Loading...
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * CLC (connection layer control) handshake over initial TCP socket to
6 * prepare for RDMA traffic
7 *
8 * Copyright IBM Corp. 2016, 2018
9 *
10 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
11 */
12
13#include <linux/in.h>
14#include <linux/inetdevice.h>
15#include <linux/if_ether.h>
16#include <linux/sched/signal.h>
17#include <linux/utsname.h>
18#include <linux/ctype.h>
19
20#include <net/addrconf.h>
21#include <net/sock.h>
22#include <net/tcp.h>
23
24#include "smc.h"
25#include "smc_core.h"
26#include "smc_clc.h"
27#include "smc_ib.h"
28#include "smc_ism.h"
29#include "smc_netlink.h"
30
31#define SMCR_CLC_ACCEPT_CONFIRM_LEN 68
32#define SMCD_CLC_ACCEPT_CONFIRM_LEN 48
33#define SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 78
34#define SMCR_CLC_ACCEPT_CONFIRM_LEN_V2 108
35#define SMC_CLC_RECV_BUF_LEN 100
36
37/* eye catcher "SMCR" EBCDIC for CLC messages */
38static const char SMC_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xd9'};
39/* eye catcher "SMCD" EBCDIC for CLC messages */
40static const char SMCD_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xc4'};
41
42static u8 smc_hostname[SMC_MAX_HOSTNAME_LEN];
43
44struct smc_clc_eid_table {
45 rwlock_t lock;
46 struct list_head list;
47 u8 ueid_cnt;
48 u8 seid_enabled;
49};
50
51static struct smc_clc_eid_table smc_clc_eid_table;
52
53struct smc_clc_eid_entry {
54 struct list_head list;
55 u8 eid[SMC_MAX_EID_LEN];
56};
57
58/* The size of a user EID is 32 characters.
59 * Valid characters should be (single-byte character set) A-Z, 0-9, '.' and '-'.
60 * Blanks should only be used to pad to the expected size.
61 * First character must be alphanumeric.
62 */
63static bool smc_clc_ueid_valid(char *ueid)
64{
65 char *end = ueid + SMC_MAX_EID_LEN;
66
67 while (--end >= ueid && isspace(*end))
68 ;
69 if (end < ueid)
70 return false;
71 if (!isalnum(*ueid) || islower(*ueid))
72 return false;
73 while (ueid <= end) {
74 if ((!isalnum(*ueid) || islower(*ueid)) && *ueid != '.' &&
75 *ueid != '-')
76 return false;
77 ueid++;
78 }
79 return true;
80}
81
82static int smc_clc_ueid_add(char *ueid)
83{
84 struct smc_clc_eid_entry *new_ueid, *tmp_ueid;
85 int rc;
86
87 if (!smc_clc_ueid_valid(ueid))
88 return -EINVAL;
89
90 /* add a new ueid entry to the ueid table if there isn't one */
91 new_ueid = kzalloc(sizeof(*new_ueid), GFP_KERNEL);
92 if (!new_ueid)
93 return -ENOMEM;
94 memcpy(new_ueid->eid, ueid, SMC_MAX_EID_LEN);
95
96 write_lock(&smc_clc_eid_table.lock);
97 if (smc_clc_eid_table.ueid_cnt >= SMC_MAX_UEID) {
98 rc = -ERANGE;
99 goto err_out;
100 }
101 list_for_each_entry(tmp_ueid, &smc_clc_eid_table.list, list) {
102 if (!memcmp(tmp_ueid->eid, ueid, SMC_MAX_EID_LEN)) {
103 rc = -EEXIST;
104 goto err_out;
105 }
106 }
107 list_add_tail(&new_ueid->list, &smc_clc_eid_table.list);
108 smc_clc_eid_table.ueid_cnt++;
109 write_unlock(&smc_clc_eid_table.lock);
110 return 0;
111
112err_out:
113 write_unlock(&smc_clc_eid_table.lock);
114 kfree(new_ueid);
115 return rc;
116}
117
118int smc_clc_ueid_count(void)
119{
120 int count;
121
122 read_lock(&smc_clc_eid_table.lock);
123 count = smc_clc_eid_table.ueid_cnt;
124 read_unlock(&smc_clc_eid_table.lock);
125
126 return count;
127}
128
129int smc_nl_add_ueid(struct sk_buff *skb, struct genl_info *info)
130{
131 struct nlattr *nla_ueid = info->attrs[SMC_NLA_EID_TABLE_ENTRY];
132 char *ueid;
133
134 if (!nla_ueid || nla_len(nla_ueid) != SMC_MAX_EID_LEN + 1)
135 return -EINVAL;
136 ueid = (char *)nla_data(nla_ueid);
137
138 return smc_clc_ueid_add(ueid);
139}
140
141/* remove one or all ueid entries from the table */
142static int smc_clc_ueid_remove(char *ueid)
143{
144 struct smc_clc_eid_entry *lst_ueid, *tmp_ueid;
145 int rc = -ENOENT;
146
147 /* remove table entry */
148 write_lock(&smc_clc_eid_table.lock);
149 list_for_each_entry_safe(lst_ueid, tmp_ueid, &smc_clc_eid_table.list,
150 list) {
151 if (!ueid || !memcmp(lst_ueid->eid, ueid, SMC_MAX_EID_LEN)) {
152 list_del(&lst_ueid->list);
153 smc_clc_eid_table.ueid_cnt--;
154 kfree(lst_ueid);
155 rc = 0;
156 }
157 }
158#if IS_ENABLED(CONFIG_S390)
159 if (!rc && !smc_clc_eid_table.ueid_cnt) {
160 smc_clc_eid_table.seid_enabled = 1;
161 rc = -EAGAIN; /* indicate success and enabling of seid */
162 }
163#endif
164 write_unlock(&smc_clc_eid_table.lock);
165 return rc;
166}
167
168int smc_nl_remove_ueid(struct sk_buff *skb, struct genl_info *info)
169{
170 struct nlattr *nla_ueid = info->attrs[SMC_NLA_EID_TABLE_ENTRY];
171 char *ueid;
172
173 if (!nla_ueid || nla_len(nla_ueid) != SMC_MAX_EID_LEN + 1)
174 return -EINVAL;
175 ueid = (char *)nla_data(nla_ueid);
176
177 return smc_clc_ueid_remove(ueid);
178}
179
180int smc_nl_flush_ueid(struct sk_buff *skb, struct genl_info *info)
181{
182 smc_clc_ueid_remove(NULL);
183 return 0;
184}
185
186static int smc_nl_ueid_dumpinfo(struct sk_buff *skb, u32 portid, u32 seq,
187 u32 flags, char *ueid)
188{
189 char ueid_str[SMC_MAX_EID_LEN + 1];
190 void *hdr;
191
192 hdr = genlmsg_put(skb, portid, seq, &smc_gen_nl_family,
193 flags, SMC_NETLINK_DUMP_UEID);
194 if (!hdr)
195 return -ENOMEM;
196 memcpy(ueid_str, ueid, SMC_MAX_EID_LEN);
197 ueid_str[SMC_MAX_EID_LEN] = 0;
198 if (nla_put_string(skb, SMC_NLA_EID_TABLE_ENTRY, ueid_str)) {
199 genlmsg_cancel(skb, hdr);
200 return -EMSGSIZE;
201 }
202 genlmsg_end(skb, hdr);
203 return 0;
204}
205
206static int _smc_nl_ueid_dump(struct sk_buff *skb, u32 portid, u32 seq,
207 int start_idx)
208{
209 struct smc_clc_eid_entry *lst_ueid;
210 int idx = 0;
211
212 read_lock(&smc_clc_eid_table.lock);
213 list_for_each_entry(lst_ueid, &smc_clc_eid_table.list, list) {
214 if (idx++ < start_idx)
215 continue;
216 if (smc_nl_ueid_dumpinfo(skb, portid, seq, NLM_F_MULTI,
217 lst_ueid->eid)) {
218 --idx;
219 break;
220 }
221 }
222 read_unlock(&smc_clc_eid_table.lock);
223 return idx;
224}
225
226int smc_nl_dump_ueid(struct sk_buff *skb, struct netlink_callback *cb)
227{
228 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
229 int idx;
230
231 idx = _smc_nl_ueid_dump(skb, NETLINK_CB(cb->skb).portid,
232 cb->nlh->nlmsg_seq, cb_ctx->pos[0]);
233
234 cb_ctx->pos[0] = idx;
235 return skb->len;
236}
237
238int smc_nl_dump_seid(struct sk_buff *skb, struct netlink_callback *cb)
239{
240 struct smc_nl_dmp_ctx *cb_ctx = smc_nl_dmp_ctx(cb);
241 char seid_str[SMC_MAX_EID_LEN + 1];
242 u8 seid_enabled;
243 void *hdr;
244 u8 *seid;
245
246 if (cb_ctx->pos[0])
247 return skb->len;
248
249 hdr = genlmsg_put(skb, NETLINK_CB(cb->skb).portid, cb->nlh->nlmsg_seq,
250 &smc_gen_nl_family, NLM_F_MULTI,
251 SMC_NETLINK_DUMP_SEID);
252 if (!hdr)
253 return -ENOMEM;
254 if (!smc_ism_is_v2_capable())
255 goto end;
256
257 smc_ism_get_system_eid(&seid);
258 memcpy(seid_str, seid, SMC_MAX_EID_LEN);
259 seid_str[SMC_MAX_EID_LEN] = 0;
260 if (nla_put_string(skb, SMC_NLA_SEID_ENTRY, seid_str))
261 goto err;
262 read_lock(&smc_clc_eid_table.lock);
263 seid_enabled = smc_clc_eid_table.seid_enabled;
264 read_unlock(&smc_clc_eid_table.lock);
265 if (nla_put_u8(skb, SMC_NLA_SEID_ENABLED, seid_enabled))
266 goto err;
267end:
268 genlmsg_end(skb, hdr);
269 cb_ctx->pos[0]++;
270 return skb->len;
271err:
272 genlmsg_cancel(skb, hdr);
273 return -EMSGSIZE;
274}
275
276int smc_nl_enable_seid(struct sk_buff *skb, struct genl_info *info)
277{
278#if IS_ENABLED(CONFIG_S390)
279 write_lock(&smc_clc_eid_table.lock);
280 smc_clc_eid_table.seid_enabled = 1;
281 write_unlock(&smc_clc_eid_table.lock);
282 return 0;
283#else
284 return -EOPNOTSUPP;
285#endif
286}
287
288int smc_nl_disable_seid(struct sk_buff *skb, struct genl_info *info)
289{
290 int rc = 0;
291
292#if IS_ENABLED(CONFIG_S390)
293 write_lock(&smc_clc_eid_table.lock);
294 if (!smc_clc_eid_table.ueid_cnt)
295 rc = -ENOENT;
296 else
297 smc_clc_eid_table.seid_enabled = 0;
298 write_unlock(&smc_clc_eid_table.lock);
299#else
300 rc = -EOPNOTSUPP;
301#endif
302 return rc;
303}
304
305static bool _smc_clc_match_ueid(u8 *peer_ueid)
306{
307 struct smc_clc_eid_entry *tmp_ueid;
308
309 list_for_each_entry(tmp_ueid, &smc_clc_eid_table.list, list) {
310 if (!memcmp(tmp_ueid->eid, peer_ueid, SMC_MAX_EID_LEN))
311 return true;
312 }
313 return false;
314}
315
316bool smc_clc_match_eid(u8 *negotiated_eid,
317 struct smc_clc_v2_extension *smc_v2_ext,
318 u8 *peer_eid, u8 *local_eid)
319{
320 bool match = false;
321 int i;
322
323 negotiated_eid[0] = 0;
324 read_lock(&smc_clc_eid_table.lock);
325 if (peer_eid && local_eid &&
326 smc_clc_eid_table.seid_enabled &&
327 smc_v2_ext->hdr.flag.seid &&
328 !memcmp(peer_eid, local_eid, SMC_MAX_EID_LEN)) {
329 memcpy(negotiated_eid, peer_eid, SMC_MAX_EID_LEN);
330 match = true;
331 goto out;
332 }
333
334 for (i = 0; i < smc_v2_ext->hdr.eid_cnt; i++) {
335 if (_smc_clc_match_ueid(smc_v2_ext->user_eids[i])) {
336 memcpy(negotiated_eid, smc_v2_ext->user_eids[i],
337 SMC_MAX_EID_LEN);
338 match = true;
339 goto out;
340 }
341 }
342out:
343 read_unlock(&smc_clc_eid_table.lock);
344 return match;
345}
346
347/* check arriving CLC proposal */
348static bool smc_clc_msg_prop_valid(struct smc_clc_msg_proposal *pclc)
349{
350 struct smc_clc_msg_proposal_prefix *pclc_prfx;
351 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
352 struct smc_clc_msg_hdr *hdr = &pclc->hdr;
353 struct smc_clc_v2_extension *v2_ext;
354
355 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
356 if (!pclc_prfx ||
357 pclc_prfx->ipv6_prefixes_cnt > SMC_CLC_MAX_V6_PREFIX)
358 return false;
359
360 if (hdr->version == SMC_V1) {
361 if (hdr->typev1 == SMC_TYPE_N)
362 return false;
363 if (ntohs(hdr->length) !=
364 sizeof(*pclc) + ntohs(pclc->iparea_offset) +
365 sizeof(*pclc_prfx) +
366 pclc_prfx->ipv6_prefixes_cnt *
367 sizeof(struct smc_clc_ipv6_prefix) +
368 sizeof(struct smc_clc_msg_trail))
369 return false;
370 } else {
371 v2_ext = smc_get_clc_v2_ext(pclc);
372 if ((hdr->typev2 != SMC_TYPE_N &&
373 (!v2_ext || v2_ext->hdr.eid_cnt > SMC_CLC_MAX_UEID)) ||
374 (smcd_indicated(hdr->typev2) &&
375 v2_ext->hdr.ism_gid_cnt > SMCD_CLC_MAX_V2_GID_ENTRIES))
376 return false;
377
378 if (ntohs(hdr->length) !=
379 sizeof(*pclc) +
380 sizeof(struct smc_clc_msg_smcd) +
381 (hdr->typev1 != SMC_TYPE_N ?
382 sizeof(*pclc_prfx) +
383 pclc_prfx->ipv6_prefixes_cnt *
384 sizeof(struct smc_clc_ipv6_prefix) : 0) +
385 (hdr->typev2 != SMC_TYPE_N ?
386 sizeof(*v2_ext) +
387 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN : 0) +
388 (smcd_indicated(hdr->typev2) ?
389 sizeof(*smcd_v2_ext) + v2_ext->hdr.ism_gid_cnt *
390 sizeof(struct smc_clc_smcd_gid_chid) :
391 0) +
392 sizeof(struct smc_clc_msg_trail))
393 return false;
394 }
395 return true;
396}
397
398/* check arriving CLC accept or confirm */
399static bool
400smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm *clc)
401{
402 struct smc_clc_msg_hdr *hdr = &clc->hdr;
403
404 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
405 return false;
406 if (hdr->version == SMC_V1) {
407 if ((hdr->typev1 == SMC_TYPE_R &&
408 ntohs(hdr->length) != SMCR_CLC_ACCEPT_CONFIRM_LEN) ||
409 (hdr->typev1 == SMC_TYPE_D &&
410 ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN))
411 return false;
412 } else {
413 if (hdr->typev1 == SMC_TYPE_D &&
414 ntohs(hdr->length) < SMCD_CLC_ACCEPT_CONFIRM_LEN_V2)
415 return false;
416 if (hdr->typev1 == SMC_TYPE_R &&
417 ntohs(hdr->length) < SMCR_CLC_ACCEPT_CONFIRM_LEN_V2)
418 return false;
419 }
420 return true;
421}
422
423/* check arriving CLC decline */
424static bool
425smc_clc_msg_decl_valid(struct smc_clc_msg_decline *dclc)
426{
427 struct smc_clc_msg_hdr *hdr = &dclc->hdr;
428
429 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
430 return false;
431 if (hdr->version == SMC_V1) {
432 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline))
433 return false;
434 } else {
435 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline_v2))
436 return false;
437 }
438 return true;
439}
440
441static int smc_clc_fill_fce_v2x(struct smc_clc_first_contact_ext_v2x *fce_v2x,
442 struct smc_init_info *ini)
443{
444 int ret = sizeof(*fce_v2x);
445
446 memset(fce_v2x, 0, sizeof(*fce_v2x));
447 fce_v2x->fce_v2_base.os_type = SMC_CLC_OS_LINUX;
448 fce_v2x->fce_v2_base.release = ini->release_nr;
449 memcpy(fce_v2x->fce_v2_base.hostname,
450 smc_hostname, sizeof(smc_hostname));
451 if (ini->is_smcd && ini->release_nr < SMC_RELEASE_1) {
452 ret = sizeof(struct smc_clc_first_contact_ext);
453 goto out;
454 }
455
456 if (ini->release_nr >= SMC_RELEASE_1) {
457 if (!ini->is_smcd) {
458 fce_v2x->max_conns = ini->max_conns;
459 fce_v2x->max_links = ini->max_links;
460 }
461 fce_v2x->feature_mask = htons(ini->feature_mask);
462 }
463
464out:
465 return ret;
466}
467
468/* check if received message has a correct header length and contains valid
469 * heading and trailing eyecatchers
470 */
471static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm, bool check_trl)
472{
473 struct smc_clc_msg_accept_confirm *clc;
474 struct smc_clc_msg_proposal *pclc;
475 struct smc_clc_msg_decline *dclc;
476 struct smc_clc_msg_trail *trl;
477
478 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
479 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
480 return false;
481 switch (clcm->type) {
482 case SMC_CLC_PROPOSAL:
483 pclc = (struct smc_clc_msg_proposal *)clcm;
484 if (!smc_clc_msg_prop_valid(pclc))
485 return false;
486 trl = (struct smc_clc_msg_trail *)
487 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
488 break;
489 case SMC_CLC_ACCEPT:
490 case SMC_CLC_CONFIRM:
491 clc = (struct smc_clc_msg_accept_confirm *)clcm;
492 if (!smc_clc_msg_acc_conf_valid(clc))
493 return false;
494 trl = (struct smc_clc_msg_trail *)
495 ((u8 *)clc + ntohs(clc->hdr.length) - sizeof(*trl));
496 break;
497 case SMC_CLC_DECLINE:
498 dclc = (struct smc_clc_msg_decline *)clcm;
499 if (!smc_clc_msg_decl_valid(dclc))
500 return false;
501 check_trl = false;
502 break;
503 default:
504 return false;
505 }
506 if (check_trl &&
507 memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
508 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
509 return false;
510 return true;
511}
512
513/* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
514static int smc_clc_prfx_set4_rcu(struct dst_entry *dst, __be32 ipv4,
515 struct smc_clc_msg_proposal_prefix *prop)
516{
517 struct in_device *in_dev = __in_dev_get_rcu(dst->dev);
518 const struct in_ifaddr *ifa;
519
520 if (!in_dev)
521 return -ENODEV;
522
523 in_dev_for_each_ifa_rcu(ifa, in_dev) {
524 if (!inet_ifa_match(ipv4, ifa))
525 continue;
526 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
527 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
528 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
529 return 0;
530 }
531 return -ENOENT;
532}
533
534/* fill CLC proposal msg with ipv6 prefixes from device */
535static int smc_clc_prfx_set6_rcu(struct dst_entry *dst,
536 struct smc_clc_msg_proposal_prefix *prop,
537 struct smc_clc_ipv6_prefix *ipv6_prfx)
538{
539#if IS_ENABLED(CONFIG_IPV6)
540 struct inet6_dev *in6_dev = __in6_dev_get(dst->dev);
541 struct inet6_ifaddr *ifa;
542 int cnt = 0;
543
544 if (!in6_dev)
545 return -ENODEV;
546 /* use a maximum of 8 IPv6 prefixes from device */
547 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
548 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
549 continue;
550 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
551 &ifa->addr, ifa->prefix_len);
552 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
553 cnt++;
554 if (cnt == SMC_CLC_MAX_V6_PREFIX)
555 break;
556 }
557 prop->ipv6_prefixes_cnt = cnt;
558 if (cnt)
559 return 0;
560#endif
561 return -ENOENT;
562}
563
564/* retrieve and set prefixes in CLC proposal msg */
565static int smc_clc_prfx_set(struct socket *clcsock,
566 struct smc_clc_msg_proposal_prefix *prop,
567 struct smc_clc_ipv6_prefix *ipv6_prfx)
568{
569 struct dst_entry *dst = sk_dst_get(clcsock->sk);
570 struct sockaddr_storage addrs;
571 struct sockaddr_in6 *addr6;
572 struct sockaddr_in *addr;
573 int rc = -ENOENT;
574
575 if (!dst) {
576 rc = -ENOTCONN;
577 goto out;
578 }
579 if (!dst->dev) {
580 rc = -ENODEV;
581 goto out_rel;
582 }
583 /* get address to which the internal TCP socket is bound */
584 if (kernel_getsockname(clcsock, (struct sockaddr *)&addrs) < 0)
585 goto out_rel;
586 /* analyze IP specific data of net_device belonging to TCP socket */
587 addr6 = (struct sockaddr_in6 *)&addrs;
588 rcu_read_lock();
589 if (addrs.ss_family == PF_INET) {
590 /* IPv4 */
591 addr = (struct sockaddr_in *)&addrs;
592 rc = smc_clc_prfx_set4_rcu(dst, addr->sin_addr.s_addr, prop);
593 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
594 /* mapped IPv4 address - peer is IPv4 only */
595 rc = smc_clc_prfx_set4_rcu(dst, addr6->sin6_addr.s6_addr32[3],
596 prop);
597 } else {
598 /* IPv6 */
599 rc = smc_clc_prfx_set6_rcu(dst, prop, ipv6_prfx);
600 }
601 rcu_read_unlock();
602out_rel:
603 dst_release(dst);
604out:
605 return rc;
606}
607
608/* match ipv4 addrs of dev against addr in CLC proposal */
609static int smc_clc_prfx_match4_rcu(struct net_device *dev,
610 struct smc_clc_msg_proposal_prefix *prop)
611{
612 struct in_device *in_dev = __in_dev_get_rcu(dev);
613 const struct in_ifaddr *ifa;
614
615 if (!in_dev)
616 return -ENODEV;
617 in_dev_for_each_ifa_rcu(ifa, in_dev) {
618 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
619 inet_ifa_match(prop->outgoing_subnet, ifa))
620 return 0;
621 }
622
623 return -ENOENT;
624}
625
626/* match ipv6 addrs of dev against addrs in CLC proposal */
627static int smc_clc_prfx_match6_rcu(struct net_device *dev,
628 struct smc_clc_msg_proposal_prefix *prop)
629{
630#if IS_ENABLED(CONFIG_IPV6)
631 struct inet6_dev *in6_dev = __in6_dev_get(dev);
632 struct smc_clc_ipv6_prefix *ipv6_prfx;
633 struct inet6_ifaddr *ifa;
634 int i, max;
635
636 if (!in6_dev)
637 return -ENODEV;
638 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
639 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
640 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
641 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
642 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
643 continue;
644 for (i = 0; i < max; i++) {
645 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
646 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
647 ifa->prefix_len))
648 return 0;
649 }
650 }
651#endif
652 return -ENOENT;
653}
654
655/* check if proposed prefixes match one of our device prefixes */
656int smc_clc_prfx_match(struct socket *clcsock,
657 struct smc_clc_msg_proposal_prefix *prop)
658{
659 struct dst_entry *dst = sk_dst_get(clcsock->sk);
660 int rc;
661
662 if (!dst) {
663 rc = -ENOTCONN;
664 goto out;
665 }
666 if (!dst->dev) {
667 rc = -ENODEV;
668 goto out_rel;
669 }
670 rcu_read_lock();
671 if (!prop->ipv6_prefixes_cnt)
672 rc = smc_clc_prfx_match4_rcu(dst->dev, prop);
673 else
674 rc = smc_clc_prfx_match6_rcu(dst->dev, prop);
675 rcu_read_unlock();
676out_rel:
677 dst_release(dst);
678out:
679 return rc;
680}
681
682/* Wait for data on the tcp-socket, analyze received data
683 * Returns:
684 * 0 if success and it was not a decline that we received.
685 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
686 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
687 */
688int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
689 u8 expected_type, unsigned long timeout)
690{
691 long rcvtimeo = smc->clcsock->sk->sk_rcvtimeo;
692 struct sock *clc_sk = smc->clcsock->sk;
693 struct smc_clc_msg_hdr *clcm = buf;
694 struct msghdr msg = {NULL, 0};
695 int reason_code = 0;
696 struct kvec vec = {buf, buflen};
697 int len, datlen, recvlen;
698 bool check_trl = true;
699 int krflags;
700
701 /* peek the first few bytes to determine length of data to receive
702 * so we don't consume any subsequent CLC message or payload data
703 * in the TCP byte stream
704 */
705 /*
706 * Caller must make sure that buflen is no less than
707 * sizeof(struct smc_clc_msg_hdr)
708 */
709 krflags = MSG_PEEK | MSG_WAITALL;
710 clc_sk->sk_rcvtimeo = timeout;
711 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1,
712 sizeof(struct smc_clc_msg_hdr));
713 len = sock_recvmsg(smc->clcsock, &msg, krflags);
714 if (signal_pending(current)) {
715 reason_code = -EINTR;
716 clc_sk->sk_err = EINTR;
717 smc->sk.sk_err = EINTR;
718 goto out;
719 }
720 if (clc_sk->sk_err) {
721 reason_code = -clc_sk->sk_err;
722 if (clc_sk->sk_err == EAGAIN &&
723 expected_type == SMC_CLC_DECLINE)
724 clc_sk->sk_err = 0; /* reset for fallback usage */
725 else
726 smc->sk.sk_err = clc_sk->sk_err;
727 goto out;
728 }
729 if (!len) { /* peer has performed orderly shutdown */
730 smc->sk.sk_err = ECONNRESET;
731 reason_code = -ECONNRESET;
732 goto out;
733 }
734 if (len < 0) {
735 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
736 smc->sk.sk_err = -len;
737 reason_code = len;
738 goto out;
739 }
740 datlen = ntohs(clcm->length);
741 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
742 (clcm->version < SMC_V1) ||
743 ((clcm->type != SMC_CLC_DECLINE) &&
744 (clcm->type != expected_type))) {
745 smc->sk.sk_err = EPROTO;
746 reason_code = -EPROTO;
747 goto out;
748 }
749
750 /* receive the complete CLC message */
751 memset(&msg, 0, sizeof(struct msghdr));
752 if (datlen > buflen) {
753 check_trl = false;
754 recvlen = buflen;
755 } else {
756 recvlen = datlen;
757 }
758 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
759 krflags = MSG_WAITALL;
760 len = sock_recvmsg(smc->clcsock, &msg, krflags);
761 if (len < recvlen || !smc_clc_msg_hdr_valid(clcm, check_trl)) {
762 smc->sk.sk_err = EPROTO;
763 reason_code = -EPROTO;
764 goto out;
765 }
766 datlen -= len;
767 while (datlen) {
768 u8 tmp[SMC_CLC_RECV_BUF_LEN];
769
770 vec.iov_base = &tmp;
771 vec.iov_len = SMC_CLC_RECV_BUF_LEN;
772 /* receive remaining proposal message */
773 recvlen = datlen > SMC_CLC_RECV_BUF_LEN ?
774 SMC_CLC_RECV_BUF_LEN : datlen;
775 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
776 len = sock_recvmsg(smc->clcsock, &msg, krflags);
777 if (len < recvlen) {
778 smc->sk.sk_err = EPROTO;
779 reason_code = -EPROTO;
780 goto out;
781 }
782 datlen -= len;
783 }
784 if (clcm->type == SMC_CLC_DECLINE) {
785 struct smc_clc_msg_decline *dclc;
786
787 dclc = (struct smc_clc_msg_decline *)clcm;
788 reason_code = SMC_CLC_DECL_PEERDECL;
789 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
790 if (((struct smc_clc_msg_decline *)buf)->hdr.typev2 &
791 SMC_FIRST_CONTACT_MASK) {
792 smc->conn.lgr->sync_err = 1;
793 smc_lgr_terminate_sched(smc->conn.lgr);
794 }
795 }
796
797out:
798 clc_sk->sk_rcvtimeo = rcvtimeo;
799 return reason_code;
800}
801
802/* send CLC DECLINE message across internal TCP socket */
803int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version)
804{
805 struct smc_clc_msg_decline *dclc_v1;
806 struct smc_clc_msg_decline_v2 dclc;
807 struct msghdr msg;
808 int len, send_len;
809 struct kvec vec;
810
811 dclc_v1 = (struct smc_clc_msg_decline *)&dclc;
812 memset(&dclc, 0, sizeof(dclc));
813 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
814 dclc.hdr.type = SMC_CLC_DECLINE;
815 dclc.hdr.version = version;
816 dclc.os_type = version == SMC_V1 ? 0 : SMC_CLC_OS_LINUX;
817 dclc.hdr.typev2 = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ?
818 SMC_FIRST_CONTACT_MASK : 0;
819 if ((!smc_conn_lgr_valid(&smc->conn) || !smc->conn.lgr->is_smcd) &&
820 smc_ib_is_valid_local_systemid())
821 memcpy(dclc.id_for_peer, local_systemid,
822 sizeof(local_systemid));
823 dclc.peer_diagnosis = htonl(peer_diag_info);
824 if (version == SMC_V1) {
825 memcpy(dclc_v1->trl.eyecatcher, SMC_EYECATCHER,
826 sizeof(SMC_EYECATCHER));
827 send_len = sizeof(*dclc_v1);
828 } else {
829 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER,
830 sizeof(SMC_EYECATCHER));
831 send_len = sizeof(dclc);
832 }
833 dclc.hdr.length = htons(send_len);
834
835 memset(&msg, 0, sizeof(msg));
836 vec.iov_base = &dclc;
837 vec.iov_len = send_len;
838 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, send_len);
839 if (len < 0 || len < send_len)
840 len = -EPROTO;
841 return len > 0 ? 0 : len;
842}
843
844/* send CLC PROPOSAL message across internal TCP socket */
845int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini)
846{
847 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
848 struct smc_clc_msg_proposal_prefix *pclc_prfx;
849 struct smc_clc_msg_proposal *pclc_base;
850 struct smc_clc_smcd_gid_chid *gidchids;
851 struct smc_clc_msg_proposal_area *pclc;
852 struct smc_clc_ipv6_prefix *ipv6_prfx;
853 struct net *net = sock_net(&smc->sk);
854 struct smc_clc_v2_extension *v2_ext;
855 struct smc_clc_msg_smcd *pclc_smcd;
856 struct smc_clc_msg_trail *trl;
857 struct smcd_dev *smcd;
858 int len, i, plen, rc;
859 int reason_code = 0;
860 struct kvec vec[8];
861 struct msghdr msg;
862
863 pclc = kzalloc(sizeof(*pclc), GFP_KERNEL);
864 if (!pclc)
865 return -ENOMEM;
866
867 pclc_base = &pclc->pclc_base;
868 pclc_smcd = &pclc->pclc_smcd;
869 pclc_prfx = &pclc->pclc_prfx;
870 ipv6_prfx = pclc->pclc_prfx_ipv6;
871 v2_ext = container_of(&pclc->pclc_v2_ext,
872 struct smc_clc_v2_extension, fixed);
873 smcd_v2_ext = container_of(&pclc->pclc_smcd_v2_ext,
874 struct smc_clc_smcd_v2_extension, fixed);
875 gidchids = pclc->pclc_gidchids;
876 trl = &pclc->pclc_trl;
877
878 pclc_base->hdr.version = SMC_V2;
879 pclc_base->hdr.typev1 = ini->smc_type_v1;
880 pclc_base->hdr.typev2 = ini->smc_type_v2;
881 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
882
883 /* retrieve ip prefixes for CLC proposal msg */
884 if (ini->smc_type_v1 != SMC_TYPE_N) {
885 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
886 if (rc) {
887 if (ini->smc_type_v2 == SMC_TYPE_N) {
888 kfree(pclc);
889 return SMC_CLC_DECL_CNFERR;
890 }
891 pclc_base->hdr.typev1 = SMC_TYPE_N;
892 } else {
893 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
894 plen += sizeof(*pclc_prfx) +
895 pclc_prfx->ipv6_prefixes_cnt *
896 sizeof(ipv6_prfx[0]);
897 }
898 }
899
900 /* build SMC Proposal CLC message */
901 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
902 sizeof(SMC_EYECATCHER));
903 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
904 if (smcr_indicated(ini->smc_type_v1)) {
905 /* add SMC-R specifics */
906 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
907 sizeof(local_systemid));
908 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
909 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
910 ETH_ALEN);
911 }
912 if (smcd_indicated(ini->smc_type_v1)) {
913 struct smcd_gid smcd_gid;
914
915 /* add SMC-D specifics */
916 if (ini->ism_dev[0]) {
917 smcd = ini->ism_dev[0];
918 smcd->ops->get_local_gid(smcd, &smcd_gid);
919 pclc_smcd->ism.gid = htonll(smcd_gid.gid);
920 pclc_smcd->ism.chid =
921 htons(smc_ism_get_chid(ini->ism_dev[0]));
922 }
923 }
924 if (ini->smc_type_v2 == SMC_TYPE_N) {
925 pclc_smcd->v2_ext_offset = 0;
926 } else {
927 struct smc_clc_eid_entry *ueident;
928 u16 v2_ext_offset;
929
930 v2_ext->hdr.flag.release = SMC_RELEASE;
931 v2_ext_offset = sizeof(*pclc_smcd) -
932 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
933 if (ini->smc_type_v1 != SMC_TYPE_N)
934 v2_ext_offset += sizeof(*pclc_prfx) +
935 pclc_prfx->ipv6_prefixes_cnt *
936 sizeof(ipv6_prfx[0]);
937 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
938 plen += sizeof(*v2_ext);
939
940 v2_ext->feature_mask = htons(SMC_FEATURE_MASK);
941 read_lock(&smc_clc_eid_table.lock);
942 v2_ext->hdr.eid_cnt = smc_clc_eid_table.ueid_cnt;
943 plen += smc_clc_eid_table.ueid_cnt * SMC_MAX_EID_LEN;
944 i = 0;
945 list_for_each_entry(ueident, &smc_clc_eid_table.list, list) {
946 memcpy(v2_ext->user_eids[i++], ueident->eid,
947 sizeof(ueident->eid));
948 }
949 read_unlock(&smc_clc_eid_table.lock);
950 }
951 if (smcd_indicated(ini->smc_type_v2)) {
952 struct smcd_gid smcd_gid;
953 u8 *eid = NULL;
954 int entry = 0;
955
956 v2_ext->hdr.flag.seid = smc_clc_eid_table.seid_enabled;
957 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
958 offsetofend(struct smc_clnt_opts_area_hdr,
959 smcd_v2_ext_offset) +
960 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
961 smc_ism_get_system_eid(&eid);
962 if (eid && v2_ext->hdr.flag.seid)
963 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
964 plen += sizeof(*smcd_v2_ext);
965 if (ini->ism_offered_cnt) {
966 for (i = 1; i <= ini->ism_offered_cnt; i++) {
967 smcd = ini->ism_dev[i];
968 smcd->ops->get_local_gid(smcd, &smcd_gid);
969 gidchids[entry].chid =
970 htons(smc_ism_get_chid(ini->ism_dev[i]));
971 gidchids[entry].gid = htonll(smcd_gid.gid);
972 if (smc_ism_is_emulated(smcd)) {
973 /* an Emulated-ISM device takes two
974 * entries. CHID of the second entry
975 * repeats that of the first entry.
976 */
977 gidchids[entry + 1].chid =
978 gidchids[entry].chid;
979 gidchids[entry + 1].gid =
980 htonll(smcd_gid.gid_ext);
981 entry++;
982 }
983 entry++;
984 }
985 plen += entry * sizeof(struct smc_clc_smcd_gid_chid);
986 }
987 v2_ext->hdr.ism_gid_cnt = entry;
988 }
989 if (smcr_indicated(ini->smc_type_v2)) {
990 memcpy(v2_ext->roce, ini->smcrv2.ib_gid_v2, SMC_GID_SIZE);
991 v2_ext->max_conns = net->smc.sysctl_max_conns_per_lgr;
992 v2_ext->max_links = net->smc.sysctl_max_links_per_lgr;
993 }
994
995 pclc_base->hdr.length = htons(plen);
996 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
997
998 /* send SMC Proposal CLC message */
999 memset(&msg, 0, sizeof(msg));
1000 i = 0;
1001 vec[i].iov_base = pclc_base;
1002 vec[i++].iov_len = sizeof(*pclc_base);
1003 vec[i].iov_base = pclc_smcd;
1004 vec[i++].iov_len = sizeof(*pclc_smcd);
1005 if (ini->smc_type_v1 != SMC_TYPE_N) {
1006 vec[i].iov_base = pclc_prfx;
1007 vec[i++].iov_len = sizeof(*pclc_prfx);
1008 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
1009 vec[i].iov_base = ipv6_prfx;
1010 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
1011 sizeof(ipv6_prfx[0]);
1012 }
1013 }
1014 if (ini->smc_type_v2 != SMC_TYPE_N) {
1015 vec[i].iov_base = v2_ext;
1016 vec[i++].iov_len = sizeof(*v2_ext) +
1017 (v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
1018 if (smcd_indicated(ini->smc_type_v2)) {
1019 vec[i].iov_base = smcd_v2_ext;
1020 vec[i++].iov_len = sizeof(*smcd_v2_ext);
1021 if (ini->ism_offered_cnt) {
1022 vec[i].iov_base = gidchids;
1023 vec[i++].iov_len = v2_ext->hdr.ism_gid_cnt *
1024 sizeof(struct smc_clc_smcd_gid_chid);
1025 }
1026 }
1027 }
1028 vec[i].iov_base = trl;
1029 vec[i++].iov_len = sizeof(*trl);
1030 /* due to the few bytes needed for clc-handshake this cannot block */
1031 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
1032 if (len < 0) {
1033 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1034 reason_code = -smc->sk.sk_err;
1035 } else if (len < ntohs(pclc_base->hdr.length)) {
1036 reason_code = -ENETUNREACH;
1037 smc->sk.sk_err = -reason_code;
1038 }
1039
1040 kfree(pclc);
1041 return reason_code;
1042}
1043
1044static void
1045smcd_clc_prep_confirm_accept(struct smc_connection *conn,
1046 struct smc_clc_msg_accept_confirm *clc,
1047 int first_contact, u8 version,
1048 u8 *eid, struct smc_init_info *ini,
1049 int *fce_len,
1050 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1051 struct smc_clc_msg_trail *trl)
1052{
1053 struct smcd_dev *smcd = conn->lgr->smcd;
1054 struct smcd_gid smcd_gid;
1055 u16 chid;
1056 int len;
1057
1058 /* SMC-D specific settings */
1059 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
1060 sizeof(SMCD_EYECATCHER));
1061 smcd->ops->get_local_gid(smcd, &smcd_gid);
1062 clc->hdr.typev1 = SMC_TYPE_D;
1063 clc->d0.gid = htonll(smcd_gid.gid);
1064 clc->d0.token = htonll(conn->rmb_desc->token);
1065 clc->d0.dmbe_size = conn->rmbe_size_comp;
1066 clc->d0.dmbe_idx = 0;
1067 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
1068 if (version == SMC_V1) {
1069 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
1070 } else {
1071 chid = smc_ism_get_chid(smcd);
1072 clc->d1.chid = htons(chid);
1073 if (eid && eid[0])
1074 memcpy(clc->d1.eid, eid, SMC_MAX_EID_LEN);
1075 if (__smc_ism_is_emulated(chid))
1076 clc->d1.gid_ext = htonll(smcd_gid.gid_ext);
1077 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
1078 if (first_contact) {
1079 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1080 len += *fce_len;
1081 }
1082 clc->hdr.length = htons(len);
1083 }
1084 memcpy(trl->eyecatcher, SMCD_EYECATCHER,
1085 sizeof(SMCD_EYECATCHER));
1086}
1087
1088static void
1089smcr_clc_prep_confirm_accept(struct smc_connection *conn,
1090 struct smc_clc_msg_accept_confirm *clc,
1091 int first_contact, u8 version,
1092 u8 *eid, struct smc_init_info *ini,
1093 int *fce_len,
1094 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1095 struct smc_clc_fce_gid_ext *gle,
1096 struct smc_clc_msg_trail *trl)
1097{
1098 struct smc_link *link = conn->lnk;
1099 int len;
1100
1101 /* SMC-R specific settings */
1102 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
1103 sizeof(SMC_EYECATCHER));
1104 clc->hdr.typev1 = SMC_TYPE_R;
1105 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
1106 sizeof(local_systemid));
1107 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
1108 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
1109 ETH_ALEN);
1110 hton24(clc->r0.qpn, link->roce_qp->qp_num);
1111 clc->r0.rmb_rkey =
1112 htonl(conn->rmb_desc->mr[link->link_idx]->rkey);
1113 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
1114 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
1115 switch (clc->hdr.type) {
1116 case SMC_CLC_ACCEPT:
1117 clc->r0.qp_mtu = link->path_mtu;
1118 break;
1119 case SMC_CLC_CONFIRM:
1120 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
1121 break;
1122 }
1123 clc->r0.rmbe_size = conn->rmbe_size_comp;
1124 clc->r0.rmb_dma_addr = conn->rmb_desc->is_vm ?
1125 cpu_to_be64((uintptr_t)conn->rmb_desc->cpu_addr) :
1126 cpu_to_be64((u64)sg_dma_address
1127 (conn->rmb_desc->sgt[link->link_idx].sgl));
1128 hton24(clc->r0.psn, link->psn_initial);
1129 if (version == SMC_V1) {
1130 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1131 } else {
1132 if (eid && eid[0])
1133 memcpy(clc->r1.eid, eid, SMC_MAX_EID_LEN);
1134 len = SMCR_CLC_ACCEPT_CONFIRM_LEN_V2;
1135 if (first_contact) {
1136 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1137 len += *fce_len;
1138 fce_v2x->fce_v2_base.v2_direct =
1139 !link->lgr->uses_gateway;
1140 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1141 memset(gle, 0, sizeof(*gle));
1142 gle->gid_cnt = ini->smcrv2.gidlist.len;
1143 len += sizeof(*gle);
1144 len += gle->gid_cnt * sizeof(gle->gid[0]);
1145 }
1146 }
1147 clc->hdr.length = htons(len);
1148 }
1149 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
1150}
1151
1152/* build and send CLC CONFIRM / ACCEPT message */
1153static int smc_clc_send_confirm_accept(struct smc_sock *smc,
1154 struct smc_clc_msg_accept_confirm *clc,
1155 int first_contact, u8 version,
1156 u8 *eid, struct smc_init_info *ini)
1157{
1158 struct smc_clc_first_contact_ext_v2x fce_v2x;
1159 struct smc_connection *conn = &smc->conn;
1160 struct smc_clc_fce_gid_ext gle;
1161 struct smc_clc_msg_trail trl;
1162 int i, fce_len;
1163 struct kvec vec[5];
1164 struct msghdr msg;
1165
1166 /* send SMC Confirm CLC msg */
1167 clc->hdr.version = version; /* SMC version */
1168 if (first_contact)
1169 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
1170 if (conn->lgr->is_smcd)
1171 smcd_clc_prep_confirm_accept(conn, clc, first_contact,
1172 version, eid, ini, &fce_len,
1173 &fce_v2x, &trl);
1174 else
1175 smcr_clc_prep_confirm_accept(conn, clc, first_contact,
1176 version, eid, ini, &fce_len,
1177 &fce_v2x, &gle, &trl);
1178 memset(&msg, 0, sizeof(msg));
1179 i = 0;
1180 vec[i].iov_base = clc;
1181 if (version > SMC_V1)
1182 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1183 SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 :
1184 SMCR_CLC_ACCEPT_CONFIRM_LEN_V2) -
1185 sizeof(trl);
1186 else
1187 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1188 SMCD_CLC_ACCEPT_CONFIRM_LEN :
1189 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
1190 sizeof(trl);
1191 if (version > SMC_V1 && first_contact) {
1192 vec[i].iov_base = &fce_v2x;
1193 vec[i++].iov_len = fce_len;
1194 if (!conn->lgr->is_smcd) {
1195 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1196 vec[i].iov_base = &gle;
1197 vec[i++].iov_len = sizeof(gle);
1198 vec[i].iov_base = &ini->smcrv2.gidlist.list;
1199 vec[i++].iov_len = gle.gid_cnt *
1200 sizeof(gle.gid[0]);
1201 }
1202 }
1203 }
1204 vec[i].iov_base = &trl;
1205 vec[i++].iov_len = sizeof(trl);
1206 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
1207 ntohs(clc->hdr.length));
1208}
1209
1210/* send CLC CONFIRM message across internal TCP socket */
1211int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
1212 u8 version, u8 *eid, struct smc_init_info *ini)
1213{
1214 struct smc_clc_msg_accept_confirm cclc;
1215 int reason_code = 0;
1216 int len;
1217
1218 /* send SMC Confirm CLC msg */
1219 memset(&cclc, 0, sizeof(cclc));
1220 cclc.hdr.type = SMC_CLC_CONFIRM;
1221 len = smc_clc_send_confirm_accept(smc, &cclc, clnt_first_contact,
1222 version, eid, ini);
1223 if (len < ntohs(cclc.hdr.length)) {
1224 if (len >= 0) {
1225 reason_code = -ENETUNREACH;
1226 smc->sk.sk_err = -reason_code;
1227 } else {
1228 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1229 reason_code = -smc->sk.sk_err;
1230 }
1231 }
1232 return reason_code;
1233}
1234
1235/* send CLC ACCEPT message across internal TCP socket */
1236int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
1237 u8 version, u8 *negotiated_eid, struct smc_init_info *ini)
1238{
1239 struct smc_clc_msg_accept_confirm aclc;
1240 int len;
1241
1242 memset(&aclc, 0, sizeof(aclc));
1243 aclc.hdr.type = SMC_CLC_ACCEPT;
1244 len = smc_clc_send_confirm_accept(new_smc, &aclc, srv_first_contact,
1245 version, negotiated_eid, ini);
1246 if (len < ntohs(aclc.hdr.length))
1247 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
1248
1249 return len > 0 ? 0 : len;
1250}
1251
1252int smc_clc_srv_v2x_features_validate(struct smc_sock *smc,
1253 struct smc_clc_msg_proposal *pclc,
1254 struct smc_init_info *ini)
1255{
1256 struct smc_clc_v2_extension *pclc_v2_ext;
1257 struct net *net = sock_net(&smc->sk);
1258
1259 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1260 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1261 ini->feature_mask = SMC_FEATURE_MASK;
1262
1263 if ((!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) ||
1264 ini->release_nr < SMC_RELEASE_1)
1265 return 0;
1266
1267 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1268 if (!pclc_v2_ext)
1269 return SMC_CLC_DECL_NOV2EXT;
1270
1271 if (ini->smcr_version & SMC_V2) {
1272 ini->max_conns = min_t(u8, pclc_v2_ext->max_conns,
1273 net->smc.sysctl_max_conns_per_lgr);
1274 if (ini->max_conns < SMC_CONN_PER_LGR_MIN)
1275 return SMC_CLC_DECL_MAXCONNERR;
1276
1277 ini->max_links = min_t(u8, pclc_v2_ext->max_links,
1278 net->smc.sysctl_max_links_per_lgr);
1279 if (ini->max_links < SMC_LINKS_ADD_LNK_MIN)
1280 return SMC_CLC_DECL_MAXLINKERR;
1281 }
1282
1283 return 0;
1284}
1285
1286int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
1287 struct smc_init_info *ini)
1288{
1289 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1290 (struct smc_clc_first_contact_ext_v2x *)fce;
1291
1292 if (ini->release_nr < SMC_RELEASE_1)
1293 return 0;
1294
1295 if (!ini->is_smcd) {
1296 if (fce_v2x->max_conns < SMC_CONN_PER_LGR_MIN)
1297 return SMC_CLC_DECL_MAXCONNERR;
1298 ini->max_conns = fce_v2x->max_conns;
1299
1300 if (fce_v2x->max_links > SMC_LINKS_ADD_LNK_MAX ||
1301 fce_v2x->max_links < SMC_LINKS_ADD_LNK_MIN)
1302 return SMC_CLC_DECL_MAXLINKERR;
1303 ini->max_links = fce_v2x->max_links;
1304 }
1305 /* common supplemental features of server and client */
1306 ini->feature_mask = ntohs(fce_v2x->feature_mask) & SMC_FEATURE_MASK;
1307
1308 return 0;
1309}
1310
1311int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
1312 struct smc_init_info *ini)
1313{
1314 struct smc_clc_first_contact_ext *fce =
1315 smc_get_clc_first_contact_ext(cclc, ini->is_smcd);
1316 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1317 (struct smc_clc_first_contact_ext_v2x *)fce;
1318
1319 if (cclc->hdr.version == SMC_V1 ||
1320 !(cclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
1321 return 0;
1322
1323 if (ini->release_nr != fce->release)
1324 return SMC_CLC_DECL_RELEASEERR;
1325
1326 if (fce->release < SMC_RELEASE_1)
1327 return 0;
1328
1329 if (!ini->is_smcd) {
1330 if (fce_v2x->max_conns != ini->max_conns)
1331 return SMC_CLC_DECL_MAXCONNERR;
1332 if (fce_v2x->max_links != ini->max_links)
1333 return SMC_CLC_DECL_MAXLINKERR;
1334 }
1335 /* common supplemental features returned by client */
1336 ini->feature_mask = ntohs(fce_v2x->feature_mask);
1337
1338 return 0;
1339}
1340
1341void smc_clc_get_hostname(u8 **host)
1342{
1343 *host = &smc_hostname[0];
1344}
1345
1346void __init smc_clc_init(void)
1347{
1348 struct new_utsname *u;
1349
1350 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
1351 u = utsname();
1352 memcpy(smc_hostname, u->nodename,
1353 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
1354
1355 INIT_LIST_HEAD(&smc_clc_eid_table.list);
1356 rwlock_init(&smc_clc_eid_table.lock);
1357 smc_clc_eid_table.ueid_cnt = 0;
1358#if IS_ENABLED(CONFIG_S390)
1359 smc_clc_eid_table.seid_enabled = 1;
1360#else
1361 smc_clc_eid_table.seid_enabled = 0;
1362#endif
1363}
1364
1365void smc_clc_exit(void)
1366{
1367 smc_clc_ueid_remove(NULL);
1368}
1// SPDX-License-Identifier: GPL-2.0
2/*
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
4 *
5 * CLC (connection layer control) handshake over initial TCP socket to
6 * prepare for RDMA traffic
7 *
8 * Copyright IBM Corp. 2016, 2018
9 *
10 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
11 */
12
13#include <linux/in.h>
14#include <linux/inetdevice.h>
15#include <linux/if_ether.h>
16#include <linux/sched/signal.h>
17
18#include <net/addrconf.h>
19#include <net/sock.h>
20#include <net/tcp.h>
21
22#include "smc.h"
23#include "smc_core.h"
24#include "smc_clc.h"
25#include "smc_ib.h"
26#include "smc_ism.h"
27
28#define SMCR_CLC_ACCEPT_CONFIRM_LEN 68
29#define SMCD_CLC_ACCEPT_CONFIRM_LEN 48
30
31/* eye catcher "SMCR" EBCDIC for CLC messages */
32static const char SMC_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xd9'};
33/* eye catcher "SMCD" EBCDIC for CLC messages */
34static const char SMCD_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xc4'};
35
36/* check if received message has a correct header length and contains valid
37 * heading and trailing eyecatchers
38 */
39static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm)
40{
41 struct smc_clc_msg_proposal_prefix *pclc_prfx;
42 struct smc_clc_msg_accept_confirm *clc;
43 struct smc_clc_msg_proposal *pclc;
44 struct smc_clc_msg_decline *dclc;
45 struct smc_clc_msg_trail *trl;
46
47 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
48 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
49 return false;
50 switch (clcm->type) {
51 case SMC_CLC_PROPOSAL:
52 if (clcm->path != SMC_TYPE_R && clcm->path != SMC_TYPE_D &&
53 clcm->path != SMC_TYPE_B)
54 return false;
55 pclc = (struct smc_clc_msg_proposal *)clcm;
56 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
57 if (ntohs(pclc->hdr.length) !=
58 sizeof(*pclc) + ntohs(pclc->iparea_offset) +
59 sizeof(*pclc_prfx) +
60 pclc_prfx->ipv6_prefixes_cnt *
61 sizeof(struct smc_clc_ipv6_prefix) +
62 sizeof(*trl))
63 return false;
64 trl = (struct smc_clc_msg_trail *)
65 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
66 break;
67 case SMC_CLC_ACCEPT:
68 case SMC_CLC_CONFIRM:
69 if (clcm->path != SMC_TYPE_R && clcm->path != SMC_TYPE_D)
70 return false;
71 clc = (struct smc_clc_msg_accept_confirm *)clcm;
72 if ((clcm->path == SMC_TYPE_R &&
73 ntohs(clc->hdr.length) != SMCR_CLC_ACCEPT_CONFIRM_LEN) ||
74 (clcm->path == SMC_TYPE_D &&
75 ntohs(clc->hdr.length) != SMCD_CLC_ACCEPT_CONFIRM_LEN))
76 return false;
77 trl = (struct smc_clc_msg_trail *)
78 ((u8 *)clc + ntohs(clc->hdr.length) - sizeof(*trl));
79 break;
80 case SMC_CLC_DECLINE:
81 dclc = (struct smc_clc_msg_decline *)clcm;
82 if (ntohs(dclc->hdr.length) != sizeof(*dclc))
83 return false;
84 trl = &dclc->trl;
85 break;
86 default:
87 return false;
88 }
89 if (memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
90 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
91 return false;
92 return true;
93}
94
95/* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
96static int smc_clc_prfx_set4_rcu(struct dst_entry *dst, __be32 ipv4,
97 struct smc_clc_msg_proposal_prefix *prop)
98{
99 struct in_device *in_dev = __in_dev_get_rcu(dst->dev);
100 const struct in_ifaddr *ifa;
101
102 if (!in_dev)
103 return -ENODEV;
104
105 in_dev_for_each_ifa_rcu(ifa, in_dev) {
106 if (!inet_ifa_match(ipv4, ifa))
107 continue;
108 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
109 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
110 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
111 return 0;
112 }
113 return -ENOENT;
114}
115
116/* fill CLC proposal msg with ipv6 prefixes from device */
117static int smc_clc_prfx_set6_rcu(struct dst_entry *dst,
118 struct smc_clc_msg_proposal_prefix *prop,
119 struct smc_clc_ipv6_prefix *ipv6_prfx)
120{
121#if IS_ENABLED(CONFIG_IPV6)
122 struct inet6_dev *in6_dev = __in6_dev_get(dst->dev);
123 struct inet6_ifaddr *ifa;
124 int cnt = 0;
125
126 if (!in6_dev)
127 return -ENODEV;
128 /* use a maximum of 8 IPv6 prefixes from device */
129 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
130 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
131 continue;
132 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
133 &ifa->addr, ifa->prefix_len);
134 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
135 cnt++;
136 if (cnt == SMC_CLC_MAX_V6_PREFIX)
137 break;
138 }
139 prop->ipv6_prefixes_cnt = cnt;
140 if (cnt)
141 return 0;
142#endif
143 return -ENOENT;
144}
145
146/* retrieve and set prefixes in CLC proposal msg */
147static int smc_clc_prfx_set(struct socket *clcsock,
148 struct smc_clc_msg_proposal_prefix *prop,
149 struct smc_clc_ipv6_prefix *ipv6_prfx)
150{
151 struct dst_entry *dst = sk_dst_get(clcsock->sk);
152 struct sockaddr_storage addrs;
153 struct sockaddr_in6 *addr6;
154 struct sockaddr_in *addr;
155 int rc = -ENOENT;
156
157 memset(prop, 0, sizeof(*prop));
158 if (!dst) {
159 rc = -ENOTCONN;
160 goto out;
161 }
162 if (!dst->dev) {
163 rc = -ENODEV;
164 goto out_rel;
165 }
166 /* get address to which the internal TCP socket is bound */
167 kernel_getsockname(clcsock, (struct sockaddr *)&addrs);
168 /* analyze IP specific data of net_device belonging to TCP socket */
169 addr6 = (struct sockaddr_in6 *)&addrs;
170 rcu_read_lock();
171 if (addrs.ss_family == PF_INET) {
172 /* IPv4 */
173 addr = (struct sockaddr_in *)&addrs;
174 rc = smc_clc_prfx_set4_rcu(dst, addr->sin_addr.s_addr, prop);
175 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
176 /* mapped IPv4 address - peer is IPv4 only */
177 rc = smc_clc_prfx_set4_rcu(dst, addr6->sin6_addr.s6_addr32[3],
178 prop);
179 } else {
180 /* IPv6 */
181 rc = smc_clc_prfx_set6_rcu(dst, prop, ipv6_prfx);
182 }
183 rcu_read_unlock();
184out_rel:
185 dst_release(dst);
186out:
187 return rc;
188}
189
190/* match ipv4 addrs of dev against addr in CLC proposal */
191static int smc_clc_prfx_match4_rcu(struct net_device *dev,
192 struct smc_clc_msg_proposal_prefix *prop)
193{
194 struct in_device *in_dev = __in_dev_get_rcu(dev);
195 const struct in_ifaddr *ifa;
196
197 if (!in_dev)
198 return -ENODEV;
199 in_dev_for_each_ifa_rcu(ifa, in_dev) {
200 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
201 inet_ifa_match(prop->outgoing_subnet, ifa))
202 return 0;
203 }
204
205 return -ENOENT;
206}
207
208/* match ipv6 addrs of dev against addrs in CLC proposal */
209static int smc_clc_prfx_match6_rcu(struct net_device *dev,
210 struct smc_clc_msg_proposal_prefix *prop)
211{
212#if IS_ENABLED(CONFIG_IPV6)
213 struct inet6_dev *in6_dev = __in6_dev_get(dev);
214 struct smc_clc_ipv6_prefix *ipv6_prfx;
215 struct inet6_ifaddr *ifa;
216 int i, max;
217
218 if (!in6_dev)
219 return -ENODEV;
220 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
221 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
222 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
223 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
224 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
225 continue;
226 for (i = 0; i < max; i++) {
227 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
228 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
229 ifa->prefix_len))
230 return 0;
231 }
232 }
233#endif
234 return -ENOENT;
235}
236
237/* check if proposed prefixes match one of our device prefixes */
238int smc_clc_prfx_match(struct socket *clcsock,
239 struct smc_clc_msg_proposal_prefix *prop)
240{
241 struct dst_entry *dst = sk_dst_get(clcsock->sk);
242 int rc;
243
244 if (!dst) {
245 rc = -ENOTCONN;
246 goto out;
247 }
248 if (!dst->dev) {
249 rc = -ENODEV;
250 goto out_rel;
251 }
252 rcu_read_lock();
253 if (!prop->ipv6_prefixes_cnt)
254 rc = smc_clc_prfx_match4_rcu(dst->dev, prop);
255 else
256 rc = smc_clc_prfx_match6_rcu(dst->dev, prop);
257 rcu_read_unlock();
258out_rel:
259 dst_release(dst);
260out:
261 return rc;
262}
263
264/* Wait for data on the tcp-socket, analyze received data
265 * Returns:
266 * 0 if success and it was not a decline that we received.
267 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
268 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
269 */
270int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
271 u8 expected_type, unsigned long timeout)
272{
273 long rcvtimeo = smc->clcsock->sk->sk_rcvtimeo;
274 struct sock *clc_sk = smc->clcsock->sk;
275 struct smc_clc_msg_hdr *clcm = buf;
276 struct msghdr msg = {NULL, 0};
277 int reason_code = 0;
278 struct kvec vec = {buf, buflen};
279 int len, datlen;
280 int krflags;
281
282 /* peek the first few bytes to determine length of data to receive
283 * so we don't consume any subsequent CLC message or payload data
284 * in the TCP byte stream
285 */
286 /*
287 * Caller must make sure that buflen is no less than
288 * sizeof(struct smc_clc_msg_hdr)
289 */
290 krflags = MSG_PEEK | MSG_WAITALL;
291 clc_sk->sk_rcvtimeo = timeout;
292 iov_iter_kvec(&msg.msg_iter, READ, &vec, 1,
293 sizeof(struct smc_clc_msg_hdr));
294 len = sock_recvmsg(smc->clcsock, &msg, krflags);
295 if (signal_pending(current)) {
296 reason_code = -EINTR;
297 clc_sk->sk_err = EINTR;
298 smc->sk.sk_err = EINTR;
299 goto out;
300 }
301 if (clc_sk->sk_err) {
302 reason_code = -clc_sk->sk_err;
303 if (clc_sk->sk_err == EAGAIN &&
304 expected_type == SMC_CLC_DECLINE)
305 clc_sk->sk_err = 0; /* reset for fallback usage */
306 else
307 smc->sk.sk_err = clc_sk->sk_err;
308 goto out;
309 }
310 if (!len) { /* peer has performed orderly shutdown */
311 smc->sk.sk_err = ECONNRESET;
312 reason_code = -ECONNRESET;
313 goto out;
314 }
315 if (len < 0) {
316 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
317 smc->sk.sk_err = -len;
318 reason_code = len;
319 goto out;
320 }
321 datlen = ntohs(clcm->length);
322 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
323 (datlen > buflen) ||
324 (clcm->version != SMC_CLC_V1) ||
325 (clcm->path != SMC_TYPE_R && clcm->path != SMC_TYPE_D &&
326 clcm->path != SMC_TYPE_B) ||
327 ((clcm->type != SMC_CLC_DECLINE) &&
328 (clcm->type != expected_type))) {
329 smc->sk.sk_err = EPROTO;
330 reason_code = -EPROTO;
331 goto out;
332 }
333
334 /* receive the complete CLC message */
335 memset(&msg, 0, sizeof(struct msghdr));
336 iov_iter_kvec(&msg.msg_iter, READ, &vec, 1, datlen);
337 krflags = MSG_WAITALL;
338 len = sock_recvmsg(smc->clcsock, &msg, krflags);
339 if (len < datlen || !smc_clc_msg_hdr_valid(clcm)) {
340 smc->sk.sk_err = EPROTO;
341 reason_code = -EPROTO;
342 goto out;
343 }
344 if (clcm->type == SMC_CLC_DECLINE) {
345 struct smc_clc_msg_decline *dclc;
346
347 dclc = (struct smc_clc_msg_decline *)clcm;
348 reason_code = SMC_CLC_DECL_PEERDECL;
349 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
350 if (((struct smc_clc_msg_decline *)buf)->hdr.flag) {
351 smc->conn.lgr->sync_err = 1;
352 smc_lgr_terminate(smc->conn.lgr);
353 }
354 }
355
356out:
357 clc_sk->sk_rcvtimeo = rcvtimeo;
358 return reason_code;
359}
360
361/* send CLC DECLINE message across internal TCP socket */
362int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info)
363{
364 struct smc_clc_msg_decline dclc;
365 struct msghdr msg;
366 struct kvec vec;
367 int len;
368
369 memset(&dclc, 0, sizeof(dclc));
370 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
371 dclc.hdr.type = SMC_CLC_DECLINE;
372 dclc.hdr.length = htons(sizeof(struct smc_clc_msg_decline));
373 dclc.hdr.version = SMC_CLC_V1;
374 dclc.hdr.flag = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ? 1 : 0;
375 memcpy(dclc.id_for_peer, local_systemid, sizeof(local_systemid));
376 dclc.peer_diagnosis = htonl(peer_diag_info);
377 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
378
379 memset(&msg, 0, sizeof(msg));
380 vec.iov_base = &dclc;
381 vec.iov_len = sizeof(struct smc_clc_msg_decline);
382 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1,
383 sizeof(struct smc_clc_msg_decline));
384 if (len < 0 || len < sizeof(struct smc_clc_msg_decline))
385 len = -EPROTO;
386 return len > 0 ? 0 : len;
387}
388
389/* send CLC PROPOSAL message across internal TCP socket */
390int smc_clc_send_proposal(struct smc_sock *smc, int smc_type,
391 struct smc_init_info *ini)
392{
393 struct smc_clc_ipv6_prefix ipv6_prfx[SMC_CLC_MAX_V6_PREFIX];
394 struct smc_clc_msg_proposal_prefix pclc_prfx;
395 struct smc_clc_msg_smcd pclc_smcd;
396 struct smc_clc_msg_proposal pclc;
397 struct smc_clc_msg_trail trl;
398 int len, i, plen, rc;
399 int reason_code = 0;
400 struct kvec vec[5];
401 struct msghdr msg;
402
403 /* retrieve ip prefixes for CLC proposal msg */
404 rc = smc_clc_prfx_set(smc->clcsock, &pclc_prfx, ipv6_prfx);
405 if (rc)
406 return SMC_CLC_DECL_CNFERR; /* configuration error */
407
408 /* send SMC Proposal CLC message */
409 plen = sizeof(pclc) + sizeof(pclc_prfx) +
410 (pclc_prfx.ipv6_prefixes_cnt * sizeof(ipv6_prfx[0])) +
411 sizeof(trl);
412 memset(&pclc, 0, sizeof(pclc));
413 memcpy(pclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
414 pclc.hdr.type = SMC_CLC_PROPOSAL;
415 pclc.hdr.version = SMC_CLC_V1; /* SMC version */
416 pclc.hdr.path = smc_type;
417 if (smc_type == SMC_TYPE_R || smc_type == SMC_TYPE_B) {
418 /* add SMC-R specifics */
419 memcpy(pclc.lcl.id_for_peer, local_systemid,
420 sizeof(local_systemid));
421 memcpy(&pclc.lcl.gid, ini->ib_gid, SMC_GID_SIZE);
422 memcpy(&pclc.lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
423 ETH_ALEN);
424 pclc.iparea_offset = htons(0);
425 }
426 if (smc_type == SMC_TYPE_D || smc_type == SMC_TYPE_B) {
427 /* add SMC-D specifics */
428 memset(&pclc_smcd, 0, sizeof(pclc_smcd));
429 plen += sizeof(pclc_smcd);
430 pclc.iparea_offset = htons(SMC_CLC_PROPOSAL_MAX_OFFSET);
431 pclc_smcd.gid = ini->ism_dev->local_gid;
432 }
433 pclc.hdr.length = htons(plen);
434
435 memcpy(trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
436 memset(&msg, 0, sizeof(msg));
437 i = 0;
438 vec[i].iov_base = &pclc;
439 vec[i++].iov_len = sizeof(pclc);
440 if (smc_type == SMC_TYPE_D || smc_type == SMC_TYPE_B) {
441 vec[i].iov_base = &pclc_smcd;
442 vec[i++].iov_len = sizeof(pclc_smcd);
443 }
444 vec[i].iov_base = &pclc_prfx;
445 vec[i++].iov_len = sizeof(pclc_prfx);
446 if (pclc_prfx.ipv6_prefixes_cnt > 0) {
447 vec[i].iov_base = &ipv6_prfx[0];
448 vec[i++].iov_len = pclc_prfx.ipv6_prefixes_cnt *
449 sizeof(ipv6_prfx[0]);
450 }
451 vec[i].iov_base = &trl;
452 vec[i++].iov_len = sizeof(trl);
453 /* due to the few bytes needed for clc-handshake this cannot block */
454 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
455 if (len < 0) {
456 smc->sk.sk_err = smc->clcsock->sk->sk_err;
457 reason_code = -smc->sk.sk_err;
458 } else if (len < (int)sizeof(pclc)) {
459 reason_code = -ENETUNREACH;
460 smc->sk.sk_err = -reason_code;
461 }
462
463 return reason_code;
464}
465
466/* send CLC CONFIRM message across internal TCP socket */
467int smc_clc_send_confirm(struct smc_sock *smc)
468{
469 struct smc_connection *conn = &smc->conn;
470 struct smc_clc_msg_accept_confirm cclc;
471 struct smc_link *link;
472 int reason_code = 0;
473 struct msghdr msg;
474 struct kvec vec;
475 int len;
476
477 /* send SMC Confirm CLC msg */
478 memset(&cclc, 0, sizeof(cclc));
479 cclc.hdr.type = SMC_CLC_CONFIRM;
480 cclc.hdr.version = SMC_CLC_V1; /* SMC version */
481 if (smc->conn.lgr->is_smcd) {
482 /* SMC-D specific settings */
483 memcpy(cclc.hdr.eyecatcher, SMCD_EYECATCHER,
484 sizeof(SMCD_EYECATCHER));
485 cclc.hdr.path = SMC_TYPE_D;
486 cclc.hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
487 cclc.gid = conn->lgr->smcd->local_gid;
488 cclc.token = conn->rmb_desc->token;
489 cclc.dmbe_size = conn->rmbe_size_short;
490 cclc.dmbe_idx = 0;
491 memcpy(&cclc.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
492 memcpy(cclc.smcd_trl.eyecatcher, SMCD_EYECATCHER,
493 sizeof(SMCD_EYECATCHER));
494 } else {
495 /* SMC-R specific settings */
496 link = &conn->lgr->lnk[SMC_SINGLE_LINK];
497 memcpy(cclc.hdr.eyecatcher, SMC_EYECATCHER,
498 sizeof(SMC_EYECATCHER));
499 cclc.hdr.path = SMC_TYPE_R;
500 cclc.hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
501 memcpy(cclc.lcl.id_for_peer, local_systemid,
502 sizeof(local_systemid));
503 memcpy(&cclc.lcl.gid, link->gid, SMC_GID_SIZE);
504 memcpy(&cclc.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
505 ETH_ALEN);
506 hton24(cclc.qpn, link->roce_qp->qp_num);
507 cclc.rmb_rkey =
508 htonl(conn->rmb_desc->mr_rx[SMC_SINGLE_LINK]->rkey);
509 cclc.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
510 cclc.rmbe_alert_token = htonl(conn->alert_token_local);
511 cclc.qp_mtu = min(link->path_mtu, link->peer_mtu);
512 cclc.rmbe_size = conn->rmbe_size_short;
513 cclc.rmb_dma_addr = cpu_to_be64((u64)sg_dma_address
514 (conn->rmb_desc->sgt[SMC_SINGLE_LINK].sgl));
515 hton24(cclc.psn, link->psn_initial);
516 memcpy(cclc.smcr_trl.eyecatcher, SMC_EYECATCHER,
517 sizeof(SMC_EYECATCHER));
518 }
519
520 memset(&msg, 0, sizeof(msg));
521 vec.iov_base = &cclc;
522 vec.iov_len = ntohs(cclc.hdr.length);
523 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1,
524 ntohs(cclc.hdr.length));
525 if (len < ntohs(cclc.hdr.length)) {
526 if (len >= 0) {
527 reason_code = -ENETUNREACH;
528 smc->sk.sk_err = -reason_code;
529 } else {
530 smc->sk.sk_err = smc->clcsock->sk->sk_err;
531 reason_code = -smc->sk.sk_err;
532 }
533 }
534 return reason_code;
535}
536
537/* send CLC ACCEPT message across internal TCP socket */
538int smc_clc_send_accept(struct smc_sock *new_smc, int srv_first_contact)
539{
540 struct smc_connection *conn = &new_smc->conn;
541 struct smc_clc_msg_accept_confirm aclc;
542 struct smc_link *link;
543 struct msghdr msg;
544 struct kvec vec;
545 int len;
546
547 memset(&aclc, 0, sizeof(aclc));
548 aclc.hdr.type = SMC_CLC_ACCEPT;
549 aclc.hdr.version = SMC_CLC_V1; /* SMC version */
550 if (srv_first_contact)
551 aclc.hdr.flag = 1;
552
553 if (new_smc->conn.lgr->is_smcd) {
554 /* SMC-D specific settings */
555 aclc.hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
556 memcpy(aclc.hdr.eyecatcher, SMCD_EYECATCHER,
557 sizeof(SMCD_EYECATCHER));
558 aclc.hdr.path = SMC_TYPE_D;
559 aclc.gid = conn->lgr->smcd->local_gid;
560 aclc.token = conn->rmb_desc->token;
561 aclc.dmbe_size = conn->rmbe_size_short;
562 aclc.dmbe_idx = 0;
563 memcpy(&aclc.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
564 memcpy(aclc.smcd_trl.eyecatcher, SMCD_EYECATCHER,
565 sizeof(SMCD_EYECATCHER));
566 } else {
567 /* SMC-R specific settings */
568 aclc.hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
569 memcpy(aclc.hdr.eyecatcher, SMC_EYECATCHER,
570 sizeof(SMC_EYECATCHER));
571 aclc.hdr.path = SMC_TYPE_R;
572 link = &conn->lgr->lnk[SMC_SINGLE_LINK];
573 memcpy(aclc.lcl.id_for_peer, local_systemid,
574 sizeof(local_systemid));
575 memcpy(&aclc.lcl.gid, link->gid, SMC_GID_SIZE);
576 memcpy(&aclc.lcl.mac, link->smcibdev->mac[link->ibport - 1],
577 ETH_ALEN);
578 hton24(aclc.qpn, link->roce_qp->qp_num);
579 aclc.rmb_rkey =
580 htonl(conn->rmb_desc->mr_rx[SMC_SINGLE_LINK]->rkey);
581 aclc.rmbe_idx = 1; /* as long as 1 RMB = 1 RMBE */
582 aclc.rmbe_alert_token = htonl(conn->alert_token_local);
583 aclc.qp_mtu = link->path_mtu;
584 aclc.rmbe_size = conn->rmbe_size_short,
585 aclc.rmb_dma_addr = cpu_to_be64((u64)sg_dma_address
586 (conn->rmb_desc->sgt[SMC_SINGLE_LINK].sgl));
587 hton24(aclc.psn, link->psn_initial);
588 memcpy(aclc.smcr_trl.eyecatcher, SMC_EYECATCHER,
589 sizeof(SMC_EYECATCHER));
590 }
591
592 memset(&msg, 0, sizeof(msg));
593 vec.iov_base = &aclc;
594 vec.iov_len = ntohs(aclc.hdr.length);
595 len = kernel_sendmsg(new_smc->clcsock, &msg, &vec, 1,
596 ntohs(aclc.hdr.length));
597 if (len < ntohs(aclc.hdr.length))
598 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
599
600 return len > 0 ? 0 : len;
601}