1/* SPDX-License-Identifier: ((GPL-2.0 WITH Linux-syscall-note) OR BSD-2-Clause) */ 2/* 3 * Copyright (c) 2008 Oracle. 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 11 106 107#define RDS_INFO_FIRST 10000 108#define RDS_INFO_COUNTERS 10000 109#define RDS_INFO_CONNECTIONS 10001 110/* 10002 aka RDS_INFO_FLOWS is deprecated */ 111#define RDS_INFO_SEND_MESSAGES 10003 112#define RDS_INFO_RETRANS_MESSAGES 10004 113#define RDS_INFO_RECV_MESSAGES 10005 114#define RDS_INFO_SOCKETS 10006 115#define RDS_INFO_TCP_SOCKETS 10007 116#define RDS_INFO_IB_CONNECTIONS 10008 117#define RDS_INFO_CONNECTION_STATS 10009 118#define RDS_INFO_IWARP_CONNECTIONS 10010 119#define RDS_INFO_LAST 10010 120 121struct rds_info_counter { 122 __u8 name[32]; 123 __u64 value; 124} __attribute__((packed)); 125 126#define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 127#define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 128#define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 129 130#define TRANSNAMSIZ 16 131 132struct rds_info_connection { 133 __u64 next_tx_seq; 134 __u64 next_rx_seq; 135 __be32 laddr; 136 __be32 faddr; 137 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 138 __u8 flags; 139} __attribute__((packed)); 140 141#define RDS_INFO_MESSAGE_FLAG_ACK 0x01 142#define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 143 144struct rds_info_message { 145 __u64 seq; 146 __u32 len; 147 __be32 laddr; 148 __be32 faddr; 149 __be16 lport; 150 __be16 fport; 151 __u8 flags; 152} __attribute__((packed)); 153 154struct rds_info_socket { 155 __u32 sndbuf; 156 __be32 bound_addr; 157 __be32 connected_addr; 158 __be16 bound_port; 159 __be16 connected_port; 160 __u32 rcvbuf; 161 __u64 inum; 162} __attribute__((packed)); 163 164struct rds_info_tcp_socket { 165 __be32 local_addr; 166 __be16 local_port; 167 __be32 peer_addr; 168 __be16 peer_port; 169 __u64 hdr_rem; 170 __u64 data_rem; 171 __u32 last_sent_nxt; 172 __u32 last_expected_una; 173 __u32 last_seen_una; 174} __attribute__((packed)); 175 176#define RDS_IB_GID_LEN 16 177struct rds_info_rdma_connection { 178 __be32 src_addr; 179 __be32 dst_addr; 180 __u8 src_gid[RDS_IB_GID_LEN]; 181 __u8 dst_gid[RDS_IB_GID_LEN]; 182 183 __u32 max_send_wr; 184 __u32 max_recv_wr; 185 __u32 max_send_sge; 186 __u32 rdma_mr_max; 187 __u32 rdma_mr_size; 188}; 189 190/* RDS message Receive Path Latency points */ 191enum rds_message_rxpath_latency { 192 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 193 RDS_MSG_RX_DGRAM_REASSEMBLE, 194 RDS_MSG_RX_DGRAM_DELIVERED, 195 RDS_MSG_RX_DGRAM_TRACE_MAX 196}; 197 198struct rds_rx_trace_so { 199 __u8 rx_traces; 200 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 201}; 202 203struct rds_cmsg_rx_trace { 204 __u8 rx_traces; 205 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 206 __u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 207}; 208 209/* 210 * Congestion monitoring. 211 * Congestion control in RDS happens at the host connection 212 * level by exchanging a bitmap marking congested ports. 213 * By default, a process sleeping in poll() is always woken 214 * up when the congestion map is updated. 215 * With explicit monitoring, an application can have more 216 * fine-grained control. 217 * The application installs a 64bit mask value in the socket, 218 * where each bit corresponds to a group of ports. 219 * When a congestion update arrives, RDS checks the set of 220 * ports that are now uncongested against the list bit mask 221 * installed in the socket, and if they overlap, we queue a 222 * cong_notification on the socket. 223 * 224 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 225 * with the 64bit mask. 226 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 227 * control messages. 228 * 229 * The correspondence between bits and ports is 230 * 1 << (portnum % 64) 231 */ 232#define RDS_CONG_MONITOR_SIZE 64 233#define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 234#define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 235 236/* 237 * RDMA related types 238 */ 239 240/* 241 * This encapsulates a remote memory location. 242 * In the current implementation, it contains the R_Key 243 * of the remote memory region, and the offset into it 244 * (so that the application does not have to worry about 245 * alignment). 246 */ 247typedef __u64 rds_rdma_cookie_t; 248 249struct rds_iovec { 250 __u64 addr; 251 __u64 bytes; 252}; 253 254struct rds_get_mr_args { 255 struct rds_iovec vec; 256 __u64 cookie_addr; 257 __u64 flags; 258}; 259 260struct rds_get_mr_for_dest_args { 261 struct __kernel_sockaddr_storage dest_addr; 262 struct rds_iovec vec; 263 __u64 cookie_addr; 264 __u64 flags; 265}; 266 267struct rds_free_mr_args { 268 rds_rdma_cookie_t cookie; 269 __u64 flags; 270}; 271 272struct rds_rdma_args { 273 rds_rdma_cookie_t cookie; 274 struct rds_iovec remote_vec; 275 __u64 local_vec_addr; 276 __u64 nr_local; 277 __u64 flags; 278 __u64 user_token; 279}; 280 281struct rds_atomic_args { 282 rds_rdma_cookie_t cookie; 283 __u64 local_addr; 284 __u64 remote_addr; 285 union { 286 struct { 287 __u64 compare; 288 __u64 swap; 289 } cswp; 290 struct { 291 __u64 add; 292 } fadd; 293 struct { 294 __u64 compare; 295 __u64 swap; 296 __u64 compare_mask; 297 __u64 swap_mask; 298 } m_cswp; 299 struct { 300 __u64 add; 301 __u64 nocarry_mask; 302 } m_fadd; 303 }; 304 __u64 flags; 305 __u64 user_token; 306}; 307 308struct rds_rdma_notify { 309 __u64 user_token; 310 __s32 status; 311}; 312 313#define RDS_RDMA_SUCCESS 0 314#define RDS_RDMA_REMOTE_ERROR 1 315#define RDS_RDMA_CANCELED 2 316#define RDS_RDMA_DROPPED 3 317#define RDS_RDMA_OTHER_ERROR 4 318 319/* 320 * Common set of flags for all RDMA related structs 321 */ 322#define RDS_RDMA_READWRITE 0x0001 323#define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 324#define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 325#define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 326#define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 327#define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 328#define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 329 330#endif /* IB_RDS_H */ 331