author | chegar |
Thu, 31 Jan 2019 14:12:49 +0000 | |
branch | rsocket-branch |
changeset 57129 | 1e1db86ea836 |
parent 57115 | 512e7cc6ccce |
child 57134 | 096bba76efd1 |
permissions | -rw-r--r-- |
57115 | 1 |
/* |
57129
1e1db86ea836
rsocket-branch: Review comments from Brian, A, B, and D
chegar
parents:
57115
diff
changeset
|
2 |
* Copyright (c) 2018, 2019, Oracle and/or its affiliates. All rights reserved. |
57115 | 3 |
* DO NOT ALTER OR REMOVE COPYRIGHT NOTICES OR THIS FILE HEADER. |
4 |
* |
|
5 |
* This code is free software; you can redistribute it and/or modify it |
|
6 |
* under the terms of the GNU General Public License version 2 only, as |
|
7 |
* published by the Free Software Foundation. Oracle designates this |
|
8 |
* particular file as subject to the "Classpath" exception as provided |
|
9 |
* by Oracle in the LICENSE file that accompanied this code. |
|
10 |
* |
|
11 |
* This code is distributed in the hope that it will be useful, but WITHOUT |
|
12 |
* ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or |
|
13 |
* FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License |
|
14 |
* version 2 for more details (a copy is included in the LICENSE file that |
|
15 |
* accompanied this code). |
|
16 |
* |
|
17 |
* You should have received a copy of the GNU General Public License version |
|
18 |
* 2 along with this work; if not, write to the Free Software Foundation, |
|
19 |
* Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA. |
|
20 |
* |
|
21 |
* Please contact Oracle, 500 Oracle Parkway, Redwood Shores, CA 94065 USA |
|
22 |
* or visit www.oracle.com if you need additional information or have any |
|
23 |
* questions. |
|
24 |
*/ |
|
25 |
||
26 |
#include <assert.h> |
|
27 |
#include "java_net_SocketOptions.h" |
|
28 |
#include "jdk_net_RdmaSocketOptions.h" |
|
29 |
#include "jvm.h" |
|
30 |
#include <netinet/tcp.h> // defines TCP_NODELAY |
|
31 |
#include "net_util.h" |
|
32 |
#include "rdma_util_md.h" |
|
33 |
#include "Rsocket.h" |
|
34 |
#include <stdlib.h> |
|
35 |
#include <sys/ioctl.h> |
|
36 |
#include <sys/resource.h> |
|
37 |
#include <pthread.h> |
|
38 |
||
39 |
#define BLOCKING_IO_RETURN_INT(FD, FUNC) { \ |
|
40 |
int ret; \ |
|
41 |
threadEntry_t self; \ |
|
42 |
fdEntry_t *fdEntry = getFdEntry(FD); \ |
|
43 |
if (fdEntry == NULL) { \ |
|
44 |
errno = EBADF; \ |
|
45 |
return -1; \ |
|
46 |
} \ |
|
47 |
do { \ |
|
48 |
startOp(fdEntry, &self); \ |
|
49 |
ret = FUNC; \ |
|
50 |
endOp(fdEntry, &self); \ |
|
51 |
} while (ret == -1 && errno == EINTR); \ |
|
52 |
return ret; \ |
|
53 |
} |
|
54 |
||
55 |
typedef struct threadEntry { |
|
56 |
pthread_t thr; /* this thread */ |
|
57 |
struct threadEntry *next; /* next thread */ |
|
58 |
int intr; /* interrupted */ |
|
59 |
} threadEntry_t; |
|
60 |
||
61 |
typedef struct { |
|
62 |
pthread_mutex_t lock; /* fd lock */ |
|
63 |
threadEntry_t *threads; /* threads blocked on fd */ |
|
64 |
} fdEntry_t; |
|
65 |
||
66 |
static int sigWakeup = (__SIGRTMAX - 2); |
|
67 |
||
68 |
static fdEntry_t* fdTable = NULL; |
|
69 |
||
70 |
static const int fdTableMaxSize = 0x1000; /* 4K */ |
|
71 |
||
72 |
static int fdTableLen = 0; |
|
73 |
||
74 |
static int fdLimit = 0; |
|
75 |
||
76 |
static fdEntry_t** fdOverflowTable = NULL; |
|
77 |
||
78 |
static int fdOverflowTableLen = 0; |
|
79 |
||
80 |
static const int fdOverflowTableSlabSize = 0x10000; /* 64k */ |
|
81 |
||
82 |
pthread_mutex_t fdOverflowTableLock = PTHREAD_MUTEX_INITIALIZER; |
|
83 |
||
84 |
static void sig_wakeup(int sig) { |
|
85 |
} |
|
86 |
||
87 |
static void __attribute((constructor)) init() { |
|
88 |
struct rlimit nbr_files; |
|
89 |
sigset_t sigset; |
|
90 |
struct sigaction sa; |
|
91 |
int i = 0; |
|
92 |
||
93 |
if (-1 == getrlimit(RLIMIT_NOFILE, &nbr_files)) { |
|
94 |
fprintf(stderr, "library initialization failed - " |
|
95 |
"unable to get max # of allocated fds\n"); |
|
96 |
abort(); |
|
97 |
} |
|
98 |
if (nbr_files.rlim_max != RLIM_INFINITY) { |
|
99 |
fdLimit = nbr_files.rlim_max; |
|
100 |
} else { |
|
101 |
fdLimit = INT_MAX; |
|
102 |
} |
|
103 |
||
104 |
fdTableLen = fdLimit < fdTableMaxSize ? fdLimit : fdTableMaxSize; |
|
105 |
fdTable = (fdEntry_t*) calloc(fdTableLen, sizeof(fdEntry_t)); |
|
106 |
if (fdTable == NULL) { |
|
107 |
fprintf(stderr, "library initialization failed - " |
|
108 |
"unable to allocate file descriptor table - out of memory"); |
|
109 |
abort(); |
|
110 |
} else { |
|
111 |
for (i = 0; i < fdTableLen; i ++) { |
|
112 |
pthread_mutex_init(&fdTable[i].lock, NULL); |
|
113 |
} |
|
114 |
} |
|
115 |
||
116 |
if (fdLimit > fdTableMaxSize) { |
|
117 |
fdOverflowTableLen = ((fdLimit - fdTableMaxSize) / fdOverflowTableSlabSize) + 1; |
|
118 |
fdOverflowTable = (fdEntry_t**) calloc(fdOverflowTableLen, sizeof(fdEntry_t*)); |
|
119 |
if (fdOverflowTable == NULL) { |
|
120 |
fprintf(stderr, "library initialization failed - " |
|
121 |
"unable to allocate file descriptor overflow table - out of memory"); |
|
122 |
abort(); |
|
123 |
} |
|
124 |
} |
|
125 |
||
126 |
sa.sa_handler = sig_wakeup; |
|
127 |
sa.sa_flags = 0; |
|
128 |
sigemptyset(&sa.sa_mask); |
|
129 |
sigaction(sigWakeup, &sa, NULL); |
|
130 |
||
131 |
sigemptyset(&sigset); |
|
132 |
sigaddset(&sigset, sigWakeup); |
|
133 |
sigprocmask(SIG_UNBLOCK, &sigset, NULL); |
|
134 |
} |
|
135 |
||
136 |
static inline fdEntry_t *getFdEntry(int fd) { |
|
137 |
fdEntry_t* result = NULL; |
|
138 |
||
139 |
if (fd < 0) { |
|
140 |
return NULL; |
|
141 |
} |
|
142 |
||
143 |
assert(fd < fdLimit); |
|
144 |
||
145 |
if (fd < fdTableMaxSize) { |
|
146 |
assert(fd < fdTableLen); |
|
147 |
result = &fdTable[fd]; |
|
148 |
} else { |
|
149 |
const int indexInOverflowTable = fd - fdTableMaxSize; |
|
150 |
const int rootindex = indexInOverflowTable / fdOverflowTableSlabSize; |
|
151 |
const int slabindex = indexInOverflowTable % fdOverflowTableSlabSize; |
|
152 |
fdEntry_t* slab = NULL; |
|
153 |
assert(rootindex < fdOverflowTableLen); |
|
154 |
assert(slabindex < fdOverflowTableSlabSize); |
|
155 |
pthread_mutex_lock(&fdOverflowTableLock); |
|
156 |
if (fdOverflowTable[rootindex] == NULL) { |
|
157 |
fdEntry_t* const newSlab = |
|
158 |
(fdEntry_t*)calloc(fdOverflowTableSlabSize, sizeof(fdEntry_t)); |
|
159 |
if (newSlab == NULL) { |
|
160 |
fprintf(stderr, "Unable to allocate file descriptor overflow" |
|
161 |
" table slab - out of memory"); |
|
162 |
pthread_mutex_unlock(&fdOverflowTableLock); |
|
163 |
abort(); |
|
164 |
} else { |
|
165 |
int i; |
|
166 |
for (i = 0; i < fdOverflowTableSlabSize; i ++) { |
|
167 |
pthread_mutex_init(&newSlab[i].lock, NULL); |
|
168 |
} |
|
169 |
fdOverflowTable[rootindex] = newSlab; |
|
170 |
} |
|
171 |
} |
|
172 |
pthread_mutex_unlock(&fdOverflowTableLock); |
|
173 |
slab = fdOverflowTable[rootindex]; |
|
174 |
result = &slab[slabindex]; |
|
175 |
} |
|
176 |
return result; |
|
177 |
} |
|
178 |
||
179 |
static inline void startOp(fdEntry_t *fdEntry, threadEntry_t *self) { |
|
180 |
self->thr = pthread_self(); |
|
181 |
self->intr = 0; |
|
182 |
||
183 |
pthread_mutex_lock(&(fdEntry->lock)); |
|
184 |
{ |
|
185 |
self->next = fdEntry->threads; |
|
186 |
fdEntry->threads = self; |
|
187 |
} |
|
188 |
pthread_mutex_unlock(&(fdEntry->lock)); |
|
189 |
} |
|
190 |
||
191 |
static inline void endOp (fdEntry_t *fdEntry, threadEntry_t *self) { |
|
192 |
int orig_errno = errno; |
|
193 |
pthread_mutex_lock(&(fdEntry->lock)); |
|
194 |
{ |
|
195 |
threadEntry_t *curr, *prev=NULL; |
|
196 |
curr = fdEntry->threads; |
|
197 |
while (curr != NULL) { |
|
198 |
if (curr == self) { |
|
199 |
if (curr->intr) { |
|
200 |
orig_errno = EBADF; |
|
201 |
} |
|
202 |
if (prev == NULL) { |
|
203 |
fdEntry->threads = curr->next; |
|
204 |
} else { |
|
205 |
prev->next = curr->next; |
|
206 |
} |
|
207 |
break; |
|
208 |
} |
|
209 |
prev = curr; |
|
210 |
curr = curr->next; |
|
211 |
} |
|
212 |
} |
|
213 |
pthread_mutex_unlock(&(fdEntry->lock)); |
|
214 |
errno = orig_errno; |
|
215 |
} |
|
216 |
||
217 |
#define RESTARTABLE(_cmd, _result) do { \ |
|
218 |
do { \ |
|
219 |
_result = _cmd; \ |
|
220 |
} while((_result == -1) && (errno == EINTR)); \ |
|
221 |
} while(0) |
|
222 |
||
223 |
int rdma_supported() { |
|
224 |
int one = 1; |
|
225 |
int rv, s; |
|
226 |
s = rs_socket(PF_INET, SOCK_STREAM, 0); |
|
227 |
if (s < 0) { |
|
57129
1e1db86ea836
rsocket-branch: Review comments from Brian, A, B, and D
chegar
parents:
57115
diff
changeset
|
228 |
rs_close(s); |
57115 | 229 |
return JNI_FALSE; |
230 |
} |
|
231 |
return JNI_TRUE; |
|
232 |
} |
|
233 |
||
234 |
int RDMA_MapSocketOption(jint cmd, int *level, int *optname) { |
|
235 |
static struct { |
|
236 |
jint cmd; |
|
237 |
int level; |
|
238 |
int optname; |
|
239 |
} const opts[] = { |
|
240 |
{ java_net_SocketOptions_TCP_NODELAY, IPPROTO_TCP, TCP_NODELAY }, |
|
241 |
{ java_net_SocketOptions_SO_SNDBUF, SOL_SOCKET, SO_SNDBUF }, |
|
242 |
{ java_net_SocketOptions_SO_RCVBUF, SOL_SOCKET, SO_RCVBUF }, |
|
243 |
{ java_net_SocketOptions_SO_REUSEADDR, SOL_SOCKET, SO_REUSEADDR }, |
|
244 |
{ jdk_net_RdmaSocketOptions_SQSIZE, SOL_RDMA, RDMA_SQSIZE }, |
|
245 |
{ jdk_net_RdmaSocketOptions_RQSIZE, SOL_RDMA, RDMA_RQSIZE }, |
|
246 |
{ jdk_net_RdmaSocketOptions_INLINE, SOL_RDMA, RDMA_INLINE }, |
|
247 |
}; |
|
248 |
int i; |
|
249 |
for (i=0; i<(int)(sizeof(opts) / sizeof(opts[0])); i++) { |
|
250 |
if (cmd == opts[i].cmd) { |
|
251 |
*level = opts[i].level; |
|
252 |
*optname = opts[i].optname; |
|
253 |
return 0; |
|
254 |
} |
|
255 |
} |
|
256 |
return -1; |
|
257 |
} |
|
258 |
||
259 |
int RDMA_GetSockOpt(int fd, int level, int opt, void *result, int *len) { |
|
260 |
int rv; |
|
261 |
socklen_t socklen = *len; |
|
262 |
||
263 |
rv = rs_getsockopt(fd, level, opt, result, &socklen); |
|
264 |
*len = socklen; |
|
265 |
||
266 |
if (rv < 0) { |
|
267 |
return rv; |
|
268 |
} |
|
269 |
||
270 |
if ((level == SOL_SOCKET) && ((opt == SO_SNDBUF) |
|
271 |
|| (opt == SO_RCVBUF))) { |
|
272 |
int n = *((int *)result); |
|
273 |
n /= 2; |
|
274 |
*((int *)result) = n; |
|
275 |
} |
|
276 |
return rv; |
|
277 |
} |
|
278 |
||
279 |
int RDMA_SetSockOpt(int fd, int level, int opt, const void *arg, int len) { |
|
280 |
int *bufsize; |
|
281 |
if (level == SOL_SOCKET && opt == SO_RCVBUF) { |
|
282 |
int *bufsize = (int *)arg; |
|
283 |
if (*bufsize < 1024) { |
|
284 |
*bufsize = 1024; |
|
285 |
} |
|
286 |
} |
|
287 |
||
288 |
return rs_setsockopt(fd, level, opt, arg, len); |
|
289 |
} |
|
290 |
||
291 |
int RDMA_Bind(int fd, SOCKETADDRESS *sa, int len) { |
|
292 |
int rv; |
|
293 |
int arg, alen; |
|
294 |
||
295 |
if (sa->sa.sa_family == AF_INET) { |
|
296 |
if ((ntohl(sa->sa4.sin_addr.s_addr) & 0x7f0000ff) == 0x7f0000ff) { |
|
297 |
errno = EADDRNOTAVAIL; |
|
298 |
return -1; |
|
299 |
} |
|
300 |
} |
|
301 |
rv = rs_bind(fd, &sa->sa, len); |
|
302 |
return rv; |
|
303 |
} |
|
304 |
||
305 |
jint RDMA_Wait(JNIEnv *env, jint fd, jint flags, jint timeout) { |
|
306 |
jlong prevNanoTime = JVM_NanoTime(env, 0); |
|
307 |
jlong nanoTimeout = (jlong) timeout * NET_NSEC_PER_MSEC; |
|
308 |
jint read_rv; |
|
309 |
||
310 |
while (1) { |
|
311 |
jlong newNanoTime; |
|
312 |
struct pollfd pfd; |
|
313 |
pfd.fd = fd; |
|
314 |
pfd.events = 0; |
|
315 |
if (flags & NET_WAIT_READ) |
|
316 |
pfd.events |= POLLIN; |
|
317 |
if (flags & NET_WAIT_WRITE) |
|
318 |
pfd.events |= POLLOUT; |
|
319 |
if (flags & NET_WAIT_CONNECT) |
|
320 |
pfd.events |= POLLOUT; |
|
321 |
||
322 |
errno = 0; |
|
323 |
read_rv = RDMA_Poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC); |
|
324 |
||
325 |
newNanoTime = JVM_NanoTime(env, 0); |
|
326 |
nanoTimeout -= (newNanoTime - prevNanoTime); |
|
327 |
if (nanoTimeout < NET_NSEC_PER_MSEC) { |
|
328 |
return read_rv > 0 ? 0 : -1; |
|
329 |
} |
|
330 |
prevNanoTime = newNanoTime; |
|
331 |
||
332 |
if (read_rv > 0) { |
|
333 |
break; |
|
334 |
} |
|
57129
1e1db86ea836
rsocket-branch: Review comments from Brian, A, B, and D
chegar
parents:
57115
diff
changeset
|
335 |
} |
57115 | 336 |
return (nanoTimeout / NET_NSEC_PER_MSEC); |
337 |
} |
|
338 |
||
339 |
static int rdma_closefd(int fd2) { |
|
340 |
int rv, orig_errno; |
|
341 |
fdEntry_t *fdEntry = getFdEntry(fd2); |
|
342 |
if (fdEntry == NULL) { |
|
343 |
errno = EBADF; |
|
344 |
return -1; |
|
345 |
} |
|
346 |
||
347 |
pthread_mutex_lock(&(fdEntry->lock)); |
|
348 |
do { |
|
349 |
rv = rs_close(fd2); |
|
350 |
} while (rv == -1 && errno == EINTR); |
|
351 |
||
352 |
threadEntry_t *curr = fdEntry->threads; |
|
353 |
while (curr != NULL) { |
|
354 |
curr->intr = 1; |
|
355 |
pthread_kill( curr->thr, sigWakeup ); |
|
356 |
curr = curr->next; |
|
357 |
} |
|
358 |
orig_errno = errno; |
|
359 |
pthread_mutex_unlock(&(fdEntry->lock)); |
|
360 |
errno = orig_errno; |
|
361 |
return rv; |
|
362 |
} |
|
363 |
||
364 |
int RDMA_SocketClose(int fd) { |
|
365 |
return rdma_closefd(fd); |
|
366 |
} |
|
367 |
||
368 |
int RDMA_Read(int s, void* buf, size_t len) { |
|
369 |
BLOCKING_IO_RETURN_INT(s, rs_recv(s, buf, len, 0)); |
|
370 |
} |
|
371 |
||
372 |
int RDMA_NonBlockingRead(int s, void* buf, size_t len) { |
|
373 |
BLOCKING_IO_RETURN_INT(s, rs_recv(s, buf, len, MSG_DONTWAIT)); |
|
374 |
} |
|
375 |
||
376 |
int RDMA_ReadV(int s, const struct iovec * vector, int count) { |
|
377 |
BLOCKING_IO_RETURN_INT(s, rs_readv(s, vector, count) ); |
|
378 |
} |
|
379 |
||
380 |
int RDMA_RecvFrom(int s, void *buf, int len, unsigned int flags, |
|
381 |
struct sockaddr *from, socklen_t *fromlen) { |
|
382 |
BLOCKING_IO_RETURN_INT(s, rs_recvfrom(s, buf, len, flags, from, fromlen)); |
|
383 |
} |
|
384 |
||
385 |
int RDMA_Send(int s, void *msg, int len, unsigned int flags) { |
|
386 |
BLOCKING_IO_RETURN_INT(s, rs_send(s, msg, len, flags)); |
|
387 |
} |
|
388 |
||
389 |
int RDMA_WriteV(int s, const struct iovec * vector, int count) { |
|
390 |
BLOCKING_IO_RETURN_INT(s, rs_writev(s, vector, count)); |
|
391 |
} |
|
392 |
||
393 |
int NET_RSendTo(int s, const void *msg, int len, unsigned int |
|
394 |
flags, const struct sockaddr *to, int tolen) { |
|
395 |
BLOCKING_IO_RETURN_INT(s, rs_sendto(s, msg, len, flags, to, tolen)); |
|
396 |
} |
|
397 |
||
398 |
int RDMA_Accept(int s, struct sockaddr *addr, socklen_t *addrlen) { |
|
399 |
BLOCKING_IO_RETURN_INT(s, rs_accept(s, addr, addrlen)); |
|
400 |
} |
|
401 |
||
402 |
int RDMA_Connect(int s, struct sockaddr *addr, int addrlen) { |
|
403 |
BLOCKING_IO_RETURN_INT(s, rs_connect(s, addr, addrlen)); |
|
404 |
} |
|
405 |
||
406 |
int RDMA_Poll(struct pollfd *ufds, unsigned int nfds, int timeout) { |
|
407 |
BLOCKING_IO_RETURN_INT(ufds[0].fd, rs_poll(ufds, nfds, timeout)); |
|
408 |
} |
|
409 |
||
410 |
int RDMA_Timeout(JNIEnv *env, int s, long timeout, jlong nanoTimeStamp) { |
|
411 |
jlong prevNanoTime = nanoTimeStamp; |
|
412 |
jlong nanoTimeout = (jlong)timeout * NET_NSEC_PER_MSEC; |
|
413 |
fdEntry_t *fdEntry = getFdEntry(s); |
|
414 |
||
415 |
if (fdEntry == NULL) { |
|
416 |
errno = EBADF; |
|
417 |
return -1; |
|
418 |
} |
|
419 |
||
420 |
for(;;) { |
|
421 |
struct pollfd pfd; |
|
422 |
int rv; |
|
423 |
threadEntry_t self; |
|
424 |
||
425 |
pfd.fd = s; |
|
426 |
pfd.events = POLLIN | POLLERR; |
|
427 |
||
428 |
startOp(fdEntry, &self); |
|
429 |
rv = rs_poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC); |
|
430 |
endOp(fdEntry, &self); |
|
431 |
if (rv < 0 && errno == EINTR) { |
|
432 |
jlong newNanoTime = JVM_NanoTime(env, 0); |
|
433 |
nanoTimeout -= newNanoTime - prevNanoTime; |
|
434 |
if (nanoTimeout < NET_NSEC_PER_MSEC) { |
|
435 |
return 0; |
|
436 |
} |
|
437 |
prevNanoTime = newNanoTime; |
|
438 |
} else { |
|
439 |
return rv; |
|
440 |
} |
|
441 |
} |
|
442 |
} |