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 v2_ext = smc_get_clc_v2_ext(pclc);
356 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
357 if (hdr->version == SMC_V1) {
358 if (hdr->typev1 == SMC_TYPE_N)
359 return false;
360 if (ntohs(hdr->length) !=
361 sizeof(*pclc) + ntohs(pclc->iparea_offset) +
362 sizeof(*pclc_prfx) +
363 pclc_prfx->ipv6_prefixes_cnt *
364 sizeof(struct smc_clc_ipv6_prefix) +
365 sizeof(struct smc_clc_msg_trail))
366 return false;
367 } else {
368 if (ntohs(hdr->length) !=
369 sizeof(*pclc) +
370 sizeof(struct smc_clc_msg_smcd) +
371 (hdr->typev1 != SMC_TYPE_N ?
372 sizeof(*pclc_prfx) +
373 pclc_prfx->ipv6_prefixes_cnt *
374 sizeof(struct smc_clc_ipv6_prefix) : 0) +
375 (hdr->typev2 != SMC_TYPE_N ?
376 sizeof(*v2_ext) +
377 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN : 0) +
378 (smcd_indicated(hdr->typev2) ?
379 sizeof(*smcd_v2_ext) + v2_ext->hdr.ism_gid_cnt *
380 sizeof(struct smc_clc_smcd_gid_chid) :
381 0) +
382 sizeof(struct smc_clc_msg_trail))
383 return false;
384 }
385 return true;
386}
387
388/* check arriving CLC accept or confirm */
389static bool
390smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm *clc)
391{
392 struct smc_clc_msg_hdr *hdr = &clc->hdr;
393
394 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
395 return false;
396 if (hdr->version == SMC_V1) {
397 if ((hdr->typev1 == SMC_TYPE_R &&
398 ntohs(hdr->length) != SMCR_CLC_ACCEPT_CONFIRM_LEN) ||
399 (hdr->typev1 == SMC_TYPE_D &&
400 ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN))
401 return false;
402 } else {
403 if (hdr->typev1 == SMC_TYPE_D &&
404 ntohs(hdr->length) < SMCD_CLC_ACCEPT_CONFIRM_LEN_V2)
405 return false;
406 if (hdr->typev1 == SMC_TYPE_R &&
407 ntohs(hdr->length) < SMCR_CLC_ACCEPT_CONFIRM_LEN_V2)
408 return false;
409 }
410 return true;
411}
412
413/* check arriving CLC decline */
414static bool
415smc_clc_msg_decl_valid(struct smc_clc_msg_decline *dclc)
416{
417 struct smc_clc_msg_hdr *hdr = &dclc->hdr;
418
419 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
420 return false;
421 if (hdr->version == SMC_V1) {
422 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline))
423 return false;
424 } else {
425 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline_v2))
426 return false;
427 }
428 return true;
429}
430
431static int smc_clc_fill_fce_v2x(struct smc_clc_first_contact_ext_v2x *fce_v2x,
432 struct smc_init_info *ini)
433{
434 int ret = sizeof(*fce_v2x);
435
436 memset(fce_v2x, 0, sizeof(*fce_v2x));
437 fce_v2x->fce_v2_base.os_type = SMC_CLC_OS_LINUX;
438 fce_v2x->fce_v2_base.release = ini->release_nr;
439 memcpy(fce_v2x->fce_v2_base.hostname,
440 smc_hostname, sizeof(smc_hostname));
441 if (ini->is_smcd && ini->release_nr < SMC_RELEASE_1) {
442 ret = sizeof(struct smc_clc_first_contact_ext);
443 goto out;
444 }
445
446 if (ini->release_nr >= SMC_RELEASE_1) {
447 if (!ini->is_smcd) {
448 fce_v2x->max_conns = ini->max_conns;
449 fce_v2x->max_links = ini->max_links;
450 }
451 fce_v2x->feature_mask = htons(ini->feature_mask);
452 }
453
454out:
455 return ret;
456}
457
458/* check if received message has a correct header length and contains valid
459 * heading and trailing eyecatchers
460 */
461static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm, bool check_trl)
462{
463 struct smc_clc_msg_accept_confirm *clc;
464 struct smc_clc_msg_proposal *pclc;
465 struct smc_clc_msg_decline *dclc;
466 struct smc_clc_msg_trail *trl;
467
468 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
469 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
470 return false;
471 switch (clcm->type) {
472 case SMC_CLC_PROPOSAL:
473 pclc = (struct smc_clc_msg_proposal *)clcm;
474 if (!smc_clc_msg_prop_valid(pclc))
475 return false;
476 trl = (struct smc_clc_msg_trail *)
477 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
478 break;
479 case SMC_CLC_ACCEPT:
480 case SMC_CLC_CONFIRM:
481 clc = (struct smc_clc_msg_accept_confirm *)clcm;
482 if (!smc_clc_msg_acc_conf_valid(clc))
483 return false;
484 trl = (struct smc_clc_msg_trail *)
485 ((u8 *)clc + ntohs(clc->hdr.length) - sizeof(*trl));
486 break;
487 case SMC_CLC_DECLINE:
488 dclc = (struct smc_clc_msg_decline *)clcm;
489 if (!smc_clc_msg_decl_valid(dclc))
490 return false;
491 check_trl = false;
492 break;
493 default:
494 return false;
495 }
496 if (check_trl &&
497 memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
498 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
499 return false;
500 return true;
501}
502
503/* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
504static int smc_clc_prfx_set4_rcu(struct dst_entry *dst, __be32 ipv4,
505 struct smc_clc_msg_proposal_prefix *prop)
506{
507 struct in_device *in_dev = __in_dev_get_rcu(dst->dev);
508 const struct in_ifaddr *ifa;
509
510 if (!in_dev)
511 return -ENODEV;
512
513 in_dev_for_each_ifa_rcu(ifa, in_dev) {
514 if (!inet_ifa_match(ipv4, ifa))
515 continue;
516 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
517 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
518 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
519 return 0;
520 }
521 return -ENOENT;
522}
523
524/* fill CLC proposal msg with ipv6 prefixes from device */
525static int smc_clc_prfx_set6_rcu(struct dst_entry *dst,
526 struct smc_clc_msg_proposal_prefix *prop,
527 struct smc_clc_ipv6_prefix *ipv6_prfx)
528{
529#if IS_ENABLED(CONFIG_IPV6)
530 struct inet6_dev *in6_dev = __in6_dev_get(dst->dev);
531 struct inet6_ifaddr *ifa;
532 int cnt = 0;
533
534 if (!in6_dev)
535 return -ENODEV;
536 /* use a maximum of 8 IPv6 prefixes from device */
537 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
538 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
539 continue;
540 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
541 &ifa->addr, ifa->prefix_len);
542 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
543 cnt++;
544 if (cnt == SMC_CLC_MAX_V6_PREFIX)
545 break;
546 }
547 prop->ipv6_prefixes_cnt = cnt;
548 if (cnt)
549 return 0;
550#endif
551 return -ENOENT;
552}
553
554/* retrieve and set prefixes in CLC proposal msg */
555static int smc_clc_prfx_set(struct socket *clcsock,
556 struct smc_clc_msg_proposal_prefix *prop,
557 struct smc_clc_ipv6_prefix *ipv6_prfx)
558{
559 struct dst_entry *dst = sk_dst_get(clcsock->sk);
560 struct sockaddr_storage addrs;
561 struct sockaddr_in6 *addr6;
562 struct sockaddr_in *addr;
563 int rc = -ENOENT;
564
565 if (!dst) {
566 rc = -ENOTCONN;
567 goto out;
568 }
569 if (!dst->dev) {
570 rc = -ENODEV;
571 goto out_rel;
572 }
573 /* get address to which the internal TCP socket is bound */
574 if (kernel_getsockname(clcsock, (struct sockaddr *)&addrs) < 0)
575 goto out_rel;
576 /* analyze IP specific data of net_device belonging to TCP socket */
577 addr6 = (struct sockaddr_in6 *)&addrs;
578 rcu_read_lock();
579 if (addrs.ss_family == PF_INET) {
580 /* IPv4 */
581 addr = (struct sockaddr_in *)&addrs;
582 rc = smc_clc_prfx_set4_rcu(dst, addr->sin_addr.s_addr, prop);
583 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
584 /* mapped IPv4 address - peer is IPv4 only */
585 rc = smc_clc_prfx_set4_rcu(dst, addr6->sin6_addr.s6_addr32[3],
586 prop);
587 } else {
588 /* IPv6 */
589 rc = smc_clc_prfx_set6_rcu(dst, prop, ipv6_prfx);
590 }
591 rcu_read_unlock();
592out_rel:
593 dst_release(dst);
594out:
595 return rc;
596}
597
598/* match ipv4 addrs of dev against addr in CLC proposal */
599static int smc_clc_prfx_match4_rcu(struct net_device *dev,
600 struct smc_clc_msg_proposal_prefix *prop)
601{
602 struct in_device *in_dev = __in_dev_get_rcu(dev);
603 const struct in_ifaddr *ifa;
604
605 if (!in_dev)
606 return -ENODEV;
607 in_dev_for_each_ifa_rcu(ifa, in_dev) {
608 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
609 inet_ifa_match(prop->outgoing_subnet, ifa))
610 return 0;
611 }
612
613 return -ENOENT;
614}
615
616/* match ipv6 addrs of dev against addrs in CLC proposal */
617static int smc_clc_prfx_match6_rcu(struct net_device *dev,
618 struct smc_clc_msg_proposal_prefix *prop)
619{
620#if IS_ENABLED(CONFIG_IPV6)
621 struct inet6_dev *in6_dev = __in6_dev_get(dev);
622 struct smc_clc_ipv6_prefix *ipv6_prfx;
623 struct inet6_ifaddr *ifa;
624 int i, max;
625
626 if (!in6_dev)
627 return -ENODEV;
628 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
629 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
630 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
631 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
632 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
633 continue;
634 for (i = 0; i < max; i++) {
635 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
636 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
637 ifa->prefix_len))
638 return 0;
639 }
640 }
641#endif
642 return -ENOENT;
643}
644
645/* check if proposed prefixes match one of our device prefixes */
646int smc_clc_prfx_match(struct socket *clcsock,
647 struct smc_clc_msg_proposal_prefix *prop)
648{
649 struct dst_entry *dst = sk_dst_get(clcsock->sk);
650 int rc;
651
652 if (!dst) {
653 rc = -ENOTCONN;
654 goto out;
655 }
656 if (!dst->dev) {
657 rc = -ENODEV;
658 goto out_rel;
659 }
660 rcu_read_lock();
661 if (!prop->ipv6_prefixes_cnt)
662 rc = smc_clc_prfx_match4_rcu(dst->dev, prop);
663 else
664 rc = smc_clc_prfx_match6_rcu(dst->dev, prop);
665 rcu_read_unlock();
666out_rel:
667 dst_release(dst);
668out:
669 return rc;
670}
671
672/* Wait for data on the tcp-socket, analyze received data
673 * Returns:
674 * 0 if success and it was not a decline that we received.
675 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
676 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
677 */
678int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
679 u8 expected_type, unsigned long timeout)
680{
681 long rcvtimeo = smc->clcsock->sk->sk_rcvtimeo;
682 struct sock *clc_sk = smc->clcsock->sk;
683 struct smc_clc_msg_hdr *clcm = buf;
684 struct msghdr msg = {NULL, 0};
685 int reason_code = 0;
686 struct kvec vec = {buf, buflen};
687 int len, datlen, recvlen;
688 bool check_trl = true;
689 int krflags;
690
691 /* peek the first few bytes to determine length of data to receive
692 * so we don't consume any subsequent CLC message or payload data
693 * in the TCP byte stream
694 */
695 /*
696 * Caller must make sure that buflen is no less than
697 * sizeof(struct smc_clc_msg_hdr)
698 */
699 krflags = MSG_PEEK | MSG_WAITALL;
700 clc_sk->sk_rcvtimeo = timeout;
701 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1,
702 sizeof(struct smc_clc_msg_hdr));
703 len = sock_recvmsg(smc->clcsock, &msg, krflags);
704 if (signal_pending(current)) {
705 reason_code = -EINTR;
706 clc_sk->sk_err = EINTR;
707 smc->sk.sk_err = EINTR;
708 goto out;
709 }
710 if (clc_sk->sk_err) {
711 reason_code = -clc_sk->sk_err;
712 if (clc_sk->sk_err == EAGAIN &&
713 expected_type == SMC_CLC_DECLINE)
714 clc_sk->sk_err = 0; /* reset for fallback usage */
715 else
716 smc->sk.sk_err = clc_sk->sk_err;
717 goto out;
718 }
719 if (!len) { /* peer has performed orderly shutdown */
720 smc->sk.sk_err = ECONNRESET;
721 reason_code = -ECONNRESET;
722 goto out;
723 }
724 if (len < 0) {
725 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
726 smc->sk.sk_err = -len;
727 reason_code = len;
728 goto out;
729 }
730 datlen = ntohs(clcm->length);
731 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
732 (clcm->version < SMC_V1) ||
733 ((clcm->type != SMC_CLC_DECLINE) &&
734 (clcm->type != expected_type))) {
735 smc->sk.sk_err = EPROTO;
736 reason_code = -EPROTO;
737 goto out;
738 }
739
740 /* receive the complete CLC message */
741 memset(&msg, 0, sizeof(struct msghdr));
742 if (datlen > buflen) {
743 check_trl = false;
744 recvlen = buflen;
745 } else {
746 recvlen = datlen;
747 }
748 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
749 krflags = MSG_WAITALL;
750 len = sock_recvmsg(smc->clcsock, &msg, krflags);
751 if (len < recvlen || !smc_clc_msg_hdr_valid(clcm, check_trl)) {
752 smc->sk.sk_err = EPROTO;
753 reason_code = -EPROTO;
754 goto out;
755 }
756 datlen -= len;
757 while (datlen) {
758 u8 tmp[SMC_CLC_RECV_BUF_LEN];
759
760 vec.iov_base = &tmp;
761 vec.iov_len = SMC_CLC_RECV_BUF_LEN;
762 /* receive remaining proposal message */
763 recvlen = datlen > SMC_CLC_RECV_BUF_LEN ?
764 SMC_CLC_RECV_BUF_LEN : datlen;
765 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
766 len = sock_recvmsg(smc->clcsock, &msg, krflags);
767 datlen -= len;
768 }
769 if (clcm->type == SMC_CLC_DECLINE) {
770 struct smc_clc_msg_decline *dclc;
771
772 dclc = (struct smc_clc_msg_decline *)clcm;
773 reason_code = SMC_CLC_DECL_PEERDECL;
774 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
775 if (((struct smc_clc_msg_decline *)buf)->hdr.typev2 &
776 SMC_FIRST_CONTACT_MASK) {
777 smc->conn.lgr->sync_err = 1;
778 smc_lgr_terminate_sched(smc->conn.lgr);
779 }
780 }
781
782out:
783 clc_sk->sk_rcvtimeo = rcvtimeo;
784 return reason_code;
785}
786
787/* send CLC DECLINE message across internal TCP socket */
788int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version)
789{
790 struct smc_clc_msg_decline *dclc_v1;
791 struct smc_clc_msg_decline_v2 dclc;
792 struct msghdr msg;
793 int len, send_len;
794 struct kvec vec;
795
796 dclc_v1 = (struct smc_clc_msg_decline *)&dclc;
797 memset(&dclc, 0, sizeof(dclc));
798 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
799 dclc.hdr.type = SMC_CLC_DECLINE;
800 dclc.hdr.version = version;
801 dclc.os_type = version == SMC_V1 ? 0 : SMC_CLC_OS_LINUX;
802 dclc.hdr.typev2 = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ?
803 SMC_FIRST_CONTACT_MASK : 0;
804 if ((!smc_conn_lgr_valid(&smc->conn) || !smc->conn.lgr->is_smcd) &&
805 smc_ib_is_valid_local_systemid())
806 memcpy(dclc.id_for_peer, local_systemid,
807 sizeof(local_systemid));
808 dclc.peer_diagnosis = htonl(peer_diag_info);
809 if (version == SMC_V1) {
810 memcpy(dclc_v1->trl.eyecatcher, SMC_EYECATCHER,
811 sizeof(SMC_EYECATCHER));
812 send_len = sizeof(*dclc_v1);
813 } else {
814 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER,
815 sizeof(SMC_EYECATCHER));
816 send_len = sizeof(dclc);
817 }
818 dclc.hdr.length = htons(send_len);
819
820 memset(&msg, 0, sizeof(msg));
821 vec.iov_base = &dclc;
822 vec.iov_len = send_len;
823 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, send_len);
824 if (len < 0 || len < send_len)
825 len = -EPROTO;
826 return len > 0 ? 0 : len;
827}
828
829/* send CLC PROPOSAL message across internal TCP socket */
830int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini)
831{
832 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
833 struct smc_clc_msg_proposal_prefix *pclc_prfx;
834 struct smc_clc_msg_proposal *pclc_base;
835 struct smc_clc_smcd_gid_chid *gidchids;
836 struct smc_clc_msg_proposal_area *pclc;
837 struct smc_clc_ipv6_prefix *ipv6_prfx;
838 struct net *net = sock_net(&smc->sk);
839 struct smc_clc_v2_extension *v2_ext;
840 struct smc_clc_msg_smcd *pclc_smcd;
841 struct smc_clc_msg_trail *trl;
842 struct smcd_dev *smcd;
843 int len, i, plen, rc;
844 int reason_code = 0;
845 struct kvec vec[8];
846 struct msghdr msg;
847
848 pclc = kzalloc(sizeof(*pclc), GFP_KERNEL);
849 if (!pclc)
850 return -ENOMEM;
851
852 pclc_base = &pclc->pclc_base;
853 pclc_smcd = &pclc->pclc_smcd;
854 pclc_prfx = &pclc->pclc_prfx;
855 ipv6_prfx = pclc->pclc_prfx_ipv6;
856 v2_ext = &pclc->pclc_v2_ext;
857 smcd_v2_ext = &pclc->pclc_smcd_v2_ext;
858 gidchids = pclc->pclc_gidchids;
859 trl = &pclc->pclc_trl;
860
861 pclc_base->hdr.version = SMC_V2;
862 pclc_base->hdr.typev1 = ini->smc_type_v1;
863 pclc_base->hdr.typev2 = ini->smc_type_v2;
864 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
865
866 /* retrieve ip prefixes for CLC proposal msg */
867 if (ini->smc_type_v1 != SMC_TYPE_N) {
868 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
869 if (rc) {
870 if (ini->smc_type_v2 == SMC_TYPE_N) {
871 kfree(pclc);
872 return SMC_CLC_DECL_CNFERR;
873 }
874 pclc_base->hdr.typev1 = SMC_TYPE_N;
875 } else {
876 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
877 plen += sizeof(*pclc_prfx) +
878 pclc_prfx->ipv6_prefixes_cnt *
879 sizeof(ipv6_prfx[0]);
880 }
881 }
882
883 /* build SMC Proposal CLC message */
884 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
885 sizeof(SMC_EYECATCHER));
886 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
887 if (smcr_indicated(ini->smc_type_v1)) {
888 /* add SMC-R specifics */
889 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
890 sizeof(local_systemid));
891 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
892 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
893 ETH_ALEN);
894 }
895 if (smcd_indicated(ini->smc_type_v1)) {
896 struct smcd_gid smcd_gid;
897
898 /* add SMC-D specifics */
899 if (ini->ism_dev[0]) {
900 smcd = ini->ism_dev[0];
901 smcd->ops->get_local_gid(smcd, &smcd_gid);
902 pclc_smcd->ism.gid = htonll(smcd_gid.gid);
903 pclc_smcd->ism.chid =
904 htons(smc_ism_get_chid(ini->ism_dev[0]));
905 }
906 }
907 if (ini->smc_type_v2 == SMC_TYPE_N) {
908 pclc_smcd->v2_ext_offset = 0;
909 } else {
910 struct smc_clc_eid_entry *ueident;
911 u16 v2_ext_offset;
912
913 v2_ext->hdr.flag.release = SMC_RELEASE;
914 v2_ext_offset = sizeof(*pclc_smcd) -
915 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
916 if (ini->smc_type_v1 != SMC_TYPE_N)
917 v2_ext_offset += sizeof(*pclc_prfx) +
918 pclc_prfx->ipv6_prefixes_cnt *
919 sizeof(ipv6_prfx[0]);
920 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
921 plen += sizeof(*v2_ext);
922
923 v2_ext->feature_mask = htons(SMC_FEATURE_MASK);
924 read_lock(&smc_clc_eid_table.lock);
925 v2_ext->hdr.eid_cnt = smc_clc_eid_table.ueid_cnt;
926 plen += smc_clc_eid_table.ueid_cnt * SMC_MAX_EID_LEN;
927 i = 0;
928 list_for_each_entry(ueident, &smc_clc_eid_table.list, list) {
929 memcpy(v2_ext->user_eids[i++], ueident->eid,
930 sizeof(ueident->eid));
931 }
932 read_unlock(&smc_clc_eid_table.lock);
933 }
934 if (smcd_indicated(ini->smc_type_v2)) {
935 struct smcd_gid smcd_gid;
936 u8 *eid = NULL;
937 int entry = 0;
938
939 v2_ext->hdr.flag.seid = smc_clc_eid_table.seid_enabled;
940 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
941 offsetofend(struct smc_clnt_opts_area_hdr,
942 smcd_v2_ext_offset) +
943 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
944 smc_ism_get_system_eid(&eid);
945 if (eid && v2_ext->hdr.flag.seid)
946 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
947 plen += sizeof(*smcd_v2_ext);
948 if (ini->ism_offered_cnt) {
949 for (i = 1; i <= ini->ism_offered_cnt; i++) {
950 smcd = ini->ism_dev[i];
951 smcd->ops->get_local_gid(smcd, &smcd_gid);
952 gidchids[entry].chid =
953 htons(smc_ism_get_chid(ini->ism_dev[i]));
954 gidchids[entry].gid = htonll(smcd_gid.gid);
955 if (smc_ism_is_virtual(smcd)) {
956 /* a virtual ISM device takes two
957 * entries. CHID of the second entry
958 * repeats that of the first entry.
959 */
960 gidchids[entry + 1].chid =
961 gidchids[entry].chid;
962 gidchids[entry + 1].gid =
963 htonll(smcd_gid.gid_ext);
964 entry++;
965 }
966 entry++;
967 }
968 plen += entry * sizeof(struct smc_clc_smcd_gid_chid);
969 }
970 v2_ext->hdr.ism_gid_cnt = entry;
971 }
972 if (smcr_indicated(ini->smc_type_v2)) {
973 memcpy(v2_ext->roce, ini->smcrv2.ib_gid_v2, SMC_GID_SIZE);
974 v2_ext->max_conns = net->smc.sysctl_max_conns_per_lgr;
975 v2_ext->max_links = net->smc.sysctl_max_links_per_lgr;
976 }
977
978 pclc_base->hdr.length = htons(plen);
979 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
980
981 /* send SMC Proposal CLC message */
982 memset(&msg, 0, sizeof(msg));
983 i = 0;
984 vec[i].iov_base = pclc_base;
985 vec[i++].iov_len = sizeof(*pclc_base);
986 vec[i].iov_base = pclc_smcd;
987 vec[i++].iov_len = sizeof(*pclc_smcd);
988 if (ini->smc_type_v1 != SMC_TYPE_N) {
989 vec[i].iov_base = pclc_prfx;
990 vec[i++].iov_len = sizeof(*pclc_prfx);
991 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
992 vec[i].iov_base = ipv6_prfx;
993 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
994 sizeof(ipv6_prfx[0]);
995 }
996 }
997 if (ini->smc_type_v2 != SMC_TYPE_N) {
998 vec[i].iov_base = v2_ext;
999 vec[i++].iov_len = sizeof(*v2_ext) +
1000 (v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
1001 if (smcd_indicated(ini->smc_type_v2)) {
1002 vec[i].iov_base = smcd_v2_ext;
1003 vec[i++].iov_len = sizeof(*smcd_v2_ext);
1004 if (ini->ism_offered_cnt) {
1005 vec[i].iov_base = gidchids;
1006 vec[i++].iov_len = v2_ext->hdr.ism_gid_cnt *
1007 sizeof(struct smc_clc_smcd_gid_chid);
1008 }
1009 }
1010 }
1011 vec[i].iov_base = trl;
1012 vec[i++].iov_len = sizeof(*trl);
1013 /* due to the few bytes needed for clc-handshake this cannot block */
1014 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
1015 if (len < 0) {
1016 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1017 reason_code = -smc->sk.sk_err;
1018 } else if (len < ntohs(pclc_base->hdr.length)) {
1019 reason_code = -ENETUNREACH;
1020 smc->sk.sk_err = -reason_code;
1021 }
1022
1023 kfree(pclc);
1024 return reason_code;
1025}
1026
1027static void
1028smcd_clc_prep_confirm_accept(struct smc_connection *conn,
1029 struct smc_clc_msg_accept_confirm *clc,
1030 int first_contact, u8 version,
1031 u8 *eid, struct smc_init_info *ini,
1032 int *fce_len,
1033 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1034 struct smc_clc_msg_trail *trl)
1035{
1036 struct smcd_dev *smcd = conn->lgr->smcd;
1037 struct smcd_gid smcd_gid;
1038 u16 chid;
1039 int len;
1040
1041 /* SMC-D specific settings */
1042 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
1043 sizeof(SMCD_EYECATCHER));
1044 smcd->ops->get_local_gid(smcd, &smcd_gid);
1045 clc->hdr.typev1 = SMC_TYPE_D;
1046 clc->d0.gid = htonll(smcd_gid.gid);
1047 clc->d0.token = htonll(conn->rmb_desc->token);
1048 clc->d0.dmbe_size = conn->rmbe_size_comp;
1049 clc->d0.dmbe_idx = 0;
1050 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
1051 if (version == SMC_V1) {
1052 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
1053 } else {
1054 chid = smc_ism_get_chid(smcd);
1055 clc->d1.chid = htons(chid);
1056 if (eid && eid[0])
1057 memcpy(clc->d1.eid, eid, SMC_MAX_EID_LEN);
1058 if (__smc_ism_is_virtual(chid))
1059 clc->d1.gid_ext = htonll(smcd_gid.gid_ext);
1060 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
1061 if (first_contact) {
1062 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1063 len += *fce_len;
1064 }
1065 clc->hdr.length = htons(len);
1066 }
1067 memcpy(trl->eyecatcher, SMCD_EYECATCHER,
1068 sizeof(SMCD_EYECATCHER));
1069}
1070
1071static void
1072smcr_clc_prep_confirm_accept(struct smc_connection *conn,
1073 struct smc_clc_msg_accept_confirm *clc,
1074 int first_contact, u8 version,
1075 u8 *eid, struct smc_init_info *ini,
1076 int *fce_len,
1077 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1078 struct smc_clc_fce_gid_ext *gle,
1079 struct smc_clc_msg_trail *trl)
1080{
1081 struct smc_link *link = conn->lnk;
1082 int len;
1083
1084 /* SMC-R specific settings */
1085 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
1086 sizeof(SMC_EYECATCHER));
1087 clc->hdr.typev1 = SMC_TYPE_R;
1088 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
1089 sizeof(local_systemid));
1090 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
1091 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
1092 ETH_ALEN);
1093 hton24(clc->r0.qpn, link->roce_qp->qp_num);
1094 clc->r0.rmb_rkey =
1095 htonl(conn->rmb_desc->mr[link->link_idx]->rkey);
1096 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
1097 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
1098 switch (clc->hdr.type) {
1099 case SMC_CLC_ACCEPT:
1100 clc->r0.qp_mtu = link->path_mtu;
1101 break;
1102 case SMC_CLC_CONFIRM:
1103 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
1104 break;
1105 }
1106 clc->r0.rmbe_size = conn->rmbe_size_comp;
1107 clc->r0.rmb_dma_addr = conn->rmb_desc->is_vm ?
1108 cpu_to_be64((uintptr_t)conn->rmb_desc->cpu_addr) :
1109 cpu_to_be64((u64)sg_dma_address
1110 (conn->rmb_desc->sgt[link->link_idx].sgl));
1111 hton24(clc->r0.psn, link->psn_initial);
1112 if (version == SMC_V1) {
1113 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1114 } else {
1115 if (eid && eid[0])
1116 memcpy(clc->r1.eid, eid, SMC_MAX_EID_LEN);
1117 len = SMCR_CLC_ACCEPT_CONFIRM_LEN_V2;
1118 if (first_contact) {
1119 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1120 len += *fce_len;
1121 fce_v2x->fce_v2_base.v2_direct =
1122 !link->lgr->uses_gateway;
1123 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1124 memset(gle, 0, sizeof(*gle));
1125 gle->gid_cnt = ini->smcrv2.gidlist.len;
1126 len += sizeof(*gle);
1127 len += gle->gid_cnt * sizeof(gle->gid[0]);
1128 }
1129 }
1130 clc->hdr.length = htons(len);
1131 }
1132 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
1133}
1134
1135/* build and send CLC CONFIRM / ACCEPT message */
1136static int smc_clc_send_confirm_accept(struct smc_sock *smc,
1137 struct smc_clc_msg_accept_confirm *clc,
1138 int first_contact, u8 version,
1139 u8 *eid, struct smc_init_info *ini)
1140{
1141 struct smc_clc_first_contact_ext_v2x fce_v2x;
1142 struct smc_connection *conn = &smc->conn;
1143 struct smc_clc_fce_gid_ext gle;
1144 struct smc_clc_msg_trail trl;
1145 int i, fce_len;
1146 struct kvec vec[5];
1147 struct msghdr msg;
1148
1149 /* send SMC Confirm CLC msg */
1150 clc->hdr.version = version; /* SMC version */
1151 if (first_contact)
1152 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
1153 if (conn->lgr->is_smcd)
1154 smcd_clc_prep_confirm_accept(conn, clc, first_contact,
1155 version, eid, ini, &fce_len,
1156 &fce_v2x, &trl);
1157 else
1158 smcr_clc_prep_confirm_accept(conn, clc, first_contact,
1159 version, eid, ini, &fce_len,
1160 &fce_v2x, &gle, &trl);
1161 memset(&msg, 0, sizeof(msg));
1162 i = 0;
1163 vec[i].iov_base = clc;
1164 if (version > SMC_V1)
1165 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1166 SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 :
1167 SMCR_CLC_ACCEPT_CONFIRM_LEN_V2) -
1168 sizeof(trl);
1169 else
1170 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1171 SMCD_CLC_ACCEPT_CONFIRM_LEN :
1172 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
1173 sizeof(trl);
1174 if (version > SMC_V1 && first_contact) {
1175 vec[i].iov_base = &fce_v2x;
1176 vec[i++].iov_len = fce_len;
1177 if (!conn->lgr->is_smcd) {
1178 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1179 vec[i].iov_base = &gle;
1180 vec[i++].iov_len = sizeof(gle);
1181 vec[i].iov_base = &ini->smcrv2.gidlist.list;
1182 vec[i++].iov_len = gle.gid_cnt *
1183 sizeof(gle.gid[0]);
1184 }
1185 }
1186 }
1187 vec[i].iov_base = &trl;
1188 vec[i++].iov_len = sizeof(trl);
1189 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
1190 ntohs(clc->hdr.length));
1191}
1192
1193/* send CLC CONFIRM message across internal TCP socket */
1194int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
1195 u8 version, u8 *eid, struct smc_init_info *ini)
1196{
1197 struct smc_clc_msg_accept_confirm cclc;
1198 int reason_code = 0;
1199 int len;
1200
1201 /* send SMC Confirm CLC msg */
1202 memset(&cclc, 0, sizeof(cclc));
1203 cclc.hdr.type = SMC_CLC_CONFIRM;
1204 len = smc_clc_send_confirm_accept(smc, &cclc, clnt_first_contact,
1205 version, eid, ini);
1206 if (len < ntohs(cclc.hdr.length)) {
1207 if (len >= 0) {
1208 reason_code = -ENETUNREACH;
1209 smc->sk.sk_err = -reason_code;
1210 } else {
1211 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1212 reason_code = -smc->sk.sk_err;
1213 }
1214 }
1215 return reason_code;
1216}
1217
1218/* send CLC ACCEPT message across internal TCP socket */
1219int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
1220 u8 version, u8 *negotiated_eid, struct smc_init_info *ini)
1221{
1222 struct smc_clc_msg_accept_confirm aclc;
1223 int len;
1224
1225 memset(&aclc, 0, sizeof(aclc));
1226 aclc.hdr.type = SMC_CLC_ACCEPT;
1227 len = smc_clc_send_confirm_accept(new_smc, &aclc, srv_first_contact,
1228 version, negotiated_eid, ini);
1229 if (len < ntohs(aclc.hdr.length))
1230 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
1231
1232 return len > 0 ? 0 : len;
1233}
1234
1235int smc_clc_srv_v2x_features_validate(struct smc_sock *smc,
1236 struct smc_clc_msg_proposal *pclc,
1237 struct smc_init_info *ini)
1238{
1239 struct smc_clc_v2_extension *pclc_v2_ext;
1240 struct net *net = sock_net(&smc->sk);
1241
1242 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1243 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1244 ini->feature_mask = SMC_FEATURE_MASK;
1245
1246 if ((!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) ||
1247 ini->release_nr < SMC_RELEASE_1)
1248 return 0;
1249
1250 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1251 if (!pclc_v2_ext)
1252 return SMC_CLC_DECL_NOV2EXT;
1253
1254 if (ini->smcr_version & SMC_V2) {
1255 ini->max_conns = min_t(u8, pclc_v2_ext->max_conns,
1256 net->smc.sysctl_max_conns_per_lgr);
1257 if (ini->max_conns < SMC_CONN_PER_LGR_MIN)
1258 return SMC_CLC_DECL_MAXCONNERR;
1259
1260 ini->max_links = min_t(u8, pclc_v2_ext->max_links,
1261 net->smc.sysctl_max_links_per_lgr);
1262 if (ini->max_links < SMC_LINKS_ADD_LNK_MIN)
1263 return SMC_CLC_DECL_MAXLINKERR;
1264 }
1265
1266 return 0;
1267}
1268
1269int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
1270 struct smc_init_info *ini)
1271{
1272 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1273 (struct smc_clc_first_contact_ext_v2x *)fce;
1274
1275 if (ini->release_nr < SMC_RELEASE_1)
1276 return 0;
1277
1278 if (!ini->is_smcd) {
1279 if (fce_v2x->max_conns < SMC_CONN_PER_LGR_MIN)
1280 return SMC_CLC_DECL_MAXCONNERR;
1281 ini->max_conns = fce_v2x->max_conns;
1282
1283 if (fce_v2x->max_links > SMC_LINKS_ADD_LNK_MAX ||
1284 fce_v2x->max_links < SMC_LINKS_ADD_LNK_MIN)
1285 return SMC_CLC_DECL_MAXLINKERR;
1286 ini->max_links = fce_v2x->max_links;
1287 }
1288 /* common supplemental features of server and client */
1289 ini->feature_mask = ntohs(fce_v2x->feature_mask) & SMC_FEATURE_MASK;
1290
1291 return 0;
1292}
1293
1294int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
1295 struct smc_init_info *ini)
1296{
1297 struct smc_clc_first_contact_ext *fce =
1298 smc_get_clc_first_contact_ext(cclc, ini->is_smcd);
1299 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1300 (struct smc_clc_first_contact_ext_v2x *)fce;
1301
1302 if (cclc->hdr.version == SMC_V1 ||
1303 !(cclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
1304 return 0;
1305
1306 if (ini->release_nr != fce->release)
1307 return SMC_CLC_DECL_RELEASEERR;
1308
1309 if (fce->release < SMC_RELEASE_1)
1310 return 0;
1311
1312 if (!ini->is_smcd) {
1313 if (fce_v2x->max_conns != ini->max_conns)
1314 return SMC_CLC_DECL_MAXCONNERR;
1315 if (fce_v2x->max_links != ini->max_links)
1316 return SMC_CLC_DECL_MAXLINKERR;
1317 }
1318 /* common supplemental features returned by client */
1319 ini->feature_mask = ntohs(fce_v2x->feature_mask);
1320
1321 return 0;
1322}
1323
1324void smc_clc_get_hostname(u8 **host)
1325{
1326 *host = &smc_hostname[0];
1327}
1328
1329void __init smc_clc_init(void)
1330{
1331 struct new_utsname *u;
1332
1333 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
1334 u = utsname();
1335 memcpy(smc_hostname, u->nodename,
1336 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
1337
1338 INIT_LIST_HEAD(&smc_clc_eid_table.list);
1339 rwlock_init(&smc_clc_eid_table.lock);
1340 smc_clc_eid_table.ueid_cnt = 0;
1341#if IS_ENABLED(CONFIG_S390)
1342 smc_clc_eid_table.seid_enabled = 1;
1343#else
1344 smc_clc_eid_table.seid_enabled = 0;
1345#endif
1346}
1347
1348void smc_clc_exit(void)
1349{
1350 smc_clc_ueid_remove(NULL);
1351}
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
30#define SMCR_CLC_ACCEPT_CONFIRM_LEN 68
31#define SMCD_CLC_ACCEPT_CONFIRM_LEN 48
32#define SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 78
33#define SMC_CLC_RECV_BUF_LEN 100
34
35/* eye catcher "SMCR" EBCDIC for CLC messages */
36static const char SMC_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xd9'};
37/* eye catcher "SMCD" EBCDIC for CLC messages */
38static const char SMCD_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xc4'};
39
40static u8 smc_hostname[SMC_MAX_HOSTNAME_LEN];
41
42/* check arriving CLC proposal */
43static bool smc_clc_msg_prop_valid(struct smc_clc_msg_proposal *pclc)
44{
45 struct smc_clc_msg_proposal_prefix *pclc_prfx;
46 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
47 struct smc_clc_msg_hdr *hdr = &pclc->hdr;
48 struct smc_clc_v2_extension *v2_ext;
49
50 v2_ext = smc_get_clc_v2_ext(pclc);
51 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
52 if (hdr->version == SMC_V1) {
53 if (hdr->typev1 == SMC_TYPE_N)
54 return false;
55 if (ntohs(hdr->length) !=
56 sizeof(*pclc) + ntohs(pclc->iparea_offset) +
57 sizeof(*pclc_prfx) +
58 pclc_prfx->ipv6_prefixes_cnt *
59 sizeof(struct smc_clc_ipv6_prefix) +
60 sizeof(struct smc_clc_msg_trail))
61 return false;
62 } else {
63 if (ntohs(hdr->length) !=
64 sizeof(*pclc) +
65 sizeof(struct smc_clc_msg_smcd) +
66 (hdr->typev1 != SMC_TYPE_N ?
67 sizeof(*pclc_prfx) +
68 pclc_prfx->ipv6_prefixes_cnt *
69 sizeof(struct smc_clc_ipv6_prefix) : 0) +
70 (hdr->typev2 != SMC_TYPE_N ?
71 sizeof(*v2_ext) +
72 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN : 0) +
73 (smcd_indicated(hdr->typev2) ?
74 sizeof(*smcd_v2_ext) + v2_ext->hdr.ism_gid_cnt *
75 sizeof(struct smc_clc_smcd_gid_chid) :
76 0) +
77 sizeof(struct smc_clc_msg_trail))
78 return false;
79 }
80 return true;
81}
82
83/* check arriving CLC accept or confirm */
84static bool
85smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm_v2 *clc_v2)
86{
87 struct smc_clc_msg_hdr *hdr = &clc_v2->hdr;
88
89 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
90 return false;
91 if (hdr->version == SMC_V1) {
92 if ((hdr->typev1 == SMC_TYPE_R &&
93 ntohs(hdr->length) != SMCR_CLC_ACCEPT_CONFIRM_LEN) ||
94 (hdr->typev1 == SMC_TYPE_D &&
95 ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN))
96 return false;
97 } else {
98 if (hdr->typev1 == SMC_TYPE_D &&
99 ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 &&
100 (ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 +
101 sizeof(struct smc_clc_first_contact_ext)))
102 return false;
103 }
104 return true;
105}
106
107static void smc_clc_fill_fce(struct smc_clc_first_contact_ext *fce, int *len)
108{
109 memset(fce, 0, sizeof(*fce));
110 fce->os_type = SMC_CLC_OS_LINUX;
111 fce->release = SMC_RELEASE;
112 memcpy(fce->hostname, smc_hostname, sizeof(smc_hostname));
113 (*len) += sizeof(*fce);
114}
115
116/* check if received message has a correct header length and contains valid
117 * heading and trailing eyecatchers
118 */
119static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm, bool check_trl)
120{
121 struct smc_clc_msg_accept_confirm_v2 *clc_v2;
122 struct smc_clc_msg_proposal *pclc;
123 struct smc_clc_msg_decline *dclc;
124 struct smc_clc_msg_trail *trl;
125
126 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
127 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
128 return false;
129 switch (clcm->type) {
130 case SMC_CLC_PROPOSAL:
131 pclc = (struct smc_clc_msg_proposal *)clcm;
132 if (!smc_clc_msg_prop_valid(pclc))
133 return false;
134 trl = (struct smc_clc_msg_trail *)
135 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
136 break;
137 case SMC_CLC_ACCEPT:
138 case SMC_CLC_CONFIRM:
139 clc_v2 = (struct smc_clc_msg_accept_confirm_v2 *)clcm;
140 if (!smc_clc_msg_acc_conf_valid(clc_v2))
141 return false;
142 trl = (struct smc_clc_msg_trail *)
143 ((u8 *)clc_v2 + ntohs(clc_v2->hdr.length) -
144 sizeof(*trl));
145 break;
146 case SMC_CLC_DECLINE:
147 dclc = (struct smc_clc_msg_decline *)clcm;
148 if (ntohs(dclc->hdr.length) != sizeof(*dclc))
149 return false;
150 trl = &dclc->trl;
151 break;
152 default:
153 return false;
154 }
155 if (check_trl &&
156 memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
157 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
158 return false;
159 return true;
160}
161
162/* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
163static int smc_clc_prfx_set4_rcu(struct dst_entry *dst, __be32 ipv4,
164 struct smc_clc_msg_proposal_prefix *prop)
165{
166 struct in_device *in_dev = __in_dev_get_rcu(dst->dev);
167 const struct in_ifaddr *ifa;
168
169 if (!in_dev)
170 return -ENODEV;
171
172 in_dev_for_each_ifa_rcu(ifa, in_dev) {
173 if (!inet_ifa_match(ipv4, ifa))
174 continue;
175 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
176 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
177 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
178 return 0;
179 }
180 return -ENOENT;
181}
182
183/* fill CLC proposal msg with ipv6 prefixes from device */
184static int smc_clc_prfx_set6_rcu(struct dst_entry *dst,
185 struct smc_clc_msg_proposal_prefix *prop,
186 struct smc_clc_ipv6_prefix *ipv6_prfx)
187{
188#if IS_ENABLED(CONFIG_IPV6)
189 struct inet6_dev *in6_dev = __in6_dev_get(dst->dev);
190 struct inet6_ifaddr *ifa;
191 int cnt = 0;
192
193 if (!in6_dev)
194 return -ENODEV;
195 /* use a maximum of 8 IPv6 prefixes from device */
196 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
197 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
198 continue;
199 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
200 &ifa->addr, ifa->prefix_len);
201 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
202 cnt++;
203 if (cnt == SMC_CLC_MAX_V6_PREFIX)
204 break;
205 }
206 prop->ipv6_prefixes_cnt = cnt;
207 if (cnt)
208 return 0;
209#endif
210 return -ENOENT;
211}
212
213/* retrieve and set prefixes in CLC proposal msg */
214static int smc_clc_prfx_set(struct socket *clcsock,
215 struct smc_clc_msg_proposal_prefix *prop,
216 struct smc_clc_ipv6_prefix *ipv6_prfx)
217{
218 struct dst_entry *dst = sk_dst_get(clcsock->sk);
219 struct sockaddr_storage addrs;
220 struct sockaddr_in6 *addr6;
221 struct sockaddr_in *addr;
222 int rc = -ENOENT;
223
224 if (!dst) {
225 rc = -ENOTCONN;
226 goto out;
227 }
228 if (!dst->dev) {
229 rc = -ENODEV;
230 goto out_rel;
231 }
232 /* get address to which the internal TCP socket is bound */
233 if (kernel_getsockname(clcsock, (struct sockaddr *)&addrs) < 0)
234 goto out_rel;
235 /* analyze IP specific data of net_device belonging to TCP socket */
236 addr6 = (struct sockaddr_in6 *)&addrs;
237 rcu_read_lock();
238 if (addrs.ss_family == PF_INET) {
239 /* IPv4 */
240 addr = (struct sockaddr_in *)&addrs;
241 rc = smc_clc_prfx_set4_rcu(dst, addr->sin_addr.s_addr, prop);
242 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
243 /* mapped IPv4 address - peer is IPv4 only */
244 rc = smc_clc_prfx_set4_rcu(dst, addr6->sin6_addr.s6_addr32[3],
245 prop);
246 } else {
247 /* IPv6 */
248 rc = smc_clc_prfx_set6_rcu(dst, prop, ipv6_prfx);
249 }
250 rcu_read_unlock();
251out_rel:
252 dst_release(dst);
253out:
254 return rc;
255}
256
257/* match ipv4 addrs of dev against addr in CLC proposal */
258static int smc_clc_prfx_match4_rcu(struct net_device *dev,
259 struct smc_clc_msg_proposal_prefix *prop)
260{
261 struct in_device *in_dev = __in_dev_get_rcu(dev);
262 const struct in_ifaddr *ifa;
263
264 if (!in_dev)
265 return -ENODEV;
266 in_dev_for_each_ifa_rcu(ifa, in_dev) {
267 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
268 inet_ifa_match(prop->outgoing_subnet, ifa))
269 return 0;
270 }
271
272 return -ENOENT;
273}
274
275/* match ipv6 addrs of dev against addrs in CLC proposal */
276static int smc_clc_prfx_match6_rcu(struct net_device *dev,
277 struct smc_clc_msg_proposal_prefix *prop)
278{
279#if IS_ENABLED(CONFIG_IPV6)
280 struct inet6_dev *in6_dev = __in6_dev_get(dev);
281 struct smc_clc_ipv6_prefix *ipv6_prfx;
282 struct inet6_ifaddr *ifa;
283 int i, max;
284
285 if (!in6_dev)
286 return -ENODEV;
287 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
288 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
289 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
290 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
291 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
292 continue;
293 for (i = 0; i < max; i++) {
294 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
295 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
296 ifa->prefix_len))
297 return 0;
298 }
299 }
300#endif
301 return -ENOENT;
302}
303
304/* check if proposed prefixes match one of our device prefixes */
305int smc_clc_prfx_match(struct socket *clcsock,
306 struct smc_clc_msg_proposal_prefix *prop)
307{
308 struct dst_entry *dst = sk_dst_get(clcsock->sk);
309 int rc;
310
311 if (!dst) {
312 rc = -ENOTCONN;
313 goto out;
314 }
315 if (!dst->dev) {
316 rc = -ENODEV;
317 goto out_rel;
318 }
319 rcu_read_lock();
320 if (!prop->ipv6_prefixes_cnt)
321 rc = smc_clc_prfx_match4_rcu(dst->dev, prop);
322 else
323 rc = smc_clc_prfx_match6_rcu(dst->dev, prop);
324 rcu_read_unlock();
325out_rel:
326 dst_release(dst);
327out:
328 return rc;
329}
330
331/* Wait for data on the tcp-socket, analyze received data
332 * Returns:
333 * 0 if success and it was not a decline that we received.
334 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
335 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
336 */
337int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
338 u8 expected_type, unsigned long timeout)
339{
340 long rcvtimeo = smc->clcsock->sk->sk_rcvtimeo;
341 struct sock *clc_sk = smc->clcsock->sk;
342 struct smc_clc_msg_hdr *clcm = buf;
343 struct msghdr msg = {NULL, 0};
344 int reason_code = 0;
345 struct kvec vec = {buf, buflen};
346 int len, datlen, recvlen;
347 bool check_trl = true;
348 int krflags;
349
350 /* peek the first few bytes to determine length of data to receive
351 * so we don't consume any subsequent CLC message or payload data
352 * in the TCP byte stream
353 */
354 /*
355 * Caller must make sure that buflen is no less than
356 * sizeof(struct smc_clc_msg_hdr)
357 */
358 krflags = MSG_PEEK | MSG_WAITALL;
359 clc_sk->sk_rcvtimeo = timeout;
360 iov_iter_kvec(&msg.msg_iter, READ, &vec, 1,
361 sizeof(struct smc_clc_msg_hdr));
362 len = sock_recvmsg(smc->clcsock, &msg, krflags);
363 if (signal_pending(current)) {
364 reason_code = -EINTR;
365 clc_sk->sk_err = EINTR;
366 smc->sk.sk_err = EINTR;
367 goto out;
368 }
369 if (clc_sk->sk_err) {
370 reason_code = -clc_sk->sk_err;
371 if (clc_sk->sk_err == EAGAIN &&
372 expected_type == SMC_CLC_DECLINE)
373 clc_sk->sk_err = 0; /* reset for fallback usage */
374 else
375 smc->sk.sk_err = clc_sk->sk_err;
376 goto out;
377 }
378 if (!len) { /* peer has performed orderly shutdown */
379 smc->sk.sk_err = ECONNRESET;
380 reason_code = -ECONNRESET;
381 goto out;
382 }
383 if (len < 0) {
384 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
385 smc->sk.sk_err = -len;
386 reason_code = len;
387 goto out;
388 }
389 datlen = ntohs(clcm->length);
390 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
391 (clcm->version < SMC_V1) ||
392 ((clcm->type != SMC_CLC_DECLINE) &&
393 (clcm->type != expected_type))) {
394 smc->sk.sk_err = EPROTO;
395 reason_code = -EPROTO;
396 goto out;
397 }
398
399 /* receive the complete CLC message */
400 memset(&msg, 0, sizeof(struct msghdr));
401 if (datlen > buflen) {
402 check_trl = false;
403 recvlen = buflen;
404 } else {
405 recvlen = datlen;
406 }
407 iov_iter_kvec(&msg.msg_iter, READ, &vec, 1, recvlen);
408 krflags = MSG_WAITALL;
409 len = sock_recvmsg(smc->clcsock, &msg, krflags);
410 if (len < recvlen || !smc_clc_msg_hdr_valid(clcm, check_trl)) {
411 smc->sk.sk_err = EPROTO;
412 reason_code = -EPROTO;
413 goto out;
414 }
415 datlen -= len;
416 while (datlen) {
417 u8 tmp[SMC_CLC_RECV_BUF_LEN];
418
419 vec.iov_base = &tmp;
420 vec.iov_len = SMC_CLC_RECV_BUF_LEN;
421 /* receive remaining proposal message */
422 recvlen = datlen > SMC_CLC_RECV_BUF_LEN ?
423 SMC_CLC_RECV_BUF_LEN : datlen;
424 iov_iter_kvec(&msg.msg_iter, READ, &vec, 1, recvlen);
425 len = sock_recvmsg(smc->clcsock, &msg, krflags);
426 datlen -= len;
427 }
428 if (clcm->type == SMC_CLC_DECLINE) {
429 struct smc_clc_msg_decline *dclc;
430
431 dclc = (struct smc_clc_msg_decline *)clcm;
432 reason_code = SMC_CLC_DECL_PEERDECL;
433 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
434 if (((struct smc_clc_msg_decline *)buf)->hdr.typev2 &
435 SMC_FIRST_CONTACT_MASK) {
436 smc->conn.lgr->sync_err = 1;
437 smc_lgr_terminate_sched(smc->conn.lgr);
438 }
439 }
440
441out:
442 clc_sk->sk_rcvtimeo = rcvtimeo;
443 return reason_code;
444}
445
446/* send CLC DECLINE message across internal TCP socket */
447int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version)
448{
449 struct smc_clc_msg_decline dclc;
450 struct msghdr msg;
451 struct kvec vec;
452 int len;
453
454 memset(&dclc, 0, sizeof(dclc));
455 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
456 dclc.hdr.type = SMC_CLC_DECLINE;
457 dclc.hdr.length = htons(sizeof(struct smc_clc_msg_decline));
458 dclc.hdr.version = version;
459 dclc.os_type = version == SMC_V1 ? 0 : SMC_CLC_OS_LINUX;
460 dclc.hdr.typev2 = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ?
461 SMC_FIRST_CONTACT_MASK : 0;
462 if ((!smc->conn.lgr || !smc->conn.lgr->is_smcd) &&
463 smc_ib_is_valid_local_systemid())
464 memcpy(dclc.id_for_peer, local_systemid,
465 sizeof(local_systemid));
466 dclc.peer_diagnosis = htonl(peer_diag_info);
467 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
468
469 memset(&msg, 0, sizeof(msg));
470 vec.iov_base = &dclc;
471 vec.iov_len = sizeof(struct smc_clc_msg_decline);
472 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1,
473 sizeof(struct smc_clc_msg_decline));
474 if (len < 0 || len < sizeof(struct smc_clc_msg_decline))
475 len = -EPROTO;
476 return len > 0 ? 0 : len;
477}
478
479/* send CLC PROPOSAL message across internal TCP socket */
480int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini)
481{
482 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
483 struct smc_clc_msg_proposal_prefix *pclc_prfx;
484 struct smc_clc_msg_proposal *pclc_base;
485 struct smc_clc_smcd_gid_chid *gidchids;
486 struct smc_clc_msg_proposal_area *pclc;
487 struct smc_clc_ipv6_prefix *ipv6_prfx;
488 struct smc_clc_v2_extension *v2_ext;
489 struct smc_clc_msg_smcd *pclc_smcd;
490 struct smc_clc_msg_trail *trl;
491 int len, i, plen, rc;
492 int reason_code = 0;
493 struct kvec vec[8];
494 struct msghdr msg;
495
496 pclc = kzalloc(sizeof(*pclc), GFP_KERNEL);
497 if (!pclc)
498 return -ENOMEM;
499
500 pclc_base = &pclc->pclc_base;
501 pclc_smcd = &pclc->pclc_smcd;
502 pclc_prfx = &pclc->pclc_prfx;
503 ipv6_prfx = pclc->pclc_prfx_ipv6;
504 v2_ext = &pclc->pclc_v2_ext;
505 smcd_v2_ext = &pclc->pclc_smcd_v2_ext;
506 gidchids = pclc->pclc_gidchids;
507 trl = &pclc->pclc_trl;
508
509 pclc_base->hdr.version = SMC_V2;
510 pclc_base->hdr.typev1 = ini->smc_type_v1;
511 pclc_base->hdr.typev2 = ini->smc_type_v2;
512 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
513
514 /* retrieve ip prefixes for CLC proposal msg */
515 if (ini->smc_type_v1 != SMC_TYPE_N) {
516 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
517 if (rc) {
518 if (ini->smc_type_v2 == SMC_TYPE_N) {
519 kfree(pclc);
520 return SMC_CLC_DECL_CNFERR;
521 }
522 pclc_base->hdr.typev1 = SMC_TYPE_N;
523 } else {
524 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
525 plen += sizeof(*pclc_prfx) +
526 pclc_prfx->ipv6_prefixes_cnt *
527 sizeof(ipv6_prfx[0]);
528 }
529 }
530
531 /* build SMC Proposal CLC message */
532 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
533 sizeof(SMC_EYECATCHER));
534 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
535 if (smcr_indicated(ini->smc_type_v1)) {
536 /* add SMC-R specifics */
537 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
538 sizeof(local_systemid));
539 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
540 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
541 ETH_ALEN);
542 }
543 if (smcd_indicated(ini->smc_type_v1)) {
544 /* add SMC-D specifics */
545 if (ini->ism_dev[0]) {
546 pclc_smcd->ism.gid = htonll(ini->ism_dev[0]->local_gid);
547 pclc_smcd->ism.chid =
548 htons(smc_ism_get_chid(ini->ism_dev[0]));
549 }
550 }
551 if (ini->smc_type_v2 == SMC_TYPE_N) {
552 pclc_smcd->v2_ext_offset = 0;
553 } else {
554 u16 v2_ext_offset;
555 u8 *eid = NULL;
556
557 v2_ext_offset = sizeof(*pclc_smcd) -
558 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
559 if (ini->smc_type_v1 != SMC_TYPE_N)
560 v2_ext_offset += sizeof(*pclc_prfx) +
561 pclc_prfx->ipv6_prefixes_cnt *
562 sizeof(ipv6_prfx[0]);
563 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
564 v2_ext->hdr.eid_cnt = 0;
565 v2_ext->hdr.ism_gid_cnt = ini->ism_offered_cnt;
566 v2_ext->hdr.flag.release = SMC_RELEASE;
567 v2_ext->hdr.flag.seid = 1;
568 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
569 offsetofend(struct smc_clnt_opts_area_hdr,
570 smcd_v2_ext_offset) +
571 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
572 if (ini->ism_dev[0])
573 smc_ism_get_system_eid(ini->ism_dev[0], &eid);
574 else
575 smc_ism_get_system_eid(ini->ism_dev[1], &eid);
576 if (eid)
577 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
578 plen += sizeof(*v2_ext) + sizeof(*smcd_v2_ext);
579 if (ini->ism_offered_cnt) {
580 for (i = 1; i <= ini->ism_offered_cnt; i++) {
581 gidchids[i - 1].gid =
582 htonll(ini->ism_dev[i]->local_gid);
583 gidchids[i - 1].chid =
584 htons(smc_ism_get_chid(ini->ism_dev[i]));
585 }
586 plen += ini->ism_offered_cnt *
587 sizeof(struct smc_clc_smcd_gid_chid);
588 }
589 }
590 pclc_base->hdr.length = htons(plen);
591 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
592
593 /* send SMC Proposal CLC message */
594 memset(&msg, 0, sizeof(msg));
595 i = 0;
596 vec[i].iov_base = pclc_base;
597 vec[i++].iov_len = sizeof(*pclc_base);
598 vec[i].iov_base = pclc_smcd;
599 vec[i++].iov_len = sizeof(*pclc_smcd);
600 if (ini->smc_type_v1 != SMC_TYPE_N) {
601 vec[i].iov_base = pclc_prfx;
602 vec[i++].iov_len = sizeof(*pclc_prfx);
603 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
604 vec[i].iov_base = ipv6_prfx;
605 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
606 sizeof(ipv6_prfx[0]);
607 }
608 }
609 if (ini->smc_type_v2 != SMC_TYPE_N) {
610 vec[i].iov_base = v2_ext;
611 vec[i++].iov_len = sizeof(*v2_ext);
612 vec[i].iov_base = smcd_v2_ext;
613 vec[i++].iov_len = sizeof(*smcd_v2_ext);
614 if (ini->ism_offered_cnt) {
615 vec[i].iov_base = gidchids;
616 vec[i++].iov_len = ini->ism_offered_cnt *
617 sizeof(struct smc_clc_smcd_gid_chid);
618 }
619 }
620 vec[i].iov_base = trl;
621 vec[i++].iov_len = sizeof(*trl);
622 /* due to the few bytes needed for clc-handshake this cannot block */
623 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
624 if (len < 0) {
625 smc->sk.sk_err = smc->clcsock->sk->sk_err;
626 reason_code = -smc->sk.sk_err;
627 } else if (len < ntohs(pclc_base->hdr.length)) {
628 reason_code = -ENETUNREACH;
629 smc->sk.sk_err = -reason_code;
630 }
631
632 kfree(pclc);
633 return reason_code;
634}
635
636/* build and send CLC CONFIRM / ACCEPT message */
637static int smc_clc_send_confirm_accept(struct smc_sock *smc,
638 struct smc_clc_msg_accept_confirm_v2 *clc_v2,
639 int first_contact, u8 version)
640{
641 struct smc_connection *conn = &smc->conn;
642 struct smc_clc_msg_accept_confirm *clc;
643 struct smc_clc_first_contact_ext fce;
644 struct smc_clc_msg_trail trl;
645 struct kvec vec[3];
646 struct msghdr msg;
647 int i, len;
648
649 /* send SMC Confirm CLC msg */
650 clc = (struct smc_clc_msg_accept_confirm *)clc_v2;
651 clc->hdr.version = version; /* SMC version */
652 if (first_contact)
653 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
654 if (conn->lgr->is_smcd) {
655 /* SMC-D specific settings */
656 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
657 sizeof(SMCD_EYECATCHER));
658 clc->hdr.typev1 = SMC_TYPE_D;
659 clc->d0.gid = conn->lgr->smcd->local_gid;
660 clc->d0.token = conn->rmb_desc->token;
661 clc->d0.dmbe_size = conn->rmbe_size_short;
662 clc->d0.dmbe_idx = 0;
663 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
664 if (version == SMC_V1) {
665 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
666 } else {
667 u8 *eid = NULL;
668
669 clc_v2->chid = htons(smc_ism_get_chid(conn->lgr->smcd));
670 smc_ism_get_system_eid(conn->lgr->smcd, &eid);
671 if (eid)
672 memcpy(clc_v2->eid, eid, SMC_MAX_EID_LEN);
673 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
674 if (first_contact)
675 smc_clc_fill_fce(&fce, &len);
676 clc_v2->hdr.length = htons(len);
677 }
678 memcpy(trl.eyecatcher, SMCD_EYECATCHER,
679 sizeof(SMCD_EYECATCHER));
680 } else {
681 struct smc_link *link = conn->lnk;
682
683 /* SMC-R specific settings */
684 link = conn->lnk;
685 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
686 sizeof(SMC_EYECATCHER));
687 clc->hdr.typev1 = SMC_TYPE_R;
688 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
689 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
690 sizeof(local_systemid));
691 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
692 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
693 ETH_ALEN);
694 hton24(clc->r0.qpn, link->roce_qp->qp_num);
695 clc->r0.rmb_rkey =
696 htonl(conn->rmb_desc->mr_rx[link->link_idx]->rkey);
697 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
698 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
699 switch (clc->hdr.type) {
700 case SMC_CLC_ACCEPT:
701 clc->r0.qp_mtu = link->path_mtu;
702 break;
703 case SMC_CLC_CONFIRM:
704 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
705 break;
706 }
707 clc->r0.rmbe_size = conn->rmbe_size_short;
708 clc->r0.rmb_dma_addr = cpu_to_be64((u64)sg_dma_address
709 (conn->rmb_desc->sgt[link->link_idx].sgl));
710 hton24(clc->r0.psn, link->psn_initial);
711 memcpy(trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
712 }
713
714 memset(&msg, 0, sizeof(msg));
715 i = 0;
716 vec[i].iov_base = clc_v2;
717 if (version > SMC_V1)
718 vec[i++].iov_len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 - sizeof(trl);
719 else
720 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
721 SMCD_CLC_ACCEPT_CONFIRM_LEN :
722 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
723 sizeof(trl);
724 if (version > SMC_V1 && first_contact) {
725 vec[i].iov_base = &fce;
726 vec[i++].iov_len = sizeof(fce);
727 }
728 vec[i].iov_base = &trl;
729 vec[i++].iov_len = sizeof(trl);
730 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
731 ntohs(clc->hdr.length));
732}
733
734/* send CLC CONFIRM message across internal TCP socket */
735int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
736 u8 version)
737{
738 struct smc_clc_msg_accept_confirm_v2 cclc_v2;
739 int reason_code = 0;
740 int len;
741
742 /* send SMC Confirm CLC msg */
743 memset(&cclc_v2, 0, sizeof(cclc_v2));
744 cclc_v2.hdr.type = SMC_CLC_CONFIRM;
745 len = smc_clc_send_confirm_accept(smc, &cclc_v2, clnt_first_contact,
746 version);
747 if (len < ntohs(cclc_v2.hdr.length)) {
748 if (len >= 0) {
749 reason_code = -ENETUNREACH;
750 smc->sk.sk_err = -reason_code;
751 } else {
752 smc->sk.sk_err = smc->clcsock->sk->sk_err;
753 reason_code = -smc->sk.sk_err;
754 }
755 }
756 return reason_code;
757}
758
759/* send CLC ACCEPT message across internal TCP socket */
760int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
761 u8 version)
762{
763 struct smc_clc_msg_accept_confirm_v2 aclc_v2;
764 int len;
765
766 memset(&aclc_v2, 0, sizeof(aclc_v2));
767 aclc_v2.hdr.type = SMC_CLC_ACCEPT;
768 len = smc_clc_send_confirm_accept(new_smc, &aclc_v2, srv_first_contact,
769 version);
770 if (len < ntohs(aclc_v2.hdr.length))
771 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
772
773 return len > 0 ? 0 : len;
774}
775
776void smc_clc_get_hostname(u8 **host)
777{
778 *host = &smc_hostname[0];
779}
780
781void __init smc_clc_init(void)
782{
783 struct new_utsname *u;
784
785 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
786 u = utsname();
787 memcpy(smc_hostname, u->nodename,
788 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
789}