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 /* IOCTLS commands for SOL_RDS */ 73 #define SIOCRDSSETTOS (SIOCPROTOPRIVATE) 74 #define SIOCRDSGETTOS (SIOCPROTOPRIVATE + 1) 75 76 typedef __u8 rds_tos_t; 77 78 /* 79 * Control message types for SOL_RDS. 80 * 81 * CMSG_RDMA_ARGS (sendmsg) 82 * Request a RDMA transfer to/from the specified 83 * memory ranges. 84 * The cmsg_data is a struct rds_rdma_args. 85 * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg) 86 * Kernel informs application about intended 87 * source/destination of a RDMA transfer 88 * RDS_CMSG_RDMA_MAP (sendmsg) 89 * Application asks kernel to map the given 90 * memory range into a IB MR, and send the 91 * R_Key along in an RDS extension header. 92 * The cmsg_data is a struct rds_get_mr_args, 93 * the same as for the GET_MR setsockopt. 94 * RDS_CMSG_RDMA_STATUS (recvmsg) 95 * Returns the status of a completed RDMA operation. 96 * RDS_CMSG_RXPATH_LATENCY(recvmsg) 97 * Returns rds message latencies in various stages of receive 98 * path in nS. Its set per socket using SO_RDS_MSG_RXPATH_LATENCY 99 * socket option. Legitimate points are defined in 100 * enum rds_message_rxpath_latency. More points can be added in 101 * future. CSMG format is struct rds_cmsg_rx_trace. 102 */ 103 #define RDS_CMSG_RDMA_ARGS 1 104 #define RDS_CMSG_RDMA_DEST 2 105 #define RDS_CMSG_RDMA_MAP 3 106 #define RDS_CMSG_RDMA_STATUS 4 107 #define RDS_CMSG_CONG_UPDATE 5 108 #define RDS_CMSG_ATOMIC_FADD 6 109 #define RDS_CMSG_ATOMIC_CSWP 7 110 #define RDS_CMSG_MASKED_ATOMIC_FADD 8 111 #define RDS_CMSG_MASKED_ATOMIC_CSWP 9 112 #define RDS_CMSG_RXPATH_LATENCY 11 113 #define RDS_CMSG_ZCOPY_COOKIE 12 114 #define RDS_CMSG_ZCOPY_COMPLETION 13 115 116 #define RDS_INFO_FIRST 10000 117 #define RDS_INFO_COUNTERS 10000 118 #define RDS_INFO_CONNECTIONS 10001 119 /* 10002 aka RDS_INFO_FLOWS is deprecated */ 120 #define RDS_INFO_SEND_MESSAGES 10003 121 #define RDS_INFO_RETRANS_MESSAGES 10004 122 #define RDS_INFO_RECV_MESSAGES 10005 123 #define RDS_INFO_SOCKETS 10006 124 #define RDS_INFO_TCP_SOCKETS 10007 125 #define RDS_INFO_IB_CONNECTIONS 10008 126 #define RDS_INFO_CONNECTION_STATS 10009 127 #define RDS_INFO_IWARP_CONNECTIONS 10010 128 129 /* PF_RDS6 options */ 130 #define RDS6_INFO_CONNECTIONS 10011 131 #define RDS6_INFO_SEND_MESSAGES 10012 132 #define RDS6_INFO_RETRANS_MESSAGES 10013 133 #define RDS6_INFO_RECV_MESSAGES 10014 134 #define RDS6_INFO_SOCKETS 10015 135 #define RDS6_INFO_TCP_SOCKETS 10016 136 #define RDS6_INFO_IB_CONNECTIONS 10017 137 138 #define RDS_INFO_LAST 10017 139 140 struct rds_info_counter { 141 __u8 name[32]; 142 __u64 value; 143 } __attribute__((packed)); 144 145 #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01 146 #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02 147 #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04 148 149 #define TRANSNAMSIZ 16 150 151 struct rds_info_connection { 152 __u64 next_tx_seq; 153 __u64 next_rx_seq; 154 __be32 laddr; 155 __be32 faddr; 156 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 157 __u8 flags; 158 __u8 tos; 159 } __attribute__((packed)); 160 161 struct rds6_info_connection { 162 __u64 next_tx_seq; 163 __u64 next_rx_seq; 164 struct in6_addr laddr; 165 struct in6_addr faddr; 166 __u8 transport[TRANSNAMSIZ]; /* null term ascii */ 167 __u8 flags; 168 } __attribute__((packed)); 169 170 #define RDS_INFO_MESSAGE_FLAG_ACK 0x01 171 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02 172 173 struct rds_info_message { 174 __u64 seq; 175 __u32 len; 176 __be32 laddr; 177 __be32 faddr; 178 __be16 lport; 179 __be16 fport; 180 __u8 flags; 181 __u8 tos; 182 } __attribute__((packed)); 183 184 struct rds6_info_message { 185 __u64 seq; 186 __u32 len; 187 struct in6_addr laddr; 188 struct in6_addr faddr; 189 __be16 lport; 190 __be16 fport; 191 __u8 flags; 192 __u8 tos; 193 } __attribute__((packed)); 194 195 struct rds_info_socket { 196 __u32 sndbuf; 197 __be32 bound_addr; 198 __be32 connected_addr; 199 __be16 bound_port; 200 __be16 connected_port; 201 __u32 rcvbuf; 202 __u64 inum; 203 } __attribute__((packed)); 204 205 struct rds6_info_socket { 206 __u32 sndbuf; 207 struct in6_addr bound_addr; 208 struct in6_addr connected_addr; 209 __be16 bound_port; 210 __be16 connected_port; 211 __u32 rcvbuf; 212 __u64 inum; 213 } __attribute__((packed)); 214 215 struct rds_info_tcp_socket { 216 __be32 local_addr; 217 __be16 local_port; 218 __be32 peer_addr; 219 __be16 peer_port; 220 __u64 hdr_rem; 221 __u64 data_rem; 222 __u32 last_sent_nxt; 223 __u32 last_expected_una; 224 __u32 last_seen_una; 225 __u8 tos; 226 } __attribute__((packed)); 227 228 struct rds6_info_tcp_socket { 229 struct in6_addr local_addr; 230 __be16 local_port; 231 struct in6_addr peer_addr; 232 __be16 peer_port; 233 __u64 hdr_rem; 234 __u64 data_rem; 235 __u32 last_sent_nxt; 236 __u32 last_expected_una; 237 __u32 last_seen_una; 238 } __attribute__((packed)); 239 240 #define RDS_IB_GID_LEN 16 241 struct rds_info_rdma_connection { 242 __be32 src_addr; 243 __be32 dst_addr; 244 __u8 src_gid[RDS_IB_GID_LEN]; 245 __u8 dst_gid[RDS_IB_GID_LEN]; 246 247 __u32 max_send_wr; 248 __u32 max_recv_wr; 249 __u32 max_send_sge; 250 __u32 rdma_mr_max; 251 __u32 rdma_mr_size; 252 __u8 tos; 253 __u32 cache_allocs; 254 }; 255 256 struct rds6_info_rdma_connection { 257 struct in6_addr src_addr; 258 struct in6_addr dst_addr; 259 __u8 src_gid[RDS_IB_GID_LEN]; 260 __u8 dst_gid[RDS_IB_GID_LEN]; 261 262 __u32 max_send_wr; 263 __u32 max_recv_wr; 264 __u32 max_send_sge; 265 __u32 rdma_mr_max; 266 __u32 rdma_mr_size; 267 __u8 tos; 268 __u32 cache_allocs; 269 }; 270 271 /* RDS message Receive Path Latency points */ 272 enum rds_message_rxpath_latency { 273 RDS_MSG_RX_HDR_TO_DGRAM_START = 0, 274 RDS_MSG_RX_DGRAM_REASSEMBLE, 275 RDS_MSG_RX_DGRAM_DELIVERED, 276 RDS_MSG_RX_DGRAM_TRACE_MAX 277 }; 278 279 struct rds_rx_trace_so { 280 __u8 rx_traces; 281 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 282 }; 283 284 struct rds_cmsg_rx_trace { 285 __u8 rx_traces; 286 __u8 rx_trace_pos[RDS_MSG_RX_DGRAM_TRACE_MAX]; 287 __u64 rx_trace[RDS_MSG_RX_DGRAM_TRACE_MAX]; 288 }; 289 290 /* 291 * Congestion monitoring. 292 * Congestion control in RDS happens at the host connection 293 * level by exchanging a bitmap marking congested ports. 294 * By default, a process sleeping in poll() is always woken 295 * up when the congestion map is updated. 296 * With explicit monitoring, an application can have more 297 * fine-grained control. 298 * The application installs a 64bit mask value in the socket, 299 * where each bit corresponds to a group of ports. 300 * When a congestion update arrives, RDS checks the set of 301 * ports that are now uncongested against the list bit mask 302 * installed in the socket, and if they overlap, we queue a 303 * cong_notification on the socket. 304 * 305 * To install the congestion monitor bitmask, use RDS_CONG_MONITOR 306 * with the 64bit mask. 307 * Congestion updates are received via RDS_CMSG_CONG_UPDATE 308 * control messages. 309 * 310 * The correspondence between bits and ports is 311 * 1 << (portnum % 64) 312 */ 313 #define RDS_CONG_MONITOR_SIZE 64 314 #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE) 315 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port)) 316 317 /* 318 * RDMA related types 319 */ 320 321 /* 322 * This encapsulates a remote memory location. 323 * In the current implementation, it contains the R_Key 324 * of the remote memory region, and the offset into it 325 * (so that the application does not have to worry about 326 * alignment). 327 */ 328 typedef __u64 rds_rdma_cookie_t; 329 330 struct rds_iovec { 331 __u64 addr; 332 __u64 bytes; 333 }; 334 335 struct rds_get_mr_args { 336 struct rds_iovec vec; 337 __u64 cookie_addr; 338 __u64 flags; 339 }; 340 341 struct rds_get_mr_for_dest_args { 342 struct __kernel_sockaddr_storage dest_addr; 343 struct rds_iovec vec; 344 __u64 cookie_addr; 345 __u64 flags; 346 }; 347 348 struct rds_free_mr_args { 349 rds_rdma_cookie_t cookie; 350 __u64 flags; 351 }; 352 353 struct rds_rdma_args { 354 rds_rdma_cookie_t cookie; 355 struct rds_iovec remote_vec; 356 __u64 local_vec_addr; 357 __u64 nr_local; 358 __u64 flags; 359 __u64 user_token; 360 }; 361 362 struct rds_atomic_args { 363 rds_rdma_cookie_t cookie; 364 __u64 local_addr; 365 __u64 remote_addr; 366 union { 367 struct { 368 __u64 compare; 369 __u64 swap; 370 } cswp; 371 struct { 372 __u64 add; 373 } fadd; 374 struct { 375 __u64 compare; 376 __u64 swap; 377 __u64 compare_mask; 378 __u64 swap_mask; 379 } m_cswp; 380 struct { 381 __u64 add; 382 __u64 nocarry_mask; 383 } m_fadd; 384 }; 385 __u64 flags; 386 __u64 user_token; 387 }; 388 389 struct rds_rdma_notify { 390 __u64 user_token; 391 __s32 status; 392 }; 393 394 #define RDS_RDMA_SUCCESS 0 395 #define RDS_RDMA_REMOTE_ERROR 1 396 #define RDS_RDMA_CANCELED 2 397 #define RDS_RDMA_DROPPED 3 398 #define RDS_RDMA_OTHER_ERROR 4 399 400 #define RDS_MAX_ZCOOKIES 8 401 struct rds_zcopy_cookies { 402 __u32 num; 403 __u32 cookies[RDS_MAX_ZCOOKIES]; 404 }; 405 406 /* 407 * Common set of flags for all RDMA related structs 408 */ 409 #define RDS_RDMA_READWRITE 0x0001 410 #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */ 411 #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */ 412 #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */ 413 #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */ 414 #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */ 415 #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */ 416 417 #endif /* IB_RDS_H */ 418