1 // SPDX-License-Identifier: GPL-2.0-only
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
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
10 * Initial restrictions:
11 * - support for alternate links postponed
13 * Copyright IBM Corp. 2016, 2018
15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
16 * based on prototype from Frank Blaschka
19 #define KMSG_COMPONENT "smc"
20 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
22 #include <linux/module.h>
23 #include <linux/socket.h>
24 #include <linux/workqueue.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 #include <linux/splice.h>
35 #include <asm/ioctls.h>
37 #include <net/net_namespace.h>
38 #include <net/netns/generic.h>
39 #include "smc_netns.h"
49 #include "smc_netlink.h"
52 #include "smc_close.h"
53 #include "smc_stats.h"
54 #include "smc_tracepoint.h"
55 #include "smc_sysctl.h"
57 static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group
60 static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group
64 static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */
65 struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
66 struct workqueue_struct *smc_close_wq; /* wq for close work */
68 static void smc_tcp_listen_work(struct work_struct *);
69 static void smc_connect_work(struct work_struct *);
71 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb)
73 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
79 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
80 &smc_gen_nl_family, NLM_F_MULTI,
81 SMC_NETLINK_DUMP_HS_LIMITATION);
85 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED,
86 sock_net(skb->sk)->smc.limit_smc_hs))
89 genlmsg_end(skb, hdr);
94 genlmsg_cancel(skb, hdr);
98 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
100 sock_net(skb->sk)->smc.limit_smc_hs = true;
104 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
106 sock_net(skb->sk)->smc.limit_smc_hs = false;
110 static void smc_set_keepalive(struct sock *sk, int val)
112 struct smc_sock *smc = smc_sk(sk);
114 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
117 static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk,
119 struct request_sock *req,
120 struct dst_entry *dst,
121 struct request_sock *req_unhash,
124 struct smc_sock *smc;
127 smc = smc_clcsock_user_data(sk);
129 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) >
130 sk->sk_max_ack_backlog)
133 if (sk_acceptq_is_full(&smc->sk)) {
134 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
138 /* passthrough to original syn recv sock fct */
139 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash,
141 /* child must not inherit smc or its ops */
143 rcu_assign_sk_user_data(child, NULL);
145 /* v4-mapped sockets don't inherit parent ops. Don't restore. */
146 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops)
147 inet_csk(child)->icsk_af_ops = smc->ori_af_ops;
157 static bool smc_hs_congested(const struct sock *sk)
159 const struct smc_sock *smc;
161 smc = smc_clcsock_user_data(sk);
166 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq))
172 static struct smc_hashinfo smc_v4_hashinfo = {
173 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
176 static struct smc_hashinfo smc_v6_hashinfo = {
177 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
180 int smc_hash_sk(struct sock *sk)
182 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
183 struct hlist_head *head;
187 write_lock_bh(&h->lock);
188 sk_add_node(sk, head);
189 write_unlock_bh(&h->lock);
190 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
194 EXPORT_SYMBOL_GPL(smc_hash_sk);
196 void smc_unhash_sk(struct sock *sk)
198 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
200 write_lock_bh(&h->lock);
201 if (sk_del_node_init(sk))
202 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
203 write_unlock_bh(&h->lock);
205 EXPORT_SYMBOL_GPL(smc_unhash_sk);
207 /* This will be called before user really release sock_lock. So do the
208 * work which we didn't do because of user hold the sock_lock in the
211 static void smc_release_cb(struct sock *sk)
213 struct smc_sock *smc = smc_sk(sk);
215 if (smc->conn.tx_in_release_sock) {
216 smc_tx_pending(&smc->conn);
217 smc->conn.tx_in_release_sock = false;
221 struct proto smc_proto = {
223 .owner = THIS_MODULE,
224 .keepalive = smc_set_keepalive,
226 .unhash = smc_unhash_sk,
227 .release_cb = smc_release_cb,
228 .obj_size = sizeof(struct smc_sock),
229 .h.smc_hash = &smc_v4_hashinfo,
230 .slab_flags = SLAB_TYPESAFE_BY_RCU,
232 EXPORT_SYMBOL_GPL(smc_proto);
234 struct proto smc_proto6 = {
236 .owner = THIS_MODULE,
237 .keepalive = smc_set_keepalive,
239 .unhash = smc_unhash_sk,
240 .release_cb = smc_release_cb,
241 .obj_size = sizeof(struct smc_sock),
242 .h.smc_hash = &smc_v6_hashinfo,
243 .slab_flags = SLAB_TYPESAFE_BY_RCU,
245 EXPORT_SYMBOL_GPL(smc_proto6);
247 static void smc_fback_restore_callbacks(struct smc_sock *smc)
249 struct sock *clcsk = smc->clcsock->sk;
251 write_lock_bh(&clcsk->sk_callback_lock);
252 clcsk->sk_user_data = NULL;
254 smc_clcsock_restore_cb(&clcsk->sk_state_change, &smc->clcsk_state_change);
255 smc_clcsock_restore_cb(&clcsk->sk_data_ready, &smc->clcsk_data_ready);
256 smc_clcsock_restore_cb(&clcsk->sk_write_space, &smc->clcsk_write_space);
257 smc_clcsock_restore_cb(&clcsk->sk_error_report, &smc->clcsk_error_report);
259 write_unlock_bh(&clcsk->sk_callback_lock);
262 static void smc_restore_fallback_changes(struct smc_sock *smc)
264 if (smc->clcsock->file) { /* non-accepted sockets have no file yet */
265 smc->clcsock->file->private_data = smc->sk.sk_socket;
266 smc->clcsock->file = NULL;
267 smc_fback_restore_callbacks(smc);
271 static int __smc_release(struct smc_sock *smc)
273 struct sock *sk = &smc->sk;
276 if (!smc->use_fallback) {
277 rc = smc_close_active(smc);
278 sock_set_flag(sk, SOCK_DEAD);
279 sk->sk_shutdown |= SHUTDOWN_MASK;
281 if (sk->sk_state != SMC_CLOSED) {
282 if (sk->sk_state != SMC_LISTEN &&
283 sk->sk_state != SMC_INIT)
284 sock_put(sk); /* passive closing */
285 if (sk->sk_state == SMC_LISTEN) {
286 /* wake up clcsock accept */
287 rc = kernel_sock_shutdown(smc->clcsock,
290 sk->sk_state = SMC_CLOSED;
291 sk->sk_state_change(sk);
293 smc_restore_fallback_changes(smc);
296 sk->sk_prot->unhash(sk);
298 if (sk->sk_state == SMC_CLOSED) {
301 smc_clcsock_release(smc);
304 if (!smc->use_fallback)
305 smc_conn_free(&smc->conn);
311 static int smc_release(struct socket *sock)
313 struct sock *sk = sock->sk;
314 struct smc_sock *smc;
315 int old_state, rc = 0;
320 sock_hold(sk); /* sock_put below */
323 old_state = sk->sk_state;
325 /* cleanup for a dangling non-blocking connect */
326 if (smc->connect_nonblock && old_state == SMC_INIT)
327 tcp_abort(smc->clcsock->sk, ECONNABORTED);
329 if (cancel_work_sync(&smc->connect_work))
330 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */
332 if (sk->sk_state == SMC_LISTEN)
333 /* smc_close_non_accepted() is called and acquires
334 * sock lock for child sockets again
336 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
340 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE &&
342 smc_close_active_abort(smc);
344 rc = __smc_release(smc);
351 sock_put(sk); /* sock_hold above */
352 sock_put(sk); /* final sock_put */
357 static void smc_destruct(struct sock *sk)
359 if (sk->sk_state != SMC_CLOSED)
361 if (!sock_flag(sk, SOCK_DEAD))
365 static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
368 struct smc_sock *smc;
372 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
373 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
377 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
378 sk->sk_state = SMC_INIT;
379 sk->sk_destruct = smc_destruct;
380 sk->sk_protocol = protocol;
381 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem));
382 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem));
384 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
385 INIT_WORK(&smc->connect_work, smc_connect_work);
386 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
387 INIT_LIST_HEAD(&smc->accept_q);
388 spin_lock_init(&smc->accept_q_lock);
389 spin_lock_init(&smc->conn.send_lock);
390 sk->sk_prot->hash(sk);
391 mutex_init(&smc->clcsock_release_lock);
392 smc_init_saved_callbacks(smc);
397 static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
400 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
401 struct sock *sk = sock->sk;
402 struct smc_sock *smc;
407 /* replicate tests from inet_bind(), to be safe wrt. future changes */
409 if (addr_len < sizeof(struct sockaddr_in))
413 if (addr->sin_family != AF_INET &&
414 addr->sin_family != AF_INET6 &&
415 addr->sin_family != AF_UNSPEC)
417 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
418 if (addr->sin_family == AF_UNSPEC &&
419 addr->sin_addr.s_addr != htonl(INADDR_ANY))
424 /* Check if socket is already active */
426 if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
429 smc->clcsock->sk->sk_reuse = sk->sk_reuse;
430 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport;
431 rc = kernel_bind(smc->clcsock, uaddr, addr_len);
439 /* copy only relevant settings and flags of SOL_SOCKET level from smc to
440 * clc socket (since smc is not called for these options from net/core)
443 #define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
444 (1UL << SOCK_KEEPOPEN) | \
445 (1UL << SOCK_LINGER) | \
446 (1UL << SOCK_BROADCAST) | \
447 (1UL << SOCK_TIMESTAMP) | \
448 (1UL << SOCK_DBG) | \
449 (1UL << SOCK_RCVTSTAMP) | \
450 (1UL << SOCK_RCVTSTAMPNS) | \
451 (1UL << SOCK_LOCALROUTE) | \
452 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
453 (1UL << SOCK_RXQ_OVFL) | \
454 (1UL << SOCK_WIFI_STATUS) | \
455 (1UL << SOCK_NOFCS) | \
456 (1UL << SOCK_FILTER_LOCKED) | \
457 (1UL << SOCK_TSTAMP_NEW))
459 /* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */
460 static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk,
463 struct net *nnet = sock_net(nsk);
465 nsk->sk_userlocks = osk->sk_userlocks;
466 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) {
467 nsk->sk_sndbuf = osk->sk_sndbuf;
469 if (mask == SK_FLAGS_SMC_TO_CLC)
470 WRITE_ONCE(nsk->sk_sndbuf,
471 READ_ONCE(nnet->ipv4.sysctl_tcp_wmem[1]));
473 WRITE_ONCE(nsk->sk_sndbuf,
474 2 * READ_ONCE(nnet->smc.sysctl_wmem));
476 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) {
477 nsk->sk_rcvbuf = osk->sk_rcvbuf;
479 if (mask == SK_FLAGS_SMC_TO_CLC)
480 WRITE_ONCE(nsk->sk_rcvbuf,
481 READ_ONCE(nnet->ipv4.sysctl_tcp_rmem[1]));
483 WRITE_ONCE(nsk->sk_rcvbuf,
484 2 * READ_ONCE(nnet->smc.sysctl_rmem));
488 static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
491 /* options we don't get control via setsockopt for */
492 nsk->sk_type = osk->sk_type;
493 nsk->sk_sndtimeo = osk->sk_sndtimeo;
494 nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
495 nsk->sk_mark = READ_ONCE(osk->sk_mark);
496 nsk->sk_priority = osk->sk_priority;
497 nsk->sk_rcvlowat = osk->sk_rcvlowat;
498 nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
499 nsk->sk_err = osk->sk_err;
501 nsk->sk_flags &= ~mask;
502 nsk->sk_flags |= osk->sk_flags & mask;
504 smc_adjust_sock_bufsizes(nsk, osk, mask);
507 static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
509 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
512 #define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
513 (1UL << SOCK_KEEPOPEN) | \
514 (1UL << SOCK_LINGER) | \
516 /* copy only settings and flags relevant for smc from clc to smc socket */
517 static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
519 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
522 /* register the new vzalloced sndbuf on all links */
523 static int smcr_lgr_reg_sndbufs(struct smc_link *link,
524 struct smc_buf_desc *snd_desc)
526 struct smc_link_group *lgr = link->lgr;
529 if (!snd_desc->is_vm)
532 /* protect against parallel smcr_link_reg_buf() */
533 down_write(&lgr->llc_conf_mutex);
534 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
535 if (!smc_link_active(&lgr->lnk[i]))
537 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc);
541 up_write(&lgr->llc_conf_mutex);
545 /* register the new rmb on all links */
546 static int smcr_lgr_reg_rmbs(struct smc_link *link,
547 struct smc_buf_desc *rmb_desc)
549 struct smc_link_group *lgr = link->lgr;
550 bool do_slow = false;
553 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
557 down_read(&lgr->llc_conf_mutex);
558 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
559 if (!smc_link_active(&lgr->lnk[i]))
561 if (!rmb_desc->is_reg_mr[link->link_idx]) {
562 up_read(&lgr->llc_conf_mutex);
566 /* mr register already */
570 /* protect against parallel smc_llc_cli_rkey_exchange() and
571 * parallel smcr_link_reg_buf()
573 down_write(&lgr->llc_conf_mutex);
574 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
575 if (!smc_link_active(&lgr->lnk[i]))
577 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc);
582 /* exchange confirm_rkey msg with peer */
583 rc = smc_llc_do_confirm_rkey(link, rmb_desc);
588 rmb_desc->is_conf_rkey = true;
590 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex);
591 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
595 static int smcr_clnt_conf_first_link(struct smc_sock *smc)
597 struct smc_link *link = smc->conn.lnk;
598 struct smc_llc_qentry *qentry;
601 /* receive CONFIRM LINK request from server over RoCE fabric */
602 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
603 SMC_LLC_CONFIRM_LINK);
605 struct smc_clc_msg_decline dclc;
607 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
608 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
609 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
611 smc_llc_save_peer_uid(qentry);
612 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
613 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
615 return SMC_CLC_DECL_RMBE_EC;
617 rc = smc_ib_modify_qp_rts(link);
619 return SMC_CLC_DECL_ERR_RDYLNK;
621 smc_wr_remember_qp_attr(link);
623 /* reg the sndbuf if it was vzalloced */
624 if (smc->conn.sndbuf_desc->is_vm) {
625 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
626 return SMC_CLC_DECL_ERR_REGBUF;
630 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
631 return SMC_CLC_DECL_ERR_REGBUF;
633 /* confirm_rkey is implicit on 1st contact */
634 smc->conn.rmb_desc->is_conf_rkey = true;
636 /* send CONFIRM LINK response over RoCE fabric */
637 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
639 return SMC_CLC_DECL_TIMEOUT_CL;
641 smc_llc_link_active(link);
642 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
644 /* optional 2nd link, receive ADD LINK request from server */
645 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
648 struct smc_clc_msg_decline dclc;
650 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
651 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
653 rc = 0; /* no DECLINE received, go with one link */
656 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
657 smc_llc_cli_add_link(link, qentry);
661 static bool smc_isascii(char *hostname)
665 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
666 if (!isascii(hostname[i]))
671 static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
672 struct smc_clc_msg_accept_confirm *clc)
674 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
675 (struct smc_clc_msg_accept_confirm_v2 *)clc;
676 struct smc_clc_first_contact_ext *fce;
679 if (clc->hdr.version == SMC_V1 ||
680 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
683 if (smc->conn.lgr->is_smcd) {
684 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid,
686 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2,
689 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid,
691 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2,
694 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len);
695 smc->conn.lgr->peer_os = fce->os_type;
696 smc->conn.lgr->peer_smc_release = fce->release;
697 if (smc_isascii(fce->hostname))
698 memcpy(smc->conn.lgr->peer_hostname, fce->hostname,
699 SMC_MAX_HOSTNAME_LEN);
702 static void smcr_conn_save_peer_info(struct smc_sock *smc,
703 struct smc_clc_msg_accept_confirm *clc)
705 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
707 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx;
708 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token);
709 smc->conn.peer_rmbe_size = bufsize;
710 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
711 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
714 static void smcd_conn_save_peer_info(struct smc_sock *smc,
715 struct smc_clc_msg_accept_confirm *clc)
717 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
719 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
720 smc->conn.peer_token = clc->d0.token;
721 /* msg header takes up space in the buffer */
722 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
723 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
724 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
727 static void smc_conn_save_peer_info(struct smc_sock *smc,
728 struct smc_clc_msg_accept_confirm *clc)
730 if (smc->conn.lgr->is_smcd)
731 smcd_conn_save_peer_info(smc, clc);
733 smcr_conn_save_peer_info(smc, clc);
734 smc_conn_save_peer_info_fce(smc, clc);
737 static void smc_link_save_peer_info(struct smc_link *link,
738 struct smc_clc_msg_accept_confirm *clc,
739 struct smc_init_info *ini)
741 link->peer_qpn = ntoh24(clc->r0.qpn);
742 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE);
743 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac));
744 link->peer_psn = ntoh24(clc->r0.psn);
745 link->peer_mtu = clc->r0.qp_mtu;
748 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
749 struct smc_stats_fback *fback_arr)
753 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) {
754 if (fback_arr[cnt].fback_code == smc->fallback_rsn) {
755 fback_arr[cnt].count++;
758 if (!fback_arr[cnt].fback_code) {
759 fback_arr[cnt].fback_code = smc->fallback_rsn;
760 fback_arr[cnt].count++;
766 static void smc_stat_fallback(struct smc_sock *smc)
768 struct net *net = sock_net(&smc->sk);
770 mutex_lock(&net->smc.mutex_fback_rsn);
771 if (smc->listen_smc) {
772 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv);
773 net->smc.fback_rsn->srv_fback_cnt++;
775 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
776 net->smc.fback_rsn->clnt_fback_cnt++;
778 mutex_unlock(&net->smc.mutex_fback_rsn);
781 /* must be called under rcu read lock */
782 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
784 struct socket_wq *wq;
787 wq = rcu_dereference(smc->sk.sk_wq);
788 if (!skwq_has_sleeper(wq))
791 /* wake up smc sk->sk_wq */
793 /* sk_state_change */
794 wake_up_interruptible_all(&wq->wait);
796 flags = key_to_poll(key);
797 if (flags & (EPOLLIN | EPOLLOUT))
798 /* sk_data_ready or sk_write_space */
799 wake_up_interruptible_sync_poll(&wq->wait, flags);
800 else if (flags & EPOLLERR)
801 /* sk_error_report */
802 wake_up_interruptible_poll(&wq->wait, flags);
806 static int smc_fback_mark_woken(wait_queue_entry_t *wait,
807 unsigned int mode, int sync, void *key)
809 struct smc_mark_woken *mark =
810 container_of(wait, struct smc_mark_woken, wait_entry);
817 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
818 void (*clcsock_callback)(struct sock *sk))
820 struct smc_mark_woken mark = { .woken = false };
821 struct socket_wq *wq;
823 init_waitqueue_func_entry(&mark.wait_entry,
824 smc_fback_mark_woken);
826 wq = rcu_dereference(clcsk->sk_wq);
829 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
830 clcsock_callback(clcsk);
831 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
834 smc_fback_wakeup_waitqueue(smc, mark.key);
839 static void smc_fback_state_change(struct sock *clcsk)
841 struct smc_sock *smc;
843 read_lock_bh(&clcsk->sk_callback_lock);
844 smc = smc_clcsock_user_data(clcsk);
846 smc_fback_forward_wakeup(smc, clcsk,
847 smc->clcsk_state_change);
848 read_unlock_bh(&clcsk->sk_callback_lock);
851 static void smc_fback_data_ready(struct sock *clcsk)
853 struct smc_sock *smc;
855 read_lock_bh(&clcsk->sk_callback_lock);
856 smc = smc_clcsock_user_data(clcsk);
858 smc_fback_forward_wakeup(smc, clcsk,
859 smc->clcsk_data_ready);
860 read_unlock_bh(&clcsk->sk_callback_lock);
863 static void smc_fback_write_space(struct sock *clcsk)
865 struct smc_sock *smc;
867 read_lock_bh(&clcsk->sk_callback_lock);
868 smc = smc_clcsock_user_data(clcsk);
870 smc_fback_forward_wakeup(smc, clcsk,
871 smc->clcsk_write_space);
872 read_unlock_bh(&clcsk->sk_callback_lock);
875 static void smc_fback_error_report(struct sock *clcsk)
877 struct smc_sock *smc;
879 read_lock_bh(&clcsk->sk_callback_lock);
880 smc = smc_clcsock_user_data(clcsk);
882 smc_fback_forward_wakeup(smc, clcsk,
883 smc->clcsk_error_report);
884 read_unlock_bh(&clcsk->sk_callback_lock);
887 static void smc_fback_replace_callbacks(struct smc_sock *smc)
889 struct sock *clcsk = smc->clcsock->sk;
891 write_lock_bh(&clcsk->sk_callback_lock);
892 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
894 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change,
895 &smc->clcsk_state_change);
896 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready,
897 &smc->clcsk_data_ready);
898 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space,
899 &smc->clcsk_write_space);
900 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report,
901 &smc->clcsk_error_report);
903 write_unlock_bh(&clcsk->sk_callback_lock);
906 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
910 mutex_lock(&smc->clcsock_release_lock);
916 smc->use_fallback = true;
917 smc->fallback_rsn = reason_code;
918 smc_stat_fallback(smc);
919 trace_smc_switch_to_fallback(smc, reason_code);
920 if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
921 smc->clcsock->file = smc->sk.sk_socket->file;
922 smc->clcsock->file->private_data = smc->clcsock;
923 smc->clcsock->wq.fasync_list =
924 smc->sk.sk_socket->wq.fasync_list;
926 /* There might be some wait entries remaining
927 * in smc sk->sk_wq and they should be woken up
928 * as clcsock's wait queue is woken up.
930 smc_fback_replace_callbacks(smc);
933 mutex_unlock(&smc->clcsock_release_lock);
937 /* fall back during connect */
938 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
940 struct net *net = sock_net(&smc->sk);
943 rc = smc_switch_to_fallback(smc, reason_code);
944 if (rc) { /* fallback fails */
945 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
946 if (smc->sk.sk_state == SMC_INIT)
947 sock_put(&smc->sk); /* passive closing */
950 smc_copy_sock_settings_to_clc(smc);
951 smc->connect_nonblock = 0;
952 if (smc->sk.sk_state == SMC_INIT)
953 smc->sk.sk_state = SMC_ACTIVE;
957 /* decline and fall back during connect */
958 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
961 struct net *net = sock_net(&smc->sk);
964 if (reason_code < 0) { /* error, fallback is not possible */
965 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
966 if (smc->sk.sk_state == SMC_INIT)
967 sock_put(&smc->sk); /* passive closing */
970 if (reason_code != SMC_CLC_DECL_PEERDECL) {
971 rc = smc_clc_send_decline(smc, reason_code, version);
973 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
974 if (smc->sk.sk_state == SMC_INIT)
975 sock_put(&smc->sk); /* passive closing */
979 return smc_connect_fallback(smc, reason_code);
982 static void smc_conn_abort(struct smc_sock *smc, int local_first)
984 struct smc_connection *conn = &smc->conn;
985 struct smc_link_group *lgr = conn->lgr;
986 bool lgr_valid = false;
988 if (smc_conn_lgr_valid(conn))
992 if (local_first && lgr_valid)
993 smc_lgr_cleanup_early(lgr);
996 /* check if there is a rdma device available for this connection. */
997 /* called for connect and listen */
998 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
1000 /* PNET table look up: search active ib_device and port
1001 * within same PNETID that also contains the ethernet device
1002 * used for the internal TCP socket
1004 smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
1005 if (!ini->check_smcrv2 && !ini->ib_dev)
1006 return SMC_CLC_DECL_NOSMCRDEV;
1007 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1008 return SMC_CLC_DECL_NOSMCRDEV;
1012 /* check if there is an ISM device available for this connection. */
1013 /* called for connect and listen */
1014 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1016 /* Find ISM device with same PNETID as connecting interface */
1017 smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1018 if (!ini->ism_dev[0])
1019 return SMC_CLC_DECL_NOSMCDDEV;
1021 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1025 /* is chid unique for the ism devices that are already determined? */
1026 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1029 int i = (!ini->ism_dev[0]) ? 1 : 0;
1031 for (; i < cnt; i++)
1032 if (ini->ism_chid[i] == chid)
1037 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1038 * PNETID matching net_device)
1040 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1041 struct smc_init_info *ini)
1043 int rc = SMC_CLC_DECL_NOSMCDDEV;
1044 struct smcd_dev *smcd;
1048 if (smcd_indicated(ini->smc_type_v1))
1049 rc = 0; /* already initialized for V1 */
1050 mutex_lock(&smcd_dev_list.mutex);
1051 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1052 if (smcd->going_away || smcd == ini->ism_dev[0])
1054 chid = smc_ism_get_chid(smcd);
1055 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1057 if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1058 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1059 ini->ism_dev[i] = smcd;
1060 ini->ism_chid[i] = chid;
1061 ini->is_smcd = true;
1064 if (i > SMC_MAX_ISM_DEVS)
1068 mutex_unlock(&smcd_dev_list.mutex);
1069 ini->ism_offered_cnt = i - 1;
1070 if (!ini->ism_dev[0] && !ini->ism_dev[1])
1071 ini->smcd_version = 0;
1076 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
1077 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
1078 struct smc_init_info *ini)
1080 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1081 return SMC_CLC_DECL_ISMVLANERR;
1085 static int smc_find_proposal_devices(struct smc_sock *smc,
1086 struct smc_init_info *ini)
1090 /* check if there is an ism device available */
1091 if (!(ini->smcd_version & SMC_V1) ||
1092 smc_find_ism_device(smc, ini) ||
1093 smc_connect_ism_vlan_setup(smc, ini))
1094 ini->smcd_version &= ~SMC_V1;
1095 /* else ISM V1 is supported for this connection */
1097 /* check if there is an rdma device available */
1098 if (!(ini->smcr_version & SMC_V1) ||
1099 smc_find_rdma_device(smc, ini))
1100 ini->smcr_version &= ~SMC_V1;
1101 /* else RDMA is supported for this connection */
1103 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1104 ini->smcr_version & SMC_V1);
1106 /* check if there is an ism v2 device available */
1107 if (!(ini->smcd_version & SMC_V2) ||
1108 !smc_ism_is_v2_capable() ||
1109 smc_find_ism_v2_device_clnt(smc, ini))
1110 ini->smcd_version &= ~SMC_V2;
1112 /* check if there is an rdma v2 device available */
1113 ini->check_smcrv2 = true;
1114 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1115 if (!(ini->smcr_version & SMC_V2) ||
1116 smc->clcsock->sk->sk_family != AF_INET ||
1117 !smc_clc_ueid_count() ||
1118 smc_find_rdma_device(smc, ini))
1119 ini->smcr_version &= ~SMC_V2;
1120 ini->check_smcrv2 = false;
1122 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1123 ini->smcr_version & SMC_V2);
1125 /* if neither ISM nor RDMA are supported, fallback */
1126 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1127 rc = SMC_CLC_DECL_NOSMCDEV;
1132 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1133 * used, the VLAN ID will be registered again during the connection setup.
1135 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc,
1136 struct smc_init_info *ini)
1138 if (!smcd_indicated(ini->smc_type_v1))
1140 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1141 return SMC_CLC_DECL_CNFERR;
1145 #define SMC_CLC_MAX_ACCEPT_LEN \
1146 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \
1147 sizeof(struct smc_clc_first_contact_ext) + \
1148 sizeof(struct smc_clc_msg_trail))
1150 /* CLC handshake during connect */
1151 static int smc_connect_clc(struct smc_sock *smc,
1152 struct smc_clc_msg_accept_confirm_v2 *aclc2,
1153 struct smc_init_info *ini)
1157 /* do inband token exchange */
1158 rc = smc_clc_send_proposal(smc, ini);
1161 /* receive SMC Accept CLC message */
1162 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN,
1163 SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1166 void smc_fill_gid_list(struct smc_link_group *lgr,
1167 struct smc_gidlist *gidlist,
1168 struct smc_ib_device *known_dev, u8 *known_gid)
1170 struct smc_init_info *alt_ini = NULL;
1172 memset(gidlist, 0, sizeof(*gidlist));
1173 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1175 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1179 alt_ini->vlan_id = lgr->vlan_id;
1180 alt_ini->check_smcrv2 = true;
1181 alt_ini->smcrv2.saddr = lgr->saddr;
1182 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1184 if (!alt_ini->smcrv2.ib_dev_v2)
1187 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1194 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1195 struct smc_clc_msg_accept_confirm *aclc,
1196 struct smc_init_info *ini)
1198 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1199 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1200 struct smc_clc_first_contact_ext *fce =
1201 (struct smc_clc_first_contact_ext *)
1202 (((u8 *)clc_v2) + sizeof(*clc_v2));
1204 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1207 if (fce->v2_direct) {
1208 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1209 ini->smcrv2.uses_gateway = false;
1211 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr,
1212 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1213 ini->smcrv2.nexthop_mac,
1214 &ini->smcrv2.uses_gateway))
1215 return SMC_CLC_DECL_NOROUTE;
1216 if (!ini->smcrv2.uses_gateway) {
1217 /* mismatch: peer claims indirect, but its direct */
1218 return SMC_CLC_DECL_NOINDIRECT;
1224 /* setup for RDMA connection of client */
1225 static int smc_connect_rdma(struct smc_sock *smc,
1226 struct smc_clc_msg_accept_confirm *aclc,
1227 struct smc_init_info *ini)
1229 int i, reason_code = 0;
1230 struct smc_link *link;
1233 ini->is_smcd = false;
1234 ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1235 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1236 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1237 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1238 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1240 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1244 mutex_lock(&smc_client_lgr_pending);
1245 reason_code = smc_conn_create(smc, ini);
1247 mutex_unlock(&smc_client_lgr_pending);
1251 smc_conn_save_peer_info(smc, aclc);
1253 if (ini->first_contact_local) {
1254 link = smc->conn.lnk;
1256 /* set link that was assigned by server */
1258 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1259 struct smc_link *l = &smc->conn.lgr->lnk[i];
1261 if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1262 !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1264 (aclc->hdr.version > SMC_V1 ||
1265 !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1266 sizeof(l->peer_mac)))) {
1272 reason_code = SMC_CLC_DECL_NOSRVLINK;
1275 smc_switch_link_and_count(&smc->conn, link);
1278 /* create send buffer and rmb */
1279 if (smc_buf_create(smc, false)) {
1280 reason_code = SMC_CLC_DECL_MEM;
1284 if (ini->first_contact_local)
1285 smc_link_save_peer_info(link, aclc, ini);
1287 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1288 reason_code = SMC_CLC_DECL_ERR_RTOK;
1292 smc_close_init(smc);
1295 if (ini->first_contact_local) {
1296 if (smc_ib_ready_link(link)) {
1297 reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1301 /* reg sendbufs if they were vzalloced */
1302 if (smc->conn.sndbuf_desc->is_vm) {
1303 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) {
1304 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1308 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1309 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1314 if (aclc->hdr.version > SMC_V1) {
1315 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1316 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1318 eid = clc_v2->r1.eid;
1319 if (ini->first_contact_local)
1320 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1321 link->smcibdev, link->gid);
1324 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1325 aclc->hdr.version, eid, ini);
1331 if (ini->first_contact_local) {
1332 /* QP confirmation over RoCE fabric */
1333 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1334 reason_code = smcr_clnt_conf_first_link(smc);
1335 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1339 mutex_unlock(&smc_client_lgr_pending);
1341 smc_copy_sock_settings_to_clc(smc);
1342 smc->connect_nonblock = 0;
1343 if (smc->sk.sk_state == SMC_INIT)
1344 smc->sk.sk_state = SMC_ACTIVE;
1348 smc_conn_abort(smc, ini->first_contact_local);
1349 mutex_unlock(&smc_client_lgr_pending);
1350 smc->connect_nonblock = 0;
1355 /* The server has chosen one of the proposed ISM devices for the communication.
1356 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1359 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc,
1360 struct smc_init_info *ini)
1364 for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1365 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1366 ini->ism_selected = i;
1374 /* setup for ISM connection of client */
1375 static int smc_connect_ism(struct smc_sock *smc,
1376 struct smc_clc_msg_accept_confirm *aclc,
1377 struct smc_init_info *ini)
1382 ini->is_smcd = true;
1383 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1385 if (aclc->hdr.version == SMC_V2) {
1386 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 =
1387 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1389 rc = smc_v2_determine_accepted_chid(aclc_v2, ini);
1393 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid;
1395 /* there is only one lgr role for SMC-D; use server lock */
1396 mutex_lock(&smc_server_lgr_pending);
1397 rc = smc_conn_create(smc, ini);
1399 mutex_unlock(&smc_server_lgr_pending);
1403 /* Create send and receive buffers */
1404 rc = smc_buf_create(smc, true);
1406 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1410 smc_conn_save_peer_info(smc, aclc);
1411 smc_close_init(smc);
1415 if (aclc->hdr.version > SMC_V1) {
1416 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1417 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1419 eid = clc_v2->d1.eid;
1422 rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1423 aclc->hdr.version, eid, NULL);
1426 mutex_unlock(&smc_server_lgr_pending);
1428 smc_copy_sock_settings_to_clc(smc);
1429 smc->connect_nonblock = 0;
1430 if (smc->sk.sk_state == SMC_INIT)
1431 smc->sk.sk_state = SMC_ACTIVE;
1435 smc_conn_abort(smc, ini->first_contact_local);
1436 mutex_unlock(&smc_server_lgr_pending);
1437 smc->connect_nonblock = 0;
1442 /* check if received accept type and version matches a proposed one */
1443 static int smc_connect_check_aclc(struct smc_init_info *ini,
1444 struct smc_clc_msg_accept_confirm *aclc)
1446 if (aclc->hdr.typev1 != SMC_TYPE_R &&
1447 aclc->hdr.typev1 != SMC_TYPE_D)
1448 return SMC_CLC_DECL_MODEUNSUPP;
1450 if (aclc->hdr.version >= SMC_V2) {
1451 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1452 !smcr_indicated(ini->smc_type_v2)) ||
1453 (aclc->hdr.typev1 == SMC_TYPE_D &&
1454 !smcd_indicated(ini->smc_type_v2)))
1455 return SMC_CLC_DECL_MODEUNSUPP;
1457 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1458 !smcr_indicated(ini->smc_type_v1)) ||
1459 (aclc->hdr.typev1 == SMC_TYPE_D &&
1460 !smcd_indicated(ini->smc_type_v1)))
1461 return SMC_CLC_DECL_MODEUNSUPP;
1467 /* perform steps before actually connecting */
1468 static int __smc_connect(struct smc_sock *smc)
1470 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1471 struct smc_clc_msg_accept_confirm_v2 *aclc2;
1472 struct smc_clc_msg_accept_confirm *aclc;
1473 struct smc_init_info *ini = NULL;
1477 if (smc->use_fallback)
1478 return smc_connect_fallback(smc, smc->fallback_rsn);
1480 /* if peer has not signalled SMC-capability, fall back */
1481 if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1482 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1484 /* IPSec connections opt out of SMC optimizations */
1485 if (using_ipsec(smc))
1486 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1489 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1491 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1494 ini->smcd_version = SMC_V1 | SMC_V2;
1495 ini->smcr_version = SMC_V1 | SMC_V2;
1496 ini->smc_type_v1 = SMC_TYPE_B;
1497 ini->smc_type_v2 = SMC_TYPE_B;
1499 /* get vlan id from IP device */
1500 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1501 ini->smcd_version &= ~SMC_V1;
1502 ini->smcr_version = 0;
1503 ini->smc_type_v1 = SMC_TYPE_N;
1504 if (!ini->smcd_version) {
1505 rc = SMC_CLC_DECL_GETVLANERR;
1510 rc = smc_find_proposal_devices(smc, ini);
1514 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1516 rc = SMC_CLC_DECL_MEM;
1519 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf;
1520 aclc = (struct smc_clc_msg_accept_confirm *)aclc2;
1522 /* perform CLC handshake */
1523 rc = smc_connect_clc(smc, aclc2, ini);
1525 /* -EAGAIN on timeout, see tcp_recvmsg() */
1526 if (rc == -EAGAIN) {
1528 smc->sk.sk_err = ETIMEDOUT;
1533 /* check if smc modes and versions of CLC proposal and accept match */
1534 rc = smc_connect_check_aclc(ini, aclc);
1535 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1539 /* depending on previous steps, connect using rdma or ism */
1540 if (aclc->hdr.typev1 == SMC_TYPE_R) {
1541 ini->smcr_version = version;
1542 rc = smc_connect_rdma(smc, aclc, ini);
1543 } else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1544 ini->smcd_version = version;
1545 rc = smc_connect_ism(smc, aclc, ini);
1550 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1551 smc_connect_ism_vlan_cleanup(smc, ini);
1557 smc_connect_ism_vlan_cleanup(smc, ini);
1561 return smc_connect_decline_fallback(smc, rc, version);
1564 static void smc_connect_work(struct work_struct *work)
1566 struct smc_sock *smc = container_of(work, struct smc_sock,
1568 long timeo = smc->sk.sk_sndtimeo;
1572 timeo = MAX_SCHEDULE_TIMEOUT;
1573 lock_sock(smc->clcsock->sk);
1574 if (smc->clcsock->sk->sk_err) {
1575 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1576 } else if ((1 << smc->clcsock->sk->sk_state) &
1577 (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1578 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1579 if ((rc == -EPIPE) &&
1580 ((1 << smc->clcsock->sk->sk_state) &
1581 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1584 release_sock(smc->clcsock->sk);
1585 lock_sock(&smc->sk);
1586 if (rc != 0 || smc->sk.sk_err) {
1587 smc->sk.sk_state = SMC_CLOSED;
1588 if (rc == -EPIPE || rc == -EAGAIN)
1589 smc->sk.sk_err = EPIPE;
1590 else if (rc == -ECONNREFUSED)
1591 smc->sk.sk_err = ECONNREFUSED;
1592 else if (signal_pending(current))
1593 smc->sk.sk_err = -sock_intr_errno(timeo);
1594 sock_put(&smc->sk); /* passive closing */
1598 rc = __smc_connect(smc);
1600 smc->sk.sk_err = -rc;
1603 if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1604 if (smc->sk.sk_err) {
1605 smc->sk.sk_state_change(&smc->sk);
1606 } else { /* allow polling before and after fallback decision */
1607 smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1608 smc->sk.sk_write_space(&smc->sk);
1611 release_sock(&smc->sk);
1614 static int smc_connect(struct socket *sock, struct sockaddr *addr,
1615 int alen, int flags)
1617 struct sock *sk = sock->sk;
1618 struct smc_sock *smc;
1623 /* separate smc parameter checking to be safe */
1624 if (alen < sizeof(addr->sa_family))
1626 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1630 switch (sock->state) {
1635 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1638 if (sk->sk_state == SMC_ACTIVE)
1641 case SS_UNCONNECTED:
1642 sock->state = SS_CONNECTING;
1646 switch (sk->sk_state) {
1650 rc = sock_error(sk) ? : -ECONNABORTED;
1651 sock->state = SS_UNCONNECTED;
1660 smc_copy_sock_settings_to_clc(smc);
1661 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1662 if (smc->connect_nonblock) {
1666 rc = kernel_connect(smc->clcsock, addr, alen, flags);
1667 if (rc && rc != -EINPROGRESS)
1670 if (smc->use_fallback) {
1671 sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1674 sock_hold(&smc->sk); /* sock put in passive closing */
1675 if (flags & O_NONBLOCK) {
1676 if (queue_work(smc_hs_wq, &smc->connect_work))
1677 smc->connect_nonblock = 1;
1681 rc = __smc_connect(smc);
1688 sock->state = SS_CONNECTED;
1695 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1697 struct socket *new_clcsock = NULL;
1698 struct sock *lsk = &lsmc->sk;
1699 struct sock *new_sk;
1703 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1706 lsk->sk_err = ENOMEM;
1711 *new_smc = smc_sk(new_sk);
1713 mutex_lock(&lsmc->clcsock_release_lock);
1715 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1716 mutex_unlock(&lsmc->clcsock_release_lock);
1718 if (rc < 0 && rc != -EAGAIN)
1720 if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1721 new_sk->sk_prot->unhash(new_sk);
1723 sock_release(new_clcsock);
1724 new_sk->sk_state = SMC_CLOSED;
1725 sock_set_flag(new_sk, SOCK_DEAD);
1726 sock_put(new_sk); /* final */
1731 /* new clcsock has inherited the smc listen-specific sk_data_ready
1732 * function; switch it back to the original sk_data_ready function
1734 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1736 /* if new clcsock has also inherited the fallback-specific callback
1737 * functions, switch them back to the original ones.
1739 if (lsmc->use_fallback) {
1740 if (lsmc->clcsk_state_change)
1741 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1742 if (lsmc->clcsk_write_space)
1743 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1744 if (lsmc->clcsk_error_report)
1745 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1748 (*new_smc)->clcsock = new_clcsock;
1753 /* add a just created sock to the accept queue of the listen sock as
1754 * candidate for a following socket accept call from user space
1756 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1758 struct smc_sock *par = smc_sk(parent);
1760 sock_hold(sk); /* sock_put in smc_accept_unlink () */
1761 spin_lock(&par->accept_q_lock);
1762 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1763 spin_unlock(&par->accept_q_lock);
1764 sk_acceptq_added(parent);
1767 /* remove a socket from the accept queue of its parental listening socket */
1768 static void smc_accept_unlink(struct sock *sk)
1770 struct smc_sock *par = smc_sk(sk)->listen_smc;
1772 spin_lock(&par->accept_q_lock);
1773 list_del_init(&smc_sk(sk)->accept_q);
1774 spin_unlock(&par->accept_q_lock);
1775 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1776 sock_put(sk); /* sock_hold in smc_accept_enqueue */
1779 /* remove a sock from the accept queue to bind it to a new socket created
1780 * for a socket accept call from user space
1782 struct sock *smc_accept_dequeue(struct sock *parent,
1783 struct socket *new_sock)
1785 struct smc_sock *isk, *n;
1786 struct sock *new_sk;
1788 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1789 new_sk = (struct sock *)isk;
1791 smc_accept_unlink(new_sk);
1792 if (new_sk->sk_state == SMC_CLOSED) {
1793 new_sk->sk_prot->unhash(new_sk);
1795 sock_release(isk->clcsock);
1796 isk->clcsock = NULL;
1798 sock_put(new_sk); /* final */
1802 sock_graft(new_sk, new_sock);
1803 new_sock->state = SS_CONNECTED;
1804 if (isk->use_fallback) {
1805 smc_sk(new_sk)->clcsock->file = new_sock->file;
1806 isk->clcsock->file->private_data = isk->clcsock;
1814 /* clean up for a created but never accepted sock */
1815 void smc_close_non_accepted(struct sock *sk)
1817 struct smc_sock *smc = smc_sk(sk);
1819 sock_hold(sk); /* sock_put below */
1821 if (!sk->sk_lingertime)
1822 /* wait for peer closing */
1823 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
1826 sock_put(sk); /* sock_hold above */
1827 sock_put(sk); /* final sock_put */
1830 static int smcr_serv_conf_first_link(struct smc_sock *smc)
1832 struct smc_link *link = smc->conn.lnk;
1833 struct smc_llc_qentry *qentry;
1836 /* reg the sndbuf if it was vzalloced*/
1837 if (smc->conn.sndbuf_desc->is_vm) {
1838 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1839 return SMC_CLC_DECL_ERR_REGBUF;
1843 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1844 return SMC_CLC_DECL_ERR_REGBUF;
1846 /* send CONFIRM LINK request to client over the RoCE fabric */
1847 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1849 return SMC_CLC_DECL_TIMEOUT_CL;
1851 /* receive CONFIRM LINK response from client over the RoCE fabric */
1852 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1853 SMC_LLC_CONFIRM_LINK);
1855 struct smc_clc_msg_decline dclc;
1857 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1858 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1859 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1861 smc_llc_save_peer_uid(qentry);
1862 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1863 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1865 return SMC_CLC_DECL_RMBE_EC;
1867 /* confirm_rkey is implicit on 1st contact */
1868 smc->conn.rmb_desc->is_conf_rkey = true;
1870 smc_llc_link_active(link);
1871 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1873 down_write(&link->lgr->llc_conf_mutex);
1874 /* initial contact - try to establish second link */
1875 smc_llc_srv_add_link(link, NULL);
1876 up_write(&link->lgr->llc_conf_mutex);
1880 /* listen worker: finish */
1881 static void smc_listen_out(struct smc_sock *new_smc)
1883 struct smc_sock *lsmc = new_smc->listen_smc;
1884 struct sock *newsmcsk = &new_smc->sk;
1886 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1887 atomic_dec(&lsmc->queued_smc_hs);
1889 if (lsmc->sk.sk_state == SMC_LISTEN) {
1890 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1891 smc_accept_enqueue(&lsmc->sk, newsmcsk);
1892 release_sock(&lsmc->sk);
1893 } else { /* no longer listening */
1894 smc_close_non_accepted(newsmcsk);
1897 /* Wake up accept */
1898 lsmc->sk.sk_data_ready(&lsmc->sk);
1899 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1902 /* listen worker: finish in state connected */
1903 static void smc_listen_out_connected(struct smc_sock *new_smc)
1905 struct sock *newsmcsk = &new_smc->sk;
1907 if (newsmcsk->sk_state == SMC_INIT)
1908 newsmcsk->sk_state = SMC_ACTIVE;
1910 smc_listen_out(new_smc);
1913 /* listen worker: finish in error state */
1914 static void smc_listen_out_err(struct smc_sock *new_smc)
1916 struct sock *newsmcsk = &new_smc->sk;
1917 struct net *net = sock_net(newsmcsk);
1919 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1920 if (newsmcsk->sk_state == SMC_INIT)
1921 sock_put(&new_smc->sk); /* passive closing */
1922 newsmcsk->sk_state = SMC_CLOSED;
1924 smc_listen_out(new_smc);
1927 /* listen worker: decline and fall back if possible */
1928 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1929 int local_first, u8 version)
1931 /* RDMA setup failed, switch back to TCP */
1932 smc_conn_abort(new_smc, local_first);
1933 if (reason_code < 0 ||
1934 smc_switch_to_fallback(new_smc, reason_code)) {
1935 /* error, no fallback possible */
1936 smc_listen_out_err(new_smc);
1939 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1940 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1941 smc_listen_out_err(new_smc);
1945 smc_listen_out_connected(new_smc);
1948 /* listen worker: version checking */
1949 static int smc_listen_v2_check(struct smc_sock *new_smc,
1950 struct smc_clc_msg_proposal *pclc,
1951 struct smc_init_info *ini)
1953 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
1954 struct smc_clc_v2_extension *pclc_v2_ext;
1955 int rc = SMC_CLC_DECL_PEERNOSMC;
1957 ini->smc_type_v1 = pclc->hdr.typev1;
1958 ini->smc_type_v2 = pclc->hdr.typev2;
1959 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1960 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1961 if (pclc->hdr.version > SMC_V1) {
1962 if (smcd_indicated(ini->smc_type_v2))
1963 ini->smcd_version |= SMC_V2;
1964 if (smcr_indicated(ini->smc_type_v2))
1965 ini->smcr_version |= SMC_V2;
1967 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
1968 rc = SMC_CLC_DECL_PEERNOSMC;
1971 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1973 ini->smcd_version &= ~SMC_V2;
1974 ini->smcr_version &= ~SMC_V2;
1975 rc = SMC_CLC_DECL_NOV2EXT;
1978 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
1979 if (ini->smcd_version & SMC_V2) {
1980 if (!smc_ism_is_v2_capable()) {
1981 ini->smcd_version &= ~SMC_V2;
1982 rc = SMC_CLC_DECL_NOISM2SUPP;
1983 } else if (!pclc_smcd_v2_ext) {
1984 ini->smcd_version &= ~SMC_V2;
1985 rc = SMC_CLC_DECL_NOV2DEXT;
1986 } else if (!pclc_v2_ext->hdr.eid_cnt &&
1987 !pclc_v2_ext->hdr.flag.seid) {
1988 ini->smcd_version &= ~SMC_V2;
1989 rc = SMC_CLC_DECL_NOUEID;
1992 if (ini->smcr_version & SMC_V2) {
1993 if (!pclc_v2_ext->hdr.eid_cnt) {
1994 ini->smcr_version &= ~SMC_V2;
1995 rc = SMC_CLC_DECL_NOUEID;
2000 if (!ini->smcd_version && !ini->smcr_version)
2006 /* listen worker: check prefixes */
2007 static int smc_listen_prfx_check(struct smc_sock *new_smc,
2008 struct smc_clc_msg_proposal *pclc)
2010 struct smc_clc_msg_proposal_prefix *pclc_prfx;
2011 struct socket *newclcsock = new_smc->clcsock;
2013 if (pclc->hdr.typev1 == SMC_TYPE_N)
2015 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2016 if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2017 return SMC_CLC_DECL_DIFFPREFIX;
2022 /* listen worker: initialize connection and buffers */
2023 static int smc_listen_rdma_init(struct smc_sock *new_smc,
2024 struct smc_init_info *ini)
2028 /* allocate connection / link group */
2029 rc = smc_conn_create(new_smc, ini);
2033 /* create send buffer and rmb */
2034 if (smc_buf_create(new_smc, false)) {
2035 smc_conn_abort(new_smc, ini->first_contact_local);
2036 return SMC_CLC_DECL_MEM;
2042 /* listen worker: initialize connection and buffers for SMC-D */
2043 static int smc_listen_ism_init(struct smc_sock *new_smc,
2044 struct smc_init_info *ini)
2048 rc = smc_conn_create(new_smc, ini);
2052 /* Create send and receive buffers */
2053 rc = smc_buf_create(new_smc, true);
2055 smc_conn_abort(new_smc, ini->first_contact_local);
2056 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2063 static bool smc_is_already_selected(struct smcd_dev *smcd,
2064 struct smc_init_info *ini,
2069 for (i = 0; i < matches; i++)
2070 if (smcd == ini->ism_dev[i])
2076 /* check for ISM devices matching proposed ISM devices */
2077 static void smc_check_ism_v2_match(struct smc_init_info *ini,
2078 u16 proposed_chid, u64 proposed_gid,
2079 unsigned int *matches)
2081 struct smcd_dev *smcd;
2083 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2084 if (smcd->going_away)
2086 if (smc_is_already_selected(smcd, ini, *matches))
2088 if (smc_ism_get_chid(smcd) == proposed_chid &&
2089 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2090 ini->ism_peer_gid[*matches] = proposed_gid;
2091 ini->ism_dev[*matches] = smcd;
2098 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini)
2104 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc,
2105 struct smc_clc_msg_proposal *pclc,
2106 struct smc_init_info *ini)
2108 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
2109 struct smc_clc_v2_extension *smc_v2_ext;
2110 struct smc_clc_msg_smcd *pclc_smcd;
2111 unsigned int matches = 0;
2116 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2119 pclc_smcd = smc_get_clc_msg_smcd(pclc);
2120 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2121 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2123 mutex_lock(&smcd_dev_list.mutex);
2124 if (pclc_smcd->ism.chid)
2125 /* check for ISM device matching proposed native ISM device */
2126 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2127 ntohll(pclc_smcd->ism.gid), &matches);
2128 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) {
2129 /* check for ISM devices matching proposed non-native ISM
2132 smc_check_ism_v2_match(ini,
2133 ntohs(smcd_v2_ext->gidchid[i - 1].chid),
2134 ntohll(smcd_v2_ext->gidchid[i - 1].gid),
2137 mutex_unlock(&smcd_dev_list.mutex);
2139 if (!ini->ism_dev[0]) {
2140 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2144 smc_ism_get_system_eid(&eid);
2145 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2146 smcd_v2_ext->system_eid, eid))
2149 /* separate - outside the smcd_dev_list.lock */
2150 smcd_version = ini->smcd_version;
2151 for (i = 0; i < matches; i++) {
2152 ini->smcd_version = SMC_V2;
2153 ini->is_smcd = true;
2154 ini->ism_selected = i;
2155 rc = smc_listen_ism_init(new_smc, ini);
2157 smc_find_ism_store_rc(rc, ini);
2158 /* try next active ISM device */
2161 return; /* matching and usable V2 ISM device found */
2163 /* no V2 ISM device could be initialized */
2164 ini->smcd_version = smcd_version; /* restore original value */
2165 ini->negotiated_eid[0] = 0;
2168 ini->smcd_version &= ~SMC_V2;
2169 ini->ism_dev[0] = NULL;
2170 ini->is_smcd = false;
2173 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2174 struct smc_clc_msg_proposal *pclc,
2175 struct smc_init_info *ini)
2177 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2180 /* check if ISM V1 is available */
2181 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1))
2183 ini->is_smcd = true; /* prepare ISM check */
2184 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid);
2185 rc = smc_find_ism_device(new_smc, ini);
2188 ini->ism_selected = 0;
2189 rc = smc_listen_ism_init(new_smc, ini);
2191 return; /* V1 ISM device found */
2194 smc_find_ism_store_rc(rc, ini);
2195 ini->smcd_version &= ~SMC_V1;
2196 ini->ism_dev[0] = NULL;
2197 ini->is_smcd = false;
2200 /* listen worker: register buffers */
2201 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2203 struct smc_connection *conn = &new_smc->conn;
2206 /* reg sendbufs if they were vzalloced */
2207 if (conn->sndbuf_desc->is_vm) {
2208 if (smcr_lgr_reg_sndbufs(conn->lnk,
2210 return SMC_CLC_DECL_ERR_REGBUF;
2212 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2213 return SMC_CLC_DECL_ERR_REGBUF;
2219 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2220 struct smc_clc_msg_proposal *pclc,
2221 struct smc_init_info *ini)
2223 struct smc_clc_v2_extension *smc_v2_ext;
2227 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2230 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2231 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2234 /* prepare RDMA check */
2235 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2236 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2237 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2238 ini->check_smcrv2 = true;
2239 ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2240 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2241 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2242 rc = smc_find_rdma_device(new_smc, ini);
2244 smc_find_ism_store_rc(rc, ini);
2247 if (!ini->smcrv2.uses_gateway)
2248 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2250 smcr_version = ini->smcr_version;
2251 ini->smcr_version = SMC_V2;
2252 rc = smc_listen_rdma_init(new_smc, ini);
2254 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2256 smc_conn_abort(new_smc, ini->first_contact_local);
2260 ini->smcr_version = smcr_version;
2261 smc_find_ism_store_rc(rc, ini);
2264 ini->smcr_version &= ~SMC_V2;
2265 ini->smcrv2.ib_dev_v2 = NULL;
2266 ini->check_smcrv2 = false;
2269 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2270 struct smc_clc_msg_proposal *pclc,
2271 struct smc_init_info *ini)
2275 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2276 return SMC_CLC_DECL_NOSMCDEV;
2278 /* prepare RDMA check */
2279 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2280 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2281 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2282 rc = smc_find_rdma_device(new_smc, ini);
2284 /* no RDMA device found */
2285 return SMC_CLC_DECL_NOSMCDEV;
2287 rc = smc_listen_rdma_init(new_smc, ini);
2290 return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2293 /* determine the local device matching to proposal */
2294 static int smc_listen_find_device(struct smc_sock *new_smc,
2295 struct smc_clc_msg_proposal *pclc,
2296 struct smc_init_info *ini)
2300 /* check for ISM device matching V2 proposed device */
2301 smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2302 if (ini->ism_dev[0])
2305 /* check for matching IP prefix and subnet length (V1) */
2306 prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2308 smc_find_ism_store_rc(prfx_rc, ini);
2310 /* get vlan id from IP device */
2311 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2312 return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2314 /* check for ISM device matching V1 proposed device */
2316 smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2317 if (ini->ism_dev[0])
2320 if (!smcr_indicated(pclc->hdr.typev1) &&
2321 !smcr_indicated(pclc->hdr.typev2))
2322 /* skip RDMA and decline */
2323 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2325 /* check if RDMA V2 is available */
2326 smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2327 if (ini->smcrv2.ib_dev_v2)
2330 /* check if RDMA V1 is available */
2334 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2335 smc_find_ism_store_rc(rc, ini);
2336 return (!rc) ? 0 : ini->rc;
2338 return SMC_CLC_DECL_NOSMCDEV;
2341 /* listen worker: finish RDMA setup */
2342 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2343 struct smc_clc_msg_accept_confirm *cclc,
2345 struct smc_init_info *ini)
2347 struct smc_link *link = new_smc->conn.lnk;
2348 int reason_code = 0;
2351 smc_link_save_peer_info(link, cclc, ini);
2353 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2354 return SMC_CLC_DECL_ERR_RTOK;
2357 if (smc_ib_ready_link(link))
2358 return SMC_CLC_DECL_ERR_RDYLNK;
2359 /* QP confirmation over RoCE fabric */
2360 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2361 reason_code = smcr_serv_conf_first_link(new_smc);
2362 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2367 /* setup for connection of server */
2368 static void smc_listen_work(struct work_struct *work)
2370 struct smc_sock *new_smc = container_of(work, struct smc_sock,
2372 struct socket *newclcsock = new_smc->clcsock;
2373 struct smc_clc_msg_accept_confirm *cclc;
2374 struct smc_clc_msg_proposal_area *buf;
2375 struct smc_clc_msg_proposal *pclc;
2376 struct smc_init_info *ini = NULL;
2377 u8 proposal_version = SMC_V1;
2381 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2382 return smc_listen_out_err(new_smc);
2384 if (new_smc->use_fallback) {
2385 smc_listen_out_connected(new_smc);
2389 /* check if peer is smc capable */
2390 if (!tcp_sk(newclcsock->sk)->syn_smc) {
2391 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2393 smc_listen_out_err(new_smc);
2395 smc_listen_out_connected(new_smc);
2399 /* do inband token exchange -
2400 * wait for and receive SMC Proposal CLC message
2402 buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2404 rc = SMC_CLC_DECL_MEM;
2407 pclc = (struct smc_clc_msg_proposal *)buf;
2408 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2409 SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2413 if (pclc->hdr.version > SMC_V1)
2414 proposal_version = SMC_V2;
2416 /* IPSec connections opt out of SMC optimizations */
2417 if (using_ipsec(new_smc)) {
2418 rc = SMC_CLC_DECL_IPSEC;
2422 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2424 rc = SMC_CLC_DECL_MEM;
2428 /* initial version checking */
2429 rc = smc_listen_v2_check(new_smc, pclc, ini);
2433 mutex_lock(&smc_server_lgr_pending);
2434 smc_close_init(new_smc);
2435 smc_rx_init(new_smc);
2436 smc_tx_init(new_smc);
2438 /* determine ISM or RoCE device used for connection */
2439 rc = smc_listen_find_device(new_smc, pclc, ini);
2443 /* send SMC Accept CLC message */
2444 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2445 rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2446 accept_version, ini->negotiated_eid);
2450 /* SMC-D does not need this lock any more */
2452 mutex_unlock(&smc_server_lgr_pending);
2454 /* receive SMC Confirm CLC message */
2455 memset(buf, 0, sizeof(*buf));
2456 cclc = (struct smc_clc_msg_accept_confirm *)buf;
2457 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2458 SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2466 if (!ini->is_smcd) {
2467 rc = smc_listen_rdma_finish(new_smc, cclc,
2468 ini->first_contact_local, ini);
2471 mutex_unlock(&smc_server_lgr_pending);
2473 smc_conn_save_peer_info(new_smc, cclc);
2474 smc_listen_out_connected(new_smc);
2475 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2479 mutex_unlock(&smc_server_lgr_pending);
2481 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2488 static void smc_tcp_listen_work(struct work_struct *work)
2490 struct smc_sock *lsmc = container_of(work, struct smc_sock,
2492 struct sock *lsk = &lsmc->sk;
2493 struct smc_sock *new_smc;
2497 while (lsk->sk_state == SMC_LISTEN) {
2498 rc = smc_clcsock_accept(lsmc, &new_smc);
2499 if (rc) /* clcsock accept queue empty or error */
2504 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2505 atomic_inc(&lsmc->queued_smc_hs);
2507 new_smc->listen_smc = lsmc;
2508 new_smc->use_fallback = lsmc->use_fallback;
2509 new_smc->fallback_rsn = lsmc->fallback_rsn;
2510 sock_hold(lsk); /* sock_put in smc_listen_work */
2511 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2512 smc_copy_sock_settings_to_smc(new_smc);
2513 sock_hold(&new_smc->sk); /* sock_put in passive closing */
2514 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2515 sock_put(&new_smc->sk);
2520 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2523 static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2525 struct smc_sock *lsmc;
2527 read_lock_bh(&listen_clcsock->sk_callback_lock);
2528 lsmc = smc_clcsock_user_data(listen_clcsock);
2531 lsmc->clcsk_data_ready(listen_clcsock);
2532 if (lsmc->sk.sk_state == SMC_LISTEN) {
2533 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2534 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2535 sock_put(&lsmc->sk);
2538 read_unlock_bh(&listen_clcsock->sk_callback_lock);
2541 static int smc_listen(struct socket *sock, int backlog)
2543 struct sock *sk = sock->sk;
2544 struct smc_sock *smc;
2551 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2552 smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2556 if (sk->sk_state == SMC_LISTEN) {
2557 sk->sk_max_ack_backlog = backlog;
2560 /* some socket options are handled in core, so we could not apply
2561 * them to the clc socket -- copy smc socket options to clc socket
2563 smc_copy_sock_settings_to_clc(smc);
2564 if (!smc->use_fallback)
2565 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2567 /* save original sk_data_ready function and establish
2568 * smc-specific sk_data_ready function
2570 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2571 smc->clcsock->sk->sk_user_data =
2572 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
2573 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2574 smc_clcsock_data_ready, &smc->clcsk_data_ready);
2575 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2577 /* save original ops */
2578 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2580 smc->af_ops = *smc->ori_af_ops;
2581 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2583 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2585 if (smc->limit_smc_hs)
2586 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2588 rc = kernel_listen(smc->clcsock, backlog);
2590 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2591 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2592 &smc->clcsk_data_ready);
2593 smc->clcsock->sk->sk_user_data = NULL;
2594 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2597 sk->sk_max_ack_backlog = backlog;
2598 sk->sk_ack_backlog = 0;
2599 sk->sk_state = SMC_LISTEN;
2606 static int smc_accept(struct socket *sock, struct socket *new_sock,
2607 int flags, bool kern)
2609 struct sock *sk = sock->sk, *nsk;
2610 DECLARE_WAITQUEUE(wait, current);
2611 struct smc_sock *lsmc;
2616 sock_hold(sk); /* sock_put below */
2619 if (lsmc->sk.sk_state != SMC_LISTEN) {
2625 /* Wait for an incoming connection */
2626 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
2627 add_wait_queue_exclusive(sk_sleep(sk), &wait);
2628 while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2629 set_current_state(TASK_INTERRUPTIBLE);
2635 timeo = schedule_timeout(timeo);
2636 /* wakeup by sk_data_ready in smc_listen_work() */
2637 sched_annotate_sleep();
2639 if (signal_pending(current)) {
2640 rc = sock_intr_errno(timeo);
2644 set_current_state(TASK_RUNNING);
2645 remove_wait_queue(sk_sleep(sk), &wait);
2648 rc = sock_error(nsk);
2653 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
2654 /* wait till data arrives on the socket */
2655 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
2657 if (smc_sk(nsk)->use_fallback) {
2658 struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2661 if (skb_queue_empty(&clcsk->sk_receive_queue))
2662 sk_wait_data(clcsk, &timeo, NULL);
2663 release_sock(clcsk);
2664 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2666 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
2672 sock_put(sk); /* sock_hold above */
2676 static int smc_getname(struct socket *sock, struct sockaddr *addr,
2679 struct smc_sock *smc;
2681 if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2682 (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2685 smc = smc_sk(sock->sk);
2687 return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2690 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2692 struct sock *sk = sock->sk;
2693 struct smc_sock *smc;
2699 /* SMC does not support connect with fastopen */
2700 if (msg->msg_flags & MSG_FASTOPEN) {
2701 /* not connected yet, fallback */
2702 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2703 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2710 } else if ((sk->sk_state != SMC_ACTIVE) &&
2711 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2712 (sk->sk_state != SMC_INIT)) {
2717 if (smc->use_fallback) {
2718 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2720 rc = smc_tx_sendmsg(smc, msg, len);
2721 SMC_STAT_TX_PAYLOAD(smc, len, rc);
2728 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2731 struct sock *sk = sock->sk;
2732 struct smc_sock *smc;
2737 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2738 /* socket was connected before, no more data to read */
2742 if ((sk->sk_state == SMC_INIT) ||
2743 (sk->sk_state == SMC_LISTEN) ||
2744 (sk->sk_state == SMC_CLOSED))
2747 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2752 if (smc->use_fallback) {
2753 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2755 msg->msg_namelen = 0;
2756 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2757 SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2765 static __poll_t smc_accept_poll(struct sock *parent)
2767 struct smc_sock *isk = smc_sk(parent);
2770 spin_lock(&isk->accept_q_lock);
2771 if (!list_empty(&isk->accept_q))
2772 mask = EPOLLIN | EPOLLRDNORM;
2773 spin_unlock(&isk->accept_q_lock);
2778 static __poll_t smc_poll(struct file *file, struct socket *sock,
2781 struct sock *sk = sock->sk;
2782 struct smc_sock *smc;
2788 smc = smc_sk(sock->sk);
2789 if (smc->use_fallback) {
2790 /* delegate to CLC child sock */
2791 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2792 sk->sk_err = smc->clcsock->sk->sk_err;
2794 if (sk->sk_state != SMC_CLOSED)
2795 sock_poll_wait(file, sock, wait);
2798 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2799 (sk->sk_state == SMC_CLOSED))
2801 if (sk->sk_state == SMC_LISTEN) {
2802 /* woken up by sk_data_ready in smc_listen_work() */
2803 mask |= smc_accept_poll(sk);
2804 } else if (smc->use_fallback) { /* as result of connect_work()*/
2805 mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2807 sk->sk_err = smc->clcsock->sk->sk_err;
2809 if ((sk->sk_state != SMC_INIT &&
2810 atomic_read(&smc->conn.sndbuf_space)) ||
2811 sk->sk_shutdown & SEND_SHUTDOWN) {
2812 mask |= EPOLLOUT | EPOLLWRNORM;
2814 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2815 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2817 if (atomic_read(&smc->conn.bytes_to_rcv))
2818 mask |= EPOLLIN | EPOLLRDNORM;
2819 if (sk->sk_shutdown & RCV_SHUTDOWN)
2820 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2821 if (sk->sk_state == SMC_APPCLOSEWAIT1)
2823 if (smc->conn.urg_state == SMC_URG_VALID)
2831 static int smc_shutdown(struct socket *sock, int how)
2833 struct sock *sk = sock->sk;
2834 bool do_shutdown = true;
2835 struct smc_sock *smc;
2842 if ((how < SHUT_RD) || (how > SHUT_RDWR))
2847 if (sock->state == SS_CONNECTING) {
2848 if (sk->sk_state == SMC_ACTIVE)
2849 sock->state = SS_CONNECTED;
2850 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2851 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2852 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2853 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2854 sk->sk_state == SMC_APPFINCLOSEWAIT)
2855 sock->state = SS_DISCONNECTING;
2859 if ((sk->sk_state != SMC_ACTIVE) &&
2860 (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2861 (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2862 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2863 (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2864 (sk->sk_state != SMC_APPFINCLOSEWAIT))
2866 if (smc->use_fallback) {
2867 rc = kernel_sock_shutdown(smc->clcsock, how);
2868 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2869 if (sk->sk_shutdown == SHUTDOWN_MASK) {
2870 sk->sk_state = SMC_CLOSED;
2871 sk->sk_socket->state = SS_UNCONNECTED;
2877 case SHUT_RDWR: /* shutdown in both directions */
2878 old_state = sk->sk_state;
2879 rc = smc_close_active(smc);
2880 if (old_state == SMC_ACTIVE &&
2881 sk->sk_state == SMC_PEERCLOSEWAIT1)
2882 do_shutdown = false;
2885 rc = smc_close_shutdown_write(smc);
2889 /* nothing more to do because peer is not involved */
2892 if (do_shutdown && smc->clcsock)
2893 rc1 = kernel_sock_shutdown(smc->clcsock, how);
2894 /* map sock_shutdown_cmd constants to sk_shutdown value range */
2895 sk->sk_shutdown |= how + 1;
2897 if (sk->sk_state == SMC_CLOSED)
2898 sock->state = SS_UNCONNECTED;
2900 sock->state = SS_DISCONNECTING;
2903 return rc ? rc : rc1;
2906 static int __smc_getsockopt(struct socket *sock, int level, int optname,
2907 char __user *optval, int __user *optlen)
2909 struct smc_sock *smc;
2912 smc = smc_sk(sock->sk);
2914 if (get_user(len, optlen))
2917 len = min_t(int, len, sizeof(int));
2924 val = smc->limit_smc_hs;
2930 if (put_user(len, optlen))
2932 if (copy_to_user(optval, &val, len))
2938 static int __smc_setsockopt(struct socket *sock, int level, int optname,
2939 sockptr_t optval, unsigned int optlen)
2941 struct sock *sk = sock->sk;
2942 struct smc_sock *smc;
2950 if (optlen < sizeof(int)) {
2954 if (copy_from_sockptr(&val, optval, sizeof(int))) {
2959 smc->limit_smc_hs = !!val;
2971 static int smc_setsockopt(struct socket *sock, int level, int optname,
2972 sockptr_t optval, unsigned int optlen)
2974 struct sock *sk = sock->sk;
2975 struct smc_sock *smc;
2978 if (level == SOL_TCP && optname == TCP_ULP)
2980 else if (level == SOL_SMC)
2981 return __smc_setsockopt(sock, level, optname, optval, optlen);
2985 /* generic setsockopts reaching us here always apply to the
2988 mutex_lock(&smc->clcsock_release_lock);
2989 if (!smc->clcsock) {
2990 mutex_unlock(&smc->clcsock_release_lock);
2993 if (unlikely(!smc->clcsock->ops->setsockopt))
2996 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
2998 if (smc->clcsock->sk->sk_err) {
2999 sk->sk_err = smc->clcsock->sk->sk_err;
3000 sk_error_report(sk);
3002 mutex_unlock(&smc->clcsock_release_lock);
3004 if (optlen < sizeof(int))
3006 if (copy_from_sockptr(&val, optval, sizeof(int)))
3010 if (rc || smc->use_fallback)
3014 case TCP_FASTOPEN_CONNECT:
3015 case TCP_FASTOPEN_KEY:
3016 case TCP_FASTOPEN_NO_COOKIE:
3017 /* option not supported by SMC */
3018 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3019 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3025 if (sk->sk_state != SMC_INIT &&
3026 sk->sk_state != SMC_LISTEN &&
3027 sk->sk_state != SMC_CLOSED) {
3029 SMC_STAT_INC(smc, ndly_cnt);
3030 smc_tx_pending(&smc->conn);
3031 cancel_delayed_work(&smc->conn.tx_work);
3036 if (sk->sk_state != SMC_INIT &&
3037 sk->sk_state != SMC_LISTEN &&
3038 sk->sk_state != SMC_CLOSED) {
3040 SMC_STAT_INC(smc, cork_cnt);
3041 smc_tx_pending(&smc->conn);
3042 cancel_delayed_work(&smc->conn.tx_work);
3046 case TCP_DEFER_ACCEPT:
3047 smc->sockopt_defer_accept = val;
3058 static int smc_getsockopt(struct socket *sock, int level, int optname,
3059 char __user *optval, int __user *optlen)
3061 struct smc_sock *smc;
3064 if (level == SOL_SMC)
3065 return __smc_getsockopt(sock, level, optname, optval, optlen);
3067 smc = smc_sk(sock->sk);
3068 mutex_lock(&smc->clcsock_release_lock);
3069 if (!smc->clcsock) {
3070 mutex_unlock(&smc->clcsock_release_lock);
3073 /* socket options apply to the CLC socket */
3074 if (unlikely(!smc->clcsock->ops->getsockopt)) {
3075 mutex_unlock(&smc->clcsock_release_lock);
3078 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3080 mutex_unlock(&smc->clcsock_release_lock);
3084 static int smc_ioctl(struct socket *sock, unsigned int cmd,
3087 union smc_host_cursor cons, urg;
3088 struct smc_connection *conn;
3089 struct smc_sock *smc;
3092 smc = smc_sk(sock->sk);
3094 lock_sock(&smc->sk);
3095 if (smc->use_fallback) {
3096 if (!smc->clcsock) {
3097 release_sock(&smc->sk);
3100 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3101 release_sock(&smc->sk);
3105 case SIOCINQ: /* same as FIONREAD */
3106 if (smc->sk.sk_state == SMC_LISTEN) {
3107 release_sock(&smc->sk);
3110 if (smc->sk.sk_state == SMC_INIT ||
3111 smc->sk.sk_state == SMC_CLOSED)
3114 answ = atomic_read(&smc->conn.bytes_to_rcv);
3117 /* output queue size (not send + not acked) */
3118 if (smc->sk.sk_state == SMC_LISTEN) {
3119 release_sock(&smc->sk);
3122 if (smc->sk.sk_state == SMC_INIT ||
3123 smc->sk.sk_state == SMC_CLOSED)
3126 answ = smc->conn.sndbuf_desc->len -
3127 atomic_read(&smc->conn.sndbuf_space);
3130 /* output queue size (not send only) */
3131 if (smc->sk.sk_state == SMC_LISTEN) {
3132 release_sock(&smc->sk);
3135 if (smc->sk.sk_state == SMC_INIT ||
3136 smc->sk.sk_state == SMC_CLOSED)
3139 answ = smc_tx_prepared_sends(&smc->conn);
3142 if (smc->sk.sk_state == SMC_LISTEN) {
3143 release_sock(&smc->sk);
3146 if (smc->sk.sk_state == SMC_INIT ||
3147 smc->sk.sk_state == SMC_CLOSED) {
3150 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3151 smc_curs_copy(&urg, &conn->urg_curs, conn);
3152 answ = smc_curs_diff(conn->rmb_desc->len,
3157 release_sock(&smc->sk);
3158 return -ENOIOCTLCMD;
3160 release_sock(&smc->sk);
3162 return put_user(answ, (int __user *)arg);
3165 /* Map the affected portions of the rmbe into an spd, note the number of bytes
3166 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3167 * updates till whenever a respective page has been fully processed.
3168 * Note that subsequent recv() calls have to wait till all splice() processing
3171 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3172 struct pipe_inode_info *pipe, size_t len,
3175 struct sock *sk = sock->sk;
3176 struct smc_sock *smc;
3181 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3182 /* socket was connected before, no more data to read */
3186 if (sk->sk_state == SMC_INIT ||
3187 sk->sk_state == SMC_LISTEN ||
3188 sk->sk_state == SMC_CLOSED)
3191 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3196 if (smc->use_fallback) {
3197 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3204 if (flags & SPLICE_F_NONBLOCK)
3205 flags = MSG_DONTWAIT;
3208 SMC_STAT_INC(smc, splice_cnt);
3209 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3217 /* must look like tcp */
3218 static const struct proto_ops smc_sock_ops = {
3220 .owner = THIS_MODULE,
3221 .release = smc_release,
3223 .connect = smc_connect,
3224 .socketpair = sock_no_socketpair,
3225 .accept = smc_accept,
3226 .getname = smc_getname,
3229 .listen = smc_listen,
3230 .shutdown = smc_shutdown,
3231 .setsockopt = smc_setsockopt,
3232 .getsockopt = smc_getsockopt,
3233 .sendmsg = smc_sendmsg,
3234 .recvmsg = smc_recvmsg,
3235 .mmap = sock_no_mmap,
3236 .splice_read = smc_splice_read,
3239 static int __smc_create(struct net *net, struct socket *sock, int protocol,
3240 int kern, struct socket *clcsock)
3242 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3243 struct smc_sock *smc;
3247 rc = -ESOCKTNOSUPPORT;
3248 if (sock->type != SOCK_STREAM)
3251 rc = -EPROTONOSUPPORT;
3252 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3256 sock->ops = &smc_sock_ops;
3257 sock->state = SS_UNCONNECTED;
3258 sk = smc_sock_alloc(net, sock, protocol);
3262 /* create internal TCP socket for CLC handshake and fallback */
3264 smc->use_fallback = false; /* assume rdma capability first */
3265 smc->fallback_rsn = 0;
3267 /* default behavior from limit_smc_hs in every net namespace */
3268 smc->limit_smc_hs = net->smc.limit_smc_hs;
3272 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3275 sk_common_release(sk);
3279 /* smc_clcsock_release() does not wait smc->clcsock->sk's
3280 * destruction; its sk_state might not be TCP_CLOSE after
3281 * smc->sk is close()d, and TCP timers can be fired later,
3282 * which need net ref.
3284 sk = smc->clcsock->sk;
3285 __netns_tracker_free(net, &sk->ns_tracker, false);
3286 sk->sk_net_refcnt = 1;
3287 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
3288 sock_inuse_add(net, 1);
3290 smc->clcsock = clcsock;
3297 static int smc_create(struct net *net, struct socket *sock, int protocol,
3300 return __smc_create(net, sock, protocol, kern, NULL);
3303 static const struct net_proto_family smc_sock_family_ops = {
3305 .owner = THIS_MODULE,
3306 .create = smc_create,
3309 static int smc_ulp_init(struct sock *sk)
3311 struct socket *tcp = sk->sk_socket;
3312 struct net *net = sock_net(sk);
3313 struct socket *smcsock;
3316 /* only TCP can be replaced */
3317 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP ||
3318 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6))
3319 return -ESOCKTNOSUPPORT;
3320 /* don't handle wq now */
3321 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list)
3324 if (sk->sk_family == AF_INET)
3325 protocol = SMCPROTO_SMC;
3327 protocol = SMCPROTO_SMC6;
3329 smcsock = sock_alloc();
3333 smcsock->type = SOCK_STREAM;
3334 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */
3335 ret = __smc_create(net, smcsock, protocol, 1, tcp);
3337 sock_release(smcsock); /* module_put() which ops won't be NULL */
3341 /* replace tcp socket to smc */
3342 smcsock->file = tcp->file;
3343 smcsock->file->private_data = smcsock;
3344 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */
3345 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */
3351 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk,
3352 const gfp_t priority)
3354 struct inet_connection_sock *icsk = inet_csk(newsk);
3356 /* don't inherit ulp ops to child when listen */
3357 icsk->icsk_ulp_ops = NULL;
3360 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = {
3362 .owner = THIS_MODULE,
3363 .init = smc_ulp_init,
3364 .clone = smc_ulp_clone,
3367 unsigned int smc_net_id;
3369 static __net_init int smc_net_init(struct net *net)
3373 rc = smc_sysctl_net_init(net);
3376 return smc_pnet_net_init(net);
3379 static void __net_exit smc_net_exit(struct net *net)
3381 smc_sysctl_net_exit(net);
3382 smc_pnet_net_exit(net);
3385 static __net_init int smc_net_stat_init(struct net *net)
3387 return smc_stats_init(net);
3390 static void __net_exit smc_net_stat_exit(struct net *net)
3392 smc_stats_exit(net);
3395 static struct pernet_operations smc_net_ops = {
3396 .init = smc_net_init,
3397 .exit = smc_net_exit,
3399 .size = sizeof(struct smc_net),
3402 static struct pernet_operations smc_net_stat_ops = {
3403 .init = smc_net_stat_init,
3404 .exit = smc_net_stat_exit,
3407 static int __init smc_init(void)
3411 rc = register_pernet_subsys(&smc_net_ops);
3415 rc = register_pernet_subsys(&smc_net_stat_ops);
3417 goto out_pernet_subsys;
3419 rc = smc_ism_init();
3421 goto out_pernet_subsys_stat;
3428 rc = smc_pnet_init();
3434 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0);
3438 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0);
3440 goto out_alloc_tcp_ls_wq;
3442 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0);
3444 goto out_alloc_hs_wq;
3446 rc = smc_core_init();
3448 pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3452 rc = smc_llc_init();
3454 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3458 rc = smc_cdc_init();
3460 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3464 rc = proto_register(&smc_proto, 1);
3466 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3470 rc = proto_register(&smc_proto6, 1);
3472 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3476 rc = sock_register(&smc_sock_family_ops);
3478 pr_err("%s: sock_register fails with %d\n", __func__, rc);
3481 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3482 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3484 rc = smc_ib_register_client();
3486 pr_err("%s: ib_register fails with %d\n", __func__, rc);
3490 rc = tcp_register_ulp(&smc_ulp_ops);
3492 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc);
3496 static_branch_enable(&tcp_have_smc);
3500 smc_ib_unregister_client();
3502 sock_unregister(PF_SMC);
3504 proto_unregister(&smc_proto6);
3506 proto_unregister(&smc_proto);
3510 destroy_workqueue(smc_close_wq);
3512 destroy_workqueue(smc_hs_wq);
3513 out_alloc_tcp_ls_wq:
3514 destroy_workqueue(smc_tcp_ls_wq);
3522 out_pernet_subsys_stat:
3523 unregister_pernet_subsys(&smc_net_stat_ops);
3525 unregister_pernet_subsys(&smc_net_ops);
3530 static void __exit smc_exit(void)
3532 static_branch_disable(&tcp_have_smc);
3533 tcp_unregister_ulp(&smc_ulp_ops);
3534 sock_unregister(PF_SMC);
3536 smc_ib_unregister_client();
3538 destroy_workqueue(smc_close_wq);
3539 destroy_workqueue(smc_tcp_ls_wq);
3540 destroy_workqueue(smc_hs_wq);
3541 proto_unregister(&smc_proto6);
3542 proto_unregister(&smc_proto);
3546 unregister_pernet_subsys(&smc_net_stat_ops);
3547 unregister_pernet_subsys(&smc_net_ops);
3551 module_init(smc_init);
3552 module_exit(smc_exit);
3554 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3555 MODULE_DESCRIPTION("smc socket address family");
3556 MODULE_LICENSE("GPL");
3557 MODULE_ALIAS_NETPROTO(PF_SMC);
3558 MODULE_ALIAS_TCP_ULP("smc");
3559 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);