evdns.c revision 290001
1/* Copyright 2006-2007 Niels Provos
2 * Copyright 2007-2012 Nick Mathewson and Niels Provos
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
27/* Based on software by Adam Langly. Adam's original message:
28 *
29 * Async DNS Library
30 * Adam Langley <agl@imperialviolet.org>
31 * http://www.imperialviolet.org/eventdns.html
32 * Public Domain code
33 *
34 * This software is Public Domain. To view a copy of the public domain dedication,
35 * visit http://creativecommons.org/licenses/publicdomain/ or send a letter to
36 * Creative Commons, 559 Nathan Abbott Way, Stanford, California 94305, USA.
37 *
38 * I ask and expect, but do not require, that all derivative works contain an
39 * attribution similar to:
40 *	Parts developed by Adam Langley <agl@imperialviolet.org>
41 *
42 * You may wish to replace the word "Parts" with something else depending on
43 * the amount of original code.
44 *
45 * (Derivative works does not include programs which link against, run or include
46 * the source verbatim in their source distributions)
47 *
48 * Version: 0.1b
49 */
50
51#include "event2/event-config.h"
52#include "evconfig-private.h"
53
54#include <sys/types.h>
55
56#ifndef _FORTIFY_SOURCE
57#define _FORTIFY_SOURCE 3
58#endif
59
60#include <string.h>
61#include <fcntl.h>
62#ifdef EVENT__HAVE_SYS_TIME_H
63#include <sys/time.h>
64#endif
65#ifdef EVENT__HAVE_STDINT_H
66#include <stdint.h>
67#endif
68#include <stdlib.h>
69#include <string.h>
70#include <errno.h>
71#ifdef EVENT__HAVE_UNISTD_H
72#include <unistd.h>
73#endif
74#include <limits.h>
75#include <sys/stat.h>
76#include <stdio.h>
77#include <stdarg.h>
78#ifdef _WIN32
79#include <winsock2.h>
80#include <ws2tcpip.h>
81#ifndef _WIN32_IE
82#define _WIN32_IE 0x400
83#endif
84#include <shlobj.h>
85#endif
86
87#include "event2/dns.h"
88#include "event2/dns_struct.h"
89#include "event2/dns_compat.h"
90#include "event2/util.h"
91#include "event2/event.h"
92#include "event2/event_struct.h"
93#include "event2/thread.h"
94
95#include "defer-internal.h"
96#include "log-internal.h"
97#include "mm-internal.h"
98#include "strlcpy-internal.h"
99#include "ipv6-internal.h"
100#include "util-internal.h"
101#include "evthread-internal.h"
102#ifdef _WIN32
103#include <ctype.h>
104#include <winsock2.h>
105#include <windows.h>
106#include <iphlpapi.h>
107#include <io.h>
108#else
109#include <sys/socket.h>
110#include <netinet/in.h>
111#include <arpa/inet.h>
112#endif
113
114#ifdef EVENT__HAVE_NETINET_IN6_H
115#include <netinet/in6.h>
116#endif
117
118#define EVDNS_LOG_DEBUG EVENT_LOG_DEBUG
119#define EVDNS_LOG_WARN EVENT_LOG_WARN
120#define EVDNS_LOG_MSG EVENT_LOG_MSG
121
122#ifndef HOST_NAME_MAX
123#define HOST_NAME_MAX 255
124#endif
125
126#include <stdio.h>
127
128#undef MIN
129#define MIN(a,b) ((a)<(b)?(a):(b))
130
131#define ASSERT_VALID_REQUEST(req) \
132	EVUTIL_ASSERT((req)->handle && (req)->handle->current_req == (req))
133
134#define u64 ev_uint64_t
135#define u32 ev_uint32_t
136#define u16 ev_uint16_t
137#define u8  ev_uint8_t
138
139/* maximum number of addresses from a single packet */
140/* that we bother recording */
141#define MAX_V4_ADDRS 32
142#define MAX_V6_ADDRS 32
143
144
145#define TYPE_A	       EVDNS_TYPE_A
146#define TYPE_CNAME     5
147#define TYPE_PTR       EVDNS_TYPE_PTR
148#define TYPE_SOA       EVDNS_TYPE_SOA
149#define TYPE_AAAA      EVDNS_TYPE_AAAA
150
151#define CLASS_INET     EVDNS_CLASS_INET
152
153/* Persistent handle.  We keep this separate from 'struct request' since we
154 * need some object to last for as long as an evdns_request is outstanding so
155 * that it can be canceled, whereas a search request can lead to multiple
156 * 'struct request' instances being created over its lifetime. */
157struct evdns_request {
158	struct request *current_req;
159	struct evdns_base *base;
160
161	int pending_cb; /* Waiting for its callback to be invoked; not
162			 * owned by event base any more. */
163
164	/* elements used by the searching code */
165	int search_index;
166	struct search_state *search_state;
167	char *search_origname;	/* needs to be free()ed */
168	int search_flags;
169};
170
171struct request {
172	u8 *request;  /* the dns packet data */
173	u8 request_type; /* TYPE_PTR or TYPE_A or TYPE_AAAA */
174	unsigned int request_len;
175	int reissue_count;
176	int tx_count;  /* the number of times that this packet has been sent */
177	void *user_pointer;  /* the pointer given to us for this request */
178	evdns_callback_type user_callback;
179	struct nameserver *ns;	/* the server which we last sent it */
180
181	/* these objects are kept in a circular list */
182	/* XXX We could turn this into a CIRCLEQ. */
183	struct request *next, *prev;
184
185	struct event timeout_event;
186
187	u16 trans_id;  /* the transaction id */
188	unsigned request_appended :1;	/* true if the request pointer is data which follows this struct */
189	unsigned transmit_me :1;  /* needs to be transmitted */
190
191	/* XXXX This is a horrible hack. */
192	char **put_cname_in_ptr; /* store the cname here if we get one. */
193
194	struct evdns_base *base;
195
196	struct evdns_request *handle;
197};
198
199struct reply {
200	unsigned int type;
201	unsigned int have_answer : 1;
202	union {
203		struct {
204			u32 addrcount;
205			u32 addresses[MAX_V4_ADDRS];
206		} a;
207		struct {
208			u32 addrcount;
209			struct in6_addr addresses[MAX_V6_ADDRS];
210		} aaaa;
211		struct {
212			char name[HOST_NAME_MAX];
213		} ptr;
214	} data;
215};
216
217struct nameserver {
218	evutil_socket_t socket;	 /* a connected UDP socket */
219	struct sockaddr_storage address;
220	ev_socklen_t addrlen;
221	int failed_times;  /* number of times which we have given this server a chance */
222	int timedout;  /* number of times in a row a request has timed out */
223	struct event event;
224	/* these objects are kept in a circular list */
225	struct nameserver *next, *prev;
226	struct event timeout_event;  /* used to keep the timeout for */
227				     /* when we next probe this server. */
228				     /* Valid if state == 0 */
229	/* Outstanding probe request for this nameserver, if any */
230	struct evdns_request *probe_request;
231	char state;  /* zero if we think that this server is down */
232	char choked;  /* true if we have an EAGAIN from this server's socket */
233	char write_waiting;  /* true if we are waiting for EV_WRITE events */
234	struct evdns_base *base;
235
236	/* Number of currently inflight requests: used
237	 * to track when we should add/del the event. */
238	int requests_inflight;
239};
240
241
242/* Represents a local port where we're listening for DNS requests. Right now, */
243/* only UDP is supported. */
244struct evdns_server_port {
245	evutil_socket_t socket; /* socket we use to read queries and write replies. */
246	int refcnt; /* reference count. */
247	char choked; /* Are we currently blocked from writing? */
248	char closing; /* Are we trying to close this port, pending writes? */
249	evdns_request_callback_fn_type user_callback; /* Fn to handle requests */
250	void *user_data; /* Opaque pointer passed to user_callback */
251	struct event event; /* Read/write event */
252	/* circular list of replies that we want to write. */
253	struct server_request *pending_replies;
254	struct event_base *event_base;
255
256#ifndef EVENT__DISABLE_THREAD_SUPPORT
257	void *lock;
258#endif
259};
260
261/* Represents part of a reply being built.	(That is, a single RR.) */
262struct server_reply_item {
263	struct server_reply_item *next; /* next item in sequence. */
264	char *name; /* name part of the RR */
265	u16 type; /* The RR type */
266	u16 class; /* The RR class (usually CLASS_INET) */
267	u32 ttl; /* The RR TTL */
268	char is_name; /* True iff data is a label */
269	u16 datalen; /* Length of data; -1 if data is a label */
270	void *data; /* The contents of the RR */
271};
272
273/* Represents a request that we've received as a DNS server, and holds */
274/* the components of the reply as we're constructing it. */
275struct server_request {
276	/* Pointers to the next and previous entries on the list of replies */
277	/* that we're waiting to write.	 Only set if we have tried to respond */
278	/* and gotten EAGAIN. */
279	struct server_request *next_pending;
280	struct server_request *prev_pending;
281
282	u16 trans_id; /* Transaction id. */
283	struct evdns_server_port *port; /* Which port received this request on? */
284	struct sockaddr_storage addr; /* Where to send the response */
285	ev_socklen_t addrlen; /* length of addr */
286
287	int n_answer; /* how many answer RRs have been set? */
288	int n_authority; /* how many authority RRs have been set? */
289	int n_additional; /* how many additional RRs have been set? */
290
291	struct server_reply_item *answer; /* linked list of answer RRs */
292	struct server_reply_item *authority; /* linked list of authority RRs */
293	struct server_reply_item *additional; /* linked list of additional RRs */
294
295	/* Constructed response.  Only set once we're ready to send a reply. */
296	/* Once this is set, the RR fields are cleared, and no more should be set. */
297	char *response;
298	size_t response_len;
299
300	/* Caller-visible fields: flags, questions. */
301	struct evdns_server_request base;
302};
303
304struct evdns_base {
305	/* An array of n_req_heads circular lists for inflight requests.
306	 * Each inflight request req is in req_heads[req->trans_id % n_req_heads].
307	 */
308	struct request **req_heads;
309	/* A circular list of requests that we're waiting to send, but haven't
310	 * sent yet because there are too many requests inflight */
311	struct request *req_waiting_head;
312	/* A circular list of nameservers. */
313	struct nameserver *server_head;
314	int n_req_heads;
315
316	struct event_base *event_base;
317
318	/* The number of good nameservers that we have */
319	int global_good_nameservers;
320
321	/* inflight requests are contained in the req_head list */
322	/* and are actually going out across the network */
323	int global_requests_inflight;
324	/* requests which aren't inflight are in the waiting list */
325	/* and are counted here */
326	int global_requests_waiting;
327
328	int global_max_requests_inflight;
329
330	struct timeval global_timeout;	/* 5 seconds by default */
331	int global_max_reissues;  /* a reissue occurs when we get some errors from the server */
332	int global_max_retransmits;  /* number of times we'll retransmit a request which timed out */
333	/* number of timeouts in a row before we consider this server to be down */
334	int global_max_nameserver_timeout;
335	/* true iff we will use the 0x20 hack to prevent poisoning attacks. */
336	int global_randomize_case;
337
338	/* The first time that a nameserver fails, how long do we wait before
339	 * probing to see if it has returned?  */
340	struct timeval global_nameserver_probe_initial_timeout;
341
342	/** Port to bind to for outgoing DNS packets. */
343	struct sockaddr_storage global_outgoing_address;
344	/** ev_socklen_t for global_outgoing_address. 0 if it isn't set. */
345	ev_socklen_t global_outgoing_addrlen;
346
347	struct timeval global_getaddrinfo_allow_skew;
348
349	int getaddrinfo_ipv4_timeouts;
350	int getaddrinfo_ipv6_timeouts;
351	int getaddrinfo_ipv4_answered;
352	int getaddrinfo_ipv6_answered;
353
354	struct search_state *global_search_state;
355
356	TAILQ_HEAD(hosts_list, hosts_entry) hostsdb;
357
358#ifndef EVENT__DISABLE_THREAD_SUPPORT
359	void *lock;
360#endif
361
362	int disable_when_inactive;
363};
364
365struct hosts_entry {
366	TAILQ_ENTRY(hosts_entry) next;
367	union {
368		struct sockaddr sa;
369		struct sockaddr_in sin;
370		struct sockaddr_in6 sin6;
371	} addr;
372	int addrlen;
373	char hostname[1];
374};
375
376static struct evdns_base *current_base = NULL;
377
378struct evdns_base *
379evdns_get_global_base(void)
380{
381	return current_base;
382}
383
384/* Given a pointer to an evdns_server_request, get the corresponding */
385/* server_request. */
386#define TO_SERVER_REQUEST(base_ptr)					\
387	((struct server_request*)					\
388	  (((char*)(base_ptr) - evutil_offsetof(struct server_request, base))))
389
390#define REQ_HEAD(base, id) ((base)->req_heads[id % (base)->n_req_heads])
391
392static struct nameserver *nameserver_pick(struct evdns_base *base);
393static void evdns_request_insert(struct request *req, struct request **head);
394static void evdns_request_remove(struct request *req, struct request **head);
395static void nameserver_ready_callback(evutil_socket_t fd, short events, void *arg);
396static int evdns_transmit(struct evdns_base *base);
397static int evdns_request_transmit(struct request *req);
398static void nameserver_send_probe(struct nameserver *const ns);
399static void search_request_finished(struct evdns_request *const);
400static int search_try_next(struct evdns_request *const req);
401static struct request *search_request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *const name, int flags, evdns_callback_type user_callback, void *user_arg);
402static void evdns_requests_pump_waiting_queue(struct evdns_base *base);
403static u16 transaction_id_pick(struct evdns_base *base);
404static struct request *request_new(struct evdns_base *base, struct evdns_request *handle, int type, const char *name, int flags, evdns_callback_type callback, void *ptr);
405static void request_submit(struct request *const req);
406
407static int server_request_free(struct server_request *req);
408static void server_request_free_answers(struct server_request *req);
409static void server_port_free(struct evdns_server_port *port);
410static void server_port_ready_callback(evutil_socket_t fd, short events, void *arg);
411static int evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename);
412static int evdns_base_set_option_impl(struct evdns_base *base,
413    const char *option, const char *val, int flags);
414static void evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests);
415static void evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg);
416
417static int strtoint(const char *const str);
418
419#ifdef EVENT__DISABLE_THREAD_SUPPORT
420#define EVDNS_LOCK(base)  EVUTIL_NIL_STMT_
421#define EVDNS_UNLOCK(base) EVUTIL_NIL_STMT_
422#define ASSERT_LOCKED(base) EVUTIL_NIL_STMT_
423#else
424#define EVDNS_LOCK(base)			\
425	EVLOCK_LOCK((base)->lock, 0)
426#define EVDNS_UNLOCK(base)			\
427	EVLOCK_UNLOCK((base)->lock, 0)
428#define ASSERT_LOCKED(base)			\
429	EVLOCK_ASSERT_LOCKED((base)->lock)
430#endif
431
432static evdns_debug_log_fn_type evdns_log_fn = NULL;
433
434void
435evdns_set_log_fn(evdns_debug_log_fn_type fn)
436{
437	evdns_log_fn = fn;
438}
439
440#ifdef __GNUC__
441#define EVDNS_LOG_CHECK	 __attribute__ ((format(printf, 2, 3)))
442#else
443#define EVDNS_LOG_CHECK
444#endif
445
446static void evdns_log_(int severity, const char *fmt, ...) EVDNS_LOG_CHECK;
447static void
448evdns_log_(int severity, const char *fmt, ...)
449{
450	va_list args;
451	va_start(args,fmt);
452	if (evdns_log_fn) {
453		char buf[512];
454		int is_warn = (severity == EVDNS_LOG_WARN);
455		evutil_vsnprintf(buf, sizeof(buf), fmt, args);
456		evdns_log_fn(is_warn, buf);
457	} else {
458		event_logv_(severity, NULL, fmt, args);
459	}
460	va_end(args);
461}
462
463#define log evdns_log_
464
465/* This walks the list of inflight requests to find the */
466/* one with a matching transaction id. Returns NULL on */
467/* failure */
468static struct request *
469request_find_from_trans_id(struct evdns_base *base, u16 trans_id) {
470	struct request *req = REQ_HEAD(base, trans_id);
471	struct request *const started_at = req;
472
473	ASSERT_LOCKED(base);
474
475	if (req) {
476		do {
477			if (req->trans_id == trans_id) return req;
478			req = req->next;
479		} while (req != started_at);
480	}
481
482	return NULL;
483}
484
485/* a libevent callback function which is called when a nameserver */
486/* has gone down and we want to test if it has came back to life yet */
487static void
488nameserver_prod_callback(evutil_socket_t fd, short events, void *arg) {
489	struct nameserver *const ns = (struct nameserver *) arg;
490	(void)fd;
491	(void)events;
492
493	EVDNS_LOCK(ns->base);
494	nameserver_send_probe(ns);
495	EVDNS_UNLOCK(ns->base);
496}
497
498/* a libevent callback which is called when a nameserver probe (to see if */
499/* it has come back to life) times out. We increment the count of failed_times */
500/* and wait longer to send the next probe packet. */
501static void
502nameserver_probe_failed(struct nameserver *const ns) {
503	struct timeval timeout;
504	int i;
505
506	ASSERT_LOCKED(ns->base);
507	(void) evtimer_del(&ns->timeout_event);
508	if (ns->state == 1) {
509		/* This can happen if the nameserver acts in a way which makes us mark */
510		/* it as bad and then starts sending good replies. */
511		return;
512	}
513
514#define MAX_PROBE_TIMEOUT 3600
515#define TIMEOUT_BACKOFF_FACTOR 3
516
517	memcpy(&timeout, &ns->base->global_nameserver_probe_initial_timeout,
518	    sizeof(struct timeval));
519	for (i=ns->failed_times; i > 0 && timeout.tv_sec < MAX_PROBE_TIMEOUT; --i) {
520		timeout.tv_sec *= TIMEOUT_BACKOFF_FACTOR;
521		timeout.tv_usec *= TIMEOUT_BACKOFF_FACTOR;
522		if (timeout.tv_usec > 1000000) {
523			timeout.tv_sec += timeout.tv_usec / 1000000;
524			timeout.tv_usec %= 1000000;
525		}
526	}
527	if (timeout.tv_sec > MAX_PROBE_TIMEOUT) {
528		timeout.tv_sec = MAX_PROBE_TIMEOUT;
529		timeout.tv_usec = 0;
530	}
531
532	ns->failed_times++;
533
534	if (evtimer_add(&ns->timeout_event, &timeout) < 0) {
535		char addrbuf[128];
536		log(EVDNS_LOG_WARN,
537		    "Error from libevent when adding timer event for %s",
538		    evutil_format_sockaddr_port_(
539			    (struct sockaddr *)&ns->address,
540			    addrbuf, sizeof(addrbuf)));
541	}
542}
543
544static void
545request_swap_ns(struct request *req, struct nameserver *ns) {
546	if (ns && req->ns != ns) {
547		EVUTIL_ASSERT(req->ns->requests_inflight > 0);
548		req->ns->requests_inflight--;
549		ns->requests_inflight++;
550
551		req->ns = ns;
552	}
553}
554
555/* called when a nameserver has been deemed to have failed. For example, too */
556/* many packets have timed out etc */
557static void
558nameserver_failed(struct nameserver *const ns, const char *msg) {
559	struct request *req, *started_at;
560	struct evdns_base *base = ns->base;
561	int i;
562	char addrbuf[128];
563
564	ASSERT_LOCKED(base);
565	/* if this nameserver has already been marked as failed */
566	/* then don't do anything */
567	if (!ns->state) return;
568
569	log(EVDNS_LOG_MSG, "Nameserver %s has failed: %s",
570	    evutil_format_sockaddr_port_(
571		    (struct sockaddr *)&ns->address,
572		    addrbuf, sizeof(addrbuf)),
573	    msg);
574
575	base->global_good_nameservers--;
576	EVUTIL_ASSERT(base->global_good_nameservers >= 0);
577	if (base->global_good_nameservers == 0) {
578		log(EVDNS_LOG_MSG, "All nameservers have failed");
579	}
580
581	ns->state = 0;
582	ns->failed_times = 1;
583
584	if (evtimer_add(&ns->timeout_event,
585		&base->global_nameserver_probe_initial_timeout) < 0) {
586		log(EVDNS_LOG_WARN,
587		    "Error from libevent when adding timer event for %s",
588		    evutil_format_sockaddr_port_(
589			    (struct sockaddr *)&ns->address,
590			    addrbuf, sizeof(addrbuf)));
591		/* ???? Do more? */
592	}
593
594	/* walk the list of inflight requests to see if any can be reassigned to */
595	/* a different server. Requests in the waiting queue don't have a */
596	/* nameserver assigned yet */
597
598	/* if we don't have *any* good nameservers then there's no point */
599	/* trying to reassign requests to one */
600	if (!base->global_good_nameservers) return;
601
602	for (i = 0; i < base->n_req_heads; ++i) {
603		req = started_at = base->req_heads[i];
604		if (req) {
605			do {
606				if (req->tx_count == 0 && req->ns == ns) {
607					/* still waiting to go out, can be moved */
608					/* to another server */
609					request_swap_ns(req, nameserver_pick(base));
610				}
611				req = req->next;
612			} while (req != started_at);
613		}
614	}
615}
616
617static void
618nameserver_up(struct nameserver *const ns)
619{
620	char addrbuf[128];
621	ASSERT_LOCKED(ns->base);
622	if (ns->state) return;
623	log(EVDNS_LOG_MSG, "Nameserver %s is back up",
624	    evutil_format_sockaddr_port_(
625		    (struct sockaddr *)&ns->address,
626		    addrbuf, sizeof(addrbuf)));
627	evtimer_del(&ns->timeout_event);
628	if (ns->probe_request) {
629		evdns_cancel_request(ns->base, ns->probe_request);
630		ns->probe_request = NULL;
631	}
632	ns->state = 1;
633	ns->failed_times = 0;
634	ns->timedout = 0;
635	ns->base->global_good_nameservers++;
636}
637
638static void
639request_trans_id_set(struct request *const req, const u16 trans_id) {
640	req->trans_id = trans_id;
641	*((u16 *) req->request) = htons(trans_id);
642}
643
644/* Called to remove a request from a list and dealloc it. */
645/* head is a pointer to the head of the list it should be */
646/* removed from or NULL if the request isn't in a list. */
647/* when free_handle is one, free the handle as well. */
648static void
649request_finished(struct request *const req, struct request **head, int free_handle) {
650	struct evdns_base *base = req->base;
651	int was_inflight = (head != &base->req_waiting_head);
652	EVDNS_LOCK(base);
653	ASSERT_VALID_REQUEST(req);
654
655	if (head)
656		evdns_request_remove(req, head);
657
658	log(EVDNS_LOG_DEBUG, "Removing timeout for request %p", req);
659	if (was_inflight) {
660		evtimer_del(&req->timeout_event);
661		base->global_requests_inflight--;
662		req->ns->requests_inflight--;
663	} else {
664		base->global_requests_waiting--;
665	}
666	/* it was initialized during request_new / evtimer_assign */
667	event_debug_unassign(&req->timeout_event);
668
669	if (req->ns &&
670	    req->ns->requests_inflight == 0 &&
671	    req->base->disable_when_inactive) {
672		event_del(&req->ns->event);
673		evtimer_del(&req->ns->timeout_event);
674	}
675
676	if (!req->request_appended) {
677		/* need to free the request data on it's own */
678		mm_free(req->request);
679	} else {
680		/* the request data is appended onto the header */
681		/* so everything gets free()ed when we: */
682	}
683
684	if (req->handle) {
685		EVUTIL_ASSERT(req->handle->current_req == req);
686
687		if (free_handle) {
688			search_request_finished(req->handle);
689			req->handle->current_req = NULL;
690			if (! req->handle->pending_cb) {
691				/* If we're planning to run the callback,
692				 * don't free the handle until later. */
693				mm_free(req->handle);
694			}
695			req->handle = NULL; /* If we have a bug, let's crash
696					     * early */
697		} else {
698			req->handle->current_req = NULL;
699		}
700	}
701
702	mm_free(req);
703
704	evdns_requests_pump_waiting_queue(base);
705	EVDNS_UNLOCK(base);
706}
707
708/* This is called when a server returns a funny error code. */
709/* We try the request again with another server. */
710/* */
711/* return: */
712/*   0 ok */
713/*   1 failed/reissue is pointless */
714static int
715request_reissue(struct request *req) {
716	const struct nameserver *const last_ns = req->ns;
717	ASSERT_LOCKED(req->base);
718	ASSERT_VALID_REQUEST(req);
719	/* the last nameserver should have been marked as failing */
720	/* by the caller of this function, therefore pick will try */
721	/* not to return it */
722	request_swap_ns(req, nameserver_pick(req->base));
723	if (req->ns == last_ns) {
724		/* ... but pick did return it */
725		/* not a lot of point in trying again with the */
726		/* same server */
727		return 1;
728	}
729
730	req->reissue_count++;
731	req->tx_count = 0;
732	req->transmit_me = 1;
733
734	return 0;
735}
736
737/* this function looks for space on the inflight queue and promotes */
738/* requests from the waiting queue if it can. */
739/* */
740/* TODO: */
741/* add return code, see at nameserver_pick() and other functions. */
742static void
743evdns_requests_pump_waiting_queue(struct evdns_base *base) {
744	ASSERT_LOCKED(base);
745	while (base->global_requests_inflight < base->global_max_requests_inflight &&
746		   base->global_requests_waiting) {
747		struct request *req;
748
749		EVUTIL_ASSERT(base->req_waiting_head);
750		req = base->req_waiting_head;
751
752		req->ns = nameserver_pick(base);
753		if (!req->ns)
754			return;
755
756		/* move a request from the waiting queue to the inflight queue */
757		req->ns->requests_inflight++;
758
759		evdns_request_remove(req, &base->req_waiting_head);
760
761		base->global_requests_waiting--;
762		base->global_requests_inflight++;
763
764		request_trans_id_set(req, transaction_id_pick(base));
765
766		evdns_request_insert(req, &REQ_HEAD(base, req->trans_id));
767		evdns_request_transmit(req);
768		evdns_transmit(base);
769	}
770}
771
772/* TODO(nickm) document */
773struct deferred_reply_callback {
774	struct event_callback deferred;
775	struct evdns_request *handle;
776	u8 request_type;
777	u8 have_reply;
778	u32 ttl;
779	u32 err;
780	evdns_callback_type user_callback;
781	struct reply reply;
782};
783
784static void
785reply_run_callback(struct event_callback *d, void *user_pointer)
786{
787	struct deferred_reply_callback *cb =
788	    EVUTIL_UPCAST(d, struct deferred_reply_callback, deferred);
789
790	switch (cb->request_type) {
791	case TYPE_A:
792		if (cb->have_reply)
793			cb->user_callback(DNS_ERR_NONE, DNS_IPv4_A,
794			    cb->reply.data.a.addrcount, cb->ttl,
795			    cb->reply.data.a.addresses,
796			    user_pointer);
797		else
798			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
799		break;
800	case TYPE_PTR:
801		if (cb->have_reply) {
802			char *name = cb->reply.data.ptr.name;
803			cb->user_callback(DNS_ERR_NONE, DNS_PTR, 1, cb->ttl,
804			    &name, user_pointer);
805		} else {
806			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
807		}
808		break;
809	case TYPE_AAAA:
810		if (cb->have_reply)
811			cb->user_callback(DNS_ERR_NONE, DNS_IPv6_AAAA,
812			    cb->reply.data.aaaa.addrcount, cb->ttl,
813			    cb->reply.data.aaaa.addresses,
814			    user_pointer);
815		else
816			cb->user_callback(cb->err, 0, 0, cb->ttl, NULL, user_pointer);
817		break;
818	default:
819		EVUTIL_ASSERT(0);
820	}
821
822	if (cb->handle && cb->handle->pending_cb) {
823		mm_free(cb->handle);
824	}
825
826	mm_free(cb);
827}
828
829static void
830reply_schedule_callback(struct request *const req, u32 ttl, u32 err, struct reply *reply)
831{
832	struct deferred_reply_callback *d = mm_calloc(1, sizeof(*d));
833
834	if (!d) {
835		event_warn("%s: Couldn't allocate space for deferred callback.",
836		    __func__);
837		return;
838	}
839
840	ASSERT_LOCKED(req->base);
841
842	d->request_type = req->request_type;
843	d->user_callback = req->user_callback;
844	d->ttl = ttl;
845	d->err = err;
846	if (reply) {
847		d->have_reply = 1;
848		memcpy(&d->reply, reply, sizeof(struct reply));
849	}
850
851	if (req->handle) {
852		req->handle->pending_cb = 1;
853		d->handle = req->handle;
854	}
855
856	event_deferred_cb_init_(
857	    &d->deferred,
858	    event_get_priority(&req->timeout_event),
859	    reply_run_callback,
860	    req->user_pointer);
861	event_deferred_cb_schedule_(
862		req->base->event_base,
863		&d->deferred);
864}
865
866/* this processes a parsed reply packet */
867static void
868reply_handle(struct request *const req, u16 flags, u32 ttl, struct reply *reply) {
869	int error;
870	char addrbuf[128];
871	static const int error_codes[] = {
872		DNS_ERR_FORMAT, DNS_ERR_SERVERFAILED, DNS_ERR_NOTEXIST,
873		DNS_ERR_NOTIMPL, DNS_ERR_REFUSED
874	};
875
876	ASSERT_LOCKED(req->base);
877	ASSERT_VALID_REQUEST(req);
878
879	if (flags & 0x020f || !reply || !reply->have_answer) {
880		/* there was an error */
881		if (flags & 0x0200) {
882			error = DNS_ERR_TRUNCATED;
883		} else if (flags & 0x000f) {
884			u16 error_code = (flags & 0x000f) - 1;
885			if (error_code > 4) {
886				error = DNS_ERR_UNKNOWN;
887			} else {
888				error = error_codes[error_code];
889			}
890		} else if (reply && !reply->have_answer) {
891			error = DNS_ERR_NODATA;
892		} else {
893			error = DNS_ERR_UNKNOWN;
894		}
895
896		switch (error) {
897		case DNS_ERR_NOTIMPL:
898		case DNS_ERR_REFUSED:
899			/* we regard these errors as marking a bad nameserver */
900			if (req->reissue_count < req->base->global_max_reissues) {
901				char msg[64];
902				evutil_snprintf(msg, sizeof(msg), "Bad response %d (%s)",
903					 error, evdns_err_to_string(error));
904				nameserver_failed(req->ns, msg);
905				if (!request_reissue(req)) return;
906			}
907			break;
908		case DNS_ERR_SERVERFAILED:
909			/* rcode 2 (servfailed) sometimes means "we
910			 * are broken" and sometimes (with some binds)
911			 * means "that request was very confusing."
912			 * Treat this as a timeout, not a failure.
913			 */
914			log(EVDNS_LOG_DEBUG, "Got a SERVERFAILED from nameserver"
915				"at %s; will allow the request to time out.",
916			    evutil_format_sockaddr_port_(
917				    (struct sockaddr *)&req->ns->address,
918				    addrbuf, sizeof(addrbuf)));
919			/* Call the timeout function */
920			evdns_request_timeout_callback(0, 0, req);
921			return;
922		default:
923			/* we got a good reply from the nameserver: it is up. */
924			if (req->handle == req->ns->probe_request) {
925				/* Avoid double-free */
926				req->ns->probe_request = NULL;
927			}
928
929			nameserver_up(req->ns);
930		}
931
932		if (req->handle->search_state &&
933		    req->request_type != TYPE_PTR) {
934			/* if we have a list of domains to search in,
935			 * try the next one */
936			if (!search_try_next(req->handle)) {
937				/* a new request was issued so this
938				 * request is finished and */
939				/* the user callback will be made when
940				 * that request (or a */
941				/* child of it) finishes. */
942				return;
943			}
944		}
945
946		/* all else failed. Pass the failure up */
947		reply_schedule_callback(req, ttl, error, NULL);
948		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
949	} else {
950		/* all ok, tell the user */
951		reply_schedule_callback(req, ttl, 0, reply);
952		if (req->handle == req->ns->probe_request)
953			req->ns->probe_request = NULL; /* Avoid double-free */
954		nameserver_up(req->ns);
955		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
956	}
957}
958
959static int
960name_parse(u8 *packet, int length, int *idx, char *name_out, int name_out_len) {
961	int name_end = -1;
962	int j = *idx;
963	int ptr_count = 0;
964#define GET32(x) do { if (j + 4 > length) goto err; memcpy(&t32_, packet + j, 4); j += 4; x = ntohl(t32_); } while (0)
965#define GET16(x) do { if (j + 2 > length) goto err; memcpy(&t_, packet + j, 2); j += 2; x = ntohs(t_); } while (0)
966#define GET8(x) do { if (j >= length) goto err; x = packet[j++]; } while (0)
967
968	char *cp = name_out;
969	const char *const end = name_out + name_out_len;
970
971	/* Normally, names are a series of length prefixed strings terminated */
972	/* with a length of 0 (the lengths are u8's < 63). */
973	/* However, the length can start with a pair of 1 bits and that */
974	/* means that the next 14 bits are a pointer within the current */
975	/* packet. */
976
977	for (;;) {
978		u8 label_len;
979		if (j >= length) return -1;
980		GET8(label_len);
981		if (!label_len) break;
982		if (label_len & 0xc0) {
983			u8 ptr_low;
984			GET8(ptr_low);
985			if (name_end < 0) name_end = j;
986			j = (((int)label_len & 0x3f) << 8) + ptr_low;
987			/* Make sure that the target offset is in-bounds. */
988			if (j < 0 || j >= length) return -1;
989			/* If we've jumped more times than there are characters in the
990			 * message, we must have a loop. */
991			if (++ptr_count > length) return -1;
992			continue;
993		}
994		if (label_len > 63) return -1;
995		if (cp != name_out) {
996			if (cp + 1 >= end) return -1;
997			*cp++ = '.';
998		}
999		if (cp + label_len >= end) return -1;
1000		memcpy(cp, packet + j, label_len);
1001		cp += label_len;
1002		j += label_len;
1003	}
1004	if (cp >= end) return -1;
1005	*cp = '\0';
1006	if (name_end < 0)
1007		*idx = j;
1008	else
1009		*idx = name_end;
1010	return 0;
1011 err:
1012	return -1;
1013}
1014
1015/* parses a raw request from a nameserver */
1016static int
1017reply_parse(struct evdns_base *base, u8 *packet, int length) {
1018	int j = 0, k = 0;  /* index into packet */
1019	u16 t_;	 /* used by the macros */
1020	u32 t32_;  /* used by the macros */
1021	char tmp_name[256], cmp_name[256]; /* used by the macros */
1022	int name_matches = 0;
1023
1024	u16 trans_id, questions, answers, authority, additional, datalength;
1025	u16 flags = 0;
1026	u32 ttl, ttl_r = 0xffffffff;
1027	struct reply reply;
1028	struct request *req = NULL;
1029	unsigned int i;
1030
1031	ASSERT_LOCKED(base);
1032
1033	GET16(trans_id);
1034	GET16(flags);
1035	GET16(questions);
1036	GET16(answers);
1037	GET16(authority);
1038	GET16(additional);
1039	(void) authority; /* suppress "unused variable" warnings. */
1040	(void) additional; /* suppress "unused variable" warnings. */
1041
1042	req = request_find_from_trans_id(base, trans_id);
1043	if (!req) return -1;
1044	EVUTIL_ASSERT(req->base == base);
1045
1046	memset(&reply, 0, sizeof(reply));
1047
1048	/* If it's not an answer, it doesn't correspond to any request. */
1049	if (!(flags & 0x8000)) return -1;  /* must be an answer */
1050	if ((flags & 0x020f) && (flags & 0x020f) != DNS_ERR_NOTEXIST) {
1051		/* there was an error and it's not NXDOMAIN */
1052		goto err;
1053	}
1054	/* if (!answers) return; */  /* must have an answer of some form */
1055
1056	/* This macro skips a name in the DNS reply. */
1057#define SKIP_NAME						\
1058	do { tmp_name[0] = '\0';				\
1059		if (name_parse(packet, length, &j, tmp_name,	\
1060			sizeof(tmp_name))<0)			\
1061			goto err;				\
1062	} while (0)
1063#define TEST_NAME							\
1064	do { tmp_name[0] = '\0';					\
1065		cmp_name[0] = '\0';					\
1066		k = j;							\
1067		if (name_parse(packet, length, &j, tmp_name,		\
1068			sizeof(tmp_name))<0)				\
1069			goto err;					\
1070		if (name_parse(req->request, req->request_len, &k,	\
1071			cmp_name, sizeof(cmp_name))<0)			\
1072			goto err;					\
1073		if (base->global_randomize_case) {			\
1074			if (strcmp(tmp_name, cmp_name) == 0)		\
1075				name_matches = 1;			\
1076		} else {						\
1077			if (evutil_ascii_strcasecmp(tmp_name, cmp_name) == 0) \
1078				name_matches = 1;			\
1079		}							\
1080	} while (0)
1081
1082	reply.type = req->request_type;
1083
1084	/* skip over each question in the reply */
1085	for (i = 0; i < questions; ++i) {
1086		/* the question looks like
1087		 *   <label:name><u16:type><u16:class>
1088		 */
1089		TEST_NAME;
1090		j += 4;
1091		if (j > length) goto err;
1092	}
1093
1094	if (!name_matches)
1095		goto err;
1096
1097	/* now we have the answer section which looks like
1098	 * <label:name><u16:type><u16:class><u32:ttl><u16:len><data...>
1099	 */
1100
1101	for (i = 0; i < answers; ++i) {
1102		u16 type, class;
1103
1104		SKIP_NAME;
1105		GET16(type);
1106		GET16(class);
1107		GET32(ttl);
1108		GET16(datalength);
1109
1110		if (type == TYPE_A && class == CLASS_INET) {
1111			int addrcount, addrtocopy;
1112			if (req->request_type != TYPE_A) {
1113				j += datalength; continue;
1114			}
1115			if ((datalength & 3) != 0) /* not an even number of As. */
1116			    goto err;
1117			addrcount = datalength >> 2;
1118			addrtocopy = MIN(MAX_V4_ADDRS - reply.data.a.addrcount, (unsigned)addrcount);
1119
1120			ttl_r = MIN(ttl_r, ttl);
1121			/* we only bother with the first four addresses. */
1122			if (j + 4*addrtocopy > length) goto err;
1123			memcpy(&reply.data.a.addresses[reply.data.a.addrcount],
1124				   packet + j, 4*addrtocopy);
1125			j += 4*addrtocopy;
1126			reply.data.a.addrcount += addrtocopy;
1127			reply.have_answer = 1;
1128			if (reply.data.a.addrcount == MAX_V4_ADDRS) break;
1129		} else if (type == TYPE_PTR && class == CLASS_INET) {
1130			if (req->request_type != TYPE_PTR) {
1131				j += datalength; continue;
1132			}
1133			if (name_parse(packet, length, &j, reply.data.ptr.name,
1134						   sizeof(reply.data.ptr.name))<0)
1135				goto err;
1136			ttl_r = MIN(ttl_r, ttl);
1137			reply.have_answer = 1;
1138			break;
1139		} else if (type == TYPE_CNAME) {
1140			char cname[HOST_NAME_MAX];
1141			if (!req->put_cname_in_ptr || *req->put_cname_in_ptr) {
1142				j += datalength; continue;
1143			}
1144			if (name_parse(packet, length, &j, cname,
1145				sizeof(cname))<0)
1146				goto err;
1147			*req->put_cname_in_ptr = mm_strdup(cname);
1148		} else if (type == TYPE_AAAA && class == CLASS_INET) {
1149			int addrcount, addrtocopy;
1150			if (req->request_type != TYPE_AAAA) {
1151				j += datalength; continue;
1152			}
1153			if ((datalength & 15) != 0) /* not an even number of AAAAs. */
1154				goto err;
1155			addrcount = datalength >> 4;  /* each address is 16 bytes long */
1156			addrtocopy = MIN(MAX_V6_ADDRS - reply.data.aaaa.addrcount, (unsigned)addrcount);
1157			ttl_r = MIN(ttl_r, ttl);
1158
1159			/* we only bother with the first four addresses. */
1160			if (j + 16*addrtocopy > length) goto err;
1161			memcpy(&reply.data.aaaa.addresses[reply.data.aaaa.addrcount],
1162				   packet + j, 16*addrtocopy);
1163			reply.data.aaaa.addrcount += addrtocopy;
1164			j += 16*addrtocopy;
1165			reply.have_answer = 1;
1166			if (reply.data.aaaa.addrcount == MAX_V6_ADDRS) break;
1167		} else {
1168			/* skip over any other type of resource */
1169			j += datalength;
1170		}
1171	}
1172
1173	if (!reply.have_answer) {
1174		for (i = 0; i < authority; ++i) {
1175			u16 type, class;
1176			SKIP_NAME;
1177			GET16(type);
1178			GET16(class);
1179			GET32(ttl);
1180			GET16(datalength);
1181			if (type == TYPE_SOA && class == CLASS_INET) {
1182				u32 serial, refresh, retry, expire, minimum;
1183				SKIP_NAME;
1184				SKIP_NAME;
1185				GET32(serial);
1186				GET32(refresh);
1187				GET32(retry);
1188				GET32(expire);
1189				GET32(minimum);
1190				(void)expire;
1191				(void)retry;
1192				(void)refresh;
1193				(void)serial;
1194				ttl_r = MIN(ttl_r, ttl);
1195				ttl_r = MIN(ttl_r, minimum);
1196			} else {
1197				/* skip over any other type of resource */
1198				j += datalength;
1199			}
1200		}
1201	}
1202
1203	if (ttl_r == 0xffffffff)
1204		ttl_r = 0;
1205
1206	reply_handle(req, flags, ttl_r, &reply);
1207	return 0;
1208 err:
1209	if (req)
1210		reply_handle(req, flags, 0, NULL);
1211	return -1;
1212}
1213
1214/* Parse a raw request (packet,length) sent to a nameserver port (port) from */
1215/* a DNS client (addr,addrlen), and if it's well-formed, call the corresponding */
1216/* callback. */
1217static int
1218request_parse(u8 *packet, int length, struct evdns_server_port *port, struct sockaddr *addr, ev_socklen_t addrlen)
1219{
1220	int j = 0;	/* index into packet */
1221	u16 t_;	 /* used by the macros */
1222	char tmp_name[256]; /* used by the macros */
1223
1224	int i;
1225	u16 trans_id, flags, questions, answers, authority, additional;
1226	struct server_request *server_req = NULL;
1227
1228	ASSERT_LOCKED(port);
1229
1230	/* Get the header fields */
1231	GET16(trans_id);
1232	GET16(flags);
1233	GET16(questions);
1234	GET16(answers);
1235	GET16(authority);
1236	GET16(additional);
1237	(void)answers;
1238	(void)additional;
1239	(void)authority;
1240
1241	if (flags & 0x8000) return -1; /* Must not be an answer. */
1242	flags &= 0x0110; /* Only RD and CD get preserved. */
1243
1244	server_req = mm_malloc(sizeof(struct server_request));
1245	if (server_req == NULL) return -1;
1246	memset(server_req, 0, sizeof(struct server_request));
1247
1248	server_req->trans_id = trans_id;
1249	memcpy(&server_req->addr, addr, addrlen);
1250	server_req->addrlen = addrlen;
1251
1252	server_req->base.flags = flags;
1253	server_req->base.nquestions = 0;
1254	server_req->base.questions = mm_calloc(sizeof(struct evdns_server_question *), questions);
1255	if (server_req->base.questions == NULL)
1256		goto err;
1257
1258	for (i = 0; i < questions; ++i) {
1259		u16 type, class;
1260		struct evdns_server_question *q;
1261		int namelen;
1262		if (name_parse(packet, length, &j, tmp_name, sizeof(tmp_name))<0)
1263			goto err;
1264		GET16(type);
1265		GET16(class);
1266		namelen = (int)strlen(tmp_name);
1267		q = mm_malloc(sizeof(struct evdns_server_question) + namelen);
1268		if (!q)
1269			goto err;
1270		q->type = type;
1271		q->dns_question_class = class;
1272		memcpy(q->name, tmp_name, namelen+1);
1273		server_req->base.questions[server_req->base.nquestions++] = q;
1274	}
1275
1276	/* Ignore answers, authority, and additional. */
1277
1278	server_req->port = port;
1279	port->refcnt++;
1280
1281	/* Only standard queries are supported. */
1282	if (flags & 0x7800) {
1283		evdns_server_request_respond(&(server_req->base), DNS_ERR_NOTIMPL);
1284		return -1;
1285	}
1286
1287	port->user_callback(&(server_req->base), port->user_data);
1288
1289	return 0;
1290err:
1291	if (server_req) {
1292		if (server_req->base.questions) {
1293			for (i = 0; i < server_req->base.nquestions; ++i)
1294				mm_free(server_req->base.questions[i]);
1295			mm_free(server_req->base.questions);
1296		}
1297		mm_free(server_req);
1298	}
1299	return -1;
1300
1301#undef SKIP_NAME
1302#undef GET32
1303#undef GET16
1304#undef GET8
1305}
1306
1307
1308void
1309evdns_set_transaction_id_fn(ev_uint16_t (*fn)(void))
1310{
1311}
1312
1313void
1314evdns_set_random_bytes_fn(void (*fn)(char *, size_t))
1315{
1316}
1317
1318/* Try to choose a strong transaction id which isn't already in flight */
1319static u16
1320transaction_id_pick(struct evdns_base *base) {
1321	ASSERT_LOCKED(base);
1322	for (;;) {
1323		u16 trans_id;
1324		evutil_secure_rng_get_bytes(&trans_id, sizeof(trans_id));
1325
1326		if (trans_id == 0xffff) continue;
1327		/* now check to see if that id is already inflight */
1328		if (request_find_from_trans_id(base, trans_id) == NULL)
1329			return trans_id;
1330	}
1331}
1332
1333/* choose a namesever to use. This function will try to ignore */
1334/* nameservers which we think are down and load balance across the rest */
1335/* by updating the server_head global each time. */
1336static struct nameserver *
1337nameserver_pick(struct evdns_base *base) {
1338	struct nameserver *started_at = base->server_head, *picked;
1339	ASSERT_LOCKED(base);
1340	if (!base->server_head) return NULL;
1341
1342	/* if we don't have any good nameservers then there's no */
1343	/* point in trying to find one. */
1344	if (!base->global_good_nameservers) {
1345		base->server_head = base->server_head->next;
1346		return base->server_head;
1347	}
1348
1349	/* remember that nameservers are in a circular list */
1350	for (;;) {
1351		if (base->server_head->state) {
1352			/* we think this server is currently good */
1353			picked = base->server_head;
1354			base->server_head = base->server_head->next;
1355			return picked;
1356		}
1357
1358		base->server_head = base->server_head->next;
1359		if (base->server_head == started_at) {
1360			/* all the nameservers seem to be down */
1361			/* so we just return this one and hope for the */
1362			/* best */
1363			EVUTIL_ASSERT(base->global_good_nameservers == 0);
1364			picked = base->server_head;
1365			base->server_head = base->server_head->next;
1366			return picked;
1367		}
1368	}
1369}
1370
1371/* this is called when a namesever socket is ready for reading */
1372static void
1373nameserver_read(struct nameserver *ns) {
1374	struct sockaddr_storage ss;
1375	ev_socklen_t addrlen = sizeof(ss);
1376	u8 packet[1500];
1377	char addrbuf[128];
1378	ASSERT_LOCKED(ns->base);
1379
1380	for (;;) {
1381		const int r = recvfrom(ns->socket, (void*)packet,
1382		    sizeof(packet), 0,
1383		    (struct sockaddr*)&ss, &addrlen);
1384		if (r < 0) {
1385			int err = evutil_socket_geterror(ns->socket);
1386			if (EVUTIL_ERR_RW_RETRIABLE(err))
1387				return;
1388			nameserver_failed(ns,
1389			    evutil_socket_error_to_string(err));
1390			return;
1391		}
1392		if (evutil_sockaddr_cmp((struct sockaddr*)&ss,
1393			(struct sockaddr*)&ns->address, 0)) {
1394			log(EVDNS_LOG_WARN, "Address mismatch on received "
1395			    "DNS packet.  Apparent source was %s",
1396			    evutil_format_sockaddr_port_(
1397				    (struct sockaddr *)&ss,
1398				    addrbuf, sizeof(addrbuf)));
1399			return;
1400		}
1401
1402		ns->timedout = 0;
1403		reply_parse(ns->base, packet, r);
1404	}
1405}
1406
1407/* Read a packet from a DNS client on a server port s, parse it, and */
1408/* act accordingly. */
1409static void
1410server_port_read(struct evdns_server_port *s) {
1411	u8 packet[1500];
1412	struct sockaddr_storage addr;
1413	ev_socklen_t addrlen;
1414	int r;
1415	ASSERT_LOCKED(s);
1416
1417	for (;;) {
1418		addrlen = sizeof(struct sockaddr_storage);
1419		r = recvfrom(s->socket, (void*)packet, sizeof(packet), 0,
1420					 (struct sockaddr*) &addr, &addrlen);
1421		if (r < 0) {
1422			int err = evutil_socket_geterror(s->socket);
1423			if (EVUTIL_ERR_RW_RETRIABLE(err))
1424				return;
1425			log(EVDNS_LOG_WARN,
1426			    "Error %s (%d) while reading request.",
1427			    evutil_socket_error_to_string(err), err);
1428			return;
1429		}
1430		request_parse(packet, r, s, (struct sockaddr*) &addr, addrlen);
1431	}
1432}
1433
1434/* Try to write all pending replies on a given DNS server port. */
1435static void
1436server_port_flush(struct evdns_server_port *port)
1437{
1438	struct server_request *req = port->pending_replies;
1439	ASSERT_LOCKED(port);
1440	while (req) {
1441		int r = sendto(port->socket, req->response, (int)req->response_len, 0,
1442			   (struct sockaddr*) &req->addr, (ev_socklen_t)req->addrlen);
1443		if (r < 0) {
1444			int err = evutil_socket_geterror(port->socket);
1445			if (EVUTIL_ERR_RW_RETRIABLE(err))
1446				return;
1447			log(EVDNS_LOG_WARN, "Error %s (%d) while writing response to port; dropping", evutil_socket_error_to_string(err), err);
1448		}
1449		if (server_request_free(req)) {
1450			/* we released the last reference to req->port. */
1451			return;
1452		} else {
1453			EVUTIL_ASSERT(req != port->pending_replies);
1454			req = port->pending_replies;
1455		}
1456	}
1457
1458	/* We have no more pending requests; stop listening for 'writeable' events. */
1459	(void) event_del(&port->event);
1460	event_assign(&port->event, port->event_base,
1461				 port->socket, EV_READ | EV_PERSIST,
1462				 server_port_ready_callback, port);
1463
1464	if (event_add(&port->event, NULL) < 0) {
1465		log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server.");
1466		/* ???? Do more? */
1467	}
1468}
1469
1470/* set if we are waiting for the ability to write to this server. */
1471/* if waiting is true then we ask libevent for EV_WRITE events, otherwise */
1472/* we stop these events. */
1473static void
1474nameserver_write_waiting(struct nameserver *ns, char waiting) {
1475	ASSERT_LOCKED(ns->base);
1476	if (ns->write_waiting == waiting) return;
1477
1478	ns->write_waiting = waiting;
1479	(void) event_del(&ns->event);
1480	event_assign(&ns->event, ns->base->event_base,
1481	    ns->socket, EV_READ | (waiting ? EV_WRITE : 0) | EV_PERSIST,
1482	    nameserver_ready_callback, ns);
1483	if (event_add(&ns->event, NULL) < 0) {
1484		char addrbuf[128];
1485		log(EVDNS_LOG_WARN, "Error from libevent when adding event for %s",
1486		    evutil_format_sockaddr_port_(
1487			    (struct sockaddr *)&ns->address,
1488			    addrbuf, sizeof(addrbuf)));
1489		/* ???? Do more? */
1490	}
1491}
1492
1493/* a callback function. Called by libevent when the kernel says that */
1494/* a nameserver socket is ready for writing or reading */
1495static void
1496nameserver_ready_callback(evutil_socket_t fd, short events, void *arg) {
1497	struct nameserver *ns = (struct nameserver *) arg;
1498	(void)fd;
1499
1500	EVDNS_LOCK(ns->base);
1501	if (events & EV_WRITE) {
1502		ns->choked = 0;
1503		if (!evdns_transmit(ns->base)) {
1504			nameserver_write_waiting(ns, 0);
1505		}
1506	}
1507	if (events & EV_READ) {
1508		nameserver_read(ns);
1509	}
1510	EVDNS_UNLOCK(ns->base);
1511}
1512
1513/* a callback function. Called by libevent when the kernel says that */
1514/* a server socket is ready for writing or reading. */
1515static void
1516server_port_ready_callback(evutil_socket_t fd, short events, void *arg) {
1517	struct evdns_server_port *port = (struct evdns_server_port *) arg;
1518	(void) fd;
1519
1520	EVDNS_LOCK(port);
1521	if (events & EV_WRITE) {
1522		port->choked = 0;
1523		server_port_flush(port);
1524	}
1525	if (events & EV_READ) {
1526		server_port_read(port);
1527	}
1528	EVDNS_UNLOCK(port);
1529}
1530
1531/* This is an inefficient representation; only use it via the dnslabel_table_*
1532 * functions, so that is can be safely replaced with something smarter later. */
1533#define MAX_LABELS 128
1534/* Structures used to implement name compression */
1535struct dnslabel_entry { char *v; off_t pos; };
1536struct dnslabel_table {
1537	int n_labels; /* number of current entries */
1538	/* map from name to position in message */
1539	struct dnslabel_entry labels[MAX_LABELS];
1540};
1541
1542/* Initialize dnslabel_table. */
1543static void
1544dnslabel_table_init(struct dnslabel_table *table)
1545{
1546	table->n_labels = 0;
1547}
1548
1549/* Free all storage held by table, but not the table itself. */
1550static void
1551dnslabel_clear(struct dnslabel_table *table)
1552{
1553	int i;
1554	for (i = 0; i < table->n_labels; ++i)
1555		mm_free(table->labels[i].v);
1556	table->n_labels = 0;
1557}
1558
1559/* return the position of the label in the current message, or -1 if the label */
1560/* hasn't been used yet. */
1561static int
1562dnslabel_table_get_pos(const struct dnslabel_table *table, const char *label)
1563{
1564	int i;
1565	for (i = 0; i < table->n_labels; ++i) {
1566		if (!strcmp(label, table->labels[i].v))
1567			return table->labels[i].pos;
1568	}
1569	return -1;
1570}
1571
1572/* remember that we've used the label at position pos */
1573static int
1574dnslabel_table_add(struct dnslabel_table *table, const char *label, off_t pos)
1575{
1576	char *v;
1577	int p;
1578	if (table->n_labels == MAX_LABELS)
1579		return (-1);
1580	v = mm_strdup(label);
1581	if (v == NULL)
1582		return (-1);
1583	p = table->n_labels++;
1584	table->labels[p].v = v;
1585	table->labels[p].pos = pos;
1586
1587	return (0);
1588}
1589
1590/* Converts a string to a length-prefixed set of DNS labels, starting */
1591/* at buf[j]. name and buf must not overlap. name_len should be the length */
1592/* of name.	 table is optional, and is used for compression. */
1593/* */
1594/* Input: abc.def */
1595/* Output: <3>abc<3>def<0> */
1596/* */
1597/* Returns the first index after the encoded name, or negative on error. */
1598/*	 -1	 label was > 63 bytes */
1599/*	 -2	 name too long to fit in buffer. */
1600/* */
1601static off_t
1602dnsname_to_labels(u8 *const buf, size_t buf_len, off_t j,
1603				  const char *name, const size_t name_len,
1604				  struct dnslabel_table *table) {
1605	const char *end = name + name_len;
1606	int ref = 0;
1607	u16 t_;
1608
1609#define APPEND16(x) do {						\
1610		if (j + 2 > (off_t)buf_len)				\
1611			goto overflow;					\
1612		t_ = htons(x);						\
1613		memcpy(buf + j, &t_, 2);				\
1614		j += 2;							\
1615	} while (0)
1616#define APPEND32(x) do {						\
1617		if (j + 4 > (off_t)buf_len)				\
1618			goto overflow;					\
1619		t32_ = htonl(x);					\
1620		memcpy(buf + j, &t32_, 4);				\
1621		j += 4;							\
1622	} while (0)
1623
1624	if (name_len > 255) return -2;
1625
1626	for (;;) {
1627		const char *const start = name;
1628		if (table && (ref = dnslabel_table_get_pos(table, name)) >= 0) {
1629			APPEND16(ref | 0xc000);
1630			return j;
1631		}
1632		name = strchr(name, '.');
1633		if (!name) {
1634			const size_t label_len = end - start;
1635			if (label_len > 63) return -1;
1636			if ((size_t)(j+label_len+1) > buf_len) return -2;
1637			if (table) dnslabel_table_add(table, start, j);
1638			buf[j++] = (ev_uint8_t)label_len;
1639
1640			memcpy(buf + j, start, label_len);
1641			j += (int) label_len;
1642			break;
1643		} else {
1644			/* append length of the label. */
1645			const size_t label_len = name - start;
1646			if (label_len > 63) return -1;
1647			if ((size_t)(j+label_len+1) > buf_len) return -2;
1648			if (table) dnslabel_table_add(table, start, j);
1649			buf[j++] = (ev_uint8_t)label_len;
1650
1651			memcpy(buf + j, start, label_len);
1652			j += (int) label_len;
1653			/* hop over the '.' */
1654			name++;
1655		}
1656	}
1657
1658	/* the labels must be terminated by a 0. */
1659	/* It's possible that the name ended in a . */
1660	/* in which case the zero is already there */
1661	if (!j || buf[j-1]) buf[j++] = 0;
1662	return j;
1663 overflow:
1664	return (-2);
1665}
1666
1667/* Finds the length of a dns request for a DNS name of the given */
1668/* length. The actual request may be smaller than the value returned */
1669/* here */
1670static size_t
1671evdns_request_len(const size_t name_len) {
1672	return 96 + /* length of the DNS standard header */
1673		name_len + 2 +
1674		4;  /* space for the resource type */
1675}
1676
1677/* build a dns request packet into buf. buf should be at least as long */
1678/* as evdns_request_len told you it should be. */
1679/* */
1680/* Returns the amount of space used. Negative on error. */
1681static int
1682evdns_request_data_build(const char *const name, const size_t name_len,
1683    const u16 trans_id, const u16 type, const u16 class,
1684    u8 *const buf, size_t buf_len) {
1685	off_t j = 0;  /* current offset into buf */
1686	u16 t_;	 /* used by the macros */
1687
1688	APPEND16(trans_id);
1689	APPEND16(0x0100);  /* standard query, recusion needed */
1690	APPEND16(1);  /* one question */
1691	APPEND16(0);  /* no answers */
1692	APPEND16(0);  /* no authority */
1693	APPEND16(0);  /* no additional */
1694
1695	j = dnsname_to_labels(buf, buf_len, j, name, name_len, NULL);
1696	if (j < 0) {
1697		return (int)j;
1698	}
1699
1700	APPEND16(type);
1701	APPEND16(class);
1702
1703	return (int)j;
1704 overflow:
1705	return (-1);
1706}
1707
1708/* exported function */
1709struct evdns_server_port *
1710evdns_add_server_port_with_base(struct event_base *base, evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data)
1711{
1712	struct evdns_server_port *port;
1713	if (flags)
1714		return NULL; /* flags not yet implemented */
1715	if (!(port = mm_malloc(sizeof(struct evdns_server_port))))
1716		return NULL;
1717	memset(port, 0, sizeof(struct evdns_server_port));
1718
1719
1720	port->socket = socket;
1721	port->refcnt = 1;
1722	port->choked = 0;
1723	port->closing = 0;
1724	port->user_callback = cb;
1725	port->user_data = user_data;
1726	port->pending_replies = NULL;
1727	port->event_base = base;
1728
1729	event_assign(&port->event, port->event_base,
1730				 port->socket, EV_READ | EV_PERSIST,
1731				 server_port_ready_callback, port);
1732	if (event_add(&port->event, NULL) < 0) {
1733		mm_free(port);
1734		return NULL;
1735	}
1736	EVTHREAD_ALLOC_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
1737	return port;
1738}
1739
1740struct evdns_server_port *
1741evdns_add_server_port(evutil_socket_t socket, int flags, evdns_request_callback_fn_type cb, void *user_data)
1742{
1743	return evdns_add_server_port_with_base(NULL, socket, flags, cb, user_data);
1744}
1745
1746/* exported function */
1747void
1748evdns_close_server_port(struct evdns_server_port *port)
1749{
1750	EVDNS_LOCK(port);
1751	if (--port->refcnt == 0) {
1752		EVDNS_UNLOCK(port);
1753		server_port_free(port);
1754	} else {
1755		port->closing = 1;
1756	}
1757}
1758
1759/* exported function */
1760int
1761evdns_server_request_add_reply(struct evdns_server_request *req_, int section, const char *name, int type, int class, int ttl, int datalen, int is_name, const char *data)
1762{
1763	struct server_request *req = TO_SERVER_REQUEST(req_);
1764	struct server_reply_item **itemp, *item;
1765	int *countp;
1766	int result = -1;
1767
1768	EVDNS_LOCK(req->port);
1769	if (req->response) /* have we already answered? */
1770		goto done;
1771
1772	switch (section) {
1773	case EVDNS_ANSWER_SECTION:
1774		itemp = &req->answer;
1775		countp = &req->n_answer;
1776		break;
1777	case EVDNS_AUTHORITY_SECTION:
1778		itemp = &req->authority;
1779		countp = &req->n_authority;
1780		break;
1781	case EVDNS_ADDITIONAL_SECTION:
1782		itemp = &req->additional;
1783		countp = &req->n_additional;
1784		break;
1785	default:
1786		goto done;
1787	}
1788	while (*itemp) {
1789		itemp = &((*itemp)->next);
1790	}
1791	item = mm_malloc(sizeof(struct server_reply_item));
1792	if (!item)
1793		goto done;
1794	item->next = NULL;
1795	if (!(item->name = mm_strdup(name))) {
1796		mm_free(item);
1797		goto done;
1798	}
1799	item->type = type;
1800	item->dns_question_class = class;
1801	item->ttl = ttl;
1802	item->is_name = is_name != 0;
1803	item->datalen = 0;
1804	item->data = NULL;
1805	if (data) {
1806		if (item->is_name) {
1807			if (!(item->data = mm_strdup(data))) {
1808				mm_free(item->name);
1809				mm_free(item);
1810				goto done;
1811			}
1812			item->datalen = (u16)-1;
1813		} else {
1814			if (!(item->data = mm_malloc(datalen))) {
1815				mm_free(item->name);
1816				mm_free(item);
1817				goto done;
1818			}
1819			item->datalen = datalen;
1820			memcpy(item->data, data, datalen);
1821		}
1822	}
1823
1824	*itemp = item;
1825	++(*countp);
1826	result = 0;
1827done:
1828	EVDNS_UNLOCK(req->port);
1829	return result;
1830}
1831
1832/* exported function */
1833int
1834evdns_server_request_add_a_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl)
1835{
1836	return evdns_server_request_add_reply(
1837		  req, EVDNS_ANSWER_SECTION, name, TYPE_A, CLASS_INET,
1838		  ttl, n*4, 0, addrs);
1839}
1840
1841/* exported function */
1842int
1843evdns_server_request_add_aaaa_reply(struct evdns_server_request *req, const char *name, int n, const void *addrs, int ttl)
1844{
1845	return evdns_server_request_add_reply(
1846		  req, EVDNS_ANSWER_SECTION, name, TYPE_AAAA, CLASS_INET,
1847		  ttl, n*16, 0, addrs);
1848}
1849
1850/* exported function */
1851int
1852evdns_server_request_add_ptr_reply(struct evdns_server_request *req, struct in_addr *in, const char *inaddr_name, const char *hostname, int ttl)
1853{
1854	u32 a;
1855	char buf[32];
1856	if (in && inaddr_name)
1857		return -1;
1858	else if (!in && !inaddr_name)
1859		return -1;
1860	if (in) {
1861		a = ntohl(in->s_addr);
1862		evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
1863				(int)(u8)((a	)&0xff),
1864				(int)(u8)((a>>8 )&0xff),
1865				(int)(u8)((a>>16)&0xff),
1866				(int)(u8)((a>>24)&0xff));
1867		inaddr_name = buf;
1868	}
1869	return evdns_server_request_add_reply(
1870		  req, EVDNS_ANSWER_SECTION, inaddr_name, TYPE_PTR, CLASS_INET,
1871		  ttl, -1, 1, hostname);
1872}
1873
1874/* exported function */
1875int
1876evdns_server_request_add_cname_reply(struct evdns_server_request *req, const char *name, const char *cname, int ttl)
1877{
1878	return evdns_server_request_add_reply(
1879		  req, EVDNS_ANSWER_SECTION, name, TYPE_CNAME, CLASS_INET,
1880		  ttl, -1, 1, cname);
1881}
1882
1883/* exported function */
1884void
1885evdns_server_request_set_flags(struct evdns_server_request *exreq, int flags)
1886{
1887	struct server_request *req = TO_SERVER_REQUEST(exreq);
1888	req->base.flags &= ~(EVDNS_FLAGS_AA|EVDNS_FLAGS_RD);
1889	req->base.flags |= flags;
1890}
1891
1892static int
1893evdns_server_request_format_response(struct server_request *req, int err)
1894{
1895	unsigned char buf[1500];
1896	size_t buf_len = sizeof(buf);
1897	off_t j = 0, r;
1898	u16 t_;
1899	u32 t32_;
1900	int i;
1901	u16 flags;
1902	struct dnslabel_table table;
1903
1904	if (err < 0 || err > 15) return -1;
1905
1906	/* Set response bit and error code; copy OPCODE and RD fields from
1907	 * question; copy RA and AA if set by caller. */
1908	flags = req->base.flags;
1909	flags |= (0x8000 | err);
1910
1911	dnslabel_table_init(&table);
1912	APPEND16(req->trans_id);
1913	APPEND16(flags);
1914	APPEND16(req->base.nquestions);
1915	APPEND16(req->n_answer);
1916	APPEND16(req->n_authority);
1917	APPEND16(req->n_additional);
1918
1919	/* Add questions. */
1920	for (i=0; i < req->base.nquestions; ++i) {
1921		const char *s = req->base.questions[i]->name;
1922		j = dnsname_to_labels(buf, buf_len, j, s, strlen(s), &table);
1923		if (j < 0) {
1924			dnslabel_clear(&table);
1925			return (int) j;
1926		}
1927		APPEND16(req->base.questions[i]->type);
1928		APPEND16(req->base.questions[i]->dns_question_class);
1929	}
1930
1931	/* Add answer, authority, and additional sections. */
1932	for (i=0; i<3; ++i) {
1933		struct server_reply_item *item;
1934		if (i==0)
1935			item = req->answer;
1936		else if (i==1)
1937			item = req->authority;
1938		else
1939			item = req->additional;
1940		while (item) {
1941			r = dnsname_to_labels(buf, buf_len, j, item->name, strlen(item->name), &table);
1942			if (r < 0)
1943				goto overflow;
1944			j = r;
1945
1946			APPEND16(item->type);
1947			APPEND16(item->dns_question_class);
1948			APPEND32(item->ttl);
1949			if (item->is_name) {
1950				off_t len_idx = j, name_start;
1951				j += 2;
1952				name_start = j;
1953				r = dnsname_to_labels(buf, buf_len, j, item->data, strlen(item->data), &table);
1954				if (r < 0)
1955					goto overflow;
1956				j = r;
1957				t_ = htons( (short) (j-name_start) );
1958				memcpy(buf+len_idx, &t_, 2);
1959			} else {
1960				APPEND16(item->datalen);
1961				if (j+item->datalen > (off_t)buf_len)
1962					goto overflow;
1963				memcpy(buf+j, item->data, item->datalen);
1964				j += item->datalen;
1965			}
1966			item = item->next;
1967		}
1968	}
1969
1970	if (j > 512) {
1971overflow:
1972		j = 512;
1973		buf[2] |= 0x02; /* set the truncated bit. */
1974	}
1975
1976	req->response_len = j;
1977
1978	if (!(req->response = mm_malloc(req->response_len))) {
1979		server_request_free_answers(req);
1980		dnslabel_clear(&table);
1981		return (-1);
1982	}
1983	memcpy(req->response, buf, req->response_len);
1984	server_request_free_answers(req);
1985	dnslabel_clear(&table);
1986	return (0);
1987}
1988
1989/* exported function */
1990int
1991evdns_server_request_respond(struct evdns_server_request *req_, int err)
1992{
1993	struct server_request *req = TO_SERVER_REQUEST(req_);
1994	struct evdns_server_port *port = req->port;
1995	int r = -1;
1996
1997	EVDNS_LOCK(port);
1998	if (!req->response) {
1999		if ((r = evdns_server_request_format_response(req, err))<0)
2000			goto done;
2001	}
2002
2003	r = sendto(port->socket, req->response, (int)req->response_len, 0,
2004			   (struct sockaddr*) &req->addr, (ev_socklen_t)req->addrlen);
2005	if (r<0) {
2006		int sock_err = evutil_socket_geterror(port->socket);
2007		if (EVUTIL_ERR_RW_RETRIABLE(sock_err))
2008			goto done;
2009
2010		if (port->pending_replies) {
2011			req->prev_pending = port->pending_replies->prev_pending;
2012			req->next_pending = port->pending_replies;
2013			req->prev_pending->next_pending =
2014				req->next_pending->prev_pending = req;
2015		} else {
2016			req->prev_pending = req->next_pending = req;
2017			port->pending_replies = req;
2018			port->choked = 1;
2019
2020			(void) event_del(&port->event);
2021			event_assign(&port->event, port->event_base, port->socket, (port->closing?0:EV_READ) | EV_WRITE | EV_PERSIST, server_port_ready_callback, port);
2022
2023			if (event_add(&port->event, NULL) < 0) {
2024				log(EVDNS_LOG_WARN, "Error from libevent when adding event for DNS server");
2025			}
2026
2027		}
2028
2029		r = 1;
2030		goto done;
2031	}
2032	if (server_request_free(req)) {
2033		r = 0;
2034		goto done;
2035	}
2036
2037	if (port->pending_replies)
2038		server_port_flush(port);
2039
2040	r = 0;
2041done:
2042	EVDNS_UNLOCK(port);
2043	return r;
2044}
2045
2046/* Free all storage held by RRs in req. */
2047static void
2048server_request_free_answers(struct server_request *req)
2049{
2050	struct server_reply_item *victim, *next, **list;
2051	int i;
2052	for (i = 0; i < 3; ++i) {
2053		if (i==0)
2054			list = &req->answer;
2055		else if (i==1)
2056			list = &req->authority;
2057		else
2058			list = &req->additional;
2059
2060		victim = *list;
2061		while (victim) {
2062			next = victim->next;
2063			mm_free(victim->name);
2064			if (victim->data)
2065				mm_free(victim->data);
2066			mm_free(victim);
2067			victim = next;
2068		}
2069		*list = NULL;
2070	}
2071}
2072
2073/* Free all storage held by req, and remove links to it. */
2074/* return true iff we just wound up freeing the server_port. */
2075static int
2076server_request_free(struct server_request *req)
2077{
2078	int i, rc=1, lock=0;
2079	if (req->base.questions) {
2080		for (i = 0; i < req->base.nquestions; ++i)
2081			mm_free(req->base.questions[i]);
2082		mm_free(req->base.questions);
2083	}
2084
2085	if (req->port) {
2086		EVDNS_LOCK(req->port);
2087		lock=1;
2088		if (req->port->pending_replies == req) {
2089			if (req->next_pending && req->next_pending != req)
2090				req->port->pending_replies = req->next_pending;
2091			else
2092				req->port->pending_replies = NULL;
2093		}
2094		rc = --req->port->refcnt;
2095	}
2096
2097	if (req->response) {
2098		mm_free(req->response);
2099	}
2100
2101	server_request_free_answers(req);
2102
2103	if (req->next_pending && req->next_pending != req) {
2104		req->next_pending->prev_pending = req->prev_pending;
2105		req->prev_pending->next_pending = req->next_pending;
2106	}
2107
2108	if (rc == 0) {
2109		EVDNS_UNLOCK(req->port); /* ????? nickm */
2110		server_port_free(req->port);
2111		mm_free(req);
2112		return (1);
2113	}
2114	if (lock)
2115		EVDNS_UNLOCK(req->port);
2116	mm_free(req);
2117	return (0);
2118}
2119
2120/* Free all storage held by an evdns_server_port.  Only called when  */
2121static void
2122server_port_free(struct evdns_server_port *port)
2123{
2124	EVUTIL_ASSERT(port);
2125	EVUTIL_ASSERT(!port->refcnt);
2126	EVUTIL_ASSERT(!port->pending_replies);
2127	if (port->socket > 0) {
2128		evutil_closesocket(port->socket);
2129		port->socket = -1;
2130	}
2131	(void) event_del(&port->event);
2132	event_debug_unassign(&port->event);
2133	EVTHREAD_FREE_LOCK(port->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
2134	mm_free(port);
2135}
2136
2137/* exported function */
2138int
2139evdns_server_request_drop(struct evdns_server_request *req_)
2140{
2141	struct server_request *req = TO_SERVER_REQUEST(req_);
2142	server_request_free(req);
2143	return 0;
2144}
2145
2146/* exported function */
2147int
2148evdns_server_request_get_requesting_addr(struct evdns_server_request *req_, struct sockaddr *sa, int addr_len)
2149{
2150	struct server_request *req = TO_SERVER_REQUEST(req_);
2151	if (addr_len < (int)req->addrlen)
2152		return -1;
2153	memcpy(sa, &(req->addr), req->addrlen);
2154	return req->addrlen;
2155}
2156
2157#undef APPEND16
2158#undef APPEND32
2159
2160/* this is a libevent callback function which is called when a request */
2161/* has timed out. */
2162static void
2163evdns_request_timeout_callback(evutil_socket_t fd, short events, void *arg) {
2164	struct request *const req = (struct request *) arg;
2165	struct evdns_base *base = req->base;
2166
2167	(void) fd;
2168	(void) events;
2169
2170	log(EVDNS_LOG_DEBUG, "Request %p timed out", arg);
2171	EVDNS_LOCK(base);
2172
2173	if (req->tx_count >= req->base->global_max_retransmits) {
2174		struct nameserver *ns = req->ns;
2175		/* this request has failed */
2176		log(EVDNS_LOG_DEBUG, "Giving up on request %p; tx_count==%d",
2177		    arg, req->tx_count);
2178		reply_schedule_callback(req, 0, DNS_ERR_TIMEOUT, NULL);
2179
2180		request_finished(req, &REQ_HEAD(req->base, req->trans_id), 1);
2181		nameserver_failed(ns, "request timed out.");
2182	} else {
2183		/* retransmit it */
2184		log(EVDNS_LOG_DEBUG, "Retransmitting request %p; tx_count==%d",
2185		    arg, req->tx_count);
2186		(void) evtimer_del(&req->timeout_event);
2187		request_swap_ns(req, nameserver_pick(base));
2188		evdns_request_transmit(req);
2189
2190		req->ns->timedout++;
2191		if (req->ns->timedout > req->base->global_max_nameserver_timeout) {
2192			req->ns->timedout = 0;
2193			nameserver_failed(req->ns, "request timed out.");
2194		}
2195	}
2196
2197	EVDNS_UNLOCK(base);
2198}
2199
2200/* try to send a request to a given server. */
2201/* */
2202/* return: */
2203/*   0 ok */
2204/*   1 temporary failure */
2205/*   2 other failure */
2206static int
2207evdns_request_transmit_to(struct request *req, struct nameserver *server) {
2208	int r;
2209	ASSERT_LOCKED(req->base);
2210	ASSERT_VALID_REQUEST(req);
2211
2212	if (server->requests_inflight == 1 &&
2213		req->base->disable_when_inactive &&
2214		event_add(&server->event, NULL) < 0) {
2215		return 1;
2216	}
2217
2218	r = sendto(server->socket, (void*)req->request, req->request_len, 0,
2219	    (struct sockaddr *)&server->address, server->addrlen);
2220	if (r < 0) {
2221		int err = evutil_socket_geterror(server->socket);
2222		if (EVUTIL_ERR_RW_RETRIABLE(err))
2223			return 1;
2224		nameserver_failed(req->ns, evutil_socket_error_to_string(err));
2225		return 2;
2226	} else if (r != (int)req->request_len) {
2227		return 1;  /* short write */
2228	} else {
2229		return 0;
2230	}
2231}
2232
2233/* try to send a request, updating the fields of the request */
2234/* as needed */
2235/* */
2236/* return: */
2237/*   0 ok */
2238/*   1 failed */
2239static int
2240evdns_request_transmit(struct request *req) {
2241	int retcode = 0, r;
2242
2243	ASSERT_LOCKED(req->base);
2244	ASSERT_VALID_REQUEST(req);
2245	/* if we fail to send this packet then this flag marks it */
2246	/* for evdns_transmit */
2247	req->transmit_me = 1;
2248	EVUTIL_ASSERT(req->trans_id != 0xffff);
2249
2250	if (!req->ns)
2251	{
2252		/* unable to transmit request if no nameservers */
2253		return 1;
2254	}
2255
2256	if (req->ns->choked) {
2257		/* don't bother trying to write to a socket */
2258		/* which we have had EAGAIN from */
2259		return 1;
2260	}
2261
2262	r = evdns_request_transmit_to(req, req->ns);
2263	switch (r) {
2264	case 1:
2265		/* temp failure */
2266		req->ns->choked = 1;
2267		nameserver_write_waiting(req->ns, 1);
2268		return 1;
2269	case 2:
2270		/* failed to transmit the request entirely. */
2271		retcode = 1;
2272		/* fall through: we'll set a timeout, which will time out,
2273		 * and make us retransmit the request anyway. */
2274	default:
2275		/* all ok */
2276		log(EVDNS_LOG_DEBUG,
2277		    "Setting timeout for request %p, sent to nameserver %p", req, req->ns);
2278		if (evtimer_add(&req->timeout_event, &req->base->global_timeout) < 0) {
2279			log(EVDNS_LOG_WARN,
2280		      "Error from libevent when adding timer for request %p",
2281			    req);
2282			/* ???? Do more? */
2283		}
2284		req->tx_count++;
2285		req->transmit_me = 0;
2286		return retcode;
2287	}
2288}
2289
2290static void
2291nameserver_probe_callback(int result, char type, int count, int ttl, void *addresses, void *arg) {
2292	struct nameserver *const ns = (struct nameserver *) arg;
2293	(void) type;
2294	(void) count;
2295	(void) ttl;
2296	(void) addresses;
2297
2298	if (result == DNS_ERR_CANCEL) {
2299		/* We canceled this request because the nameserver came up
2300		 * for some other reason.  Do not change our opinion about
2301		 * the nameserver. */
2302		return;
2303	}
2304
2305	EVDNS_LOCK(ns->base);
2306	ns->probe_request = NULL;
2307	if (result == DNS_ERR_NONE || result == DNS_ERR_NOTEXIST) {
2308		/* this is a good reply */
2309		nameserver_up(ns);
2310	} else {
2311		nameserver_probe_failed(ns);
2312	}
2313	EVDNS_UNLOCK(ns->base);
2314}
2315
2316static void
2317nameserver_send_probe(struct nameserver *const ns) {
2318	struct evdns_request *handle;
2319	struct request *req;
2320	char addrbuf[128];
2321	/* here we need to send a probe to a given nameserver */
2322	/* in the hope that it is up now. */
2323
2324	ASSERT_LOCKED(ns->base);
2325	log(EVDNS_LOG_DEBUG, "Sending probe to %s",
2326	    evutil_format_sockaddr_port_(
2327		    (struct sockaddr *)&ns->address,
2328		    addrbuf, sizeof(addrbuf)));
2329	handle = mm_calloc(1, sizeof(*handle));
2330	if (!handle) return;
2331	req = request_new(ns->base, handle, TYPE_A, "google.com", DNS_QUERY_NO_SEARCH, nameserver_probe_callback, ns);
2332	if (!req) {
2333		mm_free(handle);
2334		return;
2335	}
2336	ns->probe_request = handle;
2337	/* we force this into the inflight queue no matter what */
2338	request_trans_id_set(req, transaction_id_pick(ns->base));
2339	req->ns = ns;
2340	request_submit(req);
2341}
2342
2343/* returns: */
2344/*   0 didn't try to transmit anything */
2345/*   1 tried to transmit something */
2346static int
2347evdns_transmit(struct evdns_base *base) {
2348	char did_try_to_transmit = 0;
2349	int i;
2350
2351	ASSERT_LOCKED(base);
2352	for (i = 0; i < base->n_req_heads; ++i) {
2353		if (base->req_heads[i]) {
2354			struct request *const started_at = base->req_heads[i], *req = started_at;
2355			/* first transmit all the requests which are currently waiting */
2356			do {
2357				if (req->transmit_me) {
2358					did_try_to_transmit = 1;
2359					evdns_request_transmit(req);
2360				}
2361
2362				req = req->next;
2363			} while (req != started_at);
2364		}
2365	}
2366
2367	return did_try_to_transmit;
2368}
2369
2370/* exported function */
2371int
2372evdns_base_count_nameservers(struct evdns_base *base)
2373{
2374	const struct nameserver *server;
2375	int n = 0;
2376
2377	EVDNS_LOCK(base);
2378	server = base->server_head;
2379	if (!server)
2380		goto done;
2381	do {
2382		++n;
2383		server = server->next;
2384	} while (server != base->server_head);
2385done:
2386	EVDNS_UNLOCK(base);
2387	return n;
2388}
2389
2390int
2391evdns_count_nameservers(void)
2392{
2393	return evdns_base_count_nameservers(current_base);
2394}
2395
2396/* exported function */
2397int
2398evdns_base_clear_nameservers_and_suspend(struct evdns_base *base)
2399{
2400	struct nameserver *server, *started_at;
2401	int i;
2402
2403	EVDNS_LOCK(base);
2404	server = base->server_head;
2405	started_at = base->server_head;
2406	if (!server) {
2407		EVDNS_UNLOCK(base);
2408		return 0;
2409	}
2410	while (1) {
2411		struct nameserver *next = server->next;
2412		(void) event_del(&server->event);
2413		if (evtimer_initialized(&server->timeout_event))
2414			(void) evtimer_del(&server->timeout_event);
2415		if (server->probe_request) {
2416			evdns_cancel_request(server->base, server->probe_request);
2417			server->probe_request = NULL;
2418		}
2419		if (server->socket >= 0)
2420			evutil_closesocket(server->socket);
2421		mm_free(server);
2422		if (next == started_at)
2423			break;
2424		server = next;
2425	}
2426	base->server_head = NULL;
2427	base->global_good_nameservers = 0;
2428
2429	for (i = 0; i < base->n_req_heads; ++i) {
2430		struct request *req, *req_started_at;
2431		req = req_started_at = base->req_heads[i];
2432		while (req) {
2433			struct request *next = req->next;
2434			req->tx_count = req->reissue_count = 0;
2435			req->ns = NULL;
2436			/* ???? What to do about searches? */
2437			(void) evtimer_del(&req->timeout_event);
2438			req->trans_id = 0;
2439			req->transmit_me = 0;
2440
2441			base->global_requests_waiting++;
2442			evdns_request_insert(req, &base->req_waiting_head);
2443			/* We want to insert these suspended elements at the front of
2444			 * the waiting queue, since they were pending before any of
2445			 * the waiting entries were added.  This is a circular list,
2446			 * so we can just shift the start back by one.*/
2447			base->req_waiting_head = base->req_waiting_head->prev;
2448
2449			if (next == req_started_at)
2450				break;
2451			req = next;
2452		}
2453		base->req_heads[i] = NULL;
2454	}
2455
2456	base->global_requests_inflight = 0;
2457
2458	EVDNS_UNLOCK(base);
2459	return 0;
2460}
2461
2462int
2463evdns_clear_nameservers_and_suspend(void)
2464{
2465	return evdns_base_clear_nameservers_and_suspend(current_base);
2466}
2467
2468
2469/* exported function */
2470int
2471evdns_base_resume(struct evdns_base *base)
2472{
2473	EVDNS_LOCK(base);
2474	evdns_requests_pump_waiting_queue(base);
2475	EVDNS_UNLOCK(base);
2476
2477	return 0;
2478}
2479
2480int
2481evdns_resume(void)
2482{
2483	return evdns_base_resume(current_base);
2484}
2485
2486static int
2487evdns_nameserver_add_impl_(struct evdns_base *base, const struct sockaddr *address, int addrlen) {
2488	/* first check to see if we already have this nameserver */
2489
2490	const struct nameserver *server = base->server_head, *const started_at = base->server_head;
2491	struct nameserver *ns;
2492	int err = 0;
2493	char addrbuf[128];
2494
2495	ASSERT_LOCKED(base);
2496	if (server) {
2497		do {
2498			if (!evutil_sockaddr_cmp((struct sockaddr*)&server->address, address, 1)) return 3;
2499			server = server->next;
2500		} while (server != started_at);
2501	}
2502	if (addrlen > (int)sizeof(ns->address)) {
2503		log(EVDNS_LOG_DEBUG, "Addrlen %d too long.", (int)addrlen);
2504		return 2;
2505	}
2506
2507	ns = (struct nameserver *) mm_malloc(sizeof(struct nameserver));
2508	if (!ns) return -1;
2509
2510	memset(ns, 0, sizeof(struct nameserver));
2511	ns->base = base;
2512
2513	evtimer_assign(&ns->timeout_event, ns->base->event_base, nameserver_prod_callback, ns);
2514
2515	ns->socket = evutil_socket_(address->sa_family,
2516	    SOCK_DGRAM|EVUTIL_SOCK_NONBLOCK|EVUTIL_SOCK_CLOEXEC, 0);
2517	if (ns->socket < 0) { err = 1; goto out1; }
2518
2519	if (base->global_outgoing_addrlen &&
2520	    !evutil_sockaddr_is_loopback_(address)) {
2521		if (bind(ns->socket,
2522			(struct sockaddr*)&base->global_outgoing_address,
2523			base->global_outgoing_addrlen) < 0) {
2524			log(EVDNS_LOG_WARN,"Couldn't bind to outgoing address");
2525			err = 2;
2526			goto out2;
2527		}
2528	}
2529
2530	memcpy(&ns->address, address, addrlen);
2531	ns->addrlen = addrlen;
2532	ns->state = 1;
2533	event_assign(&ns->event, ns->base->event_base, ns->socket,
2534				 EV_READ | EV_PERSIST, nameserver_ready_callback, ns);
2535	if (!base->disable_when_inactive && event_add(&ns->event, NULL) < 0) {
2536		err = 2;
2537		goto out2;
2538	}
2539
2540	log(EVDNS_LOG_DEBUG, "Added nameserver %s as %p",
2541	    evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), ns);
2542
2543	/* insert this nameserver into the list of them */
2544	if (!base->server_head) {
2545		ns->next = ns->prev = ns;
2546		base->server_head = ns;
2547	} else {
2548		ns->next = base->server_head->next;
2549		ns->prev = base->server_head;
2550		base->server_head->next = ns;
2551		ns->next->prev = ns;
2552	}
2553
2554	base->global_good_nameservers++;
2555
2556	return 0;
2557
2558out2:
2559	evutil_closesocket(ns->socket);
2560out1:
2561	event_debug_unassign(&ns->event);
2562	mm_free(ns);
2563	log(EVDNS_LOG_WARN, "Unable to add nameserver %s: error %d",
2564	    evutil_format_sockaddr_port_(address, addrbuf, sizeof(addrbuf)), err);
2565	return err;
2566}
2567
2568/* exported function */
2569int
2570evdns_base_nameserver_add(struct evdns_base *base, unsigned long int address)
2571{
2572	struct sockaddr_in sin;
2573	int res;
2574	memset(&sin, 0, sizeof(sin));
2575	sin.sin_addr.s_addr = address;
2576	sin.sin_port = htons(53);
2577	sin.sin_family = AF_INET;
2578	EVDNS_LOCK(base);
2579	res = evdns_nameserver_add_impl_(base, (struct sockaddr*)&sin, sizeof(sin));
2580	EVDNS_UNLOCK(base);
2581	return res;
2582}
2583
2584int
2585evdns_nameserver_add(unsigned long int address) {
2586	if (!current_base)
2587		current_base = evdns_base_new(NULL, 0);
2588	return evdns_base_nameserver_add(current_base, address);
2589}
2590
2591static void
2592sockaddr_setport(struct sockaddr *sa, ev_uint16_t port)
2593{
2594	if (sa->sa_family == AF_INET) {
2595		((struct sockaddr_in *)sa)->sin_port = htons(port);
2596	} else if (sa->sa_family == AF_INET6) {
2597		((struct sockaddr_in6 *)sa)->sin6_port = htons(port);
2598	}
2599}
2600
2601static ev_uint16_t
2602sockaddr_getport(struct sockaddr *sa)
2603{
2604	if (sa->sa_family == AF_INET) {
2605		return ntohs(((struct sockaddr_in *)sa)->sin_port);
2606	} else if (sa->sa_family == AF_INET6) {
2607		return ntohs(((struct sockaddr_in6 *)sa)->sin6_port);
2608	} else {
2609		return 0;
2610	}
2611}
2612
2613/* exported function */
2614int
2615evdns_base_nameserver_ip_add(struct evdns_base *base, const char *ip_as_string) {
2616	struct sockaddr_storage ss;
2617	struct sockaddr *sa;
2618	int len = sizeof(ss);
2619	int res;
2620	if (evutil_parse_sockaddr_port(ip_as_string, (struct sockaddr *)&ss,
2621		&len)) {
2622		log(EVDNS_LOG_WARN, "Unable to parse nameserver address %s",
2623			ip_as_string);
2624		return 4;
2625	}
2626	sa = (struct sockaddr *) &ss;
2627	if (sockaddr_getport(sa) == 0)
2628		sockaddr_setport(sa, 53);
2629
2630	EVDNS_LOCK(base);
2631	res = evdns_nameserver_add_impl_(base, sa, len);
2632	EVDNS_UNLOCK(base);
2633	return res;
2634}
2635
2636int
2637evdns_nameserver_ip_add(const char *ip_as_string) {
2638	if (!current_base)
2639		current_base = evdns_base_new(NULL, 0);
2640	return evdns_base_nameserver_ip_add(current_base, ip_as_string);
2641}
2642
2643int
2644evdns_base_nameserver_sockaddr_add(struct evdns_base *base,
2645    const struct sockaddr *sa, ev_socklen_t len, unsigned flags)
2646{
2647	int res;
2648	EVUTIL_ASSERT(base);
2649	EVDNS_LOCK(base);
2650	res = evdns_nameserver_add_impl_(base, sa, len);
2651	EVDNS_UNLOCK(base);
2652	return res;
2653}
2654
2655int
2656evdns_base_get_nameserver_addr(struct evdns_base *base, int idx,
2657    struct sockaddr *sa, ev_socklen_t len)
2658{
2659	int result = -1;
2660	int i;
2661	struct nameserver *server;
2662	EVDNS_LOCK(base);
2663	server = base->server_head;
2664	for (i = 0; i < idx && server; ++i, server = server->next) {
2665		if (server->next == base->server_head)
2666			goto done;
2667	}
2668	if (! server)
2669		goto done;
2670
2671	if (server->addrlen > len) {
2672		result = (int) server->addrlen;
2673		goto done;
2674	}
2675
2676	memcpy(sa, &server->address, server->addrlen);
2677	result = (int) server->addrlen;
2678done:
2679	EVDNS_UNLOCK(base);
2680	return result;
2681}
2682
2683/* remove from the queue */
2684static void
2685evdns_request_remove(struct request *req, struct request **head)
2686{
2687	ASSERT_LOCKED(req->base);
2688	ASSERT_VALID_REQUEST(req);
2689
2690#if 0
2691	{
2692		struct request *ptr;
2693		int found = 0;
2694		EVUTIL_ASSERT(*head != NULL);
2695
2696		ptr = *head;
2697		do {
2698			if (ptr == req) {
2699				found = 1;
2700				break;
2701			}
2702			ptr = ptr->next;
2703		} while (ptr != *head);
2704		EVUTIL_ASSERT(found);
2705
2706		EVUTIL_ASSERT(req->next);
2707	}
2708#endif
2709
2710	if (req->next == req) {
2711		/* only item in the list */
2712		*head = NULL;
2713	} else {
2714		req->next->prev = req->prev;
2715		req->prev->next = req->next;
2716		if (*head == req) *head = req->next;
2717	}
2718	req->next = req->prev = NULL;
2719}
2720
2721/* insert into the tail of the queue */
2722static void
2723evdns_request_insert(struct request *req, struct request **head) {
2724	ASSERT_LOCKED(req->base);
2725	ASSERT_VALID_REQUEST(req);
2726	if (!*head) {
2727		*head = req;
2728		req->next = req->prev = req;
2729		return;
2730	}
2731
2732	req->prev = (*head)->prev;
2733	req->prev->next = req;
2734	req->next = *head;
2735	(*head)->prev = req;
2736}
2737
2738static int
2739string_num_dots(const char *s) {
2740	int count = 0;
2741	while ((s = strchr(s, '.'))) {
2742		s++;
2743		count++;
2744	}
2745	return count;
2746}
2747
2748static struct request *
2749request_new(struct evdns_base *base, struct evdns_request *handle, int type,
2750	    const char *name, int flags, evdns_callback_type callback,
2751	    void *user_ptr) {
2752
2753	const char issuing_now =
2754	    (base->global_requests_inflight < base->global_max_requests_inflight) ? 1 : 0;
2755
2756	const size_t name_len = strlen(name);
2757	const size_t request_max_len = evdns_request_len(name_len);
2758	const u16 trans_id = issuing_now ? transaction_id_pick(base) : 0xffff;
2759	/* the request data is alloced in a single block with the header */
2760	struct request *const req =
2761	    mm_malloc(sizeof(struct request) + request_max_len);
2762	int rlen;
2763	char namebuf[256];
2764	(void) flags;
2765
2766	ASSERT_LOCKED(base);
2767
2768	if (!req) return NULL;
2769
2770	if (name_len >= sizeof(namebuf)) {
2771		mm_free(req);
2772		return NULL;
2773	}
2774
2775	memset(req, 0, sizeof(struct request));
2776	req->base = base;
2777
2778	evtimer_assign(&req->timeout_event, req->base->event_base, evdns_request_timeout_callback, req);
2779
2780	if (base->global_randomize_case) {
2781		unsigned i;
2782		char randbits[(sizeof(namebuf)+7)/8];
2783		strlcpy(namebuf, name, sizeof(namebuf));
2784		evutil_secure_rng_get_bytes(randbits, (name_len+7)/8);
2785		for (i = 0; i < name_len; ++i) {
2786			if (EVUTIL_ISALPHA_(namebuf[i])) {
2787				if ((randbits[i >> 3] & (1<<(i & 7))))
2788					namebuf[i] |= 0x20;
2789				else
2790					namebuf[i] &= ~0x20;
2791			}
2792		}
2793		name = namebuf;
2794	}
2795
2796	/* request data lives just after the header */
2797	req->request = ((u8 *) req) + sizeof(struct request);
2798	/* denotes that the request data shouldn't be free()ed */
2799	req->request_appended = 1;
2800	rlen = evdns_request_data_build(name, name_len, trans_id,
2801	    type, CLASS_INET, req->request, request_max_len);
2802	if (rlen < 0)
2803		goto err1;
2804
2805	req->request_len = rlen;
2806	req->trans_id = trans_id;
2807	req->tx_count = 0;
2808	req->request_type = type;
2809	req->user_pointer = user_ptr;
2810	req->user_callback = callback;
2811	req->ns = issuing_now ? nameserver_pick(base) : NULL;
2812	req->next = req->prev = NULL;
2813	req->handle = handle;
2814	if (handle) {
2815		handle->current_req = req;
2816		handle->base = base;
2817	}
2818
2819	return req;
2820err1:
2821	mm_free(req);
2822	return NULL;
2823}
2824
2825static void
2826request_submit(struct request *const req) {
2827	struct evdns_base *base = req->base;
2828	ASSERT_LOCKED(base);
2829	ASSERT_VALID_REQUEST(req);
2830	if (req->ns) {
2831		/* if it has a nameserver assigned then this is going */
2832		/* straight into the inflight queue */
2833		evdns_request_insert(req, &REQ_HEAD(base, req->trans_id));
2834
2835		base->global_requests_inflight++;
2836		req->ns->requests_inflight++;
2837
2838		evdns_request_transmit(req);
2839	} else {
2840		evdns_request_insert(req, &base->req_waiting_head);
2841		base->global_requests_waiting++;
2842	}
2843}
2844
2845/* exported function */
2846void
2847evdns_cancel_request(struct evdns_base *base, struct evdns_request *handle)
2848{
2849	struct request *req;
2850
2851	if (!handle->current_req)
2852		return;
2853
2854	if (!base) {
2855		/* This redundancy is silly; can we fix it? (Not for 2.0) XXXX */
2856		base = handle->base;
2857		if (!base)
2858			base = handle->current_req->base;
2859	}
2860
2861	EVDNS_LOCK(base);
2862	if (handle->pending_cb) {
2863		EVDNS_UNLOCK(base);
2864		return;
2865	}
2866
2867	req = handle->current_req;
2868	ASSERT_VALID_REQUEST(req);
2869
2870	reply_schedule_callback(req, 0, DNS_ERR_CANCEL, NULL);
2871	if (req->ns) {
2872		/* remove from inflight queue */
2873		request_finished(req, &REQ_HEAD(base, req->trans_id), 1);
2874	} else {
2875		/* remove from global_waiting head */
2876		request_finished(req, &base->req_waiting_head, 1);
2877	}
2878	EVDNS_UNLOCK(base);
2879}
2880
2881/* exported function */
2882struct evdns_request *
2883evdns_base_resolve_ipv4(struct evdns_base *base, const char *name, int flags,
2884    evdns_callback_type callback, void *ptr) {
2885	struct evdns_request *handle;
2886	struct request *req;
2887	log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
2888	handle = mm_calloc(1, sizeof(*handle));
2889	if (handle == NULL)
2890		return NULL;
2891	EVDNS_LOCK(base);
2892	if (flags & DNS_QUERY_NO_SEARCH) {
2893		req =
2894			request_new(base, handle, TYPE_A, name, flags,
2895				    callback, ptr);
2896		if (req)
2897			request_submit(req);
2898	} else {
2899		search_request_new(base, handle, TYPE_A, name, flags,
2900		    callback, ptr);
2901	}
2902	if (handle->current_req == NULL) {
2903		mm_free(handle);
2904		handle = NULL;
2905	}
2906	EVDNS_UNLOCK(base);
2907	return handle;
2908}
2909
2910int evdns_resolve_ipv4(const char *name, int flags,
2911					   evdns_callback_type callback, void *ptr)
2912{
2913	return evdns_base_resolve_ipv4(current_base, name, flags, callback, ptr)
2914		? 0 : -1;
2915}
2916
2917
2918/* exported function */
2919struct evdns_request *
2920evdns_base_resolve_ipv6(struct evdns_base *base,
2921    const char *name, int flags,
2922    evdns_callback_type callback, void *ptr)
2923{
2924	struct evdns_request *handle;
2925	struct request *req;
2926	log(EVDNS_LOG_DEBUG, "Resolve requested for %s", name);
2927	handle = mm_calloc(1, sizeof(*handle));
2928	if (handle == NULL)
2929		return NULL;
2930	EVDNS_LOCK(base);
2931	if (flags & DNS_QUERY_NO_SEARCH) {
2932		req = request_new(base, handle, TYPE_AAAA, name, flags,
2933				  callback, ptr);
2934		if (req)
2935			request_submit(req);
2936	} else {
2937		search_request_new(base, handle, TYPE_AAAA, name, flags,
2938		    callback, ptr);
2939	}
2940	if (handle->current_req == NULL) {
2941		mm_free(handle);
2942		handle = NULL;
2943	}
2944	EVDNS_UNLOCK(base);
2945	return handle;
2946}
2947
2948int evdns_resolve_ipv6(const char *name, int flags,
2949    evdns_callback_type callback, void *ptr) {
2950	return evdns_base_resolve_ipv6(current_base, name, flags, callback, ptr)
2951		? 0 : -1;
2952}
2953
2954struct evdns_request *
2955evdns_base_resolve_reverse(struct evdns_base *base, const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) {
2956	char buf[32];
2957	struct evdns_request *handle;
2958	struct request *req;
2959	u32 a;
2960	EVUTIL_ASSERT(in);
2961	a = ntohl(in->s_addr);
2962	evutil_snprintf(buf, sizeof(buf), "%d.%d.%d.%d.in-addr.arpa",
2963			(int)(u8)((a	)&0xff),
2964			(int)(u8)((a>>8 )&0xff),
2965			(int)(u8)((a>>16)&0xff),
2966			(int)(u8)((a>>24)&0xff));
2967	handle = mm_calloc(1, sizeof(*handle));
2968	if (handle == NULL)
2969		return NULL;
2970	log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
2971	EVDNS_LOCK(base);
2972	req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr);
2973	if (req)
2974		request_submit(req);
2975	if (handle->current_req == NULL) {
2976		mm_free(handle);
2977		handle = NULL;
2978	}
2979	EVDNS_UNLOCK(base);
2980	return (handle);
2981}
2982
2983int evdns_resolve_reverse(const struct in_addr *in, int flags, evdns_callback_type callback, void *ptr) {
2984	return evdns_base_resolve_reverse(current_base, in, flags, callback, ptr)
2985		? 0 : -1;
2986}
2987
2988struct evdns_request *
2989evdns_base_resolve_reverse_ipv6(struct evdns_base *base, const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) {
2990	/* 32 nybbles, 32 periods, "ip6.arpa", NUL. */
2991	char buf[73];
2992	char *cp;
2993	struct evdns_request *handle;
2994	struct request *req;
2995	int i;
2996	EVUTIL_ASSERT(in);
2997	cp = buf;
2998	for (i=15; i >= 0; --i) {
2999		u8 byte = in->s6_addr[i];
3000		*cp++ = "0123456789abcdef"[byte & 0x0f];
3001		*cp++ = '.';
3002		*cp++ = "0123456789abcdef"[byte >> 4];
3003		*cp++ = '.';
3004	}
3005	EVUTIL_ASSERT(cp + strlen("ip6.arpa") < buf+sizeof(buf));
3006	memcpy(cp, "ip6.arpa", strlen("ip6.arpa")+1);
3007	handle = mm_calloc(1, sizeof(*handle));
3008	if (handle == NULL)
3009		return NULL;
3010	log(EVDNS_LOG_DEBUG, "Resolve requested for %s (reverse)", buf);
3011	EVDNS_LOCK(base);
3012	req = request_new(base, handle, TYPE_PTR, buf, flags, callback, ptr);
3013	if (req)
3014		request_submit(req);
3015	if (handle->current_req == NULL) {
3016		mm_free(handle);
3017		handle = NULL;
3018	}
3019	EVDNS_UNLOCK(base);
3020	return (handle);
3021}
3022
3023int evdns_resolve_reverse_ipv6(const struct in6_addr *in, int flags, evdns_callback_type callback, void *ptr) {
3024	return evdns_base_resolve_reverse_ipv6(current_base, in, flags, callback, ptr)
3025		? 0 : -1;
3026}
3027
3028/* ================================================================= */
3029/* Search support */
3030/* */
3031/* the libc resolver has support for searching a number of domains */
3032/* to find a name. If nothing else then it takes the single domain */
3033/* from the gethostname() call. */
3034/* */
3035/* It can also be configured via the domain and search options in a */
3036/* resolv.conf. */
3037/* */
3038/* The ndots option controls how many dots it takes for the resolver */
3039/* to decide that a name is non-local and so try a raw lookup first. */
3040
3041struct search_domain {
3042	int len;
3043	struct search_domain *next;
3044	/* the text string is appended to this structure */
3045};
3046
3047struct search_state {
3048	int refcount;
3049	int ndots;
3050	int num_domains;
3051	struct search_domain *head;
3052};
3053
3054static void
3055search_state_decref(struct search_state *const state) {
3056	if (!state) return;
3057	state->refcount--;
3058	if (!state->refcount) {
3059		struct search_domain *next, *dom;
3060		for (dom = state->head; dom; dom = next) {
3061			next = dom->next;
3062			mm_free(dom);
3063		}
3064		mm_free(state);
3065	}
3066}
3067
3068static struct search_state *
3069search_state_new(void) {
3070	struct search_state *state = (struct search_state *) mm_malloc(sizeof(struct search_state));
3071	if (!state) return NULL;
3072	memset(state, 0, sizeof(struct search_state));
3073	state->refcount = 1;
3074	state->ndots = 1;
3075
3076	return state;
3077}
3078
3079static void
3080search_postfix_clear(struct evdns_base *base) {
3081	search_state_decref(base->global_search_state);
3082
3083	base->global_search_state = search_state_new();
3084}
3085
3086/* exported function */
3087void
3088evdns_base_search_clear(struct evdns_base *base)
3089{
3090	EVDNS_LOCK(base);
3091	search_postfix_clear(base);
3092	EVDNS_UNLOCK(base);
3093}
3094
3095void
3096evdns_search_clear(void) {
3097	evdns_base_search_clear(current_base);
3098}
3099
3100static void
3101search_postfix_add(struct evdns_base *base, const char *domain) {
3102	size_t domain_len;
3103	struct search_domain *sdomain;
3104	while (domain[0] == '.') domain++;
3105	domain_len = strlen(domain);
3106
3107	ASSERT_LOCKED(base);
3108	if (!base->global_search_state) base->global_search_state = search_state_new();
3109	if (!base->global_search_state) return;
3110	base->global_search_state->num_domains++;
3111
3112	sdomain = (struct search_domain *) mm_malloc(sizeof(struct search_domain) + domain_len);
3113	if (!sdomain) return;
3114	memcpy( ((u8 *) sdomain) + sizeof(struct search_domain), domain, domain_len);
3115	sdomain->next = base->global_search_state->head;
3116	sdomain->len = (int) domain_len;
3117
3118	base->global_search_state->head = sdomain;
3119}
3120
3121/* reverse the order of members in the postfix list. This is needed because, */
3122/* when parsing resolv.conf we push elements in the wrong order */
3123static void
3124search_reverse(struct evdns_base *base) {
3125	struct search_domain *cur, *prev = NULL, *next;
3126	ASSERT_LOCKED(base);
3127	cur = base->global_search_state->head;
3128	while (cur) {
3129		next = cur->next;
3130		cur->next = prev;
3131		prev = cur;
3132		cur = next;
3133	}
3134
3135	base->global_search_state->head = prev;
3136}
3137
3138/* exported function */
3139void
3140evdns_base_search_add(struct evdns_base *base, const char *domain) {
3141	EVDNS_LOCK(base);
3142	search_postfix_add(base, domain);
3143	EVDNS_UNLOCK(base);
3144}
3145void
3146evdns_search_add(const char *domain) {
3147	evdns_base_search_add(current_base, domain);
3148}
3149
3150/* exported function */
3151void
3152evdns_base_search_ndots_set(struct evdns_base *base, const int ndots) {
3153	EVDNS_LOCK(base);
3154	if (!base->global_search_state) base->global_search_state = search_state_new();
3155	if (base->global_search_state)
3156		base->global_search_state->ndots = ndots;
3157	EVDNS_UNLOCK(base);
3158}
3159void
3160evdns_search_ndots_set(const int ndots) {
3161	evdns_base_search_ndots_set(current_base, ndots);
3162}
3163
3164static void
3165search_set_from_hostname(struct evdns_base *base) {
3166	char hostname[HOST_NAME_MAX + 1], *domainname;
3167
3168	ASSERT_LOCKED(base);
3169	search_postfix_clear(base);
3170	if (gethostname(hostname, sizeof(hostname))) return;
3171	domainname = strchr(hostname, '.');
3172	if (!domainname) return;
3173	search_postfix_add(base, domainname);
3174}
3175
3176/* warning: returns malloced string */
3177static char *
3178search_make_new(const struct search_state *const state, int n, const char *const base_name) {
3179	const size_t base_len = strlen(base_name);
3180	const char need_to_append_dot = base_name[base_len - 1] == '.' ? 0 : 1;
3181	struct search_domain *dom;
3182
3183	for (dom = state->head; dom; dom = dom->next) {
3184		if (!n--) {
3185			/* this is the postfix we want */
3186			/* the actual postfix string is kept at the end of the structure */
3187			const u8 *const postfix = ((u8 *) dom) + sizeof(struct search_domain);
3188			const int postfix_len = dom->len;
3189			char *const newname = (char *) mm_malloc(base_len + need_to_append_dot + postfix_len + 1);
3190			if (!newname) return NULL;
3191			memcpy(newname, base_name, base_len);
3192			if (need_to_append_dot) newname[base_len] = '.';
3193			memcpy(newname + base_len + need_to_append_dot, postfix, postfix_len);
3194			newname[base_len + need_to_append_dot + postfix_len] = 0;
3195			return newname;
3196		}
3197	}
3198
3199	/* we ran off the end of the list and still didn't find the requested string */
3200	EVUTIL_ASSERT(0);
3201	return NULL; /* unreachable; stops warnings in some compilers. */
3202}
3203
3204static struct request *
3205search_request_new(struct evdns_base *base, struct evdns_request *handle,
3206		   int type, const char *const name, int flags,
3207		   evdns_callback_type user_callback, void *user_arg) {
3208	ASSERT_LOCKED(base);
3209	EVUTIL_ASSERT(type == TYPE_A || type == TYPE_AAAA);
3210	EVUTIL_ASSERT(handle->current_req == NULL);
3211	if ( ((flags & DNS_QUERY_NO_SEARCH) == 0) &&
3212	     base->global_search_state &&
3213		 base->global_search_state->num_domains) {
3214		/* we have some domains to search */
3215		struct request *req;
3216		if (string_num_dots(name) >= base->global_search_state->ndots) {
3217			req = request_new(base, handle, type, name, flags, user_callback, user_arg);
3218			if (!req) return NULL;
3219			handle->search_index = -1;
3220		} else {
3221			char *const new_name = search_make_new(base->global_search_state, 0, name);
3222			if (!new_name) return NULL;
3223			req = request_new(base, handle, type, new_name, flags, user_callback, user_arg);
3224			mm_free(new_name);
3225			if (!req) return NULL;
3226			handle->search_index = 0;
3227		}
3228		EVUTIL_ASSERT(handle->search_origname == NULL);
3229		handle->search_origname = mm_strdup(name);
3230		if (handle->search_origname == NULL) {
3231			/* XXX Should we dealloc req? If yes, how? */
3232			if (req)
3233				mm_free(req);
3234			return NULL;
3235		}
3236		handle->search_state = base->global_search_state;
3237		handle->search_flags = flags;
3238		base->global_search_state->refcount++;
3239		request_submit(req);
3240		return req;
3241	} else {
3242		struct request *const req = request_new(base, handle, type, name, flags, user_callback, user_arg);
3243		if (!req) return NULL;
3244		request_submit(req);
3245		return req;
3246	}
3247}
3248
3249/* this is called when a request has failed to find a name. We need to check */
3250/* if it is part of a search and, if so, try the next name in the list */
3251/* returns: */
3252/*   0 another request has been submitted */
3253/*   1 no more requests needed */
3254static int
3255search_try_next(struct evdns_request *const handle) {
3256	struct request *req = handle->current_req;
3257	struct evdns_base *base = req->base;
3258	struct request *newreq;
3259	ASSERT_LOCKED(base);
3260	if (handle->search_state) {
3261		/* it is part of a search */
3262		char *new_name;
3263		handle->search_index++;
3264		if (handle->search_index >= handle->search_state->num_domains) {
3265			/* no more postfixes to try, however we may need to try */
3266			/* this name without a postfix */
3267			if (string_num_dots(handle->search_origname) < handle->search_state->ndots) {
3268				/* yep, we need to try it raw */
3269				newreq = request_new(base, NULL, req->request_type, handle->search_origname, handle->search_flags, req->user_callback, req->user_pointer);
3270				log(EVDNS_LOG_DEBUG, "Search: trying raw query %s", handle->search_origname);
3271				if (newreq) {
3272					search_request_finished(handle);
3273					goto submit_next;
3274				}
3275			}
3276			return 1;
3277		}
3278
3279		new_name = search_make_new(handle->search_state, handle->search_index, handle->search_origname);
3280		if (!new_name) return 1;
3281		log(EVDNS_LOG_DEBUG, "Search: now trying %s (%d)", new_name, handle->search_index);
3282		newreq = request_new(base, NULL, req->request_type, new_name, handle->search_flags, req->user_callback, req->user_pointer);
3283		mm_free(new_name);
3284		if (!newreq) return 1;
3285		goto submit_next;
3286	}
3287	return 1;
3288
3289submit_next:
3290	request_finished(req, &REQ_HEAD(req->base, req->trans_id), 0);
3291	handle->current_req = newreq;
3292	newreq->handle = handle;
3293	request_submit(newreq);
3294	return 0;
3295}
3296
3297static void
3298search_request_finished(struct evdns_request *const handle) {
3299	ASSERT_LOCKED(handle->current_req->base);
3300	if (handle->search_state) {
3301		search_state_decref(handle->search_state);
3302		handle->search_state = NULL;
3303	}
3304	if (handle->search_origname) {
3305		mm_free(handle->search_origname);
3306		handle->search_origname = NULL;
3307	}
3308}
3309
3310/* ================================================================= */
3311/* Parsing resolv.conf files */
3312
3313static void
3314evdns_resolv_set_defaults(struct evdns_base *base, int flags) {
3315	/* if the file isn't found then we assume a local resolver */
3316	ASSERT_LOCKED(base);
3317	if (flags & DNS_OPTION_SEARCH) search_set_from_hostname(base);
3318	if (flags & DNS_OPTION_NAMESERVERS) evdns_base_nameserver_ip_add(base,"127.0.0.1");
3319}
3320
3321#ifndef EVENT__HAVE_STRTOK_R
3322static char *
3323strtok_r(char *s, const char *delim, char **state) {
3324	char *cp, *start;
3325	start = cp = s ? s : *state;
3326	if (!cp)
3327		return NULL;
3328	while (*cp && !strchr(delim, *cp))
3329		++cp;
3330	if (!*cp) {
3331		if (cp == start)
3332			return NULL;
3333		*state = NULL;
3334		return start;
3335	} else {
3336		*cp++ = '\0';
3337		*state = cp;
3338		return start;
3339	}
3340}
3341#endif
3342
3343/* helper version of atoi which returns -1 on error */
3344static int
3345strtoint(const char *const str)
3346{
3347	char *endptr;
3348	const int r = strtol(str, &endptr, 10);
3349	if (*endptr) return -1;
3350	return r;
3351}
3352
3353/* Parse a number of seconds into a timeval; return -1 on error. */
3354static int
3355evdns_strtotimeval(const char *const str, struct timeval *out)
3356{
3357	double d;
3358	char *endptr;
3359	d = strtod(str, &endptr);
3360	if (*endptr) return -1;
3361	if (d < 0) return -1;
3362	out->tv_sec = (int) d;
3363	out->tv_usec = (int) ((d - (int) d)*1000000);
3364	if (out->tv_sec == 0 && out->tv_usec < 1000) /* less than 1 msec */
3365		return -1;
3366	return 0;
3367}
3368
3369/* helper version of atoi that returns -1 on error and clips to bounds. */
3370static int
3371strtoint_clipped(const char *const str, int min, int max)
3372{
3373	int r = strtoint(str);
3374	if (r == -1)
3375		return r;
3376	else if (r<min)
3377		return min;
3378	else if (r>max)
3379		return max;
3380	else
3381		return r;
3382}
3383
3384static int
3385evdns_base_set_max_requests_inflight(struct evdns_base *base, int maxinflight)
3386{
3387	int old_n_heads = base->n_req_heads, n_heads;
3388	struct request **old_heads = base->req_heads, **new_heads, *req;
3389	int i;
3390
3391	ASSERT_LOCKED(base);
3392	if (maxinflight < 1)
3393		maxinflight = 1;
3394	n_heads = (maxinflight+4) / 5;
3395	EVUTIL_ASSERT(n_heads > 0);
3396	new_heads = mm_calloc(n_heads, sizeof(struct request*));
3397	if (!new_heads)
3398		return (-1);
3399	if (old_heads) {
3400		for (i = 0; i < old_n_heads; ++i) {
3401			while (old_heads[i]) {
3402				req = old_heads[i];
3403				evdns_request_remove(req, &old_heads[i]);
3404				evdns_request_insert(req, &new_heads[req->trans_id % n_heads]);
3405			}
3406		}
3407		mm_free(old_heads);
3408	}
3409	base->req_heads = new_heads;
3410	base->n_req_heads = n_heads;
3411	base->global_max_requests_inflight = maxinflight;
3412	return (0);
3413}
3414
3415/* exported function */
3416int
3417evdns_base_set_option(struct evdns_base *base,
3418    const char *option, const char *val)
3419{
3420	int res;
3421	EVDNS_LOCK(base);
3422	res = evdns_base_set_option_impl(base, option, val, DNS_OPTIONS_ALL);
3423	EVDNS_UNLOCK(base);
3424	return res;
3425}
3426
3427static inline int
3428str_matches_option(const char *s1, const char *optionname)
3429{
3430	/* Option names are given as "option:" We accept either 'option' in
3431	 * s1, or 'option:randomjunk'.  The latter form is to implement the
3432	 * resolv.conf parser. */
3433	size_t optlen = strlen(optionname);
3434	size_t slen = strlen(s1);
3435	if (slen == optlen || slen == optlen - 1)
3436		return !strncmp(s1, optionname, slen);
3437	else if (slen > optlen)
3438		return !strncmp(s1, optionname, optlen);
3439	else
3440		return 0;
3441}
3442
3443static int
3444evdns_base_set_option_impl(struct evdns_base *base,
3445    const char *option, const char *val, int flags)
3446{
3447	ASSERT_LOCKED(base);
3448	if (str_matches_option(option, "ndots:")) {
3449		const int ndots = strtoint(val);
3450		if (ndots == -1) return -1;
3451		if (!(flags & DNS_OPTION_SEARCH)) return 0;
3452		log(EVDNS_LOG_DEBUG, "Setting ndots to %d", ndots);
3453		if (!base->global_search_state) base->global_search_state = search_state_new();
3454		if (!base->global_search_state) return -1;
3455		base->global_search_state->ndots = ndots;
3456	} else if (str_matches_option(option, "timeout:")) {
3457		struct timeval tv;
3458		if (evdns_strtotimeval(val, &tv) == -1) return -1;
3459		if (!(flags & DNS_OPTION_MISC)) return 0;
3460		log(EVDNS_LOG_DEBUG, "Setting timeout to %s", val);
3461		memcpy(&base->global_timeout, &tv, sizeof(struct timeval));
3462	} else if (str_matches_option(option, "getaddrinfo-allow-skew:")) {
3463		struct timeval tv;
3464		if (evdns_strtotimeval(val, &tv) == -1) return -1;
3465		if (!(flags & DNS_OPTION_MISC)) return 0;
3466		log(EVDNS_LOG_DEBUG, "Setting getaddrinfo-allow-skew to %s",
3467		    val);
3468		memcpy(&base->global_getaddrinfo_allow_skew, &tv,
3469		    sizeof(struct timeval));
3470	} else if (str_matches_option(option, "max-timeouts:")) {
3471		const int maxtimeout = strtoint_clipped(val, 1, 255);
3472		if (maxtimeout == -1) return -1;
3473		if (!(flags & DNS_OPTION_MISC)) return 0;
3474		log(EVDNS_LOG_DEBUG, "Setting maximum allowed timeouts to %d",
3475			maxtimeout);
3476		base->global_max_nameserver_timeout = maxtimeout;
3477	} else if (str_matches_option(option, "max-inflight:")) {
3478		const int maxinflight = strtoint_clipped(val, 1, 65000);
3479		if (maxinflight == -1) return -1;
3480		if (!(flags & DNS_OPTION_MISC)) return 0;
3481		log(EVDNS_LOG_DEBUG, "Setting maximum inflight requests to %d",
3482			maxinflight);
3483		evdns_base_set_max_requests_inflight(base, maxinflight);
3484	} else if (str_matches_option(option, "attempts:")) {
3485		int retries = strtoint(val);
3486		if (retries == -1) return -1;
3487		if (retries > 255) retries = 255;
3488		if (!(flags & DNS_OPTION_MISC)) return 0;
3489		log(EVDNS_LOG_DEBUG, "Setting retries to %d", retries);
3490		base->global_max_retransmits = retries;
3491	} else if (str_matches_option(option, "randomize-case:")) {
3492		int randcase = strtoint(val);
3493		if (!(flags & DNS_OPTION_MISC)) return 0;
3494		base->global_randomize_case = randcase;
3495	} else if (str_matches_option(option, "bind-to:")) {
3496		/* XXX This only applies to successive nameservers, not
3497		 * to already-configured ones.	We might want to fix that. */
3498		int len = sizeof(base->global_outgoing_address);
3499		if (!(flags & DNS_OPTION_NAMESERVERS)) return 0;
3500		if (evutil_parse_sockaddr_port(val,
3501			(struct sockaddr*)&base->global_outgoing_address, &len))
3502			return -1;
3503		base->global_outgoing_addrlen = len;
3504	} else if (str_matches_option(option, "initial-probe-timeout:")) {
3505		struct timeval tv;
3506		if (evdns_strtotimeval(val, &tv) == -1) return -1;
3507		if (tv.tv_sec > 3600)
3508			tv.tv_sec = 3600;
3509		if (!(flags & DNS_OPTION_MISC)) return 0;
3510		log(EVDNS_LOG_DEBUG, "Setting initial probe timeout to %s",
3511		    val);
3512		memcpy(&base->global_nameserver_probe_initial_timeout, &tv,
3513		    sizeof(tv));
3514	}
3515	return 0;
3516}
3517
3518int
3519evdns_set_option(const char *option, const char *val, int flags)
3520{
3521	if (!current_base)
3522		current_base = evdns_base_new(NULL, 0);
3523	return evdns_base_set_option(current_base, option, val);
3524}
3525
3526static void
3527resolv_conf_parse_line(struct evdns_base *base, char *const start, int flags) {
3528	char *strtok_state;
3529	static const char *const delims = " \t";
3530#define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state)
3531
3532
3533	char *const first_token = strtok_r(start, delims, &strtok_state);
3534	ASSERT_LOCKED(base);
3535	if (!first_token) return;
3536
3537	if (!strcmp(first_token, "nameserver") && (flags & DNS_OPTION_NAMESERVERS)) {
3538		const char *const nameserver = NEXT_TOKEN;
3539
3540		if (nameserver)
3541			evdns_base_nameserver_ip_add(base, nameserver);
3542	} else if (!strcmp(first_token, "domain") && (flags & DNS_OPTION_SEARCH)) {
3543		const char *const domain = NEXT_TOKEN;
3544		if (domain) {
3545			search_postfix_clear(base);
3546			search_postfix_add(base, domain);
3547		}
3548	} else if (!strcmp(first_token, "search") && (flags & DNS_OPTION_SEARCH)) {
3549		const char *domain;
3550		search_postfix_clear(base);
3551
3552		while ((domain = NEXT_TOKEN)) {
3553			search_postfix_add(base, domain);
3554		}
3555		search_reverse(base);
3556	} else if (!strcmp(first_token, "options")) {
3557		const char *option;
3558		while ((option = NEXT_TOKEN)) {
3559			const char *val = strchr(option, ':');
3560			evdns_base_set_option_impl(base, option, val ? val+1 : "", flags);
3561		}
3562	}
3563#undef NEXT_TOKEN
3564}
3565
3566/* exported function */
3567/* returns: */
3568/*   0 no errors */
3569/*   1 failed to open file */
3570/*   2 failed to stat file */
3571/*   3 file too large */
3572/*   4 out of memory */
3573/*   5 short read from file */
3574int
3575evdns_base_resolv_conf_parse(struct evdns_base *base, int flags, const char *const filename) {
3576	int res;
3577	EVDNS_LOCK(base);
3578	res = evdns_base_resolv_conf_parse_impl(base, flags, filename);
3579	EVDNS_UNLOCK(base);
3580	return res;
3581}
3582
3583static char *
3584evdns_get_default_hosts_filename(void)
3585{
3586#ifdef _WIN32
3587	/* Windows is a little coy about where it puts its configuration
3588	 * files.  Sure, they're _usually_ in C:\windows\system32, but
3589	 * there's no reason in principle they couldn't be in
3590	 * W:\hoboken chicken emergency\
3591	 */
3592	char path[MAX_PATH+1];
3593	static const char hostfile[] = "\\drivers\\etc\\hosts";
3594	char *path_out;
3595	size_t len_out;
3596
3597	if (! SHGetSpecialFolderPathA(NULL, path, CSIDL_SYSTEM, 0))
3598		return NULL;
3599	len_out = strlen(path)+strlen(hostfile)+1;
3600	path_out = mm_malloc(len_out);
3601	evutil_snprintf(path_out, len_out, "%s%s", path, hostfile);
3602	return path_out;
3603#else
3604	return mm_strdup("/etc/hosts");
3605#endif
3606}
3607
3608static int
3609evdns_base_resolv_conf_parse_impl(struct evdns_base *base, int flags, const char *const filename) {
3610	size_t n;
3611	char *resolv;
3612	char *start;
3613	int err = 0;
3614
3615	log(EVDNS_LOG_DEBUG, "Parsing resolv.conf file %s", filename);
3616
3617	if (flags & DNS_OPTION_HOSTSFILE) {
3618		char *fname = evdns_get_default_hosts_filename();
3619		evdns_base_load_hosts(base, fname);
3620		if (fname)
3621			mm_free(fname);
3622	}
3623
3624	if ((err = evutil_read_file_(filename, &resolv, &n, 0)) < 0) {
3625		if (err == -1) {
3626			/* No file. */
3627			evdns_resolv_set_defaults(base, flags);
3628			return 1;
3629		} else {
3630			return 2;
3631		}
3632	}
3633
3634	start = resolv;
3635	for (;;) {
3636		char *const newline = strchr(start, '\n');
3637		if (!newline) {
3638			resolv_conf_parse_line(base, start, flags);
3639			break;
3640		} else {
3641			*newline = 0;
3642			resolv_conf_parse_line(base, start, flags);
3643			start = newline + 1;
3644		}
3645	}
3646
3647	if (!base->server_head && (flags & DNS_OPTION_NAMESERVERS)) {
3648		/* no nameservers were configured. */
3649		evdns_base_nameserver_ip_add(base, "127.0.0.1");
3650		err = 6;
3651	}
3652	if (flags & DNS_OPTION_SEARCH && (!base->global_search_state || base->global_search_state->num_domains == 0)) {
3653		search_set_from_hostname(base);
3654	}
3655
3656	mm_free(resolv);
3657	return err;
3658}
3659
3660int
3661evdns_resolv_conf_parse(int flags, const char *const filename) {
3662	if (!current_base)
3663		current_base = evdns_base_new(NULL, 0);
3664	return evdns_base_resolv_conf_parse(current_base, flags, filename);
3665}
3666
3667
3668#ifdef _WIN32
3669/* Add multiple nameservers from a space-or-comma-separated list. */
3670static int
3671evdns_nameserver_ip_add_line(struct evdns_base *base, const char *ips) {
3672	const char *addr;
3673	char *buf;
3674	int r;
3675	ASSERT_LOCKED(base);
3676	while (*ips) {
3677		while (isspace(*ips) || *ips == ',' || *ips == '\t')
3678			++ips;
3679		addr = ips;
3680		while (isdigit(*ips) || *ips == '.' || *ips == ':' ||
3681		    *ips=='[' || *ips==']')
3682			++ips;
3683		buf = mm_malloc(ips-addr+1);
3684		if (!buf) return 4;
3685		memcpy(buf, addr, ips-addr);
3686		buf[ips-addr] = '\0';
3687		r = evdns_base_nameserver_ip_add(base, buf);
3688		mm_free(buf);
3689		if (r) return r;
3690	}
3691	return 0;
3692}
3693
3694typedef DWORD(WINAPI *GetNetworkParams_fn_t)(FIXED_INFO *, DWORD*);
3695
3696/* Use the windows GetNetworkParams interface in iphlpapi.dll to */
3697/* figure out what our nameservers are. */
3698static int
3699load_nameservers_with_getnetworkparams(struct evdns_base *base)
3700{
3701	/* Based on MSDN examples and inspection of  c-ares code. */
3702	FIXED_INFO *fixed;
3703	HMODULE handle = 0;
3704	ULONG size = sizeof(FIXED_INFO);
3705	void *buf = NULL;
3706	int status = 0, r, added_any;
3707	IP_ADDR_STRING *ns;
3708	GetNetworkParams_fn_t fn;
3709
3710	ASSERT_LOCKED(base);
3711	if (!(handle = evutil_load_windows_system_library_(
3712			TEXT("iphlpapi.dll")))) {
3713		log(EVDNS_LOG_WARN, "Could not open iphlpapi.dll");
3714		status = -1;
3715		goto done;
3716	}
3717	if (!(fn = (GetNetworkParams_fn_t) GetProcAddress(handle, "GetNetworkParams"))) {
3718		log(EVDNS_LOG_WARN, "Could not get address of function.");
3719		status = -1;
3720		goto done;
3721	}
3722
3723	buf = mm_malloc(size);
3724	if (!buf) { status = 4; goto done; }
3725	fixed = buf;
3726	r = fn(fixed, &size);
3727	if (r != ERROR_SUCCESS && r != ERROR_BUFFER_OVERFLOW) {
3728		status = -1;
3729		goto done;
3730	}
3731	if (r != ERROR_SUCCESS) {
3732		mm_free(buf);
3733		buf = mm_malloc(size);
3734		if (!buf) { status = 4; goto done; }
3735		fixed = buf;
3736		r = fn(fixed, &size);
3737		if (r != ERROR_SUCCESS) {
3738			log(EVDNS_LOG_DEBUG, "fn() failed.");
3739			status = -1;
3740			goto done;
3741		}
3742	}
3743
3744	EVUTIL_ASSERT(fixed);
3745	added_any = 0;
3746	ns = &(fixed->DnsServerList);
3747	while (ns) {
3748		r = evdns_nameserver_ip_add_line(base, ns->IpAddress.String);
3749		if (r) {
3750			log(EVDNS_LOG_DEBUG,"Could not add nameserver %s to list,error: %d",
3751				(ns->IpAddress.String),(int)GetLastError());
3752			status = r;
3753		} else {
3754			++added_any;
3755			log(EVDNS_LOG_DEBUG,"Successfully added %s as nameserver",ns->IpAddress.String);
3756		}
3757
3758		ns = ns->Next;
3759	}
3760
3761	if (!added_any) {
3762		log(EVDNS_LOG_DEBUG, "No nameservers added.");
3763		if (status == 0)
3764			status = -1;
3765	} else {
3766		status = 0;
3767	}
3768
3769 done:
3770	if (buf)
3771		mm_free(buf);
3772	if (handle)
3773		FreeLibrary(handle);
3774	return status;
3775}
3776
3777static int
3778config_nameserver_from_reg_key(struct evdns_base *base, HKEY key, const TCHAR *subkey)
3779{
3780	char *buf;
3781	DWORD bufsz = 0, type = 0;
3782	int status = 0;
3783
3784	ASSERT_LOCKED(base);
3785	if (RegQueryValueEx(key, subkey, 0, &type, NULL, &bufsz)
3786	    != ERROR_MORE_DATA)
3787		return -1;
3788	if (!(buf = mm_malloc(bufsz)))
3789		return -1;
3790
3791	if (RegQueryValueEx(key, subkey, 0, &type, (LPBYTE)buf, &bufsz)
3792	    == ERROR_SUCCESS && bufsz > 1) {
3793		status = evdns_nameserver_ip_add_line(base,buf);
3794	}
3795
3796	mm_free(buf);
3797	return status;
3798}
3799
3800#define SERVICES_KEY TEXT("System\\CurrentControlSet\\Services\\")
3801#define WIN_NS_9X_KEY  SERVICES_KEY TEXT("VxD\\MSTCP")
3802#define WIN_NS_NT_KEY  SERVICES_KEY TEXT("Tcpip\\Parameters")
3803
3804static int
3805load_nameservers_from_registry(struct evdns_base *base)
3806{
3807	int found = 0;
3808	int r;
3809#define TRY(k, name) \
3810	if (!found && config_nameserver_from_reg_key(base,k,TEXT(name)) == 0) { \
3811		log(EVDNS_LOG_DEBUG,"Found nameservers in %s/%s",#k,name); \
3812		found = 1;						\
3813	} else if (!found) {						\
3814		log(EVDNS_LOG_DEBUG,"Didn't find nameservers in %s/%s", \
3815		    #k,#name);						\
3816	}
3817
3818	ASSERT_LOCKED(base);
3819
3820	if (((int)GetVersion()) > 0) { /* NT */
3821		HKEY nt_key = 0, interfaces_key = 0;
3822
3823		if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_NT_KEY, 0,
3824				 KEY_READ, &nt_key) != ERROR_SUCCESS) {
3825			log(EVDNS_LOG_DEBUG,"Couldn't open nt key, %d",(int)GetLastError());
3826			return -1;
3827		}
3828		r = RegOpenKeyEx(nt_key, TEXT("Interfaces"), 0,
3829			     KEY_QUERY_VALUE|KEY_ENUMERATE_SUB_KEYS,
3830			     &interfaces_key);
3831		if (r != ERROR_SUCCESS) {
3832			log(EVDNS_LOG_DEBUG,"Couldn't open interfaces key, %d",(int)GetLastError());
3833			return -1;
3834		}
3835		TRY(nt_key, "NameServer");
3836		TRY(nt_key, "DhcpNameServer");
3837		TRY(interfaces_key, "NameServer");
3838		TRY(interfaces_key, "DhcpNameServer");
3839		RegCloseKey(interfaces_key);
3840		RegCloseKey(nt_key);
3841	} else {
3842		HKEY win_key = 0;
3843		if (RegOpenKeyEx(HKEY_LOCAL_MACHINE, WIN_NS_9X_KEY, 0,
3844				 KEY_READ, &win_key) != ERROR_SUCCESS) {
3845			log(EVDNS_LOG_DEBUG, "Couldn't open registry key, %d", (int)GetLastError());
3846			return -1;
3847		}
3848		TRY(win_key, "NameServer");
3849		RegCloseKey(win_key);
3850	}
3851
3852	if (found == 0) {
3853		log(EVDNS_LOG_WARN,"Didn't find any nameservers.");
3854	}
3855
3856	return found ? 0 : -1;
3857#undef TRY
3858}
3859
3860int
3861evdns_base_config_windows_nameservers(struct evdns_base *base)
3862{
3863	int r;
3864	char *fname;
3865	if (base == NULL)
3866		base = current_base;
3867	if (base == NULL)
3868		return -1;
3869	EVDNS_LOCK(base);
3870	fname = evdns_get_default_hosts_filename();
3871	log(EVDNS_LOG_DEBUG, "Loading hosts entries from %s", fname);
3872	evdns_base_load_hosts(base, fname);
3873	if (fname)
3874		mm_free(fname);
3875
3876	if (load_nameservers_with_getnetworkparams(base) == 0) {
3877		EVDNS_UNLOCK(base);
3878		return 0;
3879	}
3880	r = load_nameservers_from_registry(base);
3881
3882	EVDNS_UNLOCK(base);
3883	return r;
3884}
3885
3886int
3887evdns_config_windows_nameservers(void)
3888{
3889	if (!current_base) {
3890		current_base = evdns_base_new(NULL, 1);
3891		return current_base == NULL ? -1 : 0;
3892	} else {
3893		return evdns_base_config_windows_nameservers(current_base);
3894	}
3895}
3896#endif
3897
3898struct evdns_base *
3899evdns_base_new(struct event_base *event_base, int flags)
3900{
3901	struct evdns_base *base;
3902
3903	if (evutil_secure_rng_init() < 0) {
3904		log(EVDNS_LOG_WARN, "Unable to seed random number generator; "
3905		    "DNS can't run.");
3906		return NULL;
3907	}
3908
3909	/* Give the evutil library a hook into its evdns-enabled
3910	 * functionality.  We can't just call evdns_getaddrinfo directly or
3911	 * else libevent-core will depend on libevent-extras. */
3912	evutil_set_evdns_getaddrinfo_fn_(evdns_getaddrinfo);
3913
3914	base = mm_malloc(sizeof(struct evdns_base));
3915	if (base == NULL)
3916		return (NULL);
3917	memset(base, 0, sizeof(struct evdns_base));
3918	base->req_waiting_head = NULL;
3919
3920	EVTHREAD_ALLOC_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
3921	EVDNS_LOCK(base);
3922
3923	/* Set max requests inflight and allocate req_heads. */
3924	base->req_heads = NULL;
3925
3926	evdns_base_set_max_requests_inflight(base, 64);
3927
3928	base->server_head = NULL;
3929	base->event_base = event_base;
3930	base->global_good_nameservers = base->global_requests_inflight =
3931		base->global_requests_waiting = 0;
3932
3933	base->global_timeout.tv_sec = 5;
3934	base->global_timeout.tv_usec = 0;
3935	base->global_max_reissues = 1;
3936	base->global_max_retransmits = 3;
3937	base->global_max_nameserver_timeout = 3;
3938	base->global_search_state = NULL;
3939	base->global_randomize_case = 1;
3940	base->global_getaddrinfo_allow_skew.tv_sec = 3;
3941	base->global_getaddrinfo_allow_skew.tv_usec = 0;
3942	base->global_nameserver_probe_initial_timeout.tv_sec = 10;
3943	base->global_nameserver_probe_initial_timeout.tv_usec = 0;
3944
3945	TAILQ_INIT(&base->hostsdb);
3946
3947#define EVDNS_BASE_ALL_FLAGS (0x8001)
3948	if (flags & ~EVDNS_BASE_ALL_FLAGS) {
3949		flags = EVDNS_BASE_INITIALIZE_NAMESERVERS;
3950		log(EVDNS_LOG_WARN,
3951		    "Unrecognized flag passed to evdns_base_new(). Assuming "
3952		    "you meant EVDNS_BASE_INITIALIZE_NAMESERVERS.");
3953	}
3954#undef EVDNS_BASE_ALL_FLAGS
3955
3956	if (flags & EVDNS_BASE_INITIALIZE_NAMESERVERS) {
3957		int r;
3958#ifdef _WIN32
3959		r = evdns_base_config_windows_nameservers(base);
3960#else
3961		r = evdns_base_resolv_conf_parse(base, DNS_OPTIONS_ALL, "/etc/resolv.conf");
3962#endif
3963		if (r == -1) {
3964			evdns_base_free_and_unlock(base, 0);
3965			return NULL;
3966		}
3967	}
3968	if (flags & EVDNS_BASE_DISABLE_WHEN_INACTIVE) {
3969		base->disable_when_inactive = 1;
3970	}
3971
3972	EVDNS_UNLOCK(base);
3973	return base;
3974}
3975
3976int
3977evdns_init(void)
3978{
3979	struct evdns_base *base = evdns_base_new(NULL, 1);
3980	if (base) {
3981		current_base = base;
3982		return 0;
3983	} else {
3984		return -1;
3985	}
3986}
3987
3988const char *
3989evdns_err_to_string(int err)
3990{
3991    switch (err) {
3992	case DNS_ERR_NONE: return "no error";
3993	case DNS_ERR_FORMAT: return "misformatted query";
3994	case DNS_ERR_SERVERFAILED: return "server failed";
3995	case DNS_ERR_NOTEXIST: return "name does not exist";
3996	case DNS_ERR_NOTIMPL: return "query not implemented";
3997	case DNS_ERR_REFUSED: return "refused";
3998
3999	case DNS_ERR_TRUNCATED: return "reply truncated or ill-formed";
4000	case DNS_ERR_UNKNOWN: return "unknown";
4001	case DNS_ERR_TIMEOUT: return "request timed out";
4002	case DNS_ERR_SHUTDOWN: return "dns subsystem shut down";
4003	case DNS_ERR_CANCEL: return "dns request canceled";
4004	case DNS_ERR_NODATA: return "no records in the reply";
4005	default: return "[Unknown error code]";
4006    }
4007}
4008
4009static void
4010evdns_nameserver_free(struct nameserver *server)
4011{
4012	if (server->socket >= 0)
4013	evutil_closesocket(server->socket);
4014	(void) event_del(&server->event);
4015	event_debug_unassign(&server->event);
4016	if (server->state == 0)
4017		(void) event_del(&server->timeout_event);
4018	if (server->probe_request) {
4019		evdns_cancel_request(server->base, server->probe_request);
4020		server->probe_request = NULL;
4021	}
4022	event_debug_unassign(&server->timeout_event);
4023	mm_free(server);
4024}
4025
4026static void
4027evdns_base_free_and_unlock(struct evdns_base *base, int fail_requests)
4028{
4029	struct nameserver *server, *server_next;
4030	struct search_domain *dom, *dom_next;
4031	int i;
4032
4033	/* Requires that we hold the lock. */
4034
4035	/* TODO(nickm) we might need to refcount here. */
4036
4037	for (server = base->server_head; server; server = server_next) {
4038		server_next = server->next;
4039		evdns_nameserver_free(server);
4040		if (server_next == base->server_head)
4041			break;
4042	}
4043	base->server_head = NULL;
4044	base->global_good_nameservers = 0;
4045
4046	for (i = 0; i < base->n_req_heads; ++i) {
4047		while (base->req_heads[i]) {
4048			if (fail_requests)
4049				reply_schedule_callback(base->req_heads[i], 0, DNS_ERR_SHUTDOWN, NULL);
4050			request_finished(base->req_heads[i], &REQ_HEAD(base, base->req_heads[i]->trans_id), 1);
4051		}
4052	}
4053	while (base->req_waiting_head) {
4054		if (fail_requests)
4055			reply_schedule_callback(base->req_waiting_head, 0, DNS_ERR_SHUTDOWN, NULL);
4056		request_finished(base->req_waiting_head, &base->req_waiting_head, 1);
4057	}
4058	base->global_requests_inflight = base->global_requests_waiting = 0;
4059
4060
4061	if (base->global_search_state) {
4062		for (dom = base->global_search_state->head; dom; dom = dom_next) {
4063			dom_next = dom->next;
4064			mm_free(dom);
4065		}
4066		mm_free(base->global_search_state);
4067		base->global_search_state = NULL;
4068	}
4069
4070	{
4071		struct hosts_entry *victim;
4072		while ((victim = TAILQ_FIRST(&base->hostsdb))) {
4073			TAILQ_REMOVE(&base->hostsdb, victim, next);
4074			mm_free(victim);
4075		}
4076	}
4077
4078	mm_free(base->req_heads);
4079
4080	EVDNS_UNLOCK(base);
4081	EVTHREAD_FREE_LOCK(base->lock, EVTHREAD_LOCKTYPE_RECURSIVE);
4082
4083	mm_free(base);
4084}
4085
4086void
4087evdns_base_free(struct evdns_base *base, int fail_requests)
4088{
4089	EVDNS_LOCK(base);
4090	evdns_base_free_and_unlock(base, fail_requests);
4091}
4092
4093void
4094evdns_base_clear_host_addresses(struct evdns_base *base)
4095{
4096	struct hosts_entry *victim;
4097	EVDNS_LOCK(base);
4098	while ((victim = TAILQ_FIRST(&base->hostsdb))) {
4099		TAILQ_REMOVE(&base->hostsdb, victim, next);
4100		mm_free(victim);
4101	}
4102	EVDNS_UNLOCK(base);
4103}
4104
4105void
4106evdns_shutdown(int fail_requests)
4107{
4108	if (current_base) {
4109		struct evdns_base *b = current_base;
4110		current_base = NULL;
4111		evdns_base_free(b, fail_requests);
4112	}
4113	evdns_log_fn = NULL;
4114}
4115
4116static int
4117evdns_base_parse_hosts_line(struct evdns_base *base, char *line)
4118{
4119	char *strtok_state;
4120	static const char *const delims = " \t";
4121	char *const addr = strtok_r(line, delims, &strtok_state);
4122	char *hostname, *hash;
4123	struct sockaddr_storage ss;
4124	int socklen = sizeof(ss);
4125	ASSERT_LOCKED(base);
4126
4127#define NEXT_TOKEN strtok_r(NULL, delims, &strtok_state)
4128
4129	if (!addr || *addr == '#')
4130		return 0;
4131
4132	memset(&ss, 0, sizeof(ss));
4133	if (evutil_parse_sockaddr_port(addr, (struct sockaddr*)&ss, &socklen)<0)
4134		return -1;
4135	if (socklen > (int)sizeof(struct sockaddr_in6))
4136		return -1;
4137
4138	if (sockaddr_getport((struct sockaddr*)&ss))
4139		return -1;
4140
4141	while ((hostname = NEXT_TOKEN)) {
4142		struct hosts_entry *he;
4143		size_t namelen;
4144		if ((hash = strchr(hostname, '#'))) {
4145			if (hash == hostname)
4146				return 0;
4147			*hash = '\0';
4148		}
4149
4150		namelen = strlen(hostname);
4151
4152		he = mm_calloc(1, sizeof(struct hosts_entry)+namelen);
4153		if (!he)
4154			return -1;
4155		EVUTIL_ASSERT(socklen <= (int)sizeof(he->addr));
4156		memcpy(&he->addr, &ss, socklen);
4157		memcpy(he->hostname, hostname, namelen+1);
4158		he->addrlen = socklen;
4159
4160		TAILQ_INSERT_TAIL(&base->hostsdb, he, next);
4161
4162		if (hash)
4163			return 0;
4164	}
4165
4166	return 0;
4167#undef NEXT_TOKEN
4168}
4169
4170static int
4171evdns_base_load_hosts_impl(struct evdns_base *base, const char *hosts_fname)
4172{
4173	char *str=NULL, *cp, *eol;
4174	size_t len;
4175	int err=0;
4176
4177	ASSERT_LOCKED(base);
4178
4179	if (hosts_fname == NULL ||
4180	    (err = evutil_read_file_(hosts_fname, &str, &len, 0)) < 0) {
4181		char tmp[64];
4182		strlcpy(tmp, "127.0.0.1   localhost", sizeof(tmp));
4183		evdns_base_parse_hosts_line(base, tmp);
4184		strlcpy(tmp, "::1   localhost", sizeof(tmp));
4185		evdns_base_parse_hosts_line(base, tmp);
4186		return err ? -1 : 0;
4187	}
4188
4189	/* This will break early if there is a NUL in the hosts file.
4190	 * Probably not a problem.*/
4191	cp = str;
4192	for (;;) {
4193		eol = strchr(cp, '\n');
4194
4195		if (eol) {
4196			*eol = '\0';
4197			evdns_base_parse_hosts_line(base, cp);
4198			cp = eol+1;
4199		} else {
4200			evdns_base_parse_hosts_line(base, cp);
4201			break;
4202		}
4203	}
4204
4205	mm_free(str);
4206	return 0;
4207}
4208
4209int
4210evdns_base_load_hosts(struct evdns_base *base, const char *hosts_fname)
4211{
4212	int res;
4213	if (!base)
4214		base = current_base;
4215	EVDNS_LOCK(base);
4216	res = evdns_base_load_hosts_impl(base, hosts_fname);
4217	EVDNS_UNLOCK(base);
4218	return res;
4219}
4220
4221/* A single request for a getaddrinfo, either v4 or v6. */
4222struct getaddrinfo_subrequest {
4223	struct evdns_request *r;
4224	ev_uint32_t type;
4225};
4226
4227/* State data used to implement an in-progress getaddrinfo. */
4228struct evdns_getaddrinfo_request {
4229	struct evdns_base *evdns_base;
4230	/* Copy of the modified 'hints' data that we'll use to build
4231	 * answers. */
4232	struct evutil_addrinfo hints;
4233	/* The callback to invoke when we're done */
4234	evdns_getaddrinfo_cb user_cb;
4235	/* User-supplied data to give to the callback. */
4236	void *user_data;
4237	/* The port to use when building sockaddrs. */
4238	ev_uint16_t port;
4239	/* The sub_request for an A record (if any) */
4240	struct getaddrinfo_subrequest ipv4_request;
4241	/* The sub_request for an AAAA record (if any) */
4242	struct getaddrinfo_subrequest ipv6_request;
4243
4244	/* The cname result that we were told (if any) */
4245	char *cname_result;
4246
4247	/* If we have one request answered and one request still inflight,
4248	 * then this field holds the answer from the first request... */
4249	struct evutil_addrinfo *pending_result;
4250	/* And this event is a timeout that will tell us to cancel the second
4251	 * request if it's taking a long time. */
4252	struct event timeout;
4253
4254	/* And this field holds the error code from the first request... */
4255	int pending_error;
4256	/* If this is set, the user canceled this request. */
4257	unsigned user_canceled : 1;
4258	/* If this is set, the user can no longer cancel this request; we're
4259	 * just waiting for the free. */
4260	unsigned request_done : 1;
4261};
4262
4263/* Convert an evdns errors to the equivalent getaddrinfo error. */
4264static int
4265evdns_err_to_getaddrinfo_err(int e1)
4266{
4267	/* XXX Do this better! */
4268	if (e1 == DNS_ERR_NONE)
4269		return 0;
4270	else if (e1 == DNS_ERR_NOTEXIST)
4271		return EVUTIL_EAI_NONAME;
4272	else
4273		return EVUTIL_EAI_FAIL;
4274}
4275
4276/* Return the more informative of two getaddrinfo errors. */
4277static int
4278getaddrinfo_merge_err(int e1, int e2)
4279{
4280	/* XXXX be cleverer here. */
4281	if (e1 == 0)
4282		return e2;
4283	else
4284		return e1;
4285}
4286
4287static void
4288free_getaddrinfo_request(struct evdns_getaddrinfo_request *data)
4289{
4290	/* DO NOT CALL this if either of the requests is pending.  Only once
4291	 * both callbacks have been invoked is it safe to free the request */
4292	if (data->pending_result)
4293		evutil_freeaddrinfo(data->pending_result);
4294	if (data->cname_result)
4295		mm_free(data->cname_result);
4296	event_del(&data->timeout);
4297	mm_free(data);
4298	return;
4299}
4300
4301static void
4302add_cname_to_reply(struct evdns_getaddrinfo_request *data,
4303    struct evutil_addrinfo *ai)
4304{
4305	if (data->cname_result && ai) {
4306		ai->ai_canonname = data->cname_result;
4307		data->cname_result = NULL;
4308	}
4309}
4310
4311/* Callback: invoked when one request in a mixed-format A/AAAA getaddrinfo
4312 * request has finished, but the other one took too long to answer. Pass
4313 * along the answer we got, and cancel the other request.
4314 */
4315static void
4316evdns_getaddrinfo_timeout_cb(evutil_socket_t fd, short what, void *ptr)
4317{
4318	int v4_timedout = 0, v6_timedout = 0;
4319	struct evdns_getaddrinfo_request *data = ptr;
4320
4321	/* Cancel any pending requests, and note which one */
4322	if (data->ipv4_request.r) {
4323		/* XXXX This does nothing if the request's callback is already
4324		 * running (pending_cb is set). */
4325		evdns_cancel_request(NULL, data->ipv4_request.r);
4326		v4_timedout = 1;
4327		EVDNS_LOCK(data->evdns_base);
4328		++data->evdns_base->getaddrinfo_ipv4_timeouts;
4329		EVDNS_UNLOCK(data->evdns_base);
4330	}
4331	if (data->ipv6_request.r) {
4332		/* XXXX This does nothing if the request's callback is already
4333		 * running (pending_cb is set). */
4334		evdns_cancel_request(NULL, data->ipv6_request.r);
4335		v6_timedout = 1;
4336		EVDNS_LOCK(data->evdns_base);
4337		++data->evdns_base->getaddrinfo_ipv6_timeouts;
4338		EVDNS_UNLOCK(data->evdns_base);
4339	}
4340
4341	/* We only use this timeout callback when we have an answer for
4342	 * one address. */
4343	EVUTIL_ASSERT(!v4_timedout || !v6_timedout);
4344
4345	/* Report the outcome of the other request that didn't time out. */
4346	if (data->pending_result) {
4347		add_cname_to_reply(data, data->pending_result);
4348		data->user_cb(0, data->pending_result, data->user_data);
4349		data->pending_result = NULL;
4350	} else {
4351		int e = data->pending_error;
4352		if (!e)
4353			e = EVUTIL_EAI_AGAIN;
4354		data->user_cb(e, NULL, data->user_data);
4355	}
4356
4357	data->user_cb = NULL; /* prevent double-call if evdns callbacks are
4358			       * in-progress. XXXX It would be better if this
4359			       * weren't necessary. */
4360
4361	if (!v4_timedout && !v6_timedout) {
4362		/* should be impossible? XXXX */
4363		free_getaddrinfo_request(data);
4364	}
4365}
4366
4367static int
4368evdns_getaddrinfo_set_timeout(struct evdns_base *evdns_base,
4369    struct evdns_getaddrinfo_request *data)
4370{
4371	return event_add(&data->timeout, &evdns_base->global_getaddrinfo_allow_skew);
4372}
4373
4374static inline int
4375evdns_result_is_answer(int result)
4376{
4377	return (result != DNS_ERR_NOTIMPL && result != DNS_ERR_REFUSED &&
4378	    result != DNS_ERR_SERVERFAILED && result != DNS_ERR_CANCEL);
4379}
4380
4381static void
4382evdns_getaddrinfo_gotresolve(int result, char type, int count,
4383    int ttl, void *addresses, void *arg)
4384{
4385	int i;
4386	struct getaddrinfo_subrequest *req = arg;
4387	struct getaddrinfo_subrequest *other_req;
4388	struct evdns_getaddrinfo_request *data;
4389
4390	struct evutil_addrinfo *res;
4391
4392	struct sockaddr_in sin;
4393	struct sockaddr_in6 sin6;
4394	struct sockaddr *sa;
4395	int socklen, addrlen;
4396	void *addrp;
4397	int err;
4398	int user_canceled;
4399
4400	EVUTIL_ASSERT(req->type == DNS_IPv4_A || req->type == DNS_IPv6_AAAA);
4401	if (req->type == DNS_IPv4_A) {
4402		data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv4_request);
4403		other_req = &data->ipv6_request;
4404	} else {
4405		data = EVUTIL_UPCAST(req, struct evdns_getaddrinfo_request, ipv6_request);
4406		other_req = &data->ipv4_request;
4407	}
4408
4409	EVDNS_LOCK(data->evdns_base);
4410	if (evdns_result_is_answer(result)) {
4411		if (req->type == DNS_IPv4_A)
4412			++data->evdns_base->getaddrinfo_ipv4_answered;
4413		else
4414			++data->evdns_base->getaddrinfo_ipv6_answered;
4415	}
4416	user_canceled = data->user_canceled;
4417	if (other_req->r == NULL)
4418		data->request_done = 1;
4419	EVDNS_UNLOCK(data->evdns_base);
4420
4421	req->r = NULL;
4422
4423	if (result == DNS_ERR_CANCEL && ! user_canceled) {
4424		/* Internal cancel request from timeout or internal error.
4425		 * we already answered the user. */
4426		if (other_req->r == NULL)
4427			free_getaddrinfo_request(data);
4428		return;
4429	}
4430
4431	if (data->user_cb == NULL) {
4432		/* We already answered.  XXXX This shouldn't be needed; see
4433		 * comments in evdns_getaddrinfo_timeout_cb */
4434		free_getaddrinfo_request(data);
4435		return;
4436	}
4437
4438	if (result == DNS_ERR_NONE) {
4439		if (count == 0)
4440			err = EVUTIL_EAI_NODATA;
4441		else
4442			err = 0;
4443	} else {
4444		err = evdns_err_to_getaddrinfo_err(result);
4445	}
4446
4447	if (err) {
4448		/* Looks like we got an error. */
4449		if (other_req->r) {
4450			/* The other request is still working; maybe it will
4451			 * succeed. */
4452			/* XXXX handle failure from set_timeout */
4453			evdns_getaddrinfo_set_timeout(data->evdns_base, data);
4454			data->pending_error = err;
4455			return;
4456		}
4457
4458		if (user_canceled) {
4459			data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data);
4460		} else if (data->pending_result) {
4461			/* If we have an answer waiting, and we weren't
4462			 * canceled, ignore this error. */
4463			add_cname_to_reply(data, data->pending_result);
4464			data->user_cb(0, data->pending_result, data->user_data);
4465			data->pending_result = NULL;
4466		} else {
4467			if (data->pending_error)
4468				err = getaddrinfo_merge_err(err,
4469				    data->pending_error);
4470			data->user_cb(err, NULL, data->user_data);
4471		}
4472		free_getaddrinfo_request(data);
4473		return;
4474	} else if (user_canceled) {
4475		if (other_req->r) {
4476			/* The other request is still working; let it hit this
4477			 * callback with EVUTIL_EAI_CANCEL callback and report
4478			 * the failure. */
4479			return;
4480		}
4481		data->user_cb(EVUTIL_EAI_CANCEL, NULL, data->user_data);
4482		free_getaddrinfo_request(data);
4483		return;
4484	}
4485
4486	/* Looks like we got some answers. We should turn them into addrinfos
4487	 * and then either queue those or return them all. */
4488	EVUTIL_ASSERT(type == DNS_IPv4_A || type == DNS_IPv6_AAAA);
4489
4490	if (type == DNS_IPv4_A) {
4491		memset(&sin, 0, sizeof(sin));
4492		sin.sin_family = AF_INET;
4493		sin.sin_port = htons(data->port);
4494
4495		sa = (struct sockaddr *)&sin;
4496		socklen = sizeof(sin);
4497		addrlen = 4;
4498		addrp = &sin.sin_addr.s_addr;
4499	} else {
4500		memset(&sin6, 0, sizeof(sin6));
4501		sin6.sin6_family = AF_INET6;
4502		sin6.sin6_port = htons(data->port);
4503
4504		sa = (struct sockaddr *)&sin6;
4505		socklen = sizeof(sin6);
4506		addrlen = 16;
4507		addrp = &sin6.sin6_addr.s6_addr;
4508	}
4509
4510	res = NULL;
4511	for (i=0; i < count; ++i) {
4512		struct evutil_addrinfo *ai;
4513		memcpy(addrp, ((char*)addresses)+i*addrlen, addrlen);
4514		ai = evutil_new_addrinfo_(sa, socklen, &data->hints);
4515		if (!ai) {
4516			if (other_req->r) {
4517				evdns_cancel_request(NULL, other_req->r);
4518			}
4519			data->user_cb(EVUTIL_EAI_MEMORY, NULL, data->user_data);
4520			if (res)
4521				evutil_freeaddrinfo(res);
4522
4523			if (other_req->r == NULL)
4524				free_getaddrinfo_request(data);
4525			return;
4526		}
4527		res = evutil_addrinfo_append_(res, ai);
4528	}
4529
4530	if (other_req->r) {
4531		/* The other request is still in progress; wait for it */
4532		/* XXXX handle failure from set_timeout */
4533		evdns_getaddrinfo_set_timeout(data->evdns_base, data);
4534		data->pending_result = res;
4535		return;
4536	} else {
4537		/* The other request is done or never started; append its
4538		 * results (if any) and return them. */
4539		if (data->pending_result) {
4540			if (req->type == DNS_IPv4_A)
4541				res = evutil_addrinfo_append_(res,
4542				    data->pending_result);
4543			else
4544				res = evutil_addrinfo_append_(
4545				    data->pending_result, res);
4546			data->pending_result = NULL;
4547		}
4548
4549		/* Call the user callback. */
4550		add_cname_to_reply(data, res);
4551		data->user_cb(0, res, data->user_data);
4552
4553		/* Free data. */
4554		free_getaddrinfo_request(data);
4555	}
4556}
4557
4558static struct hosts_entry *
4559find_hosts_entry(struct evdns_base *base, const char *hostname,
4560    struct hosts_entry *find_after)
4561{
4562	struct hosts_entry *e;
4563
4564	if (find_after)
4565		e = TAILQ_NEXT(find_after, next);
4566	else
4567		e = TAILQ_FIRST(&base->hostsdb);
4568
4569	for (; e; e = TAILQ_NEXT(e, next)) {
4570		if (!evutil_ascii_strcasecmp(e->hostname, hostname))
4571			return e;
4572	}
4573	return NULL;
4574}
4575
4576static int
4577evdns_getaddrinfo_fromhosts(struct evdns_base *base,
4578    const char *nodename, struct evutil_addrinfo *hints, ev_uint16_t port,
4579    struct evutil_addrinfo **res)
4580{
4581	int n_found = 0;
4582	struct hosts_entry *e;
4583	struct evutil_addrinfo *ai=NULL;
4584	int f = hints->ai_family;
4585
4586	EVDNS_LOCK(base);
4587	for (e = find_hosts_entry(base, nodename, NULL); e;
4588	    e = find_hosts_entry(base, nodename, e)) {
4589		struct evutil_addrinfo *ai_new;
4590		++n_found;
4591		if ((e->addr.sa.sa_family == AF_INET && f == PF_INET6) ||
4592		    (e->addr.sa.sa_family == AF_INET6 && f == PF_INET))
4593			continue;
4594		ai_new = evutil_new_addrinfo_(&e->addr.sa, e->addrlen, hints);
4595		if (!ai_new) {
4596			n_found = 0;
4597			goto out;
4598		}
4599		sockaddr_setport(ai_new->ai_addr, port);
4600		ai = evutil_addrinfo_append_(ai, ai_new);
4601	}
4602	EVDNS_UNLOCK(base);
4603out:
4604	if (n_found) {
4605		/* Note that we return an empty answer if we found entries for
4606		 * this hostname but none were of the right address type. */
4607		*res = ai;
4608		return 0;
4609	} else {
4610		if (ai)
4611			evutil_freeaddrinfo(ai);
4612		return -1;
4613	}
4614}
4615
4616struct evdns_getaddrinfo_request *
4617evdns_getaddrinfo(struct evdns_base *dns_base,
4618    const char *nodename, const char *servname,
4619    const struct evutil_addrinfo *hints_in,
4620    evdns_getaddrinfo_cb cb, void *arg)
4621{
4622	struct evdns_getaddrinfo_request *data;
4623	struct evutil_addrinfo hints;
4624	struct evutil_addrinfo *res = NULL;
4625	int err;
4626	int port = 0;
4627	int want_cname = 0;
4628
4629	if (!dns_base) {
4630		dns_base = current_base;
4631		if (!dns_base) {
4632			log(EVDNS_LOG_WARN,
4633			    "Call to getaddrinfo_async with no "
4634			    "evdns_base configured.");
4635			cb(EVUTIL_EAI_FAIL, NULL, arg); /* ??? better error? */
4636			return NULL;
4637		}
4638	}
4639
4640	/* If we _must_ answer this immediately, do so. */
4641	if ((hints_in && (hints_in->ai_flags & EVUTIL_AI_NUMERICHOST))) {
4642		res = NULL;
4643		err = evutil_getaddrinfo(nodename, servname, hints_in, &res);
4644		cb(err, res, arg);
4645		return NULL;
4646	}
4647
4648	if (hints_in) {
4649		memcpy(&hints, hints_in, sizeof(hints));
4650	} else {
4651		memset(&hints, 0, sizeof(hints));
4652		hints.ai_family = PF_UNSPEC;
4653	}
4654
4655	evutil_adjust_hints_for_addrconfig_(&hints);
4656
4657	/* Now try to see if we _can_ answer immediately. */
4658	/* (It would be nice to do this by calling getaddrinfo directly, with
4659	 * AI_NUMERICHOST, on plaforms that have it, but we can't: there isn't
4660	 * a reliable way to distinguish the "that wasn't a numeric host!" case
4661	 * from any other EAI_NONAME cases.) */
4662	err = evutil_getaddrinfo_common_(nodename, servname, &hints, &res, &port);
4663	if (err != EVUTIL_EAI_NEED_RESOLVE) {
4664		cb(err, res, arg);
4665		return NULL;
4666	}
4667
4668	/* If there is an entry in the hosts file, we should give it now. */
4669	if (!evdns_getaddrinfo_fromhosts(dns_base, nodename, &hints, port, &res)) {
4670		cb(0, res, arg);
4671		return NULL;
4672	}
4673
4674	/* Okay, things are serious now. We're going to need to actually
4675	 * launch a request.
4676	 */
4677	data = mm_calloc(1,sizeof(struct evdns_getaddrinfo_request));
4678	if (!data) {
4679		cb(EVUTIL_EAI_MEMORY, NULL, arg);
4680		return NULL;
4681	}
4682
4683	memcpy(&data->hints, &hints, sizeof(data->hints));
4684	data->port = (ev_uint16_t)port;
4685	data->ipv4_request.type = DNS_IPv4_A;
4686	data->ipv6_request.type = DNS_IPv6_AAAA;
4687	data->user_cb = cb;
4688	data->user_data = arg;
4689	data->evdns_base = dns_base;
4690
4691	want_cname = (hints.ai_flags & EVUTIL_AI_CANONNAME);
4692
4693	/* If we are asked for a PF_UNSPEC address, we launch two requests in
4694	 * parallel: one for an A address and one for an AAAA address.  We
4695	 * can't send just one request, since many servers only answer one
4696	 * question per DNS request.
4697	 *
4698	 * Once we have the answer to one request, we allow for a short
4699	 * timeout before we report it, to see if the other one arrives.  If
4700	 * they both show up in time, then we report both the answers.
4701	 *
4702	 * If too many addresses of one type time out or fail, we should stop
4703	 * launching those requests. (XXX we don't do that yet.)
4704	 */
4705
4706	if (hints.ai_family != PF_INET6) {
4707		log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv4 as %p",
4708		    nodename, &data->ipv4_request);
4709
4710		data->ipv4_request.r = evdns_base_resolve_ipv4(dns_base,
4711		    nodename, 0, evdns_getaddrinfo_gotresolve,
4712		    &data->ipv4_request);
4713		if (want_cname && data->ipv4_request.r)
4714			data->ipv4_request.r->current_req->put_cname_in_ptr =
4715			    &data->cname_result;
4716	}
4717	if (hints.ai_family != PF_INET) {
4718		log(EVDNS_LOG_DEBUG, "Sending request for %s on ipv6 as %p",
4719		    nodename, &data->ipv6_request);
4720
4721		data->ipv6_request.r = evdns_base_resolve_ipv6(dns_base,
4722		    nodename, 0, evdns_getaddrinfo_gotresolve,
4723		    &data->ipv6_request);
4724		if (want_cname && data->ipv6_request.r)
4725			data->ipv6_request.r->current_req->put_cname_in_ptr =
4726			    &data->cname_result;
4727	}
4728
4729	evtimer_assign(&data->timeout, dns_base->event_base,
4730	    evdns_getaddrinfo_timeout_cb, data);
4731
4732	if (data->ipv4_request.r || data->ipv6_request.r) {
4733		return data;
4734	} else {
4735		mm_free(data);
4736		cb(EVUTIL_EAI_FAIL, NULL, arg);
4737		return NULL;
4738	}
4739}
4740
4741void
4742evdns_getaddrinfo_cancel(struct evdns_getaddrinfo_request *data)
4743{
4744	EVDNS_LOCK(data->evdns_base);
4745	if (data->request_done) {
4746		EVDNS_UNLOCK(data->evdns_base);
4747		return;
4748	}
4749	event_del(&data->timeout);
4750	data->user_canceled = 1;
4751	if (data->ipv4_request.r)
4752		evdns_cancel_request(data->evdns_base, data->ipv4_request.r);
4753	if (data->ipv6_request.r)
4754		evdns_cancel_request(data->evdns_base, data->ipv6_request.r);
4755	EVDNS_UNLOCK(data->evdns_base);
4756}
4757