rds.h 7.7 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286
  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. #ifndef _LINUX_RDS_H
  34. #define _LINUX_RDS_H
  35. #include <linux/types.h>
  36. #define RDS_IB_ABI_VERSION 0x301
  37. /*
  38. * setsockopt/getsockopt for SOL_RDS
  39. */
  40. #define RDS_CANCEL_SENT_TO 1
  41. #define RDS_GET_MR 2
  42. #define RDS_FREE_MR 3
  43. /* deprecated: RDS_BARRIER 4 */
  44. #define RDS_RECVERR 5
  45. #define RDS_CONG_MONITOR 6
  46. #define RDS_GET_MR_FOR_DEST 7
  47. /*
  48. * Control message types for SOL_RDS.
  49. *
  50. * CMSG_RDMA_ARGS (sendmsg)
  51. * Request a RDMA transfer to/from the specified
  52. * memory ranges.
  53. * The cmsg_data is a struct rds_rdma_args.
  54. * RDS_CMSG_RDMA_DEST (recvmsg, sendmsg)
  55. * Kernel informs application about intended
  56. * source/destination of a RDMA transfer
  57. * RDS_CMSG_RDMA_MAP (sendmsg)
  58. * Application asks kernel to map the given
  59. * memory range into a IB MR, and send the
  60. * R_Key along in an RDS extension header.
  61. * The cmsg_data is a struct rds_get_mr_args,
  62. * the same as for the GET_MR setsockopt.
  63. * RDS_CMSG_RDMA_STATUS (recvmsg)
  64. * Returns the status of a completed RDMA operation.
  65. */
  66. #define RDS_CMSG_RDMA_ARGS 1
  67. #define RDS_CMSG_RDMA_DEST 2
  68. #define RDS_CMSG_RDMA_MAP 3
  69. #define RDS_CMSG_RDMA_STATUS 4
  70. #define RDS_CMSG_CONG_UPDATE 5
  71. #define RDS_CMSG_ATOMIC_FADD 6
  72. #define RDS_CMSG_ATOMIC_CSWP 7
  73. #define RDS_CMSG_MASKED_ATOMIC_FADD 8
  74. #define RDS_CMSG_MASKED_ATOMIC_CSWP 9
  75. #define RDS_INFO_FIRST 10000
  76. #define RDS_INFO_COUNTERS 10000
  77. #define RDS_INFO_CONNECTIONS 10001
  78. /* 10002 aka RDS_INFO_FLOWS is deprecated */
  79. #define RDS_INFO_SEND_MESSAGES 10003
  80. #define RDS_INFO_RETRANS_MESSAGES 10004
  81. #define RDS_INFO_RECV_MESSAGES 10005
  82. #define RDS_INFO_SOCKETS 10006
  83. #define RDS_INFO_TCP_SOCKETS 10007
  84. #define RDS_INFO_IB_CONNECTIONS 10008
  85. #define RDS_INFO_CONNECTION_STATS 10009
  86. #define RDS_INFO_IWARP_CONNECTIONS 10010
  87. #define RDS_INFO_LAST 10010
  88. struct rds_info_counter {
  89. uint8_t name[32];
  90. uint64_t value;
  91. } __attribute__((packed));
  92. #define RDS_INFO_CONNECTION_FLAG_SENDING 0x01
  93. #define RDS_INFO_CONNECTION_FLAG_CONNECTING 0x02
  94. #define RDS_INFO_CONNECTION_FLAG_CONNECTED 0x04
  95. #define TRANSNAMSIZ 16
  96. struct rds_info_connection {
  97. uint64_t next_tx_seq;
  98. uint64_t next_rx_seq;
  99. __be32 laddr;
  100. __be32 faddr;
  101. uint8_t transport[TRANSNAMSIZ]; /* null term ascii */
  102. uint8_t flags;
  103. } __attribute__((packed));
  104. #define RDS_INFO_MESSAGE_FLAG_ACK 0x01
  105. #define RDS_INFO_MESSAGE_FLAG_FAST_ACK 0x02
  106. struct rds_info_message {
  107. uint64_t seq;
  108. uint32_t len;
  109. __be32 laddr;
  110. __be32 faddr;
  111. __be16 lport;
  112. __be16 fport;
  113. uint8_t flags;
  114. } __attribute__((packed));
  115. struct rds_info_socket {
  116. uint32_t sndbuf;
  117. __be32 bound_addr;
  118. __be32 connected_addr;
  119. __be16 bound_port;
  120. __be16 connected_port;
  121. uint32_t rcvbuf;
  122. uint64_t inum;
  123. } __attribute__((packed));
  124. struct rds_info_tcp_socket {
  125. __be32 local_addr;
  126. __be16 local_port;
  127. __be32 peer_addr;
  128. __be16 peer_port;
  129. uint64_t hdr_rem;
  130. uint64_t data_rem;
  131. uint32_t last_sent_nxt;
  132. uint32_t last_expected_una;
  133. uint32_t last_seen_una;
  134. } __attribute__((packed));
  135. #define RDS_IB_GID_LEN 16
  136. struct rds_info_rdma_connection {
  137. __be32 src_addr;
  138. __be32 dst_addr;
  139. uint8_t src_gid[RDS_IB_GID_LEN];
  140. uint8_t dst_gid[RDS_IB_GID_LEN];
  141. uint32_t max_send_wr;
  142. uint32_t max_recv_wr;
  143. uint32_t max_send_sge;
  144. uint32_t rdma_mr_max;
  145. uint32_t rdma_mr_size;
  146. };
  147. /*
  148. * Congestion monitoring.
  149. * Congestion control in RDS happens at the host connection
  150. * level by exchanging a bitmap marking congested ports.
  151. * By default, a process sleeping in poll() is always woken
  152. * up when the congestion map is updated.
  153. * With explicit monitoring, an application can have more
  154. * fine-grained control.
  155. * The application installs a 64bit mask value in the socket,
  156. * where each bit corresponds to a group of ports.
  157. * When a congestion update arrives, RDS checks the set of
  158. * ports that are now uncongested against the list bit mask
  159. * installed in the socket, and if they overlap, we queue a
  160. * cong_notification on the socket.
  161. *
  162. * To install the congestion monitor bitmask, use RDS_CONG_MONITOR
  163. * with the 64bit mask.
  164. * Congestion updates are received via RDS_CMSG_CONG_UPDATE
  165. * control messages.
  166. *
  167. * The correspondence between bits and ports is
  168. * 1 << (portnum % 64)
  169. */
  170. #define RDS_CONG_MONITOR_SIZE 64
  171. #define RDS_CONG_MONITOR_BIT(port) (((unsigned int) port) % RDS_CONG_MONITOR_SIZE)
  172. #define RDS_CONG_MONITOR_MASK(port) (1ULL << RDS_CONG_MONITOR_BIT(port))
  173. /*
  174. * RDMA related types
  175. */
  176. /*
  177. * This encapsulates a remote memory location.
  178. * In the current implementation, it contains the R_Key
  179. * of the remote memory region, and the offset into it
  180. * (so that the application does not have to worry about
  181. * alignment).
  182. */
  183. typedef uint64_t rds_rdma_cookie_t;
  184. struct rds_iovec {
  185. uint64_t addr;
  186. uint64_t bytes;
  187. };
  188. struct rds_get_mr_args {
  189. struct rds_iovec vec;
  190. uint64_t cookie_addr;
  191. uint64_t flags;
  192. };
  193. struct rds_get_mr_for_dest_args {
  194. struct sockaddr_storage dest_addr;
  195. struct rds_iovec vec;
  196. uint64_t cookie_addr;
  197. uint64_t flags;
  198. };
  199. struct rds_free_mr_args {
  200. rds_rdma_cookie_t cookie;
  201. uint64_t flags;
  202. };
  203. struct rds_rdma_args {
  204. rds_rdma_cookie_t cookie;
  205. struct rds_iovec remote_vec;
  206. uint64_t local_vec_addr;
  207. uint64_t nr_local;
  208. uint64_t flags;
  209. uint64_t user_token;
  210. };
  211. struct rds_atomic_args {
  212. rds_rdma_cookie_t cookie;
  213. uint64_t local_addr;
  214. uint64_t remote_addr;
  215. union {
  216. struct {
  217. uint64_t compare;
  218. uint64_t swap;
  219. } cswp;
  220. struct {
  221. uint64_t add;
  222. } fadd;
  223. struct {
  224. uint64_t compare;
  225. uint64_t swap;
  226. uint64_t compare_mask;
  227. uint64_t swap_mask;
  228. } m_cswp;
  229. struct {
  230. uint64_t add;
  231. uint64_t nocarry_mask;
  232. } m_fadd;
  233. };
  234. uint64_t flags;
  235. uint64_t user_token;
  236. };
  237. struct rds_rdma_notify {
  238. uint64_t user_token;
  239. int32_t status;
  240. };
  241. #define RDS_RDMA_SUCCESS 0
  242. #define RDS_RDMA_REMOTE_ERROR 1
  243. #define RDS_RDMA_CANCELED 2
  244. #define RDS_RDMA_DROPPED 3
  245. #define RDS_RDMA_OTHER_ERROR 4
  246. /*
  247. * Common set of flags for all RDMA related structs
  248. */
  249. #define RDS_RDMA_READWRITE 0x0001
  250. #define RDS_RDMA_FENCE 0x0002 /* use FENCE for immediate send */
  251. #define RDS_RDMA_INVALIDATE 0x0004 /* invalidate R_Key after freeing MR */
  252. #define RDS_RDMA_USE_ONCE 0x0008 /* free MR after use */
  253. #define RDS_RDMA_DONTWAIT 0x0010 /* Don't wait in SET_BARRIER */
  254. #define RDS_RDMA_NOTIFY_ME 0x0020 /* Notify when operation completes */
  255. #define RDS_RDMA_SILENT 0x0040 /* Do not interrupt remote */
  256. #endif /* IB_RDS_H */