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 25 #define SMC_RELEASE_0 0 26 #define SMC_RELEASE_1 1 27 #define SMC_RELEASE SMC_RELEASE_1 /* the latest release version */ 28 29 #define SMCPROTO_SMC 0 /* SMC protocol, IPv4 */ 30 #define SMCPROTO_SMC6 1 /* SMC protocol, IPv6 */ 31 32 #define SMC_AUTOCORKING_DEFAULT_SIZE 0x10000 /* 64K by default */ 33 34 extern struct proto smc_proto; 35 extern struct proto smc_proto6; 36 37 /* smc sock initialization */ 38 void smc_sk_init(struct net *net, struct sock *sk, int protocol); 39 /* clcsock initialization */ 40 int smc_create_clcsk(struct net *net, struct sock *sk, int family); 41 42 #ifdef ATOMIC64_INIT 43 #define KERNEL_HAS_ATOMIC64 44 #endif 45 46 enum smc_state { /* possible states of an SMC socket */ 47 SMC_ACTIVE = 1, 48 SMC_INIT = 2, 49 SMC_CLOSED = 7, 50 SMC_LISTEN = 10, 51 /* normal close */ 52 SMC_PEERCLOSEWAIT1 = 20, 53 SMC_PEERCLOSEWAIT2 = 21, 54 SMC_APPFINCLOSEWAIT = 24, 55 SMC_APPCLOSEWAIT1 = 22, 56 SMC_APPCLOSEWAIT2 = 23, 57 SMC_PEERFINCLOSEWAIT = 25, 58 /* abnormal close */ 59 SMC_PEERABORTWAIT = 26, 60 SMC_PROCESSABORT = 27, 61 }; 62 63 struct smc_link_group; 64 65 struct smc_wr_rx_hdr { /* common prefix part of LLC and CDC to demultiplex */ 66 union { 67 u8 type; 68 #if defined(__BIG_ENDIAN_BITFIELD) 69 struct { 70 u8 llc_version:4, 71 llc_type:4; 72 }; 73 #elif defined(__LITTLE_ENDIAN_BITFIELD) 74 struct { 75 u8 llc_type:4, 76 llc_version:4; 77 }; 78 #endif 79 }; 80 } __aligned(1); 81 82 struct smc_cdc_conn_state_flags { 83 #if defined(__BIG_ENDIAN_BITFIELD) 84 u8 peer_done_writing : 1; /* Sending done indicator */ 85 u8 peer_conn_closed : 1; /* Peer connection closed indicator */ 86 u8 peer_conn_abort : 1; /* Abnormal close indicator */ 87 u8 reserved : 5; 88 #elif defined(__LITTLE_ENDIAN_BITFIELD) 89 u8 reserved : 5; 90 u8 peer_conn_abort : 1; 91 u8 peer_conn_closed : 1; 92 u8 peer_done_writing : 1; 93 #endif 94 }; 95 96 struct smc_cdc_producer_flags { 97 #if defined(__BIG_ENDIAN_BITFIELD) 98 u8 write_blocked : 1; /* Writing Blocked, no rx buf space */ 99 u8 urg_data_pending : 1; /* Urgent Data Pending */ 100 u8 urg_data_present : 1; /* Urgent Data Present */ 101 u8 cons_curs_upd_req : 1; /* cursor update requested */ 102 u8 failover_validation : 1;/* message replay due to failover */ 103 u8 reserved : 3; 104 #elif defined(__LITTLE_ENDIAN_BITFIELD) 105 u8 reserved : 3; 106 u8 failover_validation : 1; 107 u8 cons_curs_upd_req : 1; 108 u8 urg_data_present : 1; 109 u8 urg_data_pending : 1; 110 u8 write_blocked : 1; 111 #endif 112 }; 113 114 /* in host byte order */ 115 union smc_host_cursor { /* SMC cursor - an offset in an RMBE */ 116 struct { 117 u16 reserved; 118 u16 wrap; /* window wrap sequence number */ 119 u32 count; /* cursor (= offset) part */ 120 }; 121 #ifdef KERNEL_HAS_ATOMIC64 122 atomic64_t acurs; /* for atomic processing */ 123 #else 124 u64 acurs; /* for atomic processing */ 125 #endif 126 } __aligned(8); 127 128 /* in host byte order, except for flag bitfields in network byte order */ 129 struct smc_host_cdc_msg { /* Connection Data Control message */ 130 struct smc_wr_rx_hdr common; /* .type = 0xFE */ 131 u8 len; /* length = 44 */ 132 u16 seqno; /* connection seq # */ 133 u32 token; /* alert_token */ 134 union smc_host_cursor prod; /* producer cursor */ 135 union smc_host_cursor cons; /* consumer cursor, 136 * piggy backed "ack" 137 */ 138 struct smc_cdc_producer_flags prod_flags; /* conn. tx/rx status */ 139 struct smc_cdc_conn_state_flags conn_state_flags; /* peer conn. status*/ 140 u8 reserved[18]; 141 } __aligned(8); 142 143 enum smc_urg_state { 144 SMC_URG_VALID = 1, /* data present */ 145 SMC_URG_NOTYET = 2, /* data pending */ 146 SMC_URG_READ = 3, /* data was already read */ 147 }; 148 149 struct smc_mark_woken { 150 bool woken; 151 void *key; 152 wait_queue_entry_t wait_entry; 153 }; 154 155 struct smc_connection { 156 struct rb_node alert_node; 157 struct smc_link_group *lgr; /* link group of connection */ 158 struct smc_link *lnk; /* assigned SMC-R link */ 159 u32 alert_token_local; /* unique conn. id */ 160 u8 peer_rmbe_idx; /* from tcp handshake */ 161 int peer_rmbe_size; /* size of peer rx buffer */ 162 atomic_t peer_rmbe_space;/* remaining free bytes in peer 163 * rmbe 164 */ 165 int rtoken_idx; /* idx to peer RMB rkey/addr */ 166 167 struct smc_buf_desc *sndbuf_desc; /* send buffer descriptor */ 168 struct smc_buf_desc *rmb_desc; /* RMBE descriptor */ 169 int rmbe_size_comp; /* compressed notation */ 170 int rmbe_update_limit; 171 /* lower limit for consumer 172 * cursor update 173 */ 174 175 struct smc_host_cdc_msg local_tx_ctrl; /* host byte order staging 176 * buffer for CDC msg send 177 * .prod cf. TCP snd_nxt 178 * .cons cf. TCP sends ack 179 */ 180 union smc_host_cursor local_tx_ctrl_fin; 181 /* prod crsr - confirmed by peer 182 */ 183 union smc_host_cursor tx_curs_prep; /* tx - prepared data 184 * snd_max..wmem_alloc 185 */ 186 union smc_host_cursor tx_curs_sent; /* tx - sent data 187 * snd_nxt ? 188 */ 189 union smc_host_cursor tx_curs_fin; /* tx - confirmed by peer 190 * snd-wnd-begin ? 191 */ 192 atomic_t sndbuf_space; /* remaining space in sndbuf */ 193 u16 tx_cdc_seq; /* sequence # for CDC send */ 194 u16 tx_cdc_seq_fin; /* sequence # - tx completed */ 195 spinlock_t send_lock; /* protect wr_sends */ 196 atomic_t cdc_pend_tx_wr; /* number of pending tx CDC wqe 197 * - inc when post wqe, 198 * - dec on polled tx cqe 199 */ 200 wait_queue_head_t cdc_pend_tx_wq; /* wakeup on no cdc_pend_tx_wr*/ 201 atomic_t tx_pushing; /* nr_threads trying tx push */ 202 struct delayed_work tx_work; /* retry of smc_cdc_msg_send */ 203 u32 tx_off; /* base offset in peer rmb */ 204 205 struct smc_host_cdc_msg local_rx_ctrl; /* filled during event_handl. 206 * .prod cf. TCP rcv_nxt 207 * .cons cf. TCP snd_una 208 */ 209 union smc_host_cursor rx_curs_confirmed; /* confirmed to peer 210 * source of snd_una ? 211 */ 212 union smc_host_cursor urg_curs; /* points at urgent byte */ 213 enum smc_urg_state urg_state; 214 bool urg_tx_pend; /* urgent data staged */ 215 bool urg_rx_skip_pend; 216 /* indicate urgent oob data 217 * read, but previous regular 218 * data still pending 219 */ 220 char urg_rx_byte; /* urgent byte */ 221 bool tx_in_release_sock; 222 /* flush pending tx data in 223 * sock release_cb() 224 */ 225 atomic_t bytes_to_rcv; /* arrived data, 226 * not yet received 227 */ 228 atomic_t splice_pending; /* number of spliced bytes 229 * pending processing 230 */ 231 #ifndef KERNEL_HAS_ATOMIC64 232 spinlock_t acurs_lock; /* protect cursors */ 233 #endif 234 struct work_struct close_work; /* peer sent some closing */ 235 struct work_struct abort_work; /* abort the connection */ 236 struct tasklet_struct rx_tsklet; /* Receiver tasklet for SMC-D */ 237 u8 rx_off; /* receive offset: 238 * 0 for SMC-R, 32 for SMC-D 239 */ 240 u64 peer_token; /* SMC-D token of peer */ 241 u8 killed : 1; /* abnormal termination */ 242 u8 freed : 1; /* normal termiation */ 243 u8 out_of_sync : 1; /* out of sync with peer */ 244 }; 245 246 struct smc_sock { /* smc sock container */ 247 struct sock sk; 248 struct socket *clcsock; /* internal tcp socket */ 249 void (*clcsk_state_change)(struct sock *sk); 250 /* original stat_change fct. */ 251 void (*clcsk_data_ready)(struct sock *sk); 252 /* original data_ready fct. */ 253 void (*clcsk_write_space)(struct sock *sk); 254 /* original write_space fct. */ 255 void (*clcsk_error_report)(struct sock *sk); 256 /* original error_report fct. */ 257 struct smc_connection conn; /* smc connection */ 258 struct smc_sock *listen_smc; /* listen parent */ 259 struct work_struct connect_work; /* handle non-blocking connect*/ 260 struct work_struct tcp_listen_work;/* handle tcp socket accepts */ 261 struct work_struct smc_listen_work;/* prepare new accept socket */ 262 struct list_head accept_q; /* sockets to be accepted */ 263 spinlock_t accept_q_lock; /* protects accept_q */ 264 bool limit_smc_hs; /* put constraint on handshake */ 265 bool use_fallback; /* fallback to tcp */ 266 int fallback_rsn; /* reason for fallback */ 267 u32 peer_diagnosis; /* decline reason from peer */ 268 atomic_t queued_smc_hs; /* queued smc handshakes */ 269 struct inet_connection_sock_af_ops af_ops; 270 const struct inet_connection_sock_af_ops *ori_af_ops; 271 /* original af ops */ 272 int sockopt_defer_accept; 273 /* sockopt TCP_DEFER_ACCEPT 274 * value 275 */ 276 u8 wait_close_tx_prepared : 1; 277 /* shutdown wr or close 278 * started, waiting for unsent 279 * data to be sent 280 */ 281 u8 connect_nonblock : 1; 282 /* non-blocking connect in 283 * flight 284 */ 285 struct mutex clcsock_release_lock; 286 /* protects clcsock of a listen 287 * socket 288 * */ 289 }; 290 291 #define smc_sk(ptr) container_of_const(ptr, struct smc_sock, sk) 292 293 static inline void smc_init_saved_callbacks(struct smc_sock *smc) 294 { 295 smc->clcsk_state_change = NULL; 296 smc->clcsk_data_ready = NULL; 297 smc->clcsk_write_space = NULL; 298 smc->clcsk_error_report = NULL; 299 } 300 301 static inline struct smc_sock *smc_clcsock_user_data(const struct sock *clcsk) 302 { 303 return (struct smc_sock *) 304 ((uintptr_t)clcsk->sk_user_data & ~SK_USER_DATA_NOCOPY); 305 } 306 307 /* save target_cb in saved_cb, and replace target_cb with new_cb */ 308 static inline void smc_clcsock_replace_cb(void (**target_cb)(struct sock *), 309 void (*new_cb)(struct sock *), 310 void (**saved_cb)(struct sock *)) 311 { 312 /* only save once */ 313 if (!*saved_cb) 314 *saved_cb = *target_cb; 315 *target_cb = new_cb; 316 } 317 318 /* restore target_cb to saved_cb, and reset saved_cb to NULL */ 319 static inline void smc_clcsock_restore_cb(void (**target_cb)(struct sock *), 320 void (**saved_cb)(struct sock *)) 321 { 322 if (!*saved_cb) 323 return; 324 *target_cb = *saved_cb; 325 *saved_cb = NULL; 326 } 327 328 extern struct workqueue_struct *smc_hs_wq; /* wq for handshake work */ 329 extern struct workqueue_struct *smc_close_wq; /* wq for close work */ 330 331 #define SMC_SYSTEMID_LEN 8 332 333 extern u8 local_systemid[SMC_SYSTEMID_LEN]; /* unique system identifier */ 334 335 #define ntohll(x) be64_to_cpu(x) 336 #define htonll(x) cpu_to_be64(x) 337 338 /* convert an u32 value into network byte order, store it into a 3 byte field */ 339 static inline void hton24(u8 *net, u32 host) 340 { 341 __be32 t; 342 343 t = cpu_to_be32(host); 344 memcpy(net, ((u8 *)&t) + 1, 3); 345 } 346 347 /* convert a received 3 byte field into host byte order*/ 348 static inline u32 ntoh24(u8 *net) 349 { 350 __be32 t = 0; 351 352 memcpy(((u8 *)&t) + 1, net, 3); 353 return be32_to_cpu(t); 354 } 355 356 #ifdef CONFIG_XFRM 357 static inline bool using_ipsec(struct smc_sock *smc) 358 { 359 return (smc->clcsock->sk->sk_policy[0] || 360 smc->clcsock->sk->sk_policy[1]) ? true : false; 361 } 362 #else 363 static inline bool using_ipsec(struct smc_sock *smc) 364 { 365 return false; 366 } 367 #endif 368 369 struct smc_gidlist; 370 371 struct sock *smc_accept_dequeue(struct sock *parent, struct socket *new_sock); 372 void smc_close_non_accepted(struct sock *sk); 373 void smc_fill_gid_list(struct smc_link_group *lgr, 374 struct smc_gidlist *gidlist, 375 struct smc_ib_device *known_dev, u8 *known_gid); 376 377 /* smc handshake limitation interface for netlink */ 378 int smc_nl_dump_hs_limitation(struct sk_buff *skb, struct netlink_callback *cb); 379 int smc_nl_enable_hs_limitation(struct sk_buff *skb, struct genl_info *info); 380 int smc_nl_disable_hs_limitation(struct sk_buff *skb, struct genl_info *info); 381 382 static inline void smc_sock_set_flag(struct sock *sk, enum sock_flags flag) 383 { 384 set_bit(flag, &sk->sk_flags); 385 } 386 387 #endif /* __SMC_H */ 388