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 #include "smc_sysctl.h" 55 56 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 57 * creation on server 58 */ 59 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 60 * creation on client 61 */ 62 63 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */ 64 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 65 struct workqueue_struct *smc_close_wq; /* wq for close work */ 66 67 static void smc_tcp_listen_work(struct work_struct *); 68 static void smc_connect_work(struct work_struct *); 69 70 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb) 71 { 72 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb); 73 void *hdr; 74 75 if (cb_ctx->pos[0]) 76 goto out; 77 78 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq, 79 &smc_gen_nl_family, NLM_F_MULTI, 80 SMC_NETLINK_DUMP_HS_LIMITATION); 81 if (!hdr) 82 return -ENOMEM; 83 84 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED, 85 sock_net(skb->sk)->smc.limit_smc_hs)) 86 goto err; 87 88 genlmsg_end(skb, hdr); 89 cb_ctx->pos[0] = 1; 90 out: 91 return skb->len; 92 err: 93 genlmsg_cancel(skb, hdr); 94 return -EMSGSIZE; 95 } 96 97 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 98 { 99 sock_net(skb->sk)->smc.limit_smc_hs = true; 100 return 0; 101 } 102 103 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info) 104 { 105 sock_net(skb->sk)->smc.limit_smc_hs = false; 106 return 0; 107 } 108 109 static void smc_set_keepalive(struct sock *sk, int val) 110 { 111 struct smc_sock *smc = smc_sk(sk); 112 113 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 114 } 115 116 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk, 117 struct sk_buff *skb, 118 struct request_sock *req, 119 struct dst_entry *dst, 120 struct request_sock *req_unhash, 121 bool *own_req) 122 { 123 struct smc_sock *smc; 124 struct sock *child; 125 126 smc = smc_clcsock_user_data(sk); 127 128 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) > 129 sk->sk_max_ack_backlog) 130 goto drop; 131 132 if (sk_acceptq_is_full(&smc->sk)) { 133 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS); 134 goto drop; 135 } 136 137 /* passthrough to original syn recv sock fct */ 138 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash, 139 own_req); 140 /* child must not inherit smc or its ops */ 141 if (child) { 142 rcu_assign_sk_user_data(child, NULL); 143 144 /* v4-mapped sockets don't inherit parent ops. Don't restore. */ 145 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops) 146 inet_csk(child)->icsk_af_ops = smc->ori_af_ops; 147 } 148 return child; 149 150 drop: 151 dst_release(dst); 152 tcp_listendrop(sk); 153 return NULL; 154 } 155 156 static bool smc_hs_congested(const struct sock *sk) 157 { 158 const struct smc_sock *smc; 159 160 smc = smc_clcsock_user_data(sk); 161 162 if (!smc) 163 return true; 164 165 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq)) 166 return true; 167 168 return false; 169 } 170 171 static struct smc_hashinfo smc_v4_hashinfo = { 172 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 173 }; 174 175 static struct smc_hashinfo smc_v6_hashinfo = { 176 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 177 }; 178 179 int smc_hash_sk(struct sock *sk) 180 { 181 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 182 struct hlist_head *head; 183 184 head = &h->ht; 185 186 write_lock_bh(&h->lock); 187 sk_add_node(sk, head); 188 write_unlock_bh(&h->lock); 189 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 190 191 return 0; 192 } 193 EXPORT_SYMBOL_GPL(smc_hash_sk); 194 195 void smc_unhash_sk(struct sock *sk) 196 { 197 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 198 199 write_lock_bh(&h->lock); 200 if (sk_del_node_init(sk)) 201 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 202 write_unlock_bh(&h->lock); 203 } 204 EXPORT_SYMBOL_GPL(smc_unhash_sk); 205 206 /* This will be called before user really release sock_lock. So do the 207 * work which we didn't do because of user hold the sock_lock in the 208 * BH context 209 */ 210 static void smc_release_cb(struct sock *sk) 211 { 212 struct smc_sock *smc = smc_sk(sk); 213 214 if (smc->conn.tx_in_release_sock) { 215 smc_tx_pending(&smc->conn); 216 smc->conn.tx_in_release_sock = false; 217 } 218 } 219 220 struct proto smc_proto = { 221 .name = "SMC", 222 .owner = THIS_MODULE, 223 .keepalive = smc_set_keepalive, 224 .hash = smc_hash_sk, 225 .unhash = smc_unhash_sk, 226 .release_cb = smc_release_cb, 227 .obj_size = sizeof(struct smc_sock), 228 .h.smc_hash = &smc_v4_hashinfo, 229 .slab_flags = SLAB_TYPESAFE_BY_RCU, 230 }; 231 EXPORT_SYMBOL_GPL(smc_proto); 232 233 struct proto smc_proto6 = { 234 .name = "SMC6", 235 .owner = THIS_MODULE, 236 .keepalive = smc_set_keepalive, 237 .hash = smc_hash_sk, 238 .unhash = smc_unhash_sk, 239 .release_cb = smc_release_cb, 240 .obj_size = sizeof(struct smc_sock), 241 .h.smc_hash = &smc_v6_hashinfo, 242 .slab_flags = SLAB_TYPESAFE_BY_RCU, 243 }; 244 EXPORT_SYMBOL_GPL(smc_proto6); 245 246 static void smc_fback_restore_callbacks(struct smc_sock *smc) 247 { 248 struct sock *clcsk = smc->clcsock->sk; 249 250 write_lock_bh(&clcsk->sk_callback_lock); 251 clcsk->sk_user_data = NULL; 252 253 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change); 254 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready); 255 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space); 256 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report); 257 258 write_unlock_bh(&clcsk->sk_callback_lock); 259 } 260 261 static void smc_restore_fallback_changes(struct smc_sock *smc) 262 { 263 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */ 264 smc->clcsock->file->private_data = smc->sk.sk_socket; 265 smc->clcsock->file = NULL; 266 smc_fback_restore_callbacks(smc); 267 } 268 } 269 270 static int __smc_release(struct smc_sock *smc) 271 { 272 struct sock *sk = &smc->sk; 273 int rc = 0; 274 275 if (!smc->use_fallback) { 276 rc = smc_close_active(smc); 277 sock_set_flag(sk, SOCK_DEAD); 278 sk->sk_shutdown |= SHUTDOWN_MASK; 279 } else { 280 if (sk->sk_state != SMC_CLOSED) { 281 if (sk->sk_state != SMC_LISTEN && 282 sk->sk_state != SMC_INIT) 283 sock_put(sk); /* passive closing */ 284 if (sk->sk_state == SMC_LISTEN) { 285 /* wake up clcsock accept */ 286 rc = kernel_sock_shutdown(smc->clcsock, 287 SHUT_RDWR); 288 } 289 sk->sk_state = SMC_CLOSED; 290 sk->sk_state_change(sk); 291 } 292 smc_restore_fallback_changes(smc); 293 } 294 295 sk->sk_prot->unhash(sk); 296 297 if (sk->sk_state == SMC_CLOSED) { 298 if (smc->clcsock) { 299 release_sock(sk); 300 smc_clcsock_release(smc); 301 lock_sock(sk); 302 } 303 if (!smc->use_fallback) 304 smc_conn_free(&smc->conn); 305 } 306 307 return rc; 308 } 309 310 static int smc_release(struct socket *sock) 311 { 312 struct sock *sk = sock->sk; 313 struct smc_sock *smc; 314 int old_state, rc = 0; 315 316 if (!sk) 317 goto out; 318 319 sock_hold(sk); /* sock_put below */ 320 smc = smc_sk(sk); 321 322 old_state = sk->sk_state; 323 324 /* cleanup for a dangling non-blocking connect */ 325 if (smc->connect_nonblock && old_state == SMC_INIT) 326 tcp_abort(smc->clcsock->sk, ECONNABORTED); 327 328 if (cancel_work_sync(&smc->connect_work)) 329 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */ 330 331 if (sk->sk_state == SMC_LISTEN) 332 /* smc_close_non_accepted() is called and acquires 333 * sock lock for child sockets again 334 */ 335 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 336 else 337 lock_sock(sk); 338 339 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE && 340 !smc->use_fallback) 341 smc_close_active_abort(smc); 342 343 rc = __smc_release(smc); 344 345 /* detach socket */ 346 sock_orphan(sk); 347 sock->sk = NULL; 348 release_sock(sk); 349 350 sock_put(sk); /* sock_hold above */ 351 sock_put(sk); /* final sock_put */ 352 out: 353 return rc; 354 } 355 356 static void smc_destruct(struct sock *sk) 357 { 358 if (sk->sk_state != SMC_CLOSED) 359 return; 360 if (!sock_flag(sk, SOCK_DEAD)) 361 return; 362 363 sk_refcnt_debug_dec(sk); 364 } 365 366 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 367 int protocol) 368 { 369 struct smc_sock *smc; 370 struct proto *prot; 371 struct sock *sk; 372 373 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 374 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 375 if (!sk) 376 return NULL; 377 378 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 379 sk->sk_state = SMC_INIT; 380 sk->sk_destruct = smc_destruct; 381 sk->sk_protocol = protocol; 382 smc = smc_sk(sk); 383 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 384 INIT_WORK(&smc->connect_work, smc_connect_work); 385 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 386 INIT_LIST_HEAD(&smc->accept_q); 387 spin_lock_init(&smc->accept_q_lock); 388 spin_lock_init(&smc->conn.send_lock); 389 sk->sk_prot->hash(sk); 390 sk_refcnt_debug_inc(sk); 391 mutex_init(&smc->clcsock_release_lock); 392 smc_init_saved_callbacks(smc); 393 394 return sk; 395 } 396 397 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 398 int addr_len) 399 { 400 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 401 struct sock *sk = sock->sk; 402 struct smc_sock *smc; 403 int rc; 404 405 smc = smc_sk(sk); 406 407 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 408 rc = -EINVAL; 409 if (addr_len < sizeof(struct sockaddr_in)) 410 goto out; 411 412 rc = -EAFNOSUPPORT; 413 if (addr->sin_family != AF_INET && 414 addr->sin_family != AF_INET6 && 415 addr->sin_family != AF_UNSPEC) 416 goto out; 417 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 418 if (addr->sin_family == AF_UNSPEC && 419 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 420 goto out; 421 422 lock_sock(sk); 423 424 /* Check if socket is already active */ 425 rc = -EINVAL; 426 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 427 goto out_rel; 428 429 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 430 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 431 432 out_rel: 433 release_sock(sk); 434 out: 435 return rc; 436 } 437 438 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 439 unsigned long mask) 440 { 441 /* options we don't get control via setsockopt for */ 442 nsk->sk_type = osk->sk_type; 443 nsk->sk_sndbuf = osk->sk_sndbuf; 444 nsk->sk_rcvbuf = osk->sk_rcvbuf; 445 nsk->sk_sndtimeo = osk->sk_sndtimeo; 446 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 447 nsk->sk_mark = osk->sk_mark; 448 nsk->sk_priority = osk->sk_priority; 449 nsk->sk_rcvlowat = osk->sk_rcvlowat; 450 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 451 nsk->sk_err = osk->sk_err; 452 453 nsk->sk_flags &= ~mask; 454 nsk->sk_flags |= osk->sk_flags & mask; 455 } 456 457 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 458 (1UL << SOCK_KEEPOPEN) | \ 459 (1UL << SOCK_LINGER) | \ 460 (1UL << SOCK_BROADCAST) | \ 461 (1UL << SOCK_TIMESTAMP) | \ 462 (1UL << SOCK_DBG) | \ 463 (1UL << SOCK_RCVTSTAMP) | \ 464 (1UL << SOCK_RCVTSTAMPNS) | \ 465 (1UL << SOCK_LOCALROUTE) | \ 466 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 467 (1UL << SOCK_RXQ_OVFL) | \ 468 (1UL << SOCK_WIFI_STATUS) | \ 469 (1UL << SOCK_NOFCS) | \ 470 (1UL << SOCK_FILTER_LOCKED) | \ 471 (1UL << SOCK_TSTAMP_NEW)) 472 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 473 * clc socket (since smc is not called for these options from net/core) 474 */ 475 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 476 { 477 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 478 } 479 480 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 481 (1UL << SOCK_KEEPOPEN) | \ 482 (1UL << SOCK_LINGER) | \ 483 (1UL << SOCK_DBG)) 484 /* copy only settings and flags relevant for smc from clc to smc socket */ 485 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 486 { 487 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 488 } 489 490 /* register the new vzalloced sndbuf on all links */ 491 static int smcr_lgr_reg_sndbufs(struct smc_link *link, 492 struct smc_buf_desc *snd_desc) 493 { 494 struct smc_link_group *lgr = link->lgr; 495 int i, rc = 0; 496 497 if (!snd_desc->is_vm) 498 return -EINVAL; 499 500 /* protect against parallel smcr_link_reg_buf() */ 501 mutex_lock(&lgr->llc_conf_mutex); 502 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 503 if (!smc_link_active(&lgr->lnk[i])) 504 continue; 505 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc); 506 if (rc) 507 break; 508 } 509 mutex_unlock(&lgr->llc_conf_mutex); 510 return rc; 511 } 512 513 /* register the new rmb on all links */ 514 static int smcr_lgr_reg_rmbs(struct smc_link *link, 515 struct smc_buf_desc *rmb_desc) 516 { 517 struct smc_link_group *lgr = link->lgr; 518 int i, rc = 0; 519 520 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY); 521 if (rc) 522 return rc; 523 /* protect against parallel smc_llc_cli_rkey_exchange() and 524 * parallel smcr_link_reg_buf() 525 */ 526 mutex_lock(&lgr->llc_conf_mutex); 527 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 528 if (!smc_link_active(&lgr->lnk[i])) 529 continue; 530 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc); 531 if (rc) 532 goto out; 533 } 534 535 /* exchange confirm_rkey msg with peer */ 536 rc = smc_llc_do_confirm_rkey(link, rmb_desc); 537 if (rc) { 538 rc = -EFAULT; 539 goto out; 540 } 541 rmb_desc->is_conf_rkey = true; 542 out: 543 mutex_unlock(&lgr->llc_conf_mutex); 544 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl); 545 return rc; 546 } 547 548 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 549 { 550 struct smc_link *link = smc->conn.lnk; 551 struct smc_llc_qentry *qentry; 552 int rc; 553 554 /* receive CONFIRM LINK request from server over RoCE fabric */ 555 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 556 SMC_LLC_CONFIRM_LINK); 557 if (!qentry) { 558 struct smc_clc_msg_decline dclc; 559 560 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 561 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 562 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 563 } 564 smc_llc_save_peer_uid(qentry); 565 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 566 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 567 if (rc) 568 return SMC_CLC_DECL_RMBE_EC; 569 570 rc = smc_ib_modify_qp_rts(link); 571 if (rc) 572 return SMC_CLC_DECL_ERR_RDYLNK; 573 574 smc_wr_remember_qp_attr(link); 575 576 /* reg the sndbuf if it was vzalloced */ 577 if (smc->conn.sndbuf_desc->is_vm) { 578 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 579 return SMC_CLC_DECL_ERR_REGBUF; 580 } 581 582 /* reg the rmb */ 583 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 584 return SMC_CLC_DECL_ERR_REGBUF; 585 586 /* confirm_rkey is implicit on 1st contact */ 587 smc->conn.rmb_desc->is_conf_rkey = true; 588 589 /* send CONFIRM LINK response over RoCE fabric */ 590 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 591 if (rc < 0) 592 return SMC_CLC_DECL_TIMEOUT_CL; 593 594 smc_llc_link_active(link); 595 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 596 597 /* optional 2nd link, receive ADD LINK request from server */ 598 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 599 SMC_LLC_ADD_LINK); 600 if (!qentry) { 601 struct smc_clc_msg_decline dclc; 602 603 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 604 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 605 if (rc == -EAGAIN) 606 rc = 0; /* no DECLINE received, go with one link */ 607 return rc; 608 } 609 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 610 smc_llc_cli_add_link(link, qentry); 611 return 0; 612 } 613 614 static bool smc_isascii(char *hostname) 615 { 616 int i; 617 618 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++) 619 if (!isascii(hostname[i])) 620 return false; 621 return true; 622 } 623 624 static void smc_conn_save_peer_info_fce(struct smc_sock *smc, 625 struct smc_clc_msg_accept_confirm *clc) 626 { 627 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 628 (struct smc_clc_msg_accept_confirm_v2 *)clc; 629 struct smc_clc_first_contact_ext *fce; 630 int clc_v2_len; 631 632 if (clc->hdr.version == SMC_V1 || 633 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK)) 634 return; 635 636 if (smc->conn.lgr->is_smcd) { 637 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid, 638 SMC_MAX_EID_LEN); 639 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 640 d1); 641 } else { 642 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid, 643 SMC_MAX_EID_LEN); 644 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2, 645 r1); 646 } 647 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len); 648 smc->conn.lgr->peer_os = fce->os_type; 649 smc->conn.lgr->peer_smc_release = fce->release; 650 if (smc_isascii(fce->hostname)) 651 memcpy(smc->conn.lgr->peer_hostname, fce->hostname, 652 SMC_MAX_HOSTNAME_LEN); 653 } 654 655 static void smcr_conn_save_peer_info(struct smc_sock *smc, 656 struct smc_clc_msg_accept_confirm *clc) 657 { 658 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size); 659 660 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx; 661 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token); 662 smc->conn.peer_rmbe_size = bufsize; 663 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 664 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 665 } 666 667 static void smcd_conn_save_peer_info(struct smc_sock *smc, 668 struct smc_clc_msg_accept_confirm *clc) 669 { 670 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size); 671 672 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx; 673 smc->conn.peer_token = clc->d0.token; 674 /* msg header takes up space in the buffer */ 675 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 676 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 677 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 678 } 679 680 static void smc_conn_save_peer_info(struct smc_sock *smc, 681 struct smc_clc_msg_accept_confirm *clc) 682 { 683 if (smc->conn.lgr->is_smcd) 684 smcd_conn_save_peer_info(smc, clc); 685 else 686 smcr_conn_save_peer_info(smc, clc); 687 smc_conn_save_peer_info_fce(smc, clc); 688 } 689 690 static void smc_link_save_peer_info(struct smc_link *link, 691 struct smc_clc_msg_accept_confirm *clc, 692 struct smc_init_info *ini) 693 { 694 link->peer_qpn = ntoh24(clc->r0.qpn); 695 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE); 696 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac)); 697 link->peer_psn = ntoh24(clc->r0.psn); 698 link->peer_mtu = clc->r0.qp_mtu; 699 } 700 701 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc, 702 struct smc_stats_fback *fback_arr) 703 { 704 int cnt; 705 706 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) { 707 if (fback_arr[cnt].fback_code == smc->fallback_rsn) { 708 fback_arr[cnt].count++; 709 break; 710 } 711 if (!fback_arr[cnt].fback_code) { 712 fback_arr[cnt].fback_code = smc->fallback_rsn; 713 fback_arr[cnt].count++; 714 break; 715 } 716 } 717 } 718 719 static void smc_stat_fallback(struct smc_sock *smc) 720 { 721 struct net *net = sock_net(&smc->sk); 722 723 mutex_lock(&net->smc.mutex_fback_rsn); 724 if (smc->listen_smc) { 725 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv); 726 net->smc.fback_rsn->srv_fback_cnt++; 727 } else { 728 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt); 729 net->smc.fback_rsn->clnt_fback_cnt++; 730 } 731 mutex_unlock(&net->smc.mutex_fback_rsn); 732 } 733 734 /* must be called under rcu read lock */ 735 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key) 736 { 737 struct socket_wq *wq; 738 __poll_t flags; 739 740 wq = rcu_dereference(smc->sk.sk_wq); 741 if (!skwq_has_sleeper(wq)) 742 return; 743 744 /* wake up smc sk->sk_wq */ 745 if (!key) { 746 /* sk_state_change */ 747 wake_up_interruptible_all(&wq->wait); 748 } else { 749 flags = key_to_poll(key); 750 if (flags & (EPOLLIN | EPOLLOUT)) 751 /* sk_data_ready or sk_write_space */ 752 wake_up_interruptible_sync_poll(&wq->wait, flags); 753 else if (flags & EPOLLERR) 754 /* sk_error_report */ 755 wake_up_interruptible_poll(&wq->wait, flags); 756 } 757 } 758 759 static int smc_fback_mark_woken(wait_queue_entry_t *wait, 760 unsigned int mode, int sync, void *key) 761 { 762 struct smc_mark_woken *mark = 763 container_of(wait, struct smc_mark_woken, wait_entry); 764 765 mark->woken = true; 766 mark->key = key; 767 return 0; 768 } 769 770 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk, 771 void (*clcsock_callback)(struct sock *sk)) 772 { 773 struct smc_mark_woken mark = { .woken = false }; 774 struct socket_wq *wq; 775 776 init_waitqueue_func_entry(&mark.wait_entry, 777 smc_fback_mark_woken); 778 rcu_read_lock(); 779 wq = rcu_dereference(clcsk->sk_wq); 780 if (!wq) 781 goto out; 782 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 783 clcsock_callback(clcsk); 784 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry); 785 786 if (mark.woken) 787 smc_fback_wakeup_waitqueue(smc, mark.key); 788 out: 789 rcu_read_unlock(); 790 } 791 792 static void smc_fback_state_change(struct sock *clcsk) 793 { 794 struct smc_sock *smc; 795 796 read_lock_bh(&clcsk->sk_callback_lock); 797 smc = smc_clcsock_user_data(clcsk); 798 if (smc) 799 smc_fback_forward_wakeup(smc, clcsk, 800 smc->clcsk_state_change); 801 read_unlock_bh(&clcsk->sk_callback_lock); 802 } 803 804 static void smc_fback_data_ready(struct sock *clcsk) 805 { 806 struct smc_sock *smc; 807 808 read_lock_bh(&clcsk->sk_callback_lock); 809 smc = smc_clcsock_user_data(clcsk); 810 if (smc) 811 smc_fback_forward_wakeup(smc, clcsk, 812 smc->clcsk_data_ready); 813 read_unlock_bh(&clcsk->sk_callback_lock); 814 } 815 816 static void smc_fback_write_space(struct sock *clcsk) 817 { 818 struct smc_sock *smc; 819 820 read_lock_bh(&clcsk->sk_callback_lock); 821 smc = smc_clcsock_user_data(clcsk); 822 if (smc) 823 smc_fback_forward_wakeup(smc, clcsk, 824 smc->clcsk_write_space); 825 read_unlock_bh(&clcsk->sk_callback_lock); 826 } 827 828 static void smc_fback_error_report(struct sock *clcsk) 829 { 830 struct smc_sock *smc; 831 832 read_lock_bh(&clcsk->sk_callback_lock); 833 smc = smc_clcsock_user_data(clcsk); 834 if (smc) 835 smc_fback_forward_wakeup(smc, clcsk, 836 smc->clcsk_error_report); 837 read_unlock_bh(&clcsk->sk_callback_lock); 838 } 839 840 static void smc_fback_replace_callbacks(struct smc_sock *smc) 841 { 842 struct sock *clcsk = smc->clcsock->sk; 843 844 write_lock_bh(&clcsk->sk_callback_lock); 845 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 846 847 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change, 848 &smc->clcsk_state_change); 849 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready, 850 &smc->clcsk_data_ready); 851 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space, 852 &smc->clcsk_write_space); 853 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report, 854 &smc->clcsk_error_report); 855 856 write_unlock_bh(&clcsk->sk_callback_lock); 857 } 858 859 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code) 860 { 861 int rc = 0; 862 863 mutex_lock(&smc->clcsock_release_lock); 864 if (!smc->clcsock) { 865 rc = -EBADF; 866 goto out; 867 } 868 869 smc->use_fallback = true; 870 smc->fallback_rsn = reason_code; 871 smc_stat_fallback(smc); 872 trace_smc_switch_to_fallback(smc, reason_code); 873 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 874 smc->clcsock->file = smc->sk.sk_socket->file; 875 smc->clcsock->file->private_data = smc->clcsock; 876 smc->clcsock->wq.fasync_list = 877 smc->sk.sk_socket->wq.fasync_list; 878 879 /* There might be some wait entries remaining 880 * in smc sk->sk_wq and they should be woken up 881 * as clcsock's wait queue is woken up. 882 */ 883 smc_fback_replace_callbacks(smc); 884 } 885 out: 886 mutex_unlock(&smc->clcsock_release_lock); 887 return rc; 888 } 889 890 /* fall back during connect */ 891 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 892 { 893 struct net *net = sock_net(&smc->sk); 894 int rc = 0; 895 896 rc = smc_switch_to_fallback(smc, reason_code); 897 if (rc) { /* fallback fails */ 898 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 899 if (smc->sk.sk_state == SMC_INIT) 900 sock_put(&smc->sk); /* passive closing */ 901 return rc; 902 } 903 smc_copy_sock_settings_to_clc(smc); 904 smc->connect_nonblock = 0; 905 if (smc->sk.sk_state == SMC_INIT) 906 smc->sk.sk_state = SMC_ACTIVE; 907 return 0; 908 } 909 910 /* decline and fall back during connect */ 911 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code, 912 u8 version) 913 { 914 struct net *net = sock_net(&smc->sk); 915 int rc; 916 917 if (reason_code < 0) { /* error, fallback is not possible */ 918 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 919 if (smc->sk.sk_state == SMC_INIT) 920 sock_put(&smc->sk); /* passive closing */ 921 return reason_code; 922 } 923 if (reason_code != SMC_CLC_DECL_PEERDECL) { 924 rc = smc_clc_send_decline(smc, reason_code, version); 925 if (rc < 0) { 926 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt); 927 if (smc->sk.sk_state == SMC_INIT) 928 sock_put(&smc->sk); /* passive closing */ 929 return rc; 930 } 931 } 932 return smc_connect_fallback(smc, reason_code); 933 } 934 935 static void smc_conn_abort(struct smc_sock *smc, int local_first) 936 { 937 struct smc_connection *conn = &smc->conn; 938 struct smc_link_group *lgr = conn->lgr; 939 bool lgr_valid = false; 940 941 if (smc_conn_lgr_valid(conn)) 942 lgr_valid = true; 943 944 smc_conn_free(conn); 945 if (local_first && lgr_valid) 946 smc_lgr_cleanup_early(lgr); 947 } 948 949 /* check if there is a rdma device available for this connection. */ 950 /* called for connect and listen */ 951 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 952 { 953 /* PNET table look up: search active ib_device and port 954 * within same PNETID that also contains the ethernet device 955 * used for the internal TCP socket 956 */ 957 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 958 if (!ini->check_smcrv2 && !ini->ib_dev) 959 return SMC_CLC_DECL_NOSMCRDEV; 960 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2) 961 return SMC_CLC_DECL_NOSMCRDEV; 962 return 0; 963 } 964 965 /* check if there is an ISM device available for this connection. */ 966 /* called for connect and listen */ 967 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 968 { 969 /* Find ISM device with same PNETID as connecting interface */ 970 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 971 if (!ini->ism_dev[0]) 972 return SMC_CLC_DECL_NOSMCDDEV; 973 else 974 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]); 975 return 0; 976 } 977 978 /* is chid unique for the ism devices that are already determined? */ 979 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini, 980 int cnt) 981 { 982 int i = (!ini->ism_dev[0]) ? 1 : 0; 983 984 for (; i < cnt; i++) 985 if (ini->ism_chid[i] == chid) 986 return false; 987 return true; 988 } 989 990 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus 991 * PNETID matching net_device) 992 */ 993 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc, 994 struct smc_init_info *ini) 995 { 996 int rc = SMC_CLC_DECL_NOSMCDDEV; 997 struct smcd_dev *smcd; 998 int i = 1; 999 u16 chid; 1000 1001 if (smcd_indicated(ini->smc_type_v1)) 1002 rc = 0; /* already initialized for V1 */ 1003 mutex_lock(&smcd_dev_list.mutex); 1004 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 1005 if (smcd->going_away || smcd == ini->ism_dev[0]) 1006 continue; 1007 chid = smc_ism_get_chid(smcd); 1008 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i)) 1009 continue; 1010 if (!smc_pnet_is_pnetid_set(smcd->pnetid) || 1011 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) { 1012 ini->ism_dev[i] = smcd; 1013 ini->ism_chid[i] = chid; 1014 ini->is_smcd = true; 1015 rc = 0; 1016 i++; 1017 if (i > SMC_MAX_ISM_DEVS) 1018 break; 1019 } 1020 } 1021 mutex_unlock(&smcd_dev_list.mutex); 1022 ini->ism_offered_cnt = i - 1; 1023 if (!ini->ism_dev[0] && !ini->ism_dev[1]) 1024 ini->smcd_version = 0; 1025 1026 return rc; 1027 } 1028 1029 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 1030 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 1031 struct smc_init_info *ini) 1032 { 1033 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id)) 1034 return SMC_CLC_DECL_ISMVLANERR; 1035 return 0; 1036 } 1037 1038 static int smc_find_proposal_devices(struct smc_sock *smc, 1039 struct smc_init_info *ini) 1040 { 1041 int rc = 0; 1042 1043 /* check if there is an ism device available */ 1044 if (!(ini->smcd_version & SMC_V1) || 1045 smc_find_ism_device(smc, ini) || 1046 smc_connect_ism_vlan_setup(smc, ini)) 1047 ini->smcd_version &= ~SMC_V1; 1048 /* else ISM V1 is supported for this connection */ 1049 1050 /* check if there is an rdma device available */ 1051 if (!(ini->smcr_version & SMC_V1) || 1052 smc_find_rdma_device(smc, ini)) 1053 ini->smcr_version &= ~SMC_V1; 1054 /* else RDMA is supported for this connection */ 1055 1056 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1, 1057 ini->smcr_version & SMC_V1); 1058 1059 /* check if there is an ism v2 device available */ 1060 if (!(ini->smcd_version & SMC_V2) || 1061 !smc_ism_is_v2_capable() || 1062 smc_find_ism_v2_device_clnt(smc, ini)) 1063 ini->smcd_version &= ~SMC_V2; 1064 1065 /* check if there is an rdma v2 device available */ 1066 ini->check_smcrv2 = true; 1067 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr; 1068 if (!(ini->smcr_version & SMC_V2) || 1069 smc->clcsock->sk->sk_family != AF_INET || 1070 !smc_clc_ueid_count() || 1071 smc_find_rdma_device(smc, ini)) 1072 ini->smcr_version &= ~SMC_V2; 1073 ini->check_smcrv2 = false; 1074 1075 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2, 1076 ini->smcr_version & SMC_V2); 1077 1078 /* if neither ISM nor RDMA are supported, fallback */ 1079 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N) 1080 rc = SMC_CLC_DECL_NOSMCDEV; 1081 1082 return rc; 1083 } 1084 1085 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 1086 * used, the VLAN ID will be registered again during the connection setup. 1087 */ 1088 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, 1089 struct smc_init_info *ini) 1090 { 1091 if (!smcd_indicated(ini->smc_type_v1)) 1092 return 0; 1093 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id)) 1094 return SMC_CLC_DECL_CNFERR; 1095 return 0; 1096 } 1097 1098 #define SMC_CLC_MAX_ACCEPT_LEN \ 1099 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \ 1100 sizeof(struct smc_clc_first_contact_ext) + \ 1101 sizeof(struct smc_clc_msg_trail)) 1102 1103 /* CLC handshake during connect */ 1104 static int smc_connect_clc(struct smc_sock *smc, 1105 struct smc_clc_msg_accept_confirm_v2 *aclc2, 1106 struct smc_init_info *ini) 1107 { 1108 int rc = 0; 1109 1110 /* do inband token exchange */ 1111 rc = smc_clc_send_proposal(smc, ini); 1112 if (rc) 1113 return rc; 1114 /* receive SMC Accept CLC message */ 1115 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN, 1116 SMC_CLC_ACCEPT, CLC_WAIT_TIME); 1117 } 1118 1119 void smc_fill_gid_list(struct smc_link_group *lgr, 1120 struct smc_gidlist *gidlist, 1121 struct smc_ib_device *known_dev, u8 *known_gid) 1122 { 1123 struct smc_init_info *alt_ini = NULL; 1124 1125 memset(gidlist, 0, sizeof(*gidlist)); 1126 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE); 1127 1128 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL); 1129 if (!alt_ini) 1130 goto out; 1131 1132 alt_ini->vlan_id = lgr->vlan_id; 1133 alt_ini->check_smcrv2 = true; 1134 alt_ini->smcrv2.saddr = lgr->saddr; 1135 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev); 1136 1137 if (!alt_ini->smcrv2.ib_dev_v2) 1138 goto out; 1139 1140 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2, 1141 SMC_GID_SIZE); 1142 1143 out: 1144 kfree(alt_ini); 1145 } 1146 1147 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc, 1148 struct smc_clc_msg_accept_confirm *aclc, 1149 struct smc_init_info *ini) 1150 { 1151 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1152 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1153 struct smc_clc_first_contact_ext *fce = 1154 (struct smc_clc_first_contact_ext *) 1155 (((u8 *)clc_v2) + sizeof(*clc_v2)); 1156 1157 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1) 1158 return 0; 1159 1160 if (fce->v2_direct) { 1161 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN); 1162 ini->smcrv2.uses_gateway = false; 1163 } else { 1164 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr, 1165 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid), 1166 ini->smcrv2.nexthop_mac, 1167 &ini->smcrv2.uses_gateway)) 1168 return SMC_CLC_DECL_NOROUTE; 1169 if (!ini->smcrv2.uses_gateway) { 1170 /* mismatch: peer claims indirect, but its direct */ 1171 return SMC_CLC_DECL_NOINDIRECT; 1172 } 1173 } 1174 return 0; 1175 } 1176 1177 /* setup for RDMA connection of client */ 1178 static int smc_connect_rdma(struct smc_sock *smc, 1179 struct smc_clc_msg_accept_confirm *aclc, 1180 struct smc_init_info *ini) 1181 { 1182 int i, reason_code = 0; 1183 struct smc_link *link; 1184 u8 *eid = NULL; 1185 1186 ini->is_smcd = false; 1187 ini->ib_clcqpn = ntoh24(aclc->r0.qpn); 1188 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1189 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN); 1190 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE); 1191 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN); 1192 1193 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini); 1194 if (reason_code) 1195 return reason_code; 1196 1197 mutex_lock(&smc_client_lgr_pending); 1198 reason_code = smc_conn_create(smc, ini); 1199 if (reason_code) { 1200 mutex_unlock(&smc_client_lgr_pending); 1201 return reason_code; 1202 } 1203 1204 smc_conn_save_peer_info(smc, aclc); 1205 1206 if (ini->first_contact_local) { 1207 link = smc->conn.lnk; 1208 } else { 1209 /* set link that was assigned by server */ 1210 link = NULL; 1211 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 1212 struct smc_link *l = &smc->conn.lgr->lnk[i]; 1213 1214 if (l->peer_qpn == ntoh24(aclc->r0.qpn) && 1215 !memcmp(l->peer_gid, &aclc->r0.lcl.gid, 1216 SMC_GID_SIZE) && 1217 (aclc->hdr.version > SMC_V1 || 1218 !memcmp(l->peer_mac, &aclc->r0.lcl.mac, 1219 sizeof(l->peer_mac)))) { 1220 link = l; 1221 break; 1222 } 1223 } 1224 if (!link) { 1225 reason_code = SMC_CLC_DECL_NOSRVLINK; 1226 goto connect_abort; 1227 } 1228 smc_switch_link_and_count(&smc->conn, link); 1229 } 1230 1231 /* create send buffer and rmb */ 1232 if (smc_buf_create(smc, false)) { 1233 reason_code = SMC_CLC_DECL_MEM; 1234 goto connect_abort; 1235 } 1236 1237 if (ini->first_contact_local) 1238 smc_link_save_peer_info(link, aclc, ini); 1239 1240 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) { 1241 reason_code = SMC_CLC_DECL_ERR_RTOK; 1242 goto connect_abort; 1243 } 1244 1245 smc_close_init(smc); 1246 smc_rx_init(smc); 1247 1248 if (ini->first_contact_local) { 1249 if (smc_ib_ready_link(link)) { 1250 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1251 goto connect_abort; 1252 } 1253 } else { 1254 /* reg sendbufs if they were vzalloced */ 1255 if (smc->conn.sndbuf_desc->is_vm) { 1256 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) { 1257 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1258 goto connect_abort; 1259 } 1260 } 1261 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) { 1262 reason_code = SMC_CLC_DECL_ERR_REGBUF; 1263 goto connect_abort; 1264 } 1265 } 1266 1267 if (aclc->hdr.version > SMC_V1) { 1268 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1269 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1270 1271 eid = clc_v2->r1.eid; 1272 if (ini->first_contact_local) 1273 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist, 1274 link->smcibdev, link->gid); 1275 } 1276 1277 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local, 1278 aclc->hdr.version, eid, ini); 1279 if (reason_code) 1280 goto connect_abort; 1281 1282 smc_tx_init(smc); 1283 1284 if (ini->first_contact_local) { 1285 /* QP confirmation over RoCE fabric */ 1286 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1287 reason_code = smcr_clnt_conf_first_link(smc); 1288 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1289 if (reason_code) 1290 goto connect_abort; 1291 } 1292 mutex_unlock(&smc_client_lgr_pending); 1293 1294 smc_copy_sock_settings_to_clc(smc); 1295 smc->connect_nonblock = 0; 1296 if (smc->sk.sk_state == SMC_INIT) 1297 smc->sk.sk_state = SMC_ACTIVE; 1298 1299 return 0; 1300 connect_abort: 1301 smc_conn_abort(smc, ini->first_contact_local); 1302 mutex_unlock(&smc_client_lgr_pending); 1303 smc->connect_nonblock = 0; 1304 1305 return reason_code; 1306 } 1307 1308 /* The server has chosen one of the proposed ISM devices for the communication. 1309 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen. 1310 */ 1311 static int 1312 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc, 1313 struct smc_init_info *ini) 1314 { 1315 int i; 1316 1317 for (i = 0; i < ini->ism_offered_cnt + 1; i++) { 1318 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) { 1319 ini->ism_selected = i; 1320 return 0; 1321 } 1322 } 1323 1324 return -EPROTO; 1325 } 1326 1327 /* setup for ISM connection of client */ 1328 static int smc_connect_ism(struct smc_sock *smc, 1329 struct smc_clc_msg_accept_confirm *aclc, 1330 struct smc_init_info *ini) 1331 { 1332 u8 *eid = NULL; 1333 int rc = 0; 1334 1335 ini->is_smcd = true; 1336 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK; 1337 1338 if (aclc->hdr.version == SMC_V2) { 1339 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 = 1340 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1341 1342 rc = smc_v2_determine_accepted_chid(aclc_v2, ini); 1343 if (rc) 1344 return rc; 1345 } 1346 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid; 1347 1348 /* there is only one lgr role for SMC-D; use server lock */ 1349 mutex_lock(&smc_server_lgr_pending); 1350 rc = smc_conn_create(smc, ini); 1351 if (rc) { 1352 mutex_unlock(&smc_server_lgr_pending); 1353 return rc; 1354 } 1355 1356 /* Create send and receive buffers */ 1357 rc = smc_buf_create(smc, true); 1358 if (rc) { 1359 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM; 1360 goto connect_abort; 1361 } 1362 1363 smc_conn_save_peer_info(smc, aclc); 1364 smc_close_init(smc); 1365 smc_rx_init(smc); 1366 smc_tx_init(smc); 1367 1368 if (aclc->hdr.version > SMC_V1) { 1369 struct smc_clc_msg_accept_confirm_v2 *clc_v2 = 1370 (struct smc_clc_msg_accept_confirm_v2 *)aclc; 1371 1372 eid = clc_v2->d1.eid; 1373 } 1374 1375 rc = smc_clc_send_confirm(smc, ini->first_contact_local, 1376 aclc->hdr.version, eid, NULL); 1377 if (rc) 1378 goto connect_abort; 1379 mutex_unlock(&smc_server_lgr_pending); 1380 1381 smc_copy_sock_settings_to_clc(smc); 1382 smc->connect_nonblock = 0; 1383 if (smc->sk.sk_state == SMC_INIT) 1384 smc->sk.sk_state = SMC_ACTIVE; 1385 1386 return 0; 1387 connect_abort: 1388 smc_conn_abort(smc, ini->first_contact_local); 1389 mutex_unlock(&smc_server_lgr_pending); 1390 smc->connect_nonblock = 0; 1391 1392 return rc; 1393 } 1394 1395 /* check if received accept type and version matches a proposed one */ 1396 static int smc_connect_check_aclc(struct smc_init_info *ini, 1397 struct smc_clc_msg_accept_confirm *aclc) 1398 { 1399 if (aclc->hdr.typev1 != SMC_TYPE_R && 1400 aclc->hdr.typev1 != SMC_TYPE_D) 1401 return SMC_CLC_DECL_MODEUNSUPP; 1402 1403 if (aclc->hdr.version >= SMC_V2) { 1404 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1405 !smcr_indicated(ini->smc_type_v2)) || 1406 (aclc->hdr.typev1 == SMC_TYPE_D && 1407 !smcd_indicated(ini->smc_type_v2))) 1408 return SMC_CLC_DECL_MODEUNSUPP; 1409 } else { 1410 if ((aclc->hdr.typev1 == SMC_TYPE_R && 1411 !smcr_indicated(ini->smc_type_v1)) || 1412 (aclc->hdr.typev1 == SMC_TYPE_D && 1413 !smcd_indicated(ini->smc_type_v1))) 1414 return SMC_CLC_DECL_MODEUNSUPP; 1415 } 1416 1417 return 0; 1418 } 1419 1420 /* perform steps before actually connecting */ 1421 static int __smc_connect(struct smc_sock *smc) 1422 { 1423 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1; 1424 struct smc_clc_msg_accept_confirm_v2 *aclc2; 1425 struct smc_clc_msg_accept_confirm *aclc; 1426 struct smc_init_info *ini = NULL; 1427 u8 *buf = NULL; 1428 int rc = 0; 1429 1430 if (smc->use_fallback) 1431 return smc_connect_fallback(smc, smc->fallback_rsn); 1432 1433 /* if peer has not signalled SMC-capability, fall back */ 1434 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 1435 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 1436 1437 /* IPSec connections opt out of SMC optimizations */ 1438 if (using_ipsec(smc)) 1439 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC, 1440 version); 1441 1442 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 1443 if (!ini) 1444 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM, 1445 version); 1446 1447 ini->smcd_version = SMC_V1 | SMC_V2; 1448 ini->smcr_version = SMC_V1 | SMC_V2; 1449 ini->smc_type_v1 = SMC_TYPE_B; 1450 ini->smc_type_v2 = SMC_TYPE_B; 1451 1452 /* get vlan id from IP device */ 1453 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) { 1454 ini->smcd_version &= ~SMC_V1; 1455 ini->smcr_version = 0; 1456 ini->smc_type_v1 = SMC_TYPE_N; 1457 if (!ini->smcd_version) { 1458 rc = SMC_CLC_DECL_GETVLANERR; 1459 goto fallback; 1460 } 1461 } 1462 1463 rc = smc_find_proposal_devices(smc, ini); 1464 if (rc) 1465 goto fallback; 1466 1467 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL); 1468 if (!buf) { 1469 rc = SMC_CLC_DECL_MEM; 1470 goto fallback; 1471 } 1472 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf; 1473 aclc = (struct smc_clc_msg_accept_confirm *)aclc2; 1474 1475 /* perform CLC handshake */ 1476 rc = smc_connect_clc(smc, aclc2, ini); 1477 if (rc) { 1478 /* -EAGAIN on timeout, see tcp_recvmsg() */ 1479 if (rc == -EAGAIN) { 1480 rc = -ETIMEDOUT; 1481 smc->sk.sk_err = ETIMEDOUT; 1482 } 1483 goto vlan_cleanup; 1484 } 1485 1486 /* check if smc modes and versions of CLC proposal and accept match */ 1487 rc = smc_connect_check_aclc(ini, aclc); 1488 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2; 1489 if (rc) 1490 goto vlan_cleanup; 1491 1492 /* depending on previous steps, connect using rdma or ism */ 1493 if (aclc->hdr.typev1 == SMC_TYPE_R) { 1494 ini->smcr_version = version; 1495 rc = smc_connect_rdma(smc, aclc, ini); 1496 } else if (aclc->hdr.typev1 == SMC_TYPE_D) { 1497 ini->smcd_version = version; 1498 rc = smc_connect_ism(smc, aclc, ini); 1499 } 1500 if (rc) 1501 goto vlan_cleanup; 1502 1503 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc); 1504 smc_connect_ism_vlan_cleanup(smc, ini); 1505 kfree(buf); 1506 kfree(ini); 1507 return 0; 1508 1509 vlan_cleanup: 1510 smc_connect_ism_vlan_cleanup(smc, ini); 1511 kfree(buf); 1512 fallback: 1513 kfree(ini); 1514 return smc_connect_decline_fallback(smc, rc, version); 1515 } 1516 1517 static void smc_connect_work(struct work_struct *work) 1518 { 1519 struct smc_sock *smc = container_of(work, struct smc_sock, 1520 connect_work); 1521 long timeo = smc->sk.sk_sndtimeo; 1522 int rc = 0; 1523 1524 if (!timeo) 1525 timeo = MAX_SCHEDULE_TIMEOUT; 1526 lock_sock(smc->clcsock->sk); 1527 if (smc->clcsock->sk->sk_err) { 1528 smc->sk.sk_err = smc->clcsock->sk->sk_err; 1529 } else if ((1 << smc->clcsock->sk->sk_state) & 1530 (TCPF_SYN_SENT | TCPF_SYN_RECV)) { 1531 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 1532 if ((rc == -EPIPE) && 1533 ((1 << smc->clcsock->sk->sk_state) & 1534 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 1535 rc = 0; 1536 } 1537 release_sock(smc->clcsock->sk); 1538 lock_sock(&smc->sk); 1539 if (rc != 0 || smc->sk.sk_err) { 1540 smc->sk.sk_state = SMC_CLOSED; 1541 if (rc == -EPIPE || rc == -EAGAIN) 1542 smc->sk.sk_err = EPIPE; 1543 else if (rc == -ECONNREFUSED) 1544 smc->sk.sk_err = ECONNREFUSED; 1545 else if (signal_pending(current)) 1546 smc->sk.sk_err = -sock_intr_errno(timeo); 1547 sock_put(&smc->sk); /* passive closing */ 1548 goto out; 1549 } 1550 1551 rc = __smc_connect(smc); 1552 if (rc < 0) 1553 smc->sk.sk_err = -rc; 1554 1555 out: 1556 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 1557 if (smc->sk.sk_err) { 1558 smc->sk.sk_state_change(&smc->sk); 1559 } else { /* allow polling before and after fallback decision */ 1560 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 1561 smc->sk.sk_write_space(&smc->sk); 1562 } 1563 } 1564 release_sock(&smc->sk); 1565 } 1566 1567 static int smc_connect(struct socket *sock, struct sockaddr *addr, 1568 int alen, int flags) 1569 { 1570 struct sock *sk = sock->sk; 1571 struct smc_sock *smc; 1572 int rc = -EINVAL; 1573 1574 smc = smc_sk(sk); 1575 1576 /* separate smc parameter checking to be safe */ 1577 if (alen < sizeof(addr->sa_family)) 1578 goto out_err; 1579 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 1580 goto out_err; 1581 1582 lock_sock(sk); 1583 switch (sock->state) { 1584 default: 1585 rc = -EINVAL; 1586 goto out; 1587 case SS_CONNECTED: 1588 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL; 1589 goto out; 1590 case SS_CONNECTING: 1591 if (sk->sk_state == SMC_ACTIVE) 1592 goto connected; 1593 break; 1594 case SS_UNCONNECTED: 1595 sock->state = SS_CONNECTING; 1596 break; 1597 } 1598 1599 switch (sk->sk_state) { 1600 default: 1601 goto out; 1602 case SMC_CLOSED: 1603 rc = sock_error(sk) ? : -ECONNABORTED; 1604 sock->state = SS_UNCONNECTED; 1605 goto out; 1606 case SMC_ACTIVE: 1607 rc = -EISCONN; 1608 goto out; 1609 case SMC_INIT: 1610 break; 1611 } 1612 1613 smc_copy_sock_settings_to_clc(smc); 1614 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1615 if (smc->connect_nonblock) { 1616 rc = -EALREADY; 1617 goto out; 1618 } 1619 rc = kernel_connect(smc->clcsock, addr, alen, flags); 1620 if (rc && rc != -EINPROGRESS) 1621 goto out; 1622 1623 if (smc->use_fallback) { 1624 sock->state = rc ? SS_CONNECTING : SS_CONNECTED; 1625 goto out; 1626 } 1627 sock_hold(&smc->sk); /* sock put in passive closing */ 1628 if (flags & O_NONBLOCK) { 1629 if (queue_work(smc_hs_wq, &smc->connect_work)) 1630 smc->connect_nonblock = 1; 1631 rc = -EINPROGRESS; 1632 goto out; 1633 } else { 1634 rc = __smc_connect(smc); 1635 if (rc < 0) 1636 goto out; 1637 } 1638 1639 connected: 1640 rc = 0; 1641 sock->state = SS_CONNECTED; 1642 out: 1643 release_sock(sk); 1644 out_err: 1645 return rc; 1646 } 1647 1648 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 1649 { 1650 struct socket *new_clcsock = NULL; 1651 struct sock *lsk = &lsmc->sk; 1652 struct sock *new_sk; 1653 int rc = -EINVAL; 1654 1655 release_sock(lsk); 1656 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 1657 if (!new_sk) { 1658 rc = -ENOMEM; 1659 lsk->sk_err = ENOMEM; 1660 *new_smc = NULL; 1661 lock_sock(lsk); 1662 goto out; 1663 } 1664 *new_smc = smc_sk(new_sk); 1665 1666 mutex_lock(&lsmc->clcsock_release_lock); 1667 if (lsmc->clcsock) 1668 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK); 1669 mutex_unlock(&lsmc->clcsock_release_lock); 1670 lock_sock(lsk); 1671 if (rc < 0 && rc != -EAGAIN) 1672 lsk->sk_err = -rc; 1673 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 1674 new_sk->sk_prot->unhash(new_sk); 1675 if (new_clcsock) 1676 sock_release(new_clcsock); 1677 new_sk->sk_state = SMC_CLOSED; 1678 sock_set_flag(new_sk, SOCK_DEAD); 1679 sock_put(new_sk); /* final */ 1680 *new_smc = NULL; 1681 goto out; 1682 } 1683 1684 /* new clcsock has inherited the smc listen-specific sk_data_ready 1685 * function; switch it back to the original sk_data_ready function 1686 */ 1687 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready; 1688 1689 /* if new clcsock has also inherited the fallback-specific callback 1690 * functions, switch them back to the original ones. 1691 */ 1692 if (lsmc->use_fallback) { 1693 if (lsmc->clcsk_state_change) 1694 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change; 1695 if (lsmc->clcsk_write_space) 1696 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space; 1697 if (lsmc->clcsk_error_report) 1698 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report; 1699 } 1700 1701 (*new_smc)->clcsock = new_clcsock; 1702 out: 1703 return rc; 1704 } 1705 1706 /* add a just created sock to the accept queue of the listen sock as 1707 * candidate for a following socket accept call from user space 1708 */ 1709 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 1710 { 1711 struct smc_sock *par = smc_sk(parent); 1712 1713 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 1714 spin_lock(&par->accept_q_lock); 1715 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 1716 spin_unlock(&par->accept_q_lock); 1717 sk_acceptq_added(parent); 1718 } 1719 1720 /* remove a socket from the accept queue of its parental listening socket */ 1721 static void smc_accept_unlink(struct sock *sk) 1722 { 1723 struct smc_sock *par = smc_sk(sk)->listen_smc; 1724 1725 spin_lock(&par->accept_q_lock); 1726 list_del_init(&smc_sk(sk)->accept_q); 1727 spin_unlock(&par->accept_q_lock); 1728 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 1729 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 1730 } 1731 1732 /* remove a sock from the accept queue to bind it to a new socket created 1733 * for a socket accept call from user space 1734 */ 1735 struct sock *smc_accept_dequeue(struct sock *parent, 1736 struct socket *new_sock) 1737 { 1738 struct smc_sock *isk, *n; 1739 struct sock *new_sk; 1740 1741 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1742 new_sk = (struct sock *)isk; 1743 1744 smc_accept_unlink(new_sk); 1745 if (new_sk->sk_state == SMC_CLOSED) { 1746 new_sk->sk_prot->unhash(new_sk); 1747 if (isk->clcsock) { 1748 sock_release(isk->clcsock); 1749 isk->clcsock = NULL; 1750 } 1751 sock_put(new_sk); /* final */ 1752 continue; 1753 } 1754 if (new_sock) { 1755 sock_graft(new_sk, new_sock); 1756 new_sock->state = SS_CONNECTED; 1757 if (isk->use_fallback) { 1758 smc_sk(new_sk)->clcsock->file = new_sock->file; 1759 isk->clcsock->file->private_data = isk->clcsock; 1760 } 1761 } 1762 return new_sk; 1763 } 1764 return NULL; 1765 } 1766 1767 /* clean up for a created but never accepted sock */ 1768 void smc_close_non_accepted(struct sock *sk) 1769 { 1770 struct smc_sock *smc = smc_sk(sk); 1771 1772 sock_hold(sk); /* sock_put below */ 1773 lock_sock(sk); 1774 if (!sk->sk_lingertime) 1775 /* wait for peer closing */ 1776 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1777 __smc_release(smc); 1778 release_sock(sk); 1779 sock_put(sk); /* sock_hold above */ 1780 sock_put(sk); /* final sock_put */ 1781 } 1782 1783 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1784 { 1785 struct smc_link *link = smc->conn.lnk; 1786 struct smc_llc_qentry *qentry; 1787 int rc; 1788 1789 /* reg the sndbuf if it was vzalloced*/ 1790 if (smc->conn.sndbuf_desc->is_vm) { 1791 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc)) 1792 return SMC_CLC_DECL_ERR_REGBUF; 1793 } 1794 1795 /* reg the rmb */ 1796 if (smcr_link_reg_buf(link, smc->conn.rmb_desc)) 1797 return SMC_CLC_DECL_ERR_REGBUF; 1798 1799 /* send CONFIRM LINK request to client over the RoCE fabric */ 1800 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1801 if (rc < 0) 1802 return SMC_CLC_DECL_TIMEOUT_CL; 1803 1804 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1805 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1806 SMC_LLC_CONFIRM_LINK); 1807 if (!qentry) { 1808 struct smc_clc_msg_decline dclc; 1809 1810 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1811 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1812 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1813 } 1814 smc_llc_save_peer_uid(qentry); 1815 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1816 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1817 if (rc) 1818 return SMC_CLC_DECL_RMBE_EC; 1819 1820 /* confirm_rkey is implicit on 1st contact */ 1821 smc->conn.rmb_desc->is_conf_rkey = true; 1822 1823 smc_llc_link_active(link); 1824 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE); 1825 1826 /* initial contact - try to establish second link */ 1827 smc_llc_srv_add_link(link, NULL); 1828 return 0; 1829 } 1830 1831 /* listen worker: finish */ 1832 static void smc_listen_out(struct smc_sock *new_smc) 1833 { 1834 struct smc_sock *lsmc = new_smc->listen_smc; 1835 struct sock *newsmcsk = &new_smc->sk; 1836 1837 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 1838 atomic_dec(&lsmc->queued_smc_hs); 1839 1840 if (lsmc->sk.sk_state == SMC_LISTEN) { 1841 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1842 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1843 release_sock(&lsmc->sk); 1844 } else { /* no longer listening */ 1845 smc_close_non_accepted(newsmcsk); 1846 } 1847 1848 /* Wake up accept */ 1849 lsmc->sk.sk_data_ready(&lsmc->sk); 1850 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1851 } 1852 1853 /* listen worker: finish in state connected */ 1854 static void smc_listen_out_connected(struct smc_sock *new_smc) 1855 { 1856 struct sock *newsmcsk = &new_smc->sk; 1857 1858 sk_refcnt_debug_inc(newsmcsk); 1859 if (newsmcsk->sk_state == SMC_INIT) 1860 newsmcsk->sk_state = SMC_ACTIVE; 1861 1862 smc_listen_out(new_smc); 1863 } 1864 1865 /* listen worker: finish in error state */ 1866 static void smc_listen_out_err(struct smc_sock *new_smc) 1867 { 1868 struct sock *newsmcsk = &new_smc->sk; 1869 struct net *net = sock_net(newsmcsk); 1870 1871 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt); 1872 if (newsmcsk->sk_state == SMC_INIT) 1873 sock_put(&new_smc->sk); /* passive closing */ 1874 newsmcsk->sk_state = SMC_CLOSED; 1875 1876 smc_listen_out(new_smc); 1877 } 1878 1879 /* listen worker: decline and fall back if possible */ 1880 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1881 int local_first, u8 version) 1882 { 1883 /* RDMA setup failed, switch back to TCP */ 1884 smc_conn_abort(new_smc, local_first); 1885 if (reason_code < 0 || 1886 smc_switch_to_fallback(new_smc, reason_code)) { 1887 /* error, no fallback possible */ 1888 smc_listen_out_err(new_smc); 1889 return; 1890 } 1891 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1892 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) { 1893 smc_listen_out_err(new_smc); 1894 return; 1895 } 1896 } 1897 smc_listen_out_connected(new_smc); 1898 } 1899 1900 /* listen worker: version checking */ 1901 static int smc_listen_v2_check(struct smc_sock *new_smc, 1902 struct smc_clc_msg_proposal *pclc, 1903 struct smc_init_info *ini) 1904 { 1905 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext; 1906 struct smc_clc_v2_extension *pclc_v2_ext; 1907 int rc = SMC_CLC_DECL_PEERNOSMC; 1908 1909 ini->smc_type_v1 = pclc->hdr.typev1; 1910 ini->smc_type_v2 = pclc->hdr.typev2; 1911 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1912 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0; 1913 if (pclc->hdr.version > SMC_V1) { 1914 if (smcd_indicated(ini->smc_type_v2)) 1915 ini->smcd_version |= SMC_V2; 1916 if (smcr_indicated(ini->smc_type_v2)) 1917 ini->smcr_version |= SMC_V2; 1918 } 1919 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) { 1920 rc = SMC_CLC_DECL_PEERNOSMC; 1921 goto out; 1922 } 1923 pclc_v2_ext = smc_get_clc_v2_ext(pclc); 1924 if (!pclc_v2_ext) { 1925 ini->smcd_version &= ~SMC_V2; 1926 ini->smcr_version &= ~SMC_V2; 1927 rc = SMC_CLC_DECL_NOV2EXT; 1928 goto out; 1929 } 1930 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext); 1931 if (ini->smcd_version & SMC_V2) { 1932 if (!smc_ism_is_v2_capable()) { 1933 ini->smcd_version &= ~SMC_V2; 1934 rc = SMC_CLC_DECL_NOISM2SUPP; 1935 } else if (!pclc_smcd_v2_ext) { 1936 ini->smcd_version &= ~SMC_V2; 1937 rc = SMC_CLC_DECL_NOV2DEXT; 1938 } else if (!pclc_v2_ext->hdr.eid_cnt && 1939 !pclc_v2_ext->hdr.flag.seid) { 1940 ini->smcd_version &= ~SMC_V2; 1941 rc = SMC_CLC_DECL_NOUEID; 1942 } 1943 } 1944 if (ini->smcr_version & SMC_V2) { 1945 if (!pclc_v2_ext->hdr.eid_cnt) { 1946 ini->smcr_version &= ~SMC_V2; 1947 rc = SMC_CLC_DECL_NOUEID; 1948 } 1949 } 1950 1951 out: 1952 if (!ini->smcd_version && !ini->smcr_version) 1953 return rc; 1954 1955 return 0; 1956 } 1957 1958 /* listen worker: check prefixes */ 1959 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1960 struct smc_clc_msg_proposal *pclc) 1961 { 1962 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1963 struct socket *newclcsock = new_smc->clcsock; 1964 1965 if (pclc->hdr.typev1 == SMC_TYPE_N) 1966 return 0; 1967 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1968 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1969 return SMC_CLC_DECL_DIFFPREFIX; 1970 1971 return 0; 1972 } 1973 1974 /* listen worker: initialize connection and buffers */ 1975 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1976 struct smc_init_info *ini) 1977 { 1978 int rc; 1979 1980 /* allocate connection / link group */ 1981 rc = smc_conn_create(new_smc, ini); 1982 if (rc) 1983 return rc; 1984 1985 /* create send buffer and rmb */ 1986 if (smc_buf_create(new_smc, false)) 1987 return SMC_CLC_DECL_MEM; 1988 1989 return 0; 1990 } 1991 1992 /* listen worker: initialize connection and buffers for SMC-D */ 1993 static int smc_listen_ism_init(struct smc_sock *new_smc, 1994 struct smc_init_info *ini) 1995 { 1996 int rc; 1997 1998 rc = smc_conn_create(new_smc, ini); 1999 if (rc) 2000 return rc; 2001 2002 /* Create send and receive buffers */ 2003 rc = smc_buf_create(new_smc, true); 2004 if (rc) { 2005 smc_conn_abort(new_smc, ini->first_contact_local); 2006 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : 2007 SMC_CLC_DECL_MEM; 2008 } 2009 2010 return 0; 2011 } 2012 2013 static bool smc_is_already_selected(struct smcd_dev *smcd, 2014 struct smc_init_info *ini, 2015 int matches) 2016 { 2017 int i; 2018 2019 for (i = 0; i < matches; i++) 2020 if (smcd == ini->ism_dev[i]) 2021 return true; 2022 2023 return false; 2024 } 2025 2026 /* check for ISM devices matching proposed ISM devices */ 2027 static void smc_check_ism_v2_match(struct smc_init_info *ini, 2028 u16 proposed_chid, u64 proposed_gid, 2029 unsigned int *matches) 2030 { 2031 struct smcd_dev *smcd; 2032 2033 list_for_each_entry(smcd, &smcd_dev_list.list, list) { 2034 if (smcd->going_away) 2035 continue; 2036 if (smc_is_already_selected(smcd, ini, *matches)) 2037 continue; 2038 if (smc_ism_get_chid(smcd) == proposed_chid && 2039 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) { 2040 ini->ism_peer_gid[*matches] = proposed_gid; 2041 ini->ism_dev[*matches] = smcd; 2042 (*matches)++; 2043 break; 2044 } 2045 } 2046 } 2047 2048 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini) 2049 { 2050 if (!ini->rc) 2051 ini->rc = rc; 2052 } 2053 2054 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc, 2055 struct smc_clc_msg_proposal *pclc, 2056 struct smc_init_info *ini) 2057 { 2058 struct smc_clc_smcd_v2_extension *smcd_v2_ext; 2059 struct smc_clc_v2_extension *smc_v2_ext; 2060 struct smc_clc_msg_smcd *pclc_smcd; 2061 unsigned int matches = 0; 2062 u8 smcd_version; 2063 u8 *eid = NULL; 2064 int i, rc; 2065 2066 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2)) 2067 goto not_found; 2068 2069 pclc_smcd = smc_get_clc_msg_smcd(pclc); 2070 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2071 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext); 2072 2073 mutex_lock(&smcd_dev_list.mutex); 2074 if (pclc_smcd->ism.chid) 2075 /* check for ISM device matching proposed native ISM device */ 2076 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid), 2077 ntohll(pclc_smcd->ism.gid), &matches); 2078 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) { 2079 /* check for ISM devices matching proposed non-native ISM 2080 * devices 2081 */ 2082 smc_check_ism_v2_match(ini, 2083 ntohs(smcd_v2_ext->gidchid[i - 1].chid), 2084 ntohll(smcd_v2_ext->gidchid[i - 1].gid), 2085 &matches); 2086 } 2087 mutex_unlock(&smcd_dev_list.mutex); 2088 2089 if (!ini->ism_dev[0]) { 2090 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini); 2091 goto not_found; 2092 } 2093 2094 smc_ism_get_system_eid(&eid); 2095 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, 2096 smcd_v2_ext->system_eid, eid)) 2097 goto not_found; 2098 2099 /* separate - outside the smcd_dev_list.lock */ 2100 smcd_version = ini->smcd_version; 2101 for (i = 0; i < matches; i++) { 2102 ini->smcd_version = SMC_V2; 2103 ini->is_smcd = true; 2104 ini->ism_selected = i; 2105 rc = smc_listen_ism_init(new_smc, ini); 2106 if (rc) { 2107 smc_find_ism_store_rc(rc, ini); 2108 /* try next active ISM device */ 2109 continue; 2110 } 2111 return; /* matching and usable V2 ISM device found */ 2112 } 2113 /* no V2 ISM device could be initialized */ 2114 ini->smcd_version = smcd_version; /* restore original value */ 2115 ini->negotiated_eid[0] = 0; 2116 2117 not_found: 2118 ini->smcd_version &= ~SMC_V2; 2119 ini->ism_dev[0] = NULL; 2120 ini->is_smcd = false; 2121 } 2122 2123 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc, 2124 struct smc_clc_msg_proposal *pclc, 2125 struct smc_init_info *ini) 2126 { 2127 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc); 2128 int rc = 0; 2129 2130 /* check if ISM V1 is available */ 2131 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1)) 2132 goto not_found; 2133 ini->is_smcd = true; /* prepare ISM check */ 2134 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid); 2135 rc = smc_find_ism_device(new_smc, ini); 2136 if (rc) 2137 goto not_found; 2138 ini->ism_selected = 0; 2139 rc = smc_listen_ism_init(new_smc, ini); 2140 if (!rc) 2141 return; /* V1 ISM device found */ 2142 2143 not_found: 2144 smc_find_ism_store_rc(rc, ini); 2145 ini->smcd_version &= ~SMC_V1; 2146 ini->ism_dev[0] = NULL; 2147 ini->is_smcd = false; 2148 } 2149 2150 /* listen worker: register buffers */ 2151 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first) 2152 { 2153 struct smc_connection *conn = &new_smc->conn; 2154 2155 if (!local_first) { 2156 /* reg sendbufs if they were vzalloced */ 2157 if (conn->sndbuf_desc->is_vm) { 2158 if (smcr_lgr_reg_sndbufs(conn->lnk, 2159 conn->sndbuf_desc)) 2160 return SMC_CLC_DECL_ERR_REGBUF; 2161 } 2162 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc)) 2163 return SMC_CLC_DECL_ERR_REGBUF; 2164 } 2165 2166 return 0; 2167 } 2168 2169 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc, 2170 struct smc_clc_msg_proposal *pclc, 2171 struct smc_init_info *ini) 2172 { 2173 struct smc_clc_v2_extension *smc_v2_ext; 2174 u8 smcr_version; 2175 int rc; 2176 2177 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2)) 2178 goto not_found; 2179 2180 smc_v2_ext = smc_get_clc_v2_ext(pclc); 2181 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL)) 2182 goto not_found; 2183 2184 /* prepare RDMA check */ 2185 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2186 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE); 2187 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2188 ini->check_smcrv2 = true; 2189 ini->smcrv2.clc_sk = new_smc->clcsock->sk; 2190 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr; 2191 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce); 2192 rc = smc_find_rdma_device(new_smc, ini); 2193 if (rc) { 2194 smc_find_ism_store_rc(rc, ini); 2195 goto not_found; 2196 } 2197 if (!ini->smcrv2.uses_gateway) 2198 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN); 2199 2200 smcr_version = ini->smcr_version; 2201 ini->smcr_version = SMC_V2; 2202 rc = smc_listen_rdma_init(new_smc, ini); 2203 if (!rc) 2204 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2205 if (!rc) 2206 return; 2207 ini->smcr_version = smcr_version; 2208 smc_find_ism_store_rc(rc, ini); 2209 2210 not_found: 2211 ini->smcr_version &= ~SMC_V2; 2212 ini->smcrv2.ib_dev_v2 = NULL; 2213 ini->check_smcrv2 = false; 2214 } 2215 2216 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc, 2217 struct smc_clc_msg_proposal *pclc, 2218 struct smc_init_info *ini) 2219 { 2220 int rc; 2221 2222 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1)) 2223 return SMC_CLC_DECL_NOSMCDEV; 2224 2225 /* prepare RDMA check */ 2226 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN); 2227 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE); 2228 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN); 2229 rc = smc_find_rdma_device(new_smc, ini); 2230 if (rc) { 2231 /* no RDMA device found */ 2232 return SMC_CLC_DECL_NOSMCDEV; 2233 } 2234 rc = smc_listen_rdma_init(new_smc, ini); 2235 if (rc) 2236 return rc; 2237 return smc_listen_rdma_reg(new_smc, ini->first_contact_local); 2238 } 2239 2240 /* determine the local device matching to proposal */ 2241 static int smc_listen_find_device(struct smc_sock *new_smc, 2242 struct smc_clc_msg_proposal *pclc, 2243 struct smc_init_info *ini) 2244 { 2245 int prfx_rc; 2246 2247 /* check for ISM device matching V2 proposed device */ 2248 smc_find_ism_v2_device_serv(new_smc, pclc, ini); 2249 if (ini->ism_dev[0]) 2250 return 0; 2251 2252 /* check for matching IP prefix and subnet length (V1) */ 2253 prfx_rc = smc_listen_prfx_check(new_smc, pclc); 2254 if (prfx_rc) 2255 smc_find_ism_store_rc(prfx_rc, ini); 2256 2257 /* get vlan id from IP device */ 2258 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini)) 2259 return ini->rc ?: SMC_CLC_DECL_GETVLANERR; 2260 2261 /* check for ISM device matching V1 proposed device */ 2262 if (!prfx_rc) 2263 smc_find_ism_v1_device_serv(new_smc, pclc, ini); 2264 if (ini->ism_dev[0]) 2265 return 0; 2266 2267 if (!smcr_indicated(pclc->hdr.typev1) && 2268 !smcr_indicated(pclc->hdr.typev2)) 2269 /* skip RDMA and decline */ 2270 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV; 2271 2272 /* check if RDMA V2 is available */ 2273 smc_find_rdma_v2_device_serv(new_smc, pclc, ini); 2274 if (ini->smcrv2.ib_dev_v2) 2275 return 0; 2276 2277 /* check if RDMA V1 is available */ 2278 if (!prfx_rc) { 2279 int rc; 2280 2281 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini); 2282 smc_find_ism_store_rc(rc, ini); 2283 return (!rc) ? 0 : ini->rc; 2284 } 2285 return SMC_CLC_DECL_NOSMCDEV; 2286 } 2287 2288 /* listen worker: finish RDMA setup */ 2289 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 2290 struct smc_clc_msg_accept_confirm *cclc, 2291 bool local_first, 2292 struct smc_init_info *ini) 2293 { 2294 struct smc_link *link = new_smc->conn.lnk; 2295 int reason_code = 0; 2296 2297 if (local_first) 2298 smc_link_save_peer_info(link, cclc, ini); 2299 2300 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) 2301 return SMC_CLC_DECL_ERR_RTOK; 2302 2303 if (local_first) { 2304 if (smc_ib_ready_link(link)) 2305 return SMC_CLC_DECL_ERR_RDYLNK; 2306 /* QP confirmation over RoCE fabric */ 2307 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 2308 reason_code = smcr_serv_conf_first_link(new_smc); 2309 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 2310 } 2311 return reason_code; 2312 } 2313 2314 /* setup for connection of server */ 2315 static void smc_listen_work(struct work_struct *work) 2316 { 2317 struct smc_sock *new_smc = container_of(work, struct smc_sock, 2318 smc_listen_work); 2319 struct socket *newclcsock = new_smc->clcsock; 2320 struct smc_clc_msg_accept_confirm *cclc; 2321 struct smc_clc_msg_proposal_area *buf; 2322 struct smc_clc_msg_proposal *pclc; 2323 struct smc_init_info *ini = NULL; 2324 u8 proposal_version = SMC_V1; 2325 u8 accept_version; 2326 int rc = 0; 2327 2328 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 2329 return smc_listen_out_err(new_smc); 2330 2331 if (new_smc->use_fallback) { 2332 smc_listen_out_connected(new_smc); 2333 return; 2334 } 2335 2336 /* check if peer is smc capable */ 2337 if (!tcp_sk(newclcsock->sk)->syn_smc) { 2338 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC); 2339 if (rc) 2340 smc_listen_out_err(new_smc); 2341 else 2342 smc_listen_out_connected(new_smc); 2343 return; 2344 } 2345 2346 /* do inband token exchange - 2347 * wait for and receive SMC Proposal CLC message 2348 */ 2349 buf = kzalloc(sizeof(*buf), GFP_KERNEL); 2350 if (!buf) { 2351 rc = SMC_CLC_DECL_MEM; 2352 goto out_decl; 2353 } 2354 pclc = (struct smc_clc_msg_proposal *)buf; 2355 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf), 2356 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 2357 if (rc) 2358 goto out_decl; 2359 2360 if (pclc->hdr.version > SMC_V1) 2361 proposal_version = SMC_V2; 2362 2363 /* IPSec connections opt out of SMC optimizations */ 2364 if (using_ipsec(new_smc)) { 2365 rc = SMC_CLC_DECL_IPSEC; 2366 goto out_decl; 2367 } 2368 2369 ini = kzalloc(sizeof(*ini), GFP_KERNEL); 2370 if (!ini) { 2371 rc = SMC_CLC_DECL_MEM; 2372 goto out_decl; 2373 } 2374 2375 /* initial version checking */ 2376 rc = smc_listen_v2_check(new_smc, pclc, ini); 2377 if (rc) 2378 goto out_decl; 2379 2380 mutex_lock(&smc_server_lgr_pending); 2381 smc_close_init(new_smc); 2382 smc_rx_init(new_smc); 2383 smc_tx_init(new_smc); 2384 2385 /* determine ISM or RoCE device used for connection */ 2386 rc = smc_listen_find_device(new_smc, pclc, ini); 2387 if (rc) 2388 goto out_unlock; 2389 2390 /* send SMC Accept CLC message */ 2391 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version; 2392 rc = smc_clc_send_accept(new_smc, ini->first_contact_local, 2393 accept_version, ini->negotiated_eid); 2394 if (rc) 2395 goto out_unlock; 2396 2397 /* SMC-D does not need this lock any more */ 2398 if (ini->is_smcd) 2399 mutex_unlock(&smc_server_lgr_pending); 2400 2401 /* receive SMC Confirm CLC message */ 2402 memset(buf, 0, sizeof(*buf)); 2403 cclc = (struct smc_clc_msg_accept_confirm *)buf; 2404 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf), 2405 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 2406 if (rc) { 2407 if (!ini->is_smcd) 2408 goto out_unlock; 2409 goto out_decl; 2410 } 2411 2412 /* finish worker */ 2413 if (!ini->is_smcd) { 2414 rc = smc_listen_rdma_finish(new_smc, cclc, 2415 ini->first_contact_local, ini); 2416 if (rc) 2417 goto out_unlock; 2418 mutex_unlock(&smc_server_lgr_pending); 2419 } 2420 smc_conn_save_peer_info(new_smc, cclc); 2421 smc_listen_out_connected(new_smc); 2422 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini); 2423 goto out_free; 2424 2425 out_unlock: 2426 mutex_unlock(&smc_server_lgr_pending); 2427 out_decl: 2428 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0, 2429 proposal_version); 2430 out_free: 2431 kfree(ini); 2432 kfree(buf); 2433 } 2434 2435 static void smc_tcp_listen_work(struct work_struct *work) 2436 { 2437 struct smc_sock *lsmc = container_of(work, struct smc_sock, 2438 tcp_listen_work); 2439 struct sock *lsk = &lsmc->sk; 2440 struct smc_sock *new_smc; 2441 int rc = 0; 2442 2443 lock_sock(lsk); 2444 while (lsk->sk_state == SMC_LISTEN) { 2445 rc = smc_clcsock_accept(lsmc, &new_smc); 2446 if (rc) /* clcsock accept queue empty or error */ 2447 goto out; 2448 if (!new_smc) 2449 continue; 2450 2451 if (tcp_sk(new_smc->clcsock->sk)->syn_smc) 2452 atomic_inc(&lsmc->queued_smc_hs); 2453 2454 new_smc->listen_smc = lsmc; 2455 new_smc->use_fallback = lsmc->use_fallback; 2456 new_smc->fallback_rsn = lsmc->fallback_rsn; 2457 sock_hold(lsk); /* sock_put in smc_listen_work */ 2458 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 2459 smc_copy_sock_settings_to_smc(new_smc); 2460 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 2461 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 2462 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 2463 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work)) 2464 sock_put(&new_smc->sk); 2465 } 2466 2467 out: 2468 release_sock(lsk); 2469 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */ 2470 } 2471 2472 static void smc_clcsock_data_ready(struct sock *listen_clcsock) 2473 { 2474 struct smc_sock *lsmc; 2475 2476 read_lock_bh(&listen_clcsock->sk_callback_lock); 2477 lsmc = smc_clcsock_user_data(listen_clcsock); 2478 if (!lsmc) 2479 goto out; 2480 lsmc->clcsk_data_ready(listen_clcsock); 2481 if (lsmc->sk.sk_state == SMC_LISTEN) { 2482 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */ 2483 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work)) 2484 sock_put(&lsmc->sk); 2485 } 2486 out: 2487 read_unlock_bh(&listen_clcsock->sk_callback_lock); 2488 } 2489 2490 static int smc_listen(struct socket *sock, int backlog) 2491 { 2492 struct sock *sk = sock->sk; 2493 struct smc_sock *smc; 2494 int rc; 2495 2496 smc = smc_sk(sk); 2497 lock_sock(sk); 2498 2499 rc = -EINVAL; 2500 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 2501 smc->connect_nonblock || sock->state != SS_UNCONNECTED) 2502 goto out; 2503 2504 rc = 0; 2505 if (sk->sk_state == SMC_LISTEN) { 2506 sk->sk_max_ack_backlog = backlog; 2507 goto out; 2508 } 2509 /* some socket options are handled in core, so we could not apply 2510 * them to the clc socket -- copy smc socket options to clc socket 2511 */ 2512 smc_copy_sock_settings_to_clc(smc); 2513 if (!smc->use_fallback) 2514 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 2515 2516 /* save original sk_data_ready function and establish 2517 * smc-specific sk_data_ready function 2518 */ 2519 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2520 smc->clcsock->sk->sk_user_data = 2521 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY); 2522 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready, 2523 smc_clcsock_data_ready, &smc->clcsk_data_ready); 2524 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2525 2526 /* save original ops */ 2527 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops; 2528 2529 smc->af_ops = *smc->ori_af_ops; 2530 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock; 2531 2532 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops; 2533 2534 if (smc->limit_smc_hs) 2535 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested; 2536 2537 rc = kernel_listen(smc->clcsock, backlog); 2538 if (rc) { 2539 write_lock_bh(&smc->clcsock->sk->sk_callback_lock); 2540 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready, 2541 &smc->clcsk_data_ready); 2542 smc->clcsock->sk->sk_user_data = NULL; 2543 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock); 2544 goto out; 2545 } 2546 sk->sk_max_ack_backlog = backlog; 2547 sk->sk_ack_backlog = 0; 2548 sk->sk_state = SMC_LISTEN; 2549 2550 out: 2551 release_sock(sk); 2552 return rc; 2553 } 2554 2555 static int smc_accept(struct socket *sock, struct socket *new_sock, 2556 int flags, bool kern) 2557 { 2558 struct sock *sk = sock->sk, *nsk; 2559 DECLARE_WAITQUEUE(wait, current); 2560 struct smc_sock *lsmc; 2561 long timeo; 2562 int rc = 0; 2563 2564 lsmc = smc_sk(sk); 2565 sock_hold(sk); /* sock_put below */ 2566 lock_sock(sk); 2567 2568 if (lsmc->sk.sk_state != SMC_LISTEN) { 2569 rc = -EINVAL; 2570 release_sock(sk); 2571 goto out; 2572 } 2573 2574 /* Wait for an incoming connection */ 2575 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 2576 add_wait_queue_exclusive(sk_sleep(sk), &wait); 2577 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 2578 set_current_state(TASK_INTERRUPTIBLE); 2579 if (!timeo) { 2580 rc = -EAGAIN; 2581 break; 2582 } 2583 release_sock(sk); 2584 timeo = schedule_timeout(timeo); 2585 /* wakeup by sk_data_ready in smc_listen_work() */ 2586 sched_annotate_sleep(); 2587 lock_sock(sk); 2588 if (signal_pending(current)) { 2589 rc = sock_intr_errno(timeo); 2590 break; 2591 } 2592 } 2593 set_current_state(TASK_RUNNING); 2594 remove_wait_queue(sk_sleep(sk), &wait); 2595 2596 if (!rc) 2597 rc = sock_error(nsk); 2598 release_sock(sk); 2599 if (rc) 2600 goto out; 2601 2602 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 2603 /* wait till data arrives on the socket */ 2604 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 2605 MSEC_PER_SEC); 2606 if (smc_sk(nsk)->use_fallback) { 2607 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 2608 2609 lock_sock(clcsk); 2610 if (skb_queue_empty(&clcsk->sk_receive_queue)) 2611 sk_wait_data(clcsk, &timeo, NULL); 2612 release_sock(clcsk); 2613 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 2614 lock_sock(nsk); 2615 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 2616 release_sock(nsk); 2617 } 2618 } 2619 2620 out: 2621 sock_put(sk); /* sock_hold above */ 2622 return rc; 2623 } 2624 2625 static int smc_getname(struct socket *sock, struct sockaddr *addr, 2626 int peer) 2627 { 2628 struct smc_sock *smc; 2629 2630 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 2631 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 2632 return -ENOTCONN; 2633 2634 smc = smc_sk(sock->sk); 2635 2636 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 2637 } 2638 2639 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 2640 { 2641 struct sock *sk = sock->sk; 2642 struct smc_sock *smc; 2643 int rc = -EPIPE; 2644 2645 smc = smc_sk(sk); 2646 lock_sock(sk); 2647 if ((sk->sk_state != SMC_ACTIVE) && 2648 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2649 (sk->sk_state != SMC_INIT)) 2650 goto out; 2651 2652 if (msg->msg_flags & MSG_FASTOPEN) { 2653 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2654 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2655 if (rc) 2656 goto out; 2657 } else { 2658 rc = -EINVAL; 2659 goto out; 2660 } 2661 } 2662 2663 if (smc->use_fallback) { 2664 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 2665 } else { 2666 rc = smc_tx_sendmsg(smc, msg, len); 2667 SMC_STAT_TX_PAYLOAD(smc, len, rc); 2668 } 2669 out: 2670 release_sock(sk); 2671 return rc; 2672 } 2673 2674 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 2675 int flags) 2676 { 2677 struct sock *sk = sock->sk; 2678 struct smc_sock *smc; 2679 int rc = -ENOTCONN; 2680 2681 smc = smc_sk(sk); 2682 lock_sock(sk); 2683 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 2684 /* socket was connected before, no more data to read */ 2685 rc = 0; 2686 goto out; 2687 } 2688 if ((sk->sk_state == SMC_INIT) || 2689 (sk->sk_state == SMC_LISTEN) || 2690 (sk->sk_state == SMC_CLOSED)) 2691 goto out; 2692 2693 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 2694 rc = 0; 2695 goto out; 2696 } 2697 2698 if (smc->use_fallback) { 2699 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 2700 } else { 2701 msg->msg_namelen = 0; 2702 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 2703 SMC_STAT_RX_PAYLOAD(smc, rc, rc); 2704 } 2705 2706 out: 2707 release_sock(sk); 2708 return rc; 2709 } 2710 2711 static __poll_t smc_accept_poll(struct sock *parent) 2712 { 2713 struct smc_sock *isk = smc_sk(parent); 2714 __poll_t mask = 0; 2715 2716 spin_lock(&isk->accept_q_lock); 2717 if (!list_empty(&isk->accept_q)) 2718 mask = EPOLLIN | EPOLLRDNORM; 2719 spin_unlock(&isk->accept_q_lock); 2720 2721 return mask; 2722 } 2723 2724 static __poll_t smc_poll(struct file *file, struct socket *sock, 2725 poll_table *wait) 2726 { 2727 struct sock *sk = sock->sk; 2728 struct smc_sock *smc; 2729 __poll_t mask = 0; 2730 2731 if (!sk) 2732 return EPOLLNVAL; 2733 2734 smc = smc_sk(sock->sk); 2735 if (smc->use_fallback) { 2736 /* delegate to CLC child sock */ 2737 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 2738 sk->sk_err = smc->clcsock->sk->sk_err; 2739 } else { 2740 if (sk->sk_state != SMC_CLOSED) 2741 sock_poll_wait(file, sock, wait); 2742 if (sk->sk_err) 2743 mask |= EPOLLERR; 2744 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 2745 (sk->sk_state == SMC_CLOSED)) 2746 mask |= EPOLLHUP; 2747 if (sk->sk_state == SMC_LISTEN) { 2748 /* woken up by sk_data_ready in smc_listen_work() */ 2749 mask |= smc_accept_poll(sk); 2750 } else if (smc->use_fallback) { /* as result of connect_work()*/ 2751 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 2752 wait); 2753 sk->sk_err = smc->clcsock->sk->sk_err; 2754 } else { 2755 if ((sk->sk_state != SMC_INIT && 2756 atomic_read(&smc->conn.sndbuf_space)) || 2757 sk->sk_shutdown & SEND_SHUTDOWN) { 2758 mask |= EPOLLOUT | EPOLLWRNORM; 2759 } else { 2760 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 2761 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 2762 } 2763 if (atomic_read(&smc->conn.bytes_to_rcv)) 2764 mask |= EPOLLIN | EPOLLRDNORM; 2765 if (sk->sk_shutdown & RCV_SHUTDOWN) 2766 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 2767 if (sk->sk_state == SMC_APPCLOSEWAIT1) 2768 mask |= EPOLLIN; 2769 if (smc->conn.urg_state == SMC_URG_VALID) 2770 mask |= EPOLLPRI; 2771 } 2772 } 2773 2774 return mask; 2775 } 2776 2777 static int smc_shutdown(struct socket *sock, int how) 2778 { 2779 struct sock *sk = sock->sk; 2780 bool do_shutdown = true; 2781 struct smc_sock *smc; 2782 int rc = -EINVAL; 2783 int old_state; 2784 int rc1 = 0; 2785 2786 smc = smc_sk(sk); 2787 2788 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 2789 return rc; 2790 2791 lock_sock(sk); 2792 2793 if (sock->state == SS_CONNECTING) { 2794 if (sk->sk_state == SMC_ACTIVE) 2795 sock->state = SS_CONNECTED; 2796 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 || 2797 sk->sk_state == SMC_PEERCLOSEWAIT2 || 2798 sk->sk_state == SMC_APPCLOSEWAIT1 || 2799 sk->sk_state == SMC_APPCLOSEWAIT2 || 2800 sk->sk_state == SMC_APPFINCLOSEWAIT) 2801 sock->state = SS_DISCONNECTING; 2802 } 2803 2804 rc = -ENOTCONN; 2805 if ((sk->sk_state != SMC_ACTIVE) && 2806 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 2807 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 2808 (sk->sk_state != SMC_APPCLOSEWAIT1) && 2809 (sk->sk_state != SMC_APPCLOSEWAIT2) && 2810 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 2811 goto out; 2812 if (smc->use_fallback) { 2813 rc = kernel_sock_shutdown(smc->clcsock, how); 2814 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 2815 if (sk->sk_shutdown == SHUTDOWN_MASK) { 2816 sk->sk_state = SMC_CLOSED; 2817 sk->sk_socket->state = SS_UNCONNECTED; 2818 sock_put(sk); 2819 } 2820 goto out; 2821 } 2822 switch (how) { 2823 case SHUT_RDWR: /* shutdown in both directions */ 2824 old_state = sk->sk_state; 2825 rc = smc_close_active(smc); 2826 if (old_state == SMC_ACTIVE && 2827 sk->sk_state == SMC_PEERCLOSEWAIT1) 2828 do_shutdown = false; 2829 break; 2830 case SHUT_WR: 2831 rc = smc_close_shutdown_write(smc); 2832 break; 2833 case SHUT_RD: 2834 rc = 0; 2835 /* nothing more to do because peer is not involved */ 2836 break; 2837 } 2838 if (do_shutdown && smc->clcsock) 2839 rc1 = kernel_sock_shutdown(smc->clcsock, how); 2840 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 2841 sk->sk_shutdown |= how + 1; 2842 2843 if (sk->sk_state == SMC_CLOSED) 2844 sock->state = SS_UNCONNECTED; 2845 else 2846 sock->state = SS_DISCONNECTING; 2847 out: 2848 release_sock(sk); 2849 return rc ? rc : rc1; 2850 } 2851 2852 static int __smc_getsockopt(struct socket *sock, int level, int optname, 2853 char __user *optval, int __user *optlen) 2854 { 2855 struct smc_sock *smc; 2856 int val, len; 2857 2858 smc = smc_sk(sock->sk); 2859 2860 if (get_user(len, optlen)) 2861 return -EFAULT; 2862 2863 len = min_t(int, len, sizeof(int)); 2864 2865 if (len < 0) 2866 return -EINVAL; 2867 2868 switch (optname) { 2869 case SMC_LIMIT_HS: 2870 val = smc->limit_smc_hs; 2871 break; 2872 default: 2873 return -EOPNOTSUPP; 2874 } 2875 2876 if (put_user(len, optlen)) 2877 return -EFAULT; 2878 if (copy_to_user(optval, &val, len)) 2879 return -EFAULT; 2880 2881 return 0; 2882 } 2883 2884 static int __smc_setsockopt(struct socket *sock, int level, int optname, 2885 sockptr_t optval, unsigned int optlen) 2886 { 2887 struct sock *sk = sock->sk; 2888 struct smc_sock *smc; 2889 int val, rc; 2890 2891 smc = smc_sk(sk); 2892 2893 lock_sock(sk); 2894 switch (optname) { 2895 case SMC_LIMIT_HS: 2896 if (optlen < sizeof(int)) { 2897 rc = -EINVAL; 2898 break; 2899 } 2900 if (copy_from_sockptr(&val, optval, sizeof(int))) { 2901 rc = -EFAULT; 2902 break; 2903 } 2904 2905 smc->limit_smc_hs = !!val; 2906 rc = 0; 2907 break; 2908 default: 2909 rc = -EOPNOTSUPP; 2910 break; 2911 } 2912 release_sock(sk); 2913 2914 return rc; 2915 } 2916 2917 static int smc_setsockopt(struct socket *sock, int level, int optname, 2918 sockptr_t optval, unsigned int optlen) 2919 { 2920 struct sock *sk = sock->sk; 2921 struct smc_sock *smc; 2922 int val, rc; 2923 2924 if (level == SOL_TCP && optname == TCP_ULP) 2925 return -EOPNOTSUPP; 2926 else if (level == SOL_SMC) 2927 return __smc_setsockopt(sock, level, optname, optval, optlen); 2928 2929 smc = smc_sk(sk); 2930 2931 /* generic setsockopts reaching us here always apply to the 2932 * CLC socket 2933 */ 2934 mutex_lock(&smc->clcsock_release_lock); 2935 if (!smc->clcsock) { 2936 mutex_unlock(&smc->clcsock_release_lock); 2937 return -EBADF; 2938 } 2939 if (unlikely(!smc->clcsock->ops->setsockopt)) 2940 rc = -EOPNOTSUPP; 2941 else 2942 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 2943 optval, optlen); 2944 if (smc->clcsock->sk->sk_err) { 2945 sk->sk_err = smc->clcsock->sk->sk_err; 2946 sk_error_report(sk); 2947 } 2948 mutex_unlock(&smc->clcsock_release_lock); 2949 2950 if (optlen < sizeof(int)) 2951 return -EINVAL; 2952 if (copy_from_sockptr(&val, optval, sizeof(int))) 2953 return -EFAULT; 2954 2955 lock_sock(sk); 2956 if (rc || smc->use_fallback) 2957 goto out; 2958 switch (optname) { 2959 case TCP_FASTOPEN: 2960 case TCP_FASTOPEN_CONNECT: 2961 case TCP_FASTOPEN_KEY: 2962 case TCP_FASTOPEN_NO_COOKIE: 2963 /* option not supported by SMC */ 2964 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 2965 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP); 2966 } else { 2967 rc = -EINVAL; 2968 } 2969 break; 2970 case TCP_NODELAY: 2971 if (sk->sk_state != SMC_INIT && 2972 sk->sk_state != SMC_LISTEN && 2973 sk->sk_state != SMC_CLOSED) { 2974 if (val) { 2975 SMC_STAT_INC(smc, ndly_cnt); 2976 smc_tx_pending(&smc->conn); 2977 cancel_delayed_work(&smc->conn.tx_work); 2978 } 2979 } 2980 break; 2981 case TCP_CORK: 2982 if (sk->sk_state != SMC_INIT && 2983 sk->sk_state != SMC_LISTEN && 2984 sk->sk_state != SMC_CLOSED) { 2985 if (!val) { 2986 SMC_STAT_INC(smc, cork_cnt); 2987 smc_tx_pending(&smc->conn); 2988 cancel_delayed_work(&smc->conn.tx_work); 2989 } 2990 } 2991 break; 2992 case TCP_DEFER_ACCEPT: 2993 smc->sockopt_defer_accept = val; 2994 break; 2995 default: 2996 break; 2997 } 2998 out: 2999 release_sock(sk); 3000 3001 return rc; 3002 } 3003 3004 static int smc_getsockopt(struct socket *sock, int level, int optname, 3005 char __user *optval, int __user *optlen) 3006 { 3007 struct smc_sock *smc; 3008 int rc; 3009 3010 if (level == SOL_SMC) 3011 return __smc_getsockopt(sock, level, optname, optval, optlen); 3012 3013 smc = smc_sk(sock->sk); 3014 mutex_lock(&smc->clcsock_release_lock); 3015 if (!smc->clcsock) { 3016 mutex_unlock(&smc->clcsock_release_lock); 3017 return -EBADF; 3018 } 3019 /* socket options apply to the CLC socket */ 3020 if (unlikely(!smc->clcsock->ops->getsockopt)) { 3021 mutex_unlock(&smc->clcsock_release_lock); 3022 return -EOPNOTSUPP; 3023 } 3024 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 3025 optval, optlen); 3026 mutex_unlock(&smc->clcsock_release_lock); 3027 return rc; 3028 } 3029 3030 static int smc_ioctl(struct socket *sock, unsigned int cmd, 3031 unsigned long arg) 3032 { 3033 union smc_host_cursor cons, urg; 3034 struct smc_connection *conn; 3035 struct smc_sock *smc; 3036 int answ; 3037 3038 smc = smc_sk(sock->sk); 3039 conn = &smc->conn; 3040 lock_sock(&smc->sk); 3041 if (smc->use_fallback) { 3042 if (!smc->clcsock) { 3043 release_sock(&smc->sk); 3044 return -EBADF; 3045 } 3046 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 3047 release_sock(&smc->sk); 3048 return answ; 3049 } 3050 switch (cmd) { 3051 case SIOCINQ: /* same as FIONREAD */ 3052 if (smc->sk.sk_state == SMC_LISTEN) { 3053 release_sock(&smc->sk); 3054 return -EINVAL; 3055 } 3056 if (smc->sk.sk_state == SMC_INIT || 3057 smc->sk.sk_state == SMC_CLOSED) 3058 answ = 0; 3059 else 3060 answ = atomic_read(&smc->conn.bytes_to_rcv); 3061 break; 3062 case SIOCOUTQ: 3063 /* output queue size (not send + not acked) */ 3064 if (smc->sk.sk_state == SMC_LISTEN) { 3065 release_sock(&smc->sk); 3066 return -EINVAL; 3067 } 3068 if (smc->sk.sk_state == SMC_INIT || 3069 smc->sk.sk_state == SMC_CLOSED) 3070 answ = 0; 3071 else 3072 answ = smc->conn.sndbuf_desc->len - 3073 atomic_read(&smc->conn.sndbuf_space); 3074 break; 3075 case SIOCOUTQNSD: 3076 /* output queue size (not send only) */ 3077 if (smc->sk.sk_state == SMC_LISTEN) { 3078 release_sock(&smc->sk); 3079 return -EINVAL; 3080 } 3081 if (smc->sk.sk_state == SMC_INIT || 3082 smc->sk.sk_state == SMC_CLOSED) 3083 answ = 0; 3084 else 3085 answ = smc_tx_prepared_sends(&smc->conn); 3086 break; 3087 case SIOCATMARK: 3088 if (smc->sk.sk_state == SMC_LISTEN) { 3089 release_sock(&smc->sk); 3090 return -EINVAL; 3091 } 3092 if (smc->sk.sk_state == SMC_INIT || 3093 smc->sk.sk_state == SMC_CLOSED) { 3094 answ = 0; 3095 } else { 3096 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 3097 smc_curs_copy(&urg, &conn->urg_curs, conn); 3098 answ = smc_curs_diff(conn->rmb_desc->len, 3099 &cons, &urg) == 1; 3100 } 3101 break; 3102 default: 3103 release_sock(&smc->sk); 3104 return -ENOIOCTLCMD; 3105 } 3106 release_sock(&smc->sk); 3107 3108 return put_user(answ, (int __user *)arg); 3109 } 3110 3111 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 3112 int offset, size_t size, int flags) 3113 { 3114 struct sock *sk = sock->sk; 3115 struct smc_sock *smc; 3116 int rc = -EPIPE; 3117 3118 smc = smc_sk(sk); 3119 lock_sock(sk); 3120 if (sk->sk_state != SMC_ACTIVE) { 3121 release_sock(sk); 3122 goto out; 3123 } 3124 release_sock(sk); 3125 if (smc->use_fallback) { 3126 rc = kernel_sendpage(smc->clcsock, page, offset, 3127 size, flags); 3128 } else { 3129 lock_sock(sk); 3130 rc = smc_tx_sendpage(smc, page, offset, size, flags); 3131 release_sock(sk); 3132 SMC_STAT_INC(smc, sendpage_cnt); 3133 } 3134 3135 out: 3136 return rc; 3137 } 3138 3139 /* Map the affected portions of the rmbe into an spd, note the number of bytes 3140 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 3141 * updates till whenever a respective page has been fully processed. 3142 * Note that subsequent recv() calls have to wait till all splice() processing 3143 * completed. 3144 */ 3145 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 3146 struct pipe_inode_info *pipe, size_t len, 3147 unsigned int flags) 3148 { 3149 struct sock *sk = sock->sk; 3150 struct smc_sock *smc; 3151 int rc = -ENOTCONN; 3152 3153 smc = smc_sk(sk); 3154 lock_sock(sk); 3155 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 3156 /* socket was connected before, no more data to read */ 3157 rc = 0; 3158 goto out; 3159 } 3160 if (sk->sk_state == SMC_INIT || 3161 sk->sk_state == SMC_LISTEN || 3162 sk->sk_state == SMC_CLOSED) 3163 goto out; 3164 3165 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 3166 rc = 0; 3167 goto out; 3168 } 3169 3170 if (smc->use_fallback) { 3171 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 3172 pipe, len, flags); 3173 } else { 3174 if (*ppos) { 3175 rc = -ESPIPE; 3176 goto out; 3177 } 3178 if (flags & SPLICE_F_NONBLOCK) 3179 flags = MSG_DONTWAIT; 3180 else 3181 flags = 0; 3182 SMC_STAT_INC(smc, splice_cnt); 3183 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 3184 } 3185 out: 3186 release_sock(sk); 3187 3188 return rc; 3189 } 3190 3191 /* must look like tcp */ 3192 static const struct proto_ops smc_sock_ops = { 3193 .family = PF_SMC, 3194 .owner = THIS_MODULE, 3195 .release = smc_release, 3196 .bind = smc_bind, 3197 .connect = smc_connect, 3198 .socketpair = sock_no_socketpair, 3199 .accept = smc_accept, 3200 .getname = smc_getname, 3201 .poll = smc_poll, 3202 .ioctl = smc_ioctl, 3203 .listen = smc_listen, 3204 .shutdown = smc_shutdown, 3205 .setsockopt = smc_setsockopt, 3206 .getsockopt = smc_getsockopt, 3207 .sendmsg = smc_sendmsg, 3208 .recvmsg = smc_recvmsg, 3209 .mmap = sock_no_mmap, 3210 .sendpage = smc_sendpage, 3211 .splice_read = smc_splice_read, 3212 }; 3213 3214 static int __smc_create(struct net *net, struct socket *sock, int protocol, 3215 int kern, struct socket *clcsock) 3216 { 3217 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 3218 struct smc_sock *smc; 3219 struct sock *sk; 3220 int rc; 3221 3222 rc = -ESOCKTNOSUPPORT; 3223 if (sock->type != SOCK_STREAM) 3224 goto out; 3225 3226 rc = -EPROTONOSUPPORT; 3227 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 3228 goto out; 3229 3230 rc = -ENOBUFS; 3231 sock->ops = &smc_sock_ops; 3232 sock->state = SS_UNCONNECTED; 3233 sk = smc_sock_alloc(net, sock, protocol); 3234 if (!sk) 3235 goto out; 3236 3237 /* create internal TCP socket for CLC handshake and fallback */ 3238 smc = smc_sk(sk); 3239 smc->use_fallback = false; /* assume rdma capability first */ 3240 smc->fallback_rsn = 0; 3241 3242 /* default behavior from limit_smc_hs in every net namespace */ 3243 smc->limit_smc_hs = net->smc.limit_smc_hs; 3244 3245 rc = 0; 3246 if (!clcsock) { 3247 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 3248 &smc->clcsock); 3249 if (rc) { 3250 sk_common_release(sk); 3251 goto out; 3252 } 3253 } else { 3254 smc->clcsock = clcsock; 3255 } 3256 3257 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 3258 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 3259 3260 out: 3261 return rc; 3262 } 3263 3264 static int smc_create(struct net *net, struct socket *sock, int protocol, 3265 int kern) 3266 { 3267 return __smc_create(net, sock, protocol, kern, NULL); 3268 } 3269 3270 static const struct net_proto_family smc_sock_family_ops = { 3271 .family = PF_SMC, 3272 .owner = THIS_MODULE, 3273 .create = smc_create, 3274 }; 3275 3276 static int smc_ulp_init(struct sock *sk) 3277 { 3278 struct socket *tcp = sk->sk_socket; 3279 struct net *net = sock_net(sk); 3280 struct socket *smcsock; 3281 int protocol, ret; 3282 3283 /* only TCP can be replaced */ 3284 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP || 3285 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6)) 3286 return -ESOCKTNOSUPPORT; 3287 /* don't handle wq now */ 3288 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list) 3289 return -ENOTCONN; 3290 3291 if (sk->sk_family == AF_INET) 3292 protocol = SMCPROTO_SMC; 3293 else 3294 protocol = SMCPROTO_SMC6; 3295 3296 smcsock = sock_alloc(); 3297 if (!smcsock) 3298 return -ENFILE; 3299 3300 smcsock->type = SOCK_STREAM; 3301 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */ 3302 ret = __smc_create(net, smcsock, protocol, 1, tcp); 3303 if (ret) { 3304 sock_release(smcsock); /* module_put() which ops won't be NULL */ 3305 return ret; 3306 } 3307 3308 /* replace tcp socket to smc */ 3309 smcsock->file = tcp->file; 3310 smcsock->file->private_data = smcsock; 3311 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */ 3312 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */ 3313 tcp->file = NULL; 3314 3315 return ret; 3316 } 3317 3318 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk, 3319 const gfp_t priority) 3320 { 3321 struct inet_connection_sock *icsk = inet_csk(newsk); 3322 3323 /* don't inherit ulp ops to child when listen */ 3324 icsk->icsk_ulp_ops = NULL; 3325 } 3326 3327 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = { 3328 .name = "smc", 3329 .owner = THIS_MODULE, 3330 .init = smc_ulp_init, 3331 .clone = smc_ulp_clone, 3332 }; 3333 3334 unsigned int smc_net_id; 3335 3336 static __net_init int smc_net_init(struct net *net) 3337 { 3338 int rc; 3339 3340 rc = smc_sysctl_net_init(net); 3341 if (rc) 3342 return rc; 3343 return smc_pnet_net_init(net); 3344 } 3345 3346 static void __net_exit smc_net_exit(struct net *net) 3347 { 3348 smc_sysctl_net_exit(net); 3349 smc_pnet_net_exit(net); 3350 } 3351 3352 static __net_init int smc_net_stat_init(struct net *net) 3353 { 3354 return smc_stats_init(net); 3355 } 3356 3357 static void __net_exit smc_net_stat_exit(struct net *net) 3358 { 3359 smc_stats_exit(net); 3360 } 3361 3362 static struct pernet_operations smc_net_ops = { 3363 .init = smc_net_init, 3364 .exit = smc_net_exit, 3365 .id = &smc_net_id, 3366 .size = sizeof(struct smc_net), 3367 }; 3368 3369 static struct pernet_operations smc_net_stat_ops = { 3370 .init = smc_net_stat_init, 3371 .exit = smc_net_stat_exit, 3372 }; 3373 3374 static int __init smc_init(void) 3375 { 3376 int rc; 3377 3378 rc = register_pernet_subsys(&smc_net_ops); 3379 if (rc) 3380 return rc; 3381 3382 rc = register_pernet_subsys(&smc_net_stat_ops); 3383 if (rc) 3384 return rc; 3385 3386 smc_ism_init(); 3387 smc_clc_init(); 3388 3389 rc = smc_nl_init(); 3390 if (rc) 3391 goto out_pernet_subsys; 3392 3393 rc = smc_pnet_init(); 3394 if (rc) 3395 goto out_nl; 3396 3397 rc = -ENOMEM; 3398 3399 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0); 3400 if (!smc_tcp_ls_wq) 3401 goto out_pnet; 3402 3403 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0); 3404 if (!smc_hs_wq) 3405 goto out_alloc_tcp_ls_wq; 3406 3407 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0); 3408 if (!smc_close_wq) 3409 goto out_alloc_hs_wq; 3410 3411 rc = smc_core_init(); 3412 if (rc) { 3413 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 3414 goto out_alloc_wqs; 3415 } 3416 3417 rc = smc_llc_init(); 3418 if (rc) { 3419 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 3420 goto out_core; 3421 } 3422 3423 rc = smc_cdc_init(); 3424 if (rc) { 3425 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 3426 goto out_core; 3427 } 3428 3429 rc = proto_register(&smc_proto, 1); 3430 if (rc) { 3431 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 3432 goto out_core; 3433 } 3434 3435 rc = proto_register(&smc_proto6, 1); 3436 if (rc) { 3437 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 3438 goto out_proto; 3439 } 3440 3441 rc = sock_register(&smc_sock_family_ops); 3442 if (rc) { 3443 pr_err("%s: sock_register fails with %d\n", __func__, rc); 3444 goto out_proto6; 3445 } 3446 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 3447 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 3448 3449 rc = smc_ib_register_client(); 3450 if (rc) { 3451 pr_err("%s: ib_register fails with %d\n", __func__, rc); 3452 goto out_sock; 3453 } 3454 3455 rc = tcp_register_ulp(&smc_ulp_ops); 3456 if (rc) { 3457 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc); 3458 goto out_ib; 3459 } 3460 3461 static_branch_enable(&tcp_have_smc); 3462 return 0; 3463 3464 out_ib: 3465 smc_ib_unregister_client(); 3466 out_sock: 3467 sock_unregister(PF_SMC); 3468 out_proto6: 3469 proto_unregister(&smc_proto6); 3470 out_proto: 3471 proto_unregister(&smc_proto); 3472 out_core: 3473 smc_core_exit(); 3474 out_alloc_wqs: 3475 destroy_workqueue(smc_close_wq); 3476 out_alloc_hs_wq: 3477 destroy_workqueue(smc_hs_wq); 3478 out_alloc_tcp_ls_wq: 3479 destroy_workqueue(smc_tcp_ls_wq); 3480 out_pnet: 3481 smc_pnet_exit(); 3482 out_nl: 3483 smc_nl_exit(); 3484 out_pernet_subsys: 3485 unregister_pernet_subsys(&smc_net_ops); 3486 3487 return rc; 3488 } 3489 3490 static void __exit smc_exit(void) 3491 { 3492 static_branch_disable(&tcp_have_smc); 3493 tcp_unregister_ulp(&smc_ulp_ops); 3494 sock_unregister(PF_SMC); 3495 smc_core_exit(); 3496 smc_ib_unregister_client(); 3497 destroy_workqueue(smc_close_wq); 3498 destroy_workqueue(smc_tcp_ls_wq); 3499 destroy_workqueue(smc_hs_wq); 3500 proto_unregister(&smc_proto6); 3501 proto_unregister(&smc_proto); 3502 smc_pnet_exit(); 3503 smc_nl_exit(); 3504 smc_clc_exit(); 3505 unregister_pernet_subsys(&smc_net_stat_ops); 3506 unregister_pernet_subsys(&smc_net_ops); 3507 rcu_barrier(); 3508 } 3509 3510 module_init(smc_init); 3511 module_exit(smc_exit); 3512 3513 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 3514 MODULE_DESCRIPTION("smc socket address family"); 3515 MODULE_LICENSE("GPL"); 3516 MODULE_ALIAS_NETPROTO(PF_SMC); 3517 MODULE_ALIAS_TCP_ULP("smc"); 3518 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME); 3519