1 /* SPDX-License-Identifier: GPL-2.0 */ 2 /* 3 * Shared Memory Communications over RDMA (SMC-R) and RoCE 4 * 5 * Definitions for the SMC module (socket related) 6 * 7 * Copyright IBM Corp. 2016 8 * 9 * Author(s): Ursula Braun <ubraun@linux.vnet.ibm.com> 10 */ 11 #ifndef __SMC_H 12 #define __SMC_H 13 14 #include <linux/socket.h> 15 #include <linux/types.h> 16 #include <linux/compiler.h> /* __aligned */ 17 #include <net/sock.h> 18 19 #include "smc_ib.h" 20 21 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */ 22 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */ 23 24 #define SMC_MAX_PORTS 2 /* Max # of ports */ 25 26 extern struct proto smc_proto; 27 extern struct proto smc_proto6; 28 29 #ifdef ATOMIC64_INIT 30 #define KERNEL_HAS_ATOMIC64 31 #endif 32 33 enum smc_state { /* possible states of an SMC socket */ 34 SMC_ACTIVE = 1, 35 SMC_INIT = 2, 36 SMC_CLOSED = 7, 37 SMC_LISTEN = 10, 38 /* normal close */ 39 SMC_PEERCLOSEWAIT1 = 20, 40 SMC_PEERCLOSEWAIT2 = 21, 41 SMC_APPFINCLOSEWAIT = 24, 42 SMC_APPCLOSEWAIT1 = 22, 43 SMC_APPCLOSEWAIT2 = 23, 44 SMC_PEERFINCLOSEWAIT = 25, 45 /* abnormal close */ 46 SMC_PEERABORTWAIT = 26, 47 SMC_PROCESSABORT = 27, 48 }; 49 50 struct smc_link_group; 51 52 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */ 53 u8 type; 54 } __aligned(1); 55 56 struct smc_cdc_conn_state_flags { 57 #if defined(__BIG_ENDIAN_BITFIELD) 58 u8 peer_done_writing : 1; /* Sending done indicator */ 59 u8 peer_conn_closed : 1; /* Peer connection closed indicator */ 60 u8 peer_conn_abort : 1; /* Abnormal close indicator */ 61 u8 reserved : 5; 62 #elif defined(__LITTLE_ENDIAN_BITFIELD) 63 u8 reserved : 5; 64 u8 peer_conn_abort : 1; 65 u8 peer_conn_closed : 1; 66 u8 peer_done_writing : 1; 67 #endif 68 }; 69 70 struct smc_cdc_producer_flags { 71 #if defined(__BIG_ENDIAN_BITFIELD) 72 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */ 73 u8 urg_data_pending : 1; /* Urgent Data Pending */ 74 u8 urg_data_present : 1; /* Urgent Data Present */ 75 u8 cons_curs_upd_req : 1; /* cursor update requested */ 76 u8 failover_validation : 1;/* message replay due to failover */ 77 u8 reserved : 3; 78 #elif defined(__LITTLE_ENDIAN_BITFIELD) 79 u8 reserved : 3; 80 u8 failover_validation : 1; 81 u8 cons_curs_upd_req : 1; 82 u8 urg_data_present : 1; 83 u8 urg_data_pending : 1; 84 u8 write_blocked : 1; 85 #endif 86 }; 87 88 /* in host byte order */ 89 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */ 90 struct { 91 u16 reserved; 92 u16 wrap; /* window wrap sequence number */ 93 u32 count; /* cursor (= offset) part */ 94 }; 95 #ifdef KERNEL_HAS_ATOMIC64 96 atomic64_t acurs; /* for atomic processing */ 97 #else 98 u64 acurs; /* for atomic processing */ 99 #endif 100 } __aligned(8); 101 102 /* in host byte order, except for flag bitfields in network byte order */ 103 struct smc_host_cdc_msg { /* Connection Data Control message */ 104 struct smc_wr_rx_hdr common; /* .type = 0xFE */ 105 u8 len; /* length = 44 */ 106 u16 seqno; /* connection seq # */ 107 u32 token; /* alert_token */ 108 union smc_host_cursor prod; /* producer cursor */ 109 union smc_host_cursor cons; /* consumer cursor, 110 * piggy backed "ack" 111 */ 112 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */ 113 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/ 114 u8 reserved[18]; 115 } __aligned(8); 116 117 enum smc_urg_state { 118 SMC_URG_VALID, /* data present */ 119 SMC_URG_NOTYET, /* data pending */ 120 SMC_URG_READ /* data was already read */ 121 }; 122 123 struct smc_connection { 124 struct rb_node alert_node; 125 struct smc_link_group *lgr; /* link group of connection */ 126 u32 alert_token_local; /* unique conn. id */ 127 u8 peer_rmbe_idx; /* from tcp handshake */ 128 int peer_rmbe_size; /* size of peer rx buffer */ 129 atomic_t peer_rmbe_space;/* remaining free bytes in peer 130 * rmbe 131 */ 132 int rtoken_idx; /* idx to peer RMB rkey/addr */ 133 134 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 135 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 136 int rmbe_size_short;/* compressed notation */ 137 int rmbe_update_limit; 138 /* lower limit for consumer 139 * cursor update 140 */ 141 142 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 143 * buffer for CDC msg send 144 * .prod cf. TCP snd_nxt 145 * .cons cf. TCP sends ack 146 */ 147 union smc_host_cursor tx_curs_prep; /* tx - prepared data 148 * snd_max..wmem_alloc 149 */ 150 union smc_host_cursor tx_curs_sent; /* tx - sent data 151 * snd_nxt ? 152 */ 153 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 154 * snd-wnd-begin ? 155 */ 156 atomic_t sndbuf_space; /* remaining space in sndbuf */ 157 u16 tx_cdc_seq; /* sequence # for CDC send */ 158 spinlock_t send_lock; /* protect wr_sends */ 159 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 160 u32 tx_off; /* base offset in peer rmb */ 161 162 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 163 * .prod cf. TCP rcv_nxt 164 * .cons cf. TCP snd_una 165 */ 166 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 167 * source of snd_una ? 168 */ 169 union smc_host_cursor urg_curs; /* points at urgent byte */ 170 enum smc_urg_state urg_state; 171 bool urg_tx_pend; /* urgent data staged */ 172 bool urg_rx_skip_pend; 173 /* indicate urgent oob data 174 * read, but previous regular 175 * data still pending 176 */ 177 char urg_rx_byte; /* urgent byte */ 178 atomic_t bytes_to_rcv; /* arrived data, 179 * not yet received 180 */ 181 atomic_t splice_pending; /* number of spliced bytes 182 * pending processing 183 */ 184 #ifndef KERNEL_HAS_ATOMIC64 185 spinlock_t acurs_lock; /* protect cursors */ 186 #endif 187 struct work_struct close_work; /* peer sent some closing */ 188 }; 189 190 struct smc_connect_info { 191 int flags; 192 int alen; 193 struct sockaddr addr; 194 }; 195 196 struct smc_sock { /* smc sock container */ 197 struct sock sk; 198 struct socket *clcsock; /* internal tcp socket */ 199 struct smc_connection conn; /* smc connection */ 200 struct smc_sock *listen_smc; /* listen parent */ 201 struct smc_connect_info *connect_info; /* connect address & flags */ 202 struct work_struct connect_work; /* handle non-blocking connect*/ 203 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 204 struct work_struct smc_listen_work;/* prepare new accept socket */ 205 struct list_head accept_q; /* sockets to be accepted */ 206 spinlock_t accept_q_lock; /* protects accept_q */ 207 bool use_fallback; /* fallback to tcp */ 208 int sockopt_defer_accept; 209 /* sockopt TCP_DEFER_ACCEPT 210 * value 211 */ 212 u8 wait_close_tx_prepared : 1; 213 /* shutdown wr or close 214 * started, waiting for unsent 215 * data to be sent 216 */ 217 }; 218 219 static inline struct smc_sock *smc_sk(const struct sock *sk) 220 { 221 return (struct smc_sock *)sk; 222 } 223 224 #define SMC_SYSTEMID_LEN 8 225 226 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 227 228 /* convert an u32 value into network byte order, store it into a 3 byte field */ 229 static inline void hton24(u8 *net, u32 host) 230 { 231 __be32 t; 232 233 t = cpu_to_be32(host); 234 memcpy(net, ((u8 *)&t) + 1, 3); 235 } 236 237 /* convert a received 3 byte field into host byte order*/ 238 static inline u32 ntoh24(u8 *net) 239 { 240 __be32 t = 0; 241 242 memcpy(((u8 *)&t) + 1, net, 3); 243 return be32_to_cpu(t); 244 } 245 246 #ifdef CONFIG_XFRM 247 static inline bool using_ipsec(struct smc_sock *smc) 248 { 249 return (smc->clcsock->sk->sk_policy[0] || 250 smc->clcsock->sk->sk_policy[1]) ? true : false; 251 } 252 #else 253 static inline bool using_ipsec(struct smc_sock *smc) 254 { 255 return false; 256 } 257 #endif 258 259 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 260 void smc_close_non_accepted(struct sock *sk); 261 262 #endif /* __SMC_H */ 263