?  search:
summary | log | graphiclog1 | graphiclog2 | commit | commitdiff | tree | refs | edit | fork
blame | history | raw | HEAD
Patrick Welche <prlw1@cam.ac.uk>
[netbsd-mini2440.git] / external / bsd / ntp / dist / lib / isc / unix / socket.c
blobc656c9a9304020547681b5b28dc437740fb12933
1 /* $NetBSD$ */
2
3 /*
4 * Copyright (C) 2004-2009 Internet Systems Consortium, Inc. ("ISC")
5 * Copyright (C) 1998-2003 Internet Software Consortium.
6 *
7 * Permission to use, copy, modify, and/or distribute this software for any
8 * purpose with or without fee is hereby granted, provided that the above
9 * copyright notice and this permission notice appear in all copies.
10 *
11 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
12 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
13 * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
14 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
15 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
16 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
17 * PERFORMANCE OF THIS SOFTWARE.
18 */
19
20 /* Id: socket.c,v 1.308.12.8 2009/04/18 01:29:26 jinmei Exp */
21
22 /*! \file */
23
24 #include <config.h>
25
26 #include <sys/param.h>
27 #include <sys/types.h>
28 #include <sys/socket.h>
29 #include <sys/stat.h>
30 #include <sys/time.h>
31 #include <sys/uio.h>
32
33 #include <errno.h>
34 #include <fcntl.h>
35 #include <stddef.h>
36 #include <stdlib.h>
37 #include <string.h>
38 #include <unistd.h>
39
40 #include <isc/buffer.h>
41 #include <isc/bufferlist.h>
42 #include <isc/condition.h>
43 #include <isc/formatcheck.h>
44 #include <isc/list.h>
45 #include <isc/log.h>
46 #include <isc/mem.h>
47 #include <isc/msgs.h>
48 #include <isc/mutex.h>
49 #include <isc/net.h>
50 #include <isc/once.h>
51 #include <isc/platform.h>
52 #include <isc/print.h>
53 #include <isc/region.h>
54 #include <isc/socket.h>
55 #include <isc/stats.h>
56 #include <isc/strerror.h>
57 #include <isc/task.h>
58 #include <isc/thread.h>
59 #include <isc/util.h>
60 #include <isc/xml.h>
61
62 #ifdef ISC_PLATFORM_HAVESYSUNH
63 #include <sys/un.h>
64 #endif
65 #ifdef ISC_PLATFORM_HAVEKQUEUE
66 #include <sys/event.h>
67 #endif
68 #ifdef ISC_PLATFORM_HAVEEPOLL
69 #include <sys/epoll.h>
70 #endif
71 #ifdef ISC_PLATFORM_HAVEDEVPOLL
72 #include <sys/devpoll.h>
73 #endif
74
75 #include "errno2result.h"
76
77 #ifndef ISC_PLATFORM_USETHREADS
78 #include "socket_p.h"
79 #endif /* ISC_PLATFORM_USETHREADS */
80
81 #if defined(SO_BSDCOMPAT) && defined(__linux__)
82 #include <sys/utsname.h>
83 #endif
84
85 /*%
86 * Choose the most preferable multiplex method.
87 */
88 #ifdef ISC_PLATFORM_HAVEKQUEUE
89 #define USE_KQUEUE
90 #elif defined (ISC_PLATFORM_HAVEEPOLL)
91 #define USE_EPOLL
92 #elif defined (ISC_PLATFORM_HAVEDEVPOLL)
93 #define USE_DEVPOLL
94 typedef struct {
95 unsigned int want_read : 1,
96 want_write : 1;
97 } pollinfo_t;
98 #else
99 #define USE_SELECT
100 #endif /* ISC_PLATFORM_HAVEKQUEUE */
101
102 #ifndef ISC_PLATFORM_USETHREADS
103 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
104 struct isc_socketwait {
105 int nevents;
106 };
107 #elif defined (USE_SELECT)
108 struct isc_socketwait {
109 fd_set *readset;
110 fd_set *writeset;
111 int nfds;
112 int maxfd;
113 };
114 #endif /* USE_KQUEUE */
115 #endif /* !ISC_PLATFORM_USETHREADS */
116
117 /*%
118 * Maximum number of allowable open sockets. This is also the maximum
119 * allowable socket file descriptor.
120 *
121 * Care should be taken before modifying this value for select():
122 * The API standard doesn't ensure select() accept more than (the system default
123 * of) FD_SETSIZE descriptors, and the default size should in fact be fine in
124 * the vast majority of cases. This constant should therefore be increased only
125 * when absolutely necessary and possible, i.e., the server is exhausting all
126 * available file descriptors (up to FD_SETSIZE) and the select() function
127 * and FD_xxx macros support larger values than FD_SETSIZE (which may not
128 * always by true, but we keep using some of them to ensure as much
129 * portability as possible). Note also that overall server performance
130 * may be rather worsened with a larger value of this constant due to
131 * inherent scalability problems of select().
132 *
133 * As a special note, this value shouldn't have to be touched if
134 * this is a build for an authoritative only DNS server.
135 */
136 #ifndef ISC_SOCKET_MAXSOCKETS
137 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
138 #define ISC_SOCKET_MAXSOCKETS 4096
139 #elif defined(USE_SELECT)
140 #define ISC_SOCKET_MAXSOCKETS FD_SETSIZE
141 #endif /* USE_KQUEUE... */
142 #endif /* ISC_SOCKET_MAXSOCKETS */
143
144 #ifdef USE_SELECT
145 /*%
146 * Mac OS X needs a special definition to support larger values in select().
147 * We always define this because a larger value can be specified run-time.
148 */
149 #ifdef __APPLE__
150 #define _DARWIN_UNLIMITED_SELECT
151 #endif /* __APPLE__ */
152 #endif /* USE_SELECT */
153
154 #ifdef ISC_SOCKET_USE_POLLWATCH
155 /*%
156 * If this macro is defined, enable workaround for a Solaris /dev/poll kernel
157 * bug: DP_POLL ioctl could keep sleeping even if socket I/O is possible for
158 * some of the specified FD. The idea is based on the observation that it's
159 * likely for a busy server to keep receiving packets. It specifically works
160 * as follows: the socket watcher is first initialized with the state of
161 * "poll_idle". While it's in the idle state it keeps sleeping until a socket
162 * event occurs. When it wakes up for a socket I/O event, it moves to the
163 * poll_active state, and sets the poll timeout to a short period
164 * (ISC_SOCKET_POLLWATCH_TIMEOUT msec). If timeout occurs in this state, the
165 * watcher goes to the poll_checking state with the same timeout period.
166 * In this state, the watcher tries to detect whether this is a break
167 * during intermittent events or the kernel bug is triggered. If the next
168 * polling reports an event within the short period, the previous timeout is
169 * likely to be a kernel bug, and so the watcher goes back to the active state.
170 * Otherwise, it moves to the idle state again.
171 *
172 * It's not clear whether this is a thread-related bug, but since we've only
173 * seen this with threads, this workaround is used only when enabling threads.
174 */
175
176 typedef enum { poll_idle, poll_active, poll_checking } pollstate_t;
177
178 #ifndef ISC_SOCKET_POLLWATCH_TIMEOUT
179 #define ISC_SOCKET_POLLWATCH_TIMEOUT 10
180 #endif /* ISC_SOCKET_POLLWATCH_TIMEOUT */
181 #endif /* ISC_SOCKET_USE_POLLWATCH */
182
183 /*%
184 * Size of per-FD lock buckets.
185 */
186 #ifdef ISC_PLATFORM_USETHREADS
187 #define FDLOCK_COUNT 1024
188 #define FDLOCK_ID(fd) ((fd) % FDLOCK_COUNT)
189 #else
190 #define FDLOCK_COUNT 1
191 #define FDLOCK_ID(fd) 0
192 #endif /* ISC_PLATFORM_USETHREADS */
193
194 /*%
195 * Maximum number of events communicated with the kernel. There should normally
196 * be no need for having a large number.
197 */
198 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
199 #ifndef ISC_SOCKET_MAXEVENTS
200 #define ISC_SOCKET_MAXEVENTS 64
201 #endif
202 #endif
203
204 /*%
205 * Some systems define the socket length argument as an int, some as size_t,
206 * some as socklen_t. This is here so it can be easily changed if needed.
207 */
208 #ifndef ISC_SOCKADDR_LEN_T
209 #define ISC_SOCKADDR_LEN_T unsigned int
210 #endif
211
212 /*%
213 * Define what the possible "soft" errors can be. These are non-fatal returns
214 * of various network related functions, like recv() and so on.
215 *
216 * For some reason, BSDI (and perhaps others) will sometimes return <0
217 * from recv() but will have errno==0. This is broken, but we have to
218 * work around it here.
219 */
220 #define SOFT_ERROR(e) ((e) == EAGAIN || \
221 (e) == EWOULDBLOCK || \
222 (e) == EINTR || \
223 (e) == 0)
224
225 #define DLVL(x) ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_SOCKET, ISC_LOG_DEBUG(x)
226
227 /*!<
228 * DLVL(90) -- Function entry/exit and other tracing.
229 * DLVL(70) -- Socket "correctness" -- including returning of events, etc.
230 * DLVL(60) -- Socket data send/receive
231 * DLVL(50) -- Event tracing, including receiving/sending completion events.
232 * DLVL(20) -- Socket creation/destruction.
233 */
234 #define TRACE_LEVEL 90
235 #define CORRECTNESS_LEVEL 70
236 #define IOEVENT_LEVEL 60
237 #define EVENT_LEVEL 50
238 #define CREATION_LEVEL 20
239
240 #define TRACE DLVL(TRACE_LEVEL)
241 #define CORRECTNESS DLVL(CORRECTNESS_LEVEL)
242 #define IOEVENT DLVL(IOEVENT_LEVEL)
243 #define EVENT DLVL(EVENT_LEVEL)
244 #define CREATION DLVL(CREATION_LEVEL)
245
246 typedef isc_event_t intev_t;
247
248 #define SOCKET_MAGIC ISC_MAGIC('I', 'O', 'i', 'o')
249 #define VALID_SOCKET(t) ISC_MAGIC_VALID(t, SOCKET_MAGIC)
250
251 /*!
252 * IPv6 control information. If the socket is an IPv6 socket we want
253 * to collect the destination address and interface so the client can
254 * set them on outgoing packets.
255 */
256 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
257 #ifndef USE_CMSG
258 #define USE_CMSG 1
259 #endif
260 #endif
261
262 /*%
263 * NetBSD and FreeBSD can timestamp packets. XXXMLG Should we have
264 * a setsockopt() like interface to request timestamps, and if the OS
265 * doesn't do it for us, call gettimeofday() on every UDP receive?
266 */
267 #ifdef SO_TIMESTAMP
268 #ifndef USE_CMSG
269 #define USE_CMSG 1
270 #endif
271 #endif
272
273 /*%
274 * The size to raise the receive buffer to (from BIND 8).
275 */
276 #define RCVBUFSIZE (32*1024)
277
278 /*%
279 * The number of times a send operation is repeated if the result is EINTR.
280 */
281 #define NRETRIES 10
282
283 struct isc_socket {
284 /* Not locked. */
285 unsigned int magic;
286 isc_socketmgr_t *manager;
287 isc_mutex_t lock;
288 isc_sockettype_t type;
289 const isc_statscounter_t *statsindex;
290
291 /* Locked by socket lock. */
292 ISC_LINK(isc_socket_t) link;
293 unsigned int references;
294 int fd;
295 int pf;
296 char name[16];
297 void * tag;
298
299 ISC_LIST(isc_socketevent_t) send_list;
300 ISC_LIST(isc_socketevent_t) recv_list;
301 ISC_LIST(isc_socket_newconnev_t) accept_list;
302 isc_socket_connev_t *connect_ev;
303
304 /*
305 * Internal events. Posted when a descriptor is readable or
306 * writable. These are statically allocated and never freed.
307 * They will be set to non-purgable before use.
308 */
309 intev_t readable_ev;
310 intev_t writable_ev;
311
312 isc_sockaddr_t peer_address; /* remote address */
313
314 unsigned int pending_recv : 1,
315 pending_send : 1,
316 pending_accept : 1,
317 listener : 1, /* listener socket */
318 connected : 1,
319 connecting : 1, /* connect pending */
320 bound : 1; /* bound to local addr */
321
322 #ifdef ISC_NET_RECVOVERFLOW
323 unsigned char overflow; /* used for MSG_TRUNC fake */
324 #endif
325
326 char *recvcmsgbuf;
327 ISC_SOCKADDR_LEN_T recvcmsgbuflen;
328 char *sendcmsgbuf;
329 ISC_SOCKADDR_LEN_T sendcmsgbuflen;
330
331 void *fdwatcharg;
332 isc_sockfdwatch_t fdwatchcb;
333 int fdwatchflags;
334 isc_task_t *fdwatchtask;
335 };
336
337 #define SOCKET_MANAGER_MAGIC ISC_MAGIC('I', 'O', 'm', 'g')
338 #define VALID_MANAGER(m) ISC_MAGIC_VALID(m, SOCKET_MANAGER_MAGIC)
339
340 struct isc_socketmgr {
341 /* Not locked. */
342 unsigned int magic;
343 isc_mem_t *mctx;
344 isc_mutex_t lock;
345 isc_mutex_t *fdlock;
346 isc_stats_t *stats;
347 #ifdef USE_KQUEUE
348 int kqueue_fd;
349 int nevents;
350 struct kevent *events;
351 #endif /* USE_KQUEUE */
352 #ifdef USE_EPOLL
353 int epoll_fd;
354 int nevents;
355 struct epoll_event *events;
356 #endif /* USE_EPOLL */
357 #ifdef USE_DEVPOLL
358 int devpoll_fd;
359 int nevents;
360 struct pollfd *events;
361 #endif /* USE_DEVPOLL */
362 #ifdef USE_SELECT
363 int fd_bufsize;
364 #endif /* USE_SELECT */
365 unsigned int maxsocks;
366 #ifdef ISC_PLATFORM_USETHREADS
367 int pipe_fds[2];
368 #endif
369
370 /* Locked by fdlock. */
371 isc_socket_t **fds;
372 int *fdstate;
373 #ifdef USE_DEVPOLL
374 pollinfo_t *fdpollinfo;
375 #endif
376
377 /* Locked by manager lock. */
378 ISC_LIST(isc_socket_t) socklist;
379 #ifdef USE_SELECT
380 fd_set *read_fds;
381 fd_set *read_fds_copy;
382 fd_set *write_fds;
383 fd_set *write_fds_copy;
384 int maxfd;
385 #endif /* USE_SELECT */
386 int reserved; /* unlocked */
387 #ifdef ISC_PLATFORM_USETHREADS
388 isc_thread_t watcher;
389 isc_condition_t shutdown_ok;
390 #else /* ISC_PLATFORM_USETHREADS */
391 unsigned int refs;
392 #endif /* ISC_PLATFORM_USETHREADS */
393 };
394
395 #ifndef ISC_PLATFORM_USETHREADS
396 static isc_socketmgr_t *socketmgr = NULL;
397 #endif /* ISC_PLATFORM_USETHREADS */
398
399 #define CLOSED 0 /* this one must be zero */
400 #define MANAGED 1
401 #define CLOSE_PENDING 2
402
403 /*
404 * send() and recv() iovec counts
405 */
406 #define MAXSCATTERGATHER_SEND (ISC_SOCKET_MAXSCATTERGATHER)
407 #ifdef ISC_NET_RECVOVERFLOW
408 # define MAXSCATTERGATHER_RECV (ISC_SOCKET_MAXSCATTERGATHER + 1)
409 #else
410 # define MAXSCATTERGATHER_RECV (ISC_SOCKET_MAXSCATTERGATHER)
411 #endif
412
413 static void send_recvdone_event(isc_socket_t *, isc_socketevent_t **);
414 static void send_senddone_event(isc_socket_t *, isc_socketevent_t **);
415 static void free_socket(isc_socket_t **);
416 static isc_result_t allocate_socket(isc_socketmgr_t *, isc_sockettype_t,
417 isc_socket_t **);
418 static void destroy(isc_socket_t **);
419 static void internal_accept(isc_task_t *, isc_event_t *);
420 static void internal_connect(isc_task_t *, isc_event_t *);
421 static void internal_recv(isc_task_t *, isc_event_t *);
422 static void internal_send(isc_task_t *, isc_event_t *);
423 static void internal_fdwatch_write(isc_task_t *, isc_event_t *);
424 static void internal_fdwatch_read(isc_task_t *, isc_event_t *);
425 static void process_cmsg(isc_socket_t *, struct msghdr *, isc_socketevent_t *);
426 static void build_msghdr_send(isc_socket_t *, isc_socketevent_t *,
427 struct msghdr *, struct iovec *, size_t *);
428 static void build_msghdr_recv(isc_socket_t *, isc_socketevent_t *,
429 struct msghdr *, struct iovec *, size_t *);
430 #ifdef ISC_PLATFORM_USETHREADS
431 static isc_boolean_t process_ctlfd(isc_socketmgr_t *manager);
432 #endif
433
434 #define SELECT_POKE_SHUTDOWN (-1)
435 #define SELECT_POKE_NOTHING (-2)
436 #define SELECT_POKE_READ (-3)
437 #define SELECT_POKE_ACCEPT (-3) /*%< Same as _READ */
438 #define SELECT_POKE_WRITE (-4)
439 #define SELECT_POKE_CONNECT (-4) /*%< Same as _WRITE */
440 #define SELECT_POKE_CLOSE (-5)
441
442 #define SOCK_DEAD(s) ((s)->references == 0)
443
444 /*%
445 * Shortcut index arrays to get access to statistics counters.
446 */
447 enum {
448 STATID_OPEN = 0,
449 STATID_OPENFAIL = 1,
450 STATID_CLOSE = 2,
451 STATID_BINDFAIL = 3,
452 STATID_CONNECTFAIL = 4,
453 STATID_CONNECT = 5,
454 STATID_ACCEPTFAIL = 6,
455 STATID_ACCEPT = 7,
456 STATID_SENDFAIL = 8,
457 STATID_RECVFAIL = 9
458 };
459 static const isc_statscounter_t upd4statsindex[] = {
460 isc_sockstatscounter_udp4open,
461 isc_sockstatscounter_udp4openfail,
462 isc_sockstatscounter_udp4close,
463 isc_sockstatscounter_udp4bindfail,
464 isc_sockstatscounter_udp4connectfail,
465 isc_sockstatscounter_udp4connect,
466 -1,
467 -1,
468 isc_sockstatscounter_udp4sendfail,
469 isc_sockstatscounter_udp4recvfail
470 };
471 static const isc_statscounter_t upd6statsindex[] = {
472 isc_sockstatscounter_udp6open,
473 isc_sockstatscounter_udp6openfail,
474 isc_sockstatscounter_udp6close,
475 isc_sockstatscounter_udp6bindfail,
476 isc_sockstatscounter_udp6connectfail,
477 isc_sockstatscounter_udp6connect,
478 -1,
479 -1,
480 isc_sockstatscounter_udp6sendfail,
481 isc_sockstatscounter_udp6recvfail
482 };
483 static const isc_statscounter_t tcp4statsindex[] = {
484 isc_sockstatscounter_tcp4open,
485 isc_sockstatscounter_tcp4openfail,
486 isc_sockstatscounter_tcp4close,
487 isc_sockstatscounter_tcp4bindfail,
488 isc_sockstatscounter_tcp4connectfail,
489 isc_sockstatscounter_tcp4connect,
490 isc_sockstatscounter_tcp4acceptfail,
491 isc_sockstatscounter_tcp4accept,
492 isc_sockstatscounter_tcp4sendfail,
493 isc_sockstatscounter_tcp4recvfail
494 };
495 static const isc_statscounter_t tcp6statsindex[] = {
496 isc_sockstatscounter_tcp6open,
497 isc_sockstatscounter_tcp6openfail,
498 isc_sockstatscounter_tcp6close,
499 isc_sockstatscounter_tcp6bindfail,
500 isc_sockstatscounter_tcp6connectfail,
501 isc_sockstatscounter_tcp6connect,
502 isc_sockstatscounter_tcp6acceptfail,
503 isc_sockstatscounter_tcp6accept,
504 isc_sockstatscounter_tcp6sendfail,
505 isc_sockstatscounter_tcp6recvfail
506 };
507 static const isc_statscounter_t unixstatsindex[] = {
508 isc_sockstatscounter_unixopen,
509 isc_sockstatscounter_unixopenfail,
510 isc_sockstatscounter_unixclose,
511 isc_sockstatscounter_unixbindfail,
512 isc_sockstatscounter_unixconnectfail,
513 isc_sockstatscounter_unixconnect,
514 isc_sockstatscounter_unixacceptfail,
515 isc_sockstatscounter_unixaccept,
516 isc_sockstatscounter_unixsendfail,
517 isc_sockstatscounter_unixrecvfail
518 };
519 static const isc_statscounter_t fdwatchstatsindex[] = {
520 -1,
521 -1,
522 isc_sockstatscounter_fdwatchclose,
523 isc_sockstatscounter_fdwatchbindfail,
524 isc_sockstatscounter_fdwatchconnectfail,
525 isc_sockstatscounter_fdwatchconnect,
526 -1,
527 -1,
528 isc_sockstatscounter_fdwatchsendfail,
529 isc_sockstatscounter_fdwatchrecvfail
530 };
531
532 static void
533 manager_log(isc_socketmgr_t *sockmgr,
534 isc_logcategory_t *category, isc_logmodule_t *module, int level,
535 const char *fmt, ...) ISC_FORMAT_PRINTF(5, 6);
536 static void
537 manager_log(isc_socketmgr_t *sockmgr,
538 isc_logcategory_t *category, isc_logmodule_t *module, int level,
539 const char *fmt, ...)
540 {
541 char msgbuf[2048];
542 va_list ap;
543
544 if (! isc_log_wouldlog(isc_lctx, level))
545 return;
546
547 va_start(ap, fmt);
548 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
549 va_end(ap);
550
551 isc_log_write(isc_lctx, category, module, level,
552 "sockmgr %p: %s", sockmgr, msgbuf);
553 }
554
555 static void
556 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
557 isc_logcategory_t *category, isc_logmodule_t *module, int level,
558 isc_msgcat_t *msgcat, int msgset, int message,
559 const char *fmt, ...) ISC_FORMAT_PRINTF(9, 10);
560 static void
561 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
562 isc_logcategory_t *category, isc_logmodule_t *module, int level,
563 isc_msgcat_t *msgcat, int msgset, int message,
564 const char *fmt, ...)
565 {
566 char msgbuf[2048];
567 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
568 va_list ap;
569
570 if (! isc_log_wouldlog(isc_lctx, level))
571 return;
572
573 va_start(ap, fmt);
574 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
575 va_end(ap);
576
577 if (address == NULL) {
578 isc_log_iwrite(isc_lctx, category, module, level,
579 msgcat, msgset, message,
580 "socket %p: %s", sock, msgbuf);
581 } else {
582 isc_sockaddr_format(address, peerbuf, sizeof(peerbuf));
583 isc_log_iwrite(isc_lctx, category, module, level,
584 msgcat, msgset, message,
585 "socket %p %s: %s", sock, peerbuf, msgbuf);
586 }
587 }
588
589 #if defined(_AIX) && defined(ISC_NET_BSD44MSGHDR) && \
590 defined(USE_CMSG) && defined(IPV6_RECVPKTINFO)
591 /*
592 * AIX has a kernel bug where IPV6_RECVPKTINFO gets cleared by
593 * setting IPV6_V6ONLY.
594 */
595 static void
596 FIX_IPV6_RECVPKTINFO(isc_socket_t *sock)
597 {
598 char strbuf[ISC_STRERRORSIZE];
599 int on = 1;
600
601 if (sock->pf != AF_INET6 || sock->type != isc_sockettype_udp)
602 return;
603
604 if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
605 (void *)&on, sizeof(on)) < 0) {
606
607 UNEXPECTED_ERROR(__FILE__, __LINE__,
608 "setsockopt(%d, IPV6_RECVPKTINFO) "
609 "%s: %s", sock->fd,
610 isc_msgcat_get(isc_msgcat,
611 ISC_MSGSET_GENERAL,
612 ISC_MSG_FAILED,
613 "failed"),
614 strbuf);
615 }
616 }
617 #else
618 #define FIX_IPV6_RECVPKTINFO(sock) (void)0
619 #endif
620
621 /*%
622 * Increment socket-related statistics counters.
623 */
624 static inline void
625 inc_stats(isc_stats_t *stats, isc_statscounter_t counterid) {
626 REQUIRE(counterid != -1);
627
628 if (stats != NULL)
629 isc_stats_increment(stats, counterid);
630 }
631
632 static inline isc_result_t
633 watch_fd(isc_socketmgr_t *manager, int fd, int msg) {
634 isc_result_t result = ISC_R_SUCCESS;
635
636 #ifdef USE_KQUEUE
637 struct kevent evchange;
638
639 memset(&evchange, 0, sizeof(evchange));
640 if (msg == SELECT_POKE_READ)
641 evchange.filter = EVFILT_READ;
642 else
643 evchange.filter = EVFILT_WRITE;
644 evchange.flags = EV_ADD;
645 evchange.ident = fd;
646 if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
647 result = isc__errno2result(errno);
648
649 return (result);
650 #elif defined(USE_EPOLL)
651 struct epoll_event event;
652
653 if (msg == SELECT_POKE_READ)
654 event.events = EPOLLIN;
655 else
656 event.events = EPOLLOUT;
657 event.data.fd = fd;
658 if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_ADD, fd, &event) == -1 &&
659 errno != EEXIST) {
660 result = isc__errno2result(errno);
661 }
662
663 return (result);
664 #elif defined(USE_DEVPOLL)
665 struct pollfd pfd;
666 int lockid = FDLOCK_ID(fd);
667
668 memset(&pfd, 0, sizeof(pfd));
669 if (msg == SELECT_POKE_READ)
670 pfd.events = POLLIN;
671 else
672 pfd.events = POLLOUT;
673 pfd.fd = fd;
674 pfd.revents = 0;
675 LOCK(&manager->fdlock[lockid]);
676 if (write(manager->devpoll_fd, &pfd, sizeof(pfd)) == -1)
677 result = isc__errno2result(errno);
678 else {
679 if (msg == SELECT_POKE_READ)
680 manager->fdpollinfo[fd].want_read = 1;
681 else
682 manager->fdpollinfo[fd].want_write = 1;
683 }
684 UNLOCK(&manager->fdlock[lockid]);
685
686 return (result);
687 #elif defined(USE_SELECT)
688 LOCK(&manager->lock);
689 if (msg == SELECT_POKE_READ)
690 FD_SET(fd, manager->read_fds);
691 if (msg == SELECT_POKE_WRITE)
692 FD_SET(fd, manager->write_fds);
693 UNLOCK(&manager->lock);
694
695 return (result);
696 #endif
697 }
698
699 static inline isc_result_t
700 unwatch_fd(isc_socketmgr_t *manager, int fd, int msg) {
701 isc_result_t result = ISC_R_SUCCESS;
702
703 #ifdef USE_KQUEUE
704 struct kevent evchange;
705
706 memset(&evchange, 0, sizeof(evchange));
707 if (msg == SELECT_POKE_READ)
708 evchange.filter = EVFILT_READ;
709 else
710 evchange.filter = EVFILT_WRITE;
711 evchange.flags = EV_DELETE;
712 evchange.ident = fd;
713 if (kevent(manager->kqueue_fd, &evchange, 1, NULL, 0, NULL) != 0)
714 result = isc__errno2result(errno);
715
716 return (result);
717 #elif defined(USE_EPOLL)
718 struct epoll_event event;
719
720 if (msg == SELECT_POKE_READ)
721 event.events = EPOLLIN;
722 else
723 event.events = EPOLLOUT;
724 event.data.fd = fd;
725 if (epoll_ctl(manager->epoll_fd, EPOLL_CTL_DEL, fd, &event) == -1 &&
726 errno != ENOENT) {
727 char strbuf[ISC_STRERRORSIZE];
728 isc__strerror(errno, strbuf, sizeof(strbuf));
729 UNEXPECTED_ERROR(__FILE__, __LINE__,
730 "epoll_ctl(DEL), %d: %s", fd, strbuf);
731 result = ISC_R_UNEXPECTED;
732 }
733 return (result);
734 #elif defined(USE_DEVPOLL)
735 struct pollfd pfds[2];
736 size_t writelen = sizeof(pfds[0]);
737 int lockid = FDLOCK_ID(fd);
738
739 memset(pfds, 0, sizeof(pfds));
740 pfds[0].events = POLLREMOVE;
741 pfds[0].fd = fd;
742
743 /*
744 * Canceling read or write polling via /dev/poll is tricky. Since it
745 * only provides a way of canceling per FD, we may need to re-poll the
746 * socket for the other operation.
747 */
748 LOCK(&manager->fdlock[lockid]);
749 if (msg == SELECT_POKE_READ &&
750 manager->fdpollinfo[fd].want_write == 1) {
751 pfds[1].events = POLLOUT;
752 pfds[1].fd = fd;
753 writelen += sizeof(pfds[1]);
754 }
755 if (msg == SELECT_POKE_WRITE &&
756 manager->fdpollinfo[fd].want_read == 1) {
757 pfds[1].events = POLLIN;
758 pfds[1].fd = fd;
759 writelen += sizeof(pfds[1]);
760 }
761
762 if (write(manager->devpoll_fd, pfds, writelen) == -1)
763 result = isc__errno2result(errno);
764 else {
765 if (msg == SELECT_POKE_READ)
766 manager->fdpollinfo[fd].want_read = 0;
767 else
768 manager->fdpollinfo[fd].want_write = 0;
769 }
770 UNLOCK(&manager->fdlock[lockid]);
771
772 return (result);
773 #elif defined(USE_SELECT)
774 LOCK(&manager->lock);
775 if (msg == SELECT_POKE_READ)
776 FD_CLR(fd, manager->read_fds);
777 else if (msg == SELECT_POKE_WRITE)
778 FD_CLR(fd, manager->write_fds);
779 UNLOCK(&manager->lock);
780
781 return (result);
782 #endif
783 }
784
785 static void
786 wakeup_socket(isc_socketmgr_t *manager, int fd, int msg) {
787 isc_result_t result;
788 int lockid = FDLOCK_ID(fd);
789
790 /*
791 * This is a wakeup on a socket. If the socket is not in the
792 * process of being closed, start watching it for either reads
793 * or writes.
794 */
795
796 INSIST(fd >= 0 && fd < (int)manager->maxsocks);
797
798 if (msg == SELECT_POKE_CLOSE) {
799 /* No one should be updating fdstate, so no need to lock it */
800 INSIST(manager->fdstate[fd] == CLOSE_PENDING);
801 manager->fdstate[fd] = CLOSED;
802 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
803 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
804 (void)close(fd);
805 return;
806 }
807
808 LOCK(&manager->fdlock[lockid]);
809 if (manager->fdstate[fd] == CLOSE_PENDING) {
810 UNLOCK(&manager->fdlock[lockid]);
811
812 /*
813 * We accept (and ignore) any error from unwatch_fd() as we are
814 * closing the socket, hoping it doesn't leave dangling state in
815 * the kernel.
816 * Note that unwatch_fd() must be called after releasing the
817 * fdlock; otherwise it could cause deadlock due to a lock order
818 * reversal.
819 */
820 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
821 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
822 return;
823 }
824 if (manager->fdstate[fd] != MANAGED) {
825 UNLOCK(&manager->fdlock[lockid]);
826 return;
827 }
828 UNLOCK(&manager->fdlock[lockid]);
829
830 /*
831 * Set requested bit.
832 */
833 result = watch_fd(manager, fd, msg);
834 if (result != ISC_R_SUCCESS) {
835 /*
836 * XXXJT: what should we do? Ignoring the failure of watching
837 * a socket will make the application dysfunctional, but there
838 * seems to be no reasonable recovery process.
839 */
840 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
841 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
842 "failed to start watching FD (%d): %s",
843 fd, isc_result_totext(result));
844 }
845 }
846
847 #ifdef ISC_PLATFORM_USETHREADS
848 /*
849 * Poke the select loop when there is something for us to do.
850 * The write is required (by POSIX) to complete. That is, we
851 * will not get partial writes.
852 */
853 static void
854 select_poke(isc_socketmgr_t *mgr, int fd, int msg) {
855 int cc;
856 int buf[2];
857 char strbuf[ISC_STRERRORSIZE];
858
859 buf[0] = fd;
860 buf[1] = msg;
861
862 do {
863 cc = write(mgr->pipe_fds[1], buf, sizeof(buf));
864 #ifdef ENOSR
865 /*
866 * Treat ENOSR as EAGAIN but loop slowly as it is
867 * unlikely to clear fast.
868 */
869 if (cc < 0 && errno == ENOSR) {
870 sleep(1);
871 errno = EAGAIN;
872 }
873 #endif
874 } while (cc < 0 && SOFT_ERROR(errno));
875
876 if (cc < 0) {
877 isc__strerror(errno, strbuf, sizeof(strbuf));
878 FATAL_ERROR(__FILE__, __LINE__,
879 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
880 ISC_MSG_WRITEFAILED,
881 "write() failed "
882 "during watcher poke: %s"),
883 strbuf);
884 }
885
886 INSIST(cc == sizeof(buf));
887 }
888
889 /*
890 * Read a message on the internal fd.
891 */
892 static void
893 select_readmsg(isc_socketmgr_t *mgr, int *fd, int *msg) {
894 int buf[2];
895 int cc;
896 char strbuf[ISC_STRERRORSIZE];
897
898 cc = read(mgr->pipe_fds[0], buf, sizeof(buf));
899 if (cc < 0) {
900 *msg = SELECT_POKE_NOTHING;
901 *fd = -1; /* Silence compiler. */
902 if (SOFT_ERROR(errno))
903 return;
904
905 isc__strerror(errno, strbuf, sizeof(strbuf));
906 FATAL_ERROR(__FILE__, __LINE__,
907 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
908 ISC_MSG_READFAILED,
909 "read() failed "
910 "during watcher poke: %s"),
911 strbuf);
912
913 return;
914 }
915 INSIST(cc == sizeof(buf));
916
917 *fd = buf[0];
918 *msg = buf[1];
919 }
920 #else /* ISC_PLATFORM_USETHREADS */
921 /*
922 * Update the state of the socketmgr when something changes.
923 */
924 static void
925 select_poke(isc_socketmgr_t *manager, int fd, int msg) {
926 if (msg == SELECT_POKE_SHUTDOWN)
927 return;
928 else if (fd >= 0)
929 wakeup_socket(manager, fd, msg);
930 return;
931 }
932 #endif /* ISC_PLATFORM_USETHREADS */
933
934 /*
935 * Make a fd non-blocking.
936 */
937 static isc_result_t
938 make_nonblock(int fd) {
939 int ret;
940 int flags;
941 char strbuf[ISC_STRERRORSIZE];
942 #ifdef USE_FIONBIO_IOCTL
943 int on = 1;
944
945 ret = ioctl(fd, FIONBIO, (char *)&on);
946 #else
947 flags = fcntl(fd, F_GETFL, 0);
948 flags |= PORT_NONBLOCK;
949 ret = fcntl(fd, F_SETFL, flags);
950 #endif
951
952 if (ret == -1) {
953 isc__strerror(errno, strbuf, sizeof(strbuf));
954 UNEXPECTED_ERROR(__FILE__, __LINE__,
955 #ifdef USE_FIONBIO_IOCTL
956 "ioctl(%d, FIONBIO, &on): %s", fd,
957 #else
958 "fcntl(%d, F_SETFL, %d): %s", fd, flags,
959 #endif
960 strbuf);
961
962 return (ISC_R_UNEXPECTED);
963 }
964
965 return (ISC_R_SUCCESS);
966 }
967
968 #ifdef USE_CMSG
969 /*
970 * Not all OSes support advanced CMSG macros: CMSG_LEN and CMSG_SPACE.
971 * In order to ensure as much portability as possible, we provide wrapper
972 * functions of these macros.
973 * Note that cmsg_space() could run slow on OSes that do not have
974 * CMSG_SPACE.
975 */
976 static inline ISC_SOCKADDR_LEN_T
977 cmsg_len(ISC_SOCKADDR_LEN_T len) {
978 #ifdef CMSG_LEN
979 return (CMSG_LEN(len));
980 #else
981 ISC_SOCKADDR_LEN_T hdrlen;
982
983 /*
984 * Cast NULL so that any pointer arithmetic performed by CMSG_DATA
985 * is correct.
986 */
987 hdrlen = (ISC_SOCKADDR_LEN_T)CMSG_DATA(((struct cmsghdr *)NULL));
988 return (hdrlen + len);
989 #endif
990 }
991
992 static inline ISC_SOCKADDR_LEN_T
993 cmsg_space(ISC_SOCKADDR_LEN_T len) {
994 #ifdef CMSG_SPACE
995 return (CMSG_SPACE(len));
996 #else
997 struct msghdr msg;
998 struct cmsghdr *cmsgp;
999 /*
1000 * XXX: The buffer length is an ad-hoc value, but should be enough
1001 * in a practical sense.
1002 */
1003 char dummybuf[sizeof(struct cmsghdr) + 1024];
1004
1005 memset(&msg, 0, sizeof(msg));
1006 msg.msg_control = dummybuf;
1007 msg.msg_controllen = sizeof(dummybuf);
1008
1009 cmsgp = (struct cmsghdr *)dummybuf;
1010 cmsgp->cmsg_len = cmsg_len(len);
1011
1012 cmsgp = CMSG_NXTHDR(&msg, cmsgp);
1013 if (cmsgp != NULL)
1014 return ((char *)cmsgp - (char *)msg.msg_control);
1015 else
1016 return (0);
1017 #endif
1018 }
1019 #endif /* USE_CMSG */
1020
1021 /*
1022 * Process control messages received on a socket.
1023 */
1024 static void
1025 process_cmsg(isc_socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
1026 #ifdef USE_CMSG
1027 struct cmsghdr *cmsgp;
1028 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1029 struct in6_pktinfo *pktinfop;
1030 #endif
1031 #ifdef SO_TIMESTAMP
1032 struct timeval *timevalp;
1033 #endif
1034 #endif
1035
1036 /*
1037 * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
1038 * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
1039 * They are all here, outside of the CPP tests, because it is
1040 * more consistent with the usual ISC coding style.
1041 */
1042 UNUSED(sock);
1043 UNUSED(msg);
1044 UNUSED(dev);
1045
1046 #ifdef ISC_NET_BSD44MSGHDR
1047
1048 #ifdef MSG_TRUNC
1049 if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
1050 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1051 #endif
1052
1053 #ifdef MSG_CTRUNC
1054 if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
1055 dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
1056 #endif
1057
1058 #ifndef USE_CMSG
1059 return;
1060 #else
1061 if (msg->msg_controllen == 0U || msg->msg_control == NULL)
1062 return;
1063
1064 #ifdef SO_TIMESTAMP
1065 timevalp = NULL;
1066 #endif
1067 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1068 pktinfop = NULL;
1069 #endif
1070
1071 cmsgp = CMSG_FIRSTHDR(msg);
1072 while (cmsgp != NULL) {
1073 socket_log(sock, NULL, TRACE,
1074 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PROCESSCMSG,
1075 "processing cmsg %p", cmsgp);
1076
1077 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1078 if (cmsgp->cmsg_level == IPPROTO_IPV6
1079 && cmsgp->cmsg_type == IPV6_PKTINFO) {
1080
1081 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1082 memcpy(&dev->pktinfo, pktinfop,
1083 sizeof(struct in6_pktinfo));
1084 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
1085 socket_log(sock, NULL, TRACE,
1086 isc_msgcat, ISC_MSGSET_SOCKET,
1087 ISC_MSG_IFRECEIVED,
1088 "interface received on ifindex %u",
1089 dev->pktinfo.ipi6_ifindex);
1090 if (IN6_IS_ADDR_MULTICAST(&pktinfop->ipi6_addr))
1091 dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;
1092 goto next;
1093 }
1094 #endif
1095
1096 #ifdef SO_TIMESTAMP
1097 if (cmsgp->cmsg_level == SOL_SOCKET
1098 && cmsgp->cmsg_type == SCM_TIMESTAMP) {
1099 timevalp = (struct timeval *)CMSG_DATA(cmsgp);
1100 dev->timestamp.seconds = timevalp->tv_sec;
1101 dev->timestamp.nanoseconds = timevalp->tv_usec * 1000;
1102 dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
1103 goto next;
1104 }
1105 #endif
1106
1107 next:
1108 cmsgp = CMSG_NXTHDR(msg, cmsgp);
1109 }
1110 #endif /* USE_CMSG */
1111
1112 #endif /* ISC_NET_BSD44MSGHDR */
1113 }
1114
1115 /*
1116 * Construct an iov array and attach it to the msghdr passed in. This is
1117 * the SEND constructor, which will use the used region of the buffer
1118 * (if using a buffer list) or will use the internal region (if a single
1119 * buffer I/O is requested).
1120 *
1121 * Nothing can be NULL, and the done event must list at least one buffer
1122 * on the buffer linked list for this function to be meaningful.
1123 *
1124 * If write_countp != NULL, *write_countp will hold the number of bytes
1125 * this transaction can send.
1126 */
1127 static void
1128 build_msghdr_send(isc_socket_t *sock, isc_socketevent_t *dev,
1129 struct msghdr *msg, struct iovec *iov, size_t *write_countp)
1130 {
1131 unsigned int iovcount;
1132 isc_buffer_t *buffer;
1133 isc_region_t used;
1134 size_t write_count;
1135 size_t skip_count;
1136
1137 memset(msg, 0, sizeof(*msg));
1138
1139 if (!sock->connected) {
1140 msg->msg_name = (void *)&dev->address.type.sa;
1141 msg->msg_namelen = dev->address.length;
1142 } else {
1143 msg->msg_name = NULL;
1144 msg->msg_namelen = 0;
1145 }
1146
1147 buffer = ISC_LIST_HEAD(dev->bufferlist);
1148 write_count = 0;
1149 iovcount = 0;
1150
1151 /*
1152 * Single buffer I/O? Skip what we've done so far in this region.
1153 */
1154 if (buffer == NULL) {
1155 write_count = dev->region.length - dev->n;
1156 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1157 iov[0].iov_len = write_count;
1158 iovcount = 1;
1159
1160 goto config;
1161 }
1162
1163 /*
1164 * Multibuffer I/O.
1165 * Skip the data in the buffer list that we have already written.
1166 */
1167 skip_count = dev->n;
1168 while (buffer != NULL) {
1169 REQUIRE(ISC_BUFFER_VALID(buffer));
1170 if (skip_count < isc_buffer_usedlength(buffer))
1171 break;
1172 skip_count -= isc_buffer_usedlength(buffer);
1173 buffer = ISC_LIST_NEXT(buffer, link);
1174 }
1175
1176 while (buffer != NULL) {
1177 INSIST(iovcount < MAXSCATTERGATHER_SEND);
1178
1179 isc_buffer_usedregion(buffer, &used);
1180
1181 if (used.length > 0) {
1182 iov[iovcount].iov_base = (void *)(used.base
1183 + skip_count);
1184 iov[iovcount].iov_len = used.length - skip_count;
1185 write_count += (used.length - skip_count);
1186 skip_count = 0;
1187 iovcount++;
1188 }
1189 buffer = ISC_LIST_NEXT(buffer, link);
1190 }
1191
1192 INSIST(skip_count == 0U);
1193
1194 config:
1195 msg->msg_iov = iov;
1196 msg->msg_iovlen = iovcount;
1197
1198 #ifdef ISC_NET_BSD44MSGHDR
1199 msg->msg_control = NULL;
1200 msg->msg_controllen = 0;
1201 msg->msg_flags = 0;
1202 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1203 if ((sock->type == isc_sockettype_udp)
1204 && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
1205 struct cmsghdr *cmsgp;
1206 struct in6_pktinfo *pktinfop;
1207
1208 socket_log(sock, NULL, TRACE,
1209 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_SENDTODATA,
1210 "sendto pktinfo data, ifindex %u",
1211 dev->pktinfo.ipi6_ifindex);
1212
1213 msg->msg_controllen = cmsg_space(sizeof(struct in6_pktinfo));
1214 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
1215 msg->msg_control = (void *)sock->sendcmsgbuf;
1216
1217 cmsgp = (struct cmsghdr *)sock->sendcmsgbuf;
1218 cmsgp->cmsg_level = IPPROTO_IPV6;
1219 cmsgp->cmsg_type = IPV6_PKTINFO;
1220 cmsgp->cmsg_len = cmsg_len(sizeof(struct in6_pktinfo));
1221 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
1222 memcpy(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
1223 }
1224 #endif /* USE_CMSG && ISC_PLATFORM_HAVEIPV6 */
1225 #else /* ISC_NET_BSD44MSGHDR */
1226 msg->msg_accrights = NULL;
1227 msg->msg_accrightslen = 0;
1228 #endif /* ISC_NET_BSD44MSGHDR */
1229
1230 if (write_countp != NULL)
1231 *write_countp = write_count;
1232 }
1233
1234 /*
1235 * Construct an iov array and attach it to the msghdr passed in. This is
1236 * the RECV constructor, which will use the available region of the buffer
1237 * (if using a buffer list) or will use the internal region (if a single
1238 * buffer I/O is requested).
1239 *
1240 * Nothing can be NULL, and the done event must list at least one buffer
1241 * on the buffer linked list for this function to be meaningful.
1242 *
1243 * If read_countp != NULL, *read_countp will hold the number of bytes
1244 * this transaction can receive.
1245 */
1246 static void
1247 build_msghdr_recv(isc_socket_t *sock, isc_socketevent_t *dev,
1248 struct msghdr *msg, struct iovec *iov, size_t *read_countp)
1249 {
1250 unsigned int iovcount;
1251 isc_buffer_t *buffer;
1252 isc_region_t available;
1253 size_t read_count;
1254
1255 memset(msg, 0, sizeof(struct msghdr));
1256
1257 if (sock->type == isc_sockettype_udp) {
1258 memset(&dev->address, 0, sizeof(dev->address));
1259 #ifdef BROKEN_RECVMSG
1260 if (sock->pf == AF_INET) {
1261 msg->msg_name = (void *)&dev->address.type.sin;
1262 msg->msg_namelen = sizeof(dev->address.type.sin6);
1263 } else if (sock->pf == AF_INET6) {
1264 msg->msg_name = (void *)&dev->address.type.sin6;
1265 msg->msg_namelen = sizeof(dev->address.type.sin6);
1266 #ifdef ISC_PLATFORM_HAVESYSUNH
1267 } else if (sock->pf == AF_UNIX) {
1268 msg->msg_name = (void *)&dev->address.type.sunix;
1269 msg->msg_namelen = sizeof(dev->address.type.sunix);
1270 #endif
1271 } else {
1272 msg->msg_name = (void *)&dev->address.type.sa;
1273 msg->msg_namelen = sizeof(dev->address.type);
1274 }
1275 #else
1276 msg->msg_name = (void *)&dev->address.type.sa;
1277 msg->msg_namelen = sizeof(dev->address.type);
1278 #endif
1279 #ifdef ISC_NET_RECVOVERFLOW
1280 /* If needed, steal one iovec for overflow detection. */
1281 maxiov--;
1282 #endif
1283 } else { /* TCP */
1284 msg->msg_name = NULL;
1285 msg->msg_namelen = 0;
1286 dev->address = sock->peer_address;
1287 }
1288
1289 buffer = ISC_LIST_HEAD(dev->bufferlist);
1290 read_count = 0;
1291
1292 /*
1293 * Single buffer I/O? Skip what we've done so far in this region.
1294 */
1295 if (buffer == NULL) {
1296 read_count = dev->region.length - dev->n;
1297 iov[0].iov_base = (void *)(dev->region.base + dev->n);
1298 iov[0].iov_len = read_count;
1299 iovcount = 1;
1300
1301 goto config;
1302 }
1303
1304 /*
1305 * Multibuffer I/O.
1306 * Skip empty buffers.
1307 */
1308 while (buffer != NULL) {
1309 REQUIRE(ISC_BUFFER_VALID(buffer));
1310 if (isc_buffer_availablelength(buffer) != 0)
1311 break;
1312 buffer = ISC_LIST_NEXT(buffer, link);
1313 }
1314
1315 iovcount = 0;
1316 while (buffer != NULL) {
1317 INSIST(iovcount < MAXSCATTERGATHER_RECV);
1318
1319 isc_buffer_availableregion(buffer, &available);
1320
1321 if (available.length > 0) {
1322 iov[iovcount].iov_base = (void *)(available.base);
1323 iov[iovcount].iov_len = available.length;
1324 read_count += available.length;
1325 iovcount++;
1326 }
1327 buffer = ISC_LIST_NEXT(buffer, link);
1328 }
1329
1330 config:
1331
1332 /*
1333 * If needed, set up to receive that one extra byte. Note that
1334 * we know there is at least one iov left, since we stole it
1335 * at the top of this function.
1336 */
1337 #ifdef ISC_NET_RECVOVERFLOW
1338 if (sock->type == isc_sockettype_udp) {
1339 iov[iovcount].iov_base = (void *)(&sock->overflow);
1340 iov[iovcount].iov_len = 1;
1341 iovcount++;
1342 }
1343 #endif
1344
1345 msg->msg_iov = iov;
1346 msg->msg_iovlen = iovcount;
1347
1348 #ifdef ISC_NET_BSD44MSGHDR
1349 msg->msg_control = NULL;
1350 msg->msg_controllen = 0;
1351 msg->msg_flags = 0;
1352 #if defined(USE_CMSG)
1353 if (sock->type == isc_sockettype_udp) {
1354 msg->msg_control = sock->recvcmsgbuf;
1355 msg->msg_controllen = sock->recvcmsgbuflen;
1356 }
1357 #endif /* USE_CMSG */
1358 #else /* ISC_NET_BSD44MSGHDR */
1359 msg->msg_accrights = NULL;
1360 msg->msg_accrightslen = 0;
1361 #endif /* ISC_NET_BSD44MSGHDR */
1362
1363 if (read_countp != NULL)
1364 *read_countp = read_count;
1365 }
1366
1367 static void
1368 set_dev_address(isc_sockaddr_t *address, isc_socket_t *sock,
1369 isc_socketevent_t *dev)
1370 {
1371 if (sock->type == isc_sockettype_udp) {
1372 if (address != NULL)
1373 dev->address = *address;
1374 else
1375 dev->address = sock->peer_address;
1376 } else if (sock->type == isc_sockettype_tcp) {
1377 INSIST(address == NULL);
1378 dev->address = sock->peer_address;
1379 }
1380 }
1381
1382 static void
1383 destroy_socketevent(isc_event_t *event) {
1384 isc_socketevent_t *ev = (isc_socketevent_t *)event;
1385
1386 INSIST(ISC_LIST_EMPTY(ev->bufferlist));
1387
1388 (ev->destroy)(event);
1389 }
1390
1391 static isc_socketevent_t *
1392 allocate_socketevent(isc_socket_t *sock, isc_eventtype_t eventtype,
1393 isc_taskaction_t action, const void *arg)
1394 {
1395 isc_socketevent_t *ev;
1396
1397 ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
1398 sock, eventtype,
1399 action, arg,
1400 sizeof(*ev));
1401
1402 if (ev == NULL)
1403 return (NULL);
1404
1405 ev->result = ISC_R_UNEXPECTED;
1406 ISC_LINK_INIT(ev, ev_link);
1407 ISC_LIST_INIT(ev->bufferlist);
1408 ev->region.base = NULL;
1409 ev->n = 0;
1410 ev->offset = 0;
1411 ev->attributes = 0;
1412 ev->destroy = ev->ev_destroy;
1413 ev->ev_destroy = destroy_socketevent;
1414
1415 return (ev);
1416 }
1417
1418 #if defined(ISC_SOCKET_DEBUG)
1419 static void
1420 dump_msg(struct msghdr *msg) {
1421 unsigned int i;
1422
1423 printf("MSGHDR %p\n", msg);
1424 printf("\tname %p, namelen %ld\n", msg->msg_name,
1425 (long) msg->msg_namelen);
1426 printf("\tiov %p, iovlen %ld\n", msg->msg_iov,
1427 (long) msg->msg_iovlen);
1428 for (i = 0; i < (unsigned int)msg->msg_iovlen; i++)
1429 printf("\t\t%d\tbase %p, len %ld\n", i,
1430 msg->msg_iov[i].iov_base,
1431 (long) msg->msg_iov[i].iov_len);
1432 #ifdef ISC_NET_BSD44MSGHDR
1433 printf("\tcontrol %p, controllen %ld\n", msg->msg_control,
1434 (long) msg->msg_controllen);
1435 #endif
1436 }
1437 #endif
1438
1439 #define DOIO_SUCCESS 0 /* i/o ok, event sent */
1440 #define DOIO_SOFT 1 /* i/o ok, soft error, no event sent */
1441 #define DOIO_HARD 2 /* i/o error, event sent */
1442 #define DOIO_EOF 3 /* EOF, no event sent */
1443
1444 static int
1445 doio_recv(isc_socket_t *sock, isc_socketevent_t *dev) {
1446 int cc;
1447 struct iovec iov[MAXSCATTERGATHER_RECV];
1448 size_t read_count;
1449 size_t actual_count;
1450 struct msghdr msghdr;
1451 isc_buffer_t *buffer;
1452 int recv_errno;
1453 char strbuf[ISC_STRERRORSIZE];
1454
1455 build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);
1456
1457 #if defined(ISC_SOCKET_DEBUG)
1458 dump_msg(&msghdr);
1459 #endif
1460
1461 cc = recvmsg(sock->fd, &msghdr, 0);
1462 recv_errno = errno;
1463
1464 #if defined(ISC_SOCKET_DEBUG)
1465 dump_msg(&msghdr);
1466 #endif
1467
1468 if (cc < 0) {
1469 if (SOFT_ERROR(recv_errno))
1470 return (DOIO_SOFT);
1471
1472 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1473 isc__strerror(recv_errno, strbuf, sizeof(strbuf));
1474 socket_log(sock, NULL, IOEVENT,
1475 isc_msgcat, ISC_MSGSET_SOCKET,
1476 ISC_MSG_DOIORECV,
1477 "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
1478 sock->fd, cc, recv_errno, strbuf);
1479 }
1480
1481 #define SOFT_OR_HARD(_system, _isc) \
1482 if (recv_errno == _system) { \
1483 if (sock->connected) { \
1484 dev->result = _isc; \
1485 inc_stats(sock->manager->stats, \
1486 sock->statsindex[STATID_RECVFAIL]); \
1487 return (DOIO_HARD); \
1488 } \
1489 return (DOIO_SOFT); \
1490 }
1491 #define ALWAYS_HARD(_system, _isc) \
1492 if (recv_errno == _system) { \
1493 dev->result = _isc; \
1494 inc_stats(sock->manager->stats, \
1495 sock->statsindex[STATID_RECVFAIL]); \
1496 return (DOIO_HARD); \
1497 }
1498
1499 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1500 SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1501 SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1502 SOFT_OR_HARD(EHOSTDOWN, ISC_R_HOSTDOWN);
1503 /* HPUX 11.11 can return EADDRNOTAVAIL. */
1504 SOFT_OR_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1505 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1506 /*
1507 * HPUX returns EPROTO and EINVAL on receiving some ICMP/ICMPv6
1508 * errors.
1509 */
1510 #ifdef EPROTO
1511 SOFT_OR_HARD(EPROTO, ISC_R_HOSTUNREACH);
1512 #endif
1513 SOFT_OR_HARD(EINVAL, ISC_R_HOSTUNREACH);
1514
1515 #undef SOFT_OR_HARD
1516 #undef ALWAYS_HARD
1517
1518 dev->result = isc__errno2result(recv_errno);
1519 inc_stats(sock->manager->stats,
1520 sock->statsindex[STATID_RECVFAIL]);
1521 return (DOIO_HARD);
1522 }
1523
1524 /*
1525 * On TCP, zero length reads indicate EOF, while on
1526 * UDP, zero length reads are perfectly valid, although
1527 * strange.
1528 */
1529 if ((sock->type == isc_sockettype_tcp) && (cc == 0))
1530 return (DOIO_EOF);
1531
1532 if (sock->type == isc_sockettype_udp) {
1533 dev->address.length = msghdr.msg_namelen;
1534 if (isc_sockaddr_getport(&dev->address) == 0) {
1535 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1536 socket_log(sock, &dev->address, IOEVENT,
1537 isc_msgcat, ISC_MSGSET_SOCKET,
1538 ISC_MSG_ZEROPORT,
1539 "dropping source port zero packet");
1540 }
1541 return (DOIO_SOFT);
1542 }
1543 }
1544
1545 socket_log(sock, &dev->address, IOEVENT,
1546 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PKTRECV,
1547 "packet received correctly");
1548
1549 /*
1550 * Overflow bit detection. If we received MORE bytes than we should,
1551 * this indicates an overflow situation. Set the flag in the
1552 * dev entry and adjust how much we read by one.
1553 */
1554 #ifdef ISC_NET_RECVOVERFLOW
1555 if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
1556 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1557 cc--;
1558 }
1559 #endif
1560
1561 /*
1562 * If there are control messages attached, run through them and pull
1563 * out the interesting bits.
1564 */
1565 if (sock->type == isc_sockettype_udp)
1566 process_cmsg(sock, &msghdr, dev);
1567
1568 /*
1569 * update the buffers (if any) and the i/o count
1570 */
1571 dev->n += cc;
1572 actual_count = cc;
1573 buffer = ISC_LIST_HEAD(dev->bufferlist);
1574 while (buffer != NULL && actual_count > 0U) {
1575 REQUIRE(ISC_BUFFER_VALID(buffer));
1576 if (isc_buffer_availablelength(buffer) <= actual_count) {
1577 actual_count -= isc_buffer_availablelength(buffer);
1578 isc_buffer_add(buffer,
1579 isc_buffer_availablelength(buffer));
1580 } else {
1581 isc_buffer_add(buffer, actual_count);
1582 actual_count = 0;
1583 break;
1584 }
1585 buffer = ISC_LIST_NEXT(buffer, link);
1586 if (buffer == NULL) {
1587 INSIST(actual_count == 0U);
1588 }
1589 }
1590
1591 /*
1592 * If we read less than we expected, update counters,
1593 * and let the upper layer poke the descriptor.
1594 */
1595 if (((size_t)cc != read_count) && (dev->n < dev->minimum))
1596 return (DOIO_SOFT);
1597
1598 /*
1599 * Full reads are posted, or partials if partials are ok.
1600 */
1601 dev->result = ISC_R_SUCCESS;
1602 return (DOIO_SUCCESS);
1603 }
1604
1605 /*
1606 * Returns:
1607 * DOIO_SUCCESS The operation succeeded. dev->result contains
1608 * ISC_R_SUCCESS.
1609 *
1610 * DOIO_HARD A hard or unexpected I/O error was encountered.
1611 * dev->result contains the appropriate error.
1612 *
1613 * DOIO_SOFT A soft I/O error was encountered. No senddone
1614 * event was sent. The operation should be retried.
1615 *
1616 * No other return values are possible.
1617 */
1618 static int
1619 doio_send(isc_socket_t *sock, isc_socketevent_t *dev) {
1620 int cc;
1621 struct iovec iov[MAXSCATTERGATHER_SEND];
1622 size_t write_count;
1623 struct msghdr msghdr;
1624 char addrbuf[ISC_SOCKADDR_FORMATSIZE];
1625 int attempts = 0;
1626 int send_errno;
1627 char strbuf[ISC_STRERRORSIZE];
1628
1629 build_msghdr_send(sock, dev, &msghdr, iov, &write_count);
1630
1631 resend:
1632 cc = sendmsg(sock->fd, &msghdr, 0);
1633 send_errno = errno;
1634
1635 /*
1636 * Check for error or block condition.
1637 */
1638 if (cc < 0) {
1639 if (send_errno == EINTR && ++attempts < NRETRIES)
1640 goto resend;
1641
1642 if (SOFT_ERROR(send_errno))
1643 return (DOIO_SOFT);
1644
1645 #define SOFT_OR_HARD(_system, _isc) \
1646 if (send_errno == _system) { \
1647 if (sock->connected) { \
1648 dev->result = _isc; \
1649 inc_stats(sock->manager->stats, \
1650 sock->statsindex[STATID_SENDFAIL]); \
1651 return (DOIO_HARD); \
1652 } \
1653 return (DOIO_SOFT); \
1654 }
1655 #define ALWAYS_HARD(_system, _isc) \
1656 if (send_errno == _system) { \
1657 dev->result = _isc; \
1658 inc_stats(sock->manager->stats, \
1659 sock->statsindex[STATID_SENDFAIL]); \
1660 return (DOIO_HARD); \
1661 }
1662
1663 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1664 ALWAYS_HARD(EACCES, ISC_R_NOPERM);
1665 ALWAYS_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
1666 ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1667 ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1668 #ifdef EHOSTDOWN
1669 ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
1670 #endif
1671 ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1672 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1673 ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
1674 ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);
1675 ALWAYS_HARD(ECONNRESET, ISC_R_CONNECTIONRESET);
1676
1677 #undef SOFT_OR_HARD
1678 #undef ALWAYS_HARD
1679
1680 /*
1681 * The other error types depend on whether or not the
1682 * socket is UDP or TCP. If it is UDP, some errors
1683 * that we expect to be fatal under TCP are merely
1684 * annoying, and are really soft errors.
1685 *
1686 * However, these soft errors are still returned as
1687 * a status.
1688 */
1689 isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
1690 isc__strerror(send_errno, strbuf, sizeof(strbuf));
1691 UNEXPECTED_ERROR(__FILE__, __LINE__, "internal_send: %s: %s",
1692 addrbuf, strbuf);
1693 dev->result = isc__errno2result(send_errno);
1694 inc_stats(sock->manager->stats,
1695 sock->statsindex[STATID_SENDFAIL]);
1696 return (DOIO_HARD);
1697 }
1698
1699 if (cc == 0) {
1700 inc_stats(sock->manager->stats,
1701 sock->statsindex[STATID_SENDFAIL]);
1702 UNEXPECTED_ERROR(__FILE__, __LINE__,
1703 "doio_send: send() %s 0",
1704 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1705 ISC_MSG_RETURNED, "returned"));
1706 }
1707
1708 /*
1709 * If we write less than we expected, update counters, poke.
1710 */
1711 dev->n += cc;
1712 if ((size_t)cc != write_count)
1713 return (DOIO_SOFT);
1714
1715 /*
1716 * Exactly what we wanted to write. We're done with this
1717 * entry. Post its completion event.
1718 */
1719 dev->result = ISC_R_SUCCESS;
1720 return (DOIO_SUCCESS);
1721 }
1722
1723 /*
1724 * Kill.
1725 *
1726 * Caller must ensure that the socket is not locked and no external
1727 * references exist.
1728 */
1729 static void
1730 closesocket(isc_socketmgr_t *manager, isc_socket_t *sock, int fd) {
1731 isc_sockettype_t type = sock->type;
1732 int lockid = FDLOCK_ID(fd);
1733
1734 /*
1735 * No one has this socket open, so the watcher doesn't have to be
1736 * poked, and the socket doesn't have to be locked.
1737 */
1738 LOCK(&manager->fdlock[lockid]);
1739 manager->fds[fd] = NULL;
1740 if (type == isc_sockettype_fdwatch)
1741 manager->fdstate[fd] = CLOSED;
1742 else
1743 manager->fdstate[fd] = CLOSE_PENDING;
1744 UNLOCK(&manager->fdlock[lockid]);
1745 if (type == isc_sockettype_fdwatch) {
1746 /*
1747 * The caller may close the socket once this function returns,
1748 * and `fd' may be reassigned for a new socket. So we do
1749 * unwatch_fd() here, rather than defer it via select_poke().
1750 * Note: this may complicate data protection among threads and
1751 * may reduce performance due to additional locks. One way to
1752 * solve this would be to dup() the watched descriptor, but we
1753 * take a simpler approach at this moment.
1754 */
1755 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
1756 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
1757 } else
1758 select_poke(manager, fd, SELECT_POKE_CLOSE);
1759
1760 inc_stats(manager->stats, sock->statsindex[STATID_CLOSE]);
1761
1762 /*
1763 * update manager->maxfd here (XXX: this should be implemented more
1764 * efficiently)
1765 */
1766 #ifdef USE_SELECT
1767 LOCK(&manager->lock);
1768 if (manager->maxfd == fd) {
1769 int i;
1770
1771 manager->maxfd = 0;
1772 for (i = fd - 1; i >= 0; i--) {
1773 lockid = FDLOCK_ID(i);
1774
1775 LOCK(&manager->fdlock[lockid]);
1776 if (manager->fdstate[i] == MANAGED) {
1777 manager->maxfd = i;
1778 UNLOCK(&manager->fdlock[lockid]);
1779 break;
1780 }
1781 UNLOCK(&manager->fdlock[lockid]);
1782 }
1783 #ifdef ISC_PLATFORM_USETHREADS
1784 if (manager->maxfd < manager->pipe_fds[0])
1785 manager->maxfd = manager->pipe_fds[0];
1786 #endif
1787 }
1788 UNLOCK(&manager->lock);
1789 #endif /* USE_SELECT */
1790 }
1791
1792 static void
1793 destroy(isc_socket_t **sockp) {
1794 int fd;
1795 isc_socket_t *sock = *sockp;
1796 isc_socketmgr_t *manager = sock->manager;
1797
1798 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1799 ISC_MSG_DESTROYING, "destroying");
1800
1801 INSIST(ISC_LIST_EMPTY(sock->accept_list));
1802 INSIST(ISC_LIST_EMPTY(sock->recv_list));
1803 INSIST(ISC_LIST_EMPTY(sock->send_list));
1804 INSIST(sock->connect_ev == NULL);
1805 REQUIRE(sock->fd == -1 || sock->fd < (int)manager->maxsocks);
1806
1807 if (sock->fd >= 0) {
1808 fd = sock->fd;
1809 sock->fd = -1;
1810 closesocket(manager, sock, fd);
1811 }
1812
1813 LOCK(&manager->lock);
1814
1815 ISC_LIST_UNLINK(manager->socklist, sock, link);
1816
1817 #ifdef ISC_PLATFORM_USETHREADS
1818 if (ISC_LIST_EMPTY(manager->socklist))
1819 SIGNAL(&manager->shutdown_ok);
1820 #endif /* ISC_PLATFORM_USETHREADS */
1821
1822 UNLOCK(&manager->lock);
1823
1824 free_socket(sockp);
1825 }
1826
1827 static isc_result_t
1828 allocate_socket(isc_socketmgr_t *manager, isc_sockettype_t type,
1829 isc_socket_t **socketp)
1830 {
1831 isc_socket_t *sock;
1832 isc_result_t result;
1833 ISC_SOCKADDR_LEN_T cmsgbuflen;
1834
1835 sock = isc_mem_get(manager->mctx, sizeof(*sock));
1836
1837 if (sock == NULL)
1838 return (ISC_R_NOMEMORY);
1839
1840 result = ISC_R_UNEXPECTED;
1841
1842 sock->magic = 0;
1843 sock->references = 0;
1844
1845 sock->manager = manager;
1846 sock->type = type;
1847 sock->fd = -1;
1848 sock->statsindex = NULL;
1849
1850 ISC_LINK_INIT(sock, link);
1851
1852 sock->recvcmsgbuf = NULL;
1853 sock->sendcmsgbuf = NULL;
1854
1855 /*
1856 * set up cmsg buffers
1857 */
1858 cmsgbuflen = 0;
1859 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1860 cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1861 #endif
1862 #if defined(USE_CMSG) && defined(SO_TIMESTAMP)
1863 cmsgbuflen += cmsg_space(sizeof(struct timeval));
1864 #endif
1865 sock->recvcmsgbuflen = cmsgbuflen;
1866 if (sock->recvcmsgbuflen != 0U) {
1867 sock->recvcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1868 if (sock->recvcmsgbuf == NULL)
1869 goto error;
1870 }
1871
1872 cmsgbuflen = 0;
1873 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1874 cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1875 #endif
1876 sock->sendcmsgbuflen = cmsgbuflen;
1877 if (sock->sendcmsgbuflen != 0U) {
1878 sock->sendcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1879 if (sock->sendcmsgbuf == NULL)
1880 goto error;
1881 }
1882
1883 memset(sock->name, 0, sizeof(sock->name));
1884 sock->tag = NULL;
1885
1886 /*
1887 * set up list of readers and writers to be initially empty
1888 */
1889 ISC_LIST_INIT(sock->recv_list);
1890 ISC_LIST_INIT(sock->send_list);
1891 ISC_LIST_INIT(sock->accept_list);
1892 sock->connect_ev = NULL;
1893 sock->pending_recv = 0;
1894 sock->pending_send = 0;
1895 sock->pending_accept = 0;
1896 sock->listener = 0;
1897 sock->connected = 0;
1898 sock->connecting = 0;
1899 sock->bound = 0;
1900
1901 /*
1902 * initialize the lock
1903 */
1904 result = isc_mutex_init(&sock->lock);
1905 if (result != ISC_R_SUCCESS) {
1906 sock->magic = 0;
1907 goto error;
1908 }
1909
1910 /*
1911 * Initialize readable and writable events
1912 */
1913 ISC_EVENT_INIT(&sock->readable_ev, sizeof(intev_t),
1914 ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTR,
1915 NULL, sock, sock, NULL, NULL);
1916 ISC_EVENT_INIT(&sock->writable_ev, sizeof(intev_t),
1917 ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTW,
1918 NULL, sock, sock, NULL, NULL);
1919
1920 sock->magic = SOCKET_MAGIC;
1921 *socketp = sock;
1922
1923 return (ISC_R_SUCCESS);
1924
1925 error:
1926 if (sock->recvcmsgbuf != NULL)
1927 isc_mem_put(manager->mctx, sock->recvcmsgbuf,
1928 sock->recvcmsgbuflen);
1929 if (sock->sendcmsgbuf != NULL)
1930 isc_mem_put(manager->mctx, sock->sendcmsgbuf,
1931 sock->sendcmsgbuflen);
1932 isc_mem_put(manager->mctx, sock, sizeof(*sock));
1933
1934 return (result);
1935 }
1936
1937 /*
1938 * This event requires that the various lists be empty, that the reference
1939 * count be 1, and that the magic number is valid. The other socket bits,
1940 * like the lock, must be initialized as well. The fd associated must be
1941 * marked as closed, by setting it to -1 on close, or this routine will
1942 * also close the socket.
1943 */
1944 static void
1945 free_socket(isc_socket_t **socketp) {
1946 isc_socket_t *sock = *socketp;
1947
1948 INSIST(sock->references == 0);
1949 INSIST(VALID_SOCKET(sock));
1950 INSIST(!sock->connecting);
1951 INSIST(!sock->pending_recv);
1952 INSIST(!sock->pending_send);
1953 INSIST(!sock->pending_accept);
1954 INSIST(ISC_LIST_EMPTY(sock->recv_list));
1955 INSIST(ISC_LIST_EMPTY(sock->send_list));
1956 INSIST(ISC_LIST_EMPTY(sock->accept_list));
1957 INSIST(!ISC_LINK_LINKED(sock, link));
1958
1959 if (sock->recvcmsgbuf != NULL)
1960 isc_mem_put(sock->manager->mctx, sock->recvcmsgbuf,
1961 sock->recvcmsgbuflen);
1962 if (sock->sendcmsgbuf != NULL)
1963 isc_mem_put(sock->manager->mctx, sock->sendcmsgbuf,
1964 sock->sendcmsgbuflen);
1965
1966 sock->magic = 0;
1967
1968 DESTROYLOCK(&sock->lock);
1969
1970 isc_mem_put(sock->manager->mctx, sock, sizeof(*sock));
1971
1972 *socketp = NULL;
1973 }
1974
1975 #ifdef SO_BSDCOMPAT
1976 /*
1977 * This really should not be necessary to do. Having to workout
1978 * which kernel version we are on at run time so that we don't cause
1979 * the kernel to issue a warning about us using a deprecated socket option.
1980 * Such warnings should *never* be on by default in production kernels.
1981 *
1982 * We can't do this a build time because executables are moved between
1983 * machines and hence kernels.
1984 *
1985 * We can't just not set SO_BSDCOMAT because some kernels require it.
1986 */
1987
1988 static isc_once_t bsdcompat_once = ISC_ONCE_INIT;
1989 isc_boolean_t bsdcompat = ISC_TRUE;
1990
1991 static void
1992 clear_bsdcompat(void) {
1993 #ifdef __linux__
1994 struct utsname buf;
1995 char *endp;
1996 long int major;
1997 long int minor;
1998
1999 uname(&buf); /* Can only fail if buf is bad in Linux. */
2000
2001 /* Paranoia in parsing can be increased, but we trust uname(). */
2002 major = strtol(buf.release, &endp, 10);
2003 if (*endp == '.') {
2004 minor = strtol(endp+1, &endp, 10);
2005 if ((major > 2) || ((major == 2) && (minor >= 4))) {
2006 bsdcompat = ISC_FALSE;
2007 }
2008 }
2009 #endif /* __linux __ */
2010 }
2011 #endif
2012
2013 static isc_result_t
2014 opensocket(isc_socketmgr_t *manager, isc_socket_t *sock) {
2015 char strbuf[ISC_STRERRORSIZE];
2016 const char *err = "socket";
2017 int tries = 0;
2018 #if defined(USE_CMSG) || defined(SO_BSDCOMPAT)
2019 int on = 1;
2020 #endif
2021 #if defined(SO_RCVBUF)
2022 ISC_SOCKADDR_LEN_T optlen;
2023 int size;
2024 #endif
2025
2026 again:
2027 switch (sock->type) {
2028 case isc_sockettype_udp:
2029 sock->fd = socket(sock->pf, SOCK_DGRAM, IPPROTO_UDP);
2030 break;
2031 case isc_sockettype_tcp:
2032 sock->fd = socket(sock->pf, SOCK_STREAM, IPPROTO_TCP);
2033 break;
2034 case isc_sockettype_unix:
2035 sock->fd = socket(sock->pf, SOCK_STREAM, 0);
2036 break;
2037 case isc_sockettype_fdwatch:
2038 /*
2039 * We should not be called for isc_sockettype_fdwatch sockets.
2040 */
2041 INSIST(0);
2042 break;
2043 }
2044 if (sock->fd == -1 && errno == EINTR && tries++ < 42)
2045 goto again;
2046
2047 #ifdef F_DUPFD
2048 /*
2049 * Leave a space for stdio and TCP to work in.
2050 */
2051 if (manager->reserved != 0 && sock->type == isc_sockettype_udp &&
2052 sock->fd >= 0 && sock->fd < manager->reserved) {
2053 int new, tmp;
2054 new = fcntl(sock->fd, F_DUPFD, manager->reserved);
2055 tmp = errno;
2056 (void)close(sock->fd);
2057 errno = tmp;
2058 sock->fd = new;
2059 err = "isc_socket_create: fcntl/reserved";
2060 } else if (sock->fd >= 0 && sock->fd < 20) {
2061 int new, tmp;
2062 new = fcntl(sock->fd, F_DUPFD, 20);
2063 tmp = errno;
2064 (void)close(sock->fd);
2065 errno = tmp;
2066 sock->fd = new;
2067 err = "isc_socket_create: fcntl";
2068 }
2069 #endif
2070
2071 if (sock->fd >= (int)manager->maxsocks) {
2072 (void)close(sock->fd);
2073 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2074 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2075 isc_msgcat, ISC_MSGSET_SOCKET,
2076 ISC_MSG_TOOMANYFDS,
2077 "socket: file descriptor exceeds limit (%d/%u)",
2078 sock->fd, manager->maxsocks);
2079 return (ISC_R_NORESOURCES);
2080 }
2081
2082 if (sock->fd < 0) {
2083 switch (errno) {
2084 case EMFILE:
2085 case ENFILE:
2086 isc__strerror(errno, strbuf, sizeof(strbuf));
2087 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2088 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2089 isc_msgcat, ISC_MSGSET_SOCKET,
2090 ISC_MSG_TOOMANYFDS,
2091 "%s: %s", err, strbuf);
2092 /* fallthrough */
2093 case ENOBUFS:
2094 return (ISC_R_NORESOURCES);
2095
2096 case EPROTONOSUPPORT:
2097 case EPFNOSUPPORT:
2098 case EAFNOSUPPORT:
2099 /*
2100 * Linux 2.2 (and maybe others) return EINVAL instead of
2101 * EAFNOSUPPORT.
2102 */
2103 case EINVAL:
2104 return (ISC_R_FAMILYNOSUPPORT);
2105
2106 default:
2107 isc__strerror(errno, strbuf, sizeof(strbuf));
2108 UNEXPECTED_ERROR(__FILE__, __LINE__,
2109 "%s() %s: %s", err,
2110 isc_msgcat_get(isc_msgcat,
2111 ISC_MSGSET_GENERAL,
2112 ISC_MSG_FAILED,
2113 "failed"),
2114 strbuf);
2115 return (ISC_R_UNEXPECTED);
2116 }
2117 }
2118
2119 if (make_nonblock(sock->fd) != ISC_R_SUCCESS) {
2120 (void)close(sock->fd);
2121 return (ISC_R_UNEXPECTED);
2122 }
2123
2124 #ifdef SO_BSDCOMPAT
2125 RUNTIME_CHECK(isc_once_do(&bsdcompat_once,
2126 clear_bsdcompat) == ISC_R_SUCCESS);
2127 if (sock->type != isc_sockettype_unix && bsdcompat &&
2128 setsockopt(sock->fd, SOL_SOCKET, SO_BSDCOMPAT,
2129 (void *)&on, sizeof(on)) < 0) {
2130 isc__strerror(errno, strbuf, sizeof(strbuf));
2131 UNEXPECTED_ERROR(__FILE__, __LINE__,
2132 "setsockopt(%d, SO_BSDCOMPAT) %s: %s",
2133 sock->fd,
2134 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2135 ISC_MSG_FAILED, "failed"),
2136 strbuf);
2137 /* Press on... */
2138 }
2139 #endif
2140
2141 #ifdef SO_NOSIGPIPE
2142 if (setsockopt(sock->fd, SOL_SOCKET, SO_NOSIGPIPE,
2143 (void *)&on, sizeof(on)) < 0) {
2144 isc__strerror(errno, strbuf, sizeof(strbuf));
2145 UNEXPECTED_ERROR(__FILE__, __LINE__,
2146 "setsockopt(%d, SO_NOSIGPIPE) %s: %s",
2147 sock->fd,
2148 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2149 ISC_MSG_FAILED, "failed"),
2150 strbuf);
2151 /* Press on... */
2152 }
2153 #endif
2154
2155 #if defined(USE_CMSG) || defined(SO_RCVBUF)
2156 if (sock->type == isc_sockettype_udp) {
2157
2158 #if defined(USE_CMSG)
2159 #if defined(SO_TIMESTAMP)
2160 if (setsockopt(sock->fd, SOL_SOCKET, SO_TIMESTAMP,
2161 (void *)&on, sizeof(on)) < 0
2162 && errno != ENOPROTOOPT) {
2163 isc__strerror(errno, strbuf, sizeof(strbuf));
2164 UNEXPECTED_ERROR(__FILE__, __LINE__,
2165 "setsockopt(%d, SO_TIMESTAMP) %s: %s",
2166 sock->fd,
2167 isc_msgcat_get(isc_msgcat,
2168 ISC_MSGSET_GENERAL,
2169 ISC_MSG_FAILED,
2170 "failed"),
2171 strbuf);
2172 /* Press on... */
2173 }
2174 #endif /* SO_TIMESTAMP */
2175
2176 #if defined(ISC_PLATFORM_HAVEIPV6)
2177 if (sock->pf == AF_INET6 && sock->recvcmsgbuflen == 0U) {
2178 /*
2179 * Warn explicitly because this anomaly can be hidden
2180 * in usual operation (and unexpectedly appear later).
2181 */
2182 UNEXPECTED_ERROR(__FILE__, __LINE__,
2183 "No buffer available to receive "
2184 "IPv6 destination");
2185 }
2186 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
2187 #ifdef IPV6_RECVPKTINFO
2188 /* RFC 3542 */
2189 if ((sock->pf == AF_INET6)
2190 && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
2191 (void *)&on, sizeof(on)) < 0)) {
2192 isc__strerror(errno, strbuf, sizeof(strbuf));
2193 UNEXPECTED_ERROR(__FILE__, __LINE__,
2194 "setsockopt(%d, IPV6_RECVPKTINFO) "
2195 "%s: %s", sock->fd,
2196 isc_msgcat_get(isc_msgcat,
2197 ISC_MSGSET_GENERAL,
2198 ISC_MSG_FAILED,
2199 "failed"),
2200 strbuf);
2201 }
2202 #else
2203 /* RFC 2292 */
2204 if ((sock->pf == AF_INET6)
2205 && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_PKTINFO,
2206 (void *)&on, sizeof(on)) < 0)) {
2207 isc__strerror(errno, strbuf, sizeof(strbuf));
2208 UNEXPECTED_ERROR(__FILE__, __LINE__,
2209 "setsockopt(%d, IPV6_PKTINFO) %s: %s",
2210 sock->fd,
2211 isc_msgcat_get(isc_msgcat,
2212 ISC_MSGSET_GENERAL,
2213 ISC_MSG_FAILED,
2214 "failed"),
2215 strbuf);
2216 }
2217 #endif /* IPV6_RECVPKTINFO */
2218 #endif /* ISC_PLATFORM_HAVEIN6PKTINFO */
2219 #ifdef IPV6_USE_MIN_MTU /* RFC 3542, not too common yet*/
2220 /* use minimum MTU */
2221 if (sock->pf == AF_INET6) {
2222 (void)setsockopt(sock->fd, IPPROTO_IPV6,
2223 IPV6_USE_MIN_MTU,
2224 (void *)&on, sizeof(on));
2225 }
2226 #endif
2227 #endif /* ISC_PLATFORM_HAVEIPV6 */
2228 #endif /* defined(USE_CMSG) */
2229
2230 #if defined(IP_MTU_DISCOVER) && defined(IP_PMTUDISC_DONT)
2231 /*
2232 * Turn off Path MTU discovery on IPv4/UDP sockets.
2233 */
2234 if (sock->pf == AF_INET) {
2235 int action = IP_PMTUDISC_DONT;
2236 (void)setsockopt(sock->fd, IPPROTO_IP, IP_MTU_DISCOVER,
2237 &action, sizeof(action));
2238 }
2239 #endif
2240 #if defined(IP_DONTFRAG)
2241 /*
2242 * Turn off Path MTU discovery on IPv4/UDP sockets.
2243 */
2244 if (sock->pf == AF_INET) {
2245 int off = 0;
2246 (void)setsockopt(sock->fd, IPPROTO_IP, IP_DONTFRAG,
2247 &off, sizeof(off));
2248 }
2249 #endif
2250
2251 #if defined(SO_RCVBUF)
2252 optlen = sizeof(size);
2253 if (getsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2254 (void *)&size, &optlen) >= 0 &&
2255 size < RCVBUFSIZE) {
2256 size = RCVBUFSIZE;
2257 if (setsockopt(sock->fd, SOL_SOCKET, SO_RCVBUF,
2258 (void *)&size, sizeof(size)) == -1) {
2259 isc__strerror(errno, strbuf, sizeof(strbuf));
2260 UNEXPECTED_ERROR(__FILE__, __LINE__,
2261 "setsockopt(%d, SO_RCVBUF, %d) %s: %s",
2262 sock->fd, size,
2263 isc_msgcat_get(isc_msgcat,
2264 ISC_MSGSET_GENERAL,
2265 ISC_MSG_FAILED,
2266 "failed"),
2267 strbuf);
2268 }
2269 }
2270 #endif
2271 }
2272 #endif /* defined(USE_CMSG) || defined(SO_RCVBUF) */
2273
2274 inc_stats(manager->stats, sock->statsindex[STATID_OPEN]);
2275
2276 return (ISC_R_SUCCESS);
2277 }
2278
2279 /*%
2280 * Create a new 'type' socket managed by 'manager'. Events
2281 * will be posted to 'task' and when dispatched 'action' will be
2282 * called with 'arg' as the arg value. The new socket is returned
2283 * in 'socketp'.
2284 */
2285 isc_result_t
2286 isc_socket_create(isc_socketmgr_t *manager, int pf, isc_sockettype_t type,
2287 isc_socket_t **socketp)
2288 {
2289 isc_socket_t *sock = NULL;
2290 isc_result_t result;
2291 int lockid;
2292
2293 REQUIRE(VALID_MANAGER(manager));
2294 REQUIRE(socketp != NULL && *socketp == NULL);
2295 REQUIRE(type != isc_sockettype_fdwatch);
2296
2297 result = allocate_socket(manager, type, &sock);
2298 if (result != ISC_R_SUCCESS)
2299 return (result);
2300
2301 switch (sock->type) {
2302 case isc_sockettype_udp:
2303 sock->statsindex =
2304 (pf == AF_INET) ? upd4statsindex : upd6statsindex;
2305 break;
2306 case isc_sockettype_tcp:
2307 sock->statsindex =
2308 (pf == AF_INET) ? tcp4statsindex : tcp6statsindex;
2309 break;
2310 case isc_sockettype_unix:
2311 sock->statsindex = unixstatsindex;
2312 break;
2313 default:
2314 INSIST(0);
2315 }
2316
2317 sock->pf = pf;
2318 result = opensocket(manager, sock);
2319 if (result != ISC_R_SUCCESS) {
2320 inc_stats(manager->stats, sock->statsindex[STATID_OPENFAIL]);
2321 free_socket(&sock);
2322 return (result);
2323 }
2324
2325 sock->references = 1;
2326 *socketp = sock;
2327
2328 /*
2329 * Note we don't have to lock the socket like we normally would because
2330 * there are no external references to it yet.
2331 */
2332
2333 lockid = FDLOCK_ID(sock->fd);
2334 LOCK(&manager->fdlock[lockid]);
2335 manager->fds[sock->fd] = sock;
2336 manager->fdstate[sock->fd] = MANAGED;
2337 #ifdef USE_DEVPOLL
2338 INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2339 sock->manager->fdpollinfo[sock->fd].want_write == 0);
2340 #endif
2341 UNLOCK(&manager->fdlock[lockid]);
2342
2343 LOCK(&manager->lock);
2344 ISC_LIST_APPEND(manager->socklist, sock, link);
2345 #ifdef USE_SELECT
2346 if (manager->maxfd < sock->fd)
2347 manager->maxfd = sock->fd;
2348 #endif
2349 UNLOCK(&manager->lock);
2350
2351 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2352 ISC_MSG_CREATED, "created");
2353
2354 return (ISC_R_SUCCESS);
2355 }
2356
2357 isc_result_t
2358 isc_socket_open(isc_socket_t *sock) {
2359 isc_result_t result;
2360
2361 REQUIRE(VALID_SOCKET(sock));
2362
2363 LOCK(&sock->lock);
2364 REQUIRE(sock->references == 1);
2365 REQUIRE(sock->type != isc_sockettype_fdwatch);
2366 UNLOCK(&sock->lock);
2367 /*
2368 * We don't need to retain the lock hereafter, since no one else has
2369 * this socket.
2370 */
2371 REQUIRE(sock->fd == -1);
2372
2373 result = opensocket(sock->manager, sock);
2374 if (result != ISC_R_SUCCESS)
2375 sock->fd = -1;
2376
2377 if (result == ISC_R_SUCCESS) {
2378 int lockid = FDLOCK_ID(sock->fd);
2379
2380 LOCK(&sock->manager->fdlock[lockid]);
2381 sock->manager->fds[sock->fd] = sock;
2382 sock->manager->fdstate[sock->fd] = MANAGED;
2383 #ifdef USE_DEVPOLL
2384 INSIST(sock->manager->fdpollinfo[sock->fd].want_read == 0 &&
2385 sock->manager->fdpollinfo[sock->fd].want_write == 0);
2386 #endif
2387 UNLOCK(&sock->manager->fdlock[lockid]);
2388
2389 #ifdef USE_SELECT
2390 LOCK(&sock->manager->lock);
2391 if (sock->manager->maxfd < sock->fd)
2392 sock->manager->maxfd = sock->fd;
2393 UNLOCK(&sock->manager->lock);
2394 #endif
2395 }
2396
2397 return (result);
2398 }
2399
2400 /*
2401 * Create a new 'type' socket managed by 'manager'. Events
2402 * will be posted to 'task' and when dispatched 'action' will be
2403 * called with 'arg' as the arg value. The new socket is returned
2404 * in 'socketp'.
2405 */
2406 isc_result_t
2407 isc_socket_fdwatchcreate(isc_socketmgr_t *manager, int fd, int flags,
2408 isc_sockfdwatch_t callback, void *cbarg,
2409 isc_task_t *task, isc_socket_t **socketp)
2410 {
2411 isc_socket_t *sock = NULL;
2412 isc_result_t result;
2413 int lockid;
2414
2415 REQUIRE(VALID_MANAGER(manager));
2416 REQUIRE(socketp != NULL && *socketp == NULL);
2417
2418 result = allocate_socket(manager, isc_sockettype_fdwatch, &sock);
2419 if (result != ISC_R_SUCCESS)
2420 return (result);
2421
2422 sock->fd = fd;
2423 sock->fdwatcharg = cbarg;
2424 sock->fdwatchcb = callback;
2425 sock->fdwatchflags = flags;
2426 sock->fdwatchtask = task;
2427 sock->statsindex = fdwatchstatsindex;
2428
2429 sock->references = 1;
2430 *socketp = sock;
2431
2432 /*
2433 * Note we don't have to lock the socket like we normally would because
2434 * there are no external references to it yet.
2435 */
2436
2437 lockid = FDLOCK_ID(sock->fd);
2438 LOCK(&manager->fdlock[lockid]);
2439 manager->fds[sock->fd] = sock;
2440 manager->fdstate[sock->fd] = MANAGED;
2441 UNLOCK(&manager->fdlock[lockid]);
2442
2443 LOCK(&manager->lock);
2444 ISC_LIST_APPEND(manager->socklist, sock, link);
2445 #ifdef USE_SELECT
2446 if (manager->maxfd < sock->fd)
2447 manager->maxfd = sock->fd;
2448 #endif
2449 UNLOCK(&manager->lock);
2450
2451 if (flags & ISC_SOCKFDWATCH_READ)
2452 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2453 if (flags & ISC_SOCKFDWATCH_WRITE)
2454 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
2455
2456 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
2457 ISC_MSG_CREATED, "fdwatch-created");
2458
2459 return (ISC_R_SUCCESS);
2460 }
2461
2462 /*
2463 * Attach to a socket. Caller must explicitly detach when it is done.
2464 */
2465 void
2466 isc_socket_attach(isc_socket_t *sock, isc_socket_t **socketp) {
2467 REQUIRE(VALID_SOCKET(sock));
2468 REQUIRE(socketp != NULL && *socketp == NULL);
2469
2470 LOCK(&sock->lock);
2471 sock->references++;
2472 UNLOCK(&sock->lock);
2473
2474 *socketp = sock;
2475 }
2476
2477 /*
2478 * Dereference a socket. If this is the last reference to it, clean things
2479 * up by destroying the socket.
2480 */
2481 void
2482 isc_socket_detach(isc_socket_t **socketp) {
2483 isc_socket_t *sock;
2484 isc_boolean_t kill_socket = ISC_FALSE;
2485
2486 REQUIRE(socketp != NULL);
2487 sock = *socketp;
2488 REQUIRE(VALID_SOCKET(sock));
2489
2490 LOCK(&sock->lock);
2491 REQUIRE(sock->references > 0);
2492 sock->references--;
2493 if (sock->references == 0)
2494 kill_socket = ISC_TRUE;
2495 UNLOCK(&sock->lock);
2496
2497 if (kill_socket)
2498 destroy(&sock);
2499
2500 *socketp = NULL;
2501 }
2502
2503 isc_result_t
2504 isc_socket_close(isc_socket_t *sock) {
2505 int fd;
2506 isc_socketmgr_t *manager;
2507 isc_sockettype_t type;
2508
2509 REQUIRE(VALID_SOCKET(sock));
2510
2511 LOCK(&sock->lock);
2512
2513 REQUIRE(sock->references == 1);
2514 REQUIRE(sock->type != isc_sockettype_fdwatch);
2515 REQUIRE(sock->fd >= 0 && sock->fd < (int)sock->manager->maxsocks);
2516
2517 INSIST(!sock->connecting);
2518 INSIST(!sock->pending_recv);
2519 INSIST(!sock->pending_send);
2520 INSIST(!sock->pending_accept);
2521 INSIST(ISC_LIST_EMPTY(sock->recv_list));
2522 INSIST(ISC_LIST_EMPTY(sock->send_list));
2523 INSIST(ISC_LIST_EMPTY(sock->accept_list));
2524 INSIST(sock->connect_ev == NULL);
2525
2526 manager = sock->manager;
2527 type = sock->type;
2528 fd = sock->fd;
2529 sock->fd = -1;
2530 memset(sock->name, 0, sizeof(sock->name));
2531 sock->tag = NULL;
2532 sock->listener = 0;
2533 sock->connected = 0;
2534 sock->connecting = 0;
2535 sock->bound = 0;
2536 isc_sockaddr_any(&sock->peer_address);
2537
2538 UNLOCK(&sock->lock);
2539
2540 closesocket(manager, sock, fd);
2541
2542 return (ISC_R_SUCCESS);
2543 }
2544
2545 /*
2546 * I/O is possible on a given socket. Schedule an event to this task that
2547 * will call an internal function to do the I/O. This will charge the
2548 * task with the I/O operation and let our select loop handler get back
2549 * to doing something real as fast as possible.
2550 *
2551 * The socket and manager must be locked before calling this function.
2552 */
2553 static void
2554 dispatch_recv(isc_socket_t *sock) {
2555 intev_t *iev;
2556 isc_socketevent_t *ev;
2557 isc_task_t *sender;
2558
2559 INSIST(!sock->pending_recv);
2560
2561 if (sock->type != isc_sockettype_fdwatch) {
2562 ev = ISC_LIST_HEAD(sock->recv_list);
2563 if (ev == NULL)
2564 return;
2565 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2566 "dispatch_recv: event %p -> task %p",
2567 ev, ev->ev_sender);
2568 sender = ev->ev_sender;
2569 } else {
2570 sender = sock->fdwatchtask;
2571 }
2572
2573 sock->pending_recv = 1;
2574 iev = &sock->readable_ev;
2575
2576 sock->references++;
2577 iev->ev_sender = sock;
2578 if (sock->type == isc_sockettype_fdwatch)
2579 iev->ev_action = internal_fdwatch_read;
2580 else
2581 iev->ev_action = internal_recv;
2582 iev->ev_arg = sock;
2583
2584 isc_task_send(sender, (isc_event_t **)&iev);
2585 }
2586
2587 static void
2588 dispatch_send(isc_socket_t *sock) {
2589 intev_t *iev;
2590 isc_socketevent_t *ev;
2591 isc_task_t *sender;
2592
2593 INSIST(!sock->pending_send);
2594
2595 if (sock->type != isc_sockettype_fdwatch) {
2596 ev = ISC_LIST_HEAD(sock->send_list);
2597 if (ev == NULL)
2598 return;
2599 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2600 "dispatch_send: event %p -> task %p",
2601 ev, ev->ev_sender);
2602 sender = ev->ev_sender;
2603 } else {
2604 sender = sock->fdwatchtask;
2605 }
2606
2607 sock->pending_send = 1;
2608 iev = &sock->writable_ev;
2609
2610 sock->references++;
2611 iev->ev_sender = sock;
2612 if (sock->type == isc_sockettype_fdwatch)
2613 iev->ev_action = internal_fdwatch_write;
2614 else
2615 iev->ev_action = internal_send;
2616 iev->ev_arg = sock;
2617
2618 isc_task_send(sender, (isc_event_t **)&iev);
2619 }
2620
2621 /*
2622 * Dispatch an internal accept event.
2623 */
2624 static void
2625 dispatch_accept(isc_socket_t *sock) {
2626 intev_t *iev;
2627 isc_socket_newconnev_t *ev;
2628
2629 INSIST(!sock->pending_accept);
2630
2631 /*
2632 * Are there any done events left, or were they all canceled
2633 * before the manager got the socket lock?
2634 */
2635 ev = ISC_LIST_HEAD(sock->accept_list);
2636 if (ev == NULL)
2637 return;
2638
2639 sock->pending_accept = 1;
2640 iev = &sock->readable_ev;
2641
2642 sock->references++; /* keep socket around for this internal event */
2643 iev->ev_sender = sock;
2644 iev->ev_action = internal_accept;
2645 iev->ev_arg = sock;
2646
2647 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
2648 }
2649
2650 static void
2651 dispatch_connect(isc_socket_t *sock) {
2652 intev_t *iev;
2653 isc_socket_connev_t *ev;
2654
2655 iev = &sock->writable_ev;
2656
2657 ev = sock->connect_ev;
2658 INSIST(ev != NULL); /* XXX */
2659
2660 INSIST(sock->connecting);
2661
2662 sock->references++; /* keep socket around for this internal event */
2663 iev->ev_sender = sock;
2664 iev->ev_action = internal_connect;
2665 iev->ev_arg = sock;
2666
2667 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
2668 }
2669
2670 /*
2671 * Dequeue an item off the given socket's read queue, set the result code
2672 * in the done event to the one provided, and send it to the task it was
2673 * destined for.
2674 *
2675 * If the event to be sent is on a list, remove it before sending. If
2676 * asked to, send and detach from the socket as well.
2677 *
2678 * Caller must have the socket locked if the event is attached to the socket.
2679 */
2680 static void
2681 send_recvdone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
2682 isc_task_t *task;
2683
2684 task = (*dev)->ev_sender;
2685
2686 (*dev)->ev_sender = sock;
2687
2688 if (ISC_LINK_LINKED(*dev, ev_link))
2689 ISC_LIST_DEQUEUE(sock->recv_list, *dev, ev_link);
2690
2691 if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
2692 == ISC_SOCKEVENTATTR_ATTACHED)
2693 isc_task_sendanddetach(&task, (isc_event_t **)dev);
2694 else
2695 isc_task_send(task, (isc_event_t **)dev);
2696 }
2697
2698 /*
2699 * See comments for send_recvdone_event() above.
2700 *
2701 * Caller must have the socket locked if the event is attached to the socket.
2702 */
2703 static void
2704 send_senddone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
2705 isc_task_t *task;
2706
2707 INSIST(dev != NULL && *dev != NULL);
2708
2709 task = (*dev)->ev_sender;
2710 (*dev)->ev_sender = sock;
2711
2712 if (ISC_LINK_LINKED(*dev, ev_link))
2713 ISC_LIST_DEQUEUE(sock->send_list, *dev, ev_link);
2714
2715 if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
2716 == ISC_SOCKEVENTATTR_ATTACHED)
2717 isc_task_sendanddetach(&task, (isc_event_t **)dev);
2718 else
2719 isc_task_send(task, (isc_event_t **)dev);
2720 }
2721
2722 /*
2723 * Call accept() on a socket, to get the new file descriptor. The listen
2724 * socket is used as a prototype to create a new isc_socket_t. The new
2725 * socket has one outstanding reference. The task receiving the event
2726 * will be detached from just after the event is delivered.
2727 *
2728 * On entry to this function, the event delivered is the internal
2729 * readable event, and the first item on the accept_list should be
2730 * the done event we want to send. If the list is empty, this is a no-op,
2731 * so just unlock and return.
2732 */
2733 static void
2734 internal_accept(isc_task_t *me, isc_event_t *ev) {
2735 isc_socket_t *sock;
2736 isc_socketmgr_t *manager;
2737 isc_socket_newconnev_t *dev;
2738 isc_task_t *task;
2739 ISC_SOCKADDR_LEN_T addrlen;
2740 int fd;
2741 isc_result_t result = ISC_R_SUCCESS;
2742 char strbuf[ISC_STRERRORSIZE];
2743 const char *err = "accept";
2744
2745 UNUSED(me);
2746
2747 sock = ev->ev_sender;
2748 INSIST(VALID_SOCKET(sock));
2749
2750 LOCK(&sock->lock);
2751 socket_log(sock, NULL, TRACE,
2752 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTLOCK,
2753 "internal_accept called, locked socket");
2754
2755 manager = sock->manager;
2756 INSIST(VALID_MANAGER(manager));
2757
2758 INSIST(sock->listener);
2759 INSIST(sock->pending_accept == 1);
2760 sock->pending_accept = 0;
2761
2762 INSIST(sock->references > 0);
2763 sock->references--; /* the internal event is done with this socket */
2764 if (sock->references == 0) {
2765 UNLOCK(&sock->lock);
2766 destroy(&sock);
2767 return;
2768 }
2769
2770 /*
2771 * Get the first item off the accept list.
2772 * If it is empty, unlock the socket and return.
2773 */
2774 dev = ISC_LIST_HEAD(sock->accept_list);
2775 if (dev == NULL) {
2776 UNLOCK(&sock->lock);
2777 return;
2778 }
2779
2780 /*
2781 * Try to accept the new connection. If the accept fails with
2782 * EAGAIN or EINTR, simply poke the watcher to watch this socket
2783 * again. Also ignore ECONNRESET, which has been reported to
2784 * be spuriously returned on Linux 2.2.19 although it is not
2785 * a documented error for accept(). ECONNABORTED has been
2786 * reported for Solaris 8. The rest are thrown in not because
2787 * we have seen them but because they are ignored by other
2788 * daemons such as BIND 8 and Apache.
2789 */
2790
2791 addrlen = sizeof(dev->newsocket->peer_address.type);
2792 memset(&dev->newsocket->peer_address.type, 0, addrlen);
2793 fd = accept(sock->fd, &dev->newsocket->peer_address.type.sa,
2794 (void *)&addrlen);
2795
2796 #ifdef F_DUPFD
2797 /*
2798 * Leave a space for stdio to work in.
2799 */
2800 if (fd >= 0 && fd < 20) {
2801 int new, tmp;
2802 new = fcntl(fd, F_DUPFD, 20);
2803 tmp = errno;
2804 (void)close(fd);
2805 errno = tmp;
2806 fd = new;
2807 err = "accept/fcntl";
2808 }
2809 #endif
2810
2811 if (fd < 0) {
2812 if (SOFT_ERROR(errno))
2813 goto soft_error;
2814 switch (errno) {
2815 case ENFILE:
2816 case EMFILE:
2817 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2818 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2819 isc_msgcat, ISC_MSGSET_SOCKET,
2820 ISC_MSG_TOOMANYFDS,
2821 "%s: too many open file descriptors",
2822 err);
2823 goto soft_error;
2824
2825 case ENOBUFS:
2826 case ENOMEM:
2827 case ECONNRESET:
2828 case ECONNABORTED:
2829 case EHOSTUNREACH:
2830 case EHOSTDOWN:
2831 case ENETUNREACH:
2832 case ENETDOWN:
2833 case ECONNREFUSED:
2834 #ifdef EPROTO
2835 case EPROTO:
2836 #endif
2837 #ifdef ENONET
2838 case ENONET:
2839 #endif
2840 goto soft_error;
2841 default:
2842 break;
2843 }
2844 isc__strerror(errno, strbuf, sizeof(strbuf));
2845 UNEXPECTED_ERROR(__FILE__, __LINE__,
2846 "internal_accept: %s() %s: %s", err,
2847 isc_msgcat_get(isc_msgcat,
2848 ISC_MSGSET_GENERAL,
2849 ISC_MSG_FAILED,
2850 "failed"),
2851 strbuf);
2852 fd = -1;
2853 result = ISC_R_UNEXPECTED;
2854 } else {
2855 if (addrlen == 0U) {
2856 UNEXPECTED_ERROR(__FILE__, __LINE__,
2857 "internal_accept(): "
2858 "accept() failed to return "
2859 "remote address");
2860
2861 (void)close(fd);
2862 goto soft_error;
2863 } else if (dev->newsocket->peer_address.type.sa.sa_family !=
2864 sock->pf)
2865 {
2866 UNEXPECTED_ERROR(__FILE__, __LINE__,
2867 "internal_accept(): "
2868 "accept() returned peer address "
2869 "family %u (expected %u)",
2870 dev->newsocket->peer_address.
2871 type.sa.sa_family,
2872 sock->pf);
2873 (void)close(fd);
2874 goto soft_error;
2875 } else if (fd >= (int)manager->maxsocks) {
2876 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2877 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2878 isc_msgcat, ISC_MSGSET_SOCKET,
2879 ISC_MSG_TOOMANYFDS,
2880 "accept: "
2881 "file descriptor exceeds limit (%d/%u)",
2882 fd, manager->maxsocks);
2883 (void)close(fd);
2884 goto soft_error;
2885 }
2886 }
2887
2888 if (fd != -1) {
2889 dev->newsocket->peer_address.length = addrlen;
2890 dev->newsocket->pf = sock->pf;
2891 }
2892
2893 /*
2894 * Pull off the done event.
2895 */
2896 ISC_LIST_UNLINK(sock->accept_list, dev, ev_link);
2897
2898 /*
2899 * Poke watcher if there are more pending accepts.
2900 */
2901 if (!ISC_LIST_EMPTY(sock->accept_list))
2902 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
2903
2904 UNLOCK(&sock->lock);
2905
2906 if (fd != -1 && (make_nonblock(fd) != ISC_R_SUCCESS)) {
2907 (void)close(fd);
2908 fd = -1;
2909 result = ISC_R_UNEXPECTED;
2910 }
2911
2912 /*
2913 * -1 means the new socket didn't happen.
2914 */
2915 if (fd != -1) {
2916 int lockid = FDLOCK_ID(fd);
2917
2918 LOCK(&manager->fdlock[lockid]);
2919 manager->fds[fd] = dev->newsocket;
2920 manager->fdstate[fd] = MANAGED;
2921 UNLOCK(&manager->fdlock[lockid]);
2922
2923 LOCK(&manager->lock);
2924 ISC_LIST_APPEND(manager->socklist, dev->newsocket, link);
2925
2926 dev->newsocket->fd = fd;
2927 dev->newsocket->bound = 1;
2928 dev->newsocket->connected = 1;
2929
2930 /*
2931 * Save away the remote address
2932 */
2933 dev->address = dev->newsocket->peer_address;
2934
2935 #ifdef USE_SELECT
2936 if (manager->maxfd < fd)
2937 manager->maxfd = fd;
2938 #endif
2939
2940 socket_log(sock, &dev->newsocket->peer_address, CREATION,
2941 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTEDCXN,
2942 "accepted connection, new socket %p",
2943 dev->newsocket);
2944
2945 UNLOCK(&manager->lock);
2946
2947 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPT]);
2948 } else {
2949 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
2950 dev->newsocket->references--;
2951 free_socket(&dev->newsocket);
2952 }
2953
2954 /*
2955 * Fill in the done event details and send it off.
2956 */
2957 dev->result = result;
2958 task = dev->ev_sender;
2959 dev->ev_sender = sock;
2960
2961 isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
2962 return;
2963
2964 soft_error:
2965 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
2966 UNLOCK(&sock->lock);
2967
2968 inc_stats(manager->stats, sock->statsindex[STATID_ACCEPTFAIL]);
2969 return;
2970 }
2971
2972 static void
2973 internal_recv(isc_task_t *me, isc_event_t *ev) {
2974 isc_socketevent_t *dev;
2975 isc_socket_t *sock;
2976
2977 INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
2978
2979 sock = ev->ev_sender;
2980 INSIST(VALID_SOCKET(sock));
2981
2982 LOCK(&sock->lock);
2983 socket_log(sock, NULL, IOEVENT,
2984 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
2985 "internal_recv: task %p got event %p", me, ev);
2986
2987 INSIST(sock->pending_recv == 1);
2988 sock->pending_recv = 0;
2989
2990 INSIST(sock->references > 0);
2991 sock->references--; /* the internal event is done with this socket */
2992 if (sock->references == 0) {
2993 UNLOCK(&sock->lock);
2994 destroy(&sock);
2995 return;
2996 }
2997
2998 /*
2999 * Try to do as much I/O as possible on this socket. There are no
3000 * limits here, currently.
3001 */
3002 dev = ISC_LIST_HEAD(sock->recv_list);
3003 while (dev != NULL) {
3004 switch (doio_recv(sock, dev)) {
3005 case DOIO_SOFT:
3006 goto poke;
3007
3008 case DOIO_EOF:
3009 /*
3010 * read of 0 means the remote end was closed.
3011 * Run through the event queue and dispatch all
3012 * the events with an EOF result code.
3013 */
3014 do {
3015 dev->result = ISC_R_EOF;
3016 send_recvdone_event(sock, &dev);
3017 dev = ISC_LIST_HEAD(sock->recv_list);
3018 } while (dev != NULL);
3019 goto poke;
3020
3021 case DOIO_SUCCESS:
3022 case DOIO_HARD:
3023 send_recvdone_event(sock, &dev);
3024 break;
3025 }
3026
3027 dev = ISC_LIST_HEAD(sock->recv_list);
3028 }
3029
3030 poke:
3031 if (!ISC_LIST_EMPTY(sock->recv_list))
3032 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3033
3034 UNLOCK(&sock->lock);
3035 }
3036
3037 static void
3038 internal_send(isc_task_t *me, isc_event_t *ev) {
3039 isc_socketevent_t *dev;
3040 isc_socket_t *sock;
3041
3042 INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3043
3044 /*
3045 * Find out what socket this is and lock it.
3046 */
3047 sock = (isc_socket_t *)ev->ev_sender;
3048 INSIST(VALID_SOCKET(sock));
3049
3050 LOCK(&sock->lock);
3051 socket_log(sock, NULL, IOEVENT,
3052 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3053 "internal_send: task %p got event %p", me, ev);
3054
3055 INSIST(sock->pending_send == 1);
3056 sock->pending_send = 0;
3057
3058 INSIST(sock->references > 0);
3059 sock->references--; /* the internal event is done with this socket */
3060 if (sock->references == 0) {
3061 UNLOCK(&sock->lock);
3062 destroy(&sock);
3063 return;
3064 }
3065
3066 /*
3067 * Try to do as much I/O as possible on this socket. There are no
3068 * limits here, currently.
3069 */
3070 dev = ISC_LIST_HEAD(sock->send_list);
3071 while (dev != NULL) {
3072 switch (doio_send(sock, dev)) {
3073 case DOIO_SOFT:
3074 goto poke;
3075
3076 case DOIO_HARD:
3077 case DOIO_SUCCESS:
3078 send_senddone_event(sock, &dev);
3079 break;
3080 }
3081
3082 dev = ISC_LIST_HEAD(sock->send_list);
3083 }
3084
3085 poke:
3086 if (!ISC_LIST_EMPTY(sock->send_list))
3087 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3088
3089 UNLOCK(&sock->lock);
3090 }
3091
3092 static void
3093 internal_fdwatch_write(isc_task_t *me, isc_event_t *ev) {
3094 isc_socket_t *sock;
3095 int more_data;
3096
3097 INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3098
3099 /*
3100 * Find out what socket this is and lock it.
3101 */
3102 sock = (isc_socket_t *)ev->ev_sender;
3103 INSIST(VALID_SOCKET(sock));
3104
3105 LOCK(&sock->lock);
3106 socket_log(sock, NULL, IOEVENT,
3107 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
3108 "internal_fdwatch_write: task %p got event %p", me, ev);
3109
3110 INSIST(sock->pending_send == 1);
3111
3112 UNLOCK(&sock->lock);
3113 more_data = (sock->fdwatchcb)(me, sock, sock->fdwatcharg);
3114 LOCK(&sock->lock);
3115
3116 sock->pending_send = 0;
3117
3118 INSIST(sock->references > 0);
3119 sock->references--; /* the internal event is done with this socket */
3120 if (sock->references == 0) {
3121 UNLOCK(&sock->lock);
3122 destroy(&sock);
3123 return;
3124 }
3125
3126 if (more_data)
3127 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
3128
3129 UNLOCK(&sock->lock);
3130 }
3131
3132 static void
3133 internal_fdwatch_read(isc_task_t *me, isc_event_t *ev) {
3134 isc_socket_t *sock;
3135 int more_data;
3136
3137 INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
3138
3139 /*
3140 * Find out what socket this is and lock it.
3141 */
3142 sock = (isc_socket_t *)ev->ev_sender;
3143 INSIST(VALID_SOCKET(sock));
3144
3145 LOCK(&sock->lock);
3146 socket_log(sock, NULL, IOEVENT,
3147 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
3148 "internal_fdwatch_read: task %p got event %p", me, ev);
3149
3150 INSIST(sock->pending_recv == 1);
3151
3152 UNLOCK(&sock->lock);
3153 more_data = (sock->fdwatchcb)(me, sock, sock->fdwatcharg);
3154 LOCK(&sock->lock);
3155
3156 sock->pending_recv = 0;
3157
3158 INSIST(sock->references > 0);
3159 sock->references--; /* the internal event is done with this socket */
3160 if (sock->references == 0) {
3161 UNLOCK(&sock->lock);
3162 destroy(&sock);
3163 return;
3164 }
3165
3166 if (more_data)
3167 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
3168
3169 UNLOCK(&sock->lock);
3170 }
3171
3172 /*
3173 * Process read/writes on each fd here. Avoid locking
3174 * and unlocking twice if both reads and writes are possible.
3175 */
3176 static void
3177 process_fd(isc_socketmgr_t *manager, int fd, isc_boolean_t readable,
3178 isc_boolean_t writeable)
3179 {
3180 isc_socket_t *sock;
3181 isc_boolean_t unlock_sock;
3182 isc_boolean_t unwatch_read = ISC_FALSE, unwatch_write = ISC_FALSE;
3183 int lockid = FDLOCK_ID(fd);
3184
3185 /*
3186 * If the socket is going to be closed, don't do more I/O.
3187 */
3188 LOCK(&manager->fdlock[lockid]);
3189 if (manager->fdstate[fd] == CLOSE_PENDING) {
3190 UNLOCK(&manager->fdlock[lockid]);
3191
3192 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3193 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3194 return;
3195 }
3196
3197 sock = manager->fds[fd];
3198 unlock_sock = ISC_FALSE;
3199 if (readable) {
3200 if (sock == NULL) {
3201 unwatch_read = ISC_TRUE;
3202 goto check_write;
3203 }
3204 unlock_sock = ISC_TRUE;
3205 LOCK(&sock->lock);
3206 if (!SOCK_DEAD(sock)) {
3207 if (sock->listener)
3208 dispatch_accept(sock);
3209 else
3210 dispatch_recv(sock);
3211 }
3212 unwatch_read = ISC_TRUE;
3213 }
3214 check_write:
3215 if (writeable) {
3216 if (sock == NULL) {
3217 unwatch_write = ISC_TRUE;
3218 goto unlock_fd;
3219 }
3220 if (!unlock_sock) {
3221 unlock_sock = ISC_TRUE;
3222 LOCK(&sock->lock);
3223 }
3224 if (!SOCK_DEAD(sock)) {
3225 if (sock->connecting)
3226 dispatch_connect(sock);
3227 else
3228 dispatch_send(sock);
3229 }
3230 unwatch_write = ISC_TRUE;
3231 }
3232 if (unlock_sock)
3233 UNLOCK(&sock->lock);
3234
3235 unlock_fd:
3236 UNLOCK(&manager->fdlock[lockid]);
3237 if (unwatch_read)
3238 (void)unwatch_fd(manager, fd, SELECT_POKE_READ);
3239 if (unwatch_write)
3240 (void)unwatch_fd(manager, fd, SELECT_POKE_WRITE);
3241
3242 }
3243
3244 #ifdef USE_KQUEUE
3245 static isc_boolean_t
3246 process_fds(isc_socketmgr_t *manager, struct kevent *events, int nevents) {
3247 int i;
3248 isc_boolean_t readable, writable;
3249 isc_boolean_t done = ISC_FALSE;
3250 #ifdef ISC_PLATFORM_USETHREADS
3251 isc_boolean_t have_ctlevent = ISC_FALSE;
3252 #endif
3253
3254 if (nevents == manager->nevents) {
3255 /*
3256 * This is not an error, but something unexpected. If this
3257 * happens, it may indicate the need for increasing
3258 * ISC_SOCKET_MAXEVENTS.
3259 */
3260 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3261 ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3262 "maximum number of FD events (%d) received",
3263 nevents);
3264 }
3265
3266 for (i = 0; i < nevents; i++) {
3267 REQUIRE(events[i].ident < manager->maxsocks);
3268 #ifdef ISC_PLATFORM_USETHREADS
3269 if (events[i].ident == (uintptr_t)manager->pipe_fds[0]) {
3270 have_ctlevent = ISC_TRUE;
3271 continue;
3272 }
3273 #endif
3274 readable = ISC_TF(events[i].filter == EVFILT_READ);
3275 writable = ISC_TF(events[i].filter == EVFILT_WRITE);
3276 process_fd(manager, events[i].ident, readable, writable);
3277 }
3278
3279 #ifdef ISC_PLATFORM_USETHREADS
3280 if (have_ctlevent)
3281 done = process_ctlfd(manager);
3282 #endif
3283
3284 return (done);
3285 }
3286 #elif defined(USE_EPOLL)
3287 static isc_boolean_t
3288 process_fds(isc_socketmgr_t *manager, struct epoll_event *events, int nevents) {
3289 int i;
3290 isc_boolean_t done = ISC_FALSE;
3291 #ifdef ISC_PLATFORM_USETHREADS
3292 isc_boolean_t have_ctlevent = ISC_FALSE;
3293 #endif
3294
3295 if (nevents == manager->nevents) {
3296 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3297 ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3298 "maximum number of FD events (%d) received",
3299 nevents);
3300 }
3301
3302 for (i = 0; i < nevents; i++) {
3303 REQUIRE(events[i].data.fd < (int)manager->maxsocks);
3304 #ifdef ISC_PLATFORM_USETHREADS
3305 if (events[i].data.fd == manager->pipe_fds[0]) {
3306 have_ctlevent = ISC_TRUE;
3307 continue;
3308 }
3309 #endif
3310 if ((events[i].events & EPOLLERR) != 0 ||
3311 (events[i].events & EPOLLHUP) != 0) {
3312 /*
3313 * epoll does not set IN/OUT bits on an erroneous
3314 * condition, so we need to try both anyway. This is a
3315 * bit inefficient, but should be okay for such rare
3316 * events. Note also that the read or write attempt
3317 * won't block because we use non-blocking sockets.
3318 */
3319 events[i].events |= (EPOLLIN | EPOLLOUT);
3320 }
3321 process_fd(manager, events[i].data.fd,
3322 (events[i].events & EPOLLIN) != 0,
3323 (events[i].events & EPOLLOUT) != 0);
3324 }
3325
3326 #ifdef ISC_PLATFORM_USETHREADS
3327 if (have_ctlevent)
3328 done = process_ctlfd(manager);
3329 #endif
3330
3331 return (done);
3332 }
3333 #elif defined(USE_DEVPOLL)
3334 static isc_boolean_t
3335 process_fds(isc_socketmgr_t *manager, struct pollfd *events, int nevents) {
3336 int i;
3337 isc_boolean_t done = ISC_FALSE;
3338 #ifdef ISC_PLATFORM_USETHREADS
3339 isc_boolean_t have_ctlevent = ISC_FALSE;
3340 #endif
3341
3342 if (nevents == manager->nevents) {
3343 manager_log(manager, ISC_LOGCATEGORY_GENERAL,
3344 ISC_LOGMODULE_SOCKET, ISC_LOG_INFO,
3345 "maximum number of FD events (%d) received",
3346 nevents);
3347 }
3348
3349 for (i = 0; i < nevents; i++) {
3350 REQUIRE(events[i].fd < (int)manager->maxsocks);
3351 #ifdef ISC_PLATFORM_USETHREADS
3352 if (events[i].fd == manager->pipe_fds[0]) {
3353 have_ctlevent = ISC_TRUE;
3354 continue;
3355 }
3356 #endif
3357 process_fd(manager, events[i].fd,
3358 (events[i].events & POLLIN) != 0,
3359 (events[i].events & POLLOUT) != 0);
3360 }
3361
3362 #ifdef ISC_PLATFORM_USETHREADS
3363 if (have_ctlevent)
3364 done = process_ctlfd(manager);
3365 #endif
3366
3367 return (done);
3368 }
3369 #elif defined(USE_SELECT)
3370 static void
3371 process_fds(isc_socketmgr_t *manager, int maxfd,
3372 fd_set *readfds, fd_set *writefds)
3373 {
3374 int i;
3375
3376 REQUIRE(maxfd <= (int)manager->maxsocks);
3377
3378 for (i = 0; i < maxfd; i++) {
3379 #ifdef ISC_PLATFORM_USETHREADS
3380 if (i == manager->pipe_fds[0] || i == manager->pipe_fds[1])
3381 continue;
3382 #endif /* ISC_PLATFORM_USETHREADS */
3383 process_fd(manager, i, FD_ISSET(i, readfds),
3384 FD_ISSET(i, writefds));
3385 }
3386 }
3387 #endif
3388
3389 #ifdef ISC_PLATFORM_USETHREADS
3390 static isc_boolean_t
3391 process_ctlfd(isc_socketmgr_t *manager) {
3392 int msg, fd;
3393
3394 for (;;) {
3395 select_readmsg(manager, &fd, &msg);
3396
3397 manager_log(manager, IOEVENT,
3398 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
3399 ISC_MSG_WATCHERMSG,
3400 "watcher got message %d "
3401 "for socket %d"), msg, fd);
3402
3403 /*
3404 * Nothing to read?
3405 */
3406 if (msg == SELECT_POKE_NOTHING)
3407 break;
3408
3409 /*
3410 * Handle shutdown message. We really should
3411 * jump out of this loop right away, but
3412 * it doesn't matter if we have to do a little
3413 * more work first.
3414 */
3415 if (msg == SELECT_POKE_SHUTDOWN)
3416 return (ISC_TRUE);
3417
3418 /*
3419 * This is a wakeup on a socket. Look
3420 * at the event queue for both read and write,
3421 * and decide if we need to watch on it now
3422 * or not.
3423 */
3424 wakeup_socket(manager, fd, msg);
3425 }
3426
3427 return (ISC_FALSE);
3428 }
3429
3430 /*
3431 * This is the thread that will loop forever, always in a select or poll
3432 * call.
3433 *
3434 * When select returns something to do, track down what thread gets to do
3435 * this I/O and post the event to it.
3436 */
3437 static isc_threadresult_t
3438 watcher(void *uap) {
3439 isc_socketmgr_t *manager = uap;
3440 isc_boolean_t done;
3441 int ctlfd;
3442 int cc;
3443 #ifdef USE_KQUEUE
3444 const char *fnname = "kevent()";
3445 #elif defined (USE_EPOLL)
3446 const char *fnname = "epoll_wait()";
3447 #elif defined(USE_DEVPOLL)
3448 const char *fnname = "ioctl(DP_POLL)";
3449 struct dvpoll dvp;
3450 #elif defined (USE_SELECT)
3451 const char *fnname = "select()";
3452 int maxfd;
3453 #endif
3454 char strbuf[ISC_STRERRORSIZE];
3455 #ifdef ISC_SOCKET_USE_POLLWATCH
3456 pollstate_t pollstate = poll_idle;
3457 #endif
3458
3459 /*
3460 * Get the control fd here. This will never change.
3461 */
3462 ctlfd = manager->pipe_fds[0];
3463 done = ISC_FALSE;
3464 while (!done) {
3465 do {
3466 #ifdef USE_KQUEUE
3467 cc = kevent(manager->kqueue_fd, NULL, 0,
3468 manager->events, manager->nevents, NULL);
3469 #elif defined(USE_EPOLL)
3470 cc = epoll_wait(manager->epoll_fd, manager->events,
3471 manager->nevents, -1);
3472 #elif defined(USE_DEVPOLL)
3473 dvp.dp_fds = manager->events;
3474 dvp.dp_nfds = manager->nevents;
3475 #ifndef ISC_SOCKET_USE_POLLWATCH
3476 dvp.dp_timeout = -1;
3477 #else
3478 if (pollstate == poll_idle)
3479 dvp.dp_timeout = -1;
3480 else
3481 dvp.dp_timeout = ISC_SOCKET_POLLWATCH_TIMEOUT;
3482 #endif /* ISC_SOCKET_USE_POLLWATCH */
3483 cc = ioctl(manager->devpoll_fd, DP_POLL, &dvp);
3484 #elif defined(USE_SELECT)
3485 LOCK(&manager->lock);
3486 memcpy(manager->read_fds_copy, manager->read_fds,
3487 manager->fd_bufsize);
3488 memcpy(manager->write_fds_copy, manager->write_fds,
3489 manager->fd_bufsize);
3490 maxfd = manager->maxfd + 1;
3491 UNLOCK(&manager->lock);
3492
3493 cc = select(maxfd, manager->read_fds_copy,
3494 manager->write_fds_copy, NULL, NULL);
3495 #endif /* USE_KQUEUE */
3496
3497 if (cc < 0 && !SOFT_ERROR(errno)) {
3498 isc__strerror(errno, strbuf, sizeof(strbuf));
3499 FATAL_ERROR(__FILE__, __LINE__,
3500 "%s %s: %s", fnname,
3501 isc_msgcat_get(isc_msgcat,
3502 ISC_MSGSET_GENERAL,
3503 ISC_MSG_FAILED,
3504 "failed"), strbuf);
3505 }
3506
3507 #if defined(USE_DEVPOLL) && defined(ISC_SOCKET_USE_POLLWATCH)
3508 if (cc == 0) {
3509 if (pollstate == poll_active)
3510 pollstate = poll_checking;
3511 else if (pollstate == poll_checking)
3512 pollstate = poll_idle;
3513 } else if (cc > 0) {
3514 if (pollstate == poll_checking) {
3515 /*
3516 * XXX: We'd like to use a more
3517 * verbose log level as it's actually an
3518 * unexpected event, but the kernel bug
3519 * reportedly happens pretty frequently
3520 * (and it can also be a false positive)
3521 * so it would be just too noisy.
3522 */
3523 manager_log(manager,
3524 ISC_LOGCATEGORY_GENERAL,
3525 ISC_LOGMODULE_SOCKET,
3526 ISC_LOG_DEBUG(1),
3527 "unexpected POLL timeout");
3528 }
3529 pollstate = poll_active;
3530 }
3531 #endif
3532 } while (cc < 0);
3533
3534 #if defined(USE_KQUEUE) || defined (USE_EPOLL) || defined (USE_DEVPOLL)
3535 done = process_fds(manager, manager->events, cc);
3536 #elif defined(USE_SELECT)
3537 process_fds(manager, maxfd, manager->read_fds_copy,
3538 manager->write_fds_copy);
3539
3540 /*
3541 * Process reads on internal, control fd.
3542 */
3543 if (FD_ISSET(ctlfd, manager->read_fds_copy))
3544 done = process_ctlfd(manager);
3545 #endif
3546 }
3547
3548 manager_log(manager, TRACE, "%s",
3549 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3550 ISC_MSG_EXITING, "watcher exiting"));
3551
3552 return ((isc_threadresult_t)0);
3553 }
3554 #endif /* ISC_PLATFORM_USETHREADS */
3555
3556 void
3557 isc__socketmgr_setreserved(isc_socketmgr_t *manager, isc_uint32_t reserved) {
3558
3559 REQUIRE(VALID_MANAGER(manager));
3560
3561 manager->reserved = reserved;
3562 }
3563
3564 /*
3565 * Create a new socket manager.
3566 */
3567
3568 static isc_result_t
3569 setup_watcher(isc_mem_t *mctx, isc_socketmgr_t *manager) {
3570 isc_result_t result;
3571 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
3572 char strbuf[ISC_STRERRORSIZE];
3573 #endif
3574
3575 #ifdef USE_KQUEUE
3576 manager->nevents = ISC_SOCKET_MAXEVENTS;
3577 manager->events = isc_mem_get(mctx, sizeof(struct kevent) *
3578 manager->nevents);
3579 if (manager->events == NULL)
3580 return (ISC_R_NOMEMORY);
3581 manager->kqueue_fd = kqueue();
3582 if (manager->kqueue_fd == -1) {
3583 result = isc__errno2result(errno);
3584 isc__strerror(errno, strbuf, sizeof(strbuf));
3585 UNEXPECTED_ERROR(__FILE__, __LINE__,
3586 "kqueue %s: %s",
3587 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3588 ISC_MSG_FAILED, "failed"),
3589 strbuf);
3590 isc_mem_put(mctx, manager->events,
3591 sizeof(struct kevent) * manager->nevents);
3592 return (result);
3593 }
3594
3595 #ifdef ISC_PLATFORM_USETHREADS
3596 result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3597 if (result != ISC_R_SUCCESS) {
3598 close(manager->kqueue_fd);
3599 isc_mem_put(mctx, manager->events,
3600 sizeof(struct kevent) * manager->nevents);
3601 return (result);
3602 }
3603 #endif /* ISC_PLATFORM_USETHREADS */
3604 #elif defined(USE_EPOLL)
3605 manager->nevents = ISC_SOCKET_MAXEVENTS;
3606 manager->events = isc_mem_get(mctx, sizeof(struct epoll_event) *
3607 manager->nevents);
3608 if (manager->events == NULL)
3609 return (ISC_R_NOMEMORY);
3610 manager->epoll_fd = epoll_create(manager->nevents);
3611 if (manager->epoll_fd == -1) {
3612 result = isc__errno2result(errno);
3613 isc__strerror(errno, strbuf, sizeof(strbuf));
3614 UNEXPECTED_ERROR(__FILE__, __LINE__,
3615 "epoll_create %s: %s",
3616 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3617 ISC_MSG_FAILED, "failed"),
3618 strbuf);
3619 isc_mem_put(mctx, manager->events,
3620 sizeof(struct epoll_event) * manager->nevents);
3621 return (result);
3622 }
3623 #ifdef ISC_PLATFORM_USETHREADS
3624 result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3625 if (result != ISC_R_SUCCESS) {
3626 close(manager->epoll_fd);
3627 isc_mem_put(mctx, manager->events,
3628 sizeof(struct epoll_event) * manager->nevents);
3629 return (result);
3630 }
3631 #endif /* ISC_PLATFORM_USETHREADS */
3632 #elif defined(USE_DEVPOLL)
3633 /*
3634 * XXXJT: /dev/poll seems to reject large numbers of events,
3635 * so we should be careful about redefining ISC_SOCKET_MAXEVENTS.
3636 */
3637 manager->nevents = ISC_SOCKET_MAXEVENTS;
3638 manager->events = isc_mem_get(mctx, sizeof(struct pollfd) *
3639 manager->nevents);
3640 if (manager->events == NULL)
3641 return (ISC_R_NOMEMORY);
3642 /*
3643 * Note: fdpollinfo should be able to support all possible FDs, so
3644 * it must have maxsocks entries (not nevents).
3645 */
3646 manager->fdpollinfo = isc_mem_get(mctx, sizeof(pollinfo_t) *
3647 manager->maxsocks);
3648 if (manager->fdpollinfo == NULL) {
3649 isc_mem_put(mctx, manager->events,
3650 sizeof(pollinfo_t) * manager->maxsocks);
3651 return (ISC_R_NOMEMORY);
3652 }
3653 memset(manager->fdpollinfo, 0, sizeof(pollinfo_t) * manager->maxsocks);
3654 manager->devpoll_fd = open("/dev/poll", O_RDWR);
3655 if (manager->devpoll_fd == -1) {
3656 result = isc__errno2result(errno);
3657 isc__strerror(errno, strbuf, sizeof(strbuf));
3658 UNEXPECTED_ERROR(__FILE__, __LINE__,
3659 "open(/dev/poll) %s: %s",
3660 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3661 ISC_MSG_FAILED, "failed"),
3662 strbuf);
3663 isc_mem_put(mctx, manager->events,
3664 sizeof(struct pollfd) * manager->nevents);
3665 isc_mem_put(mctx, manager->fdpollinfo,
3666 sizeof(pollinfo_t) * manager->maxsocks);
3667 return (result);
3668 }
3669 #ifdef ISC_PLATFORM_USETHREADS
3670 result = watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3671 if (result != ISC_R_SUCCESS) {
3672 close(manager->devpoll_fd);
3673 isc_mem_put(mctx, manager->events,
3674 sizeof(struct pollfd) * manager->nevents);
3675 isc_mem_put(mctx, manager->fdpollinfo,
3676 sizeof(pollinfo_t) * manager->maxsocks);
3677 return (result);
3678 }
3679 #endif /* ISC_PLATFORM_USETHREADS */
3680 #elif defined(USE_SELECT)
3681 UNUSED(result);
3682
3683 #if ISC_SOCKET_MAXSOCKETS > FD_SETSIZE
3684 /*
3685 * Note: this code should also cover the case of MAXSOCKETS <=
3686 * FD_SETSIZE, but we separate the cases to avoid possible portability
3687 * issues regarding howmany() and the actual representation of fd_set.
3688 */
3689 manager->fd_bufsize = howmany(manager->maxsocks, NFDBITS) *
3690 sizeof(fd_mask);
3691 #else
3692 manager->fd_bufsize = sizeof(fd_set);
3693 #endif
3694
3695 manager->read_fds = NULL;
3696 manager->read_fds_copy = NULL;
3697 manager->write_fds = NULL;
3698 manager->write_fds_copy = NULL;
3699
3700 manager->read_fds = isc_mem_get(mctx, manager->fd_bufsize);
3701 if (manager->read_fds != NULL)
3702 manager->read_fds_copy = isc_mem_get(mctx, manager->fd_bufsize);
3703 if (manager->read_fds_copy != NULL)
3704 manager->write_fds = isc_mem_get(mctx, manager->fd_bufsize);
3705 if (manager->write_fds != NULL) {
3706 manager->write_fds_copy = isc_mem_get(mctx,
3707 manager->fd_bufsize);
3708 }
3709 if (manager->write_fds_copy == NULL) {
3710 if (manager->write_fds != NULL) {
3711 isc_mem_put(mctx, manager->write_fds,
3712 manager->fd_bufsize);
3713 }
3714 if (manager->read_fds_copy != NULL) {
3715 isc_mem_put(mctx, manager->read_fds_copy,
3716 manager->fd_bufsize);
3717 }
3718 if (manager->read_fds != NULL) {
3719 isc_mem_put(mctx, manager->read_fds,
3720 manager->fd_bufsize);
3721 }
3722 return (ISC_R_NOMEMORY);
3723 }
3724 memset(manager->read_fds, 0, manager->fd_bufsize);
3725 memset(manager->write_fds, 0, manager->fd_bufsize);
3726
3727 #ifdef ISC_PLATFORM_USETHREADS
3728 (void)watch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3729 manager->maxfd = manager->pipe_fds[0];
3730 #else /* ISC_PLATFORM_USETHREADS */
3731 manager->maxfd = 0;
3732 #endif /* ISC_PLATFORM_USETHREADS */
3733 #endif /* USE_KQUEUE */
3734
3735 return (ISC_R_SUCCESS);
3736 }
3737
3738 static void
3739 cleanup_watcher(isc_mem_t *mctx, isc_socketmgr_t *manager) {
3740 #ifdef ISC_PLATFORM_USETHREADS
3741 isc_result_t result;
3742
3743 result = unwatch_fd(manager, manager->pipe_fds[0], SELECT_POKE_READ);
3744 if (result != ISC_R_SUCCESS) {
3745 UNEXPECTED_ERROR(__FILE__, __LINE__,
3746 "epoll_ctl(DEL) %s",
3747 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3748 ISC_MSG_FAILED, "failed"));
3749 }
3750 #endif /* ISC_PLATFORM_USETHREADS */
3751
3752 #ifdef USE_KQUEUE
3753 close(manager->kqueue_fd);
3754 isc_mem_put(mctx, manager->events,
3755 sizeof(struct kevent) * manager->nevents);
3756 #elif defined(USE_EPOLL)
3757 close(manager->epoll_fd);
3758 isc_mem_put(mctx, manager->events,
3759 sizeof(struct epoll_event) * manager->nevents);
3760 #elif defined(USE_DEVPOLL)
3761 close(manager->devpoll_fd);
3762 isc_mem_put(mctx, manager->events,
3763 sizeof(struct pollfd) * manager->nevents);
3764 isc_mem_put(mctx, manager->fdpollinfo,
3765 sizeof(pollinfo_t) * manager->maxsocks);
3766 #elif defined(USE_SELECT)
3767 if (manager->read_fds != NULL)
3768 isc_mem_put(mctx, manager->read_fds, manager->fd_bufsize);
3769 if (manager->read_fds_copy != NULL)
3770 isc_mem_put(mctx, manager->read_fds_copy, manager->fd_bufsize);
3771 if (manager->write_fds != NULL)
3772 isc_mem_put(mctx, manager->write_fds, manager->fd_bufsize);
3773 if (manager->write_fds_copy != NULL)
3774 isc_mem_put(mctx, manager->write_fds_copy, manager->fd_bufsize);
3775 #endif /* USE_KQUEUE */
3776 }
3777
3778 isc_result_t
3779 isc_socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp) {
3780 return (isc_socketmgr_create2(mctx, managerp, 0));
3781 }
3782
3783 isc_result_t
3784 isc_socketmgr_create2(isc_mem_t *mctx, isc_socketmgr_t **managerp,
3785 unsigned int maxsocks)
3786 {
3787 int i;
3788 isc_socketmgr_t *manager;
3789 #ifdef ISC_PLATFORM_USETHREADS
3790 char strbuf[ISC_STRERRORSIZE];
3791 #endif
3792 isc_result_t result;
3793
3794 REQUIRE(managerp != NULL && *managerp == NULL);
3795
3796 #ifndef ISC_PLATFORM_USETHREADS
3797 if (socketmgr != NULL) {
3798 /* Don't allow maxsocks to be updated */
3799 if (maxsocks > 0 && socketmgr->maxsocks != maxsocks)
3800 return (ISC_R_EXISTS);
3801
3802 socketmgr->refs++;
3803 *managerp = socketmgr;
3804 return (ISC_R_SUCCESS);
3805 }
3806 #endif /* ISC_PLATFORM_USETHREADS */
3807
3808 if (maxsocks == 0)
3809 maxsocks = ISC_SOCKET_MAXSOCKETS;
3810
3811 manager = isc_mem_get(mctx, sizeof(*manager));
3812 if (manager == NULL)
3813 return (ISC_R_NOMEMORY);
3814
3815 /* zero-clear so that necessary cleanup on failure will be easy */
3816 memset(manager, 0, sizeof(*manager));
3817 manager->maxsocks = maxsocks;
3818 manager->reserved = 0;
3819 manager->fds = isc_mem_get(mctx,
3820 manager->maxsocks * sizeof(isc_socket_t *));
3821 if (manager->fds == NULL) {
3822 result = ISC_R_NOMEMORY;
3823 goto free_manager;
3824 }
3825 manager->fdstate = isc_mem_get(mctx, manager->maxsocks * sizeof(int));
3826 if (manager->fdstate == NULL) {
3827 result = ISC_R_NOMEMORY;
3828 goto free_manager;
3829 }
3830 manager->stats = NULL;
3831
3832 manager->magic = SOCKET_MANAGER_MAGIC;
3833 manager->mctx = NULL;
3834 memset(manager->fds, 0, manager->maxsocks * sizeof(isc_socket_t *));
3835 ISC_LIST_INIT(manager->socklist);
3836 result = isc_mutex_init(&manager->lock);
3837 if (result != ISC_R_SUCCESS)
3838 goto free_manager;
3839 manager->fdlock = isc_mem_get(mctx, FDLOCK_COUNT * sizeof(isc_mutex_t));
3840 if (manager->fdlock == NULL) {
3841 result = ISC_R_NOMEMORY;
3842 goto cleanup_lock;
3843 }
3844 for (i = 0; i < FDLOCK_COUNT; i++) {
3845 result = isc_mutex_init(&manager->fdlock[i]);
3846 if (result != ISC_R_SUCCESS) {
3847 while (--i >= 0)
3848 DESTROYLOCK(&manager->fdlock[i]);
3849 isc_mem_put(mctx, manager->fdlock,
3850 FDLOCK_COUNT * sizeof(isc_mutex_t));
3851 manager->fdlock = NULL;
3852 goto cleanup_lock;
3853 }
3854 }
3855
3856 #ifdef ISC_PLATFORM_USETHREADS
3857 if (isc_condition_init(&manager->shutdown_ok) != ISC_R_SUCCESS) {
3858 UNEXPECTED_ERROR(__FILE__, __LINE__,
3859 "isc_condition_init() %s",
3860 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3861 ISC_MSG_FAILED, "failed"));
3862 result = ISC_R_UNEXPECTED;
3863 goto cleanup_lock;
3864 }
3865
3866 /*
3867 * Create the special fds that will be used to wake up the
3868 * select/poll loop when something internal needs to be done.
3869 */
3870 if (pipe(manager->pipe_fds) != 0) {
3871 isc__strerror(errno, strbuf, sizeof(strbuf));
3872 UNEXPECTED_ERROR(__FILE__, __LINE__,
3873 "pipe() %s: %s",
3874 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3875 ISC_MSG_FAILED, "failed"),
3876 strbuf);
3877 result = ISC_R_UNEXPECTED;
3878 goto cleanup_condition;
3879 }
3880
3881 RUNTIME_CHECK(make_nonblock(manager->pipe_fds[0]) == ISC_R_SUCCESS);
3882 #if 0
3883 RUNTIME_CHECK(make_nonblock(manager->pipe_fds[1]) == ISC_R_SUCCESS);
3884 #endif
3885 #else /* ISC_PLATFORM_USETHREADS */
3886 manager->refs = 1;
3887 #endif /* ISC_PLATFORM_USETHREADS */
3888
3889 /*
3890 * Set up initial state for the select loop
3891 */
3892 result = setup_watcher(mctx, manager);
3893 if (result != ISC_R_SUCCESS)
3894 goto cleanup;
3895 memset(manager->fdstate, 0, manager->maxsocks * sizeof(int));
3896 #ifdef ISC_PLATFORM_USETHREADS
3897 /*
3898 * Start up the select/poll thread.
3899 */
3900 if (isc_thread_create(watcher, manager, &manager->watcher) !=
3901 ISC_R_SUCCESS) {
3902 UNEXPECTED_ERROR(__FILE__, __LINE__,
3903 "isc_thread_create() %s",
3904 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3905 ISC_MSG_FAILED, "failed"));
3906 cleanup_watcher(mctx, manager);
3907 result = ISC_R_UNEXPECTED;
3908 goto cleanup;
3909 }
3910 #endif /* ISC_PLATFORM_USETHREADS */
3911 isc_mem_attach(mctx, &manager->mctx);
3912
3913 #ifndef ISC_PLATFORM_USETHREADS
3914 socketmgr = manager;
3915 #endif /* ISC_PLATFORM_USETHREADS */
3916 *managerp = manager;
3917
3918 return (ISC_R_SUCCESS);
3919
3920 cleanup:
3921 #ifdef ISC_PLATFORM_USETHREADS
3922 (void)close(manager->pipe_fds[0]);
3923 (void)close(manager->pipe_fds[1]);
3924 #endif /* ISC_PLATFORM_USETHREADS */
3925
3926 #ifdef ISC_PLATFORM_USETHREADS
3927 cleanup_condition:
3928 (void)isc_condition_destroy(&manager->shutdown_ok);
3929 #endif /* ISC_PLATFORM_USETHREADS */
3930
3931
3932 cleanup_lock:
3933 if (manager->fdlock != NULL) {
3934 for (i = 0; i < FDLOCK_COUNT; i++)
3935 DESTROYLOCK(&manager->fdlock[i]);
3936 }
3937 DESTROYLOCK(&manager->lock);
3938
3939 free_manager:
3940 if (manager->fdlock != NULL) {
3941 isc_mem_put(mctx, manager->fdlock,
3942 FDLOCK_COUNT * sizeof(isc_mutex_t));
3943 }
3944 if (manager->fdstate != NULL) {
3945 isc_mem_put(mctx, manager->fdstate,
3946 manager->maxsocks * sizeof(int));
3947 }
3948 if (manager->fds != NULL) {
3949 isc_mem_put(mctx, manager->fds,
3950 manager->maxsocks * sizeof(isc_socket_t *));
3951 }
3952 isc_mem_put(mctx, manager, sizeof(*manager));
3953
3954 return (result);
3955 }
3956
3957 isc_result_t
3958 isc_socketmgr_getmaxsockets(isc_socketmgr_t *manager, unsigned int *nsockp) {
3959 REQUIRE(VALID_MANAGER(manager));
3960 REQUIRE(nsockp != NULL);
3961
3962 *nsockp = manager->maxsocks;
3963
3964 return (ISC_R_SUCCESS);
3965 }
3966
3967 void
3968 isc_socketmgr_setstats(isc_socketmgr_t *manager, isc_stats_t *stats) {
3969 REQUIRE(VALID_MANAGER(manager));
3970 REQUIRE(ISC_LIST_EMPTY(manager->socklist));
3971 REQUIRE(manager->stats == NULL);
3972 REQUIRE(isc_stats_ncounters(stats) == isc_sockstatscounter_max);
3973
3974 isc_stats_attach(stats, &manager->stats);
3975 }
3976
3977 void
3978 isc_socketmgr_destroy(isc_socketmgr_t **managerp) {
3979 isc_socketmgr_t *manager;
3980 int i;
3981 isc_mem_t *mctx;
3982
3983 /*
3984 * Destroy a socket manager.
3985 */
3986
3987 REQUIRE(managerp != NULL);
3988 manager = *managerp;
3989 REQUIRE(VALID_MANAGER(manager));
3990
3991 #ifndef ISC_PLATFORM_USETHREADS
3992 if (manager->refs > 1) {
3993 manager->refs--;
3994 *managerp = NULL;
3995 return;
3996 }
3997 #endif /* ISC_PLATFORM_USETHREADS */
3998
3999 LOCK(&manager->lock);
4000
4001 #ifdef ISC_PLATFORM_USETHREADS
4002 /*
4003 * Wait for all sockets to be destroyed.
4004 */
4005 while (!ISC_LIST_EMPTY(manager->socklist)) {
4006 manager_log(manager, CREATION, "%s",
4007 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
4008 ISC_MSG_SOCKETSREMAIN,
4009 "sockets exist"));
4010 WAIT(&manager->shutdown_ok, &manager->lock);
4011 }
4012 #else /* ISC_PLATFORM_USETHREADS */
4013 /*
4014 * Hope all sockets have been destroyed.
4015 */
4016 if (!ISC_LIST_EMPTY(manager->socklist)) {
4017 manager_log(manager, CREATION, "%s",
4018 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
4019 ISC_MSG_SOCKETSREMAIN,
4020 "sockets exist"));
4021 INSIST(0);
4022 }
4023 #endif /* ISC_PLATFORM_USETHREADS */
4024
4025 UNLOCK(&manager->lock);
4026
4027 /*
4028 * Here, poke our select/poll thread. Do this by closing the write
4029 * half of the pipe, which will send EOF to the read half.
4030 * This is currently a no-op in the non-threaded case.
4031 */
4032 select_poke(manager, 0, SELECT_POKE_SHUTDOWN);
4033
4034 #ifdef ISC_PLATFORM_USETHREADS
4035 /*
4036 * Wait for thread to exit.
4037 */
4038 if (isc_thread_join(manager->watcher, NULL) != ISC_R_SUCCESS)
4039 UNEXPECTED_ERROR(__FILE__, __LINE__,
4040 "isc_thread_join() %s",
4041 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4042 ISC_MSG_FAILED, "failed"));
4043 #endif /* ISC_PLATFORM_USETHREADS */
4044
4045 /*
4046 * Clean up.
4047 */
4048 cleanup_watcher(manager->mctx, manager);
4049
4050 #ifdef ISC_PLATFORM_USETHREADS
4051 (void)close(manager->pipe_fds[0]);
4052 (void)close(manager->pipe_fds[1]);
4053 (void)isc_condition_destroy(&manager->shutdown_ok);
4054 #endif /* ISC_PLATFORM_USETHREADS */
4055
4056 for (i = 0; i < (int)manager->maxsocks; i++)
4057 if (manager->fdstate[i] == CLOSE_PENDING) /* no need to lock */
4058 (void)close(i);
4059
4060 isc_mem_put(manager->mctx, manager->fds,
4061 manager->maxsocks * sizeof(isc_socket_t *));
4062 isc_mem_put(manager->mctx, manager->fdstate,
4063 manager->maxsocks * sizeof(int));
4064
4065 if (manager->stats != NULL)
4066 isc_stats_detach(&manager->stats);
4067
4068 if (manager->fdlock != NULL) {
4069 for (i = 0; i < FDLOCK_COUNT; i++)
4070 DESTROYLOCK(&manager->fdlock[i]);
4071 isc_mem_put(manager->mctx, manager->fdlock,
4072 FDLOCK_COUNT * sizeof(isc_mutex_t));
4073 }
4074 DESTROYLOCK(&manager->lock);
4075 manager->magic = 0;
4076 mctx= manager->mctx;
4077 isc_mem_put(mctx, manager, sizeof(*manager));
4078
4079 isc_mem_detach(&mctx);
4080
4081 *managerp = NULL;
4082 }
4083
4084 static isc_result_t
4085 socket_recv(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4086 unsigned int flags)
4087 {
4088 int io_state;
4089 isc_boolean_t have_lock = ISC_FALSE;
4090 isc_task_t *ntask = NULL;
4091 isc_result_t result = ISC_R_SUCCESS;
4092
4093 dev->ev_sender = task;
4094
4095 if (sock->type == isc_sockettype_udp) {
4096 io_state = doio_recv(sock, dev);
4097 } else {
4098 LOCK(&sock->lock);
4099 have_lock = ISC_TRUE;
4100
4101 if (ISC_LIST_EMPTY(sock->recv_list))
4102 io_state = doio_recv(sock, dev);
4103 else
4104 io_state = DOIO_SOFT;
4105 }
4106
4107 switch (io_state) {
4108 case DOIO_SOFT:
4109 /*
4110 * We couldn't read all or part of the request right now, so
4111 * queue it.
4112 *
4113 * Attach to socket and to task
4114 */
4115 isc_task_attach(task, &ntask);
4116 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4117
4118 if (!have_lock) {
4119 LOCK(&sock->lock);
4120 have_lock = ISC_TRUE;
4121 }
4122
4123 /*
4124 * Enqueue the request. If the socket was previously not being
4125 * watched, poke the watcher to start paying attention to it.
4126 */
4127 if (ISC_LIST_EMPTY(sock->recv_list) && !sock->pending_recv)
4128 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
4129 ISC_LIST_ENQUEUE(sock->recv_list, dev, ev_link);
4130
4131 socket_log(sock, NULL, EVENT, NULL, 0, 0,
4132 "socket_recv: event %p -> task %p",
4133 dev, ntask);
4134
4135 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4136 result = ISC_R_INPROGRESS;
4137 break;
4138
4139 case DOIO_EOF:
4140 dev->result = ISC_R_EOF;
4141 /* fallthrough */
4142
4143 case DOIO_HARD:
4144 case DOIO_SUCCESS:
4145 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4146 send_recvdone_event(sock, &dev);
4147 break;
4148 }
4149
4150 if (have_lock)
4151 UNLOCK(&sock->lock);
4152
4153 return (result);
4154 }
4155
4156 isc_result_t
4157 isc_socket_recvv(isc_socket_t *sock, isc_bufferlist_t *buflist,
4158 unsigned int minimum, isc_task_t *task,
4159 isc_taskaction_t action, const void *arg)
4160 {
4161 isc_socketevent_t *dev;
4162 isc_socketmgr_t *manager;
4163 unsigned int iocount;
4164 isc_buffer_t *buffer;
4165
4166 REQUIRE(VALID_SOCKET(sock));
4167 REQUIRE(buflist != NULL);
4168 REQUIRE(!ISC_LIST_EMPTY(*buflist));
4169 REQUIRE(task != NULL);
4170 REQUIRE(action != NULL);
4171
4172 manager = sock->manager;
4173 REQUIRE(VALID_MANAGER(manager));
4174
4175 iocount = isc_bufferlist_availablecount(buflist);
4176 REQUIRE(iocount > 0);
4177
4178 INSIST(sock->bound);
4179
4180 dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4181 if (dev == NULL) {
4182 return (ISC_R_NOMEMORY);
4183 }
4184
4185 /*
4186 * UDP sockets are always partial read
4187 */
4188 if (sock->type == isc_sockettype_udp)
4189 dev->minimum = 1;
4190 else {
4191 if (minimum == 0)
4192 dev->minimum = iocount;
4193 else
4194 dev->minimum = minimum;
4195 }
4196
4197 /*
4198 * Move each buffer from the passed in list to our internal one.
4199 */
4200 buffer = ISC_LIST_HEAD(*buflist);
4201 while (buffer != NULL) {
4202 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4203 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4204 buffer = ISC_LIST_HEAD(*buflist);
4205 }
4206
4207 return (socket_recv(sock, dev, task, 0));
4208 }
4209
4210 isc_result_t
4211 isc_socket_recv(isc_socket_t *sock, isc_region_t *region, unsigned int minimum,
4212 isc_task_t *task, isc_taskaction_t action, const void *arg)
4213 {
4214 isc_socketevent_t *dev;
4215 isc_socketmgr_t *manager;
4216
4217 REQUIRE(VALID_SOCKET(sock));
4218 REQUIRE(action != NULL);
4219
4220 manager = sock->manager;
4221 REQUIRE(VALID_MANAGER(manager));
4222
4223 INSIST(sock->bound);
4224
4225 dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
4226 if (dev == NULL)
4227 return (ISC_R_NOMEMORY);
4228
4229 return (isc_socket_recv2(sock, region, minimum, task, dev, 0));
4230 }
4231
4232 isc_result_t
4233 isc_socket_recv2(isc_socket_t *sock, isc_region_t *region,
4234 unsigned int minimum, isc_task_t *task,
4235 isc_socketevent_t *event, unsigned int flags)
4236 {
4237 event->ev_sender = sock;
4238 event->result = ISC_R_UNEXPECTED;
4239 ISC_LIST_INIT(event->bufferlist);
4240 event->region = *region;
4241 event->n = 0;
4242 event->offset = 0;
4243 event->attributes = 0;
4244
4245 /*
4246 * UDP sockets are always partial read.
4247 */
4248 if (sock->type == isc_sockettype_udp)
4249 event->minimum = 1;
4250 else {
4251 if (minimum == 0)
4252 event->minimum = region->length;
4253 else
4254 event->minimum = minimum;
4255 }
4256
4257 return (socket_recv(sock, event, task, flags));
4258 }
4259
4260 static isc_result_t
4261 socket_send(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
4262 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4263 unsigned int flags)
4264 {
4265 int io_state;
4266 isc_boolean_t have_lock = ISC_FALSE;
4267 isc_task_t *ntask = NULL;
4268 isc_result_t result = ISC_R_SUCCESS;
4269
4270 dev->ev_sender = task;
4271
4272 set_dev_address(address, sock, dev);
4273 if (pktinfo != NULL) {
4274 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
4275 dev->pktinfo = *pktinfo;
4276
4277 if (!isc_sockaddr_issitelocal(&dev->address) &&
4278 !isc_sockaddr_islinklocal(&dev->address)) {
4279 socket_log(sock, NULL, TRACE, isc_msgcat,
4280 ISC_MSGSET_SOCKET, ISC_MSG_PKTINFOPROVIDED,
4281 "pktinfo structure provided, ifindex %u "
4282 "(set to 0)", pktinfo->ipi6_ifindex);
4283
4284 /*
4285 * Set the pktinfo index to 0 here, to let the
4286 * kernel decide what interface it should send on.
4287 */
4288 dev->pktinfo.ipi6_ifindex = 0;
4289 }
4290 }
4291
4292 if (sock->type == isc_sockettype_udp)
4293 io_state = doio_send(sock, dev);
4294 else {
4295 LOCK(&sock->lock);
4296 have_lock = ISC_TRUE;
4297
4298 if (ISC_LIST_EMPTY(sock->send_list))
4299 io_state = doio_send(sock, dev);
4300 else
4301 io_state = DOIO_SOFT;
4302 }
4303
4304 switch (io_state) {
4305 case DOIO_SOFT:
4306 /*
4307 * We couldn't send all or part of the request right now, so
4308 * queue it unless ISC_SOCKFLAG_NORETRY is set.
4309 */
4310 if ((flags & ISC_SOCKFLAG_NORETRY) == 0) {
4311 isc_task_attach(task, &ntask);
4312 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
4313
4314 if (!have_lock) {
4315 LOCK(&sock->lock);
4316 have_lock = ISC_TRUE;
4317 }
4318
4319 /*
4320 * Enqueue the request. If the socket was previously
4321 * not being watched, poke the watcher to start
4322 * paying attention to it.
4323 */
4324 if (ISC_LIST_EMPTY(sock->send_list) &&
4325 !sock->pending_send)
4326 select_poke(sock->manager, sock->fd,
4327 SELECT_POKE_WRITE);
4328 ISC_LIST_ENQUEUE(sock->send_list, dev, ev_link);
4329
4330 socket_log(sock, NULL, EVENT, NULL, 0, 0,
4331 "socket_send: event %p -> task %p",
4332 dev, ntask);
4333
4334 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
4335 result = ISC_R_INPROGRESS;
4336 break;
4337 }
4338
4339 case DOIO_HARD:
4340 case DOIO_SUCCESS:
4341 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
4342 send_senddone_event(sock, &dev);
4343 break;
4344 }
4345
4346 if (have_lock)
4347 UNLOCK(&sock->lock);
4348
4349 return (result);
4350 }
4351
4352 isc_result_t
4353 isc_socket_send(isc_socket_t *sock, isc_region_t *region,
4354 isc_task_t *task, isc_taskaction_t action, const void *arg)
4355 {
4356 /*
4357 * REQUIRE() checking is performed in isc_socket_sendto().
4358 */
4359 return (isc_socket_sendto(sock, region, task, action, arg, NULL,
4360 NULL));
4361 }
4362
4363 isc_result_t
4364 isc_socket_sendto(isc_socket_t *sock, isc_region_t *region,
4365 isc_task_t *task, isc_taskaction_t action, const void *arg,
4366 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4367 {
4368 isc_socketevent_t *dev;
4369 isc_socketmgr_t *manager;
4370
4371 REQUIRE(VALID_SOCKET(sock));
4372 REQUIRE(region != NULL);
4373 REQUIRE(task != NULL);
4374 REQUIRE(action != NULL);
4375
4376 manager = sock->manager;
4377 REQUIRE(VALID_MANAGER(manager));
4378
4379 INSIST(sock->bound);
4380
4381 dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4382 if (dev == NULL) {
4383 return (ISC_R_NOMEMORY);
4384 }
4385
4386 dev->region = *region;
4387
4388 return (socket_send(sock, dev, task, address, pktinfo, 0));
4389 }
4390
4391 isc_result_t
4392 isc_socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
4393 isc_task_t *task, isc_taskaction_t action, const void *arg)
4394 {
4395 return (isc_socket_sendtov(sock, buflist, task, action, arg, NULL,
4396 NULL));
4397 }
4398
4399 isc_result_t
4400 isc_socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
4401 isc_task_t *task, isc_taskaction_t action, const void *arg,
4402 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
4403 {
4404 isc_socketevent_t *dev;
4405 isc_socketmgr_t *manager;
4406 unsigned int iocount;
4407 isc_buffer_t *buffer;
4408
4409 REQUIRE(VALID_SOCKET(sock));
4410 REQUIRE(buflist != NULL);
4411 REQUIRE(!ISC_LIST_EMPTY(*buflist));
4412 REQUIRE(task != NULL);
4413 REQUIRE(action != NULL);
4414
4415 manager = sock->manager;
4416 REQUIRE(VALID_MANAGER(manager));
4417
4418 iocount = isc_bufferlist_usedcount(buflist);
4419 REQUIRE(iocount > 0);
4420
4421 dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
4422 if (dev == NULL) {
4423 return (ISC_R_NOMEMORY);
4424 }
4425
4426 /*
4427 * Move each buffer from the passed in list to our internal one.
4428 */
4429 buffer = ISC_LIST_HEAD(*buflist);
4430 while (buffer != NULL) {
4431 ISC_LIST_DEQUEUE(*buflist, buffer, link);
4432 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
4433 buffer = ISC_LIST_HEAD(*buflist);
4434 }
4435
4436 return (socket_send(sock, dev, task, address, pktinfo, 0));
4437 }
4438
4439 isc_result_t
4440 isc_socket_sendto2(isc_socket_t *sock, isc_region_t *region,
4441 isc_task_t *task,
4442 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
4443 isc_socketevent_t *event, unsigned int flags)
4444 {
4445 REQUIRE((flags & ~(ISC_SOCKFLAG_IMMEDIATE|ISC_SOCKFLAG_NORETRY)) == 0);
4446 if ((flags & ISC_SOCKFLAG_NORETRY) != 0)
4447 REQUIRE(sock->type == isc_sockettype_udp);
4448 event->ev_sender = sock;
4449 event->result = ISC_R_UNEXPECTED;
4450 ISC_LIST_INIT(event->bufferlist);
4451 event->region = *region;
4452 event->n = 0;
4453 event->offset = 0;
4454 event->attributes = 0;
4455
4456 return (socket_send(sock, event, task, address, pktinfo, flags));
4457 }
4458
4459 void
4460 isc_socket_cleanunix(isc_sockaddr_t *sockaddr, isc_boolean_t active) {
4461 #ifdef ISC_PLATFORM_HAVESYSUNH
4462 int s;
4463 struct stat sb;
4464 char strbuf[ISC_STRERRORSIZE];
4465
4466 if (sockaddr->type.sa.sa_family != AF_UNIX)
4467 return;
4468
4469 #ifndef S_ISSOCK
4470 #if defined(S_IFMT) && defined(S_IFSOCK)
4471 #define S_ISSOCK(mode) ((mode & S_IFMT)==S_IFSOCK)
4472 #elif defined(_S_IFMT) && defined(S_IFSOCK)
4473 #define S_ISSOCK(mode) ((mode & _S_IFMT)==S_IFSOCK)
4474 #endif
4475 #endif
4476
4477 #ifndef S_ISFIFO
4478 #if defined(S_IFMT) && defined(S_IFIFO)
4479 #define S_ISFIFO(mode) ((mode & S_IFMT)==S_IFIFO)
4480 #elif defined(_S_IFMT) && defined(S_IFIFO)
4481 #define S_ISFIFO(mode) ((mode & _S_IFMT)==S_IFIFO)
4482 #endif
4483 #endif
4484
4485 #if !defined(S_ISFIFO) && !defined(S_ISSOCK)
4486 #error You need to define S_ISFIFO and S_ISSOCK as appropriate for your platform. See <sys/stat.h>.
4487 #endif
4488
4489 #ifndef S_ISFIFO
4490 #define S_ISFIFO(mode) 0
4491 #endif
4492
4493 #ifndef S_ISSOCK
4494 #define S_ISSOCK(mode) 0
4495 #endif
4496
4497 if (active) {
4498 if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4499 isc__strerror(errno, strbuf, sizeof(strbuf));
4500 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4501 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4502 "isc_socket_cleanunix: stat(%s): %s",
4503 sockaddr->type.sunix.sun_path, strbuf);
4504 return;
4505 }
4506 if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4507 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4508 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4509 "isc_socket_cleanunix: %s: not a socket",
4510 sockaddr->type.sunix.sun_path);
4511 return;
4512 }
4513 if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4514 isc__strerror(errno, strbuf, sizeof(strbuf));
4515 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4516 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4517 "isc_socket_cleanunix: unlink(%s): %s",
4518 sockaddr->type.sunix.sun_path, strbuf);
4519 }
4520 return;
4521 }
4522
4523 s = socket(AF_UNIX, SOCK_STREAM, 0);
4524 if (s < 0) {
4525 isc__strerror(errno, strbuf, sizeof(strbuf));
4526 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4527 ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4528 "isc_socket_cleanunix: socket(%s): %s",
4529 sockaddr->type.sunix.sun_path, strbuf);
4530 return;
4531 }
4532
4533 if (stat(sockaddr->type.sunix.sun_path, &sb) < 0) {
4534 switch (errno) {
4535 case ENOENT: /* We exited cleanly last time */
4536 break;
4537 default:
4538 isc__strerror(errno, strbuf, sizeof(strbuf));
4539 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4540 ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4541 "isc_socket_cleanunix: stat(%s): %s",
4542 sockaddr->type.sunix.sun_path, strbuf);
4543 break;
4544 }
4545 goto cleanup;
4546 }
4547
4548 if (!(S_ISSOCK(sb.st_mode) || S_ISFIFO(sb.st_mode))) {
4549 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4550 ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4551 "isc_socket_cleanunix: %s: not a socket",
4552 sockaddr->type.sunix.sun_path);
4553 goto cleanup;
4554 }
4555
4556 if (connect(s, (struct sockaddr *)&sockaddr->type.sunix,
4557 sizeof(sockaddr->type.sunix)) < 0) {
4558 switch (errno) {
4559 case ECONNREFUSED:
4560 case ECONNRESET:
4561 if (unlink(sockaddr->type.sunix.sun_path) < 0) {
4562 isc__strerror(errno, strbuf, sizeof(strbuf));
4563 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4564 ISC_LOGMODULE_SOCKET,
4565 ISC_LOG_WARNING,
4566 "isc_socket_cleanunix: "
4567 "unlink(%s): %s",
4568 sockaddr->type.sunix.sun_path,
4569 strbuf);
4570 }
4571 break;
4572 default:
4573 isc__strerror(errno, strbuf, sizeof(strbuf));
4574 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4575 ISC_LOGMODULE_SOCKET, ISC_LOG_WARNING,
4576 "isc_socket_cleanunix: connect(%s): %s",
4577 sockaddr->type.sunix.sun_path, strbuf);
4578 break;
4579 }
4580 }
4581 cleanup:
4582 close(s);
4583 #else
4584 UNUSED(sockaddr);
4585 UNUSED(active);
4586 #endif
4587 }
4588
4589 isc_result_t
4590 isc_socket_permunix(isc_sockaddr_t *sockaddr, isc_uint32_t perm,
4591 isc_uint32_t owner, isc_uint32_t group)
4592 {
4593 #ifdef ISC_PLATFORM_HAVESYSUNH
4594 isc_result_t result = ISC_R_SUCCESS;
4595 char strbuf[ISC_STRERRORSIZE];
4596 char path[sizeof(sockaddr->type.sunix.sun_path)];
4597 #ifdef NEED_SECURE_DIRECTORY
4598 char *slash;
4599 #endif
4600
4601 REQUIRE(sockaddr->type.sa.sa_family == AF_UNIX);
4602 INSIST(strlen(sockaddr->type.sunix.sun_path) < sizeof(path));
4603 strcpy(path, sockaddr->type.sunix.sun_path);
4604
4605 #ifdef NEED_SECURE_DIRECTORY
4606 slash = strrchr(path, '/');
4607 if (slash != NULL) {
4608 if (slash != path)
4609 *slash = '\0';
4610 else
4611 strcpy(path, "/");
4612 } else
4613 strcpy(path, ".");
4614 #endif
4615
4616 if (chmod(path, perm) < 0) {
4617 isc__strerror(errno, strbuf, sizeof(strbuf));
4618 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4619 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4620 "isc_socket_permunix: chmod(%s, %d): %s",
4621 path, perm, strbuf);
4622 result = ISC_R_FAILURE;
4623 }
4624 if (chown(path, owner, group) < 0) {
4625 isc__strerror(errno, strbuf, sizeof(strbuf));
4626 isc_log_write(isc_lctx, ISC_LOGCATEGORY_GENERAL,
4627 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
4628 "isc_socket_permunix: chown(%s, %d, %d): %s",
4629 path, owner, group,
4630 strbuf);
4631 result = ISC_R_FAILURE;
4632 }
4633 return (result);
4634 #else
4635 UNUSED(sockaddr);
4636 UNUSED(perm);
4637 UNUSED(owner);
4638 UNUSED(group);
4639 return (ISC_R_NOTIMPLEMENTED);
4640 #endif
4641 }
4642
4643 isc_result_t
4644 isc_socket_bind(isc_socket_t *sock, isc_sockaddr_t *sockaddr,
4645 unsigned int options) {
4646 char strbuf[ISC_STRERRORSIZE];
4647 int on = 1;
4648
4649 LOCK(&sock->lock);
4650
4651 INSIST(!sock->bound);
4652
4653 if (sock->pf != sockaddr->type.sa.sa_family) {
4654 UNLOCK(&sock->lock);
4655 return (ISC_R_FAMILYMISMATCH);
4656 }
4657 /*
4658 * Only set SO_REUSEADDR when we want a specific port.
4659 */
4660 #ifdef AF_UNIX
4661 if (sock->pf == AF_UNIX)
4662 goto bind_socket;
4663 #endif
4664 if ((options & ISC_SOCKET_REUSEADDRESS) != 0 &&
4665 isc_sockaddr_getport(sockaddr) != (in_port_t)0 &&
4666 setsockopt(sock->fd, SOL_SOCKET, SO_REUSEADDR, (void *)&on,
4667 sizeof(on)) < 0) {
4668 UNEXPECTED_ERROR(__FILE__, __LINE__,
4669 "setsockopt(%d) %s", sock->fd,
4670 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
4671 ISC_MSG_FAILED, "failed"));
4672 /* Press on... */
4673 }
4674 #ifdef AF_UNIX
4675 bind_socket:
4676 #endif
4677 if (bind(sock->fd, &sockaddr->type.sa, sockaddr->length) < 0) {
4678 inc_stats(sock->manager->stats,
4679 sock->statsindex[STATID_BINDFAIL]);
4680
4681 UNLOCK(&sock->lock);
4682 switch (errno) {
4683 case EACCES:
4684 return (ISC_R_NOPERM);
4685 case EADDRNOTAVAIL:
4686 return (ISC_R_ADDRNOTAVAIL);
4687 case EADDRINUSE:
4688 return (ISC_R_ADDRINUSE);
4689 case EINVAL:
4690 return (ISC_R_BOUND);
4691 default:
4692 isc__strerror(errno, strbuf, sizeof(strbuf));
4693 UNEXPECTED_ERROR(__FILE__, __LINE__, "bind: %s",
4694 strbuf);
4695 return (ISC_R_UNEXPECTED);
4696 }
4697 }
4698
4699 socket_log(sock, sockaddr, TRACE,
4700 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_BOUND, "bound");
4701 sock->bound = 1;
4702
4703 UNLOCK(&sock->lock);
4704 return (ISC_R_SUCCESS);
4705 }
4706
4707 isc_result_t
4708 isc_socket_filter(isc_socket_t *sock, const char *filter) {
4709 #ifdef SO_ACCEPTFILTER
4710 char strbuf[ISC_STRERRORSIZE];
4711 struct accept_filter_arg afa;
4712 #else
4713 UNUSED(sock);
4714 UNUSED(filter);
4715 #endif
4716
4717 REQUIRE(VALID_SOCKET(sock));
4718
4719 #ifdef SO_ACCEPTFILTER
4720 bzero(&afa, sizeof(afa));
4721 strncpy(afa.af_name, filter, sizeof(afa.af_name));
4722 if (setsockopt(sock->fd, SOL_SOCKET, SO_ACCEPTFILTER,
4723 &afa, sizeof(afa)) == -1) {
4724 isc__strerror(errno, strbuf, sizeof(strbuf));
4725 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
4726 ISC_MSG_FILTER, "setsockopt(SO_ACCEPTFILTER): %s",
4727 strbuf);
4728 return (ISC_R_FAILURE);
4729 }
4730 return (ISC_R_SUCCESS);
4731 #else
4732 return (ISC_R_NOTIMPLEMENTED);
4733 #endif
4734 }
4735
4736 /*
4737 * Set up to listen on a given socket. We do this by creating an internal
4738 * event that will be dispatched when the socket has read activity. The
4739 * watcher will send the internal event to the task when there is a new
4740 * connection.
4741 *
4742 * Unlike in read, we don't preallocate a done event here. Every time there
4743 * is a new connection we'll have to allocate a new one anyway, so we might
4744 * as well keep things simple rather than having to track them.
4745 */
4746 isc_result_t
4747 isc_socket_listen(isc_socket_t *sock, unsigned int backlog) {
4748 char strbuf[ISC_STRERRORSIZE];
4749
4750 REQUIRE(VALID_SOCKET(sock));
4751
4752 LOCK(&sock->lock);
4753
4754 REQUIRE(!sock->listener);
4755 REQUIRE(sock->bound);
4756 REQUIRE(sock->type == isc_sockettype_tcp ||
4757 sock->type == isc_sockettype_unix);
4758
4759 if (backlog == 0)
4760 backlog = SOMAXCONN;
4761
4762 if (listen(sock->fd, (int)backlog) < 0) {
4763 UNLOCK(&sock->lock);
4764 isc__strerror(errno, strbuf, sizeof(strbuf));
4765
4766 UNEXPECTED_ERROR(__FILE__, __LINE__, "listen: %s", strbuf);
4767
4768 return (ISC_R_UNEXPECTED);
4769 }
4770
4771 sock->listener = 1;
4772
4773 UNLOCK(&sock->lock);
4774 return (ISC_R_SUCCESS);
4775 }
4776
4777 /*
4778 * This should try to do aggressive accept() XXXMLG
4779 */
4780 isc_result_t
4781 isc_socket_accept(isc_socket_t *sock,
4782 isc_task_t *task, isc_taskaction_t action, const void *arg)
4783 {
4784 isc_socket_newconnev_t *dev;
4785 isc_socketmgr_t *manager;
4786 isc_task_t *ntask = NULL;
4787 isc_socket_t *nsock;
4788 isc_result_t result;
4789 isc_boolean_t do_poke = ISC_FALSE;
4790
4791 REQUIRE(VALID_SOCKET(sock));
4792 manager = sock->manager;
4793 REQUIRE(VALID_MANAGER(manager));
4794
4795 LOCK(&sock->lock);
4796
4797 REQUIRE(sock->listener);
4798
4799 /*
4800 * Sender field is overloaded here with the task we will be sending
4801 * this event to. Just before the actual event is delivered the
4802 * actual ev_sender will be touched up to be the socket.
4803 */
4804 dev = (isc_socket_newconnev_t *)
4805 isc_event_allocate(manager->mctx, task, ISC_SOCKEVENT_NEWCONN,
4806 action, arg, sizeof(*dev));
4807 if (dev == NULL) {
4808 UNLOCK(&sock->lock);
4809 return (ISC_R_NOMEMORY);
4810 }
4811 ISC_LINK_INIT(dev, ev_link);
4812
4813 result = allocate_socket(manager, sock->type, &nsock);
4814 if (result != ISC_R_SUCCESS) {
4815 isc_event_free(ISC_EVENT_PTR(&dev));
4816 UNLOCK(&sock->lock);
4817 return (result);
4818 }
4819
4820 /*
4821 * Attach to socket and to task.
4822 */
4823 isc_task_attach(task, &ntask);
4824 nsock->references++;
4825 nsock->statsindex = sock->statsindex;
4826
4827 dev->ev_sender = ntask;
4828 dev->newsocket = nsock;
4829
4830 /*
4831 * Poke watcher here. We still have the socket locked, so there
4832 * is no race condition. We will keep the lock for such a short
4833 * bit of time waking it up now or later won't matter all that much.
4834 */
4835 if (ISC_LIST_EMPTY(sock->accept_list))
4836 do_poke = ISC_TRUE;
4837
4838 ISC_LIST_ENQUEUE(sock->accept_list, dev, ev_link);
4839
4840 if (do_poke)
4841 select_poke(manager, sock->fd, SELECT_POKE_ACCEPT);
4842
4843 UNLOCK(&sock->lock);
4844 return (ISC_R_SUCCESS);
4845 }
4846
4847 isc_result_t
4848 isc_socket_connect(isc_socket_t *sock, isc_sockaddr_t *addr,
4849 isc_task_t *task, isc_taskaction_t action, const void *arg)
4850 {
4851 isc_socket_connev_t *dev;
4852 isc_task_t *ntask = NULL;
4853 isc_socketmgr_t *manager;
4854 int cc;
4855 char strbuf[ISC_STRERRORSIZE];
4856
4857 REQUIRE(VALID_SOCKET(sock));
4858 REQUIRE(addr != NULL);
4859 REQUIRE(task != NULL);
4860 REQUIRE(action != NULL);
4861
4862 manager = sock->manager;
4863 REQUIRE(VALID_MANAGER(manager));
4864 REQUIRE(addr != NULL);
4865
4866 if (isc_sockaddr_ismulticast(addr))
4867 return (ISC_R_MULTICAST);
4868
4869 LOCK(&sock->lock);
4870
4871 REQUIRE(!sock->connecting);
4872
4873 dev = (isc_socket_connev_t *)isc_event_allocate(manager->mctx, sock,
4874 ISC_SOCKEVENT_CONNECT,
4875 action, arg,
4876 sizeof(*dev));
4877 if (dev == NULL) {
4878 UNLOCK(&sock->lock);
4879 return (ISC_R_NOMEMORY);
4880 }
4881 ISC_LINK_INIT(dev, ev_link);
4882
4883 /*
4884 * Try to do the connect right away, as there can be only one
4885 * outstanding, and it might happen to complete.
4886 */
4887 sock->peer_address = *addr;
4888 cc = connect(sock->fd, &addr->type.sa, addr->length);
4889 if (cc < 0) {
4890 /*
4891 * HP-UX "fails" to connect a UDP socket and sets errno to
4892 * EINPROGRESS if it's non-blocking. We'd rather regard this as
4893 * a success and let the user detect it if it's really an error
4894 * at the time of sending a packet on the socket.
4895 */
4896 if (sock->type == isc_sockettype_udp && errno == EINPROGRESS) {
4897 cc = 0;
4898 goto success;
4899 }
4900 if (SOFT_ERROR(errno) || errno == EINPROGRESS)
4901 goto queue;
4902
4903 switch (errno) {
4904 #define ERROR_MATCH(a, b) case a: dev->result = b; goto err_exit;
4905 ERROR_MATCH(EACCES, ISC_R_NOPERM);
4906 ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
4907 ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
4908 ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
4909 ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
4910 #ifdef EHOSTDOWN
4911 ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
4912 #endif
4913 ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
4914 ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
4915 ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
4916 ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
4917 ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
4918 #undef ERROR_MATCH
4919 }
4920
4921 sock->connected = 0;
4922
4923 isc__strerror(errno, strbuf, sizeof(strbuf));
4924 UNEXPECTED_ERROR(__FILE__, __LINE__, "%d/%s", errno, strbuf);
4925
4926 UNLOCK(&sock->lock);
4927 inc_stats(sock->manager->stats,
4928 sock->statsindex[STATID_CONNECTFAIL]);
4929 isc_event_free(ISC_EVENT_PTR(&dev));
4930 return (ISC_R_UNEXPECTED);
4931
4932 err_exit:
4933 sock->connected = 0;
4934 isc_task_send(task, ISC_EVENT_PTR(&dev));
4935
4936 UNLOCK(&sock->lock);
4937 inc_stats(sock->manager->stats,
4938 sock->statsindex[STATID_CONNECTFAIL]);
4939 return (ISC_R_SUCCESS);
4940 }
4941
4942 /*
4943 * If connect completed, fire off the done event.
4944 */
4945 success:
4946 if (cc == 0) {
4947 sock->connected = 1;
4948 sock->bound = 1;
4949 dev->result = ISC_R_SUCCESS;
4950 isc_task_send(task, ISC_EVENT_PTR(&dev));
4951
4952 UNLOCK(&sock->lock);
4953
4954 inc_stats(sock->manager->stats,
4955 sock->statsindex[STATID_CONNECT]);
4956
4957 return (ISC_R_SUCCESS);
4958 }
4959
4960 queue:
4961
4962 /*
4963 * Attach to task.
4964 */
4965 isc_task_attach(task, &ntask);
4966
4967 sock->connecting = 1;
4968
4969 dev->ev_sender = ntask;
4970
4971 /*
4972 * Poke watcher here. We still have the socket locked, so there
4973 * is no race condition. We will keep the lock for such a short
4974 * bit of time waking it up now or later won't matter all that much.
4975 */
4976 if (sock->connect_ev == NULL)
4977 select_poke(manager, sock->fd, SELECT_POKE_CONNECT);
4978
4979 sock->connect_ev = dev;
4980
4981 UNLOCK(&sock->lock);
4982 return (ISC_R_SUCCESS);
4983 }
4984
4985 /*
4986 * Called when a socket with a pending connect() finishes.
4987 */
4988 static void
4989 internal_connect(isc_task_t *me, isc_event_t *ev) {
4990 isc_socket_t *sock;
4991 isc_socket_connev_t *dev;
4992 isc_task_t *task;
4993 int cc;
4994 ISC_SOCKADDR_LEN_T optlen;
4995 char strbuf[ISC_STRERRORSIZE];
4996 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
4997
4998 UNUSED(me);
4999 INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
5000
5001 sock = ev->ev_sender;
5002 INSIST(VALID_SOCKET(sock));
5003
5004 LOCK(&sock->lock);
5005
5006 /*
5007 * When the internal event was sent the reference count was bumped
5008 * to keep the socket around for us. Decrement the count here.
5009 */
5010 INSIST(sock->references > 0);
5011 sock->references--;
5012 if (sock->references == 0) {
5013 UNLOCK(&sock->lock);
5014 destroy(&sock);
5015 return;
5016 }
5017
5018 /*
5019 * Has this event been canceled?
5020 */
5021 dev = sock->connect_ev;
5022 if (dev == NULL) {
5023 INSIST(!sock->connecting);
5024 UNLOCK(&sock->lock);
5025 return;
5026 }
5027
5028 INSIST(sock->connecting);
5029 sock->connecting = 0;
5030
5031 /*
5032 * Get any possible error status here.
5033 */
5034 optlen = sizeof(cc);
5035 if (getsockopt(sock->fd, SOL_SOCKET, SO_ERROR,
5036 (void *)&cc, (void *)&optlen) < 0)
5037 cc = errno;
5038 else
5039 errno = cc;
5040
5041 if (errno != 0) {
5042 /*
5043 * If the error is EAGAIN, just re-select on this
5044 * fd and pretend nothing strange happened.
5045 */
5046 if (SOFT_ERROR(errno) || errno == EINPROGRESS) {
5047 sock->connecting = 1;
5048 select_poke(sock->manager, sock->fd,
5049 SELECT_POKE_CONNECT);
5050 UNLOCK(&sock->lock);
5051
5052 return;
5053 }
5054
5055 inc_stats(sock->manager->stats,
5056 sock->statsindex[STATID_CONNECTFAIL]);
5057
5058 /*
5059 * Translate other errors into ISC_R_* flavors.
5060 */
5061 switch (errno) {
5062 #define ERROR_MATCH(a, b) case a: dev->result = b; break;
5063 ERROR_MATCH(EACCES, ISC_R_NOPERM);
5064 ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
5065 ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
5066 ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
5067 ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
5068 #ifdef EHOSTDOWN
5069 ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
5070 #endif
5071 ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
5072 ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
5073 ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
5074 ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
5075 ERROR_MATCH(ETIMEDOUT, ISC_R_TIMEDOUT);
5076 ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
5077 #undef ERROR_MATCH
5078 default:
5079 dev->result = ISC_R_UNEXPECTED;
5080 isc_sockaddr_format(&sock->peer_address, peerbuf,
5081 sizeof(peerbuf));
5082 isc__strerror(errno, strbuf, sizeof(strbuf));
5083 UNEXPECTED_ERROR(__FILE__, __LINE__,
5084 "internal_connect: connect(%s) %s",
5085 peerbuf, strbuf);
5086 }
5087 } else {
5088 inc_stats(sock->manager->stats,
5089 sock->statsindex[STATID_CONNECT]);
5090 dev->result = ISC_R_SUCCESS;
5091 sock->connected = 1;
5092 sock->bound = 1;
5093 }
5094
5095 sock->connect_ev = NULL;
5096
5097 UNLOCK(&sock->lock);
5098
5099 task = dev->ev_sender;
5100 dev->ev_sender = sock;
5101 isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
5102 }
5103
5104 isc_result_t
5105 isc_socket_getpeername(isc_socket_t *sock, isc_sockaddr_t *addressp) {
5106 isc_result_t result;
5107
5108 REQUIRE(VALID_SOCKET(sock));
5109 REQUIRE(addressp != NULL);
5110
5111 LOCK(&sock->lock);
5112
5113 if (sock->connected) {
5114 *addressp = sock->peer_address;
5115 result = ISC_R_SUCCESS;
5116 } else {
5117 result = ISC_R_NOTCONNECTED;
5118 }
5119
5120 UNLOCK(&sock->lock);
5121
5122 return (result);
5123 }
5124
5125 isc_result_t
5126 isc_socket_getsockname(isc_socket_t *sock, isc_sockaddr_t *addressp) {
5127 ISC_SOCKADDR_LEN_T len;
5128 isc_result_t result;
5129 char strbuf[ISC_STRERRORSIZE];
5130
5131 REQUIRE(VALID_SOCKET(sock));
5132 REQUIRE(addressp != NULL);
5133
5134 LOCK(&sock->lock);
5135
5136 if (!sock->bound) {
5137 result = ISC_R_NOTBOUND;
5138 goto out;
5139 }
5140
5141 result = ISC_R_SUCCESS;
5142
5143 len = sizeof(addressp->type);
5144 if (getsockname(sock->fd, &addressp->type.sa, (void *)&len) < 0) {
5145 isc__strerror(errno, strbuf, sizeof(strbuf));
5146 UNEXPECTED_ERROR(__FILE__, __LINE__, "getsockname: %s",
5147 strbuf);
5148 result = ISC_R_UNEXPECTED;
5149 goto out;
5150 }
5151 addressp->length = (unsigned int)len;
5152
5153 out:
5154 UNLOCK(&sock->lock);
5155
5156 return (result);
5157 }
5158
5159 /*
5160 * Run through the list of events on this socket, and cancel the ones
5161 * queued for task "task" of type "how". "how" is a bitmask.
5162 */
5163 void
5164 isc_socket_cancel(isc_socket_t *sock, isc_task_t *task, unsigned int how) {
5165
5166 REQUIRE(VALID_SOCKET(sock));
5167
5168 /*
5169 * Quick exit if there is nothing to do. Don't even bother locking
5170 * in this case.
5171 */
5172 if (how == 0)
5173 return;
5174
5175 LOCK(&sock->lock);
5176
5177 /*
5178 * All of these do the same thing, more or less.
5179 * Each will:
5180 * o If the internal event is marked as "posted" try to
5181 * remove it from the task's queue. If this fails, mark it
5182 * as canceled instead, and let the task clean it up later.
5183 * o For each I/O request for that task of that type, post
5184 * its done event with status of "ISC_R_CANCELED".
5185 * o Reset any state needed.
5186 */
5187 if (((how & ISC_SOCKCANCEL_RECV) == ISC_SOCKCANCEL_RECV)
5188 && !ISC_LIST_EMPTY(sock->recv_list)) {
5189 isc_socketevent_t *dev;
5190 isc_socketevent_t *next;
5191 isc_task_t *current_task;
5192
5193 dev = ISC_LIST_HEAD(sock->recv_list);
5194
5195 while (dev != NULL) {
5196 current_task = dev->ev_sender;
5197 next = ISC_LIST_NEXT(dev, ev_link);
5198
5199 if ((task == NULL) || (task == current_task)) {
5200 dev->result = ISC_R_CANCELED;
5201 send_recvdone_event(sock, &dev);
5202 }
5203 dev = next;
5204 }
5205 }
5206
5207 if (((how & ISC_SOCKCANCEL_SEND) == ISC_SOCKCANCEL_SEND)
5208 && !ISC_LIST_EMPTY(sock->send_list)) {
5209 isc_socketevent_t *dev;
5210 isc_socketevent_t *next;
5211 isc_task_t *current_task;
5212
5213 dev = ISC_LIST_HEAD(sock->send_list);
5214
5215 while (dev != NULL) {
5216 current_task = dev->ev_sender;
5217 next = ISC_LIST_NEXT(dev, ev_link);
5218
5219 if ((task == NULL) || (task == current_task)) {
5220 dev->result = ISC_R_CANCELED;
5221 send_senddone_event(sock, &dev);
5222 }
5223 dev = next;
5224 }
5225 }
5226
5227 if (((how & ISC_SOCKCANCEL_ACCEPT) == ISC_SOCKCANCEL_ACCEPT)
5228 && !ISC_LIST_EMPTY(sock->accept_list)) {
5229 isc_socket_newconnev_t *dev;
5230 isc_socket_newconnev_t *next;
5231 isc_task_t *current_task;
5232
5233 dev = ISC_LIST_HEAD(sock->accept_list);
5234 while (dev != NULL) {
5235 current_task = dev->ev_sender;
5236 next = ISC_LIST_NEXT(dev, ev_link);
5237
5238 if ((task == NULL) || (task == current_task)) {
5239
5240 ISC_LIST_UNLINK(sock->accept_list, dev,
5241 ev_link);
5242
5243 dev->newsocket->references--;
5244 free_socket(&dev->newsocket);
5245
5246 dev->result = ISC_R_CANCELED;
5247 dev->ev_sender = sock;
5248 isc_task_sendanddetach(&current_task,
5249 ISC_EVENT_PTR(&dev));
5250 }
5251
5252 dev = next;
5253 }
5254 }
5255
5256 /*
5257 * Connecting is not a list.
5258 */
5259 if (((how & ISC_SOCKCANCEL_CONNECT) == ISC_SOCKCANCEL_CONNECT)
5260 && sock->connect_ev != NULL) {
5261 isc_socket_connev_t *dev;
5262 isc_task_t *current_task;
5263
5264 INSIST(sock->connecting);
5265 sock->connecting = 0;
5266
5267 dev = sock->connect_ev;
5268 current_task = dev->ev_sender;
5269
5270 if ((task == NULL) || (task == current_task)) {
5271 sock->connect_ev = NULL;
5272
5273 dev->result = ISC_R_CANCELED;
5274 dev->ev_sender = sock;
5275 isc_task_sendanddetach(&current_task,
5276 ISC_EVENT_PTR(&dev));
5277 }
5278 }
5279
5280 UNLOCK(&sock->lock);
5281 }
5282
5283 isc_sockettype_t
5284 isc_socket_gettype(isc_socket_t *sock) {
5285 REQUIRE(VALID_SOCKET(sock));
5286
5287 return (sock->type);
5288 }
5289
5290 isc_boolean_t
5291 isc_socket_isbound(isc_socket_t *sock) {
5292 isc_boolean_t val;
5293
5294 LOCK(&sock->lock);
5295 val = ((sock->bound) ? ISC_TRUE : ISC_FALSE);
5296 UNLOCK(&sock->lock);
5297
5298 return (val);
5299 }
5300
5301 void
5302 isc_socket_ipv6only(isc_socket_t *sock, isc_boolean_t yes) {
5303 #if defined(IPV6_V6ONLY)
5304 int onoff = yes ? 1 : 0;
5305 #else
5306 UNUSED(yes);
5307 UNUSED(sock);
5308 #endif
5309
5310 REQUIRE(VALID_SOCKET(sock));
5311
5312 #ifdef IPV6_V6ONLY
5313 if (sock->pf == AF_INET6) {
5314 if (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_V6ONLY,
5315 (void *)&onoff, sizeof(int)) < 0) {
5316 char strbuf[ISC_STRERRORSIZE];
5317
5318 UNEXPECTED_ERROR(__FILE__, __LINE__,
5319 "setsockopt(%d, IPV6_V6ONLY) "
5320 "%s: %s", sock->fd,
5321 isc_msgcat_get(isc_msgcat,
5322 ISC_MSGSET_GENERAL,
5323 ISC_MSG_FAILED,
5324 "failed"),
5325 strbuf);
5326 }
5327 }
5328 FIX_IPV6_RECVPKTINFO(sock); /* AIX */
5329 #endif
5330 }
5331
5332 #ifndef ISC_PLATFORM_USETHREADS
5333 /* In our assumed scenario, we can simply use a single static object. */
5334 static isc_socketwait_t swait_private;
5335
5336 int
5337 isc__socketmgr_waitevents(struct timeval *tvp, isc_socketwait_t **swaitp) {
5338 int n;
5339 #ifdef USE_KQUEUE
5340 struct timespec ts, *tsp;
5341 #endif
5342 #ifdef USE_EPOLL
5343 int timeout;
5344 #endif
5345 #ifdef USE_DEVPOLL
5346 struct dvpoll dvp;
5347 #endif
5348
5349 REQUIRE(swaitp != NULL && *swaitp == NULL);
5350
5351 if (socketmgr == NULL)
5352 return (0);
5353
5354 #ifdef USE_KQUEUE
5355 if (tvp != NULL) {
5356 ts.tv_sec = tvp->tv_sec;
5357 ts.tv_nsec = tvp->tv_usec * 1000;
5358 tsp = &ts;
5359 } else
5360 tsp = NULL;
5361 swait_private.nevents = kevent(socketmgr->kqueue_fd, NULL, 0,
5362 socketmgr->events, socketmgr->nevents,
5363 tsp);
5364 n = swait_private.nevents;
5365 #elif defined(USE_EPOLL)
5366 if (tvp != NULL)
5367 timeout = tvp->tv_sec * 1000 + (tvp->tv_usec + 999) / 1000;
5368 else
5369 timeout = -1;
5370 swait_private.nevents = epoll_wait(socketmgr->epoll_fd,
5371 socketmgr->events,
5372 socketmgr->nevents, timeout);
5373 n = swait_private.nevents;
5374 #elif defined(USE_DEVPOLL)
5375 dvp.dp_fds = socketmgr->events;
5376 dvp.dp_nfds = socketmgr->nevents;
5377 if (tvp != NULL) {
5378 dvp.dp_timeout = tvp->tv_sec * 1000 +
5379 (tvp->tv_usec + 999) / 1000;
5380 } else
5381 dvp.dp_timeout = -1;
5382 swait_private.nevents = ioctl(socketmgr->devpoll_fd, DP_POLL, &dvp);
5383 n = swait_private.nevents;
5384 #elif defined(USE_SELECT)
5385 memcpy(socketmgr->read_fds_copy, socketmgr->read_fds,
5386 socketmgr->fd_bufsize);
5387 memcpy(socketmgr->write_fds_copy, socketmgr->write_fds,
5388 socketmgr->fd_bufsize);
5389
5390 swait_private.readset = socketmgr->read_fds_copy;
5391 swait_private.writeset = socketmgr->write_fds_copy;
5392 swait_private.maxfd = socketmgr->maxfd + 1;
5393
5394 n = select(swait_private.maxfd, swait_private.readset,
5395 swait_private.writeset, NULL, tvp);
5396 #endif
5397
5398 *swaitp = &swait_private;
5399 return (n);
5400 }
5401
5402 isc_result_t
5403 isc__socketmgr_dispatch(isc_socketwait_t *swait) {
5404 REQUIRE(swait == &swait_private);
5405
5406 if (socketmgr == NULL)
5407 return (ISC_R_NOTFOUND);
5408
5409 #if defined(USE_KQUEUE) || defined(USE_EPOLL) || defined(USE_DEVPOLL)
5410 (void)process_fds(socketmgr, socketmgr->events, swait->nevents);
5411 return (ISC_R_SUCCESS);
5412 #elif defined(USE_SELECT)
5413 process_fds(socketmgr, swait->maxfd, swait->readset, swait->writeset);
5414 return (ISC_R_SUCCESS);
5415 #endif
5416 }
5417 #endif /* ISC_PLATFORM_USETHREADS */
5418
5419 void
5420 isc_socket_setname(isc_socket_t *socket, const char *name, void *tag) {
5421
5422 /*
5423 * Name 'socket'.
5424 */
5425
5426 REQUIRE(VALID_SOCKET(socket));
5427
5428 LOCK(&socket->lock);
5429 memset(socket->name, 0, sizeof(socket->name));
5430 strncpy(socket->name, name, sizeof(socket->name) - 1);
5431 socket->tag = tag;
5432 UNLOCK(&socket->lock);
5433 }
5434
5435 const char *
5436 isc_socket_getname(isc_socket_t *socket) {
5437 return (socket->name);
5438 }
5439
5440 void *
5441 isc_socket_gettag(isc_socket_t *socket) {
5442 return (socket->tag);
5443 }
5444
5445 #ifdef HAVE_LIBXML2
5446
5447 static const char *
5448 _socktype(isc_sockettype_t type)
5449 {
5450 if (type == isc_sockettype_udp)
5451 return ("udp");
5452 else if (type == isc_sockettype_tcp)
5453 return ("tcp");
5454 else if (type == isc_sockettype_unix)
5455 return ("unix");
5456 else if (type == isc_sockettype_fdwatch)
5457 return ("fdwatch");
5458 else
5459 return ("not-initialized");
5460 }
5461
5462 void
5463 isc_socketmgr_renderxml(isc_socketmgr_t *mgr, xmlTextWriterPtr writer)
5464 {
5465 isc_socket_t *sock;
5466 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
5467 isc_sockaddr_t addr;
5468 ISC_SOCKADDR_LEN_T len;
5469
5470 LOCK(&mgr->lock);
5471
5472 #ifndef ISC_PLATFORM_USETHREADS
5473 xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
5474 xmlTextWriterWriteFormatString(writer, "%d", mgr->refs);
5475 xmlTextWriterEndElement(writer);
5476 #endif
5477
5478 xmlTextWriterStartElement(writer, ISC_XMLCHAR "sockets");
5479 sock = ISC_LIST_HEAD(mgr->socklist);
5480 while (sock != NULL) {
5481 LOCK(&sock->lock);
5482 xmlTextWriterStartElement(writer, ISC_XMLCHAR "socket");
5483
5484 xmlTextWriterStartElement(writer, ISC_XMLCHAR "id");
5485 xmlTextWriterWriteFormatString(writer, "%p", sock);
5486 xmlTextWriterEndElement(writer);
5487
5488 if (sock->name[0] != 0) {
5489 xmlTextWriterStartElement(writer, ISC_XMLCHAR "name");
5490 xmlTextWriterWriteFormatString(writer, "%s",
5491 sock->name);
5492 xmlTextWriterEndElement(writer); /* name */
5493 }
5494
5495 xmlTextWriterStartElement(writer, ISC_XMLCHAR "references");
5496 xmlTextWriterWriteFormatString(writer, "%d", sock->references);
5497 xmlTextWriterEndElement(writer);
5498
5499 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "type",
5500 ISC_XMLCHAR _socktype(sock->type));
5501
5502 if (sock->connected) {
5503 isc_sockaddr_format(&sock->peer_address, peerbuf,
5504 sizeof(peerbuf));
5505 xmlTextWriterWriteElement(writer,
5506 ISC_XMLCHAR "peer-address",
5507 ISC_XMLCHAR peerbuf);
5508 }
5509
5510 len = sizeof(addr);
5511 if (getsockname(sock->fd, &addr.type.sa, (void *)&len) == 0) {
5512 isc_sockaddr_format(&addr, peerbuf, sizeof(peerbuf));
5513 xmlTextWriterWriteElement(writer,
5514 ISC_XMLCHAR "local-address",
5515 ISC_XMLCHAR peerbuf);
5516 }
5517
5518 xmlTextWriterStartElement(writer, ISC_XMLCHAR "states");
5519 if (sock->pending_recv)
5520 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5521 ISC_XMLCHAR "pending-receive");
5522 if (sock->pending_send)
5523 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5524 ISC_XMLCHAR "pending-send");
5525 if (sock->pending_accept)
5526 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5527 ISC_XMLCHAR "pending_accept");
5528 if (sock->listener)
5529 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5530 ISC_XMLCHAR "listener");
5531 if (sock->connected)
5532 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5533 ISC_XMLCHAR "connected");
5534 if (sock->connecting)
5535 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5536 ISC_XMLCHAR "connecting");
5537 if (sock->bound)
5538 xmlTextWriterWriteElement(writer, ISC_XMLCHAR "state",
5539 ISC_XMLCHAR "bound");
5540
5541 xmlTextWriterEndElement(writer); /* states */
5542
5543 xmlTextWriterEndElement(writer); /* socket */
5544
5545 UNLOCK(&sock->lock);
5546 sock = ISC_LIST_NEXT(sock, link);
5547 }
5548 xmlTextWriterEndElement(writer); /* sockets */
5549
5550 UNLOCK(&mgr->lock);
5551 }
5552 #endif /* HAVE_LIBXML2 */
NetBSD on the FriendlyARM MINI2440