xref: /openbsd-src/usr.sbin/unbound/util/winsock_event.h (revision 77079be7e74cd516ce5a60021c2a03ba383e594d)
1933707f3Ssthen /*
2933707f3Ssthen  * util/winsock_event.h - unbound event handling for winsock on windows
3933707f3Ssthen  *
4933707f3Ssthen  * Copyright (c) 2008, NLnet Labs. All rights reserved.
5933707f3Ssthen  *
6933707f3Ssthen  * This software is open source.
7933707f3Ssthen  *
8933707f3Ssthen  * Redistribution and use in source and binary forms, with or without
9933707f3Ssthen  * modification, are permitted provided that the following conditions
10933707f3Ssthen  * are met:
11933707f3Ssthen  *
12933707f3Ssthen  * Redistributions of source code must retain the above copyright notice,
13933707f3Ssthen  * this list of conditions and the following disclaimer.
14933707f3Ssthen  *
15933707f3Ssthen  * Redistributions in binary form must reproduce the above copyright notice,
16933707f3Ssthen  * this list of conditions and the following disclaimer in the documentation
17933707f3Ssthen  * and/or other materials provided with the distribution.
18933707f3Ssthen  *
19933707f3Ssthen  * Neither the name of the NLNET LABS nor the names of its contributors may
20933707f3Ssthen  * be used to endorse or promote products derived from this software without
21933707f3Ssthen  * specific prior written permission.
22933707f3Ssthen  *
23933707f3Ssthen  * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
245d76a658Ssthen  * "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
255d76a658Ssthen  * LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
265d76a658Ssthen  * A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
275d76a658Ssthen  * HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
285d76a658Ssthen  * SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED
295d76a658Ssthen  * TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
305d76a658Ssthen  * PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
315d76a658Ssthen  * LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
325d76a658Ssthen  * NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
335d76a658Ssthen  * SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
34933707f3Ssthen  */
35933707f3Ssthen 
36933707f3Ssthen /**
37933707f3Ssthen  * \file
38933707f3Ssthen  *
39933707f3Ssthen  * This file contains interface functions with the WinSock2 API on Windows.
40933707f3Ssthen  * It uses the winsock WSAWaitForMultipleEvents interface on a number of
41933707f3Ssthen  * sockets.
42933707f3Ssthen  *
43933707f3Ssthen  * Note that windows can only wait for max 64 events at one time.
44933707f3Ssthen  *
45933707f3Ssthen  * Also, file descriptors cannot be waited for.
46933707f3Ssthen  *
47933707f3Ssthen  * Named pipes are not easily available (and are not usable in select() ).
48933707f3Ssthen  * For interprocess communication, it is possible to wait for a hEvent to
49933707f3Ssthen  * be signaled by another thread.
50933707f3Ssthen  *
51933707f3Ssthen  * When a socket becomes readable, then it will not be flagged as
52933707f3Ssthen  * readable again until you have gotten WOULDBLOCK from a recv routine.
53933707f3Ssthen  * That means the event handler must store the readability (edge notify)
54933707f3Ssthen  * and process the incoming data until it blocks.
55933707f3Ssthen  * The function performing recv then has to inform the event handler that
56933707f3Ssthen  * the socket has blocked, and the event handler can mark it as such.
57933707f3Ssthen  * Thus, this file transforms the edge notify from windows to a level notify
58933707f3Ssthen  * that is compatible with UNIX.
59933707f3Ssthen  * The WSAEventSelect page says that it does do level notify, as long
60933707f3Ssthen  * as you call a recv/write/accept at least once when it is signalled.
61933707f3Ssthen  * This last bit is not true, even though documented in server2008 api docs
62933707f3Ssthen  * from microsoft, it does not happen at all. Instead you have to test for
63933707f3Ssthen  * WSAEWOULDBLOCK on a tcp stream, and only then retest the socket.
64933707f3Ssthen  * And before that remember the previous result as still valid.
65933707f3Ssthen  *
66933707f3Ssthen  * To stay 'fair', instead of emptying a socket completely, the event handler
67933707f3Ssthen  * can test the other (marked as blocking) sockets for new events.
68933707f3Ssthen  *
69933707f3Ssthen  * Additionally, TCP accept sockets get special event support.
70933707f3Ssthen  *
71933707f3Ssthen  * Socket numbers are not starting small, they can be any number (say 33060).
72933707f3Ssthen  * Therefore, bitmaps are not used, but arrays.
73933707f3Ssthen  *
74933707f3Ssthen  * on winsock, you must use recv() and send() for TCP reads and writes,
75933707f3Ssthen  * not read() and write(), those work only on files.
76933707f3Ssthen  *
77933707f3Ssthen  * Also fseek and fseeko do not work if a FILE is not fopen-ed in binary mode.
78933707f3Ssthen  *
79933707f3Ssthen  * When under a high load windows gives out lots of errors, from recvfrom
80933707f3Ssthen  * on udp sockets for example (WSAECONNRESET). Even though the udp socket
81933707f3Ssthen  * has no connection per se.
82933707f3Ssthen  */
83933707f3Ssthen 
84933707f3Ssthen #ifndef UTIL_WINSOCK_EVENT_H
85933707f3Ssthen #define UTIL_WINSOCK_EVENT_H
86933707f3Ssthen 
87933707f3Ssthen #ifdef USE_WINSOCK
88933707f3Ssthen 
89933707f3Ssthen #ifndef HAVE_EVENT_BASE_FREE
90933707f3Ssthen #define HAVE_EVENT_BASE_FREE
91933707f3Ssthen #endif
92933707f3Ssthen 
9398f3ca02Sbrad /* redefine the calls to different names so that there is no name
9498f3ca02Sbrad  * collision with other code that uses libevent names. (that uses libunbound)*/
9598f3ca02Sbrad #define event_init winsockevent_init
9698f3ca02Sbrad #define event_get_version winsockevent_get_version
9798f3ca02Sbrad #define event_get_method winsockevent_get_method
9898f3ca02Sbrad #define event_base_dispatch winsockevent_base_dispatch
9998f3ca02Sbrad #define event_base_loopexit winsockevent_base_loopexit
10098f3ca02Sbrad #define event_base_free winsockevent_base_free
10198f3ca02Sbrad #define event_set winsockevent_set
10298f3ca02Sbrad #define event_base_set winsockevent_base_set
10398f3ca02Sbrad #define event_add winsockevent_add
10498f3ca02Sbrad #define event_del winsockevent_del
10598f3ca02Sbrad #define signal_add winsocksignal_add
10698f3ca02Sbrad #define signal_del winsocksignal_del
10798f3ca02Sbrad 
108933707f3Ssthen /** event timeout */
109933707f3Ssthen #define EV_TIMEOUT      0x01
110933707f3Ssthen /** event fd readable */
111933707f3Ssthen #define EV_READ         0x02
112933707f3Ssthen /** event fd writable */
113933707f3Ssthen #define EV_WRITE        0x04
114933707f3Ssthen /** event signal */
115933707f3Ssthen #define EV_SIGNAL       0x08
116933707f3Ssthen /** event must persist */
117933707f3Ssthen #define EV_PERSIST      0x10
118933707f3Ssthen 
119933707f3Ssthen /* needs our redblack tree */
120933707f3Ssthen #include "rbtree.h"
121933707f3Ssthen 
122933707f3Ssthen /** max number of signals to support */
123933707f3Ssthen #define MAX_SIG 32
124933707f3Ssthen 
125933707f3Ssthen /** The number of items that the winsock event handler can service.
126933707f3Ssthen  * Windows cannot handle more anyway */
127933707f3Ssthen #define WSK_MAX_ITEMS 64
128933707f3Ssthen 
129933707f3Ssthen /**
130933707f3Ssthen  * event base for winsock event handler
131933707f3Ssthen  */
132933707f3Ssthen struct event_base
133933707f3Ssthen {
134933707f3Ssthen 	/** sorted by timeout (absolute), ptr */
135*77079be7Ssthen 	rbtree_type* times;
136933707f3Ssthen 	/** array (first part in use) of handles to work on */
137933707f3Ssthen 	struct event** items;
138933707f3Ssthen 	/** number of items in use in array */
139933707f3Ssthen 	int max;
140933707f3Ssthen 	/** capacity of array, size of array in items */
141933707f3Ssthen 	int cap;
142933707f3Ssthen 	/** array of 0 - maxsig of ptr to event for it */
143933707f3Ssthen         struct event** signals;
144933707f3Ssthen 	/** if we need to exit */
145933707f3Ssthen 	int need_to_exit;
146933707f3Ssthen 	/** where to store time in seconds */
147229e174cSsthen 	time_t* time_secs;
148933707f3Ssthen 	/** where to store time in microseconds */
149933707f3Ssthen 	struct timeval* time_tv;
150933707f3Ssthen 	/**
151933707f3Ssthen 	 * TCP streams have sticky events to them, these are not
152933707f3Ssthen 	 * reported by the windows event system anymore, we have to
153933707f3Ssthen 	 * keep reporting those events as present until wouldblock() is
154933707f3Ssthen 	 * signalled by the handler back to use.
155933707f3Ssthen 	 */
156933707f3Ssthen 	int tcp_stickies;
157933707f3Ssthen 	/**
158933707f3Ssthen 	 * should next cycle process reinvigorated stickies,
159933707f3Ssthen 	 * these are stickies that have been stored, but due to a new
160933707f3Ssthen 	 * event_add a sudden interest in the event has incepted.
161933707f3Ssthen 	 */
162933707f3Ssthen 	int tcp_reinvigorated;
163933707f3Ssthen 	/** The list of events that is currently being processed. */
164933707f3Ssthen 	WSAEVENT waitfor[WSK_MAX_ITEMS];
165933707f3Ssthen };
166933707f3Ssthen 
167933707f3Ssthen /**
168933707f3Ssthen  * Event structure. Has some of the event elements.
169933707f3Ssthen  */
170933707f3Ssthen struct event {
171933707f3Ssthen         /** node in timeout rbtree */
172*77079be7Ssthen         rbnode_type node;
173933707f3Ssthen         /** is event already added */
174933707f3Ssthen         int added;
175933707f3Ssthen 
176933707f3Ssthen         /** event base it belongs to */
177933707f3Ssthen         struct event_base *ev_base;
178933707f3Ssthen         /** fd to poll or -1 for timeouts. signal number for sigs. */
179933707f3Ssthen         int ev_fd;
180933707f3Ssthen         /** what events this event is interested in, see EV_.. above. */
181933707f3Ssthen         short ev_events;
182933707f3Ssthen         /** timeout value */
183933707f3Ssthen         struct timeval ev_timeout;
184933707f3Ssthen 
185933707f3Ssthen         /** callback to call: fd, eventbits, userarg */
186933707f3Ssthen         void (*ev_callback)(int, short, void *);
187933707f3Ssthen         /** callback user arg */
188933707f3Ssthen         void *ev_arg;
189933707f3Ssthen 
190933707f3Ssthen 	/* ----- nonpublic part, for winsock_event only ----- */
191933707f3Ssthen 	/** index of this event in the items array (if added) */
192933707f3Ssthen 	int idx;
193933707f3Ssthen 	/** the event handle to wait for new events to become ready */
194933707f3Ssthen 	WSAEVENT hEvent;
195933707f3Ssthen 	/** true if this filedes is a TCP socket and needs special attention */
196933707f3Ssthen 	int is_tcp;
197933707f3Ssthen 	/** remembered EV_ values */
198933707f3Ssthen 	short old_events;
199933707f3Ssthen 	/** should remembered EV_ values be used for TCP streams.
200933707f3Ssthen 	 * Reset after WOULDBLOCK is signaled using the function. */
201933707f3Ssthen 	int stick_events;
202933707f3Ssthen 
203933707f3Ssthen 	/** true if this event is a signaling WSAEvent by the user.
2044bfc71b0Ssthen 	 * User created and user closed WSAEvent. Only signaled/unsignaled,
205933707f3Ssthen 	 * no read/write/distinctions needed. */
206933707f3Ssthen 	int is_signal;
207933707f3Ssthen 	/** used during callbacks to see which events were just checked */
208933707f3Ssthen 	int just_checked;
209933707f3Ssthen };
210933707f3Ssthen 
211933707f3Ssthen /** create event base */
212229e174cSsthen void *event_init(time_t* time_secs, struct timeval* time_tv);
213933707f3Ssthen /** get version */
214933707f3Ssthen const char *event_get_version(void);
215933707f3Ssthen /** get polling method (select,epoll) */
216933707f3Ssthen const char *event_get_method(void);
217933707f3Ssthen /** run select in a loop */
218933707f3Ssthen int event_base_dispatch(struct event_base *);
219933707f3Ssthen /** exit that loop */
220933707f3Ssthen int event_base_loopexit(struct event_base *, struct timeval *);
221933707f3Ssthen /** free event base. Free events yourself */
222933707f3Ssthen void event_base_free(struct event_base *);
223933707f3Ssthen /** set content of event */
224933707f3Ssthen void event_set(struct event *, int, short, void (*)(int, short, void *), void *);
225933707f3Ssthen 
226933707f3Ssthen /** add event to a base. You *must* call this for every event. */
227933707f3Ssthen int event_base_set(struct event_base *, struct event *);
228933707f3Ssthen /** add event to make it active. You may not change it with event_set anymore */
229933707f3Ssthen int event_add(struct event *, struct timeval *);
230933707f3Ssthen /** remove event. You may change it again */
231933707f3Ssthen int event_del(struct event *);
232933707f3Ssthen 
233933707f3Ssthen #define evtimer_add(ev, tv)             event_add(ev, tv)
234933707f3Ssthen #define evtimer_del(ev)                 event_del(ev)
235933707f3Ssthen 
236933707f3Ssthen /* uses different implementation. Cannot mix fd/timeouts and signals inside
237933707f3Ssthen  * the same struct event. create several event structs for that.  */
238933707f3Ssthen /** install signal handler */
239933707f3Ssthen int signal_add(struct event *, struct timeval *);
240933707f3Ssthen /** set signal event contents */
241933707f3Ssthen #define signal_set(ev, x, cb, arg)      \
242933707f3Ssthen         event_set(ev, x, EV_SIGNAL|EV_PERSIST, cb, arg)
243933707f3Ssthen /** remove signal handler */
244933707f3Ssthen int signal_del(struct event *);
245933707f3Ssthen 
246933707f3Ssthen /** compare events in tree, based on timevalue, ptr for uniqueness */
247933707f3Ssthen int mini_ev_cmp(const void* a, const void* b);
248933707f3Ssthen 
249933707f3Ssthen /**
250933707f3Ssthen  * Routine for windows only, where the handling layer can signal that
251933707f3Ssthen  * a TCP stream encountered WSAEWOULDBLOCK for a stream and thus needs
252933707f3Ssthen  * retesting the event.
253933707f3Ssthen  * Pass if EV_READ or EV_WRITE gave wouldblock.
254933707f3Ssthen  */
255933707f3Ssthen void winsock_tcp_wouldblock(struct event* ev, int eventbit);
256933707f3Ssthen 
257933707f3Ssthen /**
258933707f3Ssthen  * Routine for windows only. where you pass a signal WSAEvent that
259933707f3Ssthen  * you wait for. When the event is signaled, the callback gets called.
260933707f3Ssthen  * The callback has to WSAResetEvent to disable the signal.
261933707f3Ssthen  * @param base: the event base.
262933707f3Ssthen  * @param ev: the event structure for data storage
263933707f3Ssthen  * 	can be passed uninitialised.
264933707f3Ssthen  * @param wsaevent: the WSAEvent that gets signaled.
265933707f3Ssthen  * @param cb: callback routine.
266933707f3Ssthen  * @param arg: user argument to callback routine.
267933707f3Ssthen  * @return false on error.
268933707f3Ssthen  */
269933707f3Ssthen int winsock_register_wsaevent(struct event_base* base, struct event* ev,
270933707f3Ssthen 	WSAEVENT wsaevent, void (*cb)(int, short, void*), void* arg);
271933707f3Ssthen 
272933707f3Ssthen /**
273933707f3Ssthen  * Unregister a wsaevent. User has to close the WSAEVENT itself.
274933707f3Ssthen  * @param ev: event data storage.
275933707f3Ssthen  */
276933707f3Ssthen void winsock_unregister_wsaevent(struct event* ev);
277933707f3Ssthen 
278933707f3Ssthen #endif /* USE_WINSOCK */
279933707f3Ssthen #endif /* UTIL_WINSOCK_EVENT_H */
280