1
2 /*
3 * Copyright (c) 2004-2007 The Trustees of Indiana University and Indiana
4 * University Research and Technology
5 * Corporation. All rights reserved.
6 * Copyright (c) 2004-2009 The University of Tennessee and The University
7 * of Tennessee Research Foundation. All rights
8 * reserved.
9 * Copyright (c) 2004-2005 High Performance Computing Center Stuttgart,
10 * University of Stuttgart. All rights reserved.
11 * Copyright (c) 2004-2005 The Regents of the University of California.
12 * All rights reserved.
13 * Copyright (c) 2011 Cisco Systems, Inc. All rights reserved.
14 * $COPYRIGHT$
15 *
16 * Additional copyrights may follow
17 *
18 * $HEADER$
19 */
20 /**
21 * @file
22 */
23 #ifndef MCA_BTL_TCP_H
24 #define MCA_BTL_TCP_H
25
26 #include "ompi_config.h"
27 #ifdef HAVE_SYS_TYPES_H
28 #include <sys/types.h>
29 #endif
30 #ifdef HAVE_SYS_SOCKET_H
31 #include <sys/socket.h>
32 #endif
33 #ifdef HAVE_NETINET_IN_H
34 #include <netinet/in.h>
35 #endif
36
37 /* Open MPI includes */
38 #include "opal/mca/event/event.h"
39 #include "ompi/class/ompi_free_list.h"
40 #include "ompi/mca/btl/btl.h"
41 #include "ompi/mca/btl/base/base.h"
42 #include "ompi/mca/mpool/mpool.h"
43 #include "ompi/mca/btl/btl.h"
44 #include "opal/class/opal_hash_table.h"
45
46 #define MCA_BTL_TCP_STATISTICS 0
47 BEGIN_C_DECLS
48
49
50 /**
51 * TCP BTL component.
52 */
53
54 struct mca_btl_tcp2_component_t {
55 mca_btl_base_component_2_0_0_t super; /**< base BTL component */
56 uint32_t tcp_addr_count; /**< total number of addresses */
57 uint32_t tcp_num_btls; /**< number of hcas available to the TCP component */
58 uint32_t tcp_num_links; /**< number of logical links per physical device */
59 struct mca_btl_tcp2_module_t **tcp_btls; /**< array of available BTL modules */
60 struct mca_btl_tcp2_proc_t* tcp_local; /**< local proc struct */
61 int tcp_free_list_num; /**< initial size of free lists */
62 int tcp_free_list_max; /**< maximum size of free lists */
63 int tcp_free_list_inc; /**< number of elements to alloc when growing free lists */
64 int tcp_endpoint_cache; /**< amount of cache on each endpoint */
65 opal_hash_table_t tcp_procs; /**< hash table of tcp proc structures */
66 opal_list_t tcp_events; /**< list of pending tcp events */
67 opal_mutex_t tcp_lock; /**< lock for accessing module state */
68
69 opal_event_t tcp_recv_event; /**< recv event for IPv4 listen socket */
70 int tcp_listen_sd; /**< IPv4 listen socket for incoming connection requests */
71 unsigned short tcp_listen_port; /**< IPv4 listen port */
72 int32_t tcp_port_min; /**< IPv4 minimum port */
73 int32_t tcp_port_range; /**< IPv4 port range */
74 #if OPAL_ENABLE_IPV6
75 opal_event_t tcp6_recv_event; /**< recv event for IPv6 listen socket */
76 int tcp6_listen_sd; /**< IPv6 listen socket for incoming connection requests */
77 unsigned short tcp6_listen_port; /**< IPv6 listen port */
78 int32_t tcp6_port_min; /**< IPv4 minimum port */
79 int32_t tcp6_port_range; /**< IPv4 port range */
80 #endif
81 /* Port range restriction */
82
83 char* tcp_if_include; /**< comma seperated list of interface to include */
84 char* tcp_if_exclude; /**< comma seperated list of interface to exclude */
85 int tcp_sndbuf; /**< socket sndbuf size */
86 int tcp_rcvbuf; /**< socket rcvbuf size */
87 int tcp_disable_family; /**< disabled AF_family */
88
89 /* free list of fragment descriptors */
90 ompi_free_list_t tcp_frag_eager;
91 ompi_free_list_t tcp_frag_max;
92 ompi_free_list_t tcp_frag_user;
93
94 /* Do we want to use TCP_NODELAY? */
95 int tcp_use_nodelay;
96 };
97 typedef struct mca_btl_tcp2_component_t mca_btl_tcp2_component_t;
98
99 OMPI_MODULE_DECLSPEC extern mca_btl_tcp2_component_t mca_btl_tcp2_component;
100
101 /**
102 * BTL Module Interface
103 */
104 struct mca_btl_tcp2_module_t {
105 mca_btl_base_module_t super; /**< base BTL interface */
106 uint16_t tcp_ifkindex; /** <BTL kernel interface index */
107 #if 0
108 int tcp_ifindex; /**< PTL interface index */
109 #endif
110 struct sockaddr_storage tcp_ifaddr; /**< PTL interface address */
111 uint32_t tcp_ifmask; /**< PTL interface netmask */
112 opal_list_t tcp_endpoints;
113 #if MCA_BTL_TCP_STATISTICS
114 size_t tcp_bytes_sent;
115 size_t tcp_bytes_recv;
116 size_t tcp_send_handler;
117 #endif
118 };
119 typedef struct mca_btl_tcp2_module_t mca_btl_tcp2_module_t;
120 extern mca_btl_tcp2_module_t mca_btl_tcp2_module;
121
122 #define CLOSE_THE_SOCKET(socket) close(socket)
123
124 /**
125 * Register TCP component parameters with the MCA framework
126 */
127 extern int mca_btl_tcp2_component_open(void);
128
129 /**
130 * Any final cleanup before being unloaded.
131 */
132 extern int mca_btl_tcp2_component_close(void);
133
134 /**
135 * TCP component initialization.
136 *
137 * @param num_btl_modules (OUT) Number of BTLs returned in BTL array.
138 * @param allow_multi_user_threads (OUT) Flag indicating wether BTL supports user threads (TRUE)
139 * @param have_hidden_threads (OUT) Flag indicating wether BTL uses threads (TRUE)
140 */
141 extern mca_btl_base_module_t** mca_btl_tcp2_component_init(
142 int *num_btl_modules,
143 bool allow_multi_user_threads,
144 bool have_hidden_threads
145 );
146
147 /**
148 * TCP component control.
149 */
150 int mca_btl_tcp2_component_control(
151 int param,
152 void* value,
153 size_t size
154 );
155
156
157 /**
158 * TCP component progress.
159 */
160 extern int mca_btl_tcp2_component_progress(void);
161
162
163
164 /**
165 * Cleanup any resources held by the BTL.
166 *
167 * @param btl BTL instance.
168 * @return OMPI_SUCCESS or error status on failure.
169 */
170
171 extern int mca_btl_tcp2_finalize(
172 struct mca_btl_base_module_t* btl
173 );
174
175
176 /**
177 * PML->BTL notification of change in the process list.
178 *
179 * @param btl (IN)
180 * @param nprocs (IN) Number of processes
181 * @param procs (IN) Set of processes
182 * @param peers (OUT) Set of (optional) peer addressing info.
183 * @param peers (IN/OUT) Set of processes that are reachable via this BTL.
184 * @return OMPI_SUCCESS or error status on failure.
185 *
186 */
187
188 extern int mca_btl_tcp2_add_procs(
189 struct mca_btl_base_module_t* btl,
190 size_t nprocs,
191 struct ompi_proc_t **procs,
192 struct mca_btl_base_endpoint_t** peers,
193 opal_bitmap_t* reachable
194 );
195
196 /**
197 * PML->BTL notification of change in the process list.
198 *
199 * @param btl (IN) BTL instance
200 * @param nproc (IN) Number of processes.
201 * @param procs (IN) Set of processes.
202 * @param peers (IN) Set of peer data structures.
203 * @return Status indicating if cleanup was successful
204 *
205 */
206
207 extern int mca_btl_tcp2_del_procs(
208 struct mca_btl_base_module_t* btl,
209 size_t nprocs,
210 struct ompi_proc_t **procs,
211 struct mca_btl_base_endpoint_t** peers
212 );
213
214
215 /**
216 * Initiate an asynchronous send.
217 *
218 * @param btl (IN) BTL module
219 * @param endpoint (IN) BTL addressing information
220 * @param descriptor (IN) Description of the data to be transfered
221 * @param tag (IN) The tag value used to notify the peer.
222 */
223
224 extern int mca_btl_tcp2_send(
225 struct mca_btl_base_module_t* btl,
226 struct mca_btl_base_endpoint_t* btl_peer,
227 struct mca_btl_base_descriptor_t* descriptor,
228 mca_btl_base_tag_t tag
229 );
230
231
232 /**
233 * Initiate an asynchronous put.
234 *
235 * @param btl (IN) BTL module
236 * @param endpoint (IN) BTL addressing information
237 * @param descriptor (IN) Description of the data to be transferred
238 */
239
240 extern int mca_btl_tcp2_put(
241 struct mca_btl_base_module_t* btl,
242 struct mca_btl_base_endpoint_t* btl_peer,
243 struct mca_btl_base_descriptor_t* decriptor
244 );
245
246
247 /**
248 * Initiate an asynchronous get.
249 *
250 * @param btl (IN) BTL module
251 * @param endpoint (IN) BTL addressing information
252 * @param descriptor (IN) Description of the data to be transferred
253 */
254
255 extern int mca_btl_tcp2_get(
256 struct mca_btl_base_module_t* btl,
257 struct mca_btl_base_endpoint_t* btl_peer,
258 struct mca_btl_base_descriptor_t* decriptor
259 );
260
261 /**
262 * Allocate a descriptor with a segment of the requested size.
263 * Note that the BTL layer may choose to return a smaller size
264 * if it cannot support the request.
265 *
266 * @param btl (IN) BTL module
267 * @param size (IN) Request segment size.
268 */
269
270 extern mca_btl_base_descriptor_t* mca_btl_tcp2_alloc(
271 struct mca_btl_base_module_t* btl,
272 struct mca_btl_base_endpoint_t* endpoint,
273 uint8_t order,
274 size_t size,
275 uint32_t flags);
276
277
278 /**
279 * Return a segment allocated by this BTL.
280 *
281 * @param btl (IN) BTL module
282 * @param descriptor (IN) Allocated descriptor.
283 */
284
285 extern int mca_btl_tcp2_free(
286 struct mca_btl_base_module_t* btl,
287 mca_btl_base_descriptor_t* des);
288
289
290 /**
291 * Prepare a descriptor for send/rdma using the supplied
292 * convertor. If the convertor references data that is contigous,
293 * the descriptor may simply point to the user buffer. Otherwise,
294 * this routine is responsible for allocating buffer space and
295 * packing if required.
296 *
297 * @param btl (IN) BTL module
298 * @param endpoint (IN) BTL peer addressing
299 * @param convertor (IN) Data type convertor
300 * @param reserve (IN) Additional bytes requested by upper layer to precede user data
301 * @param size (IN/OUT) Number of bytes to prepare (IN), number of bytes actually prepared (OUT)
302 */
303
304 mca_btl_base_descriptor_t* mca_btl_tcp2_prepare_src(
305 struct mca_btl_base_module_t* btl,
306 struct mca_btl_base_endpoint_t* peer,
307 struct mca_mpool_base_registration_t*,
308 struct opal_convertor_t* convertor,
309 uint8_t order,
310 size_t reserve,
311 size_t* size,
312 uint32_t flags
313 );
314
315 extern mca_btl_base_descriptor_t* mca_btl_tcp2_prepare_dst(
316 struct mca_btl_base_module_t* btl,
317 struct mca_btl_base_endpoint_t* peer,
318 struct mca_mpool_base_registration_t*,
319 struct opal_convertor_t* convertor,
320 uint8_t order,
321 size_t reserve,
322 size_t* size,
323 uint32_t flags);
324
325
326 /**
327 * Fault Tolerance Event Notification Function
328 * @param state Checkpoint Stae
329 * @return OMPI_SUCCESS or failure status
330 */
331 int mca_btl_tcp2_ft_event(int state);
332
333 END_C_DECLS
334 #endif