1 /*
2  * Copyright (c) 2008-2012 Niels Provos and Nick Mathewson
3  *
4  * Redistribution and use in source and binary forms, with or without
5  * modification, are permitted provided that the following conditions
6  * are met:
7  * 1. Redistributions of source code must retain the above copyright
8  *    notice, this list of conditions and the following disclaimer.
9  * 2. Redistributions in binary form must reproduce the above copyright
10  *    notice, this list of conditions and the following disclaimer in the
11  *    documentation and/or other materials provided with the distribution.
12  * 3. The name of the author may not be used to endorse or promote products
13  *    derived from this software without specific prior written permission.
14  *
15  * THIS SOFTWARE IS PROVIDED BY THE AUTHOR ``AS IS'' AND ANY EXPRESS OR
16  * IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES
17  * OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED.
18  * IN NO EVENT SHALL THE AUTHOR BE LIABLE FOR ANY DIRECT, INDIRECT,
19  * INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT
20  * NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
21  * DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
22  * THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
23  * (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF
24  * THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
25  */
26 #ifndef BUFFEREVENT_INTERNAL_H_INCLUDED_
27 #define BUFFEREVENT_INTERNAL_H_INCLUDED_
28 
29 #ifdef __cplusplus
30 extern "C" {
31 #endif
32 
33 #include "event2/event-config.h"
34 #include "event2/event_struct.h"
35 #include "evconfig-private.h"
36 #include "event2/util.h"
37 #include "defer-internal.h"
38 #include "evthread-internal.h"
39 #include "event2/thread.h"
40 #include "ratelim-internal.h"
41 #include "event2/bufferevent_struct.h"
42 
43 #include "ipv6-internal.h"
44 #ifdef _WIN32
45 #include <ws2tcpip.h>
46 #endif
47 #ifdef EVENT__HAVE_NETINET_IN_H
48 #include <netinet/in.h>
49 #endif
50 #ifdef EVENT__HAVE_NETINET_IN6_H
51 #include <netinet/in6.h>
52 #endif
53 
54 /* These flags are reasons that we might be declining to actually enable
55    reading or writing on a bufferevent.
56  */
57 
58 /* On a all bufferevents, for reading: used when we have read up to the
59    watermark value.
60 
61    On a filtering bufferevent, for writing: used when the underlying
62    bufferevent's write buffer has been filled up to its watermark
63    value.
64 */
65 #define BEV_SUSPEND_WM 0x01
66 /* On a base bufferevent: when we have emptied a bandwidth buckets */
67 #define BEV_SUSPEND_BW 0x02
68 /* On a base bufferevent: when we have emptied the group's bandwidth bucket. */
69 #define BEV_SUSPEND_BW_GROUP 0x04
70 /* On a socket bufferevent: can't do any operations while we're waiting for
71  * name lookup to finish. */
72 #define BEV_SUSPEND_LOOKUP 0x08
73 /* On a base bufferevent, for reading: used when a filter has choked this
74  * (underlying) bufferevent because it has stopped reading from it. */
75 #define BEV_SUSPEND_FILT_READ 0x10
76 
77 typedef ev_uint16_t bufferevent_suspend_flags;
78 
79 struct bufferevent_rate_limit_group {
80 	/** List of all members in the group */
81 	LIST_HEAD(rlim_group_member_list, bufferevent_private) members;
82 	/** Current limits for the group. */
83 	struct ev_token_bucket rate_limit;
84 	struct ev_token_bucket_cfg rate_limit_cfg;
85 
86 	/** True iff we don't want to read from any member of the group.until
87 	 * the token bucket refills.  */
88 	unsigned read_suspended : 1;
89 	/** True iff we don't want to write from any member of the group.until
90 	 * the token bucket refills.  */
91 	unsigned write_suspended : 1;
92 	/** True iff we were unable to suspend one of the bufferevents in the
93 	 * group for reading the last time we tried, and we should try
94 	 * again. */
95 	unsigned pending_unsuspend_read : 1;
96 	/** True iff we were unable to suspend one of the bufferevents in the
97 	 * group for writing the last time we tried, and we should try
98 	 * again. */
99 	unsigned pending_unsuspend_write : 1;
100 
101 	/*@{*/
102 	/** Total number of bytes read or written in this group since last
103 	 * reset. */
104 	ev_uint64_t total_read;
105 	ev_uint64_t total_written;
106 	/*@}*/
107 
108 	/** The number of bufferevents in the group. */
109 	int n_members;
110 
111 	/** The smallest number of bytes that any member of the group should
112 	 * be limited to read or write at a time. */
113 	ev_ssize_t min_share;
114 	ev_ssize_t configured_min_share;
115 
116 	/** Timeout event that goes off once a tick, when the bucket is ready
117 	 * to refill. */
118 	struct event master_refill_event;
119 
120 	/** Seed for weak random number generator. Protected by 'lock' */
121 	struct evutil_weakrand_state weakrand_seed;
122 
123 	/** Lock to protect the members of this group.  This lock should nest
124 	 * within every bufferevent lock: if you are holding this lock, do
125 	 * not assume you can lock another bufferevent. */
126 	void *lock;
127 };
128 
129 /** Fields for rate-limiting a single bufferevent. */
130 struct bufferevent_rate_limit {
131 	/* Linked-list elements for storing this bufferevent_private in a
132 	 * group.
133 	 *
134 	 * Note that this field is supposed to be protected by the group
135 	 * lock */
136 	LIST_ENTRY(bufferevent_private) next_in_group;
137 	/** The rate-limiting group for this bufferevent, or NULL if it is
138 	 * only rate-limited on its own. */
139 	struct bufferevent_rate_limit_group *group;
140 
141 	/* This bufferevent's current limits. */
142 	struct ev_token_bucket limit;
143 	/* Pointer to the rate-limit configuration for this bufferevent.
144 	 * Can be shared.  XXX reference-count this? */
145 	struct ev_token_bucket_cfg *cfg;
146 
147 	/* Timeout event used when one this bufferevent's buckets are
148 	 * empty. */
149 	struct event refill_bucket_event;
150 };
151 
152 /** Parts of the bufferevent structure that are shared among all bufferevent
153  * types, but not exposed in bufferevent_struct.h. */
154 struct bufferevent_private {
155 	/** The underlying bufferevent structure. */
156 	struct bufferevent bev;
157 
158 	/** Evbuffer callback to enforce watermarks on input. */
159 	struct evbuffer_cb_entry *read_watermarks_cb;
160 
161 	/** If set, we should free the lock when we free the bufferevent. */
162 	unsigned own_lock : 1;
163 
164 	/** Flag: set if we have deferred callbacks and a read callback is
165 	 * pending. */
166 	unsigned readcb_pending : 1;
167 	/** Flag: set if we have deferred callbacks and a write callback is
168 	 * pending. */
169 	unsigned writecb_pending : 1;
170 	/** Flag: set if we are currently busy connecting. */
171 	unsigned connecting : 1;
172 	/** Flag: set if a connect failed prematurely; this is a hack for
173 	 * getting around the bufferevent abstraction. */
174 	unsigned connection_refused : 1;
175 	/** Set to the events pending if we have deferred callbacks and
176 	 * an events callback is pending. */
177 	short eventcb_pending;
178 
179 	/** If set, read is suspended until one or more conditions are over.
180 	 * The actual value here is a bitfield of those conditions; see the
181 	 * BEV_SUSPEND_* flags above. */
182 	bufferevent_suspend_flags read_suspended;
183 
184 	/** If set, writing is suspended until one or more conditions are over.
185 	 * The actual value here is a bitfield of those conditions; see the
186 	 * BEV_SUSPEND_* flags above. */
187 	bufferevent_suspend_flags write_suspended;
188 
189 	/** Set to the current socket errno if we have deferred callbacks and
190 	 * an events callback is pending. */
191 	int errno_pending;
192 
193 	/** The DNS error code for bufferevent_socket_connect_hostname */
194 	int dns_error;
195 
196 	/** Used to implement deferred callbacks */
197 	struct event_callback deferred;
198 
199 	/** The options this bufferevent was constructed with */
200 	enum bufferevent_options options;
201 
202 	/** Current reference count for this bufferevent. */
203 	int refcnt;
204 
205 	/** Lock for this bufferevent.  Shared by the inbuf and the outbuf.
206 	 * If NULL, locking is disabled. */
207 	void *lock;
208 
209 	/** No matter how big our bucket gets, don't try to read more than this
210 	 * much in a single read operation. */
211 	ev_ssize_t max_single_read;
212 
213 	/** No matter how big our bucket gets, don't try to write more than this
214 	 * much in a single write operation. */
215 	ev_ssize_t max_single_write;
216 
217 	/** Rate-limiting information for this bufferevent */
218 	struct bufferevent_rate_limit *rate_limiting;
219 
220 	/* Saved conn_addr, to extract IP address from it.
221 	 *
222 	 * Because some servers may reset/close connection without waiting clients,
223 	 * in that case we can't extract IP address even in close_cb.
224 	 * So we need to save it, just after we connected to remote server, or
225 	 * after resolving (to avoid extra dns requests during retrying, since UDP
226 	 * is slow) */
227 	union {
228 		struct sockaddr_in6 in6;
229 		struct sockaddr_in in;
230 	} conn_address;
231 
232 	struct evdns_getaddrinfo_request *dns_request;
233 };
234 
235 /** Possible operations for a control callback. */
236 enum bufferevent_ctrl_op {
237 	BEV_CTRL_SET_FD,
238 	BEV_CTRL_GET_FD,
239 	BEV_CTRL_GET_UNDERLYING,
240 	BEV_CTRL_CANCEL_ALL
241 };
242 
243 /** Possible data types for a control callback */
244 union bufferevent_ctrl_data {
245 	void *ptr;
246 	evutil_socket_t fd;
247 };
248 
249 /**
250    Implementation table for a bufferevent: holds function pointers and other
251    information to make the various bufferevent types work.
252 */
253 struct bufferevent_ops {
254 	/** The name of the bufferevent's type. */
255 	const char *type;
256 	/** At what offset into the implementation type will we find a
257 	    bufferevent structure?
258 
259 	    Example: if the type is implemented as
260 	    struct bufferevent_x {
261 	       int extra_data;
262 	       struct bufferevent bev;
263 	    }
264 	    then mem_offset should be offsetof(struct bufferevent_x, bev)
265 	*/
266 	off_t mem_offset;
267 
268 	/** Enables one or more of EV_READ|EV_WRITE on a bufferevent.  Does
269 	    not need to adjust the 'enabled' field.  Returns 0 on success, -1
270 	    on failure.
271 	 */
272 	int (*enable)(struct bufferevent *, short);
273 
274 	/** Disables one or more of EV_READ|EV_WRITE on a bufferevent.  Does
275 	    not need to adjust the 'enabled' field.  Returns 0 on success, -1
276 	    on failure.
277 	 */
278 	int (*disable)(struct bufferevent *, short);
279 
280 	/** Detatches the bufferevent from related data structures. Called as
281 	 * soon as its reference count reaches 0. */
282 	void (*unlink)(struct bufferevent *);
283 
284 	/** Free any storage and deallocate any extra data or structures used
285 	    in this implementation. Called when the bufferevent is
286 	    finalized.
287 	 */
288 	void (*destruct)(struct bufferevent *);
289 
290 	/** Called when the timeouts on the bufferevent have changed.*/
291 	int (*adj_timeouts)(struct bufferevent *);
292 
293 	/** Called to flush data. */
294 	int (*flush)(struct bufferevent *, short, enum bufferevent_flush_mode);
295 
296 	/** Called to access miscellaneous fields. */
297 	int (*ctrl)(struct bufferevent *, enum bufferevent_ctrl_op, union bufferevent_ctrl_data *);
298 
299 };
300 
301 extern const struct bufferevent_ops bufferevent_ops_socket;
302 extern const struct bufferevent_ops bufferevent_ops_filter;
303 extern const struct bufferevent_ops bufferevent_ops_pair;
304 
305 #define BEV_IS_SOCKET(bevp) ((bevp)->be_ops == &bufferevent_ops_socket)
306 #define BEV_IS_FILTER(bevp) ((bevp)->be_ops == &bufferevent_ops_filter)
307 #define BEV_IS_PAIR(bevp) ((bevp)->be_ops == &bufferevent_ops_pair)
308 
309 #ifdef _WIN32
310 extern const struct bufferevent_ops bufferevent_ops_async;
311 #define BEV_IS_ASYNC(bevp) ((bevp)->be_ops == &bufferevent_ops_async)
312 #else
313 #define BEV_IS_ASYNC(bevp) 0
314 #endif
315 
316 /** Initialize the shared parts of a bufferevent. */
317 int bufferevent_init_common_(struct bufferevent_private *, struct event_base *, const struct bufferevent_ops *, enum bufferevent_options options);
318 
319 /** For internal use: temporarily stop all reads on bufev, until the conditions
320  * in 'what' are over. */
321 void bufferevent_suspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what);
322 /** For internal use: clear the conditions 'what' on bufev, and re-enable
323  * reading if there are no conditions left. */
324 void bufferevent_unsuspend_read_(struct bufferevent *bufev, bufferevent_suspend_flags what);
325 
326 /** For internal use: temporarily stop all writes on bufev, until the conditions
327  * in 'what' are over. */
328 void bufferevent_suspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what);
329 /** For internal use: clear the conditions 'what' on bufev, and re-enable
330  * writing if there are no conditions left. */
331 void bufferevent_unsuspend_write_(struct bufferevent *bufev, bufferevent_suspend_flags what);
332 
333 #define bufferevent_wm_suspend_read(b) \
334 	bufferevent_suspend_read_((b), BEV_SUSPEND_WM)
335 #define bufferevent_wm_unsuspend_read(b) \
336 	bufferevent_unsuspend_read_((b), BEV_SUSPEND_WM)
337 
338 /*
339   Disable a bufferevent.  Equivalent to bufferevent_disable(), but
340   first resets 'connecting' flag to force EV_WRITE down for sure.
341 
342   XXXX this method will go away in the future; try not to add new users.
343     See comment in evhttp_connection_reset_() for discussion.
344 
345   @param bufev the bufferevent to be disabled
346   @param event any combination of EV_READ | EV_WRITE.
347   @return 0 if successful, or -1 if an error occurred
348   @see bufferevent_disable()
349  */
350 int bufferevent_disable_hard_(struct bufferevent *bufev, short event);
351 
352 /** Internal: Set up locking on a bufferevent.  If lock is set, use it.
353  * Otherwise, use a new lock. */
354 int bufferevent_enable_locking_(struct bufferevent *bufev, void *lock);
355 /** Internal: backwards compat macro for the now public function
356  * Increment the reference count on bufev. */
357 #define bufferevent_incref_(bufev) bufferevent_incref(bufev)
358 /** Internal: Lock bufev and increase its reference count.
359  * unlocking it otherwise. */
360 void bufferevent_incref_and_lock_(struct bufferevent *bufev);
361 /** Internal: backwards compat macro for the now public function
362  * Decrement the reference count on bufev.  Returns 1 if it freed
363  * the bufferevent.*/
364 #define bufferevent_decref_(bufev) bufferevent_decref(bufev)
365 
366 /** Internal: Drop the reference count on bufev, freeing as necessary, and
367  * unlocking it otherwise.  Returns 1 if it freed the bufferevent. */
368 int bufferevent_decref_and_unlock_(struct bufferevent *bufev);
369 
370 /** Internal: If callbacks are deferred and we have a read callback, schedule
371  * a readcb.  Otherwise just run the readcb. Ignores watermarks. */
372 void bufferevent_run_readcb_(struct bufferevent *bufev, int options);
373 /** Internal: If callbacks are deferred and we have a write callback, schedule
374  * a writecb.  Otherwise just run the writecb. Ignores watermarks. */
375 void bufferevent_run_writecb_(struct bufferevent *bufev, int options);
376 /** Internal: If callbacks are deferred and we have an eventcb, schedule
377  * it to run with events "what".  Otherwise just run the eventcb.
378  * See bufferevent_trigger_event for meaning of "options". */
379 void bufferevent_run_eventcb_(struct bufferevent *bufev, short what, int options);
380 
381 /** Internal: Run or schedule (if deferred or options contain
382  * BEV_TRIG_DEFER_CALLBACKS) I/O callbacks specified in iotype.
383  * Must already hold the bufev lock. Honors watermarks unless
384  * BEV_TRIG_IGNORE_WATERMARKS is in options. */
385 static inline void bufferevent_trigger_nolock_(struct bufferevent *bufev, short iotype, int options);
386 
387 /* Making this inline since all of the common-case calls to this function in
388  * libevent use constant arguments. */
389 static inline void
bufferevent_trigger_nolock_(struct bufferevent * bufev,short iotype,int options)390 bufferevent_trigger_nolock_(struct bufferevent *bufev, short iotype, int options)
391 {
392 	if ((iotype & EV_READ) && ((options & BEV_TRIG_IGNORE_WATERMARKS) ||
393 	    evbuffer_get_length(bufev->input) >= bufev->wm_read.low))
394 		bufferevent_run_readcb_(bufev, options);
395 	if ((iotype & EV_WRITE) && ((options & BEV_TRIG_IGNORE_WATERMARKS) ||
396 	    evbuffer_get_length(bufev->output) <= bufev->wm_write.low))
397 		bufferevent_run_writecb_(bufev, options);
398 }
399 
400 /** Internal: Add the event 'ev' with timeout tv, unless tv is set to 0, in
401  * which case add ev with no timeout. */
402 int bufferevent_add_event_(struct event *ev, const struct timeval *tv);
403 
404 /* =========
405  * These next functions implement timeouts for bufferevents that aren't doing
406  * anything else with ev_read and ev_write, to handle timeouts.
407  * ========= */
408 /** Internal use: Set up the ev_read and ev_write callbacks so that
409  * the other "generic_timeout" functions will work on it.  Call this from
410  * the constructor function. */
411 void bufferevent_init_generic_timeout_cbs_(struct bufferevent *bev);
412 /** Internal use: Add or delete the generic timeout events as appropriate.
413  * (If an event is enabled and a timeout is set, we add the event.  Otherwise
414  * we delete it.)  Call this from anything that changes the timeout values,
415  * that enabled EV_READ or EV_WRITE, or that disables EV_READ or EV_WRITE. */
416 int bufferevent_generic_adj_timeouts_(struct bufferevent *bev);
417 int bufferevent_generic_adj_existing_timeouts_(struct bufferevent *bev);
418 
419 enum bufferevent_options bufferevent_get_options_(struct bufferevent *bev);
420 
421 const struct sockaddr*
422 bufferevent_socket_get_conn_address_(struct bufferevent *bev);
423 
424 /** Internal use: We have just successfully read data into an inbuf, so
425  * reset the read timeout (if any). */
426 #define BEV_RESET_GENERIC_READ_TIMEOUT(bev)				\
427 	do {								\
428 		if (evutil_timerisset(&(bev)->timeout_read))		\
429 			event_add(&(bev)->ev_read, &(bev)->timeout_read); \
430 	} while (0)
431 /** Internal use: We have just successfully written data from an inbuf, so
432  * reset the read timeout (if any). */
433 #define BEV_RESET_GENERIC_WRITE_TIMEOUT(bev)				\
434 	do {								\
435 		if (evutil_timerisset(&(bev)->timeout_write))		\
436 			event_add(&(bev)->ev_write, &(bev)->timeout_write); \
437 	} while (0)
438 #define BEV_DEL_GENERIC_READ_TIMEOUT(bev)	\
439 		event_del(&(bev)->ev_read)
440 #define BEV_DEL_GENERIC_WRITE_TIMEOUT(bev)	\
441 		event_del(&(bev)->ev_write)
442 
443 
444 /** Internal: Given a bufferevent, return its corresponding
445  * bufferevent_private. */
446 #define BEV_UPCAST(b) EVUTIL_UPCAST((b), struct bufferevent_private, bev)
447 
448 #ifdef EVENT__DISABLE_THREAD_SUPPORT
449 #define BEV_LOCK(b) EVUTIL_NIL_STMT_
450 #define BEV_UNLOCK(b) EVUTIL_NIL_STMT_
451 #else
452 /** Internal: Grab the lock (if any) on a bufferevent */
453 #define BEV_LOCK(b) do {						\
454 		struct bufferevent_private *locking =  BEV_UPCAST(b);	\
455 		EVLOCK_LOCK(locking->lock, 0);				\
456 	} while (0)
457 
458 /** Internal: Release the lock (if any) on a bufferevent */
459 #define BEV_UNLOCK(b) do {						\
460 		struct bufferevent_private *locking =  BEV_UPCAST(b);	\
461 		EVLOCK_UNLOCK(locking->lock, 0);			\
462 	} while (0)
463 #endif
464 
465 
466 /* ==== For rate-limiting. */
467 
468 int bufferevent_decrement_write_buckets_(struct bufferevent_private *bev,
469     ev_ssize_t bytes);
470 int bufferevent_decrement_read_buckets_(struct bufferevent_private *bev,
471     ev_ssize_t bytes);
472 ev_ssize_t bufferevent_get_read_max_(struct bufferevent_private *bev);
473 ev_ssize_t bufferevent_get_write_max_(struct bufferevent_private *bev);
474 
475 int bufferevent_ratelim_init_(struct bufferevent_private *bev);
476 
477 #ifdef __cplusplus
478 }
479 #endif
480 
481 
482 #endif /* BUFFEREVENT_INTERNAL_H_INCLUDED_ */
483