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 */ 22 23 #define SMC_MAX_PORTS 2 /* Max # of ports */ 24 25 extern struct proto smc_proto; 26 27 #ifdef ATOMIC64_INIT 28 #define KERNEL_HAS_ATOMIC64 29 #endif 30 31 enum smc_state { /* possible states of an SMC socket */ 32 SMC_ACTIVE = 1, 33 SMC_INIT = 2, 34 SMC_CLOSED = 7, 35 SMC_LISTEN = 10, 36 /* normal close */ 37 SMC_PEERCLOSEWAIT1 = 20, 38 SMC_PEERCLOSEWAIT2 = 21, 39 SMC_APPFINCLOSEWAIT = 24, 40 SMC_APPCLOSEWAIT1 = 22, 41 SMC_APPCLOSEWAIT2 = 23, 42 SMC_PEERFINCLOSEWAIT = 25, 43 /* abnormal close */ 44 SMC_PEERABORTWAIT = 26, 45 SMC_PROCESSABORT = 27, 46 }; 47 48 struct smc_link_group; 49 50 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */ 51 u8 type; 52 } __aligned(1); 53 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 */ 59 u8 reserved : 5; 60 #elif defined(__LITTLE_ENDIAN_BITFIELD) 61 u8 reserved : 5; 62 u8 peer_conn_abort : 1; 63 u8 peer_conn_closed : 1; 64 u8 peer_done_writing : 1; 65 #endif 66 }; 67 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 */ 75 u8 reserved : 3; 76 #elif defined(__LITTLE_ENDIAN_BITFIELD) 77 u8 reserved : 3; 78 u8 failover_validation : 1; 79 u8 cons_curs_upd_req : 1; 80 u8 urg_data_present : 1; 81 u8 urg_data_pending : 1; 82 u8 write_blocked : 1; 83 #endif 84 }; 85 86 /* in host byte order */ 87 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */ 88 struct { 89 u16 reserved; 90 u16 wrap; /* window wrap sequence number */ 91 u32 count; /* cursor (= offset) part */ 92 }; 93 #ifdef KERNEL_HAS_ATOMIC64 94 atomic64_t acurs; /* for atomic processing */ 95 #else 96 u64 acurs; /* for atomic processing */ 97 #endif 98 } __aligned(8); 99 100 /* 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, 108 * piggy backed "ack" 109 */ 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*/ 112 u8 reserved[18]; 113 } __aligned(8); 114 115 struct smc_connection { 116 struct rb_node alert_node; 117 struct smc_link_group *lgr; /* link group of connection */ 118 u32 alert_token_local; /* unique conn. id */ 119 u8 peer_conn_idx; /* from tcp handshake */ 120 int peer_rmbe_size; /* size of peer rx buffer */ 121 atomic_t peer_rmbe_space;/* remaining free bytes in peer 122 * rmbe 123 */ 124 int rtoken_idx; /* idx to peer RMB rkey/addr */ 125 126 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 127 int sndbuf_size; /* sndbuf size <== sock wmem */ 128 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 129 int rmbe_size; /* RMBE size <== sock rmem */ 130 int rmbe_size_short;/* compressed notation */ 131 int rmbe_update_limit; 132 /* lower limit for consumer 133 * cursor update 134 */ 135 136 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 137 * buffer for CDC msg send 138 * .prod cf. TCP snd_nxt 139 * .cons cf. TCP sends ack 140 */ 141 union smc_host_cursor tx_curs_prep; /* tx - prepared data 142 * snd_max..wmem_alloc 143 */ 144 union smc_host_cursor tx_curs_sent; /* tx - sent data 145 * snd_nxt ? 146 */ 147 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 148 * snd-wnd-begin ? 149 */ 150 atomic_t sndbuf_space; /* remaining space in sndbuf */ 151 u16 tx_cdc_seq; /* sequence # for CDC send */ 152 spinlock_t send_lock; /* protect wr_sends */ 153 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 154 155 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 156 * .prod cf. TCP rcv_nxt 157 * .cons cf. TCP snd_una 158 */ 159 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 160 * source of snd_una ? 161 */ 162 atomic_t bytes_to_rcv; /* arrived data, 163 * not yet received 164 */ 165 #ifndef KERNEL_HAS_ATOMIC64 166 spinlock_t acurs_lock; /* protect cursors */ 167 #endif 168 struct work_struct close_work; /* peer sent some closing */ 169 }; 170 171 struct smc_sock { /* smc sock container */ 172 struct sock sk; 173 struct socket *clcsock; /* internal tcp socket */ 174 struct smc_connection conn; /* smc connection */ 175 struct sockaddr *addr; /* inet connect address */ 176 struct smc_sock *listen_smc; /* listen parent */ 177 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 178 struct work_struct smc_listen_work;/* prepare new accept socket */ 179 struct list_head accept_q; /* sockets to be accepted */ 180 spinlock_t accept_q_lock; /* protects accept_q */ 181 bool use_fallback; /* fallback to tcp */ 182 u8 wait_close_tx_prepared : 1; 183 /* shutdown wr or close 184 * started, waiting for unsent 185 * data to be sent 186 */ 187 }; 188 189 static inline struct smc_sock *smc_sk(const struct sock *sk) 190 { 191 return (struct smc_sock *)sk; 192 } 193 194 #define SMC_SYSTEMID_LEN 8 195 196 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 197 198 /* convert an u32 value into network byte order, store it into a 3 byte field */ 199 static inline void hton24(u8 *net, u32 host) 200 { 201 __be32 t; 202 203 t = cpu_to_be32(host); 204 memcpy(net, ((u8 *)&t) + 1, 3); 205 } 206 207 /* convert a received 3 byte field into host byte order*/ 208 static inline u32 ntoh24(u8 *net) 209 { 210 __be32 t = 0; 211 212 memcpy(((u8 *)&t) + 1, net, 3); 213 return be32_to_cpu(t); 214 } 215 216 #define SMC_BUF_MIN_SIZE 16384 /* minimum size of an RMB */ 217 218 #define SMC_RMBE_SIZES 16 /* number of distinct sizes for an RMBE */ 219 /* theoretically, the RFC states that largest size would be 512K, 220 * i.e. compressed 5 and thus 6 sizes (0..5), despite 221 * struct smc_clc_msg_accept_confirm.rmbe_size being a 4 bit value (0..15) 222 */ 223 224 /* convert the RMB size into the compressed notation - minimum 16K. 225 * In contrast to plain ilog2, this rounds towards the next power of 2, 226 * so the socket application gets at least its desired sndbuf / rcvbuf size. 227 */ 228 static inline u8 smc_compress_bufsize(int size) 229 { 230 u8 compressed; 231 232 if (size <= SMC_BUF_MIN_SIZE) 233 return 0; 234 235 size = (size - 1) >> 14; 236 compressed = ilog2(size) + 1; 237 if (compressed >= SMC_RMBE_SIZES) 238 compressed = SMC_RMBE_SIZES - 1; 239 return compressed; 240 } 241 242 /* convert the RMB size from compressed notation into integer */ 243 static inline int smc_uncompress_bufsize(u8 compressed) 244 { 245 u32 size; 246 247 size = 0x00000001 << (((int)compressed) + 14); 248 return (int)size; 249 } 250 251 #ifdef CONFIG_XFRM 252 static inline bool using_ipsec(struct smc_sock *smc) 253 { 254 return (smc->clcsock->sk->sk_policy[0] || 255 smc->clcsock->sk->sk_policy[1]) ? true : false; 256 } 257 #else 258 static inline bool using_ipsec(struct smc_sock *smc) 259 { 260 return false; 261 } 262 #endif 263 264 struct smc_clc_msg_local; 265 266 int smc_netinfo_by_tcpsk(struct socket *clcsock, __be32 *subnet, 267 u8 *prefix_len); 268 void smc_conn_free(struct smc_connection *conn); 269 int smc_conn_create(struct smc_sock *smc, __be32 peer_in_addr, 270 struct smc_ib_device *smcibdev, u8 ibport, 271 struct smc_clc_msg_local *lcl, int srv_first_contact); 272 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 273 void smc_close_non_accepted(struct sock *sk); 274 275 #endif /* __SMC_H */ 276