projects / linux-2.6-block.git / blame
? search:
summary | shortlog | log | commit | commitdiff | tree
blob | blame (incremental) | history | HEAD
tcp: TCP experimental option for SMC
[linux-2.6-block.git] / net / smc / af_smc.c
CommitLineData
ac713874
UB		 1/*
2 * Shared Memory Communications over RDMA (SMC-R) and RoCE
3 *
4 * AF_SMC protocol family socket handler keeping the AF_INET sock address type
5 * applies to SOCK_STREAM sockets only
6 * offers an alternative communication option for TCP-protocol sockets
7 * applicable with RoCE-cards only
8 *
a046d57d
UB		 9 * Initial restrictions:
10 * - non-blocking connect postponed
11 * - IPv6 support postponed
12 * - support for alternate links postponed
13 * - partial support for non-blocking sockets only
14 * - support for urgent data postponed
15 *
ac713874
UB		 16 * Copyright IBM Corp. 2016
17 *
18 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
19 * based on prototype from Frank Blaschka
20 */
21
22#define KMSG_COMPONENT "smc"
23#define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
24
25#include <linux/module.h>
26#include <linux/socket.h>
a046d57d
UB		 27#include <linux/inetdevice.h>
28#include <linux/workqueue.h>
5f08318f 29#include <linux/in.h>
c3edc401
IM		 30#include <linux/sched/signal.h>
31
ac713874 32#include <net/sock.h>
a046d57d 33#include <net/tcp.h>
f16a7dd5 34#include <net/smc.h>
ac713874
UB		 35
36#include "smc.h"
a046d57d 37#include "smc_clc.h"
9bf9abea 38#include "smc_llc.h"
5f08318f 39#include "smc_cdc.h"
0cfdd8f9 40#include "smc_core.h"
a4cf0443 41#include "smc_ib.h"
6812baab 42#include "smc_pnet.h"
e6727f39 43#include "smc_tx.h"
952310cc 44#include "smc_rx.h"
b38d7324 45#include "smc_close.h"
ac713874 46
0cfdd8f9
UB		 47static DEFINE_MUTEX(smc_create_lgr_pending); /* serialize link group
48 * creation
49 */
50
51struct smc_lgr_list smc_lgr_list = { /* established link groups */
52 .lock = __SPIN_LOCK_UNLOCKED(smc_lgr_list.lock),
53 .list = LIST_HEAD_INIT(smc_lgr_list.list),
54};
55
a046d57d
UB		 56static void smc_tcp_listen_work(struct work_struct *);
57
ac713874
UB		 58static void smc_set_keepalive(struct sock *sk, int val)
59{
60 struct smc_sock *smc = smc_sk(sk);
61
62 smc->clcsock->sk->sk_prot->keepalive(smc->clcsock->sk, val);
63}
64
f16a7dd5
UB		 65static struct smc_hashinfo smc_v4_hashinfo = {
66 .lock = __RW_LOCK_UNLOCKED(smc_v4_hashinfo.lock),
67};
68
69int smc_hash_sk(struct sock *sk)
70{
71 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
72 struct hlist_head *head;
73
74 head = &h->ht;
75
76 write_lock_bh(&h->lock);
77 sk_add_node(sk, head);
78 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, 1);
79 write_unlock_bh(&h->lock);
80
81 return 0;
82}
83EXPORT_SYMBOL_GPL(smc_hash_sk);
84
85void smc_unhash_sk(struct sock *sk)
86{
87 struct smc_hashinfo *h = sk->sk_prot->h.smc_hash;
88
89 write_lock_bh(&h->lock);
90 if (sk_del_node_init(sk))
91 sock_prot_inuse_add(sock_net(sk), sk->sk_prot, -1);
92 write_unlock_bh(&h->lock);
93}
94EXPORT_SYMBOL_GPL(smc_unhash_sk);
95
96struct proto smc_proto = {
ac713874
UB		 97 .name = "SMC",
98 .owner = THIS_MODULE,
99 .keepalive = smc_set_keepalive,
f16a7dd5
UB		 100 .hash = smc_hash_sk,
101 .unhash = smc_unhash_sk,
ac713874 102 .obj_size = sizeof(struct smc_sock),
f16a7dd5 103 .h.smc_hash = &smc_v4_hashinfo,
5f0d5a3a 104 .slab_flags = SLAB_TYPESAFE_BY_RCU,
ac713874 105};
f16a7dd5 106EXPORT_SYMBOL_GPL(smc_proto);
ac713874
UB		 107
108static int smc_release(struct socket *sock)
109{
110 struct sock *sk = sock->sk;
111 struct smc_sock *smc;
b38d7324 112 int rc = 0;
ac713874
UB		 113
114 if (!sk)
115 goto out;
116
117 smc = smc_sk(sk);
b38d7324
UB		 118 sock_hold(sk);
119 if (sk->sk_state == SMC_LISTEN)
120 /* smc_close_non_accepted() is called and acquires
121 * sock lock for child sockets again
122 */
123 lock_sock_nested(sk, SINGLE_DEPTH_NESTING);
124 else
125 lock_sock(sk);
ac713874 126
b38d7324
UB		 127 if (smc->use_fallback) {
128 sk->sk_state = SMC_CLOSED;
129 sk->sk_state_change(sk);
130 } else {
131 rc = smc_close_active(smc);
132 sock_set_flag(sk, SOCK_DEAD);
133 sk->sk_shutdown |= SHUTDOWN_MASK;
134 }
ac713874
UB		 135 if (smc->clcsock) {
136 sock_release(smc->clcsock);
137 smc->clcsock = NULL;
138 }
139
140 /* detach socket */
141 sock_orphan(sk);
142 sock->sk = NULL;
b38d7324
UB		 143 if (smc->use_fallback) {
144 schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
145 } else if (sk->sk_state == SMC_CLOSED) {
146 smc_conn_free(&smc->conn);
147 schedule_delayed_work(&smc->sock_put_work,
148 SMC_CLOSE_SOCK_PUT_DELAY);
149 }
ac713874
UB		 150 release_sock(sk);
151
152 sock_put(sk);
153out:
b38d7324 154 return rc;
ac713874
UB		 155}
156
157static void smc_destruct(struct sock *sk)
158{
159 if (sk->sk_state != SMC_CLOSED)
160 return;
161 if (!sock_flag(sk, SOCK_DEAD))
162 return;
163
164 sk_refcnt_debug_dec(sk);
165}
166
167static struct sock *smc_sock_alloc(struct net *net, struct socket *sock)
168{
169 struct smc_sock *smc;
170 struct sock *sk;
171
172 sk = sk_alloc(net, PF_SMC, GFP_KERNEL, &smc_proto, 0);
173 if (!sk)
174 return NULL;
175
176 sock_init_data(sock, sk); /* sets sk_refcnt to 1 */
177 sk->sk_state = SMC_INIT;
178 sk->sk_destruct = smc_destruct;
179 sk->sk_protocol = SMCPROTO_SMC;
ac713874 180 smc = smc_sk(sk);
a046d57d
UB		 181 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
182 INIT_LIST_HEAD(&smc->accept_q);
183 spin_lock_init(&smc->accept_q_lock);
b38d7324 184 INIT_DELAYED_WORK(&smc->sock_put_work, smc_close_sock_put_work);
f16a7dd5 185 sk->sk_prot->hash(sk);
a046d57d 186 sk_refcnt_debug_inc(sk);
ac713874
UB		 187
188 return sk;
189}
190
191static int smc_bind(struct socket *sock, struct sockaddr *uaddr,
192 int addr_len)
193{
194 struct sockaddr_in *addr = (struct sockaddr_in *)uaddr;
195 struct sock *sk = sock->sk;
196 struct smc_sock *smc;
197 int rc;
198
199 smc = smc_sk(sk);
200
201 /* replicate tests from inet_bind(), to be safe wrt. future changes */
202 rc = -EINVAL;
203 if (addr_len < sizeof(struct sockaddr_in))
204 goto out;
205
206 rc = -EAFNOSUPPORT;
207 /* accept AF_UNSPEC (mapped to AF_INET) only if s_addr is INADDR_ANY */
208 if ((addr->sin_family != AF_INET) &&
209 ((addr->sin_family != AF_UNSPEC) ||
210 (addr->sin_addr.s_addr != htonl(INADDR_ANY))))
211 goto out;
212
213 lock_sock(sk);
214
215 /* Check if socket is already active */
216 rc = -EINVAL;
217 if (sk->sk_state != SMC_INIT)
218 goto out_rel;
219
220 smc->clcsock->sk->sk_reuse = sk->sk_reuse;
221 rc = kernel_bind(smc->clcsock, uaddr, addr_len);
222
223out_rel:
224 release_sock(sk);
225out:
226 return rc;
227}
228
229static void smc_copy_sock_settings(struct sock *nsk, struct sock *osk,
230 unsigned long mask)
231{
232 /* options we don't get control via setsockopt for */
233 nsk->sk_type = osk->sk_type;
234 nsk->sk_sndbuf = osk->sk_sndbuf;
235 nsk->sk_rcvbuf = osk->sk_rcvbuf;
236 nsk->sk_sndtimeo = osk->sk_sndtimeo;
237 nsk->sk_rcvtimeo = osk->sk_rcvtimeo;
238 nsk->sk_mark = osk->sk_mark;
239 nsk->sk_priority = osk->sk_priority;
240 nsk->sk_rcvlowat = osk->sk_rcvlowat;
241 nsk->sk_bound_dev_if = osk->sk_bound_dev_if;
242 nsk->sk_err = osk->sk_err;
243
244 nsk->sk_flags &= ~mask;
245 nsk->sk_flags |= osk->sk_flags & mask;
246}
247
248#define SK_FLAGS_SMC_TO_CLC ((1UL << SOCK_URGINLINE) | \
249 (1UL << SOCK_KEEPOPEN) | \
250 (1UL << SOCK_LINGER) | \
251 (1UL << SOCK_BROADCAST) | \
252 (1UL << SOCK_TIMESTAMP) | \
253 (1UL << SOCK_DBG) | \
254 (1UL << SOCK_RCVTSTAMP) | \
255 (1UL << SOCK_RCVTSTAMPNS) | \
256 (1UL << SOCK_LOCALROUTE) | \
257 (1UL << SOCK_TIMESTAMPING_RX_SOFTWARE) | \
258 (1UL << SOCK_RXQ_OVFL) | \
259 (1UL << SOCK_WIFI_STATUS) | \
260 (1UL << SOCK_NOFCS) | \
261 (1UL << SOCK_FILTER_LOCKED))
262/* copy only relevant settings and flags of SOL_SOCKET level from smc to
263 * clc socket (since smc is not called for these options from net/core)
264 */
265static void smc_copy_sock_settings_to_clc(struct smc_sock *smc)
266{
267 smc_copy_sock_settings(smc->clcsock->sk, &smc->sk, SK_FLAGS_SMC_TO_CLC);
268}
269
270#define SK_FLAGS_CLC_TO_SMC ((1UL << SOCK_URGINLINE) | \
271 (1UL << SOCK_KEEPOPEN) | \
272 (1UL << SOCK_LINGER) | \
273 (1UL << SOCK_DBG))
274/* copy only settings and flags relevant for smc from clc to smc socket */
275static void smc_copy_sock_settings_to_smc(struct smc_sock *smc)
276{
277 smc_copy_sock_settings(&smc->sk, smc->clcsock->sk, SK_FLAGS_CLC_TO_SMC);
278}
279
a046d57d
UB		 280/* determine subnet and mask of internal TCP socket */
281int smc_netinfo_by_tcpsk(struct socket *clcsock,
282 __be32 *subnet, u8 *prefix_len)
283{
284 struct dst_entry *dst = sk_dst_get(clcsock->sk);
731b0085 285 struct in_device *in_dev;
a046d57d
UB		 286 struct sockaddr_in addr;
287 int rc = -ENOENT;
288 int len;
289
290 if (!dst) {
291 rc = -ENOTCONN;
292 goto out;
293 }
294 if (!dst->dev) {
295 rc = -ENODEV;
296 goto out_rel;
297 }
298
299 /* get address to which the internal TCP socket is bound */
300 kernel_getsockname(clcsock, (struct sockaddr *)&addr, &len);
301 /* analyze IPv4 specific data of net_device belonging to TCP socket */
731b0085
UB		 302 rcu_read_lock();
303 in_dev = __in_dev_get_rcu(dst->dev);
304 for_ifa(in_dev) {
305 if (!inet_ifa_match(addr.sin_addr.s_addr, ifa))
a046d57d
UB		 306 continue;
307 *prefix_len = inet_mask_len(ifa->ifa_mask);
308 *subnet = ifa->ifa_address & ifa->ifa_mask;
309 rc = 0;
310 break;
731b0085
UB		 311 } endfor_ifa(in_dev);
312 rcu_read_unlock();
a046d57d
UB		 313
314out_rel:
315 dst_release(dst);
316out:
317 return rc;
318}
319
9bf9abea
UB		 320static int smc_clnt_conf_first_link(struct smc_sock *smc, union ib_gid *gid)
321{
322 struct smc_link_group *lgr = smc->conn.lgr;
323 struct smc_link *link;
324 int rest;
325 int rc;
326
327 link = &lgr->lnk[SMC_SINGLE_LINK];
328 /* receive CONFIRM LINK request from server over RoCE fabric */
329 rest = wait_for_completion_interruptible_timeout(
330 &link->llc_confirm,
331 SMC_LLC_WAIT_FIRST_TIME);
332 if (rest <= 0) {
333 struct smc_clc_msg_decline dclc;
334
335 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
336 SMC_CLC_DECLINE);
337 return rc;
338 }
339
340 rc = smc_ib_modify_qp_rts(link);
341 if (rc)
342 return SMC_CLC_DECL_INTERR;
343
344 smc_wr_remember_qp_attr(link);
652a1e41
UB		 345
346 rc = smc_wr_reg_send(link,
347 smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
348 if (rc)
349 return SMC_CLC_DECL_INTERR;
350
9bf9abea
UB		 351 /* send CONFIRM LINK response over RoCE fabric */
352 rc = smc_llc_send_confirm_link(link,
353 link->smcibdev->mac[link->ibport - 1],
354 gid, SMC_LLC_RESP);
355 if (rc < 0)
356 return SMC_CLC_DECL_TCL;
357
358 return rc;
359}
360
0cfdd8f9
UB		 361static void smc_conn_save_peer_info(struct smc_sock *smc,
362 struct smc_clc_msg_accept_confirm *clc)
363{
364 smc->conn.peer_conn_idx = clc->conn_idx;
5f08318f 365 smc->conn.local_tx_ctrl.token = ntohl(clc->rmbe_alert_token);
cd6851f3
UB		 366 smc->conn.peer_rmbe_size = smc_uncompress_bufsize(clc->rmbe_size);
367 atomic_set(&smc->conn.peer_rmbe_space, smc->conn.peer_rmbe_size);
0cfdd8f9
UB		 368}
369
370static void smc_link_save_peer_info(struct smc_link *link,
371 struct smc_clc_msg_accept_confirm *clc)
372{
373 link->peer_qpn = ntoh24(clc->qpn);
374 memcpy(link->peer_gid, clc->lcl.gid, SMC_GID_SIZE);
375 memcpy(link->peer_mac, clc->lcl.mac, sizeof(link->peer_mac));
376 link->peer_psn = ntoh24(clc->psn);
377 link->peer_mtu = clc->qp_mtu;
378}
379
a046d57d
UB		 380/* setup for RDMA connection of client */
381static int smc_connect_rdma(struct smc_sock *smc)
382{
0cfdd8f9 383 struct sockaddr_in *inaddr = (struct sockaddr_in *)smc->addr;
a046d57d 384 struct smc_clc_msg_accept_confirm aclc;
0cfdd8f9 385 int local_contact = SMC_FIRST_CONTACT;
a046d57d 386 struct smc_ib_device *smcibdev;
0cfdd8f9
UB		 387 struct smc_link *link;
388 u8 srv_first_contact;
a046d57d
UB		 389 int reason_code = 0;
390 int rc = 0;
391 u8 ibport;
392
393 /* IPSec connections opt out of SMC-R optimizations */
394 if (using_ipsec(smc)) {
395 reason_code = SMC_CLC_DECL_IPSEC;
396 goto decline_rdma;
397 }
398
399 /* PNET table look up: search active ib_device and port
400 * within same PNETID that also contains the ethernet device
401 * used for the internal TCP socket
402 */
403 smc_pnet_find_roce_resource(smc->clcsock->sk, &smcibdev, &ibport);
404 if (!smcibdev) {
405 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
406 goto decline_rdma;
407 }
408
409 /* do inband token exchange */
410 reason_code = smc_clc_send_proposal(smc, smcibdev, ibport);
411 if (reason_code < 0) {
412 rc = reason_code;
413 goto out_err;
414 }
415 if (reason_code > 0) /* configuration error */
416 goto decline_rdma;
417 /* receive SMC Accept CLC message */
418 reason_code = smc_clc_wait_msg(smc, &aclc, sizeof(aclc),
419 SMC_CLC_ACCEPT);
420 if (reason_code < 0) {
421 rc = reason_code;
422 goto out_err;
423 }
424 if (reason_code > 0)
425 goto decline_rdma;
426
0cfdd8f9
UB		 427 srv_first_contact = aclc.hdr.flag;
428 mutex_lock(&smc_create_lgr_pending);
429 local_contact = smc_conn_create(smc, inaddr->sin_addr.s_addr, smcibdev,
430 ibport, &aclc.lcl, srv_first_contact);
431 if (local_contact < 0) {
432 rc = local_contact;
433 if (rc == -ENOMEM)
434 reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
435 else if (rc == -ENOLINK)
436 reason_code = SMC_CLC_DECL_SYNCERR; /* synchr. error */
437 goto decline_rdma_unlock;
438 }
439 link = &smc->conn.lgr->lnk[SMC_SINGLE_LINK];
a046d57d 440
0cfdd8f9 441 smc_conn_save_peer_info(smc, &aclc);
cd6851f3 442
3e034725
UB		 443 /* create send buffer and rmb */
444 rc = smc_buf_create(smc);
cd6851f3
UB		 445 if (rc) {
446 reason_code = SMC_CLC_DECL_MEM;
447 goto decline_rdma_unlock;
448 }
449
0cfdd8f9
UB		 450 if (local_contact == SMC_FIRST_CONTACT)
451 smc_link_save_peer_info(link, &aclc);
bd4ad577
UB		 452
453 rc = smc_rmb_rtoken_handling(&smc->conn, &aclc);
454 if (rc) {
455 reason_code = SMC_CLC_DECL_INTERR;
456 goto decline_rdma_unlock;
457 }
458
46c28dbd
UB		 459 smc_close_init(smc);
460 smc_rx_init(smc);
461
bd4ad577
UB		 462 if (local_contact == SMC_FIRST_CONTACT) {
463 rc = smc_ib_ready_link(link);
464 if (rc) {
465 reason_code = SMC_CLC_DECL_INTERR;
466 goto decline_rdma_unlock;
467 }
652a1e41
UB		 468 } else {
469 struct smc_buf_desc *buf_desc = smc->conn.rmb_desc;
470
471 if (!buf_desc->reused) {
472 /* register memory region for new rmb */
473 rc = smc_wr_reg_send(link,
474 buf_desc->mr_rx[SMC_SINGLE_LINK]);
475 if (rc) {
476 reason_code = SMC_CLC_DECL_INTERR;
477 goto decline_rdma_unlock;
478 }
479 }
bd4ad577 480 }
10428dd8 481 smc_rmb_sync_sg_for_device(&smc->conn);
a046d57d
UB		 482
483 rc = smc_clc_send_confirm(smc);
484 if (rc)
0cfdd8f9 485 goto out_err_unlock;
a046d57d 486
9bf9abea
UB		 487 if (local_contact == SMC_FIRST_CONTACT) {
488 /* QP confirmation over RoCE fabric */
489 reason_code = smc_clnt_conf_first_link(
490 smc, &smcibdev->gid[ibport - 1]);
491 if (reason_code < 0) {
492 rc = reason_code;
493 goto out_err_unlock;
494 }
495 if (reason_code > 0)
496 goto decline_rdma_unlock;
497 }
a046d57d 498
0cfdd8f9 499 mutex_unlock(&smc_create_lgr_pending);
e6727f39
UB		 500 smc_tx_init(smc);
501
a046d57d
UB		 502out_connected:
503 smc_copy_sock_settings_to_clc(smc);
b38d7324
UB		 504 if (smc->sk.sk_state == SMC_INIT)
505 smc->sk.sk_state = SMC_ACTIVE;
a046d57d 506
0cfdd8f9 507 return rc ? rc : local_contact;
a046d57d 508
0cfdd8f9
UB		 509decline_rdma_unlock:
510 mutex_unlock(&smc_create_lgr_pending);
511 smc_conn_free(&smc->conn);
a046d57d
UB		 512decline_rdma:
513 /* RDMA setup failed, switch back to TCP */
514 smc->use_fallback = true;
515 if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
bfbedfd3 516 rc = smc_clc_send_decline(smc, reason_code);
a046d57d
UB		 517 if (rc < sizeof(struct smc_clc_msg_decline))
518 goto out_err;
519 }
520 goto out_connected;
521
0cfdd8f9
UB		 522out_err_unlock:
523 mutex_unlock(&smc_create_lgr_pending);
524 smc_conn_free(&smc->conn);
a046d57d
UB		 525out_err:
526 return rc;
527}
528
ac713874
UB		 529static int smc_connect(struct socket *sock, struct sockaddr *addr,
530 int alen, int flags)
531{
532 struct sock *sk = sock->sk;
533 struct smc_sock *smc;
534 int rc = -EINVAL;
535
536 smc = smc_sk(sk);
537
538 /* separate smc parameter checking to be safe */
539 if (alen < sizeof(addr->sa_family))
540 goto out_err;
541 if (addr->sa_family != AF_INET)
542 goto out_err;
a046d57d 543 smc->addr = addr; /* needed for nonblocking connect */
ac713874
UB		 544
545 lock_sock(sk);
546 switch (sk->sk_state) {
547 default:
548 goto out;
549 case SMC_ACTIVE:
550 rc = -EISCONN;
551 goto out;
552 case SMC_INIT:
553 rc = 0;
554 break;
555 }
556
557 smc_copy_sock_settings_to_clc(smc);
558 rc = kernel_connect(smc->clcsock, addr, alen, flags);
559 if (rc)
560 goto out;
561
a046d57d
UB		 562 /* setup RDMA connection */
563 rc = smc_connect_rdma(smc);
564 if (rc < 0)
565 goto out;
566 else
567 rc = 0; /* success cases including fallback */
ac713874
UB		 568
569out:
570 release_sock(sk);
571out_err:
572 return rc;
573}
574
575static int smc_clcsock_accept(struct smc_sock *lsmc, struct smc_sock **new_smc)
576{
577 struct sock *sk = &lsmc->sk;
578 struct socket *new_clcsock;
579 struct sock *new_sk;
580 int rc;
581
a046d57d 582 release_sock(&lsmc->sk);
ac713874
UB		 583 new_sk = smc_sock_alloc(sock_net(sk), NULL);
584 if (!new_sk) {
585 rc = -ENOMEM;
586 lsmc->sk.sk_err = ENOMEM;
587 *new_smc = NULL;
a046d57d 588 lock_sock(&lsmc->sk);
ac713874
UB		 589 goto out;
590 }
591 *new_smc = smc_sk(new_sk);
592
593 rc = kernel_accept(lsmc->clcsock, &new_clcsock, 0);
a046d57d
UB		 594 lock_sock(&lsmc->sk);
595 if (rc < 0) {
596 lsmc->sk.sk_err = -rc;
597 new_sk->sk_state = SMC_CLOSED;
598 sock_set_flag(new_sk, SOCK_DEAD);
f16a7dd5 599 sk->sk_prot->unhash(new_sk);
a046d57d
UB		 600 sock_put(new_sk);
601 *new_smc = NULL;
602 goto out;
603 }
604 if (lsmc->sk.sk_state == SMC_CLOSED) {
605 if (new_clcsock)
606 sock_release(new_clcsock);
607 new_sk->sk_state = SMC_CLOSED;
608 sock_set_flag(new_sk, SOCK_DEAD);
f16a7dd5 609 sk->sk_prot->unhash(new_sk);
ac713874
UB		 610 sock_put(new_sk);
611 *new_smc = NULL;
612 goto out;
613 }
614
615 (*new_smc)->clcsock = new_clcsock;
616out:
617 return rc;
618}
619
a046d57d
UB		 620/* add a just created sock to the accept queue of the listen sock as
621 * candidate for a following socket accept call from user space
622 */
623static void smc_accept_enqueue(struct sock *parent, struct sock *sk)
624{
625 struct smc_sock *par = smc_sk(parent);
626
627 sock_hold(sk);
628 spin_lock(&par->accept_q_lock);
629 list_add_tail(&smc_sk(sk)->accept_q, &par->accept_q);
630 spin_unlock(&par->accept_q_lock);
631 sk_acceptq_added(parent);
632}
633
634/* remove a socket from the accept queue of its parental listening socket */
635static void smc_accept_unlink(struct sock *sk)
636{
637 struct smc_sock *par = smc_sk(sk)->listen_smc;
638
639 spin_lock(&par->accept_q_lock);
640 list_del_init(&smc_sk(sk)->accept_q);
641 spin_unlock(&par->accept_q_lock);
642 sk_acceptq_removed(&smc_sk(sk)->listen_smc->sk);
643 sock_put(sk);
644}
645
646/* remove a sock from the accept queue to bind it to a new socket created
647 * for a socket accept call from user space
648 */
b38d7324
UB		 649struct sock *smc_accept_dequeue(struct sock *parent,
650 struct socket *new_sock)
a046d57d
UB		 651{
652 struct smc_sock *isk, *n;
653 struct sock *new_sk;
654
655 list_for_each_entry_safe(isk, n, &smc_sk(parent)->accept_q, accept_q) {
656 new_sk = (struct sock *)isk;
657
658 smc_accept_unlink(new_sk);
659 if (new_sk->sk_state == SMC_CLOSED) {
288c8390
UB		 660 new_sk->sk_prot->unhash(new_sk);
661 sock_put(new_sk);
a046d57d
UB		 662 continue;
663 }
664 if (new_sock)
665 sock_graft(new_sk, new_sock);
666 return new_sk;
667 }
668 return NULL;
669}
670
671/* clean up for a created but never accepted sock */
b38d7324 672void smc_close_non_accepted(struct sock *sk)
a046d57d
UB		 673{
674 struct smc_sock *smc = smc_sk(sk);
675
676 sock_hold(sk);
b38d7324
UB		 677 lock_sock(sk);
678 if (!sk->sk_lingertime)
679 /* wait for peer closing */
680 sk->sk_lingertime = SMC_MAX_STREAM_WAIT_TIMEOUT;
288c8390
UB		 681 if (smc->use_fallback) {
682 sk->sk_state = SMC_CLOSED;
683 } else {
b38d7324 684 smc_close_active(smc);
288c8390
UB		 685 sock_set_flag(sk, SOCK_DEAD);
686 sk->sk_shutdown |= SHUTDOWN_MASK;
687 }
a046d57d
UB		 688 if (smc->clcsock) {
689 struct socket *tcp;
690
691 tcp = smc->clcsock;
692 smc->clcsock = NULL;
693 sock_release(tcp);
694 }
b38d7324
UB		 695 if (smc->use_fallback) {
696 schedule_delayed_work(&smc->sock_put_work, TCP_TIMEWAIT_LEN);
288c8390 697 } else if (sk->sk_state == SMC_CLOSED) {
b38d7324
UB		 698 smc_conn_free(&smc->conn);
699 schedule_delayed_work(&smc->sock_put_work,
700 SMC_CLOSE_SOCK_PUT_DELAY);
701 }
702 release_sock(sk);
a046d57d
UB		 703 sock_put(sk);
704}
705
9bf9abea
UB		 706static int smc_serv_conf_first_link(struct smc_sock *smc)
707{
708 struct smc_link_group *lgr = smc->conn.lgr;
709 struct smc_link *link;
710 int rest;
711 int rc;
712
713 link = &lgr->lnk[SMC_SINGLE_LINK];
652a1e41
UB		 714
715 rc = smc_wr_reg_send(link,
716 smc->conn.rmb_desc->mr_rx[SMC_SINGLE_LINK]);
717 if (rc)
718 return SMC_CLC_DECL_INTERR;
719
9bf9abea
UB		 720 /* send CONFIRM LINK request to client over the RoCE fabric */
721 rc = smc_llc_send_confirm_link(link,
722 link->smcibdev->mac[link->ibport - 1],
723 &link->smcibdev->gid[link->ibport - 1],
724 SMC_LLC_REQ);
725 if (rc < 0)
726 return SMC_CLC_DECL_TCL;
727
728 /* receive CONFIRM LINK response from client over the RoCE fabric */
729 rest = wait_for_completion_interruptible_timeout(
730 &link->llc_confirm_resp,
731 SMC_LLC_WAIT_FIRST_TIME);
732 if (rest <= 0) {
733 struct smc_clc_msg_decline dclc;
734
735 rc = smc_clc_wait_msg(smc, &dclc, sizeof(dclc),
736 SMC_CLC_DECLINE);
737 }
738
739 return rc;
740}
741
a046d57d
UB		 742/* setup for RDMA connection of server */
743static void smc_listen_work(struct work_struct *work)
744{
745 struct smc_sock *new_smc = container_of(work, struct smc_sock,
746 smc_listen_work);
747 struct socket *newclcsock = new_smc->clcsock;
748 struct smc_sock *lsmc = new_smc->listen_smc;
749 struct smc_clc_msg_accept_confirm cclc;
0cfdd8f9 750 int local_contact = SMC_REUSE_CONTACT;
a046d57d
UB		 751 struct sock *newsmcsk = &new_smc->sk;
752 struct smc_clc_msg_proposal pclc;
753 struct smc_ib_device *smcibdev;
754 struct sockaddr_in peeraddr;
0cfdd8f9 755 struct smc_link *link;
a046d57d
UB		 756 int reason_code = 0;
757 int rc = 0, len;
758 __be32 subnet;
759 u8 prefix_len;
760 u8 ibport;
761
762 /* do inband token exchange -
763 *wait for and receive SMC Proposal CLC message
764 */
765 reason_code = smc_clc_wait_msg(new_smc, &pclc, sizeof(pclc),
766 SMC_CLC_PROPOSAL);
767 if (reason_code < 0)
768 goto out_err;
769 if (reason_code > 0)
770 goto decline_rdma;
771
772 /* IPSec connections opt out of SMC-R optimizations */
773 if (using_ipsec(new_smc)) {
774 reason_code = SMC_CLC_DECL_IPSEC;
775 goto decline_rdma;
776 }
777
778 /* PNET table look up: search active ib_device and port
779 * within same PNETID that also contains the ethernet device
780 * used for the internal TCP socket
781 */
782 smc_pnet_find_roce_resource(newclcsock->sk, &smcibdev, &ibport);
783 if (!smcibdev) {
784 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
785 goto decline_rdma;
786 }
787
788 /* determine subnet and mask from internal TCP socket */
789 rc = smc_netinfo_by_tcpsk(newclcsock, &subnet, &prefix_len);
790 if (rc) {
791 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
792 goto decline_rdma;
793 }
794 if ((pclc.outgoing_subnet != subnet) ||
795 (pclc.prefix_len != prefix_len)) {
796 reason_code = SMC_CLC_DECL_CNFERR; /* configuration error */
797 goto decline_rdma;
798 }
799
800 /* get address of the peer connected to the internal TCP socket */
801 kernel_getpeername(newclcsock, (struct sockaddr *)&peeraddr, &len);
802
0cfdd8f9
UB		 803 /* allocate connection / link group */
804 mutex_lock(&smc_create_lgr_pending);
805 local_contact = smc_conn_create(new_smc, peeraddr.sin_addr.s_addr,
806 smcibdev, ibport, &pclc.lcl, 0);
0cfdd8f9
UB		 807 if (local_contact < 0) {
808 rc = local_contact;
809 if (rc == -ENOMEM)
810 reason_code = SMC_CLC_DECL_MEM;/* insufficient memory*/
145686ba 811 goto decline_rdma_unlock;
0cfdd8f9
UB		 812 }
813 link = &new_smc->conn.lgr->lnk[SMC_SINGLE_LINK];
a046d57d 814
3e034725
UB		 815 /* create send buffer and rmb */
816 rc = smc_buf_create(new_smc);
cd6851f3
UB		 817 if (rc) {
818 reason_code = SMC_CLC_DECL_MEM;
145686ba 819 goto decline_rdma_unlock;
cd6851f3 820 }
a046d57d 821
46c28dbd
UB		 822 smc_close_init(new_smc);
823 smc_rx_init(new_smc);
824
652a1e41
UB		 825 if (local_contact != SMC_FIRST_CONTACT) {
826 struct smc_buf_desc *buf_desc = new_smc->conn.rmb_desc;
827
828 if (!buf_desc->reused) {
829 /* register memory region for new rmb */
830 rc = smc_wr_reg_send(link,
831 buf_desc->mr_rx[SMC_SINGLE_LINK]);
832 if (rc) {
833 reason_code = SMC_CLC_DECL_INTERR;
145686ba 834 goto decline_rdma_unlock;
652a1e41
UB		 835 }
836 }
837 }
10428dd8 838 smc_rmb_sync_sg_for_device(&new_smc->conn);
652a1e41 839
0cfdd8f9 840 rc = smc_clc_send_accept(new_smc, local_contact);
a046d57d 841 if (rc)
145686ba 842 goto out_err_unlock;
a046d57d
UB		 843
844 /* receive SMC Confirm CLC message */
845 reason_code = smc_clc_wait_msg(new_smc, &cclc, sizeof(cclc),
846 SMC_CLC_CONFIRM);
847 if (reason_code < 0)
145686ba 848 goto out_err_unlock;
a046d57d 849 if (reason_code > 0)
145686ba 850 goto decline_rdma_unlock;
0cfdd8f9
UB		 851 smc_conn_save_peer_info(new_smc, &cclc);
852 if (local_contact == SMC_FIRST_CONTACT)
853 smc_link_save_peer_info(link, &cclc);
a046d57d 854
bd4ad577
UB		 855 rc = smc_rmb_rtoken_handling(&new_smc->conn, &cclc);
856 if (rc) {
857 reason_code = SMC_CLC_DECL_INTERR;
145686ba 858 goto decline_rdma_unlock;
bd4ad577
UB		 859 }
860
bd4ad577
UB		 861 if (local_contact == SMC_FIRST_CONTACT) {
862 rc = smc_ib_ready_link(link);
863 if (rc) {
864 reason_code = SMC_CLC_DECL_INTERR;
145686ba 865 goto decline_rdma_unlock;
bd4ad577 866 }
9bf9abea
UB		 867 /* QP confirmation over RoCE fabric */
868 reason_code = smc_serv_conf_first_link(new_smc);
869 if (reason_code < 0) {
870 /* peer is not aware of a problem */
871 rc = reason_code;
145686ba 872 goto out_err_unlock;
9bf9abea
UB		 873 }
874 if (reason_code > 0)
145686ba 875 goto decline_rdma_unlock;
bd4ad577 876 }
a046d57d 877
e6727f39 878 smc_tx_init(new_smc);
145686ba 879 mutex_unlock(&smc_create_lgr_pending);
e6727f39 880
a046d57d
UB		 881out_connected:
882 sk_refcnt_debug_inc(newsmcsk);
b38d7324
UB		 883 if (newsmcsk->sk_state == SMC_INIT)
884 newsmcsk->sk_state = SMC_ACTIVE;
a046d57d 885enqueue:
b38d7324 886 lock_sock_nested(&lsmc->sk, SINGLE_DEPTH_NESTING);
a046d57d
UB		 887 if (lsmc->sk.sk_state == SMC_LISTEN) {
888 smc_accept_enqueue(&lsmc->sk, newsmcsk);
889 } else { /* no longer listening */
890 smc_close_non_accepted(newsmcsk);
891 }
892 release_sock(&lsmc->sk);
893
894 /* Wake up accept */
895 lsmc->sk.sk_data_ready(&lsmc->sk);
896 sock_put(&lsmc->sk); /* sock_hold in smc_tcp_listen_work */
897 return;
898
145686ba
UB		 899decline_rdma_unlock:
900 mutex_unlock(&smc_create_lgr_pending);
a046d57d
UB		 901decline_rdma:
902 /* RDMA setup failed, switch back to TCP */
0cfdd8f9 903 smc_conn_free(&new_smc->conn);
a046d57d
UB		 904 new_smc->use_fallback = true;
905 if (reason_code && (reason_code != SMC_CLC_DECL_REPLY)) {
bfbedfd3 906 rc = smc_clc_send_decline(new_smc, reason_code);
a046d57d
UB		 907 if (rc < sizeof(struct smc_clc_msg_decline))
908 goto out_err;
909 }
910 goto out_connected;
911
145686ba
UB		 912out_err_unlock:
913 mutex_unlock(&smc_create_lgr_pending);
a046d57d
UB		 914out_err:
915 newsmcsk->sk_state = SMC_CLOSED;
b38d7324 916 smc_conn_free(&new_smc->conn);
a046d57d
UB		 917 goto enqueue; /* queue new sock with sk_err set */
918}
919
920static void smc_tcp_listen_work(struct work_struct *work)
921{
922 struct smc_sock *lsmc = container_of(work, struct smc_sock,
923 tcp_listen_work);
924 struct smc_sock *new_smc;
925 int rc = 0;
926
927 lock_sock(&lsmc->sk);
928 while (lsmc->sk.sk_state == SMC_LISTEN) {
929 rc = smc_clcsock_accept(lsmc, &new_smc);
930 if (rc)
931 goto out;
932 if (!new_smc)
933 continue;
934
935 new_smc->listen_smc = lsmc;
936 new_smc->use_fallback = false; /* assume rdma capability first*/
937 sock_hold(&lsmc->sk); /* sock_put in smc_listen_work */
938 INIT_WORK(&new_smc->smc_listen_work, smc_listen_work);
939 smc_copy_sock_settings_to_smc(new_smc);
940 schedule_work(&new_smc->smc_listen_work);
941 }
942
943out:
944 release_sock(&lsmc->sk);
945 lsmc->sk.sk_data_ready(&lsmc->sk); /* no more listening, wake accept */
946}
947
ac713874
UB		 948static int smc_listen(struct socket *sock, int backlog)
949{
950 struct sock *sk = sock->sk;
951 struct smc_sock *smc;
952 int rc;
953
954 smc = smc_sk(sk);
955 lock_sock(sk);
956
957 rc = -EINVAL;
958 if ((sk->sk_state != SMC_INIT) && (sk->sk_state != SMC_LISTEN))
959 goto out;
960
961 rc = 0;
962 if (sk->sk_state == SMC_LISTEN) {
963 sk->sk_max_ack_backlog = backlog;
964 goto out;
965 }
966 /* some socket options are handled in core, so we could not apply
967 * them to the clc socket -- copy smc socket options to clc socket
968 */
969 smc_copy_sock_settings_to_clc(smc);
970
971 rc = kernel_listen(smc->clcsock, backlog);
972 if (rc)
973 goto out;
974 sk->sk_max_ack_backlog = backlog;
975 sk->sk_ack_backlog = 0;
976 sk->sk_state = SMC_LISTEN;
a046d57d
UB		 977 INIT_WORK(&smc->tcp_listen_work, smc_tcp_listen_work);
978 schedule_work(&smc->tcp_listen_work);
ac713874
UB		 979
980out:
981 release_sock(sk);
982 return rc;
983}
984
985static int smc_accept(struct socket *sock, struct socket *new_sock,
cdfbabfb 986 int flags, bool kern)
ac713874 987{
a046d57d
UB		 988 struct sock *sk = sock->sk, *nsk;
989 DECLARE_WAITQUEUE(wait, current);
ac713874 990 struct smc_sock *lsmc;
a046d57d
UB		 991 long timeo;
992 int rc = 0;
ac713874
UB		 993
994 lsmc = smc_sk(sk);
995 lock_sock(sk);
996
997 if (lsmc->sk.sk_state != SMC_LISTEN) {
998 rc = -EINVAL;
999 goto out;
1000 }
1001
a046d57d
UB		 1002 /* Wait for an incoming connection */
1003 timeo = sock_rcvtimeo(sk, flags & O_NONBLOCK);
1004 add_wait_queue_exclusive(sk_sleep(sk), &wait);
1005 while (!(nsk = smc_accept_dequeue(sk, new_sock))) {
1006 set_current_state(TASK_INTERRUPTIBLE);
1007 if (!timeo) {
1008 rc = -EAGAIN;
1009 break;
1010 }
1011 release_sock(sk);
1012 timeo = schedule_timeout(timeo);
1013 /* wakeup by sk_data_ready in smc_listen_work() */
1014 sched_annotate_sleep();
1015 lock_sock(sk);
1016 if (signal_pending(current)) {
1017 rc = sock_intr_errno(timeo);
1018 break;
1019 }
1020 }
1021 set_current_state(TASK_RUNNING);
1022 remove_wait_queue(sk_sleep(sk), &wait);
ac713874 1023
a046d57d
UB		 1024 if (!rc)
1025 rc = sock_error(nsk);
ac713874
UB		 1026
1027out:
1028 release_sock(sk);
1029 return rc;
1030}
1031
1032static int smc_getname(struct socket *sock, struct sockaddr *addr,
1033 int *len, int peer)
1034{
1035 struct smc_sock *smc;
1036
b38d7324
UB		 1037 if (peer && (sock->sk->sk_state != SMC_ACTIVE) &&
1038 (sock->sk->sk_state != SMC_APPCLOSEWAIT1))
ac713874
UB		 1039 return -ENOTCONN;
1040
1041 smc = smc_sk(sock->sk);
1042
1043 return smc->clcsock->ops->getname(smc->clcsock, addr, len, peer);
1044}
1045
1046static int smc_sendmsg(struct socket *sock, struct msghdr *msg, size_t len)
1047{
1048 struct sock *sk = sock->sk;
1049 struct smc_sock *smc;
1050 int rc = -EPIPE;
1051
1052 smc = smc_sk(sk);
1053 lock_sock(sk);
b38d7324
UB		 1054 if ((sk->sk_state != SMC_ACTIVE) &&
1055 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1056 (sk->sk_state != SMC_INIT))
ac713874
UB		 1057 goto out;
1058 if (smc->use_fallback)
1059 rc = smc->clcsock->ops->sendmsg(smc->clcsock, msg, len);
1060 else
e6727f39 1061 rc = smc_tx_sendmsg(smc, msg, len);
ac713874
UB		 1062out:
1063 release_sock(sk);
1064 return rc;
1065}
1066
1067static int smc_recvmsg(struct socket *sock, struct msghdr *msg, size_t len,
1068 int flags)
1069{
1070 struct sock *sk = sock->sk;
1071 struct smc_sock *smc;
1072 int rc = -ENOTCONN;
1073
1074 smc = smc_sk(sk);
1075 lock_sock(sk);
b38d7324
UB		 1076 if ((sk->sk_state == SMC_INIT) ||
1077 (sk->sk_state == SMC_LISTEN) ||
1078 (sk->sk_state == SMC_CLOSED))
ac713874
UB		 1079 goto out;
1080
b38d7324
UB		 1081 if (sk->sk_state == SMC_PEERFINCLOSEWAIT) {
1082 rc = 0;
1083 goto out;
1084 }
1085
ac713874
UB		 1086 if (smc->use_fallback)
1087 rc = smc->clcsock->ops->recvmsg(smc->clcsock, msg, len, flags);
1088 else
952310cc 1089 rc = smc_rx_recvmsg(smc, msg, len, flags);
b38d7324 1090
ac713874
UB		 1091out:
1092 release_sock(sk);
1093 return rc;
1094}
1095
a046d57d
UB		 1096static unsigned int smc_accept_poll(struct sock *parent)
1097{
1098 struct smc_sock *isk;
1099 struct sock *sk;
1100
1101 lock_sock(parent);
1102 list_for_each_entry(isk, &smc_sk(parent)->accept_q, accept_q) {
1103 sk = (struct sock *)isk;
1104
1105 if (sk->sk_state == SMC_ACTIVE) {
1106 release_sock(parent);
1107 return POLLIN | POLLRDNORM;
1108 }
1109 }
1110 release_sock(parent);
1111
1112 return 0;
1113}
1114
ac713874
UB		 1115static unsigned int smc_poll(struct file *file, struct socket *sock,
1116 poll_table *wait)
1117{
1118 struct sock *sk = sock->sk;
1119 unsigned int mask = 0;
1120 struct smc_sock *smc;
a046d57d 1121 int rc;
ac713874
UB		 1122
1123 smc = smc_sk(sock->sk);
a046d57d
UB		 1124 if ((sk->sk_state == SMC_INIT) || smc->use_fallback) {
1125 /* delegate to CLC child sock */
ac713874
UB		 1126 mask = smc->clcsock->ops->poll(file, smc->clcsock, wait);
1127 /* if non-blocking connect finished ... */
1128 lock_sock(sk);
1129 if ((sk->sk_state == SMC_INIT) && (mask & POLLOUT)) {
a046d57d
UB		 1130 sk->sk_err = smc->clcsock->sk->sk_err;
1131 if (sk->sk_err) {
1132 mask |= POLLERR;
1133 } else {
1134 rc = smc_connect_rdma(smc);
1135 if (rc < 0)
1136 mask |= POLLERR;
1137 else
1138 /* success cases including fallback */
1139 mask |= POLLOUT | POLLWRNORM;
1140 }
ac713874
UB		 1141 }
1142 release_sock(sk);
1143 } else {
a046d57d
UB		 1144 sock_poll_wait(file, sk_sleep(sk), wait);
1145 if (sk->sk_state == SMC_LISTEN)
1146 /* woken up by sk_data_ready in smc_listen_work() */
1147 mask |= smc_accept_poll(sk);
1148 if (sk->sk_err)
1149 mask |= POLLERR;
b38d7324
UB		 1150 if (atomic_read(&smc->conn.sndbuf_space) ||
1151 (sk->sk_shutdown & SEND_SHUTDOWN)) {
e6727f39
UB		 1152 mask |= POLLOUT | POLLWRNORM;
1153 } else {
1154 sk_set_bit(SOCKWQ_ASYNC_NOSPACE, sk);
1155 set_bit(SOCK_NOSPACE, &sk->sk_socket->flags);
1156 }
952310cc
UB		 1157 if (atomic_read(&smc->conn.bytes_to_rcv))
1158 mask |= POLLIN | POLLRDNORM;
b38d7324
UB		 1159 if ((sk->sk_shutdown == SHUTDOWN_MASK) ||
1160 (sk->sk_state == SMC_CLOSED))
1161 mask |= POLLHUP;
1162 if (sk->sk_shutdown & RCV_SHUTDOWN)
1163 mask |= POLLIN | POLLRDNORM | POLLRDHUP;
1164 if (sk->sk_state == SMC_APPCLOSEWAIT1)
1165 mask |= POLLIN;
1166
ac713874
UB		 1167 }
1168
1169 return mask;
1170}
1171
1172static int smc_shutdown(struct socket *sock, int how)
1173{
1174 struct sock *sk = sock->sk;
1175 struct smc_sock *smc;
1176 int rc = -EINVAL;
b38d7324 1177 int rc1 = 0;
ac713874
UB		 1178
1179 smc = smc_sk(sk);
1180
1181 if ((how < SHUT_RD) || (how > SHUT_RDWR))
b38d7324 1182 return rc;
ac713874
UB		 1183
1184 lock_sock(sk);
1185
1186 rc = -ENOTCONN;
b38d7324
UB		 1187 if ((sk->sk_state != SMC_LISTEN) &&
1188 (sk->sk_state != SMC_ACTIVE) &&
1189 (sk->sk_state != SMC_PEERCLOSEWAIT1) &&
1190 (sk->sk_state != SMC_PEERCLOSEWAIT2) &&
1191 (sk->sk_state != SMC_APPCLOSEWAIT1) &&
1192 (sk->sk_state != SMC_APPCLOSEWAIT2) &&
1193 (sk->sk_state != SMC_APPFINCLOSEWAIT))
ac713874
UB		 1194 goto out;
1195 if (smc->use_fallback) {
1196 rc = kernel_sock_shutdown(smc->clcsock, how);
1197 sk->sk_shutdown = smc->clcsock->sk->sk_shutdown;
1198 if (sk->sk_shutdown == SHUTDOWN_MASK)
1199 sk->sk_state = SMC_CLOSED;
b38d7324 1200 goto out;
ac713874 1201 }
b38d7324
UB		 1202 switch (how) {
1203 case SHUT_RDWR: /* shutdown in both directions */
1204 rc = smc_close_active(smc);
1205 break;
1206 case SHUT_WR:
1207 rc = smc_close_shutdown_write(smc);
1208 break;
1209 case SHUT_RD:
1210 if (sk->sk_state == SMC_LISTEN)
1211 rc = smc_close_active(smc);
1212 else
1213 rc = 0;
1214 /* nothing more to do because peer is not involved */
1215 break;
1216 }
1217 rc1 = kernel_sock_shutdown(smc->clcsock, how);
1218 /* map sock_shutdown_cmd constants to sk_shutdown value range */
1219 sk->sk_shutdown |= how + 1;
ac713874
UB		 1220
1221out:
1222 release_sock(sk);
b38d7324 1223 return rc ? rc : rc1;
ac713874
UB		 1224}
1225
1226static int smc_setsockopt(struct socket *sock, int level, int optname,
1227 char __user *optval, unsigned int optlen)
1228{
1229 struct sock *sk = sock->sk;
1230 struct smc_sock *smc;
1231
1232 smc = smc_sk(sk);
1233
1234 /* generic setsockopts reaching us here always apply to the
1235 * CLC socket
1236 */
1237 return smc->clcsock->ops->setsockopt(smc->clcsock, level, optname,
1238 optval, optlen);
1239}
1240
1241static int smc_getsockopt(struct socket *sock, int level, int optname,
1242 char __user *optval, int __user *optlen)
1243{
1244 struct smc_sock *smc;
1245
1246 smc = smc_sk(sock->sk);
1247 /* socket options apply to the CLC socket */
1248 return smc->clcsock->ops->getsockopt(smc->clcsock, level, optname,
1249 optval, optlen);
1250}
1251
1252static int smc_ioctl(struct socket *sock, unsigned int cmd,
1253 unsigned long arg)
1254{
1255 struct smc_sock *smc;
1256
1257 smc = smc_sk(sock->sk);
1258 if (smc->use_fallback)
1259 return smc->clcsock->ops->ioctl(smc->clcsock, cmd, arg);
1260 else
1261 return sock_no_ioctl(sock, cmd, arg);
1262}
1263
1264static ssize_t smc_sendpage(struct socket *sock, struct page *page,
1265 int offset, size_t size, int flags)
1266{
1267 struct sock *sk = sock->sk;
1268 struct smc_sock *smc;
1269 int rc = -EPIPE;
1270
1271 smc = smc_sk(sk);
1272 lock_sock(sk);
1273 if (sk->sk_state != SMC_ACTIVE)
1274 goto out;
1275 if (smc->use_fallback)
1276 rc = kernel_sendpage(smc->clcsock, page, offset,
1277 size, flags);
1278 else
1279 rc = sock_no_sendpage(sock, page, offset, size, flags);
1280
1281out:
1282 release_sock(sk);
1283 return rc;
1284}
1285
1286static ssize_t smc_splice_read(struct socket *sock, loff_t *ppos,
1287 struct pipe_inode_info *pipe, size_t len,
1288 unsigned int flags)
1289{
1290 struct sock *sk = sock->sk;
1291 struct smc_sock *smc;
1292 int rc = -ENOTCONN;
1293
1294 smc = smc_sk(sk);
1295 lock_sock(sk);
1296 if ((sk->sk_state != SMC_ACTIVE) && (sk->sk_state != SMC_CLOSED))
1297 goto out;
1298 if (smc->use_fallback) {
1299 rc = smc->clcsock->ops->splice_read(smc->clcsock, ppos,
1300 pipe, len, flags);
1301 } else {
1302 rc = -EOPNOTSUPP;
1303 }
1304out:
1305 release_sock(sk);
1306 return rc;
1307}
1308
1309/* must look like tcp */
1310static const struct proto_ops smc_sock_ops = {
1311 .family = PF_SMC,
1312 .owner = THIS_MODULE,
1313 .release = smc_release,
1314 .bind = smc_bind,
1315 .connect = smc_connect,
1316 .socketpair = sock_no_socketpair,
1317 .accept = smc_accept,
1318 .getname = smc_getname,
1319 .poll = smc_poll,
1320 .ioctl = smc_ioctl,
1321 .listen = smc_listen,
1322 .shutdown = smc_shutdown,
1323 .setsockopt = smc_setsockopt,
1324 .getsockopt = smc_getsockopt,
1325 .sendmsg = smc_sendmsg,
1326 .recvmsg = smc_recvmsg,
1327 .mmap = sock_no_mmap,
1328 .sendpage = smc_sendpage,
1329 .splice_read = smc_splice_read,
1330};
1331
1332static int smc_create(struct net *net, struct socket *sock, int protocol,
1333 int kern)
1334{
1335 struct smc_sock *smc;
1336 struct sock *sk;
1337 int rc;
1338
1339 rc = -ESOCKTNOSUPPORT;
1340 if (sock->type != SOCK_STREAM)
1341 goto out;
1342
1343 rc = -EPROTONOSUPPORT;
1344 if ((protocol != IPPROTO_IP) && (protocol != IPPROTO_TCP))
1345 goto out;
1346
1347 rc = -ENOBUFS;
1348 sock->ops = &smc_sock_ops;
1349 sk = smc_sock_alloc(net, sock);
1350 if (!sk)
1351 goto out;
1352
1353 /* create internal TCP socket for CLC handshake and fallback */
1354 smc = smc_sk(sk);
a046d57d 1355 smc->use_fallback = false; /* assume rdma capability first */
ac713874
UB		 1356 rc = sock_create_kern(net, PF_INET, SOCK_STREAM,
1357 IPPROTO_TCP, &smc->clcsock);
1358 if (rc)
1359 sk_common_release(sk);
cd6851f3
UB		 1360 smc->sk.sk_sndbuf = max(smc->clcsock->sk->sk_sndbuf, SMC_BUF_MIN_SIZE);
1361 smc->sk.sk_rcvbuf = max(smc->clcsock->sk->sk_rcvbuf, SMC_BUF_MIN_SIZE);
ac713874
UB		 1362
1363out:
1364 return rc;
1365}
1366
1367static const struct net_proto_family smc_sock_family_ops = {
1368 .family = PF_SMC,
1369 .owner = THIS_MODULE,
1370 .create = smc_create,
1371};
1372
1373static int __init smc_init(void)
1374{
1375 int rc;
1376
6812baab
TR		 1377 rc = smc_pnet_init();
1378 if (rc)
1379 return rc;
1380
9bf9abea
UB		 1381 rc = smc_llc_init();
1382 if (rc) {
1383 pr_err("%s: smc_llc_init fails with %d\n", __func__, rc);
1384 goto out_pnet;
1385 }
1386
5f08318f
UB		 1387 rc = smc_cdc_init();
1388 if (rc) {
1389 pr_err("%s: smc_cdc_init fails with %d\n", __func__, rc);
1390 goto out_pnet;
1391 }
1392
ac713874
UB		 1393 rc = proto_register(&smc_proto, 1);
1394 if (rc) {
1395 pr_err("%s: proto_register fails with %d\n", __func__, rc);
6812baab 1396 goto out_pnet;
ac713874
UB		 1397 }
1398
1399 rc = sock_register(&smc_sock_family_ops);
1400 if (rc) {
1401 pr_err("%s: sock_register fails with %d\n", __func__, rc);
1402 goto out_proto;
1403 }
f16a7dd5 1404 INIT_HLIST_HEAD(&smc_v4_hashinfo.ht);
ac713874 1405
a4cf0443
UB		 1406 rc = smc_ib_register_client();
1407 if (rc) {
1408 pr_err("%s: ib_register fails with %d\n", __func__, rc);
1409 goto out_sock;
1410 }
1411
ac713874
UB		 1412 return 0;
1413
a4cf0443
UB		 1414out_sock:
1415 sock_unregister(PF_SMC);
ac713874
UB		 1416out_proto:
1417 proto_unregister(&smc_proto);
6812baab
TR		 1418out_pnet:
1419 smc_pnet_exit();
ac713874
UB		 1420 return rc;
1421}
1422
1423static void __exit smc_exit(void)
1424{
0cfdd8f9
UB		 1425 struct smc_link_group *lgr, *lg;
1426 LIST_HEAD(lgr_freeing_list);
1427
1428 spin_lock_bh(&smc_lgr_list.lock);
1429 if (!list_empty(&smc_lgr_list.list))
1430 list_splice_init(&smc_lgr_list.list, &lgr_freeing_list);
1431 spin_unlock_bh(&smc_lgr_list.lock);
1432 list_for_each_entry_safe(lgr, lg, &lgr_freeing_list, list) {
1433 list_del_init(&lgr->list);
1434 smc_lgr_free(lgr); /* free link group */
1435 }
a4cf0443 1436 smc_ib_unregister_client();
ac713874
UB		 1437 sock_unregister(PF_SMC);
1438 proto_unregister(&smc_proto);
6812baab 1439 smc_pnet_exit();
ac713874
UB		 1440}
1441
1442module_init(smc_init);
1443module_exit(smc_exit);
1444
1445MODULE_AUTHOR("Ursula Braun <ubraun@linux.vnet.ibm.com>");
1446MODULE_DESCRIPTION("smc socket address family");
1447MODULE_LICENSE("GPL");
1448MODULE_ALIAS_NETPROTO(PF_SMC);
Linux 6.x block layer and io_uring tree(s)