2 * Copyright (C) 2004-2008 Internet Systems Consortium, Inc. ("ISC")
3 * Copyright (C) 1998-2003 Internet Software Consortium.
5 * Permission to use, copy, modify, and/or distribute this software for any
6 * purpose with or without fee is hereby granted, provided that the above
7 * copyright notice and this permission notice appear in all copies.
9 * THE SOFTWARE IS PROVIDED "AS IS" AND ISC DISCLAIMS ALL WARRANTIES WITH
10 * REGARD TO THIS SOFTWARE INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY
11 * AND FITNESS. IN NO EVENT SHALL ISC BE LIABLE FOR ANY SPECIAL, DIRECT,
12 * INDIRECT, OR CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM
13 * LOSS OF USE, DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE
14 * OR OTHER TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
15 * PERFORMANCE OF THIS SOFTWARE.
18 /* $Id: socket.c,v 1.207.2.19.2.35.4.6 2008/07/29 04:43:57 each Exp $ */
22 #include <sys/param.h>
23 #include <sys/types.h>
24 #include <sys/socket.h>
35 #include <isc/buffer.h>
36 #include <isc/bufferlist.h>
37 #include <isc/condition.h>
38 #include <isc/formatcheck.h>
43 #include <isc/mutex.h>
46 #include <isc/platform.h>
47 #include <isc/print.h>
48 #include <isc/region.h>
49 #include <isc/socket.h>
50 #include <isc/strerror.h>
52 #include <isc/thread.h>
55 #include "errno2result.h"
57 #ifndef ISC_PLATFORM_USETHREADS
59 #endif /* ISC_PLATFORM_USETHREADS */
61 #if defined(SO_BSDCOMPAT) && defined(__linux__)
62 #include <sys/utsname.h>
66 * Max number of open sockets. In the vast majority of cases the default size
67 * of FD_SETSIZE should be fine, and this constant should be increased only
68 * when absolutely necessary and possible, i.e., the server is exhausting all
69 * available file descriptors (up to FD_SETSIZE) and the select() function
70 * and FD_xxx macros support larger values than FD_SETSIZE (which may not
71 * always by true, but we keep using some of them to ensure as much
72 * portability as possible). Note also that overall server performance
73 * may be rather worsened with a larger value of this constant due to
74 * inherent scalability problems of select().
76 * As a special note, this value shouldn't have to be touched if
77 * this is a build for an authoritative only DNS server.
80 #ifndef ISC_SOCKET_FDSETSIZE
81 #define ISC_SOCKET_FDSETSIZE FD_SETSIZE
85 * Mac OS X needs a special definition to support larger values in select()
87 #if ISC_SOCKET_FDSETSIZE > FD_SETSIZE
89 #define _DARWIN_UNLIMITED_SELECT
90 #endif /* __APPLE__ */
94 * Some systems define the socket length argument as an int, some as size_t,
95 * some as socklen_t. This is here so it can be easily changed if needed.
97 #ifndef ISC_SOCKADDR_LEN_T
98 #define ISC_SOCKADDR_LEN_T unsigned int
102 * Define what the possible "soft" errors can be. These are non-fatal returns
103 * of various network related functions, like recv() and so on.
105 * For some reason, BSDI (and perhaps others) will sometimes return <0
106 * from recv() but will have errno==0. This is broken, but we have to
107 * work around it here.
109 #define SOFT_ERROR(e) ((e) == EAGAIN || \
110 (e) == EWOULDBLOCK || \
114 #define DLVL(x) ISC_LOGCATEGORY_GENERAL, ISC_LOGMODULE_SOCKET, ISC_LOG_DEBUG(x)
117 * DLVL(90) -- Function entry/exit and other tracing.
118 * DLVL(70) -- Socket "correctness" -- including returning of events, etc.
119 * DLVL(60) -- Socket data send/receive
120 * DLVL(50) -- Event tracing, including receiving/sending completion events.
121 * DLVL(20) -- Socket creation/destruction.
123 #define TRACE_LEVEL 90
124 #define CORRECTNESS_LEVEL 70
125 #define IOEVENT_LEVEL 60
126 #define EVENT_LEVEL 50
127 #define CREATION_LEVEL 20
129 #define TRACE DLVL(TRACE_LEVEL)
130 #define CORRECTNESS DLVL(CORRECTNESS_LEVEL)
131 #define IOEVENT DLVL(IOEVENT_LEVEL)
132 #define EVENT DLVL(EVENT_LEVEL)
133 #define CREATION DLVL(CREATION_LEVEL)
135 typedef isc_event_t intev_t;
137 #define SOCKET_MAGIC ISC_MAGIC('I', 'O', 'i', 'o')
138 #define VALID_SOCKET(t) ISC_MAGIC_VALID(t, SOCKET_MAGIC)
141 * IPv6 control information. If the socket is an IPv6 socket we want
142 * to collect the destination address and interface so the client can
143 * set them on outgoing packets.
145 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
152 * NetBSD and FreeBSD can timestamp packets. XXXMLG Should we have
153 * a setsockopt() like interface to request timestamps, and if the OS
154 * doesn't do it for us, call gettimeofday() on every UDP receive?
163 * The number of times a send operation is repeated if the result is EINTR.
170 isc_socketmgr_t *manager;
172 isc_sockettype_t type;
174 /* Locked by socket lock. */
175 ISC_LINK(isc_socket_t) link;
176 unsigned int references;
180 ISC_LIST(isc_socketevent_t) send_list;
181 ISC_LIST(isc_socketevent_t) recv_list;
182 ISC_LIST(isc_socket_newconnev_t) accept_list;
183 isc_socket_connev_t *connect_ev;
186 * Internal events. Posted when a descriptor is readable or
187 * writable. These are statically allocated and never freed.
188 * They will be set to non-purgable before use.
193 isc_sockaddr_t address; /* remote address */
195 unsigned int pending_recv : 1,
198 listener : 1, /* listener socket */
200 connecting : 1, /* connect pending */
201 bound : 1; /* bound to local addr */
203 #ifdef ISC_NET_RECVOVERFLOW
204 unsigned char overflow; /* used for MSG_TRUNC fake */
208 ISC_SOCKADDR_LEN_T recvcmsgbuflen;
210 ISC_SOCKADDR_LEN_T sendcmsgbuflen;
213 #define SOCKET_MANAGER_MAGIC ISC_MAGIC('I', 'O', 'm', 'g')
214 #define VALID_MANAGER(m) ISC_MAGIC_VALID(m, SOCKET_MANAGER_MAGIC)
216 struct isc_socketmgr {
223 /* Locked by manager lock. */
224 ISC_LIST(isc_socket_t) socklist;
226 fd_set *read_fds_copy;
228 fd_set *write_fds_copy;
232 int reserved; /* unlocked */
233 #ifdef ISC_PLATFORM_USETHREADS
234 isc_thread_t watcher;
235 isc_condition_t shutdown_ok;
237 #else /* ISC_PLATFORM_USETHREADS */
239 #endif /* ISC_PLATFORM_USETHREADS */
242 #ifndef ISC_PLATFORM_USETHREADS
243 static isc_socketmgr_t *socketmgr = NULL;
244 #endif /* ISC_PLATFORM_USETHREADS */
246 #define CLOSED 0 /* this one must be zero */
248 #define CLOSE_PENDING 2
251 * send() and recv() iovec counts
253 #define MAXSCATTERGATHER_SEND (ISC_SOCKET_MAXSCATTERGATHER)
254 #ifdef ISC_NET_RECVOVERFLOW
255 # define MAXSCATTERGATHER_RECV (ISC_SOCKET_MAXSCATTERGATHER + 1)
257 # define MAXSCATTERGATHER_RECV (ISC_SOCKET_MAXSCATTERGATHER)
260 static void send_recvdone_event(isc_socket_t *, isc_socketevent_t **);
261 static void send_senddone_event(isc_socket_t *, isc_socketevent_t **);
262 static void free_socket(isc_socket_t **);
263 static isc_result_t allocate_socket(isc_socketmgr_t *, isc_sockettype_t,
265 static void destroy(isc_socket_t **);
266 static void internal_accept(isc_task_t *, isc_event_t *);
267 static void internal_connect(isc_task_t *, isc_event_t *);
268 static void internal_recv(isc_task_t *, isc_event_t *);
269 static void internal_send(isc_task_t *, isc_event_t *);
270 static void process_cmsg(isc_socket_t *, struct msghdr *, isc_socketevent_t *);
271 static void build_msghdr_send(isc_socket_t *, isc_socketevent_t *,
272 struct msghdr *, struct iovec *, size_t *);
273 static void build_msghdr_recv(isc_socket_t *, isc_socketevent_t *,
274 struct msghdr *, struct iovec *, size_t *);
275 static void cleanup_fdsets(isc_socketmgr_t *, isc_mem_t *);
276 static isc_result_t create_fdsets(isc_socketmgr_t *, isc_mem_t *);
278 #define SELECT_POKE_SHUTDOWN (-1)
279 #define SELECT_POKE_NOTHING (-2)
280 #define SELECT_POKE_READ (-3)
281 #define SELECT_POKE_ACCEPT (-3) /* Same as _READ */
282 #define SELECT_POKE_WRITE (-4)
283 #define SELECT_POKE_CONNECT (-4) /* Same as _WRITE */
284 #define SELECT_POKE_CLOSE (-5)
286 #define SOCK_DEAD(s) ((s)->references == 0)
289 manager_log(isc_socketmgr_t *sockmgr,
290 isc_logcategory_t *category, isc_logmodule_t *module, int level,
291 const char *fmt, ...) ISC_FORMAT_PRINTF(5, 6);
293 manager_log(isc_socketmgr_t *sockmgr,
294 isc_logcategory_t *category, isc_logmodule_t *module, int level,
295 const char *fmt, ...)
300 if (! isc_log_wouldlog(isc_lctx, level))
304 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
307 isc_log_write(isc_lctx, category, module, level,
308 "sockmgr %p: %s", sockmgr, msgbuf);
312 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
313 isc_logcategory_t *category, isc_logmodule_t *module, int level,
314 isc_msgcat_t *msgcat, int msgset, int message,
315 const char *fmt, ...) ISC_FORMAT_PRINTF(9, 10);
317 socket_log(isc_socket_t *sock, isc_sockaddr_t *address,
318 isc_logcategory_t *category, isc_logmodule_t *module, int level,
319 isc_msgcat_t *msgcat, int msgset, int message,
320 const char *fmt, ...)
323 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
326 if (! isc_log_wouldlog(isc_lctx, level))
330 vsnprintf(msgbuf, sizeof(msgbuf), fmt, ap);
333 if (address == NULL) {
334 isc_log_iwrite(isc_lctx, category, module, level,
335 msgcat, msgset, message,
336 "socket %p: %s", sock, msgbuf);
338 isc_sockaddr_format(address, peerbuf, sizeof(peerbuf));
339 isc_log_iwrite(isc_lctx, category, module, level,
340 msgcat, msgset, message,
341 "socket %p %s: %s", sock, peerbuf, msgbuf);
346 wakeup_socket(isc_socketmgr_t *manager, int fd, int msg) {
350 * This is a wakeup on a socket. If the socket is not in the
351 * process of being closed, start watching it for either reads
355 INSIST(fd >= 0 && fd < manager->fdsize);
357 if (manager->fdstate[fd] == CLOSE_PENDING) {
358 manager->fdstate[fd] = CLOSED;
359 FD_CLR(fd, manager->read_fds);
360 FD_CLR(fd, manager->write_fds);
364 if (manager->fdstate[fd] != MANAGED)
367 sock = manager->fds[fd];
372 if (msg == SELECT_POKE_READ)
373 FD_SET(sock->fd, manager->read_fds);
374 if (msg == SELECT_POKE_WRITE)
375 FD_SET(sock->fd, manager->write_fds);
378 #ifdef ISC_PLATFORM_USETHREADS
380 * Poke the select loop when there is something for us to do.
381 * The write is required (by POSIX) to complete. That is, we
382 * will not get partial writes.
385 select_poke(isc_socketmgr_t *mgr, int fd, int msg) {
388 char strbuf[ISC_STRERRORSIZE];
394 cc = write(mgr->pipe_fds[1], buf, sizeof(buf));
397 * Treat ENOSR as EAGAIN but loop slowly as it is
398 * unlikely to clear fast.
400 if (cc < 0 && errno == ENOSR) {
405 } while (cc < 0 && SOFT_ERROR(errno));
408 isc__strerror(errno, strbuf, sizeof(strbuf));
409 FATAL_ERROR(__FILE__, __LINE__,
410 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
413 "during watcher poke: %s"),
417 INSIST(cc == sizeof(buf));
421 * Read a message on the internal fd.
424 select_readmsg(isc_socketmgr_t *mgr, int *fd, int *msg) {
427 char strbuf[ISC_STRERRORSIZE];
429 cc = read(mgr->pipe_fds[0], buf, sizeof(buf));
431 *msg = SELECT_POKE_NOTHING;
432 *fd = -1; /* Silence compiler. */
433 if (SOFT_ERROR(errno))
436 isc__strerror(errno, strbuf, sizeof(strbuf));
437 FATAL_ERROR(__FILE__, __LINE__,
438 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
441 "during watcher poke: %s"),
446 INSIST(cc == sizeof(buf));
451 #else /* ISC_PLATFORM_USETHREADS */
453 * Update the state of the socketmgr when something changes.
456 select_poke(isc_socketmgr_t *manager, int fd, int msg) {
457 if (msg == SELECT_POKE_SHUTDOWN)
460 wakeup_socket(manager, fd, msg);
463 #endif /* ISC_PLATFORM_USETHREADS */
466 * Make a fd non-blocking.
469 make_nonblock(int fd) {
472 char strbuf[ISC_STRERRORSIZE];
473 #ifdef USE_FIONBIO_IOCTL
476 ret = ioctl(fd, FIONBIO, (char *)&on);
478 flags = fcntl(fd, F_GETFL, 0);
479 flags |= PORT_NONBLOCK;
480 ret = fcntl(fd, F_SETFL, flags);
484 isc__strerror(errno, strbuf, sizeof(strbuf));
485 UNEXPECTED_ERROR(__FILE__, __LINE__,
486 #ifdef USE_FIONBIO_IOCTL
487 "ioctl(%d, FIONBIO, &on): %s", fd,
489 "fcntl(%d, F_SETFL, %d): %s", fd, flags,
493 return (ISC_R_UNEXPECTED);
496 return (ISC_R_SUCCESS);
501 * Not all OSes support advanced CMSG macros: CMSG_LEN and CMSG_SPACE.
502 * In order to ensure as much portability as possible, we provide wrapper
503 * functions of these macros.
504 * Note that cmsg_space() could run slow on OSes that do not have
507 static inline ISC_SOCKADDR_LEN_T
508 cmsg_len(ISC_SOCKADDR_LEN_T len) {
510 return (CMSG_LEN(len));
512 ISC_SOCKADDR_LEN_T hdrlen;
515 * Cast NULL so that any pointer arithmetic performed by CMSG_DATA
518 hdrlen = (ISC_SOCKADDR_LEN_T)CMSG_DATA(((struct cmsghdr *)NULL));
519 return (hdrlen + len);
523 static inline ISC_SOCKADDR_LEN_T
524 cmsg_space(ISC_SOCKADDR_LEN_T len) {
526 return (CMSG_SPACE(len));
529 struct cmsghdr *cmsgp;
531 * XXX: The buffer length is an ad-hoc value, but should be enough
532 * in a practical sense.
534 char dummybuf[sizeof(struct cmsghdr) + 1024];
536 memset(&msg, 0, sizeof(msg));
537 msg.msg_control = dummybuf;
538 msg.msg_controllen = sizeof(dummybuf);
540 cmsgp = (struct cmsghdr *)dummybuf;
541 cmsgp->cmsg_len = cmsg_len(len);
543 cmsgp = CMSG_NXTHDR(&msg, cmsgp);
545 return ((char *)cmsgp - (char *)msg.msg_control);
550 #endif /* USE_CMSG */
553 * Process control messages received on a socket.
556 process_cmsg(isc_socket_t *sock, struct msghdr *msg, isc_socketevent_t *dev) {
558 struct cmsghdr *cmsgp;
559 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
560 struct in6_pktinfo *pktinfop;
563 struct timeval *timevalp;
568 * sock is used only when ISC_NET_BSD44MSGHDR and USE_CMSG are defined.
569 * msg and dev are used only when ISC_NET_BSD44MSGHDR is defined.
570 * They are all here, outside of the CPP tests, because it is
571 * more consistent with the usual ISC coding style.
577 #ifdef ISC_NET_BSD44MSGHDR
580 if ((msg->msg_flags & MSG_TRUNC) == MSG_TRUNC)
581 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
585 if ((msg->msg_flags & MSG_CTRUNC) == MSG_CTRUNC)
586 dev->attributes |= ISC_SOCKEVENTATTR_CTRUNC;
592 if (msg->msg_controllen == 0U || msg->msg_control == NULL)
598 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
602 cmsgp = CMSG_FIRSTHDR(msg);
603 while (cmsgp != NULL) {
604 socket_log(sock, NULL, TRACE,
605 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PROCESSCMSG,
606 "processing cmsg %p", cmsgp);
608 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
609 if (cmsgp->cmsg_level == IPPROTO_IPV6
610 && cmsgp->cmsg_type == IPV6_PKTINFO) {
612 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
613 memcpy(&dev->pktinfo, pktinfop,
614 sizeof(struct in6_pktinfo));
615 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
616 socket_log(sock, NULL, TRACE,
617 isc_msgcat, ISC_MSGSET_SOCKET,
619 "interface received on ifindex %u",
620 dev->pktinfo.ipi6_ifindex);
621 if (IN6_IS_ADDR_MULTICAST(&pktinfop->ipi6_addr))
622 dev->attributes |= ISC_SOCKEVENTATTR_MULTICAST;
628 if (cmsgp->cmsg_level == SOL_SOCKET
629 && cmsgp->cmsg_type == SCM_TIMESTAMP) {
630 timevalp = (struct timeval *)CMSG_DATA(cmsgp);
631 dev->timestamp.seconds = timevalp->tv_sec;
632 dev->timestamp.nanoseconds = timevalp->tv_usec * 1000;
633 dev->attributes |= ISC_SOCKEVENTATTR_TIMESTAMP;
639 cmsgp = CMSG_NXTHDR(msg, cmsgp);
641 #endif /* USE_CMSG */
643 #endif /* ISC_NET_BSD44MSGHDR */
647 * Construct an iov array and attach it to the msghdr passed in. This is
648 * the SEND constructor, which will use the used region of the buffer
649 * (if using a buffer list) or will use the internal region (if a single
650 * buffer I/O is requested).
652 * Nothing can be NULL, and the done event must list at least one buffer
653 * on the buffer linked list for this function to be meaningful.
655 * If write_countp != NULL, *write_countp will hold the number of bytes
656 * this transaction can send.
659 build_msghdr_send(isc_socket_t *sock, isc_socketevent_t *dev,
660 struct msghdr *msg, struct iovec *iov, size_t *write_countp)
662 unsigned int iovcount;
663 isc_buffer_t *buffer;
668 memset(msg, 0, sizeof(*msg));
670 if (sock->type == isc_sockettype_udp) {
671 msg->msg_name = (void *)&dev->address.type.sa;
672 msg->msg_namelen = dev->address.length;
674 msg->msg_name = NULL;
675 msg->msg_namelen = 0;
678 buffer = ISC_LIST_HEAD(dev->bufferlist);
683 * Single buffer I/O? Skip what we've done so far in this region.
685 if (buffer == NULL) {
686 write_count = dev->region.length - dev->n;
687 iov[0].iov_base = (void *)(dev->region.base + dev->n);
688 iov[0].iov_len = write_count;
696 * Skip the data in the buffer list that we have already written.
699 while (buffer != NULL) {
700 REQUIRE(ISC_BUFFER_VALID(buffer));
701 if (skip_count < isc_buffer_usedlength(buffer))
703 skip_count -= isc_buffer_usedlength(buffer);
704 buffer = ISC_LIST_NEXT(buffer, link);
707 while (buffer != NULL) {
708 INSIST(iovcount < MAXSCATTERGATHER_SEND);
710 isc_buffer_usedregion(buffer, &used);
712 if (used.length > 0) {
713 iov[iovcount].iov_base = (void *)(used.base
715 iov[iovcount].iov_len = used.length - skip_count;
716 write_count += (used.length - skip_count);
720 buffer = ISC_LIST_NEXT(buffer, link);
723 INSIST(skip_count == 0U);
727 msg->msg_iovlen = iovcount;
729 #ifdef ISC_NET_BSD44MSGHDR
730 msg->msg_control = NULL;
731 msg->msg_controllen = 0;
733 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
734 if ((sock->type == isc_sockettype_udp)
735 && ((dev->attributes & ISC_SOCKEVENTATTR_PKTINFO) != 0)) {
736 struct cmsghdr *cmsgp;
737 struct in6_pktinfo *pktinfop;
739 socket_log(sock, NULL, TRACE,
740 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_SENDTODATA,
741 "sendto pktinfo data, ifindex %u",
742 dev->pktinfo.ipi6_ifindex);
744 msg->msg_controllen = cmsg_space(sizeof(struct in6_pktinfo));
745 INSIST(msg->msg_controllen <= sock->sendcmsgbuflen);
746 msg->msg_control = (void *)sock->sendcmsgbuf;
748 cmsgp = (struct cmsghdr *)sock->sendcmsgbuf;
749 cmsgp->cmsg_level = IPPROTO_IPV6;
750 cmsgp->cmsg_type = IPV6_PKTINFO;
751 cmsgp->cmsg_len = cmsg_len(sizeof(struct in6_pktinfo));
752 pktinfop = (struct in6_pktinfo *)CMSG_DATA(cmsgp);
753 memcpy(pktinfop, &dev->pktinfo, sizeof(struct in6_pktinfo));
755 #endif /* USE_CMSG && ISC_PLATFORM_HAVEIPV6 */
756 #else /* ISC_NET_BSD44MSGHDR */
757 msg->msg_accrights = NULL;
758 msg->msg_accrightslen = 0;
759 #endif /* ISC_NET_BSD44MSGHDR */
761 if (write_countp != NULL)
762 *write_countp = write_count;
766 * Construct an iov array and attach it to the msghdr passed in. This is
767 * the RECV constructor, which will use the avialable region of the buffer
768 * (if using a buffer list) or will use the internal region (if a single
769 * buffer I/O is requested).
771 * Nothing can be NULL, and the done event must list at least one buffer
772 * on the buffer linked list for this function to be meaningful.
774 * If read_countp != NULL, *read_countp will hold the number of bytes
775 * this transaction can receive.
778 build_msghdr_recv(isc_socket_t *sock, isc_socketevent_t *dev,
779 struct msghdr *msg, struct iovec *iov, size_t *read_countp)
781 unsigned int iovcount;
782 isc_buffer_t *buffer;
783 isc_region_t available;
786 memset(msg, 0, sizeof(struct msghdr));
788 if (sock->type == isc_sockettype_udp) {
789 memset(&dev->address, 0, sizeof(dev->address));
790 #ifdef BROKEN_RECVMSG
791 if (sock->pf == AF_INET) {
792 msg->msg_name = (void *)&dev->address.type.sin;
793 msg->msg_namelen = sizeof(dev->address.type.sin6);
794 } else if (sock->pf == AF_INET6) {
795 msg->msg_name = (void *)&dev->address.type.sin6;
796 msg->msg_namelen = sizeof(dev->address.type.sin6);
797 #ifdef ISC_PLATFORM_HAVESYSUNH
798 } else if (sock->pf == AF_UNIX) {
799 msg->msg_name = (void *)&dev->address.type.sunix;
800 msg->msg_namelen = sizeof(dev->address.type.sunix);
803 msg->msg_name = (void *)&dev->address.type.sa;
804 msg->msg_namelen = sizeof(dev->address.type);
807 msg->msg_name = (void *)&dev->address.type.sa;
808 msg->msg_namelen = sizeof(dev->address.type);
810 #ifdef ISC_NET_RECVOVERFLOW
811 /* If needed, steal one iovec for overflow detection. */
815 msg->msg_name = NULL;
816 msg->msg_namelen = 0;
817 dev->address = sock->address;
820 buffer = ISC_LIST_HEAD(dev->bufferlist);
824 * Single buffer I/O? Skip what we've done so far in this region.
826 if (buffer == NULL) {
827 read_count = dev->region.length - dev->n;
828 iov[0].iov_base = (void *)(dev->region.base + dev->n);
829 iov[0].iov_len = read_count;
837 * Skip empty buffers.
839 while (buffer != NULL) {
840 REQUIRE(ISC_BUFFER_VALID(buffer));
841 if (isc_buffer_availablelength(buffer) != 0)
843 buffer = ISC_LIST_NEXT(buffer, link);
847 while (buffer != NULL) {
848 INSIST(iovcount < MAXSCATTERGATHER_RECV);
850 isc_buffer_availableregion(buffer, &available);
852 if (available.length > 0) {
853 iov[iovcount].iov_base = (void *)(available.base);
854 iov[iovcount].iov_len = available.length;
855 read_count += available.length;
858 buffer = ISC_LIST_NEXT(buffer, link);
864 * If needed, set up to receive that one extra byte. Note that
865 * we know there is at least one iov left, since we stole it
866 * at the top of this function.
868 #ifdef ISC_NET_RECVOVERFLOW
869 if (sock->type == isc_sockettype_udp) {
870 iov[iovcount].iov_base = (void *)(&sock->overflow);
871 iov[iovcount].iov_len = 1;
877 msg->msg_iovlen = iovcount;
879 #ifdef ISC_NET_BSD44MSGHDR
880 msg->msg_control = NULL;
881 msg->msg_controllen = 0;
883 #if defined(USE_CMSG)
884 if (sock->type == isc_sockettype_udp) {
885 msg->msg_control = sock->recvcmsgbuf;
886 msg->msg_controllen = sock->recvcmsgbuflen;
888 #endif /* USE_CMSG */
889 #else /* ISC_NET_BSD44MSGHDR */
890 msg->msg_accrights = NULL;
891 msg->msg_accrightslen = 0;
892 #endif /* ISC_NET_BSD44MSGHDR */
894 if (read_countp != NULL)
895 *read_countp = read_count;
899 set_dev_address(isc_sockaddr_t *address, isc_socket_t *sock,
900 isc_socketevent_t *dev)
902 if (sock->type == isc_sockettype_udp) {
904 dev->address = *address;
906 dev->address = sock->address;
907 } else if (sock->type == isc_sockettype_tcp) {
908 INSIST(address == NULL);
909 dev->address = sock->address;
913 static isc_socketevent_t *
914 allocate_socketevent(isc_socket_t *sock, isc_eventtype_t eventtype,
915 isc_taskaction_t action, const void *arg)
917 isc_socketevent_t *ev;
919 ev = (isc_socketevent_t *)isc_event_allocate(sock->manager->mctx,
927 ev->result = ISC_R_UNEXPECTED;
928 ISC_LINK_INIT(ev, ev_link);
929 ISC_LIST_INIT(ev->bufferlist);
930 ev->region.base = NULL;
938 #if defined(ISC_SOCKET_DEBUG)
940 dump_msg(struct msghdr *msg) {
943 printf("MSGHDR %p\n", msg);
944 printf("\tname %p, namelen %d\n", msg->msg_name, msg->msg_namelen);
945 printf("\tiov %p, iovlen %d\n", msg->msg_iov, msg->msg_iovlen);
946 for (i = 0; i < (unsigned int)msg->msg_iovlen; i++)
947 printf("\t\t%d\tbase %p, len %d\n", i,
948 msg->msg_iov[i].iov_base,
949 msg->msg_iov[i].iov_len);
950 #ifdef ISC_NET_BSD44MSGHDR
951 printf("\tcontrol %p, controllen %d\n", msg->msg_control,
952 msg->msg_controllen);
957 #define DOIO_SUCCESS 0 /* i/o ok, event sent */
958 #define DOIO_SOFT 1 /* i/o ok, soft error, no event sent */
959 #define DOIO_HARD 2 /* i/o error, event sent */
960 #define DOIO_EOF 3 /* EOF, no event sent */
963 doio_recv(isc_socket_t *sock, isc_socketevent_t *dev) {
965 struct iovec iov[MAXSCATTERGATHER_RECV];
968 struct msghdr msghdr;
969 isc_buffer_t *buffer;
971 char strbuf[ISC_STRERRORSIZE];
973 build_msghdr_recv(sock, dev, &msghdr, iov, &read_count);
975 #if defined(ISC_SOCKET_DEBUG)
979 cc = recvmsg(sock->fd, &msghdr, 0);
982 #if defined(ISC_SOCKET_DEBUG)
987 if (SOFT_ERROR(recv_errno))
990 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
991 isc__strerror(recv_errno, strbuf, sizeof(strbuf));
992 socket_log(sock, NULL, IOEVENT,
993 isc_msgcat, ISC_MSGSET_SOCKET,
995 "doio_recv: recvmsg(%d) %d bytes, err %d/%s",
996 sock->fd, cc, recv_errno, strbuf);
999 #define SOFT_OR_HARD(_system, _isc) \
1000 if (recv_errno == _system) { \
1001 if (sock->connected) { \
1002 dev->result = _isc; \
1003 return (DOIO_HARD); \
1005 return (DOIO_SOFT); \
1007 #define ALWAYS_HARD(_system, _isc) \
1008 if (recv_errno == _system) { \
1009 dev->result = _isc; \
1010 return (DOIO_HARD); \
1013 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1014 SOFT_OR_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1015 SOFT_OR_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1016 SOFT_OR_HARD(EHOSTDOWN, ISC_R_HOSTDOWN);
1017 /* HPUX 11.11 can return EADDRNOTAVAIL. */
1018 SOFT_OR_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1019 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1024 dev->result = isc__errno2result(recv_errno);
1029 * On TCP, zero length reads indicate EOF, while on
1030 * UDP, zero length reads are perfectly valid, although
1033 if ((sock->type == isc_sockettype_tcp) && (cc == 0))
1036 if (sock->type == isc_sockettype_udp) {
1037 dev->address.length = msghdr.msg_namelen;
1038 if (isc_sockaddr_getport(&dev->address) == 0) {
1039 if (isc_log_wouldlog(isc_lctx, IOEVENT_LEVEL)) {
1040 socket_log(sock, &dev->address, IOEVENT,
1041 isc_msgcat, ISC_MSGSET_SOCKET,
1043 "dropping source port zero packet");
1049 socket_log(sock, &dev->address, IOEVENT,
1050 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_PKTRECV,
1051 "packet received correctly");
1054 * Overflow bit detection. If we received MORE bytes than we should,
1055 * this indicates an overflow situation. Set the flag in the
1056 * dev entry and adjust how much we read by one.
1058 #ifdef ISC_NET_RECVOVERFLOW
1059 if ((sock->type == isc_sockettype_udp) && ((size_t)cc > read_count)) {
1060 dev->attributes |= ISC_SOCKEVENTATTR_TRUNC;
1066 * If there are control messages attached, run through them and pull
1067 * out the interesting bits.
1069 if (sock->type == isc_sockettype_udp)
1070 process_cmsg(sock, &msghdr, dev);
1073 * update the buffers (if any) and the i/o count
1077 buffer = ISC_LIST_HEAD(dev->bufferlist);
1078 while (buffer != NULL && actual_count > 0U) {
1079 REQUIRE(ISC_BUFFER_VALID(buffer));
1080 if (isc_buffer_availablelength(buffer) <= actual_count) {
1081 actual_count -= isc_buffer_availablelength(buffer);
1082 isc_buffer_add(buffer,
1083 isc_buffer_availablelength(buffer));
1085 isc_buffer_add(buffer, actual_count);
1089 buffer = ISC_LIST_NEXT(buffer, link);
1090 if (buffer == NULL) {
1091 INSIST(actual_count == 0U);
1096 * If we read less than we expected, update counters,
1097 * and let the upper layer poke the descriptor.
1099 if (((size_t)cc != read_count) && (dev->n < dev->minimum))
1103 * Full reads are posted, or partials if partials are ok.
1105 dev->result = ISC_R_SUCCESS;
1106 return (DOIO_SUCCESS);
1111 * DOIO_SUCCESS The operation succeeded. dev->result contains
1114 * DOIO_HARD A hard or unexpected I/O error was encountered.
1115 * dev->result contains the appropriate error.
1117 * DOIO_SOFT A soft I/O error was encountered. No senddone
1118 * event was sent. The operation should be retried.
1120 * No other return values are possible.
1123 doio_send(isc_socket_t *sock, isc_socketevent_t *dev) {
1125 struct iovec iov[MAXSCATTERGATHER_SEND];
1127 struct msghdr msghdr;
1128 char addrbuf[ISC_SOCKADDR_FORMATSIZE];
1131 char strbuf[ISC_STRERRORSIZE];
1133 build_msghdr_send(sock, dev, &msghdr, iov, &write_count);
1136 cc = sendmsg(sock->fd, &msghdr, 0);
1140 * Check for error or block condition.
1143 if (send_errno == EINTR && ++attempts < NRETRIES)
1146 if (SOFT_ERROR(send_errno))
1149 #define SOFT_OR_HARD(_system, _isc) \
1150 if (send_errno == _system) { \
1151 if (sock->connected) { \
1152 dev->result = _isc; \
1153 return (DOIO_HARD); \
1155 return (DOIO_SOFT); \
1157 #define ALWAYS_HARD(_system, _isc) \
1158 if (send_errno == _system) { \
1159 dev->result = _isc; \
1160 return (DOIO_HARD); \
1163 SOFT_OR_HARD(ECONNREFUSED, ISC_R_CONNREFUSED);
1164 ALWAYS_HARD(EACCES, ISC_R_NOPERM);
1165 ALWAYS_HARD(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
1166 ALWAYS_HARD(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
1167 ALWAYS_HARD(EHOSTUNREACH, ISC_R_HOSTUNREACH);
1169 ALWAYS_HARD(EHOSTDOWN, ISC_R_HOSTUNREACH);
1171 ALWAYS_HARD(ENETUNREACH, ISC_R_NETUNREACH);
1172 ALWAYS_HARD(ENOBUFS, ISC_R_NORESOURCES);
1173 ALWAYS_HARD(EPERM, ISC_R_HOSTUNREACH);
1174 ALWAYS_HARD(EPIPE, ISC_R_NOTCONNECTED);
1175 ALWAYS_HARD(ECONNRESET, ISC_R_CONNECTIONRESET);
1181 * The other error types depend on whether or not the
1182 * socket is UDP or TCP. If it is UDP, some errors
1183 * that we expect to be fatal under TCP are merely
1184 * annoying, and are really soft errors.
1186 * However, these soft errors are still returned as
1189 isc_sockaddr_format(&dev->address, addrbuf, sizeof(addrbuf));
1190 isc__strerror(send_errno, strbuf, sizeof(strbuf));
1191 UNEXPECTED_ERROR(__FILE__, __LINE__, "internal_send: %s: %s",
1193 dev->result = isc__errno2result(send_errno);
1198 UNEXPECTED_ERROR(__FILE__, __LINE__,
1199 "internal_send: send() %s 0",
1200 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1201 ISC_MSG_RETURNED, "returned"));
1204 * If we write less than we expected, update counters, poke.
1207 if ((size_t)cc != write_count)
1211 * Exactly what we wanted to write. We're done with this
1212 * entry. Post its completion event.
1214 dev->result = ISC_R_SUCCESS;
1215 return (DOIO_SUCCESS);
1221 * Caller must ensure that the socket is not locked and no external
1225 destroy(isc_socket_t **sockp) {
1226 isc_socket_t *sock = *sockp;
1227 isc_socketmgr_t *manager = sock->manager;
1229 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1230 ISC_MSG_DESTROYING, "destroying");
1232 INSIST(ISC_LIST_EMPTY(sock->accept_list));
1233 INSIST(ISC_LIST_EMPTY(sock->recv_list));
1234 INSIST(ISC_LIST_EMPTY(sock->send_list));
1235 INSIST(sock->connect_ev == NULL);
1236 REQUIRE(sock->fd >= 0 && sock->fd < (int)manager->fdsize);
1238 LOCK(&manager->lock);
1241 * No one has this socket open, so the watcher doesn't have to be
1242 * poked, and the socket doesn't have to be locked.
1244 manager->fds[sock->fd] = NULL;
1245 manager->fdstate[sock->fd] = CLOSE_PENDING;
1246 select_poke(manager, sock->fd, SELECT_POKE_CLOSE);
1247 ISC_LIST_UNLINK(manager->socklist, sock, link);
1249 #ifdef ISC_PLATFORM_USETHREADS
1250 if (ISC_LIST_EMPTY(manager->socklist))
1251 SIGNAL(&manager->shutdown_ok);
1252 #endif /* ISC_PLATFORM_USETHREADS */
1255 * XXX should reset manager->maxfd here
1258 UNLOCK(&manager->lock);
1264 allocate_socket(isc_socketmgr_t *manager, isc_sockettype_t type,
1265 isc_socket_t **socketp)
1269 ISC_SOCKADDR_LEN_T cmsgbuflen;
1271 sock = isc_mem_get(manager->mctx, sizeof(*sock));
1274 return (ISC_R_NOMEMORY);
1276 ret = ISC_R_UNEXPECTED;
1279 sock->references = 0;
1281 sock->manager = manager;
1285 ISC_LINK_INIT(sock, link);
1287 sock->recvcmsgbuf = NULL;
1288 sock->sendcmsgbuf = NULL;
1291 * set up cmsg buffers
1294 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1295 cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1297 #if defined(USE_CMSG) && defined(SO_TIMESTAMP)
1298 cmsgbuflen += cmsg_space(sizeof(struct timeval));
1300 sock->recvcmsgbuflen = cmsgbuflen;
1301 if (sock->recvcmsgbuflen != 0U) {
1302 sock->recvcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1303 if (sock->recvcmsgbuf == NULL)
1308 #if defined(USE_CMSG) && defined(ISC_PLATFORM_HAVEIN6PKTINFO)
1309 cmsgbuflen = cmsg_space(sizeof(struct in6_pktinfo));
1311 sock->sendcmsgbuflen = cmsgbuflen;
1312 if (sock->sendcmsgbuflen != 0U) {
1313 sock->sendcmsgbuf = isc_mem_get(manager->mctx, cmsgbuflen);
1314 if (sock->sendcmsgbuf == NULL)
1319 * set up list of readers and writers to be initially empty
1321 ISC_LIST_INIT(sock->recv_list);
1322 ISC_LIST_INIT(sock->send_list);
1323 ISC_LIST_INIT(sock->accept_list);
1324 sock->connect_ev = NULL;
1325 sock->pending_recv = 0;
1326 sock->pending_send = 0;
1327 sock->pending_accept = 0;
1329 sock->connected = 0;
1330 sock->connecting = 0;
1334 * initialize the lock
1336 if (isc_mutex_init(&sock->lock) != ISC_R_SUCCESS) {
1338 UNEXPECTED_ERROR(__FILE__, __LINE__,
1339 "isc_mutex_init() %s",
1340 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1341 ISC_MSG_FAILED, "failed"));
1342 ret = ISC_R_UNEXPECTED;
1347 * Initialize readable and writable events
1349 ISC_EVENT_INIT(&sock->readable_ev, sizeof(intev_t),
1350 ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTR,
1351 NULL, sock, sock, NULL, NULL);
1352 ISC_EVENT_INIT(&sock->writable_ev, sizeof(intev_t),
1353 ISC_EVENTATTR_NOPURGE, NULL, ISC_SOCKEVENT_INTW,
1354 NULL, sock, sock, NULL, NULL);
1356 sock->magic = SOCKET_MAGIC;
1359 return (ISC_R_SUCCESS);
1362 if (sock->recvcmsgbuf != NULL)
1363 isc_mem_put(manager->mctx, sock->recvcmsgbuf,
1364 sock->recvcmsgbuflen);
1365 if (sock->sendcmsgbuf != NULL)
1366 isc_mem_put(manager->mctx, sock->sendcmsgbuf,
1367 sock->sendcmsgbuflen);
1368 isc_mem_put(manager->mctx, sock, sizeof(*sock));
1374 * This event requires that the various lists be empty, that the reference
1375 * count be 1, and that the magic number is valid. The other socket bits,
1376 * like the lock, must be initialized as well. The fd associated must be
1377 * marked as closed, by setting it to -1 on close, or this routine will
1378 * also close the socket.
1381 free_socket(isc_socket_t **socketp) {
1382 isc_socket_t *sock = *socketp;
1384 INSIST(sock->references == 0);
1385 INSIST(VALID_SOCKET(sock));
1386 INSIST(!sock->connecting);
1387 INSIST(!sock->pending_recv);
1388 INSIST(!sock->pending_send);
1389 INSIST(!sock->pending_accept);
1390 INSIST(ISC_LIST_EMPTY(sock->recv_list));
1391 INSIST(ISC_LIST_EMPTY(sock->send_list));
1392 INSIST(ISC_LIST_EMPTY(sock->accept_list));
1393 INSIST(!ISC_LINK_LINKED(sock, link));
1395 if (sock->recvcmsgbuf != NULL)
1396 isc_mem_put(sock->manager->mctx, sock->recvcmsgbuf,
1397 sock->recvcmsgbuflen);
1398 if (sock->sendcmsgbuf != NULL)
1399 isc_mem_put(sock->manager->mctx, sock->sendcmsgbuf,
1400 sock->sendcmsgbuflen);
1404 DESTROYLOCK(&sock->lock);
1406 isc_mem_put(sock->manager->mctx, sock, sizeof(*sock));
1413 * This really should not be necessary to do. Having to workout
1414 * which kernel version we are on at run time so that we don't cause
1415 * the kernel to issue a warning about us using a deprecated socket option.
1416 * Such warnings should *never* be on by default in production kernels.
1418 * We can't do this a build time because executables are moved between
1419 * machines and hence kernels.
1421 * We can't just not set SO_BSDCOMAT because some kernels require it.
1424 static isc_once_t bsdcompat_once = ISC_ONCE_INIT;
1425 isc_boolean_t bsdcompat = ISC_TRUE;
1428 clear_bsdcompat(void) {
1435 uname(&buf); /* Can only fail if buf is bad in Linux. */
1437 /* Paranoia in parsing can be increased, but we trust uname(). */
1438 major = strtol(buf.release, &endp, 10);
1440 minor = strtol(endp+1, &endp, 10);
1441 if ((major > 2) || ((major == 2) && (minor >= 4))) {
1442 bsdcompat = ISC_FALSE;
1445 #endif /* __linux __ */
1450 * Create a new 'type' socket managed by 'manager'. Events
1451 * will be posted to 'task' and when dispatched 'action' will be
1452 * called with 'arg' as the arg value. The new socket is returned
1456 isc_socket_create(isc_socketmgr_t *manager, int pf, isc_sockettype_t type,
1457 isc_socket_t **socketp)
1459 isc_socket_t *sock = NULL;
1461 #if defined(USE_CMSG) || defined(SO_BSDCOMPAT)
1464 char strbuf[ISC_STRERRORSIZE];
1465 const char *err = "socket";
1468 REQUIRE(VALID_MANAGER(manager));
1469 REQUIRE(socketp != NULL && *socketp == NULL);
1471 ret = allocate_socket(manager, type, &sock);
1472 if (ret != ISC_R_SUCCESS)
1478 case isc_sockettype_udp:
1479 sock->fd = socket(pf, SOCK_DGRAM, IPPROTO_UDP);
1481 case isc_sockettype_tcp:
1482 sock->fd = socket(pf, SOCK_STREAM, IPPROTO_TCP);
1485 if (sock->fd == -1 && errno == EINTR && tries++ < 42)
1490 * Leave a space for stdio and TCP to work in.
1492 if (manager->reserved != 0 && type == isc_sockettype_udp &&
1493 sock->fd >= 0 && sock->fd < manager->reserved) {
1495 new = fcntl(sock->fd, F_DUPFD, manager->reserved);
1497 (void)close(sock->fd);
1500 err = "isc_socket_create: fcntl/reserved";
1501 } else if (sock->fd >= 0 && sock->fd < 20) {
1503 new = fcntl(sock->fd, F_DUPFD, 20);
1505 (void)close(sock->fd);
1508 err = "isc_socket_create: fcntl";
1512 if (sock->fd >= (int)manager->fdsize) {
1513 (void)close(sock->fd);
1514 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1515 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1516 isc_msgcat, ISC_MSGSET_SOCKET,
1518 "%s: too many open file descriptors", "socket");
1520 return (ISC_R_NORESOURCES);
1530 return (ISC_R_NORESOURCES);
1532 case EPROTONOSUPPORT:
1536 * Linux 2.2 (and maybe others) return EINVAL instead of
1540 return (ISC_R_FAMILYNOSUPPORT);
1543 isc__strerror(errno, strbuf, sizeof(strbuf));
1544 UNEXPECTED_ERROR(__FILE__, __LINE__,
1546 isc_msgcat_get(isc_msgcat,
1551 return (ISC_R_UNEXPECTED);
1555 if (make_nonblock(sock->fd) != ISC_R_SUCCESS) {
1556 (void)close(sock->fd);
1558 return (ISC_R_UNEXPECTED);
1562 RUNTIME_CHECK(isc_once_do(&bsdcompat_once,
1563 clear_bsdcompat) == ISC_R_SUCCESS);
1564 if (bsdcompat && setsockopt(sock->fd, SOL_SOCKET, SO_BSDCOMPAT,
1565 (void *)&on, sizeof(on)) < 0) {
1566 isc__strerror(errno, strbuf, sizeof(strbuf));
1567 UNEXPECTED_ERROR(__FILE__, __LINE__,
1568 "setsockopt(%d, SO_BSDCOMPAT) %s: %s",
1570 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1571 ISC_MSG_FAILED, "failed"),
1578 if (setsockopt(sock->fd, SOL_SOCKET, SO_NOSIGPIPE,
1579 (void *)&on, sizeof(on)) < 0) {
1580 isc__strerror(errno, strbuf, sizeof(strbuf));
1581 UNEXPECTED_ERROR(__FILE__, __LINE__,
1582 "setsockopt(%d, SO_NOSIGPIPE) %s: %s",
1584 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
1585 ISC_MSG_FAILED, "failed"),
1591 #if defined(USE_CMSG)
1592 if (type == isc_sockettype_udp) {
1594 #if defined(SO_TIMESTAMP)
1595 if (setsockopt(sock->fd, SOL_SOCKET, SO_TIMESTAMP,
1596 (void *)&on, sizeof(on)) < 0
1597 && errno != ENOPROTOOPT) {
1598 isc__strerror(errno, strbuf, sizeof(strbuf));
1599 UNEXPECTED_ERROR(__FILE__, __LINE__,
1600 "setsockopt(%d, SO_TIMESTAMP) %s: %s",
1602 isc_msgcat_get(isc_msgcat,
1609 #endif /* SO_TIMESTAMP */
1611 #if defined(ISC_PLATFORM_HAVEIPV6)
1612 if (pf == AF_INET6 && sock->recvcmsgbuflen == 0U) {
1614 * Warn explicitly because this anomaly can be hidden
1615 * in usual operation (and unexpectedly appear later).
1617 UNEXPECTED_ERROR(__FILE__, __LINE__,
1618 "No buffer available to receive "
1619 "IPv6 destination");
1621 #ifdef ISC_PLATFORM_HAVEIN6PKTINFO
1622 #ifdef IPV6_RECVPKTINFO
1624 if ((pf == AF_INET6)
1625 && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_RECVPKTINFO,
1626 (void *)&on, sizeof(on)) < 0)) {
1627 isc__strerror(errno, strbuf, sizeof(strbuf));
1628 UNEXPECTED_ERROR(__FILE__, __LINE__,
1629 "setsockopt(%d, IPV6_RECVPKTINFO) "
1631 isc_msgcat_get(isc_msgcat,
1639 if ((pf == AF_INET6)
1640 && (setsockopt(sock->fd, IPPROTO_IPV6, IPV6_PKTINFO,
1641 (void *)&on, sizeof(on)) < 0)) {
1642 isc__strerror(errno, strbuf, sizeof(strbuf));
1643 UNEXPECTED_ERROR(__FILE__, __LINE__,
1644 "setsockopt(%d, IPV6_PKTINFO) %s: %s",
1646 isc_msgcat_get(isc_msgcat,
1652 #endif /* IPV6_RECVPKTINFO */
1653 #endif /* ISC_PLATFORM_HAVEIN6PKTINFO */
1654 #ifdef IPV6_USE_MIN_MTU /* RFC 3542, not too common yet*/
1655 /* use minimum MTU */
1656 if (pf == AF_INET6) {
1657 (void)setsockopt(sock->fd, IPPROTO_IPV6,
1659 (void *)&on, sizeof(on));
1662 #endif /* ISC_PLATFORM_HAVEIPV6 */
1665 #endif /* USE_CMSG */
1667 sock->references = 1;
1670 LOCK(&manager->lock);
1673 * Note we don't have to lock the socket like we normally would because
1674 * there are no external references to it yet.
1677 manager->fds[sock->fd] = sock;
1678 manager->fdstate[sock->fd] = MANAGED;
1679 ISC_LIST_APPEND(manager->socklist, sock, link);
1680 if (manager->maxfd < sock->fd)
1681 manager->maxfd = sock->fd;
1683 UNLOCK(&manager->lock);
1685 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
1686 ISC_MSG_CREATED, "created");
1688 return (ISC_R_SUCCESS);
1692 * Attach to a socket. Caller must explicitly detach when it is done.
1695 isc_socket_attach(isc_socket_t *sock, isc_socket_t **socketp) {
1696 REQUIRE(VALID_SOCKET(sock));
1697 REQUIRE(socketp != NULL && *socketp == NULL);
1701 UNLOCK(&sock->lock);
1707 * Dereference a socket. If this is the last reference to it, clean things
1708 * up by destroying the socket.
1711 isc_socket_detach(isc_socket_t **socketp) {
1713 isc_boolean_t kill_socket = ISC_FALSE;
1715 REQUIRE(socketp != NULL);
1717 REQUIRE(VALID_SOCKET(sock));
1720 REQUIRE(sock->references > 0);
1722 if (sock->references == 0)
1723 kill_socket = ISC_TRUE;
1724 UNLOCK(&sock->lock);
1733 * I/O is possible on a given socket. Schedule an event to this task that
1734 * will call an internal function to do the I/O. This will charge the
1735 * task with the I/O operation and let our select loop handler get back
1736 * to doing something real as fast as possible.
1738 * The socket and manager must be locked before calling this function.
1741 dispatch_recv(isc_socket_t *sock) {
1743 isc_socketevent_t *ev;
1745 INSIST(!sock->pending_recv);
1747 ev = ISC_LIST_HEAD(sock->recv_list);
1751 sock->pending_recv = 1;
1752 iev = &sock->readable_ev;
1754 socket_log(sock, NULL, EVENT, NULL, 0, 0,
1755 "dispatch_recv: event %p -> task %p", ev, ev->ev_sender);
1758 iev->ev_sender = sock;
1759 iev->ev_action = internal_recv;
1762 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1766 dispatch_send(isc_socket_t *sock) {
1768 isc_socketevent_t *ev;
1770 INSIST(!sock->pending_send);
1772 ev = ISC_LIST_HEAD(sock->send_list);
1776 sock->pending_send = 1;
1777 iev = &sock->writable_ev;
1779 socket_log(sock, NULL, EVENT, NULL, 0, 0,
1780 "dispatch_send: event %p -> task %p", ev, ev->ev_sender);
1783 iev->ev_sender = sock;
1784 iev->ev_action = internal_send;
1787 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1791 * Dispatch an internal accept event.
1794 dispatch_accept(isc_socket_t *sock) {
1796 isc_socket_newconnev_t *ev;
1798 INSIST(!sock->pending_accept);
1801 * Are there any done events left, or were they all canceled
1802 * before the manager got the socket lock?
1804 ev = ISC_LIST_HEAD(sock->accept_list);
1808 sock->pending_accept = 1;
1809 iev = &sock->readable_ev;
1811 sock->references++; /* keep socket around for this internal event */
1812 iev->ev_sender = sock;
1813 iev->ev_action = internal_accept;
1816 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1820 dispatch_connect(isc_socket_t *sock) {
1822 isc_socket_connev_t *ev;
1824 iev = &sock->writable_ev;
1826 ev = sock->connect_ev;
1827 INSIST(ev != NULL); /* XXX */
1829 INSIST(sock->connecting);
1831 sock->references++; /* keep socket around for this internal event */
1832 iev->ev_sender = sock;
1833 iev->ev_action = internal_connect;
1836 isc_task_send(ev->ev_sender, (isc_event_t **)&iev);
1840 * Dequeue an item off the given socket's read queue, set the result code
1841 * in the done event to the one provided, and send it to the task it was
1844 * If the event to be sent is on a list, remove it before sending. If
1845 * asked to, send and detach from the socket as well.
1847 * Caller must have the socket locked if the event is attached to the socket.
1850 send_recvdone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
1853 task = (*dev)->ev_sender;
1855 (*dev)->ev_sender = sock;
1857 if (ISC_LINK_LINKED(*dev, ev_link))
1858 ISC_LIST_DEQUEUE(sock->recv_list, *dev, ev_link);
1860 if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
1861 == ISC_SOCKEVENTATTR_ATTACHED)
1862 isc_task_sendanddetach(&task, (isc_event_t **)dev);
1864 isc_task_send(task, (isc_event_t **)dev);
1868 * See comments for send_recvdone_event() above.
1870 * Caller must have the socket locked if the event is attached to the socket.
1873 send_senddone_event(isc_socket_t *sock, isc_socketevent_t **dev) {
1876 INSIST(dev != NULL && *dev != NULL);
1878 task = (*dev)->ev_sender;
1879 (*dev)->ev_sender = sock;
1881 if (ISC_LINK_LINKED(*dev, ev_link))
1882 ISC_LIST_DEQUEUE(sock->send_list, *dev, ev_link);
1884 if (((*dev)->attributes & ISC_SOCKEVENTATTR_ATTACHED)
1885 == ISC_SOCKEVENTATTR_ATTACHED)
1886 isc_task_sendanddetach(&task, (isc_event_t **)dev);
1888 isc_task_send(task, (isc_event_t **)dev);
1892 * Call accept() on a socket, to get the new file descriptor. The listen
1893 * socket is used as a prototype to create a new isc_socket_t. The new
1894 * socket has one outstanding reference. The task receiving the event
1895 * will be detached from just after the event is delivered.
1897 * On entry to this function, the event delivered is the internal
1898 * readable event, and the first item on the accept_list should be
1899 * the done event we want to send. If the list is empty, this is a no-op,
1900 * so just unlock and return.
1903 internal_accept(isc_task_t *me, isc_event_t *ev) {
1905 isc_socketmgr_t *manager;
1906 isc_socket_newconnev_t *dev;
1908 ISC_SOCKADDR_LEN_T addrlen;
1910 isc_result_t result = ISC_R_SUCCESS;
1911 char strbuf[ISC_STRERRORSIZE];
1912 const char *err = "accept";
1916 sock = ev->ev_sender;
1917 INSIST(VALID_SOCKET(sock));
1920 socket_log(sock, NULL, TRACE,
1921 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTLOCK,
1922 "internal_accept called, locked socket");
1924 manager = sock->manager;
1925 INSIST(VALID_MANAGER(manager));
1927 INSIST(sock->listener);
1928 INSIST(sock->pending_accept == 1);
1929 sock->pending_accept = 0;
1931 INSIST(sock->references > 0);
1932 sock->references--; /* the internal event is done with this socket */
1933 if (sock->references == 0) {
1934 UNLOCK(&sock->lock);
1940 * Get the first item off the accept list.
1941 * If it is empty, unlock the socket and return.
1943 dev = ISC_LIST_HEAD(sock->accept_list);
1945 UNLOCK(&sock->lock);
1950 * Try to accept the new connection. If the accept fails with
1951 * EAGAIN or EINTR, simply poke the watcher to watch this socket
1952 * again. Also ignore ECONNRESET, which has been reported to
1953 * be spuriously returned on Linux 2.2.19 although it is not
1954 * a documented error for accept(). ECONNABORTED has been
1955 * reported for Solaris 8. The rest are thrown in not because
1956 * we have seen them but because they are ignored by other
1957 * deamons such as BIND 8 and Apache.
1960 addrlen = sizeof(dev->newsocket->address.type);
1961 memset(&dev->newsocket->address.type, 0, addrlen);
1962 fd = accept(sock->fd, &dev->newsocket->address.type.sa,
1967 * Leave a space for stdio to work in.
1969 if (fd >= 0 && fd < 20) {
1971 new = fcntl(fd, F_DUPFD, 20);
1976 err = "accept/fcntl";
1981 if (SOFT_ERROR(errno))
1986 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
1987 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
1988 isc_msgcat, ISC_MSGSET_SOCKET,
1990 "%s: too many open file descriptors",
2013 isc__strerror(errno, strbuf, sizeof(strbuf));
2014 UNEXPECTED_ERROR(__FILE__, __LINE__,
2015 "internal_accept: %s() %s: %s", err,
2016 isc_msgcat_get(isc_msgcat,
2022 result = ISC_R_UNEXPECTED;
2024 if (addrlen == 0U) {
2025 UNEXPECTED_ERROR(__FILE__, __LINE__,
2026 "internal_accept(): "
2027 "accept() failed to return "
2032 } else if (dev->newsocket->address.type.sa.sa_family !=
2035 UNEXPECTED_ERROR(__FILE__, __LINE__,
2036 "internal_accept(): "
2037 "accept() returned peer address "
2038 "family %u (expected %u)",
2039 dev->newsocket->address.
2044 } else if (fd >= (int)manager->fdsize) {
2045 isc_log_iwrite(isc_lctx, ISC_LOGCATEGORY_GENERAL,
2046 ISC_LOGMODULE_SOCKET, ISC_LOG_ERROR,
2047 isc_msgcat, ISC_MSGSET_SOCKET,
2049 "%s: too many open file descriptors",
2057 dev->newsocket->address.length = addrlen;
2058 dev->newsocket->pf = sock->pf;
2062 * Pull off the done event.
2064 ISC_LIST_UNLINK(sock->accept_list, dev, ev_link);
2067 * Poke watcher if there are more pending accepts.
2069 if (!ISC_LIST_EMPTY(sock->accept_list))
2070 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
2072 UNLOCK(&sock->lock);
2074 if (fd != -1 && (make_nonblock(fd) != ISC_R_SUCCESS)) {
2077 result = ISC_R_UNEXPECTED;
2081 * -1 means the new socket didn't happen.
2084 LOCK(&manager->lock);
2085 ISC_LIST_APPEND(manager->socklist, dev->newsocket, link);
2087 dev->newsocket->fd = fd;
2088 dev->newsocket->bound = 1;
2089 dev->newsocket->connected = 1;
2092 * Save away the remote address
2094 dev->address = dev->newsocket->address;
2096 manager->fds[fd] = dev->newsocket;
2097 manager->fdstate[fd] = MANAGED;
2098 if (manager->maxfd < fd)
2099 manager->maxfd = fd;
2101 socket_log(sock, &dev->newsocket->address, CREATION,
2102 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_ACCEPTEDCXN,
2103 "accepted connection, new socket %p",
2106 UNLOCK(&manager->lock);
2108 dev->newsocket->references--;
2109 free_socket(&dev->newsocket);
2113 * Fill in the done event details and send it off.
2115 dev->result = result;
2116 task = dev->ev_sender;
2117 dev->ev_sender = sock;
2119 isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
2123 select_poke(sock->manager, sock->fd, SELECT_POKE_ACCEPT);
2124 UNLOCK(&sock->lock);
2129 internal_recv(isc_task_t *me, isc_event_t *ev) {
2130 isc_socketevent_t *dev;
2133 INSIST(ev->ev_type == ISC_SOCKEVENT_INTR);
2135 sock = ev->ev_sender;
2136 INSIST(VALID_SOCKET(sock));
2139 socket_log(sock, NULL, IOEVENT,
2140 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALRECV,
2141 "internal_recv: task %p got event %p", me, ev);
2143 INSIST(sock->pending_recv == 1);
2144 sock->pending_recv = 0;
2146 INSIST(sock->references > 0);
2147 sock->references--; /* the internal event is done with this socket */
2148 if (sock->references == 0) {
2149 UNLOCK(&sock->lock);
2155 * Try to do as much I/O as possible on this socket. There are no
2156 * limits here, currently.
2158 dev = ISC_LIST_HEAD(sock->recv_list);
2159 while (dev != NULL) {
2160 switch (doio_recv(sock, dev)) {
2166 * read of 0 means the remote end was closed.
2167 * Run through the event queue and dispatch all
2168 * the events with an EOF result code.
2171 dev->result = ISC_R_EOF;
2172 send_recvdone_event(sock, &dev);
2173 dev = ISC_LIST_HEAD(sock->recv_list);
2174 } while (dev != NULL);
2179 send_recvdone_event(sock, &dev);
2183 dev = ISC_LIST_HEAD(sock->recv_list);
2187 if (!ISC_LIST_EMPTY(sock->recv_list))
2188 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2190 UNLOCK(&sock->lock);
2194 internal_send(isc_task_t *me, isc_event_t *ev) {
2195 isc_socketevent_t *dev;
2198 INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
2201 * Find out what socket this is and lock it.
2203 sock = (isc_socket_t *)ev->ev_sender;
2204 INSIST(VALID_SOCKET(sock));
2207 socket_log(sock, NULL, IOEVENT,
2208 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_INTERNALSEND,
2209 "internal_send: task %p got event %p", me, ev);
2211 INSIST(sock->pending_send == 1);
2212 sock->pending_send = 0;
2214 INSIST(sock->references > 0);
2215 sock->references--; /* the internal event is done with this socket */
2216 if (sock->references == 0) {
2217 UNLOCK(&sock->lock);
2223 * Try to do as much I/O as possible on this socket. There are no
2224 * limits here, currently.
2226 dev = ISC_LIST_HEAD(sock->send_list);
2227 while (dev != NULL) {
2228 switch (doio_send(sock, dev)) {
2234 send_senddone_event(sock, &dev);
2238 dev = ISC_LIST_HEAD(sock->send_list);
2242 if (!ISC_LIST_EMPTY(sock->send_list))
2243 select_poke(sock->manager, sock->fd, SELECT_POKE_WRITE);
2245 UNLOCK(&sock->lock);
2249 process_fds(isc_socketmgr_t *manager, int maxfd,
2250 fd_set *readfds, fd_set *writefds)
2254 isc_boolean_t unlock_sock;
2256 REQUIRE(maxfd <= (int)manager->fdsize);
2259 * Process read/writes on other fds here. Avoid locking
2260 * and unlocking twice if both reads and writes are possible.
2262 for (i = 0; i < maxfd; i++) {
2263 #ifdef ISC_PLATFORM_USETHREADS
2264 if (i == manager->pipe_fds[0] || i == manager->pipe_fds[1])
2266 #endif /* ISC_PLATFORM_USETHREADS */
2268 if (manager->fdstate[i] == CLOSE_PENDING) {
2269 manager->fdstate[i] = CLOSED;
2270 FD_CLR(i, manager->read_fds);
2271 FD_CLR(i, manager->write_fds);
2278 sock = manager->fds[i];
2279 unlock_sock = ISC_FALSE;
2280 if (FD_ISSET(i, readfds)) {
2282 FD_CLR(i, manager->read_fds);
2285 unlock_sock = ISC_TRUE;
2287 if (!SOCK_DEAD(sock)) {
2289 dispatch_accept(sock);
2291 dispatch_recv(sock);
2293 FD_CLR(i, manager->read_fds);
2296 if (FD_ISSET(i, writefds)) {
2298 FD_CLR(i, manager->write_fds);
2302 unlock_sock = ISC_TRUE;
2305 if (!SOCK_DEAD(sock)) {
2306 if (sock->connecting)
2307 dispatch_connect(sock);
2309 dispatch_send(sock);
2311 FD_CLR(i, manager->write_fds);
2314 UNLOCK(&sock->lock);
2318 #ifdef ISC_PLATFORM_USETHREADS
2320 * This is the thread that will loop forever, always in a select or poll
2323 * When select returns something to do, track down what thread gets to do
2324 * this I/O and post the event to it.
2326 static isc_threadresult_t
2327 watcher(void *uap) {
2328 isc_socketmgr_t *manager = uap;
2334 char strbuf[ISC_STRERRORSIZE];
2337 * Get the control fd here. This will never change.
2339 LOCK(&manager->lock);
2340 ctlfd = manager->pipe_fds[0];
2345 memcpy(manager->read_fds_copy, manager->read_fds,
2346 manager->fd_bufsize);
2347 memcpy(manager->write_fds_copy, manager->write_fds,
2348 manager->fd_bufsize);
2349 maxfd = manager->maxfd + 1;
2351 UNLOCK(&manager->lock);
2353 cc = select(maxfd, manager->read_fds_copy,
2354 manager->write_fds_copy, NULL, NULL);
2356 if (!SOFT_ERROR(errno)) {
2357 isc__strerror(errno, strbuf,
2359 FATAL_ERROR(__FILE__, __LINE__,
2361 isc_msgcat_get(isc_msgcat,
2369 LOCK(&manager->lock);
2374 * Process reads on internal, control fd.
2376 if (FD_ISSET(ctlfd, manager->read_fds_copy)) {
2378 select_readmsg(manager, &fd, &msg);
2380 manager_log(manager, IOEVENT,
2381 isc_msgcat_get(isc_msgcat,
2384 "watcher got message %d"),
2390 if (msg == SELECT_POKE_NOTHING)
2394 * Handle shutdown message. We really should
2395 * jump out of this loop right away, but
2396 * it doesn't matter if we have to do a little
2399 if (msg == SELECT_POKE_SHUTDOWN) {
2406 * This is a wakeup on a socket. Look
2407 * at the event queue for both read and write,
2408 * and decide if we need to watch on it now
2411 wakeup_socket(manager, fd, msg);
2415 process_fds(manager, maxfd, manager->read_fds_copy,
2416 manager->write_fds_copy);
2419 manager_log(manager, TRACE,
2420 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2421 ISC_MSG_EXITING, "watcher exiting"));
2423 UNLOCK(&manager->lock);
2424 return ((isc_threadresult_t)0);
2426 #endif /* ISC_PLATFORM_USETHREADS */
2429 isc__socketmgr_setreserved(isc_socketmgr_t *manager, isc_uint32_t reserved) {
2431 REQUIRE(VALID_MANAGER(manager));
2433 manager->reserved = reserved;
2437 * Initialize fdsets in socketmgr structure.
2440 create_fdsets(isc_socketmgr_t *manager, isc_mem_t *mctx) {
2441 #if ISC_SOCKET_FDSETSIZE > FD_SETSIZE
2442 manager->fdsize = ISC_SOCKET_FDSETSIZE;
2443 manager->fd_bufsize = howmany(ISC_SOCKET_FDSETSIZE, NFDBITS) *
2446 manager->fdsize = FD_SETSIZE;
2447 manager->fd_bufsize = sizeof(fd_set);
2450 manager->fds = NULL;
2451 manager->fdstate = NULL;
2452 manager->read_fds = NULL;
2453 manager->read_fds_copy = NULL;
2454 manager->write_fds = NULL;
2455 manager->write_fds_copy = NULL;
2457 manager->fds = isc_mem_get(mctx,
2458 manager->fdsize * sizeof(manager->fds[0]));
2459 if (manager->fds == NULL)
2462 manager->fdstate = isc_mem_get(mctx, manager->fdsize *
2463 sizeof(manager->fdstate[0]));
2464 if (manager->fdstate == NULL)
2467 manager->read_fds = isc_mem_get(mctx, manager->fd_bufsize);
2468 if (manager->read_fds == NULL)
2470 manager->read_fds_copy = isc_mem_get(mctx, manager->fd_bufsize);
2471 if (manager->read_fds_copy == NULL)
2473 manager->write_fds = isc_mem_get(mctx, manager->fd_bufsize);
2474 if (manager->write_fds == NULL)
2476 manager->write_fds_copy = isc_mem_get(mctx, manager->fd_bufsize);
2477 if (manager->write_fds_copy == NULL)
2480 return (ISC_R_SUCCESS);
2483 cleanup_fdsets(manager, mctx);
2484 return (ISC_R_NOMEMORY);
2488 * Clean up fdsets in socketmgr structure.
2491 cleanup_fdsets(isc_socketmgr_t *manager, isc_mem_t *mctx) {
2492 if (manager->fds != NULL) {
2493 isc_mem_put(mctx, manager->fds,
2494 manager->fdsize * sizeof(manager->fds[0]));
2496 if (manager->fdstate != NULL) {
2497 isc_mem_put(mctx, manager->fdstate,
2498 manager->fdsize * sizeof(manager->fdstate[0]));
2500 if (manager->read_fds != NULL)
2501 isc_mem_put(mctx, manager->read_fds, manager->fd_bufsize);
2502 if (manager->read_fds_copy != NULL)
2503 isc_mem_put(mctx, manager->read_fds_copy, manager->fd_bufsize);
2504 if (manager->write_fds != NULL)
2505 isc_mem_put(mctx, manager->write_fds, manager->fd_bufsize);
2506 if (manager->write_fds_copy != NULL)
2507 isc_mem_put(mctx, manager->write_fds_copy, manager->fd_bufsize);
2511 * Create a new socket manager.
2514 isc_socketmgr_create(isc_mem_t *mctx, isc_socketmgr_t **managerp) {
2515 isc_socketmgr_t *manager;
2516 #ifdef ISC_PLATFORM_USETHREADS
2517 char strbuf[ISC_STRERRORSIZE];
2519 isc_result_t result;
2521 REQUIRE(managerp != NULL && *managerp == NULL);
2523 #ifndef ISC_PLATFORM_USETHREADS
2524 if (socketmgr != NULL) {
2526 *managerp = socketmgr;
2527 return (ISC_R_SUCCESS);
2529 #endif /* ISC_PLATFORM_USETHREADS */
2531 manager = isc_mem_get(mctx, sizeof(*manager));
2532 if (manager == NULL)
2533 return (ISC_R_NOMEMORY);
2535 result = create_fdsets(manager, mctx);
2536 if (result != ISC_R_SUCCESS) {
2537 cleanup_fdsets(manager, mctx);
2538 isc_mem_put(mctx, manager, sizeof(*manager));
2542 manager->magic = SOCKET_MANAGER_MAGIC;
2543 manager->mctx = NULL;
2544 ISC_LIST_INIT(manager->socklist);
2545 result = isc_mutex_init(&manager->lock);
2546 if (result != ISC_R_SUCCESS) {
2547 cleanup_fdsets(manager, mctx);
2548 isc_mem_put(mctx, manager, sizeof(*manager));
2549 UNEXPECTED_ERROR(__FILE__, __LINE__,
2550 "isc_mutex_init() %s",
2551 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2552 ISC_MSG_FAILED, "failed"));
2553 return (ISC_R_UNEXPECTED);
2555 #ifdef ISC_PLATFORM_USETHREADS
2556 if (isc_condition_init(&manager->shutdown_ok) != ISC_R_SUCCESS) {
2557 cleanup_fdsets(manager, mctx);
2558 DESTROYLOCK(&manager->lock);
2559 isc_mem_put(mctx, manager, sizeof(*manager));
2560 UNEXPECTED_ERROR(__FILE__, __LINE__,
2561 "isc_condition_init() %s",
2562 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2563 ISC_MSG_FAILED, "failed"));
2564 return (ISC_R_UNEXPECTED);
2568 * Create the special fds that will be used to wake up the
2569 * select/poll loop when something internal needs to be done.
2571 if (pipe(manager->pipe_fds) != 0) {
2572 cleanup_fdsets(manager, mctx);
2573 DESTROYLOCK(&manager->lock);
2574 isc_mem_put(mctx, manager, sizeof(*manager));
2575 isc__strerror(errno, strbuf, sizeof(strbuf));
2576 UNEXPECTED_ERROR(__FILE__, __LINE__,
2578 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2579 ISC_MSG_FAILED, "failed"),
2582 return (ISC_R_UNEXPECTED);
2585 RUNTIME_CHECK(make_nonblock(manager->pipe_fds[0]) == ISC_R_SUCCESS);
2587 RUNTIME_CHECK(make_nonblock(manager->pipe_fds[1]) == ISC_R_SUCCESS);
2589 #else /* ISC_PLATFORM_USETHREADS */
2591 #endif /* ISC_PLATFORM_USETHREADS */
2594 * Set up initial state for the select loop
2596 memset(manager->read_fds, 0, manager->fd_bufsize);
2597 memset(manager->write_fds, 0, manager->fd_bufsize);
2598 #ifdef ISC_PLATFORM_USETHREADS
2599 FD_SET(manager->pipe_fds[0], manager->read_fds);
2600 manager->maxfd = manager->pipe_fds[0];
2601 #else /* ISC_PLATFORM_USETHREADS */
2603 #endif /* ISC_PLATFORM_USETHREADS */
2604 manager->reserved = 0;
2605 memset(manager->fdstate, 0,
2606 manager->fdsize * sizeof(manager->fdstate[0]));
2608 #ifdef ISC_PLATFORM_USETHREADS
2610 * Start up the select/poll thread.
2612 if (isc_thread_create(watcher, manager, &manager->watcher) !=
2614 (void)close(manager->pipe_fds[0]);
2615 (void)close(manager->pipe_fds[1]);
2616 DESTROYLOCK(&manager->lock);
2617 isc_mem_put(mctx, manager, sizeof(*manager));
2618 UNEXPECTED_ERROR(__FILE__, __LINE__,
2619 "isc_thread_create() %s",
2620 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2621 ISC_MSG_FAILED, "failed"));
2622 return (ISC_R_UNEXPECTED);
2624 #endif /* ISC_PLATFORM_USETHREADS */
2625 isc_mem_attach(mctx, &manager->mctx);
2627 #ifndef ISC_PLATFORM_USETHREADS
2628 socketmgr = manager;
2629 #endif /* ISC_PLATFORM_USETHREADS */
2630 *managerp = manager;
2632 return (ISC_R_SUCCESS);
2636 isc_socketmgr_destroy(isc_socketmgr_t **managerp) {
2637 isc_socketmgr_t *manager;
2642 * Destroy a socket manager.
2645 REQUIRE(managerp != NULL);
2646 manager = *managerp;
2647 REQUIRE(VALID_MANAGER(manager));
2649 #ifndef ISC_PLATFORM_USETHREADS
2650 if (manager->refs > 1) {
2655 #endif /* ISC_PLATFORM_USETHREADS */
2657 LOCK(&manager->lock);
2659 #ifdef ISC_PLATFORM_USETHREADS
2661 * Wait for all sockets to be destroyed.
2663 while (!ISC_LIST_EMPTY(manager->socklist)) {
2664 manager_log(manager, CREATION,
2665 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
2666 ISC_MSG_SOCKETSREMAIN,
2668 WAIT(&manager->shutdown_ok, &manager->lock);
2670 #else /* ISC_PLATFORM_USETHREADS */
2672 * Hope all sockets have been destroyed.
2674 if (!ISC_LIST_EMPTY(manager->socklist)) {
2675 manager_log(manager, CREATION,
2676 isc_msgcat_get(isc_msgcat, ISC_MSGSET_SOCKET,
2677 ISC_MSG_SOCKETSREMAIN,
2681 #endif /* ISC_PLATFORM_USETHREADS */
2683 UNLOCK(&manager->lock);
2686 * Here, poke our select/poll thread. Do this by closing the write
2687 * half of the pipe, which will send EOF to the read half.
2688 * This is currently a no-op in the non-threaded case.
2690 select_poke(manager, 0, SELECT_POKE_SHUTDOWN);
2692 #ifdef ISC_PLATFORM_USETHREADS
2694 * Wait for thread to exit.
2696 if (isc_thread_join(manager->watcher, NULL) != ISC_R_SUCCESS)
2697 UNEXPECTED_ERROR(__FILE__, __LINE__,
2698 "isc_thread_join() %s",
2699 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
2700 ISC_MSG_FAILED, "failed"));
2701 #endif /* ISC_PLATFORM_USETHREADS */
2706 #ifdef ISC_PLATFORM_USETHREADS
2707 (void)close(manager->pipe_fds[0]);
2708 (void)close(manager->pipe_fds[1]);
2709 (void)isc_condition_destroy(&manager->shutdown_ok);
2710 #endif /* ISC_PLATFORM_USETHREADS */
2712 for (i = 0; i < (int)manager->fdsize; i++)
2713 if (manager->fdstate[i] == CLOSE_PENDING)
2716 DESTROYLOCK(&manager->lock);
2717 cleanup_fdsets(manager, manager->mctx);
2719 mctx= manager->mctx;
2720 isc_mem_put(mctx, manager, sizeof(*manager));
2722 isc_mem_detach(&mctx);
2728 socket_recv(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
2732 isc_boolean_t have_lock = ISC_FALSE;
2733 isc_task_t *ntask = NULL;
2734 isc_result_t result = ISC_R_SUCCESS;
2736 dev->ev_sender = task;
2738 if (sock->type == isc_sockettype_udp) {
2739 io_state = doio_recv(sock, dev);
2742 have_lock = ISC_TRUE;
2744 if (ISC_LIST_EMPTY(sock->recv_list))
2745 io_state = doio_recv(sock, dev);
2747 io_state = DOIO_SOFT;
2753 * We couldn't read all or part of the request right now, so
2756 * Attach to socket and to task
2758 isc_task_attach(task, &ntask);
2759 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
2763 have_lock = ISC_TRUE;
2767 * Enqueue the request. If the socket was previously not being
2768 * watched, poke the watcher to start paying attention to it.
2770 if (ISC_LIST_EMPTY(sock->recv_list))
2771 select_poke(sock->manager, sock->fd, SELECT_POKE_READ);
2772 ISC_LIST_ENQUEUE(sock->recv_list, dev, ev_link);
2774 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2775 "socket_recv: event %p -> task %p",
2778 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
2779 result = ISC_R_INPROGRESS;
2783 dev->result = ISC_R_EOF;
2788 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
2789 send_recvdone_event(sock, &dev);
2794 UNLOCK(&sock->lock);
2800 isc_socket_recvv(isc_socket_t *sock, isc_bufferlist_t *buflist,
2801 unsigned int minimum, isc_task_t *task,
2802 isc_taskaction_t action, const void *arg)
2804 isc_socketevent_t *dev;
2805 isc_socketmgr_t *manager;
2806 unsigned int iocount;
2807 isc_buffer_t *buffer;
2809 REQUIRE(VALID_SOCKET(sock));
2810 REQUIRE(buflist != NULL);
2811 REQUIRE(!ISC_LIST_EMPTY(*buflist));
2812 REQUIRE(task != NULL);
2813 REQUIRE(action != NULL);
2815 manager = sock->manager;
2816 REQUIRE(VALID_MANAGER(manager));
2818 iocount = isc_bufferlist_availablecount(buflist);
2819 REQUIRE(iocount > 0);
2821 INSIST(sock->bound);
2823 dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
2825 return (ISC_R_NOMEMORY);
2829 * UDP sockets are always partial read
2831 if (sock->type == isc_sockettype_udp)
2835 dev->minimum = iocount;
2837 dev->minimum = minimum;
2841 * Move each buffer from the passed in list to our internal one.
2843 buffer = ISC_LIST_HEAD(*buflist);
2844 while (buffer != NULL) {
2845 ISC_LIST_DEQUEUE(*buflist, buffer, link);
2846 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
2847 buffer = ISC_LIST_HEAD(*buflist);
2850 return (socket_recv(sock, dev, task, 0));
2854 isc_socket_recv(isc_socket_t *sock, isc_region_t *region, unsigned int minimum,
2855 isc_task_t *task, isc_taskaction_t action, const void *arg)
2857 isc_socketevent_t *dev;
2858 isc_socketmgr_t *manager;
2860 REQUIRE(VALID_SOCKET(sock));
2861 REQUIRE(action != NULL);
2863 manager = sock->manager;
2864 REQUIRE(VALID_MANAGER(manager));
2866 INSIST(sock->bound);
2868 dev = allocate_socketevent(sock, ISC_SOCKEVENT_RECVDONE, action, arg);
2870 return (ISC_R_NOMEMORY);
2872 return (isc_socket_recv2(sock, region, minimum, task, dev, 0));
2876 isc_socket_recv2(isc_socket_t *sock, isc_region_t *region,
2877 unsigned int minimum, isc_task_t *task,
2878 isc_socketevent_t *event, unsigned int flags)
2880 event->ev_sender = sock;
2881 event->result = ISC_R_UNEXPECTED;
2882 ISC_LIST_INIT(event->bufferlist);
2883 event->region = *region;
2886 event->attributes = 0;
2889 * UDP sockets are always partial read.
2891 if (sock->type == isc_sockettype_udp)
2895 event->minimum = region->length;
2897 event->minimum = minimum;
2900 return (socket_recv(sock, event, task, flags));
2904 socket_send(isc_socket_t *sock, isc_socketevent_t *dev, isc_task_t *task,
2905 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
2909 isc_boolean_t have_lock = ISC_FALSE;
2910 isc_task_t *ntask = NULL;
2911 isc_result_t result = ISC_R_SUCCESS;
2913 dev->ev_sender = task;
2915 set_dev_address(address, sock, dev);
2916 if (pktinfo != NULL) {
2917 dev->attributes |= ISC_SOCKEVENTATTR_PKTINFO;
2918 dev->pktinfo = *pktinfo;
2920 if (!isc_sockaddr_issitelocal(&dev->address) &&
2921 !isc_sockaddr_islinklocal(&dev->address)) {
2922 socket_log(sock, NULL, TRACE, isc_msgcat,
2923 ISC_MSGSET_SOCKET, ISC_MSG_PKTINFOPROVIDED,
2924 "pktinfo structure provided, ifindex %u "
2925 "(set to 0)", pktinfo->ipi6_ifindex);
2928 * Set the pktinfo index to 0 here, to let the
2929 * kernel decide what interface it should send on.
2931 dev->pktinfo.ipi6_ifindex = 0;
2935 if (sock->type == isc_sockettype_udp)
2936 io_state = doio_send(sock, dev);
2939 have_lock = ISC_TRUE;
2941 if (ISC_LIST_EMPTY(sock->send_list))
2942 io_state = doio_send(sock, dev);
2944 io_state = DOIO_SOFT;
2950 * We couldn't send all or part of the request right now, so
2951 * queue it unless ISC_SOCKFLAG_NORETRY is set.
2953 if ((flags & ISC_SOCKFLAG_NORETRY) == 0) {
2954 isc_task_attach(task, &ntask);
2955 dev->attributes |= ISC_SOCKEVENTATTR_ATTACHED;
2959 have_lock = ISC_TRUE;
2963 * Enqueue the request. If the socket was previously
2964 * not being watched, poke the watcher to start
2965 * paying attention to it.
2967 if (ISC_LIST_EMPTY(sock->send_list))
2968 select_poke(sock->manager, sock->fd,
2970 ISC_LIST_ENQUEUE(sock->send_list, dev, ev_link);
2972 socket_log(sock, NULL, EVENT, NULL, 0, 0,
2973 "socket_send: event %p -> task %p",
2976 if ((flags & ISC_SOCKFLAG_IMMEDIATE) != 0)
2977 result = ISC_R_INPROGRESS;
2983 if ((flags & ISC_SOCKFLAG_IMMEDIATE) == 0)
2984 send_senddone_event(sock, &dev);
2989 UNLOCK(&sock->lock);
2995 isc_socket_send(isc_socket_t *sock, isc_region_t *region,
2996 isc_task_t *task, isc_taskaction_t action, const void *arg)
2999 * REQUIRE() checking is performed in isc_socket_sendto().
3001 return (isc_socket_sendto(sock, region, task, action, arg, NULL,
3006 isc_socket_sendto(isc_socket_t *sock, isc_region_t *region,
3007 isc_task_t *task, isc_taskaction_t action, const void *arg,
3008 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
3010 isc_socketevent_t *dev;
3011 isc_socketmgr_t *manager;
3013 REQUIRE(VALID_SOCKET(sock));
3014 REQUIRE(region != NULL);
3015 REQUIRE(task != NULL);
3016 REQUIRE(action != NULL);
3018 manager = sock->manager;
3019 REQUIRE(VALID_MANAGER(manager));
3021 INSIST(sock->bound);
3023 dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
3025 return (ISC_R_NOMEMORY);
3028 dev->region = *region;
3030 return (socket_send(sock, dev, task, address, pktinfo, 0));
3034 isc_socket_sendv(isc_socket_t *sock, isc_bufferlist_t *buflist,
3035 isc_task_t *task, isc_taskaction_t action, const void *arg)
3037 return (isc_socket_sendtov(sock, buflist, task, action, arg, NULL,
3042 isc_socket_sendtov(isc_socket_t *sock, isc_bufferlist_t *buflist,
3043 isc_task_t *task, isc_taskaction_t action, const void *arg,
3044 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo)
3046 isc_socketevent_t *dev;
3047 isc_socketmgr_t *manager;
3048 unsigned int iocount;
3049 isc_buffer_t *buffer;
3051 REQUIRE(VALID_SOCKET(sock));
3052 REQUIRE(buflist != NULL);
3053 REQUIRE(!ISC_LIST_EMPTY(*buflist));
3054 REQUIRE(task != NULL);
3055 REQUIRE(action != NULL);
3057 manager = sock->manager;
3058 REQUIRE(VALID_MANAGER(manager));
3060 iocount = isc_bufferlist_usedcount(buflist);
3061 REQUIRE(iocount > 0);
3063 dev = allocate_socketevent(sock, ISC_SOCKEVENT_SENDDONE, action, arg);
3065 return (ISC_R_NOMEMORY);
3069 * Move each buffer from the passed in list to our internal one.
3071 buffer = ISC_LIST_HEAD(*buflist);
3072 while (buffer != NULL) {
3073 ISC_LIST_DEQUEUE(*buflist, buffer, link);
3074 ISC_LIST_ENQUEUE(dev->bufferlist, buffer, link);
3075 buffer = ISC_LIST_HEAD(*buflist);
3078 return (socket_send(sock, dev, task, address, pktinfo, 0));
3082 isc_socket_sendto2(isc_socket_t *sock, isc_region_t *region,
3084 isc_sockaddr_t *address, struct in6_pktinfo *pktinfo,
3085 isc_socketevent_t *event, unsigned int flags)
3087 REQUIRE((flags & ~(ISC_SOCKFLAG_IMMEDIATE|ISC_SOCKFLAG_NORETRY)) == 0);
3088 if ((flags & ISC_SOCKFLAG_NORETRY) != 0)
3089 REQUIRE(sock->type == isc_sockettype_udp);
3090 event->ev_sender = sock;
3091 event->result = ISC_R_UNEXPECTED;
3092 ISC_LIST_INIT(event->bufferlist);
3093 event->region = *region;
3096 event->attributes = 0;
3098 return (socket_send(sock, event, task, address, pktinfo, flags));
3102 isc_socket_bind(isc_socket_t *sock, isc_sockaddr_t *sockaddr,
3103 unsigned int options) {
3104 char strbuf[ISC_STRERRORSIZE];
3109 INSIST(!sock->bound);
3111 if (sock->pf != sockaddr->type.sa.sa_family) {
3112 UNLOCK(&sock->lock);
3113 return (ISC_R_FAMILYMISMATCH);
3116 * Only set SO_REUSEADDR when we want a specific port.
3118 if ((options & ISC_SOCKET_REUSEADDRESS) != 0 &&
3119 isc_sockaddr_getport(sockaddr) != (in_port_t)0 &&
3120 setsockopt(sock->fd, SOL_SOCKET, SO_REUSEADDR, (void *)&on,
3122 UNEXPECTED_ERROR(__FILE__, __LINE__,
3123 "setsockopt(%d) %s", sock->fd,
3124 isc_msgcat_get(isc_msgcat, ISC_MSGSET_GENERAL,
3125 ISC_MSG_FAILED, "failed"));
3128 if (bind(sock->fd, &sockaddr->type.sa, sockaddr->length) < 0) {
3129 UNLOCK(&sock->lock);
3132 return (ISC_R_NOPERM);
3134 return (ISC_R_ADDRNOTAVAIL);
3136 return (ISC_R_ADDRINUSE);
3138 return (ISC_R_BOUND);
3140 isc__strerror(errno, strbuf, sizeof(strbuf));
3141 UNEXPECTED_ERROR(__FILE__, __LINE__, "bind: %s",
3143 return (ISC_R_UNEXPECTED);
3147 socket_log(sock, sockaddr, TRACE,
3148 isc_msgcat, ISC_MSGSET_SOCKET, ISC_MSG_BOUND, "bound");
3151 UNLOCK(&sock->lock);
3152 return (ISC_R_SUCCESS);
3156 isc_socket_filter(isc_socket_t *sock, const char *filter) {
3157 #ifdef SO_ACCEPTFILTER
3158 char strbuf[ISC_STRERRORSIZE];
3159 struct accept_filter_arg afa;
3165 REQUIRE(VALID_SOCKET(sock));
3167 #ifdef SO_ACCEPTFILTER
3168 bzero(&afa, sizeof(afa));
3169 strncpy(afa.af_name, filter, sizeof(afa.af_name));
3170 if (setsockopt(sock->fd, SOL_SOCKET, SO_ACCEPTFILTER,
3171 &afa, sizeof(afa)) == -1) {
3172 isc__strerror(errno, strbuf, sizeof(strbuf));
3173 socket_log(sock, NULL, CREATION, isc_msgcat, ISC_MSGSET_SOCKET,
3174 ISC_MSG_FILTER, "setsockopt(SO_ACCEPTFILTER): %s",
3176 return (ISC_R_FAILURE);
3178 return (ISC_R_SUCCESS);
3180 return (ISC_R_NOTIMPLEMENTED);
3185 * Set up to listen on a given socket. We do this by creating an internal
3186 * event that will be dispatched when the socket has read activity. The
3187 * watcher will send the internal event to the task when there is a new
3190 * Unlike in read, we don't preallocate a done event here. Every time there
3191 * is a new connection we'll have to allocate a new one anyway, so we might
3192 * as well keep things simple rather than having to track them.
3195 isc_socket_listen(isc_socket_t *sock, unsigned int backlog) {
3196 char strbuf[ISC_STRERRORSIZE];
3198 REQUIRE(VALID_SOCKET(sock));
3202 REQUIRE(!sock->listener);
3203 REQUIRE(sock->bound);
3204 REQUIRE(sock->type == isc_sockettype_tcp);
3207 backlog = SOMAXCONN;
3209 if (listen(sock->fd, (int)backlog) < 0) {
3210 UNLOCK(&sock->lock);
3211 isc__strerror(errno, strbuf, sizeof(strbuf));
3213 UNEXPECTED_ERROR(__FILE__, __LINE__, "listen: %s", strbuf);
3215 return (ISC_R_UNEXPECTED);
3220 UNLOCK(&sock->lock);
3221 return (ISC_R_SUCCESS);
3225 * This should try to do agressive accept() XXXMLG
3228 isc_socket_accept(isc_socket_t *sock,
3229 isc_task_t *task, isc_taskaction_t action, const void *arg)
3231 isc_socket_newconnev_t *dev;
3232 isc_socketmgr_t *manager;
3233 isc_task_t *ntask = NULL;
3234 isc_socket_t *nsock;
3236 isc_boolean_t do_poke = ISC_FALSE;
3238 REQUIRE(VALID_SOCKET(sock));
3239 manager = sock->manager;
3240 REQUIRE(VALID_MANAGER(manager));
3244 REQUIRE(sock->listener);
3247 * Sender field is overloaded here with the task we will be sending
3248 * this event to. Just before the actual event is delivered the
3249 * actual ev_sender will be touched up to be the socket.
3251 dev = (isc_socket_newconnev_t *)
3252 isc_event_allocate(manager->mctx, task, ISC_SOCKEVENT_NEWCONN,
3253 action, arg, sizeof(*dev));
3255 UNLOCK(&sock->lock);
3256 return (ISC_R_NOMEMORY);
3258 ISC_LINK_INIT(dev, ev_link);
3260 ret = allocate_socket(manager, sock->type, &nsock);
3261 if (ret != ISC_R_SUCCESS) {
3262 isc_event_free(ISC_EVENT_PTR(&dev));
3263 UNLOCK(&sock->lock);
3268 * Attach to socket and to task.
3270 isc_task_attach(task, &ntask);
3271 nsock->references++;
3273 dev->ev_sender = ntask;
3274 dev->newsocket = nsock;
3277 * Poke watcher here. We still have the socket locked, so there
3278 * is no race condition. We will keep the lock for such a short
3279 * bit of time waking it up now or later won't matter all that much.
3281 if (ISC_LIST_EMPTY(sock->accept_list))
3284 ISC_LIST_ENQUEUE(sock->accept_list, dev, ev_link);
3287 select_poke(manager, sock->fd, SELECT_POKE_ACCEPT);
3289 UNLOCK(&sock->lock);
3290 return (ISC_R_SUCCESS);
3294 isc_socket_connect(isc_socket_t *sock, isc_sockaddr_t *addr,
3295 isc_task_t *task, isc_taskaction_t action, const void *arg)
3297 isc_socket_connev_t *dev;
3298 isc_task_t *ntask = NULL;
3299 isc_socketmgr_t *manager;
3301 char strbuf[ISC_STRERRORSIZE];
3303 REQUIRE(VALID_SOCKET(sock));
3304 REQUIRE(addr != NULL);
3305 REQUIRE(task != NULL);
3306 REQUIRE(action != NULL);
3308 manager = sock->manager;
3309 REQUIRE(VALID_MANAGER(manager));
3310 REQUIRE(addr != NULL);
3312 if (isc_sockaddr_ismulticast(addr))
3313 return (ISC_R_MULTICAST);
3317 REQUIRE(!sock->connecting);
3319 dev = (isc_socket_connev_t *)isc_event_allocate(manager->mctx, sock,
3320 ISC_SOCKEVENT_CONNECT,
3324 UNLOCK(&sock->lock);
3325 return (ISC_R_NOMEMORY);
3327 ISC_LINK_INIT(dev, ev_link);
3330 * Try to do the connect right away, as there can be only one
3331 * outstanding, and it might happen to complete.
3333 sock->address = *addr;
3334 cc = connect(sock->fd, &addr->type.sa, addr->length);
3336 if (SOFT_ERROR(errno) || errno == EINPROGRESS)
3340 #define ERROR_MATCH(a, b) case a: dev->result = b; goto err_exit;
3341 ERROR_MATCH(EACCES, ISC_R_NOPERM);
3342 ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
3343 ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
3344 ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
3345 ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
3347 ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
3349 ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
3350 ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
3351 ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
3352 ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
3353 ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
3357 sock->connected = 0;
3359 isc__strerror(errno, strbuf, sizeof(strbuf));
3360 UNEXPECTED_ERROR(__FILE__, __LINE__, "%d/%s", errno, strbuf);
3362 UNLOCK(&sock->lock);
3363 isc_event_free(ISC_EVENT_PTR(&dev));
3364 return (ISC_R_UNEXPECTED);
3367 sock->connected = 0;
3368 isc_task_send(task, ISC_EVENT_PTR(&dev));
3370 UNLOCK(&sock->lock);
3371 return (ISC_R_SUCCESS);
3375 * If connect completed, fire off the done event.
3378 sock->connected = 1;
3380 dev->result = ISC_R_SUCCESS;
3381 isc_task_send(task, ISC_EVENT_PTR(&dev));
3383 UNLOCK(&sock->lock);
3384 return (ISC_R_SUCCESS);
3392 isc_task_attach(task, &ntask);
3394 sock->connecting = 1;
3396 dev->ev_sender = ntask;
3399 * Poke watcher here. We still have the socket locked, so there
3400 * is no race condition. We will keep the lock for such a short
3401 * bit of time waking it up now or later won't matter all that much.
3403 if (sock->connect_ev == NULL)
3404 select_poke(manager, sock->fd, SELECT_POKE_CONNECT);
3406 sock->connect_ev = dev;
3408 UNLOCK(&sock->lock);
3409 return (ISC_R_SUCCESS);
3413 * Called when a socket with a pending connect() finishes.
3416 internal_connect(isc_task_t *me, isc_event_t *ev) {
3418 isc_socket_connev_t *dev;
3421 ISC_SOCKADDR_LEN_T optlen;
3422 char strbuf[ISC_STRERRORSIZE];
3423 char peerbuf[ISC_SOCKADDR_FORMATSIZE];
3426 INSIST(ev->ev_type == ISC_SOCKEVENT_INTW);
3428 sock = ev->ev_sender;
3429 INSIST(VALID_SOCKET(sock));
3434 * When the internal event was sent the reference count was bumped
3435 * to keep the socket around for us. Decrement the count here.
3437 INSIST(sock->references > 0);
3439 if (sock->references == 0) {
3440 UNLOCK(&sock->lock);
3446 * Has this event been canceled?
3448 dev = sock->connect_ev;
3450 INSIST(!sock->connecting);
3451 UNLOCK(&sock->lock);
3455 INSIST(sock->connecting);
3456 sock->connecting = 0;
3459 * Get any possible error status here.
3461 optlen = sizeof(cc);
3462 if (getsockopt(sock->fd, SOL_SOCKET, SO_ERROR,
3463 (void *)&cc, (void *)&optlen) < 0)
3470 * If the error is EAGAIN, just re-select on this
3471 * fd and pretend nothing strange happened.
3473 if (SOFT_ERROR(errno) || errno == EINPROGRESS) {
3474 sock->connecting = 1;
3475 select_poke(sock->manager, sock->fd,
3476 SELECT_POKE_CONNECT);
3477 UNLOCK(&sock->lock);
3483 * Translate other errors into ISC_R_* flavors.
3486 #define ERROR_MATCH(a, b) case a: dev->result = b; break;
3487 ERROR_MATCH(EACCES, ISC_R_NOPERM);
3488 ERROR_MATCH(EADDRNOTAVAIL, ISC_R_ADDRNOTAVAIL);
3489 ERROR_MATCH(EAFNOSUPPORT, ISC_R_ADDRNOTAVAIL);
3490 ERROR_MATCH(ECONNREFUSED, ISC_R_CONNREFUSED);
3491 ERROR_MATCH(EHOSTUNREACH, ISC_R_HOSTUNREACH);
3493 ERROR_MATCH(EHOSTDOWN, ISC_R_HOSTUNREACH);
3495 ERROR_MATCH(ENETUNREACH, ISC_R_NETUNREACH);
3496 ERROR_MATCH(ENOBUFS, ISC_R_NORESOURCES);
3497 ERROR_MATCH(EPERM, ISC_R_HOSTUNREACH);
3498 ERROR_MATCH(EPIPE, ISC_R_NOTCONNECTED);
3499 ERROR_MATCH(ETIMEDOUT, ISC_R_TIMEDOUT);
3500 ERROR_MATCH(ECONNRESET, ISC_R_CONNECTIONRESET);
3503 dev->result = ISC_R_UNEXPECTED;
3504 isc_sockaddr_format(&sock->address, peerbuf,
3506 isc__strerror(errno, strbuf, sizeof(strbuf));
3507 UNEXPECTED_ERROR(__FILE__, __LINE__,
3508 "internal_connect: connect(%s) %s",
3512 dev->result = ISC_R_SUCCESS;
3513 sock->connected = 1;
3517 sock->connect_ev = NULL;
3519 UNLOCK(&sock->lock);
3521 task = dev->ev_sender;
3522 dev->ev_sender = sock;
3523 isc_task_sendanddetach(&task, ISC_EVENT_PTR(&dev));
3527 isc_socket_getpeername(isc_socket_t *sock, isc_sockaddr_t *addressp) {
3530 REQUIRE(VALID_SOCKET(sock));
3531 REQUIRE(addressp != NULL);
3535 if (sock->connected) {
3536 *addressp = sock->address;
3537 ret = ISC_R_SUCCESS;
3539 ret = ISC_R_NOTCONNECTED;
3542 UNLOCK(&sock->lock);
3548 isc_socket_getsockname(isc_socket_t *sock, isc_sockaddr_t *addressp) {
3549 ISC_SOCKADDR_LEN_T len;
3551 char strbuf[ISC_STRERRORSIZE];
3553 REQUIRE(VALID_SOCKET(sock));
3554 REQUIRE(addressp != NULL);
3559 ret = ISC_R_NOTBOUND;
3563 ret = ISC_R_SUCCESS;
3565 len = sizeof(addressp->type);
3566 if (getsockname(sock->fd, &addressp->type.sa, (void *)&len) < 0) {
3567 isc__strerror(errno, strbuf, sizeof(strbuf));
3568 UNEXPECTED_ERROR(__FILE__, __LINE__, "getsockname: %s",
3570 ret = ISC_R_UNEXPECTED;
3573 addressp->length = (unsigned int)len;
3576 UNLOCK(&sock->lock);
3582 * Run through the list of events on this socket, and cancel the ones
3583 * queued for task "task" of type "how". "how" is a bitmask.
3586 isc_socket_cancel(isc_socket_t *sock, isc_task_t *task, unsigned int how) {
3588 REQUIRE(VALID_SOCKET(sock));
3591 * Quick exit if there is nothing to do. Don't even bother locking
3600 * All of these do the same thing, more or less.
3602 * o If the internal event is marked as "posted" try to
3603 * remove it from the task's queue. If this fails, mark it
3604 * as canceled instead, and let the task clean it up later.
3605 * o For each I/O request for that task of that type, post
3606 * its done event with status of "ISC_R_CANCELED".
3607 * o Reset any state needed.
3609 if (((how & ISC_SOCKCANCEL_RECV) == ISC_SOCKCANCEL_RECV)
3610 && !ISC_LIST_EMPTY(sock->recv_list)) {
3611 isc_socketevent_t *dev;
3612 isc_socketevent_t *next;
3613 isc_task_t *current_task;
3615 dev = ISC_LIST_HEAD(sock->recv_list);
3617 while (dev != NULL) {
3618 current_task = dev->ev_sender;
3619 next = ISC_LIST_NEXT(dev, ev_link);
3621 if ((task == NULL) || (task == current_task)) {
3622 dev->result = ISC_R_CANCELED;
3623 send_recvdone_event(sock, &dev);
3629 if (((how & ISC_SOCKCANCEL_SEND) == ISC_SOCKCANCEL_SEND)
3630 && !ISC_LIST_EMPTY(sock->send_list)) {
3631 isc_socketevent_t *dev;
3632 isc_socketevent_t *next;
3633 isc_task_t *current_task;
3635 dev = ISC_LIST_HEAD(sock->send_list);
3637 while (dev != NULL) {
3638 current_task = dev->ev_sender;
3639 next = ISC_LIST_NEXT(dev, ev_link);
3641 if ((task == NULL) || (task == current_task)) {
3642 dev->result = ISC_R_CANCELED;
3643 send_senddone_event(sock, &dev);
3649 if (((how & ISC_SOCKCANCEL_ACCEPT) == ISC_SOCKCANCEL_ACCEPT)
3650 && !ISC_LIST_EMPTY(sock->accept_list)) {
3651 isc_socket_newconnev_t *dev;
3652 isc_socket_newconnev_t *next;
3653 isc_task_t *current_task;
3655 dev = ISC_LIST_HEAD(sock->accept_list);
3656 while (dev != NULL) {
3657 current_task = dev->ev_sender;
3658 next = ISC_LIST_NEXT(dev, ev_link);
3660 if ((task == NULL) || (task == current_task)) {
3662 ISC_LIST_UNLINK(sock->accept_list, dev,
3665 dev->newsocket->references--;
3666 free_socket(&dev->newsocket);
3668 dev->result = ISC_R_CANCELED;
3669 dev->ev_sender = sock;
3670 isc_task_sendanddetach(¤t_task,
3671 ISC_EVENT_PTR(&dev));
3679 * Connecting is not a list.
3681 if (((how & ISC_SOCKCANCEL_CONNECT) == ISC_SOCKCANCEL_CONNECT)
3682 && sock->connect_ev != NULL) {
3683 isc_socket_connev_t *dev;
3684 isc_task_t *current_task;
3686 INSIST(sock->connecting);
3687 sock->connecting = 0;
3689 dev = sock->connect_ev;
3690 current_task = dev->ev_sender;
3692 if ((task == NULL) || (task == current_task)) {
3693 sock->connect_ev = NULL;
3695 dev->result = ISC_R_CANCELED;
3696 dev->ev_sender = sock;
3697 isc_task_sendanddetach(¤t_task,
3698 ISC_EVENT_PTR(&dev));
3702 UNLOCK(&sock->lock);
3706 isc_socket_gettype(isc_socket_t *sock) {
3707 REQUIRE(VALID_SOCKET(sock));
3709 return (sock->type);
3713 isc_socket_isbound(isc_socket_t *sock) {
3717 val = ((sock->bound) ? ISC_TRUE : ISC_FALSE);
3718 UNLOCK(&sock->lock);
3724 isc_socket_ipv6only(isc_socket_t *sock, isc_boolean_t yes) {
3725 #if defined(IPV6_V6ONLY)
3726 int onoff = yes ? 1 : 0;
3732 REQUIRE(VALID_SOCKET(sock));
3735 if (sock->pf == AF_INET6) {
3736 (void)setsockopt(sock->fd, IPPROTO_IPV6, IPV6_V6ONLY,
3737 (void *)&onoff, sizeof(onoff));
3742 #ifndef ISC_PLATFORM_USETHREADS
3744 isc__socketmgr_getfdsets(fd_set **readset, fd_set **writeset, int *maxfd) {
3745 if (socketmgr == NULL)
3748 /* Prepare duplicates of fd_sets, as select() will modify */
3749 memcpy(socketmgr->read_fds_copy, socketmgr->read_fds,
3750 socketmgr->fd_bufsize);
3751 memcpy(socketmgr->write_fds_copy, socketmgr->write_fds,
3752 socketmgr->fd_bufsize);
3753 *readset = socketmgr->read_fds_copy;
3754 *writeset = socketmgr->write_fds_copy;
3755 *maxfd = socketmgr->maxfd + 1;
3760 isc__socketmgr_dispatch(fd_set *readset, fd_set *writeset, int maxfd) {
3761 isc_socketmgr_t *manager = socketmgr;
3763 if (manager == NULL)
3764 return (ISC_R_NOTFOUND);
3766 process_fds(manager, maxfd, readset, writeset);
3767 return (ISC_R_SUCCESS);
3769 #endif /* ISC_PLATFORM_USETHREADS */