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 SMC_V1 1 /* SMC version V1 */ 22 #define SMC_V2 2 /* SMC version V2 */ 23 #define SMC_RELEASE 0 24 25 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */ 26 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */ 27 28 #define SMC_MAX_ISM_DEVS 8 /* max # of proposed non-native ISM 29 * devices 30 */ 31 32 extern struct proto smc_proto; 33 extern struct proto smc_proto6; 34 35 #ifdef ATOMIC64_INIT 36 #define KERNEL_HAS_ATOMIC64 37 #endif 38 39 enum smc_state { /* possible states of an SMC socket */ 40 SMC_ACTIVE = 1, 41 SMC_INIT = 2, 42 SMC_CLOSED = 7, 43 SMC_LISTEN = 10, 44 /* normal close */ 45 SMC_PEERCLOSEWAIT1 = 20, 46 SMC_PEERCLOSEWAIT2 = 21, 47 SMC_APPFINCLOSEWAIT = 24, 48 SMC_APPCLOSEWAIT1 = 22, 49 SMC_APPCLOSEWAIT2 = 23, 50 SMC_PEERFINCLOSEWAIT = 25, 51 /* abnormal close */ 52 SMC_PEERABORTWAIT = 26, 53 SMC_PROCESSABORT = 27, 54 }; 55 56 struct smc_link_group; 57 58 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */ 59 union { 60 u8 type; 61 #if defined(__BIG_ENDIAN_BITFIELD) 62 struct { 63 u8 llc_version:4, 64 llc_type:4; 65 }; 66 #elif defined(__LITTLE_ENDIAN_BITFIELD) 67 struct { 68 u8 llc_type:4, 69 llc_version:4; 70 }; 71 #endif 72 }; 73 } __aligned(1); 74 75 struct smc_cdc_conn_state_flags { 76 #if defined(__BIG_ENDIAN_BITFIELD) 77 u8 peer_done_writing : 1; /* Sending done indicator */ 78 u8 peer_conn_closed : 1; /* Peer connection closed indicator */ 79 u8 peer_conn_abort : 1; /* Abnormal close indicator */ 80 u8 reserved : 5; 81 #elif defined(__LITTLE_ENDIAN_BITFIELD) 82 u8 reserved : 5; 83 u8 peer_conn_abort : 1; 84 u8 peer_conn_closed : 1; 85 u8 peer_done_writing : 1; 86 #endif 87 }; 88 89 struct smc_cdc_producer_flags { 90 #if defined(__BIG_ENDIAN_BITFIELD) 91 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */ 92 u8 urg_data_pending : 1; /* Urgent Data Pending */ 93 u8 urg_data_present : 1; /* Urgent Data Present */ 94 u8 cons_curs_upd_req : 1; /* cursor update requested */ 95 u8 failover_validation : 1;/* message replay due to failover */ 96 u8 reserved : 3; 97 #elif defined(__LITTLE_ENDIAN_BITFIELD) 98 u8 reserved : 3; 99 u8 failover_validation : 1; 100 u8 cons_curs_upd_req : 1; 101 u8 urg_data_present : 1; 102 u8 urg_data_pending : 1; 103 u8 write_blocked : 1; 104 #endif 105 }; 106 107 /* in host byte order */ 108 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */ 109 struct { 110 u16 reserved; 111 u16 wrap; /* window wrap sequence number */ 112 u32 count; /* cursor (= offset) part */ 113 }; 114 #ifdef KERNEL_HAS_ATOMIC64 115 atomic64_t acurs; /* for atomic processing */ 116 #else 117 u64 acurs; /* for atomic processing */ 118 #endif 119 } __aligned(8); 120 121 /* in host byte order, except for flag bitfields in network byte order */ 122 struct smc_host_cdc_msg { /* Connection Data Control message */ 123 struct smc_wr_rx_hdr common; /* .type = 0xFE */ 124 u8 len; /* length = 44 */ 125 u16 seqno; /* connection seq # */ 126 u32 token; /* alert_token */ 127 union smc_host_cursor prod; /* producer cursor */ 128 union smc_host_cursor cons; /* consumer cursor, 129 * piggy backed "ack" 130 */ 131 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */ 132 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/ 133 u8 reserved[18]; 134 } __aligned(8); 135 136 enum smc_urg_state { 137 SMC_URG_VALID = 1, /* data present */ 138 SMC_URG_NOTYET = 2, /* data pending */ 139 SMC_URG_READ = 3, /* data was already read */ 140 }; 141 142 struct smc_mark_woken { 143 bool woken; 144 void *key; 145 wait_queue_entry_t wait_entry; 146 }; 147 148 struct smc_connection { 149 struct rb_node alert_node; 150 struct smc_link_group *lgr; /* link group of connection */ 151 struct smc_link *lnk; /* assigned SMC-R link */ 152 u32 alert_token_local; /* unique conn. id */ 153 u8 peer_rmbe_idx; /* from tcp handshake */ 154 int peer_rmbe_size; /* size of peer rx buffer */ 155 atomic_t peer_rmbe_space;/* remaining free bytes in peer 156 * rmbe 157 */ 158 int rtoken_idx; /* idx to peer RMB rkey/addr */ 159 160 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 161 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 162 int rmbe_size_short;/* compressed notation */ 163 int rmbe_update_limit; 164 /* lower limit for consumer 165 * cursor update 166 */ 167 168 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 169 * buffer for CDC msg send 170 * .prod cf. TCP snd_nxt 171 * .cons cf. TCP sends ack 172 */ 173 union smc_host_cursor local_tx_ctrl_fin; 174 /* prod crsr - confirmed by peer 175 */ 176 union smc_host_cursor tx_curs_prep; /* tx - prepared data 177 * snd_max..wmem_alloc 178 */ 179 union smc_host_cursor tx_curs_sent; /* tx - sent data 180 * snd_nxt ? 181 */ 182 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 183 * snd-wnd-begin ? 184 */ 185 atomic_t sndbuf_space; /* remaining space in sndbuf */ 186 u16 tx_cdc_seq; /* sequence # for CDC send */ 187 u16 tx_cdc_seq_fin; /* sequence # - tx completed */ 188 spinlock_t send_lock; /* protect wr_sends */ 189 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe 190 * - inc when post wqe, 191 * - dec on polled tx cqe 192 */ 193 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/ 194 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 195 u32 tx_off; /* base offset in peer rmb */ 196 197 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 198 * .prod cf. TCP rcv_nxt 199 * .cons cf. TCP snd_una 200 */ 201 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 202 * source of snd_una ? 203 */ 204 union smc_host_cursor urg_curs; /* points at urgent byte */ 205 enum smc_urg_state urg_state; 206 bool urg_tx_pend; /* urgent data staged */ 207 bool urg_rx_skip_pend; 208 /* indicate urgent oob data 209 * read, but previous regular 210 * data still pending 211 */ 212 char urg_rx_byte; /* urgent byte */ 213 atomic_t bytes_to_rcv; /* arrived data, 214 * not yet received 215 */ 216 atomic_t splice_pending; /* number of spliced bytes 217 * pending processing 218 */ 219 #ifndef KERNEL_HAS_ATOMIC64 220 spinlock_t acurs_lock; /* protect cursors */ 221 #endif 222 struct work_struct close_work; /* peer sent some closing */ 223 struct work_struct abort_work; /* abort the connection */ 224 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */ 225 u8 rx_off; /* receive offset: 226 * 0 for SMC-R, 32 for SMC-D 227 */ 228 u64 peer_token; /* SMC-D token of peer */ 229 u8 killed : 1; /* abnormal termination */ 230 u8 freed : 1; /* normal termiation */ 231 u8 out_of_sync : 1; /* out of sync with peer */ 232 }; 233 234 struct smc_sock { /* smc sock container */ 235 struct sock sk; 236 struct socket *clcsock; /* internal tcp socket */ 237 void (*clcsk_state_change)(struct sock *sk); 238 /* original stat_change fct. */ 239 void (*clcsk_data_ready)(struct sock *sk); 240 /* original data_ready fct. */ 241 void (*clcsk_write_space)(struct sock *sk); 242 /* original write_space fct. */ 243 void (*clcsk_error_report)(struct sock *sk); 244 /* original error_report fct. */ 245 struct smc_connection conn; /* smc connection */ 246 struct smc_sock *listen_smc; /* listen parent */ 247 struct work_struct connect_work; /* handle non-blocking connect*/ 248 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 249 struct work_struct smc_listen_work;/* prepare new accept socket */ 250 struct list_head accept_q; /* sockets to be accepted */ 251 spinlock_t accept_q_lock; /* protects accept_q */ 252 bool use_fallback; /* fallback to tcp */ 253 int fallback_rsn; /* reason for fallback */ 254 u32 peer_diagnosis; /* decline reason from peer */ 255 int sockopt_defer_accept; 256 /* sockopt TCP_DEFER_ACCEPT 257 * value 258 */ 259 u8 wait_close_tx_prepared : 1; 260 /* shutdown wr or close 261 * started, waiting for unsent 262 * data to be sent 263 */ 264 u8 connect_nonblock : 1; 265 /* non-blocking connect in 266 * flight 267 */ 268 struct mutex clcsock_release_lock; 269 /* protects clcsock of a listen 270 * socket 271 * */ 272 }; 273 274 static inline struct smc_sock *smc_sk(const struct sock *sk) 275 { 276 return (struct smc_sock *)sk; 277 } 278 279 static inline struct smc_sock *smc_clcsock_user_data(struct sock *clcsk) 280 { 281 return (struct smc_sock *) 282 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY); 283 } 284 285 extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 286 extern struct workqueue_struct *smc_close_wq; /* wq for close work */ 287 288 #define SMC_SYSTEMID_LEN 8 289 290 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 291 292 #define ntohll(x) be64_to_cpu(x) 293 #define htonll(x) cpu_to_be64(x) 294 295 /* convert an u32 value into network byte order, store it into a 3 byte field */ 296 static inline void hton24(u8 *net, u32 host) 297 { 298 __be32 t; 299 300 t = cpu_to_be32(host); 301 memcpy(net, ((u8 *)&t) + 1, 3); 302 } 303 304 /* convert a received 3 byte field into host byte order*/ 305 static inline u32 ntoh24(u8 *net) 306 { 307 __be32 t = 0; 308 309 memcpy(((u8 *)&t) + 1, net, 3); 310 return be32_to_cpu(t); 311 } 312 313 #ifdef CONFIG_XFRM 314 static inline bool using_ipsec(struct smc_sock *smc) 315 { 316 return (smc->clcsock->sk->sk_policy[0] || 317 smc->clcsock->sk->sk_policy[1]) ? true : false; 318 } 319 #else 320 static inline bool using_ipsec(struct smc_sock *smc) 321 { 322 return false; 323 } 324 #endif 325 326 struct smc_gidlist; 327 328 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 329 void smc_close_non_accepted(struct sock *sk); 330 void smc_fill_gid_list(struct smc_link_group *lgr, 331 struct smc_gidlist *gidlist, 332 struct smc_ib_device *known_dev, u8 *known_gid); 333 334 #endif /* __SMC_H */ 335