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 * clc)390 smc_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 */
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_v2x(struct smc_clc_first_contact_ext_v2x * fce_v2x,struct smc_init_info * ini)431 static 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
454 out:
455 return ret;
456 }
457
458 /* check if received message has a correct header length and contains valid
459 * heading and trailing eyecatchers
460 */
smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr * clcm,bool check_trl)461 static 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 */
smc_clc_prfx_set4_rcu(struct dst_entry * dst,__be32 ipv4,struct smc_clc_msg_proposal_prefix * prop)504 static 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 */
smc_clc_prfx_set6_rcu(struct dst_entry * dst,struct smc_clc_msg_proposal_prefix * prop,struct smc_clc_ipv6_prefix * ipv6_prfx)525 static 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 */
smc_clc_prfx_set(struct socket * clcsock,struct smc_clc_msg_proposal_prefix * prop,struct smc_clc_ipv6_prefix * ipv6_prfx)555 static 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();
592 out_rel:
593 dst_release(dst);
594 out:
595 return rc;
596 }
597
598 /* 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)599 static 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 */
smc_clc_prfx_match6_rcu(struct net_device * dev,struct smc_clc_msg_proposal_prefix * prop)617 static 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 */
smc_clc_prfx_match(struct socket * clcsock,struct smc_clc_msg_proposal_prefix * prop)646 int 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();
666 out_rel:
667 dst_release(dst);
668 out:
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 */
smc_clc_wait_msg(struct smc_sock * smc,void * buf,int buflen,u8 expected_type,unsigned long timeout)678 int 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
782 out:
783 clc_sk->sk_rcvtimeo = rcvtimeo;
784 return reason_code;
785 }
786
787 /* send CLC DECLINE message across internal TCP socket */
smc_clc_send_decline(struct smc_sock * smc,u32 peer_diag_info,u8 version)788 int 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 */
smc_clc_send_proposal(struct smc_sock * smc,struct smc_init_info * ini)830 int 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 = container_of(&pclc->pclc_v2_ext,
857 struct smc_clc_v2_extension, fixed);
858 smcd_v2_ext = container_of(&pclc->pclc_smcd_v2_ext,
859 struct smc_clc_smcd_v2_extension, fixed);
860 gidchids = pclc->pclc_gidchids;
861 trl = &pclc->pclc_trl;
862
863 pclc_base->hdr.version = SMC_V2;
864 pclc_base->hdr.typev1 = ini->smc_type_v1;
865 pclc_base->hdr.typev2 = ini->smc_type_v2;
866 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
867
868 /* retrieve ip prefixes for CLC proposal msg */
869 if (ini->smc_type_v1 != SMC_TYPE_N) {
870 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
871 if (rc) {
872 if (ini->smc_type_v2 == SMC_TYPE_N) {
873 kfree(pclc);
874 return SMC_CLC_DECL_CNFERR;
875 }
876 pclc_base->hdr.typev1 = SMC_TYPE_N;
877 } else {
878 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
879 plen += sizeof(*pclc_prfx) +
880 pclc_prfx->ipv6_prefixes_cnt *
881 sizeof(ipv6_prfx[0]);
882 }
883 }
884
885 /* build SMC Proposal CLC message */
886 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
887 sizeof(SMC_EYECATCHER));
888 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
889 if (smcr_indicated(ini->smc_type_v1)) {
890 /* add SMC-R specifics */
891 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
892 sizeof(local_systemid));
893 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
894 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
895 ETH_ALEN);
896 }
897 if (smcd_indicated(ini->smc_type_v1)) {
898 struct smcd_gid smcd_gid;
899
900 /* add SMC-D specifics */
901 if (ini->ism_dev[0]) {
902 smcd = ini->ism_dev[0];
903 smcd->ops->get_local_gid(smcd, &smcd_gid);
904 pclc_smcd->ism.gid = htonll(smcd_gid.gid);
905 pclc_smcd->ism.chid =
906 htons(smc_ism_get_chid(ini->ism_dev[0]));
907 }
908 }
909 if (ini->smc_type_v2 == SMC_TYPE_N) {
910 pclc_smcd->v2_ext_offset = 0;
911 } else {
912 struct smc_clc_eid_entry *ueident;
913 u16 v2_ext_offset;
914
915 v2_ext->hdr.flag.release = SMC_RELEASE;
916 v2_ext_offset = sizeof(*pclc_smcd) -
917 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
918 if (ini->smc_type_v1 != SMC_TYPE_N)
919 v2_ext_offset += sizeof(*pclc_prfx) +
920 pclc_prfx->ipv6_prefixes_cnt *
921 sizeof(ipv6_prfx[0]);
922 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
923 plen += sizeof(*v2_ext);
924
925 v2_ext->feature_mask = htons(SMC_FEATURE_MASK);
926 read_lock(&smc_clc_eid_table.lock);
927 v2_ext->hdr.eid_cnt = smc_clc_eid_table.ueid_cnt;
928 plen += smc_clc_eid_table.ueid_cnt * SMC_MAX_EID_LEN;
929 i = 0;
930 list_for_each_entry(ueident, &smc_clc_eid_table.list, list) {
931 memcpy(v2_ext->user_eids[i++], ueident->eid,
932 sizeof(ueident->eid));
933 }
934 read_unlock(&smc_clc_eid_table.lock);
935 }
936 if (smcd_indicated(ini->smc_type_v2)) {
937 struct smcd_gid smcd_gid;
938 u8 *eid = NULL;
939 int entry = 0;
940
941 v2_ext->hdr.flag.seid = smc_clc_eid_table.seid_enabled;
942 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
943 offsetofend(struct smc_clnt_opts_area_hdr,
944 smcd_v2_ext_offset) +
945 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
946 smc_ism_get_system_eid(&eid);
947 if (eid && v2_ext->hdr.flag.seid)
948 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
949 plen += sizeof(*smcd_v2_ext);
950 if (ini->ism_offered_cnt) {
951 for (i = 1; i <= ini->ism_offered_cnt; i++) {
952 smcd = ini->ism_dev[i];
953 smcd->ops->get_local_gid(smcd, &smcd_gid);
954 gidchids[entry].chid =
955 htons(smc_ism_get_chid(ini->ism_dev[i]));
956 gidchids[entry].gid = htonll(smcd_gid.gid);
957 if (smc_ism_is_emulated(smcd)) {
958 /* an Emulated-ISM device takes two
959 * entries. CHID of the second entry
960 * repeats that of the first entry.
961 */
962 gidchids[entry + 1].chid =
963 gidchids[entry].chid;
964 gidchids[entry + 1].gid =
965 htonll(smcd_gid.gid_ext);
966 entry++;
967 }
968 entry++;
969 }
970 plen += entry * sizeof(struct smc_clc_smcd_gid_chid);
971 }
972 v2_ext->hdr.ism_gid_cnt = entry;
973 }
974 if (smcr_indicated(ini->smc_type_v2)) {
975 memcpy(v2_ext->roce, ini->smcrv2.ib_gid_v2, SMC_GID_SIZE);
976 v2_ext->max_conns = net->smc.sysctl_max_conns_per_lgr;
977 v2_ext->max_links = net->smc.sysctl_max_links_per_lgr;
978 }
979
980 pclc_base->hdr.length = htons(plen);
981 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
982
983 /* send SMC Proposal CLC message */
984 memset(&msg, 0, sizeof(msg));
985 i = 0;
986 vec[i].iov_base = pclc_base;
987 vec[i++].iov_len = sizeof(*pclc_base);
988 vec[i].iov_base = pclc_smcd;
989 vec[i++].iov_len = sizeof(*pclc_smcd);
990 if (ini->smc_type_v1 != SMC_TYPE_N) {
991 vec[i].iov_base = pclc_prfx;
992 vec[i++].iov_len = sizeof(*pclc_prfx);
993 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
994 vec[i].iov_base = ipv6_prfx;
995 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
996 sizeof(ipv6_prfx[0]);
997 }
998 }
999 if (ini->smc_type_v2 != SMC_TYPE_N) {
1000 vec[i].iov_base = v2_ext;
1001 vec[i++].iov_len = sizeof(*v2_ext) +
1002 (v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
1003 if (smcd_indicated(ini->smc_type_v2)) {
1004 vec[i].iov_base = smcd_v2_ext;
1005 vec[i++].iov_len = sizeof(*smcd_v2_ext);
1006 if (ini->ism_offered_cnt) {
1007 vec[i].iov_base = gidchids;
1008 vec[i++].iov_len = v2_ext->hdr.ism_gid_cnt *
1009 sizeof(struct smc_clc_smcd_gid_chid);
1010 }
1011 }
1012 }
1013 vec[i].iov_base = trl;
1014 vec[i++].iov_len = sizeof(*trl);
1015 /* due to the few bytes needed for clc-handshake this cannot block */
1016 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
1017 if (len < 0) {
1018 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1019 reason_code = -smc->sk.sk_err;
1020 } else if (len < ntohs(pclc_base->hdr.length)) {
1021 reason_code = -ENETUNREACH;
1022 smc->sk.sk_err = -reason_code;
1023 }
1024
1025 kfree(pclc);
1026 return reason_code;
1027 }
1028
1029 static void
smcd_clc_prep_confirm_accept(struct smc_connection * conn,struct smc_clc_msg_accept_confirm * clc,int first_contact,u8 version,u8 * eid,struct smc_init_info * ini,int * fce_len,struct smc_clc_first_contact_ext_v2x * fce_v2x,struct smc_clc_msg_trail * trl)1030 smcd_clc_prep_confirm_accept(struct smc_connection *conn,
1031 struct smc_clc_msg_accept_confirm *clc,
1032 int first_contact, u8 version,
1033 u8 *eid, struct smc_init_info *ini,
1034 int *fce_len,
1035 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1036 struct smc_clc_msg_trail *trl)
1037 {
1038 struct smcd_dev *smcd = conn->lgr->smcd;
1039 struct smcd_gid smcd_gid;
1040 u16 chid;
1041 int len;
1042
1043 /* SMC-D specific settings */
1044 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
1045 sizeof(SMCD_EYECATCHER));
1046 smcd->ops->get_local_gid(smcd, &smcd_gid);
1047 clc->hdr.typev1 = SMC_TYPE_D;
1048 clc->d0.gid = htonll(smcd_gid.gid);
1049 clc->d0.token = htonll(conn->rmb_desc->token);
1050 clc->d0.dmbe_size = conn->rmbe_size_comp;
1051 clc->d0.dmbe_idx = 0;
1052 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
1053 if (version == SMC_V1) {
1054 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
1055 } else {
1056 chid = smc_ism_get_chid(smcd);
1057 clc->d1.chid = htons(chid);
1058 if (eid && eid[0])
1059 memcpy(clc->d1.eid, eid, SMC_MAX_EID_LEN);
1060 if (__smc_ism_is_emulated(chid))
1061 clc->d1.gid_ext = htonll(smcd_gid.gid_ext);
1062 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
1063 if (first_contact) {
1064 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1065 len += *fce_len;
1066 }
1067 clc->hdr.length = htons(len);
1068 }
1069 memcpy(trl->eyecatcher, SMCD_EYECATCHER,
1070 sizeof(SMCD_EYECATCHER));
1071 }
1072
1073 static void
smcr_clc_prep_confirm_accept(struct smc_connection * conn,struct smc_clc_msg_accept_confirm * clc,int first_contact,u8 version,u8 * eid,struct smc_init_info * ini,int * fce_len,struct smc_clc_first_contact_ext_v2x * fce_v2x,struct smc_clc_fce_gid_ext * gle,struct smc_clc_msg_trail * trl)1074 smcr_clc_prep_confirm_accept(struct smc_connection *conn,
1075 struct smc_clc_msg_accept_confirm *clc,
1076 int first_contact, u8 version,
1077 u8 *eid, struct smc_init_info *ini,
1078 int *fce_len,
1079 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1080 struct smc_clc_fce_gid_ext *gle,
1081 struct smc_clc_msg_trail *trl)
1082 {
1083 struct smc_link *link = conn->lnk;
1084 int len;
1085
1086 /* SMC-R specific settings */
1087 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
1088 sizeof(SMC_EYECATCHER));
1089 clc->hdr.typev1 = SMC_TYPE_R;
1090 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
1091 sizeof(local_systemid));
1092 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
1093 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
1094 ETH_ALEN);
1095 hton24(clc->r0.qpn, link->roce_qp->qp_num);
1096 clc->r0.rmb_rkey =
1097 htonl(conn->rmb_desc->mr[link->link_idx]->rkey);
1098 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
1099 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
1100 switch (clc->hdr.type) {
1101 case SMC_CLC_ACCEPT:
1102 clc->r0.qp_mtu = link->path_mtu;
1103 break;
1104 case SMC_CLC_CONFIRM:
1105 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
1106 break;
1107 }
1108 clc->r0.rmbe_size = conn->rmbe_size_comp;
1109 clc->r0.rmb_dma_addr = conn->rmb_desc->is_vm ?
1110 cpu_to_be64((uintptr_t)conn->rmb_desc->cpu_addr) :
1111 cpu_to_be64((u64)sg_dma_address
1112 (conn->rmb_desc->sgt[link->link_idx].sgl));
1113 hton24(clc->r0.psn, link->psn_initial);
1114 if (version == SMC_V1) {
1115 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1116 } else {
1117 if (eid && eid[0])
1118 memcpy(clc->r1.eid, eid, SMC_MAX_EID_LEN);
1119 len = SMCR_CLC_ACCEPT_CONFIRM_LEN_V2;
1120 if (first_contact) {
1121 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1122 len += *fce_len;
1123 fce_v2x->fce_v2_base.v2_direct =
1124 !link->lgr->uses_gateway;
1125 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1126 memset(gle, 0, sizeof(*gle));
1127 gle->gid_cnt = ini->smcrv2.gidlist.len;
1128 len += sizeof(*gle);
1129 len += gle->gid_cnt * sizeof(gle->gid[0]);
1130 }
1131 }
1132 clc->hdr.length = htons(len);
1133 }
1134 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
1135 }
1136
1137 /* build and send CLC CONFIRM / ACCEPT message */
smc_clc_send_confirm_accept(struct smc_sock * smc,struct smc_clc_msg_accept_confirm * clc,int first_contact,u8 version,u8 * eid,struct smc_init_info * ini)1138 static int smc_clc_send_confirm_accept(struct smc_sock *smc,
1139 struct smc_clc_msg_accept_confirm *clc,
1140 int first_contact, u8 version,
1141 u8 *eid, struct smc_init_info *ini)
1142 {
1143 struct smc_clc_first_contact_ext_v2x fce_v2x;
1144 struct smc_connection *conn = &smc->conn;
1145 struct smc_clc_fce_gid_ext gle;
1146 struct smc_clc_msg_trail trl;
1147 int i, fce_len;
1148 struct kvec vec[5];
1149 struct msghdr msg;
1150
1151 /* send SMC Confirm CLC msg */
1152 clc->hdr.version = version; /* SMC version */
1153 if (first_contact)
1154 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
1155 if (conn->lgr->is_smcd)
1156 smcd_clc_prep_confirm_accept(conn, clc, first_contact,
1157 version, eid, ini, &fce_len,
1158 &fce_v2x, &trl);
1159 else
1160 smcr_clc_prep_confirm_accept(conn, clc, first_contact,
1161 version, eid, ini, &fce_len,
1162 &fce_v2x, &gle, &trl);
1163 memset(&msg, 0, sizeof(msg));
1164 i = 0;
1165 vec[i].iov_base = clc;
1166 if (version > SMC_V1)
1167 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1168 SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 :
1169 SMCR_CLC_ACCEPT_CONFIRM_LEN_V2) -
1170 sizeof(trl);
1171 else
1172 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1173 SMCD_CLC_ACCEPT_CONFIRM_LEN :
1174 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
1175 sizeof(trl);
1176 if (version > SMC_V1 && first_contact) {
1177 vec[i].iov_base = &fce_v2x;
1178 vec[i++].iov_len = fce_len;
1179 if (!conn->lgr->is_smcd) {
1180 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1181 vec[i].iov_base = &gle;
1182 vec[i++].iov_len = sizeof(gle);
1183 vec[i].iov_base = &ini->smcrv2.gidlist.list;
1184 vec[i++].iov_len = gle.gid_cnt *
1185 sizeof(gle.gid[0]);
1186 }
1187 }
1188 }
1189 vec[i].iov_base = &trl;
1190 vec[i++].iov_len = sizeof(trl);
1191 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
1192 ntohs(clc->hdr.length));
1193 }
1194
1195 /* 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)1196 int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
1197 u8 version, u8 *eid, struct smc_init_info *ini)
1198 {
1199 struct smc_clc_msg_accept_confirm cclc;
1200 int reason_code = 0;
1201 int len;
1202
1203 /* send SMC Confirm CLC msg */
1204 memset(&cclc, 0, sizeof(cclc));
1205 cclc.hdr.type = SMC_CLC_CONFIRM;
1206 len = smc_clc_send_confirm_accept(smc, &cclc, clnt_first_contact,
1207 version, eid, ini);
1208 if (len < ntohs(cclc.hdr.length)) {
1209 if (len >= 0) {
1210 reason_code = -ENETUNREACH;
1211 smc->sk.sk_err = -reason_code;
1212 } else {
1213 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1214 reason_code = -smc->sk.sk_err;
1215 }
1216 }
1217 return reason_code;
1218 }
1219
1220 /* 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)1221 int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
1222 u8 version, u8 *negotiated_eid, struct smc_init_info *ini)
1223 {
1224 struct smc_clc_msg_accept_confirm aclc;
1225 int len;
1226
1227 memset(&aclc, 0, sizeof(aclc));
1228 aclc.hdr.type = SMC_CLC_ACCEPT;
1229 len = smc_clc_send_confirm_accept(new_smc, &aclc, srv_first_contact,
1230 version, negotiated_eid, ini);
1231 if (len < ntohs(aclc.hdr.length))
1232 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
1233
1234 return len > 0 ? 0 : len;
1235 }
1236
smc_clc_srv_v2x_features_validate(struct smc_sock * smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)1237 int smc_clc_srv_v2x_features_validate(struct smc_sock *smc,
1238 struct smc_clc_msg_proposal *pclc,
1239 struct smc_init_info *ini)
1240 {
1241 struct smc_clc_v2_extension *pclc_v2_ext;
1242 struct net *net = sock_net(&smc->sk);
1243
1244 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1245 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1246 ini->feature_mask = SMC_FEATURE_MASK;
1247
1248 if ((!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) ||
1249 ini->release_nr < SMC_RELEASE_1)
1250 return 0;
1251
1252 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1253 if (!pclc_v2_ext)
1254 return SMC_CLC_DECL_NOV2EXT;
1255
1256 if (ini->smcr_version & SMC_V2) {
1257 ini->max_conns = min_t(u8, pclc_v2_ext->max_conns,
1258 net->smc.sysctl_max_conns_per_lgr);
1259 if (ini->max_conns < SMC_CONN_PER_LGR_MIN)
1260 return SMC_CLC_DECL_MAXCONNERR;
1261
1262 ini->max_links = min_t(u8, pclc_v2_ext->max_links,
1263 net->smc.sysctl_max_links_per_lgr);
1264 if (ini->max_links < SMC_LINKS_ADD_LNK_MIN)
1265 return SMC_CLC_DECL_MAXLINKERR;
1266 }
1267
1268 return 0;
1269 }
1270
smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext * fce,struct smc_init_info * ini)1271 int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
1272 struct smc_init_info *ini)
1273 {
1274 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1275 (struct smc_clc_first_contact_ext_v2x *)fce;
1276
1277 if (ini->release_nr < SMC_RELEASE_1)
1278 return 0;
1279
1280 if (!ini->is_smcd) {
1281 if (fce_v2x->max_conns < SMC_CONN_PER_LGR_MIN)
1282 return SMC_CLC_DECL_MAXCONNERR;
1283 ini->max_conns = fce_v2x->max_conns;
1284
1285 if (fce_v2x->max_links > SMC_LINKS_ADD_LNK_MAX ||
1286 fce_v2x->max_links < SMC_LINKS_ADD_LNK_MIN)
1287 return SMC_CLC_DECL_MAXLINKERR;
1288 ini->max_links = fce_v2x->max_links;
1289 }
1290 /* common supplemental features of server and client */
1291 ini->feature_mask = ntohs(fce_v2x->feature_mask) & SMC_FEATURE_MASK;
1292
1293 return 0;
1294 }
1295
smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm * cclc,struct smc_init_info * ini)1296 int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
1297 struct smc_init_info *ini)
1298 {
1299 struct smc_clc_first_contact_ext *fce =
1300 smc_get_clc_first_contact_ext(cclc, ini->is_smcd);
1301 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1302 (struct smc_clc_first_contact_ext_v2x *)fce;
1303
1304 if (cclc->hdr.version == SMC_V1 ||
1305 !(cclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
1306 return 0;
1307
1308 if (ini->release_nr != fce->release)
1309 return SMC_CLC_DECL_RELEASEERR;
1310
1311 if (fce->release < SMC_RELEASE_1)
1312 return 0;
1313
1314 if (!ini->is_smcd) {
1315 if (fce_v2x->max_conns != ini->max_conns)
1316 return SMC_CLC_DECL_MAXCONNERR;
1317 if (fce_v2x->max_links != ini->max_links)
1318 return SMC_CLC_DECL_MAXLINKERR;
1319 }
1320 /* common supplemental features returned by client */
1321 ini->feature_mask = ntohs(fce_v2x->feature_mask);
1322
1323 return 0;
1324 }
1325
smc_clc_get_hostname(u8 ** host)1326 void smc_clc_get_hostname(u8 **host)
1327 {
1328 *host = &smc_hostname[0];
1329 }
1330
smc_clc_init(void)1331 void __init smc_clc_init(void)
1332 {
1333 struct new_utsname *u;
1334
1335 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
1336 u = utsname();
1337 memcpy(smc_hostname, u->nodename,
1338 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
1339
1340 INIT_LIST_HEAD(&smc_clc_eid_table.list);
1341 rwlock_init(&smc_clc_eid_table.lock);
1342 smc_clc_eid_table.ueid_cnt = 0;
1343 #if IS_ENABLED(CONFIG_S390)
1344 smc_clc_eid_table.seid_enabled = 1;
1345 #else
1346 smc_clc_eid_table.seid_enabled = 0;
1347 #endif
1348 }
1349
smc_clc_exit(void)1350 void smc_clc_exit(void)
1351 {
1352 smc_clc_ueid_remove(NULL);
1353 }
1354