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