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 41 #define RDS_IB_ABI_VERSION 0x301 42 43 #define SOL_RDS 276 44 45 /* 46 * setsockopt/getsockopt for SOL_RDS 47 */ 48 #define RDS_CANCEL_SENT_TO 1 49 #define RDS_GET_MR 2 50 #define RDS_FREE_MR 3 51 /* deprecated: RDS_BARRIER 4 */ 52 #define RDS_RECVERR 5 53 #define RDS_CONG_MONITOR 6 54 #define RDS_GET_MR_FOR_DEST 7 55 #define SO_RDS_TRANSPORT 8 56 57 /* Socket option to tap receive path latency 58 * SO_RDS: SO_RDS_MSG_RXPATH_LATENCY 59 * Format used struct rds_rx_trace_so 60 */ 61 #define SO_RDS_MSG_RXPATH_LATENCY 10 62 63 64 /* supported values for SO_RDS_TRANSPORT */ 65 #define RDS_TRANS_IB 0 66 #define RDS_TRANS_IWARP 1 67 #define RDS_TRANS_TCP 2 68 #define RDS_TRANS_COUNT 3 69 #define RDS_TRANS_NONE (~0) 70 71 /* 72 * Control message types for SOL_RDS. 73 * 74 * CMSG_RDMA_ARGS (sendmsg) 75 * Request a RDMA transfer to/from the specified 76 * memory ranges. 77 * The cmsg_data is a struct rds_rdma_args. 78 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 79 * Kernel informs application about intended 80 * source/destination of a RDMA transfer 81 * RDS_CMSG_RDMA_MAP (sendmsg) 82 * Application asks kernel to map the given 83 * memory range into a IB MR, and send the 84 * R_Key along in an RDS extension header. 85 * The cmsg_data is a struct rds_get_mr_args, 86 * the same as for the GET_MR setsockopt. 87 * RDS_CMSG_RDMA_STATUS (recvmsg) 88 * Returns the status of a completed RDMA operation. 89 * RDS_CMSG_RXPATH_LATENCY(recvmsg) 90 * Returns rds message latencies in various stages of receive 91 * path in nS. Its set per socket using SO_RDS_MSG_RXPATH_LATENCY 92 * socket option. Legitimate points are defined in 93 * enum rds_message_rxpath_latency. More points can be added in 94 * future. CSMG format is struct rds_cmsg_rx_trace. 95 */ 96 #define RDS_CMSG_RDMA_ARGS 1 97 #define RDS_CMSG_RDMA_DEST 2 98 #define RDS_CMSG_RDMA_MAP 3 99 #define RDS_CMSG_RDMA_STATUS 4 100 #define RDS_CMSG_CONG_UPDATE 5 101 #define RDS_CMSG_ATOMIC_FADD 6 102 #define RDS_CMSG_ATOMIC_CSWP 7 103 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 104 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 105 #define RDS_CMSG_RXPATH_LATENCY 11 106 #define RDS_CMSG_ZCOPY_COOKIE 12 107 #define RDS_CMSG_ZCOPY_COMPLETION 13 108 109 #define RDS_INFO_FIRST 10000 110 #define RDS_INFO_COUNTERS 10000 111 #define RDS_INFO_CONNECTIONS 10001 112 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 113 #define RDS_INFO_SEND_MESSAGES 10003 114 #define RDS_INFO_RETRANS_MESSAGES 10004 115 #define RDS_INFO_RECV_MESSAGES 10005 116 #define RDS_INFO_SOCKETS 10006 117 #define RDS_INFO_TCP_SOCKETS 10007 118 #define RDS_INFO_IB_CONNECTIONS 10008 119 #define RDS_INFO_CONNECTION_STATS 10009 120 #define RDS_INFO_IWARP_CONNECTIONS 10010 121 122 /* PF_RDS6 options */ 123 #define RDS6_INFO_CONNECTIONS 10011 124 #define RDS6_INFO_SEND_MESSAGES 10012 125 #define RDS6_INFO_RETRANS_MESSAGES 10013 126 #define RDS6_INFO_RECV_MESSAGES 10014 127 #define RDS6_INFO_SOCKETS 10015 128 #define RDS6_INFO_TCP_SOCKETS 10016 129 #define RDS6_INFO_IB_CONNECTIONS 10017 130 131 #define RDS_INFO_LAST 10017 132 133 struct rds_info_counter { 134 __u8 name[32]; 135 __u64 value; 136 } __attribute__((packed)); 137 138 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 139 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 140 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 141 142 #define TRANSNAMSIZ 16 143 144 struct rds_info_connection { 145 __u64 next_tx_seq; 146 __u64 next_rx_seq; 147 __be32 laddr; 148 __be32 faddr; 149 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 150 __u8 flags; 151 } __attribute__((packed)); 152 153 struct rds6_info_connection { 154 __u64 next_tx_seq; 155 __u64 next_rx_seq; 156 struct in6_addr laddr; 157 struct in6_addr faddr; 158 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 159 __u8 flags; 160 } __attribute__((packed)); 161 162 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 163 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 164 165 struct rds_info_message { 166 __u64 seq; 167 __u32 len; 168 __be32 laddr; 169 __be32 faddr; 170 __be16 lport; 171 __be16 fport; 172 __u8 flags; 173 } __attribute__((packed)); 174 175 struct rds6_info_message { 176 __u64 seq; 177 __u32 len; 178 struct in6_addr laddr; 179 struct in6_addr faddr; 180 __be16 lport; 181 __be16 fport; 182 __u8 flags; 183 __u8 tos; 184 } __attribute__((packed)); 185 186 struct rds_info_socket { 187 __u32 sndbuf; 188 __be32 bound_addr; 189 __be32 connected_addr; 190 __be16 bound_port; 191 __be16 connected_port; 192 __u32 rcvbuf; 193 __u64 inum; 194 } __attribute__((packed)); 195 196 struct rds6_info_socket { 197 __u32 sndbuf; 198 struct in6_addr bound_addr; 199 struct in6_addr connected_addr; 200 __be16 bound_port; 201 __be16 connected_port; 202 __u32 rcvbuf; 203 __u64 inum; 204 } __attribute__((packed)); 205 206 struct rds_info_tcp_socket { 207 __be32 local_addr; 208 __be16 local_port; 209 __be32 peer_addr; 210 __be16 peer_port; 211 __u64 hdr_rem; 212 __u64 data_rem; 213 __u32 last_sent_nxt; 214 __u32 last_expected_una; 215 __u32 last_seen_una; 216 } __attribute__((packed)); 217 218 struct rds6_info_tcp_socket { 219 struct in6_addr local_addr; 220 __be16 local_port; 221 struct in6_addr peer_addr; 222 __be16 peer_port; 223 __u64 hdr_rem; 224 __u64 data_rem; 225 __u32 last_sent_nxt; 226 __u32 last_expected_una; 227 __u32 last_seen_una; 228 } __attribute__((packed)); 229 230 #define RDS_IB_GID_LEN 16 231 struct rds_info_rdma_connection { 232 __be32 src_addr; 233 __be32 dst_addr; 234 __u8 src_gid[RDS_IB_GID_LEN]; 235 __u8 dst_gid[RDS_IB_GID_LEN]; 236 237 __u32 max_send_wr; 238 __u32 max_recv_wr; 239 __u32 max_send_sge; 240 __u32 rdma_mr_max; 241 __u32 rdma_mr_size; 242 }; 243 244 struct rds6_info_rdma_connection { 245 struct in6_addr src_addr; 246 struct in6_addr dst_addr; 247 __u8 src_gid[RDS_IB_GID_LEN]; 248 __u8 dst_gid[RDS_IB_GID_LEN]; 249 250 __u32 max_send_wr; 251 __u32 max_recv_wr; 252 __u32 max_send_sge; 253 __u32 rdma_mr_max; 254 __u32 rdma_mr_size; 255 }; 256 257 /* RDS message Receive Path Latency points */ 258 enum rds_message_rxpath_latency { 259 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 260 RDS_MSG_RX_DGRAM_REASSEMBLE, 261 RDS_MSG_RX_DGRAM_DELIVERED, 262 RDS_MSG_RX_DGRAM_TRACE_MAX 263 }; 264 265 struct rds_rx_trace_so { 266 __u8 rx_traces; 267 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 268 }; 269 270 struct rds_cmsg_rx_trace { 271 __u8 rx_traces; 272 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 273 __u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 274 }; 275 276 /* 277 * Congestion monitoring. 278 * Congestion control in RDS happens at the host connection 279 * level by exchanging a bitmap marking congested ports. 280 * By default, a process sleeping in poll() is always woken 281 * up when the congestion map is updated. 282 * With explicit monitoring, an application can have more 283 * fine-grained control. 284 * The application installs a 64bit mask value in the socket, 285 * where each bit corresponds to a group of ports. 286 * When a congestion update arrives, RDS checks the set of 287 * ports that are now uncongested against the list bit mask 288 * installed in the socket, and if they overlap, we queue a 289 * cong_notification on the socket. 290 * 291 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 292 * with the 64bit mask. 293 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 294 * control messages. 295 * 296 * The correspondence between bits and ports is 297 * 1 << (portnum % 64) 298 */ 299 #define RDS_CONG_MONITOR_SIZE 64 300 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 301 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 302 303 /* 304 * RDMA related types 305 */ 306 307 /* 308 * This encapsulates a remote memory location. 309 * In the current implementation, it contains the R_Key 310 * of the remote memory region, and the offset into it 311 * (so that the application does not have to worry about 312 * alignment). 313 */ 314 typedef __u64 rds_rdma_cookie_t; 315 316 struct rds_iovec { 317 __u64 addr; 318 __u64 bytes; 319 }; 320 321 struct rds_get_mr_args { 322 struct rds_iovec vec; 323 __u64 cookie_addr; 324 __u64 flags; 325 }; 326 327 struct rds_get_mr_for_dest_args { 328 struct __kernel_sockaddr_storage dest_addr; 329 struct rds_iovec vec; 330 __u64 cookie_addr; 331 __u64 flags; 332 }; 333 334 struct rds_free_mr_args { 335 rds_rdma_cookie_t cookie; 336 __u64 flags; 337 }; 338 339 struct rds_rdma_args { 340 rds_rdma_cookie_t cookie; 341 struct rds_iovec remote_vec; 342 __u64 local_vec_addr; 343 __u64 nr_local; 344 __u64 flags; 345 __u64 user_token; 346 }; 347 348 struct rds_atomic_args { 349 rds_rdma_cookie_t cookie; 350 __u64 local_addr; 351 __u64 remote_addr; 352 union { 353 struct { 354 __u64 compare; 355 __u64 swap; 356 } cswp; 357 struct { 358 __u64 add; 359 } fadd; 360 struct { 361 __u64 compare; 362 __u64 swap; 363 __u64 compare_mask; 364 __u64 swap_mask; 365 } m_cswp; 366 struct { 367 __u64 add; 368 __u64 nocarry_mask; 369 } m_fadd; 370 }; 371 __u64 flags; 372 __u64 user_token; 373 }; 374 375 struct rds_rdma_notify { 376 __u64 user_token; 377 __s32 status; 378 }; 379 380 #define RDS_RDMA_SUCCESS 0 381 #define RDS_RDMA_REMOTE_ERROR 1 382 #define RDS_RDMA_CANCELED 2 383 #define RDS_RDMA_DROPPED 3 384 #define RDS_RDMA_OTHER_ERROR 4 385 386 #define RDS_MAX_ZCOOKIES 8 387 struct rds_zcopy_cookies { 388 __u32 num; 389 __u32 cookies[RDS_MAX_ZCOOKIES]; 390 }; 391 392 /* 393 * Common set of flags for all RDMA related structs 394 */ 395 #define RDS_RDMA_READWRITE 0x0001 396 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 397 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 398 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 399 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 400 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 401 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 402 403 #endif /* IB_RDS_H */ 404