57115
|
1 |
/*
|
|
2 |
* Copyright (c) 2018, Oracle and/or its affiliates. All rights reserved.
|
|
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) {
|
|
228 |
return JNI_FALSE;
|
|
229 |
}
|
|
230 |
return JNI_TRUE;
|
|
231 |
}
|
|
232 |
|
|
233 |
int RDMA_MapSocketOption(jint cmd, int *level, int *optname) {
|
|
234 |
static struct {
|
|
235 |
jint cmd;
|
|
236 |
int level;
|
|
237 |
int optname;
|
|
238 |
} const opts[] = {
|
|
239 |
{ java_net_SocketOptions_TCP_NODELAY, IPPROTO_TCP, TCP_NODELAY },
|
|
240 |
{ java_net_SocketOptions_SO_SNDBUF, SOL_SOCKET, SO_SNDBUF },
|
|
241 |
{ java_net_SocketOptions_SO_RCVBUF, SOL_SOCKET, SO_RCVBUF },
|
|
242 |
{ java_net_SocketOptions_SO_REUSEADDR, SOL_SOCKET, SO_REUSEADDR },
|
|
243 |
{ jdk_net_RdmaSocketOptions_SQSIZE, SOL_RDMA, RDMA_SQSIZE },
|
|
244 |
{ jdk_net_RdmaSocketOptions_RQSIZE, SOL_RDMA, RDMA_RQSIZE },
|
|
245 |
{ jdk_net_RdmaSocketOptions_INLINE, SOL_RDMA, RDMA_INLINE },
|
|
246 |
};
|
|
247 |
int i;
|
|
248 |
for (i=0; i<(int)(sizeof(opts) / sizeof(opts[0])); i++) {
|
|
249 |
if (cmd == opts[i].cmd) {
|
|
250 |
*level = opts[i].level;
|
|
251 |
*optname = opts[i].optname;
|
|
252 |
return 0;
|
|
253 |
}
|
|
254 |
}
|
|
255 |
return -1;
|
|
256 |
}
|
|
257 |
|
|
258 |
int RDMA_GetSockOpt(int fd, int level, int opt, void *result, int *len) {
|
|
259 |
int rv;
|
|
260 |
socklen_t socklen = *len;
|
|
261 |
|
|
262 |
rv = rs_getsockopt(fd, level, opt, result, &socklen);
|
|
263 |
*len = socklen;
|
|
264 |
|
|
265 |
if (rv < 0) {
|
|
266 |
return rv;
|
|
267 |
}
|
|
268 |
|
|
269 |
if ((level == SOL_SOCKET) && ((opt == SO_SNDBUF)
|
|
270 |
|| (opt == SO_RCVBUF))) {
|
|
271 |
int n = *((int *)result);
|
|
272 |
n /= 2;
|
|
273 |
*((int *)result) = n;
|
|
274 |
}
|
|
275 |
return rv;
|
|
276 |
}
|
|
277 |
|
|
278 |
int RDMA_SetSockOpt(int fd, int level, int opt, const void *arg, int len) {
|
|
279 |
int *bufsize;
|
|
280 |
if (level == SOL_SOCKET && opt == SO_RCVBUF) {
|
|
281 |
int *bufsize = (int *)arg;
|
|
282 |
if (*bufsize < 1024) {
|
|
283 |
*bufsize = 1024;
|
|
284 |
}
|
|
285 |
}
|
|
286 |
|
|
287 |
return rs_setsockopt(fd, level, opt, arg, len);
|
|
288 |
}
|
|
289 |
|
|
290 |
int RDMA_Bind(int fd, SOCKETADDRESS *sa, int len) {
|
|
291 |
int rv;
|
|
292 |
int arg, alen;
|
|
293 |
|
|
294 |
if (sa->sa.sa_family == AF_INET) {
|
|
295 |
if ((ntohl(sa->sa4.sin_addr.s_addr) & 0x7f0000ff) == 0x7f0000ff) {
|
|
296 |
errno = EADDRNOTAVAIL;
|
|
297 |
return -1;
|
|
298 |
}
|
|
299 |
}
|
|
300 |
rv = rs_bind(fd, &sa->sa, len);
|
|
301 |
return rv;
|
|
302 |
}
|
|
303 |
|
|
304 |
jint RDMA_Wait(JNIEnv *env, jint fd, jint flags, jint timeout) {
|
|
305 |
jlong prevNanoTime = JVM_NanoTime(env, 0);
|
|
306 |
jlong nanoTimeout = (jlong) timeout * NET_NSEC_PER_MSEC;
|
|
307 |
jint read_rv;
|
|
308 |
|
|
309 |
while (1) {
|
|
310 |
jlong newNanoTime;
|
|
311 |
struct pollfd pfd;
|
|
312 |
pfd.fd = fd;
|
|
313 |
pfd.events = 0;
|
|
314 |
if (flags & NET_WAIT_READ)
|
|
315 |
pfd.events |= POLLIN;
|
|
316 |
if (flags & NET_WAIT_WRITE)
|
|
317 |
pfd.events |= POLLOUT;
|
|
318 |
if (flags & NET_WAIT_CONNECT)
|
|
319 |
pfd.events |= POLLOUT;
|
|
320 |
|
|
321 |
errno = 0;
|
|
322 |
read_rv = RDMA_Poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC);
|
|
323 |
|
|
324 |
newNanoTime = JVM_NanoTime(env, 0);
|
|
325 |
nanoTimeout -= (newNanoTime - prevNanoTime);
|
|
326 |
if (nanoTimeout < NET_NSEC_PER_MSEC) {
|
|
327 |
return read_rv > 0 ? 0 : -1;
|
|
328 |
}
|
|
329 |
prevNanoTime = newNanoTime;
|
|
330 |
|
|
331 |
if (read_rv > 0) {
|
|
332 |
break;
|
|
333 |
}
|
|
334 |
}
|
|
335 |
return (nanoTimeout / NET_NSEC_PER_MSEC);
|
|
336 |
}
|
|
337 |
|
|
338 |
static int rdma_closefd(int fd2) {
|
|
339 |
int rv, orig_errno;
|
|
340 |
fdEntry_t *fdEntry = getFdEntry(fd2);
|
|
341 |
if (fdEntry == NULL) {
|
|
342 |
errno = EBADF;
|
|
343 |
return -1;
|
|
344 |
}
|
|
345 |
|
|
346 |
pthread_mutex_lock(&(fdEntry->lock));
|
|
347 |
do {
|
|
348 |
rv = rs_close(fd2);
|
|
349 |
} while (rv == -1 && errno == EINTR);
|
|
350 |
|
|
351 |
threadEntry_t *curr = fdEntry->threads;
|
|
352 |
while (curr != NULL) {
|
|
353 |
curr->intr = 1;
|
|
354 |
pthread_kill( curr->thr, sigWakeup );
|
|
355 |
curr = curr->next;
|
|
356 |
}
|
|
357 |
orig_errno = errno;
|
|
358 |
pthread_mutex_unlock(&(fdEntry->lock));
|
|
359 |
errno = orig_errno;
|
|
360 |
return rv;
|
|
361 |
}
|
|
362 |
|
|
363 |
int RDMA_SocketClose(int fd) {
|
|
364 |
return rdma_closefd(fd);
|
|
365 |
}
|
|
366 |
|
|
367 |
int RDMA_Read(int s, void* buf, size_t len) {
|
|
368 |
BLOCKING_IO_RETURN_INT(s, rs_recv(s, buf, len, 0));
|
|
369 |
}
|
|
370 |
|
|
371 |
int RDMA_NonBlockingRead(int s, void* buf, size_t len) {
|
|
372 |
BLOCKING_IO_RETURN_INT(s, rs_recv(s, buf, len, MSG_DONTWAIT));
|
|
373 |
}
|
|
374 |
|
|
375 |
int RDMA_ReadV(int s, const struct iovec * vector, int count) {
|
|
376 |
BLOCKING_IO_RETURN_INT(s, rs_readv(s, vector, count) );
|
|
377 |
}
|
|
378 |
|
|
379 |
int RDMA_RecvFrom(int s, void *buf, int len, unsigned int flags,
|
|
380 |
struct sockaddr *from, socklen_t *fromlen) {
|
|
381 |
BLOCKING_IO_RETURN_INT(s, rs_recvfrom(s, buf, len, flags, from, fromlen));
|
|
382 |
}
|
|
383 |
|
|
384 |
int RDMA_Send(int s, void *msg, int len, unsigned int flags) {
|
|
385 |
BLOCKING_IO_RETURN_INT(s, rs_send(s, msg, len, flags));
|
|
386 |
}
|
|
387 |
|
|
388 |
int RDMA_WriteV(int s, const struct iovec * vector, int count) {
|
|
389 |
BLOCKING_IO_RETURN_INT(s, rs_writev(s, vector, count));
|
|
390 |
}
|
|
391 |
|
|
392 |
int NET_RSendTo(int s, const void *msg, int len, unsigned int
|
|
393 |
flags, const struct sockaddr *to, int tolen) {
|
|
394 |
BLOCKING_IO_RETURN_INT(s, rs_sendto(s, msg, len, flags, to, tolen));
|
|
395 |
}
|
|
396 |
|
|
397 |
int RDMA_Accept(int s, struct sockaddr *addr, socklen_t *addrlen) {
|
|
398 |
BLOCKING_IO_RETURN_INT(s, rs_accept(s, addr, addrlen));
|
|
399 |
}
|
|
400 |
|
|
401 |
int RDMA_Connect(int s, struct sockaddr *addr, int addrlen) {
|
|
402 |
BLOCKING_IO_RETURN_INT(s, rs_connect(s, addr, addrlen));
|
|
403 |
}
|
|
404 |
|
|
405 |
int RDMA_Poll(struct pollfd *ufds, unsigned int nfds, int timeout) {
|
|
406 |
BLOCKING_IO_RETURN_INT(ufds[0].fd, rs_poll(ufds, nfds, timeout));
|
|
407 |
}
|
|
408 |
|
|
409 |
int RDMA_Timeout(JNIEnv *env, int s, long timeout, jlong nanoTimeStamp) {
|
|
410 |
jlong prevNanoTime = nanoTimeStamp;
|
|
411 |
jlong nanoTimeout = (jlong)timeout * NET_NSEC_PER_MSEC;
|
|
412 |
fdEntry_t *fdEntry = getFdEntry(s);
|
|
413 |
|
|
414 |
if (fdEntry == NULL) {
|
|
415 |
errno = EBADF;
|
|
416 |
return -1;
|
|
417 |
}
|
|
418 |
|
|
419 |
for(;;) {
|
|
420 |
struct pollfd pfd;
|
|
421 |
int rv;
|
|
422 |
threadEntry_t self;
|
|
423 |
|
|
424 |
pfd.fd = s;
|
|
425 |
pfd.events = POLLIN | POLLERR;
|
|
426 |
|
|
427 |
startOp(fdEntry, &self);
|
|
428 |
rv = rs_poll(&pfd, 1, nanoTimeout / NET_NSEC_PER_MSEC);
|
|
429 |
endOp(fdEntry, &self);
|
|
430 |
if (rv < 0 && errno == EINTR) {
|
|
431 |
jlong newNanoTime = JVM_NanoTime(env, 0);
|
|
432 |
nanoTimeout -= newNanoTime - prevNanoTime;
|
|
433 |
if (nanoTimeout < NET_NSEC_PER_MSEC) {
|
|
434 |
return 0;
|
|
435 |
}
|
|
436 |
prevNanoTime = newNanoTime;
|
|
437 |
} else {
|
|
438 |
return rv;
|
|
439 |
}
|
|
440 |
}
|
|
441 |
}
|