1 /* Copyright Joyent, Inc. and other Node contributors. All rights reserved.
2 *
3 * Permission is hereby granted, free of charge, to any person obtaining a copy
4 * of this software and associated documentation files (the "Software"), to
5 * deal in the Software without restriction, including without limitation the
6 * rights to use, copy, modify, merge, publish, distribute, sublicense, and/or
7 * sell copies of the Software, and to permit persons to whom the Software is
8 * furnished to do so, subject to the following conditions:
9 *
10 * The above copyright notice and this permission notice shall be included in
11 * all copies or substantial portions of the Software.
12 *
13 * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
14 * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
15 * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
16 * AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
17 * LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING
18 * FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS
19 * IN THE SOFTWARE.
20 */
21
22 #include <assert.h>
23 #include <stdlib.h>
24
25 #include "uv.h"
26 #include "internal.h"
27 #include "handle-inl.h"
28 #include "stream-inl.h"
29 #include "req-inl.h"
30
31
32 /*
33 * Number of simultaneous pending AcceptEx calls.
34 */
35 const unsigned int uv_simultaneous_server_accepts = 32;
36
37 /* A zero-size buffer for use by uv_tcp_read */
38 static char uv_zero_[] = "";
39
uv__tcp_nodelay(uv_tcp_t * handle,SOCKET socket,int enable)40 static int uv__tcp_nodelay(uv_tcp_t* handle, SOCKET socket, int enable) {
41 if (setsockopt(socket,
42 IPPROTO_TCP,
43 TCP_NODELAY,
44 (const char*)&enable,
45 sizeof enable) == -1) {
46 return WSAGetLastError();
47 }
48 return 0;
49 }
50
51
uv__tcp_keepalive(uv_tcp_t * handle,SOCKET socket,int enable,unsigned int delay)52 static int uv__tcp_keepalive(uv_tcp_t* handle, SOCKET socket, int enable, unsigned int delay) {
53 if (setsockopt(socket,
54 SOL_SOCKET,
55 SO_KEEPALIVE,
56 (const char*)&enable,
57 sizeof enable) == -1) {
58 return WSAGetLastError();
59 }
60
61 if (!enable)
62 return 0;
63
64 if (delay < 1)
65 return UV_EINVAL;
66
67 if (setsockopt(socket,
68 IPPROTO_TCP,
69 TCP_KEEPALIVE,
70 (const char*)&delay,
71 sizeof delay) == -1) {
72 return WSAGetLastError();
73 }
74
75 return 0;
76 }
77
78
uv__tcp_set_socket(uv_loop_t * loop,uv_tcp_t * handle,SOCKET socket,int family,int imported)79 static int uv__tcp_set_socket(uv_loop_t* loop,
80 uv_tcp_t* handle,
81 SOCKET socket,
82 int family,
83 int imported) {
84 DWORD yes = 1;
85 int non_ifs_lsp;
86 int err;
87
88 if (handle->socket != INVALID_SOCKET)
89 return UV_EBUSY;
90
91 /* Set the socket to nonblocking mode */
92 if (ioctlsocket(socket, FIONBIO, &yes) == SOCKET_ERROR) {
93 return WSAGetLastError();
94 }
95
96 /* Make the socket non-inheritable */
97 if (!SetHandleInformation((HANDLE) socket, HANDLE_FLAG_INHERIT, 0))
98 return GetLastError();
99
100 /* Associate it with the I/O completion port. Use uv_handle_t pointer as
101 * completion key. */
102 if (CreateIoCompletionPort((HANDLE)socket,
103 loop->iocp,
104 (ULONG_PTR)socket,
105 0) == NULL) {
106 if (imported) {
107 handle->flags |= UV_HANDLE_EMULATE_IOCP;
108 } else {
109 return GetLastError();
110 }
111 }
112
113 if (family == AF_INET6) {
114 non_ifs_lsp = uv_tcp_non_ifs_lsp_ipv6;
115 } else {
116 non_ifs_lsp = uv_tcp_non_ifs_lsp_ipv4;
117 }
118
119 if (!(handle->flags & UV_HANDLE_EMULATE_IOCP) && !non_ifs_lsp) {
120 UCHAR sfcnm_flags =
121 FILE_SKIP_SET_EVENT_ON_HANDLE | FILE_SKIP_COMPLETION_PORT_ON_SUCCESS;
122 if (!SetFileCompletionNotificationModes((HANDLE) socket, sfcnm_flags))
123 return GetLastError();
124 handle->flags |= UV_HANDLE_SYNC_BYPASS_IOCP;
125 }
126
127 if (handle->flags & UV_HANDLE_TCP_NODELAY) {
128 err = uv__tcp_nodelay(handle, socket, 1);
129 if (err)
130 return err;
131 }
132
133 /* TODO: Use stored delay. */
134 if (handle->flags & UV_HANDLE_TCP_KEEPALIVE) {
135 err = uv__tcp_keepalive(handle, socket, 1, 60);
136 if (err)
137 return err;
138 }
139
140 handle->socket = socket;
141
142 if (family == AF_INET6) {
143 handle->flags |= UV_HANDLE_IPV6;
144 } else {
145 assert(!(handle->flags & UV_HANDLE_IPV6));
146 }
147
148 return 0;
149 }
150
151
uv_tcp_init_ex(uv_loop_t * loop,uv_tcp_t * handle,unsigned int flags)152 int uv_tcp_init_ex(uv_loop_t* loop, uv_tcp_t* handle, unsigned int flags) {
153 int domain;
154
155 /* Use the lower 8 bits for the domain */
156 domain = flags & 0xFF;
157 if (domain != AF_INET && domain != AF_INET6 && domain != AF_UNSPEC)
158 return UV_EINVAL;
159
160 if (flags & ~0xFF)
161 return UV_EINVAL;
162
163 uv__stream_init(loop, (uv_stream_t*) handle, UV_TCP);
164 handle->tcp.serv.accept_reqs = NULL;
165 handle->tcp.serv.pending_accepts = NULL;
166 handle->socket = INVALID_SOCKET;
167 handle->reqs_pending = 0;
168 handle->tcp.serv.func_acceptex = NULL;
169 handle->tcp.conn.func_connectex = NULL;
170 handle->tcp.serv.processed_accepts = 0;
171 handle->delayed_error = 0;
172
173 /* If anything fails beyond this point we need to remove the handle from
174 * the handle queue, since it was added by uv__handle_init in uv__stream_init.
175 */
176
177 if (domain != AF_UNSPEC) {
178 SOCKET sock;
179 DWORD err;
180
181 sock = socket(domain, SOCK_STREAM, 0);
182 if (sock == INVALID_SOCKET) {
183 err = WSAGetLastError();
184 uv__queue_remove(&handle->handle_queue);
185 return uv_translate_sys_error(err);
186 }
187
188 err = uv__tcp_set_socket(handle->loop, handle, sock, domain, 0);
189 if (err) {
190 closesocket(sock);
191 uv__queue_remove(&handle->handle_queue);
192 return uv_translate_sys_error(err);
193 }
194
195 }
196
197 return 0;
198 }
199
200
uv_tcp_init(uv_loop_t * loop,uv_tcp_t * handle)201 int uv_tcp_init(uv_loop_t* loop, uv_tcp_t* handle) {
202 return uv_tcp_init_ex(loop, handle, AF_UNSPEC);
203 }
204
205
uv__process_tcp_shutdown_req(uv_loop_t * loop,uv_tcp_t * stream,uv_shutdown_t * req)206 void uv__process_tcp_shutdown_req(uv_loop_t* loop, uv_tcp_t* stream, uv_shutdown_t *req) {
207 int err;
208
209 assert(req);
210 assert(stream->stream.conn.write_reqs_pending == 0);
211 assert(!(stream->flags & UV_HANDLE_SHUT));
212 assert(stream->flags & UV_HANDLE_CONNECTION);
213
214 stream->stream.conn.shutdown_req = NULL;
215 UNREGISTER_HANDLE_REQ(loop, stream, req);
216
217 err = 0;
218 if (stream->flags & UV_HANDLE_CLOSING)
219 /* The user destroyed the stream before we got to do the shutdown. */
220 err = UV_ECANCELED;
221 else if (shutdown(stream->socket, SD_SEND) == SOCKET_ERROR)
222 err = uv_translate_sys_error(WSAGetLastError());
223 else /* Success. */
224 stream->flags |= UV_HANDLE_SHUT;
225
226 if (req->cb)
227 req->cb(req, err);
228
229 DECREASE_PENDING_REQ_COUNT(stream);
230 }
231
232
uv__tcp_endgame(uv_loop_t * loop,uv_tcp_t * handle)233 void uv__tcp_endgame(uv_loop_t* loop, uv_tcp_t* handle) {
234 unsigned int i;
235 uv_tcp_accept_t* req;
236
237 assert(handle->flags & UV_HANDLE_CLOSING);
238 assert(handle->reqs_pending == 0);
239 assert(!(handle->flags & UV_HANDLE_CLOSED));
240 assert(handle->socket == INVALID_SOCKET);
241
242 if (!(handle->flags & UV_HANDLE_CONNECTION) && handle->tcp.serv.accept_reqs) {
243 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
244 for (i = 0; i < uv_simultaneous_server_accepts; i++) {
245 req = &handle->tcp.serv.accept_reqs[i];
246 if (req->wait_handle != INVALID_HANDLE_VALUE) {
247 UnregisterWait(req->wait_handle);
248 req->wait_handle = INVALID_HANDLE_VALUE;
249 }
250 if (req->event_handle != NULL) {
251 CloseHandle(req->event_handle);
252 req->event_handle = NULL;
253 }
254 }
255 }
256
257 uv__free(handle->tcp.serv.accept_reqs);
258 handle->tcp.serv.accept_reqs = NULL;
259 }
260
261 if (handle->flags & UV_HANDLE_CONNECTION &&
262 handle->flags & UV_HANDLE_EMULATE_IOCP) {
263 if (handle->read_req.wait_handle != INVALID_HANDLE_VALUE) {
264 UnregisterWait(handle->read_req.wait_handle);
265 handle->read_req.wait_handle = INVALID_HANDLE_VALUE;
266 }
267 if (handle->read_req.event_handle != NULL) {
268 CloseHandle(handle->read_req.event_handle);
269 handle->read_req.event_handle = NULL;
270 }
271 }
272
273 uv__handle_close(handle);
274 }
275
276
277 /* Unlike on Unix, here we don't set SO_REUSEADDR, because it doesn't just
278 * allow binding to addresses that are in use by sockets in TIME_WAIT, it
279 * effectively allows 'stealing' a port which is in use by another application.
280 *
281 * SO_EXCLUSIVEADDRUSE is also not good here because it does check all sockets,
282 * regardless of state, so we'd get an error even if the port is in use by a
283 * socket in TIME_WAIT state.
284 *
285 * See issue #1360.
286 *
287 */
uv__tcp_try_bind(uv_tcp_t * handle,const struct sockaddr * addr,unsigned int addrlen,unsigned int flags)288 static int uv__tcp_try_bind(uv_tcp_t* handle,
289 const struct sockaddr* addr,
290 unsigned int addrlen,
291 unsigned int flags) {
292 DWORD err;
293 int r;
294
295 if (handle->socket == INVALID_SOCKET) {
296 SOCKET sock;
297
298 /* Cannot set IPv6-only mode on non-IPv6 socket. */
299 if ((flags & UV_TCP_IPV6ONLY) && addr->sa_family != AF_INET6)
300 return ERROR_INVALID_PARAMETER;
301
302 sock = socket(addr->sa_family, SOCK_STREAM, 0);
303 if (sock == INVALID_SOCKET) {
304 return WSAGetLastError();
305 }
306
307 err = uv__tcp_set_socket(handle->loop, handle, sock, addr->sa_family, 0);
308 if (err) {
309 closesocket(sock);
310 return err;
311 }
312 }
313
314 #ifdef IPV6_V6ONLY
315 if (addr->sa_family == AF_INET6) {
316 int on;
317
318 on = (flags & UV_TCP_IPV6ONLY) != 0;
319
320 /* TODO: how to handle errors? This may fail if there is no ipv4 stack
321 * available, or when run on XP/2003 which have no support for dualstack
322 * sockets. For now we're silently ignoring the error. */
323 setsockopt(handle->socket,
324 IPPROTO_IPV6,
325 IPV6_V6ONLY,
326 (const char*)&on,
327 sizeof on);
328 }
329 #endif
330
331 r = bind(handle->socket, addr, addrlen);
332
333 if (r == SOCKET_ERROR) {
334 err = WSAGetLastError();
335 if (err == WSAEADDRINUSE) {
336 /* Some errors are not to be reported until connect() or listen() */
337 handle->delayed_error = err;
338 } else {
339 return err;
340 }
341 }
342
343 handle->flags |= UV_HANDLE_BOUND;
344
345 return 0;
346 }
347
348
post_completion(void * context,BOOLEAN timed_out)349 static void CALLBACK post_completion(void* context, BOOLEAN timed_out) {
350 uv_req_t* req;
351 uv_tcp_t* handle;
352
353 req = (uv_req_t*) context;
354 assert(req != NULL);
355 handle = (uv_tcp_t*)req->data;
356 assert(handle != NULL);
357 assert(!timed_out);
358
359 if (!PostQueuedCompletionStatus(handle->loop->iocp,
360 req->u.io.overlapped.InternalHigh,
361 0,
362 &req->u.io.overlapped)) {
363 uv_fatal_error(GetLastError(), "PostQueuedCompletionStatus");
364 }
365 }
366
367
post_write_completion(void * context,BOOLEAN timed_out)368 static void CALLBACK post_write_completion(void* context, BOOLEAN timed_out) {
369 uv_write_t* req;
370 uv_tcp_t* handle;
371
372 req = (uv_write_t*) context;
373 assert(req != NULL);
374 handle = (uv_tcp_t*)req->handle;
375 assert(handle != NULL);
376 assert(!timed_out);
377
378 if (!PostQueuedCompletionStatus(handle->loop->iocp,
379 req->u.io.overlapped.InternalHigh,
380 0,
381 &req->u.io.overlapped)) {
382 uv_fatal_error(GetLastError(), "PostQueuedCompletionStatus");
383 }
384 }
385
386
uv__tcp_queue_accept(uv_tcp_t * handle,uv_tcp_accept_t * req)387 static void uv__tcp_queue_accept(uv_tcp_t* handle, uv_tcp_accept_t* req) {
388 uv_loop_t* loop = handle->loop;
389 BOOL success;
390 DWORD bytes;
391 SOCKET accept_socket;
392 short family;
393
394 assert(handle->flags & UV_HANDLE_LISTENING);
395 assert(req->accept_socket == INVALID_SOCKET);
396
397 /* choose family and extension function */
398 if (handle->flags & UV_HANDLE_IPV6) {
399 family = AF_INET6;
400 } else {
401 family = AF_INET;
402 }
403
404 /* Open a socket for the accepted connection. */
405 accept_socket = socket(family, SOCK_STREAM, 0);
406 if (accept_socket == INVALID_SOCKET) {
407 SET_REQ_ERROR(req, WSAGetLastError());
408 uv__insert_pending_req(loop, (uv_req_t*)req);
409 handle->reqs_pending++;
410 return;
411 }
412
413 /* Make the socket non-inheritable */
414 if (!SetHandleInformation((HANDLE) accept_socket, HANDLE_FLAG_INHERIT, 0)) {
415 SET_REQ_ERROR(req, GetLastError());
416 uv__insert_pending_req(loop, (uv_req_t*)req);
417 handle->reqs_pending++;
418 closesocket(accept_socket);
419 return;
420 }
421
422 /* Prepare the overlapped structure. */
423 memset(&(req->u.io.overlapped), 0, sizeof(req->u.io.overlapped));
424 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
425 assert(req->event_handle != NULL);
426 req->u.io.overlapped.hEvent = (HANDLE) ((ULONG_PTR) req->event_handle | 1);
427 }
428
429 success = handle->tcp.serv.func_acceptex(handle->socket,
430 accept_socket,
431 (void*)req->accept_buffer,
432 0,
433 sizeof(struct sockaddr_storage),
434 sizeof(struct sockaddr_storage),
435 &bytes,
436 &req->u.io.overlapped);
437
438 if (UV_SUCCEEDED_WITHOUT_IOCP(success)) {
439 /* Process the req without IOCP. */
440 req->accept_socket = accept_socket;
441 handle->reqs_pending++;
442 uv__insert_pending_req(loop, (uv_req_t*)req);
443 } else if (UV_SUCCEEDED_WITH_IOCP(success)) {
444 /* The req will be processed with IOCP. */
445 req->accept_socket = accept_socket;
446 handle->reqs_pending++;
447 if (handle->flags & UV_HANDLE_EMULATE_IOCP &&
448 req->wait_handle == INVALID_HANDLE_VALUE &&
449 !RegisterWaitForSingleObject(&req->wait_handle,
450 req->event_handle, post_completion, (void*) req,
451 INFINITE, WT_EXECUTEINWAITTHREAD)) {
452 SET_REQ_ERROR(req, GetLastError());
453 uv__insert_pending_req(loop, (uv_req_t*)req);
454 }
455 } else {
456 /* Make this req pending reporting an error. */
457 SET_REQ_ERROR(req, WSAGetLastError());
458 uv__insert_pending_req(loop, (uv_req_t*)req);
459 handle->reqs_pending++;
460 /* Destroy the preallocated client socket. */
461 closesocket(accept_socket);
462 /* Destroy the event handle */
463 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
464 CloseHandle(req->event_handle);
465 req->event_handle = NULL;
466 }
467 }
468 }
469
470
uv__tcp_queue_read(uv_loop_t * loop,uv_tcp_t * handle)471 static void uv__tcp_queue_read(uv_loop_t* loop, uv_tcp_t* handle) {
472 uv_read_t* req;
473 uv_buf_t buf;
474 int result;
475 DWORD bytes, flags;
476
477 assert(handle->flags & UV_HANDLE_READING);
478 assert(!(handle->flags & UV_HANDLE_READ_PENDING));
479
480 req = &handle->read_req;
481 memset(&req->u.io.overlapped, 0, sizeof(req->u.io.overlapped));
482
483 handle->flags |= UV_HANDLE_ZERO_READ;
484 buf.base = (char*) &uv_zero_;
485 buf.len = 0;
486
487 /* Prepare the overlapped structure. */
488 memset(&(req->u.io.overlapped), 0, sizeof(req->u.io.overlapped));
489 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
490 assert(req->event_handle != NULL);
491 req->u.io.overlapped.hEvent = (HANDLE) ((ULONG_PTR) req->event_handle | 1);
492 }
493
494 flags = 0;
495 result = WSARecv(handle->socket,
496 (WSABUF*)&buf,
497 1,
498 &bytes,
499 &flags,
500 &req->u.io.overlapped,
501 NULL);
502
503 handle->flags |= UV_HANDLE_READ_PENDING;
504 handle->reqs_pending++;
505
506 if (UV_SUCCEEDED_WITHOUT_IOCP(result == 0)) {
507 /* Process the req without IOCP. */
508 req->u.io.overlapped.InternalHigh = bytes;
509 uv__insert_pending_req(loop, (uv_req_t*)req);
510 } else if (UV_SUCCEEDED_WITH_IOCP(result == 0)) {
511 /* The req will be processed with IOCP. */
512 if (handle->flags & UV_HANDLE_EMULATE_IOCP &&
513 req->wait_handle == INVALID_HANDLE_VALUE &&
514 !RegisterWaitForSingleObject(&req->wait_handle,
515 req->event_handle, post_completion, (void*) req,
516 INFINITE, WT_EXECUTEINWAITTHREAD)) {
517 SET_REQ_ERROR(req, GetLastError());
518 uv__insert_pending_req(loop, (uv_req_t*)req);
519 }
520 } else {
521 /* Make this req pending reporting an error. */
522 SET_REQ_ERROR(req, WSAGetLastError());
523 uv__insert_pending_req(loop, (uv_req_t*)req);
524 }
525 }
526
527
uv_tcp_close_reset(uv_tcp_t * handle,uv_close_cb close_cb)528 int uv_tcp_close_reset(uv_tcp_t* handle, uv_close_cb close_cb) {
529 struct linger l = { 1, 0 };
530
531 /* Disallow setting SO_LINGER to zero due to some platform inconsistencies */
532 if (uv__is_stream_shutting(handle))
533 return UV_EINVAL;
534
535 if (0 != setsockopt(handle->socket, SOL_SOCKET, SO_LINGER, (const char*)&l, sizeof(l)))
536 return uv_translate_sys_error(WSAGetLastError());
537
538 uv_close((uv_handle_t*) handle, close_cb);
539 return 0;
540 }
541
542
uv__tcp_listen(uv_tcp_t * handle,int backlog,uv_connection_cb cb)543 int uv__tcp_listen(uv_tcp_t* handle, int backlog, uv_connection_cb cb) {
544 unsigned int i, simultaneous_accepts;
545 uv_tcp_accept_t* req;
546 int err;
547
548 assert(backlog > 0);
549
550 if (handle->flags & UV_HANDLE_LISTENING) {
551 handle->stream.serv.connection_cb = cb;
552 }
553
554 if (handle->flags & UV_HANDLE_READING) {
555 return WSAEISCONN;
556 }
557
558 if (handle->delayed_error) {
559 return handle->delayed_error;
560 }
561
562 if (!(handle->flags & UV_HANDLE_BOUND)) {
563 err = uv__tcp_try_bind(handle,
564 (const struct sockaddr*) &uv_addr_ip4_any_,
565 sizeof(uv_addr_ip4_any_),
566 0);
567 if (err)
568 return err;
569 if (handle->delayed_error)
570 return handle->delayed_error;
571 }
572
573 if (!handle->tcp.serv.func_acceptex) {
574 if (!uv__get_acceptex_function(handle->socket, &handle->tcp.serv.func_acceptex)) {
575 return WSAEAFNOSUPPORT;
576 }
577 }
578
579 /* If this flag is set, we already made this listen call in xfer. */
580 if (!(handle->flags & UV_HANDLE_SHARED_TCP_SOCKET) &&
581 listen(handle->socket, backlog) == SOCKET_ERROR) {
582 return WSAGetLastError();
583 }
584
585 handle->flags |= UV_HANDLE_LISTENING;
586 handle->stream.serv.connection_cb = cb;
587 INCREASE_ACTIVE_COUNT(loop, handle);
588
589 simultaneous_accepts = handle->flags & UV_HANDLE_TCP_SINGLE_ACCEPT ? 1
590 : uv_simultaneous_server_accepts;
591
592 if (handle->tcp.serv.accept_reqs == NULL) {
593 handle->tcp.serv.accept_reqs =
594 uv__malloc(uv_simultaneous_server_accepts * sizeof(uv_tcp_accept_t));
595 if (!handle->tcp.serv.accept_reqs) {
596 uv_fatal_error(ERROR_OUTOFMEMORY, "uv__malloc");
597 }
598
599 for (i = 0; i < simultaneous_accepts; i++) {
600 req = &handle->tcp.serv.accept_reqs[i];
601 UV_REQ_INIT(req, UV_ACCEPT);
602 req->accept_socket = INVALID_SOCKET;
603 req->data = handle;
604
605 req->wait_handle = INVALID_HANDLE_VALUE;
606 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
607 req->event_handle = CreateEvent(NULL, 0, 0, NULL);
608 if (req->event_handle == NULL) {
609 uv_fatal_error(GetLastError(), "CreateEvent");
610 }
611 } else {
612 req->event_handle = NULL;
613 }
614
615 uv__tcp_queue_accept(handle, req);
616 }
617
618 /* Initialize other unused requests too, because uv_tcp_endgame doesn't
619 * know how many requests were initialized, so it will try to clean up
620 * {uv_simultaneous_server_accepts} requests. */
621 for (i = simultaneous_accepts; i < uv_simultaneous_server_accepts; i++) {
622 req = &handle->tcp.serv.accept_reqs[i];
623 UV_REQ_INIT(req, UV_ACCEPT);
624 req->accept_socket = INVALID_SOCKET;
625 req->data = handle;
626 req->wait_handle = INVALID_HANDLE_VALUE;
627 req->event_handle = NULL;
628 }
629 }
630
631 return 0;
632 }
633
634
uv__tcp_accept(uv_tcp_t * server,uv_tcp_t * client)635 int uv__tcp_accept(uv_tcp_t* server, uv_tcp_t* client) {
636 int err = 0;
637 int family;
638
639 uv_tcp_accept_t* req = server->tcp.serv.pending_accepts;
640
641 if (!req) {
642 /* No valid connections found, so we error out. */
643 return WSAEWOULDBLOCK;
644 }
645
646 if (req->accept_socket == INVALID_SOCKET) {
647 return WSAENOTCONN;
648 }
649
650 if (server->flags & UV_HANDLE_IPV6) {
651 family = AF_INET6;
652 } else {
653 family = AF_INET;
654 }
655
656 err = uv__tcp_set_socket(client->loop,
657 client,
658 req->accept_socket,
659 family,
660 0);
661 if (err) {
662 closesocket(req->accept_socket);
663 } else {
664 uv__connection_init((uv_stream_t*) client);
665 /* AcceptEx() implicitly binds the accepted socket. */
666 client->flags |= UV_HANDLE_BOUND | UV_HANDLE_READABLE | UV_HANDLE_WRITABLE;
667 }
668
669 /* Prepare the req to pick up a new connection */
670 server->tcp.serv.pending_accepts = req->next_pending;
671 req->next_pending = NULL;
672 req->accept_socket = INVALID_SOCKET;
673
674 if (!(server->flags & UV_HANDLE_CLOSING)) {
675 /* Check if we're in a middle of changing the number of pending accepts. */
676 if (!(server->flags & UV_HANDLE_TCP_ACCEPT_STATE_CHANGING)) {
677 uv__tcp_queue_accept(server, req);
678 } else {
679 /* We better be switching to a single pending accept. */
680 assert(server->flags & UV_HANDLE_TCP_SINGLE_ACCEPT);
681
682 server->tcp.serv.processed_accepts++;
683
684 if (server->tcp.serv.processed_accepts >= uv_simultaneous_server_accepts) {
685 server->tcp.serv.processed_accepts = 0;
686 /*
687 * All previously queued accept requests are now processed.
688 * We now switch to queueing just a single accept.
689 */
690 uv__tcp_queue_accept(server, &server->tcp.serv.accept_reqs[0]);
691 server->flags &= ~UV_HANDLE_TCP_ACCEPT_STATE_CHANGING;
692 server->flags |= UV_HANDLE_TCP_SINGLE_ACCEPT;
693 }
694 }
695 }
696
697 return err;
698 }
699
700
uv__tcp_read_start(uv_tcp_t * handle,uv_alloc_cb alloc_cb,uv_read_cb read_cb)701 int uv__tcp_read_start(uv_tcp_t* handle, uv_alloc_cb alloc_cb,
702 uv_read_cb read_cb) {
703 uv_loop_t* loop = handle->loop;
704
705 handle->flags |= UV_HANDLE_READING;
706 handle->read_cb = read_cb;
707 handle->alloc_cb = alloc_cb;
708 INCREASE_ACTIVE_COUNT(loop, handle);
709
710 /* If reading was stopped and then started again, there could still be a read
711 * request pending. */
712 if (!(handle->flags & UV_HANDLE_READ_PENDING)) {
713 if (handle->flags & UV_HANDLE_EMULATE_IOCP &&
714 handle->read_req.event_handle == NULL) {
715 handle->read_req.event_handle = CreateEvent(NULL, 0, 0, NULL);
716 if (handle->read_req.event_handle == NULL) {
717 uv_fatal_error(GetLastError(), "CreateEvent");
718 }
719 }
720 uv__tcp_queue_read(loop, handle);
721 }
722
723 return 0;
724 }
725
uv__is_loopback(const struct sockaddr_storage * storage)726 static int uv__is_loopback(const struct sockaddr_storage* storage) {
727 const struct sockaddr_in* in4;
728 const struct sockaddr_in6* in6;
729 int i;
730
731 if (storage->ss_family == AF_INET) {
732 in4 = (const struct sockaddr_in*) storage;
733 return in4->sin_addr.S_un.S_un_b.s_b1 == 127;
734 }
735 if (storage->ss_family == AF_INET6) {
736 in6 = (const struct sockaddr_in6*) storage;
737 for (i = 0; i < 7; ++i) {
738 if (in6->sin6_addr.u.Word[i] != 0)
739 return 0;
740 }
741 return in6->sin6_addr.u.Word[7] == htons(1);
742 }
743 return 0;
744 }
745
746 // Check if Windows version is 10.0.16299 or later
uv__is_fast_loopback_fail_supported(void)747 static int uv__is_fast_loopback_fail_supported(void) {
748 OSVERSIONINFOW os_info;
749 if (!pRtlGetVersion)
750 return 0;
751 pRtlGetVersion(&os_info);
752 if (os_info.dwMajorVersion < 10)
753 return 0;
754 if (os_info.dwMajorVersion > 10)
755 return 1;
756 if (os_info.dwMinorVersion > 0)
757 return 1;
758 return os_info.dwBuildNumber >= 16299;
759 }
760
uv__tcp_try_connect(uv_connect_t * req,uv_tcp_t * handle,const struct sockaddr * addr,unsigned int addrlen,uv_connect_cb cb)761 static int uv__tcp_try_connect(uv_connect_t* req,
762 uv_tcp_t* handle,
763 const struct sockaddr* addr,
764 unsigned int addrlen,
765 uv_connect_cb cb) {
766 uv_loop_t* loop = handle->loop;
767 TCP_INITIAL_RTO_PARAMETERS retransmit_ioctl;
768 const struct sockaddr* bind_addr;
769 struct sockaddr_storage converted;
770 BOOL success;
771 DWORD bytes;
772 int err;
773
774 err = uv__convert_to_localhost_if_unspecified(addr, &converted);
775 if (err)
776 return err;
777
778 if (handle->delayed_error != 0)
779 goto out;
780
781 if (!(handle->flags & UV_HANDLE_BOUND)) {
782 if (addrlen == sizeof(uv_addr_ip4_any_)) {
783 bind_addr = (const struct sockaddr*) &uv_addr_ip4_any_;
784 } else if (addrlen == sizeof(uv_addr_ip6_any_)) {
785 bind_addr = (const struct sockaddr*) &uv_addr_ip6_any_;
786 } else {
787 abort();
788 }
789 err = uv__tcp_try_bind(handle, bind_addr, addrlen, 0);
790 if (err)
791 return err;
792 if (handle->delayed_error != 0)
793 goto out;
794 }
795
796 if (!handle->tcp.conn.func_connectex) {
797 if (!uv__get_connectex_function(handle->socket, &handle->tcp.conn.func_connectex)) {
798 return WSAEAFNOSUPPORT;
799 }
800 }
801
802 /* This makes connect() fail instantly if the target port on the localhost
803 * is not reachable, instead of waiting for 2s. We do not care if this fails.
804 * This only works on Windows version 10.0.16299 and later.
805 */
806 if (uv__is_fast_loopback_fail_supported() && uv__is_loopback(&converted)) {
807 memset(&retransmit_ioctl, 0, sizeof(retransmit_ioctl));
808 retransmit_ioctl.Rtt = TCP_INITIAL_RTO_NO_SYN_RETRANSMISSIONS;
809 retransmit_ioctl.MaxSynRetransmissions = TCP_INITIAL_RTO_NO_SYN_RETRANSMISSIONS;
810 WSAIoctl(handle->socket,
811 SIO_TCP_INITIAL_RTO,
812 &retransmit_ioctl,
813 sizeof(retransmit_ioctl),
814 NULL,
815 0,
816 &bytes,
817 NULL,
818 NULL);
819 }
820
821 out:
822
823 UV_REQ_INIT(req, UV_CONNECT);
824 req->handle = (uv_stream_t*) handle;
825 req->cb = cb;
826 memset(&req->u.io.overlapped, 0, sizeof(req->u.io.overlapped));
827
828 if (handle->delayed_error != 0) {
829 /* Process the req without IOCP. */
830 handle->reqs_pending++;
831 REGISTER_HANDLE_REQ(loop, handle, req);
832 uv__insert_pending_req(loop, (uv_req_t*)req);
833 return 0;
834 }
835
836 success = handle->tcp.conn.func_connectex(handle->socket,
837 (const struct sockaddr*) &converted,
838 addrlen,
839 NULL,
840 0,
841 &bytes,
842 &req->u.io.overlapped);
843
844 if (UV_SUCCEEDED_WITHOUT_IOCP(success)) {
845 /* Process the req without IOCP. */
846 handle->reqs_pending++;
847 REGISTER_HANDLE_REQ(loop, handle, req);
848 uv__insert_pending_req(loop, (uv_req_t*)req);
849 } else if (UV_SUCCEEDED_WITH_IOCP(success)) {
850 /* The req will be processed with IOCP. */
851 handle->reqs_pending++;
852 REGISTER_HANDLE_REQ(loop, handle, req);
853 } else {
854 return WSAGetLastError();
855 }
856
857 return 0;
858 }
859
860
uv_tcp_getsockname(const uv_tcp_t * handle,struct sockaddr * name,int * namelen)861 int uv_tcp_getsockname(const uv_tcp_t* handle,
862 struct sockaddr* name,
863 int* namelen) {
864
865 return uv__getsockpeername((const uv_handle_t*) handle,
866 getsockname,
867 name,
868 namelen,
869 handle->delayed_error);
870 }
871
872
uv_tcp_getpeername(const uv_tcp_t * handle,struct sockaddr * name,int * namelen)873 int uv_tcp_getpeername(const uv_tcp_t* handle,
874 struct sockaddr* name,
875 int* namelen) {
876
877 return uv__getsockpeername((const uv_handle_t*) handle,
878 getpeername,
879 name,
880 namelen,
881 handle->delayed_error);
882 }
883
884
uv__tcp_write(uv_loop_t * loop,uv_write_t * req,uv_tcp_t * handle,const uv_buf_t bufs[],unsigned int nbufs,uv_write_cb cb)885 int uv__tcp_write(uv_loop_t* loop,
886 uv_write_t* req,
887 uv_tcp_t* handle,
888 const uv_buf_t bufs[],
889 unsigned int nbufs,
890 uv_write_cb cb) {
891 int result;
892 DWORD bytes;
893
894 UV_REQ_INIT(req, UV_WRITE);
895 req->handle = (uv_stream_t*) handle;
896 req->cb = cb;
897
898 /* Prepare the overlapped structure. */
899 memset(&(req->u.io.overlapped), 0, sizeof(req->u.io.overlapped));
900 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
901 req->event_handle = CreateEvent(NULL, 0, 0, NULL);
902 if (req->event_handle == NULL) {
903 uv_fatal_error(GetLastError(), "CreateEvent");
904 }
905 req->u.io.overlapped.hEvent = (HANDLE) ((ULONG_PTR) req->event_handle | 1);
906 req->wait_handle = INVALID_HANDLE_VALUE;
907 }
908
909 result = WSASend(handle->socket,
910 (WSABUF*) bufs,
911 nbufs,
912 &bytes,
913 0,
914 &req->u.io.overlapped,
915 NULL);
916
917 if (UV_SUCCEEDED_WITHOUT_IOCP(result == 0)) {
918 /* Request completed immediately. */
919 req->u.io.queued_bytes = 0;
920 handle->reqs_pending++;
921 handle->stream.conn.write_reqs_pending++;
922 REGISTER_HANDLE_REQ(loop, handle, req);
923 uv__insert_pending_req(loop, (uv_req_t*) req);
924 } else if (UV_SUCCEEDED_WITH_IOCP(result == 0)) {
925 /* Request queued by the kernel. */
926 req->u.io.queued_bytes = uv__count_bufs(bufs, nbufs);
927 handle->reqs_pending++;
928 handle->stream.conn.write_reqs_pending++;
929 REGISTER_HANDLE_REQ(loop, handle, req);
930 handle->write_queue_size += req->u.io.queued_bytes;
931 if (handle->flags & UV_HANDLE_EMULATE_IOCP &&
932 !RegisterWaitForSingleObject(&req->wait_handle,
933 req->event_handle, post_write_completion, (void*) req,
934 INFINITE, WT_EXECUTEINWAITTHREAD | WT_EXECUTEONLYONCE)) {
935 SET_REQ_ERROR(req, GetLastError());
936 uv__insert_pending_req(loop, (uv_req_t*)req);
937 }
938 } else {
939 /* Send failed due to an error, report it later */
940 req->u.io.queued_bytes = 0;
941 handle->reqs_pending++;
942 handle->stream.conn.write_reqs_pending++;
943 REGISTER_HANDLE_REQ(loop, handle, req);
944 SET_REQ_ERROR(req, WSAGetLastError());
945 uv__insert_pending_req(loop, (uv_req_t*) req);
946 }
947
948 return 0;
949 }
950
951
uv__tcp_try_write(uv_tcp_t * handle,const uv_buf_t bufs[],unsigned int nbufs)952 int uv__tcp_try_write(uv_tcp_t* handle,
953 const uv_buf_t bufs[],
954 unsigned int nbufs) {
955 int result;
956 DWORD bytes;
957
958 if (handle->stream.conn.write_reqs_pending > 0)
959 return UV_EAGAIN;
960
961 result = WSASend(handle->socket,
962 (WSABUF*) bufs,
963 nbufs,
964 &bytes,
965 0,
966 NULL,
967 NULL);
968
969 if (result == SOCKET_ERROR)
970 return uv_translate_sys_error(WSAGetLastError());
971 else
972 return bytes;
973 }
974
975
uv__process_tcp_read_req(uv_loop_t * loop,uv_tcp_t * handle,uv_req_t * req)976 void uv__process_tcp_read_req(uv_loop_t* loop, uv_tcp_t* handle,
977 uv_req_t* req) {
978 DWORD bytes, flags, err;
979 uv_buf_t buf;
980 int count;
981
982 assert(handle->type == UV_TCP);
983
984 handle->flags &= ~UV_HANDLE_READ_PENDING;
985
986 if (!REQ_SUCCESS(req)) {
987 /* An error occurred doing the read. */
988 if ((handle->flags & UV_HANDLE_READING) ||
989 !(handle->flags & UV_HANDLE_ZERO_READ)) {
990 handle->flags &= ~UV_HANDLE_READING;
991 DECREASE_ACTIVE_COUNT(loop, handle);
992 buf = (handle->flags & UV_HANDLE_ZERO_READ) ?
993 uv_buf_init(NULL, 0) : handle->tcp.conn.read_buffer;
994
995 err = GET_REQ_SOCK_ERROR(req);
996
997 if (err == WSAECONNABORTED) {
998 /* Turn WSAECONNABORTED into UV_ECONNRESET to be consistent with Unix.
999 */
1000 err = WSAECONNRESET;
1001 }
1002 handle->flags &= ~(UV_HANDLE_READABLE | UV_HANDLE_WRITABLE);
1003
1004 handle->read_cb((uv_stream_t*)handle,
1005 uv_translate_sys_error(err),
1006 &buf);
1007 }
1008 } else {
1009 if (!(handle->flags & UV_HANDLE_ZERO_READ)) {
1010 /* The read was done with a non-zero buffer length. */
1011 if (req->u.io.overlapped.InternalHigh > 0) {
1012 /* Successful read */
1013 handle->read_cb((uv_stream_t*)handle,
1014 req->u.io.overlapped.InternalHigh,
1015 &handle->tcp.conn.read_buffer);
1016 /* Read again only if bytes == buf.len */
1017 if (req->u.io.overlapped.InternalHigh < handle->tcp.conn.read_buffer.len) {
1018 goto done;
1019 }
1020 } else {
1021 /* Connection closed */
1022 if (handle->flags & UV_HANDLE_READING) {
1023 handle->flags &= ~UV_HANDLE_READING;
1024 DECREASE_ACTIVE_COUNT(loop, handle);
1025 }
1026
1027 buf.base = 0;
1028 buf.len = 0;
1029 handle->read_cb((uv_stream_t*)handle, UV_EOF, &handle->tcp.conn.read_buffer);
1030 goto done;
1031 }
1032 }
1033
1034 /* Do nonblocking reads until the buffer is empty */
1035 count = 32;
1036 while ((handle->flags & UV_HANDLE_READING) && (count-- > 0)) {
1037 buf = uv_buf_init(NULL, 0);
1038 handle->alloc_cb((uv_handle_t*) handle, 65536, &buf);
1039 if (buf.base == NULL || buf.len == 0) {
1040 handle->read_cb((uv_stream_t*) handle, UV_ENOBUFS, &buf);
1041 break;
1042 }
1043 assert(buf.base != NULL);
1044
1045 flags = 0;
1046 if (WSARecv(handle->socket,
1047 (WSABUF*)&buf,
1048 1,
1049 &bytes,
1050 &flags,
1051 NULL,
1052 NULL) != SOCKET_ERROR) {
1053 if (bytes > 0) {
1054 /* Successful read */
1055 handle->read_cb((uv_stream_t*)handle, bytes, &buf);
1056 /* Read again only if bytes == buf.len */
1057 if (bytes < buf.len) {
1058 break;
1059 }
1060 } else {
1061 /* Connection closed */
1062 handle->flags &= ~UV_HANDLE_READING;
1063 DECREASE_ACTIVE_COUNT(loop, handle);
1064
1065 handle->read_cb((uv_stream_t*)handle, UV_EOF, &buf);
1066 break;
1067 }
1068 } else {
1069 err = WSAGetLastError();
1070 if (err == WSAEWOULDBLOCK) {
1071 /* Read buffer was completely empty, report a 0-byte read. */
1072 handle->read_cb((uv_stream_t*)handle, 0, &buf);
1073 } else {
1074 /* Ouch! serious error. */
1075 handle->flags &= ~UV_HANDLE_READING;
1076 DECREASE_ACTIVE_COUNT(loop, handle);
1077
1078 if (err == WSAECONNABORTED) {
1079 /* Turn WSAECONNABORTED into UV_ECONNRESET to be consistent with
1080 * Unix. */
1081 err = WSAECONNRESET;
1082 }
1083 handle->flags &= ~(UV_HANDLE_READABLE | UV_HANDLE_WRITABLE);
1084
1085 handle->read_cb((uv_stream_t*)handle,
1086 uv_translate_sys_error(err),
1087 &buf);
1088 }
1089 break;
1090 }
1091 }
1092
1093 done:
1094 /* Post another read if still reading and not closing. */
1095 if ((handle->flags & UV_HANDLE_READING) &&
1096 !(handle->flags & UV_HANDLE_READ_PENDING)) {
1097 uv__tcp_queue_read(loop, handle);
1098 }
1099 }
1100
1101 DECREASE_PENDING_REQ_COUNT(handle);
1102 }
1103
1104
uv__process_tcp_write_req(uv_loop_t * loop,uv_tcp_t * handle,uv_write_t * req)1105 void uv__process_tcp_write_req(uv_loop_t* loop, uv_tcp_t* handle,
1106 uv_write_t* req) {
1107 int err;
1108
1109 assert(handle->type == UV_TCP);
1110
1111 assert(handle->write_queue_size >= req->u.io.queued_bytes);
1112 handle->write_queue_size -= req->u.io.queued_bytes;
1113
1114 UNREGISTER_HANDLE_REQ(loop, handle, req);
1115
1116 if (handle->flags & UV_HANDLE_EMULATE_IOCP) {
1117 if (req->wait_handle != INVALID_HANDLE_VALUE) {
1118 UnregisterWait(req->wait_handle);
1119 req->wait_handle = INVALID_HANDLE_VALUE;
1120 }
1121 if (req->event_handle != NULL) {
1122 CloseHandle(req->event_handle);
1123 req->event_handle = NULL;
1124 }
1125 }
1126
1127 if (req->cb) {
1128 err = uv_translate_sys_error(GET_REQ_SOCK_ERROR(req));
1129 if (err == UV_ECONNABORTED) {
1130 /* use UV_ECANCELED for consistency with Unix */
1131 err = UV_ECANCELED;
1132 }
1133 req->cb(req, err);
1134 }
1135
1136 handle->stream.conn.write_reqs_pending--;
1137 if (handle->stream.conn.write_reqs_pending == 0) {
1138 if (handle->flags & UV_HANDLE_CLOSING) {
1139 closesocket(handle->socket);
1140 handle->socket = INVALID_SOCKET;
1141 }
1142 if (uv__is_stream_shutting(handle))
1143 uv__process_tcp_shutdown_req(loop,
1144 handle,
1145 handle->stream.conn.shutdown_req);
1146 }
1147
1148 DECREASE_PENDING_REQ_COUNT(handle);
1149 }
1150
1151
uv__process_tcp_accept_req(uv_loop_t * loop,uv_tcp_t * handle,uv_req_t * raw_req)1152 void uv__process_tcp_accept_req(uv_loop_t* loop, uv_tcp_t* handle,
1153 uv_req_t* raw_req) {
1154 uv_tcp_accept_t* req = (uv_tcp_accept_t*) raw_req;
1155 int err;
1156
1157 assert(handle->type == UV_TCP);
1158
1159 /* If handle->accepted_socket is not a valid socket, then uv_queue_accept
1160 * must have failed. This is a serious error. We stop accepting connections
1161 * and report this error to the connection callback. */
1162 if (req->accept_socket == INVALID_SOCKET) {
1163 if (handle->flags & UV_HANDLE_LISTENING) {
1164 handle->flags &= ~UV_HANDLE_LISTENING;
1165 DECREASE_ACTIVE_COUNT(loop, handle);
1166 if (handle->stream.serv.connection_cb) {
1167 err = GET_REQ_SOCK_ERROR(req);
1168 handle->stream.serv.connection_cb((uv_stream_t*)handle,
1169 uv_translate_sys_error(err));
1170 }
1171 }
1172 } else if (REQ_SUCCESS(req) &&
1173 setsockopt(req->accept_socket,
1174 SOL_SOCKET,
1175 SO_UPDATE_ACCEPT_CONTEXT,
1176 (char*)&handle->socket,
1177 sizeof(handle->socket)) == 0) {
1178 req->next_pending = handle->tcp.serv.pending_accepts;
1179 handle->tcp.serv.pending_accepts = req;
1180
1181 /* Accept and SO_UPDATE_ACCEPT_CONTEXT were successful. */
1182 if (handle->stream.serv.connection_cb) {
1183 handle->stream.serv.connection_cb((uv_stream_t*)handle, 0);
1184 }
1185 } else {
1186 /* Error related to accepted socket is ignored because the server socket
1187 * may still be healthy. If the server socket is broken uv_queue_accept
1188 * will detect it. */
1189 closesocket(req->accept_socket);
1190 req->accept_socket = INVALID_SOCKET;
1191 if (handle->flags & UV_HANDLE_LISTENING) {
1192 uv__tcp_queue_accept(handle, req);
1193 }
1194 }
1195
1196 DECREASE_PENDING_REQ_COUNT(handle);
1197 }
1198
1199
uv__process_tcp_connect_req(uv_loop_t * loop,uv_tcp_t * handle,uv_connect_t * req)1200 void uv__process_tcp_connect_req(uv_loop_t* loop, uv_tcp_t* handle,
1201 uv_connect_t* req) {
1202 int err;
1203
1204 assert(handle->type == UV_TCP);
1205
1206 UNREGISTER_HANDLE_REQ(loop, handle, req);
1207
1208 err = 0;
1209 if (handle->delayed_error) {
1210 /* To smooth over the differences between unixes errors that
1211 * were reported synchronously on the first connect can be delayed
1212 * until the next tick--which is now.
1213 */
1214 err = handle->delayed_error;
1215 handle->delayed_error = 0;
1216 } else if (REQ_SUCCESS(req)) {
1217 if (handle->flags & UV_HANDLE_CLOSING) {
1218 /* use UV_ECANCELED for consistency with Unix */
1219 err = ERROR_OPERATION_ABORTED;
1220 } else if (setsockopt(handle->socket,
1221 SOL_SOCKET,
1222 SO_UPDATE_CONNECT_CONTEXT,
1223 NULL,
1224 0) == 0) {
1225 uv__connection_init((uv_stream_t*)handle);
1226 handle->flags |= UV_HANDLE_READABLE | UV_HANDLE_WRITABLE;
1227 } else {
1228 err = WSAGetLastError();
1229 }
1230 } else {
1231 err = GET_REQ_SOCK_ERROR(req);
1232 }
1233 req->cb(req, uv_translate_sys_error(err));
1234
1235 DECREASE_PENDING_REQ_COUNT(handle);
1236 }
1237
1238
uv__tcp_xfer_export(uv_tcp_t * handle,int target_pid,uv__ipc_socket_xfer_type_t * xfer_type,uv__ipc_socket_xfer_info_t * xfer_info)1239 int uv__tcp_xfer_export(uv_tcp_t* handle,
1240 int target_pid,
1241 uv__ipc_socket_xfer_type_t* xfer_type,
1242 uv__ipc_socket_xfer_info_t* xfer_info) {
1243 if (handle->flags & UV_HANDLE_CONNECTION) {
1244 *xfer_type = UV__IPC_SOCKET_XFER_TCP_CONNECTION;
1245 } else {
1246 *xfer_type = UV__IPC_SOCKET_XFER_TCP_SERVER;
1247 /* We're about to share the socket with another process. Because this is a
1248 * listening socket, we assume that the other process will be accepting
1249 * connections on it. Thus, before sharing the socket with another process,
1250 * we call listen here in the parent process. */
1251 if (!(handle->flags & UV_HANDLE_LISTENING)) {
1252 if (!(handle->flags & UV_HANDLE_BOUND)) {
1253 return ERROR_NOT_SUPPORTED;
1254 }
1255 if (handle->delayed_error == 0 &&
1256 listen(handle->socket, SOMAXCONN) == SOCKET_ERROR) {
1257 handle->delayed_error = WSAGetLastError();
1258 }
1259 }
1260 }
1261
1262 if (WSADuplicateSocketW(handle->socket, target_pid, &xfer_info->socket_info))
1263 return WSAGetLastError();
1264 xfer_info->delayed_error = handle->delayed_error;
1265
1266 /* Mark the local copy of the handle as 'shared' so we behave in a way that's
1267 * friendly to the process(es) that we share the socket with. */
1268 handle->flags |= UV_HANDLE_SHARED_TCP_SOCKET;
1269
1270 return 0;
1271 }
1272
1273
uv__tcp_xfer_import(uv_tcp_t * tcp,uv__ipc_socket_xfer_type_t xfer_type,uv__ipc_socket_xfer_info_t * xfer_info)1274 int uv__tcp_xfer_import(uv_tcp_t* tcp,
1275 uv__ipc_socket_xfer_type_t xfer_type,
1276 uv__ipc_socket_xfer_info_t* xfer_info) {
1277 int err;
1278 SOCKET socket;
1279
1280 assert(xfer_type == UV__IPC_SOCKET_XFER_TCP_SERVER ||
1281 xfer_type == UV__IPC_SOCKET_XFER_TCP_CONNECTION);
1282
1283 socket = WSASocketW(FROM_PROTOCOL_INFO,
1284 FROM_PROTOCOL_INFO,
1285 FROM_PROTOCOL_INFO,
1286 &xfer_info->socket_info,
1287 0,
1288 WSA_FLAG_OVERLAPPED);
1289
1290 if (socket == INVALID_SOCKET) {
1291 return WSAGetLastError();
1292 }
1293
1294 err = uv__tcp_set_socket(
1295 tcp->loop, tcp, socket, xfer_info->socket_info.iAddressFamily, 1);
1296 if (err) {
1297 closesocket(socket);
1298 return err;
1299 }
1300
1301 tcp->delayed_error = xfer_info->delayed_error;
1302 tcp->flags |= UV_HANDLE_BOUND | UV_HANDLE_SHARED_TCP_SOCKET;
1303
1304 if (xfer_type == UV__IPC_SOCKET_XFER_TCP_CONNECTION) {
1305 uv__connection_init((uv_stream_t*)tcp);
1306 tcp->flags |= UV_HANDLE_READABLE | UV_HANDLE_WRITABLE;
1307 }
1308
1309 return 0;
1310 }
1311
1312
uv_tcp_nodelay(uv_tcp_t * handle,int enable)1313 int uv_tcp_nodelay(uv_tcp_t* handle, int enable) {
1314 int err;
1315
1316 if (handle->socket != INVALID_SOCKET) {
1317 err = uv__tcp_nodelay(handle, handle->socket, enable);
1318 if (err)
1319 return uv_translate_sys_error(err);
1320 }
1321
1322 if (enable) {
1323 handle->flags |= UV_HANDLE_TCP_NODELAY;
1324 } else {
1325 handle->flags &= ~UV_HANDLE_TCP_NODELAY;
1326 }
1327
1328 return 0;
1329 }
1330
1331
uv_tcp_keepalive(uv_tcp_t * handle,int enable,unsigned int delay)1332 int uv_tcp_keepalive(uv_tcp_t* handle, int enable, unsigned int delay) {
1333 int err;
1334
1335 if (handle->socket != INVALID_SOCKET) {
1336 err = uv__tcp_keepalive(handle, handle->socket, enable, delay);
1337 if (err)
1338 return uv_translate_sys_error(err);
1339 }
1340
1341 if (enable) {
1342 handle->flags |= UV_HANDLE_TCP_KEEPALIVE;
1343 } else {
1344 handle->flags &= ~UV_HANDLE_TCP_KEEPALIVE;
1345 }
1346
1347 /* TODO: Store delay if handle->socket isn't created yet. */
1348
1349 return 0;
1350 }
1351
1352
uv_tcp_simultaneous_accepts(uv_tcp_t * handle,int enable)1353 int uv_tcp_simultaneous_accepts(uv_tcp_t* handle, int enable) {
1354 if (handle->flags & UV_HANDLE_CONNECTION) {
1355 return UV_EINVAL;
1356 }
1357
1358 /* Check if we're already in the desired mode. */
1359 if ((enable && !(handle->flags & UV_HANDLE_TCP_SINGLE_ACCEPT)) ||
1360 (!enable && handle->flags & UV_HANDLE_TCP_SINGLE_ACCEPT)) {
1361 return 0;
1362 }
1363
1364 /* Don't allow switching from single pending accept to many. */
1365 if (enable) {
1366 return UV_ENOTSUP;
1367 }
1368
1369 /* Check if we're in a middle of changing the number of pending accepts. */
1370 if (handle->flags & UV_HANDLE_TCP_ACCEPT_STATE_CHANGING) {
1371 return 0;
1372 }
1373
1374 handle->flags |= UV_HANDLE_TCP_SINGLE_ACCEPT;
1375
1376 /* Flip the changing flag if we have already queued multiple accepts. */
1377 if (handle->flags & UV_HANDLE_LISTENING) {
1378 handle->flags |= UV_HANDLE_TCP_ACCEPT_STATE_CHANGING;
1379 }
1380
1381 return 0;
1382 }
1383
1384
uv__tcp_try_cancel_reqs(uv_tcp_t * tcp)1385 static void uv__tcp_try_cancel_reqs(uv_tcp_t* tcp) {
1386 SOCKET socket;
1387 int non_ifs_lsp;
1388 int reading;
1389 int writing;
1390
1391 socket = tcp->socket;
1392 reading = tcp->flags & UV_HANDLE_READ_PENDING;
1393 writing = tcp->stream.conn.write_reqs_pending > 0;
1394 if (!reading && !writing)
1395 return;
1396
1397 /* TODO: in libuv v2, keep explicit track of write_reqs, so we can cancel
1398 * them each explicitly with CancelIoEx (like unix). */
1399 if (reading)
1400 CancelIoEx((HANDLE) socket, &tcp->read_req.u.io.overlapped);
1401 if (writing)
1402 CancelIo((HANDLE) socket);
1403
1404 /* Check if we have any non-IFS LSPs stacked on top of TCP */
1405 non_ifs_lsp = (tcp->flags & UV_HANDLE_IPV6) ? uv_tcp_non_ifs_lsp_ipv6 :
1406 uv_tcp_non_ifs_lsp_ipv4;
1407
1408 /* If there are non-ifs LSPs then try to obtain a base handle for the socket.
1409 */
1410 if (non_ifs_lsp) {
1411 DWORD bytes;
1412 if (WSAIoctl(socket,
1413 SIO_BASE_HANDLE,
1414 NULL,
1415 0,
1416 &socket,
1417 sizeof socket,
1418 &bytes,
1419 NULL,
1420 NULL) != 0) {
1421 /* Failed. We can't do CancelIo. */
1422 return;
1423 }
1424 }
1425
1426 assert(socket != 0 && socket != INVALID_SOCKET);
1427
1428 if (socket != tcp->socket) {
1429 if (reading)
1430 CancelIoEx((HANDLE) socket, &tcp->read_req.u.io.overlapped);
1431 if (writing)
1432 CancelIo((HANDLE) socket);
1433 }
1434 }
1435
1436
uv__tcp_close(uv_loop_t * loop,uv_tcp_t * tcp)1437 void uv__tcp_close(uv_loop_t* loop, uv_tcp_t* tcp) {
1438 if (tcp->flags & UV_HANDLE_CONNECTION) {
1439 if (tcp->flags & UV_HANDLE_READING) {
1440 uv_read_stop((uv_stream_t*) tcp);
1441 }
1442 uv__tcp_try_cancel_reqs(tcp);
1443 } else {
1444 if (tcp->tcp.serv.accept_reqs != NULL) {
1445 /* First close the incoming sockets to cancel the accept operations before
1446 * we free their resources. */
1447 unsigned int i;
1448 for (i = 0; i < uv_simultaneous_server_accepts; i++) {
1449 uv_tcp_accept_t* req = &tcp->tcp.serv.accept_reqs[i];
1450 if (req->accept_socket != INVALID_SOCKET) {
1451 closesocket(req->accept_socket);
1452 req->accept_socket = INVALID_SOCKET;
1453 }
1454 }
1455 }
1456 assert(!(tcp->flags & UV_HANDLE_READING));
1457 }
1458
1459 if (tcp->flags & UV_HANDLE_LISTENING) {
1460 tcp->flags &= ~UV_HANDLE_LISTENING;
1461 DECREASE_ACTIVE_COUNT(loop, tcp);
1462 }
1463
1464 tcp->flags &= ~(UV_HANDLE_READABLE | UV_HANDLE_WRITABLE);
1465 uv__handle_closing(tcp);
1466
1467 /* If any overlapped req failed to cancel, calling `closesocket` now would
1468 * cause Win32 to send an RST packet. Try to avoid that for writes, if
1469 * possibly applicable, by waiting to process the completion notifications
1470 * first (which typically should be cancellations). There's not much we can
1471 * do about canceled reads, which also will generate an RST packet. */
1472 if (!(tcp->flags & UV_HANDLE_CONNECTION) ||
1473 tcp->stream.conn.write_reqs_pending == 0) {
1474 closesocket(tcp->socket);
1475 tcp->socket = INVALID_SOCKET;
1476 }
1477
1478 if (tcp->reqs_pending == 0)
1479 uv__want_endgame(loop, (uv_handle_t*) tcp);
1480 }
1481
1482
uv_tcp_open(uv_tcp_t * handle,uv_os_sock_t sock)1483 int uv_tcp_open(uv_tcp_t* handle, uv_os_sock_t sock) {
1484 WSAPROTOCOL_INFOW protocol_info;
1485 int opt_len;
1486 int err;
1487 struct sockaddr_storage saddr;
1488 int saddr_len;
1489
1490 /* Detect the address family of the socket. */
1491 opt_len = (int) sizeof protocol_info;
1492 if (getsockopt(sock,
1493 SOL_SOCKET,
1494 SO_PROTOCOL_INFOW,
1495 (char*) &protocol_info,
1496 &opt_len) == SOCKET_ERROR) {
1497 return uv_translate_sys_error(GetLastError());
1498 }
1499
1500 err = uv__tcp_set_socket(handle->loop,
1501 handle,
1502 sock,
1503 protocol_info.iAddressFamily,
1504 1);
1505 if (err) {
1506 return uv_translate_sys_error(err);
1507 }
1508
1509 /* Support already active socket. */
1510 saddr_len = sizeof(saddr);
1511 if (!uv_tcp_getsockname(handle, (struct sockaddr*) &saddr, &saddr_len)) {
1512 /* Socket is already bound. */
1513 handle->flags |= UV_HANDLE_BOUND;
1514 saddr_len = sizeof(saddr);
1515 if (!uv_tcp_getpeername(handle, (struct sockaddr*) &saddr, &saddr_len)) {
1516 /* Socket is already connected. */
1517 uv__connection_init((uv_stream_t*) handle);
1518 handle->flags |= UV_HANDLE_READABLE | UV_HANDLE_WRITABLE;
1519 }
1520 }
1521
1522 return 0;
1523 }
1524
1525
1526 /* This function is an egress point, i.e. it returns libuv errors rather than
1527 * system errors.
1528 */
uv__tcp_bind(uv_tcp_t * handle,const struct sockaddr * addr,unsigned int addrlen,unsigned int flags)1529 int uv__tcp_bind(uv_tcp_t* handle,
1530 const struct sockaddr* addr,
1531 unsigned int addrlen,
1532 unsigned int flags) {
1533 int err;
1534
1535 err = uv__tcp_try_bind(handle, addr, addrlen, flags);
1536 if (err)
1537 return uv_translate_sys_error(err);
1538
1539 return 0;
1540 }
1541
1542
1543 /* This function is an egress point, i.e. it returns libuv errors rather than
1544 * system errors.
1545 */
uv__tcp_connect(uv_connect_t * req,uv_tcp_t * handle,const struct sockaddr * addr,unsigned int addrlen,uv_connect_cb cb)1546 int uv__tcp_connect(uv_connect_t* req,
1547 uv_tcp_t* handle,
1548 const struct sockaddr* addr,
1549 unsigned int addrlen,
1550 uv_connect_cb cb) {
1551 int err;
1552
1553 err = uv__tcp_try_connect(req, handle, addr, addrlen, cb);
1554 if (err)
1555 return uv_translate_sys_error(err);
1556
1557 return 0;
1558 }
1559
1560
uv_socketpair(int type,int protocol,uv_os_sock_t fds[2],int flags0,int flags1)1561 int uv_socketpair(int type, int protocol, uv_os_sock_t fds[2], int flags0, int flags1) {
1562 SOCKET server = INVALID_SOCKET;
1563 SOCKET client0 = INVALID_SOCKET;
1564 SOCKET client1 = INVALID_SOCKET;
1565 SOCKADDR_IN name;
1566 LPFN_ACCEPTEX func_acceptex;
1567 WSAOVERLAPPED overlap;
1568 char accept_buffer[sizeof(struct sockaddr_storage) * 2 + 32];
1569 int namelen;
1570 int err;
1571 DWORD bytes;
1572 DWORD flags;
1573 DWORD client0_flags = WSA_FLAG_NO_HANDLE_INHERIT;
1574 DWORD client1_flags = WSA_FLAG_NO_HANDLE_INHERIT;
1575
1576 if (flags0 & UV_NONBLOCK_PIPE)
1577 client0_flags |= WSA_FLAG_OVERLAPPED;
1578 if (flags1 & UV_NONBLOCK_PIPE)
1579 client1_flags |= WSA_FLAG_OVERLAPPED;
1580
1581 server = WSASocketW(AF_INET, type, protocol, NULL, 0,
1582 WSA_FLAG_OVERLAPPED | WSA_FLAG_NO_HANDLE_INHERIT);
1583 if (server == INVALID_SOCKET)
1584 goto wsaerror;
1585 if (!SetHandleInformation((HANDLE) server, HANDLE_FLAG_INHERIT, 0))
1586 goto error;
1587 name.sin_family = AF_INET;
1588 name.sin_addr.s_addr = htonl(INADDR_LOOPBACK);
1589 name.sin_port = 0;
1590 if (bind(server, (SOCKADDR*) &name, sizeof(name)) != 0)
1591 goto wsaerror;
1592 if (listen(server, 1) != 0)
1593 goto wsaerror;
1594 namelen = sizeof(name);
1595 if (getsockname(server, (SOCKADDR*) &name, &namelen) != 0)
1596 goto wsaerror;
1597 client0 = WSASocketW(AF_INET, type, protocol, NULL, 0, client0_flags);
1598 if (client0 == INVALID_SOCKET)
1599 goto wsaerror;
1600 if (!SetHandleInformation((HANDLE) client0, HANDLE_FLAG_INHERIT, 0))
1601 goto error;
1602 if (connect(client0, (SOCKADDR*) &name, sizeof(name)) != 0)
1603 goto wsaerror;
1604 client1 = WSASocketW(AF_INET, type, protocol, NULL, 0, client1_flags);
1605 if (client1 == INVALID_SOCKET)
1606 goto wsaerror;
1607 if (!SetHandleInformation((HANDLE) client1, HANDLE_FLAG_INHERIT, 0))
1608 goto error;
1609 if (!uv__get_acceptex_function(server, &func_acceptex)) {
1610 err = WSAEAFNOSUPPORT;
1611 goto cleanup;
1612 }
1613 memset(&overlap, 0, sizeof(overlap));
1614 if (!func_acceptex(server,
1615 client1,
1616 accept_buffer,
1617 0,
1618 sizeof(struct sockaddr_storage),
1619 sizeof(struct sockaddr_storage),
1620 &bytes,
1621 &overlap)) {
1622 err = WSAGetLastError();
1623 if (err == ERROR_IO_PENDING) {
1624 /* Result should complete immediately, since we already called connect,
1625 * but empirically, we sometimes have to poll the kernel a couple times
1626 * until it notices that. */
1627 while (!WSAGetOverlappedResult(client1, &overlap, &bytes, FALSE, &flags)) {
1628 err = WSAGetLastError();
1629 if (err != WSA_IO_INCOMPLETE)
1630 goto cleanup;
1631 SwitchToThread();
1632 }
1633 }
1634 else {
1635 goto cleanup;
1636 }
1637 }
1638 if (setsockopt(client1, SOL_SOCKET, SO_UPDATE_ACCEPT_CONTEXT,
1639 (char*) &server, sizeof(server)) != 0) {
1640 goto wsaerror;
1641 }
1642
1643 closesocket(server);
1644
1645 fds[0] = client0;
1646 fds[1] = client1;
1647
1648 return 0;
1649
1650 wsaerror:
1651 err = WSAGetLastError();
1652 goto cleanup;
1653
1654 error:
1655 err = GetLastError();
1656 goto cleanup;
1657
1658 cleanup:
1659 if (server != INVALID_SOCKET)
1660 closesocket(server);
1661 if (client0 != INVALID_SOCKET)
1662 closesocket(client0);
1663 if (client1 != INVALID_SOCKET)
1664 closesocket(client1);
1665
1666 assert(err);
1667 return uv_translate_sys_error(err);
1668 }
1669
