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/genetlink.h> 18 #include <net/sock.h> 19 20 #include "smc_ib.h" 21 22 #define SMC_V1 1 /* SMC version V1 */ 23 #define SMC_V2 2 /* SMC version V2 */ 24 #define SMC_RELEASE 0 25 26 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */ 27 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */ 28 29 #define SMC_MAX_ISM_DEVS 8 /* max # of proposed non-native ISM 30 * devices 31 */ 32 33 extern struct proto smc_proto; 34 extern struct proto smc_proto6; 35 36 #ifdef ATOMIC64_INIT 37 #define KERNEL_HAS_ATOMIC64 38 #endif 39 40 enum smc_state { /* possible states of an SMC socket */ 41 SMC_ACTIVE = 1, 42 SMC_INIT = 2, 43 SMC_CLOSED = 7, 44 SMC_LISTEN = 10, 45 /* normal close */ 46 SMC_PEERCLOSEWAIT1 = 20, 47 SMC_PEERCLOSEWAIT2 = 21, 48 SMC_APPFINCLOSEWAIT = 24, 49 SMC_APPCLOSEWAIT1 = 22, 50 SMC_APPCLOSEWAIT2 = 23, 51 SMC_PEERFINCLOSEWAIT = 25, 52 /* abnormal close */ 53 SMC_PEERABORTWAIT = 26, 54 SMC_PROCESSABORT = 27, 55 }; 56 57 struct smc_link_group; 58 59 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */ 60 union { 61 u8 type; 62 #if defined(__BIG_ENDIAN_BITFIELD) 63 struct { 64 u8 llc_version:4, 65 llc_type:4; 66 }; 67 #elif defined(__LITTLE_ENDIAN_BITFIELD) 68 struct { 69 u8 llc_type:4, 70 llc_version:4; 71 }; 72 #endif 73 }; 74 } __aligned(1); 75 76 struct smc_cdc_conn_state_flags { 77 #if defined(__BIG_ENDIAN_BITFIELD) 78 u8 peer_done_writing : 1; /* Sending done indicator */ 79 u8 peer_conn_closed : 1; /* Peer connection closed indicator */ 80 u8 peer_conn_abort : 1; /* Abnormal close indicator */ 81 u8 reserved : 5; 82 #elif defined(__LITTLE_ENDIAN_BITFIELD) 83 u8 reserved : 5; 84 u8 peer_conn_abort : 1; 85 u8 peer_conn_closed : 1; 86 u8 peer_done_writing : 1; 87 #endif 88 }; 89 90 struct smc_cdc_producer_flags { 91 #if defined(__BIG_ENDIAN_BITFIELD) 92 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */ 93 u8 urg_data_pending : 1; /* Urgent Data Pending */ 94 u8 urg_data_present : 1; /* Urgent Data Present */ 95 u8 cons_curs_upd_req : 1; /* cursor update requested */ 96 u8 failover_validation : 1;/* message replay due to failover */ 97 u8 reserved : 3; 98 #elif defined(__LITTLE_ENDIAN_BITFIELD) 99 u8 reserved : 3; 100 u8 failover_validation : 1; 101 u8 cons_curs_upd_req : 1; 102 u8 urg_data_present : 1; 103 u8 urg_data_pending : 1; 104 u8 write_blocked : 1; 105 #endif 106 }; 107 108 /* in host byte order */ 109 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */ 110 struct { 111 u16 reserved; 112 u16 wrap; /* window wrap sequence number */ 113 u32 count; /* cursor (= offset) part */ 114 }; 115 #ifdef KERNEL_HAS_ATOMIC64 116 atomic64_t acurs; /* for atomic processing */ 117 #else 118 u64 acurs; /* for atomic processing */ 119 #endif 120 } __aligned(8); 121 122 /* in host byte order, except for flag bitfields in network byte order */ 123 struct smc_host_cdc_msg { /* Connection Data Control message */ 124 struct smc_wr_rx_hdr common; /* .type = 0xFE */ 125 u8 len; /* length = 44 */ 126 u16 seqno; /* connection seq # */ 127 u32 token; /* alert_token */ 128 union smc_host_cursor prod; /* producer cursor */ 129 union smc_host_cursor cons; /* consumer cursor, 130 * piggy backed "ack" 131 */ 132 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */ 133 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/ 134 u8 reserved[18]; 135 } __aligned(8); 136 137 enum smc_urg_state { 138 SMC_URG_VALID = 1, /* data present */ 139 SMC_URG_NOTYET = 2, /* data pending */ 140 SMC_URG_READ = 3, /* data was already read */ 141 }; 142 143 struct smc_mark_woken { 144 bool woken; 145 void *key; 146 wait_queue_entry_t wait_entry; 147 }; 148 149 struct smc_connection { 150 struct rb_node alert_node; 151 struct smc_link_group *lgr; /* link group of connection */ 152 struct smc_link *lnk; /* assigned SMC-R link */ 153 u32 alert_token_local; /* unique conn. id */ 154 u8 peer_rmbe_idx; /* from tcp handshake */ 155 int peer_rmbe_size; /* size of peer rx buffer */ 156 atomic_t peer_rmbe_space;/* remaining free bytes in peer 157 * rmbe 158 */ 159 int rtoken_idx; /* idx to peer RMB rkey/addr */ 160 161 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 162 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 163 int rmbe_size_short;/* compressed notation */ 164 int rmbe_update_limit; 165 /* lower limit for consumer 166 * cursor update 167 */ 168 169 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 170 * buffer for CDC msg send 171 * .prod cf. TCP snd_nxt 172 * .cons cf. TCP sends ack 173 */ 174 union smc_host_cursor local_tx_ctrl_fin; 175 /* prod crsr - confirmed by peer 176 */ 177 union smc_host_cursor tx_curs_prep; /* tx - prepared data 178 * snd_max..wmem_alloc 179 */ 180 union smc_host_cursor tx_curs_sent; /* tx - sent data 181 * snd_nxt ? 182 */ 183 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 184 * snd-wnd-begin ? 185 */ 186 atomic_t sndbuf_space; /* remaining space in sndbuf */ 187 u16 tx_cdc_seq; /* sequence # for CDC send */ 188 u16 tx_cdc_seq_fin; /* sequence # - tx completed */ 189 spinlock_t send_lock; /* protect wr_sends */ 190 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe 191 * - inc when post wqe, 192 * - dec on polled tx cqe 193 */ 194 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/ 195 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 196 u32 tx_off; /* base offset in peer rmb */ 197 198 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 199 * .prod cf. TCP rcv_nxt 200 * .cons cf. TCP snd_una 201 */ 202 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 203 * source of snd_una ? 204 */ 205 union smc_host_cursor urg_curs; /* points at urgent byte */ 206 enum smc_urg_state urg_state; 207 bool urg_tx_pend; /* urgent data staged */ 208 bool urg_rx_skip_pend; 209 /* indicate urgent oob data 210 * read, but previous regular 211 * data still pending 212 */ 213 char urg_rx_byte; /* urgent byte */ 214 atomic_t bytes_to_rcv; /* arrived data, 215 * not yet received 216 */ 217 atomic_t splice_pending; /* number of spliced bytes 218 * pending processing 219 */ 220 #ifndef KERNEL_HAS_ATOMIC64 221 spinlock_t acurs_lock; /* protect cursors */ 222 #endif 223 struct work_struct close_work; /* peer sent some closing */ 224 struct work_struct abort_work; /* abort the connection */ 225 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */ 226 u8 rx_off; /* receive offset: 227 * 0 for SMC-R, 32 for SMC-D 228 */ 229 u64 peer_token; /* SMC-D token of peer */ 230 u8 killed : 1; /* abnormal termination */ 231 u8 freed : 1; /* normal termiation */ 232 u8 out_of_sync : 1; /* out of sync with peer */ 233 }; 234 235 struct smc_sock { /* smc sock container */ 236 struct sock sk; 237 struct socket *clcsock; /* internal tcp socket */ 238 void (*clcsk_state_change)(struct sock *sk); 239 /* original stat_change fct. */ 240 void (*clcsk_data_ready)(struct sock *sk); 241 /* original data_ready fct. */ 242 void (*clcsk_write_space)(struct sock *sk); 243 /* original write_space fct. */ 244 void (*clcsk_error_report)(struct sock *sk); 245 /* original error_report fct. */ 246 struct smc_connection conn; /* smc connection */ 247 struct smc_sock *listen_smc; /* listen parent */ 248 struct work_struct connect_work; /* handle non-blocking connect*/ 249 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 250 struct work_struct smc_listen_work;/* prepare new accept socket */ 251 struct list_head accept_q; /* sockets to be accepted */ 252 spinlock_t accept_q_lock; /* protects accept_q */ 253 bool limit_smc_hs; /* put constraint on handshake */ 254 bool use_fallback; /* fallback to tcp */ 255 int fallback_rsn; /* reason for fallback */ 256 u32 peer_diagnosis; /* decline reason from peer */ 257 atomic_t queued_smc_hs; /* queued smc handshakes */ 258 struct inet_connection_sock_af_ops af_ops; 259 const struct inet_connection_sock_af_ops *ori_af_ops; 260 /* original af ops */ 261 int sockopt_defer_accept; 262 /* sockopt TCP_DEFER_ACCEPT 263 * value 264 */ 265 u8 wait_close_tx_prepared : 1; 266 /* shutdown wr or close 267 * started, waiting for unsent 268 * data to be sent 269 */ 270 u8 connect_nonblock : 1; 271 /* non-blocking connect in 272 * flight 273 */ 274 struct mutex clcsock_release_lock; 275 /* protects clcsock of a listen 276 * socket 277 * */ 278 }; 279 280 static inline struct smc_sock *smc_sk(const struct sock *sk) 281 { 282 return (struct smc_sock *)sk; 283 } 284 285 static inline struct smc_sock *smc_clcsock_user_data(const struct sock *clcsk) 286 { 287 return (struct smc_sock *) 288 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY); 289 } 290 291 extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 292 extern struct workqueue_struct *smc_close_wq; /* wq for close work */ 293 294 #define SMC_SYSTEMID_LEN 8 295 296 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 297 298 #define ntohll(x) be64_to_cpu(x) 299 #define htonll(x) cpu_to_be64(x) 300 301 /* convert an u32 value into network byte order, store it into a 3 byte field */ 302 static inline void hton24(u8 *net, u32 host) 303 { 304 __be32 t; 305 306 t = cpu_to_be32(host); 307 memcpy(net, ((u8 *)&t) + 1, 3); 308 } 309 310 /* convert a received 3 byte field into host byte order*/ 311 static inline u32 ntoh24(u8 *net) 312 { 313 __be32 t = 0; 314 315 memcpy(((u8 *)&t) + 1, net, 3); 316 return be32_to_cpu(t); 317 } 318 319 #ifdef CONFIG_XFRM 320 static inline bool using_ipsec(struct smc_sock *smc) 321 { 322 return (smc->clcsock->sk->sk_policy[0] || 323 smc->clcsock->sk->sk_policy[1]) ? true : false; 324 } 325 #else 326 static inline bool using_ipsec(struct smc_sock *smc) 327 { 328 return false; 329 } 330 #endif 331 332 struct smc_gidlist; 333 334 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 335 void smc_close_non_accepted(struct sock *sk); 336 void smc_fill_gid_list(struct smc_link_group *lgr, 337 struct smc_gidlist *gidlist, 338 struct smc_ib_device *known_dev, u8 *known_gid); 339 340 /* smc handshake limitation interface for netlink */ 341 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb); 342 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info); 343 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info); 344 345 #endif /* __SMC_H */ 346