Loading...
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type
6 * applies to SOCK_STREAM sockets only
7 * offers an alternative communication option for TCP-protocol sockets
8 * applicable with RoCE-cards only
9 *
10 * Initial restrictions:
11 * - support for alternate links postponed
12 *
13 * Copyright IBM Corp. 2016, 2018
14 *
15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
16 * based on prototype from Frank Blaschka
17 */
18
19#define KMSG_COMPONENT "smc"
20#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22#include <linux/module.h>
23#include <linux/socket.h>
24#include <linux/workqueue.h>
25#include <linux/in.h>
26#include <linux/sched/signal.h>
27#include <linux/if_vlan.h>
28#include <linux/rcupdate_wait.h>
29#include <linux/ctype.h>
30#include <linux/splice.h>
31
32#include <net/sock.h>
33#include <net/tcp.h>
34#include <net/smc.h>
35#include <asm/ioctls.h>
36
37#include <net/net_namespace.h>
38#include <net/netns/generic.h>
39#include "smc_netns.h"
40
41#include "smc.h"
42#include "smc_clc.h"
43#include "smc_llc.h"
44#include "smc_cdc.h"
45#include "smc_core.h"
46#include "smc_ib.h"
47#include "smc_ism.h"
48#include "smc_pnet.h"
49#include "smc_netlink.h"
50#include "smc_tx.h"
51#include "smc_rx.h"
52#include "smc_close.h"
53#include "smc_stats.h"
54#include "smc_tracepoint.h"
55#include "smc_sysctl.h"
56
57static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group
58 * creation on server
59 */
60static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group
61 * creation on client
62 */
63
64static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */
65struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
66struct workqueue_struct *smc_close_wq; /* wq for close work */
67
68static void smc_tcp_listen_work(struct work_struct *);
69static void smc_connect_work(struct work_struct *);
70
71int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb)
72{
73 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
74 void *hdr;
75
76 if (cb_ctx->pos[0])
77 goto out;
78
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);
82 if (!hdr)
83 return -ENOMEM;
84
85 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED,
86 sock_net(skb->sk)->smc.limit_smc_hs))
87 goto err;
88
89 genlmsg_end(skb, hdr);
90 cb_ctx->pos[0] = 1;
91out:
92 return skb->len;
93err:
94 genlmsg_cancel(skb, hdr);
95 return -EMSGSIZE;
96}
97
98int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
99{
100 sock_net(skb->sk)->smc.limit_smc_hs = true;
101 return 0;
102}
103
104int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
105{
106 sock_net(skb->sk)->smc.limit_smc_hs = false;
107 return 0;
108}
109
110static void smc_set_keepalive(struct sock *sk, int val)
111{
112 struct smc_sock *smc = smc_sk(sk);
113
114 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
115}
116
117static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk,
118 struct sk_buff *skb,
119 struct request_sock *req,
120 struct dst_entry *dst,
121 struct request_sock *req_unhash,
122 bool *own_req)
123{
124 struct smc_sock *smc;
125 struct sock *child;
126
127 smc = smc_clcsock_user_data(sk);
128
129 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) >
130 sk->sk_max_ack_backlog)
131 goto drop;
132
133 if (sk_acceptq_is_full(&smc->sk)) {
134 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
135 goto drop;
136 }
137
138 /* passthrough to original syn recv sock fct */
139 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash,
140 own_req);
141 /* child must not inherit smc or its ops */
142 if (child) {
143 rcu_assign_sk_user_data(child, NULL);
144
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;
148 }
149 return child;
150
151drop:
152 dst_release(dst);
153 tcp_listendrop(sk);
154 return NULL;
155}
156
157static bool smc_hs_congested(const struct sock *sk)
158{
159 const struct smc_sock *smc;
160
161 smc = smc_clcsock_user_data(sk);
162
163 if (!smc)
164 return true;
165
166 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq))
167 return true;
168
169 return false;
170}
171
172static struct smc_hashinfo smc_v4_hashinfo = {
173 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
174};
175
176static struct smc_hashinfo smc_v6_hashinfo = {
177 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
178};
179
180int smc_hash_sk(struct sock *sk)
181{
182 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
183 struct hlist_head *head;
184
185 head = &h->ht;
186
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);
191
192 return 0;
193}
194EXPORT_SYMBOL_GPL(smc_hash_sk);
195
196void smc_unhash_sk(struct sock *sk)
197{
198 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
199
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);
204}
205EXPORT_SYMBOL_GPL(smc_unhash_sk);
206
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
209 * BH context
210 */
211static void smc_release_cb(struct sock *sk)
212{
213 struct smc_sock *smc = smc_sk(sk);
214
215 if (smc->conn.tx_in_release_sock) {
216 smc_tx_pending(&smc->conn);
217 smc->conn.tx_in_release_sock = false;
218 }
219}
220
221struct proto smc_proto = {
222 .name = "SMC",
223 .owner = THIS_MODULE,
224 .keepalive = smc_set_keepalive,
225 .hash = smc_hash_sk,
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,
231};
232EXPORT_SYMBOL_GPL(smc_proto);
233
234struct proto smc_proto6 = {
235 .name = "SMC6",
236 .owner = THIS_MODULE,
237 .keepalive = smc_set_keepalive,
238 .hash = smc_hash_sk,
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,
244};
245EXPORT_SYMBOL_GPL(smc_proto6);
246
247static void smc_fback_restore_callbacks(struct smc_sock *smc)
248{
249 struct sock *clcsk = smc->clcsock->sk;
250
251 write_lock_bh(&clcsk->sk_callback_lock);
252 clcsk->sk_user_data = NULL;
253
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);
258
259 write_unlock_bh(&clcsk->sk_callback_lock);
260}
261
262static void smc_restore_fallback_changes(struct smc_sock *smc)
263{
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);
268 }
269}
270
271static int __smc_release(struct smc_sock *smc)
272{
273 struct sock *sk = &smc->sk;
274 int rc = 0;
275
276 if (!smc->use_fallback) {
277 rc = smc_close_active(smc);
278 smc_sock_set_flag(sk, SOCK_DEAD);
279 sk->sk_shutdown |= SHUTDOWN_MASK;
280 } else {
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,
288 SHUT_RDWR);
289 }
290 sk->sk_state = SMC_CLOSED;
291 sk->sk_state_change(sk);
292 }
293 smc_restore_fallback_changes(smc);
294 }
295
296 sk->sk_prot->unhash(sk);
297
298 if (sk->sk_state == SMC_CLOSED) {
299 if (smc->clcsock) {
300 release_sock(sk);
301 smc_clcsock_release(smc);
302 lock_sock(sk);
303 }
304 if (!smc->use_fallback)
305 smc_conn_free(&smc->conn);
306 }
307
308 return rc;
309}
310
311static int smc_release(struct socket *sock)
312{
313 struct sock *sk = sock->sk;
314 struct smc_sock *smc;
315 int old_state, rc = 0;
316
317 if (!sk)
318 goto out;
319
320 sock_hold(sk); /* sock_put below */
321 smc = smc_sk(sk);
322
323 old_state = sk->sk_state;
324
325 /* cleanup for a dangling non-blocking connect */
326 if (smc->connect_nonblock && old_state == SMC_INIT)
327 tcp_abort(smc->clcsock->sk, ECONNABORTED);
328
329 if (cancel_work_sync(&smc->connect_work))
330 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */
331
332 if (sk->sk_state == SMC_LISTEN)
333 /* smc_close_non_accepted() is called and acquires
334 * sock lock for child sockets again
335 */
336 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
337 else
338 lock_sock(sk);
339
340 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE &&
341 !smc->use_fallback)
342 smc_close_active_abort(smc);
343
344 rc = __smc_release(smc);
345
346 /* detach socket */
347 sock_orphan(sk);
348 sock->sk = NULL;
349 release_sock(sk);
350
351 sock_put(sk); /* sock_hold above */
352 sock_put(sk); /* final sock_put */
353out:
354 return rc;
355}
356
357static void smc_destruct(struct sock *sk)
358{
359 if (sk->sk_state != SMC_CLOSED)
360 return;
361 if (!sock_flag(sk, SOCK_DEAD))
362 return;
363}
364
365static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
366 int protocol)
367{
368 struct smc_sock *smc;
369 struct proto *prot;
370 struct sock *sk;
371
372 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
373 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
374 if (!sk)
375 return NULL;
376
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));
383 smc = smc_sk(sk);
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);
393
394 return sk;
395}
396
397static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
398 int addr_len)
399{
400 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
401 struct sock *sk = sock->sk;
402 struct smc_sock *smc;
403 int rc;
404
405 smc = smc_sk(sk);
406
407 /* replicate tests from inet_bind(), to be safe wrt. future changes */
408 rc = -EINVAL;
409 if (addr_len < sizeof(struct sockaddr_in))
410 goto out;
411
412 rc = -EAFNOSUPPORT;
413 if (addr->sin_family != AF_INET &&
414 addr->sin_family != AF_INET6 &&
415 addr->sin_family != AF_UNSPEC)
416 goto out;
417 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
418 if (addr->sin_family == AF_UNSPEC &&
419 addr->sin_addr.s_addr != htonl(INADDR_ANY))
420 goto out;
421
422 lock_sock(sk);
423
424 /* Check if socket is already active */
425 rc = -EINVAL;
426 if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
427 goto out_rel;
428
429 smc->clcsock->sk->sk_reuse = sk->sk_reuse;
430 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport;
431 rc = kernel_bind(smc->clcsock, uaddr, addr_len);
432
433out_rel:
434 release_sock(sk);
435out:
436 return rc;
437}
438
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)
441 */
442
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))
458
459/* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */
460static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk,
461 unsigned long mask)
462{
463 struct net *nnet = sock_net(nsk);
464
465 nsk->sk_userlocks = osk->sk_userlocks;
466 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK) {
467 nsk->sk_sndbuf = osk->sk_sndbuf;
468 } else {
469 if (mask == SK_FLAGS_SMC_TO_CLC)
470 WRITE_ONCE(nsk->sk_sndbuf,
471 READ_ONCE(nnet->ipv4.sysctl_tcp_wmem[1]));
472 else
473 WRITE_ONCE(nsk->sk_sndbuf,
474 2 * READ_ONCE(nnet->smc.sysctl_wmem));
475 }
476 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK) {
477 nsk->sk_rcvbuf = osk->sk_rcvbuf;
478 } else {
479 if (mask == SK_FLAGS_SMC_TO_CLC)
480 WRITE_ONCE(nsk->sk_rcvbuf,
481 READ_ONCE(nnet->ipv4.sysctl_tcp_rmem[1]));
482 else
483 WRITE_ONCE(nsk->sk_rcvbuf,
484 2 * READ_ONCE(nnet->smc.sysctl_rmem));
485 }
486}
487
488static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
489 unsigned long mask)
490{
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 = READ_ONCE(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;
500
501 nsk->sk_flags &= ~mask;
502 nsk->sk_flags |= osk->sk_flags & mask;
503
504 smc_adjust_sock_bufsizes(nsk, osk, mask);
505}
506
507static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
508{
509 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
510}
511
512#define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
513 (1UL << SOCK_KEEPOPEN) | \
514 (1UL << SOCK_LINGER) | \
515 (1UL << SOCK_DBG))
516/* copy only settings and flags relevant for smc from clc to smc socket */
517static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
518{
519 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
520}
521
522/* register the new vzalloced sndbuf on all links */
523static int smcr_lgr_reg_sndbufs(struct smc_link *link,
524 struct smc_buf_desc *snd_desc)
525{
526 struct smc_link_group *lgr = link->lgr;
527 int i, rc = 0;
528
529 if (!snd_desc->is_vm)
530 return -EINVAL;
531
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]))
536 continue;
537 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc);
538 if (rc)
539 break;
540 }
541 up_write(&lgr->llc_conf_mutex);
542 return rc;
543}
544
545/* register the new rmb on all links */
546static int smcr_lgr_reg_rmbs(struct smc_link *link,
547 struct smc_buf_desc *rmb_desc)
548{
549 struct smc_link_group *lgr = link->lgr;
550 bool do_slow = false;
551 int i, rc = 0;
552
553 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
554 if (rc)
555 return rc;
556
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]))
560 continue;
561 if (!rmb_desc->is_reg_mr[link->link_idx]) {
562 up_read(&lgr->llc_conf_mutex);
563 goto slow_path;
564 }
565 }
566 /* mr register already */
567 goto fast_path;
568slow_path:
569 do_slow = true;
570 /* protect against parallel smc_llc_cli_rkey_exchange() and
571 * parallel smcr_link_reg_buf()
572 */
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]))
576 continue;
577 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc);
578 if (rc)
579 goto out;
580 }
581fast_path:
582 /* exchange confirm_rkey msg with peer */
583 rc = smc_llc_do_confirm_rkey(link, rmb_desc);
584 if (rc) {
585 rc = -EFAULT;
586 goto out;
587 }
588 rmb_desc->is_conf_rkey = true;
589out:
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);
592 return rc;
593}
594
595static int smcr_clnt_conf_first_link(struct smc_sock *smc)
596{
597 struct smc_link *link = smc->conn.lnk;
598 struct smc_llc_qentry *qentry;
599 int rc;
600
601 /* Receive CONFIRM LINK request from server over RoCE fabric.
602 * Increasing the client's timeout by twice as much as the server's
603 * timeout by default can temporarily avoid decline messages of
604 * both sides crossing or colliding
605 */
606 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME,
607 SMC_LLC_CONFIRM_LINK);
608 if (!qentry) {
609 struct smc_clc_msg_decline dclc;
610
611 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
612 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
613 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
614 }
615 smc_llc_save_peer_uid(qentry);
616 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
617 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
618 if (rc)
619 return SMC_CLC_DECL_RMBE_EC;
620
621 rc = smc_ib_modify_qp_rts(link);
622 if (rc)
623 return SMC_CLC_DECL_ERR_RDYLNK;
624
625 smc_wr_remember_qp_attr(link);
626
627 /* reg the sndbuf if it was vzalloced */
628 if (smc->conn.sndbuf_desc->is_vm) {
629 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
630 return SMC_CLC_DECL_ERR_REGBUF;
631 }
632
633 /* reg the rmb */
634 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
635 return SMC_CLC_DECL_ERR_REGBUF;
636
637 /* confirm_rkey is implicit on 1st contact */
638 smc->conn.rmb_desc->is_conf_rkey = true;
639
640 /* send CONFIRM LINK response over RoCE fabric */
641 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
642 if (rc < 0)
643 return SMC_CLC_DECL_TIMEOUT_CL;
644
645 smc_llc_link_active(link);
646 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
647
648 if (link->lgr->max_links > 1) {
649 /* optional 2nd link, receive ADD LINK request from server */
650 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
651 SMC_LLC_ADD_LINK);
652 if (!qentry) {
653 struct smc_clc_msg_decline dclc;
654
655 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
656 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
657 if (rc == -EAGAIN)
658 rc = 0; /* no DECLINE received, go with one link */
659 return rc;
660 }
661 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
662 smc_llc_cli_add_link(link, qentry);
663 }
664 return 0;
665}
666
667static bool smc_isascii(char *hostname)
668{
669 int i;
670
671 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
672 if (!isascii(hostname[i]))
673 return false;
674 return true;
675}
676
677static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
678 struct smc_clc_msg_accept_confirm *clc)
679{
680 struct smc_clc_first_contact_ext *fce;
681 int clc_v2_len;
682
683 if (clc->hdr.version == SMC_V1 ||
684 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
685 return;
686
687 if (smc->conn.lgr->is_smcd) {
688 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid,
689 SMC_MAX_EID_LEN);
690 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1);
691 } else {
692 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid,
693 SMC_MAX_EID_LEN);
694 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1);
695 }
696 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + 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);
702}
703
704static void smcr_conn_save_peer_info(struct smc_sock *smc,
705 struct smc_clc_msg_accept_confirm *clc)
706{
707 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
708
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);
714}
715
716static void smcd_conn_save_peer_info(struct smc_sock *smc,
717 struct smc_clc_msg_accept_confirm *clc)
718{
719 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
720
721 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
722 smc->conn.peer_token = ntohll(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;
727}
728
729static void smc_conn_save_peer_info(struct smc_sock *smc,
730 struct smc_clc_msg_accept_confirm *clc)
731{
732 if (smc->conn.lgr->is_smcd)
733 smcd_conn_save_peer_info(smc, clc);
734 else
735 smcr_conn_save_peer_info(smc, clc);
736 smc_conn_save_peer_info_fce(smc, clc);
737}
738
739static void smc_link_save_peer_info(struct smc_link *link,
740 struct smc_clc_msg_accept_confirm *clc,
741 struct smc_init_info *ini)
742{
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;
748}
749
750static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
751 struct smc_stats_fback *fback_arr)
752{
753 int cnt;
754
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++;
758 break;
759 }
760 if (!fback_arr[cnt].fback_code) {
761 fback_arr[cnt].fback_code = smc->fallback_rsn;
762 fback_arr[cnt].count++;
763 break;
764 }
765 }
766}
767
768static void smc_stat_fallback(struct smc_sock *smc)
769{
770 struct net *net = sock_net(&smc->sk);
771
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++;
776 } else {
777 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
778 net->smc.fback_rsn->clnt_fback_cnt++;
779 }
780 mutex_unlock(&net->smc.mutex_fback_rsn);
781}
782
783/* must be called under rcu read lock */
784static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
785{
786 struct socket_wq *wq;
787 __poll_t flags;
788
789 wq = rcu_dereference(smc->sk.sk_wq);
790 if (!skwq_has_sleeper(wq))
791 return;
792
793 /* wake up smc sk->sk_wq */
794 if (!key) {
795 /* sk_state_change */
796 wake_up_interruptible_all(&wq->wait);
797 } else {
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);
805 }
806}
807
808static int smc_fback_mark_woken(wait_queue_entry_t *wait,
809 unsigned int mode, int sync, void *key)
810{
811 struct smc_mark_woken *mark =
812 container_of(wait, struct smc_mark_woken, wait_entry);
813
814 mark->woken = true;
815 mark->key = key;
816 return 0;
817}
818
819static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
820 void (*clcsock_callback)(struct sock *sk))
821{
822 struct smc_mark_woken mark = { .woken = false };
823 struct socket_wq *wq;
824
825 init_waitqueue_func_entry(&mark.wait_entry,
826 smc_fback_mark_woken);
827 rcu_read_lock();
828 wq = rcu_dereference(clcsk->sk_wq);
829 if (!wq)
830 goto out;
831 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
832 clcsock_callback(clcsk);
833 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
834
835 if (mark.woken)
836 smc_fback_wakeup_waitqueue(smc, mark.key);
837out:
838 rcu_read_unlock();
839}
840
841static void smc_fback_state_change(struct sock *clcsk)
842{
843 struct smc_sock *smc;
844
845 read_lock_bh(&clcsk->sk_callback_lock);
846 smc = smc_clcsock_user_data(clcsk);
847 if (smc)
848 smc_fback_forward_wakeup(smc, clcsk,
849 smc->clcsk_state_change);
850 read_unlock_bh(&clcsk->sk_callback_lock);
851}
852
853static void smc_fback_data_ready(struct sock *clcsk)
854{
855 struct smc_sock *smc;
856
857 read_lock_bh(&clcsk->sk_callback_lock);
858 smc = smc_clcsock_user_data(clcsk);
859 if (smc)
860 smc_fback_forward_wakeup(smc, clcsk,
861 smc->clcsk_data_ready);
862 read_unlock_bh(&clcsk->sk_callback_lock);
863}
864
865static void smc_fback_write_space(struct sock *clcsk)
866{
867 struct smc_sock *smc;
868
869 read_lock_bh(&clcsk->sk_callback_lock);
870 smc = smc_clcsock_user_data(clcsk);
871 if (smc)
872 smc_fback_forward_wakeup(smc, clcsk,
873 smc->clcsk_write_space);
874 read_unlock_bh(&clcsk->sk_callback_lock);
875}
876
877static void smc_fback_error_report(struct sock *clcsk)
878{
879 struct smc_sock *smc;
880
881 read_lock_bh(&clcsk->sk_callback_lock);
882 smc = smc_clcsock_user_data(clcsk);
883 if (smc)
884 smc_fback_forward_wakeup(smc, clcsk,
885 smc->clcsk_error_report);
886 read_unlock_bh(&clcsk->sk_callback_lock);
887}
888
889static void smc_fback_replace_callbacks(struct smc_sock *smc)
890{
891 struct sock *clcsk = smc->clcsock->sk;
892
893 write_lock_bh(&clcsk->sk_callback_lock);
894 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
895
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);
904
905 write_unlock_bh(&clcsk->sk_callback_lock);
906}
907
908static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
909{
910 int rc = 0;
911
912 mutex_lock(&smc->clcsock_release_lock);
913 if (!smc->clcsock) {
914 rc = -EBADF;
915 goto out;
916 }
917
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;
927 smc->sk.sk_socket->wq.fasync_list = NULL;
928
929 /* There might be some wait entries remaining
930 * in smc sk->sk_wq and they should be woken up
931 * as clcsock's wait queue is woken up.
932 */
933 smc_fback_replace_callbacks(smc);
934 }
935out:
936 mutex_unlock(&smc->clcsock_release_lock);
937 return rc;
938}
939
940/* fall back during connect */
941static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
942{
943 struct net *net = sock_net(&smc->sk);
944 int rc = 0;
945
946 rc = smc_switch_to_fallback(smc, reason_code);
947 if (rc) { /* fallback fails */
948 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
949 if (smc->sk.sk_state == SMC_INIT)
950 sock_put(&smc->sk); /* passive closing */
951 return rc;
952 }
953 smc_copy_sock_settings_to_clc(smc);
954 smc->connect_nonblock = 0;
955 if (smc->sk.sk_state == SMC_INIT)
956 smc->sk.sk_state = SMC_ACTIVE;
957 return 0;
958}
959
960/* decline and fall back during connect */
961static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
962 u8 version)
963{
964 struct net *net = sock_net(&smc->sk);
965 int rc;
966
967 if (reason_code < 0) { /* error, fallback is not possible */
968 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
969 if (smc->sk.sk_state == SMC_INIT)
970 sock_put(&smc->sk); /* passive closing */
971 return reason_code;
972 }
973 if (reason_code != SMC_CLC_DECL_PEERDECL) {
974 rc = smc_clc_send_decline(smc, reason_code, version);
975 if (rc < 0) {
976 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
977 if (smc->sk.sk_state == SMC_INIT)
978 sock_put(&smc->sk); /* passive closing */
979 return rc;
980 }
981 }
982 return smc_connect_fallback(smc, reason_code);
983}
984
985static void smc_conn_abort(struct smc_sock *smc, int local_first)
986{
987 struct smc_connection *conn = &smc->conn;
988 struct smc_link_group *lgr = conn->lgr;
989 bool lgr_valid = false;
990
991 if (smc_conn_lgr_valid(conn))
992 lgr_valid = true;
993
994 smc_conn_free(conn);
995 if (local_first && lgr_valid)
996 smc_lgr_cleanup_early(lgr);
997}
998
999/* check if there is a rdma device available for this connection. */
1000/* called for connect and listen */
1001static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
1002{
1003 /* PNET table look up: search active ib_device and port
1004 * within same PNETID that also contains the ethernet device
1005 * used for the internal TCP socket
1006 */
1007 smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
1008 if (!ini->check_smcrv2 && !ini->ib_dev)
1009 return SMC_CLC_DECL_NOSMCRDEV;
1010 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1011 return SMC_CLC_DECL_NOSMCRDEV;
1012 return 0;
1013}
1014
1015/* check if there is an ISM device available for this connection. */
1016/* called for connect and listen */
1017static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1018{
1019 /* Find ISM device with same PNETID as connecting interface */
1020 smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1021 if (!ini->ism_dev[0])
1022 return SMC_CLC_DECL_NOSMCDDEV;
1023 else
1024 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1025 return 0;
1026}
1027
1028/* is chid unique for the ism devices that are already determined? */
1029static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1030 int cnt)
1031{
1032 int i = (!ini->ism_dev[0]) ? 1 : 0;
1033
1034 for (; i < cnt; i++)
1035 if (ini->ism_chid[i] == chid)
1036 return false;
1037 return true;
1038}
1039
1040/* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1041 * PNETID matching net_device)
1042 */
1043static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1044 struct smc_init_info *ini)
1045{
1046 int rc = SMC_CLC_DECL_NOSMCDDEV;
1047 struct smcd_dev *smcd;
1048 int i = 1, entry = 1;
1049 bool is_virtual;
1050 u16 chid;
1051
1052 if (smcd_indicated(ini->smc_type_v1))
1053 rc = 0; /* already initialized for V1 */
1054 mutex_lock(&smcd_dev_list.mutex);
1055 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1056 if (smcd->going_away || smcd == ini->ism_dev[0])
1057 continue;
1058 chid = smc_ism_get_chid(smcd);
1059 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1060 continue;
1061 is_virtual = __smc_ism_is_virtual(chid);
1062 if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1063 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1064 if (is_virtual && entry == SMCD_CLC_MAX_V2_GID_ENTRIES)
1065 /* It's the last GID-CHID entry left in CLC
1066 * Proposal SMC-Dv2 extension, but a virtual
1067 * ISM device will take two entries. So give
1068 * up it and try the next potential ISM device.
1069 */
1070 continue;
1071 ini->ism_dev[i] = smcd;
1072 ini->ism_chid[i] = chid;
1073 ini->is_smcd = true;
1074 rc = 0;
1075 i++;
1076 entry = is_virtual ? entry + 2 : entry + 1;
1077 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES)
1078 break;
1079 }
1080 }
1081 mutex_unlock(&smcd_dev_list.mutex);
1082 ini->ism_offered_cnt = i - 1;
1083 if (!ini->ism_dev[0] && !ini->ism_dev[1])
1084 ini->smcd_version = 0;
1085
1086 return rc;
1087}
1088
1089/* Check for VLAN ID and register it on ISM device just for CLC handshake */
1090static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
1091 struct smc_init_info *ini)
1092{
1093 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1094 return SMC_CLC_DECL_ISMVLANERR;
1095 return 0;
1096}
1097
1098static int smc_find_proposal_devices(struct smc_sock *smc,
1099 struct smc_init_info *ini)
1100{
1101 int rc = 0;
1102
1103 /* check if there is an ism device available */
1104 if (!(ini->smcd_version & SMC_V1) ||
1105 smc_find_ism_device(smc, ini) ||
1106 smc_connect_ism_vlan_setup(smc, ini))
1107 ini->smcd_version &= ~SMC_V1;
1108 /* else ISM V1 is supported for this connection */
1109
1110 /* check if there is an rdma device available */
1111 if (!(ini->smcr_version & SMC_V1) ||
1112 smc_find_rdma_device(smc, ini))
1113 ini->smcr_version &= ~SMC_V1;
1114 /* else RDMA is supported for this connection */
1115
1116 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1117 ini->smcr_version & SMC_V1);
1118
1119 /* check if there is an ism v2 device available */
1120 if (!(ini->smcd_version & SMC_V2) ||
1121 !smc_ism_is_v2_capable() ||
1122 smc_find_ism_v2_device_clnt(smc, ini))
1123 ini->smcd_version &= ~SMC_V2;
1124
1125 /* check if there is an rdma v2 device available */
1126 ini->check_smcrv2 = true;
1127 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1128 if (!(ini->smcr_version & SMC_V2) ||
1129 smc->clcsock->sk->sk_family != AF_INET ||
1130 !smc_clc_ueid_count() ||
1131 smc_find_rdma_device(smc, ini))
1132 ini->smcr_version &= ~SMC_V2;
1133 ini->check_smcrv2 = false;
1134
1135 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1136 ini->smcr_version & SMC_V2);
1137
1138 /* if neither ISM nor RDMA are supported, fallback */
1139 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1140 rc = SMC_CLC_DECL_NOSMCDEV;
1141
1142 return rc;
1143}
1144
1145/* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1146 * used, the VLAN ID will be registered again during the connection setup.
1147 */
1148static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc,
1149 struct smc_init_info *ini)
1150{
1151 if (!smcd_indicated(ini->smc_type_v1))
1152 return 0;
1153 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1154 return SMC_CLC_DECL_CNFERR;
1155 return 0;
1156}
1157
1158#define SMC_CLC_MAX_ACCEPT_LEN \
1159 (sizeof(struct smc_clc_msg_accept_confirm) + \
1160 sizeof(struct smc_clc_first_contact_ext_v2x) + \
1161 sizeof(struct smc_clc_msg_trail))
1162
1163/* CLC handshake during connect */
1164static int smc_connect_clc(struct smc_sock *smc,
1165 struct smc_clc_msg_accept_confirm *aclc,
1166 struct smc_init_info *ini)
1167{
1168 int rc = 0;
1169
1170 /* do inband token exchange */
1171 rc = smc_clc_send_proposal(smc, ini);
1172 if (rc)
1173 return rc;
1174 /* receive SMC Accept CLC message */
1175 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN,
1176 SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1177}
1178
1179void smc_fill_gid_list(struct smc_link_group *lgr,
1180 struct smc_gidlist *gidlist,
1181 struct smc_ib_device *known_dev, u8 *known_gid)
1182{
1183 struct smc_init_info *alt_ini = NULL;
1184
1185 memset(gidlist, 0, sizeof(*gidlist));
1186 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1187
1188 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1189 if (!alt_ini)
1190 goto out;
1191
1192 alt_ini->vlan_id = lgr->vlan_id;
1193 alt_ini->check_smcrv2 = true;
1194 alt_ini->smcrv2.saddr = lgr->saddr;
1195 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1196
1197 if (!alt_ini->smcrv2.ib_dev_v2)
1198 goto out;
1199
1200 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1201 SMC_GID_SIZE);
1202
1203out:
1204 kfree(alt_ini);
1205}
1206
1207static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1208 struct smc_clc_msg_accept_confirm *aclc,
1209 struct smc_init_info *ini)
1210{
1211 struct smc_clc_first_contact_ext *fce =
1212 smc_get_clc_first_contact_ext(aclc, false);
1213 struct net *net = sock_net(&smc->sk);
1214 int rc;
1215
1216 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1217 return 0;
1218
1219 if (fce->v2_direct) {
1220 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1221 ini->smcrv2.uses_gateway = false;
1222 } else {
1223 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr,
1224 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1225 ini->smcrv2.nexthop_mac,
1226 &ini->smcrv2.uses_gateway))
1227 return SMC_CLC_DECL_NOROUTE;
1228 if (!ini->smcrv2.uses_gateway) {
1229 /* mismatch: peer claims indirect, but its direct */
1230 return SMC_CLC_DECL_NOINDIRECT;
1231 }
1232 }
1233
1234 ini->release_nr = fce->release;
1235 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1236 if (rc)
1237 return rc;
1238
1239 return 0;
1240}
1241
1242/* setup for RDMA connection of client */
1243static int smc_connect_rdma(struct smc_sock *smc,
1244 struct smc_clc_msg_accept_confirm *aclc,
1245 struct smc_init_info *ini)
1246{
1247 int i, reason_code = 0;
1248 struct smc_link *link;
1249 u8 *eid = NULL;
1250
1251 ini->is_smcd = false;
1252 ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1253 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1254 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1255 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1256 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1257 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1258 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1259
1260 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1261 if (reason_code)
1262 return reason_code;
1263
1264 mutex_lock(&smc_client_lgr_pending);
1265 reason_code = smc_conn_create(smc, ini);
1266 if (reason_code) {
1267 mutex_unlock(&smc_client_lgr_pending);
1268 return reason_code;
1269 }
1270
1271 smc_conn_save_peer_info(smc, aclc);
1272
1273 if (ini->first_contact_local) {
1274 link = smc->conn.lnk;
1275 } else {
1276 /* set link that was assigned by server */
1277 link = NULL;
1278 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1279 struct smc_link *l = &smc->conn.lgr->lnk[i];
1280
1281 if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1282 !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1283 SMC_GID_SIZE) &&
1284 (aclc->hdr.version > SMC_V1 ||
1285 !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1286 sizeof(l->peer_mac)))) {
1287 link = l;
1288 break;
1289 }
1290 }
1291 if (!link) {
1292 reason_code = SMC_CLC_DECL_NOSRVLINK;
1293 goto connect_abort;
1294 }
1295 smc_switch_link_and_count(&smc->conn, link);
1296 }
1297
1298 /* create send buffer and rmb */
1299 if (smc_buf_create(smc, false)) {
1300 reason_code = SMC_CLC_DECL_MEM;
1301 goto connect_abort;
1302 }
1303
1304 if (ini->first_contact_local)
1305 smc_link_save_peer_info(link, aclc, ini);
1306
1307 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1308 reason_code = SMC_CLC_DECL_ERR_RTOK;
1309 goto connect_abort;
1310 }
1311
1312 smc_close_init(smc);
1313 smc_rx_init(smc);
1314
1315 if (ini->first_contact_local) {
1316 if (smc_ib_ready_link(link)) {
1317 reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1318 goto connect_abort;
1319 }
1320 } else {
1321 /* reg sendbufs if they were vzalloced */
1322 if (smc->conn.sndbuf_desc->is_vm) {
1323 if (smcr_lgr_reg_sndbufs(link, smc->conn.sndbuf_desc)) {
1324 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1325 goto connect_abort;
1326 }
1327 }
1328 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1329 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1330 goto connect_abort;
1331 }
1332 }
1333
1334 if (aclc->hdr.version > SMC_V1) {
1335 eid = aclc->r1.eid;
1336 if (ini->first_contact_local)
1337 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1338 link->smcibdev, link->gid);
1339 }
1340
1341 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1342 aclc->hdr.version, eid, ini);
1343 if (reason_code)
1344 goto connect_abort;
1345
1346 smc_tx_init(smc);
1347
1348 if (ini->first_contact_local) {
1349 /* QP confirmation over RoCE fabric */
1350 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1351 reason_code = smcr_clnt_conf_first_link(smc);
1352 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1353 if (reason_code)
1354 goto connect_abort;
1355 }
1356 mutex_unlock(&smc_client_lgr_pending);
1357
1358 smc_copy_sock_settings_to_clc(smc);
1359 smc->connect_nonblock = 0;
1360 if (smc->sk.sk_state == SMC_INIT)
1361 smc->sk.sk_state = SMC_ACTIVE;
1362
1363 return 0;
1364connect_abort:
1365 smc_conn_abort(smc, ini->first_contact_local);
1366 mutex_unlock(&smc_client_lgr_pending);
1367 smc->connect_nonblock = 0;
1368
1369 return reason_code;
1370}
1371
1372/* The server has chosen one of the proposed ISM devices for the communication.
1373 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1374 */
1375static int
1376smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc,
1377 struct smc_init_info *ini)
1378{
1379 int i;
1380
1381 for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1382 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1383 ini->ism_selected = i;
1384 return 0;
1385 }
1386 }
1387
1388 return -EPROTO;
1389}
1390
1391/* setup for ISM connection of client */
1392static int smc_connect_ism(struct smc_sock *smc,
1393 struct smc_clc_msg_accept_confirm *aclc,
1394 struct smc_init_info *ini)
1395{
1396 u8 *eid = NULL;
1397 int rc = 0;
1398
1399 ini->is_smcd = true;
1400 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1401
1402 if (aclc->hdr.version == SMC_V2) {
1403 if (ini->first_contact_peer) {
1404 struct smc_clc_first_contact_ext *fce =
1405 smc_get_clc_first_contact_ext(aclc, true);
1406
1407 ini->release_nr = fce->release;
1408 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1409 if (rc)
1410 return rc;
1411 }
1412
1413 rc = smc_v2_determine_accepted_chid(aclc, ini);
1414 if (rc)
1415 return rc;
1416
1417 if (__smc_ism_is_virtual(ini->ism_chid[ini->ism_selected]))
1418 ini->ism_peer_gid[ini->ism_selected].gid_ext =
1419 ntohll(aclc->d1.gid_ext);
1420 /* for non-virtual ISM devices, peer gid_ext remains 0. */
1421 }
1422 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid);
1423
1424 /* there is only one lgr role for SMC-D; use server lock */
1425 mutex_lock(&smc_server_lgr_pending);
1426 rc = smc_conn_create(smc, ini);
1427 if (rc) {
1428 mutex_unlock(&smc_server_lgr_pending);
1429 return rc;
1430 }
1431
1432 /* Create send and receive buffers */
1433 rc = smc_buf_create(smc, true);
1434 if (rc) {
1435 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1436 goto connect_abort;
1437 }
1438
1439 smc_conn_save_peer_info(smc, aclc);
1440 smc_close_init(smc);
1441 smc_rx_init(smc);
1442 smc_tx_init(smc);
1443
1444 if (aclc->hdr.version > SMC_V1)
1445 eid = aclc->d1.eid;
1446
1447 rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1448 aclc->hdr.version, eid, ini);
1449 if (rc)
1450 goto connect_abort;
1451 mutex_unlock(&smc_server_lgr_pending);
1452
1453 smc_copy_sock_settings_to_clc(smc);
1454 smc->connect_nonblock = 0;
1455 if (smc->sk.sk_state == SMC_INIT)
1456 smc->sk.sk_state = SMC_ACTIVE;
1457
1458 return 0;
1459connect_abort:
1460 smc_conn_abort(smc, ini->first_contact_local);
1461 mutex_unlock(&smc_server_lgr_pending);
1462 smc->connect_nonblock = 0;
1463
1464 return rc;
1465}
1466
1467/* check if received accept type and version matches a proposed one */
1468static int smc_connect_check_aclc(struct smc_init_info *ini,
1469 struct smc_clc_msg_accept_confirm *aclc)
1470{
1471 if (aclc->hdr.typev1 != SMC_TYPE_R &&
1472 aclc->hdr.typev1 != SMC_TYPE_D)
1473 return SMC_CLC_DECL_MODEUNSUPP;
1474
1475 if (aclc->hdr.version >= SMC_V2) {
1476 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1477 !smcr_indicated(ini->smc_type_v2)) ||
1478 (aclc->hdr.typev1 == SMC_TYPE_D &&
1479 !smcd_indicated(ini->smc_type_v2)))
1480 return SMC_CLC_DECL_MODEUNSUPP;
1481 } else {
1482 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1483 !smcr_indicated(ini->smc_type_v1)) ||
1484 (aclc->hdr.typev1 == SMC_TYPE_D &&
1485 !smcd_indicated(ini->smc_type_v1)))
1486 return SMC_CLC_DECL_MODEUNSUPP;
1487 }
1488
1489 return 0;
1490}
1491
1492/* perform steps before actually connecting */
1493static int __smc_connect(struct smc_sock *smc)
1494{
1495 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1496 struct smc_clc_msg_accept_confirm *aclc;
1497 struct smc_init_info *ini = NULL;
1498 u8 *buf = NULL;
1499 int rc = 0;
1500
1501 if (smc->use_fallback)
1502 return smc_connect_fallback(smc, smc->fallback_rsn);
1503
1504 /* if peer has not signalled SMC-capability, fall back */
1505 if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1506 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1507
1508 /* IPSec connections opt out of SMC optimizations */
1509 if (using_ipsec(smc))
1510 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1511 version);
1512
1513 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1514 if (!ini)
1515 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1516 version);
1517
1518 ini->smcd_version = SMC_V1 | SMC_V2;
1519 ini->smcr_version = SMC_V1 | SMC_V2;
1520 ini->smc_type_v1 = SMC_TYPE_B;
1521 ini->smc_type_v2 = SMC_TYPE_B;
1522
1523 /* get vlan id from IP device */
1524 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1525 ini->smcd_version &= ~SMC_V1;
1526 ini->smcr_version = 0;
1527 ini->smc_type_v1 = SMC_TYPE_N;
1528 if (!ini->smcd_version) {
1529 rc = SMC_CLC_DECL_GETVLANERR;
1530 goto fallback;
1531 }
1532 }
1533
1534 rc = smc_find_proposal_devices(smc, ini);
1535 if (rc)
1536 goto fallback;
1537
1538 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1539 if (!buf) {
1540 rc = SMC_CLC_DECL_MEM;
1541 goto fallback;
1542 }
1543 aclc = (struct smc_clc_msg_accept_confirm *)buf;
1544
1545 /* perform CLC handshake */
1546 rc = smc_connect_clc(smc, aclc, ini);
1547 if (rc) {
1548 /* -EAGAIN on timeout, see tcp_recvmsg() */
1549 if (rc == -EAGAIN) {
1550 rc = -ETIMEDOUT;
1551 smc->sk.sk_err = ETIMEDOUT;
1552 }
1553 goto vlan_cleanup;
1554 }
1555
1556 /* check if smc modes and versions of CLC proposal and accept match */
1557 rc = smc_connect_check_aclc(ini, aclc);
1558 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1559 if (rc)
1560 goto vlan_cleanup;
1561
1562 /* depending on previous steps, connect using rdma or ism */
1563 if (aclc->hdr.typev1 == SMC_TYPE_R) {
1564 ini->smcr_version = version;
1565 rc = smc_connect_rdma(smc, aclc, ini);
1566 } else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1567 ini->smcd_version = version;
1568 rc = smc_connect_ism(smc, aclc, ini);
1569 }
1570 if (rc)
1571 goto vlan_cleanup;
1572
1573 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1574 smc_connect_ism_vlan_cleanup(smc, ini);
1575 kfree(buf);
1576 kfree(ini);
1577 return 0;
1578
1579vlan_cleanup:
1580 smc_connect_ism_vlan_cleanup(smc, ini);
1581 kfree(buf);
1582fallback:
1583 kfree(ini);
1584 return smc_connect_decline_fallback(smc, rc, version);
1585}
1586
1587static void smc_connect_work(struct work_struct *work)
1588{
1589 struct smc_sock *smc = container_of(work, struct smc_sock,
1590 connect_work);
1591 long timeo = smc->sk.sk_sndtimeo;
1592 int rc = 0;
1593
1594 if (!timeo)
1595 timeo = MAX_SCHEDULE_TIMEOUT;
1596 lock_sock(smc->clcsock->sk);
1597 if (smc->clcsock->sk->sk_err) {
1598 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1599 } else if ((1 << smc->clcsock->sk->sk_state) &
1600 (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1601 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1602 if ((rc == -EPIPE) &&
1603 ((1 << smc->clcsock->sk->sk_state) &
1604 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1605 rc = 0;
1606 }
1607 release_sock(smc->clcsock->sk);
1608 lock_sock(&smc->sk);
1609 if (rc != 0 || smc->sk.sk_err) {
1610 smc->sk.sk_state = SMC_CLOSED;
1611 if (rc == -EPIPE || rc == -EAGAIN)
1612 smc->sk.sk_err = EPIPE;
1613 else if (rc == -ECONNREFUSED)
1614 smc->sk.sk_err = ECONNREFUSED;
1615 else if (signal_pending(current))
1616 smc->sk.sk_err = -sock_intr_errno(timeo);
1617 sock_put(&smc->sk); /* passive closing */
1618 goto out;
1619 }
1620
1621 rc = __smc_connect(smc);
1622 if (rc < 0)
1623 smc->sk.sk_err = -rc;
1624
1625out:
1626 if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1627 if (smc->sk.sk_err) {
1628 smc->sk.sk_state_change(&smc->sk);
1629 } else { /* allow polling before and after fallback decision */
1630 smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1631 smc->sk.sk_write_space(&smc->sk);
1632 }
1633 }
1634 release_sock(&smc->sk);
1635}
1636
1637static int smc_connect(struct socket *sock, struct sockaddr *addr,
1638 int alen, int flags)
1639{
1640 struct sock *sk = sock->sk;
1641 struct smc_sock *smc;
1642 int rc = -EINVAL;
1643
1644 smc = smc_sk(sk);
1645
1646 /* separate smc parameter checking to be safe */
1647 if (alen < sizeof(addr->sa_family))
1648 goto out_err;
1649 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1650 goto out_err;
1651
1652 lock_sock(sk);
1653 switch (sock->state) {
1654 default:
1655 rc = -EINVAL;
1656 goto out;
1657 case SS_CONNECTED:
1658 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1659 goto out;
1660 case SS_CONNECTING:
1661 if (sk->sk_state == SMC_ACTIVE)
1662 goto connected;
1663 break;
1664 case SS_UNCONNECTED:
1665 sock->state = SS_CONNECTING;
1666 break;
1667 }
1668
1669 switch (sk->sk_state) {
1670 default:
1671 goto out;
1672 case SMC_CLOSED:
1673 rc = sock_error(sk) ? : -ECONNABORTED;
1674 sock->state = SS_UNCONNECTED;
1675 goto out;
1676 case SMC_ACTIVE:
1677 rc = -EISCONN;
1678 goto out;
1679 case SMC_INIT:
1680 break;
1681 }
1682
1683 smc_copy_sock_settings_to_clc(smc);
1684 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1685 if (smc->connect_nonblock) {
1686 rc = -EALREADY;
1687 goto out;
1688 }
1689 rc = kernel_connect(smc->clcsock, addr, alen, flags);
1690 if (rc && rc != -EINPROGRESS)
1691 goto out;
1692
1693 if (smc->use_fallback) {
1694 sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1695 goto out;
1696 }
1697 sock_hold(&smc->sk); /* sock put in passive closing */
1698 if (flags & O_NONBLOCK) {
1699 if (queue_work(smc_hs_wq, &smc->connect_work))
1700 smc->connect_nonblock = 1;
1701 rc = -EINPROGRESS;
1702 goto out;
1703 } else {
1704 rc = __smc_connect(smc);
1705 if (rc < 0)
1706 goto out;
1707 }
1708
1709connected:
1710 rc = 0;
1711 sock->state = SS_CONNECTED;
1712out:
1713 release_sock(sk);
1714out_err:
1715 return rc;
1716}
1717
1718static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1719{
1720 struct socket *new_clcsock = NULL;
1721 struct sock *lsk = &lsmc->sk;
1722 struct sock *new_sk;
1723 int rc = -EINVAL;
1724
1725 release_sock(lsk);
1726 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1727 if (!new_sk) {
1728 rc = -ENOMEM;
1729 lsk->sk_err = ENOMEM;
1730 *new_smc = NULL;
1731 lock_sock(lsk);
1732 goto out;
1733 }
1734 *new_smc = smc_sk(new_sk);
1735
1736 mutex_lock(&lsmc->clcsock_release_lock);
1737 if (lsmc->clcsock)
1738 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1739 mutex_unlock(&lsmc->clcsock_release_lock);
1740 lock_sock(lsk);
1741 if (rc < 0 && rc != -EAGAIN)
1742 lsk->sk_err = -rc;
1743 if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1744 new_sk->sk_prot->unhash(new_sk);
1745 if (new_clcsock)
1746 sock_release(new_clcsock);
1747 new_sk->sk_state = SMC_CLOSED;
1748 smc_sock_set_flag(new_sk, SOCK_DEAD);
1749 sock_put(new_sk); /* final */
1750 *new_smc = NULL;
1751 goto out;
1752 }
1753
1754 /* new clcsock has inherited the smc listen-specific sk_data_ready
1755 * function; switch it back to the original sk_data_ready function
1756 */
1757 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1758
1759 /* if new clcsock has also inherited the fallback-specific callback
1760 * functions, switch them back to the original ones.
1761 */
1762 if (lsmc->use_fallback) {
1763 if (lsmc->clcsk_state_change)
1764 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1765 if (lsmc->clcsk_write_space)
1766 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1767 if (lsmc->clcsk_error_report)
1768 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1769 }
1770
1771 (*new_smc)->clcsock = new_clcsock;
1772out:
1773 return rc;
1774}
1775
1776/* add a just created sock to the accept queue of the listen sock as
1777 * candidate for a following socket accept call from user space
1778 */
1779static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1780{
1781 struct smc_sock *par = smc_sk(parent);
1782
1783 sock_hold(sk); /* sock_put in smc_accept_unlink () */
1784 spin_lock(&par->accept_q_lock);
1785 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1786 spin_unlock(&par->accept_q_lock);
1787 sk_acceptq_added(parent);
1788}
1789
1790/* remove a socket from the accept queue of its parental listening socket */
1791static void smc_accept_unlink(struct sock *sk)
1792{
1793 struct smc_sock *par = smc_sk(sk)->listen_smc;
1794
1795 spin_lock(&par->accept_q_lock);
1796 list_del_init(&smc_sk(sk)->accept_q);
1797 spin_unlock(&par->accept_q_lock);
1798 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1799 sock_put(sk); /* sock_hold in smc_accept_enqueue */
1800}
1801
1802/* remove a sock from the accept queue to bind it to a new socket created
1803 * for a socket accept call from user space
1804 */
1805struct sock *smc_accept_dequeue(struct sock *parent,
1806 struct socket *new_sock)
1807{
1808 struct smc_sock *isk, *n;
1809 struct sock *new_sk;
1810
1811 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1812 new_sk = (struct sock *)isk;
1813
1814 smc_accept_unlink(new_sk);
1815 if (new_sk->sk_state == SMC_CLOSED) {
1816 new_sk->sk_prot->unhash(new_sk);
1817 if (isk->clcsock) {
1818 sock_release(isk->clcsock);
1819 isk->clcsock = NULL;
1820 }
1821 sock_put(new_sk); /* final */
1822 continue;
1823 }
1824 if (new_sock) {
1825 sock_graft(new_sk, new_sock);
1826 new_sock->state = SS_CONNECTED;
1827 if (isk->use_fallback) {
1828 smc_sk(new_sk)->clcsock->file = new_sock->file;
1829 isk->clcsock->file->private_data = isk->clcsock;
1830 }
1831 }
1832 return new_sk;
1833 }
1834 return NULL;
1835}
1836
1837/* clean up for a created but never accepted sock */
1838void smc_close_non_accepted(struct sock *sk)
1839{
1840 struct smc_sock *smc = smc_sk(sk);
1841
1842 sock_hold(sk); /* sock_put below */
1843 lock_sock(sk);
1844 if (!sk->sk_lingertime)
1845 /* wait for peer closing */
1846 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT);
1847 __smc_release(smc);
1848 release_sock(sk);
1849 sock_put(sk); /* sock_hold above */
1850 sock_put(sk); /* final sock_put */
1851}
1852
1853static int smcr_serv_conf_first_link(struct smc_sock *smc)
1854{
1855 struct smc_link *link = smc->conn.lnk;
1856 struct smc_llc_qentry *qentry;
1857 int rc;
1858
1859 /* reg the sndbuf if it was vzalloced*/
1860 if (smc->conn.sndbuf_desc->is_vm) {
1861 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1862 return SMC_CLC_DECL_ERR_REGBUF;
1863 }
1864
1865 /* reg the rmb */
1866 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1867 return SMC_CLC_DECL_ERR_REGBUF;
1868
1869 /* send CONFIRM LINK request to client over the RoCE fabric */
1870 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1871 if (rc < 0)
1872 return SMC_CLC_DECL_TIMEOUT_CL;
1873
1874 /* receive CONFIRM LINK response from client over the RoCE fabric */
1875 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1876 SMC_LLC_CONFIRM_LINK);
1877 if (!qentry) {
1878 struct smc_clc_msg_decline dclc;
1879
1880 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1881 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1882 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1883 }
1884 smc_llc_save_peer_uid(qentry);
1885 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1886 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1887 if (rc)
1888 return SMC_CLC_DECL_RMBE_EC;
1889
1890 /* confirm_rkey is implicit on 1st contact */
1891 smc->conn.rmb_desc->is_conf_rkey = true;
1892
1893 smc_llc_link_active(link);
1894 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1895
1896 if (link->lgr->max_links > 1) {
1897 down_write(&link->lgr->llc_conf_mutex);
1898 /* initial contact - try to establish second link */
1899 smc_llc_srv_add_link(link, NULL);
1900 up_write(&link->lgr->llc_conf_mutex);
1901 }
1902 return 0;
1903}
1904
1905/* listen worker: finish */
1906static void smc_listen_out(struct smc_sock *new_smc)
1907{
1908 struct smc_sock *lsmc = new_smc->listen_smc;
1909 struct sock *newsmcsk = &new_smc->sk;
1910
1911 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1912 atomic_dec(&lsmc->queued_smc_hs);
1913
1914 if (lsmc->sk.sk_state == SMC_LISTEN) {
1915 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1916 smc_accept_enqueue(&lsmc->sk, newsmcsk);
1917 release_sock(&lsmc->sk);
1918 } else { /* no longer listening */
1919 smc_close_non_accepted(newsmcsk);
1920 }
1921
1922 /* Wake up accept */
1923 lsmc->sk.sk_data_ready(&lsmc->sk);
1924 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1925}
1926
1927/* listen worker: finish in state connected */
1928static void smc_listen_out_connected(struct smc_sock *new_smc)
1929{
1930 struct sock *newsmcsk = &new_smc->sk;
1931
1932 if (newsmcsk->sk_state == SMC_INIT)
1933 newsmcsk->sk_state = SMC_ACTIVE;
1934
1935 smc_listen_out(new_smc);
1936}
1937
1938/* listen worker: finish in error state */
1939static void smc_listen_out_err(struct smc_sock *new_smc)
1940{
1941 struct sock *newsmcsk = &new_smc->sk;
1942 struct net *net = sock_net(newsmcsk);
1943
1944 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1945 if (newsmcsk->sk_state == SMC_INIT)
1946 sock_put(&new_smc->sk); /* passive closing */
1947 newsmcsk->sk_state = SMC_CLOSED;
1948
1949 smc_listen_out(new_smc);
1950}
1951
1952/* listen worker: decline and fall back if possible */
1953static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1954 int local_first, u8 version)
1955{
1956 /* RDMA setup failed, switch back to TCP */
1957 smc_conn_abort(new_smc, local_first);
1958 if (reason_code < 0 ||
1959 smc_switch_to_fallback(new_smc, reason_code)) {
1960 /* error, no fallback possible */
1961 smc_listen_out_err(new_smc);
1962 return;
1963 }
1964 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1965 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1966 smc_listen_out_err(new_smc);
1967 return;
1968 }
1969 }
1970 smc_listen_out_connected(new_smc);
1971}
1972
1973/* listen worker: version checking */
1974static int smc_listen_v2_check(struct smc_sock *new_smc,
1975 struct smc_clc_msg_proposal *pclc,
1976 struct smc_init_info *ini)
1977{
1978 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
1979 struct smc_clc_v2_extension *pclc_v2_ext;
1980 int rc = SMC_CLC_DECL_PEERNOSMC;
1981
1982 ini->smc_type_v1 = pclc->hdr.typev1;
1983 ini->smc_type_v2 = pclc->hdr.typev2;
1984 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1985 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1986 if (pclc->hdr.version > SMC_V1) {
1987 if (smcd_indicated(ini->smc_type_v2))
1988 ini->smcd_version |= SMC_V2;
1989 if (smcr_indicated(ini->smc_type_v2))
1990 ini->smcr_version |= SMC_V2;
1991 }
1992 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
1993 rc = SMC_CLC_DECL_PEERNOSMC;
1994 goto out;
1995 }
1996 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1997 if (!pclc_v2_ext) {
1998 ini->smcd_version &= ~SMC_V2;
1999 ini->smcr_version &= ~SMC_V2;
2000 rc = SMC_CLC_DECL_NOV2EXT;
2001 goto out;
2002 }
2003 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
2004 if (ini->smcd_version & SMC_V2) {
2005 if (!smc_ism_is_v2_capable()) {
2006 ini->smcd_version &= ~SMC_V2;
2007 rc = SMC_CLC_DECL_NOISM2SUPP;
2008 } else if (!pclc_smcd_v2_ext) {
2009 ini->smcd_version &= ~SMC_V2;
2010 rc = SMC_CLC_DECL_NOV2DEXT;
2011 } else if (!pclc_v2_ext->hdr.eid_cnt &&
2012 !pclc_v2_ext->hdr.flag.seid) {
2013 ini->smcd_version &= ~SMC_V2;
2014 rc = SMC_CLC_DECL_NOUEID;
2015 }
2016 }
2017 if (ini->smcr_version & SMC_V2) {
2018 if (!pclc_v2_ext->hdr.eid_cnt) {
2019 ini->smcr_version &= ~SMC_V2;
2020 rc = SMC_CLC_DECL_NOUEID;
2021 }
2022 }
2023
2024 ini->release_nr = pclc_v2_ext->hdr.flag.release;
2025 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE)
2026 ini->release_nr = SMC_RELEASE;
2027
2028out:
2029 if (!ini->smcd_version && !ini->smcr_version)
2030 return rc;
2031
2032 return 0;
2033}
2034
2035/* listen worker: check prefixes */
2036static int smc_listen_prfx_check(struct smc_sock *new_smc,
2037 struct smc_clc_msg_proposal *pclc)
2038{
2039 struct smc_clc_msg_proposal_prefix *pclc_prfx;
2040 struct socket *newclcsock = new_smc->clcsock;
2041
2042 if (pclc->hdr.typev1 == SMC_TYPE_N)
2043 return 0;
2044 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2045 if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2046 return SMC_CLC_DECL_DIFFPREFIX;
2047
2048 return 0;
2049}
2050
2051/* listen worker: initialize connection and buffers */
2052static int smc_listen_rdma_init(struct smc_sock *new_smc,
2053 struct smc_init_info *ini)
2054{
2055 int rc;
2056
2057 /* allocate connection / link group */
2058 rc = smc_conn_create(new_smc, ini);
2059 if (rc)
2060 return rc;
2061
2062 /* create send buffer and rmb */
2063 if (smc_buf_create(new_smc, false)) {
2064 smc_conn_abort(new_smc, ini->first_contact_local);
2065 return SMC_CLC_DECL_MEM;
2066 }
2067
2068 return 0;
2069}
2070
2071/* listen worker: initialize connection and buffers for SMC-D */
2072static int smc_listen_ism_init(struct smc_sock *new_smc,
2073 struct smc_init_info *ini)
2074{
2075 int rc;
2076
2077 rc = smc_conn_create(new_smc, ini);
2078 if (rc)
2079 return rc;
2080
2081 /* Create send and receive buffers */
2082 rc = smc_buf_create(new_smc, true);
2083 if (rc) {
2084 smc_conn_abort(new_smc, ini->first_contact_local);
2085 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2086 SMC_CLC_DECL_MEM;
2087 }
2088
2089 return 0;
2090}
2091
2092static bool smc_is_already_selected(struct smcd_dev *smcd,
2093 struct smc_init_info *ini,
2094 int matches)
2095{
2096 int i;
2097
2098 for (i = 0; i < matches; i++)
2099 if (smcd == ini->ism_dev[i])
2100 return true;
2101
2102 return false;
2103}
2104
2105/* check for ISM devices matching proposed ISM devices */
2106static void smc_check_ism_v2_match(struct smc_init_info *ini,
2107 u16 proposed_chid,
2108 struct smcd_gid *proposed_gid,
2109 unsigned int *matches)
2110{
2111 struct smcd_dev *smcd;
2112
2113 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2114 if (smcd->going_away)
2115 continue;
2116 if (smc_is_already_selected(smcd, ini, *matches))
2117 continue;
2118 if (smc_ism_get_chid(smcd) == proposed_chid &&
2119 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2120 ini->ism_peer_gid[*matches].gid = proposed_gid->gid;
2121 if (__smc_ism_is_virtual(proposed_chid))
2122 ini->ism_peer_gid[*matches].gid_ext =
2123 proposed_gid->gid_ext;
2124 /* non-virtual ISM's peer gid_ext remains 0. */
2125 ini->ism_dev[*matches] = smcd;
2126 (*matches)++;
2127 break;
2128 }
2129 }
2130}
2131
2132static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini)
2133{
2134 if (!ini->rc)
2135 ini->rc = rc;
2136}
2137
2138static void smc_find_ism_v2_device_serv(struct smc_sock *new_smc,
2139 struct smc_clc_msg_proposal *pclc,
2140 struct smc_init_info *ini)
2141{
2142 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
2143 struct smc_clc_v2_extension *smc_v2_ext;
2144 struct smc_clc_msg_smcd *pclc_smcd;
2145 unsigned int matches = 0;
2146 struct smcd_gid smcd_gid;
2147 u8 smcd_version;
2148 u8 *eid = NULL;
2149 int i, rc;
2150 u16 chid;
2151
2152 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2153 goto not_found;
2154
2155 pclc_smcd = smc_get_clc_msg_smcd(pclc);
2156 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2157 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2158
2159 mutex_lock(&smcd_dev_list.mutex);
2160 if (pclc_smcd->ism.chid) {
2161 /* check for ISM device matching proposed native ISM device */
2162 smcd_gid.gid = ntohll(pclc_smcd->ism.gid);
2163 smcd_gid.gid_ext = 0;
2164 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2165 &smcd_gid, &matches);
2166 }
2167 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) {
2168 /* check for ISM devices matching proposed non-native ISM
2169 * devices
2170 */
2171 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid);
2172 smcd_gid.gid_ext = 0;
2173 chid = ntohs(smcd_v2_ext->gidchid[i].chid);
2174 if (__smc_ism_is_virtual(chid)) {
2175 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt ||
2176 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid))
2177 /* each virtual ISM device takes two GID-CHID
2178 * entries and CHID of the second entry repeats
2179 * that of the first entry.
2180 *
2181 * So check if the next GID-CHID entry exists
2182 * and both two entries' CHIDs are the same.
2183 */
2184 continue;
2185 smcd_gid.gid_ext =
2186 ntohll(smcd_v2_ext->gidchid[++i].gid);
2187 }
2188 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches);
2189 }
2190 mutex_unlock(&smcd_dev_list.mutex);
2191
2192 if (!ini->ism_dev[0]) {
2193 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2194 goto not_found;
2195 }
2196
2197 smc_ism_get_system_eid(&eid);
2198 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2199 smcd_v2_ext->system_eid, eid))
2200 goto not_found;
2201
2202 /* separate - outside the smcd_dev_list.lock */
2203 smcd_version = ini->smcd_version;
2204 for (i = 0; i < matches; i++) {
2205 ini->smcd_version = SMC_V2;
2206 ini->is_smcd = true;
2207 ini->ism_selected = i;
2208 rc = smc_listen_ism_init(new_smc, ini);
2209 if (rc) {
2210 smc_find_ism_store_rc(rc, ini);
2211 /* try next active ISM device */
2212 continue;
2213 }
2214 return; /* matching and usable V2 ISM device found */
2215 }
2216 /* no V2 ISM device could be initialized */
2217 ini->smcd_version = smcd_version; /* restore original value */
2218 ini->negotiated_eid[0] = 0;
2219
2220not_found:
2221 ini->smcd_version &= ~SMC_V2;
2222 ini->ism_dev[0] = NULL;
2223 ini->is_smcd = false;
2224}
2225
2226static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2227 struct smc_clc_msg_proposal *pclc,
2228 struct smc_init_info *ini)
2229{
2230 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2231 int rc = 0;
2232
2233 /* check if ISM V1 is available */
2234 if (!(ini->smcd_version & SMC_V1) || !smcd_indicated(ini->smc_type_v1))
2235 goto not_found;
2236 ini->is_smcd = true; /* prepare ISM check */
2237 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid);
2238 ini->ism_peer_gid[0].gid_ext = 0;
2239 rc = smc_find_ism_device(new_smc, ini);
2240 if (rc)
2241 goto not_found;
2242 ini->ism_selected = 0;
2243 rc = smc_listen_ism_init(new_smc, ini);
2244 if (!rc)
2245 return; /* V1 ISM device found */
2246
2247not_found:
2248 smc_find_ism_store_rc(rc, ini);
2249 ini->smcd_version &= ~SMC_V1;
2250 ini->ism_dev[0] = NULL;
2251 ini->is_smcd = false;
2252}
2253
2254/* listen worker: register buffers */
2255static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2256{
2257 struct smc_connection *conn = &new_smc->conn;
2258
2259 if (!local_first) {
2260 /* reg sendbufs if they were vzalloced */
2261 if (conn->sndbuf_desc->is_vm) {
2262 if (smcr_lgr_reg_sndbufs(conn->lnk,
2263 conn->sndbuf_desc))
2264 return SMC_CLC_DECL_ERR_REGBUF;
2265 }
2266 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2267 return SMC_CLC_DECL_ERR_REGBUF;
2268 }
2269
2270 return 0;
2271}
2272
2273static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2274 struct smc_clc_msg_proposal *pclc,
2275 struct smc_init_info *ini)
2276{
2277 struct smc_clc_v2_extension *smc_v2_ext;
2278 u8 smcr_version;
2279 int rc;
2280
2281 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2282 goto not_found;
2283
2284 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2285 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2286 goto not_found;
2287
2288 /* prepare RDMA check */
2289 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2290 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2291 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2292 ini->check_smcrv2 = true;
2293 ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2294 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2295 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2296 rc = smc_find_rdma_device(new_smc, ini);
2297 if (rc) {
2298 smc_find_ism_store_rc(rc, ini);
2299 goto not_found;
2300 }
2301 if (!ini->smcrv2.uses_gateway)
2302 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2303
2304 smcr_version = ini->smcr_version;
2305 ini->smcr_version = SMC_V2;
2306 rc = smc_listen_rdma_init(new_smc, ini);
2307 if (!rc) {
2308 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2309 if (rc)
2310 smc_conn_abort(new_smc, ini->first_contact_local);
2311 }
2312 if (!rc)
2313 return;
2314 ini->smcr_version = smcr_version;
2315 smc_find_ism_store_rc(rc, ini);
2316
2317not_found:
2318 ini->smcr_version &= ~SMC_V2;
2319 ini->smcrv2.ib_dev_v2 = NULL;
2320 ini->check_smcrv2 = false;
2321}
2322
2323static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2324 struct smc_clc_msg_proposal *pclc,
2325 struct smc_init_info *ini)
2326{
2327 int rc;
2328
2329 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2330 return SMC_CLC_DECL_NOSMCDEV;
2331
2332 /* prepare RDMA check */
2333 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2334 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2335 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2336 rc = smc_find_rdma_device(new_smc, ini);
2337 if (rc) {
2338 /* no RDMA device found */
2339 return SMC_CLC_DECL_NOSMCDEV;
2340 }
2341 rc = smc_listen_rdma_init(new_smc, ini);
2342 if (rc)
2343 return rc;
2344 return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2345}
2346
2347/* determine the local device matching to proposal */
2348static int smc_listen_find_device(struct smc_sock *new_smc,
2349 struct smc_clc_msg_proposal *pclc,
2350 struct smc_init_info *ini)
2351{
2352 int prfx_rc;
2353
2354 /* check for ISM device matching V2 proposed device */
2355 smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2356 if (ini->ism_dev[0])
2357 return 0;
2358
2359 /* check for matching IP prefix and subnet length (V1) */
2360 prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2361 if (prfx_rc)
2362 smc_find_ism_store_rc(prfx_rc, ini);
2363
2364 /* get vlan id from IP device */
2365 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2366 return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2367
2368 /* check for ISM device matching V1 proposed device */
2369 if (!prfx_rc)
2370 smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2371 if (ini->ism_dev[0])
2372 return 0;
2373
2374 if (!smcr_indicated(pclc->hdr.typev1) &&
2375 !smcr_indicated(pclc->hdr.typev2))
2376 /* skip RDMA and decline */
2377 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2378
2379 /* check if RDMA V2 is available */
2380 smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2381 if (ini->smcrv2.ib_dev_v2)
2382 return 0;
2383
2384 /* check if RDMA V1 is available */
2385 if (!prfx_rc) {
2386 int rc;
2387
2388 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2389 smc_find_ism_store_rc(rc, ini);
2390 return (!rc) ? 0 : ini->rc;
2391 }
2392 return prfx_rc;
2393}
2394
2395/* listen worker: finish RDMA setup */
2396static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2397 struct smc_clc_msg_accept_confirm *cclc,
2398 bool local_first,
2399 struct smc_init_info *ini)
2400{
2401 struct smc_link *link = new_smc->conn.lnk;
2402 int reason_code = 0;
2403
2404 if (local_first)
2405 smc_link_save_peer_info(link, cclc, ini);
2406
2407 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2408 return SMC_CLC_DECL_ERR_RTOK;
2409
2410 if (local_first) {
2411 if (smc_ib_ready_link(link))
2412 return SMC_CLC_DECL_ERR_RDYLNK;
2413 /* QP confirmation over RoCE fabric */
2414 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2415 reason_code = smcr_serv_conf_first_link(new_smc);
2416 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2417 }
2418 return reason_code;
2419}
2420
2421/* setup for connection of server */
2422static void smc_listen_work(struct work_struct *work)
2423{
2424 struct smc_sock *new_smc = container_of(work, struct smc_sock,
2425 smc_listen_work);
2426 struct socket *newclcsock = new_smc->clcsock;
2427 struct smc_clc_msg_accept_confirm *cclc;
2428 struct smc_clc_msg_proposal_area *buf;
2429 struct smc_clc_msg_proposal *pclc;
2430 struct smc_init_info *ini = NULL;
2431 u8 proposal_version = SMC_V1;
2432 u8 accept_version;
2433 int rc = 0;
2434
2435 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2436 return smc_listen_out_err(new_smc);
2437
2438 if (new_smc->use_fallback) {
2439 smc_listen_out_connected(new_smc);
2440 return;
2441 }
2442
2443 /* check if peer is smc capable */
2444 if (!tcp_sk(newclcsock->sk)->syn_smc) {
2445 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2446 if (rc)
2447 smc_listen_out_err(new_smc);
2448 else
2449 smc_listen_out_connected(new_smc);
2450 return;
2451 }
2452
2453 /* do inband token exchange -
2454 * wait for and receive SMC Proposal CLC message
2455 */
2456 buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2457 if (!buf) {
2458 rc = SMC_CLC_DECL_MEM;
2459 goto out_decl;
2460 }
2461 pclc = (struct smc_clc_msg_proposal *)buf;
2462 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2463 SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2464 if (rc)
2465 goto out_decl;
2466
2467 if (pclc->hdr.version > SMC_V1)
2468 proposal_version = SMC_V2;
2469
2470 /* IPSec connections opt out of SMC optimizations */
2471 if (using_ipsec(new_smc)) {
2472 rc = SMC_CLC_DECL_IPSEC;
2473 goto out_decl;
2474 }
2475
2476 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2477 if (!ini) {
2478 rc = SMC_CLC_DECL_MEM;
2479 goto out_decl;
2480 }
2481
2482 /* initial version checking */
2483 rc = smc_listen_v2_check(new_smc, pclc, ini);
2484 if (rc)
2485 goto out_decl;
2486
2487 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini);
2488 if (rc)
2489 goto out_decl;
2490
2491 mutex_lock(&smc_server_lgr_pending);
2492 smc_close_init(new_smc);
2493 smc_rx_init(new_smc);
2494 smc_tx_init(new_smc);
2495
2496 /* determine ISM or RoCE device used for connection */
2497 rc = smc_listen_find_device(new_smc, pclc, ini);
2498 if (rc)
2499 goto out_unlock;
2500
2501 /* send SMC Accept CLC message */
2502 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2503 rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2504 accept_version, ini->negotiated_eid, ini);
2505 if (rc)
2506 goto out_unlock;
2507
2508 /* SMC-D does not need this lock any more */
2509 if (ini->is_smcd)
2510 mutex_unlock(&smc_server_lgr_pending);
2511
2512 /* receive SMC Confirm CLC message */
2513 memset(buf, 0, sizeof(*buf));
2514 cclc = (struct smc_clc_msg_accept_confirm *)buf;
2515 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2516 SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2517 if (rc) {
2518 if (!ini->is_smcd)
2519 goto out_unlock;
2520 goto out_decl;
2521 }
2522
2523 rc = smc_clc_v2x_features_confirm_check(cclc, ini);
2524 if (rc) {
2525 if (!ini->is_smcd)
2526 goto out_unlock;
2527 goto out_decl;
2528 }
2529
2530 /* fce smc release version is needed in smc_listen_rdma_finish,
2531 * so save fce info here.
2532 */
2533 smc_conn_save_peer_info_fce(new_smc, cclc);
2534
2535 /* finish worker */
2536 if (!ini->is_smcd) {
2537 rc = smc_listen_rdma_finish(new_smc, cclc,
2538 ini->first_contact_local, ini);
2539 if (rc)
2540 goto out_unlock;
2541 mutex_unlock(&smc_server_lgr_pending);
2542 }
2543 smc_conn_save_peer_info(new_smc, cclc);
2544 smc_listen_out_connected(new_smc);
2545 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2546 goto out_free;
2547
2548out_unlock:
2549 mutex_unlock(&smc_server_lgr_pending);
2550out_decl:
2551 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2552 proposal_version);
2553out_free:
2554 kfree(ini);
2555 kfree(buf);
2556}
2557
2558static void smc_tcp_listen_work(struct work_struct *work)
2559{
2560 struct smc_sock *lsmc = container_of(work, struct smc_sock,
2561 tcp_listen_work);
2562 struct sock *lsk = &lsmc->sk;
2563 struct smc_sock *new_smc;
2564 int rc = 0;
2565
2566 lock_sock(lsk);
2567 while (lsk->sk_state == SMC_LISTEN) {
2568 rc = smc_clcsock_accept(lsmc, &new_smc);
2569 if (rc) /* clcsock accept queue empty or error */
2570 goto out;
2571 if (!new_smc)
2572 continue;
2573
2574 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2575 atomic_inc(&lsmc->queued_smc_hs);
2576
2577 new_smc->listen_smc = lsmc;
2578 new_smc->use_fallback = lsmc->use_fallback;
2579 new_smc->fallback_rsn = lsmc->fallback_rsn;
2580 sock_hold(lsk); /* sock_put in smc_listen_work */
2581 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2582 smc_copy_sock_settings_to_smc(new_smc);
2583 sock_hold(&new_smc->sk); /* sock_put in passive closing */
2584 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2585 sock_put(&new_smc->sk);
2586 }
2587
2588out:
2589 release_sock(lsk);
2590 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2591}
2592
2593static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2594{
2595 struct smc_sock *lsmc;
2596
2597 read_lock_bh(&listen_clcsock->sk_callback_lock);
2598 lsmc = smc_clcsock_user_data(listen_clcsock);
2599 if (!lsmc)
2600 goto out;
2601 lsmc->clcsk_data_ready(listen_clcsock);
2602 if (lsmc->sk.sk_state == SMC_LISTEN) {
2603 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2604 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2605 sock_put(&lsmc->sk);
2606 }
2607out:
2608 read_unlock_bh(&listen_clcsock->sk_callback_lock);
2609}
2610
2611static int smc_listen(struct socket *sock, int backlog)
2612{
2613 struct sock *sk = sock->sk;
2614 struct smc_sock *smc;
2615 int rc;
2616
2617 smc = smc_sk(sk);
2618 lock_sock(sk);
2619
2620 rc = -EINVAL;
2621 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2622 smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2623 goto out;
2624
2625 rc = 0;
2626 if (sk->sk_state == SMC_LISTEN) {
2627 sk->sk_max_ack_backlog = backlog;
2628 goto out;
2629 }
2630 /* some socket options are handled in core, so we could not apply
2631 * them to the clc socket -- copy smc socket options to clc socket
2632 */
2633 smc_copy_sock_settings_to_clc(smc);
2634 if (!smc->use_fallback)
2635 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2636
2637 /* save original sk_data_ready function and establish
2638 * smc-specific sk_data_ready function
2639 */
2640 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2641 smc->clcsock->sk->sk_user_data =
2642 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
2643 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2644 smc_clcsock_data_ready, &smc->clcsk_data_ready);
2645 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2646
2647 /* save original ops */
2648 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2649
2650 smc->af_ops = *smc->ori_af_ops;
2651 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2652
2653 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2654
2655 if (smc->limit_smc_hs)
2656 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2657
2658 rc = kernel_listen(smc->clcsock, backlog);
2659 if (rc) {
2660 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2661 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2662 &smc->clcsk_data_ready);
2663 smc->clcsock->sk->sk_user_data = NULL;
2664 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2665 goto out;
2666 }
2667 sk->sk_max_ack_backlog = backlog;
2668 sk->sk_ack_backlog = 0;
2669 sk->sk_state = SMC_LISTEN;
2670
2671out:
2672 release_sock(sk);
2673 return rc;
2674}
2675
2676static int smc_accept(struct socket *sock, struct socket *new_sock,
2677 int flags, bool kern)
2678{
2679 struct sock *sk = sock->sk, *nsk;
2680 DECLARE_WAITQUEUE(wait, current);
2681 struct smc_sock *lsmc;
2682 long timeo;
2683 int rc = 0;
2684
2685 lsmc = smc_sk(sk);
2686 sock_hold(sk); /* sock_put below */
2687 lock_sock(sk);
2688
2689 if (lsmc->sk.sk_state != SMC_LISTEN) {
2690 rc = -EINVAL;
2691 release_sock(sk);
2692 goto out;
2693 }
2694
2695 /* Wait for an incoming connection */
2696 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
2697 add_wait_queue_exclusive(sk_sleep(sk), &wait);
2698 while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2699 set_current_state(TASK_INTERRUPTIBLE);
2700 if (!timeo) {
2701 rc = -EAGAIN;
2702 break;
2703 }
2704 release_sock(sk);
2705 timeo = schedule_timeout(timeo);
2706 /* wakeup by sk_data_ready in smc_listen_work() */
2707 sched_annotate_sleep();
2708 lock_sock(sk);
2709 if (signal_pending(current)) {
2710 rc = sock_intr_errno(timeo);
2711 break;
2712 }
2713 }
2714 set_current_state(TASK_RUNNING);
2715 remove_wait_queue(sk_sleep(sk), &wait);
2716
2717 if (!rc)
2718 rc = sock_error(nsk);
2719 release_sock(sk);
2720 if (rc)
2721 goto out;
2722
2723 if (lsmc->sockopt_defer_accept && !(flags & O_NONBLOCK)) {
2724 /* wait till data arrives on the socket */
2725 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
2726 MSEC_PER_SEC);
2727 if (smc_sk(nsk)->use_fallback) {
2728 struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2729
2730 lock_sock(clcsk);
2731 if (skb_queue_empty(&clcsk->sk_receive_queue))
2732 sk_wait_data(clcsk, &timeo, NULL);
2733 release_sock(clcsk);
2734 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2735 lock_sock(nsk);
2736 smc_rx_wait(smc_sk(nsk), &timeo, smc_rx_data_available);
2737 release_sock(nsk);
2738 }
2739 }
2740
2741out:
2742 sock_put(sk); /* sock_hold above */
2743 return rc;
2744}
2745
2746static int smc_getname(struct socket *sock, struct sockaddr *addr,
2747 int peer)
2748{
2749 struct smc_sock *smc;
2750
2751 if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2752 (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2753 return -ENOTCONN;
2754
2755 smc = smc_sk(sock->sk);
2756
2757 return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2758}
2759
2760static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2761{
2762 struct sock *sk = sock->sk;
2763 struct smc_sock *smc;
2764 int rc;
2765
2766 smc = smc_sk(sk);
2767 lock_sock(sk);
2768
2769 /* SMC does not support connect with fastopen */
2770 if (msg->msg_flags & MSG_FASTOPEN) {
2771 /* not connected yet, fallback */
2772 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2773 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2774 if (rc)
2775 goto out;
2776 } else {
2777 rc = -EINVAL;
2778 goto out;
2779 }
2780 } else if ((sk->sk_state != SMC_ACTIVE) &&
2781 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2782 (sk->sk_state != SMC_INIT)) {
2783 rc = -EPIPE;
2784 goto out;
2785 }
2786
2787 if (smc->use_fallback) {
2788 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2789 } else {
2790 rc = smc_tx_sendmsg(smc, msg, len);
2791 SMC_STAT_TX_PAYLOAD(smc, len, rc);
2792 }
2793out:
2794 release_sock(sk);
2795 return rc;
2796}
2797
2798static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2799 int flags)
2800{
2801 struct sock *sk = sock->sk;
2802 struct smc_sock *smc;
2803 int rc = -ENOTCONN;
2804
2805 smc = smc_sk(sk);
2806 lock_sock(sk);
2807 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2808 /* socket was connected before, no more data to read */
2809 rc = 0;
2810 goto out;
2811 }
2812 if ((sk->sk_state == SMC_INIT) ||
2813 (sk->sk_state == SMC_LISTEN) ||
2814 (sk->sk_state == SMC_CLOSED))
2815 goto out;
2816
2817 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2818 rc = 0;
2819 goto out;
2820 }
2821
2822 if (smc->use_fallback) {
2823 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2824 } else {
2825 msg->msg_namelen = 0;
2826 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2827 SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2828 }
2829
2830out:
2831 release_sock(sk);
2832 return rc;
2833}
2834
2835static __poll_t smc_accept_poll(struct sock *parent)
2836{
2837 struct smc_sock *isk = smc_sk(parent);
2838 __poll_t mask = 0;
2839
2840 spin_lock(&isk->accept_q_lock);
2841 if (!list_empty(&isk->accept_q))
2842 mask = EPOLLIN | EPOLLRDNORM;
2843 spin_unlock(&isk->accept_q_lock);
2844
2845 return mask;
2846}
2847
2848static __poll_t smc_poll(struct file *file, struct socket *sock,
2849 poll_table *wait)
2850{
2851 struct sock *sk = sock->sk;
2852 struct smc_sock *smc;
2853 __poll_t mask = 0;
2854
2855 if (!sk)
2856 return EPOLLNVAL;
2857
2858 smc = smc_sk(sock->sk);
2859 if (smc->use_fallback) {
2860 /* delegate to CLC child sock */
2861 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2862 sk->sk_err = smc->clcsock->sk->sk_err;
2863 } else {
2864 if (sk->sk_state != SMC_CLOSED)
2865 sock_poll_wait(file, sock, wait);
2866 if (sk->sk_err)
2867 mask |= EPOLLERR;
2868 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2869 (sk->sk_state == SMC_CLOSED))
2870 mask |= EPOLLHUP;
2871 if (sk->sk_state == SMC_LISTEN) {
2872 /* woken up by sk_data_ready in smc_listen_work() */
2873 mask |= smc_accept_poll(sk);
2874 } else if (smc->use_fallback) { /* as result of connect_work()*/
2875 mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2876 wait);
2877 sk->sk_err = smc->clcsock->sk->sk_err;
2878 } else {
2879 if ((sk->sk_state != SMC_INIT &&
2880 atomic_read(&smc->conn.sndbuf_space)) ||
2881 sk->sk_shutdown & SEND_SHUTDOWN) {
2882 mask |= EPOLLOUT | EPOLLWRNORM;
2883 } else {
2884 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2885 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2886 }
2887 if (atomic_read(&smc->conn.bytes_to_rcv))
2888 mask |= EPOLLIN | EPOLLRDNORM;
2889 if (sk->sk_shutdown & RCV_SHUTDOWN)
2890 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2891 if (sk->sk_state == SMC_APPCLOSEWAIT1)
2892 mask |= EPOLLIN;
2893 if (smc->conn.urg_state == SMC_URG_VALID)
2894 mask |= EPOLLPRI;
2895 }
2896 }
2897
2898 return mask;
2899}
2900
2901static int smc_shutdown(struct socket *sock, int how)
2902{
2903 struct sock *sk = sock->sk;
2904 bool do_shutdown = true;
2905 struct smc_sock *smc;
2906 int rc = -EINVAL;
2907 int old_state;
2908 int rc1 = 0;
2909
2910 smc = smc_sk(sk);
2911
2912 if ((how < SHUT_RD) || (how > SHUT_RDWR))
2913 return rc;
2914
2915 lock_sock(sk);
2916
2917 if (sock->state == SS_CONNECTING) {
2918 if (sk->sk_state == SMC_ACTIVE)
2919 sock->state = SS_CONNECTED;
2920 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2921 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2922 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2923 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2924 sk->sk_state == SMC_APPFINCLOSEWAIT)
2925 sock->state = SS_DISCONNECTING;
2926 }
2927
2928 rc = -ENOTCONN;
2929 if ((sk->sk_state != SMC_ACTIVE) &&
2930 (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2931 (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2932 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2933 (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2934 (sk->sk_state != SMC_APPFINCLOSEWAIT))
2935 goto out;
2936 if (smc->use_fallback) {
2937 rc = kernel_sock_shutdown(smc->clcsock, how);
2938 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2939 if (sk->sk_shutdown == SHUTDOWN_MASK) {
2940 sk->sk_state = SMC_CLOSED;
2941 sk->sk_socket->state = SS_UNCONNECTED;
2942 sock_put(sk);
2943 }
2944 goto out;
2945 }
2946 switch (how) {
2947 case SHUT_RDWR: /* shutdown in both directions */
2948 old_state = sk->sk_state;
2949 rc = smc_close_active(smc);
2950 if (old_state == SMC_ACTIVE &&
2951 sk->sk_state == SMC_PEERCLOSEWAIT1)
2952 do_shutdown = false;
2953 break;
2954 case SHUT_WR:
2955 rc = smc_close_shutdown_write(smc);
2956 break;
2957 case SHUT_RD:
2958 rc = 0;
2959 /* nothing more to do because peer is not involved */
2960 break;
2961 }
2962 if (do_shutdown && smc->clcsock)
2963 rc1 = kernel_sock_shutdown(smc->clcsock, how);
2964 /* map sock_shutdown_cmd constants to sk_shutdown value range */
2965 sk->sk_shutdown |= how + 1;
2966
2967 if (sk->sk_state == SMC_CLOSED)
2968 sock->state = SS_UNCONNECTED;
2969 else
2970 sock->state = SS_DISCONNECTING;
2971out:
2972 release_sock(sk);
2973 return rc ? rc : rc1;
2974}
2975
2976static int __smc_getsockopt(struct socket *sock, int level, int optname,
2977 char __user *optval, int __user *optlen)
2978{
2979 struct smc_sock *smc;
2980 int val, len;
2981
2982 smc = smc_sk(sock->sk);
2983
2984 if (get_user(len, optlen))
2985 return -EFAULT;
2986
2987 len = min_t(int, len, sizeof(int));
2988
2989 if (len < 0)
2990 return -EINVAL;
2991
2992 switch (optname) {
2993 case SMC_LIMIT_HS:
2994 val = smc->limit_smc_hs;
2995 break;
2996 default:
2997 return -EOPNOTSUPP;
2998 }
2999
3000 if (put_user(len, optlen))
3001 return -EFAULT;
3002 if (copy_to_user(optval, &val, len))
3003 return -EFAULT;
3004
3005 return 0;
3006}
3007
3008static int __smc_setsockopt(struct socket *sock, int level, int optname,
3009 sockptr_t optval, unsigned int optlen)
3010{
3011 struct sock *sk = sock->sk;
3012 struct smc_sock *smc;
3013 int val, rc;
3014
3015 smc = smc_sk(sk);
3016
3017 lock_sock(sk);
3018 switch (optname) {
3019 case SMC_LIMIT_HS:
3020 if (optlen < sizeof(int)) {
3021 rc = -EINVAL;
3022 break;
3023 }
3024 if (copy_from_sockptr(&val, optval, sizeof(int))) {
3025 rc = -EFAULT;
3026 break;
3027 }
3028
3029 smc->limit_smc_hs = !!val;
3030 rc = 0;
3031 break;
3032 default:
3033 rc = -EOPNOTSUPP;
3034 break;
3035 }
3036 release_sock(sk);
3037
3038 return rc;
3039}
3040
3041static int smc_setsockopt(struct socket *sock, int level, int optname,
3042 sockptr_t optval, unsigned int optlen)
3043{
3044 struct sock *sk = sock->sk;
3045 struct smc_sock *smc;
3046 int val, rc;
3047
3048 if (level == SOL_TCP && optname == TCP_ULP)
3049 return -EOPNOTSUPP;
3050 else if (level == SOL_SMC)
3051 return __smc_setsockopt(sock, level, optname, optval, optlen);
3052
3053 smc = smc_sk(sk);
3054
3055 /* generic setsockopts reaching us here always apply to the
3056 * CLC socket
3057 */
3058 mutex_lock(&smc->clcsock_release_lock);
3059 if (!smc->clcsock) {
3060 mutex_unlock(&smc->clcsock_release_lock);
3061 return -EBADF;
3062 }
3063 if (unlikely(!smc->clcsock->ops->setsockopt))
3064 rc = -EOPNOTSUPP;
3065 else
3066 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
3067 optval, optlen);
3068 if (smc->clcsock->sk->sk_err) {
3069 sk->sk_err = smc->clcsock->sk->sk_err;
3070 sk_error_report(sk);
3071 }
3072 mutex_unlock(&smc->clcsock_release_lock);
3073
3074 if (optlen < sizeof(int))
3075 return -EINVAL;
3076 if (copy_from_sockptr(&val, optval, sizeof(int)))
3077 return -EFAULT;
3078
3079 lock_sock(sk);
3080 if (rc || smc->use_fallback)
3081 goto out;
3082 switch (optname) {
3083 case TCP_FASTOPEN:
3084 case TCP_FASTOPEN_CONNECT:
3085 case TCP_FASTOPEN_KEY:
3086 case TCP_FASTOPEN_NO_COOKIE:
3087 /* option not supported by SMC */
3088 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3089 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3090 } else {
3091 rc = -EINVAL;
3092 }
3093 break;
3094 case TCP_NODELAY:
3095 if (sk->sk_state != SMC_INIT &&
3096 sk->sk_state != SMC_LISTEN &&
3097 sk->sk_state != SMC_CLOSED) {
3098 if (val) {
3099 SMC_STAT_INC(smc, ndly_cnt);
3100 smc_tx_pending(&smc->conn);
3101 cancel_delayed_work(&smc->conn.tx_work);
3102 }
3103 }
3104 break;
3105 case TCP_CORK:
3106 if (sk->sk_state != SMC_INIT &&
3107 sk->sk_state != SMC_LISTEN &&
3108 sk->sk_state != SMC_CLOSED) {
3109 if (!val) {
3110 SMC_STAT_INC(smc, cork_cnt);
3111 smc_tx_pending(&smc->conn);
3112 cancel_delayed_work(&smc->conn.tx_work);
3113 }
3114 }
3115 break;
3116 case TCP_DEFER_ACCEPT:
3117 smc->sockopt_defer_accept = val;
3118 break;
3119 default:
3120 break;
3121 }
3122out:
3123 release_sock(sk);
3124
3125 return rc;
3126}
3127
3128static int smc_getsockopt(struct socket *sock, int level, int optname,
3129 char __user *optval, int __user *optlen)
3130{
3131 struct smc_sock *smc;
3132 int rc;
3133
3134 if (level == SOL_SMC)
3135 return __smc_getsockopt(sock, level, optname, optval, optlen);
3136
3137 smc = smc_sk(sock->sk);
3138 mutex_lock(&smc->clcsock_release_lock);
3139 if (!smc->clcsock) {
3140 mutex_unlock(&smc->clcsock_release_lock);
3141 return -EBADF;
3142 }
3143 /* socket options apply to the CLC socket */
3144 if (unlikely(!smc->clcsock->ops->getsockopt)) {
3145 mutex_unlock(&smc->clcsock_release_lock);
3146 return -EOPNOTSUPP;
3147 }
3148 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3149 optval, optlen);
3150 mutex_unlock(&smc->clcsock_release_lock);
3151 return rc;
3152}
3153
3154static int smc_ioctl(struct socket *sock, unsigned int cmd,
3155 unsigned long arg)
3156{
3157 union smc_host_cursor cons, urg;
3158 struct smc_connection *conn;
3159 struct smc_sock *smc;
3160 int answ;
3161
3162 smc = smc_sk(sock->sk);
3163 conn = &smc->conn;
3164 lock_sock(&smc->sk);
3165 if (smc->use_fallback) {
3166 if (!smc->clcsock) {
3167 release_sock(&smc->sk);
3168 return -EBADF;
3169 }
3170 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3171 release_sock(&smc->sk);
3172 return answ;
3173 }
3174 switch (cmd) {
3175 case SIOCINQ: /* same as FIONREAD */
3176 if (smc->sk.sk_state == SMC_LISTEN) {
3177 release_sock(&smc->sk);
3178 return -EINVAL;
3179 }
3180 if (smc->sk.sk_state == SMC_INIT ||
3181 smc->sk.sk_state == SMC_CLOSED)
3182 answ = 0;
3183 else
3184 answ = atomic_read(&smc->conn.bytes_to_rcv);
3185 break;
3186 case SIOCOUTQ:
3187 /* output queue size (not send + not acked) */
3188 if (smc->sk.sk_state == SMC_LISTEN) {
3189 release_sock(&smc->sk);
3190 return -EINVAL;
3191 }
3192 if (smc->sk.sk_state == SMC_INIT ||
3193 smc->sk.sk_state == SMC_CLOSED)
3194 answ = 0;
3195 else
3196 answ = smc->conn.sndbuf_desc->len -
3197 atomic_read(&smc->conn.sndbuf_space);
3198 break;
3199 case SIOCOUTQNSD:
3200 /* output queue size (not send only) */
3201 if (smc->sk.sk_state == SMC_LISTEN) {
3202 release_sock(&smc->sk);
3203 return -EINVAL;
3204 }
3205 if (smc->sk.sk_state == SMC_INIT ||
3206 smc->sk.sk_state == SMC_CLOSED)
3207 answ = 0;
3208 else
3209 answ = smc_tx_prepared_sends(&smc->conn);
3210 break;
3211 case SIOCATMARK:
3212 if (smc->sk.sk_state == SMC_LISTEN) {
3213 release_sock(&smc->sk);
3214 return -EINVAL;
3215 }
3216 if (smc->sk.sk_state == SMC_INIT ||
3217 smc->sk.sk_state == SMC_CLOSED) {
3218 answ = 0;
3219 } else {
3220 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3221 smc_curs_copy(&urg, &conn->urg_curs, conn);
3222 answ = smc_curs_diff(conn->rmb_desc->len,
3223 &cons, &urg) == 1;
3224 }
3225 break;
3226 default:
3227 release_sock(&smc->sk);
3228 return -ENOIOCTLCMD;
3229 }
3230 release_sock(&smc->sk);
3231
3232 return put_user(answ, (int __user *)arg);
3233}
3234
3235/* Map the affected portions of the rmbe into an spd, note the number of bytes
3236 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3237 * updates till whenever a respective page has been fully processed.
3238 * Note that subsequent recv() calls have to wait till all splice() processing
3239 * completed.
3240 */
3241static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3242 struct pipe_inode_info *pipe, size_t len,
3243 unsigned int flags)
3244{
3245 struct sock *sk = sock->sk;
3246 struct smc_sock *smc;
3247 int rc = -ENOTCONN;
3248
3249 smc = smc_sk(sk);
3250 lock_sock(sk);
3251 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3252 /* socket was connected before, no more data to read */
3253 rc = 0;
3254 goto out;
3255 }
3256 if (sk->sk_state == SMC_INIT ||
3257 sk->sk_state == SMC_LISTEN ||
3258 sk->sk_state == SMC_CLOSED)
3259 goto out;
3260
3261 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3262 rc = 0;
3263 goto out;
3264 }
3265
3266 if (smc->use_fallback) {
3267 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3268 pipe, len, flags);
3269 } else {
3270 if (*ppos) {
3271 rc = -ESPIPE;
3272 goto out;
3273 }
3274 if (flags & SPLICE_F_NONBLOCK)
3275 flags = MSG_DONTWAIT;
3276 else
3277 flags = 0;
3278 SMC_STAT_INC(smc, splice_cnt);
3279 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3280 }
3281out:
3282 release_sock(sk);
3283
3284 return rc;
3285}
3286
3287/* must look like tcp */
3288static const struct proto_ops smc_sock_ops = {
3289 .family = PF_SMC,
3290 .owner = THIS_MODULE,
3291 .release = smc_release,
3292 .bind = smc_bind,
3293 .connect = smc_connect,
3294 .socketpair = sock_no_socketpair,
3295 .accept = smc_accept,
3296 .getname = smc_getname,
3297 .poll = smc_poll,
3298 .ioctl = smc_ioctl,
3299 .listen = smc_listen,
3300 .shutdown = smc_shutdown,
3301 .setsockopt = smc_setsockopt,
3302 .getsockopt = smc_getsockopt,
3303 .sendmsg = smc_sendmsg,
3304 .recvmsg = smc_recvmsg,
3305 .mmap = sock_no_mmap,
3306 .splice_read = smc_splice_read,
3307};
3308
3309static int __smc_create(struct net *net, struct socket *sock, int protocol,
3310 int kern, struct socket *clcsock)
3311{
3312 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3313 struct smc_sock *smc;
3314 struct sock *sk;
3315 int rc;
3316
3317 rc = -ESOCKTNOSUPPORT;
3318 if (sock->type != SOCK_STREAM)
3319 goto out;
3320
3321 rc = -EPROTONOSUPPORT;
3322 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3323 goto out;
3324
3325 rc = -ENOBUFS;
3326 sock->ops = &smc_sock_ops;
3327 sock->state = SS_UNCONNECTED;
3328 sk = smc_sock_alloc(net, sock, protocol);
3329 if (!sk)
3330 goto out;
3331
3332 /* create internal TCP socket for CLC handshake and fallback */
3333 smc = smc_sk(sk);
3334 smc->use_fallback = false; /* assume rdma capability first */
3335 smc->fallback_rsn = 0;
3336
3337 /* default behavior from limit_smc_hs in every net namespace */
3338 smc->limit_smc_hs = net->smc.limit_smc_hs;
3339
3340 rc = 0;
3341 if (!clcsock) {
3342 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3343 &smc->clcsock);
3344 if (rc) {
3345 sk_common_release(sk);
3346 goto out;
3347 }
3348
3349 /* smc_clcsock_release() does not wait smc->clcsock->sk's
3350 * destruction; its sk_state might not be TCP_CLOSE after
3351 * smc->sk is close()d, and TCP timers can be fired later,
3352 * which need net ref.
3353 */
3354 sk = smc->clcsock->sk;
3355 __netns_tracker_free(net, &sk->ns_tracker, false);
3356 sk->sk_net_refcnt = 1;
3357 get_net_track(net, &sk->ns_tracker, GFP_KERNEL);
3358 sock_inuse_add(net, 1);
3359 } else {
3360 smc->clcsock = clcsock;
3361 }
3362
3363out:
3364 return rc;
3365}
3366
3367static int smc_create(struct net *net, struct socket *sock, int protocol,
3368 int kern)
3369{
3370 return __smc_create(net, sock, protocol, kern, NULL);
3371}
3372
3373static const struct net_proto_family smc_sock_family_ops = {
3374 .family = PF_SMC,
3375 .owner = THIS_MODULE,
3376 .create = smc_create,
3377};
3378
3379static int smc_ulp_init(struct sock *sk)
3380{
3381 struct socket *tcp = sk->sk_socket;
3382 struct net *net = sock_net(sk);
3383 struct socket *smcsock;
3384 int protocol, ret;
3385
3386 /* only TCP can be replaced */
3387 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP ||
3388 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6))
3389 return -ESOCKTNOSUPPORT;
3390 /* don't handle wq now */
3391 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list)
3392 return -ENOTCONN;
3393
3394 if (sk->sk_family == AF_INET)
3395 protocol = SMCPROTO_SMC;
3396 else
3397 protocol = SMCPROTO_SMC6;
3398
3399 smcsock = sock_alloc();
3400 if (!smcsock)
3401 return -ENFILE;
3402
3403 smcsock->type = SOCK_STREAM;
3404 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */
3405 ret = __smc_create(net, smcsock, protocol, 1, tcp);
3406 if (ret) {
3407 sock_release(smcsock); /* module_put() which ops won't be NULL */
3408 return ret;
3409 }
3410
3411 /* replace tcp socket to smc */
3412 smcsock->file = tcp->file;
3413 smcsock->file->private_data = smcsock;
3414 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */
3415 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */
3416 tcp->file = NULL;
3417
3418 return ret;
3419}
3420
3421static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk,
3422 const gfp_t priority)
3423{
3424 struct inet_connection_sock *icsk = inet_csk(newsk);
3425
3426 /* don't inherit ulp ops to child when listen */
3427 icsk->icsk_ulp_ops = NULL;
3428}
3429
3430static struct tcp_ulp_ops smc_ulp_ops __read_mostly = {
3431 .name = "smc",
3432 .owner = THIS_MODULE,
3433 .init = smc_ulp_init,
3434 .clone = smc_ulp_clone,
3435};
3436
3437unsigned int smc_net_id;
3438
3439static __net_init int smc_net_init(struct net *net)
3440{
3441 int rc;
3442
3443 rc = smc_sysctl_net_init(net);
3444 if (rc)
3445 return rc;
3446 return smc_pnet_net_init(net);
3447}
3448
3449static void __net_exit smc_net_exit(struct net *net)
3450{
3451 smc_sysctl_net_exit(net);
3452 smc_pnet_net_exit(net);
3453}
3454
3455static __net_init int smc_net_stat_init(struct net *net)
3456{
3457 return smc_stats_init(net);
3458}
3459
3460static void __net_exit smc_net_stat_exit(struct net *net)
3461{
3462 smc_stats_exit(net);
3463}
3464
3465static struct pernet_operations smc_net_ops = {
3466 .init = smc_net_init,
3467 .exit = smc_net_exit,
3468 .id = &smc_net_id,
3469 .size = sizeof(struct smc_net),
3470};
3471
3472static struct pernet_operations smc_net_stat_ops = {
3473 .init = smc_net_stat_init,
3474 .exit = smc_net_stat_exit,
3475};
3476
3477static int __init smc_init(void)
3478{
3479 int rc;
3480
3481 rc = register_pernet_subsys(&smc_net_ops);
3482 if (rc)
3483 return rc;
3484
3485 rc = register_pernet_subsys(&smc_net_stat_ops);
3486 if (rc)
3487 goto out_pernet_subsys;
3488
3489 rc = smc_ism_init();
3490 if (rc)
3491 goto out_pernet_subsys_stat;
3492 smc_clc_init();
3493
3494 rc = smc_nl_init();
3495 if (rc)
3496 goto out_ism;
3497
3498 rc = smc_pnet_init();
3499 if (rc)
3500 goto out_nl;
3501
3502 rc = -ENOMEM;
3503
3504 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0);
3505 if (!smc_tcp_ls_wq)
3506 goto out_pnet;
3507
3508 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0);
3509 if (!smc_hs_wq)
3510 goto out_alloc_tcp_ls_wq;
3511
3512 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0);
3513 if (!smc_close_wq)
3514 goto out_alloc_hs_wq;
3515
3516 rc = smc_core_init();
3517 if (rc) {
3518 pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3519 goto out_alloc_wqs;
3520 }
3521
3522 rc = smc_llc_init();
3523 if (rc) {
3524 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3525 goto out_core;
3526 }
3527
3528 rc = smc_cdc_init();
3529 if (rc) {
3530 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3531 goto out_core;
3532 }
3533
3534 rc = proto_register(&smc_proto, 1);
3535 if (rc) {
3536 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3537 goto out_core;
3538 }
3539
3540 rc = proto_register(&smc_proto6, 1);
3541 if (rc) {
3542 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3543 goto out_proto;
3544 }
3545
3546 rc = sock_register(&smc_sock_family_ops);
3547 if (rc) {
3548 pr_err("%s: sock_register fails with %d\n", __func__, rc);
3549 goto out_proto6;
3550 }
3551 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3552 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3553
3554 rc = smc_ib_register_client();
3555 if (rc) {
3556 pr_err("%s: ib_register fails with %d\n", __func__, rc);
3557 goto out_sock;
3558 }
3559
3560 rc = tcp_register_ulp(&smc_ulp_ops);
3561 if (rc) {
3562 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc);
3563 goto out_ib;
3564 }
3565
3566 static_branch_enable(&tcp_have_smc);
3567 return 0;
3568
3569out_ib:
3570 smc_ib_unregister_client();
3571out_sock:
3572 sock_unregister(PF_SMC);
3573out_proto6:
3574 proto_unregister(&smc_proto6);
3575out_proto:
3576 proto_unregister(&smc_proto);
3577out_core:
3578 smc_core_exit();
3579out_alloc_wqs:
3580 destroy_workqueue(smc_close_wq);
3581out_alloc_hs_wq:
3582 destroy_workqueue(smc_hs_wq);
3583out_alloc_tcp_ls_wq:
3584 destroy_workqueue(smc_tcp_ls_wq);
3585out_pnet:
3586 smc_pnet_exit();
3587out_nl:
3588 smc_nl_exit();
3589out_ism:
3590 smc_clc_exit();
3591 smc_ism_exit();
3592out_pernet_subsys_stat:
3593 unregister_pernet_subsys(&smc_net_stat_ops);
3594out_pernet_subsys:
3595 unregister_pernet_subsys(&smc_net_ops);
3596
3597 return rc;
3598}
3599
3600static void __exit smc_exit(void)
3601{
3602 static_branch_disable(&tcp_have_smc);
3603 tcp_unregister_ulp(&smc_ulp_ops);
3604 sock_unregister(PF_SMC);
3605 smc_core_exit();
3606 smc_ib_unregister_client();
3607 smc_ism_exit();
3608 destroy_workqueue(smc_close_wq);
3609 destroy_workqueue(smc_tcp_ls_wq);
3610 destroy_workqueue(smc_hs_wq);
3611 proto_unregister(&smc_proto6);
3612 proto_unregister(&smc_proto);
3613 smc_pnet_exit();
3614 smc_nl_exit();
3615 smc_clc_exit();
3616 unregister_pernet_subsys(&smc_net_stat_ops);
3617 unregister_pernet_subsys(&smc_net_ops);
3618 rcu_barrier();
3619}
3620
3621module_init(smc_init);
3622module_exit(smc_exit);
3623
3624MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3625MODULE_DESCRIPTION("smc socket address family");
3626MODULE_LICENSE("GPL");
3627MODULE_ALIAS_NETPROTO(PF_SMC);
3628MODULE_ALIAS_TCP_ULP("smc");
3629MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);
1// SPDX-License-Identifier: GPL-2.0-only
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * AF_SMC protocol family socket handler keeping the AF_INET sock address type
6 * applies to SOCK_STREAM sockets only
7 * offers an alternative communication option for TCP-protocol sockets
8 * applicable with RoCE-cards only
9 *
10 * Initial restrictions:
11 * - support for alternate links postponed
12 *
13 * Copyright IBM Corp. 2016, 2018
14 *
15 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
16 * based on prototype from Frank Blaschka
17 */
18
19#define KMSG_COMPONENT "smc"
20#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
21
22#include <linux/module.h>
23#include <linux/socket.h>
24#include <linux/workqueue.h>
25#include <linux/in.h>
26#include <linux/sched/signal.h>
27#include <linux/if_vlan.h>
28#include <linux/rcupdate_wait.h>
29#include <linux/ctype.h>
30#include <linux/splice.h>
31
32#include <net/sock.h>
33#include <net/tcp.h>
34#include <net/smc.h>
35#include <asm/ioctls.h>
36
37#include <net/net_namespace.h>
38#include <net/netns/generic.h>
39#include "smc_netns.h"
40
41#include "smc.h"
42#include "smc_clc.h"
43#include "smc_llc.h"
44#include "smc_cdc.h"
45#include "smc_core.h"
46#include "smc_ib.h"
47#include "smc_ism.h"
48#include "smc_pnet.h"
49#include "smc_netlink.h"
50#include "smc_tx.h"
51#include "smc_rx.h"
52#include "smc_close.h"
53#include "smc_stats.h"
54#include "smc_tracepoint.h"
55#include "smc_sysctl.h"
56#include "smc_loopback.h"
57#include "smc_inet.h"
58
59static DEFINE_MUTEX(smc_server_lgr_pending); /* serialize link group
60 * creation on server
61 */
62static DEFINE_MUTEX(smc_client_lgr_pending); /* serialize link group
63 * creation on client
64 */
65
66static struct workqueue_struct *smc_tcp_ls_wq; /* wq for tcp listen work */
67struct workqueue_struct *smc_hs_wq; /* wq for handshake work */
68struct workqueue_struct *smc_close_wq; /* wq for close work */
69
70static void smc_tcp_listen_work(struct work_struct *);
71static void smc_connect_work(struct work_struct *);
72
73int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb)
74{
75 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
76 void *hdr;
77
78 if (cb_ctx->pos[0])
79 goto out;
80
81 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
82 &smc_gen_nl_family, NLM_F_MULTI,
83 SMC_NETLINK_DUMP_HS_LIMITATION);
84 if (!hdr)
85 return -ENOMEM;
86
87 if (nla_put_u8(skb, SMC_NLA_HS_LIMITATION_ENABLED,
88 sock_net(skb->sk)->smc.limit_smc_hs))
89 goto err;
90
91 genlmsg_end(skb, hdr);
92 cb_ctx->pos[0] = 1;
93out:
94 return skb->len;
95err:
96 genlmsg_cancel(skb, hdr);
97 return -EMSGSIZE;
98}
99
100int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
101{
102 sock_net(skb->sk)->smc.limit_smc_hs = true;
103 return 0;
104}
105
106int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info)
107{
108 sock_net(skb->sk)->smc.limit_smc_hs = false;
109 return 0;
110}
111
112static void smc_set_keepalive(struct sock *sk, int val)
113{
114 struct smc_sock *smc = smc_sk(sk);
115
116 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
117}
118
119static struct sock *smc_tcp_syn_recv_sock(const struct sock *sk,
120 struct sk_buff *skb,
121 struct request_sock *req,
122 struct dst_entry *dst,
123 struct request_sock *req_unhash,
124 bool *own_req)
125{
126 struct smc_sock *smc;
127 struct sock *child;
128
129 smc = smc_clcsock_user_data(sk);
130
131 if (READ_ONCE(sk->sk_ack_backlog) + atomic_read(&smc->queued_smc_hs) >
132 sk->sk_max_ack_backlog)
133 goto drop;
134
135 if (sk_acceptq_is_full(&smc->sk)) {
136 NET_INC_STATS(sock_net(sk), LINUX_MIB_LISTENOVERFLOWS);
137 goto drop;
138 }
139
140 /* passthrough to original syn recv sock fct */
141 child = smc->ori_af_ops->syn_recv_sock(sk, skb, req, dst, req_unhash,
142 own_req);
143 /* child must not inherit smc or its ops */
144 if (child) {
145 rcu_assign_sk_user_data(child, NULL);
146
147 /* v4-mapped sockets don't inherit parent ops. Don't restore. */
148 if (inet_csk(child)->icsk_af_ops == inet_csk(sk)->icsk_af_ops)
149 inet_csk(child)->icsk_af_ops = smc->ori_af_ops;
150 }
151 return child;
152
153drop:
154 dst_release(dst);
155 tcp_listendrop(sk);
156 return NULL;
157}
158
159static bool smc_hs_congested(const struct sock *sk)
160{
161 const struct smc_sock *smc;
162
163 smc = smc_clcsock_user_data(sk);
164
165 if (!smc)
166 return true;
167
168 if (workqueue_congested(WORK_CPU_UNBOUND, smc_hs_wq))
169 return true;
170
171 return false;
172}
173
174struct smc_hashinfo smc_v4_hashinfo = {
175 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
176};
177
178struct smc_hashinfo smc_v6_hashinfo = {
179 .lock = __RW_LOCK_UNLOCKED(smc_v6_hashinfo.lock),
180};
181
182int smc_hash_sk(struct sock *sk)
183{
184 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
185 struct hlist_head *head;
186
187 head = &h->ht;
188
189 write_lock_bh(&h->lock);
190 sk_add_node(sk, head);
191 write_unlock_bh(&h->lock);
192 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
193
194 return 0;
195}
196
197void smc_unhash_sk(struct sock *sk)
198{
199 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
200
201 write_lock_bh(&h->lock);
202 if (sk_del_node_init(sk))
203 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
204 write_unlock_bh(&h->lock);
205}
206
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
209 * BH context
210 */
211void smc_release_cb(struct sock *sk)
212{
213 struct smc_sock *smc = smc_sk(sk);
214
215 if (smc->conn.tx_in_release_sock) {
216 smc_tx_pending(&smc->conn);
217 smc->conn.tx_in_release_sock = false;
218 }
219}
220
221struct proto smc_proto = {
222 .name = "SMC",
223 .owner = THIS_MODULE,
224 .keepalive = smc_set_keepalive,
225 .hash = smc_hash_sk,
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,
231};
232EXPORT_SYMBOL_GPL(smc_proto);
233
234struct proto smc_proto6 = {
235 .name = "SMC6",
236 .owner = THIS_MODULE,
237 .keepalive = smc_set_keepalive,
238 .hash = smc_hash_sk,
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,
244};
245EXPORT_SYMBOL_GPL(smc_proto6);
246
247static void smc_fback_restore_callbacks(struct smc_sock *smc)
248{
249 struct sock *clcsk = smc->clcsock->sk;
250
251 write_lock_bh(&clcsk->sk_callback_lock);
252 clcsk->sk_user_data = NULL;
253
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);
258
259 write_unlock_bh(&clcsk->sk_callback_lock);
260}
261
262static void smc_restore_fallback_changes(struct smc_sock *smc)
263{
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);
268 }
269}
270
271static int __smc_release(struct smc_sock *smc)
272{
273 struct sock *sk = &smc->sk;
274 int rc = 0;
275
276 if (!smc->use_fallback) {
277 rc = smc_close_active(smc);
278 smc_sock_set_flag(sk, SOCK_DEAD);
279 sk->sk_shutdown |= SHUTDOWN_MASK;
280 } else {
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,
288 SHUT_RDWR);
289 }
290 sk->sk_state = SMC_CLOSED;
291 sk->sk_state_change(sk);
292 }
293 smc_restore_fallback_changes(smc);
294 }
295
296 sk->sk_prot->unhash(sk);
297
298 if (sk->sk_state == SMC_CLOSED) {
299 if (smc->clcsock) {
300 release_sock(sk);
301 smc_clcsock_release(smc);
302 lock_sock(sk);
303 }
304 if (!smc->use_fallback)
305 smc_conn_free(&smc->conn);
306 }
307
308 return rc;
309}
310
311int smc_release(struct socket *sock)
312{
313 struct sock *sk = sock->sk;
314 struct smc_sock *smc;
315 int old_state, rc = 0;
316
317 if (!sk)
318 goto out;
319
320 sock_hold(sk); /* sock_put below */
321 smc = smc_sk(sk);
322
323 old_state = sk->sk_state;
324
325 /* cleanup for a dangling non-blocking connect */
326 if (smc->connect_nonblock && old_state == SMC_INIT)
327 tcp_abort(smc->clcsock->sk, ECONNABORTED);
328
329 if (cancel_work_sync(&smc->connect_work))
330 sock_put(&smc->sk); /* sock_hold in smc_connect for passive closing */
331
332 if (sk->sk_state == SMC_LISTEN)
333 /* smc_close_non_accepted() is called and acquires
334 * sock lock for child sockets again
335 */
336 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
337 else
338 lock_sock(sk);
339
340 if (old_state == SMC_INIT && sk->sk_state == SMC_ACTIVE &&
341 !smc->use_fallback)
342 smc_close_active_abort(smc);
343
344 rc = __smc_release(smc);
345
346 /* detach socket */
347 sock_orphan(sk);
348 sock->sk = NULL;
349 release_sock(sk);
350
351 sock_put(sk); /* sock_hold above */
352 sock_put(sk); /* final sock_put */
353out:
354 return rc;
355}
356
357static void smc_destruct(struct sock *sk)
358{
359 if (sk->sk_state != SMC_CLOSED)
360 return;
361 if (!sock_flag(sk, SOCK_DEAD))
362 return;
363}
364
365void smc_sk_init(struct net *net, struct sock *sk, int protocol)
366{
367 struct smc_sock *smc = smc_sk(sk);
368
369 sk->sk_state = SMC_INIT;
370 sk->sk_destruct = smc_destruct;
371 sk->sk_protocol = protocol;
372 WRITE_ONCE(sk->sk_sndbuf, 2 * READ_ONCE(net->smc.sysctl_wmem));
373 WRITE_ONCE(sk->sk_rcvbuf, 2 * READ_ONCE(net->smc.sysctl_rmem));
374 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
375 INIT_WORK(&smc->connect_work, smc_connect_work);
376 INIT_DELAYED_WORK(&smc->conn.tx_work, smc_tx_work);
377 INIT_LIST_HEAD(&smc->accept_q);
378 spin_lock_init(&smc->accept_q_lock);
379 spin_lock_init(&smc->conn.send_lock);
380 sk->sk_prot->hash(sk);
381 mutex_init(&smc->clcsock_release_lock);
382 smc_init_saved_callbacks(smc);
383 smc->limit_smc_hs = net->smc.limit_smc_hs;
384 smc->use_fallback = false; /* assume rdma capability first */
385 smc->fallback_rsn = 0;
386 smc_close_init(smc);
387}
388
389static struct sock *smc_sock_alloc(struct net *net, struct socket *sock,
390 int protocol)
391{
392 struct proto *prot;
393 struct sock *sk;
394
395 prot = (protocol == SMCPROTO_SMC6) ? &smc_proto6 : &smc_proto;
396 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, prot, 0);
397 if (!sk)
398 return NULL;
399
400 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
401 smc_sk_init(net, sk, protocol);
402
403 return sk;
404}
405
406int smc_bind(struct socket *sock, struct sockaddr *uaddr,
407 int addr_len)
408{
409 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
410 struct sock *sk = sock->sk;
411 struct smc_sock *smc;
412 int rc;
413
414 smc = smc_sk(sk);
415
416 /* replicate tests from inet_bind(), to be safe wrt. future changes */
417 rc = -EINVAL;
418 if (addr_len < sizeof(struct sockaddr_in))
419 goto out;
420
421 rc = -EAFNOSUPPORT;
422 if (addr->sin_family != AF_INET &&
423 addr->sin_family != AF_INET6 &&
424 addr->sin_family != AF_UNSPEC)
425 goto out;
426 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
427 if (addr->sin_family == AF_UNSPEC &&
428 addr->sin_addr.s_addr != htonl(INADDR_ANY))
429 goto out;
430
431 lock_sock(sk);
432
433 /* Check if socket is already active */
434 rc = -EINVAL;
435 if (sk->sk_state != SMC_INIT || smc->connect_nonblock)
436 goto out_rel;
437
438 smc->clcsock->sk->sk_reuse = sk->sk_reuse;
439 smc->clcsock->sk->sk_reuseport = sk->sk_reuseport;
440 rc = kernel_bind(smc->clcsock, uaddr, addr_len);
441
442out_rel:
443 release_sock(sk);
444out:
445 return rc;
446}
447
448/* copy only relevant settings and flags of SOL_SOCKET level from smc to
449 * clc socket (since smc is not called for these options from net/core)
450 */
451
452#define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
453 (1UL << SOCK_KEEPOPEN) | \
454 (1UL << SOCK_LINGER) | \
455 (1UL << SOCK_BROADCAST) | \
456 (1UL << SOCK_TIMESTAMP) | \
457 (1UL << SOCK_DBG) | \
458 (1UL << SOCK_RCVTSTAMP) | \
459 (1UL << SOCK_RCVTSTAMPNS) | \
460 (1UL << SOCK_LOCALROUTE) | \
461 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
462 (1UL << SOCK_RXQ_OVFL) | \
463 (1UL << SOCK_WIFI_STATUS) | \
464 (1UL << SOCK_NOFCS) | \
465 (1UL << SOCK_FILTER_LOCKED) | \
466 (1UL << SOCK_TSTAMP_NEW))
467
468/* if set, use value set by setsockopt() - else use IPv4 or SMC sysctl value */
469static void smc_adjust_sock_bufsizes(struct sock *nsk, struct sock *osk,
470 unsigned long mask)
471{
472 nsk->sk_userlocks = osk->sk_userlocks;
473 if (osk->sk_userlocks & SOCK_SNDBUF_LOCK)
474 nsk->sk_sndbuf = osk->sk_sndbuf;
475 if (osk->sk_userlocks & SOCK_RCVBUF_LOCK)
476 nsk->sk_rcvbuf = osk->sk_rcvbuf;
477}
478
479static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
480 unsigned long mask)
481{
482 /* options we don't get control via setsockopt for */
483 nsk->sk_type = osk->sk_type;
484 nsk->sk_sndtimeo = osk->sk_sndtimeo;
485 nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
486 nsk->sk_mark = READ_ONCE(osk->sk_mark);
487 nsk->sk_priority = READ_ONCE(osk->sk_priority);
488 nsk->sk_rcvlowat = osk->sk_rcvlowat;
489 nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
490 nsk->sk_err = osk->sk_err;
491
492 nsk->sk_flags &= ~mask;
493 nsk->sk_flags |= osk->sk_flags & mask;
494
495 smc_adjust_sock_bufsizes(nsk, osk, mask);
496}
497
498static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
499{
500 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
501}
502
503#define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
504 (1UL << SOCK_KEEPOPEN) | \
505 (1UL << SOCK_LINGER) | \
506 (1UL << SOCK_DBG))
507/* copy only settings and flags relevant for smc from clc to smc socket */
508static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
509{
510 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
511}
512
513/* register the new vzalloced sndbuf on all links */
514static int smcr_lgr_reg_sndbufs(struct smc_link *link,
515 struct smc_buf_desc *snd_desc)
516{
517 struct smc_link_group *lgr = link->lgr;
518 int i, rc = 0;
519
520 if (!snd_desc->is_vm)
521 return -EINVAL;
522
523 /* protect against parallel smcr_link_reg_buf() */
524 down_write(&lgr->llc_conf_mutex);
525 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
526 if (!smc_link_active(&lgr->lnk[i]))
527 continue;
528 rc = smcr_link_reg_buf(&lgr->lnk[i], snd_desc);
529 if (rc)
530 break;
531 }
532 up_write(&lgr->llc_conf_mutex);
533 return rc;
534}
535
536/* register the new rmb on all links */
537static int smcr_lgr_reg_rmbs(struct smc_link *link,
538 struct smc_buf_desc *rmb_desc)
539{
540 struct smc_link_group *lgr = link->lgr;
541 bool do_slow = false;
542 int i, rc = 0;
543
544 rc = smc_llc_flow_initiate(lgr, SMC_LLC_FLOW_RKEY);
545 if (rc)
546 return rc;
547
548 down_read(&lgr->llc_conf_mutex);
549 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
550 if (!smc_link_active(&lgr->lnk[i]))
551 continue;
552 if (!rmb_desc->is_reg_mr[link->link_idx]) {
553 up_read(&lgr->llc_conf_mutex);
554 goto slow_path;
555 }
556 }
557 /* mr register already */
558 goto fast_path;
559slow_path:
560 do_slow = true;
561 /* protect against parallel smc_llc_cli_rkey_exchange() and
562 * parallel smcr_link_reg_buf()
563 */
564 down_write(&lgr->llc_conf_mutex);
565 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
566 if (!smc_link_active(&lgr->lnk[i]))
567 continue;
568 rc = smcr_link_reg_buf(&lgr->lnk[i], rmb_desc);
569 if (rc)
570 goto out;
571 }
572fast_path:
573 /* exchange confirm_rkey msg with peer */
574 rc = smc_llc_do_confirm_rkey(link, rmb_desc);
575 if (rc) {
576 rc = -EFAULT;
577 goto out;
578 }
579 rmb_desc->is_conf_rkey = true;
580out:
581 do_slow ? up_write(&lgr->llc_conf_mutex) : up_read(&lgr->llc_conf_mutex);
582 smc_llc_flow_stop(lgr, &lgr->llc_flow_lcl);
583 return rc;
584}
585
586static int smcr_clnt_conf_first_link(struct smc_sock *smc)
587{
588 struct smc_link *link = smc->conn.lnk;
589 struct smc_llc_qentry *qentry;
590 int rc;
591
592 /* Receive CONFIRM LINK request from server over RoCE fabric.
593 * Increasing the client's timeout by twice as much as the server's
594 * timeout by default can temporarily avoid decline messages of
595 * both sides crossing or colliding
596 */
597 qentry = smc_llc_wait(link->lgr, NULL, 2 * SMC_LLC_WAIT_TIME,
598 SMC_LLC_CONFIRM_LINK);
599 if (!qentry) {
600 struct smc_clc_msg_decline dclc;
601
602 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
603 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
604 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
605 }
606 smc_llc_save_peer_uid(qentry);
607 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_REQ);
608 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
609 if (rc)
610 return SMC_CLC_DECL_RMBE_EC;
611
612 rc = smc_ib_modify_qp_rts(link);
613 if (rc)
614 return SMC_CLC_DECL_ERR_RDYLNK;
615
616 smc_wr_remember_qp_attr(link);
617
618 /* reg the sndbuf if it was vzalloced */
619 if (smc->conn.sndbuf_desc->is_vm) {
620 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
621 return SMC_CLC_DECL_ERR_REGBUF;
622 }
623
624 /* reg the rmb */
625 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
626 return SMC_CLC_DECL_ERR_REGBUF;
627
628 /* confirm_rkey is implicit on 1st contact */
629 smc->conn.rmb_desc->is_conf_rkey = true;
630
631 /* send CONFIRM LINK response over RoCE fabric */
632 rc = smc_llc_send_confirm_link(link, SMC_LLC_RESP);
633 if (rc < 0)
634 return SMC_CLC_DECL_TIMEOUT_CL;
635
636 smc_llc_link_active(link);
637 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
638
639 if (link->lgr->max_links > 1) {
640 /* optional 2nd link, receive ADD LINK request from server */
641 qentry = smc_llc_wait(link->lgr, NULL, SMC_LLC_WAIT_TIME,
642 SMC_LLC_ADD_LINK);
643 if (!qentry) {
644 struct smc_clc_msg_decline dclc;
645
646 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
647 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
648 if (rc == -EAGAIN)
649 rc = 0; /* no DECLINE received, go with one link */
650 return rc;
651 }
652 smc_llc_flow_qentry_clr(&link->lgr->llc_flow_lcl);
653 smc_llc_cli_add_link(link, qentry);
654 }
655 return 0;
656}
657
658static bool smc_isascii(char *hostname)
659{
660 int i;
661
662 for (i = 0; i < SMC_MAX_HOSTNAME_LEN; i++)
663 if (!isascii(hostname[i]))
664 return false;
665 return true;
666}
667
668static void smc_conn_save_peer_info_fce(struct smc_sock *smc,
669 struct smc_clc_msg_accept_confirm *clc)
670{
671 struct smc_clc_first_contact_ext *fce;
672 int clc_v2_len;
673
674 if (clc->hdr.version == SMC_V1 ||
675 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
676 return;
677
678 if (smc->conn.lgr->is_smcd) {
679 memcpy(smc->conn.lgr->negotiated_eid, clc->d1.eid,
680 SMC_MAX_EID_LEN);
681 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, d1);
682 } else {
683 memcpy(smc->conn.lgr->negotiated_eid, clc->r1.eid,
684 SMC_MAX_EID_LEN);
685 clc_v2_len = offsetofend(struct smc_clc_msg_accept_confirm, r1);
686 }
687 fce = (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len);
688 smc->conn.lgr->peer_os = fce->os_type;
689 smc->conn.lgr->peer_smc_release = fce->release;
690 if (smc_isascii(fce->hostname))
691 memcpy(smc->conn.lgr->peer_hostname, fce->hostname,
692 SMC_MAX_HOSTNAME_LEN);
693}
694
695static void smcr_conn_save_peer_info(struct smc_sock *smc,
696 struct smc_clc_msg_accept_confirm *clc)
697{
698 int bufsize = smc_uncompress_bufsize(clc->r0.rmbe_size);
699
700 smc->conn.peer_rmbe_idx = clc->r0.rmbe_idx;
701 smc->conn.local_tx_ctrl.token = ntohl(clc->r0.rmbe_alert_token);
702 smc->conn.peer_rmbe_size = bufsize;
703 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
704 smc->conn.tx_off = bufsize * (smc->conn.peer_rmbe_idx - 1);
705}
706
707static void smcd_conn_save_peer_info(struct smc_sock *smc,
708 struct smc_clc_msg_accept_confirm *clc)
709{
710 int bufsize = smc_uncompress_bufsize(clc->d0.dmbe_size);
711
712 smc->conn.peer_rmbe_idx = clc->d0.dmbe_idx;
713 smc->conn.peer_token = ntohll(clc->d0.token);
714 /* msg header takes up space in the buffer */
715 smc->conn.peer_rmbe_size = bufsize - sizeof(struct smcd_cdc_msg);
716 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
717 smc->conn.tx_off = bufsize * smc->conn.peer_rmbe_idx;
718}
719
720static void smc_conn_save_peer_info(struct smc_sock *smc,
721 struct smc_clc_msg_accept_confirm *clc)
722{
723 if (smc->conn.lgr->is_smcd)
724 smcd_conn_save_peer_info(smc, clc);
725 else
726 smcr_conn_save_peer_info(smc, clc);
727 smc_conn_save_peer_info_fce(smc, clc);
728}
729
730static void smc_link_save_peer_info(struct smc_link *link,
731 struct smc_clc_msg_accept_confirm *clc,
732 struct smc_init_info *ini)
733{
734 link->peer_qpn = ntoh24(clc->r0.qpn);
735 memcpy(link->peer_gid, ini->peer_gid, SMC_GID_SIZE);
736 memcpy(link->peer_mac, ini->peer_mac, sizeof(link->peer_mac));
737 link->peer_psn = ntoh24(clc->r0.psn);
738 link->peer_mtu = clc->r0.qp_mtu;
739}
740
741static void smc_stat_inc_fback_rsn_cnt(struct smc_sock *smc,
742 struct smc_stats_fback *fback_arr)
743{
744 int cnt;
745
746 for (cnt = 0; cnt < SMC_MAX_FBACK_RSN_CNT; cnt++) {
747 if (fback_arr[cnt].fback_code == smc->fallback_rsn) {
748 fback_arr[cnt].count++;
749 break;
750 }
751 if (!fback_arr[cnt].fback_code) {
752 fback_arr[cnt].fback_code = smc->fallback_rsn;
753 fback_arr[cnt].count++;
754 break;
755 }
756 }
757}
758
759static void smc_stat_fallback(struct smc_sock *smc)
760{
761 struct net *net = sock_net(&smc->sk);
762
763 mutex_lock(&net->smc.mutex_fback_rsn);
764 if (smc->listen_smc) {
765 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->srv);
766 net->smc.fback_rsn->srv_fback_cnt++;
767 } else {
768 smc_stat_inc_fback_rsn_cnt(smc, net->smc.fback_rsn->clnt);
769 net->smc.fback_rsn->clnt_fback_cnt++;
770 }
771 mutex_unlock(&net->smc.mutex_fback_rsn);
772}
773
774/* must be called under rcu read lock */
775static void smc_fback_wakeup_waitqueue(struct smc_sock *smc, void *key)
776{
777 struct socket_wq *wq;
778 __poll_t flags;
779
780 wq = rcu_dereference(smc->sk.sk_wq);
781 if (!skwq_has_sleeper(wq))
782 return;
783
784 /* wake up smc sk->sk_wq */
785 if (!key) {
786 /* sk_state_change */
787 wake_up_interruptible_all(&wq->wait);
788 } else {
789 flags = key_to_poll(key);
790 if (flags & (EPOLLIN | EPOLLOUT))
791 /* sk_data_ready or sk_write_space */
792 wake_up_interruptible_sync_poll(&wq->wait, flags);
793 else if (flags & EPOLLERR)
794 /* sk_error_report */
795 wake_up_interruptible_poll(&wq->wait, flags);
796 }
797}
798
799static int smc_fback_mark_woken(wait_queue_entry_t *wait,
800 unsigned int mode, int sync, void *key)
801{
802 struct smc_mark_woken *mark =
803 container_of(wait, struct smc_mark_woken, wait_entry);
804
805 mark->woken = true;
806 mark->key = key;
807 return 0;
808}
809
810static void smc_fback_forward_wakeup(struct smc_sock *smc, struct sock *clcsk,
811 void (*clcsock_callback)(struct sock *sk))
812{
813 struct smc_mark_woken mark = { .woken = false };
814 struct socket_wq *wq;
815
816 init_waitqueue_func_entry(&mark.wait_entry,
817 smc_fback_mark_woken);
818 rcu_read_lock();
819 wq = rcu_dereference(clcsk->sk_wq);
820 if (!wq)
821 goto out;
822 add_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
823 clcsock_callback(clcsk);
824 remove_wait_queue(sk_sleep(clcsk), &mark.wait_entry);
825
826 if (mark.woken)
827 smc_fback_wakeup_waitqueue(smc, mark.key);
828out:
829 rcu_read_unlock();
830}
831
832static void smc_fback_state_change(struct sock *clcsk)
833{
834 struct smc_sock *smc;
835
836 read_lock_bh(&clcsk->sk_callback_lock);
837 smc = smc_clcsock_user_data(clcsk);
838 if (smc)
839 smc_fback_forward_wakeup(smc, clcsk,
840 smc->clcsk_state_change);
841 read_unlock_bh(&clcsk->sk_callback_lock);
842}
843
844static void smc_fback_data_ready(struct sock *clcsk)
845{
846 struct smc_sock *smc;
847
848 read_lock_bh(&clcsk->sk_callback_lock);
849 smc = smc_clcsock_user_data(clcsk);
850 if (smc)
851 smc_fback_forward_wakeup(smc, clcsk,
852 smc->clcsk_data_ready);
853 read_unlock_bh(&clcsk->sk_callback_lock);
854}
855
856static void smc_fback_write_space(struct sock *clcsk)
857{
858 struct smc_sock *smc;
859
860 read_lock_bh(&clcsk->sk_callback_lock);
861 smc = smc_clcsock_user_data(clcsk);
862 if (smc)
863 smc_fback_forward_wakeup(smc, clcsk,
864 smc->clcsk_write_space);
865 read_unlock_bh(&clcsk->sk_callback_lock);
866}
867
868static void smc_fback_error_report(struct sock *clcsk)
869{
870 struct smc_sock *smc;
871
872 read_lock_bh(&clcsk->sk_callback_lock);
873 smc = smc_clcsock_user_data(clcsk);
874 if (smc)
875 smc_fback_forward_wakeup(smc, clcsk,
876 smc->clcsk_error_report);
877 read_unlock_bh(&clcsk->sk_callback_lock);
878}
879
880static void smc_fback_replace_callbacks(struct smc_sock *smc)
881{
882 struct sock *clcsk = smc->clcsock->sk;
883
884 write_lock_bh(&clcsk->sk_callback_lock);
885 clcsk->sk_user_data = (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
886
887 smc_clcsock_replace_cb(&clcsk->sk_state_change, smc_fback_state_change,
888 &smc->clcsk_state_change);
889 smc_clcsock_replace_cb(&clcsk->sk_data_ready, smc_fback_data_ready,
890 &smc->clcsk_data_ready);
891 smc_clcsock_replace_cb(&clcsk->sk_write_space, smc_fback_write_space,
892 &smc->clcsk_write_space);
893 smc_clcsock_replace_cb(&clcsk->sk_error_report, smc_fback_error_report,
894 &smc->clcsk_error_report);
895
896 write_unlock_bh(&clcsk->sk_callback_lock);
897}
898
899static int smc_switch_to_fallback(struct smc_sock *smc, int reason_code)
900{
901 int rc = 0;
902
903 mutex_lock(&smc->clcsock_release_lock);
904 if (!smc->clcsock) {
905 rc = -EBADF;
906 goto out;
907 }
908
909 smc->use_fallback = true;
910 smc->fallback_rsn = reason_code;
911 smc_stat_fallback(smc);
912 trace_smc_switch_to_fallback(smc, reason_code);
913 if (smc->sk.sk_socket && smc->sk.sk_socket->file) {
914 smc->clcsock->file = smc->sk.sk_socket->file;
915 smc->clcsock->file->private_data = smc->clcsock;
916 smc->clcsock->wq.fasync_list =
917 smc->sk.sk_socket->wq.fasync_list;
918 smc->sk.sk_socket->wq.fasync_list = NULL;
919
920 /* There might be some wait entries remaining
921 * in smc sk->sk_wq and they should be woken up
922 * as clcsock's wait queue is woken up.
923 */
924 smc_fback_replace_callbacks(smc);
925 }
926out:
927 mutex_unlock(&smc->clcsock_release_lock);
928 return rc;
929}
930
931/* fall back during connect */
932static int smc_connect_fallback(struct smc_sock *smc, int reason_code)
933{
934 struct net *net = sock_net(&smc->sk);
935 int rc = 0;
936
937 rc = smc_switch_to_fallback(smc, reason_code);
938 if (rc) { /* fallback fails */
939 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
940 if (smc->sk.sk_state == SMC_INIT)
941 sock_put(&smc->sk); /* passive closing */
942 return rc;
943 }
944 smc_copy_sock_settings_to_clc(smc);
945 smc->connect_nonblock = 0;
946 if (smc->sk.sk_state == SMC_INIT)
947 smc->sk.sk_state = SMC_ACTIVE;
948 return 0;
949}
950
951/* decline and fall back during connect */
952static int smc_connect_decline_fallback(struct smc_sock *smc, int reason_code,
953 u8 version)
954{
955 struct net *net = sock_net(&smc->sk);
956 int rc;
957
958 if (reason_code < 0) { /* error, fallback is not possible */
959 this_cpu_inc(net->smc.smc_stats->clnt_hshake_err_cnt);
960 if (smc->sk.sk_state == SMC_INIT)
961 sock_put(&smc->sk); /* passive closing */
962 return reason_code;
963 }
964 if (reason_code != SMC_CLC_DECL_PEERDECL) {
965 rc = smc_clc_send_decline(smc, reason_code, version);
966 if (rc < 0) {
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 */
970 return rc;
971 }
972 }
973 return smc_connect_fallback(smc, reason_code);
974}
975
976static void smc_conn_abort(struct smc_sock *smc, int local_first)
977{
978 struct smc_connection *conn = &smc->conn;
979 struct smc_link_group *lgr = conn->lgr;
980 bool lgr_valid = false;
981
982 if (smc_conn_lgr_valid(conn))
983 lgr_valid = true;
984
985 smc_conn_free(conn);
986 if (local_first && lgr_valid)
987 smc_lgr_cleanup_early(lgr);
988}
989
990/* check if there is a rdma device available for this connection. */
991/* called for connect and listen */
992static int smc_find_rdma_device(struct smc_sock *smc, struct smc_init_info *ini)
993{
994 /* PNET table look up: search active ib_device and port
995 * within same PNETID that also contains the ethernet device
996 * used for the internal TCP socket
997 */
998 smc_pnet_find_roce_resource(smc->clcsock->sk, ini);
999 if (!ini->check_smcrv2 && !ini->ib_dev)
1000 return SMC_CLC_DECL_NOSMCRDEV;
1001 if (ini->check_smcrv2 && !ini->smcrv2.ib_dev_v2)
1002 return SMC_CLC_DECL_NOSMCRDEV;
1003 return 0;
1004}
1005
1006/* check if there is an ISM device available for this connection. */
1007/* called for connect and listen */
1008static int smc_find_ism_device(struct smc_sock *smc, struct smc_init_info *ini)
1009{
1010 /* Find ISM device with same PNETID as connecting interface */
1011 smc_pnet_find_ism_resource(smc->clcsock->sk, ini);
1012 if (!ini->ism_dev[0])
1013 return SMC_CLC_DECL_NOSMCDDEV;
1014 else
1015 ini->ism_chid[0] = smc_ism_get_chid(ini->ism_dev[0]);
1016 return 0;
1017}
1018
1019/* is chid unique for the ism devices that are already determined? */
1020static bool smc_find_ism_v2_is_unique_chid(u16 chid, struct smc_init_info *ini,
1021 int cnt)
1022{
1023 int i = (!ini->ism_dev[0]) ? 1 : 0;
1024
1025 for (; i < cnt; i++)
1026 if (ini->ism_chid[i] == chid)
1027 return false;
1028 return true;
1029}
1030
1031/* determine possible V2 ISM devices (either without PNETID or with PNETID plus
1032 * PNETID matching net_device)
1033 */
1034static int smc_find_ism_v2_device_clnt(struct smc_sock *smc,
1035 struct smc_init_info *ini)
1036{
1037 int rc = SMC_CLC_DECL_NOSMCDDEV;
1038 struct smcd_dev *smcd;
1039 int i = 1, entry = 1;
1040 bool is_emulated;
1041 u16 chid;
1042
1043 if (smcd_indicated(ini->smc_type_v1))
1044 rc = 0; /* already initialized for V1 */
1045 mutex_lock(&smcd_dev_list.mutex);
1046 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
1047 if (smcd->going_away || smcd == ini->ism_dev[0])
1048 continue;
1049 chid = smc_ism_get_chid(smcd);
1050 if (!smc_find_ism_v2_is_unique_chid(chid, ini, i))
1051 continue;
1052 is_emulated = __smc_ism_is_emulated(chid);
1053 if (!smc_pnet_is_pnetid_set(smcd->pnetid) ||
1054 smc_pnet_is_ndev_pnetid(sock_net(&smc->sk), smcd->pnetid)) {
1055 if (is_emulated && entry == SMCD_CLC_MAX_V2_GID_ENTRIES)
1056 /* It's the last GID-CHID entry left in CLC
1057 * Proposal SMC-Dv2 extension, but an Emulated-
1058 * ISM device will take two entries. So give
1059 * up it and try the next potential ISM device.
1060 */
1061 continue;
1062 ini->ism_dev[i] = smcd;
1063 ini->ism_chid[i] = chid;
1064 ini->is_smcd = true;
1065 rc = 0;
1066 i++;
1067 entry = is_emulated ? entry + 2 : entry + 1;
1068 if (entry > SMCD_CLC_MAX_V2_GID_ENTRIES)
1069 break;
1070 }
1071 }
1072 mutex_unlock(&smcd_dev_list.mutex);
1073 ini->ism_offered_cnt = i - 1;
1074 if (!ini->ism_dev[0] && !ini->ism_dev[1])
1075 ini->smcd_version = 0;
1076
1077 return rc;
1078}
1079
1080/* Check for VLAN ID and register it on ISM device just for CLC handshake */
1081static int smc_connect_ism_vlan_setup(struct smc_sock *smc,
1082 struct smc_init_info *ini)
1083{
1084 if (ini->vlan_id && smc_ism_get_vlan(ini->ism_dev[0], ini->vlan_id))
1085 return SMC_CLC_DECL_ISMVLANERR;
1086 return 0;
1087}
1088
1089static int smc_find_proposal_devices(struct smc_sock *smc,
1090 struct smc_init_info *ini)
1091{
1092 int rc = 0;
1093
1094 /* check if there is an ism device available */
1095 if (!(ini->smcd_version & SMC_V1) ||
1096 smc_find_ism_device(smc, ini) ||
1097 smc_connect_ism_vlan_setup(smc, ini))
1098 ini->smcd_version &= ~SMC_V1;
1099 /* else ISM V1 is supported for this connection */
1100
1101 /* check if there is an rdma device available */
1102 if (!(ini->smcr_version & SMC_V1) ||
1103 smc_find_rdma_device(smc, ini))
1104 ini->smcr_version &= ~SMC_V1;
1105 /* else RDMA is supported for this connection */
1106
1107 ini->smc_type_v1 = smc_indicated_type(ini->smcd_version & SMC_V1,
1108 ini->smcr_version & SMC_V1);
1109
1110 /* check if there is an ism v2 device available */
1111 if (!(ini->smcd_version & SMC_V2) ||
1112 !smc_ism_is_v2_capable() ||
1113 smc_find_ism_v2_device_clnt(smc, ini))
1114 ini->smcd_version &= ~SMC_V2;
1115
1116 /* check if there is an rdma v2 device available */
1117 ini->check_smcrv2 = true;
1118 ini->smcrv2.saddr = smc->clcsock->sk->sk_rcv_saddr;
1119 if (!(ini->smcr_version & SMC_V2) ||
1120 smc->clcsock->sk->sk_family != AF_INET ||
1121 !smc_clc_ueid_count() ||
1122 smc_find_rdma_device(smc, ini))
1123 ini->smcr_version &= ~SMC_V2;
1124 ini->check_smcrv2 = false;
1125
1126 ini->smc_type_v2 = smc_indicated_type(ini->smcd_version & SMC_V2,
1127 ini->smcr_version & SMC_V2);
1128
1129 /* if neither ISM nor RDMA are supported, fallback */
1130 if (ini->smc_type_v1 == SMC_TYPE_N && ini->smc_type_v2 == SMC_TYPE_N)
1131 rc = SMC_CLC_DECL_NOSMCDEV;
1132
1133 return rc;
1134}
1135
1136/* cleanup temporary VLAN ID registration used for CLC handshake. If ISM is
1137 * used, the VLAN ID will be registered again during the connection setup.
1138 */
1139static int smc_connect_ism_vlan_cleanup(struct smc_sock *smc,
1140 struct smc_init_info *ini)
1141{
1142 if (!smcd_indicated(ini->smc_type_v1))
1143 return 0;
1144 if (ini->vlan_id && smc_ism_put_vlan(ini->ism_dev[0], ini->vlan_id))
1145 return SMC_CLC_DECL_CNFERR;
1146 return 0;
1147}
1148
1149#define SMC_CLC_MAX_ACCEPT_LEN \
1150 (sizeof(struct smc_clc_msg_accept_confirm) + \
1151 sizeof(struct smc_clc_first_contact_ext_v2x) + \
1152 sizeof(struct smc_clc_msg_trail))
1153
1154/* CLC handshake during connect */
1155static int smc_connect_clc(struct smc_sock *smc,
1156 struct smc_clc_msg_accept_confirm *aclc,
1157 struct smc_init_info *ini)
1158{
1159 int rc = 0;
1160
1161 /* do inband token exchange */
1162 rc = smc_clc_send_proposal(smc, ini);
1163 if (rc)
1164 return rc;
1165 /* receive SMC Accept CLC message */
1166 return smc_clc_wait_msg(smc, aclc, SMC_CLC_MAX_ACCEPT_LEN,
1167 SMC_CLC_ACCEPT, CLC_WAIT_TIME);
1168}
1169
1170void smc_fill_gid_list(struct smc_link_group *lgr,
1171 struct smc_gidlist *gidlist,
1172 struct smc_ib_device *known_dev, u8 *known_gid)
1173{
1174 struct smc_init_info *alt_ini = NULL;
1175
1176 memset(gidlist, 0, sizeof(*gidlist));
1177 memcpy(gidlist->list[gidlist->len++], known_gid, SMC_GID_SIZE);
1178
1179 alt_ini = kzalloc(sizeof(*alt_ini), GFP_KERNEL);
1180 if (!alt_ini)
1181 goto out;
1182
1183 alt_ini->vlan_id = lgr->vlan_id;
1184 alt_ini->check_smcrv2 = true;
1185 alt_ini->smcrv2.saddr = lgr->saddr;
1186 smc_pnet_find_alt_roce(lgr, alt_ini, known_dev);
1187
1188 if (!alt_ini->smcrv2.ib_dev_v2)
1189 goto out;
1190
1191 memcpy(gidlist->list[gidlist->len++], alt_ini->smcrv2.ib_gid_v2,
1192 SMC_GID_SIZE);
1193
1194out:
1195 kfree(alt_ini);
1196}
1197
1198static int smc_connect_rdma_v2_prepare(struct smc_sock *smc,
1199 struct smc_clc_msg_accept_confirm *aclc,
1200 struct smc_init_info *ini)
1201{
1202 struct smc_clc_first_contact_ext *fce =
1203 smc_get_clc_first_contact_ext(aclc, false);
1204 struct net *net = sock_net(&smc->sk);
1205 int rc;
1206
1207 if (!ini->first_contact_peer || aclc->hdr.version == SMC_V1)
1208 return 0;
1209
1210 if (fce->v2_direct) {
1211 memcpy(ini->smcrv2.nexthop_mac, &aclc->r0.lcl.mac, ETH_ALEN);
1212 ini->smcrv2.uses_gateway = false;
1213 } else {
1214 if (smc_ib_find_route(net, smc->clcsock->sk->sk_rcv_saddr,
1215 smc_ib_gid_to_ipv4(aclc->r0.lcl.gid),
1216 ini->smcrv2.nexthop_mac,
1217 &ini->smcrv2.uses_gateway))
1218 return SMC_CLC_DECL_NOROUTE;
1219 if (!ini->smcrv2.uses_gateway) {
1220 /* mismatch: peer claims indirect, but its direct */
1221 return SMC_CLC_DECL_NOINDIRECT;
1222 }
1223 }
1224
1225 ini->release_nr = fce->release;
1226 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1227 if (rc)
1228 return rc;
1229
1230 return 0;
1231}
1232
1233/* setup for RDMA connection of client */
1234static int smc_connect_rdma(struct smc_sock *smc,
1235 struct smc_clc_msg_accept_confirm *aclc,
1236 struct smc_init_info *ini)
1237{
1238 int i, reason_code = 0;
1239 struct smc_link *link;
1240 u8 *eid = NULL;
1241
1242 ini->is_smcd = false;
1243 ini->ib_clcqpn = ntoh24(aclc->r0.qpn);
1244 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1245 memcpy(ini->peer_systemid, aclc->r0.lcl.id_for_peer, SMC_SYSTEMID_LEN);
1246 memcpy(ini->peer_gid, aclc->r0.lcl.gid, SMC_GID_SIZE);
1247 memcpy(ini->peer_mac, aclc->r0.lcl.mac, ETH_ALEN);
1248 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1249 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1250
1251 reason_code = smc_connect_rdma_v2_prepare(smc, aclc, ini);
1252 if (reason_code)
1253 return reason_code;
1254
1255 mutex_lock(&smc_client_lgr_pending);
1256 reason_code = smc_conn_create(smc, ini);
1257 if (reason_code) {
1258 mutex_unlock(&smc_client_lgr_pending);
1259 return reason_code;
1260 }
1261
1262 smc_conn_save_peer_info(smc, aclc);
1263
1264 if (ini->first_contact_local) {
1265 link = smc->conn.lnk;
1266 } else {
1267 /* set link that was assigned by server */
1268 link = NULL;
1269 for (i = 0; i < SMC_LINKS_PER_LGR_MAX; i++) {
1270 struct smc_link *l = &smc->conn.lgr->lnk[i];
1271
1272 if (l->peer_qpn == ntoh24(aclc->r0.qpn) &&
1273 !memcmp(l->peer_gid, &aclc->r0.lcl.gid,
1274 SMC_GID_SIZE) &&
1275 (aclc->hdr.version > SMC_V1 ||
1276 !memcmp(l->peer_mac, &aclc->r0.lcl.mac,
1277 sizeof(l->peer_mac)))) {
1278 link = l;
1279 break;
1280 }
1281 }
1282 if (!link) {
1283 reason_code = SMC_CLC_DECL_NOSRVLINK;
1284 goto connect_abort;
1285 }
1286 smc_switch_link_and_count(&smc->conn, link);
1287 }
1288
1289 /* create send buffer and rmb */
1290 if (smc_buf_create(smc, false)) {
1291 reason_code = SMC_CLC_DECL_MEM;
1292 goto connect_abort;
1293 }
1294
1295 if (ini->first_contact_local)
1296 smc_link_save_peer_info(link, aclc, ini);
1297
1298 if (smc_rmb_rtoken_handling(&smc->conn, link, aclc)) {
1299 reason_code = SMC_CLC_DECL_ERR_RTOK;
1300 goto connect_abort;
1301 }
1302
1303 smc_rx_init(smc);
1304
1305 if (ini->first_contact_local) {
1306 if (smc_ib_ready_link(link)) {
1307 reason_code = SMC_CLC_DECL_ERR_RDYLNK;
1308 goto connect_abort;
1309 }
1310 } else {
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;
1315 goto connect_abort;
1316 }
1317 }
1318 if (smcr_lgr_reg_rmbs(link, smc->conn.rmb_desc)) {
1319 reason_code = SMC_CLC_DECL_ERR_REGBUF;
1320 goto connect_abort;
1321 }
1322 }
1323
1324 if (aclc->hdr.version > SMC_V1) {
1325 eid = aclc->r1.eid;
1326 if (ini->first_contact_local)
1327 smc_fill_gid_list(link->lgr, &ini->smcrv2.gidlist,
1328 link->smcibdev, link->gid);
1329 }
1330
1331 reason_code = smc_clc_send_confirm(smc, ini->first_contact_local,
1332 aclc->hdr.version, eid, ini);
1333 if (reason_code)
1334 goto connect_abort;
1335
1336 smc_tx_init(smc);
1337
1338 if (ini->first_contact_local) {
1339 /* QP confirmation over RoCE fabric */
1340 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
1341 reason_code = smcr_clnt_conf_first_link(smc);
1342 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
1343 if (reason_code)
1344 goto connect_abort;
1345 }
1346 mutex_unlock(&smc_client_lgr_pending);
1347
1348 smc_copy_sock_settings_to_clc(smc);
1349 smc->connect_nonblock = 0;
1350 if (smc->sk.sk_state == SMC_INIT)
1351 smc->sk.sk_state = SMC_ACTIVE;
1352
1353 return 0;
1354connect_abort:
1355 smc_conn_abort(smc, ini->first_contact_local);
1356 mutex_unlock(&smc_client_lgr_pending);
1357 smc->connect_nonblock = 0;
1358
1359 return reason_code;
1360}
1361
1362/* The server has chosen one of the proposed ISM devices for the communication.
1363 * Determine from the CHID of the received CLC ACCEPT the ISM device chosen.
1364 */
1365static int
1366smc_v2_determine_accepted_chid(struct smc_clc_msg_accept_confirm *aclc,
1367 struct smc_init_info *ini)
1368{
1369 int i;
1370
1371 for (i = 0; i < ini->ism_offered_cnt + 1; i++) {
1372 if (ini->ism_chid[i] == ntohs(aclc->d1.chid)) {
1373 ini->ism_selected = i;
1374 return 0;
1375 }
1376 }
1377
1378 return -EPROTO;
1379}
1380
1381/* setup for ISM connection of client */
1382static int smc_connect_ism(struct smc_sock *smc,
1383 struct smc_clc_msg_accept_confirm *aclc,
1384 struct smc_init_info *ini)
1385{
1386 u8 *eid = NULL;
1387 int rc = 0;
1388
1389 ini->is_smcd = true;
1390 ini->first_contact_peer = aclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK;
1391
1392 if (aclc->hdr.version == SMC_V2) {
1393 if (ini->first_contact_peer) {
1394 struct smc_clc_first_contact_ext *fce =
1395 smc_get_clc_first_contact_ext(aclc, true);
1396
1397 ini->release_nr = fce->release;
1398 rc = smc_clc_clnt_v2x_features_validate(fce, ini);
1399 if (rc)
1400 return rc;
1401 }
1402
1403 rc = smc_v2_determine_accepted_chid(aclc, ini);
1404 if (rc)
1405 return rc;
1406
1407 if (__smc_ism_is_emulated(ini->ism_chid[ini->ism_selected]))
1408 ini->ism_peer_gid[ini->ism_selected].gid_ext =
1409 ntohll(aclc->d1.gid_ext);
1410 /* for non-Emulated-ISM devices, peer gid_ext remains 0. */
1411 }
1412 ini->ism_peer_gid[ini->ism_selected].gid = ntohll(aclc->d0.gid);
1413
1414 /* there is only one lgr role for SMC-D; use server lock */
1415 mutex_lock(&smc_server_lgr_pending);
1416 rc = smc_conn_create(smc, ini);
1417 if (rc) {
1418 mutex_unlock(&smc_server_lgr_pending);
1419 return rc;
1420 }
1421
1422 /* Create send and receive buffers */
1423 rc = smc_buf_create(smc, true);
1424 if (rc) {
1425 rc = (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB : SMC_CLC_DECL_MEM;
1426 goto connect_abort;
1427 }
1428
1429 smc_conn_save_peer_info(smc, aclc);
1430
1431 if (smc_ism_support_dmb_nocopy(smc->conn.lgr->smcd)) {
1432 rc = smcd_buf_attach(smc);
1433 if (rc) {
1434 rc = SMC_CLC_DECL_MEM; /* try to fallback */
1435 goto connect_abort;
1436 }
1437 }
1438 smc_rx_init(smc);
1439 smc_tx_init(smc);
1440
1441 if (aclc->hdr.version > SMC_V1)
1442 eid = aclc->d1.eid;
1443
1444 rc = smc_clc_send_confirm(smc, ini->first_contact_local,
1445 aclc->hdr.version, eid, ini);
1446 if (rc)
1447 goto connect_abort;
1448 mutex_unlock(&smc_server_lgr_pending);
1449
1450 smc_copy_sock_settings_to_clc(smc);
1451 smc->connect_nonblock = 0;
1452 if (smc->sk.sk_state == SMC_INIT)
1453 smc->sk.sk_state = SMC_ACTIVE;
1454
1455 return 0;
1456connect_abort:
1457 smc_conn_abort(smc, ini->first_contact_local);
1458 mutex_unlock(&smc_server_lgr_pending);
1459 smc->connect_nonblock = 0;
1460
1461 return rc;
1462}
1463
1464/* check if received accept type and version matches a proposed one */
1465static int smc_connect_check_aclc(struct smc_init_info *ini,
1466 struct smc_clc_msg_accept_confirm *aclc)
1467{
1468 if (aclc->hdr.version >= SMC_V2) {
1469 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1470 !smcr_indicated(ini->smc_type_v2)) ||
1471 (aclc->hdr.typev1 == SMC_TYPE_D &&
1472 !smcd_indicated(ini->smc_type_v2)))
1473 return SMC_CLC_DECL_MODEUNSUPP;
1474 } else {
1475 if ((aclc->hdr.typev1 == SMC_TYPE_R &&
1476 !smcr_indicated(ini->smc_type_v1)) ||
1477 (aclc->hdr.typev1 == SMC_TYPE_D &&
1478 !smcd_indicated(ini->smc_type_v1)))
1479 return SMC_CLC_DECL_MODEUNSUPP;
1480 }
1481
1482 return 0;
1483}
1484
1485/* perform steps before actually connecting */
1486static int __smc_connect(struct smc_sock *smc)
1487{
1488 u8 version = smc_ism_is_v2_capable() ? SMC_V2 : SMC_V1;
1489 struct smc_clc_msg_accept_confirm *aclc;
1490 struct smc_init_info *ini = NULL;
1491 u8 *buf = NULL;
1492 int rc = 0;
1493
1494 if (smc->use_fallback)
1495 return smc_connect_fallback(smc, smc->fallback_rsn);
1496
1497 /* if peer has not signalled SMC-capability, fall back */
1498 if (!tcp_sk(smc->clcsock->sk)->syn_smc)
1499 return smc_connect_fallback(smc, SMC_CLC_DECL_PEERNOSMC);
1500
1501 /* IPSec connections opt out of SMC optimizations */
1502 if (using_ipsec(smc))
1503 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_IPSEC,
1504 version);
1505
1506 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
1507 if (!ini)
1508 return smc_connect_decline_fallback(smc, SMC_CLC_DECL_MEM,
1509 version);
1510
1511 ini->smcd_version = SMC_V1 | SMC_V2;
1512 ini->smcr_version = SMC_V1 | SMC_V2;
1513 ini->smc_type_v1 = SMC_TYPE_B;
1514 ini->smc_type_v2 = SMC_TYPE_B;
1515
1516 /* get vlan id from IP device */
1517 if (smc_vlan_by_tcpsk(smc->clcsock, ini)) {
1518 ini->smcd_version &= ~SMC_V1;
1519 ini->smcr_version = 0;
1520 ini->smc_type_v1 = SMC_TYPE_N;
1521 }
1522
1523 rc = smc_find_proposal_devices(smc, ini);
1524 if (rc)
1525 goto fallback;
1526
1527 buf = kzalloc(SMC_CLC_MAX_ACCEPT_LEN, GFP_KERNEL);
1528 if (!buf) {
1529 rc = SMC_CLC_DECL_MEM;
1530 goto fallback;
1531 }
1532 aclc = (struct smc_clc_msg_accept_confirm *)buf;
1533
1534 /* perform CLC handshake */
1535 rc = smc_connect_clc(smc, aclc, ini);
1536 if (rc) {
1537 /* -EAGAIN on timeout, see tcp_recvmsg() */
1538 if (rc == -EAGAIN) {
1539 rc = -ETIMEDOUT;
1540 smc->sk.sk_err = ETIMEDOUT;
1541 }
1542 goto vlan_cleanup;
1543 }
1544
1545 /* check if smc modes and versions of CLC proposal and accept match */
1546 rc = smc_connect_check_aclc(ini, aclc);
1547 version = aclc->hdr.version == SMC_V1 ? SMC_V1 : SMC_V2;
1548 if (rc)
1549 goto vlan_cleanup;
1550
1551 /* depending on previous steps, connect using rdma or ism */
1552 if (aclc->hdr.typev1 == SMC_TYPE_R) {
1553 ini->smcr_version = version;
1554 rc = smc_connect_rdma(smc, aclc, ini);
1555 } else if (aclc->hdr.typev1 == SMC_TYPE_D) {
1556 ini->smcd_version = version;
1557 rc = smc_connect_ism(smc, aclc, ini);
1558 }
1559 if (rc)
1560 goto vlan_cleanup;
1561
1562 SMC_STAT_CLNT_SUCC_INC(sock_net(smc->clcsock->sk), aclc);
1563 smc_connect_ism_vlan_cleanup(smc, ini);
1564 kfree(buf);
1565 kfree(ini);
1566 return 0;
1567
1568vlan_cleanup:
1569 smc_connect_ism_vlan_cleanup(smc, ini);
1570 kfree(buf);
1571fallback:
1572 kfree(ini);
1573 return smc_connect_decline_fallback(smc, rc, version);
1574}
1575
1576static void smc_connect_work(struct work_struct *work)
1577{
1578 struct smc_sock *smc = container_of(work, struct smc_sock,
1579 connect_work);
1580 long timeo = smc->sk.sk_sndtimeo;
1581 int rc = 0;
1582
1583 if (!timeo)
1584 timeo = MAX_SCHEDULE_TIMEOUT;
1585 lock_sock(smc->clcsock->sk);
1586 if (smc->clcsock->sk->sk_err) {
1587 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1588 } else if ((1 << smc->clcsock->sk->sk_state) &
1589 (TCPF_SYN_SENT | TCPF_SYN_RECV)) {
1590 rc = sk_stream_wait_connect(smc->clcsock->sk, &timeo);
1591 if ((rc == -EPIPE) &&
1592 ((1 << smc->clcsock->sk->sk_state) &
1593 (TCPF_ESTABLISHED | TCPF_CLOSE_WAIT)))
1594 rc = 0;
1595 }
1596 release_sock(smc->clcsock->sk);
1597 lock_sock(&smc->sk);
1598 if (rc != 0 || smc->sk.sk_err) {
1599 smc->sk.sk_state = SMC_CLOSED;
1600 if (rc == -EPIPE || rc == -EAGAIN)
1601 smc->sk.sk_err = EPIPE;
1602 else if (rc == -ECONNREFUSED)
1603 smc->sk.sk_err = ECONNREFUSED;
1604 else if (signal_pending(current))
1605 smc->sk.sk_err = -sock_intr_errno(timeo);
1606 sock_put(&smc->sk); /* passive closing */
1607 goto out;
1608 }
1609
1610 rc = __smc_connect(smc);
1611 if (rc < 0)
1612 smc->sk.sk_err = -rc;
1613
1614out:
1615 if (!sock_flag(&smc->sk, SOCK_DEAD)) {
1616 if (smc->sk.sk_err) {
1617 smc->sk.sk_state_change(&smc->sk);
1618 } else { /* allow polling before and after fallback decision */
1619 smc->clcsock->sk->sk_write_space(smc->clcsock->sk);
1620 smc->sk.sk_write_space(&smc->sk);
1621 }
1622 }
1623 release_sock(&smc->sk);
1624}
1625
1626int smc_connect(struct socket *sock, struct sockaddr *addr,
1627 int alen, int flags)
1628{
1629 struct sock *sk = sock->sk;
1630 struct smc_sock *smc;
1631 int rc = -EINVAL;
1632
1633 smc = smc_sk(sk);
1634
1635 /* separate smc parameter checking to be safe */
1636 if (alen < sizeof(addr->sa_family))
1637 goto out_err;
1638 if (addr->sa_family != AF_INET && addr->sa_family != AF_INET6)
1639 goto out_err;
1640
1641 lock_sock(sk);
1642 switch (sock->state) {
1643 default:
1644 rc = -EINVAL;
1645 goto out;
1646 case SS_CONNECTED:
1647 rc = sk->sk_state == SMC_ACTIVE ? -EISCONN : -EINVAL;
1648 goto out;
1649 case SS_CONNECTING:
1650 if (sk->sk_state == SMC_ACTIVE)
1651 goto connected;
1652 break;
1653 case SS_UNCONNECTED:
1654 sock->state = SS_CONNECTING;
1655 break;
1656 }
1657
1658 switch (sk->sk_state) {
1659 default:
1660 goto out;
1661 case SMC_CLOSED:
1662 rc = sock_error(sk) ? : -ECONNABORTED;
1663 sock->state = SS_UNCONNECTED;
1664 goto out;
1665 case SMC_ACTIVE:
1666 rc = -EISCONN;
1667 goto out;
1668 case SMC_INIT:
1669 break;
1670 }
1671
1672 smc_copy_sock_settings_to_clc(smc);
1673 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
1674 if (smc->connect_nonblock) {
1675 rc = -EALREADY;
1676 goto out;
1677 }
1678 rc = kernel_connect(smc->clcsock, addr, alen, flags);
1679 if (rc && rc != -EINPROGRESS)
1680 goto out;
1681
1682 if (smc->use_fallback) {
1683 sock->state = rc ? SS_CONNECTING : SS_CONNECTED;
1684 goto out;
1685 }
1686 sock_hold(&smc->sk); /* sock put in passive closing */
1687 if (flags & O_NONBLOCK) {
1688 if (queue_work(smc_hs_wq, &smc->connect_work))
1689 smc->connect_nonblock = 1;
1690 rc = -EINPROGRESS;
1691 goto out;
1692 } else {
1693 rc = __smc_connect(smc);
1694 if (rc < 0)
1695 goto out;
1696 }
1697
1698connected:
1699 rc = 0;
1700 sock->state = SS_CONNECTED;
1701out:
1702 release_sock(sk);
1703out_err:
1704 return rc;
1705}
1706
1707static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
1708{
1709 struct socket *new_clcsock = NULL;
1710 struct sock *lsk = &lsmc->sk;
1711 struct sock *new_sk;
1712 int rc = -EINVAL;
1713
1714 release_sock(lsk);
1715 new_sk = smc_sock_alloc(sock_net(lsk), NULL, lsk->sk_protocol);
1716 if (!new_sk) {
1717 rc = -ENOMEM;
1718 lsk->sk_err = ENOMEM;
1719 *new_smc = NULL;
1720 lock_sock(lsk);
1721 goto out;
1722 }
1723 *new_smc = smc_sk(new_sk);
1724
1725 mutex_lock(&lsmc->clcsock_release_lock);
1726 if (lsmc->clcsock)
1727 rc = kernel_accept(lsmc->clcsock, &new_clcsock, SOCK_NONBLOCK);
1728 mutex_unlock(&lsmc->clcsock_release_lock);
1729 lock_sock(lsk);
1730 if (rc < 0 && rc != -EAGAIN)
1731 lsk->sk_err = -rc;
1732 if (rc < 0 || lsk->sk_state == SMC_CLOSED) {
1733 new_sk->sk_prot->unhash(new_sk);
1734 if (new_clcsock)
1735 sock_release(new_clcsock);
1736 new_sk->sk_state = SMC_CLOSED;
1737 smc_sock_set_flag(new_sk, SOCK_DEAD);
1738 sock_put(new_sk); /* final */
1739 *new_smc = NULL;
1740 goto out;
1741 }
1742
1743 /* new clcsock has inherited the smc listen-specific sk_data_ready
1744 * function; switch it back to the original sk_data_ready function
1745 */
1746 new_clcsock->sk->sk_data_ready = lsmc->clcsk_data_ready;
1747
1748 /* if new clcsock has also inherited the fallback-specific callback
1749 * functions, switch them back to the original ones.
1750 */
1751 if (lsmc->use_fallback) {
1752 if (lsmc->clcsk_state_change)
1753 new_clcsock->sk->sk_state_change = lsmc->clcsk_state_change;
1754 if (lsmc->clcsk_write_space)
1755 new_clcsock->sk->sk_write_space = lsmc->clcsk_write_space;
1756 if (lsmc->clcsk_error_report)
1757 new_clcsock->sk->sk_error_report = lsmc->clcsk_error_report;
1758 }
1759
1760 (*new_smc)->clcsock = new_clcsock;
1761out:
1762 return rc;
1763}
1764
1765/* add a just created sock to the accept queue of the listen sock as
1766 * candidate for a following socket accept call from user space
1767 */
1768static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
1769{
1770 struct smc_sock *par = smc_sk(parent);
1771
1772 sock_hold(sk); /* sock_put in smc_accept_unlink () */
1773 spin_lock(&par->accept_q_lock);
1774 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
1775 spin_unlock(&par->accept_q_lock);
1776 sk_acceptq_added(parent);
1777}
1778
1779/* remove a socket from the accept queue of its parental listening socket */
1780static void smc_accept_unlink(struct sock *sk)
1781{
1782 struct smc_sock *par = smc_sk(sk)->listen_smc;
1783
1784 spin_lock(&par->accept_q_lock);
1785 list_del_init(&smc_sk(sk)->accept_q);
1786 spin_unlock(&par->accept_q_lock);
1787 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
1788 sock_put(sk); /* sock_hold in smc_accept_enqueue */
1789}
1790
1791/* remove a sock from the accept queue to bind it to a new socket created
1792 * for a socket accept call from user space
1793 */
1794struct sock *smc_accept_dequeue(struct sock *parent,
1795 struct socket *new_sock)
1796{
1797 struct smc_sock *isk, *n;
1798 struct sock *new_sk;
1799
1800 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
1801 new_sk = (struct sock *)isk;
1802
1803 smc_accept_unlink(new_sk);
1804 if (new_sk->sk_state == SMC_CLOSED) {
1805 new_sk->sk_prot->unhash(new_sk);
1806 if (isk->clcsock) {
1807 sock_release(isk->clcsock);
1808 isk->clcsock = NULL;
1809 }
1810 sock_put(new_sk); /* final */
1811 continue;
1812 }
1813 if (new_sock) {
1814 sock_graft(new_sk, new_sock);
1815 new_sock->state = SS_CONNECTED;
1816 if (isk->use_fallback) {
1817 smc_sk(new_sk)->clcsock->file = new_sock->file;
1818 isk->clcsock->file->private_data = isk->clcsock;
1819 }
1820 }
1821 return new_sk;
1822 }
1823 return NULL;
1824}
1825
1826/* clean up for a created but never accepted sock */
1827void smc_close_non_accepted(struct sock *sk)
1828{
1829 struct smc_sock *smc = smc_sk(sk);
1830
1831 sock_hold(sk); /* sock_put below */
1832 lock_sock(sk);
1833 if (!sk->sk_lingertime)
1834 /* wait for peer closing */
1835 WRITE_ONCE(sk->sk_lingertime, SMC_MAX_STREAM_WAIT_TIMEOUT);
1836 __smc_release(smc);
1837 release_sock(sk);
1838 sock_put(sk); /* sock_hold above */
1839 sock_put(sk); /* final sock_put */
1840}
1841
1842static int smcr_serv_conf_first_link(struct smc_sock *smc)
1843{
1844 struct smc_link *link = smc->conn.lnk;
1845 struct smc_llc_qentry *qentry;
1846 int rc;
1847
1848 /* reg the sndbuf if it was vzalloced*/
1849 if (smc->conn.sndbuf_desc->is_vm) {
1850 if (smcr_link_reg_buf(link, smc->conn.sndbuf_desc))
1851 return SMC_CLC_DECL_ERR_REGBUF;
1852 }
1853
1854 /* reg the rmb */
1855 if (smcr_link_reg_buf(link, smc->conn.rmb_desc))
1856 return SMC_CLC_DECL_ERR_REGBUF;
1857
1858 /* send CONFIRM LINK request to client over the RoCE fabric */
1859 rc = smc_llc_send_confirm_link(link, SMC_LLC_REQ);
1860 if (rc < 0)
1861 return SMC_CLC_DECL_TIMEOUT_CL;
1862
1863 /* receive CONFIRM LINK response from client over the RoCE fabric */
1864 qentry = smc_llc_wait(link->lgr, link, SMC_LLC_WAIT_TIME,
1865 SMC_LLC_CONFIRM_LINK);
1866 if (!qentry) {
1867 struct smc_clc_msg_decline dclc;
1868
1869 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
1870 SMC_CLC_DECLINE, CLC_WAIT_TIME_SHORT);
1871 return rc == -EAGAIN ? SMC_CLC_DECL_TIMEOUT_CL : rc;
1872 }
1873 smc_llc_save_peer_uid(qentry);
1874 rc = smc_llc_eval_conf_link(qentry, SMC_LLC_RESP);
1875 smc_llc_flow_qentry_del(&link->lgr->llc_flow_lcl);
1876 if (rc)
1877 return SMC_CLC_DECL_RMBE_EC;
1878
1879 /* confirm_rkey is implicit on 1st contact */
1880 smc->conn.rmb_desc->is_conf_rkey = true;
1881
1882 smc_llc_link_active(link);
1883 smcr_lgr_set_type(link->lgr, SMC_LGR_SINGLE);
1884
1885 if (link->lgr->max_links > 1) {
1886 down_write(&link->lgr->llc_conf_mutex);
1887 /* initial contact - try to establish second link */
1888 smc_llc_srv_add_link(link, NULL);
1889 up_write(&link->lgr->llc_conf_mutex);
1890 }
1891 return 0;
1892}
1893
1894/* listen worker: finish */
1895static void smc_listen_out(struct smc_sock *new_smc)
1896{
1897 struct smc_sock *lsmc = new_smc->listen_smc;
1898 struct sock *newsmcsk = &new_smc->sk;
1899
1900 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
1901 atomic_dec(&lsmc->queued_smc_hs);
1902
1903 release_sock(newsmcsk); /* lock in smc_listen_work() */
1904 if (lsmc->sk.sk_state == SMC_LISTEN) {
1905 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
1906 smc_accept_enqueue(&lsmc->sk, newsmcsk);
1907 release_sock(&lsmc->sk);
1908 } else { /* no longer listening */
1909 smc_close_non_accepted(newsmcsk);
1910 }
1911
1912 /* Wake up accept */
1913 lsmc->sk.sk_data_ready(&lsmc->sk);
1914 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
1915}
1916
1917/* listen worker: finish in state connected */
1918static void smc_listen_out_connected(struct smc_sock *new_smc)
1919{
1920 struct sock *newsmcsk = &new_smc->sk;
1921
1922 if (newsmcsk->sk_state == SMC_INIT)
1923 newsmcsk->sk_state = SMC_ACTIVE;
1924
1925 smc_listen_out(new_smc);
1926}
1927
1928/* listen worker: finish in error state */
1929static void smc_listen_out_err(struct smc_sock *new_smc)
1930{
1931 struct sock *newsmcsk = &new_smc->sk;
1932 struct net *net = sock_net(newsmcsk);
1933
1934 this_cpu_inc(net->smc.smc_stats->srv_hshake_err_cnt);
1935 if (newsmcsk->sk_state == SMC_INIT)
1936 sock_put(&new_smc->sk); /* passive closing */
1937 newsmcsk->sk_state = SMC_CLOSED;
1938
1939 smc_listen_out(new_smc);
1940}
1941
1942/* listen worker: decline and fall back if possible */
1943static void smc_listen_decline(struct smc_sock *new_smc, int reason_code,
1944 int local_first, u8 version)
1945{
1946 /* RDMA setup failed, switch back to TCP */
1947 smc_conn_abort(new_smc, local_first);
1948 if (reason_code < 0 ||
1949 smc_switch_to_fallback(new_smc, reason_code)) {
1950 /* error, no fallback possible */
1951 smc_listen_out_err(new_smc);
1952 return;
1953 }
1954 if (reason_code && reason_code != SMC_CLC_DECL_PEERDECL) {
1955 if (smc_clc_send_decline(new_smc, reason_code, version) < 0) {
1956 smc_listen_out_err(new_smc);
1957 return;
1958 }
1959 }
1960 smc_listen_out_connected(new_smc);
1961}
1962
1963/* listen worker: version checking */
1964static int smc_listen_v2_check(struct smc_sock *new_smc,
1965 struct smc_clc_msg_proposal *pclc,
1966 struct smc_init_info *ini)
1967{
1968 struct smc_clc_smcd_v2_extension *pclc_smcd_v2_ext;
1969 struct smc_clc_v2_extension *pclc_v2_ext;
1970 int rc = SMC_CLC_DECL_PEERNOSMC;
1971
1972 ini->smc_type_v1 = pclc->hdr.typev1;
1973 ini->smc_type_v2 = pclc->hdr.typev2;
1974 ini->smcd_version = smcd_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1975 ini->smcr_version = smcr_indicated(ini->smc_type_v1) ? SMC_V1 : 0;
1976 if (pclc->hdr.version > SMC_V1) {
1977 if (smcd_indicated(ini->smc_type_v2))
1978 ini->smcd_version |= SMC_V2;
1979 if (smcr_indicated(ini->smc_type_v2))
1980 ini->smcr_version |= SMC_V2;
1981 }
1982 if (!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) {
1983 rc = SMC_CLC_DECL_PEERNOSMC;
1984 goto out;
1985 }
1986 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1987 if (!pclc_v2_ext) {
1988 ini->smcd_version &= ~SMC_V2;
1989 ini->smcr_version &= ~SMC_V2;
1990 rc = SMC_CLC_DECL_NOV2EXT;
1991 goto out;
1992 }
1993 pclc_smcd_v2_ext = smc_get_clc_smcd_v2_ext(pclc_v2_ext);
1994 if (ini->smcd_version & SMC_V2) {
1995 if (!smc_ism_is_v2_capable()) {
1996 ini->smcd_version &= ~SMC_V2;
1997 rc = SMC_CLC_DECL_NOISM2SUPP;
1998 } else if (!pclc_smcd_v2_ext) {
1999 ini->smcd_version &= ~SMC_V2;
2000 rc = SMC_CLC_DECL_NOV2DEXT;
2001 } else if (!pclc_v2_ext->hdr.eid_cnt &&
2002 !pclc_v2_ext->hdr.flag.seid) {
2003 ini->smcd_version &= ~SMC_V2;
2004 rc = SMC_CLC_DECL_NOUEID;
2005 }
2006 }
2007 if (ini->smcr_version & SMC_V2) {
2008 if (!pclc_v2_ext->hdr.eid_cnt) {
2009 ini->smcr_version &= ~SMC_V2;
2010 rc = SMC_CLC_DECL_NOUEID;
2011 }
2012 }
2013
2014 ini->release_nr = pclc_v2_ext->hdr.flag.release;
2015 if (pclc_v2_ext->hdr.flag.release > SMC_RELEASE)
2016 ini->release_nr = SMC_RELEASE;
2017
2018out:
2019 if (!ini->smcd_version && !ini->smcr_version)
2020 return rc;
2021
2022 return 0;
2023}
2024
2025/* listen worker: check prefixes */
2026static int smc_listen_prfx_check(struct smc_sock *new_smc,
2027 struct smc_clc_msg_proposal *pclc)
2028{
2029 struct smc_clc_msg_proposal_prefix *pclc_prfx;
2030 struct socket *newclcsock = new_smc->clcsock;
2031
2032 if (pclc->hdr.typev1 == SMC_TYPE_N)
2033 return 0;
2034 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
2035 if (!pclc_prfx)
2036 return -EPROTO;
2037 if (smc_clc_prfx_match(newclcsock, pclc_prfx))
2038 return SMC_CLC_DECL_DIFFPREFIX;
2039
2040 return 0;
2041}
2042
2043/* listen worker: initialize connection and buffers */
2044static int smc_listen_rdma_init(struct smc_sock *new_smc,
2045 struct smc_init_info *ini)
2046{
2047 int rc;
2048
2049 /* allocate connection / link group */
2050 rc = smc_conn_create(new_smc, ini);
2051 if (rc)
2052 return rc;
2053
2054 /* create send buffer and rmb */
2055 if (smc_buf_create(new_smc, false)) {
2056 smc_conn_abort(new_smc, ini->first_contact_local);
2057 return SMC_CLC_DECL_MEM;
2058 }
2059
2060 return 0;
2061}
2062
2063/* listen worker: initialize connection and buffers for SMC-D */
2064static int smc_listen_ism_init(struct smc_sock *new_smc,
2065 struct smc_init_info *ini)
2066{
2067 int rc;
2068
2069 rc = smc_conn_create(new_smc, ini);
2070 if (rc)
2071 return rc;
2072
2073 /* Create send and receive buffers */
2074 rc = smc_buf_create(new_smc, true);
2075 if (rc) {
2076 smc_conn_abort(new_smc, ini->first_contact_local);
2077 return (rc == -ENOSPC) ? SMC_CLC_DECL_MAX_DMB :
2078 SMC_CLC_DECL_MEM;
2079 }
2080
2081 return 0;
2082}
2083
2084static bool smc_is_already_selected(struct smcd_dev *smcd,
2085 struct smc_init_info *ini,
2086 int matches)
2087{
2088 int i;
2089
2090 for (i = 0; i < matches; i++)
2091 if (smcd == ini->ism_dev[i])
2092 return true;
2093
2094 return false;
2095}
2096
2097/* check for ISM devices matching proposed ISM devices */
2098static void smc_check_ism_v2_match(struct smc_init_info *ini,
2099 u16 proposed_chid,
2100 struct smcd_gid *proposed_gid,
2101 unsigned int *matches)
2102{
2103 struct smcd_dev *smcd;
2104
2105 list_for_each_entry(smcd, &smcd_dev_list.list, list) {
2106 if (smcd->going_away)
2107 continue;
2108 if (smc_is_already_selected(smcd, ini, *matches))
2109 continue;
2110 if (smc_ism_get_chid(smcd) == proposed_chid &&
2111 !smc_ism_cantalk(proposed_gid, ISM_RESERVED_VLANID, smcd)) {
2112 ini->ism_peer_gid[*matches].gid = proposed_gid->gid;
2113 if (__smc_ism_is_emulated(proposed_chid))
2114 ini->ism_peer_gid[*matches].gid_ext =
2115 proposed_gid->gid_ext;
2116 /* non-Emulated-ISM's peer gid_ext remains 0. */
2117 ini->ism_dev[*matches] = smcd;
2118 (*matches)++;
2119 break;
2120 }
2121 }
2122}
2123
2124static void smc_find_ism_store_rc(u32 rc, struct smc_init_info *ini)
2125{
2126 if (!ini->rc)
2127 ini->rc = rc;
2128}
2129
2130static 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)
2133{
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;
2138 struct smcd_gid smcd_gid;
2139 u8 smcd_version;
2140 u8 *eid = NULL;
2141 int i, rc;
2142 u16 chid;
2143
2144 if (!(ini->smcd_version & SMC_V2) || !smcd_indicated(ini->smc_type_v2))
2145 goto not_found;
2146
2147 pclc_smcd = smc_get_clc_msg_smcd(pclc);
2148 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2149 smcd_v2_ext = smc_get_clc_smcd_v2_ext(smc_v2_ext);
2150 if (!pclc_smcd || !smc_v2_ext || !smcd_v2_ext)
2151 goto not_found;
2152
2153 mutex_lock(&smcd_dev_list.mutex);
2154 if (pclc_smcd->ism.chid) {
2155 /* check for ISM device matching proposed native ISM device */
2156 smcd_gid.gid = ntohll(pclc_smcd->ism.gid);
2157 smcd_gid.gid_ext = 0;
2158 smc_check_ism_v2_match(ini, ntohs(pclc_smcd->ism.chid),
2159 &smcd_gid, &matches);
2160 }
2161 for (i = 0; i < smc_v2_ext->hdr.ism_gid_cnt; i++) {
2162 /* check for ISM devices matching proposed non-native ISM
2163 * devices
2164 */
2165 smcd_gid.gid = ntohll(smcd_v2_ext->gidchid[i].gid);
2166 smcd_gid.gid_ext = 0;
2167 chid = ntohs(smcd_v2_ext->gidchid[i].chid);
2168 if (__smc_ism_is_emulated(chid)) {
2169 if ((i + 1) == smc_v2_ext->hdr.ism_gid_cnt ||
2170 chid != ntohs(smcd_v2_ext->gidchid[i + 1].chid))
2171 /* each Emulated-ISM device takes two GID-CHID
2172 * entries and CHID of the second entry repeats
2173 * that of the first entry.
2174 *
2175 * So check if the next GID-CHID entry exists
2176 * and both two entries' CHIDs are the same.
2177 */
2178 continue;
2179 smcd_gid.gid_ext =
2180 ntohll(smcd_v2_ext->gidchid[++i].gid);
2181 }
2182 smc_check_ism_v2_match(ini, chid, &smcd_gid, &matches);
2183 }
2184 mutex_unlock(&smcd_dev_list.mutex);
2185
2186 if (!ini->ism_dev[0]) {
2187 smc_find_ism_store_rc(SMC_CLC_DECL_NOSMCD2DEV, ini);
2188 goto not_found;
2189 }
2190
2191 smc_ism_get_system_eid(&eid);
2192 if (!smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext,
2193 smcd_v2_ext->system_eid, eid))
2194 goto not_found;
2195
2196 /* separate - outside the smcd_dev_list.lock */
2197 smcd_version = ini->smcd_version;
2198 for (i = 0; i < matches; i++) {
2199 ini->smcd_version = SMC_V2;
2200 ini->is_smcd = true;
2201 ini->ism_selected = i;
2202 rc = smc_listen_ism_init(new_smc, ini);
2203 if (rc) {
2204 smc_find_ism_store_rc(rc, ini);
2205 /* try next active ISM device */
2206 continue;
2207 }
2208 return; /* matching and usable V2 ISM device found */
2209 }
2210 /* no V2 ISM device could be initialized */
2211 ini->smcd_version = smcd_version; /* restore original value */
2212 ini->negotiated_eid[0] = 0;
2213
2214not_found:
2215 ini->smcd_version &= ~SMC_V2;
2216 ini->ism_dev[0] = NULL;
2217 ini->is_smcd = false;
2218}
2219
2220static void smc_find_ism_v1_device_serv(struct smc_sock *new_smc,
2221 struct smc_clc_msg_proposal *pclc,
2222 struct smc_init_info *ini)
2223{
2224 struct smc_clc_msg_smcd *pclc_smcd = smc_get_clc_msg_smcd(pclc);
2225 int rc = 0;
2226
2227 /* check if ISM V1 is available */
2228 if (!(ini->smcd_version & SMC_V1) ||
2229 !smcd_indicated(ini->smc_type_v1) ||
2230 !pclc_smcd)
2231 goto not_found;
2232 ini->is_smcd = true; /* prepare ISM check */
2233 ini->ism_peer_gid[0].gid = ntohll(pclc_smcd->ism.gid);
2234 ini->ism_peer_gid[0].gid_ext = 0;
2235 rc = smc_find_ism_device(new_smc, ini);
2236 if (rc)
2237 goto not_found;
2238 ini->ism_selected = 0;
2239 rc = smc_listen_ism_init(new_smc, ini);
2240 if (!rc)
2241 return; /* V1 ISM device found */
2242
2243not_found:
2244 smc_find_ism_store_rc(rc, ini);
2245 ini->smcd_version &= ~SMC_V1;
2246 ini->ism_dev[0] = NULL;
2247 ini->is_smcd = false;
2248}
2249
2250/* listen worker: register buffers */
2251static int smc_listen_rdma_reg(struct smc_sock *new_smc, bool local_first)
2252{
2253 struct smc_connection *conn = &new_smc->conn;
2254
2255 if (!local_first) {
2256 /* reg sendbufs if they were vzalloced */
2257 if (conn->sndbuf_desc->is_vm) {
2258 if (smcr_lgr_reg_sndbufs(conn->lnk,
2259 conn->sndbuf_desc))
2260 return SMC_CLC_DECL_ERR_REGBUF;
2261 }
2262 if (smcr_lgr_reg_rmbs(conn->lnk, conn->rmb_desc))
2263 return SMC_CLC_DECL_ERR_REGBUF;
2264 }
2265
2266 return 0;
2267}
2268
2269static void smc_find_rdma_v2_device_serv(struct smc_sock *new_smc,
2270 struct smc_clc_msg_proposal *pclc,
2271 struct smc_init_info *ini)
2272{
2273 struct smc_clc_v2_extension *smc_v2_ext;
2274 u8 smcr_version;
2275 int rc;
2276
2277 if (!(ini->smcr_version & SMC_V2) || !smcr_indicated(ini->smc_type_v2))
2278 goto not_found;
2279
2280 smc_v2_ext = smc_get_clc_v2_ext(pclc);
2281 if (!smc_v2_ext ||
2282 !smc_clc_match_eid(ini->negotiated_eid, smc_v2_ext, NULL, NULL))
2283 goto not_found;
2284
2285 /* prepare RDMA check */
2286 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2287 memcpy(ini->peer_gid, smc_v2_ext->roce, SMC_GID_SIZE);
2288 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2289 ini->check_smcrv2 = true;
2290 ini->smcrv2.clc_sk = new_smc->clcsock->sk;
2291 ini->smcrv2.saddr = new_smc->clcsock->sk->sk_rcv_saddr;
2292 ini->smcrv2.daddr = smc_ib_gid_to_ipv4(smc_v2_ext->roce);
2293 rc = smc_find_rdma_device(new_smc, ini);
2294 if (rc) {
2295 smc_find_ism_store_rc(rc, ini);
2296 goto not_found;
2297 }
2298 if (!ini->smcrv2.uses_gateway)
2299 memcpy(ini->smcrv2.nexthop_mac, pclc->lcl.mac, ETH_ALEN);
2300
2301 smcr_version = ini->smcr_version;
2302 ini->smcr_version = SMC_V2;
2303 rc = smc_listen_rdma_init(new_smc, ini);
2304 if (!rc) {
2305 rc = smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2306 if (rc)
2307 smc_conn_abort(new_smc, ini->first_contact_local);
2308 }
2309 if (!rc)
2310 return;
2311 ini->smcr_version = smcr_version;
2312 smc_find_ism_store_rc(rc, ini);
2313
2314not_found:
2315 ini->smcr_version &= ~SMC_V2;
2316 ini->smcrv2.ib_dev_v2 = NULL;
2317 ini->check_smcrv2 = false;
2318}
2319
2320static int smc_find_rdma_v1_device_serv(struct smc_sock *new_smc,
2321 struct smc_clc_msg_proposal *pclc,
2322 struct smc_init_info *ini)
2323{
2324 int rc;
2325
2326 if (!(ini->smcr_version & SMC_V1) || !smcr_indicated(ini->smc_type_v1))
2327 return SMC_CLC_DECL_NOSMCDEV;
2328
2329 /* prepare RDMA check */
2330 memcpy(ini->peer_systemid, pclc->lcl.id_for_peer, SMC_SYSTEMID_LEN);
2331 memcpy(ini->peer_gid, pclc->lcl.gid, SMC_GID_SIZE);
2332 memcpy(ini->peer_mac, pclc->lcl.mac, ETH_ALEN);
2333 rc = smc_find_rdma_device(new_smc, ini);
2334 if (rc) {
2335 /* no RDMA device found */
2336 return SMC_CLC_DECL_NOSMCDEV;
2337 }
2338 rc = smc_listen_rdma_init(new_smc, ini);
2339 if (rc)
2340 return rc;
2341 return smc_listen_rdma_reg(new_smc, ini->first_contact_local);
2342}
2343
2344/* determine the local device matching to proposal */
2345static int smc_listen_find_device(struct smc_sock *new_smc,
2346 struct smc_clc_msg_proposal *pclc,
2347 struct smc_init_info *ini)
2348{
2349 int prfx_rc;
2350
2351 /* check for ISM device matching V2 proposed device */
2352 smc_find_ism_v2_device_serv(new_smc, pclc, ini);
2353 if (ini->ism_dev[0])
2354 return 0;
2355
2356 /* check for matching IP prefix and subnet length (V1) */
2357 prfx_rc = smc_listen_prfx_check(new_smc, pclc);
2358 if (prfx_rc)
2359 smc_find_ism_store_rc(prfx_rc, ini);
2360
2361 /* get vlan id from IP device */
2362 if (smc_vlan_by_tcpsk(new_smc->clcsock, ini))
2363 return ini->rc ?: SMC_CLC_DECL_GETVLANERR;
2364
2365 /* check for ISM device matching V1 proposed device */
2366 if (!prfx_rc)
2367 smc_find_ism_v1_device_serv(new_smc, pclc, ini);
2368 if (ini->ism_dev[0])
2369 return 0;
2370
2371 if (!smcr_indicated(pclc->hdr.typev1) &&
2372 !smcr_indicated(pclc->hdr.typev2))
2373 /* skip RDMA and decline */
2374 return ini->rc ?: SMC_CLC_DECL_NOSMCDDEV;
2375
2376 /* check if RDMA V2 is available */
2377 smc_find_rdma_v2_device_serv(new_smc, pclc, ini);
2378 if (ini->smcrv2.ib_dev_v2)
2379 return 0;
2380
2381 /* check if RDMA V1 is available */
2382 if (!prfx_rc) {
2383 int rc;
2384
2385 rc = smc_find_rdma_v1_device_serv(new_smc, pclc, ini);
2386 smc_find_ism_store_rc(rc, ini);
2387 return (!rc) ? 0 : ini->rc;
2388 }
2389 return prfx_rc;
2390}
2391
2392/* listen worker: finish RDMA setup */
2393static int smc_listen_rdma_finish(struct smc_sock *new_smc,
2394 struct smc_clc_msg_accept_confirm *cclc,
2395 bool local_first,
2396 struct smc_init_info *ini)
2397{
2398 struct smc_link *link = new_smc->conn.lnk;
2399 int reason_code = 0;
2400
2401 if (local_first)
2402 smc_link_save_peer_info(link, cclc, ini);
2403
2404 if (smc_rmb_rtoken_handling(&new_smc->conn, link, cclc))
2405 return SMC_CLC_DECL_ERR_RTOK;
2406
2407 if (local_first) {
2408 if (smc_ib_ready_link(link))
2409 return SMC_CLC_DECL_ERR_RDYLNK;
2410 /* QP confirmation over RoCE fabric */
2411 smc_llc_flow_initiate(link->lgr, SMC_LLC_FLOW_ADD_LINK);
2412 reason_code = smcr_serv_conf_first_link(new_smc);
2413 smc_llc_flow_stop(link->lgr, &link->lgr->llc_flow_lcl);
2414 }
2415 return reason_code;
2416}
2417
2418/* setup for connection of server */
2419static void smc_listen_work(struct work_struct *work)
2420{
2421 struct smc_sock *new_smc = container_of(work, struct smc_sock,
2422 smc_listen_work);
2423 struct socket *newclcsock = new_smc->clcsock;
2424 struct smc_clc_msg_accept_confirm *cclc;
2425 struct smc_clc_msg_proposal_area *buf;
2426 struct smc_clc_msg_proposal *pclc;
2427 struct smc_init_info *ini = NULL;
2428 u8 proposal_version = SMC_V1;
2429 u8 accept_version;
2430 int rc = 0;
2431
2432 lock_sock(&new_smc->sk); /* release in smc_listen_out() */
2433 if (new_smc->listen_smc->sk.sk_state != SMC_LISTEN)
2434 return smc_listen_out_err(new_smc);
2435
2436 if (new_smc->use_fallback) {
2437 smc_listen_out_connected(new_smc);
2438 return;
2439 }
2440
2441 /* check if peer is smc capable */
2442 if (!tcp_sk(newclcsock->sk)->syn_smc) {
2443 rc = smc_switch_to_fallback(new_smc, SMC_CLC_DECL_PEERNOSMC);
2444 if (rc)
2445 smc_listen_out_err(new_smc);
2446 else
2447 smc_listen_out_connected(new_smc);
2448 return;
2449 }
2450
2451 /* do inband token exchange -
2452 * wait for and receive SMC Proposal CLC message
2453 */
2454 buf = kzalloc(sizeof(*buf), GFP_KERNEL);
2455 if (!buf) {
2456 rc = SMC_CLC_DECL_MEM;
2457 goto out_decl;
2458 }
2459 pclc = (struct smc_clc_msg_proposal *)buf;
2460 rc = smc_clc_wait_msg(new_smc, pclc, sizeof(*buf),
2461 SMC_CLC_PROPOSAL, CLC_WAIT_TIME);
2462 if (rc)
2463 goto out_decl;
2464
2465 if (pclc->hdr.version > SMC_V1)
2466 proposal_version = SMC_V2;
2467
2468 /* IPSec connections opt out of SMC optimizations */
2469 if (using_ipsec(new_smc)) {
2470 rc = SMC_CLC_DECL_IPSEC;
2471 goto out_decl;
2472 }
2473
2474 ini = kzalloc(sizeof(*ini), GFP_KERNEL);
2475 if (!ini) {
2476 rc = SMC_CLC_DECL_MEM;
2477 goto out_decl;
2478 }
2479
2480 /* initial version checking */
2481 rc = smc_listen_v2_check(new_smc, pclc, ini);
2482 if (rc)
2483 goto out_decl;
2484
2485 rc = smc_clc_srv_v2x_features_validate(new_smc, pclc, ini);
2486 if (rc)
2487 goto out_decl;
2488
2489 mutex_lock(&smc_server_lgr_pending);
2490 smc_rx_init(new_smc);
2491 smc_tx_init(new_smc);
2492
2493 /* determine ISM or RoCE device used for connection */
2494 rc = smc_listen_find_device(new_smc, pclc, ini);
2495 if (rc)
2496 goto out_unlock;
2497
2498 /* send SMC Accept CLC message */
2499 accept_version = ini->is_smcd ? ini->smcd_version : ini->smcr_version;
2500 rc = smc_clc_send_accept(new_smc, ini->first_contact_local,
2501 accept_version, ini->negotiated_eid, ini);
2502 if (rc)
2503 goto out_unlock;
2504
2505 /* SMC-D does not need this lock any more */
2506 if (ini->is_smcd)
2507 mutex_unlock(&smc_server_lgr_pending);
2508
2509 /* receive SMC Confirm CLC message */
2510 memset(buf, 0, sizeof(*buf));
2511 cclc = (struct smc_clc_msg_accept_confirm *)buf;
2512 rc = smc_clc_wait_msg(new_smc, cclc, sizeof(*buf),
2513 SMC_CLC_CONFIRM, CLC_WAIT_TIME);
2514 if (rc) {
2515 if (!ini->is_smcd)
2516 goto out_unlock;
2517 goto out_decl;
2518 }
2519
2520 rc = smc_clc_v2x_features_confirm_check(cclc, ini);
2521 if (rc) {
2522 if (!ini->is_smcd)
2523 goto out_unlock;
2524 goto out_decl;
2525 }
2526
2527 /* fce smc release version is needed in smc_listen_rdma_finish,
2528 * so save fce info here.
2529 */
2530 smc_conn_save_peer_info_fce(new_smc, cclc);
2531
2532 /* finish worker */
2533 if (!ini->is_smcd) {
2534 rc = smc_listen_rdma_finish(new_smc, cclc,
2535 ini->first_contact_local, ini);
2536 if (rc)
2537 goto out_unlock;
2538 mutex_unlock(&smc_server_lgr_pending);
2539 }
2540 smc_conn_save_peer_info(new_smc, cclc);
2541
2542 if (ini->is_smcd &&
2543 smc_ism_support_dmb_nocopy(new_smc->conn.lgr->smcd)) {
2544 rc = smcd_buf_attach(new_smc);
2545 if (rc)
2546 goto out_decl;
2547 }
2548
2549 smc_listen_out_connected(new_smc);
2550 SMC_STAT_SERV_SUCC_INC(sock_net(newclcsock->sk), ini);
2551 goto out_free;
2552
2553out_unlock:
2554 mutex_unlock(&smc_server_lgr_pending);
2555out_decl:
2556 smc_listen_decline(new_smc, rc, ini ? ini->first_contact_local : 0,
2557 proposal_version);
2558out_free:
2559 kfree(ini);
2560 kfree(buf);
2561}
2562
2563static void smc_tcp_listen_work(struct work_struct *work)
2564{
2565 struct smc_sock *lsmc = container_of(work, struct smc_sock,
2566 tcp_listen_work);
2567 struct sock *lsk = &lsmc->sk;
2568 struct smc_sock *new_smc;
2569 int rc = 0;
2570
2571 lock_sock(lsk);
2572 while (lsk->sk_state == SMC_LISTEN) {
2573 rc = smc_clcsock_accept(lsmc, &new_smc);
2574 if (rc) /* clcsock accept queue empty or error */
2575 goto out;
2576 if (!new_smc)
2577 continue;
2578
2579 if (tcp_sk(new_smc->clcsock->sk)->syn_smc)
2580 atomic_inc(&lsmc->queued_smc_hs);
2581
2582 new_smc->listen_smc = lsmc;
2583 new_smc->use_fallback = lsmc->use_fallback;
2584 new_smc->fallback_rsn = lsmc->fallback_rsn;
2585 sock_hold(lsk); /* sock_put in smc_listen_work */
2586 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
2587 smc_copy_sock_settings_to_smc(new_smc);
2588 sock_hold(&new_smc->sk); /* sock_put in passive closing */
2589 if (!queue_work(smc_hs_wq, &new_smc->smc_listen_work))
2590 sock_put(&new_smc->sk);
2591 }
2592
2593out:
2594 release_sock(lsk);
2595 sock_put(&lsmc->sk); /* sock_hold in smc_clcsock_data_ready() */
2596}
2597
2598static void smc_clcsock_data_ready(struct sock *listen_clcsock)
2599{
2600 struct smc_sock *lsmc;
2601
2602 read_lock_bh(&listen_clcsock->sk_callback_lock);
2603 lsmc = smc_clcsock_user_data(listen_clcsock);
2604 if (!lsmc)
2605 goto out;
2606 lsmc->clcsk_data_ready(listen_clcsock);
2607 if (lsmc->sk.sk_state == SMC_LISTEN) {
2608 sock_hold(&lsmc->sk); /* sock_put in smc_tcp_listen_work() */
2609 if (!queue_work(smc_tcp_ls_wq, &lsmc->tcp_listen_work))
2610 sock_put(&lsmc->sk);
2611 }
2612out:
2613 read_unlock_bh(&listen_clcsock->sk_callback_lock);
2614}
2615
2616int smc_listen(struct socket *sock, int backlog)
2617{
2618 struct sock *sk = sock->sk;
2619 struct smc_sock *smc;
2620 int rc;
2621
2622 smc = smc_sk(sk);
2623 lock_sock(sk);
2624
2625 rc = -EINVAL;
2626 if ((sk->sk_state != SMC_INIT && sk->sk_state != SMC_LISTEN) ||
2627 smc->connect_nonblock || sock->state != SS_UNCONNECTED)
2628 goto out;
2629
2630 rc = 0;
2631 if (sk->sk_state == SMC_LISTEN) {
2632 sk->sk_max_ack_backlog = backlog;
2633 goto out;
2634 }
2635 /* some socket options are handled in core, so we could not apply
2636 * them to the clc socket -- copy smc socket options to clc socket
2637 */
2638 smc_copy_sock_settings_to_clc(smc);
2639 if (!smc->use_fallback)
2640 tcp_sk(smc->clcsock->sk)->syn_smc = 1;
2641
2642 /* save original sk_data_ready function and establish
2643 * smc-specific sk_data_ready function
2644 */
2645 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2646 smc->clcsock->sk->sk_user_data =
2647 (void *)((uintptr_t)smc | SK_USER_DATA_NOCOPY);
2648 smc_clcsock_replace_cb(&smc->clcsock->sk->sk_data_ready,
2649 smc_clcsock_data_ready, &smc->clcsk_data_ready);
2650 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2651
2652 /* save original ops */
2653 smc->ori_af_ops = inet_csk(smc->clcsock->sk)->icsk_af_ops;
2654
2655 smc->af_ops = *smc->ori_af_ops;
2656 smc->af_ops.syn_recv_sock = smc_tcp_syn_recv_sock;
2657
2658 inet_csk(smc->clcsock->sk)->icsk_af_ops = &smc->af_ops;
2659
2660 if (smc->limit_smc_hs)
2661 tcp_sk(smc->clcsock->sk)->smc_hs_congested = smc_hs_congested;
2662
2663 rc = kernel_listen(smc->clcsock, backlog);
2664 if (rc) {
2665 write_lock_bh(&smc->clcsock->sk->sk_callback_lock);
2666 smc_clcsock_restore_cb(&smc->clcsock->sk->sk_data_ready,
2667 &smc->clcsk_data_ready);
2668 smc->clcsock->sk->sk_user_data = NULL;
2669 write_unlock_bh(&smc->clcsock->sk->sk_callback_lock);
2670 goto out;
2671 }
2672 sk->sk_max_ack_backlog = backlog;
2673 sk->sk_ack_backlog = 0;
2674 sk->sk_state = SMC_LISTEN;
2675
2676out:
2677 release_sock(sk);
2678 return rc;
2679}
2680
2681int smc_accept(struct socket *sock, struct socket *new_sock,
2682 struct proto_accept_arg *arg)
2683{
2684 struct sock *sk = sock->sk, *nsk;
2685 DECLARE_WAITQUEUE(wait, current);
2686 struct smc_sock *lsmc;
2687 long timeo;
2688 int rc = 0;
2689
2690 lsmc = smc_sk(sk);
2691 sock_hold(sk); /* sock_put below */
2692 lock_sock(sk);
2693
2694 if (lsmc->sk.sk_state != SMC_LISTEN) {
2695 rc = -EINVAL;
2696 release_sock(sk);
2697 goto out;
2698 }
2699
2700 /* Wait for an incoming connection */
2701 timeo = sock_rcvtimeo(sk, arg->flags & O_NONBLOCK);
2702 add_wait_queue_exclusive(sk_sleep(sk), &wait);
2703 while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
2704 set_current_state(TASK_INTERRUPTIBLE);
2705 if (!timeo) {
2706 rc = -EAGAIN;
2707 break;
2708 }
2709 release_sock(sk);
2710 timeo = schedule_timeout(timeo);
2711 /* wakeup by sk_data_ready in smc_listen_work() */
2712 sched_annotate_sleep();
2713 lock_sock(sk);
2714 if (signal_pending(current)) {
2715 rc = sock_intr_errno(timeo);
2716 break;
2717 }
2718 }
2719 set_current_state(TASK_RUNNING);
2720 remove_wait_queue(sk_sleep(sk), &wait);
2721
2722 if (!rc)
2723 rc = sock_error(nsk);
2724 release_sock(sk);
2725 if (rc)
2726 goto out;
2727
2728 if (lsmc->sockopt_defer_accept && !(arg->flags & O_NONBLOCK)) {
2729 /* wait till data arrives on the socket */
2730 timeo = msecs_to_jiffies(lsmc->sockopt_defer_accept *
2731 MSEC_PER_SEC);
2732 if (smc_sk(nsk)->use_fallback) {
2733 struct sock *clcsk = smc_sk(nsk)->clcsock->sk;
2734
2735 lock_sock(clcsk);
2736 if (skb_queue_empty(&clcsk->sk_receive_queue))
2737 sk_wait_data(clcsk, &timeo, NULL);
2738 release_sock(clcsk);
2739 } else if (!atomic_read(&smc_sk(nsk)->conn.bytes_to_rcv)) {
2740 lock_sock(nsk);
2741 smc_rx_wait(smc_sk(nsk), &timeo, 0, smc_rx_data_available);
2742 release_sock(nsk);
2743 }
2744 }
2745
2746out:
2747 sock_put(sk); /* sock_hold above */
2748 return rc;
2749}
2750
2751int smc_getname(struct socket *sock, struct sockaddr *addr,
2752 int peer)
2753{
2754 struct smc_sock *smc;
2755
2756 if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
2757 (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
2758 return -ENOTCONN;
2759
2760 smc = smc_sk(sock->sk);
2761
2762 return smc->clcsock->ops->getname(smc->clcsock, addr, peer);
2763}
2764
2765int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
2766{
2767 struct sock *sk = sock->sk;
2768 struct smc_sock *smc;
2769 int rc;
2770
2771 smc = smc_sk(sk);
2772 lock_sock(sk);
2773
2774 /* SMC does not support connect with fastopen */
2775 if (msg->msg_flags & MSG_FASTOPEN) {
2776 /* not connected yet, fallback */
2777 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
2778 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
2779 if (rc)
2780 goto out;
2781 } else {
2782 rc = -EINVAL;
2783 goto out;
2784 }
2785 } else if ((sk->sk_state != SMC_ACTIVE) &&
2786 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2787 (sk->sk_state != SMC_INIT)) {
2788 rc = -EPIPE;
2789 goto out;
2790 }
2791
2792 if (smc->use_fallback) {
2793 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
2794 } else {
2795 rc = smc_tx_sendmsg(smc, msg, len);
2796 SMC_STAT_TX_PAYLOAD(smc, len, rc);
2797 }
2798out:
2799 release_sock(sk);
2800 return rc;
2801}
2802
2803int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
2804 int flags)
2805{
2806 struct sock *sk = sock->sk;
2807 struct smc_sock *smc;
2808 int rc = -ENOTCONN;
2809
2810 smc = smc_sk(sk);
2811 lock_sock(sk);
2812 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
2813 /* socket was connected before, no more data to read */
2814 rc = 0;
2815 goto out;
2816 }
2817 if ((sk->sk_state == SMC_INIT) ||
2818 (sk->sk_state == SMC_LISTEN) ||
2819 (sk->sk_state == SMC_CLOSED))
2820 goto out;
2821
2822 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
2823 rc = 0;
2824 goto out;
2825 }
2826
2827 if (smc->use_fallback) {
2828 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
2829 } else {
2830 msg->msg_namelen = 0;
2831 rc = smc_rx_recvmsg(smc, msg, NULL, len, flags);
2832 SMC_STAT_RX_PAYLOAD(smc, rc, rc);
2833 }
2834
2835out:
2836 release_sock(sk);
2837 return rc;
2838}
2839
2840static __poll_t smc_accept_poll(struct sock *parent)
2841{
2842 struct smc_sock *isk = smc_sk(parent);
2843 __poll_t mask = 0;
2844
2845 spin_lock(&isk->accept_q_lock);
2846 if (!list_empty(&isk->accept_q))
2847 mask = EPOLLIN | EPOLLRDNORM;
2848 spin_unlock(&isk->accept_q_lock);
2849
2850 return mask;
2851}
2852
2853__poll_t smc_poll(struct file *file, struct socket *sock,
2854 poll_table *wait)
2855{
2856 struct sock *sk = sock->sk;
2857 struct smc_sock *smc;
2858 __poll_t mask = 0;
2859
2860 if (!sk)
2861 return EPOLLNVAL;
2862
2863 smc = smc_sk(sock->sk);
2864 if (smc->use_fallback) {
2865 /* delegate to CLC child sock */
2866 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
2867 sk->sk_err = smc->clcsock->sk->sk_err;
2868 } else {
2869 if (sk->sk_state != SMC_CLOSED)
2870 sock_poll_wait(file, sock, wait);
2871 if (sk->sk_err)
2872 mask |= EPOLLERR;
2873 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
2874 (sk->sk_state == SMC_CLOSED))
2875 mask |= EPOLLHUP;
2876 if (sk->sk_state == SMC_LISTEN) {
2877 /* woken up by sk_data_ready in smc_listen_work() */
2878 mask |= smc_accept_poll(sk);
2879 } else if (smc->use_fallback) { /* as result of connect_work()*/
2880 mask |= smc->clcsock->ops->poll(file, smc->clcsock,
2881 wait);
2882 sk->sk_err = smc->clcsock->sk->sk_err;
2883 } else {
2884 if ((sk->sk_state != SMC_INIT &&
2885 atomic_read(&smc->conn.sndbuf_space)) ||
2886 sk->sk_shutdown & SEND_SHUTDOWN) {
2887 mask |= EPOLLOUT | EPOLLWRNORM;
2888 } else {
2889 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
2890 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
2891
2892 if (sk->sk_state != SMC_INIT) {
2893 /* Race breaker the same way as tcp_poll(). */
2894 smp_mb__after_atomic();
2895 if (atomic_read(&smc->conn.sndbuf_space))
2896 mask |= EPOLLOUT | EPOLLWRNORM;
2897 }
2898 }
2899 if (atomic_read(&smc->conn.bytes_to_rcv))
2900 mask |= EPOLLIN | EPOLLRDNORM;
2901 if (sk->sk_shutdown & RCV_SHUTDOWN)
2902 mask |= EPOLLIN | EPOLLRDNORM | EPOLLRDHUP;
2903 if (sk->sk_state == SMC_APPCLOSEWAIT1)
2904 mask |= EPOLLIN;
2905 if (smc->conn.urg_state == SMC_URG_VALID)
2906 mask |= EPOLLPRI;
2907 }
2908 }
2909
2910 return mask;
2911}
2912
2913int smc_shutdown(struct socket *sock, int how)
2914{
2915 struct sock *sk = sock->sk;
2916 bool do_shutdown = true;
2917 struct smc_sock *smc;
2918 int rc = -EINVAL;
2919 int old_state;
2920 int rc1 = 0;
2921
2922 smc = smc_sk(sk);
2923
2924 if ((how < SHUT_RD) || (how > SHUT_RDWR))
2925 return rc;
2926
2927 lock_sock(sk);
2928
2929 if (sock->state == SS_CONNECTING) {
2930 if (sk->sk_state == SMC_ACTIVE)
2931 sock->state = SS_CONNECTED;
2932 else if (sk->sk_state == SMC_PEERCLOSEWAIT1 ||
2933 sk->sk_state == SMC_PEERCLOSEWAIT2 ||
2934 sk->sk_state == SMC_APPCLOSEWAIT1 ||
2935 sk->sk_state == SMC_APPCLOSEWAIT2 ||
2936 sk->sk_state == SMC_APPFINCLOSEWAIT)
2937 sock->state = SS_DISCONNECTING;
2938 }
2939
2940 rc = -ENOTCONN;
2941 if ((sk->sk_state != SMC_ACTIVE) &&
2942 (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
2943 (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
2944 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
2945 (sk->sk_state != SMC_APPCLOSEWAIT2) &&
2946 (sk->sk_state != SMC_APPFINCLOSEWAIT))
2947 goto out;
2948 if (smc->use_fallback) {
2949 rc = kernel_sock_shutdown(smc->clcsock, how);
2950 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
2951 if (sk->sk_shutdown == SHUTDOWN_MASK) {
2952 sk->sk_state = SMC_CLOSED;
2953 sk->sk_socket->state = SS_UNCONNECTED;
2954 sock_put(sk);
2955 }
2956 goto out;
2957 }
2958 switch (how) {
2959 case SHUT_RDWR: /* shutdown in both directions */
2960 old_state = sk->sk_state;
2961 rc = smc_close_active(smc);
2962 if (old_state == SMC_ACTIVE &&
2963 sk->sk_state == SMC_PEERCLOSEWAIT1)
2964 do_shutdown = false;
2965 break;
2966 case SHUT_WR:
2967 rc = smc_close_shutdown_write(smc);
2968 break;
2969 case SHUT_RD:
2970 rc = 0;
2971 /* nothing more to do because peer is not involved */
2972 break;
2973 }
2974 if (do_shutdown && smc->clcsock)
2975 rc1 = kernel_sock_shutdown(smc->clcsock, how);
2976 /* map sock_shutdown_cmd constants to sk_shutdown value range */
2977 sk->sk_shutdown |= how + 1;
2978
2979 if (sk->sk_state == SMC_CLOSED)
2980 sock->state = SS_UNCONNECTED;
2981 else
2982 sock->state = SS_DISCONNECTING;
2983out:
2984 release_sock(sk);
2985 return rc ? rc : rc1;
2986}
2987
2988static int __smc_getsockopt(struct socket *sock, int level, int optname,
2989 char __user *optval, int __user *optlen)
2990{
2991 struct smc_sock *smc;
2992 int val, len;
2993
2994 smc = smc_sk(sock->sk);
2995
2996 if (get_user(len, optlen))
2997 return -EFAULT;
2998
2999 len = min_t(int, len, sizeof(int));
3000
3001 if (len < 0)
3002 return -EINVAL;
3003
3004 switch (optname) {
3005 case SMC_LIMIT_HS:
3006 val = smc->limit_smc_hs;
3007 break;
3008 default:
3009 return -EOPNOTSUPP;
3010 }
3011
3012 if (put_user(len, optlen))
3013 return -EFAULT;
3014 if (copy_to_user(optval, &val, len))
3015 return -EFAULT;
3016
3017 return 0;
3018}
3019
3020static int __smc_setsockopt(struct socket *sock, int level, int optname,
3021 sockptr_t optval, unsigned int optlen)
3022{
3023 struct sock *sk = sock->sk;
3024 struct smc_sock *smc;
3025 int val, rc;
3026
3027 smc = smc_sk(sk);
3028
3029 lock_sock(sk);
3030 switch (optname) {
3031 case SMC_LIMIT_HS:
3032 if (optlen < sizeof(int)) {
3033 rc = -EINVAL;
3034 break;
3035 }
3036 if (copy_from_sockptr(&val, optval, sizeof(int))) {
3037 rc = -EFAULT;
3038 break;
3039 }
3040
3041 smc->limit_smc_hs = !!val;
3042 rc = 0;
3043 break;
3044 default:
3045 rc = -EOPNOTSUPP;
3046 break;
3047 }
3048 release_sock(sk);
3049
3050 return rc;
3051}
3052
3053int smc_setsockopt(struct socket *sock, int level, int optname,
3054 sockptr_t optval, unsigned int optlen)
3055{
3056 struct sock *sk = sock->sk;
3057 struct smc_sock *smc;
3058 int val, rc;
3059
3060 if (level == SOL_TCP && optname == TCP_ULP)
3061 return -EOPNOTSUPP;
3062 else if (level == SOL_SMC)
3063 return __smc_setsockopt(sock, level, optname, optval, optlen);
3064
3065 smc = smc_sk(sk);
3066
3067 /* generic setsockopts reaching us here always apply to the
3068 * CLC socket
3069 */
3070 mutex_lock(&smc->clcsock_release_lock);
3071 if (!smc->clcsock) {
3072 mutex_unlock(&smc->clcsock_release_lock);
3073 return -EBADF;
3074 }
3075 if (unlikely(!smc->clcsock->ops->setsockopt))
3076 rc = -EOPNOTSUPP;
3077 else
3078 rc = smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
3079 optval, optlen);
3080 if (smc->clcsock->sk->sk_err) {
3081 sk->sk_err = smc->clcsock->sk->sk_err;
3082 sk_error_report(sk);
3083 }
3084 mutex_unlock(&smc->clcsock_release_lock);
3085
3086 if (optlen < sizeof(int))
3087 return -EINVAL;
3088 if (copy_from_sockptr(&val, optval, sizeof(int)))
3089 return -EFAULT;
3090
3091 lock_sock(sk);
3092 if (rc || smc->use_fallback)
3093 goto out;
3094 switch (optname) {
3095 case TCP_FASTOPEN:
3096 case TCP_FASTOPEN_CONNECT:
3097 case TCP_FASTOPEN_KEY:
3098 case TCP_FASTOPEN_NO_COOKIE:
3099 /* option not supported by SMC */
3100 if (sk->sk_state == SMC_INIT && !smc->connect_nonblock) {
3101 rc = smc_switch_to_fallback(smc, SMC_CLC_DECL_OPTUNSUPP);
3102 } else {
3103 rc = -EINVAL;
3104 }
3105 break;
3106 case TCP_NODELAY:
3107 if (sk->sk_state != SMC_INIT &&
3108 sk->sk_state != SMC_LISTEN &&
3109 sk->sk_state != SMC_CLOSED) {
3110 if (val) {
3111 SMC_STAT_INC(smc, ndly_cnt);
3112 smc_tx_pending(&smc->conn);
3113 cancel_delayed_work(&smc->conn.tx_work);
3114 }
3115 }
3116 break;
3117 case TCP_CORK:
3118 if (sk->sk_state != SMC_INIT &&
3119 sk->sk_state != SMC_LISTEN &&
3120 sk->sk_state != SMC_CLOSED) {
3121 if (!val) {
3122 SMC_STAT_INC(smc, cork_cnt);
3123 smc_tx_pending(&smc->conn);
3124 cancel_delayed_work(&smc->conn.tx_work);
3125 }
3126 }
3127 break;
3128 case TCP_DEFER_ACCEPT:
3129 smc->sockopt_defer_accept = val;
3130 break;
3131 default:
3132 break;
3133 }
3134out:
3135 release_sock(sk);
3136
3137 return rc;
3138}
3139
3140int smc_getsockopt(struct socket *sock, int level, int optname,
3141 char __user *optval, int __user *optlen)
3142{
3143 struct smc_sock *smc;
3144 int rc;
3145
3146 if (level == SOL_SMC)
3147 return __smc_getsockopt(sock, level, optname, optval, optlen);
3148
3149 smc = smc_sk(sock->sk);
3150 mutex_lock(&smc->clcsock_release_lock);
3151 if (!smc->clcsock) {
3152 mutex_unlock(&smc->clcsock_release_lock);
3153 return -EBADF;
3154 }
3155 /* socket options apply to the CLC socket */
3156 if (unlikely(!smc->clcsock->ops->getsockopt)) {
3157 mutex_unlock(&smc->clcsock_release_lock);
3158 return -EOPNOTSUPP;
3159 }
3160 rc = smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
3161 optval, optlen);
3162 mutex_unlock(&smc->clcsock_release_lock);
3163 return rc;
3164}
3165
3166int smc_ioctl(struct socket *sock, unsigned int cmd,
3167 unsigned long arg)
3168{
3169 union smc_host_cursor cons, urg;
3170 struct smc_connection *conn;
3171 struct smc_sock *smc;
3172 int answ;
3173
3174 smc = smc_sk(sock->sk);
3175 conn = &smc->conn;
3176 lock_sock(&smc->sk);
3177 if (smc->use_fallback) {
3178 if (!smc->clcsock) {
3179 release_sock(&smc->sk);
3180 return -EBADF;
3181 }
3182 answ = smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
3183 release_sock(&smc->sk);
3184 return answ;
3185 }
3186 switch (cmd) {
3187 case SIOCINQ: /* same as FIONREAD */
3188 if (smc->sk.sk_state == SMC_LISTEN) {
3189 release_sock(&smc->sk);
3190 return -EINVAL;
3191 }
3192 if (smc->sk.sk_state == SMC_INIT ||
3193 smc->sk.sk_state == SMC_CLOSED)
3194 answ = 0;
3195 else
3196 answ = atomic_read(&smc->conn.bytes_to_rcv);
3197 break;
3198 case SIOCOUTQ:
3199 /* output queue size (not send + not acked) */
3200 if (smc->sk.sk_state == SMC_LISTEN) {
3201 release_sock(&smc->sk);
3202 return -EINVAL;
3203 }
3204 if (smc->sk.sk_state == SMC_INIT ||
3205 smc->sk.sk_state == SMC_CLOSED)
3206 answ = 0;
3207 else
3208 answ = smc->conn.sndbuf_desc->len -
3209 atomic_read(&smc->conn.sndbuf_space);
3210 break;
3211 case SIOCOUTQNSD:
3212 /* output queue size (not send only) */
3213 if (smc->sk.sk_state == SMC_LISTEN) {
3214 release_sock(&smc->sk);
3215 return -EINVAL;
3216 }
3217 if (smc->sk.sk_state == SMC_INIT ||
3218 smc->sk.sk_state == SMC_CLOSED)
3219 answ = 0;
3220 else
3221 answ = smc_tx_prepared_sends(&smc->conn);
3222 break;
3223 case SIOCATMARK:
3224 if (smc->sk.sk_state == SMC_LISTEN) {
3225 release_sock(&smc->sk);
3226 return -EINVAL;
3227 }
3228 if (smc->sk.sk_state == SMC_INIT ||
3229 smc->sk.sk_state == SMC_CLOSED) {
3230 answ = 0;
3231 } else {
3232 smc_curs_copy(&cons, &conn->local_tx_ctrl.cons, conn);
3233 smc_curs_copy(&urg, &conn->urg_curs, conn);
3234 answ = smc_curs_diff(conn->rmb_desc->len,
3235 &cons, &urg) == 1;
3236 }
3237 break;
3238 default:
3239 release_sock(&smc->sk);
3240 return -ENOIOCTLCMD;
3241 }
3242 release_sock(&smc->sk);
3243
3244 return put_user(answ, (int __user *)arg);
3245}
3246
3247/* Map the affected portions of the rmbe into an spd, note the number of bytes
3248 * to splice in conn->splice_pending, and press 'go'. Delays consumer cursor
3249 * updates till whenever a respective page has been fully processed.
3250 * Note that subsequent recv() calls have to wait till all splice() processing
3251 * completed.
3252 */
3253ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
3254 struct pipe_inode_info *pipe, size_t len,
3255 unsigned int flags)
3256{
3257 struct sock *sk = sock->sk;
3258 struct smc_sock *smc;
3259 int rc = -ENOTCONN;
3260
3261 smc = smc_sk(sk);
3262 lock_sock(sk);
3263 if (sk->sk_state == SMC_CLOSED && (sk->sk_shutdown & RCV_SHUTDOWN)) {
3264 /* socket was connected before, no more data to read */
3265 rc = 0;
3266 goto out;
3267 }
3268 if (sk->sk_state == SMC_INIT ||
3269 sk->sk_state == SMC_LISTEN ||
3270 sk->sk_state == SMC_CLOSED)
3271 goto out;
3272
3273 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
3274 rc = 0;
3275 goto out;
3276 }
3277
3278 if (smc->use_fallback) {
3279 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
3280 pipe, len, flags);
3281 } else {
3282 if (*ppos) {
3283 rc = -ESPIPE;
3284 goto out;
3285 }
3286 if (flags & SPLICE_F_NONBLOCK)
3287 flags = MSG_DONTWAIT;
3288 else
3289 flags = 0;
3290 SMC_STAT_INC(smc, splice_cnt);
3291 rc = smc_rx_recvmsg(smc, NULL, pipe, len, flags);
3292 }
3293out:
3294 release_sock(sk);
3295
3296 return rc;
3297}
3298
3299/* must look like tcp */
3300static const struct proto_ops smc_sock_ops = {
3301 .family = PF_SMC,
3302 .owner = THIS_MODULE,
3303 .release = smc_release,
3304 .bind = smc_bind,
3305 .connect = smc_connect,
3306 .socketpair = sock_no_socketpair,
3307 .accept = smc_accept,
3308 .getname = smc_getname,
3309 .poll = smc_poll,
3310 .ioctl = smc_ioctl,
3311 .listen = smc_listen,
3312 .shutdown = smc_shutdown,
3313 .setsockopt = smc_setsockopt,
3314 .getsockopt = smc_getsockopt,
3315 .sendmsg = smc_sendmsg,
3316 .recvmsg = smc_recvmsg,
3317 .mmap = sock_no_mmap,
3318 .splice_read = smc_splice_read,
3319};
3320
3321int smc_create_clcsk(struct net *net, struct sock *sk, int family)
3322{
3323 struct smc_sock *smc = smc_sk(sk);
3324 int rc;
3325
3326 rc = sock_create_kern(net, family, SOCK_STREAM, IPPROTO_TCP,
3327 &smc->clcsock);
3328 if (rc)
3329 return rc;
3330
3331 /* smc_clcsock_release() does not wait smc->clcsock->sk's
3332 * destruction; its sk_state might not be TCP_CLOSE after
3333 * smc->sk is close()d, and TCP timers can be fired later,
3334 * which need net ref.
3335 */
3336 sk = smc->clcsock->sk;
3337 sk_net_refcnt_upgrade(sk);
3338 return 0;
3339}
3340
3341static int __smc_create(struct net *net, struct socket *sock, int protocol,
3342 int kern, struct socket *clcsock)
3343{
3344 int family = (protocol == SMCPROTO_SMC6) ? PF_INET6 : PF_INET;
3345 struct smc_sock *smc;
3346 struct sock *sk;
3347 int rc;
3348
3349 rc = -ESOCKTNOSUPPORT;
3350 if (sock->type != SOCK_STREAM)
3351 goto out;
3352
3353 rc = -EPROTONOSUPPORT;
3354 if (protocol != SMCPROTO_SMC && protocol != SMCPROTO_SMC6)
3355 goto out;
3356
3357 rc = -ENOBUFS;
3358 sock->ops = &smc_sock_ops;
3359 sock->state = SS_UNCONNECTED;
3360 sk = smc_sock_alloc(net, sock, protocol);
3361 if (!sk)
3362 goto out;
3363
3364 /* create internal TCP socket for CLC handshake and fallback */
3365 smc = smc_sk(sk);
3366
3367 rc = 0;
3368 if (clcsock)
3369 smc->clcsock = clcsock;
3370 else
3371 rc = smc_create_clcsk(net, sk, family);
3372
3373 if (rc) {
3374 sk_common_release(sk);
3375 sock->sk = NULL;
3376 }
3377out:
3378 return rc;
3379}
3380
3381static int smc_create(struct net *net, struct socket *sock, int protocol,
3382 int kern)
3383{
3384 return __smc_create(net, sock, protocol, kern, NULL);
3385}
3386
3387static const struct net_proto_family smc_sock_family_ops = {
3388 .family = PF_SMC,
3389 .owner = THIS_MODULE,
3390 .create = smc_create,
3391};
3392
3393static int smc_ulp_init(struct sock *sk)
3394{
3395 struct socket *tcp = sk->sk_socket;
3396 struct net *net = sock_net(sk);
3397 struct socket *smcsock;
3398 int protocol, ret;
3399
3400 /* only TCP can be replaced */
3401 if (tcp->type != SOCK_STREAM || sk->sk_protocol != IPPROTO_TCP ||
3402 (sk->sk_family != AF_INET && sk->sk_family != AF_INET6))
3403 return -ESOCKTNOSUPPORT;
3404 /* don't handle wq now */
3405 if (tcp->state != SS_UNCONNECTED || !tcp->file || tcp->wq.fasync_list)
3406 return -ENOTCONN;
3407
3408 if (sk->sk_family == AF_INET)
3409 protocol = SMCPROTO_SMC;
3410 else
3411 protocol = SMCPROTO_SMC6;
3412
3413 smcsock = sock_alloc();
3414 if (!smcsock)
3415 return -ENFILE;
3416
3417 smcsock->type = SOCK_STREAM;
3418 __module_get(THIS_MODULE); /* tried in __tcp_ulp_find_autoload */
3419 ret = __smc_create(net, smcsock, protocol, 1, tcp);
3420 if (ret) {
3421 sock_release(smcsock); /* module_put() which ops won't be NULL */
3422 return ret;
3423 }
3424
3425 /* replace tcp socket to smc */
3426 smcsock->file = tcp->file;
3427 smcsock->file->private_data = smcsock;
3428 smcsock->file->f_inode = SOCK_INODE(smcsock); /* replace inode when sock_close */
3429 smcsock->file->f_path.dentry->d_inode = SOCK_INODE(smcsock); /* dput() in __fput */
3430 tcp->file = NULL;
3431
3432 return ret;
3433}
3434
3435static void smc_ulp_clone(const struct request_sock *req, struct sock *newsk,
3436 const gfp_t priority)
3437{
3438 struct inet_connection_sock *icsk = inet_csk(newsk);
3439
3440 /* don't inherit ulp ops to child when listen */
3441 icsk->icsk_ulp_ops = NULL;
3442}
3443
3444static struct tcp_ulp_ops smc_ulp_ops __read_mostly = {
3445 .name = "smc",
3446 .owner = THIS_MODULE,
3447 .init = smc_ulp_init,
3448 .clone = smc_ulp_clone,
3449};
3450
3451unsigned int smc_net_id;
3452
3453static __net_init int smc_net_init(struct net *net)
3454{
3455 int rc;
3456
3457 rc = smc_sysctl_net_init(net);
3458 if (rc)
3459 return rc;
3460 return smc_pnet_net_init(net);
3461}
3462
3463static void __net_exit smc_net_exit(struct net *net)
3464{
3465 smc_sysctl_net_exit(net);
3466 smc_pnet_net_exit(net);
3467}
3468
3469static __net_init int smc_net_stat_init(struct net *net)
3470{
3471 return smc_stats_init(net);
3472}
3473
3474static void __net_exit smc_net_stat_exit(struct net *net)
3475{
3476 smc_stats_exit(net);
3477}
3478
3479static struct pernet_operations smc_net_ops = {
3480 .init = smc_net_init,
3481 .exit = smc_net_exit,
3482 .id = &smc_net_id,
3483 .size = sizeof(struct smc_net),
3484};
3485
3486static struct pernet_operations smc_net_stat_ops = {
3487 .init = smc_net_stat_init,
3488 .exit = smc_net_stat_exit,
3489};
3490
3491static int __init smc_init(void)
3492{
3493 int rc;
3494
3495 rc = register_pernet_subsys(&smc_net_ops);
3496 if (rc)
3497 return rc;
3498
3499 rc = register_pernet_subsys(&smc_net_stat_ops);
3500 if (rc)
3501 goto out_pernet_subsys;
3502
3503 rc = smc_ism_init();
3504 if (rc)
3505 goto out_pernet_subsys_stat;
3506 smc_clc_init();
3507
3508 rc = smc_nl_init();
3509 if (rc)
3510 goto out_ism;
3511
3512 rc = smc_pnet_init();
3513 if (rc)
3514 goto out_nl;
3515
3516 rc = -ENOMEM;
3517
3518 smc_tcp_ls_wq = alloc_workqueue("smc_tcp_ls_wq", 0, 0);
3519 if (!smc_tcp_ls_wq)
3520 goto out_pnet;
3521
3522 smc_hs_wq = alloc_workqueue("smc_hs_wq", 0, 0);
3523 if (!smc_hs_wq)
3524 goto out_alloc_tcp_ls_wq;
3525
3526 smc_close_wq = alloc_workqueue("smc_close_wq", 0, 0);
3527 if (!smc_close_wq)
3528 goto out_alloc_hs_wq;
3529
3530 rc = smc_core_init();
3531 if (rc) {
3532 pr_err("%s: smc_core_init fails with %d\n", __func__, rc);
3533 goto out_alloc_wqs;
3534 }
3535
3536 rc = smc_llc_init();
3537 if (rc) {
3538 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
3539 goto out_core;
3540 }
3541
3542 rc = smc_cdc_init();
3543 if (rc) {
3544 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
3545 goto out_core;
3546 }
3547
3548 rc = proto_register(&smc_proto, 1);
3549 if (rc) {
3550 pr_err("%s: proto_register(v4) fails with %d\n", __func__, rc);
3551 goto out_core;
3552 }
3553
3554 rc = proto_register(&smc_proto6, 1);
3555 if (rc) {
3556 pr_err("%s: proto_register(v6) fails with %d\n", __func__, rc);
3557 goto out_proto;
3558 }
3559
3560 rc = sock_register(&smc_sock_family_ops);
3561 if (rc) {
3562 pr_err("%s: sock_register fails with %d\n", __func__, rc);
3563 goto out_proto6;
3564 }
3565 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
3566 INIT_HLIST_HEAD(&smc_v6_hashinfo.ht);
3567
3568 rc = smc_ib_register_client();
3569 if (rc) {
3570 pr_err("%s: ib_register fails with %d\n", __func__, rc);
3571 goto out_sock;
3572 }
3573
3574 rc = smc_loopback_init();
3575 if (rc) {
3576 pr_err("%s: smc_loopback_init fails with %d\n", __func__, rc);
3577 goto out_ib;
3578 }
3579
3580 rc = tcp_register_ulp(&smc_ulp_ops);
3581 if (rc) {
3582 pr_err("%s: tcp_ulp_register fails with %d\n", __func__, rc);
3583 goto out_lo;
3584 }
3585 rc = smc_inet_init();
3586 if (rc) {
3587 pr_err("%s: smc_inet_init fails with %d\n", __func__, rc);
3588 goto out_ulp;
3589 }
3590 static_branch_enable(&tcp_have_smc);
3591 return 0;
3592out_ulp:
3593 tcp_unregister_ulp(&smc_ulp_ops);
3594out_lo:
3595 smc_loopback_exit();
3596out_ib:
3597 smc_ib_unregister_client();
3598out_sock:
3599 sock_unregister(PF_SMC);
3600out_proto6:
3601 proto_unregister(&smc_proto6);
3602out_proto:
3603 proto_unregister(&smc_proto);
3604out_core:
3605 smc_core_exit();
3606out_alloc_wqs:
3607 destroy_workqueue(smc_close_wq);
3608out_alloc_hs_wq:
3609 destroy_workqueue(smc_hs_wq);
3610out_alloc_tcp_ls_wq:
3611 destroy_workqueue(smc_tcp_ls_wq);
3612out_pnet:
3613 smc_pnet_exit();
3614out_nl:
3615 smc_nl_exit();
3616out_ism:
3617 smc_clc_exit();
3618 smc_ism_exit();
3619out_pernet_subsys_stat:
3620 unregister_pernet_subsys(&smc_net_stat_ops);
3621out_pernet_subsys:
3622 unregister_pernet_subsys(&smc_net_ops);
3623
3624 return rc;
3625}
3626
3627static void __exit smc_exit(void)
3628{
3629 static_branch_disable(&tcp_have_smc);
3630 smc_inet_exit();
3631 tcp_unregister_ulp(&smc_ulp_ops);
3632 sock_unregister(PF_SMC);
3633 smc_core_exit();
3634 smc_loopback_exit();
3635 smc_ib_unregister_client();
3636 smc_ism_exit();
3637 destroy_workqueue(smc_close_wq);
3638 destroy_workqueue(smc_tcp_ls_wq);
3639 destroy_workqueue(smc_hs_wq);
3640 proto_unregister(&smc_proto6);
3641 proto_unregister(&smc_proto);
3642 smc_pnet_exit();
3643 smc_nl_exit();
3644 smc_clc_exit();
3645 unregister_pernet_subsys(&smc_net_stat_ops);
3646 unregister_pernet_subsys(&smc_net_ops);
3647 rcu_barrier();
3648}
3649
3650module_init(smc_init);
3651module_exit(smc_exit);
3652
3653MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
3654MODULE_DESCRIPTION("smc socket address family");
3655MODULE_LICENSE("GPL");
3656MODULE_ALIAS_NETPROTO(PF_SMC);
3657MODULE_ALIAS_TCP_ULP("smc");
3658/* 256 for IPPROTO_SMC and 1 for SOCK_STREAM */
3659MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET, 256, 1);
3660#if IS_ENABLED(CONFIG_IPV6)
3661MODULE_ALIAS_NET_PF_PROTO_TYPE(PF_INET6, 256, 1);
3662#endif /* CONFIG_IPV6 */
3663MODULE_ALIAS_GENL_FAMILY(SMC_GENL_FAMILY_NAME);