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_IWARP 1 68 #define RDS_TRANS_TCP 2 69 #define RDS_TRANS_COUNT 3 70 #define RDS_TRANS_NONE (~0) 71 72 /* 73 * Control message types for SOL_RDS. 74 * 75 * CMSG_RDMA_ARGS (sendmsg) 76 * Request a RDMA transfer to/from the specified 77 * memory ranges. 78 * The cmsg_data is a struct rds_rdma_args. 79 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 80 * Kernel informs application about intended 81 * source/destination of a RDMA transfer 82 * RDS_CMSG_RDMA_MAP (sendmsg) 83 * Application asks kernel to map the given 84 * memory range into a IB MR, and send the 85 * R_Key along in an RDS extension header. 86 * The cmsg_data is a struct rds_get_mr_args, 87 * the same as for the GET_MR setsockopt. 88 * RDS_CMSG_RDMA_STATUS (recvmsg) 89 * Returns the status of a completed RDMA operation. 90 * RDS_CMSG_RXPATH_LATENCY(recvmsg) 91 * Returns rds message latencies in various stages of receive 92 * path in nS. Its set per socket using SO_RDS_MSG_RXPATH_LATENCY 93 * socket option. Legitimate points are defined in 94 * enum rds_message_rxpath_latency. More points can be added in 95 * future. CSMG format is struct rds_cmsg_rx_trace. 96 */ 97 #define RDS_CMSG_RDMA_ARGS 1 98 #define RDS_CMSG_RDMA_DEST 2 99 #define RDS_CMSG_RDMA_MAP 3 100 #define RDS_CMSG_RDMA_STATUS 4 101 #define RDS_CMSG_CONG_UPDATE 5 102 #define RDS_CMSG_ATOMIC_FADD 6 103 #define RDS_CMSG_ATOMIC_CSWP 7 104 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 105 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 106 #define RDS_CMSG_RXPATH_LATENCY 11 107 #define RDS_CMSG_ZCOPY_COOKIE 12 108 #define RDS_CMSG_ZCOPY_COMPLETION 13 109 110 #define RDS_INFO_FIRST 10000 111 #define RDS_INFO_COUNTERS 10000 112 #define RDS_INFO_CONNECTIONS 10001 113 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 114 #define RDS_INFO_SEND_MESSAGES 10003 115 #define RDS_INFO_RETRANS_MESSAGES 10004 116 #define RDS_INFO_RECV_MESSAGES 10005 117 #define RDS_INFO_SOCKETS 10006 118 #define RDS_INFO_TCP_SOCKETS 10007 119 #define RDS_INFO_IB_CONNECTIONS 10008 120 #define RDS_INFO_CONNECTION_STATS 10009 121 #define RDS_INFO_IWARP_CONNECTIONS 10010 122 123 /* PF_RDS6 options */ 124 #define RDS6_INFO_CONNECTIONS 10011 125 #define RDS6_INFO_SEND_MESSAGES 10012 126 #define RDS6_INFO_RETRANS_MESSAGES 10013 127 #define RDS6_INFO_RECV_MESSAGES 10014 128 #define RDS6_INFO_SOCKETS 10015 129 #define RDS6_INFO_TCP_SOCKETS 10016 130 #define RDS6_INFO_IB_CONNECTIONS 10017 131 132 #define RDS_INFO_LAST 10017 133 134 struct rds_info_counter { 135 __u8 name[32]; 136 __u64 value; 137 } __attribute__((packed)); 138 139 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 140 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 141 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 142 143 #define TRANSNAMSIZ 16 144 145 struct rds_info_connection { 146 __u64 next_tx_seq; 147 __u64 next_rx_seq; 148 __be32 laddr; 149 __be32 faddr; 150 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 151 __u8 flags; 152 } __attribute__((packed)); 153 154 struct rds6_info_connection { 155 __u64 next_tx_seq; 156 __u64 next_rx_seq; 157 struct in6_addr laddr; 158 struct in6_addr faddr; 159 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 160 __u8 flags; 161 } __attribute__((packed)); 162 163 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 164 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 165 166 struct rds_info_message { 167 __u64 seq; 168 __u32 len; 169 __be32 laddr; 170 __be32 faddr; 171 __be16 lport; 172 __be16 fport; 173 __u8 flags; 174 } __attribute__((packed)); 175 176 struct rds6_info_message { 177 __u64 seq; 178 __u32 len; 179 struct in6_addr laddr; 180 struct in6_addr faddr; 181 __be16 lport; 182 __be16 fport; 183 __u8 flags; 184 __u8 tos; 185 } __attribute__((packed)); 186 187 struct rds_info_socket { 188 __u32 sndbuf; 189 __be32 bound_addr; 190 __be32 connected_addr; 191 __be16 bound_port; 192 __be16 connected_port; 193 __u32 rcvbuf; 194 __u64 inum; 195 } __attribute__((packed)); 196 197 struct rds6_info_socket { 198 __u32 sndbuf; 199 struct in6_addr bound_addr; 200 struct in6_addr connected_addr; 201 __be16 bound_port; 202 __be16 connected_port; 203 __u32 rcvbuf; 204 __u64 inum; 205 } __attribute__((packed)); 206 207 struct rds_info_tcp_socket { 208 __be32 local_addr; 209 __be16 local_port; 210 __be32 peer_addr; 211 __be16 peer_port; 212 __u64 hdr_rem; 213 __u64 data_rem; 214 __u32 last_sent_nxt; 215 __u32 last_expected_una; 216 __u32 last_seen_una; 217 } __attribute__((packed)); 218 219 struct rds6_info_tcp_socket { 220 struct in6_addr local_addr; 221 __be16 local_port; 222 struct in6_addr peer_addr; 223 __be16 peer_port; 224 __u64 hdr_rem; 225 __u64 data_rem; 226 __u32 last_sent_nxt; 227 __u32 last_expected_una; 228 __u32 last_seen_una; 229 } __attribute__((packed)); 230 231 #define RDS_IB_GID_LEN 16 232 struct rds_info_rdma_connection { 233 __be32 src_addr; 234 __be32 dst_addr; 235 __u8 src_gid[RDS_IB_GID_LEN]; 236 __u8 dst_gid[RDS_IB_GID_LEN]; 237 238 __u32 max_send_wr; 239 __u32 max_recv_wr; 240 __u32 max_send_sge; 241 __u32 rdma_mr_max; 242 __u32 rdma_mr_size; 243 }; 244 245 struct rds6_info_rdma_connection { 246 struct in6_addr src_addr; 247 struct in6_addr dst_addr; 248 __u8 src_gid[RDS_IB_GID_LEN]; 249 __u8 dst_gid[RDS_IB_GID_LEN]; 250 251 __u32 max_send_wr; 252 __u32 max_recv_wr; 253 __u32 max_send_sge; 254 __u32 rdma_mr_max; 255 __u32 rdma_mr_size; 256 }; 257 258 /* RDS message Receive Path Latency points */ 259 enum rds_message_rxpath_latency { 260 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 261 RDS_MSG_RX_DGRAM_REASSEMBLE, 262 RDS_MSG_RX_DGRAM_DELIVERED, 263 RDS_MSG_RX_DGRAM_TRACE_MAX 264 }; 265 266 struct rds_rx_trace_so { 267 __u8 rx_traces; 268 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 269 }; 270 271 struct rds_cmsg_rx_trace { 272 __u8 rx_traces; 273 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 274 __u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 275 }; 276 277 /* 278 * Congestion monitoring. 279 * Congestion control in RDS happens at the host connection 280 * level by exchanging a bitmap marking congested ports. 281 * By default, a process sleeping in poll() is always woken 282 * up when the congestion map is updated. 283 * With explicit monitoring, an application can have more 284 * fine-grained control. 285 * The application installs a 64bit mask value in the socket, 286 * where each bit corresponds to a group of ports. 287 * When a congestion update arrives, RDS checks the set of 288 * ports that are now uncongested against the list bit mask 289 * installed in the socket, and if they overlap, we queue a 290 * cong_notification on the socket. 291 * 292 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 293 * with the 64bit mask. 294 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 295 * control messages. 296 * 297 * The correspondence between bits and ports is 298 * 1 << (portnum % 64) 299 */ 300 #define RDS_CONG_MONITOR_SIZE 64 301 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 302 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 303 304 /* 305 * RDMA related types 306 */ 307 308 /* 309 * This encapsulates a remote memory location. 310 * In the current implementation, it contains the R_Key 311 * of the remote memory region, and the offset into it 312 * (so that the application does not have to worry about 313 * alignment). 314 */ 315 typedef __u64 rds_rdma_cookie_t; 316 317 struct rds_iovec { 318 __u64 addr; 319 __u64 bytes; 320 }; 321 322 struct rds_get_mr_args { 323 struct rds_iovec vec; 324 __u64 cookie_addr; 325 __u64 flags; 326 }; 327 328 struct rds_get_mr_for_dest_args { 329 struct __kernel_sockaddr_storage dest_addr; 330 struct rds_iovec vec; 331 __u64 cookie_addr; 332 __u64 flags; 333 }; 334 335 struct rds_free_mr_args { 336 rds_rdma_cookie_t cookie; 337 __u64 flags; 338 }; 339 340 struct rds_rdma_args { 341 rds_rdma_cookie_t cookie; 342 struct rds_iovec remote_vec; 343 __u64 local_vec_addr; 344 __u64 nr_local; 345 __u64 flags; 346 __u64 user_token; 347 }; 348 349 struct rds_atomic_args { 350 rds_rdma_cookie_t cookie; 351 __u64 local_addr; 352 __u64 remote_addr; 353 union { 354 struct { 355 __u64 compare; 356 __u64 swap; 357 } cswp; 358 struct { 359 __u64 add; 360 } fadd; 361 struct { 362 __u64 compare; 363 __u64 swap; 364 __u64 compare_mask; 365 __u64 swap_mask; 366 } m_cswp; 367 struct { 368 __u64 add; 369 __u64 nocarry_mask; 370 } m_fadd; 371 }; 372 __u64 flags; 373 __u64 user_token; 374 }; 375 376 struct rds_rdma_notify { 377 __u64 user_token; 378 __s32 status; 379 }; 380 381 #define RDS_RDMA_SUCCESS 0 382 #define RDS_RDMA_REMOTE_ERROR 1 383 #define RDS_RDMA_CANCELED 2 384 #define RDS_RDMA_DROPPED 3 385 #define RDS_RDMA_OTHER_ERROR 4 386 387 #define RDS_MAX_ZCOOKIES 8 388 struct rds_zcopy_cookies { 389 __u32 num; 390 __u32 cookies[RDS_MAX_ZCOOKIES]; 391 }; 392 393 /* 394 * Common set of flags for all RDMA related structs 395 */ 396 #define RDS_RDMA_READWRITE 0x0001 397 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 398 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 399 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 400 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 401 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 402 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 403 404 #endif /* IB_RDS_H */ 405