1 /* SPDX-License-Identifier: ((GPL-2.0 WITH Linux-syscall-note) OR Linux-OpenIB) */ 2 /* 3 * Copyright (c) 2008, 2018 Oracle and/or its affiliates. All rights reserved. 4 * 5 * This software is available to you under a choice of one of two 6 * licenses. You may choose to be licensed under the terms of the GNU 7 * General Public License (GPL) Version 2, available from the file 8 * COPYING in the main directory of this source tree, or the 9 * OpenIB.org BSD license below: 10 * 11 * Redistribution and use in source and binary forms, with or 12 * without modification, are permitted provided that the following 13 * conditions are met: 14 * 15 * - Redistributions of source code must retain the above 16 * copyright notice, this list of conditions and the following 17 * disclaimer. 18 * 19 * - Redistributions in binary form must reproduce the above 20 * copyright notice, this list of conditions and the following 21 * disclaimer in the documentation and/or other materials 22 * provided with the distribution. 23 * 24 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, 25 * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF 26 * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND 27 * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS 28 * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN 29 * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN 30 * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE 31 * SOFTWARE. 32 * 33 */ 34 35 #ifndef _LINUX_RDS_H 36 #define _LINUX_RDS_H 37 38 #include <linux/types.h> 39 #include <linux/socket.h> /* For __kernel_sockaddr_storage. */ 40 #include <linux/in6.h> /* For struct in6_addr. */ 41 42 #define RDS_IB_ABI_VERSION 0x301 43 44 #define SOL_RDS 276 45 46 /* 47 * setsockopt/getsockopt for SOL_RDS 48 */ 49 #define RDS_CANCEL_SENT_TO 1 50 #define RDS_GET_MR 2 51 #define RDS_FREE_MR 3 52 /* deprecated: RDS_BARRIER 4 */ 53 #define RDS_RECVERR 5 54 #define RDS_CONG_MONITOR 6 55 #define RDS_GET_MR_FOR_DEST 7 56 #define SO_RDS_TRANSPORT 8 57 58 /* Socket option to tap receive path latency 59 * SO_RDS: SO_RDS_MSG_RXPATH_LATENCY 60 * Format used struct rds_rx_trace_so 61 */ 62 #define SO_RDS_MSG_RXPATH_LATENCY 10 63 64 65 /* supported values for SO_RDS_TRANSPORT */ 66 #define RDS_TRANS_IB 0 67 #define RDS_TRANS_GAP 1 68 #define RDS_TRANS_TCP 2 69 #define RDS_TRANS_COUNT 3 70 #define RDS_TRANS_NONE (~0) 71 /* don't use RDS_TRANS_IWARP - it is deprecated */ 72 #define RDS_TRANS_IWARP RDS_TRANS_GAP 73 74 /* IOCTLS commands for SOL_RDS */ 75 #define SIOCRDSSETTOS (SIOCPROTOPRIVATE) 76 #define SIOCRDSGETTOS (SIOCPROTOPRIVATE + 1) 77 78 typedef uint8_t rds_tos_t; 79 80 /* 81 * Control message types for SOL_RDS. 82 * 83 * CMSG_RDMA_ARGS (sendmsg) 84 * Request a RDMA transfer to/from the specified 85 * memory ranges. 86 * The cmsg_data is a struct rds_rdma_args. 87 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 88 * Kernel informs application about intended 89 * source/destination of a RDMA transfer 90 * RDS_CMSG_RDMA_MAP (sendmsg) 91 * Application asks kernel to map the given 92 * memory range into a IB MR, and send the 93 * R_Key along in an RDS extension header. 94 * The cmsg_data is a struct rds_get_mr_args, 95 * the same as for the GET_MR setsockopt. 96 * RDS_CMSG_RDMA_STATUS (recvmsg) 97 * Returns the status of a completed RDMA operation. 98 * RDS_CMSG_RXPATH_LATENCY(recvmsg) 99 * Returns rds message latencies in various stages of receive 100 * path in nS. Its set per socket using SO_RDS_MSG_RXPATH_LATENCY 101 * socket option. Legitimate points are defined in 102 * enum rds_message_rxpath_latency. More points can be added in 103 * future. CSMG format is struct rds_cmsg_rx_trace. 104 */ 105 #define RDS_CMSG_RDMA_ARGS 1 106 #define RDS_CMSG_RDMA_DEST 2 107 #define RDS_CMSG_RDMA_MAP 3 108 #define RDS_CMSG_RDMA_STATUS 4 109 #define RDS_CMSG_CONG_UPDATE 5 110 #define RDS_CMSG_ATOMIC_FADD 6 111 #define RDS_CMSG_ATOMIC_CSWP 7 112 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 113 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 114 #define RDS_CMSG_RXPATH_LATENCY 11 115 #define RDS_CMSG_ZCOPY_COOKIE 12 116 #define RDS_CMSG_ZCOPY_COMPLETION 13 117 118 #define RDS_INFO_FIRST 10000 119 #define RDS_INFO_COUNTERS 10000 120 #define RDS_INFO_CONNECTIONS 10001 121 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 122 #define RDS_INFO_SEND_MESSAGES 10003 123 #define RDS_INFO_RETRANS_MESSAGES 10004 124 #define RDS_INFO_RECV_MESSAGES 10005 125 #define RDS_INFO_SOCKETS 10006 126 #define RDS_INFO_TCP_SOCKETS 10007 127 #define RDS_INFO_IB_CONNECTIONS 10008 128 #define RDS_INFO_CONNECTION_STATS 10009 129 #define RDS_INFO_IWARP_CONNECTIONS 10010 130 131 /* PF_RDS6 options */ 132 #define RDS6_INFO_CONNECTIONS 10011 133 #define RDS6_INFO_SEND_MESSAGES 10012 134 #define RDS6_INFO_RETRANS_MESSAGES 10013 135 #define RDS6_INFO_RECV_MESSAGES 10014 136 #define RDS6_INFO_SOCKETS 10015 137 #define RDS6_INFO_TCP_SOCKETS 10016 138 #define RDS6_INFO_IB_CONNECTIONS 10017 139 140 #define RDS_INFO_LAST 10017 141 142 struct rds_info_counter { 143 uint8_t name[32]; 144 uint64_t value; 145 } __attribute__((packed)); 146 147 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 148 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 149 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 150 151 #define TRANSNAMSIZ 16 152 153 struct rds_info_connection { 154 uint64_t next_tx_seq; 155 uint64_t next_rx_seq; 156 uint32_t laddr; 157 uint32_t faddr; 158 uint8_t transport[TRANSNAMSIZ]; /* null term ascii */ 159 uint8_t flags; 160 uint8_t tos; 161 } __attribute__((packed)); 162 163 struct rds6_info_connection { 164 uint64_t next_tx_seq; 165 uint64_t next_rx_seq; 166 struct in6_addr laddr; 167 struct in6_addr faddr; 168 uint8_t transport[TRANSNAMSIZ]; /* null term ascii */ 169 uint8_t flags; 170 } __attribute__((packed)); 171 172 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 173 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 174 175 struct rds_info_message { 176 uint64_t seq; 177 uint32_t len; 178 uint32_t laddr; 179 uint32_t faddr; 180 uint16_t lport; 181 uint16_t fport; 182 uint8_t flags; 183 uint8_t tos; 184 } __attribute__((packed)); 185 186 struct rds6_info_message { 187 uint64_t seq; 188 uint32_t len; 189 struct in6_addr laddr; 190 struct in6_addr faddr; 191 uint16_t lport; 192 uint16_t fport; 193 uint8_t flags; 194 uint8_t tos; 195 } __attribute__((packed)); 196 197 struct rds_info_socket { 198 uint32_t sndbuf; 199 uint32_t bound_addr; 200 uint32_t connected_addr; 201 uint16_t bound_port; 202 uint16_t connected_port; 203 uint32_t rcvbuf; 204 uint64_t inum; 205 } __attribute__((packed)); 206 207 struct rds6_info_socket { 208 uint32_t sndbuf; 209 struct in6_addr bound_addr; 210 struct in6_addr connected_addr; 211 uint16_t bound_port; 212 uint16_t connected_port; 213 uint32_t rcvbuf; 214 uint64_t inum; 215 } __attribute__((packed)); 216 217 struct rds_info_tcp_socket { 218 uint32_t local_addr; 219 uint16_t local_port; 220 uint32_t peer_addr; 221 uint16_t peer_port; 222 uint64_t hdr_rem; 223 uint64_t data_rem; 224 uint32_t last_sent_nxt; 225 uint32_t last_expected_una; 226 uint32_t last_seen_una; 227 uint8_t tos; 228 } __attribute__((packed)); 229 230 struct rds6_info_tcp_socket { 231 struct in6_addr local_addr; 232 uint16_t local_port; 233 struct in6_addr peer_addr; 234 uint16_t peer_port; 235 uint64_t hdr_rem; 236 uint64_t data_rem; 237 uint32_t last_sent_nxt; 238 uint32_t last_expected_una; 239 uint32_t last_seen_una; 240 } __attribute__((packed)); 241 242 #define RDS_IB_GID_LEN 16 243 struct rds_info_rdma_connection { 244 uint32_t src_addr; 245 uint32_t dst_addr; 246 uint8_t src_gid[RDS_IB_GID_LEN]; 247 uint8_t dst_gid[RDS_IB_GID_LEN]; 248 249 uint32_t max_send_wr; 250 uint32_t max_recv_wr; 251 uint32_t max_send_sge; 252 uint32_t rdma_mr_max; 253 uint32_t rdma_mr_size; 254 uint8_t tos; 255 uint8_t sl; 256 uint32_t cache_allocs; 257 }; 258 259 struct rds6_info_rdma_connection { 260 struct in6_addr src_addr; 261 struct in6_addr dst_addr; 262 uint8_t src_gid[RDS_IB_GID_LEN]; 263 uint8_t dst_gid[RDS_IB_GID_LEN]; 264 265 uint32_t max_send_wr; 266 uint32_t max_recv_wr; 267 uint32_t max_send_sge; 268 uint32_t rdma_mr_max; 269 uint32_t rdma_mr_size; 270 uint8_t tos; 271 uint8_t sl; 272 uint32_t cache_allocs; 273 }; 274 275 /* RDS message Receive Path Latency points */ 276 enum rds_message_rxpath_latency { 277 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 278 RDS_MSG_RX_DGRAM_REASSEMBLE, 279 RDS_MSG_RX_DGRAM_DELIVERED, 280 RDS_MSG_RX_DGRAM_TRACE_MAX 281 }; 282 283 struct rds_rx_trace_so { 284 uint8_t rx_traces; 285 uint8_t rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 286 }; 287 288 struct rds_cmsg_rx_trace { 289 uint8_t rx_traces; 290 uint8_t rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 291 uint64_t rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 292 }; 293 294 /* 295 * Congestion monitoring. 296 * Congestion control in RDS happens at the host connection 297 * level by exchanging a bitmap marking congested ports. 298 * By default, a process sleeping in poll() is always woken 299 * up when the congestion map is updated. 300 * With explicit monitoring, an application can have more 301 * fine-grained control. 302 * The application installs a 64bit mask value in the socket, 303 * where each bit corresponds to a group of ports. 304 * When a congestion update arrives, RDS checks the set of 305 * ports that are now uncongested against the list bit mask 306 * installed in the socket, and if they overlap, we queue a 307 * cong_notification on the socket. 308 * 309 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 310 * with the 64bit mask. 311 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 312 * control messages. 313 * 314 * The correspondence between bits and ports is 315 * 1 << (portnum % 64) 316 */ 317 #define RDS_CONG_MONITOR_SIZE 64 318 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 319 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 320 321 /* 322 * RDMA related types 323 */ 324 325 /* 326 * This encapsulates a remote memory location. 327 * In the current implementation, it contains the R_Key 328 * of the remote memory region, and the offset into it 329 * (so that the application does not have to worry about 330 * alignment). 331 */ 332 typedef uint64_t rds_rdma_cookie_t; 333 334 struct rds_iovec { 335 uint64_t addr; 336 uint64_t bytes; 337 }; 338 339 struct rds_get_mr_args { 340 struct rds_iovec vec; 341 uint64_t cookie_addr; 342 uint64_t flags; 343 }; 344 345 struct rds_get_mr_for_dest_args { 346 struct __kernel_sockaddr_storage dest_addr; 347 struct rds_iovec vec; 348 uint64_t cookie_addr; 349 uint64_t flags; 350 }; 351 352 struct rds_free_mr_args { 353 rds_rdma_cookie_t cookie; 354 uint64_t flags; 355 }; 356 357 struct rds_rdma_args { 358 rds_rdma_cookie_t cookie; 359 struct rds_iovec remote_vec; 360 uint64_t local_vec_addr; 361 uint64_t nr_local; 362 uint64_t flags; 363 uint64_t user_token; 364 }; 365 366 struct rds_atomic_args { 367 rds_rdma_cookie_t cookie; 368 uint64_t local_addr; 369 uint64_t remote_addr; 370 union { 371 struct { 372 uint64_t compare; 373 uint64_t swap; 374 } cswp; 375 struct { 376 uint64_t add; 377 } fadd; 378 struct { 379 uint64_t compare; 380 uint64_t swap; 381 uint64_t compare_mask; 382 uint64_t swap_mask; 383 } m_cswp; 384 struct { 385 uint64_t add; 386 uint64_t nocarry_mask; 387 } m_fadd; 388 }; 389 uint64_t flags; 390 uint64_t user_token; 391 }; 392 393 struct rds_rdma_notify { 394 uint64_t user_token; 395 int32_t status; 396 }; 397 398 #define RDS_RDMA_SUCCESS 0 399 #define RDS_RDMA_REMOTE_ERROR 1 400 #define RDS_RDMA_CANCELED 2 401 #define RDS_RDMA_DROPPED 3 402 #define RDS_RDMA_OTHER_ERROR 4 403 404 #define RDS_MAX_ZCOOKIES 8 405 struct rds_zcopy_cookies { 406 uint32_t num; 407 uint32_t cookies[RDS_MAX_ZCOOKIES]; 408 }; 409 410 /* 411 * Common set of flags for all RDMA related structs 412 */ 413 #define RDS_RDMA_READWRITE 0x0001 414 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 415 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 416 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 417 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 418 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 419 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 420 421 #endif /* IB_RDS_H */ 422