1 /* SPDX-License-Identifier: ((GPL-2.0 WITH Linux-syscall-note) OR Linux-OpenIB) */ 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 #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 #define RDS_INFO_LAST 10010 122 123 struct rds_info_counter { 124 __u8 name[32]; 125 __u64 value; 126 } __attribute__((packed)); 127 128 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 129 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 130 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 131 132 #define TRANSNAMSIZ 16 133 134 struct rds_info_connection { 135 __u64 next_tx_seq; 136 __u64 next_rx_seq; 137 __be32 laddr; 138 __be32 faddr; 139 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 140 __u8 flags; 141 } __attribute__((packed)); 142 143 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 144 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 145 146 struct rds_info_message { 147 __u64 seq; 148 __u32 len; 149 __be32 laddr; 150 __be32 faddr; 151 __be16 lport; 152 __be16 fport; 153 __u8 flags; 154 } __attribute__((packed)); 155 156 struct rds_info_socket { 157 __u32 sndbuf; 158 __be32 bound_addr; 159 __be32 connected_addr; 160 __be16 bound_port; 161 __be16 connected_port; 162 __u32 rcvbuf; 163 __u64 inum; 164 } __attribute__((packed)); 165 166 struct rds_info_tcp_socket { 167 __be32 local_addr; 168 __be16 local_port; 169 __be32 peer_addr; 170 __be16 peer_port; 171 __u64 hdr_rem; 172 __u64 data_rem; 173 __u32 last_sent_nxt; 174 __u32 last_expected_una; 175 __u32 last_seen_una; 176 } __attribute__((packed)); 177 178 #define RDS_IB_GID_LEN 16 179 struct rds_info_rdma_connection { 180 __be32 src_addr; 181 __be32 dst_addr; 182 __u8 src_gid[RDS_IB_GID_LEN]; 183 __u8 dst_gid[RDS_IB_GID_LEN]; 184 185 __u32 max_send_wr; 186 __u32 max_recv_wr; 187 __u32 max_send_sge; 188 __u32 rdma_mr_max; 189 __u32 rdma_mr_size; 190 }; 191 192 /* RDS message Receive Path Latency points */ 193 enum rds_message_rxpath_latency { 194 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 195 RDS_MSG_RX_DGRAM_REASSEMBLE, 196 RDS_MSG_RX_DGRAM_DELIVERED, 197 RDS_MSG_RX_DGRAM_TRACE_MAX 198 }; 199 200 struct rds_rx_trace_so { 201 __u8 rx_traces; 202 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 203 }; 204 205 struct rds_cmsg_rx_trace { 206 __u8 rx_traces; 207 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 208 __u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 209 }; 210 211 /* 212 * Congestion monitoring. 213 * Congestion control in RDS happens at the host connection 214 * level by exchanging a bitmap marking congested ports. 215 * By default, a process sleeping in poll() is always woken 216 * up when the congestion map is updated. 217 * With explicit monitoring, an application can have more 218 * fine-grained control. 219 * The application installs a 64bit mask value in the socket, 220 * where each bit corresponds to a group of ports. 221 * When a congestion update arrives, RDS checks the set of 222 * ports that are now uncongested against the list bit mask 223 * installed in the socket, and if they overlap, we queue a 224 * cong_notification on the socket. 225 * 226 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 227 * with the 64bit mask. 228 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 229 * control messages. 230 * 231 * The correspondence between bits and ports is 232 * 1 << (portnum % 64) 233 */ 234 #define RDS_CONG_MONITOR_SIZE 64 235 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 236 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 237 238 /* 239 * RDMA related types 240 */ 241 242 /* 243 * This encapsulates a remote memory location. 244 * In the current implementation, it contains the R_Key 245 * of the remote memory region, and the offset into it 246 * (so that the application does not have to worry about 247 * alignment). 248 */ 249 typedef __u64 rds_rdma_cookie_t; 250 251 struct rds_iovec { 252 __u64 addr; 253 __u64 bytes; 254 }; 255 256 struct rds_get_mr_args { 257 struct rds_iovec vec; 258 __u64 cookie_addr; 259 __u64 flags; 260 }; 261 262 struct rds_get_mr_for_dest_args { 263 struct __kernel_sockaddr_storage dest_addr; 264 struct rds_iovec vec; 265 __u64 cookie_addr; 266 __u64 flags; 267 }; 268 269 struct rds_free_mr_args { 270 rds_rdma_cookie_t cookie; 271 __u64 flags; 272 }; 273 274 struct rds_rdma_args { 275 rds_rdma_cookie_t cookie; 276 struct rds_iovec remote_vec; 277 __u64 local_vec_addr; 278 __u64 nr_local; 279 __u64 flags; 280 __u64 user_token; 281 }; 282 283 struct rds_atomic_args { 284 rds_rdma_cookie_t cookie; 285 __u64 local_addr; 286 __u64 remote_addr; 287 union { 288 struct { 289 __u64 compare; 290 __u64 swap; 291 } cswp; 292 struct { 293 __u64 add; 294 } fadd; 295 struct { 296 __u64 compare; 297 __u64 swap; 298 __u64 compare_mask; 299 __u64 swap_mask; 300 } m_cswp; 301 struct { 302 __u64 add; 303 __u64 nocarry_mask; 304 } m_fadd; 305 }; 306 __u64 flags; 307 __u64 user_token; 308 }; 309 310 struct rds_rdma_notify { 311 __u64 user_token; 312 __s32 status; 313 }; 314 315 #define RDS_RDMA_SUCCESS 0 316 #define RDS_RDMA_REMOTE_ERROR 1 317 #define RDS_RDMA_CANCELED 2 318 #define RDS_RDMA_DROPPED 3 319 #define RDS_RDMA_OTHER_ERROR 4 320 321 #define RDS_MAX_ZCOOKIES 8 322 struct rds_zcopy_cookies { 323 __u32 num; 324 __u32 cookies[RDS_MAX_ZCOOKIES]; 325 }; 326 327 /* 328 * Common set of flags for all RDMA related structs 329 */ 330 #define RDS_RDMA_READWRITE 0x0001 331 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 332 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 333 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 334 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 335 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 336 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 337 338 #endif /* IB_RDS_H */ 339