include/linux/rds.h

   1 /*
   2  * Copyright (c) 2008 Oracle.  All rights reserved.
   3  *
   4  * This software is available to you under a choice of one of two
   5  * licenses.  You may choose to be licensed under the terms of the GNU
   6  * General Public License (GPL) Version 2, available from the file
   7  * COPYING in the main directory of this source tree, or the
   8  * OpenIB.org BSD license below:
   9  *
  10  *     Redistribution and use in source and binary forms, with or
  11  *     without modification, are permitted provided that the following
  12  *     conditions are met:
  13  *
  14  *      - Redistributions of source code must retain the above
  15  *        copyright notice, this list of conditions and the following
  16  *        disclaimer.
  17  *
  18  *      - Redistributions in binary form must reproduce the above
  19  *        copyright notice, this list of conditions and the following
  20  *        disclaimer in the documentation and/or other materials
  21  *        provided with the distribution.
  22  *
  23  * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
  24  * EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
  25  * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND
  26  * NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT HOLDERS
  27  * BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
  28  * ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
  29  * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
  30  * SOFTWARE.
  31  *
  32  */
  33
  34 #ifndef _LINUX_RDS_H
  35 #define _LINUX_RDS_H
  36
  37 #include <linux/types.h>
  38
  39 #define RDS_IB_ABI_VERSION              0x301
  40
  41 /*
  42  * setsockopt/getsockopt for SOL_RDS
  43  */
  44 #define RDS_CANCEL_SENT_TO              1
  45 #define RDS_GET_MR                      2
  46 #define RDS_FREE_MR                     3
  47 /* deprecated: RDS_BARRIER 4 */
  48 #define RDS_RECVERR                     5
  49 #define RDS_CONG_MONITOR                6
  50 #define RDS_GET_MR_FOR_DEST             7
  51
  52 /*
  53  * Control message types for SOL_RDS.
  54  *
  55  * CMSG_RDMA_ARGS (sendmsg)
  56  *      Request a RDMA transfer to/from the specified
  57  *      memory ranges.
  58  *      The cmsg_data is a struct rds_rdma_args.
  59  * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg)
  60  *      Kernel informs application about intended
  61  *      source/destination of a RDMA transfer
  62  * RDS_CMSG_RDMA_MAP (sendmsg)
  63  *      Application asks kernel to map the given
  64  *      memory range into a IB MR, and send the
  65  *      R_Key along in an RDS extension header.
  66  *      The cmsg_data is a struct rds_get_mr_args,
  67  *      the same as for the GET_MR setsockopt.
  68  * RDS_CMSG_RDMA_STATUS (recvmsg)
  69  *      Returns the status of a completed RDMA operation.
  70  */
  71 #define RDS_CMSG_RDMA_ARGS              1
  72 #define RDS_CMSG_RDMA_DEST              2
  73 #define RDS_CMSG_RDMA_MAP               3
  74 #define RDS_CMSG_RDMA_STATUS            4
  75 #define RDS_CMSG_CONG_UPDATE            5
  76 #define RDS_CMSG_ATOMIC_FADD            6
  77 #define RDS_CMSG_ATOMIC_CSWP            7
  78 #define RDS_CMSG_MASKED_ATOMIC_FADD     8
  79 #define RDS_CMSG_MASKED_ATOMIC_CSWP     9
  80
  81 #define RDS_INFO_FIRST                  10000
  82 #define RDS_INFO_COUNTERS               10000
  83 #define RDS_INFO_CONNECTIONS            10001
  84 /* 10002 aka RDS_INFO_FLOWS is deprecated */
  85 #define RDS_INFO_SEND_MESSAGES          10003
  86 #define RDS_INFO_RETRANS_MESSAGES       10004
  87 #define RDS_INFO_RECV_MESSAGES          10005
  88 #define RDS_INFO_SOCKETS                10006
  89 #define RDS_INFO_TCP_SOCKETS            10007
  90 #define RDS_INFO_IB_CONNECTIONS         10008
  91 #define RDS_INFO_CONNECTION_STATS       10009
  92 #define RDS_INFO_IWARP_CONNECTIONS      10010
  93 #define RDS_INFO_LAST                   10010
  94
  95 struct rds_info_counter {
  96         uint8_t name[32];
  97         uint64_t        value;
  98 } __attribute__((packed));
  99
 100 #define RDS_INFO_CONNECTION_FLAG_SENDING        0x01
 101 #define RDS_INFO_CONNECTION_FLAG_CONNECTING     0x02
 102 #define RDS_INFO_CONNECTION_FLAG_CONNECTED      0x04
 103
 104 #define TRANSNAMSIZ     16
 105
 106 struct rds_info_connection {
 107         uint64_t        next_tx_seq;
 108         uint64_t        next_rx_seq;
 109         __be32          laddr;
 110         __be32          faddr;
 111         uint8_t transport[TRANSNAMSIZ];         /* null term ascii */
 112         uint8_t flags;
 113 } __attribute__((packed));
 114
 115 #define RDS_INFO_MESSAGE_FLAG_ACK               0x01
 116 #define RDS_INFO_MESSAGE_FLAG_FAST_ACK          0x02
 117
 118 struct rds_info_message {
 119         uint64_t        seq;
 120         uint32_t        len;
 121         __be32          laddr;
 122         __be32          faddr;
 123         __be16          lport;
 124         __be16          fport;
 125         uint8_t flags;
 126 } __attribute__((packed));
 127
 128 struct rds_info_socket {
 129         uint32_t        sndbuf;
 130         __be32          bound_addr;
 131         __be32          connected_addr;
 132         __be16          bound_port;
 133         __be16          connected_port;
 134         uint32_t        rcvbuf;
 135         uint64_t        inum;
 136 } __attribute__((packed));
 137
 138 struct rds_info_tcp_socket {
 139         __be32          local_addr;
 140         __be16          local_port;
 141         __be32          peer_addr;
 142         __be16          peer_port;
 143         uint64_t       hdr_rem;
 144         uint64_t       data_rem;
 145         uint32_t       last_sent_nxt;
 146         uint32_t       last_expected_una;
 147         uint32_t       last_seen_una;
 148 } __attribute__((packed));
 149
 150 #define RDS_IB_GID_LEN  16
 151 struct rds_info_rdma_connection {
 152         __be32          src_addr;
 153         __be32          dst_addr;
 154         uint8_t         src_gid[RDS_IB_GID_LEN];
 155         uint8_t         dst_gid[RDS_IB_GID_LEN];
 156
 157         uint32_t        max_send_wr;
 158         uint32_t        max_recv_wr;
 159         uint32_t        max_send_sge;
 160         uint32_t        rdma_mr_max;
 161         uint32_t        rdma_mr_size;
 162 };
 163
 164 /*
 165  * Congestion monitoring.
 166  * Congestion control in RDS happens at the host connection
 167  * level by exchanging a bitmap marking congested ports.
 168  * By default, a process sleeping in poll() is always woken
 169  * up when the congestion map is updated.
 170  * With explicit monitoring, an application can have more
 171  * fine-grained control.
 172  * The application installs a 64bit mask value in the socket,
 173  * where each bit corresponds to a group of ports.
 174  * When a congestion update arrives, RDS checks the set of
 175  * ports that are now uncongested against the list bit mask
 176  * installed in the socket, and if they overlap, we queue a
 177  * cong_notification on the socket.
 178  *
 179  * To install the congestion monitor bitmask, use RDS_CONG_MONITOR
 180  * with the 64bit mask.
 181  * Congestion updates are received via RDS_CMSG_CONG_UPDATE
 182  * control messages.
 183  *
 184  * The correspondence between bits and ports is
 185  *      1 << (portnum % 64)
 186  */
 187 #define RDS_CONG_MONITOR_SIZE   64
 188 #define RDS_CONG_MONITOR_BIT(port)  (((unsigned int) port) % RDS_CONG_MONITOR_SIZE)
 189 #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port))
 190
 191 /*
 192  * RDMA related types
 193  */
 194
 195 /*
 196  * This encapsulates a remote memory location.
 197  * In the current implementation, it contains the R_Key
 198  * of the remote memory region, and the offset into it
 199  * (so that the application does not have to worry about
 200  * alignment).
 201  */
 202 typedef uint64_t        rds_rdma_cookie_t;
 203
 204 struct rds_iovec {
 205         uint64_t        addr;
 206         uint64_t        bytes;
 207 };
 208
 209 struct rds_get_mr_args {
 210         struct rds_iovec vec;
 211         uint64_t        cookie_addr;
 212         uint64_t        flags;
 213 };
 214
 215 struct rds_get_mr_for_dest_args {
 216         struct sockaddr_storage dest_addr;
 217         struct rds_iovec        vec;
 218         uint64_t                cookie_addr;
 219         uint64_t                flags;
 220 };
 221
 222 struct rds_free_mr_args {
 223         rds_rdma_cookie_t cookie;
 224         uint64_t        flags;
 225 };
 226
 227 struct rds_rdma_args {
 228         rds_rdma_cookie_t cookie;
 229         struct rds_iovec remote_vec;
 230         uint64_t        local_vec_addr;
 231         uint64_t        nr_local;
 232         uint64_t        flags;
 233         uint64_t        user_token;
 234 };
 235
 236 struct rds_atomic_args {
 237         rds_rdma_cookie_t cookie;
 238         uint64_t        local_addr;
 239         uint64_t        remote_addr;
 240         union {
 241                 struct {
 242                         uint64_t        compare;
 243                         uint64_t        swap;
 244                 } cswp;
 245                 struct {
 246                         uint64_t        add;
 247                 } fadd;
 248                 struct {
 249                         uint64_t        compare;
 250                         uint64_t        swap;
 251                         uint64_t        compare_mask;
 252                         uint64_t        swap_mask;
 253                 } m_cswp;
 254                 struct {
 255                         uint64_t        add;
 256                         uint64_t        nocarry_mask;
 257                 } m_fadd;
 258         };
 259         uint64_t        flags;
 260         uint64_t        user_token;
 261 };
 262
 263 struct rds_rdma_notify {
 264         uint64_t        user_token;
 265         int32_t         status;
 266 };
 267
 268 #define RDS_RDMA_SUCCESS        0
 269 #define RDS_RDMA_REMOTE_ERROR   1
 270 #define RDS_RDMA_CANCELED       2
 271 #define RDS_RDMA_DROPPED        3
 272 #define RDS_RDMA_OTHER_ERROR    4
 273
 274 /*
 275  * Common set of flags for all RDMA related structs
 276  */
 277 #define RDS_RDMA_READWRITE      0x0001
 278 #define RDS_RDMA_FENCE          0x0002  /* use FENCE for immediate send */
 279 #define RDS_RDMA_INVALIDATE     0x0004  /* invalidate R_Key after freeing MR */
 280 #define RDS_RDMA_USE_ONCE       0x0008  /* free MR after use */
 281 #define RDS_RDMA_DONTWAIT       0x0010  /* Don't wait in SET_BARRIER */
 282 #define RDS_RDMA_NOTIFY_ME      0x0020  /* Notify when operation completes */
 283 #define RDS_RDMA_SILENT         0x0040  /* Do not interrupt remote */
 284
 285 #endif /* IB_RDS_H */