->sk_shutdown |= SHUTDOWN_MASK; 236 } else { 237 if (sk->sk_state != SMC_CLOSED) { 238 if (sk->sk_state != SMC_LISTEN && 239 sk->sk_state != SMC_INIT) 240 sock_put(sk); /* passive closing */ 241 if (sk->sk_state == SMC_LISTEN) { 242 /* wake up clcsock accept */ 243 rc = kernel_sock_shutdown(smc->clcsock, 244 SHUT_RDWR); 245 } 246 sk->sk_state = SMC_CLOSED; 247 sk->sk_state_change(sk); 248 } 249 smc_restore_fallback_changes(smc); 250 } 251 252 sk->sk_prot->unhash(sk); 253 254 if (sk->sk_state == SMC_CLOSED) { 255 if (smc->clcsock) { 256 release_sock(sk); 257 smc_clcsock_release(smc); 258 lock_sock(sk); 259 } 260 if (!smc->use_fallback) 261 smc_conn_free(&smc->conn); 262 } 263 264 return rc; 265 } 266 267 static int smc_release(struct socket *sock) 268 { 269 struct sock *sk = sock->sk; 270 struct smc_sock *smc; 271 int rc = 0; 272 273 if (!sk) 274 goto out; 275 276 sock_hold(sk); /* sock_put below */ 277 smc = smc_sk(sk); 278 279 /* cleanup for a dangling non-blocking connect */ 280 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 281 tcp_abort(smc->clcsock->sk, ECONNABORTED); 282 283 if (cancel_work_sync(&smc->connect_work)) 284 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 285 286 if (sk->sk_state == SMC_LISTEN) 287 /* smc_close_non_accepted() is called and acquires 288 * sock lock for child sockets again 289 */ 290 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 291 else 292 lock_sock(sk); 293 294 rc = __smc_release(smc); 295 296 /* detach socket */ 297 sock_orphan(sk); 298 sock->sk = NULL; 299 release_sock(sk); 300 301 sock_put(sk); /* sock_hold above */ 302 sock_put(sk); /* final sock_put */ 303 out: 304 return rc; 305 } 306 307 static void smc_destruct(struct sock *sk) 308 { 309 if (sk->sk_state != SMC_CLOSED) 310 return; 311 if (!sock_flag(sk, SOCK_DEAD)) 312 return; 313 314 sk_refcnt_debug_dec(sk); 315 } 316 317 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 318 int protocol) 319 { 320 struct smc_sock *smc; 321 struct proto *prot; 322 struct sock *sk; 323 324 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 325 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 326 if (!sk) 327 return NULL; 328 329 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 330 sk->sk_state = SMC_INIT; 331 sk->sk_destruct = smc_destruct; 332 sk->sk_protocol = protocol; 333 smc = smc_sk(sk); 334 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 335 INIT_WORK(&smc->connect_work, smc_connect_work); 336 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 337 INIT_LIST_HEAD(&smc->accept_q); 338 spin_lock_init(&smc->accept_q_lock); 339 spin_lock_init(&smc->conn.send_lock); 340 sk->sk_prot->hash(sk); 341 sk_refcnt_debug_inc(sk); 342 mutex_init(&smc->clcsock_release_lock); 343 344 return sk; 345 } 346 347 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 348 int addr_len) 349 { 350 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 351 struct sock *sk = sock->sk; 352 struct smc_sock *smc; 353 int rc; 354 355 smc = smc_sk(sk); 356 357 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 358 rc = -EINVAL; 359 if (addr_len < sizeof(struct sockaddr_in)) 360 goto out; 361 362 rc = -EAFNOSUPPORT; 363 if (addr->sin_family != AF_INET && 364 addr->sin_family != AF_INET6 && 365 addr->sin_family != AF_UNSPEC) 366 goto out; 367 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 368 if (addr->sin_family == AF_UNSPEC && 369 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 370 goto out; 371 372 lock_sock(sk); 373 374 /* Check if socket is already active */ 375 rc = -EINVAL; 376 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 377 goto out_rel; 378 379 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 380 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 381 382 out_rel: 383 release_sock(sk); 384 out: 385 return rc; 386 } 387 388 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 389 unsigned long mask) 390 { 391 /* options we don't get control via setsockopt for */ 392 nsk->sk_type = osk->sk_type; 393 nsk->sk_sndbuf = osk->sk_sndbuf; 394 nsk->sk_rcvbuf = osk->sk_rcvbuf; 395 nsk->sk_sndtimeo = osk->sk_sndtimeo; 396 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 397 nsk->sk_mark = osk->sk_mark; 398 nsk->sk_priority = osk->sk_priority; 399 nsk->sk_rcvlowat = osk->sk_rcvlowat; 400 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 401 nsk->sk_err = osk->sk_err; 402 403 nsk->sk_flags &= ~mask; 404 nsk->sk_flags |= osk->sk_flags & mask; 405 } 406 407 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 408 (1UL << SOCK_KEEPOPEN) | \ 409 (1UL << SOCK_LINGER) | \ 410 (1UL << SOCK_BROADCAST) | \ 411 (1UL << SOCK_TIMESTAMP) | \ 412 (1UL << SOCK_DBG) | \ 413 (1UL << SOCK_RCVTSTAMP) | \ 414 (1UL << SOCK_RCVTSTAMPNS) | \ 415 (1UL << SOCK_LOCALROUTE) | \ 416 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 417 (1UL << SOCK_RXQ_OVFL) | \ 418 (1UL << SOCK_WIFI_STATUS) | \ 419 (1UL << SOCK_NOFCS) | \ 420 (1UL << SOCK_FILTER_LOCKED) | \ 421 (1UL << SOCK_TSTAMP_NEW)) 422 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 423 * clc socket (since smc is not called for these options from net/core) 424 */ 425 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 426 { 427 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 428 } 429 430 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 431 (1UL << SOCK_KEEPOPEN) | \ 432 (1UL << SOCK_LINGER) | \ 433 (1UL << SOCK_DBG)) 434 /* copy only settings and flags relevant for smc from clc to smc socket */ 435 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 436 { 437 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 438 } 439 440 /* register the new rmb on all links */ 441 static int smcr_lgr_reg_rmbs(struct smc_link *link, 442 struct smc_buf_desc *rmb_desc) 443 { 444 struct smc_link_group *lgr = link->lgr; 445 int i, rc = 0; 446 447 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 448 if (rc) 449 return rc; 450 /* protect against parallel smc_llc_cli_rkey_exchange() and 451 * parallel smcr_link_reg_rmb() 452 */ 453 mutex_lock(&lgr->llc_conf_mutex); 454 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 455 if (!smc_link_active(&lgr->lnk[i])) 456 continue; 457 rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc); 458 if (rc) 459 goto out; 460 } 461 462 /* exchange confirm_rkey msg with peer */ 463 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 464 if (rc) { 465 rc = -EFAULT; 466 goto out; 467 } 468 rmb_desc->is_conf_rkey = true; 469 out: 470 mutex_unlock(&lgr->llc_conf_mutex); 471 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 472 return rc; 473 } 474 475 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 476 { 477 struct smc_link *link = smc->conn.lnk; 478 struct smc_llc_qentry *qentry; 479 int rc; 480 481 /* receive CONFIRM LINK request from server over RoCE fabric */ 482 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 483 SMC_LLC_CONFIRM_LINK); 484 if (!qentry) { 485 struct smc_clc_msg_decline dclc; 486 487 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 488 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 489 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 490 } 491 smc_llc_save_peer_uid(qentry); 492 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 493 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 494 if (rc) 495 return SMC_CLC_DECL_RMBE_EC; 496 497 rc = smc_ib_modify_qp_rts(link); 498 if (rc) 499 return SMC_CLC_DECL_ERR_RDYLNK; 500 501 smc_wr_remember_qp_attr(link); 502 503 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 504 return SMC_CLC_DECL_ERR_REGRMB; 505 506 /* confirm_rkey is implicit on 1st contact */ 507 smc->conn.rmb_desc->is_conf_rkey = true; 508 509 /* send CONFIRM LINK response over RoCE fabric */ 510 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 511 if (rc < 0) 512 return SMC_CLC_DECL_TIMEOUT_CL; 513 514 smc_llc_link_active(link); 515 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 516 517 /* optional 2nd link, receive ADD LINK request from server */ 518 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 519 SMC_LLC_ADD_LINK); 520 if (!qentry) { 521 struct smc_clc_msg_decline dclc; 522 523 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 524 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 525 if (rc == -EAGAIN) 526 rc = 0; /* no DECLINE received, go with one link */ 527 return rc; 528 } 529 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 530 smc_llc_cli_add_link(link, qentry); 531 return 0; 532 } 533 534 static bool smc_isascii(char *hostname) 535 { 536 int i; 537 538 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 539 if (!isascii(hostname[i])) 540 return false; 541 return true; 542 } 543 544 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 545 struct smc_clc_msg_accept_confirm *clc) 546 { 547 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 548 (struct smc_clc_msg_accept_confirm_v2 *)clc; 549 struct smc_clc_first_contact_ext *fce; 550 int clc_v2_len; 551 552 if (clc->hdr.version == SMC_V1 || 553 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 554 return; 555 556 if (smc->conn.lgr->is_smcd) { 557 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 558 SMC_MAX_EID_LEN); 559 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 560 d1); 561 } else { 562 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 563 SMC_MAX_EID_LEN); 564 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 565 r1); 566 } 567 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 568 smc->conn.lgr->peer_os = fce->os_type; 569 smc->conn.lgr->peer_smc_release = fce->release; 570 if (smc_isascii(fce->hostname)) 571 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 572 SMC_MAX_HOSTNAME_LEN); 573 } 574 575 static void smcr_conn_save_peer_info(struct smc_sock *smc, 576 struct smc_clc_msg_accept_confirm *clc) 577 { 578 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 579 580 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 581 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 582 smc->conn.peer_rmbe_size = bufsize; 583 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 584 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 585 } 586 587 static void smcd_conn_save_peer_info(struct smc_sock *smc, 588 struct smc_clc_msg_accept_confirm *clc) 589 { 590 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 591 592 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 593 smc->conn.peer_token = clc->d0.token; 594 /* msg header takes up space in the buffer */ 595 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 596 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 597 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 598 } 599 600 static void smc_conn_save_peer_info(struct smc_sock *smc, 601 struct smc_clc_msg_accept_confirm *clc) 602 { 603 if (smc->conn.lgr->is_smcd) 604 smcd_conn_save_peer_info(smc, clc); 605 else 606 smcr_conn_save_peer_info(smc, clc); 607 smc_conn_save_peer_info_fce(smc, clc); 608 } 609 610 static void smc_link_save_peer_info(struct smc_link *link, 611 struct smc_clc_msg_accept_confirm *clc, 612 struct smc_init_info *ini) 613 { 614 link->peer_qpn = ntoh24(clc->r0.qpn); 615 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 616 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 617 link->peer_psn = ntoh24(clc->r0.psn); 618 link->peer_mtu = clc->r0.qp_mtu; 619 } 620 621 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 622 struct smc_stats_fback *fback_arr) 623 { 624 int cnt; 625 626 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 627 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 628 fback_arr[cnt].count++; 629 break; 630 } 631 if (!fback_arr[cnt].fback_code) { 632 fback_arr[cnt].fback_code = smc->fallback_rsn; 633 fback_arr[cnt].count++; 634 break; 635 } 636 } 637 } 638 639 static void smc_stat_fallback(struct smc_sock *smc) 640 { 641 struct net *net = sock_net(&smc->sk); 642 643 mutex_lock(&net->smc.mutex_fback_rsn); 644 if (smc->listen_smc) { 645 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 646 net->smc.fback_rsn->srv_fback_cnt++; 647 } else { 648 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 649 net->smc.fback_rsn->clnt_fback_cnt++; 650 } 651 mutex_unlock(&net->smc.mutex_fback_rsn); 652 } 653 654 /* must be called under rcu read lock */ 655 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 656 { 657 struct socket_wq *wq; 658 __poll_t flags; 659 660 wq = rcu_dereference(smc->sk.sk_wq); 661 if (!skwq_has_sleeper(wq)) 662 return; 663 664 /* wake up smc sk->sk_wq */ 665 if (!key) { 666 /* sk_state_change */ 667 wake_up_interruptible_all(&wq->wait); 668 } else { 669 flags = key_to_poll(key); 670 if (flags & (EPOLLIN | EPOLLOUT)) 671 /* sk_data_ready or sk_write_space */ 672 wake_up_interruptible_sync_poll(&wq->wait, flags); 673 else if (flags & EPOLLERR) 674 /* sk_error_report */ 675 wake_up_interruptible_poll(&wq->wait, flags); 676 } 677 } 678 679 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 680 unsigned int mode, int sync, void *key) 681 { 682 struct smc_mark_woken *mark = 683 container_of(wait, struct smc_mark_woken, wait_entry); 684 685 mark->woken = true; 686 mark->key = key; 687 return 0; 688 } 689 690 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 691 void (*clcsock_callback)(struct sock *sk)) 692 { 693 struct smc_mark_woken mark = { .woken = false }; 694 struct socket_wq *wq; 695 696 init_waitqueue_func_entry(&mark.wait_entry, 697 smc_fback_mark_woken); 698 rcu_read_lock(); 699 wq = rcu_dereference(clcsk->sk_wq); 700 if (!wq) 701 goto out; 702 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 703 clcsock_callback(clcsk); 704 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 705 706 if (mark.woken) 707 smc_fback_wakeup_waitqueue(smc, mark.key); 708 out: 709 rcu_read_unlock(); 710 } 711 712 static void smc_fback_state_change(struct sock *clcsk) 713 { 714 struct smc_sock *smc = 715 smc_clcsock_user_data(clcsk); 716 717 if (!smc) 718 return; 719 smc_fback_forward_wakeup(smc, clcsk, smc->clcsk_state_change); 720 } 721 722 static void smc_fback_data_ready(struct sock *clcsk) 723 { 724 struct smc_sock *smc = 725 smc_clcsock_user_data(clcsk); 726 727 if (!smc) 728 return; 729 smc_fback_forward_wakeup(smc, clcsk, smc->clcsk_data_ready); 730 } 731 732 static void smc_fback_write_space(struct sock *clcsk) 733 { 734 struct smc_sock *smc = 735 smc_clcsock_user_data(clcsk); 736 737 if (!smc) 738 return; 739 smc_fback_forward_wakeup(smc, clcsk, smc->clcsk_write_space); 740 } 741 742 static void smc_fback_error_report(struct sock *clcsk) 743 { 744 struct smc_sock *smc = 745 smc_clcsock_user_data(clcsk); 746 747 if (!smc) 748 return; 749 smc_fback_forward_wakeup(smc, clcsk, smc->clcsk_error_report); 750 } 751 752 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 753 { 754 struct sock *clcsk; 755 int rc = 0; 756 757 mutex_lock(&smc->clcsock_release_lock); 758 if (!smc->clcsock) { 759 rc = -EBADF; 760 goto out; 761 } 762 clcsk = smc->clcsock->sk; 763 764 if (smc->use_fallback) 765 goto out; 766 smc->use_fallback = true; 767 smc->fallback_rsn = reason_code; 768 smc_stat_fallback(smc); 769 trace_smc_switch_to_fallback(smc, reason_code); 770 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 771 smc->clcsock->file = smc->sk.sk_socket->file; 772 smc->clcsock->file->private_data = smc->clcsock; 773 smc->clcsock->wq.fasync_list = 774 smc->sk.sk_socket->wq.fasync_list; 775 776 /* There might be some wait entries remaining 777 * in smc sk->sk_wq and they should be woken up 778 * as clcsock's wait queue is woken up. 779 */ 780 smc->clcsk_state_change = clcsk->sk_state_change; 781 smc->clcsk_data_ready = clcsk->sk_data_ready; 782 smc->clcsk_write_space = clcsk->sk_write_space; 783 smc->clcsk_error_report = clcsk->sk_error_report; 784 785 clcsk->sk_state_change = smc_fback_state_change; 786 clcsk->sk_data_ready = smc_fback_data_ready; 787 clcsk->sk_write_space = smc_fback_write_space; 788 clcsk->sk_error_report = smc_fback_error_report; 789 790 smc->clcsock->sk->sk_user_data = 791 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 792 } 793 out: 794 mutex_unlock(&smc->clcsock_release_lock); 795 return rc; 796 } 797 798 /* fall back during connect */ 799 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 800 { 801 struct net *net = sock_net(&smc->sk); 802 int rc = 0; 803 804 rc = smc_switch_to_fallback(smc, reason_code); 805 if (rc) { /* fallback fails */ 806 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 807 if (smc->sk.sk_state == SMC_INIT) 808 sock_put(&smc->sk); /* passive closing */ 809 return rc; 810 } 811 smc_copy_sock_settings_to_clc(smc); 812 smc->connect_nonblock = 0; 813 if (smc->sk.sk_state == SMC_INIT) 814 smc->sk.sk_state = SMC_ACTIVE; 815 return 0; 816 } 817 818 /* decline and fall back during connect */ 819 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 820 u8 version) 821 { 822 struct net *net = sock_net(&smc->sk); 823 int rc; 824 825 if (reason_code < 0) { /* error, fallback is not possible */ 826 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 827 if (smc->sk.sk_state == SMC_INIT) 828 sock_put(&smc->sk); /* passive closing */ 829 return reason_code; 830 } 831 if (reason_code != SMC_CLC_DECL_PEERDECL) { 832 rc = smc_clc_send_decline(smc, reason_code, version); 833 if (rc < 0) { 834 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 835 if (smc->sk.sk_state == SMC_INIT) 836 sock_put(&smc->sk); /* passive closing */ 837 return rc; 838 } 839 } 840 return smc_connect_fallback(smc, reason_code); 841 } 842 843 static void smc_conn_abort(struct smc_sock *smc, int local_first) 844 { 845 struct smc_connection *conn = &smc->conn; 846 struct smc_link_group *lgr = conn->lgr; 847 bool lgr_valid = false; 848 849 if (smc_conn_lgr_valid(conn)) 850 lgr_valid = true; 851 852 smc_conn_free(conn); 853 if (local_first && lgr_valid) 854 smc_lgr_cleanup_early(lgr); 855 } 856 857 /* check if there is a rdma device available for this connection. */ 858 /* called for connect and listen */ 859 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 860 { 861 /* PNET table look up: search active ib_device and port 862 * within same PNETID that also contains the ethernet device 863 * used for the internal TCP socket 864 */ 865 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 866 if (!ini->check_smcrv2 && !ini->ib_dev) 867 return SMC_CLC_DECL_NOSMCRDEV; 868 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 869 return SMC_CLC_DECL_NOSMCRDEV; 870 return 0; 871 } 872 873 /* check if there is an ISM device available for this connection. */ 874 /* called for connect and listen */ 875 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 876 { 877 /* Find ISM device with same PNETID as connecting interface */ 878 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 879 if (!ini->ism_dev[0]) 880 return SMC_CLC_DECL_NOSMCDDEV; 881 else 882 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 883 return 0; 884 } 885 886 /* is chid unique for the ism devices that are already determined? */ 887 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 888 int cnt) 889 { 890 int i = (!ini->ism_dev[0]) ? 1 : 0; 891 892 for (; i < cnt; i++) 893 if (ini->ism_chid[i] == chid) 894 return false; 895 return true; 896 } 897 898 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 899 * PNETID matching net_device) 900 */ 901 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 902 struct smc_init_info *ini) 903 { 904 int rc = SMC_CLC_DECL_NOSMCDDEV; 905 struct smcd_dev *smcd; 906 int i = 1; 907 u16 chid; 908 909 if (smcd_indicated(ini->smc_type_v1)) 910 rc = 0; /* already initialized for V1 */ 911 mutex_lock(&smcd_dev_list.mutex); 912 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 913 if (smcd->going_away || smcd == ini->ism_dev[0]) 914 continue; 915 chid = smc_ism_get_chid(smcd); 916 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 917 continue; 918 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 919 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 920 ini->ism_dev[i] = smcd; 921 ini->ism_chid[i] = chid; 922 ini->is_smcd = true; 923 rc = 0; 924 i++; 925 if (i > SMC_MAX_ISM_DEVS) 926 break; 927 } 928 } 929 mutex_unlock(&smcd_dev_list.mutex); 930 ini->ism_offered_cnt = i - 1; 931 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 932 ini->smcd_version = 0; 933 934 return rc; 935 } 936 937 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 938 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 939 struct smc_init_info *ini) 940 { 941 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 942 return SMC_CLC_DECL_ISMVLANERR; 943 return 0; 944 } 945 946 static int smc_find_proposal_devices(struct smc_sock *smc, 947 struct smc_init_info *ini) 948 { 949 int rc = 0; 950 951 /* check if there is an ism device available */ 952 if (!(ini->smcd_version & SMC_V1) || 953 smc_find_ism_device(smc, ini) || 954 smc_connect_ism_vlan_setup(smc, ini)) 955 ini->smcd_version &= ~SMC_V1; 956 /* else ISM V1 is supported for this connection */ 957 958 /* check if there is an rdma device available */ 959 if (!(ini->smcr_version & SMC_V1) || 960 smc_find_rdma_device(smc, ini)) 961 ini->smcr_version &= ~SMC_V1; 962 /* else RDMA is supported for this connection */ 963 964 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 965 ini->smcr_version & SMC_V1); 966 967 /* check if there is an ism v2 device available */ 968 if (!(ini->smcd_version & SMC_V2) || 969 !smc_ism_is_v2_capable() || 970 smc_find_ism_v2_device_clnt(smc, ini)) 971 ini->smcd_version &= ~SMC_V2; 972 973 /* check if there is an rdma v2 device available */ 974 ini->check_smcrv2 = true; 975 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 976 if (!(ini->smcr_version & SMC_V2) || 977 smc->clcsock->sk->sk_family != AF_INET || 978 !smc_clc_ueid_count() || 979 smc_find_rdma_device(smc, ini)) 980 ini->smcr_version &= ~SMC_V2; 981 ini->check_smcrv2 = false; 982 983 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 984 ini->smcr_version & SMC_V2); 985 986 /* if neither ISM nor RDMA are supported, fallback */ 987 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 988 rc = SMC_CLC_DECL_NOSMCDEV; 989 990 return rc; 991 } 992 993 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 994 * used, the VLAN ID will be registered again during the connection setup. 995 */ 996 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 997 struct smc_init_info *ini) 998 { 999 if (!smcd_indicated(ini->smc_type_v1)) 1000 return 0; 1001 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1002 return SMC_CLC_DECL_CNFERR; 1003 return 0; 1004 } 1005 1006 #define SMC_CLC_MAX_ACCEPT_LEN \ 1007 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 1008 sizeof(struct smc_clc_first_contact_ext) + \ 1009 sizeof(struct smc_clc_msg_trail)) 1010 1011 /* CLC handshake during connect */ 1012 static int smc_connect_clc(struct smc_sock *smc, 1013 struct smc_clc_msg_accept_confirm_v2 *aclc2, 1014 struct smc_init_info *ini) 1015 { 1016 int rc = 0; 1017 1018 /* do inband token exchange */ 1019 rc = smc_clc_send_proposal(smc, ini); 1020 if (rc) 1021 return rc; 1022 /* receive SMC Accept CLC message */ 1023 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 1024 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1025 } 1026 1027 void smc_fill_gid_list(struct smc_link_group *lgr, 1028 struct smc_gidlist *gidlist, 1029 struct smc_ib_device *known_dev, u8 *known_gid) 1030 { 1031 struct smc_init_info *alt_ini = NULL; 1032 1033 memset(gidlist, 0, sizeof(*gidlist)); 1034 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1035 1036 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1037 if (!alt_ini) 1038 goto out; 1039 1040 alt_ini->vlan_id = lgr->vlan_id; 1041 alt_ini->check_smcrv2 = true; 1042 alt_ini->smcrv2.saddr = lgr->saddr; 1043 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1044 1045 if (!alt_ini->smcrv2.ib_dev_v2) 1046 goto out; 1047 1048 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1049 SMC_GID_SIZE); 1050 1051 out: 1052 kfree(alt_ini); 1053 } 1054 1055 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1056 struct smc_clc_msg_accept_confirm *aclc, 1057 struct smc_init_info *ini) 1058 { 1059 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1060 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1061 struct smc_clc_first_contact_ext *fce = 1062 (struct smc_clc_first_contact_ext *) 1063 (((u8 *)clc_v2) + sizeof(*clc_v2)); 1064 1065 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1066 return 0; 1067 1068 if (fce->v2_direct) { 1069 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1070 ini->smcrv2.uses_gateway = false; 1071 } else { 1072 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr, 1073 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1074 ini->smcrv2.nexthop_mac, 1075 &ini->smcrv2.uses_gateway)) 1076 return SMC_CLC_DECL_NOROUTE; 1077 if (!ini->smcrv2.uses_gateway) { 1078 /* mismatch: peer claims indirect, but its direct */ 1079 return SMC_CLC_DECL_NOINDIRECT; 1080 } 1081 } 1082 return 0; 1083 } 1084 1085 /* setup for RDMA connection of client */ 1086 static int smc_connect_rdma(struct smc_sock *smc, 1087 struct smc_clc_msg_accept_confirm *aclc, 1088 struct smc_init_info *ini) 1089 { 1090 int i, reason_code = 0; 1091 struct smc_link *link; 1092 u8 *eid = NULL; 1093 1094 ini->is_smcd = false; 1095 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1096 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1097 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1098 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1099 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1100 1101 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1102 if (reason_code) 1103 return reason_code; 1104 1105 mutex_lock(&smc_client_lgr_pending); 1106 reason_code = smc_conn_create(smc, ini); 1107 if (reason_code) { 1108 mutex_unlock(&smc_client_lgr_pending); 1109 return reason_code; 1110 } 1111 1112 smc_conn_save_peer_info(smc, aclc); 1113 1114 if (ini->first_contact_local) { 1115 link = smc->conn.lnk; 1116 } else { 1117 /* set link that was assigned by server */ 1118 link = NULL; 1119 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1120 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1121 1122 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1123 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1124 SMC_GID_SIZE) && 1125 (aclc->hdr.version > SMC_V1 || 1126 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1127 sizeof(l->peer_mac)))) { 1128 link = l; 1129 break; 1130 } 1131 } 1132 if (!link) { 1133 reason_code = SMC_CLC_DECL_NOSRVLINK; 1134 goto connect_abort; 1135 } 1136 smc_switch_link_and_count(&smc->conn, link); 1137 } 1138 1139 /* create send buffer and rmb */ 1140 if (smc_buf_create(smc, false)) { 1141 reason_code = SMC_CLC_DECL_MEM; 1142 goto connect_abort; 1143 } 1144 1145 if (ini->first_contact_local) 1146 smc_link_save_peer_info(link, aclc, ini); 1147 1148 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1149 reason_code = SMC_CLC_DECL_ERR_RTOK; 1150 goto connect_abort; 1151 } 1152 1153 smc_close_init(smc); 1154 smc_rx_init(smc); 1155 1156 if (ini->first_contact_local) { 1157 if (smc_ib_ready_link(link)) { 1158 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1159 goto connect_abort; 1160 } 1161 } else { 1162 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1163 reason_code = SMC_CLC_DECL_ERR_REGRMB; 1164 goto connect_abort; 1165 } 1166 } 1167 smc_rmb_sync_sg_for_device(&smc->conn); 1168 1169 if (aclc->hdr.version > SMC_V1) { 1170 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1171 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1172 1173 eid = clc_v2->r1.eid; 1174 if (ini->first_contact_local) 1175 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1176 link->smcibdev, link->gid); 1177 } 1178 1179 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1180 aclc->hdr.version, eid, ini); 1181 if (reason_code) 1182 goto connect_abort; 1183 1184 smc_tx_init(smc); 1185 1186 if (ini->first_contact_local) { 1187 /* QP confirmation over RoCE fabric */ 1188 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1189 reason_code = smcr_clnt_conf_first_link(smc); 1190 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1191 if (reason_code) 1192 goto connect_abort; 1193 } 1194 mutex_unlock(&smc_client_lgr_pending); 1195 1196 smc_copy_sock_settings_to_clc(smc); 1197 smc->connect_nonblock = 0; 1198 if (smc->sk.sk_state == SMC_INIT) 1199 smc->sk.sk_state = SMC_ACTIVE; 1200 1201 return 0; 1202 connect_abort: 1203 smc_conn_abort(smc, ini->first_contact_local); 1204 mutex_unlock(&smc_client_lgr_pending); 1205 smc->connect_nonblock = 0; 1206 1207 return reason_code; 1208 } 1209 1210 /* The server has chosen one of the proposed ISM devices for the communication. 1211 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1212 */ 1213 static int 1214 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 1215 struct smc_init_info *ini) 1216 { 1217 int i; 1218 1219 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1220 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1221 ini->ism_selected = i; 1222 return 0; 1223 } 1224 } 1225 1226 return -EPROTO; 1227 } 1228 1229 /* setup for ISM connection of client */ 1230 static int smc_connect_ism(struct smc_sock *smc, 1231 struct smc_clc_msg_accept_confirm *aclc, 1232 struct smc_init_info *ini) 1233 { 1234 u8 *eid = NULL; 1235 int rc = 0; 1236 1237 ini->is_smcd = true; 1238 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1239 1240 if (aclc->hdr.version == SMC_V2) { 1241 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1242 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1243 1244 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1245 if (rc) 1246 return rc; 1247 } 1248 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 1249 1250 /* there is only one lgr role for SMC-D; use server lock */ 1251 mutex_lock(&smc_server_lgr_pending); 1252 rc = smc_conn_create(smc, ini); 1253 if (rc) { 1254 mutex_unlock(&smc_server_lgr_pending); 1255 return rc; 1256 } 1257 1258 /* Create send and receive buffers */ 1259 rc = smc_buf_create(smc, true); 1260 if (rc) { 1261 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1262 goto connect_abort; 1263 } 1264 1265 smc_conn_save_peer_info(smc, aclc); 1266 smc_close_init(smc); 1267 smc_rx_init(smc); 1268 smc_tx_init(smc); 1269 1270 if (aclc->hdr.version > SMC_V1) { 1271 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1272 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1273 1274 eid = clc_v2->d1.eid; 1275 } 1276 1277 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1278 aclc->hdr.version, eid, NULL); 1279 if (rc) 1280 goto connect_abort; 1281 mutex_unlock(&smc_server_lgr_pending); 1282 1283 smc_copy_sock_settings_to_clc(smc); 1284 smc->connect_nonblock = 0; 1285 if (smc->sk.sk_state == SMC_INIT) 1286 smc->sk.sk_state = SMC_ACTIVE; 1287 1288 return 0; 1289 connect_abort: 1290 smc_conn_abort(smc, ini->first_contact_local); 1291 mutex_unlock(&smc_server_lgr_pending); 1292 smc->connect_nonblock = 0; 1293 1294 return rc; 1295 } 1296 1297 /* check if received accept type and version matches a proposed one */ 1298 static int smc_connect_check_aclc(struct smc_init_info *ini, 1299 struct smc_clc_msg_accept_confirm *aclc) 1300 { 1301 if (aclc->hdr.typev1 != SMC_TYPE_R && 1302 aclc->hdr.typev1 != SMC_TYPE_D) 1303 return SMC_CLC_DECL_MODEUNSUPP; 1304 1305 if (aclc->hdr.version >= SMC_V2) { 1306 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1307 !smcr_indicated(ini->smc_type_v2)) || 1308 (aclc->hdr.typev1 == SMC_TYPE_D && 1309 !smcd_indicated(ini->smc_type_v2))) 1310 return SMC_CLC_DECL_MODEUNSUPP; 1311 } else { 1312 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1313 !smcr_indicated(ini->smc_type_v1)) || 1314 (aclc->hdr.typev1 == SMC_TYPE_D && 1315 !smcd_indicated(ini->smc_type_v1))) 1316 return SMC_CLC_DECL_MODEUNSUPP; 1317 } 1318 1319 return 0; 1320 } 1321 1322 /* perform steps before actually connecting */ 1323 static int __smc_connect(struct smc_sock *smc) 1324 { 1325 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1326 struct smc_clc_msg_accept_confirm_v2 *aclc2; 1327 struct smc_clc_msg_accept_confirm *aclc; 1328 struct smc_init_info *ini = NULL; 1329 u8 *buf = NULL; 1330 int rc = 0; 1331 1332 if (smc->use_fallback) 1333 return smc_connect_fallback(smc, smc->fallback_rsn); 1334 1335 /* if peer has not signalled SMC-capability, fall back */ 1336 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1337 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1338 1339 /* IPSec connections opt out of SMC optimizations */ 1340 if (using_ipsec(smc)) 1341 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1342 version); 1343 1344 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1345 if (!ini) 1346 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1347 version); 1348 1349 ini->smcd_version = SMC_V1 | SMC_V2; 1350 ini->smcr_version = SMC_V1 | SMC_V2; 1351 ini->smc_type_v1 = SMC_TYPE_B; 1352 ini->smc_type_v2 = SMC_TYPE_B; 1353 1354 /* get vlan id from IP device */ 1355 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1356 ini->smcd_version &= ~SMC_V1; 1357 ini->smcr_version = 0; 1358 ini->smc_type_v1 = SMC_TYPE_N; 1359 if (!ini->smcd_version) { 1360 rc = SMC_CLC_DECL_GETVLANERR; 1361 goto fallback; 1362 } 1363 } 1364 1365 rc = smc_find_proposal_devices(smc, ini); 1366 if (rc) 1367 goto fallback; 1368 1369 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1370 if (!buf) { 1371 rc = SMC_CLC_DECL_MEM; 1372 goto fallback; 1373 } 1374 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1375 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1376 1377 /* perform CLC handshake */ 1378 rc = smc_connect_clc(smc, aclc2, ini); 1379 if (rc) { 1380 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1381 if (rc == -EAGAIN) { 1382 rc = -ETIMEDOUT; 1383 smc->sk.sk_err = ETIMEDOUT; 1384 } 1385 goto vlan_cleanup; 1386 } 1387 1388 /* check if smc modes and versions of CLC proposal and accept match */ 1389 rc = smc_connect_check_aclc(ini, aclc); 1390 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1391 if (rc) 1392 goto vlan_cleanup; 1393 1394 /* depending on previous steps, connect using rdma or ism */ 1395 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1396 ini->smcr_version = version; 1397 rc = smc_connect_rdma(smc, aclc, ini); 1398 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1399 ini->smcd_version = version; 1400 rc = smc_connect_ism(smc, aclc, ini); 1401 } 1402 if (rc) 1403 goto vlan_cleanup; 1404 1405 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1406 smc_connect_ism_vlan_cleanup(smc, ini); 1407 kfree(buf); 1408 kfree(ini); 1409 return 0; 1410 1411 vlan_cleanup: 1412 smc_connect_ism_vlan_cleanup(smc, ini); 1413 kfree(buf); 1414 fallback: 1415 kfree(ini); 1416 return smc_connect_decline_fallback(smc, rc, version); 1417 } 1418 1419 static void smc_connect_work(struct work_struct *work) 1420 { 1421 struct smc_sock *smc = container_of(work, struct smc_sock, 1422 connect_work); 1423 long timeo = smc->sk.sk_sndtimeo; 1424 int rc = 0; 1425 1426 if (!timeo) 1427 timeo = MAX_SCHEDULE_TIMEOUT; 1428 lock_sock(smc->clcsock->sk); 1429 if (smc->clcsock->sk->sk_err) { 1430 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1431 } else if ((1 << smc->clcsock->sk->sk_state) & 1432 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1433 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1434 if ((rc == -EPIPE) && 1435 ((1 << smc->clcsock->sk->sk_state) & 1436 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1437 rc = 0; 1438 } 1439 release_sock(smc->clcsock->sk); 1440 lock_sock(&smc->sk); 1441 if (rc != 0 || smc->sk.sk_err) { 1442 smc->sk.sk_state = SMC_CLOSED; 1443 if (rc == -EPIPE || rc == -EAGAIN) 1444 smc->sk.sk_err = EPIPE; 1445 else if (signal_pending(current)) 1446 smc->sk.sk_err = -sock_intr_errno(timeo); 1447 sock_put(&smc->sk); /* passive closing */ 1448 goto out; 1449 } 1450 1451 rc = __smc_connect(smc); 1452 if (rc < 0) 1453 smc->sk.sk_err = -rc; 1454 1455 out: 1456 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1457 if (smc->sk.sk_err) { 1458 smc->sk.sk_state_change(&smc->sk); 1459 } else { /* allow polling before and after fallback decision */ 1460 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1461 smc->sk.sk_write_space(&smc->sk); 1462 } 1463 } 1464 release_sock(&smc->sk); 1465 } 1466 1467 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1468 int alen, int flags) 1469 { 1470 struct sock *sk = sock->sk; 1471 struct smc_sock *smc; 1472 int rc = -EINVAL; 1473 1474 smc = smc_sk(sk); 1475 1476 /* separate smc parameter checking to be safe */ 1477 if (alen < sizeof(addr->sa_family)) 1478 goto out_err; 1479 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1480 goto out_err; 1481 1482 lock_sock(sk); 1483 switch (sk->sk_state) { 1484 default: 1485 goto out; 1486 case SMC_ACTIVE: 1487 rc = -EISCONN; 1488 goto out; 1489 case SMC_INIT: 1490 break; 1491 } 1492 1493 smc_copy_sock_settings_to_clc(smc); 1494 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1495 if (smc->connect_nonblock) { 1496 rc = -EALREADY; 1497 goto out; 1498 } 1499 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1500 if (rc && rc != -EINPROGRESS) 1501 goto out; 1502 1503 sock_hold(&smc->sk); /* sock put in passive closing */ 1504 if (smc->use_fallback) 1505 goto out; 1506 if (flags & O_NONBLOCK) { 1507 if (queue_work(smc_hs_wq, &smc->connect_work)) 1508 smc->connect_nonblock = 1; 1509 rc = -EINPROGRESS; 1510 } else { 1511 rc = __smc_connect(smc); 1512 if (rc < 0) 1513 goto out; 1514 else 1515 rc = 0; /* success cases including fallback */ 1516 } 1517 1518 out: 1519 release_sock(sk); 1520 out_err: 1521 return rc; 1522 } 1523 1524 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1525 { 1526 struct socket *new_clcsock = NULL; 1527 struct sock *lsk = &lsmc->sk; 1528 struct sock *new_sk; 1529 int rc = -EINVAL; 1530 1531 release_sock(lsk); 1532 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1533 if (!new_sk) { 1534 rc = -ENOMEM; 1535 lsk->sk_err = ENOMEM; 1536 *new_smc = NULL; 1537 lock_sock(lsk); 1538 goto out; 1539 } 1540 *new_smc = smc_sk(new_sk); 1541 1542 mutex_lock(&lsmc->clcsock_release_lock); 1543 if (lsmc->clcsock) 1544 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1545 mutex_unlock(&lsmc->clcsock_release_lock); 1546 lock_sock(lsk); 1547 if (rc < 0 && rc != -EAGAIN) 1548 lsk->sk_err = -rc; 1549 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1550 new_sk->sk_prot->unhash(new_sk); 1551 if (new_clcsock) 1552 sock_release(new_clcsock); 1553 new_sk->sk_state = SMC_CLOSED; 1554 sock_set_flag(new_sk, SOCK_DEAD); 1555 sock_put(new_sk); /* final */ 1556 *new_smc = NULL; 1557 goto out; 1558 } 1559 1560 /* new clcsock has inherited the smc listen-specific sk_data_ready 1561 * function; switch it back to the original sk_data_ready function 1562 */ 1563 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1564 (*new_smc)->clcsock = new_clcsock; 1565 out: 1566 return rc; 1567 } 1568 1569 /* add a just created sock to the accept queue of the listen sock as 1570 * candidate for a following socket accept call from user space 1571 */ 1572 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1573 { 1574 struct smc_sock *par = smc_sk(parent); 1575 1576 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1577 spin_lock(&par->accept_q_lock); 1578 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1579 spin_unlock(&par->accept_q_lock); 1580 sk_acceptq_added(parent); 1581 } 1582 1583 /* remove a socket from the accept queue of its parental listening socket */ 1584 static void smc_accept_unlink(struct sock *sk) 1585 { 1586 struct smc_sock *par = smc_sk(sk)->listen_smc; 1587 1588 spin_lock(&par->accept_q_lock); 1589 list_del_init(&smc_sk(sk)->accept_q); 1590 spin_unlock(&par->accept_q_lock); 1591 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1592 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1593 } 1594 1595 /* remove a sock from the accept queue to bind it to a new socket created 1596 * for a socket accept call from user space 1597 */ 1598 struct sock *smc_accept_dequeue(struct sock *parent, 1599 struct socket *new_sock) 1600 { 1601 struct smc_sock *isk, *n; 1602 struct sock *new_sk; 1603 1604 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1605 new_sk = (struct sock *)isk; 1606 1607 smc_accept_unlink(new_sk); 1608 if (new_sk->sk_state == SMC_CLOSED) { 1609 new_sk->sk_prot->unhash(new_sk); 1610 if (isk->clcsock) { 1611 sock_release(isk->clcsock); 1612 isk->clcsock = NULL; 1613 } 1614 sock_put(new_sk); /* final */ 1615 continue; 1616 } 1617 if (new_sock) { 1618 sock_graft(new_sk, new_sock); 1619 if (isk->use_fallback) { 1620 smc_sk(new_sk)->clcsock->file = new_sock->file; 1621 isk->clcsock->file->private_data = isk->clcsock; 1622 } 1623 } 1624 return new_sk; 1625 } 1626 return NULL; 1627 } 1628 1629 /* clean up for a created but never accepted sock */ 1630 void smc_close_non_accepted(struct sock *sk) 1631 { 1632 struct smc_sock *smc = smc_sk(sk); 1633 1634 sock_hold(sk); /* sock_put below */ 1635 lock_sock(sk); 1636 if (!sk->sk_lingertime) 1637 /* wait for peer closing */ 1638 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1639 __smc_release(smc); 1640 release_sock(sk); 1641 sock_put(sk); /* sock_hold above */ 1642 sock_put(sk); /* final sock_put */ 1643 } 1644 1645 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1646 { 1647 struct smc_link *link = smc->conn.lnk; 1648 struct smc_llc_qentry *qentry; 1649 int rc; 1650 1651 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc)) 1652 return SMC_CLC_DECL_ERR_REGRMB; 1653 1654 /* send CONFIRM LINK request to client over the RoCE fabric */ 1655 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1656 if (rc < 0) 1657 return SMC_CLC_DECL_TIMEOUT_CL; 1658 1659 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1660 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1661 SMC_LLC_CONFIRM_LINK); 1662 if (!qentry) { 1663 struct smc_clc_msg_decline dclc; 1664 1665 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1666 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1667 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1668 } 1669 smc_llc_save_peer_uid(qentry); 1670 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1671 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1672 if (rc) 1673 return SMC_CLC_DECL_RMBE_EC; 1674 1675 /* confirm_rkey is implicit on 1st contact */ 1676 smc->conn.rmb_desc->is_conf_rkey = true; 1677 1678 smc_llc_link_active(link); 1679 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1680 1681 /* initial contact - try to establish second link */ 1682 smc_llc_srv_add_link(link, NULL); 1683 return 0; 1684 } 1685 1686 /* listen worker: finish */ 1687 static void smc_listen_out(struct smc_sock *new_smc) 1688 { 1689 struct smc_sock *lsmc = new_smc->listen_smc; 1690 struct sock *newsmcsk = &new_smc->sk; 1691 1692 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1693 atomic_dec(&lsmc->queued_smc_hs); 1694 1695 if (lsmc->sk.sk_state == SMC_LISTEN) { 1696 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1697 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1698 release_sock(&lsmc->sk); 1699 } else { /* no longer listening */ 1700 smc_close_non_accepted(newsmcsk); 1701 } 1702 1703 /* Wake up accept */ 1704 lsmc->sk.sk_data_ready(&lsmc->sk); 1705 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1706 } 1707 1708 /* listen worker: finish in state connected */ 1709 static void smc_listen_out_connected(struct smc_sock *new_smc) 1710 { 1711 struct sock *newsmcsk = &new_smc->sk; 1712 1713 sk_refcnt_debug_inc(newsmcsk); 1714 if (newsmcsk->sk_state == SMC_INIT) 1715 newsmcsk->sk_state = SMC_ACTIVE; 1716 1717 smc_listen_out(new_smc); 1718 } 1719 1720 /* listen worker: finish in error state */ 1721 static void smc_listen_out_err(struct smc_sock *new_smc) 1722 { 1723 struct sock *newsmcsk = &new_smc->sk; 1724 struct net *net = sock_net(newsmcsk); 1725 1726 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1727 if (newsmcsk->sk_state == SMC_INIT) 1728 sock_put(&new_smc->sk); /* passive closing */ 1729 newsmcsk->sk_state = SMC_CLOSED; 1730 1731 smc_listen_out(new_smc); 1732 } 1733 1734 /* listen worker: decline and fall back if possible */ 1735 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1736 int local_first, u8 version) 1737 { 1738 /* RDMA setup failed, switch back to TCP */ 1739 smc_conn_abort(new_smc, local_first); 1740 if (reason_code < 0 || 1741 smc_switch_to_fallback(new_smc, reason_code)) { 1742 /* error, no fallback possible */ 1743 smc_listen_out_err(new_smc); 1744 return; 1745 } 1746 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1747 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1748 smc_listen_out_err(new_smc); 1749 return; 1750 } 1751 } 1752 smc_listen_out_connected(new_smc); 1753 } 1754 1755 /* listen worker: version checking */ 1756 static int smc_listen_v2_check(struct smc_sock *new_smc, 1757 struct smc_clc_msg_proposal *pclc, 1758 struct smc_init_info *ini) 1759 { 1760 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1761 struct smc_clc_v2_extension *pclc_v2_ext; 1762 int rc = SMC_CLC_DECL_PEERNOSMC; 1763 1764 ini->smc_type_v1 = pclc->hdr.typev1; 1765 ini->smc_type_v2 = pclc->hdr.typev2; 1766 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1767 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1768 if (pclc->hdr.version > SMC_V1) { 1769 if (smcd_indicated(ini->smc_type_v2)) 1770 ini->smcd_version |= SMC_V2; 1771 if (smcr_indicated(ini->smc_type_v2)) 1772 ini->smcr_version |= SMC_V2; 1773 } 1774 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1775 rc = SMC_CLC_DECL_PEERNOSMC; 1776 goto out; 1777 } 1778 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1779 if (!pclc_v2_ext) { 1780 ini->smcd_version &= ~SMC_V2; 1781 ini->smcr_version &= ~SMC_V2; 1782 rc = SMC_CLC_DECL_NOV2EXT; 1783 goto out; 1784 } 1785 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1786 if (ini->smcd_version & SMC_V2) { 1787 if (!smc_ism_is_v2_capable()) { 1788 ini->smcd_version &= ~SMC_V2; 1789 rc = SMC_CLC_DECL_NOISM2SUPP; 1790 } else if (!pclc_smcd_v2_ext) { 1791 ini->smcd_version &= ~SMC_V2; 1792 rc = SMC_CLC_DECL_NOV2DEXT; 1793 } else if (!pclc_v2_ext->hdr.eid_cnt && 1794 !pclc_v2_ext->hdr.flag.seid) { 1795 ini->smcd_version &= ~SMC_V2; 1796 rc = SMC_CLC_DECL_NOUEID; 1797 } 1798 } 1799 if (ini->smcr_version & SMC_V2) { 1800 if (!pclc_v2_ext->hdr.eid_cnt) { 1801 ini->smcr_version &= ~SMC_V2; 1802 rc = SMC_CLC_DECL_NOUEID; 1803 } 1804 } 1805 1806 out: 1807 if (!ini->smcd_version && !ini->smcr_version) 1808 return rc; 1809 1810 return 0; 1811 } 1812 1813 /* listen worker: check prefixes */ 1814 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1815 struct smc_clc_msg_proposal *pclc) 1816 { 1817 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1818 struct socket *newclcsock = new_smc->clcsock; 1819 1820 if (pclc->hdr.typev1 == SMC_TYPE_N) 1821 return 0; 1822 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1823 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1824 return SMC_CLC_DECL_DIFFPREFIX; 1825 1826 return 0; 1827 } 1828 1829 /* listen worker: initialize connection and buffers */ 1830 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1831 struct smc_init_info *ini) 1832 { 1833 int rc; 1834 1835 /* allocate connection / link group */ 1836 rc = smc_conn_create(new_smc, ini); 1837 if (rc) 1838 return rc; 1839 1840 /* create send buffer and rmb */ 1841 if (smc_buf_create(new_smc, false)) 1842 return SMC_CLC_DECL_MEM; 1843 1844 return 0; 1845 } 1846 1847 /* listen worker: initialize connection and buffers for SMC-D */ 1848 static int smc_listen_ism_init(struct smc_sock *new_smc, 1849 struct smc_init_info *ini) 1850 { 1851 int rc; 1852 1853 rc = smc_conn_create(new_smc, ini); 1854 if (rc) 1855 return rc; 1856 1857 /* Create send and receive buffers */ 1858 rc = smc_buf_create(new_smc, true); 1859 if (rc) { 1860 smc_conn_abort(new_smc, ini->first_contact_local); 1861 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 1862 SMC_CLC_DECL_MEM; 1863 } 1864 1865 return 0; 1866 } 1867 1868 static bool smc_is_already_selected(struct smcd_dev *smcd, 1869 struct smc_init_info *ini, 1870 int matches) 1871 { 1872 int i; 1873 1874 for (i = 0; i < matches; i++) 1875 if (smcd == ini->ism_dev[i]) 1876 return true; 1877 1878 return false; 1879 } 1880 1881 /* check for ISM devices matching proposed ISM devices */ 1882 static void smc_check_ism_v2_match(struct smc_init_info *ini, 1883 u16 proposed_chid, u64 proposed_gid, 1884 unsigned int *matches) 1885 { 1886 struct smcd_dev *smcd; 1887 1888 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1889 if (smcd->going_away) 1890 continue; 1891 if (smc_is_already_selected(smcd, ini, *matches)) 1892 continue; 1893 if (smc_ism_get_chid(smcd) == proposed_chid && 1894 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 1895 ini->ism_peer_gid[*matches] = proposed_gid; 1896 ini->ism_dev[*matches] = smcd; 1897 (*matches)++; 1898 break; 1899 } 1900 } 1901 } 1902 1903 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 1904 { 1905 if (!ini->rc) 1906 ini->rc = rc; 1907 } 1908 1909 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 1910 struct smc_clc_msg_proposal *pclc, 1911 struct smc_init_info *ini) 1912 { 1913 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 1914 struct smc_clc_v2_extension *smc_v2_ext; 1915 struct smc_clc_msg_smcd *pclc_smcd; 1916 unsigned int matches = 0; 1917 u8 smcd_version; 1918 u8 *eid = NULL; 1919 int i, rc; 1920 1921 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 1922 goto not_found; 1923 1924 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1925 smc_v2_ext = smc_get_clc_v2_ext(pclc); 1926 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 1927 1928 mutex_lock(&smcd_dev_list.mutex); 1929 if (pclc_smcd->ism.chid) 1930 /* check for ISM device matching proposed native ISM device */ 1931 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 1932 ntohll(pclc_smcd->ism.gid), &matches); 1933 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 1934 /* check for ISM devices matching proposed non-native ISM 1935 * devices 1936 */ 1937 smc_check_ism_v2_match(ini, 1938 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 1939 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 1940 &matches); 1941 } 1942 mutex_unlock(&smcd_dev_list.mutex); 1943 1944 if (!ini->ism_dev[0]) { 1945 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 1946 goto not_found; 1947 } 1948 1949 smc_ism_get_system_eid(&eid); 1950 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 1951 smcd_v2_ext->system_eid, eid)) 1952 goto not_found; 1953 1954 /* separate - outside the smcd_dev_list.lock */ 1955 smcd_version = ini->smcd_version; 1956 for (i = 0; i < matches; i++) { 1957 ini->smcd_version = SMC_V2; 1958 ini->is_smcd = true; 1959 ini->ism_selected = i; 1960 rc = smc_listen_ism_init(new_smc, ini); 1961 if (rc) { 1962 smc_find_ism_store_rc(rc, ini); 1963 /* try next active ISM device */ 1964 continue; 1965 } 1966 return; /* matching and usable V2 ISM device found */ 1967 } 1968 /* no V2 ISM device could be initialized */ 1969 ini->smcd_version = smcd_version; /* restore original value */ 1970 ini->negotiated_eid[0] = 0; 1971 1972 not_found: 1973 ini->smcd_version &= ~SMC_V2; 1974 ini->ism_dev[0] = NULL; 1975 ini->is_smcd = false; 1976 } 1977 1978 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 1979 struct smc_clc_msg_proposal *pclc, 1980 struct smc_init_info *ini) 1981 { 1982 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 1983 int rc = 0; 1984 1985 /* check if ISM V1 is available */ 1986 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 1987 goto not_found; 1988 ini->is_smcd = true; /* prepare ISM check */ 1989 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 1990 rc = smc_find_ism_device(new_smc, ini); 1991 if (rc) 1992 goto not_found; 1993 ini->ism_selected = 0; 1994 rc = smc_listen_ism_init(new_smc, ini); 1995 if (!rc) 1996 return; /* V1 ISM device found */ 1997 1998 not_found: 1999 smc_find_ism_store_rc(rc, ini); 2000 ini->smcd_version &= ~SMC_V1; 2001 ini->ism_dev[0] = NULL; 2002 ini->is_smcd = false; 2003 } 2004 2005 /* listen worker: register buffers */ 2006 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2007 { 2008 struct smc_connection *conn = &new_smc->conn; 2009 2010 if (!local_first) { 2011 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2012 return SMC_CLC_DECL_ERR_REGRMB; 2013 } 2014 smc_rmb_sync_sg_for_device(&new_smc->conn); 2015 2016 return 0; 2017 } 2018 2019 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2020 struct smc_clc_msg_proposal *pclc, 2021 struct smc_init_info *ini) 2022 { 2023 struct smc_clc_v2_extension *smc_v2_ext; 2024 u8 smcr_version; 2025 int rc; 2026 2027 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2028 goto not_found; 2029 2030 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2031 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2032 goto not_found; 2033 2034 /* prepare RDMA check */ 2035 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2036 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2037 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2038 ini->check_smcrv2 = true; 2039 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2040 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2041 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2042 rc = smc_find_rdma_device(new_smc, ini); 2043 if (rc) { 2044 smc_find_ism_store_rc(rc, ini); 2045 goto not_found; 2046 } 2047 if (!ini->smcrv2.uses_gateway) 2048 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2049 2050 smcr_version = ini->smcr_version; 2051 ini->smcr_version = SMC_V2; 2052 rc = smc_listen_rdma_init(new_smc, ini); 2053 if (!rc) 2054 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2055 if (!rc) 2056 return; 2057 ini->smcr_version = smcr_version; 2058 smc_find_ism_store_rc(rc, ini); 2059 2060 not_found: 2061 ini->smcr_version &= ~SMC_V2; 2062 ini->check_smcrv2 = false; 2063 } 2064 2065 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2066 struct smc_clc_msg_proposal *pclc, 2067 struct smc_init_info *ini) 2068 { 2069 int rc; 2070 2071 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2072 return SMC_CLC_DECL_NOSMCDEV; 2073 2074 /* prepare RDMA check */ 2075 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2076 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2077 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2078 rc = smc_find_rdma_device(new_smc, ini); 2079 if (rc) { 2080 /* no RDMA device found */ 2081 return SMC_CLC_DECL_NOSMCDEV; 2082 } 2083 rc = smc_listen_rdma_init(new_smc, ini); 2084 if (rc) 2085 return rc; 2086 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2087 } 2088 2089 /* determine the local device matching to proposal */ 2090 static int smc_listen_find_device(struct smc_sock *new_smc, 2091 struct smc_clc_msg_proposal *pclc, 2092 struct smc_init_info *ini) 2093 { 2094 int prfx_rc; 2095 2096 /* check for ISM device matching V2 proposed device */ 2097 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2098 if (ini->ism_dev[0]) 2099 return 0; 2100 2101 /* check for matching IP prefix and subnet length (V1) */ 2102 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2103 if (prfx_rc) 2104 smc_find_ism_store_rc(prfx_rc, ini); 2105 2106 /* get vlan id from IP device */ 2107 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2108 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2109 2110 /* check for ISM device matching V1 proposed device */ 2111 if (!prfx_rc) 2112 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2113 if (ini->ism_dev[0]) 2114 return 0; 2115 2116 if (!smcr_indicated(pclc->hdr.typev1) && 2117 !smcr_indicated(pclc->hdr.typev2)) 2118 /* skip RDMA and decline */ 2119 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2120 2121 /* check if RDMA V2 is available */ 2122 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2123 if (ini->smcrv2.ib_dev_v2) 2124 return 0; 2125 2126 /* check if RDMA V1 is available */ 2127 if (!prfx_rc) { 2128 int rc; 2129 2130 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2131 smc_find_ism_store_rc(rc, ini); 2132 return (!rc) ? 0 : ini->rc; 2133 } 2134 return SMC_CLC_DECL_NOSMCDEV; 2135 } 2136 2137 /* listen worker: finish RDMA setup */ 2138 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2139 struct smc_clc_msg_accept_confirm *cclc, 2140 bool local_first, 2141 struct smc_init_info *ini) 2142 { 2143 struct smc_link *link = new_smc->conn.lnk; 2144 int reason_code = 0; 2145 2146 if (local_first) 2147 smc_link_save_peer_info(link, cclc, ini); 2148 2149 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2150 return SMC_CLC_DECL_ERR_RTOK; 2151 2152 if (local_first) { 2153 if (smc_ib_ready_link(link)) 2154 return SMC_CLC_DECL_ERR_RDYLNK; 2155 /* QP confirmation over RoCE fabric */ 2156 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2157 reason_code = smcr_serv_conf_first_link(new_smc); 2158 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2159 } 2160 return reason_code; 2161 } 2162 2163 /* setup for connection of server */ 2164 static void smc_listen_work(struct work_struct *work) 2165 { 2166 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2167 smc_listen_work); 2168 struct socket *newclcsock = new_smc->clcsock; 2169 struct smc_clc_msg_accept_confirm *cclc; 2170 struct smc_clc_msg_proposal_area *buf; 2171 struct smc_clc_msg_proposal *pclc; 2172 struct smc_init_info *ini = NULL; 2173 u8 proposal_version = SMC_V1; 2174 u8 accept_version; 2175 int rc = 0; 2176 2177 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2178 return smc_listen_out_err(new_smc); 2179 2180 if (new_smc->use_fallback) { 2181 smc_listen_out_connected(new_smc); 2182 return; 2183 } 2184 2185 /* check if peer is smc capable */ 2186 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2187 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2188 if (rc) 2189 smc_listen_out_err(new_smc); 2190 else 2191 smc_listen_out_connected(new_smc); 2192 return; 2193 } 2194 2195 /* do inband token exchange - 2196 * wait for and receive SMC Proposal CLC message 2197 */ 2198 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2199 if (!buf) { 2200 rc = SMC_CLC_DECL_MEM; 2201 goto out_decl; 2202 } 2203 pclc = (struct smc_clc_msg_proposal *)buf; 2204 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2205 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2206 if (rc) 2207 goto out_decl; 2208 2209 if (pclc->hdr.version > SMC_V1) 2210 proposal_version = SMC_V2; 2211 2212 /* IPSec connections opt out of SMC optimizations */ 2213 if (using_ipsec(new_smc)) { 2214 rc = SMC_CLC_DECL_IPSEC; 2215 goto out_decl; 2216 } 2217 2218 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2219 if (!ini) { 2220 rc = SMC_CLC_DECL_MEM; 2221 goto out_decl; 2222 } 2223 2224 /* initial version checking */ 2225 rc = smc_listen_v2_check(new_smc, pclc, ini); 2226 if (rc) 2227 goto out_decl; 2228 2229 mutex_lock(&smc_server_lgr_pending); 2230 smc_close_init(new_smc); 2231 smc_rx_init(new_smc); 2232 smc_tx_init(new_smc); 2233 2234 /* determine ISM or RoCE device used for connection */ 2235 rc = smc_listen_find_device(new_smc, pclc, ini); 2236 if (rc) 2237 goto out_unlock; 2238 2239 /* send SMC Accept CLC message */ 2240 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2241 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2242 accept_version, ini->negotiated_eid); 2243 if (rc) 2244 goto out_unlock; 2245 2246 /* SMC-D does not need this lock any more */ 2247 if (ini->is_smcd) 2248 mutex_unlock(&smc_server_lgr_pending); 2249 2250 /* receive SMC Confirm CLC message */ 2251 memset(buf, 0, sizeof(*buf)); 2252 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2253 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2254 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2255 if (rc) { 2256 if (!ini->is_smcd) 2257 goto out_unlock; 2258 goto out_decl; 2259 } 2260 2261 /* finish worker */ 2262 if (!ini->is_smcd) { 2263 rc = smc_listen_rdma_finish(new_smc, cclc, 2264 ini->first_contact_local, ini); 2265 if (rc) 2266 goto out_unlock; 2267 mutex_unlock(&smc_server_lgr_pending); 2268 } 2269 smc_conn_save_peer_info(new_smc, cclc); 2270 smc_listen_out_connected(new_smc); 2271 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2272 goto out_free; 2273 2274 out_unlock: 2275 mutex_unlock(&smc_server_lgr_pending); 2276 out_decl: 2277 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2278 proposal_version); 2279 out_free: 2280 kfree(ini); 2281 kfree(buf); 2282 } 2283 2284 static void smc_tcp_listen_work(struct work_struct *work) 2285 { 2286 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2287 tcp_listen_work); 2288 struct sock *lsk = &lsmc->sk; 2289 struct smc_sock *new_smc; 2290 int rc = 0; 2291 2292 lock_sock(lsk); 2293 while (lsk->sk_state == SMC_LISTEN) { 2294 rc = smc_clcsock_accept(lsmc, &new_smc); 2295 if (rc) /* clcsock accept queue empty or error */ 2296 goto out; 2297 if (!new_smc) 2298 continue; 2299 2300 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2301 atomic_inc(&lsmc->queued_smc_hs); 2302 2303 new_smc->listen_smc = lsmc; 2304 new_smc->use_fallback = lsmc->use_fallback; 2305 new_smc->fallback_rsn = lsmc->fallback_rsn; 2306 sock_hold(lsk); /* sock_put in smc_listen_work */ 2307 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2308 smc_copy_sock_settings_to_smc(new_smc); 2309 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 2310 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 2311 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2312 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2313 sock_put(&new_smc->sk); 2314 } 2315 2316 out: 2317 release_sock(lsk); 2318 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2319 } 2320 2321 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2322 { 2323 struct smc_sock *lsmc = 2324 smc_clcsock_user_data(listen_clcsock); 2325 2326 if (!lsmc) 2327 return; 2328 lsmc->clcsk_data_ready(listen_clcsock); 2329 if (lsmc->sk.sk_state == SMC_LISTEN) { 2330 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2331 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2332 sock_put(&lsmc->sk); 2333 } 2334 } 2335 2336 static int smc_listen(struct socket *sock, int backlog) 2337 { 2338 struct sock *sk = sock->sk; 2339 struct smc_sock *smc; 2340 int rc; 2341 2342 smc = smc_sk(sk); 2343 lock_sock(sk); 2344 2345 rc = -EINVAL; 2346 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2347 smc->connect_nonblock) 2348 goto out; 2349 2350 rc = 0; 2351 if (sk->sk_state == SMC_LISTEN) { 2352 sk->sk_max_ack_backlog = backlog; 2353 goto out; 2354 } 2355 /* some socket options are handled in core, so we could not apply 2356 * them to the clc socket -- copy smc socket options to clc socket 2357 */ 2358 smc_copy_sock_settings_to_clc(smc); 2359 if (!smc->use_fallback) 2360 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2361 2362 /* save original sk_data_ready function and establish 2363 * smc-specific sk_data_ready function 2364 */ 2365 smc->clcsk_data_ready = smc->clcsock->sk->sk_data_ready; 2366 smc->clcsock->sk->sk_data_ready = smc_clcsock_data_ready; 2367 smc->clcsock->sk->sk_user_data = 2368 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2369 2370 /* save original ops */ 2371 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2372 2373 smc->af_ops = *smc->ori_af_ops; 2374 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2375 2376 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2377 2378 if (smc->limit_smc_hs) 2379 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2380 2381 rc = kernel_listen(smc->clcsock, backlog); 2382 if (rc) { 2383 smc->clcsock->sk->sk_data_ready = smc->clcsk_data_ready; 2384 goto out; 2385 } 2386 sk->sk_max_ack_backlog = backlog; 2387 sk->sk_ack_backlog = 0; 2388 sk->sk_state = SMC_LISTEN; 2389 2390 out: 2391 release_sock(sk); 2392 return rc; 2393 } 2394 2395 static int smc_accept(struct socket *sock, struct socket *new_sock, 2396 int flags, bool kern) 2397 { 2398 struct sock *sk = sock->sk, *nsk; 2399 DECLARE_WAITQUEUE(wait, current); 2400 struct smc_sock *lsmc; 2401 long timeo; 2402 int rc = 0; 2403 2404 lsmc = smc_sk(sk); 2405 sock_hold(sk); /* sock_put below */ 2406 lock_sock(sk); 2407 2408 if (lsmc->sk.sk_state != SMC_LISTEN) { 2409 rc = -EINVAL; 2410 release_sock(sk); 2411 goto out; 2412 } 2413 2414 /* Wait for an incoming connection */ 2415 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2416 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2417 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2418 set_current_state(TASK_INTERRUPTIBLE); 2419 if (!timeo) { 2420 rc = -EAGAIN; 2421 break; 2422 } 2423 release_sock(sk); 2424 timeo = schedule_timeout(timeo); 2425 /* wakeup by sk_data_ready in smc_listen_work() */ 2426 sched_annotate_sleep(); 2427 lock_sock(sk); 2428 if (signal_pending(current)) { 2429 rc = sock_intr_errno(timeo); 2430 break; 2431 } 2432 } 2433 set_current_state(TASK_RUNNING); 2434 remove_wait_queue(sk_sleep(sk), &wait); 2435 2436 if (!rc) 2437 rc = sock_error(nsk); 2438 release_sock(sk); 2439 if (rc) 2440 goto out; 2441 2442 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2443 /* wait till data arrives on the socket */ 2444 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2445 MSEC_PER_SEC); 2446 if (smc_sk(nsk)->use_fallback) { 2447 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2448 2449 lock_sock(clcsk); 2450 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2451 sk_wait_data(clcsk, &timeo, NULL); 2452 release_sock(clcsk); 2453 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2454 lock_sock(nsk); 2455 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2456 release_sock(nsk); 2457 } 2458 } 2459 2460 out: 2461 sock_put(sk); /* sock_hold above */ 2462 return rc; 2463 } 2464 2465 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2466 int peer) 2467 { 2468 struct smc_sock *smc; 2469 2470 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2471 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2472 return -ENOTCONN; 2473 2474 smc = smc_sk(sock->sk); 2475 2476 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2477 } 2478 2479 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2480 { 2481 struct sock *sk = sock->sk; 2482 struct smc_sock *smc; 2483 int rc = -EPIPE; 2484 2485 smc = smc_sk(sk); 2486 lock_sock(sk); 2487 if ((sk->sk_state != SMC_ACTIVE) && 2488 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2489 (sk->sk_state != SMC_INIT)) 2490 goto out; 2491 2492 if (msg->msg_flags & MSG_FASTOPEN) { 2493 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2494 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2495 if (rc) 2496 goto out; 2497 } else { 2498 rc = -EINVAL; 2499 goto out; 2500 } 2501 } 2502 2503 if (smc->use_fallback) { 2504 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2505 } else { 2506 rc = smc_tx_sendmsg(smc, msg, len); 2507 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2508 } 2509 out: 2510 release_sock(sk); 2511 return rc; 2512 } 2513 2514 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2515 int flags) 2516 { 2517 struct sock *sk = sock->sk; 2518 struct smc_sock *smc; 2519 int rc = -ENOTCONN; 2520 2521 smc = smc_sk(sk); 2522 lock_sock(sk); 2523 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2524 /* socket was connected before, no more data to read */ 2525 rc = 0; 2526 goto out; 2527 } 2528 if ((sk->sk_state == SMC_INIT) || 2529 (sk->sk_state == SMC_LISTEN) || 2530 (sk->sk_state == SMC_CLOSED)) 2531 goto out; 2532 2533 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2534 rc = 0; 2535 goto out; 2536 } 2537 2538 if (smc->use_fallback) { 2539 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2540 } else { 2541 msg->msg_namelen = 0; 2542 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2543 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2544 } 2545 2546 out: 2547 release_sock(sk); 2548 return rc; 2549 } 2550 2551 static __poll_t smc_accept_poll(struct sock *parent) 2552 { 2553 struct smc_sock *isk = smc_sk(parent); 2554 __poll_t mask = 0; 2555 2556 spin_lock(&isk->accept_q_lock); 2557 if (!list_empty(&isk->accept_q)) 2558 mask = EPOLLIN | EPOLLRDNORM; 2559 spin_unlock(&isk->accept_q_lock); 2560 2561 return mask; 2562 } 2563 2564 static __poll_t smc_poll(struct file *file, struct socket *sock, 2565 poll_table *wait) 2566 { 2567 struct sock *sk = sock->sk; 2568 struct smc_sock *smc; 2569 __poll_t mask = 0; 2570 2571 if (!sk) 2572 return EPOLLNVAL; 2573 2574 smc = smc_sk(sock->sk); 2575 if (smc->use_fallback) { 2576 /* delegate to CLC child sock */ 2577 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2578 sk->sk_err = smc->clcsock->sk->sk_err; 2579 } else { 2580 if (sk->sk_state != SMC_CLOSED) 2581 sock_poll_wait(file, sock, wait); 2582 if (sk->sk_err) 2583 mask |= EPOLLERR; 2584 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2585 (sk->sk_state == SMC_CLOSED)) 2586 mask |= EPOLLHUP; 2587 if (sk->sk_state == SMC_LISTEN) { 2588 /* woken up by sk_data_ready in smc_listen_work() */ 2589 mask |= smc_accept_poll(sk); 2590 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2591 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2592 wait); 2593 sk->sk_err = smc->clcsock->sk->sk_err; 2594 } else { 2595 if ((sk->sk_state != SMC_INIT && 2596 atomic_read(&smc->conn.sndbuf_space)) || 2597 sk->sk_shutdown & SEND_SHUTDOWN) { 2598 mask |= EPOLLOUT | EPOLLWRNORM; 2599 } else { 2600 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2601 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2602 } 2603 if (atomic_read(&smc->conn.bytes_to_rcv)) 2604 mask |= EPOLLIN | EPOLLRDNORM; 2605 if (sk->sk_shutdown & RCV_SHUTDOWN) 2606 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2607 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2608 mask |= EPOLLIN; 2609 if (smc->conn.urg_state == SMC_URG_VALID) 2610 mask |= EPOLLPRI; 2611 } 2612 } 2613 2614 return mask; 2615 } 2616 2617 static int smc_shutdown(struct socket *sock, int how) 2618 { 2619 struct sock *sk = sock->sk; 2620 bool do_shutdown = true; 2621 struct smc_sock *smc; 2622 int rc = -EINVAL; 2623 int old_state; 2624 int rc1 = 0; 2625 2626 smc = smc_sk(sk); 2627 2628 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2629 return rc; 2630 2631 lock_sock(sk); 2632 2633 rc = -ENOTCONN; 2634 if ((sk->sk_state != SMC_ACTIVE) && 2635 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2636 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2637 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2638 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2639 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2640 goto out; 2641 if (smc->use_fallback) { 2642 rc = kernel_sock_shutdown(smc->clcsock, how); 2643 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2644 if (sk->sk_shutdown == SHUTDOWN_MASK) 2645 sk->sk_state = SMC_CLOSED; 2646 goto out; 2647 } 2648 switch (how) { 2649 case SHUT_RDWR: /* shutdown in both directions */ 2650 old_state = sk->sk_state; 2651 rc = smc_close_active(smc); 2652 if (old_state == SMC_ACTIVE && 2653 sk->sk_state == SMC_PEERCLOSEWAIT1) 2654 do_shutdown = false; 2655 break; 2656 case SHUT_WR: 2657 rc = smc_close_shutdown_write(smc); 2658 break; 2659 case SHUT_RD: 2660 rc = 0; 2661 /* nothing more to do because peer is not involved */ 2662 break; 2663 } 2664 if (do_shutdown && smc->clcsock) 2665 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2666 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2667 sk->sk_shutdown |= how + 1; 2668 2669 out: 2670 release_sock(sk); 2671 return rc ? rc : rc1; 2672 } 2673 2674 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2675 char __user *optval, int __user *optlen) 2676 { 2677 struct smc_sock *smc; 2678 int val, len; 2679 2680 smc = smc_sk(sock->sk); 2681 2682 if (get_user(len, optlen)) 2683 return -EFAULT; 2684 2685 len = min_t(int, len, sizeof(int)); 2686 2687 if (len < 0) 2688 return -EINVAL; 2689 2690 switch (optname) { 2691 case SMC_LIMIT_HS: 2692 val = smc->limit_smc_hs; 2693 break; 2694 default: 2695 return -EOPNOTSUPP; 2696 } 2697 2698 if (put_user(len, optlen)) 2699 return -EFAULT; 2700 if (copy_to_user(optval, &val, len)) 2701 return -EFAULT; 2702 2703 return 0; 2704 } 2705 2706 static int __smc_setsockopt(struct socket *sock, int level, int optname, 2707 sockptr_t optval, unsigned int optlen) 2708 { 2709 struct sock *sk = sock->sk; 2710 struct smc_sock *smc; 2711 int val, rc; 2712 2713 smc = smc_sk(sk); 2714 2715 lock_sock(sk); 2716 switch (optname) { 2717 case SMC_LIMIT_HS: 2718 if (optlen < sizeof(int)) { 2719 rc = -EINVAL; 2720 break; 2721 } 2722 if (copy_from_sockptr(&val, optval, sizeof(int))) { 2723 rc = -EFAULT; 2724 break; 2725 } 2726 2727 smc->limit_smc_hs = !!val; 2728 rc = 0; 2729 break; 2730 default: 2731 rc = -EOPNOTSUPP; 2732 break; 2733 } 2734 release_sock(sk); 2735 2736 return rc; 2737 } 2738 2739 static int smc_setsockopt(struct socket *sock, int level, int optname, 2740 sockptr_t optval, unsigned int optlen) 2741 { 2742 struct sock *sk = sock->sk; 2743 struct smc_sock *smc; 2744 int val, rc; 2745 2746 if (level == SOL_TCP && optname == TCP_ULP) 2747 return -EOPNOTSUPP; 2748 else if (level == SOL_SMC) 2749 return __smc_setsockopt(sock, level, optname, optval, optlen); 2750 2751 smc = smc_sk(sk); 2752 2753 /* generic setsockopts reaching us here always apply to the 2754 * CLC socket 2755 */ 2756 mutex_lock(&smc->clcsock_release_lock); 2757 if (!smc->clcsock) { 2758 mutex_unlock(&smc->clcsock_release_lock); 2759 return -EBADF; 2760 } 2761 if (unlikely(!smc->clcsock->ops->setsockopt)) 2762 rc = -EOPNOTSUPP; 2763 else 2764 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 2765 optval, optlen); 2766 if (smc->clcsock->sk->sk_err) { 2767 sk->sk_err = smc->clcsock->sk->sk_err; 2768 sk_error_report(sk); 2769 } 2770 mutex_unlock(&smc->clcsock_release_lock); 2771 2772 if (optlen < sizeof(int)) 2773 return -EINVAL; 2774 if (copy_from_sockptr(&val, optval, sizeof(int))) 2775 return -EFAULT; 2776 2777 lock_sock(sk); 2778 if (rc || smc->use_fallback) 2779 goto out; 2780 switch (optname) { 2781 case TCP_FASTOPEN: 2782 case TCP_FASTOPEN_CONNECT: 2783 case TCP_FASTOPEN_KEY: 2784 case TCP_FASTOPEN_NO_COOKIE: 2785 /* option not supported by SMC */ 2786 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2787 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2788 } else { 2789 rc = -EINVAL; 2790 } 2791 break; 2792 case TCP_NODELAY: 2793 if (sk->sk_state != SMC_INIT && 2794 sk->sk_state != SMC_LISTEN && 2795 sk->sk_state != SMC_CLOSED) { 2796 if (val) { 2797 SMC_STAT_INC(smc, ndly_cnt); 2798 mod_delayed_work(smc->conn.lgr->tx_wq, 2799 &smc->conn.tx_work, 0); 2800 } 2801 } 2802 break; 2803 case TCP_CORK: 2804 if (sk->sk_state != SMC_INIT && 2805 sk->sk_state != SMC_LISTEN && 2806 sk->sk_state != SMC_CLOSED) { 2807 if (!val) { 2808 SMC_STAT_INC(smc, cork_cnt); 2809 smc_tx_pending(&smc->conn); 2810 cancel_delayed_work(&smc->conn.tx_work); 2811 } 2812 } 2813 break; 2814 case TCP_DEFER_ACCEPT: 2815 smc->sockopt_defer_accept = val; 2816 break; 2817 default: 2818 break; 2819 } 2820 out: 2821 release_sock(sk); 2822 2823 return rc; 2824 } 2825 2826 static int smc_getsockopt(struct socket *sock, int level, int optname, 2827 char __user *optval, int __user *optlen) 2828 { 2829 struct smc_sock *smc; 2830 int rc; 2831 2832 if (level == SOL_SMC) 2833 return __smc_getsockopt(sock, level, optname, optval, optlen); 2834 2835 smc = smc_sk(sock->sk); 2836 mutex_lock(&smc->clcsock_release_lock); 2837 if (!smc->clcsock) { 2838 mutex_unlock(&smc->clcsock_release_lock); 2839 return -EBADF; 2840 } 2841 /* socket options apply to the CLC socket */ 2842 if (unlikely(!smc->clcsock->ops->getsockopt)) { 2843 mutex_unlock(&smc->clcsock_release_lock); 2844 return -EOPNOTSUPP; 2845 } 2846 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 2847 optval, optlen); 2848 mutex_unlock(&smc->clcsock_release_lock); 2849 return rc; 2850 } 2851 2852 static int smc_ioctl(struct socket *sock, unsigned int cmd, 2853 unsigned long arg) 2854 { 2855 union smc_host_cursor cons, urg; 2856 struct smc_connection *conn; 2857 struct smc_sock *smc; 2858 int answ; 2859 2860 smc = smc_sk(sock->sk); 2861 conn = &smc->conn; 2862 lock_sock(&smc->sk); 2863 if (smc->use_fallback) { 2864 if (!smc->clcsock) { 2865 release_sock(&smc->sk); 2866 return -EBADF; 2867 } 2868 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 2869 release_sock(&smc->sk); 2870 return answ; 2871 } 2872 switch (cmd) { 2873 case SIOCINQ: /* same as FIONREAD */ 2874 if (smc->sk.sk_state == SMC_LISTEN) { 2875 release_sock(&smc->sk); 2876 return -EINVAL; 2877 } 2878 if (smc->sk.sk_state == SMC_INIT || 2879 smc->sk.sk_state == SMC_CLOSED) 2880 answ = 0; 2881 else 2882 answ = atomic_read(&smc->conn.bytes_to_rcv); 2883 break; 2884 case SIOCOUTQ: 2885 /* output queue size (not send + not acked) */ 2886 if (smc->sk.sk_state == SMC_LISTEN) { 2887 release_sock(&smc->sk); 2888 return -EINVAL; 2889 } 2890 if (smc->sk.sk_state == SMC_INIT || 2891 smc->sk.sk_state == SMC_CLOSED) 2892 answ = 0; 2893 else 2894 answ = smc->conn.sndbuf_desc->len - 2895 atomic_read(&smc->conn.sndbuf_space); 2896 break; 2897 case SIOCOUTQNSD: 2898 /* output queue size (not send only) */ 2899 if (smc->sk.sk_state == SMC_LISTEN) { 2900 release_sock(&smc->sk); 2901 return -EINVAL; 2902 } 2903 if (smc->sk.sk_state == SMC_INIT || 2904 smc->sk.sk_state == SMC_CLOSED) 2905 answ = 0; 2906 else 2907 answ = smc_tx_prepared_sends(&smc->conn); 2908 break; 2909 case SIOCATMARK: 2910 if (smc->sk.sk_state == SMC_LISTEN) { 2911 release_sock(&smc->sk); 2912 return -EINVAL; 2913 } 2914 if (smc->sk.sk_state == SMC_INIT || 2915 smc->sk.sk_state == SMC_CLOSED) { 2916 answ = 0; 2917 } else { 2918 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 2919 smc_curs_copy(&urg, &conn->urg_curs, conn); 2920 answ = smc_curs_diff(conn->rmb_desc->len, 2921 &cons, &urg) == 1; 2922 } 2923 break; 2924 default: 2925 release_sock(&smc->sk); 2926 return -ENOIOCTLCMD; 2927 } 2928 release_sock(&smc->sk); 2929 2930 return put_user(answ, (int __user *)arg); 2931 } 2932 2933 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 2934 int offset, size_t size, int flags) 2935 { 2936 struct sock *sk = sock->sk; 2937 struct smc_sock *smc; 2938 int rc = -EPIPE; 2939 2940 smc = smc_sk(sk); 2941 lock_sock(sk); 2942 if (sk->sk_state != SMC_ACTIVE) { 2943 release_sock(sk); 2944 goto out; 2945 } 2946 release_sock(sk); 2947 if (smc->use_fallback) { 2948 rc = kernel_sendpage(smc->clcsock, page, offset, 2949 size, flags); 2950 } else { 2951 lock_sock(sk); 2952 rc = smc_tx_sendpage(smc, page, offset, size, flags); 2953 release_sock(sk); 2954 SMC_STAT_INC(smc, sendpage_cnt); 2955 } 2956 2957 out: 2958 return rc; 2959 } 2960 2961 /* Map the affected portions of the rmbe into an spd, note the number of bytes 2962 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 2963 * updates till whenever a respective page has been fully processed. 2964 * Note that subsequent recv() calls have to wait till all splice() processing 2965 * completed. 2966 */ 2967 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 2968 struct pipe_inode_info *pipe, size_t len, 2969 unsigned int flags) 2970 { 2971 struct sock *sk = sock->sk; 2972 struct smc_sock *smc; 2973 int rc = -ENOTCONN; 2974 2975 smc = smc_sk(sk); 2976 lock_sock(sk); 2977 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2978 /* socket was connected before, no more data to read */ 2979 rc = 0; 2980 goto out; 2981 } 2982 if (sk->sk_state == SMC_INIT || 2983 sk->sk_state == SMC_LISTEN || 2984 sk->sk_state == SMC_CLOSED) 2985 goto out; 2986 2987 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2988 rc = 0; 2989 goto out; 2990 } 2991 2992 if (smc->use_fallback) { 2993 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 2994 pipe, len, flags); 2995 } else { 2996 if (*ppos) { 2997 rc = -ESPIPE; 2998 goto out; 2999 } 3000 if (flags & SPLICE_F_NONBLOCK) 3001 flags = MSG_DONTWAIT; 3002 else 3003 flags = 0; 3004 SMC_STAT_INC(smc, splice_cnt); 3005 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3006 } 3007 out: 3008 release_sock(sk); 3009 3010 return rc; 3011 } 3012 3013 /* must look like tcp */ 3014 static const struct proto_ops smc_sock_ops = { 3015 .family = PF_SMC, 3016 .owner = THIS_MODULE, 3017 .release = smc_release, 3018 .bind = smc_bind, 3019 .connect = smc_connect, 3020 .socketpair = sock_no_socketpair, 3021 .accept = smc_accept, 3022 .getname = smc_getname, 3023 .poll = smc_poll, 3024 .ioctl = smc_ioctl, 3025 .listen = smc_listen, 3026 .shutdown = smc_shutdown, 3027 .setsockopt = smc_setsockopt, 3028 .getsockopt = smc_getsockopt, 3029 .sendmsg = smc_sendmsg, 3030 .recvmsg = smc_recvmsg, 3031 .mmap = sock_no_mmap, 3032 .sendpage = smc_sendpage, 3033 .splice_read = smc_splice_read, 3034 }; 3035 3036 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3037 int kern, struct socket *clcsock) 3038 { 3039 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3040 struct smc_sock *smc; 3041 struct sock *sk; 3042 int rc; 3043 3044 rc = -ESOCKTNOSUPPORT; 3045 if (sock->type != SOCK_STREAM) 3046 goto out; 3047 3048 rc = -EPROTONOSUPPORT; 3049 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3050 goto out; 3051 3052 rc = -ENOBUFS; 3053 sock->ops = &smc_sock_ops; 3054 sk = smc_sock_alloc(net, sock, protocol); 3055 if (!sk) 3056 goto out; 3057 3058 /* create internal TCP socket for CLC handshake and fallback */ 3059 smc = smc_sk(sk); 3060 smc->use_fallback = false; /* assume rdma capability first */ 3061 smc->fallback_rsn = 0; 3062 3063 /* default behavior from limit_smc_hs in every net namespace */ 3064 smc->limit_smc_hs = net->smc.limit_smc_hs; 3065 3066 rc = 0; 3067 if (!clcsock) { 3068 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3069 &smc->clcsock); 3070 if (rc) { 3071 sk_common_release(sk); 3072 goto out; 3073 } 3074 } else { 3075 smc->clcsock = clcsock; 3076 } 3077 3078 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 3079 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 3080 3081 out: 3082 return rc; 3083 } 3084 3085 static int smc_create(struct net *net, struct socket *sock, int protocol, 3086 int kern) 3087 { 3088 return __smc_create(net, sock, protocol, kern, NULL); 3089 } 3090 3091 static const struct net_proto_family smc_sock_family_ops = { 3092 .family = PF_SMC, 3093 .owner = THIS_MODULE, 3094 .create = smc_create, 3095 }; 3096 3097 static int smc_ulp_init(struct sock *sk) 3098 { 3099 struct socket *tcp = sk->sk_socket; 3100 struct net *net = sock_net(sk); 3101 struct socket *smcsock; 3102 int protocol, ret; 3103 3104 /* only TCP can be replaced */ 3105 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3106 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3107 return -ESOCKTNOSUPPORT; 3108 /* don't handle wq now */ 3109 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3110 return -ENOTCONN; 3111 3112 if (sk->sk_family == AF_INET) 3113 protocol = SMCPROTO_SMC; 3114 else 3115 protocol = SMCPROTO_SMC6; 3116 3117 smcsock = sock_alloc(); 3118 if (!smcsock) 3119 return -ENFILE; 3120 3121 smcsock->type = SOCK_STREAM; 3122 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3123 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3124 if (ret) { 3125 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3126 return ret; 3127 } 3128 3129 /* replace tcp socket to smc */ 3130 smcsock->file = tcp->file; 3131 smcsock->file->private_data = smcsock; 3132 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3133 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3134 tcp->file = NULL; 3135 3136 return ret; 3137 } 3138 3139 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3140 const gfp_t priority) 3141 { 3142 struct inet_connection_sock *icsk = inet_csk(newsk); 3143 3144 /* don't inherit ulp ops to child when listen */ 3145 icsk->icsk_ulp_ops = NULL; 3146 } 3147 3148 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3149 .name = "smc", 3150 .owner = THIS_MODULE, 3151 .init = smc_ulp_init, 3152 .clone = smc_ulp_clone, 3153 }; 3154 3155 unsigned int smc_net_id; 3156 3157 static __net_init int smc_net_init(struct net *net) 3158 { 3159 return smc_pnet_net_init(net); 3160 } 3161 3162 static void __net_exit smc_net_exit(struct net *net) 3163 { 3164 smc_pnet_net_exit(net); 3165 } 3166 3167 static __net_init int smc_net_stat_init(struct net *net) 3168 { 3169 return smc_stats_init(net); 3170 } 3171 3172 static void __net_exit smc_net_stat_exit(struct net *net) 3173 { 3174 smc_stats_exit(net); 3175 } 3176 3177 static struct pernet_operations smc_net_ops = { 3178 .init = smc_net_init, 3179 .exit = smc_net_exit, 3180 .id = &smc_net_id, 3181 .size = sizeof(struct smc_net), 3182 }; 3183 3184 static struct pernet_operations smc_net_stat_ops = { 3185 .init = smc_net_stat_init, 3186 .exit = smc_net_stat_exit, 3187 }; 3188 3189 static int __init smc_init(void) 3190 { 3191 int rc; 3192 3193 rc = register_pernet_subsys(&smc_net_ops); 3194 if (rc) 3195 return rc; 3196 3197 rc = register_pernet_subsys(&smc_net_stat_ops); 3198 if (rc) 3199 return rc; 3200 3201 smc_ism_init(); 3202 smc_clc_init(); 3203 3204 rc = smc_nl_init(); 3205 if (rc) 3206 goto out_pernet_subsys; 3207 3208 rc = smc_pnet_init(); 3209 if (rc) 3210 goto out_nl; 3211 3212 rc = -ENOMEM; 3213 3214 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3215 if (!smc_tcp_ls_wq) 3216 goto out_pnet; 3217 3218 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3219 if (!smc_hs_wq) 3220 goto out_alloc_tcp_ls_wq; 3221 3222 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3223 if (!smc_close_wq) 3224 goto out_alloc_hs_wq; 3225 3226 rc = smc_core_init(); 3227 if (rc) { 3228 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3229 goto out_alloc_wqs; 3230 } 3231 3232 rc = smc_llc_init(); 3233 if (rc) { 3234 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3235 goto out_core; 3236 } 3237 3238 rc = smc_cdc_init(); 3239 if (rc) { 3240 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3241 goto out_core; 3242 } 3243 3244 rc = proto_register(&smc_proto, 1); 3245 if (rc) { 3246 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3247 goto out_core; 3248 } 3249 3250 rc = proto_register(&smc_proto6, 1); 3251 if (rc) { 3252 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3253 goto out_proto; 3254 } 3255 3256 rc = sock_register(&smc_sock_family_ops); 3257 if (rc) { 3258 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3259 goto out_proto6; 3260 } 3261 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3262 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3263 3264 rc = smc_ib_register_client(); 3265 if (rc) { 3266 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3267 goto out_sock; 3268 } 3269 3270 rc = tcp_register_ulp(&smc_ulp_ops); 3271 if (rc) { 3272 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3273 goto out_sock; 3274 } 3275 3276 static_branch_enable(&tcp_have_smc); 3277 return 0; 3278 3279 out_sock: 3280 sock_unregister(PF_SMC); 3281 out_proto6: 3282 proto_unregister(&smc_proto6); 3283 out_proto: 3284 proto_unregister(&smc_proto); 3285 out_core: 3286 smc_core_exit(); 3287 out_alloc_wqs: 3288 destroy_workqueue(smc_close_wq); 3289 out_alloc_hs_wq: 3290 destroy_workqueue(smc_hs_wq); 3291 out_alloc_tcp_ls_wq: 3292 destroy_workqueue(smc_tcp_ls_wq); 3293 out_pnet: 3294 smc_pnet_exit(); 3295 out_nl: 3296 smc_nl_exit(); 3297 out_pernet_subsys: 3298 unregister_pernet_subsys(&smc_net_ops); 3299 3300 return rc; 3301 } 3302 3303 static void __exit smc_exit(void) 3304 { 3305 static_branch_disable(&tcp_have_smc); 3306 tcp_unregister_ulp(&smc_ulp_ops); 3307 sock_unregister(PF_SMC); 3308 smc_core_exit(); 3309 smc_ib_unregister_client(); 3310 destroy_workqueue(smc_close_wq); 3311 destroy_workqueue(smc_tcp_ls_wq); 3312 destroy_workqueue(smc_hs_wq); 3313 proto_unregister(&smc_proto6); 3314 proto_unregister(&smc_proto); 3315 smc_pnet_exit(); 3316 smc_nl_exit(); 3317 smc_clc_exit(); 3318 unregister_pernet_subsys(&smc_net_stat_ops); 3319 unregister_pernet_subsys(&smc_net_ops); 3320 rcu_barrier(); 3321 } 3322 3323 module_init(smc_init); 3324 module_exit(smc_exit); 3325 3326 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3327 MODULE_DESCRIPTION("smc socket address family"); 3328 MODULE_LICENSE("GPL"); 3329 MODULE_ALIAS_NETPROTO(PF_SMC); 3330 MODULE_ALIAS_TCP_ULP("smc"); 3331