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) 1996-2023, 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 #include <netinet/tcp.h>
76 #include "common/ip.h"
77 #include "libpq/libpq.h"
78 #include "miscadmin.h"
79 #include "port/pg_bswap.h"
80 #include "storage/ipc.h"
81 #include "utils/guc_hooks.h"
82 #include "utils/memutils.h"
85 * Cope with the various platform-specific ways to spell TCP keepalive socket
86 * options. This doesn't cover Windows, which as usual does its own thing.
88 #if defined(TCP_KEEPIDLE)
89 /* TCP_KEEPIDLE is the name of this option on Linux and *BSD */
90 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPIDLE
91 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPIDLE"
92 #elif defined(TCP_KEEPALIVE_THRESHOLD)
93 /* TCP_KEEPALIVE_THRESHOLD is the name of this option on Solaris >= 11 */
94 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE_THRESHOLD
95 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE_THRESHOLD"
96 #elif defined(TCP_KEEPALIVE) && defined(__darwin__)
97 /* TCP_KEEPALIVE is the name of this option on macOS */
98 /* Caution: Solaris has this symbol but it means something different */
99 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE
100 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE"
104 * Configuration options
106 int Unix_socket_permissions
;
107 char *Unix_socket_group
;
109 /* Where the Unix socket files are (list of palloc'd strings) */
110 static List
*sock_paths
= NIL
;
113 * Buffers for low-level I/O.
115 * The receive buffer is fixed size. Send buffer is usually 8k, but can be
116 * enlarged by pq_putmessage_noblock() if the message doesn't fit otherwise.
119 #define PQ_SEND_BUFFER_SIZE 8192
120 #define PQ_RECV_BUFFER_SIZE 8192
122 static char *PqSendBuffer
;
123 static int PqSendBufferSize
; /* Size send buffer */
124 static int PqSendPointer
; /* Next index to store a byte in PqSendBuffer */
125 static int PqSendStart
; /* Next index to send a byte in PqSendBuffer */
127 static char PqRecvBuffer
[PQ_RECV_BUFFER_SIZE
];
128 static int PqRecvPointer
; /* Next index to read a byte from PqRecvBuffer */
129 static int PqRecvLength
; /* End of data available in PqRecvBuffer */
134 static bool PqCommBusy
; /* busy sending data to the client */
135 static bool PqCommReadingMsg
; /* in the middle of reading a message */
138 /* Internal functions */
139 static void socket_comm_reset(void);
140 static void socket_close(int code
, Datum arg
);
141 static void socket_set_nonblocking(bool nonblocking
);
142 static int socket_flush(void);
143 static int socket_flush_if_writable(void);
144 static bool socket_is_send_pending(void);
145 static int socket_putmessage(char msgtype
, const char *s
, size_t len
);
146 static void socket_putmessage_noblock(char msgtype
, const char *s
, size_t len
);
147 static int internal_putbytes(const char *s
, size_t len
);
148 static int internal_flush(void);
150 static int Lock_AF_UNIX(const char *unixSocketDir
, const char *unixSocketPath
);
151 static int Setup_AF_UNIX(const char *sock_path
);
153 static const PQcommMethods PqCommSocketMethods
= {
156 socket_flush_if_writable
,
157 socket_is_send_pending
,
159 socket_putmessage_noblock
162 const PQcommMethods
*PqCommMethods
= &PqCommSocketMethods
;
164 WaitEventSet
*FeBeWaitSet
;
167 /* --------------------------------
168 * pq_init - initialize libpq at backend startup
169 * --------------------------------
174 int socket_pos PG_USED_FOR_ASSERTS_ONLY
;
175 int latch_pos PG_USED_FOR_ASSERTS_ONLY
;
177 /* initialize state variables */
178 PqSendBufferSize
= PQ_SEND_BUFFER_SIZE
;
179 PqSendBuffer
= MemoryContextAlloc(TopMemoryContext
, PqSendBufferSize
);
180 PqSendPointer
= PqSendStart
= PqRecvPointer
= PqRecvLength
= 0;
182 PqCommReadingMsg
= false;
184 /* set up process-exit hook to close the socket */
185 on_proc_exit(socket_close
, 0);
188 * In backends (as soon as forked) we operate the underlying socket in
189 * nonblocking mode and use latches to implement blocking semantics if
190 * needed. That allows us to provide safely interruptible reads and
193 * Use COMMERROR on failure, because ERROR would try to send the error to
194 * the client, which might require changing the mode again, leading to
195 * infinite recursion.
198 if (!pg_set_noblock(MyProcPort
->sock
))
200 (errmsg("could not set socket to nonblocking mode: %m")));
205 /* Don't give the socket to any subprograms we execute. */
206 if (fcntl(MyProcPort
->sock
, F_SETFD
, FD_CLOEXEC
) < 0)
207 elog(FATAL
, "fcntl(F_SETFD) failed on socket: %m");
210 FeBeWaitSet
= CreateWaitEventSet(TopMemoryContext
, FeBeWaitSetNEvents
);
211 socket_pos
= AddWaitEventToSet(FeBeWaitSet
, WL_SOCKET_WRITEABLE
,
212 MyProcPort
->sock
, NULL
, NULL
);
213 latch_pos
= AddWaitEventToSet(FeBeWaitSet
, WL_LATCH_SET
, PGINVALID_SOCKET
,
215 AddWaitEventToSet(FeBeWaitSet
, WL_POSTMASTER_DEATH
, PGINVALID_SOCKET
,
219 * The event positions match the order we added them, but let's sanity
220 * check them to be sure.
222 Assert(socket_pos
== FeBeWaitSetSocketPos
);
223 Assert(latch_pos
== FeBeWaitSetLatchPos
);
226 /* --------------------------------
227 * socket_comm_reset - reset libpq during error recovery
229 * This is called from error recovery at the outer idle loop. It's
230 * just to get us out of trouble if we somehow manage to elog() from
231 * inside a pqcomm.c routine (which ideally will never happen, but...)
232 * --------------------------------
235 socket_comm_reset(void)
237 /* Do not throw away pending data, but do reset the busy flag */
241 /* --------------------------------
242 * socket_close - shutdown libpq at backend exit
244 * This is the one pg_on_exit_callback in place during BackendInitialize().
245 * That function's unusual signal handling constrains that this callback be
246 * safe to run at any instant.
247 * --------------------------------
250 socket_close(int code
, Datum arg
)
252 /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
253 if (MyProcPort
!= NULL
)
257 * Shutdown GSSAPI layer. This section does nothing when interrupting
258 * BackendInitialize(), because pg_GSS_recvauth() makes first use of
261 * Note that we don't bother to free MyProcPort->gss, since we're
262 * about to exit anyway.
268 if (MyProcPort
->gss
->ctx
!= GSS_C_NO_CONTEXT
)
269 gss_delete_sec_context(&min_s
, &MyProcPort
->gss
->ctx
, NULL
);
271 if (MyProcPort
->gss
->cred
!= GSS_C_NO_CREDENTIAL
)
272 gss_release_cred(&min_s
, &MyProcPort
->gss
->cred
);
274 #endif /* ENABLE_GSS */
277 * Cleanly shut down SSL layer. Nowhere else does a postmaster child
278 * call this, so this is safe when interrupting BackendInitialize().
280 secure_close(MyProcPort
);
283 * Formerly we did an explicit close() here, but it seems better to
284 * leave the socket open until the process dies. This allows clients
285 * to perform a "synchronous close" if they care --- wait till the
286 * transport layer reports connection closure, and you can be sure the
287 * backend has exited.
289 * We do set sock to PGINVALID_SOCKET to prevent any further I/O,
292 MyProcPort
->sock
= PGINVALID_SOCKET
;
299 * Streams -- wrapper around Unix socket system calls
302 * Stream functions are used for vanilla TCP connection protocol.
307 * StreamServerPort -- open a "listening" port to accept connections.
309 * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
310 * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
311 * specified. For TCP ports, hostName is either NULL for all interfaces or
312 * the interface to listen on, and unixSocketDir is ignored (can be NULL).
314 * Successfully opened sockets are added to the ListenSocket[] array (of
315 * length MaxListen), at the first position that isn't PGINVALID_SOCKET.
317 * RETURNS: STATUS_OK or STATUS_ERROR
321 StreamServerPort(int family
, const char *hostName
, unsigned short portNumber
,
322 const char *unixSocketDir
,
323 pgsocket ListenSocket
[], int MaxListen
)
329 char portNumberStr
[32];
330 const char *familyDesc
;
331 char familyDescBuf
[64];
332 const char *addrDesc
;
333 char addrBuf
[NI_MAXHOST
];
335 struct addrinfo
*addrs
= NULL
,
337 struct addrinfo hint
;
338 int listen_index
= 0;
340 char unixSocketPath
[MAXPGPATH
];
341 #if !defined(WIN32) || defined(IPV6_V6ONLY)
345 /* Initialize hint structure */
346 MemSet(&hint
, 0, sizeof(hint
));
347 hint
.ai_family
= family
;
348 hint
.ai_flags
= AI_PASSIVE
;
349 hint
.ai_socktype
= SOCK_STREAM
;
351 if (family
== AF_UNIX
)
354 * Create unixSocketPath from portNumber and unixSocketDir and lock
357 UNIXSOCK_PATH(unixSocketPath
, portNumber
, unixSocketDir
);
358 if (strlen(unixSocketPath
) >= UNIXSOCK_PATH_BUFLEN
)
361 (errmsg("Unix-domain socket path \"%s\" is too long (maximum %d bytes)",
363 (int) (UNIXSOCK_PATH_BUFLEN
- 1))));
366 if (Lock_AF_UNIX(unixSocketDir
, unixSocketPath
) != STATUS_OK
)
368 service
= unixSocketPath
;
372 snprintf(portNumberStr
, sizeof(portNumberStr
), "%d", portNumber
);
373 service
= portNumberStr
;
376 ret
= pg_getaddrinfo_all(hostName
, service
, &hint
, &addrs
);
381 (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
382 hostName
, service
, gai_strerror(ret
))));
385 (errmsg("could not translate service \"%s\" to address: %s",
386 service
, gai_strerror(ret
))));
388 pg_freeaddrinfo_all(hint
.ai_family
, addrs
);
392 for (addr
= addrs
; addr
; addr
= addr
->ai_next
)
394 if (family
!= AF_UNIX
&& addr
->ai_family
== AF_UNIX
)
397 * Only set up a unix domain socket when they really asked for it.
398 * The service/port is different in that case.
403 /* See if there is still room to add 1 more socket. */
404 for (; listen_index
< MaxListen
; listen_index
++)
406 if (ListenSocket
[listen_index
] == PGINVALID_SOCKET
)
409 if (listen_index
>= MaxListen
)
412 (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
417 /* set up address family name for log messages */
418 switch (addr
->ai_family
)
421 familyDesc
= _("IPv4");
424 familyDesc
= _("IPv6");
427 familyDesc
= _("Unix");
430 snprintf(familyDescBuf
, sizeof(familyDescBuf
),
431 _("unrecognized address family %d"),
433 familyDesc
= familyDescBuf
;
437 /* set up text form of address for log messages */
438 if (addr
->ai_family
== AF_UNIX
)
439 addrDesc
= unixSocketPath
;
442 pg_getnameinfo_all((const struct sockaddr_storage
*) addr
->ai_addr
,
444 addrBuf
, sizeof(addrBuf
),
450 if ((fd
= socket(addr
->ai_family
, SOCK_STREAM
, 0)) == PGINVALID_SOCKET
)
453 (errcode_for_socket_access(),
454 /* translator: first %s is IPv4, IPv6, or Unix */
455 errmsg("could not create %s socket for address \"%s\": %m",
456 familyDesc
, addrDesc
)));
463 * Without the SO_REUSEADDR flag, a new postmaster can't be started
464 * right away after a stop or crash, giving "address already in use"
465 * error on TCP ports.
467 * On win32, however, this behavior only happens if the
468 * SO_EXCLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows
469 * multiple servers to listen on the same address, resulting in
470 * unpredictable behavior. With no flags at all, win32 behaves as Unix
473 if (addr
->ai_family
!= AF_UNIX
)
475 if ((setsockopt(fd
, SOL_SOCKET
, SO_REUSEADDR
,
476 (char *) &one
, sizeof(one
))) == -1)
479 (errcode_for_socket_access(),
480 /* translator: third %s is IPv4, IPv6, or Unix */
481 errmsg("%s(%s) failed for %s address \"%s\": %m",
482 "setsockopt", "SO_REUSEADDR",
483 familyDesc
, addrDesc
)));
491 if (addr
->ai_family
== AF_INET6
)
493 if (setsockopt(fd
, IPPROTO_IPV6
, IPV6_V6ONLY
,
494 (char *) &one
, sizeof(one
)) == -1)
497 (errcode_for_socket_access(),
498 /* translator: third %s is IPv4, IPv6, or Unix */
499 errmsg("%s(%s) failed for %s address \"%s\": %m",
500 "setsockopt", "IPV6_V6ONLY",
501 familyDesc
, addrDesc
)));
509 * Note: This might fail on some OS's, like Linux older than
510 * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
511 * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
514 err
= bind(fd
, addr
->ai_addr
, addr
->ai_addrlen
);
517 int saved_errno
= errno
;
520 (errcode_for_socket_access(),
521 /* translator: first %s is IPv4, IPv6, or Unix */
522 errmsg("could not bind %s address \"%s\": %m",
523 familyDesc
, addrDesc
),
524 saved_errno
== EADDRINUSE
?
525 (addr
->ai_family
== AF_UNIX
?
526 errhint("Is another postmaster already running on port %d?",
528 errhint("Is another postmaster already running on port %d?"
529 " If not, wait a few seconds and retry.",
530 (int) portNumber
)) : 0));
535 if (addr
->ai_family
== AF_UNIX
)
537 if (Setup_AF_UNIX(service
) != STATUS_OK
)
545 * Select appropriate accept-queue length limit. It seems reasonable
546 * to use a value similar to the maximum number of child processes
547 * that the postmaster will permit.
549 maxconn
= MaxConnections
* 2;
551 err
= listen(fd
, maxconn
);
555 (errcode_for_socket_access(),
556 /* translator: first %s is IPv4, IPv6, or Unix */
557 errmsg("could not listen on %s address \"%s\": %m",
558 familyDesc
, addrDesc
)));
563 if (addr
->ai_family
== AF_UNIX
)
565 (errmsg("listening on Unix socket \"%s\"",
569 /* translator: first %s is IPv4 or IPv6 */
570 (errmsg("listening on %s address \"%s\", port %d",
571 familyDesc
, addrDesc
, (int) portNumber
)));
573 ListenSocket
[listen_index
] = fd
;
577 pg_freeaddrinfo_all(hint
.ai_family
, addrs
);
587 * Lock_AF_UNIX -- configure unix socket file path
590 Lock_AF_UNIX(const char *unixSocketDir
, const char *unixSocketPath
)
592 /* no lock file for abstract sockets */
593 if (unixSocketPath
[0] == '@')
597 * Grab an interlock file associated with the socket file.
599 * Note: there are two reasons for using a socket lock file, rather than
600 * trying to interlock directly on the socket itself. First, it's a lot
601 * more portable, and second, it lets us remove any pre-existing socket
602 * file without race conditions.
604 CreateSocketLockFile(unixSocketPath
, true, unixSocketDir
);
607 * Once we have the interlock, we can safely delete any pre-existing
608 * socket file to avoid failure at bind() time.
610 (void) unlink(unixSocketPath
);
613 * Remember socket file pathnames for later maintenance.
615 sock_paths
= lappend(sock_paths
, pstrdup(unixSocketPath
));
622 * Setup_AF_UNIX -- configure unix socket permissions
625 Setup_AF_UNIX(const char *sock_path
)
627 /* no file system permissions for abstract sockets */
628 if (sock_path
[0] == '@')
632 * Fix socket ownership/permission if requested. Note we must do this
633 * before we listen() to avoid a window where unwanted connections could
636 Assert(Unix_socket_group
);
637 if (Unix_socket_group
[0] != '\0')
640 elog(WARNING
, "configuration item unix_socket_group is not supported on this platform");
646 val
= strtoul(Unix_socket_group
, &endptr
, 10);
648 { /* numeric group id */
652 { /* convert group name to id */
655 gr
= getgrnam(Unix_socket_group
);
659 (errmsg("group \"%s\" does not exist",
660 Unix_socket_group
)));
665 if (chown(sock_path
, -1, gid
) == -1)
668 (errcode_for_file_access(),
669 errmsg("could not set group of file \"%s\": %m",
676 if (chmod(sock_path
, Unix_socket_permissions
) == -1)
679 (errcode_for_file_access(),
680 errmsg("could not set permissions of file \"%s\": %m",
689 * StreamConnection -- create a new connection with client using
690 * server port. Set port->sock to the FD of the new connection.
692 * ASSUME: that this doesn't need to be non-blocking because
693 * the Postmaster waits for the socket to be ready to accept().
695 * RETURNS: STATUS_OK or STATUS_ERROR
698 StreamConnection(pgsocket server_fd
, Port
*port
)
700 /* accept connection and fill in the client (remote) address */
701 port
->raddr
.salen
= sizeof(port
->raddr
.addr
);
702 if ((port
->sock
= accept(server_fd
,
703 (struct sockaddr
*) &port
->raddr
.addr
,
704 &port
->raddr
.salen
)) == PGINVALID_SOCKET
)
707 (errcode_for_socket_access(),
708 errmsg("could not accept new connection: %m")));
711 * If accept() fails then postmaster.c will still see the server
712 * socket as read-ready, and will immediately try again. To avoid
713 * uselessly sucking lots of CPU, delay a bit before trying again.
714 * (The most likely reason for failure is being out of kernel file
715 * table slots; we can do little except hope some will get freed up.)
717 pg_usleep(100000L); /* wait 0.1 sec */
721 /* fill in the server (local) address */
722 port
->laddr
.salen
= sizeof(port
->laddr
.addr
);
723 if (getsockname(port
->sock
,
724 (struct sockaddr
*) &port
->laddr
.addr
,
725 &port
->laddr
.salen
) < 0)
728 (errmsg("%s() failed: %m", "getsockname")));
732 /* select NODELAY and KEEPALIVE options if it's a TCP connection */
733 if (port
->laddr
.addr
.ss_family
!= AF_UNIX
)
744 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_NODELAY
,
745 (char *) &on
, sizeof(on
)) < 0)
748 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_NODELAY")));
753 if (setsockopt(port
->sock
, SOL_SOCKET
, SO_KEEPALIVE
,
754 (char *) &on
, sizeof(on
)) < 0)
757 (errmsg("%s(%s) failed: %m", "setsockopt", "SO_KEEPALIVE")));
764 * This is a Win32 socket optimization. The OS send buffer should be
765 * large enough to send the whole Postgres send buffer in one go, or
766 * performance suffers. The Postgres send buffer can be enlarged if a
767 * very large message needs to be sent, but we won't attempt to
768 * enlarge the OS buffer if that happens, so somewhat arbitrarily
769 * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
770 * (That's 32kB with the current default).
772 * The default OS buffer size used to be 8kB in earlier Windows
773 * versions, but was raised to 64kB in Windows 2012. So it shouldn't
774 * be necessary to change it in later versions anymore. Changing it
775 * unnecessarily can even reduce performance, because setting
776 * SO_SNDBUF in the application disables the "dynamic send buffering"
777 * feature that was introduced in Windows 7. So before fiddling with
778 * SO_SNDBUF, check if the current buffer size is already large enough
779 * and only increase it if necessary.
781 * See https://support.microsoft.com/kb/823764/EN-US/ and
782 * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
784 optlen
= sizeof(oldopt
);
785 if (getsockopt(port
->sock
, SOL_SOCKET
, SO_SNDBUF
, (char *) &oldopt
,
789 (errmsg("%s(%s) failed: %m", "getsockopt", "SO_SNDBUF")));
792 newopt
= PQ_SEND_BUFFER_SIZE
* 4;
795 if (setsockopt(port
->sock
, SOL_SOCKET
, SO_SNDBUF
, (char *) &newopt
,
799 (errmsg("%s(%s) failed: %m", "setsockopt", "SO_SNDBUF")));
806 * Also apply the current keepalive parameters. If we fail to set a
807 * parameter, don't error out, because these aren't universally
808 * supported. (Note: you might think we need to reset the GUC
809 * variables to 0 in such a case, but it's not necessary because the
810 * show hooks for these variables report the truth anyway.)
812 (void) pq_setkeepalivesidle(tcp_keepalives_idle
, port
);
813 (void) pq_setkeepalivesinterval(tcp_keepalives_interval
, port
);
814 (void) pq_setkeepalivescount(tcp_keepalives_count
, port
);
815 (void) pq_settcpusertimeout(tcp_user_timeout
, port
);
822 * StreamClose -- close a client/backend connection
824 * NOTE: this is NOT used to terminate a session; it is just used to release
825 * the file descriptor in a process that should no longer have the socket
826 * open. (For example, the postmaster calls this after passing ownership
827 * of the connection to a child process.) It is expected that someone else
828 * still has the socket open. So, we only want to close the descriptor,
829 * we do NOT want to send anything to the far end.
832 StreamClose(pgsocket sock
)
838 * TouchSocketFiles -- mark socket files as recently accessed
840 * This routine should be called every so often to ensure that the socket
841 * files have a recent mod date (ordinary operations on sockets usually won't
842 * change the mod date). That saves them from being removed by
843 * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
844 * never have put the socket file in /tmp...)
847 TouchSocketFiles(void)
851 /* Loop through all created sockets... */
852 foreach(l
, sock_paths
)
854 char *sock_path
= (char *) lfirst(l
);
856 /* Ignore errors; there's no point in complaining */
857 (void) utime(sock_path
, NULL
);
862 * RemoveSocketFiles -- unlink socket files at postmaster shutdown
865 RemoveSocketFiles(void)
869 /* Loop through all created sockets... */
870 foreach(l
, sock_paths
)
872 char *sock_path
= (char *) lfirst(l
);
874 /* Ignore any error. */
875 (void) unlink(sock_path
);
877 /* Since we're about to exit, no need to reclaim storage */
882 /* --------------------------------
883 * Low-level I/O routines begin here.
885 * These routines communicate with a frontend client across a connection
886 * already established by the preceding routines.
887 * --------------------------------
890 /* --------------------------------
891 * socket_set_nonblocking - set socket blocking/non-blocking
893 * Sets the socket non-blocking if nonblocking is true, or sets it
894 * blocking otherwise.
895 * --------------------------------
898 socket_set_nonblocking(bool nonblocking
)
900 if (MyProcPort
== NULL
)
902 (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST
),
903 errmsg("there is no client connection")));
905 MyProcPort
->noblock
= nonblocking
;
908 /* --------------------------------
909 * pq_recvbuf - load some bytes into the input buffer
911 * returns 0 if OK, EOF if trouble
912 * --------------------------------
917 if (PqRecvPointer
> 0)
919 if (PqRecvLength
> PqRecvPointer
)
921 /* still some unread data, left-justify it in the buffer */
922 memmove(PqRecvBuffer
, PqRecvBuffer
+ PqRecvPointer
,
923 PqRecvLength
- PqRecvPointer
);
924 PqRecvLength
-= PqRecvPointer
;
928 PqRecvLength
= PqRecvPointer
= 0;
931 /* Ensure that we're in blocking mode */
932 socket_set_nonblocking(false);
934 /* Can fill buffer from PqRecvLength and upwards */
941 r
= secure_read(MyProcPort
, PqRecvBuffer
+ PqRecvLength
,
942 PQ_RECV_BUFFER_SIZE
- PqRecvLength
);
947 continue; /* Ok if interrupted */
950 * Careful: an ereport() that tries to write to the client would
951 * cause recursion to here, leading to stack overflow and core
952 * dump! This message must go *only* to the postmaster log.
954 * If errno is zero, assume it's EOF and let the caller complain.
958 (errcode_for_socket_access(),
959 errmsg("could not receive data from client: %m")));
965 * EOF detected. We used to write a log message here, but it's
966 * better to expect the ultimate caller to do that.
970 /* r contains number of bytes read, so just incr length */
976 /* --------------------------------
977 * pq_getbyte - get a single byte from connection, or return EOF
978 * --------------------------------
983 Assert(PqCommReadingMsg
);
985 while (PqRecvPointer
>= PqRecvLength
)
987 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
988 return EOF
; /* Failed to recv data */
990 return (unsigned char) PqRecvBuffer
[PqRecvPointer
++];
993 /* --------------------------------
994 * pq_peekbyte - peek at next byte from connection
996 * Same as pq_getbyte() except we don't advance the pointer.
997 * --------------------------------
1002 Assert(PqCommReadingMsg
);
1004 while (PqRecvPointer
>= PqRecvLength
)
1006 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1007 return EOF
; /* Failed to recv data */
1009 return (unsigned char) PqRecvBuffer
[PqRecvPointer
];
1012 /* --------------------------------
1013 * pq_getbyte_if_available - get a single byte from connection,
1016 * The received byte is stored in *c. Returns 1 if a byte was read,
1017 * 0 if no data was available, or EOF if trouble.
1018 * --------------------------------
1021 pq_getbyte_if_available(unsigned char *c
)
1025 Assert(PqCommReadingMsg
);
1027 if (PqRecvPointer
< PqRecvLength
)
1029 *c
= PqRecvBuffer
[PqRecvPointer
++];
1033 /* Put the socket into non-blocking mode */
1034 socket_set_nonblocking(true);
1038 r
= secure_read(MyProcPort
, c
, 1);
1042 * Ok if no data available without blocking or interrupted (though
1043 * EINTR really shouldn't happen with a non-blocking socket). Report
1046 if (errno
== EAGAIN
|| errno
== EWOULDBLOCK
|| errno
== EINTR
)
1051 * Careful: an ereport() that tries to write to the client would
1052 * cause recursion to here, leading to stack overflow and core
1053 * dump! This message must go *only* to the postmaster log.
1055 * If errno is zero, assume it's EOF and let the caller complain.
1059 (errcode_for_socket_access(),
1060 errmsg("could not receive data from client: %m")));
1073 /* --------------------------------
1074 * pq_getbytes - get a known number of bytes from connection
1076 * returns 0 if OK, EOF if trouble
1077 * --------------------------------
1080 pq_getbytes(char *s
, size_t len
)
1084 Assert(PqCommReadingMsg
);
1088 while (PqRecvPointer
>= PqRecvLength
)
1090 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1091 return EOF
; /* Failed to recv data */
1093 amount
= PqRecvLength
- PqRecvPointer
;
1096 memcpy(s
, PqRecvBuffer
+ PqRecvPointer
, amount
);
1097 PqRecvPointer
+= amount
;
1104 /* --------------------------------
1105 * pq_discardbytes - throw away a known number of bytes
1107 * same as pq_getbytes except we do not copy the data to anyplace.
1108 * this is used for resynchronizing after read errors.
1110 * returns 0 if OK, EOF if trouble
1111 * --------------------------------
1114 pq_discardbytes(size_t len
)
1118 Assert(PqCommReadingMsg
);
1122 while (PqRecvPointer
>= PqRecvLength
)
1124 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1125 return EOF
; /* Failed to recv data */
1127 amount
= PqRecvLength
- PqRecvPointer
;
1130 PqRecvPointer
+= amount
;
1136 /* --------------------------------
1137 * pq_buffer_has_data - is any buffered data available to read?
1139 * This will *not* attempt to read more data.
1140 * --------------------------------
1143 pq_buffer_has_data(void)
1145 return (PqRecvPointer
< PqRecvLength
);
1149 /* --------------------------------
1150 * pq_startmsgread - begin reading a message from the client.
1152 * This must be called before any of the pq_get* functions.
1153 * --------------------------------
1156 pq_startmsgread(void)
1159 * There shouldn't be a read active already, but let's check just to be
1162 if (PqCommReadingMsg
)
1164 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1165 errmsg("terminating connection because protocol synchronization was lost")));
1167 PqCommReadingMsg
= true;
1171 /* --------------------------------
1172 * pq_endmsgread - finish reading message.
1174 * This must be called after reading a message with pq_getbytes()
1175 * and friends, to indicate that we have read the whole message.
1176 * pq_getmessage() does this implicitly.
1177 * --------------------------------
1182 Assert(PqCommReadingMsg
);
1184 PqCommReadingMsg
= false;
1187 /* --------------------------------
1188 * pq_is_reading_msg - are we currently reading a message?
1190 * This is used in error recovery at the outer idle loop to detect if we have
1191 * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1192 * will check for that too, but it's nicer to detect it earlier.
1193 * --------------------------------
1196 pq_is_reading_msg(void)
1198 return PqCommReadingMsg
;
1201 /* --------------------------------
1202 * pq_getmessage - get a message with length word from connection
1204 * The return value is placed in an expansible StringInfo, which has
1205 * already been initialized by the caller.
1206 * Only the message body is placed in the StringInfo; the length word
1207 * is removed. Also, s->cursor is initialized to zero for convenience
1208 * in scanning the message contents.
1210 * maxlen is the upper limit on the length of the
1211 * message we are willing to accept. We abort the connection (by
1212 * returning EOF) if client tries to send more than that.
1214 * returns 0 if OK, EOF if trouble
1215 * --------------------------------
1218 pq_getmessage(StringInfo s
, int maxlen
)
1222 Assert(PqCommReadingMsg
);
1226 /* Read message length word */
1227 if (pq_getbytes((char *) &len
, 4) == EOF
)
1230 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1231 errmsg("unexpected EOF within message length word")));
1235 len
= pg_ntoh32(len
);
1237 if (len
< 4 || len
> maxlen
)
1240 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1241 errmsg("invalid message length")));
1245 len
-= 4; /* discount length itself */
1250 * Allocate space for message. If we run out of room (ridiculously
1251 * large message), we will elog(ERROR), but we want to discard the
1252 * message body so as not to lose communication sync.
1256 enlargeStringInfo(s
, len
);
1260 if (pq_discardbytes(len
) == EOF
)
1262 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1263 errmsg("incomplete message from client")));
1265 /* we discarded the rest of the message so we're back in sync. */
1266 PqCommReadingMsg
= false;
1271 /* And grab the message */
1272 if (pq_getbytes(s
->data
, len
) == EOF
)
1275 (errcode(ERRCODE_PROTOCOL_VIOLATION
),
1276 errmsg("incomplete message from client")));
1280 /* Place a trailing null per StringInfo convention */
1281 s
->data
[len
] = '\0';
1284 /* finished reading the message. */
1285 PqCommReadingMsg
= false;
1292 internal_putbytes(const char *s
, size_t len
)
1298 /* If buffer is full, then flush it out */
1299 if (PqSendPointer
>= PqSendBufferSize
)
1301 socket_set_nonblocking(false);
1302 if (internal_flush())
1305 amount
= PqSendBufferSize
- PqSendPointer
;
1308 memcpy(PqSendBuffer
+ PqSendPointer
, s
, amount
);
1309 PqSendPointer
+= amount
;
1316 /* --------------------------------
1317 * socket_flush - flush pending output
1319 * returns 0 if OK, EOF if trouble
1320 * --------------------------------
1327 /* No-op if reentrant call */
1331 socket_set_nonblocking(false);
1332 res
= internal_flush();
1337 /* --------------------------------
1338 * internal_flush - flush pending output
1340 * Returns 0 if OK (meaning everything was sent, or operation would block
1341 * and the socket is in non-blocking mode), or EOF if trouble.
1342 * --------------------------------
1345 internal_flush(void)
1347 static int last_reported_send_errno
= 0;
1349 char *bufptr
= PqSendBuffer
+ PqSendStart
;
1350 char *bufend
= PqSendBuffer
+ PqSendPointer
;
1352 while (bufptr
< bufend
)
1356 r
= secure_write(MyProcPort
, bufptr
, bufend
- bufptr
);
1361 continue; /* Ok if we were interrupted */
1364 * Ok if no data writable without blocking, and the socket is in
1365 * non-blocking mode.
1367 if (errno
== EAGAIN
||
1368 errno
== EWOULDBLOCK
)
1374 * Careful: an ereport() that tries to write to the client would
1375 * cause recursion to here, leading to stack overflow and core
1376 * dump! This message must go *only* to the postmaster log.
1378 * If a client disconnects while we're in the midst of output, we
1379 * might write quite a bit of data before we get to a safe query
1380 * abort point. So, suppress duplicate log messages.
1382 if (errno
!= last_reported_send_errno
)
1384 last_reported_send_errno
= errno
;
1386 (errcode_for_socket_access(),
1387 errmsg("could not send data to client: %m")));
1391 * We drop the buffered data anyway so that processing can
1392 * continue, even though we'll probably quit soon. We also set a
1393 * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1396 PqSendStart
= PqSendPointer
= 0;
1397 ClientConnectionLost
= 1;
1398 InterruptPending
= 1;
1402 last_reported_send_errno
= 0; /* reset after any successful send */
1407 PqSendStart
= PqSendPointer
= 0;
1411 /* --------------------------------
1412 * pq_flush_if_writable - flush pending output if writable without blocking
1414 * Returns 0 if OK, or EOF if trouble.
1415 * --------------------------------
1418 socket_flush_if_writable(void)
1422 /* Quick exit if nothing to do */
1423 if (PqSendPointer
== PqSendStart
)
1426 /* No-op if reentrant call */
1430 /* Temporarily put the socket into non-blocking mode */
1431 socket_set_nonblocking(true);
1434 res
= internal_flush();
1439 /* --------------------------------
1440 * socket_is_send_pending - is there any pending data in the output buffer?
1441 * --------------------------------
1444 socket_is_send_pending(void)
1446 return (PqSendStart
< PqSendPointer
);
1449 /* --------------------------------
1450 * Message-level I/O routines begin here.
1451 * --------------------------------
1455 /* --------------------------------
1456 * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1458 * msgtype is a message type code to place before the message body.
1460 * len is the length of the message body data at *s. A message length
1461 * word (equal to len+4 because it counts itself too) is inserted by this
1464 * We suppress messages generated while pqcomm.c is busy. This
1465 * avoids any possibility of messages being inserted within other
1466 * messages. The only known trouble case arises if SIGQUIT occurs
1467 * during a pqcomm.c routine --- quickdie() will try to send a warning
1468 * message, and the most reasonable approach seems to be to drop it.
1470 * returns 0 if OK, EOF if trouble
1471 * --------------------------------
1474 socket_putmessage(char msgtype
, const char *s
, size_t len
)
1478 Assert(msgtype
!= 0);
1483 if (internal_putbytes(&msgtype
, 1))
1486 n32
= pg_hton32((uint32
) (len
+ 4));
1487 if (internal_putbytes((char *) &n32
, 4))
1490 if (internal_putbytes(s
, len
))
1500 /* --------------------------------
1501 * pq_putmessage_noblock - like pq_putmessage, but never blocks
1503 * If the output buffer is too small to hold the message, the buffer
1507 socket_putmessage_noblock(char msgtype
, const char *s
, size_t len
)
1509 int res PG_USED_FOR_ASSERTS_ONLY
;
1513 * Ensure we have enough space in the output buffer for the message header
1514 * as well as the message itself.
1516 required
= PqSendPointer
+ 1 + 4 + len
;
1517 if (required
> PqSendBufferSize
)
1519 PqSendBuffer
= repalloc(PqSendBuffer
, required
);
1520 PqSendBufferSize
= required
;
1522 res
= pq_putmessage(msgtype
, s
, len
);
1523 Assert(res
== 0); /* should not fail when the message fits in
1527 /* --------------------------------
1528 * pq_putmessage_v2 - send a message in protocol version 2
1530 * msgtype is a message type code to place before the message body.
1532 * We no longer support protocol version 2, but we have kept this
1533 * function so that if a client tries to connect with protocol version 2,
1534 * as a courtesy we can still send the "unsupported protocol version"
1535 * error to the client in the old format.
1537 * Like in pq_putmessage(), we suppress messages generated while
1540 * returns 0 if OK, EOF if trouble
1541 * --------------------------------
1544 pq_putmessage_v2(char msgtype
, const char *s
, size_t len
)
1546 Assert(msgtype
!= 0);
1551 if (internal_putbytes(&msgtype
, 1))
1554 if (internal_putbytes(s
, len
))
1565 * Support for TCP Keepalive parameters
1569 * On Windows, we need to set both idle and interval at the same time.
1570 * We also cannot reset them to the default (setting to zero will
1571 * actually set them to zero, not default), therefore we fallback to
1572 * the out-of-the-box default instead.
1574 #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1576 pq_setkeepaliveswin32(Port
*port
, int idle
, int interval
)
1578 struct tcp_keepalive ka
;
1582 idle
= 2 * 60 * 60; /* default = 2 hours */
1584 interval
= 1; /* default = 1 second */
1587 ka
.keepalivetime
= idle
* 1000;
1588 ka
.keepaliveinterval
= interval
* 1000;
1590 if (WSAIoctl(port
->sock
,
1602 (errmsg("%s(%s) failed: error code %d",
1603 "WSAIoctl", "SIO_KEEPALIVE_VALS", WSAGetLastError())));
1604 return STATUS_ERROR
;
1606 if (port
->keepalives_idle
!= idle
)
1607 port
->keepalives_idle
= idle
;
1608 if (port
->keepalives_interval
!= interval
)
1609 port
->keepalives_interval
= interval
;
1615 pq_getkeepalivesidle(Port
*port
)
1617 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1618 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1621 if (port
->keepalives_idle
!= 0)
1622 return port
->keepalives_idle
;
1624 if (port
->default_keepalives_idle
== 0)
1627 socklen_t size
= sizeof(port
->default_keepalives_idle
);
1629 if (getsockopt(port
->sock
, IPPROTO_TCP
, PG_TCP_KEEPALIVE_IDLE
,
1630 (char *) &port
->default_keepalives_idle
,
1634 (errmsg("%s(%s) failed: %m", "getsockopt", PG_TCP_KEEPALIVE_IDLE_STR
)));
1635 port
->default_keepalives_idle
= -1; /* don't know */
1638 /* We can't get the defaults on Windows, so return "don't know" */
1639 port
->default_keepalives_idle
= -1;
1643 return port
->default_keepalives_idle
;
1650 pq_setkeepalivesidle(int idle
, Port
*port
)
1652 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1655 /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1656 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1657 if (idle
== port
->keepalives_idle
)
1661 if (port
->default_keepalives_idle
<= 0)
1663 if (pq_getkeepalivesidle(port
) < 0)
1666 return STATUS_OK
; /* default is set but unknown */
1668 return STATUS_ERROR
;
1673 idle
= port
->default_keepalives_idle
;
1675 if (setsockopt(port
->sock
, IPPROTO_TCP
, PG_TCP_KEEPALIVE_IDLE
,
1676 (char *) &idle
, sizeof(idle
)) < 0)
1679 (errmsg("%s(%s) failed: %m", "setsockopt", PG_TCP_KEEPALIVE_IDLE_STR
)));
1680 return STATUS_ERROR
;
1683 port
->keepalives_idle
= idle
;
1685 return pq_setkeepaliveswin32(port
, idle
, port
->keepalives_interval
);
1691 (errmsg("setting the keepalive idle time is not supported")));
1692 return STATUS_ERROR
;
1700 pq_getkeepalivesinterval(Port
*port
)
1702 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1703 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1706 if (port
->keepalives_interval
!= 0)
1707 return port
->keepalives_interval
;
1709 if (port
->default_keepalives_interval
== 0)
1712 socklen_t size
= sizeof(port
->default_keepalives_interval
);
1714 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPINTVL
,
1715 (char *) &port
->default_keepalives_interval
,
1719 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPINTVL")));
1720 port
->default_keepalives_interval
= -1; /* don't know */
1723 /* We can't get the defaults on Windows, so return "don't know" */
1724 port
->default_keepalives_interval
= -1;
1728 return port
->default_keepalives_interval
;
1735 pq_setkeepalivesinterval(int interval
, Port
*port
)
1737 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1740 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1741 if (interval
== port
->keepalives_interval
)
1745 if (port
->default_keepalives_interval
<= 0)
1747 if (pq_getkeepalivesinterval(port
) < 0)
1750 return STATUS_OK
; /* default is set but unknown */
1752 return STATUS_ERROR
;
1757 interval
= port
->default_keepalives_interval
;
1759 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPINTVL
,
1760 (char *) &interval
, sizeof(interval
)) < 0)
1763 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPINTVL")));
1764 return STATUS_ERROR
;
1767 port
->keepalives_interval
= interval
;
1769 return pq_setkeepaliveswin32(port
, port
->keepalives_idle
, interval
);
1775 (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPINTVL")));
1776 return STATUS_ERROR
;
1784 pq_getkeepalivescount(Port
*port
)
1787 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1790 if (port
->keepalives_count
!= 0)
1791 return port
->keepalives_count
;
1793 if (port
->default_keepalives_count
== 0)
1795 socklen_t size
= sizeof(port
->default_keepalives_count
);
1797 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPCNT
,
1798 (char *) &port
->default_keepalives_count
,
1802 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPCNT")));
1803 port
->default_keepalives_count
= -1; /* don't know */
1807 return port
->default_keepalives_count
;
1814 pq_setkeepalivescount(int count
, Port
*port
)
1816 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1820 if (count
== port
->keepalives_count
)
1823 if (port
->default_keepalives_count
<= 0)
1825 if (pq_getkeepalivescount(port
) < 0)
1828 return STATUS_OK
; /* default is set but unknown */
1830 return STATUS_ERROR
;
1835 count
= port
->default_keepalives_count
;
1837 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_KEEPCNT
,
1838 (char *) &count
, sizeof(count
)) < 0)
1841 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPCNT")));
1842 return STATUS_ERROR
;
1845 port
->keepalives_count
= count
;
1850 (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPCNT")));
1851 return STATUS_ERROR
;
1859 pq_gettcpusertimeout(Port
*port
)
1861 #ifdef TCP_USER_TIMEOUT
1862 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1865 if (port
->tcp_user_timeout
!= 0)
1866 return port
->tcp_user_timeout
;
1868 if (port
->default_tcp_user_timeout
== 0)
1870 socklen_t size
= sizeof(port
->default_tcp_user_timeout
);
1872 if (getsockopt(port
->sock
, IPPROTO_TCP
, TCP_USER_TIMEOUT
,
1873 (char *) &port
->default_tcp_user_timeout
,
1877 (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_USER_TIMEOUT")));
1878 port
->default_tcp_user_timeout
= -1; /* don't know */
1882 return port
->default_tcp_user_timeout
;
1889 pq_settcpusertimeout(int timeout
, Port
*port
)
1891 if (port
== NULL
|| port
->laddr
.addr
.ss_family
== AF_UNIX
)
1894 #ifdef TCP_USER_TIMEOUT
1895 if (timeout
== port
->tcp_user_timeout
)
1898 if (port
->default_tcp_user_timeout
<= 0)
1900 if (pq_gettcpusertimeout(port
) < 0)
1903 return STATUS_OK
; /* default is set but unknown */
1905 return STATUS_ERROR
;
1910 timeout
= port
->default_tcp_user_timeout
;
1912 if (setsockopt(port
->sock
, IPPROTO_TCP
, TCP_USER_TIMEOUT
,
1913 (char *) &timeout
, sizeof(timeout
)) < 0)
1916 (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_USER_TIMEOUT")));
1917 return STATUS_ERROR
;
1920 port
->tcp_user_timeout
= timeout
;
1925 (errmsg("%s(%s) not supported", "setsockopt", "TCP_USER_TIMEOUT")));
1926 return STATUS_ERROR
;
1934 * GUC assign_hook for tcp_keepalives_idle
1937 assign_tcp_keepalives_idle(int newval
, void *extra
)
1940 * The kernel API provides no way to test a value without setting it; and
1941 * once we set it we might fail to unset it. So there seems little point
1942 * in fully implementing the check-then-assign GUC API for these
1943 * variables. Instead we just do the assignment on demand.
1944 * pq_setkeepalivesidle reports any problems via ereport(LOG).
1946 * This approach means that the GUC value might have little to do with the
1947 * actual kernel value, so we use a show_hook that retrieves the kernel
1948 * value rather than trusting GUC's copy.
1950 (void) pq_setkeepalivesidle(newval
, MyProcPort
);
1954 * GUC show_hook for tcp_keepalives_idle
1957 show_tcp_keepalives_idle(void)
1959 /* See comments in assign_tcp_keepalives_idle */
1960 static char nbuf
[16];
1962 snprintf(nbuf
, sizeof(nbuf
), "%d", pq_getkeepalivesidle(MyProcPort
));
1967 * GUC assign_hook for tcp_keepalives_interval
1970 assign_tcp_keepalives_interval(int newval
, void *extra
)
1972 /* See comments in assign_tcp_keepalives_idle */
1973 (void) pq_setkeepalivesinterval(newval
, MyProcPort
);
1977 * GUC show_hook for tcp_keepalives_interval
1980 show_tcp_keepalives_interval(void)
1982 /* See comments in assign_tcp_keepalives_idle */
1983 static char nbuf
[16];
1985 snprintf(nbuf
, sizeof(nbuf
), "%d", pq_getkeepalivesinterval(MyProcPort
));
1990 * GUC assign_hook for tcp_keepalives_count
1993 assign_tcp_keepalives_count(int newval
, void *extra
)
1995 /* See comments in assign_tcp_keepalives_idle */
1996 (void) pq_setkeepalivescount(newval
, MyProcPort
);
2000 * GUC show_hook for tcp_keepalives_count
2003 show_tcp_keepalives_count(void)
2005 /* See comments in assign_tcp_keepalives_idle */
2006 static char nbuf
[16];
2008 snprintf(nbuf
, sizeof(nbuf
), "%d", pq_getkeepalivescount(MyProcPort
));
2013 * GUC assign_hook for tcp_user_timeout
2016 assign_tcp_user_timeout(int newval
, void *extra
)
2018 /* See comments in assign_tcp_keepalives_idle */
2019 (void) pq_settcpusertimeout(newval
, MyProcPort
);
2023 * GUC show_hook for tcp_user_timeout
2026 show_tcp_user_timeout(void)
2028 /* See comments in assign_tcp_keepalives_idle */
2029 static char nbuf
[16];
2031 snprintf(nbuf
, sizeof(nbuf
), "%d", pq_gettcpusertimeout(MyProcPort
));
2036 * Check if the client is still connected.
2039 pq_check_connection(void)
2041 WaitEvent events
[FeBeWaitSetNEvents
];
2045 * It's OK to modify the socket event filter without restoring, because
2046 * all FeBeWaitSet socket wait sites do the same.
2048 ModifyWaitEvent(FeBeWaitSet
, FeBeWaitSetSocketPos
, WL_SOCKET_CLOSED
, NULL
);
2051 rc
= WaitEventSetWait(FeBeWaitSet
, 0, events
, lengthof(events
), 0);
2052 for (int i
= 0; i
< rc
; ++i
)
2054 if (events
[i
].events
& WL_SOCKET_CLOSED
)
2056 if (events
[i
].events
& WL_LATCH_SET
)
2059 * A latch event might be preventing other events from being
2060 * reported. Reset it and poll again. No need to restore it
2061 * because no code should expect latches to survive across
2062 * CHECK_FOR_INTERRUPTS().
2064 ResetLatch(MyLatch
);