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 */
38 static const char SMC_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xd9'};
39 /* eye catcher "SMCD" EBCDIC for CLC messages */
40 static const char SMCD_EYECATCHER[4] = {'\xe2', '\xd4', '\xc3', '\xc4'};
41
42 static u8 smc_hostname[SMC_MAX_HOSTNAME_LEN];
43
44 struct smc_clc_eid_table {
45 rwlock_t lock;
46 struct list_head list;
47 u8 ueid_cnt;
48 u8 seid_enabled;
49 };
50
51 static struct smc_clc_eid_table smc_clc_eid_table;
52
53 struct 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 */
smc_clc_ueid_valid(char * ueid)63 static 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
smc_clc_ueid_add(char * ueid)82 static 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
112 err_out:
113 write_unlock(&smc_clc_eid_table.lock);
114 kfree(new_ueid);
115 return rc;
116 }
117
smc_clc_ueid_count(void)118 int 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
smc_nl_add_ueid(struct sk_buff * skb,struct genl_info * info)129 int 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 */
smc_clc_ueid_remove(char * ueid)142 static 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
smc_nl_remove_ueid(struct sk_buff * skb,struct genl_info * info)168 int 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
smc_nl_flush_ueid(struct sk_buff * skb,struct genl_info * info)180 int smc_nl_flush_ueid(struct sk_buff *skb, struct genl_info *info)
181 {
182 smc_clc_ueid_remove(NULL);
183 return 0;
184 }
185
smc_nl_ueid_dumpinfo(struct sk_buff * skb,u32 portid,u32 seq,u32 flags,char * ueid)186 static 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
_smc_nl_ueid_dump(struct sk_buff * skb,u32 portid,u32 seq,int start_idx)206 static 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
smc_nl_dump_ueid(struct sk_buff * skb,struct netlink_callback * cb)226 int 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
smc_nl_dump_seid(struct sk_buff * skb,struct netlink_callback * cb)238 int 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;
267 end:
268 genlmsg_end(skb, hdr);
269 cb_ctx->pos[0]++;
270 return skb->len;
271 err:
272 genlmsg_cancel(skb, hdr);
273 return -EMSGSIZE;
274 }
275
smc_nl_enable_seid(struct sk_buff * skb,struct genl_info * info)276 int 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
smc_nl_disable_seid(struct sk_buff * skb,struct genl_info * info)288 int 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
_smc_clc_match_ueid(u8 * peer_ueid)305 static 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
smc_clc_match_eid(u8 * negotiated_eid,struct smc_clc_v2_extension * smc_v2_ext,u8 * peer_eid,u8 * local_eid)316 bool 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 }
342 out:
343 read_unlock(&smc_clc_eid_table.lock);
344 return match;
345 }
346
347 /* check arriving CLC proposal */
smc_clc_msg_prop_valid(struct smc_clc_msg_proposal * pclc)348 static 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 */
389 static bool
smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm_v2 * clc_v2)390 smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm_v2 *clc_v2)
391 {
392 struct smc_clc_msg_hdr *hdr = &clc_v2->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 */
414 static bool
smc_clc_msg_decl_valid(struct smc_clc_msg_decline * dclc)415 smc_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
smc_clc_fill_fce(struct smc_clc_first_contact_ext_v2x * fce,struct smc_init_info * ini)431 static int smc_clc_fill_fce(struct smc_clc_first_contact_ext_v2x *fce,
432 struct smc_init_info *ini)
433 {
434 int ret = sizeof(*fce);
435
436 memset(fce, 0, sizeof(*fce));
437 fce->fce_v2_base.os_type = SMC_CLC_OS_LINUX;
438 fce->fce_v2_base.release = ini->release_nr;
439 memcpy(fce->fce_v2_base.hostname, smc_hostname, sizeof(smc_hostname));
440 if (ini->is_smcd && ini->release_nr < SMC_RELEASE_1) {
441 ret = sizeof(struct smc_clc_first_contact_ext);
442 goto out;
443 }
444
445 if (ini->release_nr >= SMC_RELEASE_1) {
446 if (!ini->is_smcd) {
447 fce->max_conns = ini->max_conns;
448 fce->max_links = ini->max_links;
449 }
450 }
451
452 out:
453 return ret;
454 }
455
456 /* check if received message has a correct header length and contains valid
457 * heading and trailing eyecatchers
458 */
smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr * clcm,bool check_trl)459 static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm, bool check_trl)
460 {
461 struct smc_clc_msg_accept_confirm_v2 *clc_v2;
462 struct smc_clc_msg_proposal *pclc;
463 struct smc_clc_msg_decline *dclc;
464 struct smc_clc_msg_trail *trl;
465
466 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
467 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
468 return false;
469 switch (clcm->type) {
470 case SMC_CLC_PROPOSAL:
471 pclc = (struct smc_clc_msg_proposal *)clcm;
472 if (!smc_clc_msg_prop_valid(pclc))
473 return false;
474 trl = (struct smc_clc_msg_trail *)
475 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
476 break;
477 case SMC_CLC_ACCEPT:
478 case SMC_CLC_CONFIRM:
479 clc_v2 = (struct smc_clc_msg_accept_confirm_v2 *)clcm;
480 if (!smc_clc_msg_acc_conf_valid(clc_v2))
481 return false;
482 trl = (struct smc_clc_msg_trail *)
483 ((u8 *)clc_v2 + ntohs(clc_v2->hdr.length) -
484 sizeof(*trl));
485 break;
486 case SMC_CLC_DECLINE:
487 dclc = (struct smc_clc_msg_decline *)clcm;
488 if (!smc_clc_msg_decl_valid(dclc))
489 return false;
490 check_trl = false;
491 break;
492 default:
493 return false;
494 }
495 if (check_trl &&
496 memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
497 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
498 return false;
499 return true;
500 }
501
502 /* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
smc_clc_prfx_set4_rcu(struct dst_entry * dst,__be32 ipv4,struct smc_clc_msg_proposal_prefix * prop)503 static int smc_clc_prfx_set4_rcu(struct dst_entry *dst, __be32 ipv4,
504 struct smc_clc_msg_proposal_prefix *prop)
505 {
506 struct in_device *in_dev = __in_dev_get_rcu(dst->dev);
507 const struct in_ifaddr *ifa;
508
509 if (!in_dev)
510 return -ENODEV;
511
512 in_dev_for_each_ifa_rcu(ifa, in_dev) {
513 if (!inet_ifa_match(ipv4, ifa))
514 continue;
515 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
516 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
517 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
518 return 0;
519 }
520 return -ENOENT;
521 }
522
523 /* fill CLC proposal msg with ipv6 prefixes from device */
smc_clc_prfx_set6_rcu(struct dst_entry * dst,struct smc_clc_msg_proposal_prefix * prop,struct smc_clc_ipv6_prefix * ipv6_prfx)524 static int smc_clc_prfx_set6_rcu(struct dst_entry *dst,
525 struct smc_clc_msg_proposal_prefix *prop,
526 struct smc_clc_ipv6_prefix *ipv6_prfx)
527 {
528 #if IS_ENABLED(CONFIG_IPV6)
529 struct inet6_dev *in6_dev = __in6_dev_get(dst->dev);
530 struct inet6_ifaddr *ifa;
531 int cnt = 0;
532
533 if (!in6_dev)
534 return -ENODEV;
535 /* use a maximum of 8 IPv6 prefixes from device */
536 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
537 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
538 continue;
539 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
540 &ifa->addr, ifa->prefix_len);
541 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
542 cnt++;
543 if (cnt == SMC_CLC_MAX_V6_PREFIX)
544 break;
545 }
546 prop->ipv6_prefixes_cnt = cnt;
547 if (cnt)
548 return 0;
549 #endif
550 return -ENOENT;
551 }
552
553 /* retrieve and set prefixes in CLC proposal msg */
smc_clc_prfx_set(struct socket * clcsock,struct smc_clc_msg_proposal_prefix * prop,struct smc_clc_ipv6_prefix * ipv6_prfx)554 static int smc_clc_prfx_set(struct socket *clcsock,
555 struct smc_clc_msg_proposal_prefix *prop,
556 struct smc_clc_ipv6_prefix *ipv6_prfx)
557 {
558 struct dst_entry *dst = sk_dst_get(clcsock->sk);
559 struct sockaddr_storage addrs;
560 struct sockaddr_in6 *addr6;
561 struct sockaddr_in *addr;
562 int rc = -ENOENT;
563
564 if (!dst) {
565 rc = -ENOTCONN;
566 goto out;
567 }
568 if (!dst->dev) {
569 rc = -ENODEV;
570 goto out_rel;
571 }
572 /* get address to which the internal TCP socket is bound */
573 if (kernel_getsockname(clcsock, (struct sockaddr *)&addrs) < 0)
574 goto out_rel;
575 /* analyze IP specific data of net_device belonging to TCP socket */
576 addr6 = (struct sockaddr_in6 *)&addrs;
577 rcu_read_lock();
578 if (addrs.ss_family == PF_INET) {
579 /* IPv4 */
580 addr = (struct sockaddr_in *)&addrs;
581 rc = smc_clc_prfx_set4_rcu(dst, addr->sin_addr.s_addr, prop);
582 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
583 /* mapped IPv4 address - peer is IPv4 only */
584 rc = smc_clc_prfx_set4_rcu(dst, addr6->sin6_addr.s6_addr32[3],
585 prop);
586 } else {
587 /* IPv6 */
588 rc = smc_clc_prfx_set6_rcu(dst, prop, ipv6_prfx);
589 }
590 rcu_read_unlock();
591 out_rel:
592 dst_release(dst);
593 out:
594 return rc;
595 }
596
597 /* match ipv4 addrs of dev against addr in CLC proposal */
smc_clc_prfx_match4_rcu(struct net_device * dev,struct smc_clc_msg_proposal_prefix * prop)598 static int smc_clc_prfx_match4_rcu(struct net_device *dev,
599 struct smc_clc_msg_proposal_prefix *prop)
600 {
601 struct in_device *in_dev = __in_dev_get_rcu(dev);
602 const struct in_ifaddr *ifa;
603
604 if (!in_dev)
605 return -ENODEV;
606 in_dev_for_each_ifa_rcu(ifa, in_dev) {
607 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
608 inet_ifa_match(prop->outgoing_subnet, ifa))
609 return 0;
610 }
611
612 return -ENOENT;
613 }
614
615 /* match ipv6 addrs of dev against addrs in CLC proposal */
smc_clc_prfx_match6_rcu(struct net_device * dev,struct smc_clc_msg_proposal_prefix * prop)616 static int smc_clc_prfx_match6_rcu(struct net_device *dev,
617 struct smc_clc_msg_proposal_prefix *prop)
618 {
619 #if IS_ENABLED(CONFIG_IPV6)
620 struct inet6_dev *in6_dev = __in6_dev_get(dev);
621 struct smc_clc_ipv6_prefix *ipv6_prfx;
622 struct inet6_ifaddr *ifa;
623 int i, max;
624
625 if (!in6_dev)
626 return -ENODEV;
627 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
628 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
629 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
630 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
631 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
632 continue;
633 for (i = 0; i < max; i++) {
634 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
635 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
636 ifa->prefix_len))
637 return 0;
638 }
639 }
640 #endif
641 return -ENOENT;
642 }
643
644 /* check if proposed prefixes match one of our device prefixes */
smc_clc_prfx_match(struct socket * clcsock,struct smc_clc_msg_proposal_prefix * prop)645 int smc_clc_prfx_match(struct socket *clcsock,
646 struct smc_clc_msg_proposal_prefix *prop)
647 {
648 struct dst_entry *dst = sk_dst_get(clcsock->sk);
649 int rc;
650
651 if (!dst) {
652 rc = -ENOTCONN;
653 goto out;
654 }
655 if (!dst->dev) {
656 rc = -ENODEV;
657 goto out_rel;
658 }
659 rcu_read_lock();
660 if (!prop->ipv6_prefixes_cnt)
661 rc = smc_clc_prfx_match4_rcu(dst->dev, prop);
662 else
663 rc = smc_clc_prfx_match6_rcu(dst->dev, prop);
664 rcu_read_unlock();
665 out_rel:
666 dst_release(dst);
667 out:
668 return rc;
669 }
670
671 /* Wait for data on the tcp-socket, analyze received data
672 * Returns:
673 * 0 if success and it was not a decline that we received.
674 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
675 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
676 */
smc_clc_wait_msg(struct smc_sock * smc,void * buf,int buflen,u8 expected_type,unsigned long timeout)677 int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
678 u8 expected_type, unsigned long timeout)
679 {
680 long rcvtimeo = smc->clcsock->sk->sk_rcvtimeo;
681 struct sock *clc_sk = smc->clcsock->sk;
682 struct smc_clc_msg_hdr *clcm = buf;
683 struct msghdr msg = {NULL, 0};
684 int reason_code = 0;
685 struct kvec vec = {buf, buflen};
686 int len, datlen, recvlen;
687 bool check_trl = true;
688 int krflags;
689
690 /* peek the first few bytes to determine length of data to receive
691 * so we don't consume any subsequent CLC message or payload data
692 * in the TCP byte stream
693 */
694 /*
695 * Caller must make sure that buflen is no less than
696 * sizeof(struct smc_clc_msg_hdr)
697 */
698 krflags = MSG_PEEK | MSG_WAITALL;
699 clc_sk->sk_rcvtimeo = timeout;
700 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1,
701 sizeof(struct smc_clc_msg_hdr));
702 len = sock_recvmsg(smc->clcsock, &msg, krflags);
703 if (signal_pending(current)) {
704 reason_code = -EINTR;
705 clc_sk->sk_err = EINTR;
706 smc->sk.sk_err = EINTR;
707 goto out;
708 }
709 if (clc_sk->sk_err) {
710 reason_code = -clc_sk->sk_err;
711 if (clc_sk->sk_err == EAGAIN &&
712 expected_type == SMC_CLC_DECLINE)
713 clc_sk->sk_err = 0; /* reset for fallback usage */
714 else
715 smc->sk.sk_err = clc_sk->sk_err;
716 goto out;
717 }
718 if (!len) { /* peer has performed orderly shutdown */
719 smc->sk.sk_err = ECONNRESET;
720 reason_code = -ECONNRESET;
721 goto out;
722 }
723 if (len < 0) {
724 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
725 smc->sk.sk_err = -len;
726 reason_code = len;
727 goto out;
728 }
729 datlen = ntohs(clcm->length);
730 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
731 (clcm->version < SMC_V1) ||
732 ((clcm->type != SMC_CLC_DECLINE) &&
733 (clcm->type != expected_type))) {
734 smc->sk.sk_err = EPROTO;
735 reason_code = -EPROTO;
736 goto out;
737 }
738
739 /* receive the complete CLC message */
740 memset(&msg, 0, sizeof(struct msghdr));
741 if (datlen > buflen) {
742 check_trl = false;
743 recvlen = buflen;
744 } else {
745 recvlen = datlen;
746 }
747 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
748 krflags = MSG_WAITALL;
749 len = sock_recvmsg(smc->clcsock, &msg, krflags);
750 if (len < recvlen || !smc_clc_msg_hdr_valid(clcm, check_trl)) {
751 smc->sk.sk_err = EPROTO;
752 reason_code = -EPROTO;
753 goto out;
754 }
755 datlen -= len;
756 while (datlen) {
757 u8 tmp[SMC_CLC_RECV_BUF_LEN];
758
759 vec.iov_base = &tmp;
760 vec.iov_len = SMC_CLC_RECV_BUF_LEN;
761 /* receive remaining proposal message */
762 recvlen = datlen > SMC_CLC_RECV_BUF_LEN ?
763 SMC_CLC_RECV_BUF_LEN : datlen;
764 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
765 len = sock_recvmsg(smc->clcsock, &msg, krflags);
766 datlen -= len;
767 }
768 if (clcm->type == SMC_CLC_DECLINE) {
769 struct smc_clc_msg_decline *dclc;
770
771 dclc = (struct smc_clc_msg_decline *)clcm;
772 reason_code = SMC_CLC_DECL_PEERDECL;
773 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
774 if (((struct smc_clc_msg_decline *)buf)->hdr.typev2 &
775 SMC_FIRST_CONTACT_MASK) {
776 smc->conn.lgr->sync_err = 1;
777 smc_lgr_terminate_sched(smc->conn.lgr);
778 }
779 }
780
781 out:
782 clc_sk->sk_rcvtimeo = rcvtimeo;
783 return reason_code;
784 }
785
786 /* send CLC DECLINE message across internal TCP socket */
smc_clc_send_decline(struct smc_sock * smc,u32 peer_diag_info,u8 version)787 int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version)
788 {
789 struct smc_clc_msg_decline *dclc_v1;
790 struct smc_clc_msg_decline_v2 dclc;
791 struct msghdr msg;
792 int len, send_len;
793 struct kvec vec;
794
795 dclc_v1 = (struct smc_clc_msg_decline *)&dclc;
796 memset(&dclc, 0, sizeof(dclc));
797 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
798 dclc.hdr.type = SMC_CLC_DECLINE;
799 dclc.hdr.version = version;
800 dclc.os_type = version == SMC_V1 ? 0 : SMC_CLC_OS_LINUX;
801 dclc.hdr.typev2 = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ?
802 SMC_FIRST_CONTACT_MASK : 0;
803 if ((!smc_conn_lgr_valid(&smc->conn) || !smc->conn.lgr->is_smcd) &&
804 smc_ib_is_valid_local_systemid())
805 memcpy(dclc.id_for_peer, local_systemid,
806 sizeof(local_systemid));
807 dclc.peer_diagnosis = htonl(peer_diag_info);
808 if (version == SMC_V1) {
809 memcpy(dclc_v1->trl.eyecatcher, SMC_EYECATCHER,
810 sizeof(SMC_EYECATCHER));
811 send_len = sizeof(*dclc_v1);
812 } else {
813 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER,
814 sizeof(SMC_EYECATCHER));
815 send_len = sizeof(dclc);
816 }
817 dclc.hdr.length = htons(send_len);
818
819 memset(&msg, 0, sizeof(msg));
820 vec.iov_base = &dclc;
821 vec.iov_len = send_len;
822 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, send_len);
823 if (len < 0 || len < send_len)
824 len = -EPROTO;
825 return len > 0 ? 0 : len;
826 }
827
828 /* send CLC PROPOSAL message across internal TCP socket */
smc_clc_send_proposal(struct smc_sock * smc,struct smc_init_info * ini)829 int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini)
830 {
831 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
832 struct smc_clc_msg_proposal_prefix *pclc_prfx;
833 struct smc_clc_msg_proposal *pclc_base;
834 struct smc_clc_smcd_gid_chid *gidchids;
835 struct smc_clc_msg_proposal_area *pclc;
836 struct smc_clc_ipv6_prefix *ipv6_prfx;
837 struct smc_clc_v2_extension *v2_ext;
838 struct smc_clc_msg_smcd *pclc_smcd;
839 struct smc_clc_msg_trail *trl;
840 struct smcd_dev *smcd;
841 int len, i, plen, rc;
842 int reason_code = 0;
843 struct kvec vec[8];
844 struct msghdr msg;
845
846 pclc = kzalloc(sizeof(*pclc), GFP_KERNEL);
847 if (!pclc)
848 return -ENOMEM;
849
850 pclc_base = &pclc->pclc_base;
851 pclc_smcd = &pclc->pclc_smcd;
852 pclc_prfx = &pclc->pclc_prfx;
853 ipv6_prfx = pclc->pclc_prfx_ipv6;
854 v2_ext = &pclc->pclc_v2_ext;
855 smcd_v2_ext = &pclc->pclc_smcd_v2_ext;
856 gidchids = pclc->pclc_gidchids;
857 trl = &pclc->pclc_trl;
858
859 pclc_base->hdr.version = SMC_V2;
860 pclc_base->hdr.typev1 = ini->smc_type_v1;
861 pclc_base->hdr.typev2 = ini->smc_type_v2;
862 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
863
864 /* retrieve ip prefixes for CLC proposal msg */
865 if (ini->smc_type_v1 != SMC_TYPE_N) {
866 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
867 if (rc) {
868 if (ini->smc_type_v2 == SMC_TYPE_N) {
869 kfree(pclc);
870 return SMC_CLC_DECL_CNFERR;
871 }
872 pclc_base->hdr.typev1 = SMC_TYPE_N;
873 } else {
874 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
875 plen += sizeof(*pclc_prfx) +
876 pclc_prfx->ipv6_prefixes_cnt *
877 sizeof(ipv6_prfx[0]);
878 }
879 }
880
881 /* build SMC Proposal CLC message */
882 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
883 sizeof(SMC_EYECATCHER));
884 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
885 if (smcr_indicated(ini->smc_type_v1)) {
886 /* add SMC-R specifics */
887 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
888 sizeof(local_systemid));
889 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
890 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
891 ETH_ALEN);
892 }
893 if (smcd_indicated(ini->smc_type_v1)) {
894 /* add SMC-D specifics */
895 if (ini->ism_dev[0]) {
896 smcd = ini->ism_dev[0];
897 pclc_smcd->ism.gid =
898 htonll(smcd->ops->get_local_gid(smcd));
899 pclc_smcd->ism.chid =
900 htons(smc_ism_get_chid(ini->ism_dev[0]));
901 }
902 }
903 if (ini->smc_type_v2 == SMC_TYPE_N) {
904 pclc_smcd->v2_ext_offset = 0;
905 } else {
906 struct smc_clc_eid_entry *ueident;
907 u16 v2_ext_offset;
908
909 v2_ext->hdr.flag.release = SMC_RELEASE;
910 v2_ext_offset = sizeof(*pclc_smcd) -
911 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
912 if (ini->smc_type_v1 != SMC_TYPE_N)
913 v2_ext_offset += sizeof(*pclc_prfx) +
914 pclc_prfx->ipv6_prefixes_cnt *
915 sizeof(ipv6_prfx[0]);
916 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
917 plen += sizeof(*v2_ext);
918
919 read_lock(&smc_clc_eid_table.lock);
920 v2_ext->hdr.eid_cnt = smc_clc_eid_table.ueid_cnt;
921 plen += smc_clc_eid_table.ueid_cnt * SMC_MAX_EID_LEN;
922 i = 0;
923 list_for_each_entry(ueident, &smc_clc_eid_table.list, list) {
924 memcpy(v2_ext->user_eids[i++], ueident->eid,
925 sizeof(ueident->eid));
926 }
927 read_unlock(&smc_clc_eid_table.lock);
928 }
929 if (smcd_indicated(ini->smc_type_v2)) {
930 u8 *eid = NULL;
931
932 v2_ext->hdr.flag.seid = smc_clc_eid_table.seid_enabled;
933 v2_ext->hdr.ism_gid_cnt = ini->ism_offered_cnt;
934 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
935 offsetofend(struct smc_clnt_opts_area_hdr,
936 smcd_v2_ext_offset) +
937 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
938 smc_ism_get_system_eid(&eid);
939 if (eid && v2_ext->hdr.flag.seid)
940 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
941 plen += sizeof(*smcd_v2_ext);
942 if (ini->ism_offered_cnt) {
943 for (i = 1; i <= ini->ism_offered_cnt; i++) {
944 smcd = ini->ism_dev[i];
945 gidchids[i - 1].gid =
946 htonll(smcd->ops->get_local_gid(smcd));
947 gidchids[i - 1].chid =
948 htons(smc_ism_get_chid(ini->ism_dev[i]));
949 }
950 plen += ini->ism_offered_cnt *
951 sizeof(struct smc_clc_smcd_gid_chid);
952 }
953 }
954 if (smcr_indicated(ini->smc_type_v2)) {
955 memcpy(v2_ext->roce, ini->smcrv2.ib_gid_v2, SMC_GID_SIZE);
956 v2_ext->max_conns = SMC_CONN_PER_LGR_PREFER;
957 v2_ext->max_links = SMC_LINKS_PER_LGR_MAX_PREFER;
958 }
959
960 pclc_base->hdr.length = htons(plen);
961 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
962
963 /* send SMC Proposal CLC message */
964 memset(&msg, 0, sizeof(msg));
965 i = 0;
966 vec[i].iov_base = pclc_base;
967 vec[i++].iov_len = sizeof(*pclc_base);
968 vec[i].iov_base = pclc_smcd;
969 vec[i++].iov_len = sizeof(*pclc_smcd);
970 if (ini->smc_type_v1 != SMC_TYPE_N) {
971 vec[i].iov_base = pclc_prfx;
972 vec[i++].iov_len = sizeof(*pclc_prfx);
973 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
974 vec[i].iov_base = ipv6_prfx;
975 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
976 sizeof(ipv6_prfx[0]);
977 }
978 }
979 if (ini->smc_type_v2 != SMC_TYPE_N) {
980 vec[i].iov_base = v2_ext;
981 vec[i++].iov_len = sizeof(*v2_ext) +
982 (v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
983 if (smcd_indicated(ini->smc_type_v2)) {
984 vec[i].iov_base = smcd_v2_ext;
985 vec[i++].iov_len = sizeof(*smcd_v2_ext);
986 if (ini->ism_offered_cnt) {
987 vec[i].iov_base = gidchids;
988 vec[i++].iov_len = ini->ism_offered_cnt *
989 sizeof(struct smc_clc_smcd_gid_chid);
990 }
991 }
992 }
993 vec[i].iov_base = trl;
994 vec[i++].iov_len = sizeof(*trl);
995 /* due to the few bytes needed for clc-handshake this cannot block */
996 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
997 if (len < 0) {
998 smc->sk.sk_err = smc->clcsock->sk->sk_err;
999 reason_code = -smc->sk.sk_err;
1000 } else if (len < ntohs(pclc_base->hdr.length)) {
1001 reason_code = -ENETUNREACH;
1002 smc->sk.sk_err = -reason_code;
1003 }
1004
1005 kfree(pclc);
1006 return reason_code;
1007 }
1008
1009 /* build and send CLC CONFIRM / ACCEPT message */
smc_clc_send_confirm_accept(struct smc_sock * smc,struct smc_clc_msg_accept_confirm_v2 * clc_v2,int first_contact,u8 version,u8 * eid,struct smc_init_info * ini)1010 static int smc_clc_send_confirm_accept(struct smc_sock *smc,
1011 struct smc_clc_msg_accept_confirm_v2 *clc_v2,
1012 int first_contact, u8 version,
1013 u8 *eid, struct smc_init_info *ini)
1014 {
1015 struct smc_connection *conn = &smc->conn;
1016 struct smc_clc_first_contact_ext_v2x fce;
1017 struct smcd_dev *smcd = conn->lgr->smcd;
1018 struct smc_clc_msg_accept_confirm *clc;
1019 struct smc_clc_fce_gid_ext gle;
1020 struct smc_clc_msg_trail trl;
1021 int i, len, fce_len;
1022 struct kvec vec[5];
1023 struct msghdr msg;
1024
1025 /* send SMC Confirm CLC msg */
1026 clc = (struct smc_clc_msg_accept_confirm *)clc_v2;
1027 clc->hdr.version = version; /* SMC version */
1028 if (first_contact)
1029 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
1030 if (conn->lgr->is_smcd) {
1031 /* SMC-D specific settings */
1032 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
1033 sizeof(SMCD_EYECATCHER));
1034 clc->hdr.typev1 = SMC_TYPE_D;
1035 clc->d0.gid = htonll(smcd->ops->get_local_gid(smcd));
1036 clc->d0.token = htonll(conn->rmb_desc->token);
1037 clc->d0.dmbe_size = conn->rmbe_size_comp;
1038 clc->d0.dmbe_idx = 0;
1039 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
1040 if (version == SMC_V1) {
1041 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
1042 } else {
1043 clc_v2->d1.chid = htons(smc_ism_get_chid(smcd));
1044 if (eid && eid[0])
1045 memcpy(clc_v2->d1.eid, eid, SMC_MAX_EID_LEN);
1046 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
1047 if (first_contact) {
1048 fce_len = smc_clc_fill_fce(&fce, ini);
1049 len += fce_len;
1050 }
1051 clc_v2->hdr.length = htons(len);
1052 }
1053 memcpy(trl.eyecatcher, SMCD_EYECATCHER,
1054 sizeof(SMCD_EYECATCHER));
1055 } else {
1056 struct smc_link *link = conn->lnk;
1057
1058 /* SMC-R specific settings */
1059 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
1060 sizeof(SMC_EYECATCHER));
1061 clc->hdr.typev1 = SMC_TYPE_R;
1062 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1063 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
1064 sizeof(local_systemid));
1065 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
1066 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
1067 ETH_ALEN);
1068 hton24(clc->r0.qpn, link->roce_qp->qp_num);
1069 clc->r0.rmb_rkey =
1070 htonl(conn->rmb_desc->mr[link->link_idx]->rkey);
1071 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
1072 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
1073 switch (clc->hdr.type) {
1074 case SMC_CLC_ACCEPT:
1075 clc->r0.qp_mtu = link->path_mtu;
1076 break;
1077 case SMC_CLC_CONFIRM:
1078 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
1079 break;
1080 }
1081 clc->r0.rmbe_size = conn->rmbe_size_comp;
1082 clc->r0.rmb_dma_addr = conn->rmb_desc->is_vm ?
1083 cpu_to_be64((uintptr_t)conn->rmb_desc->cpu_addr) :
1084 cpu_to_be64((u64)sg_dma_address
1085 (conn->rmb_desc->sgt[link->link_idx].sgl));
1086 hton24(clc->r0.psn, link->psn_initial);
1087 if (version == SMC_V1) {
1088 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1089 } else {
1090 if (eid && eid[0])
1091 memcpy(clc_v2->r1.eid, eid, SMC_MAX_EID_LEN);
1092 len = SMCR_CLC_ACCEPT_CONFIRM_LEN_V2;
1093 if (first_contact) {
1094 fce_len = smc_clc_fill_fce(&fce, ini);
1095 len += fce_len;
1096 fce.fce_v2_base.v2_direct = !link->lgr->uses_gateway;
1097 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1098 memset(&gle, 0, sizeof(gle));
1099 gle.gid_cnt = ini->smcrv2.gidlist.len;
1100 len += sizeof(gle);
1101 len += gle.gid_cnt * sizeof(gle.gid[0]);
1102 }
1103 }
1104 clc_v2->hdr.length = htons(len);
1105 }
1106 memcpy(trl.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
1107 }
1108
1109 memset(&msg, 0, sizeof(msg));
1110 i = 0;
1111 vec[i].iov_base = clc_v2;
1112 if (version > SMC_V1)
1113 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1114 SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 :
1115 SMCR_CLC_ACCEPT_CONFIRM_LEN_V2) -
1116 sizeof(trl);
1117 else
1118 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1119 SMCD_CLC_ACCEPT_CONFIRM_LEN :
1120 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
1121 sizeof(trl);
1122 if (version > SMC_V1 && first_contact) {
1123 vec[i].iov_base = &fce;
1124 vec[i++].iov_len = fce_len;
1125 if (!conn->lgr->is_smcd) {
1126 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1127 vec[i].iov_base = &gle;
1128 vec[i++].iov_len = sizeof(gle);
1129 vec[i].iov_base = &ini->smcrv2.gidlist.list;
1130 vec[i++].iov_len = gle.gid_cnt *
1131 sizeof(gle.gid[0]);
1132 }
1133 }
1134 }
1135 vec[i].iov_base = &trl;
1136 vec[i++].iov_len = sizeof(trl);
1137 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
1138 ntohs(clc->hdr.length));
1139 }
1140
1141 /* send CLC CONFIRM message across internal TCP socket */
smc_clc_send_confirm(struct smc_sock * smc,bool clnt_first_contact,u8 version,u8 * eid,struct smc_init_info * ini)1142 int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
1143 u8 version, u8 *eid, struct smc_init_info *ini)
1144 {
1145 struct smc_clc_msg_accept_confirm_v2 cclc_v2;
1146 int reason_code = 0;
1147 int len;
1148
1149 /* send SMC Confirm CLC msg */
1150 memset(&cclc_v2, 0, sizeof(cclc_v2));
1151 cclc_v2.hdr.type = SMC_CLC_CONFIRM;
1152 len = smc_clc_send_confirm_accept(smc, &cclc_v2, clnt_first_contact,
1153 version, eid, ini);
1154 if (len < ntohs(cclc_v2.hdr.length)) {
1155 if (len >= 0) {
1156 reason_code = -ENETUNREACH;
1157 smc->sk.sk_err = -reason_code;
1158 } else {
1159 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1160 reason_code = -smc->sk.sk_err;
1161 }
1162 }
1163 return reason_code;
1164 }
1165
1166 /* send CLC ACCEPT message across internal TCP socket */
smc_clc_send_accept(struct smc_sock * new_smc,bool srv_first_contact,u8 version,u8 * negotiated_eid,struct smc_init_info * ini)1167 int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
1168 u8 version, u8 *negotiated_eid, struct smc_init_info *ini)
1169 {
1170 struct smc_clc_msg_accept_confirm_v2 aclc_v2;
1171 int len;
1172
1173 memset(&aclc_v2, 0, sizeof(aclc_v2));
1174 aclc_v2.hdr.type = SMC_CLC_ACCEPT;
1175 len = smc_clc_send_confirm_accept(new_smc, &aclc_v2, srv_first_contact,
1176 version, negotiated_eid, ini);
1177 if (len < ntohs(aclc_v2.hdr.length))
1178 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
1179
1180 return len > 0 ? 0 : len;
1181 }
1182
smc_clc_srv_v2x_features_validate(struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)1183 int smc_clc_srv_v2x_features_validate(struct smc_clc_msg_proposal *pclc,
1184 struct smc_init_info *ini)
1185 {
1186 struct smc_clc_v2_extension *pclc_v2_ext;
1187
1188 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1189 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1190
1191 if ((!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) ||
1192 ini->release_nr < SMC_RELEASE_1)
1193 return 0;
1194
1195 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1196 if (!pclc_v2_ext)
1197 return SMC_CLC_DECL_NOV2EXT;
1198
1199 if (ini->smcr_version & SMC_V2) {
1200 ini->max_conns = min_t(u8, pclc_v2_ext->max_conns, SMC_CONN_PER_LGR_PREFER);
1201 if (ini->max_conns < SMC_CONN_PER_LGR_MIN)
1202 return SMC_CLC_DECL_MAXCONNERR;
1203
1204 ini->max_links = min_t(u8, pclc_v2_ext->max_links, SMC_LINKS_PER_LGR_MAX_PREFER);
1205 if (ini->max_links < SMC_LINKS_ADD_LNK_MIN)
1206 return SMC_CLC_DECL_MAXLINKERR;
1207 }
1208
1209 return 0;
1210 }
1211
smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext * fce,struct smc_init_info * ini)1212 int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
1213 struct smc_init_info *ini)
1214 {
1215 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1216 (struct smc_clc_first_contact_ext_v2x *)fce;
1217
1218 if (ini->release_nr < SMC_RELEASE_1)
1219 return 0;
1220
1221 if (!ini->is_smcd) {
1222 if (fce_v2x->max_conns < SMC_CONN_PER_LGR_MIN)
1223 return SMC_CLC_DECL_MAXCONNERR;
1224 ini->max_conns = fce_v2x->max_conns;
1225
1226 if (fce_v2x->max_links > SMC_LINKS_ADD_LNK_MAX ||
1227 fce_v2x->max_links < SMC_LINKS_ADD_LNK_MIN)
1228 return SMC_CLC_DECL_MAXLINKERR;
1229 ini->max_links = fce_v2x->max_links;
1230 }
1231
1232 return 0;
1233 }
1234
smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm * cclc,struct smc_init_info * ini)1235 int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
1236 struct smc_init_info *ini)
1237 {
1238 struct smc_clc_msg_accept_confirm_v2 *clc_v2 =
1239 (struct smc_clc_msg_accept_confirm_v2 *)cclc;
1240 struct smc_clc_first_contact_ext *fce =
1241 smc_get_clc_first_contact_ext(clc_v2, ini->is_smcd);
1242 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1243 (struct smc_clc_first_contact_ext_v2x *)fce;
1244
1245 if (cclc->hdr.version == SMC_V1 ||
1246 !(cclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
1247 return 0;
1248
1249 if (ini->release_nr != fce->release)
1250 return SMC_CLC_DECL_RELEASEERR;
1251
1252 if (fce->release < SMC_RELEASE_1)
1253 return 0;
1254
1255 if (!ini->is_smcd) {
1256 if (fce_v2x->max_conns != ini->max_conns)
1257 return SMC_CLC_DECL_MAXCONNERR;
1258 if (fce_v2x->max_links != ini->max_links)
1259 return SMC_CLC_DECL_MAXLINKERR;
1260 }
1261
1262 return 0;
1263 }
1264
smc_clc_get_hostname(u8 ** host)1265 void smc_clc_get_hostname(u8 **host)
1266 {
1267 *host = &smc_hostname[0];
1268 }
1269
smc_clc_init(void)1270 void __init smc_clc_init(void)
1271 {
1272 struct new_utsname *u;
1273
1274 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
1275 u = utsname();
1276 memcpy(smc_hostname, u->nodename,
1277 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
1278
1279 INIT_LIST_HEAD(&smc_clc_eid_table.list);
1280 rwlock_init(&smc_clc_eid_table.lock);
1281 smc_clc_eid_table.ueid_cnt = 0;
1282 #if IS_ENABLED(CONFIG_S390)
1283 smc_clc_eid_table.seid_enabled = 1;
1284 #else
1285 smc_clc_eid_table.seid_enabled = 0;
1286 #endif
1287 }
1288
smc_clc_exit(void)1289 void smc_clc_exit(void)
1290 {
1291 smc_clc_ueid_remove(NULL);
1292 }
1293