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