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 struct smc_connection { 118 struct rb_node alert_node; 119 struct smc_link_group *lgr; /* link group of connection */ 120 u32 alert_token_local; /* unique conn. id */ 121 u8 peer_conn_idx; /* from tcp handshake */ 122 int peer_rmbe_size; /* size of peer rx buffer */ 123 atomic_t peer_rmbe_space;/* remaining free bytes in peer 124 * rmbe 125 */ 126 int rtoken_idx; /* idx to peer RMB rkey/addr */ 127 128 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 129 int sndbuf_size; /* sndbuf size <== sock wmem */ 130 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 131 int rmbe_size; /* RMBE size <== sock rmem */ 132 int rmbe_size_short;/* compressed notation */ 133 int rmbe_update_limit; 134 /* lower limit for consumer 135 * cursor update 136 */ 137 138 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 139 * buffer for CDC msg send 140 * .prod cf. TCP snd_nxt 141 * .cons cf. TCP sends ack 142 */ 143 union smc_host_cursor tx_curs_prep; /* tx - prepared data 144 * snd_max..wmem_alloc 145 */ 146 union smc_host_cursor tx_curs_sent; /* tx - sent data 147 * snd_nxt ? 148 */ 149 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 150 * snd-wnd-begin ? 151 */ 152 atomic_t sndbuf_space; /* remaining space in sndbuf */ 153 u16 tx_cdc_seq; /* sequence # for CDC send */ 154 spinlock_t send_lock; /* protect wr_sends */ 155 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 156 157 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 158 * .prod cf. TCP rcv_nxt 159 * .cons cf. TCP snd_una 160 */ 161 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 162 * source of snd_una ? 163 */ 164 atomic_t bytes_to_rcv; /* arrived data, 165 * not yet received 166 */ 167 atomic_t splice_pending; /* number of spliced bytes 168 * pending processing 169 */ 170 #ifndef KERNEL_HAS_ATOMIC64 171 spinlock_t acurs_lock; /* protect cursors */ 172 #endif 173 struct work_struct close_work; /* peer sent some closing */ 174 }; 175 176 struct smc_sock { /* smc sock container */ 177 struct sock sk; 178 struct socket *clcsock; /* internal tcp socket */ 179 struct smc_connection conn; /* smc connection */ 180 struct smc_sock *listen_smc; /* listen parent */ 181 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 182 struct work_struct smc_listen_work;/* prepare new accept socket */ 183 struct list_head accept_q; /* sockets to be accepted */ 184 spinlock_t accept_q_lock; /* protects accept_q */ 185 bool use_fallback; /* fallback to tcp */ 186 int sockopt_defer_accept; 187 /* sockopt TCP_DEFER_ACCEPT 188 * value 189 */ 190 u8 wait_close_tx_prepared : 1; 191 /* shutdown wr or close 192 * started, waiting for unsent 193 * data to be sent 194 */ 195 }; 196 197 static inline struct smc_sock *smc_sk(const struct sock *sk) 198 { 199 return (struct smc_sock *)sk; 200 } 201 202 #define SMC_SYSTEMID_LEN 8 203 204 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 205 206 /* convert an u32 value into network byte order, store it into a 3 byte field */ 207 static inline void hton24(u8 *net, u32 host) 208 { 209 __be32 t; 210 211 t = cpu_to_be32(host); 212 memcpy(net, ((u8 *)&t) + 1, 3); 213 } 214 215 /* convert a received 3 byte field into host byte order*/ 216 static inline u32 ntoh24(u8 *net) 217 { 218 __be32 t = 0; 219 220 memcpy(((u8 *)&t) + 1, net, 3); 221 return be32_to_cpu(t); 222 } 223 224 #define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */ 225 226 #define SMC_RMBE_SIZES 16 /* number of distinct sizes for an RMBE */ 227 /* theoretically, the RFC states that largest size would be 512K, 228 * i.e. compressed 5 and thus 6 sizes (0..5), despite 229 * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15) 230 */ 231 232 /* convert the RMB size into the compressed notation - minimum 16K. 233 * In contrast to plain ilog2, this rounds towards the next power of 2, 234 * so the socket application gets at least its desired sndbuf / rcvbuf size. 235 */ 236 static inline u8 smc_compress_bufsize(int size) 237 { 238 u8 compressed; 239 240 if (size <= SMC_BUF_MIN_SIZE) 241 return 0; 242 243 size = (size - 1) >> 14; 244 compressed = ilog2(size) + 1; 245 if (compressed >= SMC_RMBE_SIZES) 246 compressed = SMC_RMBE_SIZES - 1; 247 return compressed; 248 } 249 250 /* convert the RMB size from compressed notation into integer */ 251 static inline int smc_uncompress_bufsize(u8 compressed) 252 { 253 u32 size; 254 255 size = 0x00000001 << (((int)compressed) + 14); 256 return (int)size; 257 } 258 259 #ifdef CONFIG_XFRM 260 static inline bool using_ipsec(struct smc_sock *smc) 261 { 262 return (smc->clcsock->sk->sk_policy[0] || 263 smc->clcsock->sk->sk_policy[1]) ? true : false; 264 } 265 #else 266 static inline bool using_ipsec(struct smc_sock *smc) 267 { 268 return false; 269 } 270 #endif 271 272 struct smc_clc_msg_local; 273 274 void smc_conn_free(struct smc_connection *conn); 275 int smc_conn_create(struct smc_sock *smc, 276 struct smc_ib_device *smcibdev, u8 ibport, 277 struct smc_clc_msg_local *lcl, int srv_first_contact); 278 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 279 void smc_close_non_accepted(struct sock *sk); 280 281 #endif /* __SMC_H */ 282