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 if (link->lgr->max_links > 1) {
645 /* optional 2nd link, receive ADD LINK request from server */
646 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
649 struct smc_clc_msg_decline dclc;
651 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
652 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
654 rc = 0; /* no DECLINE received, go with one link */
657 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
658 smc_llc_cli_add_link(link, qentry);
663 static bool smc_isascii(char *hostname)
667 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
668 if (!isascii(hostname[i]))
673 static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
674 struct smc_clc_msg_accept_confirm *clc)
676 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
677 (struct smc_clc_msg_accept_confirm_v2 *)clc;
678 struct smc_clc_first_contact_ext *fce;
681 if (clc->hdr.version == SMC_V1 ||
682 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
685 if (smc->conn.lgr->is_smcd) {
686 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->d1.eid,
688 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2,
691 memcpy(smc->conn.lgr->negotiated_eid, clc_v2->r1.eid,
693 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm_v2,
696 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc_v2) + clc_v2_len);
697 smc->conn.lgr->peer_os = fce->os_type;
698 smc->conn.lgr->peer_smc_release = fce->release;
699 if (smc_isascii(fce->hostname))
700 memcpy(smc->conn.lgr->peer_hostname, fce->hostname,
701 SMC_MAX_HOSTNAME_LEN);
704 static void smcr_conn_save_peer_info(struct smc_sock *smc,
705 struct smc_clc_msg_accept_confirm *clc)
707 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
709 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx;
710 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token);
711 smc->conn.peer_rmbe_size = bufsize;
712 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
713 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
716 static void smcd_conn_save_peer_info(struct smc_sock *smc,
717 struct smc_clc_msg_accept_confirm *clc)
719 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
721 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
722 smc->conn.peer_token = clc->d0.token;
723 /* msg header takes up space in the buffer */
724 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
725 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
726 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
729 static void smc_conn_save_peer_info(struct smc_sock *smc,
730 struct smc_clc_msg_accept_confirm *clc)
732 if (smc->conn.lgr->is_smcd)
733 smcd_conn_save_peer_info(smc, clc);
735 smcr_conn_save_peer_info(smc, clc);
736 smc_conn_save_peer_info_fce(smc, clc);
739 static void smc_link_save_peer_info(struct smc_link *link,
740 struct smc_clc_msg_accept_confirm *clc,
741 struct smc_init_info *ini)
743 link->peer_qpn = ntoh24(clc->r0.qpn);
744 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE);
745 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac));
746 link->peer_psn = ntoh24(clc->r0.psn);
747 link->peer_mtu = clc->r0.qp_mtu;
750 static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
751 struct smc_stats_fback *fback_arr)
755 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) {
756 if (fback_arr[cnt].fback_code == smc->fallback_rsn) {
757 fback_arr[cnt].count++;
760 if (!fback_arr[cnt].fback_code) {
761 fback_arr[cnt].fback_code = smc->fallback_rsn;
762 fback_arr[cnt].count++;
768 static void smc_stat_fallback(struct smc_sock *smc)
770 struct net *net = sock_net(&smc->sk);
772 mutex_lock(&net->smc.mutex_fback_rsn);
773 if (smc->listen_smc) {
774 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv);
775 net->smc.fback_rsn->srv_fback_cnt++;
777 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
778 net->smc.fback_rsn->clnt_fback_cnt++;
780 mutex_unlock(&net->smc.mutex_fback_rsn);
783 /* must be called under rcu read lock */
784 static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
786 struct socket_wq *wq;
789 wq = rcu_dereference(smc->sk.sk_wq);
790 if (!skwq_has_sleeper(wq))
793 /* wake up smc sk->sk_wq */
795 /* sk_state_change */
796 wake_up_interruptible_all(&wq->wait);
798 flags = key_to_poll(key);
799 if (flags & (EPOLLIN | EPOLLOUT))
800 /* sk_data_ready or sk_write_space */
801 wake_up_interruptible_sync_poll(&wq->wait, flags);
802 else if (flags & EPOLLERR)
803 /* sk_error_report */
804 wake_up_interruptible_poll(&wq->wait, flags);
808 static int smc_fback_mark_woken(wait_queue_entry_t *wait,
809 unsigned int mode, int sync, void *key)
811 struct smc_mark_woken *mark =
812 container_of(wait, struct smc_mark_woken, wait_entry);
819 static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
820 void (*clcsock_callback)(struct sock *sk))
822 struct smc_mark_woken mark = { .woken = false };
823 struct socket_wq *wq;
825 init_waitqueue_func_entry(&mark.wait_entry,
826 smc_fback_mark_woken);
828 wq = rcu_dereference(clcsk->sk_wq);
831 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
832 clcsock_callback(clcsk);
833 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
836 smc_fback_wakeup_waitqueue(smc, mark.key);
841 static void smc_fback_state_change(struct sock *clcsk)
843 struct smc_sock *smc;
845 read_lock_bh(&clcsk->sk_callback_lock);
846 smc = smc_clcsock_user_data(clcsk);
848 smc_fback_forward_wakeup(smc, clcsk,
849 smc->clcsk_state_change);
850 read_unlock_bh(&clcsk->sk_callback_lock);
853 static void smc_fback_data_ready(struct sock *clcsk)
855 struct smc_sock *smc;
857 read_lock_bh(&clcsk->sk_callback_lock);
858 smc = smc_clcsock_user_data(clcsk);
860 smc_fback_forward_wakeup(smc, clcsk,
861 smc->clcsk_data_ready);
862 read_unlock_bh(&clcsk->sk_callback_lock);
865 static void smc_fback_write_space(struct sock *clcsk)
867 struct smc_sock *smc;
869 read_lock_bh(&clcsk->sk_callback_lock);
870 smc = smc_clcsock_user_data(clcsk);
872 smc_fback_forward_wakeup(smc, clcsk,
873 smc->clcsk_write_space);
874 read_unlock_bh(&clcsk->sk_callback_lock);
877 static void smc_fback_error_report(struct sock *clcsk)
879 struct smc_sock *smc;
881 read_lock_bh(&clcsk->sk_callback_lock);
882 smc = smc_clcsock_user_data(clcsk);
884 smc_fback_forward_wakeup(smc, clcsk,
885 smc->clcsk_error_report);
886 read_unlock_bh(&clcsk->sk_callback_lock);
889 static void smc_fback_replace_callbacks(struct smc_sock *smc)
891 struct sock *clcsk = smc->clcsock->sk;
893 write_lock_bh(&clcsk->sk_callback_lock);
894 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
896 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change,
897 &smc->clcsk_state_change);
898 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready,
899 &smc->clcsk_data_ready);
900 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space,
901 &smc->clcsk_write_space);
902 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report,
903 &smc->clcsk_error_report);
905 write_unlock_bh(&clcsk->sk_callback_lock);
908 static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
912 mutex_lock(&smc->clcsock_release_lock);
918 smc->use_fallback = true;
919 smc->fallback_rsn = reason_code;
920 smc_stat_fallback(smc);
921 trace_smc_switch_to_fallback(smc, reason_code);
922 if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
923 smc->clcsock->file = smc->sk.sk_socket->file;
924 smc->clcsock->file->private_data = smc->clcsock;
925 smc->clcsock->wq.fasync_list =
926 smc->sk.sk_socket->wq.fasync_list;
928 /* There might be some wait entries remaining
929 * in smc sk->sk_wq and they should be woken up
930 * as clcsock's wait queue is woken up.
932 smc_fback_replace_callbacks(smc);
935 mutex_unlock(&smc->clcsock_release_lock);
939 /* fall back during connect */
940 static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
942 struct net *net = sock_net(&smc->sk);
945 rc = smc_switch_to_fallback(smc, reason_code);
946 if (rc) { /* fallback fails */
947 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
948 if (smc->sk.sk_state == SMC_INIT)
949 sock_put(&smc->sk); /* passive closing */
952 smc_copy_sock_settings_to_clc(smc);
953 smc->connect_nonblock = 0;
954 if (smc->sk.sk_state == SMC_INIT)
955 smc->sk.sk_state = SMC_ACTIVE;
959 /* decline and fall back during connect */
960 static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
963 struct net *net = sock_net(&smc->sk);
966 if (reason_code < 0) { /* error, fallback is not possible */
967 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
968 if (smc->sk.sk_state == SMC_INIT)
969 sock_put(&smc->sk); /* passive closing */
972 if (reason_code != SMC_CLC_DECL_PEERDECL) {
973 rc = smc_clc_send_decline(smc, reason_code, version);
975 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
976 if (smc->sk.sk_state == SMC_INIT)
977 sock_put(&smc->sk); /* passive closing */
981 return smc_connect_fallback(smc, reason_code);
984 static void smc_conn_abort(struct smc_sock *smc, int local_first)
986 struct smc_connection *conn = &smc->conn;
987 struct smc_link_group *lgr = conn->lgr;
988 bool lgr_valid = false;
990 if (smc_conn_lgr_valid(conn))
994 if (local_first && lgr_valid)
995 smc_lgr_cleanup_early(lgr);
998 /* check if there is a rdma device available for this connection. */
999 /* called for connect and listen */
1000 static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
1002 /* PNET table look up: search active ib_device and port
1003 * within same PNETID that also contains the ethernet device
1004 * used for the internal TCP socket
1006 smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
1007 if (!ini->check_smcrv2 && !ini->ib_dev)
1008 return SMC_CLC_DECL_NOSMCRDEV;
1009 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1010 return SMC_CLC_DECL_NOSMCRDEV;
1014 /* check if there is an ISM device available for this connection. */
1015 /* called for connect and listen */
1016 static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1018 /* Find ISM device with same PNETID as connecting interface */
1019 smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1020 if (!ini->ism_dev[0])
1021 return SMC_CLC_DECL_NOSMCDDEV;
1023 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1027 /* is chid unique for the ism devices that are already determined? */
1028 static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1031 int i = (!ini->ism_dev[0]) ? 1 : 0;
1033 for (; i < cnt; i++)
1034 if (ini->ism_chid[i] == chid)
1039 /* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1040 * PNETID matching net_device)
1042 static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1043 struct smc_init_info *ini)
1045 int rc = SMC_CLC_DECL_NOSMCDDEV;
1046 struct smcd_dev *smcd;
1050 if (smcd_indicated(ini->smc_type_v1))
1051 rc = 0; /* already initialized for V1 */
1052 mutex_lock(&smcd_dev_list.mutex);
1053 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1054 if (smcd->going_away || smcd == ini->ism_dev[0])
1056 chid = smc_ism_get_chid(smcd);
1057 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1059 if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1060 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1061 ini->ism_dev[i] = smcd;
1062 ini->ism_chid[i] = chid;
1063 ini->is_smcd = true;
1066 if (i > SMC_MAX_ISM_DEVS)
1070 mutex_unlock(&smcd_dev_list.mutex);
1071 ini->ism_offered_cnt = i - 1;
1072 if (!ini->ism_dev[0] && !ini->ism_dev[1])
1073 ini->smcd_version = 0;
1078 /* Check for VLAN ID and register it on ISM device just for CLC handshake */
1079 static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
1080 struct smc_init_info *ini)
1082 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1083 return SMC_CLC_DECL_ISMVLANERR;
1087 static int smc_find_proposal_devices(struct smc_sock *smc,
1088 struct smc_init_info *ini)
1092 /* check if there is an ism device available */
1093 if (!(ini->smcd_version & SMC_V1) ||
1094 smc_find_ism_device(smc, ini) ||
1095 smc_connect_ism_vlan_setup(smc, ini))
1096 ini->smcd_version &= ~SMC_V1;
1097 /* else ISM V1 is supported for this connection */
1099 /* check if there is an rdma device available */
1100 if (!(ini->smcr_version & SMC_V1) ||
1101 smc_find_rdma_device(smc, ini))
1102 ini->smcr_version &= ~SMC_V1;
1103 /* else RDMA is supported for this connection */
1105 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1106 ini->smcr_version & SMC_V1);
1108 /* check if there is an ism v2 device available */
1109 if (!(ini->smcd_version & SMC_V2) ||
1110 !smc_ism_is_v2_capable() ||
1111 smc_find_ism_v2_device_clnt(smc, ini))
1112 ini->smcd_version &= ~SMC_V2;
1114 /* check if there is an rdma v2 device available */
1115 ini->check_smcrv2 = true;
1116 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1117 if (!(ini->smcr_version & SMC_V2) ||
1118 smc->clcsock->sk->sk_family != AF_INET ||
1119 !smc_clc_ueid_count() ||
1120 smc_find_rdma_device(smc, ini))
1121 ini->smcr_version &= ~SMC_V2;
1122 ini->check_smcrv2 = false;
1124 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1125 ini->smcr_version & SMC_V2);
1127 /* if neither ISM nor RDMA are supported, fallback */
1128 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1129 rc = SMC_CLC_DECL_NOSMCDEV;
1134 /* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1135 * used, the VLAN ID will be registered again during the connection setup.
1137 static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc,
1138 struct smc_init_info *ini)
1140 if (!smcd_indicated(ini->smc_type_v1))
1142 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1143 return SMC_CLC_DECL_CNFERR;
1147 #define SMC_CLC_MAX_ACCEPT_LEN \
1148 (sizeof(struct smc_clc_msg_accept_confirm_v2) + \
1149 sizeof(struct smc_clc_first_contact_ext_v2x) + \
1150 sizeof(struct smc_clc_msg_trail))
1152 /* CLC handshake during connect */
1153 static int smc_connect_clc(struct smc_sock *smc,
1154 struct smc_clc_msg_accept_confirm_v2 *aclc2,
1155 struct smc_init_info *ini)
1159 /* do inband token exchange */
1160 rc = smc_clc_send_proposal(smc, ini);
1163 /* receive SMC Accept CLC message */
1164 return smc_clc_wait_msg(smc, aclc2, SMC_CLC_MAX_ACCEPT_LEN,
1165 SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1168 void smc_fill_gid_list(struct smc_link_group *lgr,
1169 struct smc_gidlist *gidlist,
1170 struct smc_ib_device *known_dev, u8 *known_gid)
1172 struct smc_init_info *alt_ini = NULL;
1174 memset(gidlist, 0, sizeof(*gidlist));
1175 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1177 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1181 alt_ini->vlan_id = lgr->vlan_id;
1182 alt_ini->check_smcrv2 = true;
1183 alt_ini->smcrv2.saddr = lgr->saddr;
1184 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1186 if (!alt_ini->smcrv2.ib_dev_v2)
1189 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1196 static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1197 struct smc_clc_msg_accept_confirm *aclc,
1198 struct smc_init_info *ini)
1200 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1201 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1202 struct smc_clc_first_contact_ext *fce =
1203 smc_get_clc_first_contact_ext(clc_v2, false);
1206 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1209 if (fce->v2_direct) {
1210 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1211 ini->smcrv2.uses_gateway = false;
1213 if (smc_ib_find_route(smc->clcsock->sk->sk_rcv_saddr,
1214 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1215 ini->smcrv2.nexthop_mac,
1216 &ini->smcrv2.uses_gateway))
1217 return SMC_CLC_DECL_NOROUTE;
1218 if (!ini->smcrv2.uses_gateway) {
1219 /* mismatch: peer claims indirect, but its direct */
1220 return SMC_CLC_DECL_NOINDIRECT;
1224 ini->release_nr = fce->release;
1225 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1232 /* setup for RDMA connection of client */
1233 static int smc_connect_rdma(struct smc_sock *smc,
1234 struct smc_clc_msg_accept_confirm *aclc,
1235 struct smc_init_info *ini)
1237 int i, reason_code = 0;
1238 struct smc_link *link;
1241 ini->is_smcd = false;
1242 ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1243 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1244 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1245 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1246 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1247 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1248 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1250 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1254 mutex_lock(&smc_client_lgr_pending);
1255 reason_code = smc_conn_create(smc, ini);
1257 mutex_unlock(&smc_client_lgr_pending);
1261 smc_conn_save_peer_info(smc, aclc);
1263 if (ini->first_contact_local) {
1264 link = smc->conn.lnk;
1266 /* set link that was assigned by server */
1268 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1269 struct smc_link *l = &smc->conn.lgr->lnk[i];
1271 if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1272 !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1274 (aclc->hdr.version > SMC_V1 ||
1275 !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1276 sizeof(l->peer_mac)))) {
1282 reason_code = SMC_CLC_DECL_NOSRVLINK;
1285 smc_switch_link_and_count(&smc->conn, link);
1288 /* create send buffer and rmb */
1289 if (smc_buf_create(smc, false)) {
1290 reason_code = SMC_CLC_DECL_MEM;
1294 if (ini->first_contact_local)
1295 smc_link_save_peer_info(link, aclc, ini);
1297 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1298 reason_code = SMC_CLC_DECL_ERR_RTOK;
1302 smc_close_init(smc);
1305 if (ini->first_contact_local) {
1306 if (smc_ib_ready_link(link)) {
1307 reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1311 /* reg sendbufs if they were vzalloced */
1312 if (smc->conn.sndbuf_desc->is_vm) {
1313 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) {
1314 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1318 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1319 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1324 if (aclc->hdr.version > SMC_V1) {
1325 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1326 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1328 eid = clc_v2->r1.eid;
1329 if (ini->first_contact_local)
1330 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1331 link->smcibdev, link->gid);
1334 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1335 aclc->hdr.version, eid, ini);
1341 if (ini->first_contact_local) {
1342 /* QP confirmation over RoCE fabric */
1343 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1344 reason_code = smcr_clnt_conf_first_link(smc);
1345 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1349 mutex_unlock(&smc_client_lgr_pending);
1351 smc_copy_sock_settings_to_clc(smc);
1352 smc->connect_nonblock = 0;
1353 if (smc->sk.sk_state == SMC_INIT)
1354 smc->sk.sk_state = SMC_ACTIVE;
1358 smc_conn_abort(smc, ini->first_contact_local);
1359 mutex_unlock(&smc_client_lgr_pending);
1360 smc->connect_nonblock = 0;
1365 /* The server has chosen one of the proposed ISM devices for the communication.
1366 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1369 smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm_v2 *aclc,
1370 struct smc_init_info *ini)
1374 for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1375 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1376 ini->ism_selected = i;
1384 /* setup for ISM connection of client */
1385 static int smc_connect_ism(struct smc_sock *smc,
1386 struct smc_clc_msg_accept_confirm *aclc,
1387 struct smc_init_info *ini)
1392 ini->is_smcd = true;
1393 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1395 if (aclc->hdr.version == SMC_V2) {
1396 struct smc_clc_msg_accept_confirm_v2 *aclc_v2 =
1397 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1399 if (ini->first_contact_peer) {
1400 struct smc_clc_first_contact_ext *fce =
1401 smc_get_clc_first_contact_ext(aclc_v2, true);
1403 ini->release_nr = fce->release;
1404 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1409 rc = smc_v2_determine_accepted_chid(aclc_v2, ini);
1413 ini->ism_peer_gid[ini->ism_selected] = aclc->d0.gid;
1415 /* there is only one lgr role for SMC-D; use server lock */
1416 mutex_lock(&smc_server_lgr_pending);
1417 rc = smc_conn_create(smc, ini);
1419 mutex_unlock(&smc_server_lgr_pending);
1423 /* Create send and receive buffers */
1424 rc = smc_buf_create(smc, true);
1426 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1430 smc_conn_save_peer_info(smc, aclc);
1431 smc_close_init(smc);
1435 if (aclc->hdr.version > SMC_V1) {
1436 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1437 (struct smc_clc_msg_accept_confirm_v2 *)aclc;
1439 eid = clc_v2->d1.eid;
1442 rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1443 aclc->hdr.version, eid, ini);
1446 mutex_unlock(&smc_server_lgr_pending);
1448 smc_copy_sock_settings_to_clc(smc);
1449 smc->connect_nonblock = 0;
1450 if (smc->sk.sk_state == SMC_INIT)
1451 smc->sk.sk_state = SMC_ACTIVE;
1455 smc_conn_abort(smc, ini->first_contact_local);
1456 mutex_unlock(&smc_server_lgr_pending);
1457 smc->connect_nonblock = 0;
1462 /* check if received accept type and version matches a proposed one */
1463 static int smc_connect_check_aclc(struct smc_init_info *ini,
1464 struct smc_clc_msg_accept_confirm *aclc)
1466 if (aclc->hdr.typev1 != SMC_TYPE_R &&
1467 aclc->hdr.typev1 != SMC_TYPE_D)
1468 return SMC_CLC_DECL_MODEUNSUPP;
1470 if (aclc->hdr.version >= SMC_V2) {
1471 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1472 !smcr_indicated(ini->smc_type_v2)) ||
1473 (aclc->hdr.typev1 == SMC_TYPE_D &&
1474 !smcd_indicated(ini->smc_type_v2)))
1475 return SMC_CLC_DECL_MODEUNSUPP;
1477 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1478 !smcr_indicated(ini->smc_type_v1)) ||
1479 (aclc->hdr.typev1 == SMC_TYPE_D &&
1480 !smcd_indicated(ini->smc_type_v1)))
1481 return SMC_CLC_DECL_MODEUNSUPP;
1487 /* perform steps before actually connecting */
1488 static int __smc_connect(struct smc_sock *smc)
1490 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1491 struct smc_clc_msg_accept_confirm_v2 *aclc2;
1492 struct smc_clc_msg_accept_confirm *aclc;
1493 struct smc_init_info *ini = NULL;
1497 if (smc->use_fallback)
1498 return smc_connect_fallback(smc, smc->fallback_rsn);
1500 /* if peer has not signalled SMC-capability, fall back */
1501 if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1502 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1504 /* IPSec connections opt out of SMC optimizations */
1505 if (using_ipsec(smc))
1506 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1509 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1511 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1514 ini->smcd_version = SMC_V1 | SMC_V2;
1515 ini->smcr_version = SMC_V1 | SMC_V2;
1516 ini->smc_type_v1 = SMC_TYPE_B;
1517 ini->smc_type_v2 = SMC_TYPE_B;
1519 /* get vlan id from IP device */
1520 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1521 ini->smcd_version &= ~SMC_V1;
1522 ini->smcr_version = 0;
1523 ini->smc_type_v1 = SMC_TYPE_N;
1524 if (!ini->smcd_version) {
1525 rc = SMC_CLC_DECL_GETVLANERR;
1530 rc = smc_find_proposal_devices(smc, ini);
1534 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1536 rc = SMC_CLC_DECL_MEM;
1539 aclc2 = (struct smc_clc_msg_accept_confirm_v2 *)buf;
1540 aclc = (struct smc_clc_msg_accept_confirm *)aclc2;
1542 /* perform CLC handshake */
1543 rc = smc_connect_clc(smc, aclc2, ini);
1545 /* -EAGAIN on timeout, see tcp_recvmsg() */
1546 if (rc == -EAGAIN) {
1548 smc->sk.sk_err = ETIMEDOUT;
1553 /* check if smc modes and versions of CLC proposal and accept match */
1554 rc = smc_connect_check_aclc(ini, aclc);
1555 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1559 /* depending on previous steps, connect using rdma or ism */
1560 if (aclc->hdr.typev1 == SMC_TYPE_R) {
1561 ini->smcr_version = version;
1562 rc = smc_connect_rdma(smc, aclc, ini);
1563 } else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1564 ini->smcd_version = version;
1565 rc = smc_connect_ism(smc, aclc, ini);
1570 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1571 smc_connect_ism_vlan_cleanup(smc, ini);
1577 smc_connect_ism_vlan_cleanup(smc, ini);
1581 return smc_connect_decline_fallback(smc, rc, version);
1584 static void smc_connect_work(struct work_struct *work)
1586 struct smc_sock *smc = container_of(work, struct smc_sock,
1588 long timeo = smc->sk.sk_sndtimeo;
1592 timeo = MAX_SCHEDULE_TIMEOUT;
1593 lock_sock(smc->clcsock->sk);
1594 if (smc->clcsock->sk->sk_err) {
1595 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1596 } else if ((1 << smc->clcsock->sk->sk_state) &
1597 (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1598 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1599 if ((rc == -EPIPE) &&
1600 ((1 << smc->clcsock->sk->sk_state) &
1601 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1604 release_sock(smc->clcsock->sk);
1605 lock_sock(&smc->sk);
1606 if (rc != 0 || smc->sk.sk_err) {
1607 smc->sk.sk_state = SMC_CLOSED;
1608 if (rc == -EPIPE || rc == -EAGAIN)
1609 smc->sk.sk_err = EPIPE;
1610 else if (rc == -ECONNREFUSED)
1611 smc->sk.sk_err = ECONNREFUSED;
1612 else if (signal_pending(current))
1613 smc->sk.sk_err = -sock_intr_errno(timeo);
1614 sock_put(&smc->sk); /* passive closing */
1618 rc = __smc_connect(smc);
1620 smc->sk.sk_err = -rc;
1623 if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1624 if (smc->sk.sk_err) {
1625 smc->sk.sk_state_change(&smc->sk);
1626 } else { /* allow polling before and after fallback decision */
1627 smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1628 smc->sk.sk_write_space(&smc->sk);
1631 release_sock(&smc->sk);
1634 static int smc_connect(struct socket *sock, struct sockaddr *addr,
1635 int alen, int flags)
1637 struct sock *sk = sock->sk;
1638 struct smc_sock *smc;
1643 /* separate smc parameter checking to be safe */
1644 if (alen < sizeof(addr->sa_family))
1646 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1650 switch (sock->state) {
1655 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1658 if (sk->sk_state == SMC_ACTIVE)
1661 case SS_UNCONNECTED:
1662 sock->state = SS_CONNECTING;
1666 switch (sk->sk_state) {
1670 rc = sock_error(sk) ? : -ECONNABORTED;
1671 sock->state = SS_UNCONNECTED;
1680 smc_copy_sock_settings_to_clc(smc);
1681 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1682 if (smc->connect_nonblock) {
1686 rc = kernel_connect(smc->clcsock, addr, alen, flags);
1687 if (rc && rc != -EINPROGRESS)
1690 if (smc->use_fallback) {
1691 sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1694 sock_hold(&smc->sk); /* sock put in passive closing */
1695 if (flags & O_NONBLOCK) {
1696 if (queue_work(smc_hs_wq, &smc->connect_work))
1697 smc->connect_nonblock = 1;
1701 rc = __smc_connect(smc);
1708 sock->state = SS_CONNECTED;
1715 static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1717 struct socket *new_clcsock = NULL;
1718 struct sock *lsk = &lsmc->sk;
1719 struct sock *new_sk;
1723 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1726 lsk->sk_err = ENOMEM;
1731 *new_smc = smc_sk(new_sk);
1733 mutex_lock(&lsmc->clcsock_release_lock);
1735 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1736 mutex_unlock(&lsmc->clcsock_release_lock);
1738 if (rc < 0 && rc != -EAGAIN)
1740 if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1741 new_sk->sk_prot->unhash(new_sk);
1743 sock_release(new_clcsock);
1744 new_sk->sk_state = SMC_CLOSED;
1745 sock_set_flag(new_sk, SOCK_DEAD);
1746 sock_put(new_sk); /* final */
1751 /* new clcsock has inherited the smc listen-specific sk_data_ready
1752 * function; switch it back to the original sk_data_ready function
1754 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1756 /* if new clcsock has also inherited the fallback-specific callback
1757 * functions, switch them back to the original ones.
1759 if (lsmc->use_fallback) {
1760 if (lsmc->clcsk_state_change)
1761 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1762 if (lsmc->clcsk_write_space)
1763 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1764 if (lsmc->clcsk_error_report)
1765 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1768 (*new_smc)->clcsock = new_clcsock;
1773 /* add a just created sock to the accept queue of the listen sock as
1774 * candidate for a following socket accept call from user space
1776 static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1778 struct smc_sock *par = smc_sk(parent);
1780 sock_hold(sk); /* sock_put in smc_accept_unlink () */
1781 spin_lock(&par->accept_q_lock);
1782 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1783 spin_unlock(&par->accept_q_lock);
1784 sk_acceptq_added(parent);
1787 /* remove a socket from the accept queue of its parental listening socket */
1788 static void smc_accept_unlink(struct sock *sk)
1790 struct smc_sock *par = smc_sk(sk)->listen_smc;
1792 spin_lock(&par->accept_q_lock);
1793 list_del_init(&smc_sk(sk)->accept_q);
1794 spin_unlock(&par->accept_q_lock);
1795 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1796 sock_put(sk); /* sock_hold in smc_accept_enqueue */
1799 /* remove a sock from the accept queue to bind it to a new socket created
1800 * for a socket accept call from user space
1802 struct sock *smc_accept_dequeue(struct sock *parent,
1803 struct socket *new_sock)
1805 struct smc_sock *isk, *n;
1806 struct sock *new_sk;
1808 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1809 new_sk = (struct sock *)isk;
1811 smc_accept_unlink(new_sk);
1812 if (new_sk->sk_state == SMC_CLOSED) {
1813 new_sk->sk_prot->unhash(new_sk);
1815 sock_release(isk->clcsock);
1816 isk->clcsock = NULL;
1818 sock_put(new_sk); /* final */
1822 sock_graft(new_sk, new_sock);
1823 new_sock->state = SS_CONNECTED;
1824 if (isk->use_fallback) {
1825 smc_sk(new_sk)->clcsock->file = new_sock->file;
1826 isk->clcsock->file->private_data = isk->clcsock;
1834 /* clean up for a created but never accepted sock */
1835 void smc_close_non_accepted(struct sock *sk)
1837 struct smc_sock *smc = smc_sk(sk);
1839 sock_hold(sk); /* sock_put below */
1841 if (!sk->sk_lingertime)
1842 /* wait for peer closing */
1843 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT);
1846 sock_put(sk); /* sock_hold above */
1847 sock_put(sk); /* final sock_put */
1850 static int smcr_serv_conf_first_link(struct smc_sock *smc)
1852 struct smc_link *link = smc->conn.lnk;
1853 struct smc_llc_qentry *qentry;
1856 /* reg the sndbuf if it was vzalloced*/
1857 if (smc->conn.sndbuf_desc->is_vm) {
1858 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1859 return SMC_CLC_DECL_ERR_REGBUF;
1863 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1864 return SMC_CLC_DECL_ERR_REGBUF;
1866 /* send CONFIRM LINK request to client over the RoCE fabric */
1867 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1869 return SMC_CLC_DECL_TIMEOUT_CL;
1871 /* receive CONFIRM LINK response from client over the RoCE fabric */
1872 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1873 SMC_LLC_CONFIRM_LINK);
1875 struct smc_clc_msg_decline dclc;
1877 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1878 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1879 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1881 smc_llc_save_peer_uid(qentry);
1882 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1883 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1885 return SMC_CLC_DECL_RMBE_EC;
1887 /* confirm_rkey is implicit on 1st contact */
1888 smc->conn.rmb_desc->is_conf_rkey = true;
1890 smc_llc_link_active(link);
1891 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1893 if (link->lgr->max_links > 1) {
1894 down_write(&link->lgr->llc_conf_mutex);
1895 /* initial contact - try to establish second link */
1896 smc_llc_srv_add_link(link, NULL);
1897 up_write(&link->lgr->llc_conf_mutex);
1902 /* listen worker: finish */
1903 static void smc_listen_out(struct smc_sock *new_smc)
1905 struct smc_sock *lsmc = new_smc->listen_smc;
1906 struct sock *newsmcsk = &new_smc->sk;
1908 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1909 atomic_dec(&lsmc->queued_smc_hs);
1911 if (lsmc->sk.sk_state == SMC_LISTEN) {
1912 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1913 smc_accept_enqueue(&lsmc->sk, newsmcsk);
1914 release_sock(&lsmc->sk);
1915 } else { /* no longer listening */
1916 smc_close_non_accepted(newsmcsk);
1919 /* Wake up accept */
1920 lsmc->sk.sk_data_ready(&lsmc->sk);
1921 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1924 /* listen worker: finish in state connected */
1925 static void smc_listen_out_connected(struct smc_sock *new_smc)
1927 struct sock *newsmcsk = &new_smc->sk;
1929 if (newsmcsk->sk_state == SMC_INIT)
1930 newsmcsk->sk_state = SMC_ACTIVE;
1932 smc_listen_out(new_smc);
1935 /* listen worker: finish in error state */
1936 static void smc_listen_out_err(struct smc_sock *new_smc)
1938 struct sock *newsmcsk = &new_smc->sk;
1939 struct net *net = sock_net(newsmcsk);
1941 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1942 if (newsmcsk->sk_state == SMC_INIT)
1943 sock_put(&new_smc->sk); /* passive closing */
1944 newsmcsk->sk_state = SMC_CLOSED;
1946 smc_listen_out(new_smc);
1949 /* listen worker: decline and fall back if possible */
1950 static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1951 int local_first, u8 version)
1953 /* RDMA setup failed, switch back to TCP */
1954 smc_conn_abort(new_smc, local_first);
1955 if (reason_code < 0 ||
1956 smc_switch_to_fallback(new_smc, reason_code)) {
1957 /* error, no fallback possible */
1958 smc_listen_out_err(new_smc);
1961 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1962 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1963 smc_listen_out_err(new_smc);
1967 smc_listen_out_connected(new_smc);
1970 /* listen worker: version checking */
1971 static int smc_listen_v2_check(struct smc_sock *new_smc,
1972 struct smc_clc_msg_proposal *pclc,
1973 struct smc_init_info *ini)
1975 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
1976 struct smc_clc_v2_extension *pclc_v2_ext;
1977 int rc = SMC_CLC_DECL_PEERNOSMC;
1979 ini->smc_type_v1 = pclc->hdr.typev1;
1980 ini->smc_type_v2 = pclc->hdr.typev2;
1981 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1982 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1983 if (pclc->hdr.version > SMC_V1) {
1984 if (smcd_indicated(ini->smc_type_v2))
1985 ini->smcd_version |= SMC_V2;
1986 if (smcr_indicated(ini->smc_type_v2))
1987 ini->smcr_version |= SMC_V2;
1989 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
1990 rc = SMC_CLC_DECL_PEERNOSMC;
1993 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1995 ini->smcd_version &= ~SMC_V2;
1996 ini->smcr_version &= ~SMC_V2;
1997 rc = SMC_CLC_DECL_NOV2EXT;
2000 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
2001 if (ini->smcd_version & SMC_V2) {
2002 if (!smc_ism_is_v2_capable()) {
2003 ini->smcd_version &= ~SMC_V2;
2004 rc = SMC_CLC_DECL_NOISM2SUPP;
2005 } else if (!pclc_smcd_v2_ext) {
2006 ini->smcd_version &= ~SMC_V2;
2007 rc = SMC_CLC_DECL_NOV2DEXT;
2008 } else if (!pclc_v2_ext->hdr.eid_cnt &&
2009 !pclc_v2_ext->hdr.flag.seid) {
2010 ini->smcd_version &= ~SMC_V2;
2011 rc = SMC_CLC_DECL_NOUEID;
2014 if (ini->smcr_version & SMC_V2) {
2015 if (!pclc_v2_ext->hdr.eid_cnt) {
2016 ini->smcr_version &= ~SMC_V2;
2017 rc = SMC_CLC_DECL_NOUEID;
2021 ini->release_nr = pclc_v2_ext->hdr.flag.release;
2022 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE)
2023 ini->release_nr = SMC_RELEASE;
2026 if (!ini->smcd_version && !ini->smcr_version)
2032 /* listen worker: check prefixes */
2033 static int smc_listen_prfx_check(struct smc_sock *new_smc,
2034 struct smc_clc_msg_proposal *pclc)
2036 struct smc_clc_msg_proposal_prefix *pclc_prfx;
2037 struct socket *newclcsock = new_smc->clcsock;
2039 if (pclc->hdr.typev1 == SMC_TYPE_N)
2041 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2042 if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2043 return SMC_CLC_DECL_DIFFPREFIX;
2048 /* listen worker: initialize connection and buffers */
2049 static int smc_listen_rdma_init(struct smc_sock *new_smc,
2050 struct smc_init_info *ini)
2054 /* allocate connection / link group */
2055 rc = smc_conn_create(new_smc, ini);
2059 /* create send buffer and rmb */
2060 if (smc_buf_create(new_smc, false)) {
2061 smc_conn_abort(new_smc, ini->first_contact_local);
2062 return SMC_CLC_DECL_MEM;
2068 /* listen worker: initialize connection and buffers for SMC-D */
2069 static int smc_listen_ism_init(struct smc_sock *new_smc,
2070 struct smc_init_info *ini)
2074 rc = smc_conn_create(new_smc, ini);
2078 /* Create send and receive buffers */
2079 rc = smc_buf_create(new_smc, true);
2081 smc_conn_abort(new_smc, ini->first_contact_local);
2082 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2089 static bool smc_is_already_selected(struct smcd_dev *smcd,
2090 struct smc_init_info *ini,
2095 for (i = 0; i < matches; i++)
2096 if (smcd == ini->ism_dev[i])
2102 /* check for ISM devices matching proposed ISM devices */
2103 static void smc_check_ism_v2_match(struct smc_init_info *ini,
2104 u16 proposed_chid, u64 proposed_gid,
2105 unsigned int *matches)
2107 struct smcd_dev *smcd;
2109 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2110 if (smcd->going_away)
2112 if (smc_is_already_selected(smcd, ini, *matches))
2114 if (smc_ism_get_chid(smcd) == proposed_chid &&
2115 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2116 ini->ism_peer_gid[*matches] = proposed_gid;
2117 ini->ism_dev[*matches] = smcd;
2124 static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini)
2130 static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc,
2131 struct smc_clc_msg_proposal *pclc,
2132 struct smc_init_info *ini)
2134 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
2135 struct smc_clc_v2_extension *smc_v2_ext;
2136 struct smc_clc_msg_smcd *pclc_smcd;
2137 unsigned int matches = 0;
2142 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2145 pclc_smcd = smc_get_clc_msg_smcd(pclc);
2146 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2147 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2149 mutex_lock(&smcd_dev_list.mutex);
2150 if (pclc_smcd->ism.chid)
2151 /* check for ISM device matching proposed native ISM device */
2152 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2153 ntohll(pclc_smcd->ism.gid), &matches);
2154 for (i = 1; i <= smc_v2_ext->hdr.ism_gid_cnt; i++) {
2155 /* check for ISM devices matching proposed non-native ISM
2158 smc_check_ism_v2_match(ini,
2159 ntohs(smcd_v2_ext->gidchid[i - 1].chid),
2160 ntohll(smcd_v2_ext->gidchid[i - 1].gid),
2163 mutex_unlock(&smcd_dev_list.mutex);
2165 if (!ini->ism_dev[0]) {
2166 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2170 smc_ism_get_system_eid(&eid);
2171 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2172 smcd_v2_ext->system_eid, eid))
2175 /* separate - outside the smcd_dev_list.lock */
2176 smcd_version = ini->smcd_version;
2177 for (i = 0; i < matches; i++) {
2178 ini->smcd_version = SMC_V2;
2179 ini->is_smcd = true;
2180 ini->ism_selected = i;
2181 rc = smc_listen_ism_init(new_smc, ini);
2183 smc_find_ism_store_rc(rc, ini);
2184 /* try next active ISM device */
2187 return; /* matching and usable V2 ISM device found */
2189 /* no V2 ISM device could be initialized */
2190 ini->smcd_version = smcd_version; /* restore original value */
2191 ini->negotiated_eid[0] = 0;
2194 ini->smcd_version &= ~SMC_V2;
2195 ini->ism_dev[0] = NULL;
2196 ini->is_smcd = false;
2199 static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2200 struct smc_clc_msg_proposal *pclc,
2201 struct smc_init_info *ini)
2203 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2206 /* check if ISM V1 is available */
2207 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1))
2209 ini->is_smcd = true; /* prepare ISM check */
2210 ini->ism_peer_gid[0] = ntohll(pclc_smcd->ism.gid);
2211 rc = smc_find_ism_device(new_smc, ini);
2214 ini->ism_selected = 0;
2215 rc = smc_listen_ism_init(new_smc, ini);
2217 return; /* V1 ISM device found */
2220 smc_find_ism_store_rc(rc, ini);
2221 ini->smcd_version &= ~SMC_V1;
2222 ini->ism_dev[0] = NULL;
2223 ini->is_smcd = false;
2226 /* listen worker: register buffers */
2227 static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2229 struct smc_connection *conn = &new_smc->conn;
2232 /* reg sendbufs if they were vzalloced */
2233 if (conn->sndbuf_desc->is_vm) {
2234 if (smcr_lgr_reg_sndbufs(conn->lnk,
2236 return SMC_CLC_DECL_ERR_REGBUF;
2238 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2239 return SMC_CLC_DECL_ERR_REGBUF;
2245 static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2246 struct smc_clc_msg_proposal *pclc,
2247 struct smc_init_info *ini)
2249 struct smc_clc_v2_extension *smc_v2_ext;
2253 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2256 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2257 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2260 /* prepare RDMA check */
2261 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2262 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2263 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2264 ini->check_smcrv2 = true;
2265 ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2266 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2267 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2268 rc = smc_find_rdma_device(new_smc, ini);
2270 smc_find_ism_store_rc(rc, ini);
2273 if (!ini->smcrv2.uses_gateway)
2274 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2276 smcr_version = ini->smcr_version;
2277 ini->smcr_version = SMC_V2;
2278 rc = smc_listen_rdma_init(new_smc, ini);
2280 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2282 smc_conn_abort(new_smc, ini->first_contact_local);
2286 ini->smcr_version = smcr_version;
2287 smc_find_ism_store_rc(rc, ini);
2290 ini->smcr_version &= ~SMC_V2;
2291 ini->smcrv2.ib_dev_v2 = NULL;
2292 ini->check_smcrv2 = false;
2295 static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2296 struct smc_clc_msg_proposal *pclc,
2297 struct smc_init_info *ini)
2301 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2302 return SMC_CLC_DECL_NOSMCDEV;
2304 /* prepare RDMA check */
2305 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2306 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2307 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2308 rc = smc_find_rdma_device(new_smc, ini);
2310 /* no RDMA device found */
2311 return SMC_CLC_DECL_NOSMCDEV;
2313 rc = smc_listen_rdma_init(new_smc, ini);
2316 return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2319 /* determine the local device matching to proposal */
2320 static int smc_listen_find_device(struct smc_sock *new_smc,
2321 struct smc_clc_msg_proposal *pclc,
2322 struct smc_init_info *ini)
2326 /* check for ISM device matching V2 proposed device */
2327 smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2328 if (ini->ism_dev[0])
2331 /* check for matching IP prefix and subnet length (V1) */
2332 prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2334 smc_find_ism_store_rc(prfx_rc, ini);
2336 /* get vlan id from IP device */
2337 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2338 return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2340 /* check for ISM device matching V1 proposed device */
2342 smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2343 if (ini->ism_dev[0])
2346 if (!smcr_indicated(pclc->hdr.typev1) &&
2347 !smcr_indicated(pclc->hdr.typev2))
2348 /* skip RDMA and decline */
2349 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2351 /* check if RDMA V2 is available */
2352 smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2353 if (ini->smcrv2.ib_dev_v2)
2356 /* check if RDMA V1 is available */
2360 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2361 smc_find_ism_store_rc(rc, ini);
2362 return (!rc) ? 0 : ini->rc;
2364 return SMC_CLC_DECL_NOSMCDEV;
2367 /* listen worker: finish RDMA setup */
2368 static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2369 struct smc_clc_msg_accept_confirm *cclc,
2371 struct smc_init_info *ini)
2373 struct smc_link *link = new_smc->conn.lnk;
2374 int reason_code = 0;
2377 smc_link_save_peer_info(link, cclc, ini);
2379 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2380 return SMC_CLC_DECL_ERR_RTOK;
2383 if (smc_ib_ready_link(link))
2384 return SMC_CLC_DECL_ERR_RDYLNK;
2385 /* QP confirmation over RoCE fabric */
2386 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2387 reason_code = smcr_serv_conf_first_link(new_smc);
2388 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2393 /* setup for connection of server */
2394 static void smc_listen_work(struct work_struct *work)
2396 struct smc_sock *new_smc = container_of(work, struct smc_sock,
2398 struct socket *newclcsock = new_smc->clcsock;
2399 struct smc_clc_msg_accept_confirm *cclc;
2400 struct smc_clc_msg_proposal_area *buf;
2401 struct smc_clc_msg_proposal *pclc;
2402 struct smc_init_info *ini = NULL;
2403 u8 proposal_version = SMC_V1;
2407 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2408 return smc_listen_out_err(new_smc);
2410 if (new_smc->use_fallback) {
2411 smc_listen_out_connected(new_smc);
2415 /* check if peer is smc capable */
2416 if (!tcp_sk(newclcsock->sk)->syn_smc) {
2417 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2419 smc_listen_out_err(new_smc);
2421 smc_listen_out_connected(new_smc);
2425 /* do inband token exchange -
2426 * wait for and receive SMC Proposal CLC message
2428 buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2430 rc = SMC_CLC_DECL_MEM;
2433 pclc = (struct smc_clc_msg_proposal *)buf;
2434 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2435 SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2439 if (pclc->hdr.version > SMC_V1)
2440 proposal_version = SMC_V2;
2442 /* IPSec connections opt out of SMC optimizations */
2443 if (using_ipsec(new_smc)) {
2444 rc = SMC_CLC_DECL_IPSEC;
2448 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2450 rc = SMC_CLC_DECL_MEM;
2454 /* initial version checking */
2455 rc = smc_listen_v2_check(new_smc, pclc, ini);
2459 rc = smc_clc_srv_v2x_features_validate(pclc, ini);
2463 mutex_lock(&smc_server_lgr_pending);
2464 smc_close_init(new_smc);
2465 smc_rx_init(new_smc);
2466 smc_tx_init(new_smc);
2468 /* determine ISM or RoCE device used for connection */
2469 rc = smc_listen_find_device(new_smc, pclc, ini);
2473 /* send SMC Accept CLC message */
2474 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2475 rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2476 accept_version, ini->negotiated_eid, ini);
2480 /* SMC-D does not need this lock any more */
2482 mutex_unlock(&smc_server_lgr_pending);
2484 /* receive SMC Confirm CLC message */
2485 memset(buf, 0, sizeof(*buf));
2486 cclc = (struct smc_clc_msg_accept_confirm *)buf;
2487 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2488 SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2495 rc = smc_clc_v2x_features_confirm_check(cclc, ini);
2502 /* fce smc release version is needed in smc_listen_rdma_finish,
2503 * so save fce info here.
2505 smc_conn_save_peer_info_fce(new_smc, cclc);
2508 if (!ini->is_smcd) {
2509 rc = smc_listen_rdma_finish(new_smc, cclc,
2510 ini->first_contact_local, ini);
2513 mutex_unlock(&smc_server_lgr_pending);
2515 smc_conn_save_peer_info(new_smc, cclc);
2516 smc_listen_out_connected(new_smc);
2517 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2521 mutex_unlock(&smc_server_lgr_pending);
2523 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2530 static void smc_tcp_listen_work(struct work_struct *work)
2532 struct smc_sock *lsmc = container_of(work, struct smc_sock,
2534 struct sock *lsk = &lsmc->sk;
2535 struct smc_sock *new_smc;
2539 while (lsk->sk_state == SMC_LISTEN) {
2540 rc = smc_clcsock_accept(lsmc, &new_smc);
2541 if (rc) /* clcsock accept queue empty or error */
2546 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2547 atomic_inc(&lsmc->queued_smc_hs);
2549 new_smc->listen_smc = lsmc;
2550 new_smc->use_fallback = lsmc->use_fallback;
2551 new_smc->fallback_rsn = lsmc->fallback_rsn;
2552 sock_hold(lsk); /* sock_put in smc_listen_work */
2553 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2554 smc_copy_sock_settings_to_smc(new_smc);
2555 sock_hold(&new_smc->sk); /* sock_put in passive closing */
2556 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2557 sock_put(&new_smc->sk);
2562 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2565 static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2567 struct smc_sock *lsmc;
2569 read_lock_bh(&listen_clcsock->sk_callback_lock);
2570 lsmc = smc_clcsock_user_data(listen_clcsock);
2573 lsmc->clcsk_data_ready(listen_clcsock);
2574 if (lsmc->sk.sk_state == SMC_LISTEN) {
2575 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2576 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2577 sock_put(&lsmc->sk);
2580 read_unlock_bh(&listen_clcsock->sk_callback_lock);
2583 static int smc_listen(struct socket *sock, int backlog)
2585 struct sock *sk = sock->sk;
2586 struct smc_sock *smc;
2593 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2594 smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2598 if (sk->sk_state == SMC_LISTEN) {
2599 sk->sk_max_ack_backlog = backlog;
2602 /* some socket options are handled in core, so we could not apply
2603 * them to the clc socket -- copy smc socket options to clc socket
2605 smc_copy_sock_settings_to_clc(smc);
2606 if (!smc->use_fallback)
2607 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2609 /* save original sk_data_ready function and establish
2610 * smc-specific sk_data_ready function
2612 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2613 smc->clcsock->sk->sk_user_data =
2614 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
2615 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2616 smc_clcsock_data_ready, &smc->clcsk_data_ready);
2617 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2619 /* save original ops */
2620 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2622 smc->af_ops = *smc->ori_af_ops;
2623 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2625 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2627 if (smc->limit_smc_hs)
2628 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2630 rc = kernel_listen(smc->clcsock, backlog);
2632 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2633 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2634 &smc->clcsk_data_ready);
2635 smc->clcsock->sk->sk_user_data = NULL;
2636 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2639 sk->sk_max_ack_backlog = backlog;
2640 sk->sk_ack_backlog = 0;
2641 sk->sk_state = SMC_LISTEN;
2648 static int smc_accept(struct socket *sock, struct socket *new_sock,
2649 int flags, bool kern)
2651 struct sock *sk = sock->sk, *nsk;
2652 DECLARE_WAITQUEUE(wait, current);
2653 struct smc_sock *lsmc;
2658 sock_hold(sk); /* sock_put below */
2661 if (lsmc->sk.sk_state != SMC_LISTEN) {
2667 /* Wait for an incoming connection */
2668 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
2669 add_wait_queue_exclusive(sk_sleep(sk), &wait);
2670 while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2671 set_current_state(TASK_INTERRUPTIBLE);
2677 timeo = schedule_timeout(timeo);
2678 /* wakeup by sk_data_ready in smc_listen_work() */
2679 sched_annotate_sleep();
2681 if (signal_pending(current)) {
2682 rc = sock_intr_errno(timeo);
2686 set_current_state(TASK_RUNNING);
2687 remove_wait_queue(sk_sleep(sk), &wait);
2690 rc = sock_error(nsk);
2695 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
2696 /* wait till data arrives on the socket */
2697 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
2699 if (smc_sk(nsk)->use_fallback) {
2700 struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2703 if (skb_queue_empty(&clcsk->sk_receive_queue))
2704 sk_wait_data(clcsk, &timeo, NULL);
2705 release_sock(clcsk);
2706 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2708 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
2714 sock_put(sk); /* sock_hold above */
2718 static int smc_getname(struct socket *sock, struct sockaddr *addr,
2721 struct smc_sock *smc;
2723 if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2724 (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2727 smc = smc_sk(sock->sk);
2729 return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2732 static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2734 struct sock *sk = sock->sk;
2735 struct smc_sock *smc;
2741 /* SMC does not support connect with fastopen */
2742 if (msg->msg_flags & MSG_FASTOPEN) {
2743 /* not connected yet, fallback */
2744 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2745 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2752 } else if ((sk->sk_state != SMC_ACTIVE) &&
2753 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2754 (sk->sk_state != SMC_INIT)) {
2759 if (smc->use_fallback) {
2760 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2762 rc = smc_tx_sendmsg(smc, msg, len);
2763 SMC_STAT_TX_PAYLOAD(smc, len, rc);
2770 static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2773 struct sock *sk = sock->sk;
2774 struct smc_sock *smc;
2779 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2780 /* socket was connected before, no more data to read */
2784 if ((sk->sk_state == SMC_INIT) ||
2785 (sk->sk_state == SMC_LISTEN) ||
2786 (sk->sk_state == SMC_CLOSED))
2789 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2794 if (smc->use_fallback) {
2795 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2797 msg->msg_namelen = 0;
2798 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2799 SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2807 static __poll_t smc_accept_poll(struct sock *parent)
2809 struct smc_sock *isk = smc_sk(parent);
2812 spin_lock(&isk->accept_q_lock);
2813 if (!list_empty(&isk->accept_q))
2814 mask = EPOLLIN | EPOLLRDNORM;
2815 spin_unlock(&isk->accept_q_lock);
2820 static __poll_t smc_poll(struct file *file, struct socket *sock,
2823 struct sock *sk = sock->sk;
2824 struct smc_sock *smc;
2830 smc = smc_sk(sock->sk);
2831 if (smc->use_fallback) {
2832 /* delegate to CLC child sock */
2833 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2834 sk->sk_err = smc->clcsock->sk->sk_err;
2836 if (sk->sk_state != SMC_CLOSED)
2837 sock_poll_wait(file, sock, wait);
2840 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2841 (sk->sk_state == SMC_CLOSED))
2843 if (sk->sk_state == SMC_LISTEN) {
2844 /* woken up by sk_data_ready in smc_listen_work() */
2845 mask |= smc_accept_poll(sk);
2846 } else if (smc->use_fallback) { /* as result of connect_work()*/
2847 mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2849 sk->sk_err = smc->clcsock->sk->sk_err;
2851 if ((sk->sk_state != SMC_INIT &&
2852 atomic_read(&smc->conn.sndbuf_space)) ||
2853 sk->sk_shutdown & SEND_SHUTDOWN) {
2854 mask |= EPOLLOUT | EPOLLWRNORM;
2856 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2857 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2859 if (atomic_read(&smc->conn.bytes_to_rcv))
2860 mask |= EPOLLIN | EPOLLRDNORM;
2861 if (sk->sk_shutdown & RCV_SHUTDOWN)
2862 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2863 if (sk->sk_state == SMC_APPCLOSEWAIT1)
2865 if (smc->conn.urg_state == SMC_URG_VALID)
2873 static int smc_shutdown(struct socket *sock, int how)
2875 struct sock *sk = sock->sk;
2876 bool do_shutdown = true;
2877 struct smc_sock *smc;
2884 if ((how < SHUT_RD) || (how > SHUT_RDWR))
2889 if (sock->state == SS_CONNECTING) {
2890 if (sk->sk_state == SMC_ACTIVE)
2891 sock->state = SS_CONNECTED;
2892 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2893 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2894 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2895 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2896 sk->sk_state == SMC_APPFINCLOSEWAIT)
2897 sock->state = SS_DISCONNECTING;
2901 if ((sk->sk_state != SMC_ACTIVE) &&
2902 (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2903 (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2904 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2905 (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2906 (sk->sk_state != SMC_APPFINCLOSEWAIT))
2908 if (smc->use_fallback) {
2909 rc = kernel_sock_shutdown(smc->clcsock, how);
2910 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2911 if (sk->sk_shutdown == SHUTDOWN_MASK) {
2912 sk->sk_state = SMC_CLOSED;
2913 sk->sk_socket->state = SS_UNCONNECTED;
2919 case SHUT_RDWR: /* shutdown in both directions */
2920 old_state = sk->sk_state;
2921 rc = smc_close_active(smc);
2922 if (old_state == SMC_ACTIVE &&
2923 sk->sk_state == SMC_PEERCLOSEWAIT1)
2924 do_shutdown = false;
2927 rc = smc_close_shutdown_write(smc);
2931 /* nothing more to do because peer is not involved */
2934 if (do_shutdown && smc->clcsock)
2935 rc1 = kernel_sock_shutdown(smc->clcsock, how);
2936 /* map sock_shutdown_cmd constants to sk_shutdown value range */
2937 sk->sk_shutdown |= how + 1;
2939 if (sk->sk_state == SMC_CLOSED)
2940 sock->state = SS_UNCONNECTED;
2942 sock->state = SS_DISCONNECTING;
2945 return rc ? rc : rc1;
2948 static int __smc_getsockopt(struct socket *sock, int level, int optname,
2949 char __user *optval, int __user *optlen)
2951 struct smc_sock *smc;
2954 smc = smc_sk(sock->sk);
2956 if (get_user(len, optlen))
2959 len = min_t(int, len, sizeof(int));
2966 val = smc->limit_smc_hs;
2972 if (put_user(len, optlen))
2974 if (copy_to_user(optval, &val, len))
2980 static int __smc_setsockopt(struct socket *sock, int level, int optname,
2981 sockptr_t optval, unsigned int optlen)
2983 struct sock *sk = sock->sk;
2984 struct smc_sock *smc;
2992 if (optlen < sizeof(int)) {
2996 if (copy_from_sockptr(&val, optval, sizeof(int))) {
3001 smc->limit_smc_hs = !!val;
3013 static int smc_setsockopt(struct socket *sock, int level, int optname,
3014 sockptr_t optval, unsigned int optlen)
3016 struct sock *sk = sock->sk;
3017 struct smc_sock *smc;
3020 if (level == SOL_TCP && optname == TCP_ULP)
3022 else if (level == SOL_SMC)
3023 return __smc_setsockopt(sock, level, optname, optval, optlen);
3027 /* generic setsockopts reaching us here always apply to the
3030 mutex_lock(&smc->clcsock_release_lock);
3031 if (!smc->clcsock) {
3032 mutex_unlock(&smc->clcsock_release_lock);
3035 if (unlikely(!smc->clcsock->ops->setsockopt))
3038 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
3040 if (smc->clcsock->sk->sk_err) {
3041 sk->sk_err = smc->clcsock->sk->sk_err;
3042 sk_error_report(sk);
3044 mutex_unlock(&smc->clcsock_release_lock);
3046 if (optlen < sizeof(int))
3048 if (copy_from_sockptr(&val, optval, sizeof(int)))
3052 if (rc || smc->use_fallback)
3056 case TCP_FASTOPEN_CONNECT:
3057 case TCP_FASTOPEN_KEY:
3058 case TCP_FASTOPEN_NO_COOKIE:
3059 /* option not supported by SMC */
3060 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3061 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3067 if (sk->sk_state != SMC_INIT &&
3068 sk->sk_state != SMC_LISTEN &&
3069 sk->sk_state != SMC_CLOSED) {
3071 SMC_STAT_INC(smc, ndly_cnt);
3072 smc_tx_pending(&smc->conn);
3073 cancel_delayed_work(&smc->conn.tx_work);
3078 if (sk->sk_state != SMC_INIT &&
3079 sk->sk_state != SMC_LISTEN &&
3080 sk->sk_state != SMC_CLOSED) {
3082 SMC_STAT_INC(smc, cork_cnt);
3083 smc_tx_pending(&smc->conn);
3084 cancel_delayed_work(&smc->conn.tx_work);
3088 case TCP_DEFER_ACCEPT:
3089 smc->sockopt_defer_accept = val;
3100 static int smc_getsockopt(struct socket *sock, int level, int optname,
3101 char __user *optval, int __user *optlen)
3103 struct smc_sock *smc;
3106 if (level == SOL_SMC)
3107 return __smc_getsockopt(sock, level, optname, optval, optlen);
3109 smc = smc_sk(sock->sk);
3110 mutex_lock(&smc->clcsock_release_lock);
3111 if (!smc->clcsock) {
3112 mutex_unlock(&smc->clcsock_release_lock);
3115 /* socket options apply to the CLC socket */
3116 if (unlikely(!smc->clcsock->ops->getsockopt)) {
3117 mutex_unlock(&smc->clcsock_release_lock);
3120 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3122 mutex_unlock(&smc->clcsock_release_lock);
3126 static int smc_ioctl(struct socket *sock, unsigned int cmd,
3129 union smc_host_cursor cons, urg;
3130 struct smc_connection *conn;
3131 struct smc_sock *smc;
3134 smc = smc_sk(sock->sk);
3136 lock_sock(&smc->sk);
3137 if (smc->use_fallback) {
3138 if (!smc->clcsock) {
3139 release_sock(&smc->sk);
3142 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3143 release_sock(&smc->sk);
3147 case SIOCINQ: /* same as FIONREAD */
3148 if (smc->sk.sk_state == SMC_LISTEN) {
3149 release_sock(&smc->sk);
3152 if (smc->sk.sk_state == SMC_INIT ||
3153 smc->sk.sk_state == SMC_CLOSED)
3156 answ = atomic_read(&smc->conn.bytes_to_rcv);
3159 /* output queue size (not send + not acked) */
3160 if (smc->sk.sk_state == SMC_LISTEN) {
3161 release_sock(&smc->sk);
3164 if (smc->sk.sk_state == SMC_INIT ||
3165 smc->sk.sk_state == SMC_CLOSED)
3168 answ = smc->conn.sndbuf_desc->len -
3169 atomic_read(&smc->conn.sndbuf_space);
3172 /* output queue size (not send only) */
3173 if (smc->sk.sk_state == SMC_LISTEN) {
3174 release_sock(&smc->sk);
3177 if (smc->sk.sk_state == SMC_INIT ||
3178 smc->sk.sk_state == SMC_CLOSED)
3181 answ = smc_tx_prepared_sends(&smc->conn);
3184 if (smc->sk.sk_state == SMC_LISTEN) {
3185 release_sock(&smc->sk);
3188 if (smc->sk.sk_state == SMC_INIT ||
3189 smc->sk.sk_state == SMC_CLOSED) {
3192 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3193 smc_curs_copy(&urg, &conn->urg_curs, conn);
3194 answ = smc_curs_diff(conn->rmb_desc->len,
3199 release_sock(&smc->sk);
3200 return -ENOIOCTLCMD;
3202 release_sock(&smc->sk);
3204 return put_user(answ, (int __user *)arg);
3207 /* Map the affected portions of the rmbe into an spd, note the number of bytes
3208 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3209 * updates till whenever a respective page has been fully processed.
3210 * Note that subsequent recv() calls have to wait till all splice() processing
3213 static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3214 struct pipe_inode_info *pipe, size_t len,
3217 struct sock *sk = sock->sk;
3218 struct smc_sock *smc;
3223 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3224 /* socket was connected before, no more data to read */
3228 if (sk->sk_state == SMC_INIT ||
3229 sk->sk_state == SMC_LISTEN ||
3230 sk->sk_state == SMC_CLOSED)
3233 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3238 if (smc->use_fallback) {
3239 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3246 if (flags & SPLICE_F_NONBLOCK)
3247 flags = MSG_DONTWAIT;
3250 SMC_STAT_INC(smc, splice_cnt);
3251 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3259 /* must look like tcp */
3260 static const struct proto_ops smc_sock_ops = {
3262 .owner = THIS_MODULE,
3263 .release = smc_release,
3265 .connect = smc_connect,
3266 .socketpair = sock_no_socketpair,
3267 .accept = smc_accept,
3268 .getname = smc_getname,
3271 .listen = smc_listen,
3272 .shutdown = smc_shutdown,
3273 .setsockopt = smc_setsockopt,
3274 .getsockopt = smc_getsockopt,
3275 .sendmsg = smc_sendmsg,
3276 .recvmsg = smc_recvmsg,
3277 .mmap = sock_no_mmap,
3278 .splice_read = smc_splice_read,
3281 static int __smc_create(struct net *net, struct socket *sock, int protocol,
3282 int kern, struct socket *clcsock)
3284 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3285 struct smc_sock *smc;
3289 rc = -ESOCKTNOSUPPORT;
3290 if (sock->type != SOCK_STREAM)
3293 rc = -EPROTONOSUPPORT;
3294 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3298 sock->ops = &smc_sock_ops;
3299 sock->state = SS_UNCONNECTED;
3300 sk = smc_sock_alloc(net, sock, protocol);
3304 /* create internal TCP socket for CLC handshake and fallback */
3306 smc->use_fallback = false; /* assume rdma capability first */
3307 smc->fallback_rsn = 0;
3309 /* default behavior from limit_smc_hs in every net namespace */
3310 smc->limit_smc_hs = net->smc.limit_smc_hs;
3314 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3317 sk_common_release(sk);
3321 /* smc_clcsock_release() does not wait smc->clcsock->sk's
3322 * destruction; its sk_state might not be TCP_CLOSE after
3323 * smc->sk is close()d, and TCP timers can be fired later,
3324 * which need net ref.
3326 sk = smc->clcsock->sk;
3327 __netns_tracker_free(net, &sk->ns_tracker, false);
3328 sk->sk_net_refcnt = 1;
3329 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
3330 sock_inuse_add(net, 1);
3332 smc->clcsock = clcsock;
3339 static int smc_create(struct net *net, struct socket *sock, int protocol,
3342 return __smc_create(net, sock, protocol, kern, NULL);
3345 static const struct net_proto_family smc_sock_family_ops = {
3347 .owner = THIS_MODULE,
3348 .create = smc_create,
3351 static int smc_ulp_init(struct sock *sk)
3353 struct socket *tcp = sk->sk_socket;
3354 struct net *net = sock_net(sk);
3355 struct socket *smcsock;
3358 /* only TCP can be replaced */
3359 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP ||
3360 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6))
3361 return -ESOCKTNOSUPPORT;
3362 /* don't handle wq now */
3363 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list)
3366 if (sk->sk_family == AF_INET)
3367 protocol = SMCPROTO_SMC;
3369 protocol = SMCPROTO_SMC6;
3371 smcsock = sock_alloc();
3375 smcsock->type = SOCK_STREAM;
3376 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */
3377 ret = __smc_create(net, smcsock, protocol, 1, tcp);
3379 sock_release(smcsock); /* module_put() which ops won't be NULL */
3383 /* replace tcp socket to smc */
3384 smcsock->file = tcp->file;
3385 smcsock->file->private_data = smcsock;
3386 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */
3387 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */
3393 static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk,
3394 const gfp_t priority)
3396 struct inet_connection_sock *icsk = inet_csk(newsk);
3398 /* don't inherit ulp ops to child when listen */
3399 icsk->icsk_ulp_ops = NULL;
3402 static struct tcp_ulp_ops smc_ulp_ops __read_mostly = {
3404 .owner = THIS_MODULE,
3405 .init = smc_ulp_init,
3406 .clone = smc_ulp_clone,
3409 unsigned int smc_net_id;
3411 static __net_init int smc_net_init(struct net *net)
3415 rc = smc_sysctl_net_init(net);
3418 return smc_pnet_net_init(net);
3421 static void __net_exit smc_net_exit(struct net *net)
3423 smc_sysctl_net_exit(net);
3424 smc_pnet_net_exit(net);
3427 static __net_init int smc_net_stat_init(struct net *net)
3429 return smc_stats_init(net);
3432 static void __net_exit smc_net_stat_exit(struct net *net)
3434 smc_stats_exit(net);
3437 static struct pernet_operations smc_net_ops = {
3438 .init = smc_net_init,
3439 .exit = smc_net_exit,
3441 .size = sizeof(struct smc_net),
3444 static struct pernet_operations smc_net_stat_ops = {
3445 .init = smc_net_stat_init,
3446 .exit = smc_net_stat_exit,
3449 static int __init smc_init(void)
3453 rc = register_pernet_subsys(&smc_net_ops);
3457 rc = register_pernet_subsys(&smc_net_stat_ops);
3459 goto out_pernet_subsys;
3461 rc = smc_ism_init();
3463 goto out_pernet_subsys_stat;
3470 rc = smc_pnet_init();
3476 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0);
3480 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0);
3482 goto out_alloc_tcp_ls_wq;
3484 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0);
3486 goto out_alloc_hs_wq;
3488 rc = smc_core_init();
3490 pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3494 rc = smc_llc_init();
3496 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3500 rc = smc_cdc_init();
3502 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3506 rc = proto_register(&smc_proto, 1);
3508 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3512 rc = proto_register(&smc_proto6, 1);
3514 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3518 rc = sock_register(&smc_sock_family_ops);
3520 pr_err("%s: sock_register fails with %d\n", __func__, rc);
3523 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3524 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3526 rc = smc_ib_register_client();
3528 pr_err("%s: ib_register fails with %d\n", __func__, rc);
3532 rc = tcp_register_ulp(&smc_ulp_ops);
3534 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc);
3538 static_branch_enable(&tcp_have_smc);
3542 smc_ib_unregister_client();
3544 sock_unregister(PF_SMC);
3546 proto_unregister(&smc_proto6);
3548 proto_unregister(&smc_proto);
3552 destroy_workqueue(smc_close_wq);
3554 destroy_workqueue(smc_hs_wq);
3555 out_alloc_tcp_ls_wq:
3556 destroy_workqueue(smc_tcp_ls_wq);
3564 out_pernet_subsys_stat:
3565 unregister_pernet_subsys(&smc_net_stat_ops);
3567 unregister_pernet_subsys(&smc_net_ops);
3572 static void __exit smc_exit(void)
3574 static_branch_disable(&tcp_have_smc);
3575 tcp_unregister_ulp(&smc_ulp_ops);
3576 sock_unregister(PF_SMC);
3578 smc_ib_unregister_client();
3580 destroy_workqueue(smc_close_wq);
3581 destroy_workqueue(smc_tcp_ls_wq);
3582 destroy_workqueue(smc_hs_wq);
3583 proto_unregister(&smc_proto6);
3584 proto_unregister(&smc_proto);
3588 unregister_pernet_subsys(&smc_net_stat_ops);
3589 unregister_pernet_subsys(&smc_net_ops);
3593 module_init(smc_init);
3594 module_exit(smc_exit);
3596 MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3597 MODULE_DESCRIPTION("smc socket address family");
3598 MODULE_LICENSE("GPL");
3599 MODULE_ALIAS_NETPROTO(PF_SMC);
3600 MODULE_ALIAS_TCP_ULP("smc");
3601 MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);