1 /* Copyright 2003-2004 Roger Dingledine.
2 * Copyright 2004 Roger Dingledine, Nick Mathewson. */
3 /* See LICENSE for licensing information */
5 const char dns_c_id
[] = "$Id$";
9 * \brief Resolve hostnames in separate processes.
12 /* See http://elvin.dstc.com/ListArchive/elvin-dev/archive/2001/09/msg00027.html
13 * for some approaches to asynchronous dns. We will want to switch once one of
14 * them becomes more commonly available.
20 /** Longest hostname we're willing to resolve. */
21 #define MAX_ADDRESSLEN 256
23 /** Maximum DNS processes to spawn. */
24 #define MAX_DNSWORKERS 100
25 /** Minimum DNS processes to spawn. */
26 #define MIN_DNSWORKERS 3
28 /** If more than this many processes are idle, shut down the extras. */
29 #define MAX_IDLE_DNSWORKERS 10
31 /** Possible outcomes from hostname lookup: permanent failure,
32 * transient (retryable) failure, and success. */
33 #define DNS_RESOLVE_FAILED_TRANSIENT 1
34 #define DNS_RESOLVE_FAILED_PERMANENT 2
35 #define DNS_RESOLVE_SUCCEEDED 3
37 /** How many dnsworkers we have running right now. */
38 static int num_dnsworkers
=0;
39 /** How many of the running dnsworkers have an assigned task right now. */
40 static int num_dnsworkers_busy
=0;
41 /** When did we last rotate the dnsworkers? */
42 static time_t last_rotation_time
=0;
44 /** Linked list of connections waiting for a DNS answer. */
45 struct pending_connection_t
{
46 struct connection_t
*conn
;
47 struct pending_connection_t
*next
;
50 /** A DNS request: possibly completed, possibly pending; cached_resolve
51 * structs are stored at the OR side in a splay tree, and as a linked
52 * list from oldest to newest.
54 struct cached_resolve
{
55 SPLAY_ENTRY(cached_resolve
) node
;
56 char address
[MAX_ADDRESSLEN
]; /**< The hostname to be resolved. */
57 uint32_t addr
; /**< IPv4 addr for <b>address</b>. */
58 char state
; /**< 0 is pending; 1 means answer is valid; 2 means resolve failed. */
59 #define CACHE_STATE_PENDING 0
60 #define CACHE_STATE_VALID 1
61 #define CACHE_STATE_FAILED 2
62 uint32_t expire
; /**< Remove items from cache after this time. */
63 struct pending_connection_t
*pending_connections
;
64 struct cached_resolve
*next
;
67 static void purge_expired_resolves(uint32_t now
);
68 static int assign_to_dnsworker(connection_t
*exitconn
);
69 static void dns_purge_resolve(struct cached_resolve
*resolve
);
70 static void dns_found_answer(char *address
, uint32_t addr
, char outcome
);
71 static int dnsworker_main(void *data
);
72 static int spawn_dnsworker(void);
73 static void spawn_enough_dnsworkers(void);
74 static void send_resolved_cell(connection_t
*conn
, uint8_t answer_type
);
76 /** Splay tree of cached_resolve objects. */
77 static SPLAY_HEAD(cache_tree
, cached_resolve
) cache_root
;
79 /** Function to compare hashed resolves on their addresses; used to
80 * implement splay trees. */
81 static int compare_cached_resolves(struct cached_resolve
*a
,
82 struct cached_resolve
*b
) {
83 /* make this smarter one day? */
84 return strncasecmp(a
->address
, b
->address
, MAX_ADDRESSLEN
);
87 SPLAY_PROTOTYPE(cache_tree
, cached_resolve
, node
, compare_cached_resolves
);
88 SPLAY_GENERATE(cache_tree
, cached_resolve
, node
, compare_cached_resolves
);
90 /** Initialize the DNS cache. */
91 static void init_cache_tree(void) {
92 SPLAY_INIT(&cache_root
);
95 /** Initialize the DNS subsystem; called by the OR process. */
98 last_rotation_time
=time(NULL
);
99 spawn_enough_dnsworkers();
102 /** Linked list of resolved addresses, oldest to newest. */
103 static struct cached_resolve
*oldest_cached_resolve
= NULL
;
104 static struct cached_resolve
*newest_cached_resolve
= NULL
;
106 /** Remove every cached_resolve whose <b>expire</b> time is before <b>now</b>
108 static void purge_expired_resolves(uint32_t now
) {
109 struct cached_resolve
*resolve
;
110 struct pending_connection_t
*pend
;
111 connection_t
*pendconn
;
113 /* this is fast because the linked list
114 * oldest_cached_resolve is ordered by when they came in.
116 while (oldest_cached_resolve
&& (oldest_cached_resolve
->expire
< now
)) {
117 resolve
= oldest_cached_resolve
;
118 log(LOG_DEBUG
,"Forgetting old cached resolve (expires %lu)", (unsigned long)resolve
->expire
);
119 if (resolve
->state
== CACHE_STATE_PENDING
) {
120 log_fn(LOG_WARN
,"Bug: Expiring a dns resolve that's still pending. Forgot to cull it?");
125 if (resolve
->pending_connections
) {
126 log_fn(LOG_WARN
, "Closing pending connections on expiring DNS resolve!");
130 while (resolve
->pending_connections
) {
131 pend
= resolve
->pending_connections
;
132 resolve
->pending_connections
= pend
->next
;
133 /* Connections should only be pending if they have no socket. */
134 tor_assert(pend
->conn
->s
== -1);
135 pendconn
= pend
->conn
;
136 connection_edge_end(pendconn
, END_STREAM_REASON_MISC
,
137 pendconn
->cpath_layer
);
138 circuit_detach_stream(circuit_get_by_conn(pendconn
), pendconn
);
139 connection_free(pendconn
);
143 oldest_cached_resolve
= resolve
->next
;
144 if (!oldest_cached_resolve
) /* if there are no more, */
145 newest_cached_resolve
= NULL
; /* then make sure the list's tail knows that too */
146 SPLAY_REMOVE(cache_tree
, &cache_root
, resolve
);
151 static void send_resolved_cell(connection_t
*conn
, uint8_t answer_type
)
153 char buf
[RELAY_PAYLOAD_SIZE
];
156 buf
[0] = answer_type
;
160 case RESOLVED_TYPE_IPV4
:
162 set_uint32(buf
+2, htonl(conn
->addr
));
165 case RESOLVED_TYPE_ERROR_TRANSIENT
:
166 case RESOLVED_TYPE_ERROR
:
167 buf
[1] = 24; /* length of "error resolving hostname" */
168 strlcpy(buf
+2, "error resolving hostname", sizeof(buf
)-2);
174 connection_edge_send_command(conn
, circuit_get_by_conn(conn
),
175 RELAY_COMMAND_RESOLVED
, buf
, buflen
,
179 /** Link <b>r</b> into the tree of address-to-result mappings, and add it to
180 * the linked list of resolves-by-age. */
182 insert_resolve(struct cached_resolve
*r
)
184 /* add us to the linked list of resolves */
185 if (!oldest_cached_resolve
) {
186 oldest_cached_resolve
= r
;
188 newest_cached_resolve
->next
= r
;
190 newest_cached_resolve
= r
;
192 SPLAY_INSERT(cache_tree
, &cache_root
, r
);
195 /** See if we have a cache entry for <b>exitconn</b>-\>address. if so,
196 * if resolve valid, put it into <b>exitconn</b>-\>addr and return 1.
197 * If resolve failed, unlink exitconn if needed, free it, and return -1.
199 * Else, if seen before and pending, add conn to the pending list,
202 * Else, if not seen before, add conn to pending list, hand to
203 * dns farm, and return 0.
205 int dns_resolve(connection_t
*exitconn
) {
206 struct cached_resolve
*resolve
;
207 struct cached_resolve search
;
208 struct pending_connection_t
*pending_connection
;
211 uint32_t now
= time(NULL
);
212 assert_connection_ok(exitconn
, 0);
213 tor_assert(exitconn
->s
== -1);
215 /* first check if exitconn->address is an IP. If so, we already
216 * know the answer. */
217 if (tor_inet_aton(exitconn
->address
, &in
) != 0) {
218 exitconn
->addr
= ntohl(in
.s_addr
);
222 /* then take this opportunity to see if there are any expired
223 * resolves in the tree. */
224 purge_expired_resolves(now
);
226 /* now check the tree to see if 'address' is already there. */
227 strlcpy(search
.address
, exitconn
->address
, sizeof(search
.address
));
228 resolve
= SPLAY_FIND(cache_tree
, &cache_root
, &search
);
229 if (resolve
) { /* already there */
230 switch (resolve
->state
) {
231 case CACHE_STATE_PENDING
:
232 /* add us to the pending list */
233 pending_connection
= tor_malloc_zero(
234 sizeof(struct pending_connection_t
));
235 pending_connection
->conn
= exitconn
;
236 pending_connection
->next
= resolve
->pending_connections
;
237 resolve
->pending_connections
= pending_connection
;
238 log_fn(LOG_DEBUG
,"Connection (fd %d) waiting for pending DNS resolve of '%s'",
239 exitconn
->s
, exitconn
->address
);
240 exitconn
->state
= EXIT_CONN_STATE_RESOLVING
;
242 case CACHE_STATE_VALID
:
243 exitconn
->addr
= resolve
->addr
;
244 log_fn(LOG_DEBUG
,"Connection (fd %d) found cached answer for '%s'",
245 exitconn
->s
, exitconn
->address
);
246 if (exitconn
->purpose
== EXIT_PURPOSE_RESOLVE
)
247 send_resolved_cell(exitconn
, RESOLVED_TYPE_IPV4
);
249 case CACHE_STATE_FAILED
:
250 log_fn(LOG_DEBUG
,"Connection (fd %d) found cached error for '%s'",
251 exitconn
->s
, exitconn
->address
);
252 if (exitconn
->purpose
== EXIT_PURPOSE_RESOLVE
)
253 send_resolved_cell(exitconn
, RESOLVED_TYPE_ERROR
);
254 circ
= circuit_get_by_conn(exitconn
);
256 circuit_detach_stream(circ
, exitconn
);
257 connection_free(exitconn
);
262 /* not there, need to add it */
263 resolve
= tor_malloc_zero(sizeof(struct cached_resolve
));
264 resolve
->state
= CACHE_STATE_PENDING
;
265 resolve
->expire
= now
+ MAX_DNS_ENTRY_AGE
;
266 strlcpy(resolve
->address
, exitconn
->address
, sizeof(resolve
->address
));
268 /* add us to the pending list */
269 pending_connection
= tor_malloc_zero(sizeof(struct pending_connection_t
));
270 pending_connection
->conn
= exitconn
;
271 resolve
->pending_connections
= pending_connection
;
272 exitconn
->state
= EXIT_CONN_STATE_RESOLVING
;
274 insert_resolve(resolve
);
275 return assign_to_dnsworker(exitconn
);
278 /** Find or spawn a dns worker process to handle resolving
279 * <b>exitconn</b>-\>address; tell that dns worker to begin resolving.
281 static int assign_to_dnsworker(connection_t
*exitconn
) {
282 connection_t
*dnsconn
;
285 tor_assert(exitconn
->state
== EXIT_CONN_STATE_RESOLVING
);
286 tor_assert(exitconn
->s
== -1);
288 spawn_enough_dnsworkers(); /* respawn here, to be sure there are enough */
290 dnsconn
= connection_get_by_type_state(CONN_TYPE_DNSWORKER
, DNSWORKER_STATE_IDLE
);
293 log_fn(LOG_WARN
,"no idle dns workers. Failing.");
294 if (exitconn
->purpose
== EXIT_PURPOSE_RESOLVE
)
295 send_resolved_cell(exitconn
, RESOLVED_TYPE_ERROR_TRANSIENT
);
296 dns_cancel_pending_resolve(exitconn
->address
); /* also sends end and frees! */
300 log_fn(LOG_DEBUG
, "Connection (fd %d) needs to resolve '%s'; assigning to DNSWorker (fd %d)",
301 exitconn
->s
, exitconn
->address
, dnsconn
->s
);
303 tor_free(dnsconn
->address
);
304 dnsconn
->address
= tor_strdup(exitconn
->address
);
305 dnsconn
->state
= DNSWORKER_STATE_BUSY
;
306 num_dnsworkers_busy
++;
308 len
= strlen(dnsconn
->address
);
309 connection_write_to_buf(&len
, 1, dnsconn
);
310 connection_write_to_buf(dnsconn
->address
, len
, dnsconn
);
312 // log_fn(LOG_DEBUG,"submitted '%s'", exitconn->address);
316 /** Remove <b>conn</b> from the list of connections waiting for conn-\>address.
318 void connection_dns_remove(connection_t
*conn
)
320 struct pending_connection_t
*pend
, *victim
;
321 struct cached_resolve search
;
322 struct cached_resolve
*resolve
;
324 tor_assert(conn
->type
== CONN_TYPE_EXIT
);
325 tor_assert(conn
->state
== EXIT_CONN_STATE_RESOLVING
);
327 strlcpy(search
.address
, conn
->address
, sizeof(search
.address
));
329 resolve
= SPLAY_FIND(cache_tree
, &cache_root
, &search
);
331 log_fn(LOG_NOTICE
,"Address '%s' is not pending. Dropping.", conn
->address
);
335 tor_assert(resolve
->pending_connections
);
336 assert_connection_ok(conn
,0);
338 pend
= resolve
->pending_connections
;
340 if (pend
->conn
== conn
) {
341 resolve
->pending_connections
= pend
->next
;
343 log_fn(LOG_DEBUG
, "First connection (fd %d) no longer waiting for resolve of '%s'",
344 conn
->s
, conn
->address
);
347 for ( ; pend
->next
; pend
= pend
->next
) {
348 if (pend
->next
->conn
== conn
) {
350 pend
->next
= victim
->next
;
352 log_fn(LOG_DEBUG
, "Connection (fd %d) no longer waiting for resolve of '%s'",
353 conn
->s
, conn
->address
);
354 return; /* more are pending */
357 tor_assert(0); /* not reachable unless onlyconn not in pending list */
361 /** Log an error and abort if conn is waiting for a DNS resolve.
363 void assert_connection_edge_not_dns_pending(connection_t
*conn
) {
364 struct pending_connection_t
*pend
;
365 struct cached_resolve
*resolve
;
367 SPLAY_FOREACH(resolve
, cache_tree
, &cache_root
) {
368 for (pend
= resolve
->pending_connections
;
371 tor_assert(pend
->conn
!= conn
);
376 /** Log an error and abort if any connection waiting for a DNS resolve is
378 void assert_all_pending_dns_resolves_ok(void) {
379 struct pending_connection_t
*pend
;
380 struct cached_resolve
*resolve
;
382 SPLAY_FOREACH(resolve
, cache_tree
, &cache_root
) {
383 for (pend
= resolve
->pending_connections
;
386 assert_connection_ok(pend
->conn
, 0);
387 tor_assert(pend
->conn
->s
== -1);
388 tor_assert(!connection_in_array(pend
->conn
));
393 /** Mark all connections waiting for <b>address</b> for close. Then cancel
394 * the resolve for <b>address</b> itself, and remove any cached results for
395 * <b>address</b> from the cache.
397 void dns_cancel_pending_resolve(char *address
) {
398 struct pending_connection_t
*pend
;
399 struct cached_resolve search
;
400 struct cached_resolve
*resolve
;
401 connection_t
*pendconn
;
404 strlcpy(search
.address
, address
, sizeof(search
.address
));
406 resolve
= SPLAY_FIND(cache_tree
, &cache_root
, &search
);
408 log_fn(LOG_NOTICE
,"Address '%s' is not pending. Dropping.", address
);
412 if (!resolve
->pending_connections
) {
413 /* XXX this should never trigger, but sometimes it does */
414 log_fn(LOG_WARN
,"Bug: Address '%s' is pending but has no pending connections!", address
);
420 tor_assert(resolve
->pending_connections
);
422 /* mark all pending connections to fail */
423 log_fn(LOG_DEBUG
, "Failing all connections waiting on DNS resolve of '%s'",
425 while (resolve
->pending_connections
) {
426 pend
= resolve
->pending_connections
;
427 pend
->conn
->state
= EXIT_CONN_STATE_RESOLVEFAILED
;
428 pendconn
= pend
->conn
;
429 tor_assert(pendconn
->s
== -1);
430 if (!pendconn
->marked_for_close
) {
431 connection_edge_end(pendconn
, END_STREAM_REASON_RESOURCELIMIT
,
432 pendconn
->cpath_layer
);
434 circ
= circuit_get_by_conn(pendconn
);
436 circuit_detach_stream(circ
, pendconn
);
437 connection_free(pendconn
);
438 resolve
->pending_connections
= pend
->next
;
442 dns_purge_resolve(resolve
);
445 /** Remove <b>resolve</b> from the cache.
447 static void dns_purge_resolve(struct cached_resolve
*resolve
) {
448 struct cached_resolve
*tmp
;
450 /* remove resolve from the linked list */
451 if (resolve
== oldest_cached_resolve
) {
452 oldest_cached_resolve
= resolve
->next
;
453 if (oldest_cached_resolve
== NULL
)
454 newest_cached_resolve
= NULL
;
456 /* FFFF make it a doubly linked list if this becomes too slow */
457 for (tmp
=oldest_cached_resolve
; tmp
&& tmp
->next
!= resolve
; tmp
=tmp
->next
) ;
458 tor_assert(tmp
); /* it's got to be in the list, or we screwed up somewhere else */
459 tmp
->next
= resolve
->next
; /* unlink it */
461 if (newest_cached_resolve
== resolve
)
462 newest_cached_resolve
= tmp
;
465 /* remove resolve from the tree */
466 SPLAY_REMOVE(cache_tree
, &cache_root
, resolve
);
471 /** Called on the OR side when a DNS worker tells us the outcome of a DNS
472 * resolve: tell all pending connections about the result of the lookup, and
473 * cache the value. (<b>address</b> is a NUL-terminated string containing the
474 * address to look up; <b>addr</b> is an IPv4 address in host order;
475 * <b>outcome</b> is one of
476 * DNS_RESOLVE_{FAILED_TRANSIENT|FAILED_PERMANENT|SUCCEEDED}.
478 static void dns_found_answer(char *address
, uint32_t addr
, char outcome
) {
479 struct pending_connection_t
*pend
;
480 struct cached_resolve search
;
481 struct cached_resolve
*resolve
;
482 connection_t
*pendconn
;
485 strlcpy(search
.address
, address
, sizeof(search
.address
));
487 resolve
= SPLAY_FIND(cache_tree
, &cache_root
, &search
);
489 log_fn(LOG_INFO
,"Resolved unasked address '%s'; caching anyway.", address
);
490 resolve
= tor_malloc_zero(sizeof(struct cached_resolve
));
491 resolve
->state
= (outcome
== DNS_RESOLVE_SUCCEEDED
) ?
492 CACHE_STATE_VALID
: CACHE_STATE_FAILED
;
493 resolve
->addr
= addr
;
494 resolve
->expire
= time(NULL
) + MAX_DNS_ENTRY_AGE
;
495 insert_resolve(resolve
);
499 if (resolve
->state
!= CACHE_STATE_PENDING
) {
500 /* XXXX Maybe update addr? or check addr for consistency? Or let
501 * VALID replace FAILED? */
502 log_fn(LOG_NOTICE
, "Resolved '%s' which was already resolved; ignoring",
504 tor_assert(resolve
->pending_connections
== NULL
);
507 /* Removed this assertion: in fact, we'll sometimes get a double answer
508 * to the same question. This can happen when we ask one worker to resolve
509 * X.Y.Z., then we cancel the request, and then we ask another worker to
511 /* tor_assert(resolve->state == CACHE_STATE_PENDING); */
513 resolve
->addr
= addr
;
514 if (outcome
== DNS_RESOLVE_SUCCEEDED
)
515 resolve
->state
= CACHE_STATE_VALID
;
517 resolve
->state
= CACHE_STATE_FAILED
;
519 while (resolve
->pending_connections
) {
520 pend
= resolve
->pending_connections
;
521 assert_connection_ok(pend
->conn
,time(NULL
));
522 pend
->conn
->addr
= resolve
->addr
;
523 pendconn
= pend
->conn
; /* don't pass complex things to the
524 connection_mark_for_close macro */
526 if (resolve
->state
== CACHE_STATE_FAILED
) {
527 /* prevent double-remove. */
528 pendconn
->state
= EXIT_CONN_STATE_RESOLVEFAILED
;
529 if (pendconn
->purpose
== EXIT_PURPOSE_CONNECT
) {
530 connection_edge_end(pendconn
, END_STREAM_REASON_RESOLVEFAILED
, pendconn
->cpath_layer
);
531 /* This detach must happen after we send the end cell. */
532 circuit_detach_stream(circuit_get_by_conn(pendconn
), pendconn
);
534 send_resolved_cell(pendconn
, RESOLVED_TYPE_ERROR
);
535 /* This detach must happen after we send the resolved cell. */
536 circuit_detach_stream(circuit_get_by_conn(pendconn
), pendconn
);
538 connection_free(pendconn
);
540 if (pendconn
->purpose
== EXIT_PURPOSE_CONNECT
) {
541 /* prevent double-remove. */
542 pend
->conn
->state
= EXIT_CONN_STATE_CONNECTING
;
544 circ
= circuit_get_by_conn(pend
->conn
);
546 /* unlink pend->conn from resolving_streams, */
547 circuit_detach_stream(circ
, pend
->conn
);
548 /* and link it to n_streams */
549 pend
->conn
->next_stream
= circ
->n_streams
;
550 circ
->n_streams
= pend
->conn
;
552 connection_exit_connect(pend
->conn
);
554 /* prevent double-remove. This isn't really an accurate state,
555 * but it does the right thing. */
556 pendconn
->state
= EXIT_CONN_STATE_RESOLVEFAILED
;
557 send_resolved_cell(pendconn
, RESOLVED_TYPE_IPV4
);
558 circ
= circuit_get_by_conn(pendconn
);
560 circuit_detach_stream(circ
, pendconn
);
561 connection_free(pendconn
);
564 resolve
->pending_connections
= pend
->next
;
568 if (outcome
== DNS_RESOLVE_FAILED_TRANSIENT
) { /* remove from cache */
569 dns_purge_resolve(resolve
);
573 /******************************************************************/
576 * Connection between OR and dnsworker
579 /** Write handler: called when we've pushed a request to a dnsworker. */
580 int connection_dns_finished_flushing(connection_t
*conn
) {
582 tor_assert(conn
->type
== CONN_TYPE_DNSWORKER
);
583 connection_stop_writing(conn
);
587 int connection_dns_reached_eof(connection_t
*conn
) {
588 log_fn(LOG_WARN
,"Read eof. Worker died unexpectedly.");
589 if (conn
->state
== DNSWORKER_STATE_BUSY
) {
590 /* don't cancel the resolve here -- it would be cancelled in
591 * connection_about_to_close_connection(), since conn is still
594 num_dnsworkers_busy
--;
597 connection_mark_for_close(conn
);
601 /** Read handler: called when we get data from a dnsworker. See
602 * if we have a complete answer. If so, call dns_found_answer on the
603 * result. If not, wait. Returns 0. */
604 int connection_dns_process_inbuf(connection_t
*conn
) {
609 tor_assert(conn
->type
== CONN_TYPE_DNSWORKER
);
611 if (conn
->state
!= DNSWORKER_STATE_BUSY
&& buf_datalen(conn
->inbuf
)) {
612 log_fn(LOG_WARN
,"Bug: read data from an idle dns worker. Please report.");
618 if (buf_datalen(conn
->inbuf
) < 5) /* entire answer available? */
619 return 0; /* not yet */
620 tor_assert(conn
->state
== DNSWORKER_STATE_BUSY
);
621 tor_assert(buf_datalen(conn
->inbuf
) == 5);
623 connection_fetch_from_buf(&success
,1,conn
);
624 connection_fetch_from_buf((char *)&addr
,sizeof(uint32_t),conn
);
626 log_fn(LOG_DEBUG
, "DNSWorker (fd %d) returned answer for '%s'",
627 conn
->s
, conn
->address
);
629 tor_assert(success
>= DNS_RESOLVE_FAILED_TRANSIENT
);
630 tor_assert(success
<= DNS_RESOLVE_SUCCEEDED
);
631 dns_found_answer(conn
->address
, ntohl(addr
), success
);
633 tor_free(conn
->address
);
634 conn
->address
= tor_strdup("<idle>");
635 conn
->state
= DNSWORKER_STATE_IDLE
;
636 num_dnsworkers_busy
--;
637 if (conn
->timestamp_created
< last_rotation_time
) {
638 connection_mark_for_close(conn
);
640 spawn_enough_dnsworkers();
645 /** Close and re-open all idle dnsworkers; schedule busy ones to be closed
646 * and re-opened once they're no longer busy.
648 void dnsworkers_rotate(void)
650 connection_t
*dnsconn
;
651 log_fn(LOG_INFO
, "Rotating DNS workers.");
652 while ((dnsconn
= connection_get_by_type_state(CONN_TYPE_DNSWORKER
,
653 DNSWORKER_STATE_IDLE
))) {
654 connection_mark_for_close(dnsconn
);
657 last_rotation_time
= time(NULL
);
658 spawn_enough_dnsworkers();
661 /** Implementation for DNS workers; this code runs in a separate
662 * execution context. It takes as its argument an fdarray as returned
663 * by socketpair(), and communicates via fdarray[1]. The protocol is
666 * - ADDRESSLEN [1 byte]
667 * - ADDRESS [ADDRESSLEN bytes]
668 * - The DNS worker does the lookup, and replies:
672 * OUTCOME is one of DNS_RESOLVE_{FAILED_TRANSIENT|FAILED_PERMANENT|SUCCEEDED}.
673 * IP is in host order.
675 * The dnsworker runs indefinitely, until its connection is closed or an error
678 static int dnsworker_main(void *data
) {
679 char address
[MAX_ADDRESSLEN
];
680 unsigned char address_len
;
687 /* log_fn(LOG_NOTICE,"After spawn: fdarray @%d has %d:%d", (int)fdarray, fdarray[0],fdarray[1]); */
689 fd
= fdarray
[1]; /* this side is ours */
690 #ifndef TOR_IS_MULTITHREADED
691 tor_close_socket(fdarray
[0]); /* this is the side of the socketpair the parent uses */
692 connection_free_all(); /* so the child doesn't hold the parent's fd's open */
693 handle_signals(0); /* ignore interrupts from the keyboard, etc */
700 if ((r
= recv(fd
, &address_len
, 1, 0)) != 1) {
702 log_fn(LOG_INFO
,"DNS worker exiting because Tor process closed connection (either pruned idle dnsworker or died).");
704 log_fn(LOG_INFO
,"DNS worker exiting because of error on connection to Tor process.");
705 log_fn(LOG_INFO
,"(Error on %d was %s)", fd
, tor_socket_strerror(tor_socket_errno(fd
)));
710 if (address_len
&& read_all(fd
, address
, address_len
, 1) != address_len
) {
711 log_fn(LOG_ERR
,"read hostname failed. Child exiting.");
714 address
[address_len
] = 0; /* null terminate it */
716 result
= tor_lookup_hostname(address
, &ip
);
717 /* Make 0.0.0.0 an error, so that we can use "0" to mean "no addr") */
722 /* XXX result can never be 1, because we set it to -1 above on error */
723 log_fn(LOG_INFO
,"Could not resolve dest addr %s (transient).",address
);
724 answer
[0] = DNS_RESOLVE_FAILED_TRANSIENT
;
727 log_fn(LOG_INFO
,"Could not resolve dest addr %s (permanent).",address
);
728 answer
[0] = DNS_RESOLVE_FAILED_PERMANENT
;
731 log_fn(LOG_INFO
,"Resolved address '%s'.",address
);
732 answer
[0] = DNS_RESOLVE_SUCCEEDED
;
735 set_uint32(answer
+1, ip
);
736 if (write_all(fd
, answer
, 5, 1) != 5) {
737 log_fn(LOG_ERR
,"writing answer failed. Child exiting.");
741 return 0; /* windows wants this function to return an int */
744 /** Launch a new DNS worker; return 0 on success, -1 on failure.
746 static int spawn_dnsworker(void) {
751 fdarray
= tor_malloc(sizeof(int)*2);
752 if (tor_socketpair(AF_UNIX
, SOCK_STREAM
, 0, fdarray
) < 0) {
753 log(LOG_ERR
, "Couldn't construct socketpair: %s",
754 tor_socket_strerror(tor_socket_errno(-1)));
760 /* log_fn(LOG_NOTICE,"Before spawn: fdarray @%d has %d:%d", (int)fdarray, fdarray[0],fdarray[1]); */
762 fd
= fdarray
[0]; /* We copy this out here, since dnsworker_main may free fdarray */
763 spawn_func(dnsworker_main
, (void*)fdarray
);
764 log_fn(LOG_DEBUG
,"just spawned a worker.");
765 #ifndef TOR_IS_MULTITHREADED
766 tor_close_socket(fdarray
[1]); /* we don't need the worker's side of the pipe */
770 conn
= connection_new(CONN_TYPE_DNSWORKER
);
772 set_socket_nonblocking(fd
);
774 /* set up conn so it's got all the data we need to remember */
776 conn
->address
= tor_strdup("<unused>");
778 if (connection_add(conn
) < 0) { /* no space, forget it */
779 log_fn(LOG_WARN
,"connection_add failed. Giving up.");
780 connection_free(conn
); /* this closes fd */
784 conn
->state
= DNSWORKER_STATE_IDLE
;
785 connection_start_reading(conn
);
787 return 0; /* success */
790 /** If we have too many or too few DNS workers, spawn or kill some.
792 static void spawn_enough_dnsworkers(void) {
793 int num_dnsworkers_needed
; /* aim to have 1 more than needed,
794 * but no less than min and no more than max */
795 connection_t
*dnsconn
;
797 /* XXX This may not be the best strategy. Maybe we should queue pending
798 * requests until the old ones finish or time out: otherwise, if
799 * the connection requests come fast enough, we never get any DNS done. -NM
800 * XXX But if we queue them, then the adversary can pile even more
801 * queries onto us, blocking legitimate requests for even longer.
802 * Maybe we should compromise and only kill if it's been at it for
803 * more than, e.g., 2 seconds. -RD
805 if (num_dnsworkers_busy
== MAX_DNSWORKERS
) {
806 /* We always want at least one worker idle.
807 * So find the oldest busy worker and kill it.
809 dnsconn
= connection_get_by_type_state_lastwritten(CONN_TYPE_DNSWORKER
,
810 DNSWORKER_STATE_BUSY
);
813 log_fn(LOG_WARN
, "%d DNS workers are spawned; all are busy. Killing one.",
816 connection_mark_for_close(dnsconn
);
817 num_dnsworkers_busy
--;
821 if (num_dnsworkers_busy
>= MIN_DNSWORKERS
)
822 num_dnsworkers_needed
= num_dnsworkers_busy
+1;
824 num_dnsworkers_needed
= MIN_DNSWORKERS
;
826 while (num_dnsworkers
< num_dnsworkers_needed
) {
827 if (spawn_dnsworker() < 0) {
828 log(LOG_WARN
,"spawn_enough_dnsworkers(): spawn failed!");
834 while (num_dnsworkers
> num_dnsworkers_busy
+MAX_IDLE_DNSWORKERS
) { /* too many idle? */
835 /* cull excess workers */
836 log_fn(LOG_NOTICE
,"%d of %d dnsworkers are idle. Killing one.",
837 num_dnsworkers
-num_dnsworkers_needed
, num_dnsworkers
);
838 dnsconn
= connection_get_by_type_state(CONN_TYPE_DNSWORKER
, DNSWORKER_STATE_IDLE
);
840 connection_mark_for_close(dnsconn
);