1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
5 * CLC (connection layer control) handshake over initial TCP socket to
6 * prepare for RDMA traffic
8 * Copyright IBM Corp. 2016
16 #include <rdma/ib_verbs.h>
17 #include <linux/smc.h>
20 #include "smc_netlink.h"
22 #define SMC_CLC_PROPOSAL 0x01
23 #define SMC_CLC_ACCEPT 0x02
24 #define SMC_CLC_CONFIRM 0x03
25 #define SMC_CLC_DECLINE 0x04
27 #define SMC_TYPE_R 0 /* SMC-R only */
28 #define SMC_TYPE_D 1 /* SMC-D only */
29 #define SMC_TYPE_N 2 /* neither SMC-R nor SMC-D */
30 #define SMC_TYPE_B 3 /* SMC-R and SMC-D */
31 #define CLC_WAIT_TIME (6 * HZ) /* max. wait time on clcsock */
32 #define CLC_WAIT_TIME_SHORT HZ /* short wait time on clcsock */
33 #define SMC_CLC_DECL_MEM 0x01010000 /* insufficient memory resources */
34 #define SMC_CLC_DECL_TIMEOUT_CL 0x02010000 /* timeout w4 QP confirm link */
35 #define SMC_CLC_DECL_TIMEOUT_AL 0x02020000 /* timeout w4 QP add link */
36 #define SMC_CLC_DECL_CNFERR 0x03000000 /* configuration error */
37 #define SMC_CLC_DECL_PEERNOSMC 0x03010000 /* peer did not indicate SMC */
38 #define SMC_CLC_DECL_IPSEC 0x03020000 /* IPsec usage */
39 #define SMC_CLC_DECL_NOSMCDEV 0x03030000 /* no SMC device found (R or D) */
40 #define SMC_CLC_DECL_NOSMCDDEV 0x03030001 /* no SMC-D device found */
41 #define SMC_CLC_DECL_NOSMCRDEV 0x03030002 /* no SMC-R device found */
42 #define SMC_CLC_DECL_NOISM2SUPP 0x03030003 /* hardware has no ISMv2 support */
43 #define SMC_CLC_DECL_NOV2EXT 0x03030004 /* peer sent no clc v2 extension */
44 #define SMC_CLC_DECL_NOV2DEXT 0x03030005 /* peer sent no clc SMC-Dv2 ext. */
45 #define SMC_CLC_DECL_NOSEID 0x03030006 /* peer sent no SEID */
46 #define SMC_CLC_DECL_NOSMCD2DEV 0x03030007 /* no SMC-Dv2 device found */
47 #define SMC_CLC_DECL_NOUEID 0x03030008 /* peer sent no UEID */
48 #define SMC_CLC_DECL_RELEASEERR 0x03030009 /* release version negotiate failed */
49 #define SMC_CLC_DECL_MAXCONNERR 0x0303000a /* max connections negotiate failed */
50 #define SMC_CLC_DECL_MAXLINKERR 0x0303000b /* max links negotiate failed */
51 #define SMC_CLC_DECL_MODEUNSUPP 0x03040000 /* smc modes do not match (R or D)*/
52 #define SMC_CLC_DECL_RMBE_EC 0x03050000 /* peer has eyecatcher in RMBE */
53 #define SMC_CLC_DECL_OPTUNSUPP 0x03060000 /* fastopen sockopt not supported */
54 #define SMC_CLC_DECL_DIFFPREFIX 0x03070000 /* IP prefix / subnet mismatch */
55 #define SMC_CLC_DECL_GETVLANERR 0x03080000 /* err to get vlan id of ip device*/
56 #define SMC_CLC_DECL_ISMVLANERR 0x03090000 /* err to reg vlan id on ism dev */
57 #define SMC_CLC_DECL_NOACTLINK 0x030a0000 /* no active smc-r link in lgr */
58 #define SMC_CLC_DECL_NOSRVLINK 0x030b0000 /* SMC-R link from srv not found */
59 #define SMC_CLC_DECL_VERSMISMAT 0x030c0000 /* SMC version mismatch */
60 #define SMC_CLC_DECL_MAX_DMB 0x030d0000 /* SMC-D DMB limit exceeded */
61 #define SMC_CLC_DECL_NOROUTE 0x030e0000 /* SMC-Rv2 conn. no route to peer */
62 #define SMC_CLC_DECL_NOINDIRECT 0x030f0000 /* SMC-Rv2 conn. indirect mismatch*/
63 #define SMC_CLC_DECL_SYNCERR 0x04000000 /* synchronization error */
64 #define SMC_CLC_DECL_PEERDECL 0x05000000 /* peer declined during handshake */
65 #define SMC_CLC_DECL_INTERR 0x09990000 /* internal error */
66 #define SMC_CLC_DECL_ERR_RTOK 0x09990001 /* rtoken handling failed */
67 #define SMC_CLC_DECL_ERR_RDYLNK 0x09990002 /* ib ready link failed */
68 #define SMC_CLC_DECL_ERR_REGBUF 0x09990003 /* reg rdma bufs failed */
70 #define SMC_FIRST_CONTACT_MASK 0b10 /* first contact bit within typev2 */
72 struct smc_clc_msg_hdr { /* header1 of clc messages */
73 u8 eyecatcher[4]; /* eye catcher */
74 u8 type; /* proposal / accept / confirm / decline */
76 #if defined(__BIG_ENDIAN_BITFIELD)
80 #elif defined(__LITTLE_ENDIAN_BITFIELD)
85 } __packed; /* format defined in RFC7609 */
87 struct smc_clc_msg_trail { /* trailer of clc messages */
91 struct smc_clc_msg_local { /* header2 of clc messages */
92 u8 id_for_peer[SMC_SYSTEMID_LEN]; /* unique system id */
93 u8 gid[16]; /* gid of ib_device port */
94 u8 mac[6]; /* mac of ib_device port */
97 /* Struct would be 4 byte aligned, but it is used in an array that is sent
98 * to peers and must conform to RFC7609, hence we need to use packed here.
100 struct smc_clc_ipv6_prefix {
101 struct in6_addr prefix;
103 } __packed; /* format defined in RFC7609 */
105 #if defined(__BIG_ENDIAN_BITFIELD)
106 struct smc_clc_v2_flag {
111 #elif defined(__LITTLE_ENDIAN_BITFIELD)
112 struct smc_clc_v2_flag {
119 struct smc_clnt_opts_area_hdr {
120 u8 eid_cnt; /* number of user defined EIDs */
121 u8 ism_gid_cnt; /* number of ISMv2 GIDs */
123 struct smc_clc_v2_flag flag;
125 __be16 smcd_v2_ext_offset; /* SMC-Dv2 Extension Offset */
128 struct smc_clc_smcd_gid_chid {
129 __be64 gid; /* ISM GID */
130 __be16 chid; /* ISMv2 CHID */
131 } __packed; /* format defined in
132 * IBM Shared Memory Communications Version 2
133 * (https://www.ibm.com/support/pages/node/6326337)
136 struct smc_clc_v2_extension {
137 /* New members must be added within the struct_group() macro below. */
138 struct_group_tagged(smc_clc_v2_extension_fixed, fixed,
139 struct smc_clnt_opts_area_hdr hdr;
140 u8 roce[16]; /* RoCEv2 GID */
146 u8 user_eids[][SMC_MAX_EID_LEN];
148 static_assert(offsetof(struct smc_clc_v2_extension, user_eids) == sizeof(struct smc_clc_v2_extension_fixed),
149 "struct member likely outside of struct_group_tagged()");
151 struct smc_clc_msg_proposal_prefix { /* prefix part of clc proposal message*/
152 __be32 outgoing_subnet; /* subnet mask */
153 u8 prefix_len; /* number of significant bits in mask */
155 u8 ipv6_prefixes_cnt; /* number of IPv6 prefixes in prefix array */
158 struct smc_clc_msg_smcd { /* SMC-D GID information */
159 struct smc_clc_smcd_gid_chid ism; /* ISM native GID+CHID of requester */
160 __be16 v2_ext_offset; /* SMC Version 2 Extension Offset */
161 u8 vendor_oui[3]; /* vendor organizationally unique identifier */
162 u8 vendor_exp_options[5];
166 struct smc_clc_smcd_v2_extension {
167 /* New members must be added within the struct_group() macro below. */
168 struct_group_tagged(smc_clc_smcd_v2_extension_fixed, fixed,
169 u8 system_eid[SMC_MAX_EID_LEN];
172 struct smc_clc_smcd_gid_chid gidchid[];
174 static_assert(offsetof(struct smc_clc_smcd_v2_extension, gidchid) == sizeof(struct smc_clc_smcd_v2_extension_fixed),
175 "struct member likely outside of struct_group_tagged()");
177 struct smc_clc_msg_proposal { /* clc proposal message sent by Linux */
178 struct smc_clc_msg_hdr hdr;
179 struct smc_clc_msg_local lcl;
180 __be16 iparea_offset; /* offset to IP address information area */
183 #define SMC_CLC_MAX_V6_PREFIX 8
184 #define SMC_CLC_MAX_UEID 8
185 #define SMCD_CLC_MAX_V2_GID_ENTRIES 8 /* max # of CHID-GID entries in CLC
186 * proposal SMC-Dv2 extension.
187 * each ISM device takes one entry and
188 * each Emulated-ISM takes two entries
191 struct smc_clc_msg_proposal_area {
192 struct smc_clc_msg_proposal pclc_base;
193 struct smc_clc_msg_smcd pclc_smcd;
194 struct smc_clc_msg_proposal_prefix pclc_prfx;
195 struct smc_clc_ipv6_prefix pclc_prfx_ipv6[SMC_CLC_MAX_V6_PREFIX];
196 struct smc_clc_v2_extension_fixed pclc_v2_ext;
197 u8 user_eids[SMC_CLC_MAX_UEID][SMC_MAX_EID_LEN];
198 struct smc_clc_smcd_v2_extension_fixed pclc_smcd_v2_ext;
199 struct smc_clc_smcd_gid_chid
200 pclc_gidchids[SMCD_CLC_MAX_V2_GID_ENTRIES];
201 struct smc_clc_msg_trail pclc_trl;
204 struct smcr_clc_msg_accept_confirm { /* SMCR accept/confirm */
205 struct smc_clc_msg_local lcl;
206 u8 qpn[3]; /* QP number */
207 __be32 rmb_rkey; /* RMB rkey */
208 u8 rmbe_idx; /* Index of RMBE in RMB */
209 __be32 rmbe_alert_token; /* unique connection id */
210 #if defined(__BIG_ENDIAN_BITFIELD)
211 u8 rmbe_size : 4, /* buf size (compressed) */
212 qp_mtu : 4; /* QP mtu */
213 #elif defined(__LITTLE_ENDIAN_BITFIELD)
218 __be64 rmb_dma_addr; /* RMB virtual address */
220 u8 psn[3]; /* packet sequence number */
223 struct smcd_clc_msg_accept_confirm_common { /* SMCD accept/confirm */
224 __be64 gid; /* Sender GID */
225 __be64 token; /* DMB token */
226 u8 dmbe_idx; /* DMBE index */
227 #if defined(__BIG_ENDIAN_BITFIELD)
228 u8 dmbe_size : 4, /* buf size (compressed) */
230 #elif defined(__LITTLE_ENDIAN_BITFIELD)
235 __be32 linkid; /* Link identifier */
238 #define SMC_CLC_OS_ZOS 1
239 #define SMC_CLC_OS_LINUX 2
240 #define SMC_CLC_OS_AIX 3
242 struct smc_clc_first_contact_ext {
243 #if defined(__BIG_ENDIAN_BITFIELD)
248 #elif defined(__LITTLE_ENDIAN_BITFIELD)
255 u8 hostname[SMC_MAX_HOSTNAME_LEN];
258 struct smc_clc_first_contact_ext_v2x {
259 struct smc_clc_first_contact_ext fce_v2_base;
262 u8 max_conns; /* for SMC-R only */
263 u8 max_links; /* for SMC-R only */
265 u8 reserved3[2]; /* for SMC-D only */
268 __be32 vendor_exp_options;
270 } __packed; /* format defined in
271 * IBM Shared Memory Communications Version 2 (Third Edition)
272 * (https://www.ibm.com/support/pages/node/7009315)
275 struct smc_clc_fce_gid_ext {
278 u8 gid[][SMC_GID_SIZE];
281 struct smc_clc_msg_accept_confirm { /* clc accept / confirm message */
282 struct smc_clc_msg_hdr hdr;
285 struct smcr_clc_msg_accept_confirm r0;
286 struct { /* v2 only */
287 u8 eid[SMC_MAX_EID_LEN];
292 struct smcd_clc_msg_accept_confirm_common d0;
293 struct { /* v2 only, but 12 bytes reserved in v1 */
295 u8 eid[SMC_MAX_EID_LEN];
302 struct smc_clc_msg_decline { /* clc decline message */
303 struct smc_clc_msg_hdr hdr;
304 u8 id_for_peer[SMC_SYSTEMID_LEN]; /* sender peer_id */
305 __be32 peer_diagnosis; /* diagnosis information */
306 #if defined(__BIG_ENDIAN_BITFIELD)
309 #elif defined(__LITTLE_ENDIAN_BITFIELD)
314 struct smc_clc_msg_trail trl; /* eye catcher "SMCD" or "SMCR" EBCDIC */
317 #define SMC_DECL_DIAG_COUNT_V2 4 /* no. of additional peer diagnosis codes */
319 struct smc_clc_msg_decline_v2 { /* clc decline message */
320 struct smc_clc_msg_hdr hdr;
321 u8 id_for_peer[SMC_SYSTEMID_LEN]; /* sender peer_id */
322 __be32 peer_diagnosis; /* diagnosis information */
323 #if defined(__BIG_ENDIAN_BITFIELD)
326 #elif defined(__LITTLE_ENDIAN_BITFIELD)
331 __be32 peer_diagnosis_v2[SMC_DECL_DIAG_COUNT_V2];
332 struct smc_clc_msg_trail trl; /* eye catcher "SMCD" or "SMCR" EBCDIC */
335 /* determine start of the prefix area within the proposal message */
336 static inline struct smc_clc_msg_proposal_prefix *
337 smc_clc_proposal_get_prefix(struct smc_clc_msg_proposal *pclc)
339 u16 offset = ntohs(pclc->iparea_offset);
341 if (offset > sizeof(struct smc_clc_msg_smcd))
343 return (struct smc_clc_msg_proposal_prefix *)
344 ((u8 *)pclc + sizeof(*pclc) + offset);
347 static inline bool smcr_indicated(int smc_type)
349 return smc_type == SMC_TYPE_R || smc_type == SMC_TYPE_B;
352 static inline bool smcd_indicated(int smc_type)
354 return smc_type == SMC_TYPE_D || smc_type == SMC_TYPE_B;
357 static inline u8 smc_indicated_type(int is_smcd, int is_smcr)
359 if (is_smcd && is_smcr)
368 /* get SMC-D info from proposal message */
369 static inline struct smc_clc_msg_smcd *
370 smc_get_clc_msg_smcd(struct smc_clc_msg_proposal *prop)
372 if (smcd_indicated(prop->hdr.typev1) &&
373 ntohs(prop->iparea_offset) != sizeof(struct smc_clc_msg_smcd))
376 return (struct smc_clc_msg_smcd *)(prop + 1);
379 static inline struct smc_clc_v2_extension *
380 smc_get_clc_v2_ext(struct smc_clc_msg_proposal *prop)
382 struct smc_clc_msg_smcd *prop_smcd = smc_get_clc_msg_smcd(prop);
385 max_offset = offsetof(struct smc_clc_msg_proposal_area, pclc_v2_ext) -
386 offsetof(struct smc_clc_msg_proposal_area, pclc_smcd) -
387 offsetofend(struct smc_clc_msg_smcd, v2_ext_offset);
389 if (!prop_smcd || !ntohs(prop_smcd->v2_ext_offset) ||
390 ntohs(prop_smcd->v2_ext_offset) > max_offset)
393 return (struct smc_clc_v2_extension *)
395 offsetof(struct smc_clc_msg_smcd, v2_ext_offset) +
396 sizeof(prop_smcd->v2_ext_offset) +
397 ntohs(prop_smcd->v2_ext_offset));
400 static inline struct smc_clc_smcd_v2_extension *
401 smc_get_clc_smcd_v2_ext(struct smc_clc_v2_extension *prop_v2ext)
403 u16 max_offset = offsetof(struct smc_clc_msg_proposal_area, pclc_smcd_v2_ext) -
404 offsetof(struct smc_clc_msg_proposal_area, pclc_v2_ext) -
405 offsetof(struct smc_clc_v2_extension, hdr) -
406 offsetofend(struct smc_clnt_opts_area_hdr, smcd_v2_ext_offset);
410 if (!ntohs(prop_v2ext->hdr.smcd_v2_ext_offset) ||
411 ntohs(prop_v2ext->hdr.smcd_v2_ext_offset) > max_offset)
414 return (struct smc_clc_smcd_v2_extension *)
416 offsetof(struct smc_clc_v2_extension, hdr) +
417 offsetof(struct smc_clnt_opts_area_hdr, smcd_v2_ext_offset) +
418 sizeof(prop_v2ext->hdr.smcd_v2_ext_offset) +
419 ntohs(prop_v2ext->hdr.smcd_v2_ext_offset));
422 static inline struct smc_clc_first_contact_ext *
423 smc_get_clc_first_contact_ext(struct smc_clc_msg_accept_confirm *clc,
428 if (clc->hdr.version == SMC_V1 ||
429 !(clc->hdr.typev2 & SMC_FIRST_CONTACT_MASK))
434 offsetofend(struct smc_clc_msg_accept_confirm, d1);
437 offsetofend(struct smc_clc_msg_accept_confirm, r1);
439 return (struct smc_clc_first_contact_ext *)(((u8 *)clc) + clc_v2_len);
443 struct smc_init_info;
445 int smc_clc_prfx_match(struct socket *clcsock,
446 struct smc_clc_msg_proposal_prefix *prop);
447 int smc_clc_wait_msg(struct smc_sock *smc, void *buf, int buflen,
448 u8 expected_type, unsigned long timeout);
449 int smc_clc_send_decline(struct smc_sock *smc, u32 peer_diag_info, u8 version);
450 int smc_clc_send_proposal(struct smc_sock *smc, struct smc_init_info *ini);
451 int smc_clc_send_confirm(struct smc_sock *smc, bool clnt_first_contact,
452 u8 version, u8 *eid, struct smc_init_info *ini);
453 int smc_clc_send_accept(struct smc_sock *smc, bool srv_first_contact,
454 u8 version, u8 *negotiated_eid, struct smc_init_info *ini);
455 int smc_clc_srv_v2x_features_validate(struct smc_sock *smc,
456 struct smc_clc_msg_proposal *pclc,
457 struct smc_init_info *ini);
458 int smc_clc_clnt_v2x_features_validate(struct smc_clc_first_contact_ext *fce,
459 struct smc_init_info *ini);
460 int smc_clc_v2x_features_confirm_check(struct smc_clc_msg_accept_confirm *cclc,
461 struct smc_init_info *ini);
462 void smc_clc_init(void) __init;
463 void smc_clc_exit(void);
464 void smc_clc_get_hostname(u8 **host);
465 bool smc_clc_match_eid(u8 *negotiated_eid,
466 struct smc_clc_v2_extension *smc_v2_ext,
467 u8 *peer_eid, u8 *local_eid);
468 int smc_clc_ueid_count(void);
469 int smc_nl_dump_ueid(struct sk_buff *skb, struct netlink_callback *cb);
470 int smc_nl_add_ueid(struct sk_buff *skb, struct genl_info *info);
471 int smc_nl_remove_ueid(struct sk_buff *skb, struct genl_info *info);
472 int smc_nl_flush_ueid(struct sk_buff *skb, struct genl_info *info);
473 int smc_nl_dump_seid(struct sk_buff *skb, struct netlink_callback *cb);
474 int smc_nl_enable_seid(struct sk_buff *skb, struct genl_info *info);
475 int smc_nl_disable_seid(struct sk_buff *skb, struct genl_info *info);