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 pclc_prfx = smc_clc_proposal_get_prefix(pclc);
356 if (!pclc_prfx ||
357 pclc_prfx->ipv6_prefixes_cnt > SMC_CLC_MAX_V6_PREFIX)
358 return false;
359
360 if (hdr->version == SMC_V1) {
361 if (hdr->typev1 == SMC_TYPE_N)
362 return false;
363 if (ntohs(hdr->length) !=
364 sizeof(*pclc) + ntohs(pclc->iparea_offset) +
365 sizeof(*pclc_prfx) +
366 pclc_prfx->ipv6_prefixes_cnt *
367 sizeof(struct smc_clc_ipv6_prefix) +
368 sizeof(struct smc_clc_msg_trail))
369 return false;
370 } else {
371 v2_ext = smc_get_clc_v2_ext(pclc);
372 if ((hdr->typev2 != SMC_TYPE_N &&
373 (!v2_ext || v2_ext->hdr.eid_cnt > SMC_CLC_MAX_UEID)) ||
374 (smcd_indicated(hdr->typev2) &&
375 v2_ext->hdr.ism_gid_cnt > SMCD_CLC_MAX_V2_GID_ENTRIES))
376 return false;
377
378 if (ntohs(hdr->length) !=
379 sizeof(*pclc) +
380 sizeof(struct smc_clc_msg_smcd) +
381 (hdr->typev1 != SMC_TYPE_N ?
382 sizeof(*pclc_prfx) +
383 pclc_prfx->ipv6_prefixes_cnt *
384 sizeof(struct smc_clc_ipv6_prefix) : 0) +
385 (hdr->typev2 != SMC_TYPE_N ?
386 sizeof(*v2_ext) +
387 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN : 0) +
388 (smcd_indicated(hdr->typev2) ?
389 sizeof(*smcd_v2_ext) + v2_ext->hdr.ism_gid_cnt *
390 sizeof(struct smc_clc_smcd_gid_chid) :
391 0) +
392 sizeof(struct smc_clc_msg_trail))
393 return false;
394 }
395 return true;
396 }
397
398 /* check arriving CLC accept or confirm */
399 static bool
smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm * clc)400 smc_clc_msg_acc_conf_valid(struct smc_clc_msg_accept_confirm *clc)
401 {
402 struct smc_clc_msg_hdr *hdr = &clc->hdr;
403
404 if (hdr->typev1 != SMC_TYPE_R && hdr->typev1 != SMC_TYPE_D)
405 return false;
406 if (hdr->version == SMC_V1) {
407 if ((hdr->typev1 == SMC_TYPE_R &&
408 ntohs(hdr->length) != SMCR_CLC_ACCEPT_CONFIRM_LEN) ||
409 (hdr->typev1 == SMC_TYPE_D &&
410 ntohs(hdr->length) != SMCD_CLC_ACCEPT_CONFIRM_LEN))
411 return false;
412 } else {
413 if (hdr->typev1 == SMC_TYPE_D &&
414 ntohs(hdr->length) < SMCD_CLC_ACCEPT_CONFIRM_LEN_V2)
415 return false;
416 if (hdr->typev1 == SMC_TYPE_R &&
417 ntohs(hdr->length) < SMCR_CLC_ACCEPT_CONFIRM_LEN_V2)
418 return false;
419 }
420 return true;
421 }
422
423 /* check arriving CLC decline */
424 static bool
smc_clc_msg_decl_valid(struct smc_clc_msg_decline * dclc)425 smc_clc_msg_decl_valid(struct smc_clc_msg_decline *dclc)
426 {
427 struct smc_clc_msg_hdr *hdr = &dclc->hdr;
428
429 if (hdr->version == SMC_V1) {
430 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline))
431 return false;
432 } else {
433 if (ntohs(hdr->length) != sizeof(struct smc_clc_msg_decline_v2))
434 return false;
435 }
436 return true;
437 }
438
smc_clc_fill_fce_v2x(struct smc_clc_first_contact_ext_v2x * fce_v2x,struct smc_init_info * ini)439 static int smc_clc_fill_fce_v2x(struct smc_clc_first_contact_ext_v2x *fce_v2x,
440 struct smc_init_info *ini)
441 {
442 int ret = sizeof(*fce_v2x);
443
444 memset(fce_v2x, 0, sizeof(*fce_v2x));
445 fce_v2x->fce_v2_base.os_type = SMC_CLC_OS_LINUX;
446 fce_v2x->fce_v2_base.release = ini->release_nr;
447 memcpy(fce_v2x->fce_v2_base.hostname,
448 smc_hostname, sizeof(smc_hostname));
449 if (ini->is_smcd && ini->release_nr < SMC_RELEASE_1) {
450 ret = sizeof(struct smc_clc_first_contact_ext);
451 goto out;
452 }
453
454 if (ini->release_nr >= SMC_RELEASE_1) {
455 if (!ini->is_smcd) {
456 fce_v2x->max_conns = ini->max_conns;
457 fce_v2x->max_links = ini->max_links;
458 }
459 fce_v2x->feature_mask = htons(ini->feature_mask);
460 }
461
462 out:
463 return ret;
464 }
465
466 /* check if received message has a correct header length and contains valid
467 * heading and trailing eyecatchers
468 */
smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr * clcm,bool check_trl)469 static bool smc_clc_msg_hdr_valid(struct smc_clc_msg_hdr *clcm, bool check_trl)
470 {
471 struct smc_clc_msg_accept_confirm *clc;
472 struct smc_clc_msg_proposal *pclc;
473 struct smc_clc_msg_decline *dclc;
474 struct smc_clc_msg_trail *trl;
475
476 if (memcmp(clcm->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
477 memcmp(clcm->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
478 return false;
479 switch (clcm->type) {
480 case SMC_CLC_PROPOSAL:
481 pclc = (struct smc_clc_msg_proposal *)clcm;
482 if (!smc_clc_msg_prop_valid(pclc))
483 return false;
484 trl = (struct smc_clc_msg_trail *)
485 ((u8 *)pclc + ntohs(pclc->hdr.length) - sizeof(*trl));
486 break;
487 case SMC_CLC_ACCEPT:
488 case SMC_CLC_CONFIRM:
489 clc = (struct smc_clc_msg_accept_confirm *)clcm;
490 if (!smc_clc_msg_acc_conf_valid(clc))
491 return false;
492 trl = (struct smc_clc_msg_trail *)
493 ((u8 *)clc + ntohs(clc->hdr.length) - sizeof(*trl));
494 break;
495 case SMC_CLC_DECLINE:
496 dclc = (struct smc_clc_msg_decline *)clcm;
497 if (!smc_clc_msg_decl_valid(dclc))
498 return false;
499 check_trl = false;
500 break;
501 default:
502 return false;
503 }
504 if (check_trl &&
505 memcmp(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER)) &&
506 memcmp(trl->eyecatcher, SMCD_EYECATCHER, sizeof(SMCD_EYECATCHER)))
507 return false;
508 return true;
509 }
510
511 /* find ipv4 addr on device and get the prefix len, fill CLC proposal msg */
smc_clc_prfx_set4_rcu(struct net_device * dev,__be32 ipv4,struct smc_clc_msg_proposal_prefix * prop)512 static int smc_clc_prfx_set4_rcu(struct net_device *dev, __be32 ipv4,
513 struct smc_clc_msg_proposal_prefix *prop)
514 {
515 struct in_device *in_dev = __in_dev_get_rcu(dev);
516 const struct in_ifaddr *ifa;
517
518 if (!in_dev)
519 return -ENODEV;
520
521 in_dev_for_each_ifa_rcu(ifa, in_dev) {
522 if (!inet_ifa_match(ipv4, ifa))
523 continue;
524 prop->prefix_len = inet_mask_len(ifa->ifa_mask);
525 prop->outgoing_subnet = ifa->ifa_address & ifa->ifa_mask;
526 /* prop->ipv6_prefixes_cnt = 0; already done by memset before */
527 return 0;
528 }
529 return -ENOENT;
530 }
531
532 /* fill CLC proposal msg with ipv6 prefixes from device */
smc_clc_prfx_set6_rcu(struct net_device * dev,struct smc_clc_msg_proposal_prefix * prop,struct smc_clc_ipv6_prefix * ipv6_prfx)533 static int smc_clc_prfx_set6_rcu(struct net_device *dev,
534 struct smc_clc_msg_proposal_prefix *prop,
535 struct smc_clc_ipv6_prefix *ipv6_prfx)
536 {
537 #if IS_ENABLED(CONFIG_IPV6)
538 struct inet6_dev *in6_dev = __in6_dev_get(dev);
539 struct inet6_ifaddr *ifa;
540 int cnt = 0;
541
542 if (!in6_dev)
543 return -ENODEV;
544 /* use a maximum of 8 IPv6 prefixes from device */
545 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
546 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
547 continue;
548 ipv6_addr_prefix(&ipv6_prfx[cnt].prefix,
549 &ifa->addr, ifa->prefix_len);
550 ipv6_prfx[cnt].prefix_len = ifa->prefix_len;
551 cnt++;
552 if (cnt == SMC_CLC_MAX_V6_PREFIX)
553 break;
554 }
555 prop->ipv6_prefixes_cnt = cnt;
556 if (cnt)
557 return 0;
558 #endif
559 return -ENOENT;
560 }
561
562 /* 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)563 static int smc_clc_prfx_set(struct socket *clcsock,
564 struct smc_clc_msg_proposal_prefix *prop,
565 struct smc_clc_ipv6_prefix *ipv6_prfx)
566 {
567 struct sockaddr_storage addrs;
568 struct sockaddr_in6 *addr6;
569 struct sockaddr_in *addr;
570 struct net_device *dev;
571 struct dst_entry *dst;
572 int rc = -ENOENT;
573
574 /* get address to which the internal TCP socket is bound */
575 if (kernel_getsockname(clcsock, (struct sockaddr *)&addrs) < 0)
576 goto out;
577
578 /* analyze IP specific data of net_device belonging to TCP socket */
579 addr6 = (struct sockaddr_in6 *)&addrs;
580
581 rcu_read_lock();
582
583 dst = __sk_dst_get(clcsock->sk);
584 dev = dst ? dst_dev_rcu(dst) : NULL;
585 if (!dev) {
586 rc = -ENODEV;
587 goto out_unlock;
588 }
589
590 if (addrs.ss_family == PF_INET) {
591 /* IPv4 */
592 addr = (struct sockaddr_in *)&addrs;
593 rc = smc_clc_prfx_set4_rcu(dev, addr->sin_addr.s_addr, prop);
594 } else if (ipv6_addr_v4mapped(&addr6->sin6_addr)) {
595 /* mapped IPv4 address - peer is IPv4 only */
596 rc = smc_clc_prfx_set4_rcu(dev, addr6->sin6_addr.s6_addr32[3],
597 prop);
598 } else {
599 /* IPv6 */
600 rc = smc_clc_prfx_set6_rcu(dev, prop, ipv6_prfx);
601 }
602
603 out_unlock:
604 rcu_read_unlock();
605 out:
606 return rc;
607 }
608
609 /* 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)610 static int smc_clc_prfx_match4_rcu(struct net_device *dev,
611 struct smc_clc_msg_proposal_prefix *prop)
612 {
613 struct in_device *in_dev = __in_dev_get_rcu(dev);
614 const struct in_ifaddr *ifa;
615
616 if (!in_dev)
617 return -ENODEV;
618 in_dev_for_each_ifa_rcu(ifa, in_dev) {
619 if (prop->prefix_len == inet_mask_len(ifa->ifa_mask) &&
620 inet_ifa_match(prop->outgoing_subnet, ifa))
621 return 0;
622 }
623
624 return -ENOENT;
625 }
626
627 /* 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)628 static int smc_clc_prfx_match6_rcu(struct net_device *dev,
629 struct smc_clc_msg_proposal_prefix *prop)
630 {
631 #if IS_ENABLED(CONFIG_IPV6)
632 struct inet6_dev *in6_dev = __in6_dev_get(dev);
633 struct smc_clc_ipv6_prefix *ipv6_prfx;
634 struct inet6_ifaddr *ifa;
635 int i, max;
636
637 if (!in6_dev)
638 return -ENODEV;
639 /* ipv6 prefix list starts behind smc_clc_msg_proposal_prefix */
640 ipv6_prfx = (struct smc_clc_ipv6_prefix *)((u8 *)prop + sizeof(*prop));
641 max = min_t(u8, prop->ipv6_prefixes_cnt, SMC_CLC_MAX_V6_PREFIX);
642 list_for_each_entry(ifa, &in6_dev->addr_list, if_list) {
643 if (ipv6_addr_type(&ifa->addr) & IPV6_ADDR_LINKLOCAL)
644 continue;
645 for (i = 0; i < max; i++) {
646 if (ifa->prefix_len == ipv6_prfx[i].prefix_len &&
647 ipv6_prefix_equal(&ifa->addr, &ipv6_prfx[i].prefix,
648 ifa->prefix_len))
649 return 0;
650 }
651 }
652 #endif
653 return -ENOENT;
654 }
655
656 /* check if proposed prefixes match one of our device prefixes */
smc_clc_prfx_match(struct socket * clcsock,struct smc_clc_msg_proposal_prefix * prop)657 int smc_clc_prfx_match(struct socket *clcsock,
658 struct smc_clc_msg_proposal_prefix *prop)
659 {
660 struct net_device *dev;
661 struct dst_entry *dst;
662 int rc;
663
664 rcu_read_lock();
665
666 dst = __sk_dst_get(clcsock->sk);
667 dev = dst ? dst_dev_rcu(dst) : NULL;
668 if (!dev) {
669 rc = -ENODEV;
670 goto out;
671 }
672
673 if (!prop->ipv6_prefixes_cnt)
674 rc = smc_clc_prfx_match4_rcu(dev, prop);
675 else
676 rc = smc_clc_prfx_match6_rcu(dev, prop);
677 out:
678 rcu_read_unlock();
679
680 return rc;
681 }
682
683 /* Wait for data on the tcp-socket, analyze received data
684 * Returns:
685 * 0 if success and it was not a decline that we received.
686 * SMC_CLC_DECL_REPLY if decline received for fallback w/o another decl send.
687 * clcsock error, -EINTR, -ECONNRESET, -EPROTO otherwise.
688 */
smc_clc_wait_msg(struct smc_sock * smc,void * buf,int buflen,u8 expected_type,unsigned long timeout)689 int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
690 u8 expected_type, unsigned long timeout)
691 {
692 long rcvtimeo = READ_ONCE(smc->clcsock->sk->sk_rcvtimeo);
693 struct sock *clc_sk = smc->clcsock->sk;
694 struct smc_clc_msg_hdr *clcm = buf;
695 struct msghdr msg = {NULL, 0};
696 int reason_code = 0;
697 struct kvec vec = {buf, buflen};
698 int len, datlen, recvlen;
699 bool check_trl = true;
700 int krflags;
701
702 /* peek the first few bytes to determine length of data to receive
703 * so we don't consume any subsequent CLC message or payload data
704 * in the TCP byte stream
705 */
706 /*
707 * Caller must make sure that buflen is no less than
708 * sizeof(struct smc_clc_msg_hdr)
709 */
710 krflags = MSG_PEEK | MSG_WAITALL;
711 WRITE_ONCE(clc_sk->sk_rcvtimeo, timeout);
712 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1,
713 sizeof(struct smc_clc_msg_hdr));
714 len = sock_recvmsg(smc->clcsock, &msg, krflags);
715 if (signal_pending(current)) {
716 reason_code = -EINTR;
717 clc_sk->sk_err = EINTR;
718 smc->sk.sk_err = EINTR;
719 goto out;
720 }
721 if (clc_sk->sk_err) {
722 reason_code = -clc_sk->sk_err;
723 if (clc_sk->sk_err == EAGAIN &&
724 expected_type == SMC_CLC_DECLINE)
725 clc_sk->sk_err = 0; /* reset for fallback usage */
726 else
727 smc->sk.sk_err = clc_sk->sk_err;
728 goto out;
729 }
730 if (!len) { /* peer has performed orderly shutdown */
731 smc->sk.sk_err = ECONNRESET;
732 reason_code = -ECONNRESET;
733 goto out;
734 }
735 if (len < 0) {
736 if (len != -EAGAIN || expected_type != SMC_CLC_DECLINE)
737 smc->sk.sk_err = -len;
738 reason_code = len;
739 goto out;
740 }
741 datlen = ntohs(clcm->length);
742 if ((len < sizeof(struct smc_clc_msg_hdr)) ||
743 (clcm->version < SMC_V1) ||
744 ((clcm->type != SMC_CLC_DECLINE) &&
745 (clcm->type != expected_type))) {
746 smc->sk.sk_err = EPROTO;
747 reason_code = -EPROTO;
748 goto out;
749 }
750
751 /* receive the complete CLC message */
752 memset(&msg, 0, sizeof(struct msghdr));
753 if (datlen > buflen) {
754 check_trl = false;
755 recvlen = buflen;
756 } else {
757 recvlen = datlen;
758 }
759 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
760 krflags = MSG_WAITALL;
761 len = sock_recvmsg(smc->clcsock, &msg, krflags);
762 if (len < recvlen || !smc_clc_msg_hdr_valid(clcm, check_trl)) {
763 smc->sk.sk_err = EPROTO;
764 reason_code = -EPROTO;
765 goto out;
766 }
767 datlen -= len;
768 while (datlen) {
769 u8 tmp[SMC_CLC_RECV_BUF_LEN];
770
771 vec.iov_base = &tmp;
772 vec.iov_len = SMC_CLC_RECV_BUF_LEN;
773 /* receive remaining proposal message */
774 recvlen = datlen > SMC_CLC_RECV_BUF_LEN ?
775 SMC_CLC_RECV_BUF_LEN : datlen;
776 iov_iter_kvec(&msg.msg_iter, ITER_DEST, &vec, 1, recvlen);
777 len = sock_recvmsg(smc->clcsock, &msg, krflags);
778 if (len < recvlen) {
779 smc->sk.sk_err = EPROTO;
780 reason_code = -EPROTO;
781 goto out;
782 }
783 datlen -= len;
784 }
785 if (clcm->type == SMC_CLC_DECLINE) {
786 struct smc_clc_msg_decline *dclc;
787
788 dclc = (struct smc_clc_msg_decline *)clcm;
789 reason_code = SMC_CLC_DECL_PEERDECL;
790 smc->peer_diagnosis = ntohl(dclc->peer_diagnosis);
791 if (((struct smc_clc_msg_decline *)buf)->hdr.typev2 &
792 SMC_FIRST_CONTACT_MASK) {
793 smc->conn.lgr->sync_err = 1;
794 smc_lgr_terminate_sched(smc->conn.lgr);
795 }
796 }
797
798 out:
799 WRITE_ONCE(clc_sk->sk_rcvtimeo, rcvtimeo);
800 return reason_code;
801 }
802
803 /* send CLC DECLINE message across internal TCP socket */
smc_clc_send_decline(struct smc_sock * smc,u32 peer_diag_info,u8 version)804 int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version)
805 {
806 struct smc_clc_msg_decline *dclc_v1;
807 struct smc_clc_msg_decline_v2 dclc;
808 struct msghdr msg;
809 int len, send_len;
810 struct kvec vec;
811
812 dclc_v1 = (struct smc_clc_msg_decline *)&dclc;
813 memset(&dclc, 0, sizeof(dclc));
814 memcpy(dclc.hdr.eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
815 dclc.hdr.type = SMC_CLC_DECLINE;
816 dclc.hdr.version = version;
817 dclc.os_type = version == SMC_V1 ? 0 : SMC_CLC_OS_LINUX;
818 dclc.hdr.typev2 = (peer_diag_info == SMC_CLC_DECL_SYNCERR) ?
819 SMC_FIRST_CONTACT_MASK : 0;
820 if ((!smc_conn_lgr_valid(&smc->conn) || !smc->conn.lgr->is_smcd) &&
821 smc_ib_is_valid_local_systemid())
822 memcpy(dclc.id_for_peer, local_systemid,
823 sizeof(local_systemid));
824 dclc.peer_diagnosis = htonl(peer_diag_info);
825 if (version == SMC_V1) {
826 memcpy(dclc_v1->trl.eyecatcher, SMC_EYECATCHER,
827 sizeof(SMC_EYECATCHER));
828 send_len = sizeof(*dclc_v1);
829 } else {
830 memcpy(dclc.trl.eyecatcher, SMC_EYECATCHER,
831 sizeof(SMC_EYECATCHER));
832 send_len = sizeof(dclc);
833 }
834 dclc.hdr.length = htons(send_len);
835
836 memset(&msg, 0, sizeof(msg));
837 vec.iov_base = &dclc;
838 vec.iov_len = send_len;
839 len = kernel_sendmsg(smc->clcsock, &msg, &vec, 1, send_len);
840 if (len < 0 || len < send_len)
841 len = -EPROTO;
842 return len > 0 ? 0 : len;
843 }
844
845 /* send CLC PROPOSAL message across internal TCP socket */
smc_clc_send_proposal(struct smc_sock * smc,struct smc_init_info * ini)846 int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini)
847 {
848 struct smc_clc_smcd_v2_extension *smcd_v2_ext;
849 struct smc_clc_msg_proposal_prefix *pclc_prfx;
850 struct smc_clc_msg_proposal *pclc_base;
851 struct smc_clc_smcd_gid_chid *gidchids;
852 struct smc_clc_msg_proposal_area *pclc;
853 struct smc_clc_ipv6_prefix *ipv6_prfx;
854 struct net *net = sock_net(&smc->sk);
855 struct smc_clc_v2_extension *v2_ext;
856 struct smc_clc_msg_smcd *pclc_smcd;
857 struct smc_clc_msg_trail *trl;
858 struct smcd_dev *smcd;
859 int len, i, plen, rc;
860 int reason_code = 0;
861 struct kvec vec[8];
862 struct msghdr msg;
863
864 pclc = kzalloc(sizeof(*pclc), GFP_KERNEL);
865 if (!pclc)
866 return -ENOMEM;
867
868 pclc_base = &pclc->pclc_base;
869 pclc_smcd = &pclc->pclc_smcd;
870 pclc_prfx = &pclc->pclc_prfx;
871 ipv6_prfx = pclc->pclc_prfx_ipv6;
872 v2_ext = container_of(&pclc->pclc_v2_ext,
873 struct smc_clc_v2_extension, fixed);
874 smcd_v2_ext = container_of(&pclc->pclc_smcd_v2_ext,
875 struct smc_clc_smcd_v2_extension, fixed);
876 gidchids = pclc->pclc_gidchids;
877 trl = &pclc->pclc_trl;
878
879 pclc_base->hdr.version = SMC_V2;
880 pclc_base->hdr.typev1 = ini->smc_type_v1;
881 pclc_base->hdr.typev2 = ini->smc_type_v2;
882 plen = sizeof(*pclc_base) + sizeof(*pclc_smcd) + sizeof(*trl);
883
884 /* retrieve ip prefixes for CLC proposal msg */
885 if (ini->smc_type_v1 != SMC_TYPE_N) {
886 rc = smc_clc_prfx_set(smc->clcsock, pclc_prfx, ipv6_prfx);
887 if (rc) {
888 if (ini->smc_type_v2 == SMC_TYPE_N) {
889 kfree(pclc);
890 return SMC_CLC_DECL_CNFERR;
891 }
892 pclc_base->hdr.typev1 = SMC_TYPE_N;
893 ini->smc_type_v1 = SMC_TYPE_N;
894 } else {
895 pclc_base->iparea_offset = htons(sizeof(*pclc_smcd));
896 plen += sizeof(*pclc_prfx) +
897 pclc_prfx->ipv6_prefixes_cnt *
898 sizeof(ipv6_prfx[0]);
899 }
900 }
901
902 /* build SMC Proposal CLC message */
903 memcpy(pclc_base->hdr.eyecatcher, SMC_EYECATCHER,
904 sizeof(SMC_EYECATCHER));
905 pclc_base->hdr.type = SMC_CLC_PROPOSAL;
906 if (smcr_indicated(ini->smc_type_v1)) {
907 /* add SMC-R specifics */
908 memcpy(pclc_base->lcl.id_for_peer, local_systemid,
909 sizeof(local_systemid));
910 memcpy(pclc_base->lcl.gid, ini->ib_gid, SMC_GID_SIZE);
911 memcpy(pclc_base->lcl.mac, &ini->ib_dev->mac[ini->ib_port - 1],
912 ETH_ALEN);
913 }
914 if (smcd_indicated(ini->smc_type_v1)) {
915 struct smcd_gid smcd_gid;
916
917 /* add SMC-D specifics */
918 if (ini->ism_dev[0]) {
919 smcd = ini->ism_dev[0];
920 copy_to_smcdgid(&smcd_gid, &smcd->dibs->gid);
921 pclc_smcd->ism.gid = htonll(smcd_gid.gid);
922 pclc_smcd->ism.chid =
923 htons(smc_ism_get_chid(ini->ism_dev[0]));
924 }
925 }
926 if (ini->smc_type_v2 == SMC_TYPE_N) {
927 pclc_smcd->v2_ext_offset = 0;
928 } else {
929 struct smc_clc_eid_entry *ueident;
930 u16 v2_ext_offset;
931
932 v2_ext->hdr.flag.release = SMC_RELEASE;
933 v2_ext_offset = sizeof(*pclc_smcd) -
934 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
935 if (ini->smc_type_v1 != SMC_TYPE_N)
936 v2_ext_offset += sizeof(*pclc_prfx) +
937 pclc_prfx->ipv6_prefixes_cnt *
938 sizeof(ipv6_prfx[0]);
939 pclc_smcd->v2_ext_offset = htons(v2_ext_offset);
940 plen += sizeof(*v2_ext);
941
942 v2_ext->feature_mask = htons(SMC_FEATURE_MASK);
943 read_lock(&smc_clc_eid_table.lock);
944 v2_ext->hdr.eid_cnt = smc_clc_eid_table.ueid_cnt;
945 plen += smc_clc_eid_table.ueid_cnt * SMC_MAX_EID_LEN;
946 i = 0;
947 list_for_each_entry(ueident, &smc_clc_eid_table.list, list) {
948 memcpy(v2_ext->user_eids[i++], ueident->eid,
949 sizeof(ueident->eid));
950 }
951 read_unlock(&smc_clc_eid_table.lock);
952 }
953 if (smcd_indicated(ini->smc_type_v2)) {
954 struct smcd_gid smcd_gid;
955 u8 *eid = NULL;
956 int entry = 0;
957
958 v2_ext->hdr.flag.seid = smc_clc_eid_table.seid_enabled;
959 v2_ext->hdr.smcd_v2_ext_offset = htons(sizeof(*v2_ext) -
960 offsetofend(struct smc_clnt_opts_area_hdr,
961 smcd_v2_ext_offset) +
962 v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
963 smc_ism_get_system_eid(&eid);
964 if (eid && v2_ext->hdr.flag.seid)
965 memcpy(smcd_v2_ext->system_eid, eid, SMC_MAX_EID_LEN);
966 plen += sizeof(*smcd_v2_ext);
967 if (ini->ism_offered_cnt) {
968 for (i = 1; i <= ini->ism_offered_cnt; i++) {
969 smcd = ini->ism_dev[i];
970 copy_to_smcdgid(&smcd_gid, &smcd->dibs->gid);
971 gidchids[entry].chid =
972 htons(smc_ism_get_chid(ini->ism_dev[i]));
973 gidchids[entry].gid = htonll(smcd_gid.gid);
974 if (smc_ism_is_emulated(smcd)) {
975 /* an Emulated-ISM device takes two
976 * entries. CHID of the second entry
977 * repeats that of the first entry.
978 */
979 gidchids[entry + 1].chid =
980 gidchids[entry].chid;
981 gidchids[entry + 1].gid =
982 htonll(smcd_gid.gid_ext);
983 entry++;
984 }
985 entry++;
986 }
987 plen += entry * sizeof(struct smc_clc_smcd_gid_chid);
988 }
989 v2_ext->hdr.ism_gid_cnt = entry;
990 }
991 if (smcr_indicated(ini->smc_type_v2)) {
992 memcpy(v2_ext->roce, ini->smcrv2.ib_gid_v2, SMC_GID_SIZE);
993 v2_ext->max_conns = net->smc.sysctl_max_conns_per_lgr;
994 v2_ext->max_links = net->smc.sysctl_max_links_per_lgr;
995 }
996
997 pclc_base->hdr.length = htons(plen);
998 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
999
1000 /* send SMC Proposal CLC message */
1001 memset(&msg, 0, sizeof(msg));
1002 i = 0;
1003 vec[i].iov_base = pclc_base;
1004 vec[i++].iov_len = sizeof(*pclc_base);
1005 vec[i].iov_base = pclc_smcd;
1006 vec[i++].iov_len = sizeof(*pclc_smcd);
1007 if (ini->smc_type_v1 != SMC_TYPE_N) {
1008 vec[i].iov_base = pclc_prfx;
1009 vec[i++].iov_len = sizeof(*pclc_prfx);
1010 if (pclc_prfx->ipv6_prefixes_cnt > 0) {
1011 vec[i].iov_base = ipv6_prfx;
1012 vec[i++].iov_len = pclc_prfx->ipv6_prefixes_cnt *
1013 sizeof(ipv6_prfx[0]);
1014 }
1015 }
1016 if (ini->smc_type_v2 != SMC_TYPE_N) {
1017 vec[i].iov_base = v2_ext;
1018 vec[i++].iov_len = sizeof(*v2_ext) +
1019 (v2_ext->hdr.eid_cnt * SMC_MAX_EID_LEN);
1020 if (smcd_indicated(ini->smc_type_v2)) {
1021 vec[i].iov_base = smcd_v2_ext;
1022 vec[i++].iov_len = sizeof(*smcd_v2_ext);
1023 if (ini->ism_offered_cnt) {
1024 vec[i].iov_base = gidchids;
1025 vec[i++].iov_len = v2_ext->hdr.ism_gid_cnt *
1026 sizeof(struct smc_clc_smcd_gid_chid);
1027 }
1028 }
1029 }
1030 vec[i].iov_base = trl;
1031 vec[i++].iov_len = sizeof(*trl);
1032 /* due to the few bytes needed for clc-handshake this cannot block */
1033 len = kernel_sendmsg(smc->clcsock, &msg, vec, i, plen);
1034 if (len < 0) {
1035 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1036 reason_code = -smc->sk.sk_err;
1037 } else if (len < ntohs(pclc_base->hdr.length)) {
1038 reason_code = -ENETUNREACH;
1039 smc->sk.sk_err = -reason_code;
1040 }
1041
1042 kfree(pclc);
1043 return reason_code;
1044 }
1045
1046 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)1047 smcd_clc_prep_confirm_accept(struct smc_connection *conn,
1048 struct smc_clc_msg_accept_confirm *clc,
1049 int first_contact, u8 version,
1050 u8 *eid, struct smc_init_info *ini,
1051 int *fce_len,
1052 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1053 struct smc_clc_msg_trail *trl)
1054 {
1055 struct smcd_dev *smcd = conn->lgr->smcd;
1056 struct smcd_gid smcd_gid;
1057 u16 chid;
1058 int len;
1059
1060 /* SMC-D specific settings */
1061 memcpy(clc->hdr.eyecatcher, SMCD_EYECATCHER,
1062 sizeof(SMCD_EYECATCHER));
1063 copy_to_smcdgid(&smcd_gid, &smcd->dibs->gid);
1064 clc->hdr.typev1 = SMC_TYPE_D;
1065 clc->d0.gid = htonll(smcd_gid.gid);
1066 clc->d0.token = htonll(conn->rmb_desc->token);
1067 clc->d0.dmbe_size = conn->rmbe_size_comp;
1068 clc->d0.dmbe_idx = 0;
1069 memcpy(&clc->d0.linkid, conn->lgr->id, SMC_LGR_ID_SIZE);
1070 if (version == SMC_V1) {
1071 clc->hdr.length = htons(SMCD_CLC_ACCEPT_CONFIRM_LEN);
1072 } else {
1073 chid = smc_ism_get_chid(smcd);
1074 clc->d1.chid = htons(chid);
1075 if (eid && eid[0])
1076 memcpy(clc->d1.eid, eid, SMC_MAX_EID_LEN);
1077 if (__smc_ism_is_emulated(chid))
1078 clc->d1.gid_ext = htonll(smcd_gid.gid_ext);
1079 len = SMCD_CLC_ACCEPT_CONFIRM_LEN_V2;
1080 if (first_contact) {
1081 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1082 len += *fce_len;
1083 }
1084 clc->hdr.length = htons(len);
1085 }
1086 memcpy(trl->eyecatcher, SMCD_EYECATCHER,
1087 sizeof(SMCD_EYECATCHER));
1088 }
1089
1090 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)1091 smcr_clc_prep_confirm_accept(struct smc_connection *conn,
1092 struct smc_clc_msg_accept_confirm *clc,
1093 int first_contact, u8 version,
1094 u8 *eid, struct smc_init_info *ini,
1095 int *fce_len,
1096 struct smc_clc_first_contact_ext_v2x *fce_v2x,
1097 struct smc_clc_fce_gid_ext *gle,
1098 struct smc_clc_msg_trail *trl)
1099 {
1100 struct smc_link *link = conn->lnk;
1101 int len;
1102
1103 /* SMC-R specific settings */
1104 memcpy(clc->hdr.eyecatcher, SMC_EYECATCHER,
1105 sizeof(SMC_EYECATCHER));
1106 clc->hdr.typev1 = SMC_TYPE_R;
1107 memcpy(clc->r0.lcl.id_for_peer, local_systemid,
1108 sizeof(local_systemid));
1109 memcpy(&clc->r0.lcl.gid, link->gid, SMC_GID_SIZE);
1110 memcpy(&clc->r0.lcl.mac, &link->smcibdev->mac[link->ibport - 1],
1111 ETH_ALEN);
1112 hton24(clc->r0.qpn, link->roce_qp->qp_num);
1113 clc->r0.rmb_rkey =
1114 htonl(conn->rmb_desc->mr[link->link_idx]->rkey);
1115 clc->r0.rmbe_idx = 1; /* for now: 1 RMB = 1 RMBE */
1116 clc->r0.rmbe_alert_token = htonl(conn->alert_token_local);
1117 switch (clc->hdr.type) {
1118 case SMC_CLC_ACCEPT:
1119 clc->r0.qp_mtu = link->path_mtu;
1120 break;
1121 case SMC_CLC_CONFIRM:
1122 clc->r0.qp_mtu = min(link->path_mtu, link->peer_mtu);
1123 break;
1124 }
1125 clc->r0.rmbe_size = conn->rmbe_size_comp;
1126 clc->r0.rmb_dma_addr = conn->rmb_desc->is_vm ?
1127 cpu_to_be64((uintptr_t)conn->rmb_desc->cpu_addr) :
1128 cpu_to_be64((u64)sg_dma_address
1129 (conn->rmb_desc->sgt[link->link_idx].sgl));
1130 hton24(clc->r0.psn, link->psn_initial);
1131 if (version == SMC_V1) {
1132 clc->hdr.length = htons(SMCR_CLC_ACCEPT_CONFIRM_LEN);
1133 } else {
1134 if (eid && eid[0])
1135 memcpy(clc->r1.eid, eid, SMC_MAX_EID_LEN);
1136 len = SMCR_CLC_ACCEPT_CONFIRM_LEN_V2;
1137 if (first_contact) {
1138 *fce_len = smc_clc_fill_fce_v2x(fce_v2x, ini);
1139 len += *fce_len;
1140 fce_v2x->fce_v2_base.v2_direct =
1141 !link->lgr->uses_gateway;
1142 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1143 memset(gle, 0, sizeof(*gle));
1144 gle->gid_cnt = ini->smcrv2.gidlist.len;
1145 len += sizeof(*gle);
1146 len += gle->gid_cnt * sizeof(gle->gid[0]);
1147 }
1148 }
1149 clc->hdr.length = htons(len);
1150 }
1151 memcpy(trl->eyecatcher, SMC_EYECATCHER, sizeof(SMC_EYECATCHER));
1152 }
1153
1154 /* 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)1155 static int smc_clc_send_confirm_accept(struct smc_sock *smc,
1156 struct smc_clc_msg_accept_confirm *clc,
1157 int first_contact, u8 version,
1158 u8 *eid, struct smc_init_info *ini)
1159 {
1160 struct smc_clc_first_contact_ext_v2x fce_v2x;
1161 struct smc_connection *conn = &smc->conn;
1162 struct smc_clc_fce_gid_ext gle;
1163 struct smc_clc_msg_trail trl;
1164 int i, fce_len;
1165 struct kvec vec[5];
1166 struct msghdr msg;
1167
1168 /* send SMC Confirm CLC msg */
1169 clc->hdr.version = version; /* SMC version */
1170 if (first_contact)
1171 clc->hdr.typev2 |= SMC_FIRST_CONTACT_MASK;
1172 if (conn->lgr->is_smcd)
1173 smcd_clc_prep_confirm_accept(conn, clc, first_contact,
1174 version, eid, ini, &fce_len,
1175 &fce_v2x, &trl);
1176 else
1177 smcr_clc_prep_confirm_accept(conn, clc, first_contact,
1178 version, eid, ini, &fce_len,
1179 &fce_v2x, &gle, &trl);
1180 memset(&msg, 0, sizeof(msg));
1181 i = 0;
1182 vec[i].iov_base = clc;
1183 if (version > SMC_V1)
1184 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1185 SMCD_CLC_ACCEPT_CONFIRM_LEN_V2 :
1186 SMCR_CLC_ACCEPT_CONFIRM_LEN_V2) -
1187 sizeof(trl);
1188 else
1189 vec[i++].iov_len = (clc->hdr.typev1 == SMC_TYPE_D ?
1190 SMCD_CLC_ACCEPT_CONFIRM_LEN :
1191 SMCR_CLC_ACCEPT_CONFIRM_LEN) -
1192 sizeof(trl);
1193 if (version > SMC_V1 && first_contact) {
1194 vec[i].iov_base = &fce_v2x;
1195 vec[i++].iov_len = fce_len;
1196 if (!conn->lgr->is_smcd) {
1197 if (clc->hdr.type == SMC_CLC_CONFIRM) {
1198 vec[i].iov_base = &gle;
1199 vec[i++].iov_len = sizeof(gle);
1200 vec[i].iov_base = &ini->smcrv2.gidlist.list;
1201 vec[i++].iov_len = gle.gid_cnt *
1202 sizeof(gle.gid[0]);
1203 }
1204 }
1205 }
1206 vec[i].iov_base = &trl;
1207 vec[i++].iov_len = sizeof(trl);
1208 return kernel_sendmsg(smc->clcsock, &msg, vec, 1,
1209 ntohs(clc->hdr.length));
1210 }
1211
1212 /* 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)1213 int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
1214 u8 version, u8 *eid, struct smc_init_info *ini)
1215 {
1216 struct smc_clc_msg_accept_confirm cclc;
1217 int reason_code = 0;
1218 int len;
1219
1220 /* send SMC Confirm CLC msg */
1221 memset(&cclc, 0, sizeof(cclc));
1222 cclc.hdr.type = SMC_CLC_CONFIRM;
1223 len = smc_clc_send_confirm_accept(smc, &cclc, clnt_first_contact,
1224 version, eid, ini);
1225 if (len < ntohs(cclc.hdr.length)) {
1226 if (len >= 0) {
1227 reason_code = -ENETUNREACH;
1228 smc->sk.sk_err = -reason_code;
1229 } else {
1230 smc->sk.sk_err = smc->clcsock->sk->sk_err;
1231 reason_code = -smc->sk.sk_err;
1232 }
1233 }
1234 return reason_code;
1235 }
1236
1237 /* 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)1238 int smc_clc_send_accept(struct smc_sock *new_smc, bool srv_first_contact,
1239 u8 version, u8 *negotiated_eid, struct smc_init_info *ini)
1240 {
1241 struct smc_clc_msg_accept_confirm aclc;
1242 int len;
1243
1244 memset(&aclc, 0, sizeof(aclc));
1245 aclc.hdr.type = SMC_CLC_ACCEPT;
1246 len = smc_clc_send_confirm_accept(new_smc, &aclc, srv_first_contact,
1247 version, negotiated_eid, ini);
1248 if (len < ntohs(aclc.hdr.length))
1249 len = len >= 0 ? -EPROTO : -new_smc->clcsock->sk->sk_err;
1250
1251 return len > 0 ? 0 : len;
1252 }
1253
smc_clc_srv_v2x_features_validate(struct smc_sock * smc,struct smc_clc_msg_proposal * pclc,struct smc_init_info * ini)1254 int smc_clc_srv_v2x_features_validate(struct smc_sock *smc,
1255 struct smc_clc_msg_proposal *pclc,
1256 struct smc_init_info *ini)
1257 {
1258 struct smc_clc_v2_extension *pclc_v2_ext;
1259 struct net *net = sock_net(&smc->sk);
1260
1261 ini->max_conns = SMC_CONN_PER_LGR_MAX;
1262 ini->max_links = SMC_LINKS_ADD_LNK_MAX;
1263 ini->feature_mask = SMC_FEATURE_MASK;
1264
1265 if ((!(ini->smcd_version & SMC_V2) && !(ini->smcr_version & SMC_V2)) ||
1266 ini->release_nr < SMC_RELEASE_1)
1267 return 0;
1268
1269 pclc_v2_ext = smc_get_clc_v2_ext(pclc);
1270 if (!pclc_v2_ext)
1271 return SMC_CLC_DECL_NOV2EXT;
1272
1273 if (ini->smcr_version & SMC_V2) {
1274 ini->max_conns = min_t(u8, pclc_v2_ext->max_conns,
1275 net->smc.sysctl_max_conns_per_lgr);
1276 if (ini->max_conns < SMC_CONN_PER_LGR_MIN)
1277 return SMC_CLC_DECL_MAXCONNERR;
1278
1279 ini->max_links = min_t(u8, pclc_v2_ext->max_links,
1280 net->smc.sysctl_max_links_per_lgr);
1281 if (ini->max_links < SMC_LINKS_ADD_LNK_MIN)
1282 return SMC_CLC_DECL_MAXLINKERR;
1283 }
1284
1285 return 0;
1286 }
1287
smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext * fce,struct smc_init_info * ini)1288 int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
1289 struct smc_init_info *ini)
1290 {
1291 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1292 (struct smc_clc_first_contact_ext_v2x *)fce;
1293
1294 if (ini->release_nr < SMC_RELEASE_1)
1295 return 0;
1296
1297 if (!ini->is_smcd) {
1298 if (fce_v2x->max_conns < SMC_CONN_PER_LGR_MIN)
1299 return SMC_CLC_DECL_MAXCONNERR;
1300 ini->max_conns = fce_v2x->max_conns;
1301
1302 if (fce_v2x->max_links > SMC_LINKS_ADD_LNK_MAX ||
1303 fce_v2x->max_links < SMC_LINKS_ADD_LNK_MIN)
1304 return SMC_CLC_DECL_MAXLINKERR;
1305 ini->max_links = fce_v2x->max_links;
1306 }
1307 /* common supplemental features of server and client */
1308 ini->feature_mask = ntohs(fce_v2x->feature_mask) & SMC_FEATURE_MASK;
1309
1310 return 0;
1311 }
1312
smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm * cclc,struct smc_init_info * ini)1313 int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
1314 struct smc_init_info *ini)
1315 {
1316 struct smc_clc_first_contact_ext *fce =
1317 smc_get_clc_first_contact_ext(cclc, ini->is_smcd);
1318 struct smc_clc_first_contact_ext_v2x *fce_v2x =
1319 (struct smc_clc_first_contact_ext_v2x *)fce;
1320
1321 if (cclc->hdr.version == SMC_V1 ||
1322 !(cclc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
1323 return 0;
1324
1325 if (ini->release_nr != fce->release)
1326 return SMC_CLC_DECL_RELEASEERR;
1327
1328 if (fce->release < SMC_RELEASE_1)
1329 return 0;
1330
1331 if (!ini->is_smcd) {
1332 if (fce_v2x->max_conns != ini->max_conns)
1333 return SMC_CLC_DECL_MAXCONNERR;
1334 if (fce_v2x->max_links != ini->max_links)
1335 return SMC_CLC_DECL_MAXLINKERR;
1336 }
1337 /* common supplemental features returned by client */
1338 ini->feature_mask = ntohs(fce_v2x->feature_mask);
1339
1340 return 0;
1341 }
1342
smc_clc_get_hostname(u8 ** host)1343 void smc_clc_get_hostname(u8 **host)
1344 {
1345 *host = &smc_hostname[0];
1346 }
1347
smc_clc_init(void)1348 void __init smc_clc_init(void)
1349 {
1350 struct new_utsname *u;
1351
1352 memset(smc_hostname, _S, sizeof(smc_hostname)); /* ASCII blanks */
1353 u = utsname();
1354 memcpy(smc_hostname, u->nodename,
1355 min_t(size_t, strlen(u->nodename), sizeof(smc_hostname)));
1356
1357 INIT_LIST_HEAD(&smc_clc_eid_table.list);
1358 rwlock_init(&smc_clc_eid_table.lock);
1359 smc_clc_eid_table.ueid_cnt = 0;
1360 #if IS_ENABLED(CONFIG_S390)
1361 smc_clc_eid_table.seid_enabled = 1;
1362 #else
1363 smc_clc_eid_table.seid_enabled = 0;
1364 #endif
1365 }
1366
smc_clc_exit(void)1367 void smc_clc_exit(void)
1368 {
1369 smc_clc_ueid_remove(NULL);
1370 }
1371