1 /*-------------------------------------------------------------------------
4 * Communication functions between the Frontend and the Backend
6 * These routines handle the low-level details of communication between
7 * frontend and backend. They just shove data across the communication
8 * channel, and are ignorant of the semantics of the data.
10 * To emit an outgoing message, use the routines in pqformat.c to construct
11 * the message in a buffer and then emit it in one call to pq_putmessage.
12 * There are no functions to send raw bytes or partial messages; this
13 * ensures that the channel will not be clogged by an incomplete message if
14 * execution is aborted by ereport(ERROR) partway through the message.
16 * At one time, libpq was shared between frontend and backend, but now
17 * the backend's "backend/libpq" is quite separate from "interfaces/libpq".
18 * All that remains is similarities of names to trap the unwary...
20 * Portions Copyright (c) 1996-2022, PostgreSQL Global Development Group
21 * Portions Copyright (c) 1994, Regents of the University of California
23 * src/backend/libpq/pqcomm.c
25 *-------------------------------------------------------------------------
28 /*------------------------
32 * StreamServerPort - Open postmaster's server port
33 * StreamConnection - Create new connection with client
34 * StreamClose - Close a client/backend connection
35 * TouchSocketFiles - Protect socket files against /tmp cleaners
36 * pq_init - initialize libpq at backend startup
37 * socket_comm_reset - reset libpq during error recovery
38 * socket_close - shutdown libpq at backend exit
41 * pq_getbytes - get a known number of bytes from connection
42 * pq_getmessage - get a message with length word from connection
43 * pq_getbyte - get next byte from connection
44 * pq_peekbyte - peek at next byte from connection
45 * pq_flush - flush pending output
46 * pq_flush_if_writable - flush pending output if writable without blocking
47 * pq_getbyte_if_available - get a byte if available without blocking
50 * pq_putmessage - send a normal message (suppressed in COPY OUT mode)
51 * pq_putmessage_noblock - buffer a normal message (suppressed in COPY OUT)
53 *------------------------
65 #include <sys/socket.h>
69 #include <netinet/in.h>
70 #ifdef HAVE_NETINET_TCP_H
71 #include <netinet/tcp.h>
74 #ifdef _MSC_VER /* mstcpip.h is missing on mingw */
78 #include "common/ip.h"
79 #include "libpq/libpq.h"
80 #include "miscadmin.h"
81 #include "port/pg_bswap.h"
82 #include "storage/ipc.h"
83 #include "utils/guc.h"
84 #include "utils/memutils.h"
87 * Cope with the various platform-specific ways to spell TCP keepalive socket
88 * options. This doesn't cover Windows, which as usual does its own thing.
90 #if defined(TCP_KEEPIDLE)
91 /* TCP_KEEPIDLE is the name of this option on Linux and *BSD */
92 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPIDLE
93 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPIDLE"
94 #elif defined(TCP_KEEPALIVE_THRESHOLD)
95 /* TCP_KEEPALIVE_THRESHOLD is the name of this option on Solaris >= 11 */
96 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE_THRESHOLD
97 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE_THRESHOLD"
98 #elif defined(TCP_KEEPALIVE) && defined(__darwin__)
99 /* TCP_KEEPALIVE is the name of this option on macOS */
100 /* Caution: Solaris has this symbol but it means something different */
101 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE
102 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE"
106 * Configuration options
108 int Unix_socket_permissions
;
109 char *Unix_socket_group
;
111 /* Where the Unix socket files are (list of palloc'd strings) */
112 static List
*sock_paths
= NIL
;
115 * Buffers for low-level I/O.
117 * The receive buffer is fixed size. Send buffer is usually 8k, but can be
118 * enlarged by pq_putmessage_noblock() if the message doesn't fit otherwise.
121 #define PQ_SEND_BUFFER_SIZE 8192
122 #define PQ_RECV_BUFFER_SIZE 8192
124 static char *PqSendBuffer
;
125 static int PqSendBufferSize
; /* Size send buffer */
126 static int PqSendPointer
; /* Next index to store a byte in PqSendBuffer */
127 static int PqSendStart
; /* Next index to send a byte in PqSendBuffer */
129 static char PqRecvBuffer
[PQ_RECV_BUFFER_SIZE
];
130 static int PqRecvPointer
; /* Next index to read a byte from PqRecvBuffer */
131 static int PqRecvLength
; /* End of data available in PqRecvBuffer */
136 static bool PqCommBusy
; /* busy sending data to the client */
137 static bool PqCommReadingMsg
; /* in the middle of reading a message */
140 /* Internal functions */
141 static void socket_comm_reset(void);
142 static void socket_close(int code
, Datum arg
);
143 static void socket_set_nonblocking(bool nonblocking
);
144 static int socket_flush(void);
145 static int socket_flush_if_writable(void);
146 static bool socket_is_send_pending(void);
147 static int socket_putmessage(char msgtype
, const char *s
, size_t len
);
148 static void socket_putmessage_noblock(char msgtype
, const char *s
, size_t len
);
149 static int internal_putbytes(const char *s
, size_t len
);
150 static int internal_flush(void);
152 #ifdef HAVE_UNIX_SOCKETS
153 static int Lock_AF_UNIX(const char *unixSocketDir
, const char *unixSocketPath
);
154 static int Setup_AF_UNIX(const char *sock_path
);
155 #endif /* HAVE_UNIX_SOCKETS */
157 static const PQcommMethods PqCommSocketMethods
= {
160 socket_flush_if_writable
,
161 socket_is_send_pending
,
163 socket_putmessage_noblock
166 const PQcommMethods
*PqCommMethods
= &PqCommSocketMethods
;
168 WaitEventSet
*FeBeWaitSet
;
171 /* --------------------------------
172 * pq_init - initialize libpq at backend startup
173 * --------------------------------
178 int socket_pos PG_USED_FOR_ASSERTS_ONLY
;
179 int latch_pos PG_USED_FOR_ASSERTS_ONLY
;
181 /* initialize state variables */
182 PqSendBufferSize
= PQ_SEND_BUFFER_SIZE
;
183 PqSendBuffer
= MemoryContextAlloc(TopMemoryContext
, PqSendBufferSize
);
184 PqSendPointer
= PqSendStart
= PqRecvPointer
= PqRecvLength
= 0;
186 PqCommReadingMsg
= false;
188 /* set up process-exit hook to close the socket */
189 on_proc_exit(socket_close
, 0);
192 * In backends (as soon as forked) we operate the underlying socket in
193 * nonblocking mode and use latches to implement blocking semantics if
194 * needed. That allows us to provide safely interruptible reads and
197 * Use COMMERROR on failure, because ERROR would try to send the error to
198 * the client, which might require changing the mode again, leading to
199 * infinite recursion.
202 if (!pg_set_noblock(MyProcPort
->sock
))
204 (errmsg("could not set socket to nonblocking mode: %m")));
207 FeBeWaitSet
= CreateWaitEventSet(TopMemoryContext
, 3);
208 socket_pos
= AddWaitEventToSet(FeBeWaitSet
, WL_SOCKET_WRITEABLE
,
209 MyProcPort
->sock
, NULL
, NULL
);
210 latch_pos
= AddWaitEventToSet(FeBeWaitSet
, WL_LATCH_SET
, PGINVALID_SOCKET
,
212 AddWaitEventToSet(FeBeWaitSet
, WL_POSTMASTER_DEATH
, PGINVALID_SOCKET
,
216 * The event positions match the order we added them, but let's sanity
217 * check them to be sure.
219 Assert(socket_pos
== FeBeWaitSetSocketPos
);
220 Assert(latch_pos
== FeBeWaitSetLatchPos
);
223 /* --------------------------------
224 * socket_comm_reset - reset libpq during error recovery
226 * This is called from error recovery at the outer idle loop. It's
227 * just to get us out of trouble if we somehow manage to elog() from
228 * inside a pqcomm.c routine (which ideally will never happen, but...)
229 * --------------------------------
232 socket_comm_reset(void)
234 /* Do not throw away pending data, but do reset the busy flag */
238 /* --------------------------------
239 * socket_close - shutdown libpq at backend exit
241 * This is the one pg_on_exit_callback in place during BackendInitialize().
242 * That function's unusual signal handling constrains that this callback be
243 * safe to run at any instant.
244 * --------------------------------
247 socket_close(int code
, Datum arg
)
249 /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
250 if (MyProcPort
!= NULL
)
254 * Shutdown GSSAPI layer. This section does nothing when interrupting
255 * BackendInitialize(), because pg_GSS_recvauth() makes first use of
258 * Note that we don't bother to free MyProcPort->gss, since we're
259 * about to exit anyway.
265 if (MyProcPort
->gss
->ctx
!= GSS_C_NO_CONTEXT
)
266 gss_delete_sec_context(&min_s
, &MyProcPort
->gss
->ctx
, NULL
);
268 if (MyProcPort
->gss
->cred
!= GSS_C_NO_CREDENTIAL
)
269 gss_release_cred(&min_s
, &MyProcPort
->gss
->cred
);
271 #endif /* ENABLE_GSS */
274 * Cleanly shut down SSL layer. Nowhere else does a postmaster child
275 * call this, so this is safe when interrupting BackendInitialize().
277 secure_close(MyProcPort
);
280 * On most platforms, we leave the socket open until the process dies.
281 * This allows clients to perform a "synchronous close" if they care
282 * --- wait till the transport layer reports connection closure, and
283 * you can be sure the backend has exited. Saves a kernel call, too.
285 * However, that does not work on Windows: if the kernel closes the
286 * socket it will invoke an "abortive shutdown" that discards any data
287 * not yet sent to the client. (This is a flat-out violation of the
288 * TCP RFCs, but count on Microsoft not to care about that.) To get
289 * the spec-compliant "graceful shutdown" behavior, we must invoke
290 * closesocket() explicitly. When using OpenSSL, it seems that clean
291 * shutdown also requires an explicit shutdown() call.
293 * This code runs late enough during process shutdown that we should
294 * have finished all externally-visible shutdown activities, so that
295 * in principle it's good enough to act as a synchronous close on
296 * Windows too. But it's a lot more fragile than the other way.
299 shutdown(MyProcPort
->sock
, SD_SEND
);
300 closesocket(MyProcPort
->sock
);
303 /* In any case, set sock to PGINVALID_SOCKET to prevent further I/O */
304 MyProcPort
->sock
= PGINVALID_SOCKET
;
311 * Streams -- wrapper around Unix socket system calls
314 * Stream functions are used for vanilla TCP connection protocol.
319 * StreamServerPort -- open a "listening" port to accept connections.
321 * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
322 * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
323 * specified. For TCP ports, hostName is either NULL for all interfaces or
324 * the interface to listen on, and unixSocketDir is ignored (can be NULL).
326 * Successfully opened sockets are added to the ListenSocket[] array (of
327 * length MaxListen), at the first position that isn't PGINVALID_SOCKET.
329 * RETURNS: STATUS_OK or STATUS_ERROR
333 StreamServerPort(int family
, const char *hostName
, unsigned short portNumber
,
334 const char *unixSocketDir
,
335 pgsocket ListenSocket
[], int MaxListen
)
341 char portNumberStr
[32];
342 const char *familyDesc
;
343 char familyDescBuf
[64];
344 const char *addrDesc
;
345 char addrBuf
[NI_MAXHOST
];
347 struct addrinfo
*addrs
= NULL
,
349 struct addrinfo hint
;
350 int listen_index
= 0;
353 #ifdef HAVE_UNIX_SOCKETS
354 char unixSocketPath
[MAXPGPATH
];
356 #if !defined(WIN32) || defined(IPV6_V6ONLY)
360 /* Initialize hint structure */
361 MemSet(&hint
, 0, sizeof(hint
));
362 hint
.ai_family
= family
;
363 hint
.ai_flags
= AI_PASSIVE
;
364 hint
.ai_socktype
= SOCK_STREAM
;
366 #ifdef HAVE_UNIX_SOCKETS
367 if (family
== AF_UNIX
)
370 * Create unixSocketPath from portNumber and unixSocketDir and lock
373 UNIXSOCK_PATH(unixSocketPath
, portNumber
, unixSocketDir
);
374 if (strlen(unixSocketPath
) >= UNIXSOCK_PATH_BUFLEN
)
377 (errmsg("Unix-domain socket path \"%s\" is too long (maximum %d bytes)",
379 (int) (UNIXSOCK_PATH_BUFLEN
- 1))));
382 if (Lock_AF_UNIX(unixSocketDir
, unixSocketPath
) != STATUS_OK
)
384 service
= unixSocketPath
;
387 #endif /* HAVE_UNIX_SOCKETS */
389 snprintf(portNumberStr
, sizeof(portNumberStr
), "%d", portNumber
);
390 service
= portNumberStr
;
393 ret
= pg_getaddrinfo_all(hostName
, service
, &hint
, &addrs
);
398 (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
399 hostName
, service
, gai_strerror(ret
))));
402 (errmsg("could not translate service \"%s\" to address: %s",
403 service
, gai_strerror(ret
))));
405 pg_freeaddrinfo_all(hint
.ai_family
, addrs
);
409 for (addr
= addrs
; addr
; addr
= addr
->ai_next
)
411 if (!IS_AF_UNIX(family
) && IS_AF_UNIX(addr
->ai_family
))
414 * Only set up a unix domain socket when they really asked for it.
415 * The service/port is different in that case.
420 /* See if there is still room to add 1 more socket. */
421 for (; listen_index
< MaxListen
; listen_index
++)
423 if (ListenSocket
[listen_index
] == PGINVALID_SOCKET
)
426 if (listen_index
>= MaxListen
)
429 (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
434 /* set up address family name for log messages */
435 switch (addr
->ai_family
)
438 familyDesc
= _("IPv4");
442 familyDesc
= _("IPv6");
445 #ifdef HAVE_UNIX_SOCKETS
447 familyDesc
= _("Unix");
451 snprintf(familyDescBuf
, sizeof(familyDescBuf
),
452 _("unrecognized address family %d"),
454 familyDesc
= familyDescBuf
;
458 /* set up text form of address for log messages */
459 #ifdef HAVE_UNIX_SOCKETS
460 if (addr
->ai_family
== AF_UNIX
)
461 addrDesc
= unixSocketPath
;
465 pg_getnameinfo_all((const struct sockaddr_storage
*) addr
->ai_addr
,
467 addrBuf
, sizeof(addrBuf
),
473 if ((fd
= socket(addr
->ai_family
, SOCK_STREAM
, 0)) == PGINVALID_SOCKET
)
476 (errcode_for_socket_access(),
477 /* translator: first %s is IPv4, IPv6, or Unix */
478 errmsg("could not create %s socket for address \"%s\": %m",
479 familyDesc
, addrDesc
)));
486 * Without the SO_REUSEADDR flag, a new postmaster can't be started
487 * right away after a stop or crash, giving "address already in use"
488 * error on TCP ports.
490 * On win32, however, this behavior only happens if the
491 * SO_EXCLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows
492 * multiple servers to listen on the same address, resulting in
493 * unpredictable behavior. With no flags at all, win32 behaves as Unix
496 if (!IS_AF_UNIX(addr
->ai_family
))
498 if ((setsockopt(fd
, SOL_SOCKET
, SO_REUSEADDR
,
499 (char *) &one
, sizeof(one
))) == -1)
502 (errcode_for_socket_access(),
503 /* translator: third %s is IPv4, IPv6, or Unix */
504 errmsg("%s(%s) failed for %s address \"%s\": %m",
505 "setsockopt", "SO_REUSEADDR",
506 familyDesc
, addrDesc
)));
514 if (addr
->ai_family
== AF_INET6
)
516 if (setsockopt(fd
, IPPROTO_IPV6
, IPV6_V6ONLY
,
517 (char *) &one
, sizeof(one
)) == -1)
520 (errcode_for_socket_access(),
521 /* translator: third %s is IPv4, IPv6, or Unix */
522 errmsg("%s(%s) failed for %s address \"%s\": %m",
523 "setsockopt", "IPV6_V6ONLY",
524 familyDesc
, addrDesc
)));
532 * Note: This might fail on some OS's, like Linux older than
533 * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
534 * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
537 err
= bind(fd
, addr
->ai_addr
, addr
->ai_addrlen
);
540 int saved_errno
= errno
;
543 (errcode_for_socket_access(),
544 /* translator: first %s is IPv4, IPv6, or Unix */
545 errmsg("could not bind %s address \"%s\": %m",
546 familyDesc
, addrDesc
),
547 saved_errno
== EADDRINUSE
?
548 (IS_AF_UNIX(addr
->ai_family
) ?
549 errhint("Is another postmaster already running on port %d?",
551 errhint("Is another postmaster already running on port %d?"
552 " If not, wait a few seconds and retry.",
553 (int) portNumber
)) : 0));
558 #ifdef HAVE_UNIX_SOCKETS
559 if (addr
->ai_family
== AF_UNIX
)
561 if (Setup_AF_UNIX(service
) != STATUS_OK
)
570 * Select appropriate accept-queue length limit. PG_SOMAXCONN is only
571 * intended to provide a clamp on the request on platforms where an
572 * overly large request provokes a kernel error (are there any?).
574 maxconn
= MaxBackends
* 2;
575 if (maxconn
> PG_SOMAXCONN
)
576 maxconn
= PG_SOMAXCONN
;
578 err
= listen(fd
, maxconn
);
582 (errcode_for_socket_access(),
583 /* translator: first %s is IPv4, IPv6, or Unix */
584 errmsg("could not listen on %s address \"%s\": %m",
585 familyDesc
, addrDesc
)));
590 #ifdef HAVE_UNIX_SOCKETS
591 if (addr
->ai_family
== AF_UNIX
)
593 (errmsg("listening on Unix socket \"%s\"",
598 /* translator: first %s is IPv4 or IPv6 */
599 (errmsg("listening on %s address \"%s\", port %d",
600 familyDesc
, addrDesc
, (int) portNumber
)));
602 ListenSocket
[listen_index
] = fd
;
606 pg_freeaddrinfo_all(hint
.ai_family
, addrs
);
615 #ifdef HAVE_UNIX_SOCKETS
618 * Lock_AF_UNIX -- configure unix socket file path
621 Lock_AF_UNIX(const char *unixSocketDir
, const char *unixSocketPath
)
623 /* no lock file for abstract sockets */
624 if (unixSocketPath
[0] == '@')
628 * Grab an interlock file associated with the socket file.
630 * Note: there are two reasons for using a socket lock file, rather than
631 * trying to interlock directly on the socket itself. First, it's a lot
632 * more portable, and second, it lets us remove any pre-existing socket
633 * file without race conditions.
635 CreateSocketLockFile(unixSocketPath
, true, unixSocketDir
);
638 * Once we have the interlock, we can safely delete any pre-existing
639 * socket file to avoid failure at bind() time.
641 (void) unlink(unixSocketPath
);
644 * Remember socket file pathnames for later maintenance.
646 sock_paths
= lappend(sock_paths
, pstrdup(unixSocketPath
));
653 * Setup_AF_UNIX -- configure unix socket permissions
656 Setup_AF_UNIX(const char *sock_path
)
658 /* no file system permissions for abstract sockets */
659 if (sock_path
[0] == '@')
663 * Fix socket ownership/permission if requested. Note we must do this
664 * before we listen() to avoid a window where unwanted connections could
667 Assert(Unix_socket_group
);
668 if (Unix_socket_group
[0] != '\0')
671 elog(WARNING
, "configuration item unix_socket_group is not supported on this platform");
677 val
= strtoul(Unix_socket_group
, &endptr
, 10);
679 { /* numeric group id */
683 { /* convert group name to id */
686 gr
= getgrnam(Unix_socket_group
);
690 (errmsg("group \"%s\" does not exist",
691 Unix_socket_group
)));
696 if (chown(sock_path
, -1, gid
) == -1)
699 (errcode_for_file_access(),
700 errmsg("could not set group of file \"%s\": %m",
707 if (chmod(sock_path
, Unix_socket_permissions
) == -1)
710 (errcode_for_file_access(),
711 errmsg("could not set permissions of file \"%s\": %m",
717 #endif /* HAVE_UNIX_SOCKETS */
721 * StreamConnection -- create a new connection with client using
722 * server port. Set port->sock to the FD of the new connection.
724 * ASSUME: that this doesn't need to be non-blocking because
725 * the Postmaster uses select() to tell when the socket is ready for
728 * RETURNS: STATUS_OK or STATUS_ERROR
731 StreamConnection(pgsocket server_fd
, Port
*port
)
733 /* accept connection and fill in the client (remote) address */
734 port
->raddr
.salen
= sizeof(port
->raddr
.addr
);
735 if ((port
->sock
= accept(server_fd
,
736 (struct sockaddr
*) &port
->raddr
.addr
,
737 &port
->raddr
.salen
)) == PGINVALID_SOCKET
)
740 (errcode_for_socket_access(),
741 errmsg("could not accept new connection: %m")));
744 * If accept() fails then postmaster.c will still see the server
745 * socket as read-ready, and will immediately try again. To avoid
746 * uselessly sucking lots of CPU, delay a bit before trying again.
747 * (The most likely reason for failure is being out of kernel file
748 * table slots; we can do little except hope some will get freed up.)
750 pg_usleep(100000L); /* wait 0.1 sec */
754 /* fill in the server (local) address */
755 port
->laddr
.salen
= sizeof(port
->laddr
.addr
);
756 if (getsockname(port
->sock
,
757 (struct sockaddr
*) &port
->laddr
.addr
,
758 &port
->laddr
.salen
) < 0)
761 (errmsg("%s() failed: %m", "getsockname")));
765 /* select NODELAY and KEEPALIVE options if it's a TCP connection */
766 if (!IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
777 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_NODELAY
,
778 (char *) &on
, sizeof(on
)) < 0)
781 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_NODELAY")));
786 if (setsockopt(port
->sock
, SOL_SOCKET
, SO_KEEPALIVE
,
787 (char *) &on
, sizeof(on
)) < 0)
790 (errmsg("%s(%s) failed: %m", "setsockopt", "SO_KEEPALIVE")));
797 * This is a Win32 socket optimization. The OS send buffer should be
798 * large enough to send the whole Postgres send buffer in one go, or
799 * performance suffers. The Postgres send buffer can be enlarged if a
800 * very large message needs to be sent, but we won't attempt to
801 * enlarge the OS buffer if that happens, so somewhat arbitrarily
802 * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
803 * (That's 32kB with the current default).
805 * The default OS buffer size used to be 8kB in earlier Windows
806 * versions, but was raised to 64kB in Windows 2012. So it shouldn't
807 * be necessary to change it in later versions anymore. Changing it
808 * unnecessarily can even reduce performance, because setting
809 * SO_SNDBUF in the application disables the "dynamic send buffering"
810 * feature that was introduced in Windows 7. So before fiddling with
811 * SO_SNDBUF, check if the current buffer size is already large enough
812 * and only increase it if necessary.
814 * See https://support.microsoft.com/kb/823764/EN-US/ and
815 * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
817 optlen
= sizeof(oldopt
);
818 if (getsockopt(port
->sock
, SOL_SOCKET
, SO_SNDBUF
, (char *) &oldopt
,
822 (errmsg("%s(%s) failed: %m", "getsockopt", "SO_SNDBUF")));
825 newopt
= PQ_SEND_BUFFER_SIZE
* 4;
828 if (setsockopt(port
->sock
, SOL_SOCKET
, SO_SNDBUF
, (char *) &newopt
,
832 (errmsg("%s(%s) failed: %m", "setsockopt", "SO_SNDBUF")));
839 * Also apply the current keepalive parameters. If we fail to set a
840 * parameter, don't error out, because these aren't universally
841 * supported. (Note: you might think we need to reset the GUC
842 * variables to 0 in such a case, but it's not necessary because the
843 * show hooks for these variables report the truth anyway.)
845 (void) pq_setkeepalivesidle(tcp_keepalives_idle
, port
);
846 (void) pq_setkeepalivesinterval(tcp_keepalives_interval
, port
);
847 (void) pq_setkeepalivescount(tcp_keepalives_count
, port
);
848 (void) pq_settcpusertimeout(tcp_user_timeout
, port
);
855 * StreamClose -- close a client/backend connection
857 * NOTE: this is NOT used to terminate a session; it is just used to release
858 * the file descriptor in a process that should no longer have the socket
859 * open. (For example, the postmaster calls this after passing ownership
860 * of the connection to a child process.) It is expected that someone else
861 * still has the socket open. So, we only want to close the descriptor,
862 * we do NOT want to send anything to the far end.
865 StreamClose(pgsocket sock
)
871 * TouchSocketFiles -- mark socket files as recently accessed
873 * This routine should be called every so often to ensure that the socket
874 * files have a recent mod date (ordinary operations on sockets usually won't
875 * change the mod date). That saves them from being removed by
876 * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
877 * never have put the socket file in /tmp...)
880 TouchSocketFiles(void)
884 /* Loop through all created sockets... */
885 foreach(l
, sock_paths
)
887 char *sock_path
= (char *) lfirst(l
);
889 /* Ignore errors; there's no point in complaining */
890 (void) utime(sock_path
, NULL
);
895 * RemoveSocketFiles -- unlink socket files at postmaster shutdown
898 RemoveSocketFiles(void)
902 /* Loop through all created sockets... */
903 foreach(l
, sock_paths
)
905 char *sock_path
= (char *) lfirst(l
);
907 /* Ignore any error. */
908 (void) unlink(sock_path
);
910 /* Since we're about to exit, no need to reclaim storage */
915 /* --------------------------------
916 * Low-level I/O routines begin here.
918 * These routines communicate with a frontend client across a connection
919 * already established by the preceding routines.
920 * --------------------------------
923 /* --------------------------------
924 * socket_set_nonblocking - set socket blocking/non-blocking
926 * Sets the socket non-blocking if nonblocking is true, or sets it
927 * blocking otherwise.
928 * --------------------------------
931 socket_set_nonblocking(bool nonblocking
)
933 if (MyProcPort
== NULL
)
935 (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST
),
936 errmsg("there is no client connection")));
938 MyProcPort
->noblock
= nonblocking
;
941 /* --------------------------------
942 * pq_recvbuf - load some bytes into the input buffer
944 * returns 0 if OK, EOF if trouble
945 * --------------------------------
950 if (PqRecvPointer
> 0)
952 if (PqRecvLength
> PqRecvPointer
)
954 /* still some unread data, left-justify it in the buffer */
955 memmove(PqRecvBuffer
, PqRecvBuffer
+ PqRecvPointer
,
956 PqRecvLength
- PqRecvPointer
);
957 PqRecvLength
-= PqRecvPointer
;
961 PqRecvLength
= PqRecvPointer
= 0;
964 /* Ensure that we're in blocking mode */
965 socket_set_nonblocking(false);
967 /* Can fill buffer from PqRecvLength and upwards */
972 r
= secure_read(MyProcPort
, PqRecvBuffer
+ PqRecvLength
,
973 PQ_RECV_BUFFER_SIZE
- PqRecvLength
);
978 continue; /* Ok if interrupted */
981 * Careful: an ereport() that tries to write to the client would
982 * cause recursion to here, leading to stack overflow and core
983 * dump! This message must go *only* to the postmaster log.
986 (errcode_for_socket_access(),
987 errmsg("could not receive data from client: %m")));
993 * EOF detected. We used to write a log message here, but it's
994 * better to expect the ultimate caller to do that.
998 /* r contains number of bytes read, so just incr length */
1004 /* --------------------------------
1005 * pq_getbyte - get a single byte from connection, or return EOF
1006 * --------------------------------
1011 Assert(PqCommReadingMsg
);
1013 while (PqRecvPointer
>= PqRecvLength
)
1015 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1016 return EOF
; /* Failed to recv data */
1018 return (unsigned char) PqRecvBuffer
[PqRecvPointer
++];
1021 /* --------------------------------
1022 * pq_peekbyte - peek at next byte from connection
1024 * Same as pq_getbyte() except we don't advance the pointer.
1025 * --------------------------------
1030 Assert(PqCommReadingMsg
);
1032 while (PqRecvPointer
>= PqRecvLength
)
1034 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1035 return EOF
; /* Failed to recv data */
1037 return (unsigned char) PqRecvBuffer
[PqRecvPointer
];
1040 /* --------------------------------
1041 * pq_getbyte_if_available - get a single byte from connection,
1044 * The received byte is stored in *c. Returns 1 if a byte was read,
1045 * 0 if no data was available, or EOF if trouble.
1046 * --------------------------------
1049 pq_getbyte_if_available(unsigned char *c
)
1053 Assert(PqCommReadingMsg
);
1055 if (PqRecvPointer
< PqRecvLength
)
1057 *c
= PqRecvBuffer
[PqRecvPointer
++];
1061 /* Put the socket into non-blocking mode */
1062 socket_set_nonblocking(true);
1064 r
= secure_read(MyProcPort
, c
, 1);
1068 * Ok if no data available without blocking or interrupted (though
1069 * EINTR really shouldn't happen with a non-blocking socket). Report
1072 if (errno
== EAGAIN
|| errno
== EWOULDBLOCK
|| errno
== EINTR
)
1077 * Careful: an ereport() that tries to write to the client would
1078 * cause recursion to here, leading to stack overflow and core
1079 * dump! This message must go *only* to the postmaster log.
1082 (errcode_for_socket_access(),
1083 errmsg("could not receive data from client: %m")));
1096 /* --------------------------------
1097 * pq_getbytes - get a known number of bytes from connection
1099 * returns 0 if OK, EOF if trouble
1100 * --------------------------------
1103 pq_getbytes(char *s
, size_t len
)
1107 Assert(PqCommReadingMsg
);
1111 while (PqRecvPointer
>= PqRecvLength
)
1113 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1114 return EOF
; /* Failed to recv data */
1116 amount
= PqRecvLength
- PqRecvPointer
;
1119 memcpy(s
, PqRecvBuffer
+ PqRecvPointer
, amount
);
1120 PqRecvPointer
+= amount
;
1127 /* --------------------------------
1128 * pq_discardbytes - throw away a known number of bytes
1130 * same as pq_getbytes except we do not copy the data to anyplace.
1131 * this is used for resynchronizing after read errors.
1133 * returns 0 if OK, EOF if trouble
1134 * --------------------------------
1137 pq_discardbytes(size_t len
)
1141 Assert(PqCommReadingMsg
);
1145 while (PqRecvPointer
>= PqRecvLength
)
1147 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1148 return EOF
; /* Failed to recv data */
1150 amount
= PqRecvLength
- PqRecvPointer
;
1153 PqRecvPointer
+= amount
;
1159 /* --------------------------------
1160 * pq_buffer_has_data - is any buffered data available to read?
1162 * This will *not* attempt to read more data.
1163 * --------------------------------
1166 pq_buffer_has_data(void)
1168 return (PqRecvPointer
< PqRecvLength
);
1172 /* --------------------------------
1173 * pq_startmsgread - begin reading a message from the client.
1175 * This must be called before any of the pq_get* functions.
1176 * --------------------------------
1179 pq_startmsgread(void)
1182 * There shouldn't be a read active already, but let's check just to be
1185 if (PqCommReadingMsg
)
1187 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1188 errmsg("terminating connection because protocol synchronization was lost")));
1190 PqCommReadingMsg
= true;
1194 /* --------------------------------
1195 * pq_endmsgread - finish reading message.
1197 * This must be called after reading a message with pq_getbytes()
1198 * and friends, to indicate that we have read the whole message.
1199 * pq_getmessage() does this implicitly.
1200 * --------------------------------
1205 Assert(PqCommReadingMsg
);
1207 PqCommReadingMsg
= false;
1210 /* --------------------------------
1211 * pq_is_reading_msg - are we currently reading a message?
1213 * This is used in error recovery at the outer idle loop to detect if we have
1214 * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1215 * will check for that too, but it's nicer to detect it earlier.
1216 * --------------------------------
1219 pq_is_reading_msg(void)
1221 return PqCommReadingMsg
;
1224 /* --------------------------------
1225 * pq_getmessage - get a message with length word from connection
1227 * The return value is placed in an expansible StringInfo, which has
1228 * already been initialized by the caller.
1229 * Only the message body is placed in the StringInfo; the length word
1230 * is removed. Also, s->cursor is initialized to zero for convenience
1231 * in scanning the message contents.
1233 * maxlen is the upper limit on the length of the
1234 * message we are willing to accept. We abort the connection (by
1235 * returning EOF) if client tries to send more than that.
1237 * returns 0 if OK, EOF if trouble
1238 * --------------------------------
1241 pq_getmessage(StringInfo s
, int maxlen
)
1245 Assert(PqCommReadingMsg
);
1249 /* Read message length word */
1250 if (pq_getbytes((char *) &len
, 4) == EOF
)
1253 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1254 errmsg("unexpected EOF within message length word")));
1258 len
= pg_ntoh32(len
);
1260 if (len
< 4 || len
> maxlen
)
1263 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1264 errmsg("invalid message length")));
1268 len
-= 4; /* discount length itself */
1273 * Allocate space for message. If we run out of room (ridiculously
1274 * large message), we will elog(ERROR), but we want to discard the
1275 * message body so as not to lose communication sync.
1279 enlargeStringInfo(s
, len
);
1283 if (pq_discardbytes(len
) == EOF
)
1285 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1286 errmsg("incomplete message from client")));
1288 /* we discarded the rest of the message so we're back in sync. */
1289 PqCommReadingMsg
= false;
1294 /* And grab the message */
1295 if (pq_getbytes(s
->data
, len
) == EOF
)
1298 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1299 errmsg("incomplete message from client")));
1303 /* Place a trailing null per StringInfo convention */
1304 s
->data
[len
] = '\0';
1307 /* finished reading the message. */
1308 PqCommReadingMsg
= false;
1315 internal_putbytes(const char *s
, size_t len
)
1321 /* If buffer is full, then flush it out */
1322 if (PqSendPointer
>= PqSendBufferSize
)
1324 socket_set_nonblocking(false);
1325 if (internal_flush())
1328 amount
= PqSendBufferSize
- PqSendPointer
;
1331 memcpy(PqSendBuffer
+ PqSendPointer
, s
, amount
);
1332 PqSendPointer
+= amount
;
1339 /* --------------------------------
1340 * socket_flush - flush pending output
1342 * returns 0 if OK, EOF if trouble
1343 * --------------------------------
1350 /* No-op if reentrant call */
1354 socket_set_nonblocking(false);
1355 res
= internal_flush();
1360 /* --------------------------------
1361 * internal_flush - flush pending output
1363 * Returns 0 if OK (meaning everything was sent, or operation would block
1364 * and the socket is in non-blocking mode), or EOF if trouble.
1365 * --------------------------------
1368 internal_flush(void)
1370 static int last_reported_send_errno
= 0;
1372 char *bufptr
= PqSendBuffer
+ PqSendStart
;
1373 char *bufend
= PqSendBuffer
+ PqSendPointer
;
1375 while (bufptr
< bufend
)
1379 r
= secure_write(MyProcPort
, bufptr
, bufend
- bufptr
);
1384 continue; /* Ok if we were interrupted */
1387 * Ok if no data writable without blocking, and the socket is in
1388 * non-blocking mode.
1390 if (errno
== EAGAIN
||
1391 errno
== EWOULDBLOCK
)
1397 * Careful: an ereport() that tries to write to the client would
1398 * cause recursion to here, leading to stack overflow and core
1399 * dump! This message must go *only* to the postmaster log.
1401 * If a client disconnects while we're in the midst of output, we
1402 * might write quite a bit of data before we get to a safe query
1403 * abort point. So, suppress duplicate log messages.
1405 if (errno
!= last_reported_send_errno
)
1407 last_reported_send_errno
= errno
;
1409 (errcode_for_socket_access(),
1410 errmsg("could not send data to client: %m")));
1414 * We drop the buffered data anyway so that processing can
1415 * continue, even though we'll probably quit soon. We also set a
1416 * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1419 PqSendStart
= PqSendPointer
= 0;
1420 ClientConnectionLost
= 1;
1421 InterruptPending
= 1;
1425 last_reported_send_errno
= 0; /* reset after any successful send */
1430 PqSendStart
= PqSendPointer
= 0;
1434 /* --------------------------------
1435 * pq_flush_if_writable - flush pending output if writable without blocking
1437 * Returns 0 if OK, or EOF if trouble.
1438 * --------------------------------
1441 socket_flush_if_writable(void)
1445 /* Quick exit if nothing to do */
1446 if (PqSendPointer
== PqSendStart
)
1449 /* No-op if reentrant call */
1453 /* Temporarily put the socket into non-blocking mode */
1454 socket_set_nonblocking(true);
1457 res
= internal_flush();
1462 /* --------------------------------
1463 * socket_is_send_pending - is there any pending data in the output buffer?
1464 * --------------------------------
1467 socket_is_send_pending(void)
1469 return (PqSendStart
< PqSendPointer
);
1472 /* --------------------------------
1473 * Message-level I/O routines begin here.
1474 * --------------------------------
1478 /* --------------------------------
1479 * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1481 * msgtype is a message type code to place before the message body.
1483 * len is the length of the message body data at *s. A message length
1484 * word (equal to len+4 because it counts itself too) is inserted by this
1487 * We suppress messages generated while pqcomm.c is busy. This
1488 * avoids any possibility of messages being inserted within other
1489 * messages. The only known trouble case arises if SIGQUIT occurs
1490 * during a pqcomm.c routine --- quickdie() will try to send a warning
1491 * message, and the most reasonable approach seems to be to drop it.
1493 * returns 0 if OK, EOF if trouble
1494 * --------------------------------
1497 socket_putmessage(char msgtype
, const char *s
, size_t len
)
1501 Assert(msgtype
!= 0);
1506 if (internal_putbytes(&msgtype
, 1))
1509 n32
= pg_hton32((uint32
) (len
+ 4));
1510 if (internal_putbytes((char *) &n32
, 4))
1513 if (internal_putbytes(s
, len
))
1523 /* --------------------------------
1524 * pq_putmessage_noblock - like pq_putmessage, but never blocks
1526 * If the output buffer is too small to hold the message, the buffer
1530 socket_putmessage_noblock(char msgtype
, const char *s
, size_t len
)
1532 int res PG_USED_FOR_ASSERTS_ONLY
;
1536 * Ensure we have enough space in the output buffer for the message header
1537 * as well as the message itself.
1539 required
= PqSendPointer
+ 1 + 4 + len
;
1540 if (required
> PqSendBufferSize
)
1542 PqSendBuffer
= repalloc(PqSendBuffer
, required
);
1543 PqSendBufferSize
= required
;
1545 res
= pq_putmessage(msgtype
, s
, len
);
1546 Assert(res
== 0); /* should not fail when the message fits in
1550 /* --------------------------------
1551 * pq_putmessage_v2 - send a message in protocol version 2
1553 * msgtype is a message type code to place before the message body.
1555 * We no longer support protocol version 2, but we have kept this
1556 * function so that if a client tries to connect with protocol version 2,
1557 * as a courtesy we can still send the "unsupported protocol version"
1558 * error to the client in the old format.
1560 * Like in pq_putmessage(), we suppress messages generated while
1563 * returns 0 if OK, EOF if trouble
1564 * --------------------------------
1567 pq_putmessage_v2(char msgtype
, const char *s
, size_t len
)
1569 Assert(msgtype
!= 0);
1574 if (internal_putbytes(&msgtype
, 1))
1577 if (internal_putbytes(s
, len
))
1588 * Support for TCP Keepalive parameters
1592 * On Windows, we need to set both idle and interval at the same time.
1593 * We also cannot reset them to the default (setting to zero will
1594 * actually set them to zero, not default), therefore we fallback to
1595 * the out-of-the-box default instead.
1597 #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1599 pq_setkeepaliveswin32(Port
*port
, int idle
, int interval
)
1601 struct tcp_keepalive ka
;
1605 idle
= 2 * 60 * 60; /* default = 2 hours */
1607 interval
= 1; /* default = 1 second */
1610 ka
.keepalivetime
= idle
* 1000;
1611 ka
.keepaliveinterval
= interval
* 1000;
1613 if (WSAIoctl(port
->sock
,
1625 (errmsg("%s(%s) failed: error code %d",
1626 "WSAIoctl", "SIO_KEEPALIVE_VALS", WSAGetLastError())));
1627 return STATUS_ERROR
;
1629 if (port
->keepalives_idle
!= idle
)
1630 port
->keepalives_idle
= idle
;
1631 if (port
->keepalives_interval
!= interval
)
1632 port
->keepalives_interval
= interval
;
1638 pq_getkeepalivesidle(Port
*port
)
1640 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1641 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1644 if (port
->keepalives_idle
!= 0)
1645 return port
->keepalives_idle
;
1647 if (port
->default_keepalives_idle
== 0)
1650 socklen_t size
= sizeof(port
->default_keepalives_idle
);
1652 if (getsockopt(port
->sock
, IPPROTO_TCP
, PG_TCP_KEEPALIVE_IDLE
,
1653 (char *) &port
->default_keepalives_idle
,
1657 (errmsg("%s(%s) failed: %m", "getsockopt", PG_TCP_KEEPALIVE_IDLE_STR
)));
1658 port
->default_keepalives_idle
= -1; /* don't know */
1661 /* We can't get the defaults on Windows, so return "don't know" */
1662 port
->default_keepalives_idle
= -1;
1666 return port
->default_keepalives_idle
;
1673 pq_setkeepalivesidle(int idle
, Port
*port
)
1675 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1678 /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1679 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1680 if (idle
== port
->keepalives_idle
)
1684 if (port
->default_keepalives_idle
<= 0)
1686 if (pq_getkeepalivesidle(port
) < 0)
1689 return STATUS_OK
; /* default is set but unknown */
1691 return STATUS_ERROR
;
1696 idle
= port
->default_keepalives_idle
;
1698 if (setsockopt(port
->sock
, IPPROTO_TCP
, PG_TCP_KEEPALIVE_IDLE
,
1699 (char *) &idle
, sizeof(idle
)) < 0)
1702 (errmsg("%s(%s) failed: %m", "setsockopt", PG_TCP_KEEPALIVE_IDLE_STR
)));
1703 return STATUS_ERROR
;
1706 port
->keepalives_idle
= idle
;
1708 return pq_setkeepaliveswin32(port
, idle
, port
->keepalives_interval
);
1714 (errmsg("setting the keepalive idle time is not supported")));
1715 return STATUS_ERROR
;
1723 pq_getkeepalivesinterval(Port
*port
)
1725 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1726 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1729 if (port
->keepalives_interval
!= 0)
1730 return port
->keepalives_interval
;
1732 if (port
->default_keepalives_interval
== 0)
1735 socklen_t size
= sizeof(port
->default_keepalives_interval
);
1737 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPINTVL
,
1738 (char *) &port
->default_keepalives_interval
,
1742 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPINTVL")));
1743 port
->default_keepalives_interval
= -1; /* don't know */
1746 /* We can't get the defaults on Windows, so return "don't know" */
1747 port
->default_keepalives_interval
= -1;
1751 return port
->default_keepalives_interval
;
1758 pq_setkeepalivesinterval(int interval
, Port
*port
)
1760 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1763 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1764 if (interval
== port
->keepalives_interval
)
1768 if (port
->default_keepalives_interval
<= 0)
1770 if (pq_getkeepalivesinterval(port
) < 0)
1773 return STATUS_OK
; /* default is set but unknown */
1775 return STATUS_ERROR
;
1780 interval
= port
->default_keepalives_interval
;
1782 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPINTVL
,
1783 (char *) &interval
, sizeof(interval
)) < 0)
1786 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPINTVL")));
1787 return STATUS_ERROR
;
1790 port
->keepalives_interval
= interval
;
1792 return pq_setkeepaliveswin32(port
, port
->keepalives_idle
, interval
);
1798 (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPINTVL")));
1799 return STATUS_ERROR
;
1807 pq_getkeepalivescount(Port
*port
)
1810 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1813 if (port
->keepalives_count
!= 0)
1814 return port
->keepalives_count
;
1816 if (port
->default_keepalives_count
== 0)
1818 socklen_t size
= sizeof(port
->default_keepalives_count
);
1820 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPCNT
,
1821 (char *) &port
->default_keepalives_count
,
1825 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPCNT")));
1826 port
->default_keepalives_count
= -1; /* don't know */
1830 return port
->default_keepalives_count
;
1837 pq_setkeepalivescount(int count
, Port
*port
)
1839 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1843 if (count
== port
->keepalives_count
)
1846 if (port
->default_keepalives_count
<= 0)
1848 if (pq_getkeepalivescount(port
) < 0)
1851 return STATUS_OK
; /* default is set but unknown */
1853 return STATUS_ERROR
;
1858 count
= port
->default_keepalives_count
;
1860 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPCNT
,
1861 (char *) &count
, sizeof(count
)) < 0)
1864 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPCNT")));
1865 return STATUS_ERROR
;
1868 port
->keepalives_count
= count
;
1873 (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPCNT")));
1874 return STATUS_ERROR
;
1882 pq_gettcpusertimeout(Port
*port
)
1884 #ifdef TCP_USER_TIMEOUT
1885 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1888 if (port
->tcp_user_timeout
!= 0)
1889 return port
->tcp_user_timeout
;
1891 if (port
->default_tcp_user_timeout
== 0)
1893 socklen_t size
= sizeof(port
->default_tcp_user_timeout
);
1895 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_USER_TIMEOUT
,
1896 (char *) &port
->default_tcp_user_timeout
,
1900 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_USER_TIMEOUT")));
1901 port
->default_tcp_user_timeout
= -1; /* don't know */
1905 return port
->default_tcp_user_timeout
;
1912 pq_settcpusertimeout(int timeout
, Port
*port
)
1914 if (port
== NULL
|| IS_AF_UNIX(port
->laddr
.addr
.ss_family
))
1917 #ifdef TCP_USER_TIMEOUT
1918 if (timeout
== port
->tcp_user_timeout
)
1921 if (port
->default_tcp_user_timeout
<= 0)
1923 if (pq_gettcpusertimeout(port
) < 0)
1926 return STATUS_OK
; /* default is set but unknown */
1928 return STATUS_ERROR
;
1933 timeout
= port
->default_tcp_user_timeout
;
1935 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_USER_TIMEOUT
,
1936 (char *) &timeout
, sizeof(timeout
)) < 0)
1939 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_USER_TIMEOUT")));
1940 return STATUS_ERROR
;
1943 port
->tcp_user_timeout
= timeout
;
1948 (errmsg("%s(%s) not supported", "setsockopt", "TCP_USER_TIMEOUT")));
1949 return STATUS_ERROR
;
1957 * Check if the client is still connected.
1960 pq_check_connection(void)
1962 #if defined(POLLRDHUP)
1964 * POLLRDHUP is a Linux extension to poll(2) to detect sockets closed by
1965 * the other end. We don't have a portable way to do that without
1966 * actually trying to read or write data on other systems. We don't want
1967 * to read because that would be confused by pipelined queries and COPY
1968 * data. Perhaps in future we'll try to write a heartbeat message instead.
1970 struct pollfd pollfd
;
1973 pollfd
.fd
= MyProcPort
->sock
;
1974 pollfd
.events
= POLLOUT
| POLLIN
| POLLRDHUP
;
1977 rc
= poll(&pollfd
, 1, 0);
1982 (errcode_for_socket_access(),
1983 errmsg("could not poll socket: %m")));
1986 else if (rc
== 1 && (pollfd
.revents
& (POLLHUP
| POLLRDHUP
)))