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 30 #include <net/sock.h> 31 #include <net/tcp.h> 32 #include <net/smc.h> 33 #include <asm/ioctls.h> 34 35 #include <net/net_namespace.h> 36 #include <net/netns/generic.h> 37 #include "smc_netns.h" 38 39 #include "smc.h" 40 #include "smc_clc.h" 41 #include "smc_llc.h" 42 #include "smc_cdc.h" 43 #include "smc_core.h" 44 #include "smc_ib.h" 45 #include "smc_ism.h" 46 #include "smc_pnet.h" 47 #include "smc_tx.h" 48 #include "smc_rx.h" 49 #include "smc_close.h" 50 51 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group 52 * creation on server 53 */ 54 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group 55 * creation on client 56 */ 57 58 static void smc_tcp_listen_work(struct work_struct *); 59 static void smc_connect_work(struct work_struct *); 60 61 static void smc_set_keepalive(struct sock *sk, int val) 62 { 63 struct smc_sock *smc = smc_sk(sk); 64 65 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val); 66 } 67 68 static struct smc_hashinfo smc_v4_hashinfo = { 69 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock), 70 }; 71 72 static struct smc_hashinfo smc_v6_hashinfo = { 73 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock), 74 }; 75 76 int smc_hash_sk(struct sock *sk) 77 { 78 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 79 struct hlist_head *head; 80 81 head = &h->ht; 82 83 write_lock_bh(&h->lock); 84 sk_add_node(sk, head); 85 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1); 86 write_unlock_bh(&h->lock); 87 88 return 0; 89 } 90 EXPORT_SYMBOL_GPL(smc_hash_sk); 91 92 void smc_unhash_sk(struct sock *sk) 93 { 94 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash; 95 96 write_lock_bh(&h->lock); 97 if (sk_del_node_init(sk)) 98 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1); 99 write_unlock_bh(&h->lock); 100 } 101 EXPORT_SYMBOL_GPL(smc_unhash_sk); 102 103 struct proto smc_proto = { 104 .name = "SMC", 105 .owner = THIS_MODULE, 106 .keepalive = smc_set_keepalive, 107 .hash = smc_hash_sk, 108 .unhash = smc_unhash_sk, 109 .obj_size = sizeof(struct smc_sock), 110 .h.smc_hash = &smc_v4_hashinfo, 111 .slab_flags = SLAB_TYPESAFE_BY_RCU, 112 }; 113 EXPORT_SYMBOL_GPL(smc_proto); 114 115 struct proto smc_proto6 = { 116 .name = "SMC6", 117 .owner = THIS_MODULE, 118 .keepalive = smc_set_keepalive, 119 .hash = smc_hash_sk, 120 .unhash = smc_unhash_sk, 121 .obj_size = sizeof(struct smc_sock), 122 .h.smc_hash = &smc_v6_hashinfo, 123 .slab_flags = SLAB_TYPESAFE_BY_RCU, 124 }; 125 EXPORT_SYMBOL_GPL(smc_proto6); 126 127 static void smc_restore_fallback_changes(struct smc_sock *smc) 128 { 129 smc->clcsock->file->private_data = smc->sk.sk_socket; 130 smc->clcsock->file = NULL; 131 } 132 133 static int __smc_release(struct smc_sock *smc) 134 { 135 struct sock *sk = &smc->sk; 136 int rc = 0; 137 138 if (!smc->use_fallback) { 139 rc = smc_close_active(smc); 140 sock_set_flag(sk, SOCK_DEAD); 141 sk->sk_shutdown |= SHUTDOWN_MASK; 142 } else { 143 if (sk->sk_state != SMC_LISTEN && sk->sk_state != SMC_INIT) 144 sock_put(sk); /* passive closing */ 145 if (sk->sk_state == SMC_LISTEN) { 146 /* wake up clcsock accept */ 147 rc = kernel_sock_shutdown(smc->clcsock, SHUT_RDWR); 148 } 149 sk->sk_state = SMC_CLOSED; 150 sk->sk_state_change(sk); 151 smc_restore_fallback_changes(smc); 152 } 153 154 sk->sk_prot->unhash(sk); 155 156 if (sk->sk_state == SMC_CLOSED) { 157 if (smc->clcsock) { 158 release_sock(sk); 159 smc_clcsock_release(smc); 160 lock_sock(sk); 161 } 162 if (!smc->use_fallback) 163 smc_conn_free(&smc->conn); 164 } 165 166 return rc; 167 } 168 169 static int smc_release(struct socket *sock) 170 { 171 struct sock *sk = sock->sk; 172 struct smc_sock *smc; 173 int rc = 0; 174 175 if (!sk) 176 goto out; 177 178 sock_hold(sk); /* sock_put below */ 179 smc = smc_sk(sk); 180 181 /* cleanup for a dangling non-blocking connect */ 182 if (smc->connect_nonblock && sk->sk_state == SMC_INIT) 183 tcp_abort(smc->clcsock->sk, ECONNABORTED); 184 flush_work(&smc->connect_work); 185 186 if (sk->sk_state == SMC_LISTEN) 187 /* smc_close_non_accepted() is called and acquires 188 * sock lock for child sockets again 189 */ 190 lock_sock_nested(sk, SINGLE_DEPTH_NESTING); 191 else 192 lock_sock(sk); 193 194 rc = __smc_release(smc); 195 196 /* detach socket */ 197 sock_orphan(sk); 198 sock->sk = NULL; 199 release_sock(sk); 200 201 sock_put(sk); /* sock_hold above */ 202 sock_put(sk); /* final sock_put */ 203 out: 204 return rc; 205 } 206 207 static void smc_destruct(struct sock *sk) 208 { 209 if (sk->sk_state != SMC_CLOSED) 210 return; 211 if (!sock_flag(sk, SOCK_DEAD)) 212 return; 213 214 sk_refcnt_debug_dec(sk); 215 } 216 217 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock, 218 int protocol) 219 { 220 struct smc_sock *smc; 221 struct proto *prot; 222 struct sock *sk; 223 224 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto; 225 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0); 226 if (!sk) 227 return NULL; 228 229 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */ 230 sk->sk_state = SMC_INIT; 231 sk->sk_destruct = smc_destruct; 232 sk->sk_protocol = protocol; 233 smc = smc_sk(sk); 234 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work); 235 INIT_WORK(&smc->connect_work, smc_connect_work); 236 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work); 237 INIT_LIST_HEAD(&smc->accept_q); 238 spin_lock_init(&smc->accept_q_lock); 239 spin_lock_init(&smc->conn.send_lock); 240 sk->sk_prot->hash(sk); 241 sk_refcnt_debug_inc(sk); 242 mutex_init(&smc->clcsock_release_lock); 243 244 return sk; 245 } 246 247 static int smc_bind(struct socket *sock, struct sockaddr *uaddr, 248 int addr_len) 249 { 250 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr; 251 struct sock *sk = sock->sk; 252 struct smc_sock *smc; 253 int rc; 254 255 smc = smc_sk(sk); 256 257 /* replicate tests from inet_bind(), to be safe wrt. future changes */ 258 rc = -EINVAL; 259 if (addr_len < sizeof(struct sockaddr_in)) 260 goto out; 261 262 rc = -EAFNOSUPPORT; 263 if (addr->sin_family != AF_INET && 264 addr->sin_family != AF_INET6 && 265 addr->sin_family != AF_UNSPEC) 266 goto out; 267 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */ 268 if (addr->sin_family == AF_UNSPEC && 269 addr->sin_addr.s_addr != htonl(INADDR_ANY)) 270 goto out; 271 272 lock_sock(sk); 273 274 /* Check if socket is already active */ 275 rc = -EINVAL; 276 if (sk->sk_state != SMC_INIT || smc->connect_nonblock) 277 goto out_rel; 278 279 smc->clcsock->sk->sk_reuse = sk->sk_reuse; 280 rc = kernel_bind(smc->clcsock, uaddr, addr_len); 281 282 out_rel: 283 release_sock(sk); 284 out: 285 return rc; 286 } 287 288 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk, 289 unsigned long mask) 290 { 291 /* options we don't get control via setsockopt for */ 292 nsk->sk_type = osk->sk_type; 293 nsk->sk_sndbuf = osk->sk_sndbuf; 294 nsk->sk_rcvbuf = osk->sk_rcvbuf; 295 nsk->sk_sndtimeo = osk->sk_sndtimeo; 296 nsk->sk_rcvtimeo = osk->sk_rcvtimeo; 297 nsk->sk_mark = osk->sk_mark; 298 nsk->sk_priority = osk->sk_priority; 299 nsk->sk_rcvlowat = osk->sk_rcvlowat; 300 nsk->sk_bound_dev_if = osk->sk_bound_dev_if; 301 nsk->sk_err = osk->sk_err; 302 303 nsk->sk_flags &= ~mask; 304 nsk->sk_flags |= osk->sk_flags & mask; 305 } 306 307 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \ 308 (1UL << SOCK_KEEPOPEN) | \ 309 (1UL << SOCK_LINGER) | \ 310 (1UL << SOCK_BROADCAST) | \ 311 (1UL << SOCK_TIMESTAMP) | \ 312 (1UL << SOCK_DBG) | \ 313 (1UL << SOCK_RCVTSTAMP) | \ 314 (1UL << SOCK_RCVTSTAMPNS) | \ 315 (1UL << SOCK_LOCALROUTE) | \ 316 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \ 317 (1UL << SOCK_RXQ_OVFL) | \ 318 (1UL << SOCK_WIFI_STATUS) | \ 319 (1UL << SOCK_NOFCS) | \ 320 (1UL << SOCK_FILTER_LOCKED) | \ 321 (1UL << SOCK_TSTAMP_NEW)) 322 /* copy only relevant settings and flags of SOL_SOCKET level from smc to 323 * clc socket (since smc is not called for these options from net/core) 324 */ 325 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc) 326 { 327 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC); 328 } 329 330 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \ 331 (1UL << SOCK_KEEPOPEN) | \ 332 (1UL << SOCK_LINGER) | \ 333 (1UL << SOCK_DBG)) 334 /* copy only settings and flags relevant for smc from clc to smc socket */ 335 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc) 336 { 337 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC); 338 } 339 340 /* register a new rmb, send confirm_rkey msg to register with peer */ 341 static int smcr_link_reg_rmb(struct smc_link *link, 342 struct smc_buf_desc *rmb_desc, bool conf_rkey) 343 { 344 if (!rmb_desc->is_reg_mr[link->link_idx]) { 345 /* register memory region for new rmb */ 346 if (smc_wr_reg_send(link, rmb_desc->mr_rx[link->link_idx])) { 347 rmb_desc->is_reg_err = true; 348 return -EFAULT; 349 } 350 rmb_desc->is_reg_mr[link->link_idx] = true; 351 } 352 if (!conf_rkey) 353 return 0; 354 355 /* exchange confirm_rkey msg with peer */ 356 if (!rmb_desc->is_conf_rkey) { 357 if (smc_llc_do_confirm_rkey(link, rmb_desc)) { 358 rmb_desc->is_reg_err = true; 359 return -EFAULT; 360 } 361 rmb_desc->is_conf_rkey = true; 362 } 363 return 0; 364 } 365 366 /* register the new rmb on all links */ 367 static int smcr_lgr_reg_rmbs(struct smc_link_group *lgr, 368 struct smc_buf_desc *rmb_desc) 369 { 370 int i, rc; 371 372 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 373 if (lgr->lnk[i].state != SMC_LNK_ACTIVE) 374 continue; 375 rc = smcr_link_reg_rmb(&lgr->lnk[i], rmb_desc, true); 376 if (rc) 377 return rc; 378 } 379 return 0; 380 } 381 382 static int smcr_clnt_conf_first_link(struct smc_sock *smc) 383 { 384 struct smc_link *link = smc->conn.lnk; 385 struct smc_llc_qentry *qentry; 386 int rc; 387 388 link->lgr->type = SMC_LGR_SINGLE; 389 390 /* receive CONFIRM LINK request from server over RoCE fabric */ 391 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 392 SMC_LLC_CONFIRM_LINK); 393 if (!qentry) { 394 struct smc_clc_msg_decline dclc; 395 396 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 397 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 398 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 399 } 400 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ); 401 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 402 if (rc) 403 return SMC_CLC_DECL_RMBE_EC; 404 405 rc = smc_ib_modify_qp_rts(link); 406 if (rc) 407 return SMC_CLC_DECL_ERR_RDYLNK; 408 409 smc_wr_remember_qp_attr(link); 410 411 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc, false)) 412 return SMC_CLC_DECL_ERR_REGRMB; 413 414 /* confirm_rkey is implicit on 1st contact */ 415 smc->conn.rmb_desc->is_conf_rkey = true; 416 417 /* send CONFIRM LINK response over RoCE fabric */ 418 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP); 419 if (rc < 0) 420 return SMC_CLC_DECL_TIMEOUT_CL; 421 422 smc_llc_link_active(link); 423 424 /* optional 2nd link, receive ADD LINK request from server */ 425 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME, 426 SMC_LLC_ADD_LINK); 427 if (!qentry) { 428 struct smc_clc_msg_decline dclc; 429 430 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 431 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 432 if (rc == -EAGAIN) 433 rc = 0; /* no DECLINE received, go with one link */ 434 return rc; 435 } 436 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl); 437 /* tbd: call smc_llc_cli_add_link(link, qentry); */ 438 return 0; 439 } 440 441 static void smcr_conn_save_peer_info(struct smc_sock *smc, 442 struct smc_clc_msg_accept_confirm *clc) 443 { 444 int bufsize = smc_uncompress_bufsize(clc->rmbe_size); 445 446 smc->conn.peer_rmbe_idx = clc->rmbe_idx; 447 smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token); 448 smc->conn.peer_rmbe_size = bufsize; 449 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 450 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1); 451 } 452 453 static void smcd_conn_save_peer_info(struct smc_sock *smc, 454 struct smc_clc_msg_accept_confirm *clc) 455 { 456 int bufsize = smc_uncompress_bufsize(clc->dmbe_size); 457 458 smc->conn.peer_rmbe_idx = clc->dmbe_idx; 459 smc->conn.peer_token = clc->token; 460 /* msg header takes up space in the buffer */ 461 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg); 462 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size); 463 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx; 464 } 465 466 static void smc_conn_save_peer_info(struct smc_sock *smc, 467 struct smc_clc_msg_accept_confirm *clc) 468 { 469 if (smc->conn.lgr->is_smcd) 470 smcd_conn_save_peer_info(smc, clc); 471 else 472 smcr_conn_save_peer_info(smc, clc); 473 } 474 475 static void smc_link_save_peer_info(struct smc_link *link, 476 struct smc_clc_msg_accept_confirm *clc) 477 { 478 link->peer_qpn = ntoh24(clc->qpn); 479 memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE); 480 memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac)); 481 link->peer_psn = ntoh24(clc->psn); 482 link->peer_mtu = clc->qp_mtu; 483 } 484 485 static void smc_switch_to_fallback(struct smc_sock *smc) 486 { 487 smc->use_fallback = true; 488 if (smc->sk.sk_socket && smc->sk.sk_socket->file) { 489 smc->clcsock->file = smc->sk.sk_socket->file; 490 smc->clcsock->file->private_data = smc->clcsock; 491 smc->clcsock->wq.fasync_list = 492 smc->sk.sk_socket->wq.fasync_list; 493 } 494 } 495 496 /* fall back during connect */ 497 static int smc_connect_fallback(struct smc_sock *smc, int reason_code) 498 { 499 smc_switch_to_fallback(smc); 500 smc->fallback_rsn = reason_code; 501 smc_copy_sock_settings_to_clc(smc); 502 smc->connect_nonblock = 0; 503 if (smc->sk.sk_state == SMC_INIT) 504 smc->sk.sk_state = SMC_ACTIVE; 505 return 0; 506 } 507 508 /* decline and fall back during connect */ 509 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code) 510 { 511 int rc; 512 513 if (reason_code < 0) { /* error, fallback is not possible */ 514 if (smc->sk.sk_state == SMC_INIT) 515 sock_put(&smc->sk); /* passive closing */ 516 return reason_code; 517 } 518 if (reason_code != SMC_CLC_DECL_PEERDECL) { 519 rc = smc_clc_send_decline(smc, reason_code); 520 if (rc < 0) { 521 if (smc->sk.sk_state == SMC_INIT) 522 sock_put(&smc->sk); /* passive closing */ 523 return rc; 524 } 525 } 526 return smc_connect_fallback(smc, reason_code); 527 } 528 529 /* abort connecting */ 530 static int smc_connect_abort(struct smc_sock *smc, int reason_code, 531 int local_contact) 532 { 533 bool is_smcd = smc->conn.lgr->is_smcd; 534 535 if (local_contact == SMC_FIRST_CONTACT) 536 smc_lgr_cleanup_early(&smc->conn); 537 else 538 smc_conn_free(&smc->conn); 539 if (is_smcd) 540 /* there is only one lgr role for SMC-D; use server lock */ 541 mutex_unlock(&smc_server_lgr_pending); 542 else 543 mutex_unlock(&smc_client_lgr_pending); 544 545 smc->connect_nonblock = 0; 546 return reason_code; 547 } 548 549 /* check if there is a rdma device available for this connection. */ 550 /* called for connect and listen */ 551 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini) 552 { 553 /* PNET table look up: search active ib_device and port 554 * within same PNETID that also contains the ethernet device 555 * used for the internal TCP socket 556 */ 557 smc_pnet_find_roce_resource(smc->clcsock->sk, ini); 558 if (!ini->ib_dev) 559 return SMC_CLC_DECL_NOSMCRDEV; 560 return 0; 561 } 562 563 /* check if there is an ISM device available for this connection. */ 564 /* called for connect and listen */ 565 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini) 566 { 567 /* Find ISM device with same PNETID as connecting interface */ 568 smc_pnet_find_ism_resource(smc->clcsock->sk, ini); 569 if (!ini->ism_dev) 570 return SMC_CLC_DECL_NOSMCDDEV; 571 return 0; 572 } 573 574 /* Check for VLAN ID and register it on ISM device just for CLC handshake */ 575 static int smc_connect_ism_vlan_setup(struct smc_sock *smc, 576 struct smc_init_info *ini) 577 { 578 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev, ini->vlan_id)) 579 return SMC_CLC_DECL_ISMVLANERR; 580 return 0; 581 } 582 583 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is 584 * used, the VLAN ID will be registered again during the connection setup. 585 */ 586 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc, bool is_smcd, 587 struct smc_init_info *ini) 588 { 589 if (!is_smcd) 590 return 0; 591 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev, ini->vlan_id)) 592 return SMC_CLC_DECL_CNFERR; 593 return 0; 594 } 595 596 /* CLC handshake during connect */ 597 static int smc_connect_clc(struct smc_sock *smc, int smc_type, 598 struct smc_clc_msg_accept_confirm *aclc, 599 struct smc_init_info *ini) 600 { 601 int rc = 0; 602 603 /* do inband token exchange */ 604 rc = smc_clc_send_proposal(smc, smc_type, ini); 605 if (rc) 606 return rc; 607 /* receive SMC Accept CLC message */ 608 return smc_clc_wait_msg(smc, aclc, sizeof(*aclc), SMC_CLC_ACCEPT, 609 CLC_WAIT_TIME); 610 } 611 612 /* setup for RDMA connection of client */ 613 static int smc_connect_rdma(struct smc_sock *smc, 614 struct smc_clc_msg_accept_confirm *aclc, 615 struct smc_init_info *ini) 616 { 617 int i, reason_code = 0; 618 struct smc_link *link; 619 620 ini->is_smcd = false; 621 ini->ib_lcl = &aclc->lcl; 622 ini->ib_clcqpn = ntoh24(aclc->qpn); 623 ini->srv_first_contact = aclc->hdr.flag; 624 625 mutex_lock(&smc_client_lgr_pending); 626 reason_code = smc_conn_create(smc, ini); 627 if (reason_code) { 628 mutex_unlock(&smc_client_lgr_pending); 629 return reason_code; 630 } 631 632 smc_conn_save_peer_info(smc, aclc); 633 634 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 635 link = smc->conn.lnk; 636 } else { 637 /* set link that was assigned by server */ 638 link = NULL; 639 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) { 640 struct smc_link *l = &smc->conn.lgr->lnk[i]; 641 642 if (l->peer_qpn == ntoh24(aclc->qpn)) { 643 link = l; 644 break; 645 } 646 } 647 if (!link) 648 return smc_connect_abort(smc, SMC_CLC_DECL_NOSRVLINK, 649 ini->cln_first_contact); 650 smc->conn.lnk = link; 651 } 652 653 /* create send buffer and rmb */ 654 if (smc_buf_create(smc, false)) 655 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 656 ini->cln_first_contact); 657 658 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 659 smc_link_save_peer_info(link, aclc); 660 661 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) 662 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RTOK, 663 ini->cln_first_contact); 664 665 smc_close_init(smc); 666 smc_rx_init(smc); 667 668 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 669 if (smc_ib_ready_link(link)) 670 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_RDYLNK, 671 ini->cln_first_contact); 672 } else { 673 if (smcr_lgr_reg_rmbs(smc->conn.lgr, smc->conn.rmb_desc)) 674 return smc_connect_abort(smc, SMC_CLC_DECL_ERR_REGRMB, 675 ini->cln_first_contact); 676 } 677 smc_rmb_sync_sg_for_device(&smc->conn); 678 679 reason_code = smc_clc_send_confirm(smc); 680 if (reason_code) 681 return smc_connect_abort(smc, reason_code, 682 ini->cln_first_contact); 683 684 smc_tx_init(smc); 685 686 if (ini->cln_first_contact == SMC_FIRST_CONTACT) { 687 /* QP confirmation over RoCE fabric */ 688 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 689 reason_code = smcr_clnt_conf_first_link(smc); 690 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 691 if (reason_code) 692 return smc_connect_abort(smc, reason_code, 693 ini->cln_first_contact); 694 } 695 mutex_unlock(&smc_client_lgr_pending); 696 697 smc_copy_sock_settings_to_clc(smc); 698 smc->connect_nonblock = 0; 699 if (smc->sk.sk_state == SMC_INIT) 700 smc->sk.sk_state = SMC_ACTIVE; 701 702 return 0; 703 } 704 705 /* setup for ISM connection of client */ 706 static int smc_connect_ism(struct smc_sock *smc, 707 struct smc_clc_msg_accept_confirm *aclc, 708 struct smc_init_info *ini) 709 { 710 int rc = 0; 711 712 ini->is_smcd = true; 713 ini->ism_gid = aclc->gid; 714 ini->srv_first_contact = aclc->hdr.flag; 715 716 /* there is only one lgr role for SMC-D; use server lock */ 717 mutex_lock(&smc_server_lgr_pending); 718 rc = smc_conn_create(smc, ini); 719 if (rc) { 720 mutex_unlock(&smc_server_lgr_pending); 721 return rc; 722 } 723 724 /* Create send and receive buffers */ 725 if (smc_buf_create(smc, true)) 726 return smc_connect_abort(smc, SMC_CLC_DECL_MEM, 727 ini->cln_first_contact); 728 729 smc_conn_save_peer_info(smc, aclc); 730 smc_close_init(smc); 731 smc_rx_init(smc); 732 smc_tx_init(smc); 733 734 rc = smc_clc_send_confirm(smc); 735 if (rc) 736 return smc_connect_abort(smc, rc, ini->cln_first_contact); 737 mutex_unlock(&smc_server_lgr_pending); 738 739 smc_copy_sock_settings_to_clc(smc); 740 smc->connect_nonblock = 0; 741 if (smc->sk.sk_state == SMC_INIT) 742 smc->sk.sk_state = SMC_ACTIVE; 743 744 return 0; 745 } 746 747 /* perform steps before actually connecting */ 748 static int __smc_connect(struct smc_sock *smc) 749 { 750 bool ism_supported = false, rdma_supported = false; 751 struct smc_clc_msg_accept_confirm aclc; 752 struct smc_init_info ini = {0}; 753 int smc_type; 754 int rc = 0; 755 756 if (smc->use_fallback) 757 return smc_connect_fallback(smc, smc->fallback_rsn); 758 759 /* if peer has not signalled SMC-capability, fall back */ 760 if (!tcp_sk(smc->clcsock->sk)->syn_smc) 761 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC); 762 763 /* IPSec connections opt out of SMC-R optimizations */ 764 if (using_ipsec(smc)) 765 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC); 766 767 /* get vlan id from IP device */ 768 if (smc_vlan_by_tcpsk(smc->clcsock, &ini)) 769 return smc_connect_decline_fallback(smc, 770 SMC_CLC_DECL_GETVLANERR); 771 772 /* check if there is an ism device available */ 773 if (!smc_find_ism_device(smc, &ini) && 774 !smc_connect_ism_vlan_setup(smc, &ini)) { 775 /* ISM is supported for this connection */ 776 ism_supported = true; 777 smc_type = SMC_TYPE_D; 778 } 779 780 /* check if there is a rdma device available */ 781 if (!smc_find_rdma_device(smc, &ini)) { 782 /* RDMA is supported for this connection */ 783 rdma_supported = true; 784 if (ism_supported) 785 smc_type = SMC_TYPE_B; /* both */ 786 else 787 smc_type = SMC_TYPE_R; /* only RDMA */ 788 } 789 790 /* if neither ISM nor RDMA are supported, fallback */ 791 if (!rdma_supported && !ism_supported) 792 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_NOSMCDEV); 793 794 /* perform CLC handshake */ 795 rc = smc_connect_clc(smc, smc_type, &aclc, &ini); 796 if (rc) { 797 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 798 return smc_connect_decline_fallback(smc, rc); 799 } 800 801 /* depending on previous steps, connect using rdma or ism */ 802 if (rdma_supported && aclc.hdr.path == SMC_TYPE_R) 803 rc = smc_connect_rdma(smc, &aclc, &ini); 804 else if (ism_supported && aclc.hdr.path == SMC_TYPE_D) 805 rc = smc_connect_ism(smc, &aclc, &ini); 806 else 807 rc = SMC_CLC_DECL_MODEUNSUPP; 808 if (rc) { 809 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 810 return smc_connect_decline_fallback(smc, rc); 811 } 812 813 smc_connect_ism_vlan_cleanup(smc, ism_supported, &ini); 814 return 0; 815 } 816 817 static void smc_connect_work(struct work_struct *work) 818 { 819 struct smc_sock *smc = container_of(work, struct smc_sock, 820 connect_work); 821 long timeo = smc->sk.sk_sndtimeo; 822 int rc = 0; 823 824 if (!timeo) 825 timeo = MAX_SCHEDULE_TIMEOUT; 826 lock_sock(smc->clcsock->sk); 827 if (smc->clcsock->sk->sk_err) { 828 smc->sk.sk_err = smc->clcsock->sk->sk_err; 829 } else if ((1 << smc->clcsock->sk->sk_state) & 830 (TCPF_SYN_SENT | TCP_SYN_RECV)) { 831 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo); 832 if ((rc == -EPIPE) && 833 ((1 << smc->clcsock->sk->sk_state) & 834 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT))) 835 rc = 0; 836 } 837 release_sock(smc->clcsock->sk); 838 lock_sock(&smc->sk); 839 if (rc != 0 || smc->sk.sk_err) { 840 smc->sk.sk_state = SMC_CLOSED; 841 if (rc == -EPIPE || rc == -EAGAIN) 842 smc->sk.sk_err = EPIPE; 843 else if (signal_pending(current)) 844 smc->sk.sk_err = -sock_intr_errno(timeo); 845 sock_put(&smc->sk); /* passive closing */ 846 goto out; 847 } 848 849 rc = __smc_connect(smc); 850 if (rc < 0) 851 smc->sk.sk_err = -rc; 852 853 out: 854 if (!sock_flag(&smc->sk, SOCK_DEAD)) { 855 if (smc->sk.sk_err) { 856 smc->sk.sk_state_change(&smc->sk); 857 } else { /* allow polling before and after fallback decision */ 858 smc->clcsock->sk->sk_write_space(smc->clcsock->sk); 859 smc->sk.sk_write_space(&smc->sk); 860 } 861 } 862 release_sock(&smc->sk); 863 } 864 865 static int smc_connect(struct socket *sock, struct sockaddr *addr, 866 int alen, int flags) 867 { 868 struct sock *sk = sock->sk; 869 struct smc_sock *smc; 870 int rc = -EINVAL; 871 872 smc = smc_sk(sk); 873 874 /* separate smc parameter checking to be safe */ 875 if (alen < sizeof(addr->sa_family)) 876 goto out_err; 877 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6) 878 goto out_err; 879 880 lock_sock(sk); 881 switch (sk->sk_state) { 882 default: 883 goto out; 884 case SMC_ACTIVE: 885 rc = -EISCONN; 886 goto out; 887 case SMC_INIT: 888 rc = 0; 889 break; 890 } 891 892 smc_copy_sock_settings_to_clc(smc); 893 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 894 if (smc->connect_nonblock) { 895 rc = -EALREADY; 896 goto out; 897 } 898 rc = kernel_connect(smc->clcsock, addr, alen, flags); 899 if (rc && rc != -EINPROGRESS) 900 goto out; 901 902 sock_hold(&smc->sk); /* sock put in passive closing */ 903 if (smc->use_fallback) 904 goto out; 905 if (flags & O_NONBLOCK) { 906 if (schedule_work(&smc->connect_work)) 907 smc->connect_nonblock = 1; 908 rc = -EINPROGRESS; 909 } else { 910 rc = __smc_connect(smc); 911 if (rc < 0) 912 goto out; 913 else 914 rc = 0; /* success cases including fallback */ 915 } 916 917 out: 918 release_sock(sk); 919 out_err: 920 return rc; 921 } 922 923 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc) 924 { 925 struct socket *new_clcsock = NULL; 926 struct sock *lsk = &lsmc->sk; 927 struct sock *new_sk; 928 int rc = -EINVAL; 929 930 release_sock(lsk); 931 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol); 932 if (!new_sk) { 933 rc = -ENOMEM; 934 lsk->sk_err = ENOMEM; 935 *new_smc = NULL; 936 lock_sock(lsk); 937 goto out; 938 } 939 *new_smc = smc_sk(new_sk); 940 941 mutex_lock(&lsmc->clcsock_release_lock); 942 if (lsmc->clcsock) 943 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0); 944 mutex_unlock(&lsmc->clcsock_release_lock); 945 lock_sock(lsk); 946 if (rc < 0) 947 lsk->sk_err = -rc; 948 if (rc < 0 || lsk->sk_state == SMC_CLOSED) { 949 new_sk->sk_prot->unhash(new_sk); 950 if (new_clcsock) 951 sock_release(new_clcsock); 952 new_sk->sk_state = SMC_CLOSED; 953 sock_set_flag(new_sk, SOCK_DEAD); 954 sock_put(new_sk); /* final */ 955 *new_smc = NULL; 956 goto out; 957 } 958 959 (*new_smc)->clcsock = new_clcsock; 960 out: 961 return rc; 962 } 963 964 /* add a just created sock to the accept queue of the listen sock as 965 * candidate for a following socket accept call from user space 966 */ 967 static void smc_accept_enqueue(struct sock *parent, struct sock *sk) 968 { 969 struct smc_sock *par = smc_sk(parent); 970 971 sock_hold(sk); /* sock_put in smc_accept_unlink () */ 972 spin_lock(&par->accept_q_lock); 973 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q); 974 spin_unlock(&par->accept_q_lock); 975 sk_acceptq_added(parent); 976 } 977 978 /* remove a socket from the accept queue of its parental listening socket */ 979 static void smc_accept_unlink(struct sock *sk) 980 { 981 struct smc_sock *par = smc_sk(sk)->listen_smc; 982 983 spin_lock(&par->accept_q_lock); 984 list_del_init(&smc_sk(sk)->accept_q); 985 spin_unlock(&par->accept_q_lock); 986 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk); 987 sock_put(sk); /* sock_hold in smc_accept_enqueue */ 988 } 989 990 /* remove a sock from the accept queue to bind it to a new socket created 991 * for a socket accept call from user space 992 */ 993 struct sock *smc_accept_dequeue(struct sock *parent, 994 struct socket *new_sock) 995 { 996 struct smc_sock *isk, *n; 997 struct sock *new_sk; 998 999 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) { 1000 new_sk = (struct sock *)isk; 1001 1002 smc_accept_unlink(new_sk); 1003 if (new_sk->sk_state == SMC_CLOSED) { 1004 new_sk->sk_prot->unhash(new_sk); 1005 if (isk->clcsock) { 1006 sock_release(isk->clcsock); 1007 isk->clcsock = NULL; 1008 } 1009 sock_put(new_sk); /* final */ 1010 continue; 1011 } 1012 if (new_sock) { 1013 sock_graft(new_sk, new_sock); 1014 if (isk->use_fallback) { 1015 smc_sk(new_sk)->clcsock->file = new_sock->file; 1016 isk->clcsock->file->private_data = isk->clcsock; 1017 } 1018 } 1019 return new_sk; 1020 } 1021 return NULL; 1022 } 1023 1024 /* clean up for a created but never accepted sock */ 1025 void smc_close_non_accepted(struct sock *sk) 1026 { 1027 struct smc_sock *smc = smc_sk(sk); 1028 1029 sock_hold(sk); /* sock_put below */ 1030 lock_sock(sk); 1031 if (!sk->sk_lingertime) 1032 /* wait for peer closing */ 1033 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT; 1034 __smc_release(smc); 1035 release_sock(sk); 1036 sock_put(sk); /* sock_hold above */ 1037 sock_put(sk); /* final sock_put */ 1038 } 1039 1040 static int smcr_serv_conf_first_link(struct smc_sock *smc) 1041 { 1042 struct smc_link *link = smc->conn.lnk; 1043 struct smc_llc_qentry *qentry; 1044 int rc; 1045 1046 link->lgr->type = SMC_LGR_SINGLE; 1047 1048 if (smcr_link_reg_rmb(link, smc->conn.rmb_desc, false)) 1049 return SMC_CLC_DECL_ERR_REGRMB; 1050 1051 /* send CONFIRM LINK request to client over the RoCE fabric */ 1052 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ); 1053 if (rc < 0) 1054 return SMC_CLC_DECL_TIMEOUT_CL; 1055 1056 /* receive CONFIRM LINK response from client over the RoCE fabric */ 1057 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME, 1058 SMC_LLC_CONFIRM_LINK); 1059 if (!qentry) { 1060 struct smc_clc_msg_decline dclc; 1061 1062 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc), 1063 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT); 1064 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc; 1065 } 1066 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP); 1067 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl); 1068 if (rc) 1069 return SMC_CLC_DECL_RMBE_EC; 1070 1071 /* confirm_rkey is implicit on 1st contact */ 1072 smc->conn.rmb_desc->is_conf_rkey = true; 1073 1074 smc_llc_link_active(link); 1075 1076 /* initial contact - try to establish second link */ 1077 /* tbd: call smc_llc_srv_add_link(link); */ 1078 return 0; 1079 } 1080 1081 /* listen worker: finish */ 1082 static void smc_listen_out(struct smc_sock *new_smc) 1083 { 1084 struct smc_sock *lsmc = new_smc->listen_smc; 1085 struct sock *newsmcsk = &new_smc->sk; 1086 1087 if (lsmc->sk.sk_state == SMC_LISTEN) { 1088 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING); 1089 smc_accept_enqueue(&lsmc->sk, newsmcsk); 1090 release_sock(&lsmc->sk); 1091 } else { /* no longer listening */ 1092 smc_close_non_accepted(newsmcsk); 1093 } 1094 1095 /* Wake up accept */ 1096 lsmc->sk.sk_data_ready(&lsmc->sk); 1097 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */ 1098 } 1099 1100 /* listen worker: finish in state connected */ 1101 static void smc_listen_out_connected(struct smc_sock *new_smc) 1102 { 1103 struct sock *newsmcsk = &new_smc->sk; 1104 1105 sk_refcnt_debug_inc(newsmcsk); 1106 if (newsmcsk->sk_state == SMC_INIT) 1107 newsmcsk->sk_state = SMC_ACTIVE; 1108 1109 smc_listen_out(new_smc); 1110 } 1111 1112 /* listen worker: finish in error state */ 1113 static void smc_listen_out_err(struct smc_sock *new_smc) 1114 { 1115 struct sock *newsmcsk = &new_smc->sk; 1116 1117 if (newsmcsk->sk_state == SMC_INIT) 1118 sock_put(&new_smc->sk); /* passive closing */ 1119 newsmcsk->sk_state = SMC_CLOSED; 1120 1121 smc_listen_out(new_smc); 1122 } 1123 1124 /* listen worker: decline and fall back if possible */ 1125 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code, 1126 int local_contact) 1127 { 1128 /* RDMA setup failed, switch back to TCP */ 1129 if (local_contact == SMC_FIRST_CONTACT) 1130 smc_lgr_cleanup_early(&new_smc->conn); 1131 else 1132 smc_conn_free(&new_smc->conn); 1133 if (reason_code < 0) { /* error, no fallback possible */ 1134 smc_listen_out_err(new_smc); 1135 return; 1136 } 1137 smc_switch_to_fallback(new_smc); 1138 new_smc->fallback_rsn = reason_code; 1139 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) { 1140 if (smc_clc_send_decline(new_smc, reason_code) < 0) { 1141 smc_listen_out_err(new_smc); 1142 return; 1143 } 1144 } 1145 smc_listen_out_connected(new_smc); 1146 } 1147 1148 /* listen worker: check prefixes */ 1149 static int smc_listen_prfx_check(struct smc_sock *new_smc, 1150 struct smc_clc_msg_proposal *pclc) 1151 { 1152 struct smc_clc_msg_proposal_prefix *pclc_prfx; 1153 struct socket *newclcsock = new_smc->clcsock; 1154 1155 pclc_prfx = smc_clc_proposal_get_prefix(pclc); 1156 if (smc_clc_prfx_match(newclcsock, pclc_prfx)) 1157 return SMC_CLC_DECL_DIFFPREFIX; 1158 1159 return 0; 1160 } 1161 1162 /* listen worker: initialize connection and buffers */ 1163 static int smc_listen_rdma_init(struct smc_sock *new_smc, 1164 struct smc_init_info *ini) 1165 { 1166 int rc; 1167 1168 /* allocate connection / link group */ 1169 rc = smc_conn_create(new_smc, ini); 1170 if (rc) 1171 return rc; 1172 1173 /* create send buffer and rmb */ 1174 if (smc_buf_create(new_smc, false)) 1175 return SMC_CLC_DECL_MEM; 1176 1177 return 0; 1178 } 1179 1180 /* listen worker: initialize connection and buffers for SMC-D */ 1181 static int smc_listen_ism_init(struct smc_sock *new_smc, 1182 struct smc_clc_msg_proposal *pclc, 1183 struct smc_init_info *ini) 1184 { 1185 struct smc_clc_msg_smcd *pclc_smcd; 1186 int rc; 1187 1188 pclc_smcd = smc_get_clc_msg_smcd(pclc); 1189 ini->ism_gid = pclc_smcd->gid; 1190 rc = smc_conn_create(new_smc, ini); 1191 if (rc) 1192 return rc; 1193 1194 /* Check if peer can be reached via ISM device */ 1195 if (smc_ism_cantalk(new_smc->conn.lgr->peer_gid, 1196 new_smc->conn.lgr->vlan_id, 1197 new_smc->conn.lgr->smcd)) { 1198 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1199 smc_lgr_cleanup_early(&new_smc->conn); 1200 else 1201 smc_conn_free(&new_smc->conn); 1202 return SMC_CLC_DECL_SMCDNOTALK; 1203 } 1204 1205 /* Create send and receive buffers */ 1206 if (smc_buf_create(new_smc, true)) { 1207 if (ini->cln_first_contact == SMC_FIRST_CONTACT) 1208 smc_lgr_cleanup_early(&new_smc->conn); 1209 else 1210 smc_conn_free(&new_smc->conn); 1211 return SMC_CLC_DECL_MEM; 1212 } 1213 1214 return 0; 1215 } 1216 1217 /* listen worker: register buffers */ 1218 static int smc_listen_rdma_reg(struct smc_sock *new_smc, int local_contact) 1219 { 1220 struct smc_connection *conn = &new_smc->conn; 1221 1222 if (local_contact != SMC_FIRST_CONTACT) { 1223 if (smcr_lgr_reg_rmbs(conn->lgr, conn->rmb_desc)) 1224 return SMC_CLC_DECL_ERR_REGRMB; 1225 } 1226 smc_rmb_sync_sg_for_device(&new_smc->conn); 1227 1228 return 0; 1229 } 1230 1231 /* listen worker: finish RDMA setup */ 1232 static int smc_listen_rdma_finish(struct smc_sock *new_smc, 1233 struct smc_clc_msg_accept_confirm *cclc, 1234 int local_contact) 1235 { 1236 struct smc_link *link = new_smc->conn.lnk; 1237 int reason_code = 0; 1238 1239 if (local_contact == SMC_FIRST_CONTACT) 1240 smc_link_save_peer_info(link, cclc); 1241 1242 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc)) { 1243 reason_code = SMC_CLC_DECL_ERR_RTOK; 1244 goto decline; 1245 } 1246 1247 if (local_contact == SMC_FIRST_CONTACT) { 1248 if (smc_ib_ready_link(link)) { 1249 reason_code = SMC_CLC_DECL_ERR_RDYLNK; 1250 goto decline; 1251 } 1252 /* QP confirmation over RoCE fabric */ 1253 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK); 1254 reason_code = smcr_serv_conf_first_link(new_smc); 1255 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl); 1256 if (reason_code) 1257 goto decline; 1258 } 1259 return 0; 1260 1261 decline: 1262 smc_listen_decline(new_smc, reason_code, local_contact); 1263 return reason_code; 1264 } 1265 1266 /* setup for RDMA connection of server */ 1267 static void smc_listen_work(struct work_struct *work) 1268 { 1269 struct smc_sock *new_smc = container_of(work, struct smc_sock, 1270 smc_listen_work); 1271 struct socket *newclcsock = new_smc->clcsock; 1272 struct smc_clc_msg_accept_confirm cclc; 1273 struct smc_clc_msg_proposal *pclc; 1274 struct smc_init_info ini = {0}; 1275 bool ism_supported = false; 1276 u8 buf[SMC_CLC_MAX_LEN]; 1277 int rc = 0; 1278 1279 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN) 1280 return smc_listen_out_err(new_smc); 1281 1282 if (new_smc->use_fallback) { 1283 smc_listen_out_connected(new_smc); 1284 return; 1285 } 1286 1287 /* check if peer is smc capable */ 1288 if (!tcp_sk(newclcsock->sk)->syn_smc) { 1289 smc_switch_to_fallback(new_smc); 1290 new_smc->fallback_rsn = SMC_CLC_DECL_PEERNOSMC; 1291 smc_listen_out_connected(new_smc); 1292 return; 1293 } 1294 1295 /* do inband token exchange - 1296 * wait for and receive SMC Proposal CLC message 1297 */ 1298 pclc = (struct smc_clc_msg_proposal *)&buf; 1299 rc = smc_clc_wait_msg(new_smc, pclc, SMC_CLC_MAX_LEN, 1300 SMC_CLC_PROPOSAL, CLC_WAIT_TIME); 1301 if (rc) 1302 goto out_decl; 1303 1304 /* IPSec connections opt out of SMC-R optimizations */ 1305 if (using_ipsec(new_smc)) { 1306 rc = SMC_CLC_DECL_IPSEC; 1307 goto out_decl; 1308 } 1309 1310 /* check for matching IP prefix and subnet length */ 1311 rc = smc_listen_prfx_check(new_smc, pclc); 1312 if (rc) 1313 goto out_decl; 1314 1315 /* get vlan id from IP device */ 1316 if (smc_vlan_by_tcpsk(new_smc->clcsock, &ini)) { 1317 rc = SMC_CLC_DECL_GETVLANERR; 1318 goto out_decl; 1319 } 1320 1321 mutex_lock(&smc_server_lgr_pending); 1322 smc_close_init(new_smc); 1323 smc_rx_init(new_smc); 1324 smc_tx_init(new_smc); 1325 1326 /* check if ISM is available */ 1327 if (pclc->hdr.path == SMC_TYPE_D || pclc->hdr.path == SMC_TYPE_B) { 1328 ini.is_smcd = true; /* prepare ISM check */ 1329 rc = smc_find_ism_device(new_smc, &ini); 1330 if (!rc) 1331 rc = smc_listen_ism_init(new_smc, pclc, &ini); 1332 if (!rc) 1333 ism_supported = true; 1334 else if (pclc->hdr.path == SMC_TYPE_D) 1335 goto out_unlock; /* skip RDMA and decline */ 1336 } 1337 1338 /* check if RDMA is available */ 1339 if (!ism_supported) { /* SMC_TYPE_R or SMC_TYPE_B */ 1340 /* prepare RDMA check */ 1341 ini.is_smcd = false; 1342 ini.ism_dev = NULL; 1343 ini.ib_lcl = &pclc->lcl; 1344 rc = smc_find_rdma_device(new_smc, &ini); 1345 if (rc) { 1346 /* no RDMA device found */ 1347 if (pclc->hdr.path == SMC_TYPE_B) 1348 /* neither ISM nor RDMA device found */ 1349 rc = SMC_CLC_DECL_NOSMCDEV; 1350 goto out_unlock; 1351 } 1352 rc = smc_listen_rdma_init(new_smc, &ini); 1353 if (rc) 1354 goto out_unlock; 1355 rc = smc_listen_rdma_reg(new_smc, ini.cln_first_contact); 1356 if (rc) 1357 goto out_unlock; 1358 } 1359 1360 /* send SMC Accept CLC message */ 1361 rc = smc_clc_send_accept(new_smc, ini.cln_first_contact); 1362 if (rc) 1363 goto out_unlock; 1364 1365 /* SMC-D does not need this lock any more */ 1366 if (ism_supported) 1367 mutex_unlock(&smc_server_lgr_pending); 1368 1369 /* receive SMC Confirm CLC message */ 1370 rc = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc), 1371 SMC_CLC_CONFIRM, CLC_WAIT_TIME); 1372 if (rc) { 1373 if (!ism_supported) 1374 goto out_unlock; 1375 goto out_decl; 1376 } 1377 1378 /* finish worker */ 1379 if (!ism_supported) { 1380 rc = smc_listen_rdma_finish(new_smc, &cclc, 1381 ini.cln_first_contact); 1382 mutex_unlock(&smc_server_lgr_pending); 1383 if (rc) 1384 return; 1385 } 1386 smc_conn_save_peer_info(new_smc, &cclc); 1387 smc_listen_out_connected(new_smc); 1388 return; 1389 1390 out_unlock: 1391 mutex_unlock(&smc_server_lgr_pending); 1392 out_decl: 1393 smc_listen_decline(new_smc, rc, ini.cln_first_contact); 1394 } 1395 1396 static void smc_tcp_listen_work(struct work_struct *work) 1397 { 1398 struct smc_sock *lsmc = container_of(work, struct smc_sock, 1399 tcp_listen_work); 1400 struct sock *lsk = &lsmc->sk; 1401 struct smc_sock *new_smc; 1402 int rc = 0; 1403 1404 lock_sock(lsk); 1405 while (lsk->sk_state == SMC_LISTEN) { 1406 rc = smc_clcsock_accept(lsmc, &new_smc); 1407 if (rc) 1408 goto out; 1409 if (!new_smc) 1410 continue; 1411 1412 new_smc->listen_smc = lsmc; 1413 new_smc->use_fallback = lsmc->use_fallback; 1414 new_smc->fallback_rsn = lsmc->fallback_rsn; 1415 sock_hold(lsk); /* sock_put in smc_listen_work */ 1416 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work); 1417 smc_copy_sock_settings_to_smc(new_smc); 1418 new_smc->sk.sk_sndbuf = lsmc->sk.sk_sndbuf; 1419 new_smc->sk.sk_rcvbuf = lsmc->sk.sk_rcvbuf; 1420 sock_hold(&new_smc->sk); /* sock_put in passive closing */ 1421 if (!schedule_work(&new_smc->smc_listen_work)) 1422 sock_put(&new_smc->sk); 1423 } 1424 1425 out: 1426 release_sock(lsk); 1427 sock_put(&lsmc->sk); /* sock_hold in smc_listen */ 1428 } 1429 1430 static int smc_listen(struct socket *sock, int backlog) 1431 { 1432 struct sock *sk = sock->sk; 1433 struct smc_sock *smc; 1434 int rc; 1435 1436 smc = smc_sk(sk); 1437 lock_sock(sk); 1438 1439 rc = -EINVAL; 1440 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) || 1441 smc->connect_nonblock) 1442 goto out; 1443 1444 rc = 0; 1445 if (sk->sk_state == SMC_LISTEN) { 1446 sk->sk_max_ack_backlog = backlog; 1447 goto out; 1448 } 1449 /* some socket options are handled in core, so we could not apply 1450 * them to the clc socket -- copy smc socket options to clc socket 1451 */ 1452 smc_copy_sock_settings_to_clc(smc); 1453 if (!smc->use_fallback) 1454 tcp_sk(smc->clcsock->sk)->syn_smc = 1; 1455 1456 rc = kernel_listen(smc->clcsock, backlog); 1457 if (rc) 1458 goto out; 1459 sk->sk_max_ack_backlog = backlog; 1460 sk->sk_ack_backlog = 0; 1461 sk->sk_state = SMC_LISTEN; 1462 sock_hold(sk); /* sock_hold in tcp_listen_worker */ 1463 if (!schedule_work(&smc->tcp_listen_work)) 1464 sock_put(sk); 1465 1466 out: 1467 release_sock(sk); 1468 return rc; 1469 } 1470 1471 static int smc_accept(struct socket *sock, struct socket *new_sock, 1472 int flags, bool kern) 1473 { 1474 struct sock *sk = sock->sk, *nsk; 1475 DECLARE_WAITQUEUE(wait, current); 1476 struct smc_sock *lsmc; 1477 long timeo; 1478 int rc = 0; 1479 1480 lsmc = smc_sk(sk); 1481 sock_hold(sk); /* sock_put below */ 1482 lock_sock(sk); 1483 1484 if (lsmc->sk.sk_state != SMC_LISTEN) { 1485 rc = -EINVAL; 1486 release_sock(sk); 1487 goto out; 1488 } 1489 1490 /* Wait for an incoming connection */ 1491 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK); 1492 add_wait_queue_exclusive(sk_sleep(sk), &wait); 1493 while (!(nsk = smc_accept_dequeue(sk, new_sock))) { 1494 set_current_state(TASK_INTERRUPTIBLE); 1495 if (!timeo) { 1496 rc = -EAGAIN; 1497 break; 1498 } 1499 release_sock(sk); 1500 timeo = schedule_timeout(timeo); 1501 /* wakeup by sk_data_ready in smc_listen_work() */ 1502 sched_annotate_sleep(); 1503 lock_sock(sk); 1504 if (signal_pending(current)) { 1505 rc = sock_intr_errno(timeo); 1506 break; 1507 } 1508 } 1509 set_current_state(TASK_RUNNING); 1510 remove_wait_queue(sk_sleep(sk), &wait); 1511 1512 if (!rc) 1513 rc = sock_error(nsk); 1514 release_sock(sk); 1515 if (rc) 1516 goto out; 1517 1518 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) { 1519 /* wait till data arrives on the socket */ 1520 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept * 1521 MSEC_PER_SEC); 1522 if (smc_sk(nsk)->use_fallback) { 1523 struct sock *clcsk = smc_sk(nsk)->clcsock->sk; 1524 1525 lock_sock(clcsk); 1526 if (skb_queue_empty(&clcsk->sk_receive_queue)) 1527 sk_wait_data(clcsk, &timeo, NULL); 1528 release_sock(clcsk); 1529 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) { 1530 lock_sock(nsk); 1531 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available); 1532 release_sock(nsk); 1533 } 1534 } 1535 1536 out: 1537 sock_put(sk); /* sock_hold above */ 1538 return rc; 1539 } 1540 1541 static int smc_getname(struct socket *sock, struct sockaddr *addr, 1542 int peer) 1543 { 1544 struct smc_sock *smc; 1545 1546 if (peer && (sock->sk->sk_state != SMC_ACTIVE) && 1547 (sock->sk->sk_state != SMC_APPCLOSEWAIT1)) 1548 return -ENOTCONN; 1549 1550 smc = smc_sk(sock->sk); 1551 1552 return smc->clcsock->ops->getname(smc->clcsock, addr, peer); 1553 } 1554 1555 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len) 1556 { 1557 struct sock *sk = sock->sk; 1558 struct smc_sock *smc; 1559 int rc = -EPIPE; 1560 1561 smc = smc_sk(sk); 1562 lock_sock(sk); 1563 if ((sk->sk_state != SMC_ACTIVE) && 1564 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1565 (sk->sk_state != SMC_INIT)) 1566 goto out; 1567 1568 if (msg->msg_flags & MSG_FASTOPEN) { 1569 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1570 smc_switch_to_fallback(smc); 1571 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1572 } else { 1573 rc = -EINVAL; 1574 goto out; 1575 } 1576 } 1577 1578 if (smc->use_fallback) 1579 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len); 1580 else 1581 rc = smc_tx_sendmsg(smc, msg, len); 1582 out: 1583 release_sock(sk); 1584 return rc; 1585 } 1586 1587 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len, 1588 int flags) 1589 { 1590 struct sock *sk = sock->sk; 1591 struct smc_sock *smc; 1592 int rc = -ENOTCONN; 1593 1594 smc = smc_sk(sk); 1595 lock_sock(sk); 1596 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1597 /* socket was connected before, no more data to read */ 1598 rc = 0; 1599 goto out; 1600 } 1601 if ((sk->sk_state == SMC_INIT) || 1602 (sk->sk_state == SMC_LISTEN) || 1603 (sk->sk_state == SMC_CLOSED)) 1604 goto out; 1605 1606 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1607 rc = 0; 1608 goto out; 1609 } 1610 1611 if (smc->use_fallback) { 1612 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags); 1613 } else { 1614 msg->msg_namelen = 0; 1615 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags); 1616 } 1617 1618 out: 1619 release_sock(sk); 1620 return rc; 1621 } 1622 1623 static __poll_t smc_accept_poll(struct sock *parent) 1624 { 1625 struct smc_sock *isk = smc_sk(parent); 1626 __poll_t mask = 0; 1627 1628 spin_lock(&isk->accept_q_lock); 1629 if (!list_empty(&isk->accept_q)) 1630 mask = EPOLLIN | EPOLLRDNORM; 1631 spin_unlock(&isk->accept_q_lock); 1632 1633 return mask; 1634 } 1635 1636 static __poll_t smc_poll(struct file *file, struct socket *sock, 1637 poll_table *wait) 1638 { 1639 struct sock *sk = sock->sk; 1640 struct smc_sock *smc; 1641 __poll_t mask = 0; 1642 1643 if (!sk) 1644 return EPOLLNVAL; 1645 1646 smc = smc_sk(sock->sk); 1647 if (smc->use_fallback) { 1648 /* delegate to CLC child sock */ 1649 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait); 1650 sk->sk_err = smc->clcsock->sk->sk_err; 1651 } else { 1652 if (sk->sk_state != SMC_CLOSED) 1653 sock_poll_wait(file, sock, wait); 1654 if (sk->sk_err) 1655 mask |= EPOLLERR; 1656 if ((sk->sk_shutdown == SHUTDOWN_MASK) || 1657 (sk->sk_state == SMC_CLOSED)) 1658 mask |= EPOLLHUP; 1659 if (sk->sk_state == SMC_LISTEN) { 1660 /* woken up by sk_data_ready in smc_listen_work() */ 1661 mask |= smc_accept_poll(sk); 1662 } else if (smc->use_fallback) { /* as result of connect_work()*/ 1663 mask |= smc->clcsock->ops->poll(file, smc->clcsock, 1664 wait); 1665 sk->sk_err = smc->clcsock->sk->sk_err; 1666 } else { 1667 if ((sk->sk_state != SMC_INIT && 1668 atomic_read(&smc->conn.sndbuf_space)) || 1669 sk->sk_shutdown & SEND_SHUTDOWN) { 1670 mask |= EPOLLOUT | EPOLLWRNORM; 1671 } else { 1672 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk); 1673 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags); 1674 } 1675 if (atomic_read(&smc->conn.bytes_to_rcv)) 1676 mask |= EPOLLIN | EPOLLRDNORM; 1677 if (sk->sk_shutdown & RCV_SHUTDOWN) 1678 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP; 1679 if (sk->sk_state == SMC_APPCLOSEWAIT1) 1680 mask |= EPOLLIN; 1681 if (smc->conn.urg_state == SMC_URG_VALID) 1682 mask |= EPOLLPRI; 1683 } 1684 } 1685 1686 return mask; 1687 } 1688 1689 static int smc_shutdown(struct socket *sock, int how) 1690 { 1691 struct sock *sk = sock->sk; 1692 struct smc_sock *smc; 1693 int rc = -EINVAL; 1694 int rc1 = 0; 1695 1696 smc = smc_sk(sk); 1697 1698 if ((how < SHUT_RD) || (how > SHUT_RDWR)) 1699 return rc; 1700 1701 lock_sock(sk); 1702 1703 rc = -ENOTCONN; 1704 if ((sk->sk_state != SMC_ACTIVE) && 1705 (sk->sk_state != SMC_PEERCLOSEWAIT1) && 1706 (sk->sk_state != SMC_PEERCLOSEWAIT2) && 1707 (sk->sk_state != SMC_APPCLOSEWAIT1) && 1708 (sk->sk_state != SMC_APPCLOSEWAIT2) && 1709 (sk->sk_state != SMC_APPFINCLOSEWAIT)) 1710 goto out; 1711 if (smc->use_fallback) { 1712 rc = kernel_sock_shutdown(smc->clcsock, how); 1713 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown; 1714 if (sk->sk_shutdown == SHUTDOWN_MASK) 1715 sk->sk_state = SMC_CLOSED; 1716 goto out; 1717 } 1718 switch (how) { 1719 case SHUT_RDWR: /* shutdown in both directions */ 1720 rc = smc_close_active(smc); 1721 break; 1722 case SHUT_WR: 1723 rc = smc_close_shutdown_write(smc); 1724 break; 1725 case SHUT_RD: 1726 rc = 0; 1727 /* nothing more to do because peer is not involved */ 1728 break; 1729 } 1730 if (smc->clcsock) 1731 rc1 = kernel_sock_shutdown(smc->clcsock, how); 1732 /* map sock_shutdown_cmd constants to sk_shutdown value range */ 1733 sk->sk_shutdown |= how + 1; 1734 1735 out: 1736 release_sock(sk); 1737 return rc ? rc : rc1; 1738 } 1739 1740 static int smc_setsockopt(struct socket *sock, int level, int optname, 1741 char __user *optval, unsigned int optlen) 1742 { 1743 struct sock *sk = sock->sk; 1744 struct smc_sock *smc; 1745 int val, rc; 1746 1747 smc = smc_sk(sk); 1748 1749 /* generic setsockopts reaching us here always apply to the 1750 * CLC socket 1751 */ 1752 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname, 1753 optval, optlen); 1754 if (smc->clcsock->sk->sk_err) { 1755 sk->sk_err = smc->clcsock->sk->sk_err; 1756 sk->sk_error_report(sk); 1757 } 1758 1759 if (optlen < sizeof(int)) 1760 return -EINVAL; 1761 if (get_user(val, (int __user *)optval)) 1762 return -EFAULT; 1763 1764 lock_sock(sk); 1765 if (rc || smc->use_fallback) 1766 goto out; 1767 switch (optname) { 1768 case TCP_ULP: 1769 case TCP_FASTOPEN: 1770 case TCP_FASTOPEN_CONNECT: 1771 case TCP_FASTOPEN_KEY: 1772 case TCP_FASTOPEN_NO_COOKIE: 1773 /* option not supported by SMC */ 1774 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) { 1775 smc_switch_to_fallback(smc); 1776 smc->fallback_rsn = SMC_CLC_DECL_OPTUNSUPP; 1777 } else { 1778 rc = -EINVAL; 1779 } 1780 break; 1781 case TCP_NODELAY: 1782 if (sk->sk_state != SMC_INIT && 1783 sk->sk_state != SMC_LISTEN && 1784 sk->sk_state != SMC_CLOSED) { 1785 if (val) 1786 mod_delayed_work(system_wq, &smc->conn.tx_work, 1787 0); 1788 } 1789 break; 1790 case TCP_CORK: 1791 if (sk->sk_state != SMC_INIT && 1792 sk->sk_state != SMC_LISTEN && 1793 sk->sk_state != SMC_CLOSED) { 1794 if (!val) 1795 mod_delayed_work(system_wq, &smc->conn.tx_work, 1796 0); 1797 } 1798 break; 1799 case TCP_DEFER_ACCEPT: 1800 smc->sockopt_defer_accept = val; 1801 break; 1802 default: 1803 break; 1804 } 1805 out: 1806 release_sock(sk); 1807 1808 return rc; 1809 } 1810 1811 static int smc_getsockopt(struct socket *sock, int level, int optname, 1812 char __user *optval, int __user *optlen) 1813 { 1814 struct smc_sock *smc; 1815 1816 smc = smc_sk(sock->sk); 1817 /* socket options apply to the CLC socket */ 1818 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname, 1819 optval, optlen); 1820 } 1821 1822 static int smc_ioctl(struct socket *sock, unsigned int cmd, 1823 unsigned long arg) 1824 { 1825 union smc_host_cursor cons, urg; 1826 struct smc_connection *conn; 1827 struct smc_sock *smc; 1828 int answ; 1829 1830 smc = smc_sk(sock->sk); 1831 conn = &smc->conn; 1832 lock_sock(&smc->sk); 1833 if (smc->use_fallback) { 1834 if (!smc->clcsock) { 1835 release_sock(&smc->sk); 1836 return -EBADF; 1837 } 1838 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg); 1839 release_sock(&smc->sk); 1840 return answ; 1841 } 1842 switch (cmd) { 1843 case SIOCINQ: /* same as FIONREAD */ 1844 if (smc->sk.sk_state == SMC_LISTEN) { 1845 release_sock(&smc->sk); 1846 return -EINVAL; 1847 } 1848 if (smc->sk.sk_state == SMC_INIT || 1849 smc->sk.sk_state == SMC_CLOSED) 1850 answ = 0; 1851 else 1852 answ = atomic_read(&smc->conn.bytes_to_rcv); 1853 break; 1854 case SIOCOUTQ: 1855 /* output queue size (not send + not acked) */ 1856 if (smc->sk.sk_state == SMC_LISTEN) { 1857 release_sock(&smc->sk); 1858 return -EINVAL; 1859 } 1860 if (smc->sk.sk_state == SMC_INIT || 1861 smc->sk.sk_state == SMC_CLOSED) 1862 answ = 0; 1863 else 1864 answ = smc->conn.sndbuf_desc->len - 1865 atomic_read(&smc->conn.sndbuf_space); 1866 break; 1867 case SIOCOUTQNSD: 1868 /* output queue size (not send only) */ 1869 if (smc->sk.sk_state == SMC_LISTEN) { 1870 release_sock(&smc->sk); 1871 return -EINVAL; 1872 } 1873 if (smc->sk.sk_state == SMC_INIT || 1874 smc->sk.sk_state == SMC_CLOSED) 1875 answ = 0; 1876 else 1877 answ = smc_tx_prepared_sends(&smc->conn); 1878 break; 1879 case SIOCATMARK: 1880 if (smc->sk.sk_state == SMC_LISTEN) { 1881 release_sock(&smc->sk); 1882 return -EINVAL; 1883 } 1884 if (smc->sk.sk_state == SMC_INIT || 1885 smc->sk.sk_state == SMC_CLOSED) { 1886 answ = 0; 1887 } else { 1888 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn); 1889 smc_curs_copy(&urg, &conn->urg_curs, conn); 1890 answ = smc_curs_diff(conn->rmb_desc->len, 1891 &cons, &urg) == 1; 1892 } 1893 break; 1894 default: 1895 release_sock(&smc->sk); 1896 return -ENOIOCTLCMD; 1897 } 1898 release_sock(&smc->sk); 1899 1900 return put_user(answ, (int __user *)arg); 1901 } 1902 1903 static ssize_t smc_sendpage(struct socket *sock, struct page *page, 1904 int offset, size_t size, int flags) 1905 { 1906 struct sock *sk = sock->sk; 1907 struct smc_sock *smc; 1908 int rc = -EPIPE; 1909 1910 smc = smc_sk(sk); 1911 lock_sock(sk); 1912 if (sk->sk_state != SMC_ACTIVE) { 1913 release_sock(sk); 1914 goto out; 1915 } 1916 release_sock(sk); 1917 if (smc->use_fallback) 1918 rc = kernel_sendpage(smc->clcsock, page, offset, 1919 size, flags); 1920 else 1921 rc = sock_no_sendpage(sock, page, offset, size, flags); 1922 1923 out: 1924 return rc; 1925 } 1926 1927 /* Map the affected portions of the rmbe into an spd, note the number of bytes 1928 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor 1929 * updates till whenever a respective page has been fully processed. 1930 * Note that subsequent recv() calls have to wait till all splice() processing 1931 * completed. 1932 */ 1933 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos, 1934 struct pipe_inode_info *pipe, size_t len, 1935 unsigned int flags) 1936 { 1937 struct sock *sk = sock->sk; 1938 struct smc_sock *smc; 1939 int rc = -ENOTCONN; 1940 1941 smc = smc_sk(sk); 1942 lock_sock(sk); 1943 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) { 1944 /* socket was connected before, no more data to read */ 1945 rc = 0; 1946 goto out; 1947 } 1948 if (sk->sk_state == SMC_INIT || 1949 sk->sk_state == SMC_LISTEN || 1950 sk->sk_state == SMC_CLOSED) 1951 goto out; 1952 1953 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) { 1954 rc = 0; 1955 goto out; 1956 } 1957 1958 if (smc->use_fallback) { 1959 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos, 1960 pipe, len, flags); 1961 } else { 1962 if (*ppos) { 1963 rc = -ESPIPE; 1964 goto out; 1965 } 1966 if (flags & SPLICE_F_NONBLOCK) 1967 flags = MSG_DONTWAIT; 1968 else 1969 flags = 0; 1970 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags); 1971 } 1972 out: 1973 release_sock(sk); 1974 1975 return rc; 1976 } 1977 1978 /* must look like tcp */ 1979 static const struct proto_ops smc_sock_ops = { 1980 .family = PF_SMC, 1981 .owner = THIS_MODULE, 1982 .release = smc_release, 1983 .bind = smc_bind, 1984 .connect = smc_connect, 1985 .socketpair = sock_no_socketpair, 1986 .accept = smc_accept, 1987 .getname = smc_getname, 1988 .poll = smc_poll, 1989 .ioctl = smc_ioctl, 1990 .listen = smc_listen, 1991 .shutdown = smc_shutdown, 1992 .setsockopt = smc_setsockopt, 1993 .getsockopt = smc_getsockopt, 1994 .sendmsg = smc_sendmsg, 1995 .recvmsg = smc_recvmsg, 1996 .mmap = sock_no_mmap, 1997 .sendpage = smc_sendpage, 1998 .splice_read = smc_splice_read, 1999 }; 2000 2001 static int smc_create(struct net *net, struct socket *sock, int protocol, 2002 int kern) 2003 { 2004 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET; 2005 struct smc_sock *smc; 2006 struct sock *sk; 2007 int rc; 2008 2009 rc = -ESOCKTNOSUPPORT; 2010 if (sock->type != SOCK_STREAM) 2011 goto out; 2012 2013 rc = -EPROTONOSUPPORT; 2014 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6) 2015 goto out; 2016 2017 rc = -ENOBUFS; 2018 sock->ops = &smc_sock_ops; 2019 sk = smc_sock_alloc(net, sock, protocol); 2020 if (!sk) 2021 goto out; 2022 2023 /* create internal TCP socket for CLC handshake and fallback */ 2024 smc = smc_sk(sk); 2025 smc->use_fallback = false; /* assume rdma capability first */ 2026 smc->fallback_rsn = 0; 2027 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP, 2028 &smc->clcsock); 2029 if (rc) { 2030 sk_common_release(sk); 2031 goto out; 2032 } 2033 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE); 2034 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE); 2035 2036 out: 2037 return rc; 2038 } 2039 2040 static const struct net_proto_family smc_sock_family_ops = { 2041 .family = PF_SMC, 2042 .owner = THIS_MODULE, 2043 .create = smc_create, 2044 }; 2045 2046 unsigned int smc_net_id; 2047 2048 static __net_init int smc_net_init(struct net *net) 2049 { 2050 return smc_pnet_net_init(net); 2051 } 2052 2053 static void __net_exit smc_net_exit(struct net *net) 2054 { 2055 smc_pnet_net_exit(net); 2056 } 2057 2058 static struct pernet_operations smc_net_ops = { 2059 .init = smc_net_init, 2060 .exit = smc_net_exit, 2061 .id = &smc_net_id, 2062 .size = sizeof(struct smc_net), 2063 }; 2064 2065 static int __init smc_init(void) 2066 { 2067 int rc; 2068 2069 rc = register_pernet_subsys(&smc_net_ops); 2070 if (rc) 2071 return rc; 2072 2073 rc = smc_pnet_init(); 2074 if (rc) 2075 goto out_pernet_subsys; 2076 2077 rc = smc_core_init(); 2078 if (rc) { 2079 pr_err("%s: smc_core_init fails with %d\n", __func__, rc); 2080 goto out_pnet; 2081 } 2082 2083 rc = smc_llc_init(); 2084 if (rc) { 2085 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc); 2086 goto out_core; 2087 } 2088 2089 rc = smc_cdc_init(); 2090 if (rc) { 2091 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc); 2092 goto out_core; 2093 } 2094 2095 rc = proto_register(&smc_proto, 1); 2096 if (rc) { 2097 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc); 2098 goto out_core; 2099 } 2100 2101 rc = proto_register(&smc_proto6, 1); 2102 if (rc) { 2103 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc); 2104 goto out_proto; 2105 } 2106 2107 rc = sock_register(&smc_sock_family_ops); 2108 if (rc) { 2109 pr_err("%s: sock_register fails with %d\n", __func__, rc); 2110 goto out_proto6; 2111 } 2112 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht); 2113 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht); 2114 2115 rc = smc_ib_register_client(); 2116 if (rc) { 2117 pr_err("%s: ib_register fails with %d\n", __func__, rc); 2118 goto out_sock; 2119 } 2120 2121 static_branch_enable(&tcp_have_smc); 2122 return 0; 2123 2124 out_sock: 2125 sock_unregister(PF_SMC); 2126 out_proto6: 2127 proto_unregister(&smc_proto6); 2128 out_proto: 2129 proto_unregister(&smc_proto); 2130 out_core: 2131 smc_core_exit(); 2132 out_pnet: 2133 smc_pnet_exit(); 2134 out_pernet_subsys: 2135 unregister_pernet_subsys(&smc_net_ops); 2136 2137 return rc; 2138 } 2139 2140 static void __exit smc_exit(void) 2141 { 2142 static_branch_disable(&tcp_have_smc); 2143 sock_unregister(PF_SMC); 2144 smc_core_exit(); 2145 smc_ib_unregister_client(); 2146 proto_unregister(&smc_proto6); 2147 proto_unregister(&smc_proto); 2148 smc_pnet_exit(); 2149 unregister_pernet_subsys(&smc_net_ops); 2150 rcu_barrier(); 2151 } 2152 2153 module_init(smc_init); 2154 module_exit(smc_exit); 2155 2156 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>"); 2157 MODULE_DESCRIPTION("smc socket address family"); 2158 MODULE_LICENSE("GPL"); 2159 MODULE_ALIAS_NETPROTO(PF_SMC); 2160