Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * pqcomm.c
4 : * Communication functions between the Frontend and the Backend
5 : *
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.
9 : *
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.
15 : *
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...
19 : *
20 : * Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group
21 : * Portions Copyright (c) 1994, Regents of the University of California
22 : *
23 : * src/backend/libpq/pqcomm.c
24 : *
25 : *-------------------------------------------------------------------------
26 : */
27 :
28 : /*------------------------
29 : * INTERFACE ROUTINES
30 : *
31 : * setup/teardown:
32 : * ListenServerPort - Open postmaster's server port
33 : * AcceptConnection - Accept new connection with client
34 : * TouchSocketFiles - Protect socket files against /tmp cleaners
35 : * pq_init - initialize libpq at backend startup
36 : * socket_comm_reset - reset libpq during error recovery
37 : * socket_close - shutdown libpq at backend exit
38 : *
39 : * low-level I/O:
40 : * pq_getbytes - get a known number of bytes from connection
41 : * pq_getmessage - get a message with length word from connection
42 : * pq_getbyte - get next byte from connection
43 : * pq_peekbyte - peek at next byte from connection
44 : * pq_flush - flush pending output
45 : * pq_flush_if_writable - flush pending output if writable without blocking
46 : * pq_getbyte_if_available - get a byte if available without blocking
47 : *
48 : * message-level I/O
49 : * pq_putmessage - send a normal message (suppressed in COPY OUT mode)
50 : * pq_putmessage_noblock - buffer a normal message (suppressed in COPY OUT)
51 : *
52 : *------------------------
53 : */
54 : #include "postgres.h"
55 :
56 : #ifdef HAVE_POLL_H
57 : #include <poll.h>
58 : #endif
59 : #include <signal.h>
60 : #include <fcntl.h>
61 : #include <grp.h>
62 : #include <unistd.h>
63 : #include <sys/file.h>
64 : #include <sys/socket.h>
65 : #include <sys/stat.h>
66 : #include <sys/time.h>
67 : #include <netdb.h>
68 : #include <netinet/in.h>
69 : #include <netinet/tcp.h>
70 : #include <utime.h>
71 : #ifdef WIN32
72 : #include <mstcpip.h>
73 : #endif
74 :
75 : #include "common/ip.h"
76 : #include "libpq/libpq.h"
77 : #include "miscadmin.h"
78 : #include "port/pg_bswap.h"
79 : #include "postmaster/postmaster.h"
80 : #include "storage/ipc.h"
81 : #include "utils/guc_hooks.h"
82 : #include "utils/memutils.h"
83 :
84 : /*
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.
87 : */
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"
101 : #endif
102 :
103 : /*
104 : * Configuration options
105 : */
106 : int Unix_socket_permissions;
107 : char *Unix_socket_group;
108 :
109 : /* Where the Unix socket files are (list of palloc'd strings) */
110 : static List *sock_paths = NIL;
111 :
112 : /*
113 : * Buffers for low-level I/O.
114 : *
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.
117 : */
118 :
119 : #define PQ_SEND_BUFFER_SIZE 8192
120 : #define PQ_RECV_BUFFER_SIZE 8192
121 :
122 : static char *PqSendBuffer;
123 : static int PqSendBufferSize; /* Size send buffer */
124 : static size_t PqSendPointer; /* Next index to store a byte in PqSendBuffer */
125 : static size_t PqSendStart; /* Next index to send a byte in PqSendBuffer */
126 :
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 */
130 :
131 : /*
132 : * Message status
133 : */
134 : static bool PqCommBusy; /* busy sending data to the client */
135 : static bool PqCommReadingMsg; /* in the middle of reading a message */
136 :
137 :
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 inline int internal_putbytes(const void *b, size_t len);
148 : static inline int internal_flush(void);
149 : static pg_noinline int internal_flush_buffer(const char *buf, size_t *start,
150 : size_t *end);
151 :
152 : static int Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath);
153 : static int Setup_AF_UNIX(const char *sock_path);
154 :
155 : static const PQcommMethods PqCommSocketMethods = {
156 : .comm_reset = socket_comm_reset,
157 : .flush = socket_flush,
158 : .flush_if_writable = socket_flush_if_writable,
159 : .is_send_pending = socket_is_send_pending,
160 : .putmessage = socket_putmessage,
161 : .putmessage_noblock = socket_putmessage_noblock
162 : };
163 :
164 : const PQcommMethods *PqCommMethods = &PqCommSocketMethods;
165 :
166 : WaitEventSet *FeBeWaitSet;
167 :
168 :
169 : /* --------------------------------
170 : * pq_init - initialize libpq at backend startup
171 : * --------------------------------
172 : */
173 : Port *
174 25980 : pq_init(ClientSocket *client_sock)
175 : {
176 : Port *port;
177 : int socket_pos PG_USED_FOR_ASSERTS_ONLY;
178 : int latch_pos PG_USED_FOR_ASSERTS_ONLY;
179 :
180 : /* allocate the Port struct and copy the ClientSocket contents to it */
181 25980 : port = palloc0(sizeof(Port));
182 25980 : port->sock = client_sock->sock;
183 25980 : memcpy(&port->raddr.addr, &client_sock->raddr.addr, client_sock->raddr.salen);
184 25980 : port->raddr.salen = client_sock->raddr.salen;
185 :
186 : /* fill in the server (local) address */
187 25980 : port->laddr.salen = sizeof(port->laddr.addr);
188 25980 : if (getsockname(port->sock,
189 25980 : (struct sockaddr *) &port->laddr.addr,
190 : &port->laddr.salen) < 0)
191 : {
192 0 : ereport(FATAL,
193 : (errmsg("%s() failed: %m", "getsockname")));
194 : }
195 :
196 : /* select NODELAY and KEEPALIVE options if it's a TCP connection */
197 25980 : if (port->laddr.addr.ss_family != AF_UNIX)
198 : {
199 : int on;
200 : #ifdef WIN32
201 : int oldopt;
202 : int optlen;
203 : int newopt;
204 : #endif
205 :
206 : #ifdef TCP_NODELAY
207 578 : on = 1;
208 578 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_NODELAY,
209 : (char *) &on, sizeof(on)) < 0)
210 : {
211 0 : ereport(FATAL,
212 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_NODELAY")));
213 : }
214 : #endif
215 578 : on = 1;
216 578 : if (setsockopt(port->sock, SOL_SOCKET, SO_KEEPALIVE,
217 : (char *) &on, sizeof(on)) < 0)
218 : {
219 0 : ereport(FATAL,
220 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_KEEPALIVE")));
221 : }
222 :
223 : #ifdef WIN32
224 :
225 : /*
226 : * This is a Win32 socket optimization. The OS send buffer should be
227 : * large enough to send the whole Postgres send buffer in one go, or
228 : * performance suffers. The Postgres send buffer can be enlarged if a
229 : * very large message needs to be sent, but we won't attempt to
230 : * enlarge the OS buffer if that happens, so somewhat arbitrarily
231 : * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
232 : * (That's 32kB with the current default).
233 : *
234 : * The default OS buffer size used to be 8kB in earlier Windows
235 : * versions, but was raised to 64kB in Windows 2012. So it shouldn't
236 : * be necessary to change it in later versions anymore. Changing it
237 : * unnecessarily can even reduce performance, because setting
238 : * SO_SNDBUF in the application disables the "dynamic send buffering"
239 : * feature that was introduced in Windows 7. So before fiddling with
240 : * SO_SNDBUF, check if the current buffer size is already large enough
241 : * and only increase it if necessary.
242 : *
243 : * See https://support.microsoft.com/kb/823764/EN-US/ and
244 : * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
245 : */
246 : optlen = sizeof(oldopt);
247 : if (getsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &oldopt,
248 : &optlen) < 0)
249 : {
250 : ereport(FATAL,
251 : (errmsg("%s(%s) failed: %m", "getsockopt", "SO_SNDBUF")));
252 : }
253 : newopt = PQ_SEND_BUFFER_SIZE * 4;
254 : if (oldopt < newopt)
255 : {
256 : if (setsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &newopt,
257 : sizeof(newopt)) < 0)
258 : {
259 : ereport(FATAL,
260 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_SNDBUF")));
261 : }
262 : }
263 : #endif
264 :
265 : /*
266 : * Also apply the current keepalive parameters. If we fail to set a
267 : * parameter, don't error out, because these aren't universally
268 : * supported. (Note: you might think we need to reset the GUC
269 : * variables to 0 in such a case, but it's not necessary because the
270 : * show hooks for these variables report the truth anyway.)
271 : */
272 578 : (void) pq_setkeepalivesidle(tcp_keepalives_idle, port);
273 578 : (void) pq_setkeepalivesinterval(tcp_keepalives_interval, port);
274 578 : (void) pq_setkeepalivescount(tcp_keepalives_count, port);
275 578 : (void) pq_settcpusertimeout(tcp_user_timeout, port);
276 : }
277 :
278 : /* initialize state variables */
279 25980 : PqSendBufferSize = PQ_SEND_BUFFER_SIZE;
280 25980 : PqSendBuffer = MemoryContextAlloc(TopMemoryContext, PqSendBufferSize);
281 25980 : PqSendPointer = PqSendStart = PqRecvPointer = PqRecvLength = 0;
282 25980 : PqCommBusy = false;
283 25980 : PqCommReadingMsg = false;
284 :
285 : /* set up process-exit hook to close the socket */
286 25980 : on_proc_exit(socket_close, 0);
287 :
288 : /*
289 : * In backends (as soon as forked) we operate the underlying socket in
290 : * nonblocking mode and use latches to implement blocking semantics if
291 : * needed. That allows us to provide safely interruptible reads and
292 : * writes.
293 : */
294 : #ifndef WIN32
295 25980 : if (!pg_set_noblock(port->sock))
296 0 : ereport(FATAL,
297 : (errmsg("could not set socket to nonblocking mode: %m")));
298 : #endif
299 :
300 : #ifndef WIN32
301 :
302 : /* Don't give the socket to any subprograms we execute. */
303 25980 : if (fcntl(port->sock, F_SETFD, FD_CLOEXEC) < 0)
304 0 : elog(FATAL, "fcntl(F_SETFD) failed on socket: %m");
305 : #endif
306 :
307 25980 : FeBeWaitSet = CreateWaitEventSet(NULL, FeBeWaitSetNEvents);
308 25980 : socket_pos = AddWaitEventToSet(FeBeWaitSet, WL_SOCKET_WRITEABLE,
309 : port->sock, NULL, NULL);
310 25980 : latch_pos = AddWaitEventToSet(FeBeWaitSet, WL_LATCH_SET, PGINVALID_SOCKET,
311 : MyLatch, NULL);
312 25980 : AddWaitEventToSet(FeBeWaitSet, WL_POSTMASTER_DEATH, PGINVALID_SOCKET,
313 : NULL, NULL);
314 :
315 : /*
316 : * The event positions match the order we added them, but let's sanity
317 : * check them to be sure.
318 : */
319 : Assert(socket_pos == FeBeWaitSetSocketPos);
320 : Assert(latch_pos == FeBeWaitSetLatchPos);
321 :
322 25980 : return port;
323 : }
324 :
325 : /* --------------------------------
326 : * socket_comm_reset - reset libpq during error recovery
327 : *
328 : * This is called from error recovery at the outer idle loop. It's
329 : * just to get us out of trouble if we somehow manage to elog() from
330 : * inside a pqcomm.c routine (which ideally will never happen, but...)
331 : * --------------------------------
332 : */
333 : static void
334 43460 : socket_comm_reset(void)
335 : {
336 : /* Do not throw away pending data, but do reset the busy flag */
337 43460 : PqCommBusy = false;
338 43460 : }
339 :
340 : /* --------------------------------
341 : * socket_close - shutdown libpq at backend exit
342 : *
343 : * This is the one pg_on_exit_callback in place during BackendInitialize().
344 : * That function's unusual signal handling constrains that this callback be
345 : * safe to run at any instant.
346 : * --------------------------------
347 : */
348 : static void
349 25980 : socket_close(int code, Datum arg)
350 : {
351 : /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
352 25980 : if (MyProcPort != NULL)
353 : {
354 : #ifdef ENABLE_GSS
355 : /*
356 : * Shutdown GSSAPI layer. This section does nothing when interrupting
357 : * BackendInitialize(), because pg_GSS_recvauth() makes first use of
358 : * "ctx" and "cred".
359 : *
360 : * Note that we don't bother to free MyProcPort->gss, since we're
361 : * about to exit anyway.
362 : */
363 : if (MyProcPort->gss)
364 : {
365 : OM_uint32 min_s;
366 :
367 : if (MyProcPort->gss->ctx != GSS_C_NO_CONTEXT)
368 : gss_delete_sec_context(&min_s, &MyProcPort->gss->ctx, NULL);
369 :
370 : if (MyProcPort->gss->cred != GSS_C_NO_CREDENTIAL)
371 : gss_release_cred(&min_s, &MyProcPort->gss->cred);
372 : }
373 : #endif /* ENABLE_GSS */
374 :
375 : /*
376 : * Cleanly shut down SSL layer. Nowhere else does a postmaster child
377 : * call this, so this is safe when interrupting BackendInitialize().
378 : */
379 25980 : secure_close(MyProcPort);
380 :
381 : /*
382 : * Formerly we did an explicit close() here, but it seems better to
383 : * leave the socket open until the process dies. This allows clients
384 : * to perform a "synchronous close" if they care --- wait till the
385 : * transport layer reports connection closure, and you can be sure the
386 : * backend has exited.
387 : *
388 : * We do set sock to PGINVALID_SOCKET to prevent any further I/O,
389 : * though.
390 : */
391 25980 : MyProcPort->sock = PGINVALID_SOCKET;
392 : }
393 25980 : }
394 :
395 :
396 :
397 : /* --------------------------------
398 : * Postmaster functions to handle sockets.
399 : * --------------------------------
400 : */
401 :
402 : /*
403 : * ListenServerPort -- open a "listening" port to accept connections.
404 : *
405 : * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
406 : * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
407 : * specified. For TCP ports, hostName is either NULL for all interfaces or
408 : * the interface to listen on, and unixSocketDir is ignored (can be NULL).
409 : *
410 : * Successfully opened sockets are appended to the ListenSockets[] array. On
411 : * entry, *NumListenSockets holds the number of elements currently in the
412 : * array, and it is updated to reflect the opened sockets. MaxListen is the
413 : * allocated size of the array.
414 : *
415 : * RETURNS: STATUS_OK or STATUS_ERROR
416 : */
417 : int
418 1756 : ListenServerPort(int family, const char *hostName, unsigned short portNumber,
419 : const char *unixSocketDir,
420 : pgsocket ListenSockets[], int *NumListenSockets, int MaxListen)
421 : {
422 : pgsocket fd;
423 : int err;
424 : int maxconn;
425 : int ret;
426 : char portNumberStr[32];
427 : const char *familyDesc;
428 : char familyDescBuf[64];
429 : const char *addrDesc;
430 : char addrBuf[NI_MAXHOST];
431 : char *service;
432 1756 : struct addrinfo *addrs = NULL,
433 : *addr;
434 : struct addrinfo hint;
435 1756 : int added = 0;
436 : char unixSocketPath[MAXPGPATH];
437 : #if !defined(WIN32) || defined(IPV6_V6ONLY)
438 1756 : int one = 1;
439 : #endif
440 :
441 : /* Initialize hint structure */
442 12292 : MemSet(&hint, 0, sizeof(hint));
443 1756 : hint.ai_family = family;
444 1756 : hint.ai_flags = AI_PASSIVE;
445 1756 : hint.ai_socktype = SOCK_STREAM;
446 :
447 1756 : if (family == AF_UNIX)
448 : {
449 : /*
450 : * Create unixSocketPath from portNumber and unixSocketDir and lock
451 : * that file path
452 : */
453 1698 : UNIXSOCK_PATH(unixSocketPath, portNumber, unixSocketDir);
454 1698 : if (strlen(unixSocketPath) >= UNIXSOCK_PATH_BUFLEN)
455 : {
456 0 : ereport(LOG,
457 : (errmsg("Unix-domain socket path \"%s\" is too long (maximum %d bytes)",
458 : unixSocketPath,
459 : (int) (UNIXSOCK_PATH_BUFLEN - 1))));
460 0 : return STATUS_ERROR;
461 : }
462 1698 : if (Lock_AF_UNIX(unixSocketDir, unixSocketPath) != STATUS_OK)
463 0 : return STATUS_ERROR;
464 1698 : service = unixSocketPath;
465 : }
466 : else
467 : {
468 58 : snprintf(portNumberStr, sizeof(portNumberStr), "%d", portNumber);
469 58 : service = portNumberStr;
470 : }
471 :
472 1756 : ret = pg_getaddrinfo_all(hostName, service, &hint, &addrs);
473 1756 : if (ret || !addrs)
474 : {
475 0 : if (hostName)
476 0 : ereport(LOG,
477 : (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
478 : hostName, service, gai_strerror(ret))));
479 : else
480 0 : ereport(LOG,
481 : (errmsg("could not translate service \"%s\" to address: %s",
482 : service, gai_strerror(ret))));
483 0 : if (addrs)
484 0 : pg_freeaddrinfo_all(hint.ai_family, addrs);
485 0 : return STATUS_ERROR;
486 : }
487 :
488 3514 : for (addr = addrs; addr; addr = addr->ai_next)
489 : {
490 1758 : if (family != AF_UNIX && addr->ai_family == AF_UNIX)
491 : {
492 : /*
493 : * Only set up a unix domain socket when they really asked for it.
494 : * The service/port is different in that case.
495 : */
496 0 : continue;
497 : }
498 :
499 : /* See if there is still room to add 1 more socket. */
500 1758 : if (*NumListenSockets == MaxListen)
501 : {
502 0 : ereport(LOG,
503 : (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
504 : MaxListen)));
505 0 : break;
506 : }
507 :
508 : /* set up address family name for log messages */
509 1758 : switch (addr->ai_family)
510 : {
511 58 : case AF_INET:
512 58 : familyDesc = _("IPv4");
513 58 : break;
514 2 : case AF_INET6:
515 2 : familyDesc = _("IPv6");
516 2 : break;
517 1698 : case AF_UNIX:
518 1698 : familyDesc = _("Unix");
519 1698 : break;
520 0 : default:
521 0 : snprintf(familyDescBuf, sizeof(familyDescBuf),
522 0 : _("unrecognized address family %d"),
523 : addr->ai_family);
524 0 : familyDesc = familyDescBuf;
525 0 : break;
526 : }
527 :
528 : /* set up text form of address for log messages */
529 1758 : if (addr->ai_family == AF_UNIX)
530 1698 : addrDesc = unixSocketPath;
531 : else
532 : {
533 60 : pg_getnameinfo_all((const struct sockaddr_storage *) addr->ai_addr,
534 60 : addr->ai_addrlen,
535 : addrBuf, sizeof(addrBuf),
536 : NULL, 0,
537 : NI_NUMERICHOST);
538 60 : addrDesc = addrBuf;
539 : }
540 :
541 1758 : if ((fd = socket(addr->ai_family, SOCK_STREAM, 0)) == PGINVALID_SOCKET)
542 : {
543 0 : ereport(LOG,
544 : (errcode_for_socket_access(),
545 : /* translator: first %s is IPv4, IPv6, or Unix */
546 : errmsg("could not create %s socket for address \"%s\": %m",
547 : familyDesc, addrDesc)));
548 0 : continue;
549 : }
550 :
551 : #ifndef WIN32
552 : /* Don't give the listen socket to any subprograms we execute. */
553 1758 : if (fcntl(fd, F_SETFD, FD_CLOEXEC) < 0)
554 0 : elog(FATAL, "fcntl(F_SETFD) failed on socket: %m");
555 :
556 : /*
557 : * Without the SO_REUSEADDR flag, a new postmaster can't be started
558 : * right away after a stop or crash, giving "address already in use"
559 : * error on TCP ports.
560 : *
561 : * On win32, however, this behavior only happens if the
562 : * SO_EXCLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows
563 : * multiple servers to listen on the same address, resulting in
564 : * unpredictable behavior. With no flags at all, win32 behaves as Unix
565 : * with SO_REUSEADDR.
566 : */
567 1758 : if (addr->ai_family != AF_UNIX)
568 : {
569 60 : if ((setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
570 : (char *) &one, sizeof(one))) == -1)
571 : {
572 0 : ereport(LOG,
573 : (errcode_for_socket_access(),
574 : /* translator: third %s is IPv4 or IPv6 */
575 : errmsg("%s(%s) failed for %s address \"%s\": %m",
576 : "setsockopt", "SO_REUSEADDR",
577 : familyDesc, addrDesc)));
578 0 : closesocket(fd);
579 0 : continue;
580 : }
581 : }
582 : #endif
583 :
584 : #ifdef IPV6_V6ONLY
585 1758 : if (addr->ai_family == AF_INET6)
586 : {
587 2 : if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY,
588 : (char *) &one, sizeof(one)) == -1)
589 : {
590 0 : ereport(LOG,
591 : (errcode_for_socket_access(),
592 : /* translator: third %s is IPv6 */
593 : errmsg("%s(%s) failed for %s address \"%s\": %m",
594 : "setsockopt", "IPV6_V6ONLY",
595 : familyDesc, addrDesc)));
596 0 : closesocket(fd);
597 0 : continue;
598 : }
599 : }
600 : #endif
601 :
602 : /*
603 : * Note: This might fail on some OS's, like Linux older than
604 : * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
605 : * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
606 : * connections.
607 : */
608 1758 : err = bind(fd, addr->ai_addr, addr->ai_addrlen);
609 1758 : if (err < 0)
610 0 : {
611 0 : int saved_errno = errno;
612 :
613 0 : ereport(LOG,
614 : (errcode_for_socket_access(),
615 : /* translator: first %s is IPv4, IPv6, or Unix */
616 : errmsg("could not bind %s address \"%s\": %m",
617 : familyDesc, addrDesc),
618 : saved_errno == EADDRINUSE ?
619 : (addr->ai_family == AF_UNIX ?
620 : errhint("Is another postmaster already running on port %d?",
621 : (int) portNumber) :
622 : errhint("Is another postmaster already running on port %d?"
623 : " If not, wait a few seconds and retry.",
624 : (int) portNumber)) : 0));
625 0 : closesocket(fd);
626 0 : continue;
627 : }
628 :
629 1758 : if (addr->ai_family == AF_UNIX)
630 : {
631 1698 : if (Setup_AF_UNIX(service) != STATUS_OK)
632 : {
633 0 : closesocket(fd);
634 0 : break;
635 : }
636 : }
637 :
638 : /*
639 : * Select appropriate accept-queue length limit. It seems reasonable
640 : * to use a value similar to the maximum number of child processes
641 : * that the postmaster will permit.
642 : */
643 1758 : maxconn = MaxConnections * 2;
644 :
645 1758 : err = listen(fd, maxconn);
646 1758 : if (err < 0)
647 : {
648 0 : ereport(LOG,
649 : (errcode_for_socket_access(),
650 : /* translator: first %s is IPv4, IPv6, or Unix */
651 : errmsg("could not listen on %s address \"%s\": %m",
652 : familyDesc, addrDesc)));
653 0 : closesocket(fd);
654 0 : continue;
655 : }
656 :
657 1758 : if (addr->ai_family == AF_UNIX)
658 1698 : ereport(LOG,
659 : (errmsg("listening on Unix socket \"%s\"",
660 : addrDesc)));
661 : else
662 60 : ereport(LOG,
663 : /* translator: first %s is IPv4 or IPv6 */
664 : (errmsg("listening on %s address \"%s\", port %d",
665 : familyDesc, addrDesc, (int) portNumber)));
666 :
667 1758 : ListenSockets[*NumListenSockets] = fd;
668 1758 : (*NumListenSockets)++;
669 1758 : added++;
670 : }
671 :
672 1756 : pg_freeaddrinfo_all(hint.ai_family, addrs);
673 :
674 1756 : if (!added)
675 0 : return STATUS_ERROR;
676 :
677 1756 : return STATUS_OK;
678 : }
679 :
680 :
681 : /*
682 : * Lock_AF_UNIX -- configure unix socket file path
683 : */
684 : static int
685 1698 : Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath)
686 : {
687 : /* no lock file for abstract sockets */
688 1698 : if (unixSocketPath[0] == '@')
689 0 : return STATUS_OK;
690 :
691 : /*
692 : * Grab an interlock file associated with the socket file.
693 : *
694 : * Note: there are two reasons for using a socket lock file, rather than
695 : * trying to interlock directly on the socket itself. First, it's a lot
696 : * more portable, and second, it lets us remove any pre-existing socket
697 : * file without race conditions.
698 : */
699 1698 : CreateSocketLockFile(unixSocketPath, true, unixSocketDir);
700 :
701 : /*
702 : * Once we have the interlock, we can safely delete any pre-existing
703 : * socket file to avoid failure at bind() time.
704 : */
705 1698 : (void) unlink(unixSocketPath);
706 :
707 : /*
708 : * Remember socket file pathnames for later maintenance.
709 : */
710 1698 : sock_paths = lappend(sock_paths, pstrdup(unixSocketPath));
711 :
712 1698 : return STATUS_OK;
713 : }
714 :
715 :
716 : /*
717 : * Setup_AF_UNIX -- configure unix socket permissions
718 : */
719 : static int
720 1698 : Setup_AF_UNIX(const char *sock_path)
721 : {
722 : /* no file system permissions for abstract sockets */
723 1698 : if (sock_path[0] == '@')
724 0 : return STATUS_OK;
725 :
726 : /*
727 : * Fix socket ownership/permission if requested. Note we must do this
728 : * before we listen() to avoid a window where unwanted connections could
729 : * get accepted.
730 : */
731 : Assert(Unix_socket_group);
732 1698 : if (Unix_socket_group[0] != '\0')
733 : {
734 : #ifdef WIN32
735 : elog(WARNING, "configuration item \"unix_socket_group\" is not supported on this platform");
736 : #else
737 : char *endptr;
738 : unsigned long val;
739 : gid_t gid;
740 :
741 0 : val = strtoul(Unix_socket_group, &endptr, 10);
742 0 : if (*endptr == '\0')
743 : { /* numeric group id */
744 0 : gid = val;
745 : }
746 : else
747 : { /* convert group name to id */
748 : struct group *gr;
749 :
750 0 : gr = getgrnam(Unix_socket_group);
751 0 : if (!gr)
752 : {
753 0 : ereport(LOG,
754 : (errmsg("group \"%s\" does not exist",
755 : Unix_socket_group)));
756 0 : return STATUS_ERROR;
757 : }
758 0 : gid = gr->gr_gid;
759 : }
760 0 : if (chown(sock_path, -1, gid) == -1)
761 : {
762 0 : ereport(LOG,
763 : (errcode_for_file_access(),
764 : errmsg("could not set group of file \"%s\": %m",
765 : sock_path)));
766 0 : return STATUS_ERROR;
767 : }
768 : #endif
769 : }
770 :
771 1698 : if (chmod(sock_path, Unix_socket_permissions) == -1)
772 : {
773 0 : ereport(LOG,
774 : (errcode_for_file_access(),
775 : errmsg("could not set permissions of file \"%s\": %m",
776 : sock_path)));
777 0 : return STATUS_ERROR;
778 : }
779 1698 : return STATUS_OK;
780 : }
781 :
782 :
783 : /*
784 : * AcceptConnection -- accept a new connection with client using
785 : * server port. Fills *client_sock with the FD and endpoint info
786 : * of the new connection.
787 : *
788 : * ASSUME: that this doesn't need to be non-blocking because
789 : * the Postmaster waits for the socket to be ready to accept().
790 : *
791 : * RETURNS: STATUS_OK or STATUS_ERROR
792 : */
793 : int
794 26706 : AcceptConnection(pgsocket server_fd, ClientSocket *client_sock)
795 : {
796 : /* accept connection and fill in the client (remote) address */
797 26706 : client_sock->raddr.salen = sizeof(client_sock->raddr.addr);
798 26706 : if ((client_sock->sock = accept(server_fd,
799 26706 : (struct sockaddr *) &client_sock->raddr.addr,
800 : &client_sock->raddr.salen)) == PGINVALID_SOCKET)
801 : {
802 0 : ereport(LOG,
803 : (errcode_for_socket_access(),
804 : errmsg("could not accept new connection: %m")));
805 :
806 : /*
807 : * If accept() fails then postmaster.c will still see the server
808 : * socket as read-ready, and will immediately try again. To avoid
809 : * uselessly sucking lots of CPU, delay a bit before trying again.
810 : * (The most likely reason for failure is being out of kernel file
811 : * table slots; we can do little except hope some will get freed up.)
812 : */
813 0 : pg_usleep(100000L); /* wait 0.1 sec */
814 0 : return STATUS_ERROR;
815 : }
816 :
817 26706 : return STATUS_OK;
818 : }
819 :
820 : /*
821 : * TouchSocketFiles -- mark socket files as recently accessed
822 : *
823 : * This routine should be called every so often to ensure that the socket
824 : * files have a recent mod date (ordinary operations on sockets usually won't
825 : * change the mod date). That saves them from being removed by
826 : * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
827 : * never have put the socket file in /tmp...)
828 : */
829 : void
830 0 : TouchSocketFiles(void)
831 : {
832 : ListCell *l;
833 :
834 : /* Loop through all created sockets... */
835 0 : foreach(l, sock_paths)
836 : {
837 0 : char *sock_path = (char *) lfirst(l);
838 :
839 : /* Ignore errors; there's no point in complaining */
840 0 : (void) utime(sock_path, NULL);
841 : }
842 0 : }
843 :
844 : /*
845 : * RemoveSocketFiles -- unlink socket files at postmaster shutdown
846 : */
847 : void
848 1700 : RemoveSocketFiles(void)
849 : {
850 : ListCell *l;
851 :
852 : /* Loop through all created sockets... */
853 3398 : foreach(l, sock_paths)
854 : {
855 1698 : char *sock_path = (char *) lfirst(l);
856 :
857 : /* Ignore any error. */
858 1698 : (void) unlink(sock_path);
859 : }
860 : /* Since we're about to exit, no need to reclaim storage */
861 1700 : }
862 :
863 :
864 : /* --------------------------------
865 : * Low-level I/O routines begin here.
866 : *
867 : * These routines communicate with a frontend client across a connection
868 : * already established by the preceding routines.
869 : * --------------------------------
870 : */
871 :
872 : /* --------------------------------
873 : * socket_set_nonblocking - set socket blocking/non-blocking
874 : *
875 : * Sets the socket non-blocking if nonblocking is true, or sets it
876 : * blocking otherwise.
877 : * --------------------------------
878 : */
879 : static void
880 6323910 : socket_set_nonblocking(bool nonblocking)
881 : {
882 6323910 : if (MyProcPort == NULL)
883 0 : ereport(ERROR,
884 : (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST),
885 : errmsg("there is no client connection")));
886 :
887 6323910 : MyProcPort->noblock = nonblocking;
888 6323910 : }
889 :
890 : /* --------------------------------
891 : * pq_recvbuf - load some bytes into the input buffer
892 : *
893 : * returns 0 if OK, EOF if trouble
894 : * --------------------------------
895 : */
896 : static int
897 730522 : pq_recvbuf(void)
898 : {
899 730522 : if (PqRecvPointer > 0)
900 : {
901 704542 : if (PqRecvLength > PqRecvPointer)
902 : {
903 : /* still some unread data, left-justify it in the buffer */
904 0 : memmove(PqRecvBuffer, PqRecvBuffer + PqRecvPointer,
905 0 : PqRecvLength - PqRecvPointer);
906 0 : PqRecvLength -= PqRecvPointer;
907 0 : PqRecvPointer = 0;
908 : }
909 : else
910 704542 : PqRecvLength = PqRecvPointer = 0;
911 : }
912 :
913 : /* Ensure that we're in blocking mode */
914 730522 : socket_set_nonblocking(false);
915 :
916 : /* Can fill buffer from PqRecvLength and upwards */
917 : for (;;)
918 0 : {
919 : int r;
920 :
921 730522 : errno = 0;
922 :
923 1460968 : r = secure_read(MyProcPort, PqRecvBuffer + PqRecvLength,
924 730522 : PQ_RECV_BUFFER_SIZE - PqRecvLength);
925 :
926 730446 : if (r < 0)
927 : {
928 2 : if (errno == EINTR)
929 0 : continue; /* Ok if interrupted */
930 :
931 : /*
932 : * Careful: an ereport() that tries to write to the client would
933 : * cause recursion to here, leading to stack overflow and core
934 : * dump! This message must go *only* to the postmaster log.
935 : *
936 : * If errno is zero, assume it's EOF and let the caller complain.
937 : */
938 2 : if (errno != 0)
939 2 : ereport(COMMERROR,
940 : (errcode_for_socket_access(),
941 : errmsg("could not receive data from client: %m")));
942 2 : return EOF;
943 : }
944 730444 : if (r == 0)
945 : {
946 : /*
947 : * EOF detected. We used to write a log message here, but it's
948 : * better to expect the ultimate caller to do that.
949 : */
950 160 : return EOF;
951 : }
952 : /* r contains number of bytes read, so just incr length */
953 730284 : PqRecvLength += r;
954 730284 : return 0;
955 : }
956 : }
957 :
958 : /* --------------------------------
959 : * pq_getbyte - get a single byte from connection, or return EOF
960 : * --------------------------------
961 : */
962 : int
963 1116096 : pq_getbyte(void)
964 : {
965 : Assert(PqCommReadingMsg);
966 :
967 1752076 : while (PqRecvPointer >= PqRecvLength)
968 : {
969 636184 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
970 128 : return EOF; /* Failed to recv data */
971 : }
972 1115892 : return (unsigned char) PqRecvBuffer[PqRecvPointer++];
973 : }
974 :
975 : /* --------------------------------
976 : * pq_peekbyte - peek at next byte from connection
977 : *
978 : * Same as pq_getbyte() except we don't advance the pointer.
979 : * --------------------------------
980 : */
981 : int
982 25980 : pq_peekbyte(void)
983 : {
984 : Assert(PqCommReadingMsg);
985 :
986 51956 : while (PqRecvPointer >= PqRecvLength)
987 : {
988 25980 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
989 4 : return EOF; /* Failed to recv data */
990 : }
991 25976 : return (unsigned char) PqRecvBuffer[PqRecvPointer];
992 : }
993 :
994 : /* --------------------------------
995 : * pq_getbyte_if_available - get a single byte from connection,
996 : * if available
997 : *
998 : * The received byte is stored in *c. Returns 1 if a byte was read,
999 : * 0 if no data was available, or EOF if trouble.
1000 : * --------------------------------
1001 : */
1002 : int
1003 3606958 : pq_getbyte_if_available(unsigned char *c)
1004 : {
1005 : int r;
1006 :
1007 : Assert(PqCommReadingMsg);
1008 :
1009 3606958 : if (PqRecvPointer < PqRecvLength)
1010 : {
1011 119458 : *c = PqRecvBuffer[PqRecvPointer++];
1012 119458 : return 1;
1013 : }
1014 :
1015 : /* Put the socket into non-blocking mode */
1016 3487500 : socket_set_nonblocking(true);
1017 :
1018 3487500 : errno = 0;
1019 :
1020 3487500 : r = secure_read(MyProcPort, c, 1);
1021 3487500 : if (r < 0)
1022 : {
1023 : /*
1024 : * Ok if no data available without blocking or interrupted (though
1025 : * EINTR really shouldn't happen with a non-blocking socket). Report
1026 : * other errors.
1027 : */
1028 3422158 : if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
1029 3422150 : r = 0;
1030 : else
1031 : {
1032 : /*
1033 : * Careful: an ereport() that tries to write to the client would
1034 : * cause recursion to here, leading to stack overflow and core
1035 : * dump! This message must go *only* to the postmaster log.
1036 : *
1037 : * If errno is zero, assume it's EOF and let the caller complain.
1038 : */
1039 8 : if (errno != 0)
1040 8 : ereport(COMMERROR,
1041 : (errcode_for_socket_access(),
1042 : errmsg("could not receive data from client: %m")));
1043 8 : r = EOF;
1044 : }
1045 : }
1046 65342 : else if (r == 0)
1047 : {
1048 : /* EOF detected */
1049 20 : r = EOF;
1050 : }
1051 :
1052 3487500 : return r;
1053 : }
1054 :
1055 : /* --------------------------------
1056 : * pq_getbytes - get a known number of bytes from connection
1057 : *
1058 : * returns 0 if OK, EOF if trouble
1059 : * --------------------------------
1060 : */
1061 : int
1062 2630878 : pq_getbytes(void *b, size_t len)
1063 : {
1064 2630878 : char *s = b;
1065 : size_t amount;
1066 :
1067 : Assert(PqCommReadingMsg);
1068 :
1069 5264240 : while (len > 0)
1070 : {
1071 2701720 : while (PqRecvPointer >= PqRecvLength)
1072 : {
1073 68358 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1074 30 : return EOF; /* Failed to recv data */
1075 : }
1076 2633362 : amount = PqRecvLength - PqRecvPointer;
1077 2633362 : if (amount > len)
1078 1903204 : amount = len;
1079 2633362 : memcpy(s, PqRecvBuffer + PqRecvPointer, amount);
1080 2633362 : PqRecvPointer += amount;
1081 2633362 : s += amount;
1082 2633362 : len -= amount;
1083 : }
1084 2630848 : return 0;
1085 : }
1086 :
1087 : /* --------------------------------
1088 : * pq_discardbytes - throw away a known number of bytes
1089 : *
1090 : * same as pq_getbytes except we do not copy the data to anyplace.
1091 : * this is used for resynchronizing after read errors.
1092 : *
1093 : * returns 0 if OK, EOF if trouble
1094 : * --------------------------------
1095 : */
1096 : static int
1097 0 : pq_discardbytes(size_t len)
1098 : {
1099 : size_t amount;
1100 :
1101 : Assert(PqCommReadingMsg);
1102 :
1103 0 : while (len > 0)
1104 : {
1105 0 : while (PqRecvPointer >= PqRecvLength)
1106 : {
1107 0 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1108 0 : return EOF; /* Failed to recv data */
1109 : }
1110 0 : amount = PqRecvLength - PqRecvPointer;
1111 0 : if (amount > len)
1112 0 : amount = len;
1113 0 : PqRecvPointer += amount;
1114 0 : len -= amount;
1115 : }
1116 0 : return 0;
1117 : }
1118 :
1119 : /* --------------------------------
1120 : * pq_buffer_remaining_data - return number of bytes in receive buffer
1121 : *
1122 : * This will *not* attempt to read more data. And reading up to that number of
1123 : * bytes should not cause reading any more data either.
1124 : * --------------------------------
1125 : */
1126 : ssize_t
1127 754 : pq_buffer_remaining_data(void)
1128 : {
1129 : Assert(PqRecvLength >= PqRecvPointer);
1130 754 : return (PqRecvLength - PqRecvPointer);
1131 : }
1132 :
1133 :
1134 : /* --------------------------------
1135 : * pq_startmsgread - begin reading a message from the client.
1136 : *
1137 : * This must be called before any of the pq_get* functions.
1138 : * --------------------------------
1139 : */
1140 : void
1141 4775514 : pq_startmsgread(void)
1142 : {
1143 : /*
1144 : * There shouldn't be a read active already, but let's check just to be
1145 : * sure.
1146 : */
1147 4775514 : if (PqCommReadingMsg)
1148 0 : ereport(FATAL,
1149 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1150 : errmsg("terminating connection because protocol synchronization was lost")));
1151 :
1152 4775514 : PqCommReadingMsg = true;
1153 4775514 : }
1154 :
1155 :
1156 : /* --------------------------------
1157 : * pq_endmsgread - finish reading message.
1158 : *
1159 : * This must be called after reading a message with pq_getbytes()
1160 : * and friends, to indicate that we have read the whole message.
1161 : * pq_getmessage() does this implicitly.
1162 : * --------------------------------
1163 : */
1164 : void
1165 3474580 : pq_endmsgread(void)
1166 : {
1167 : Assert(PqCommReadingMsg);
1168 :
1169 3474580 : PqCommReadingMsg = false;
1170 3474580 : }
1171 :
1172 : /* --------------------------------
1173 : * pq_is_reading_msg - are we currently reading a message?
1174 : *
1175 : * This is used in error recovery at the outer idle loop to detect if we have
1176 : * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1177 : * will check for that too, but it's nicer to detect it earlier.
1178 : * --------------------------------
1179 : */
1180 : bool
1181 43460 : pq_is_reading_msg(void)
1182 : {
1183 43460 : return PqCommReadingMsg;
1184 : }
1185 :
1186 : /* --------------------------------
1187 : * pq_getmessage - get a message with length word from connection
1188 : *
1189 : * The return value is placed in an expansible StringInfo, which has
1190 : * already been initialized by the caller.
1191 : * Only the message body is placed in the StringInfo; the length word
1192 : * is removed. Also, s->cursor is initialized to zero for convenience
1193 : * in scanning the message contents.
1194 : *
1195 : * maxlen is the upper limit on the length of the
1196 : * message we are willing to accept. We abort the connection (by
1197 : * returning EOF) if client tries to send more than that.
1198 : *
1199 : * returns 0 if OK, EOF if trouble
1200 : * --------------------------------
1201 : */
1202 : int
1203 1300668 : pq_getmessage(StringInfo s, int maxlen)
1204 : {
1205 : int32 len;
1206 :
1207 : Assert(PqCommReadingMsg);
1208 :
1209 1300668 : resetStringInfo(s);
1210 :
1211 : /* Read message length word */
1212 1300668 : if (pq_getbytes(&len, 4) == EOF)
1213 : {
1214 0 : ereport(COMMERROR,
1215 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1216 : errmsg("unexpected EOF within message length word")));
1217 0 : return EOF;
1218 : }
1219 :
1220 1300668 : len = pg_ntoh32(len);
1221 :
1222 1300668 : if (len < 4 || len > maxlen)
1223 : {
1224 0 : ereport(COMMERROR,
1225 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1226 : errmsg("invalid message length")));
1227 0 : return EOF;
1228 : }
1229 :
1230 1300668 : len -= 4; /* discount length itself */
1231 :
1232 1300668 : if (len > 0)
1233 : {
1234 : /*
1235 : * Allocate space for message. If we run out of room (ridiculously
1236 : * large message), we will elog(ERROR), but we want to discard the
1237 : * message body so as not to lose communication sync.
1238 : */
1239 1250842 : PG_TRY();
1240 : {
1241 1250842 : enlargeStringInfo(s, len);
1242 : }
1243 0 : PG_CATCH();
1244 : {
1245 0 : if (pq_discardbytes(len) == EOF)
1246 0 : ereport(COMMERROR,
1247 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1248 : errmsg("incomplete message from client")));
1249 :
1250 : /* we discarded the rest of the message so we're back in sync. */
1251 0 : PqCommReadingMsg = false;
1252 0 : PG_RE_THROW();
1253 : }
1254 1250842 : PG_END_TRY();
1255 :
1256 : /* And grab the message */
1257 1250842 : if (pq_getbytes(s->data, len) == EOF)
1258 : {
1259 0 : ereport(COMMERROR,
1260 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1261 : errmsg("incomplete message from client")));
1262 0 : return EOF;
1263 : }
1264 1250842 : s->len = len;
1265 : /* Place a trailing null per StringInfo convention */
1266 1250842 : s->data[len] = '\0';
1267 : }
1268 :
1269 : /* finished reading the message. */
1270 1300668 : PqCommReadingMsg = false;
1271 :
1272 1300668 : return 0;
1273 : }
1274 :
1275 :
1276 : static inline int
1277 41972418 : internal_putbytes(const void *b, size_t len)
1278 : {
1279 41972418 : const char *s = b;
1280 :
1281 84283788 : while (len > 0)
1282 : {
1283 : /* If buffer is full, then flush it out */
1284 42311378 : if (PqSendPointer >= PqSendBufferSize)
1285 : {
1286 399262 : socket_set_nonblocking(false);
1287 399262 : if (internal_flush())
1288 8 : return EOF;
1289 : }
1290 :
1291 : /*
1292 : * If the buffer is empty and data length is larger than the buffer
1293 : * size, send it without buffering. Otherwise, copy as much data as
1294 : * possible into the buffer.
1295 : */
1296 42311370 : if (len >= PqSendBufferSize && PqSendStart == PqSendPointer)
1297 222456 : {
1298 222456 : size_t start = 0;
1299 :
1300 222456 : socket_set_nonblocking(false);
1301 222456 : if (internal_flush_buffer(s, &start, &len))
1302 0 : return EOF;
1303 : }
1304 : else
1305 : {
1306 42088914 : size_t amount = PqSendBufferSize - PqSendPointer;
1307 :
1308 42088914 : if (amount > len)
1309 41506958 : amount = len;
1310 42088914 : memcpy(PqSendBuffer + PqSendPointer, s, amount);
1311 42088914 : PqSendPointer += amount;
1312 42088914 : s += amount;
1313 42088914 : len -= amount;
1314 : }
1315 : }
1316 :
1317 41972410 : return 0;
1318 : }
1319 :
1320 : /* --------------------------------
1321 : * socket_flush - flush pending output
1322 : *
1323 : * returns 0 if OK, EOF if trouble
1324 : * --------------------------------
1325 : */
1326 : static int
1327 838240 : socket_flush(void)
1328 : {
1329 : int res;
1330 :
1331 : /* No-op if reentrant call */
1332 838240 : if (PqCommBusy)
1333 0 : return 0;
1334 838240 : PqCommBusy = true;
1335 838240 : socket_set_nonblocking(false);
1336 838240 : res = internal_flush();
1337 838240 : PqCommBusy = false;
1338 838240 : return res;
1339 : }
1340 :
1341 : /* --------------------------------
1342 : * internal_flush - flush pending output
1343 : *
1344 : * Returns 0 if OK (meaning everything was sent, or operation would block
1345 : * and the socket is in non-blocking mode), or EOF if trouble.
1346 : * --------------------------------
1347 : */
1348 : static inline int
1349 1883432 : internal_flush(void)
1350 : {
1351 1883432 : return internal_flush_buffer(PqSendBuffer, &PqSendStart, &PqSendPointer);
1352 : }
1353 :
1354 : /* --------------------------------
1355 : * internal_flush_buffer - flush the given buffer content
1356 : *
1357 : * Returns 0 if OK (meaning everything was sent, or operation would block
1358 : * and the socket is in non-blocking mode), or EOF if trouble.
1359 : * --------------------------------
1360 : */
1361 : static pg_noinline int
1362 2105888 : internal_flush_buffer(const char *buf, size_t *start, size_t *end)
1363 : {
1364 : static int last_reported_send_errno = 0;
1365 :
1366 2105888 : const char *bufptr = buf + *start;
1367 2105888 : const char *bufend = buf + *end;
1368 :
1369 4172842 : while (bufptr < bufend)
1370 : {
1371 : int r;
1372 :
1373 2137298 : r = secure_write(MyProcPort, bufptr, bufend - bufptr);
1374 :
1375 2137298 : if (r <= 0)
1376 : {
1377 70344 : if (errno == EINTR)
1378 0 : continue; /* Ok if we were interrupted */
1379 :
1380 : /*
1381 : * Ok if no data writable without blocking, and the socket is in
1382 : * non-blocking mode.
1383 : */
1384 70344 : if (errno == EAGAIN ||
1385 84 : errno == EWOULDBLOCK)
1386 : {
1387 70260 : return 0;
1388 : }
1389 :
1390 : /*
1391 : * Careful: an ereport() that tries to write to the client would
1392 : * cause recursion to here, leading to stack overflow and core
1393 : * dump! This message must go *only* to the postmaster log.
1394 : *
1395 : * If a client disconnects while we're in the midst of output, we
1396 : * might write quite a bit of data before we get to a safe query
1397 : * abort point. So, suppress duplicate log messages.
1398 : */
1399 84 : if (errno != last_reported_send_errno)
1400 : {
1401 82 : last_reported_send_errno = errno;
1402 82 : ereport(COMMERROR,
1403 : (errcode_for_socket_access(),
1404 : errmsg("could not send data to client: %m")));
1405 : }
1406 :
1407 : /*
1408 : * We drop the buffered data anyway so that processing can
1409 : * continue, even though we'll probably quit soon. We also set a
1410 : * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1411 : * the connection.
1412 : */
1413 84 : *start = *end = 0;
1414 84 : ClientConnectionLost = 1;
1415 84 : InterruptPending = 1;
1416 84 : return EOF;
1417 : }
1418 :
1419 2066954 : last_reported_send_errno = 0; /* reset after any successful send */
1420 2066954 : bufptr += r;
1421 2066954 : *start += r;
1422 : }
1423 :
1424 2035544 : *start = *end = 0;
1425 2035544 : return 0;
1426 : }
1427 :
1428 : /* --------------------------------
1429 : * pq_flush_if_writable - flush pending output if writable without blocking
1430 : *
1431 : * Returns 0 if OK, or EOF if trouble.
1432 : * --------------------------------
1433 : */
1434 : static int
1435 3778762 : socket_flush_if_writable(void)
1436 : {
1437 : int res;
1438 :
1439 : /* Quick exit if nothing to do */
1440 3778762 : if (PqSendPointer == PqSendStart)
1441 3132832 : return 0;
1442 :
1443 : /* No-op if reentrant call */
1444 645930 : if (PqCommBusy)
1445 0 : return 0;
1446 :
1447 : /* Temporarily put the socket into non-blocking mode */
1448 645930 : socket_set_nonblocking(true);
1449 :
1450 645930 : PqCommBusy = true;
1451 645930 : res = internal_flush();
1452 645930 : PqCommBusy = false;
1453 645930 : return res;
1454 : }
1455 :
1456 : /* --------------------------------
1457 : * socket_is_send_pending - is there any pending data in the output buffer?
1458 : * --------------------------------
1459 : */
1460 : static bool
1461 8734098 : socket_is_send_pending(void)
1462 : {
1463 8734098 : return (PqSendStart < PqSendPointer);
1464 : }
1465 :
1466 : /* --------------------------------
1467 : * Message-level I/O routines begin here.
1468 : * --------------------------------
1469 : */
1470 :
1471 :
1472 : /* --------------------------------
1473 : * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1474 : *
1475 : * msgtype is a message type code to place before the message body.
1476 : *
1477 : * len is the length of the message body data at *s. A message length
1478 : * word (equal to len+4 because it counts itself too) is inserted by this
1479 : * routine.
1480 : *
1481 : * We suppress messages generated while pqcomm.c is busy. This
1482 : * avoids any possibility of messages being inserted within other
1483 : * messages. The only known trouble case arises if SIGQUIT occurs
1484 : * during a pqcomm.c routine --- quickdie() will try to send a warning
1485 : * message, and the most reasonable approach seems to be to drop it.
1486 : *
1487 : * returns 0 if OK, EOF if trouble
1488 : * --------------------------------
1489 : */
1490 : static int
1491 13990806 : socket_putmessage(char msgtype, const char *s, size_t len)
1492 : {
1493 : uint32 n32;
1494 :
1495 : Assert(msgtype != 0);
1496 :
1497 13990806 : if (PqCommBusy)
1498 0 : return 0;
1499 13990806 : PqCommBusy = true;
1500 13990806 : if (internal_putbytes(&msgtype, 1))
1501 0 : goto fail;
1502 :
1503 13990806 : n32 = pg_hton32((uint32) (len + 4));
1504 13990806 : if (internal_putbytes(&n32, 4))
1505 0 : goto fail;
1506 :
1507 13990806 : if (internal_putbytes(s, len))
1508 8 : goto fail;
1509 13990798 : PqCommBusy = false;
1510 13990798 : return 0;
1511 :
1512 8 : fail:
1513 8 : PqCommBusy = false;
1514 8 : return EOF;
1515 : }
1516 :
1517 : /* --------------------------------
1518 : * pq_putmessage_noblock - like pq_putmessage, but never blocks
1519 : *
1520 : * If the output buffer is too small to hold the message, the buffer
1521 : * is enlarged.
1522 : */
1523 : static void
1524 575412 : socket_putmessage_noblock(char msgtype, const char *s, size_t len)
1525 : {
1526 : int res PG_USED_FOR_ASSERTS_ONLY;
1527 : int required;
1528 :
1529 : /*
1530 : * Ensure we have enough space in the output buffer for the message header
1531 : * as well as the message itself.
1532 : */
1533 575412 : required = PqSendPointer + 1 + 4 + len;
1534 575412 : if (required > PqSendBufferSize)
1535 : {
1536 950 : PqSendBuffer = repalloc(PqSendBuffer, required);
1537 950 : PqSendBufferSize = required;
1538 : }
1539 575412 : res = pq_putmessage(msgtype, s, len);
1540 : Assert(res == 0); /* should not fail when the message fits in
1541 : * buffer */
1542 575412 : }
1543 :
1544 : /* --------------------------------
1545 : * pq_putmessage_v2 - send a message in protocol version 2
1546 : *
1547 : * msgtype is a message type code to place before the message body.
1548 : *
1549 : * We no longer support protocol version 2, but we have kept this
1550 : * function so that if a client tries to connect with protocol version 2,
1551 : * as a courtesy we can still send the "unsupported protocol version"
1552 : * error to the client in the old format.
1553 : *
1554 : * Like in pq_putmessage(), we suppress messages generated while
1555 : * pqcomm.c is busy.
1556 : *
1557 : * returns 0 if OK, EOF if trouble
1558 : * --------------------------------
1559 : */
1560 : int
1561 0 : pq_putmessage_v2(char msgtype, const char *s, size_t len)
1562 : {
1563 : Assert(msgtype != 0);
1564 :
1565 0 : if (PqCommBusy)
1566 0 : return 0;
1567 0 : PqCommBusy = true;
1568 0 : if (internal_putbytes(&msgtype, 1))
1569 0 : goto fail;
1570 :
1571 0 : if (internal_putbytes(s, len))
1572 0 : goto fail;
1573 0 : PqCommBusy = false;
1574 0 : return 0;
1575 :
1576 0 : fail:
1577 0 : PqCommBusy = false;
1578 0 : return EOF;
1579 : }
1580 :
1581 : /*
1582 : * Support for TCP Keepalive parameters
1583 : */
1584 :
1585 : /*
1586 : * On Windows, we need to set both idle and interval at the same time.
1587 : * We also cannot reset them to the default (setting to zero will
1588 : * actually set them to zero, not default), therefore we fallback to
1589 : * the out-of-the-box default instead.
1590 : */
1591 : #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1592 : static int
1593 : pq_setkeepaliveswin32(Port *port, int idle, int interval)
1594 : {
1595 : struct tcp_keepalive ka;
1596 : DWORD retsize;
1597 :
1598 : if (idle <= 0)
1599 : idle = 2 * 60 * 60; /* default = 2 hours */
1600 : if (interval <= 0)
1601 : interval = 1; /* default = 1 second */
1602 :
1603 : ka.onoff = 1;
1604 : ka.keepalivetime = idle * 1000;
1605 : ka.keepaliveinterval = interval * 1000;
1606 :
1607 : if (WSAIoctl(port->sock,
1608 : SIO_KEEPALIVE_VALS,
1609 : (LPVOID) &ka,
1610 : sizeof(ka),
1611 : NULL,
1612 : 0,
1613 : &retsize,
1614 : NULL,
1615 : NULL)
1616 : != 0)
1617 : {
1618 : ereport(LOG,
1619 : (errmsg("%s(%s) failed: error code %d",
1620 : "WSAIoctl", "SIO_KEEPALIVE_VALS", WSAGetLastError())));
1621 : return STATUS_ERROR;
1622 : }
1623 : if (port->keepalives_idle != idle)
1624 : port->keepalives_idle = idle;
1625 : if (port->keepalives_interval != interval)
1626 : port->keepalives_interval = interval;
1627 : return STATUS_OK;
1628 : }
1629 : #endif
1630 :
1631 : int
1632 3450 : pq_getkeepalivesidle(Port *port)
1633 : {
1634 : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1635 3450 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1636 3450 : return 0;
1637 :
1638 0 : if (port->keepalives_idle != 0)
1639 0 : return port->keepalives_idle;
1640 :
1641 0 : if (port->default_keepalives_idle == 0)
1642 : {
1643 : #ifndef WIN32
1644 0 : socklen_t size = sizeof(port->default_keepalives_idle);
1645 :
1646 0 : if (getsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1647 0 : (char *) &port->default_keepalives_idle,
1648 : &size) < 0)
1649 : {
1650 0 : ereport(LOG,
1651 : (errmsg("%s(%s) failed: %m", "getsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
1652 0 : port->default_keepalives_idle = -1; /* don't know */
1653 : }
1654 : #else /* WIN32 */
1655 : /* We can't get the defaults on Windows, so return "don't know" */
1656 : port->default_keepalives_idle = -1;
1657 : #endif /* WIN32 */
1658 : }
1659 :
1660 0 : return port->default_keepalives_idle;
1661 : #else
1662 : return 0;
1663 : #endif
1664 : }
1665 :
1666 : int
1667 2782 : pq_setkeepalivesidle(int idle, Port *port)
1668 : {
1669 2782 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1670 2204 : return STATUS_OK;
1671 :
1672 : /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1673 : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1674 578 : if (idle == port->keepalives_idle)
1675 578 : return STATUS_OK;
1676 :
1677 : #ifndef WIN32
1678 0 : if (port->default_keepalives_idle <= 0)
1679 : {
1680 0 : if (pq_getkeepalivesidle(port) < 0)
1681 : {
1682 0 : if (idle == 0)
1683 0 : return STATUS_OK; /* default is set but unknown */
1684 : else
1685 0 : return STATUS_ERROR;
1686 : }
1687 : }
1688 :
1689 0 : if (idle == 0)
1690 0 : idle = port->default_keepalives_idle;
1691 :
1692 0 : if (setsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1693 : (char *) &idle, sizeof(idle)) < 0)
1694 : {
1695 0 : ereport(LOG,
1696 : (errmsg("%s(%s) failed: %m", "setsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
1697 0 : return STATUS_ERROR;
1698 : }
1699 :
1700 0 : port->keepalives_idle = idle;
1701 : #else /* WIN32 */
1702 : return pq_setkeepaliveswin32(port, idle, port->keepalives_interval);
1703 : #endif
1704 : #else
1705 : if (idle != 0)
1706 : {
1707 : ereport(LOG,
1708 : (errmsg("setting the keepalive idle time is not supported")));
1709 : return STATUS_ERROR;
1710 : }
1711 : #endif
1712 :
1713 0 : return STATUS_OK;
1714 : }
1715 :
1716 : int
1717 3450 : pq_getkeepalivesinterval(Port *port)
1718 : {
1719 : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1720 3450 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1721 3450 : return 0;
1722 :
1723 0 : if (port->keepalives_interval != 0)
1724 0 : return port->keepalives_interval;
1725 :
1726 0 : if (port->default_keepalives_interval == 0)
1727 : {
1728 : #ifndef WIN32
1729 0 : socklen_t size = sizeof(port->default_keepalives_interval);
1730 :
1731 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1732 0 : (char *) &port->default_keepalives_interval,
1733 : &size) < 0)
1734 : {
1735 0 : ereport(LOG,
1736 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPINTVL")));
1737 0 : port->default_keepalives_interval = -1; /* don't know */
1738 : }
1739 : #else
1740 : /* We can't get the defaults on Windows, so return "don't know" */
1741 : port->default_keepalives_interval = -1;
1742 : #endif /* WIN32 */
1743 : }
1744 :
1745 0 : return port->default_keepalives_interval;
1746 : #else
1747 : return 0;
1748 : #endif
1749 : }
1750 :
1751 : int
1752 2782 : pq_setkeepalivesinterval(int interval, Port *port)
1753 : {
1754 2782 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1755 2204 : return STATUS_OK;
1756 :
1757 : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1758 578 : if (interval == port->keepalives_interval)
1759 578 : return STATUS_OK;
1760 :
1761 : #ifndef WIN32
1762 0 : if (port->default_keepalives_interval <= 0)
1763 : {
1764 0 : if (pq_getkeepalivesinterval(port) < 0)
1765 : {
1766 0 : if (interval == 0)
1767 0 : return STATUS_OK; /* default is set but unknown */
1768 : else
1769 0 : return STATUS_ERROR;
1770 : }
1771 : }
1772 :
1773 0 : if (interval == 0)
1774 0 : interval = port->default_keepalives_interval;
1775 :
1776 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1777 : (char *) &interval, sizeof(interval)) < 0)
1778 : {
1779 0 : ereport(LOG,
1780 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPINTVL")));
1781 0 : return STATUS_ERROR;
1782 : }
1783 :
1784 0 : port->keepalives_interval = interval;
1785 : #else /* WIN32 */
1786 : return pq_setkeepaliveswin32(port, port->keepalives_idle, interval);
1787 : #endif
1788 : #else
1789 : if (interval != 0)
1790 : {
1791 : ereport(LOG,
1792 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPINTVL")));
1793 : return STATUS_ERROR;
1794 : }
1795 : #endif
1796 :
1797 0 : return STATUS_OK;
1798 : }
1799 :
1800 : int
1801 3450 : pq_getkeepalivescount(Port *port)
1802 : {
1803 : #ifdef TCP_KEEPCNT
1804 3450 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1805 3450 : return 0;
1806 :
1807 0 : if (port->keepalives_count != 0)
1808 0 : return port->keepalives_count;
1809 :
1810 0 : if (port->default_keepalives_count == 0)
1811 : {
1812 0 : socklen_t size = sizeof(port->default_keepalives_count);
1813 :
1814 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1815 0 : (char *) &port->default_keepalives_count,
1816 : &size) < 0)
1817 : {
1818 0 : ereport(LOG,
1819 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPCNT")));
1820 0 : port->default_keepalives_count = -1; /* don't know */
1821 : }
1822 : }
1823 :
1824 0 : return port->default_keepalives_count;
1825 : #else
1826 : return 0;
1827 : #endif
1828 : }
1829 :
1830 : int
1831 2782 : pq_setkeepalivescount(int count, Port *port)
1832 : {
1833 2782 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1834 2204 : return STATUS_OK;
1835 :
1836 : #ifdef TCP_KEEPCNT
1837 578 : if (count == port->keepalives_count)
1838 578 : return STATUS_OK;
1839 :
1840 0 : if (port->default_keepalives_count <= 0)
1841 : {
1842 0 : if (pq_getkeepalivescount(port) < 0)
1843 : {
1844 0 : if (count == 0)
1845 0 : return STATUS_OK; /* default is set but unknown */
1846 : else
1847 0 : return STATUS_ERROR;
1848 : }
1849 : }
1850 :
1851 0 : if (count == 0)
1852 0 : count = port->default_keepalives_count;
1853 :
1854 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1855 : (char *) &count, sizeof(count)) < 0)
1856 : {
1857 0 : ereport(LOG,
1858 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPCNT")));
1859 0 : return STATUS_ERROR;
1860 : }
1861 :
1862 0 : port->keepalives_count = count;
1863 : #else
1864 : if (count != 0)
1865 : {
1866 : ereport(LOG,
1867 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPCNT")));
1868 : return STATUS_ERROR;
1869 : }
1870 : #endif
1871 :
1872 0 : return STATUS_OK;
1873 : }
1874 :
1875 : int
1876 3450 : pq_gettcpusertimeout(Port *port)
1877 : {
1878 : #ifdef TCP_USER_TIMEOUT
1879 3450 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1880 3450 : return 0;
1881 :
1882 0 : if (port->tcp_user_timeout != 0)
1883 0 : return port->tcp_user_timeout;
1884 :
1885 0 : if (port->default_tcp_user_timeout == 0)
1886 : {
1887 0 : socklen_t size = sizeof(port->default_tcp_user_timeout);
1888 :
1889 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1890 0 : (char *) &port->default_tcp_user_timeout,
1891 : &size) < 0)
1892 : {
1893 0 : ereport(LOG,
1894 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_USER_TIMEOUT")));
1895 0 : port->default_tcp_user_timeout = -1; /* don't know */
1896 : }
1897 : }
1898 :
1899 0 : return port->default_tcp_user_timeout;
1900 : #else
1901 : return 0;
1902 : #endif
1903 : }
1904 :
1905 : int
1906 2782 : pq_settcpusertimeout(int timeout, Port *port)
1907 : {
1908 2782 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1909 2204 : return STATUS_OK;
1910 :
1911 : #ifdef TCP_USER_TIMEOUT
1912 578 : if (timeout == port->tcp_user_timeout)
1913 578 : return STATUS_OK;
1914 :
1915 0 : if (port->default_tcp_user_timeout <= 0)
1916 : {
1917 0 : if (pq_gettcpusertimeout(port) < 0)
1918 : {
1919 0 : if (timeout == 0)
1920 0 : return STATUS_OK; /* default is set but unknown */
1921 : else
1922 0 : return STATUS_ERROR;
1923 : }
1924 : }
1925 :
1926 0 : if (timeout == 0)
1927 0 : timeout = port->default_tcp_user_timeout;
1928 :
1929 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1930 : (char *) &timeout, sizeof(timeout)) < 0)
1931 : {
1932 0 : ereport(LOG,
1933 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_USER_TIMEOUT")));
1934 0 : return STATUS_ERROR;
1935 : }
1936 :
1937 0 : port->tcp_user_timeout = timeout;
1938 : #else
1939 : if (timeout != 0)
1940 : {
1941 : ereport(LOG,
1942 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_USER_TIMEOUT")));
1943 : return STATUS_ERROR;
1944 : }
1945 : #endif
1946 :
1947 0 : return STATUS_OK;
1948 : }
1949 :
1950 : /*
1951 : * GUC assign_hook for tcp_keepalives_idle
1952 : */
1953 : void
1954 2204 : assign_tcp_keepalives_idle(int newval, void *extra)
1955 : {
1956 : /*
1957 : * The kernel API provides no way to test a value without setting it; and
1958 : * once we set it we might fail to unset it. So there seems little point
1959 : * in fully implementing the check-then-assign GUC API for these
1960 : * variables. Instead we just do the assignment on demand.
1961 : * pq_setkeepalivesidle reports any problems via ereport(LOG).
1962 : *
1963 : * This approach means that the GUC value might have little to do with the
1964 : * actual kernel value, so we use a show_hook that retrieves the kernel
1965 : * value rather than trusting GUC's copy.
1966 : */
1967 2204 : (void) pq_setkeepalivesidle(newval, MyProcPort);
1968 2204 : }
1969 :
1970 : /*
1971 : * GUC show_hook for tcp_keepalives_idle
1972 : */
1973 : const char *
1974 3450 : show_tcp_keepalives_idle(void)
1975 : {
1976 : /* See comments in assign_tcp_keepalives_idle */
1977 : static char nbuf[16];
1978 :
1979 3450 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesidle(MyProcPort));
1980 3450 : return nbuf;
1981 : }
1982 :
1983 : /*
1984 : * GUC assign_hook for tcp_keepalives_interval
1985 : */
1986 : void
1987 2204 : assign_tcp_keepalives_interval(int newval, void *extra)
1988 : {
1989 : /* See comments in assign_tcp_keepalives_idle */
1990 2204 : (void) pq_setkeepalivesinterval(newval, MyProcPort);
1991 2204 : }
1992 :
1993 : /*
1994 : * GUC show_hook for tcp_keepalives_interval
1995 : */
1996 : const char *
1997 3450 : show_tcp_keepalives_interval(void)
1998 : {
1999 : /* See comments in assign_tcp_keepalives_idle */
2000 : static char nbuf[16];
2001 :
2002 3450 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesinterval(MyProcPort));
2003 3450 : return nbuf;
2004 : }
2005 :
2006 : /*
2007 : * GUC assign_hook for tcp_keepalives_count
2008 : */
2009 : void
2010 2204 : assign_tcp_keepalives_count(int newval, void *extra)
2011 : {
2012 : /* See comments in assign_tcp_keepalives_idle */
2013 2204 : (void) pq_setkeepalivescount(newval, MyProcPort);
2014 2204 : }
2015 :
2016 : /*
2017 : * GUC show_hook for tcp_keepalives_count
2018 : */
2019 : const char *
2020 3450 : show_tcp_keepalives_count(void)
2021 : {
2022 : /* See comments in assign_tcp_keepalives_idle */
2023 : static char nbuf[16];
2024 :
2025 3450 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivescount(MyProcPort));
2026 3450 : return nbuf;
2027 : }
2028 :
2029 : /*
2030 : * GUC assign_hook for tcp_user_timeout
2031 : */
2032 : void
2033 2204 : assign_tcp_user_timeout(int newval, void *extra)
2034 : {
2035 : /* See comments in assign_tcp_keepalives_idle */
2036 2204 : (void) pq_settcpusertimeout(newval, MyProcPort);
2037 2204 : }
2038 :
2039 : /*
2040 : * GUC show_hook for tcp_user_timeout
2041 : */
2042 : const char *
2043 3450 : show_tcp_user_timeout(void)
2044 : {
2045 : /* See comments in assign_tcp_keepalives_idle */
2046 : static char nbuf[16];
2047 :
2048 3450 : snprintf(nbuf, sizeof(nbuf), "%d", pq_gettcpusertimeout(MyProcPort));
2049 3450 : return nbuf;
2050 : }
2051 :
2052 : /*
2053 : * Check if the client is still connected.
2054 : */
2055 : bool
2056 0 : pq_check_connection(void)
2057 : {
2058 : WaitEvent events[FeBeWaitSetNEvents];
2059 : int rc;
2060 :
2061 : /*
2062 : * It's OK to modify the socket event filter without restoring, because
2063 : * all FeBeWaitSet socket wait sites do the same.
2064 : */
2065 0 : ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetSocketPos, WL_SOCKET_CLOSED, NULL);
2066 :
2067 0 : retry:
2068 0 : rc = WaitEventSetWait(FeBeWaitSet, 0, events, lengthof(events), 0);
2069 0 : for (int i = 0; i < rc; ++i)
2070 : {
2071 0 : if (events[i].events & WL_SOCKET_CLOSED)
2072 0 : return false;
2073 0 : if (events[i].events & WL_LATCH_SET)
2074 : {
2075 : /*
2076 : * A latch event might be preventing other events from being
2077 : * reported. Reset it and poll again. No need to restore it
2078 : * because no code should expect latches to survive across
2079 : * CHECK_FOR_INTERRUPTS().
2080 : */
2081 0 : ResetLatch(MyLatch);
2082 0 : goto retry;
2083 : }
2084 : }
2085 :
2086 0 : return true;
2087 : }
|
