1 /* SPDX-License-Identifier: GPL-2.0 */
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
5 * Definitions for the SMC module (socket related)
7 * Copyright IBM Corp. 2016
9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com>
14 #include <linux/socket.h>
15 #include <linux/types.h>
16 #include <linux/compiler.h> /* __aligned */
21 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */
22 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */
24 extern struct proto smc_proto
;
25 extern struct proto smc_proto6
;
28 #define KERNEL_HAS_ATOMIC64
31 enum smc_state
{ /* possible states of an SMC socket */
37 SMC_PEERCLOSEWAIT1
= 20,
38 SMC_PEERCLOSEWAIT2
= 21,
39 SMC_APPFINCLOSEWAIT
= 24,
40 SMC_APPCLOSEWAIT1
= 22,
41 SMC_APPCLOSEWAIT2
= 23,
42 SMC_PEERFINCLOSEWAIT
= 25,
44 SMC_PEERABORTWAIT
= 26,
45 SMC_PROCESSABORT
= 27,
48 struct smc_link_group
;
50 struct smc_wr_rx_hdr
{ /* common prefix part of LLC and CDC to demultiplex */
54 struct smc_cdc_conn_state_flags
{
55 #if defined(__BIG_ENDIAN_BITFIELD)
56 u8 peer_done_writing
: 1; /* Sending done indicator */
57 u8 peer_conn_closed
: 1; /* Peer connection closed indicator */
58 u8 peer_conn_abort
: 1; /* Abnormal close indicator */
60 #elif defined(__LITTLE_ENDIAN_BITFIELD)
62 u8 peer_conn_abort
: 1;
63 u8 peer_conn_closed
: 1;
64 u8 peer_done_writing
: 1;
68 struct smc_cdc_producer_flags
{
69 #if defined(__BIG_ENDIAN_BITFIELD)
70 u8 write_blocked
: 1; /* Writing Blocked, no rx buf space */
71 u8 urg_data_pending
: 1; /* Urgent Data Pending */
72 u8 urg_data_present
: 1; /* Urgent Data Present */
73 u8 cons_curs_upd_req
: 1; /* cursor update requested */
74 u8 failover_validation
: 1;/* message replay due to failover */
76 #elif defined(__LITTLE_ENDIAN_BITFIELD)
78 u8 failover_validation
: 1;
79 u8 cons_curs_upd_req
: 1;
80 u8 urg_data_present
: 1;
81 u8 urg_data_pending
: 1;
86 /* in host byte order */
87 union smc_host_cursor
{ /* SMC cursor - an offset in an RMBE */
90 u16 wrap
; /* window wrap sequence number */
91 u32 count
; /* cursor (= offset) part */
93 #ifdef KERNEL_HAS_ATOMIC64
94 atomic64_t acurs
; /* for atomic processing */
96 u64 acurs
; /* for atomic processing */
100 /* in host byte order, except for flag bitfields in network byte order */
101 struct smc_host_cdc_msg
{ /* Connection Data Control message */
102 struct smc_wr_rx_hdr common
; /* .type = 0xFE */
103 u8 len
; /* length = 44 */
104 u16 seqno
; /* connection seq # */
105 u32 token
; /* alert_token */
106 union smc_host_cursor prod
; /* producer cursor */
107 union smc_host_cursor cons
; /* consumer cursor,
110 struct smc_cdc_producer_flags prod_flags
; /* conn. tx/rx status */
111 struct smc_cdc_conn_state_flags conn_state_flags
; /* peer conn. status*/
116 SMC_URG_VALID
= 1, /* data present */
117 SMC_URG_NOTYET
= 2, /* data pending */
118 SMC_URG_READ
= 3, /* data was already read */
121 struct smc_connection
{
122 struct rb_node alert_node
;
123 struct smc_link_group
*lgr
; /* link group of connection */
124 struct smc_link
*lnk
; /* assigned SMC-R link */
125 u32 alert_token_local
; /* unique conn. id */
126 u8 peer_rmbe_idx
; /* from tcp handshake */
127 int peer_rmbe_size
; /* size of peer rx buffer */
128 atomic_t peer_rmbe_space
;/* remaining free bytes in peer
131 int rtoken_idx
; /* idx to peer RMB rkey/addr */
133 struct smc_buf_desc
*sndbuf_desc
; /* send buffer descriptor */
134 struct smc_buf_desc
*rmb_desc
; /* RMBE descriptor */
135 int rmbe_size_short
;/* compressed notation */
136 int rmbe_update_limit
;
137 /* lower limit for consumer
141 struct smc_host_cdc_msg local_tx_ctrl
; /* host byte order staging
142 * buffer for CDC msg send
143 * .prod cf. TCP snd_nxt
144 * .cons cf. TCP sends ack
146 union smc_host_cursor local_tx_ctrl_fin
;
147 /* prod crsr - confirmed by peer
149 union smc_host_cursor tx_curs_prep
; /* tx - prepared data
150 * snd_max..wmem_alloc
152 union smc_host_cursor tx_curs_sent
; /* tx - sent data
155 union smc_host_cursor tx_curs_fin
; /* tx - confirmed by peer
158 atomic_t sndbuf_space
; /* remaining space in sndbuf */
159 u16 tx_cdc_seq
; /* sequence # for CDC send */
160 u16 tx_cdc_seq_fin
; /* sequence # - tx completed */
161 spinlock_t send_lock
; /* protect wr_sends */
162 struct delayed_work tx_work
; /* retry of smc_cdc_msg_send */
163 u32 tx_off
; /* base offset in peer rmb */
165 struct smc_host_cdc_msg local_rx_ctrl
; /* filled during event_handl.
166 * .prod cf. TCP rcv_nxt
167 * .cons cf. TCP snd_una
169 union smc_host_cursor rx_curs_confirmed
; /* confirmed to peer
170 * source of snd_una ?
172 union smc_host_cursor urg_curs
; /* points at urgent byte */
173 enum smc_urg_state urg_state
;
174 bool urg_tx_pend
; /* urgent data staged */
175 bool urg_rx_skip_pend
;
176 /* indicate urgent oob data
177 * read, but previous regular
180 char urg_rx_byte
; /* urgent byte */
181 atomic_t bytes_to_rcv
; /* arrived data,
184 atomic_t splice_pending
; /* number of spliced bytes
187 #ifndef KERNEL_HAS_ATOMIC64
188 spinlock_t acurs_lock
; /* protect cursors */
190 struct work_struct close_work
; /* peer sent some closing */
191 struct work_struct abort_work
; /* abort the connection */
192 struct tasklet_struct rx_tsklet
; /* Receiver tasklet for SMC-D */
193 u8 rx_off
; /* receive offset:
194 * 0 for SMC-R, 32 for SMC-D
196 u64 peer_token
; /* SMC-D token of peer */
197 u8 killed
: 1; /* abnormal termination */
198 u8 out_of_sync
: 1; /* out of sync with peer */
201 struct smc_sock
{ /* smc sock container */
203 struct socket
*clcsock
; /* internal tcp socket */
204 struct smc_connection conn
; /* smc connection */
205 struct smc_sock
*listen_smc
; /* listen parent */
206 struct work_struct connect_work
; /* handle non-blocking connect*/
207 struct work_struct tcp_listen_work
;/* handle tcp socket accepts */
208 struct work_struct smc_listen_work
;/* prepare new accept socket */
209 struct list_head accept_q
; /* sockets to be accepted */
210 spinlock_t accept_q_lock
; /* protects accept_q */
211 bool use_fallback
; /* fallback to tcp */
212 int fallback_rsn
; /* reason for fallback */
213 u32 peer_diagnosis
; /* decline reason from peer */
214 int sockopt_defer_accept
;
215 /* sockopt TCP_DEFER_ACCEPT
218 u8 wait_close_tx_prepared
: 1;
219 /* shutdown wr or close
220 * started, waiting for unsent
223 u8 connect_nonblock
: 1;
224 /* non-blocking connect in
227 struct mutex clcsock_release_lock
;
228 /* protects clcsock of a listen
233 static inline struct smc_sock
*smc_sk(const struct sock
*sk
)
235 return (struct smc_sock
*)sk
;
238 #define SMC_SYSTEMID_LEN 8
240 extern u8 local_systemid
[SMC_SYSTEMID_LEN
]; /* unique system identifier */
242 /* convert an u32 value into network byte order, store it into a 3 byte field */
243 static inline void hton24(u8
*net
, u32 host
)
247 t
= cpu_to_be32(host
);
248 memcpy(net
, ((u8
*)&t
) + 1, 3);
251 /* convert a received 3 byte field into host byte order*/
252 static inline u32
ntoh24(u8
*net
)
256 memcpy(((u8
*)&t
) + 1, net
, 3);
257 return be32_to_cpu(t
);
261 static inline bool using_ipsec(struct smc_sock
*smc
)
263 return (smc
->clcsock
->sk
->sk_policy
[0] ||
264 smc
->clcsock
->sk
->sk_policy
[1]) ? true : false;
267 static inline bool using_ipsec(struct smc_sock
*smc
)
273 struct sock
*smc_accept_dequeue(struct sock
*parent
, struct socket
*new_sock
);
274 void smc_close_non_accepted(struct sock
*sk
);