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) 1996-2026, 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 "storage/latch.h"
82 : #include "utils/guc_hooks.h"
83 : #include "utils/memutils.h"
84 :
85 : /*
86 : * Cope with the various platform-specific ways to spell TCP keepalive socket
87 : * options. This doesn't cover Windows, which as usual does its own thing.
88 : */
89 : #if defined(TCP_KEEPIDLE)
90 : /* TCP_KEEPIDLE is the name of this option on Linux and *BSD */
91 : #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPIDLE
92 : #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPIDLE"
93 : #elif defined(TCP_KEEPALIVE_THRESHOLD)
94 : /* TCP_KEEPALIVE_THRESHOLD is the name of this option on Solaris >= 11 */
95 : #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE_THRESHOLD
96 : #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE_THRESHOLD"
97 : #elif defined(TCP_KEEPALIVE) && defined(__darwin__)
98 : /* TCP_KEEPALIVE is the name of this option on macOS */
99 : /* Caution: Solaris has this symbol but it means something different */
100 : #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE
101 : #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE"
102 : #endif
103 :
104 : /*
105 : * Configuration options
106 : */
107 : int Unix_socket_permissions;
108 : char *Unix_socket_group;
109 :
110 : /* Where the Unix socket files are (list of palloc'd strings) */
111 : static List *sock_paths = NIL;
112 :
113 : /*
114 : * Buffers for low-level I/O.
115 : *
116 : * The receive buffer is fixed size. Send buffer is usually 8k, but can be
117 : * enlarged by pq_putmessage_noblock() if the message doesn't fit otherwise.
118 : */
119 :
120 : #define PQ_SEND_BUFFER_SIZE 8192
121 : #define PQ_RECV_BUFFER_SIZE 8192
122 :
123 : static char *PqSendBuffer;
124 : static int PqSendBufferSize; /* Size send buffer */
125 : static size_t PqSendPointer; /* Next index to store a byte in PqSendBuffer */
126 : static size_t PqSendStart; /* Next index to send a byte in PqSendBuffer */
127 :
128 : static char PqRecvBuffer[PQ_RECV_BUFFER_SIZE];
129 : static int PqRecvPointer; /* Next index to read a byte from PqRecvBuffer */
130 : static int PqRecvLength; /* End of data available in PqRecvBuffer */
131 :
132 : /*
133 : * Message status
134 : */
135 : static bool PqCommBusy; /* busy sending data to the client */
136 : static bool PqCommReadingMsg; /* in the middle of reading a message */
137 :
138 :
139 : /* Internal functions */
140 : static void socket_comm_reset(void);
141 : static void socket_close(int code, Datum arg);
142 : static void socket_set_nonblocking(bool nonblocking);
143 : static int socket_flush(void);
144 : static int socket_flush_if_writable(void);
145 : static bool socket_is_send_pending(void);
146 : static int socket_putmessage(char msgtype, const char *s, size_t len);
147 : static void socket_putmessage_noblock(char msgtype, const char *s, size_t len);
148 : static inline int internal_putbytes(const void *b, size_t len);
149 : static inline int internal_flush(void);
150 : static pg_noinline int internal_flush_buffer(const char *buf, size_t *start,
151 : size_t *end);
152 :
153 : static int Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath);
154 : static int Setup_AF_UNIX(const char *sock_path);
155 :
156 : static const PQcommMethods PqCommSocketMethods = {
157 : .comm_reset = socket_comm_reset,
158 : .flush = socket_flush,
159 : .flush_if_writable = socket_flush_if_writable,
160 : .is_send_pending = socket_is_send_pending,
161 : .putmessage = socket_putmessage,
162 : .putmessage_noblock = socket_putmessage_noblock
163 : };
164 :
165 : const PQcommMethods *PqCommMethods = &PqCommSocketMethods;
166 :
167 : WaitEventSet *FeBeWaitSet;
168 :
169 :
170 : /* --------------------------------
171 : * pq_init - initialize libpq at backend startup
172 : * --------------------------------
173 : */
174 : Port *
175 14475 : pq_init(ClientSocket *client_sock)
176 : {
177 : Port *port;
178 : int socket_pos PG_USED_FOR_ASSERTS_ONLY;
179 : int latch_pos PG_USED_FOR_ASSERTS_ONLY;
180 :
181 : /* allocate the Port struct and copy the ClientSocket contents to it */
182 14475 : port = palloc0_object(Port);
183 14475 : port->sock = client_sock->sock;
184 14475 : memcpy(&port->raddr.addr, &client_sock->raddr.addr, client_sock->raddr.salen);
185 14475 : port->raddr.salen = client_sock->raddr.salen;
186 :
187 : /* fill in the server (local) address */
188 14475 : port->laddr.salen = sizeof(port->laddr.addr);
189 14475 : if (getsockname(port->sock,
190 14475 : (struct sockaddr *) &port->laddr.addr,
191 : &port->laddr.salen) < 0)
192 : {
193 0 : ereport(FATAL,
194 : (errmsg("%s() failed: %m", "getsockname")));
195 : }
196 :
197 : /* select NODELAY and KEEPALIVE options if it's a TCP connection */
198 14475 : if (port->laddr.addr.ss_family != AF_UNIX)
199 : {
200 : int on;
201 : #ifdef WIN32
202 : int oldopt;
203 : int optlen;
204 : int newopt;
205 : #endif
206 :
207 : #ifdef TCP_NODELAY
208 335 : on = 1;
209 335 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_NODELAY,
210 : (char *) &on, sizeof(on)) < 0)
211 : {
212 0 : ereport(FATAL,
213 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_NODELAY")));
214 : }
215 : #endif
216 335 : on = 1;
217 335 : if (setsockopt(port->sock, SOL_SOCKET, SO_KEEPALIVE,
218 : (char *) &on, sizeof(on)) < 0)
219 : {
220 0 : ereport(FATAL,
221 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_KEEPALIVE")));
222 : }
223 :
224 : #ifdef WIN32
225 :
226 : /*
227 : * This is a Win32 socket optimization. The OS send buffer should be
228 : * large enough to send the whole Postgres send buffer in one go, or
229 : * performance suffers. The Postgres send buffer can be enlarged if a
230 : * very large message needs to be sent, but we won't attempt to
231 : * enlarge the OS buffer if that happens, so somewhat arbitrarily
232 : * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
233 : * (That's 32kB with the current default).
234 : *
235 : * The default OS buffer size used to be 8kB in earlier Windows
236 : * versions, but was raised to 64kB in Windows 2012. So it shouldn't
237 : * be necessary to change it in later versions anymore. Changing it
238 : * unnecessarily can even reduce performance, because setting
239 : * SO_SNDBUF in the application disables the "dynamic send buffering"
240 : * feature that was introduced in Windows 7. So before fiddling with
241 : * SO_SNDBUF, check if the current buffer size is already large enough
242 : * and only increase it if necessary.
243 : *
244 : * See https://support.microsoft.com/kb/823764/EN-US/ and
245 : * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
246 : */
247 : optlen = sizeof(oldopt);
248 : if (getsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &oldopt,
249 : &optlen) < 0)
250 : {
251 : ereport(FATAL,
252 : (errmsg("%s(%s) failed: %m", "getsockopt", "SO_SNDBUF")));
253 : }
254 : newopt = PQ_SEND_BUFFER_SIZE * 4;
255 : if (oldopt < newopt)
256 : {
257 : if (setsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &newopt,
258 : sizeof(newopt)) < 0)
259 : {
260 : ereport(FATAL,
261 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_SNDBUF")));
262 : }
263 : }
264 : #endif
265 :
266 : /*
267 : * Also apply the current keepalive parameters. If we fail to set a
268 : * parameter, don't error out, because these aren't universally
269 : * supported. (Note: you might think we need to reset the GUC
270 : * variables to 0 in such a case, but it's not necessary because the
271 : * show hooks for these variables report the truth anyway.)
272 : */
273 335 : (void) pq_setkeepalivesidle(tcp_keepalives_idle, port);
274 335 : (void) pq_setkeepalivesinterval(tcp_keepalives_interval, port);
275 335 : (void) pq_setkeepalivescount(tcp_keepalives_count, port);
276 335 : (void) pq_settcpusertimeout(tcp_user_timeout, port);
277 : }
278 :
279 : /* initialize state variables */
280 14475 : PqSendBufferSize = PQ_SEND_BUFFER_SIZE;
281 14475 : PqSendBuffer = MemoryContextAlloc(TopMemoryContext, PqSendBufferSize);
282 14475 : PqSendPointer = PqSendStart = PqRecvPointer = PqRecvLength = 0;
283 14475 : PqCommBusy = false;
284 14475 : PqCommReadingMsg = false;
285 :
286 : /* set up process-exit hook to close the socket */
287 14475 : on_proc_exit(socket_close, 0);
288 :
289 : /*
290 : * In backends (as soon as forked) we operate the underlying socket in
291 : * nonblocking mode and use latches to implement blocking semantics if
292 : * needed. That allows us to provide safely interruptible reads and
293 : * writes.
294 : */
295 : #ifndef WIN32
296 14475 : if (!pg_set_noblock(port->sock))
297 0 : ereport(FATAL,
298 : (errmsg("could not set socket to nonblocking mode: %m")));
299 : #endif
300 :
301 : #ifndef WIN32
302 :
303 : /* Don't give the socket to any subprograms we execute. */
304 14475 : if (fcntl(port->sock, F_SETFD, FD_CLOEXEC) < 0)
305 0 : elog(FATAL, "fcntl(F_SETFD) failed on socket: %m");
306 : #endif
307 :
308 14475 : FeBeWaitSet = CreateWaitEventSet(NULL, FeBeWaitSetNEvents);
309 14475 : socket_pos = AddWaitEventToSet(FeBeWaitSet, WL_SOCKET_WRITEABLE,
310 : port->sock, NULL, NULL);
311 14475 : latch_pos = AddWaitEventToSet(FeBeWaitSet, WL_LATCH_SET, PGINVALID_SOCKET,
312 : MyLatch, NULL);
313 14475 : AddWaitEventToSet(FeBeWaitSet, WL_POSTMASTER_DEATH, PGINVALID_SOCKET,
314 : NULL, NULL);
315 :
316 : /*
317 : * The event positions match the order we added them, but let's sanity
318 : * check them to be sure.
319 : */
320 : Assert(socket_pos == FeBeWaitSetSocketPos);
321 : Assert(latch_pos == FeBeWaitSetLatchPos);
322 :
323 14475 : return port;
324 : }
325 :
326 : /* --------------------------------
327 : * socket_comm_reset - reset libpq during error recovery
328 : *
329 : * This is called from error recovery at the outer idle loop. It's
330 : * just to get us out of trouble if we somehow manage to elog() from
331 : * inside a pqcomm.c routine (which ideally will never happen, but...)
332 : * --------------------------------
333 : */
334 : static void
335 30210 : socket_comm_reset(void)
336 : {
337 : /* Do not throw away pending data, but do reset the busy flag */
338 30210 : PqCommBusy = false;
339 30210 : }
340 :
341 : /* --------------------------------
342 : * socket_close - shutdown libpq at backend exit
343 : *
344 : * This is the one pg_on_exit_callback in place during BackendInitialize().
345 : * That function's unusual signal handling constrains that this callback be
346 : * safe to run at any instant.
347 : * --------------------------------
348 : */
349 : static void
350 14475 : socket_close(int code, Datum arg)
351 : {
352 : /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
353 14475 : if (MyProcPort != NULL)
354 : {
355 : #ifdef ENABLE_GSS
356 : /*
357 : * Shutdown GSSAPI layer. This section does nothing when interrupting
358 : * BackendInitialize(), because pg_GSS_recvauth() makes first use of
359 : * "ctx" and "cred".
360 : *
361 : * Note that we don't bother to free MyProcPort->gss, since we're
362 : * about to exit anyway.
363 : */
364 : if (MyProcPort->gss)
365 : {
366 : OM_uint32 min_s;
367 :
368 : if (MyProcPort->gss->ctx != GSS_C_NO_CONTEXT)
369 : gss_delete_sec_context(&min_s, &MyProcPort->gss->ctx, NULL);
370 :
371 : if (MyProcPort->gss->cred != GSS_C_NO_CREDENTIAL)
372 : gss_release_cred(&min_s, &MyProcPort->gss->cred);
373 : }
374 : #endif /* ENABLE_GSS */
375 :
376 : /*
377 : * Cleanly shut down SSL layer. Nowhere else does a postmaster child
378 : * call this, so this is safe when interrupting BackendInitialize().
379 : */
380 14475 : secure_close(MyProcPort);
381 :
382 : /*
383 : * Formerly we did an explicit close() here, but it seems better to
384 : * leave the socket open until the process dies. This allows clients
385 : * to perform a "synchronous close" if they care --- wait till the
386 : * transport layer reports connection closure, and you can be sure the
387 : * backend has exited.
388 : *
389 : * We do set sock to PGINVALID_SOCKET to prevent any further I/O,
390 : * though.
391 : */
392 14475 : MyProcPort->sock = PGINVALID_SOCKET;
393 : }
394 14475 : }
395 :
396 :
397 :
398 : /* --------------------------------
399 : * Postmaster functions to handle sockets.
400 : * --------------------------------
401 : */
402 :
403 : /*
404 : * ListenServerPort -- open a "listening" port to accept connections.
405 : *
406 : * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
407 : * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
408 : * specified. For TCP ports, hostName is either NULL for all interfaces or
409 : * the interface to listen on, and unixSocketDir is ignored (can be NULL).
410 : *
411 : * Successfully opened sockets are appended to the ListenSockets[] array. On
412 : * entry, *NumListenSockets holds the number of elements currently in the
413 : * array, and it is updated to reflect the opened sockets. MaxListen is the
414 : * allocated size of the array.
415 : *
416 : * RETURNS: STATUS_OK or STATUS_ERROR
417 : */
418 : int
419 991 : ListenServerPort(int family, const char *hostName, unsigned short portNumber,
420 : const char *unixSocketDir,
421 : pgsocket ListenSockets[], int *NumListenSockets, int MaxListen)
422 : {
423 : pgsocket fd;
424 : int err;
425 : int maxconn;
426 : int ret;
427 : char portNumberStr[32];
428 : const char *familyDesc;
429 : char familyDescBuf[64];
430 : const char *addrDesc;
431 : char addrBuf[NI_MAXHOST];
432 : char *service;
433 991 : struct addrinfo *addrs = NULL,
434 : *addr;
435 : struct addrinfo hint;
436 991 : int added = 0;
437 : char unixSocketPath[MAXPGPATH];
438 : #if !defined(WIN32) || defined(IPV6_V6ONLY)
439 991 : int one = 1;
440 : #endif
441 :
442 : /* Initialize hint structure */
443 6937 : MemSet(&hint, 0, sizeof(hint));
444 991 : hint.ai_family = family;
445 991 : hint.ai_flags = AI_PASSIVE;
446 991 : hint.ai_socktype = SOCK_STREAM;
447 :
448 991 : if (family == AF_UNIX)
449 : {
450 : /*
451 : * Create unixSocketPath from portNumber and unixSocketDir and lock
452 : * that file path
453 : */
454 950 : UNIXSOCK_PATH(unixSocketPath, portNumber, unixSocketDir);
455 950 : if (strlen(unixSocketPath) >= UNIXSOCK_PATH_BUFLEN)
456 : {
457 0 : ereport(LOG,
458 : (errmsg("Unix-domain socket path \"%s\" is too long (maximum %zu bytes)",
459 : unixSocketPath,
460 : (UNIXSOCK_PATH_BUFLEN - 1))));
461 0 : return STATUS_ERROR;
462 : }
463 950 : if (Lock_AF_UNIX(unixSocketDir, unixSocketPath) != STATUS_OK)
464 0 : return STATUS_ERROR;
465 950 : service = unixSocketPath;
466 : }
467 : else
468 : {
469 41 : snprintf(portNumberStr, sizeof(portNumberStr), "%d", portNumber);
470 41 : service = portNumberStr;
471 : }
472 :
473 991 : ret = pg_getaddrinfo_all(hostName, service, &hint, &addrs);
474 991 : if (ret || !addrs)
475 : {
476 0 : if (hostName)
477 0 : ereport(LOG,
478 : (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
479 : hostName, service, gai_strerror(ret))));
480 : else
481 0 : ereport(LOG,
482 : (errmsg("could not translate service \"%s\" to address: %s",
483 : service, gai_strerror(ret))));
484 0 : if (addrs)
485 0 : pg_freeaddrinfo_all(hint.ai_family, addrs);
486 0 : return STATUS_ERROR;
487 : }
488 :
489 1983 : for (addr = addrs; addr; addr = addr->ai_next)
490 : {
491 992 : if (family != AF_UNIX && addr->ai_family == AF_UNIX)
492 : {
493 : /*
494 : * Only set up a unix domain socket when they really asked for it.
495 : * The service/port is different in that case.
496 : */
497 0 : continue;
498 : }
499 :
500 : /* See if there is still room to add 1 more socket. */
501 992 : if (*NumListenSockets == MaxListen)
502 : {
503 0 : ereport(LOG,
504 : (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
505 : MaxListen)));
506 0 : break;
507 : }
508 :
509 : /* set up address family name for log messages */
510 992 : switch (addr->ai_family)
511 : {
512 41 : case AF_INET:
513 41 : familyDesc = _("IPv4");
514 41 : break;
515 1 : case AF_INET6:
516 1 : familyDesc = _("IPv6");
517 1 : break;
518 950 : case AF_UNIX:
519 950 : familyDesc = _("Unix");
520 950 : break;
521 0 : default:
522 0 : snprintf(familyDescBuf, sizeof(familyDescBuf),
523 0 : _("unrecognized address family %d"),
524 : addr->ai_family);
525 0 : familyDesc = familyDescBuf;
526 0 : break;
527 : }
528 :
529 : /* set up text form of address for log messages */
530 992 : if (addr->ai_family == AF_UNIX)
531 950 : addrDesc = unixSocketPath;
532 : else
533 : {
534 42 : pg_getnameinfo_all((const struct sockaddr_storage *) addr->ai_addr,
535 42 : addr->ai_addrlen,
536 : addrBuf, sizeof(addrBuf),
537 : NULL, 0,
538 : NI_NUMERICHOST);
539 42 : addrDesc = addrBuf;
540 : }
541 :
542 992 : if ((fd = socket(addr->ai_family, SOCK_STREAM, 0)) == PGINVALID_SOCKET)
543 : {
544 0 : ereport(LOG,
545 : (errcode_for_socket_access(),
546 : /* translator: first %s is IPv4, IPv6, or Unix */
547 : errmsg("could not create %s socket for address \"%s\": %m",
548 : familyDesc, addrDesc)));
549 0 : continue;
550 : }
551 :
552 : #ifndef WIN32
553 : /* Don't give the listen socket to any subprograms we execute. */
554 992 : if (fcntl(fd, F_SETFD, FD_CLOEXEC) < 0)
555 0 : elog(FATAL, "fcntl(F_SETFD) failed on socket: %m");
556 :
557 : /*
558 : * Without the SO_REUSEADDR flag, a new postmaster can't be started
559 : * right away after a stop or crash, giving "address already in use"
560 : * error on TCP ports.
561 : *
562 : * On win32, however, this behavior only happens if the
563 : * SO_EXCLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows
564 : * multiple servers to listen on the same address, resulting in
565 : * unpredictable behavior. With no flags at all, win32 behaves as Unix
566 : * with SO_REUSEADDR.
567 : */
568 992 : if (addr->ai_family != AF_UNIX)
569 : {
570 42 : if ((setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
571 : (char *) &one, sizeof(one))) == -1)
572 : {
573 0 : ereport(LOG,
574 : (errcode_for_socket_access(),
575 : /* translator: third %s is IPv4 or IPv6 */
576 : errmsg("%s(%s) failed for %s address \"%s\": %m",
577 : "setsockopt", "SO_REUSEADDR",
578 : familyDesc, addrDesc)));
579 0 : closesocket(fd);
580 0 : continue;
581 : }
582 : }
583 : #endif
584 :
585 : #ifdef IPV6_V6ONLY
586 992 : if (addr->ai_family == AF_INET6)
587 : {
588 1 : if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY,
589 : (char *) &one, sizeof(one)) == -1)
590 : {
591 0 : ereport(LOG,
592 : (errcode_for_socket_access(),
593 : /* translator: third %s is IPv6 */
594 : errmsg("%s(%s) failed for %s address \"%s\": %m",
595 : "setsockopt", "IPV6_V6ONLY",
596 : familyDesc, addrDesc)));
597 0 : closesocket(fd);
598 0 : continue;
599 : }
600 : }
601 : #endif
602 :
603 : /*
604 : * Note: This might fail on some OS's, like Linux older than
605 : * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
606 : * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
607 : * connections.
608 : */
609 992 : err = bind(fd, addr->ai_addr, addr->ai_addrlen);
610 992 : if (err < 0)
611 0 : {
612 0 : int saved_errno = errno;
613 :
614 0 : ereport(LOG,
615 : (errcode_for_socket_access(),
616 : /* translator: first %s is IPv4, IPv6, or Unix */
617 : errmsg("could not bind %s address \"%s\": %m",
618 : familyDesc, addrDesc),
619 : saved_errno == EADDRINUSE ?
620 : (addr->ai_family == AF_UNIX ?
621 : errhint("Is another postmaster already running on port %d?",
622 : portNumber) :
623 : errhint("Is another postmaster already running on port %d?"
624 : " If not, wait a few seconds and retry.",
625 : portNumber)) : 0));
626 0 : closesocket(fd);
627 0 : continue;
628 : }
629 :
630 992 : if (addr->ai_family == AF_UNIX)
631 : {
632 950 : if (Setup_AF_UNIX(service) != STATUS_OK)
633 : {
634 0 : closesocket(fd);
635 0 : break;
636 : }
637 : }
638 :
639 : /*
640 : * Select appropriate accept-queue length limit. It seems reasonable
641 : * to use a value similar to the maximum number of child processes
642 : * that the postmaster will permit.
643 : */
644 992 : maxconn = MaxConnections * 2;
645 :
646 992 : err = listen(fd, maxconn);
647 992 : if (err < 0)
648 : {
649 0 : ereport(LOG,
650 : (errcode_for_socket_access(),
651 : /* translator: first %s is IPv4, IPv6, or Unix */
652 : errmsg("could not listen on %s address \"%s\": %m",
653 : familyDesc, addrDesc)));
654 0 : closesocket(fd);
655 0 : continue;
656 : }
657 :
658 992 : if (addr->ai_family == AF_UNIX)
659 950 : ereport(LOG,
660 : (errmsg("listening on Unix socket \"%s\"",
661 : addrDesc)));
662 : else
663 42 : ereport(LOG,
664 : /* translator: first %s is IPv4 or IPv6 */
665 : (errmsg("listening on %s address \"%s\", port %d",
666 : familyDesc, addrDesc, portNumber)));
667 :
668 992 : ListenSockets[*NumListenSockets] = fd;
669 992 : (*NumListenSockets)++;
670 992 : added++;
671 : }
672 :
673 991 : pg_freeaddrinfo_all(hint.ai_family, addrs);
674 :
675 991 : if (!added)
676 0 : return STATUS_ERROR;
677 :
678 991 : return STATUS_OK;
679 : }
680 :
681 :
682 : /*
683 : * Lock_AF_UNIX -- configure unix socket file path
684 : */
685 : static int
686 950 : Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath)
687 : {
688 : /* no lock file for abstract sockets */
689 950 : if (unixSocketPath[0] == '@')
690 0 : return STATUS_OK;
691 :
692 : /*
693 : * Grab an interlock file associated with the socket file.
694 : *
695 : * Note: there are two reasons for using a socket lock file, rather than
696 : * trying to interlock directly on the socket itself. First, it's a lot
697 : * more portable, and second, it lets us remove any pre-existing socket
698 : * file without race conditions.
699 : */
700 950 : CreateSocketLockFile(unixSocketPath, true, unixSocketDir);
701 :
702 : /*
703 : * Once we have the interlock, we can safely delete any pre-existing
704 : * socket file to avoid failure at bind() time.
705 : */
706 950 : (void) unlink(unixSocketPath);
707 :
708 : /*
709 : * Remember socket file pathnames for later maintenance.
710 : */
711 950 : sock_paths = lappend(sock_paths, pstrdup(unixSocketPath));
712 :
713 950 : return STATUS_OK;
714 : }
715 :
716 :
717 : /*
718 : * Setup_AF_UNIX -- configure unix socket permissions
719 : */
720 : static int
721 950 : Setup_AF_UNIX(const char *sock_path)
722 : {
723 : /* no file system permissions for abstract sockets */
724 950 : if (sock_path[0] == '@')
725 0 : return STATUS_OK;
726 :
727 : /*
728 : * Fix socket ownership/permission if requested. Note we must do this
729 : * before we listen() to avoid a window where unwanted connections could
730 : * get accepted.
731 : */
732 : Assert(Unix_socket_group);
733 950 : if (Unix_socket_group[0] != '\0')
734 : {
735 : #ifdef WIN32
736 : elog(WARNING, "configuration item \"unix_socket_group\" is not supported on this platform");
737 : #else
738 : char *endptr;
739 : unsigned long val;
740 : gid_t gid;
741 :
742 0 : val = strtoul(Unix_socket_group, &endptr, 10);
743 0 : if (*endptr == '\0')
744 : { /* numeric group id */
745 0 : gid = val;
746 : }
747 : else
748 : { /* convert group name to id */
749 : struct group *gr;
750 :
751 0 : gr = getgrnam(Unix_socket_group);
752 0 : if (!gr)
753 : {
754 0 : ereport(LOG,
755 : (errmsg("group \"%s\" does not exist",
756 : Unix_socket_group)));
757 0 : return STATUS_ERROR;
758 : }
759 0 : gid = gr->gr_gid;
760 : }
761 0 : if (chown(sock_path, -1, gid) == -1)
762 : {
763 0 : ereport(LOG,
764 : (errcode_for_file_access(),
765 : errmsg("could not set group of file \"%s\": %m",
766 : sock_path)));
767 0 : return STATUS_ERROR;
768 : }
769 : #endif
770 : }
771 :
772 950 : if (chmod(sock_path, Unix_socket_permissions) == -1)
773 : {
774 0 : ereport(LOG,
775 : (errcode_for_file_access(),
776 : errmsg("could not set permissions of file \"%s\": %m",
777 : sock_path)));
778 0 : return STATUS_ERROR;
779 : }
780 950 : return STATUS_OK;
781 : }
782 :
783 :
784 : /*
785 : * AcceptConnection -- accept a new connection with client using
786 : * server port. Fills *client_sock with the FD and endpoint info
787 : * of the new connection.
788 : *
789 : * ASSUME: that this doesn't need to be non-blocking because
790 : * the Postmaster waits for the socket to be ready to accept().
791 : *
792 : * RETURNS: STATUS_OK or STATUS_ERROR
793 : */
794 : int
795 14729 : AcceptConnection(pgsocket server_fd, ClientSocket *client_sock)
796 : {
797 : /* accept connection and fill in the client (remote) address */
798 14729 : client_sock->raddr.salen = sizeof(client_sock->raddr.addr);
799 14729 : if ((client_sock->sock = accept(server_fd,
800 14729 : (struct sockaddr *) &client_sock->raddr.addr,
801 : &client_sock->raddr.salen)) == PGINVALID_SOCKET)
802 : {
803 0 : ereport(LOG,
804 : (errcode_for_socket_access(),
805 : errmsg("could not accept new connection: %m")));
806 :
807 : /*
808 : * If accept() fails then postmaster.c will still see the server
809 : * socket as read-ready, and will immediately try again. To avoid
810 : * uselessly sucking lots of CPU, delay a bit before trying again.
811 : * (The most likely reason for failure is being out of kernel file
812 : * table slots; we can do little except hope some will get freed up.)
813 : */
814 0 : pg_usleep(100000L); /* wait 0.1 sec */
815 0 : return STATUS_ERROR;
816 : }
817 :
818 14729 : return STATUS_OK;
819 : }
820 :
821 : /*
822 : * TouchSocketFiles -- mark socket files as recently accessed
823 : *
824 : * This routine should be called every so often to ensure that the socket
825 : * files have a recent mod date (ordinary operations on sockets usually won't
826 : * change the mod date). That saves them from being removed by
827 : * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
828 : * never have put the socket file in /tmp...)
829 : */
830 : void
831 0 : TouchSocketFiles(void)
832 : {
833 : ListCell *l;
834 :
835 : /* Loop through all created sockets... */
836 0 : foreach(l, sock_paths)
837 : {
838 0 : char *sock_path = (char *) lfirst(l);
839 :
840 : /* Ignore errors; there's no point in complaining */
841 0 : (void) utime(sock_path, NULL);
842 : }
843 0 : }
844 :
845 : /*
846 : * RemoveSocketFiles -- unlink socket files at postmaster shutdown
847 : */
848 : void
849 951 : RemoveSocketFiles(void)
850 : {
851 : ListCell *l;
852 :
853 : /* Loop through all created sockets... */
854 1901 : foreach(l, sock_paths)
855 : {
856 950 : char *sock_path = (char *) lfirst(l);
857 :
858 : /* Ignore any error. */
859 950 : (void) unlink(sock_path);
860 : }
861 : /* Since we're about to exit, no need to reclaim storage */
862 951 : }
863 :
864 :
865 : /* --------------------------------
866 : * Low-level I/O routines begin here.
867 : *
868 : * These routines communicate with a frontend client across a connection
869 : * already established by the preceding routines.
870 : * --------------------------------
871 : */
872 :
873 : /* --------------------------------
874 : * socket_set_nonblocking - set socket blocking/non-blocking
875 : *
876 : * Sets the socket non-blocking if nonblocking is true, or sets it
877 : * blocking otherwise.
878 : * --------------------------------
879 : */
880 : static void
881 2612473 : socket_set_nonblocking(bool nonblocking)
882 : {
883 2612473 : if (MyProcPort == NULL)
884 0 : ereport(ERROR,
885 : (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST),
886 : errmsg("there is no client connection")));
887 :
888 2612473 : MyProcPort->noblock = nonblocking;
889 2612473 : }
890 :
891 : /* --------------------------------
892 : * pq_recvbuf - load some bytes into the input buffer
893 : *
894 : * returns 0 if OK, EOF if trouble
895 : * --------------------------------
896 : */
897 : static int
898 501566 : pq_recvbuf(void)
899 : {
900 501566 : if (PqRecvPointer > 0)
901 : {
902 487091 : if (PqRecvLength > PqRecvPointer)
903 : {
904 : /* still some unread data, left-justify it in the buffer */
905 0 : memmove(PqRecvBuffer, PqRecvBuffer + PqRecvPointer,
906 0 : PqRecvLength - PqRecvPointer);
907 0 : PqRecvLength -= PqRecvPointer;
908 0 : PqRecvPointer = 0;
909 : }
910 : else
911 487091 : PqRecvLength = PqRecvPointer = 0;
912 : }
913 :
914 : /* Ensure that we're in blocking mode */
915 501566 : socket_set_nonblocking(false);
916 :
917 : /* Can fill buffer from PqRecvLength and upwards */
918 : for (;;)
919 0 : {
920 : int r;
921 :
922 501566 : errno = 0;
923 :
924 1003112 : r = secure_read(MyProcPort, PqRecvBuffer + PqRecvLength,
925 501566 : PQ_RECV_BUFFER_SIZE - PqRecvLength);
926 :
927 501546 : if (r < 0)
928 : {
929 0 : if (errno == EINTR)
930 0 : continue; /* Ok if interrupted */
931 :
932 : /*
933 : * Careful: an ereport() that tries to write to the client would
934 : * cause recursion to here, leading to stack overflow and core
935 : * dump! This message must go *only* to the postmaster log.
936 : *
937 : * If errno is zero, assume it's EOF and let the caller complain.
938 : */
939 0 : if (errno != 0)
940 0 : ereport(COMMERROR,
941 : (errcode_for_socket_access(),
942 : errmsg("could not receive data from client: %m")));
943 0 : return EOF;
944 : }
945 501546 : if (r == 0)
946 : {
947 : /*
948 : * EOF detected. We used to write a log message here, but it's
949 : * better to expect the ultimate caller to do that.
950 : */
951 78 : return EOF;
952 : }
953 : /* r contains number of bytes read, so just incr length */
954 501468 : PqRecvLength += r;
955 501468 : return 0;
956 : }
957 : }
958 :
959 : /* --------------------------------
960 : * pq_getbyte - get a single byte from connection, or return EOF
961 : * --------------------------------
962 : */
963 : int
964 657851 : pq_getbyte(void)
965 : {
966 : Assert(PqCommReadingMsg);
967 :
968 1074940 : while (PqRecvPointer >= PqRecvLength)
969 : {
970 417170 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
971 61 : return EOF; /* Failed to recv data */
972 : }
973 657770 : return (unsigned char) PqRecvBuffer[PqRecvPointer++];
974 : }
975 :
976 : /* --------------------------------
977 : * pq_peekbyte - peek at next byte from connection
978 : *
979 : * Same as pq_getbyte() except we don't advance the pointer.
980 : * --------------------------------
981 : */
982 : int
983 14475 : pq_peekbyte(void)
984 : {
985 : Assert(PqCommReadingMsg);
986 :
987 28948 : while (PqRecvPointer >= PqRecvLength)
988 : {
989 14475 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
990 2 : return EOF; /* Failed to recv data */
991 : }
992 14473 : return (unsigned char) PqRecvBuffer[PqRecvPointer];
993 : }
994 :
995 : /* --------------------------------
996 : * pq_getbyte_if_available - get a single byte from connection,
997 : * if available
998 : *
999 : * The received byte is stored in *c. Returns 1 if a byte was read,
1000 : * 0 if no data was available, or EOF if trouble.
1001 : * --------------------------------
1002 : */
1003 : int
1004 941935 : pq_getbyte_if_available(unsigned char *c)
1005 : {
1006 : int r;
1007 :
1008 : Assert(PqCommReadingMsg);
1009 :
1010 941935 : if (PqRecvPointer < PqRecvLength)
1011 : {
1012 53494 : *c = PqRecvBuffer[PqRecvPointer++];
1013 53494 : return 1;
1014 : }
1015 :
1016 : /* Put the socket into non-blocking mode */
1017 888441 : socket_set_nonblocking(true);
1018 :
1019 888441 : errno = 0;
1020 :
1021 888441 : r = secure_read(MyProcPort, c, 1);
1022 888441 : if (r < 0)
1023 : {
1024 : /*
1025 : * Ok if no data available without blocking or interrupted (though
1026 : * EINTR really shouldn't happen with a non-blocking socket). Report
1027 : * other errors.
1028 : */
1029 820255 : if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
1030 820252 : r = 0;
1031 : else
1032 : {
1033 : /*
1034 : * Careful: an ereport() that tries to write to the client would
1035 : * cause recursion to here, leading to stack overflow and core
1036 : * dump! This message must go *only* to the postmaster log.
1037 : *
1038 : * If errno is zero, assume it's EOF and let the caller complain.
1039 : */
1040 3 : if (errno != 0)
1041 3 : ereport(COMMERROR,
1042 : (errcode_for_socket_access(),
1043 : errmsg("could not receive data from client: %m")));
1044 3 : r = EOF;
1045 : }
1046 : }
1047 68186 : else if (r == 0)
1048 : {
1049 : /* EOF detected */
1050 14 : r = EOF;
1051 : }
1052 :
1053 888441 : return r;
1054 : }
1055 :
1056 : /* --------------------------------
1057 : * pq_getbytes - get a known number of bytes from connection
1058 : *
1059 : * returns 0 if OK, EOF if trouble
1060 : * --------------------------------
1061 : */
1062 : int
1063 1576386 : pq_getbytes(void *b, size_t len)
1064 : {
1065 1576386 : char *s = b;
1066 : size_t amount;
1067 :
1068 : Assert(PqCommReadingMsg);
1069 :
1070 3154211 : while (len > 0)
1071 : {
1072 1647746 : while (PqRecvPointer >= PqRecvLength)
1073 : {
1074 69921 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1075 15 : return EOF; /* Failed to recv data */
1076 : }
1077 1577825 : amount = PqRecvLength - PqRecvPointer;
1078 1577825 : if (amount > len)
1079 1076367 : amount = len;
1080 1577825 : memcpy(s, PqRecvBuffer + PqRecvPointer, amount);
1081 1577825 : PqRecvPointer += amount;
1082 1577825 : s += amount;
1083 1577825 : len -= amount;
1084 : }
1085 1576371 : return 0;
1086 : }
1087 :
1088 : /* --------------------------------
1089 : * pq_discardbytes - throw away a known number of bytes
1090 : *
1091 : * same as pq_getbytes except we do not copy the data to anyplace.
1092 : * this is used for resynchronizing after read errors.
1093 : *
1094 : * returns 0 if OK, EOF if trouble
1095 : * --------------------------------
1096 : */
1097 : static int
1098 0 : pq_discardbytes(size_t len)
1099 : {
1100 : size_t amount;
1101 :
1102 : Assert(PqCommReadingMsg);
1103 :
1104 0 : while (len > 0)
1105 : {
1106 0 : while (PqRecvPointer >= PqRecvLength)
1107 : {
1108 0 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1109 0 : return EOF; /* Failed to recv data */
1110 : }
1111 0 : amount = PqRecvLength - PqRecvPointer;
1112 0 : if (amount > len)
1113 0 : amount = len;
1114 0 : PqRecvPointer += amount;
1115 0 : len -= amount;
1116 : }
1117 0 : return 0;
1118 : }
1119 :
1120 : /* --------------------------------
1121 : * pq_buffer_remaining_data - return number of bytes in receive buffer
1122 : *
1123 : * This will *not* attempt to read more data. And reading up to that number of
1124 : * bytes should not cause reading any more data either.
1125 : * --------------------------------
1126 : */
1127 : ssize_t
1128 456 : pq_buffer_remaining_data(void)
1129 : {
1130 : Assert(PqRecvLength >= PqRecvPointer);
1131 456 : return (PqRecvLength - PqRecvPointer);
1132 : }
1133 :
1134 :
1135 : /* --------------------------------
1136 : * pq_startmsgread - begin reading a message from the client.
1137 : *
1138 : * This must be called before any of the pq_get* functions.
1139 : * --------------------------------
1140 : */
1141 : void
1142 1629020 : pq_startmsgread(void)
1143 : {
1144 : /*
1145 : * There shouldn't be a read active already, but let's check just to be
1146 : * sure.
1147 : */
1148 1629020 : if (PqCommReadingMsg)
1149 0 : ereport(FATAL,
1150 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1151 : errmsg("terminating connection because protocol synchronization was lost")));
1152 :
1153 1629020 : PqCommReadingMsg = true;
1154 1629020 : }
1155 :
1156 :
1157 : /* --------------------------------
1158 : * pq_endmsgread - finish reading message.
1159 : *
1160 : * This must be called after reading a message with pq_getbytes()
1161 : * and friends, to indicate that we have read the whole message.
1162 : * pq_getmessage() does this implicitly.
1163 : * --------------------------------
1164 : */
1165 : void
1166 849471 : pq_endmsgread(void)
1167 : {
1168 : Assert(PqCommReadingMsg);
1169 :
1170 849471 : PqCommReadingMsg = false;
1171 849471 : }
1172 :
1173 : /* --------------------------------
1174 : * pq_is_reading_msg - are we currently reading a message?
1175 : *
1176 : * This is used in error recovery at the outer idle loop to detect if we have
1177 : * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1178 : * will check for that too, but it's nicer to detect it earlier.
1179 : * --------------------------------
1180 : */
1181 : bool
1182 30210 : pq_is_reading_msg(void)
1183 : {
1184 30210 : return PqCommReadingMsg;
1185 : }
1186 :
1187 : /* --------------------------------
1188 : * pq_getmessage - get a message with length word from connection
1189 : *
1190 : * The return value is placed in an expansible StringInfo, which has
1191 : * already been initialized by the caller.
1192 : * Only the message body is placed in the StringInfo; the length word
1193 : * is removed. Also, s->cursor is initialized to zero for convenience
1194 : * in scanning the message contents.
1195 : *
1196 : * maxlen is the upper limit on the length of the
1197 : * message we are willing to accept. We abort the connection (by
1198 : * returning EOF) if client tries to send more than that.
1199 : *
1200 : * returns 0 if OK, EOF if trouble
1201 : * --------------------------------
1202 : */
1203 : int
1204 779434 : pq_getmessage(StringInfo s, int maxlen)
1205 : {
1206 : int32 len;
1207 :
1208 : Assert(PqCommReadingMsg);
1209 :
1210 779434 : resetStringInfo(s);
1211 :
1212 : /* Read message length word */
1213 779434 : if (pq_getbytes(&len, 4) == EOF)
1214 : {
1215 0 : ereport(COMMERROR,
1216 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1217 : errmsg("unexpected EOF within message length word")));
1218 0 : return EOF;
1219 : }
1220 :
1221 779434 : len = pg_ntoh32(len);
1222 :
1223 779434 : if (len < 4 || len > maxlen)
1224 : {
1225 0 : ereport(COMMERROR,
1226 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1227 : errmsg("invalid message length")));
1228 0 : return EOF;
1229 : }
1230 :
1231 779434 : len -= 4; /* discount length itself */
1232 :
1233 779434 : if (len > 0)
1234 : {
1235 : /*
1236 : * Allocate space for message. If we run out of room (ridiculously
1237 : * large message), we will elog(ERROR), but we want to discard the
1238 : * message body so as not to lose communication sync.
1239 : */
1240 752711 : PG_TRY();
1241 : {
1242 752711 : enlargeStringInfo(s, len);
1243 : }
1244 0 : PG_CATCH();
1245 : {
1246 0 : if (pq_discardbytes(len) == EOF)
1247 0 : ereport(COMMERROR,
1248 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1249 : errmsg("incomplete message from client")));
1250 :
1251 : /* we discarded the rest of the message so we're back in sync. */
1252 0 : PqCommReadingMsg = false;
1253 0 : PG_RE_THROW();
1254 : }
1255 752711 : PG_END_TRY();
1256 :
1257 : /* And grab the message */
1258 752711 : if (pq_getbytes(s->data, len) == EOF)
1259 : {
1260 0 : ereport(COMMERROR,
1261 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1262 : errmsg("incomplete message from client")));
1263 0 : return EOF;
1264 : }
1265 752711 : s->len = len;
1266 : /* Place a trailing null per StringInfo convention */
1267 752711 : s->data[len] = '\0';
1268 : }
1269 :
1270 : /* finished reading the message. */
1271 779434 : PqCommReadingMsg = false;
1272 :
1273 779434 : return 0;
1274 : }
1275 :
1276 :
1277 : static inline int
1278 25280799 : internal_putbytes(const void *b, size_t len)
1279 : {
1280 25280799 : const char *s = b;
1281 :
1282 50755023 : while (len > 0)
1283 : {
1284 : /* If buffer is full, then flush it out */
1285 25474236 : if (PqSendPointer >= PqSendBufferSize)
1286 : {
1287 224969 : socket_set_nonblocking(false);
1288 224969 : if (internal_flush())
1289 11 : return EOF;
1290 : }
1291 :
1292 : /*
1293 : * If the buffer is empty and data length is larger than the buffer
1294 : * size, send it without buffering. Otherwise, copy as much data as
1295 : * possible into the buffer.
1296 : */
1297 25474225 : if (len >= PqSendBufferSize && PqSendStart == PqSendPointer)
1298 118087 : {
1299 118088 : size_t start = 0;
1300 :
1301 118088 : socket_set_nonblocking(false);
1302 118088 : if (internal_flush_buffer(s, &start, &len))
1303 1 : return EOF;
1304 : }
1305 : else
1306 : {
1307 25356137 : size_t amount = PqSendBufferSize - PqSendPointer;
1308 :
1309 25356137 : if (amount > len)
1310 25051621 : amount = len;
1311 25356137 : memcpy(PqSendBuffer + PqSendPointer, s, amount);
1312 25356137 : PqSendPointer += amount;
1313 25356137 : s += amount;
1314 25356137 : len -= amount;
1315 : }
1316 : }
1317 :
1318 25280787 : return 0;
1319 : }
1320 :
1321 : /* --------------------------------
1322 : * socket_flush - flush pending output
1323 : *
1324 : * returns 0 if OK, EOF if trouble
1325 : * --------------------------------
1326 : */
1327 : static int
1328 531290 : socket_flush(void)
1329 : {
1330 : int res;
1331 :
1332 : /* No-op if reentrant call */
1333 531290 : if (PqCommBusy)
1334 0 : return 0;
1335 531290 : PqCommBusy = true;
1336 531290 : socket_set_nonblocking(false);
1337 531290 : res = internal_flush();
1338 531290 : PqCommBusy = false;
1339 531290 : return res;
1340 : }
1341 :
1342 : /* --------------------------------
1343 : * internal_flush - flush pending output
1344 : *
1345 : * Returns 0 if OK (meaning everything was sent, or operation would block
1346 : * and the socket is in non-blocking mode), or EOF if trouble.
1347 : * --------------------------------
1348 : */
1349 : static inline int
1350 1104378 : internal_flush(void)
1351 : {
1352 1104378 : return internal_flush_buffer(PqSendBuffer, &PqSendStart, &PqSendPointer);
1353 : }
1354 :
1355 : /* --------------------------------
1356 : * internal_flush_buffer - flush the given buffer content
1357 : *
1358 : * Returns 0 if OK (meaning everything was sent, or operation would block
1359 : * and the socket is in non-blocking mode), or EOF if trouble.
1360 : * --------------------------------
1361 : */
1362 : static pg_noinline int
1363 1222466 : internal_flush_buffer(const char *buf, size_t *start, size_t *end)
1364 : {
1365 : static int last_reported_send_errno = 0;
1366 :
1367 1222466 : const char *bufptr = buf + *start;
1368 1222466 : const char *bufend = buf + *end;
1369 :
1370 2434979 : while (bufptr < bufend)
1371 : {
1372 : int r;
1373 :
1374 1251496 : r = secure_write(MyProcPort, bufptr, bufend - bufptr);
1375 :
1376 1251496 : if (r <= 0)
1377 : {
1378 38983 : if (errno == EINTR)
1379 0 : continue; /* Ok if we were interrupted */
1380 :
1381 : /*
1382 : * Ok if no data writable without blocking, and the socket is in
1383 : * non-blocking mode.
1384 : */
1385 38983 : if (errno == EAGAIN ||
1386 28 : errno == EWOULDBLOCK)
1387 : {
1388 38955 : return 0;
1389 : }
1390 :
1391 : /*
1392 : * Careful: an ereport() that tries to write to the client would
1393 : * cause recursion to here, leading to stack overflow and core
1394 : * dump! This message must go *only* to the postmaster log.
1395 : *
1396 : * If a client disconnects while we're in the midst of output, we
1397 : * might write quite a bit of data before we get to a safe query
1398 : * abort point. So, suppress duplicate log messages.
1399 : */
1400 28 : if (errno != last_reported_send_errno)
1401 : {
1402 20 : last_reported_send_errno = errno;
1403 20 : ereport(COMMERROR,
1404 : (errcode_for_socket_access(),
1405 : errmsg("could not send data to client: %m")));
1406 : }
1407 :
1408 : /*
1409 : * We drop the buffered data anyway so that processing can
1410 : * continue, even though we'll probably quit soon. We also set a
1411 : * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1412 : * the connection.
1413 : */
1414 28 : *start = *end = 0;
1415 28 : ClientConnectionLost = 1;
1416 28 : InterruptPending = 1;
1417 28 : return EOF;
1418 : }
1419 :
1420 1212513 : last_reported_send_errno = 0; /* reset after any successful send */
1421 1212513 : bufptr += r;
1422 1212513 : *start += r;
1423 : }
1424 :
1425 1183483 : *start = *end = 0;
1426 1183483 : return 0;
1427 : }
1428 :
1429 : /* --------------------------------
1430 : * pq_flush_if_writable - flush pending output if writable without blocking
1431 : *
1432 : * Returns 0 if OK, or EOF if trouble.
1433 : * --------------------------------
1434 : */
1435 : static int
1436 1001895 : socket_flush_if_writable(void)
1437 : {
1438 : int res;
1439 :
1440 : /* Quick exit if nothing to do */
1441 1001895 : if (PqSendPointer == PqSendStart)
1442 653776 : return 0;
1443 :
1444 : /* No-op if reentrant call */
1445 348119 : if (PqCommBusy)
1446 0 : return 0;
1447 :
1448 : /* Temporarily put the socket into non-blocking mode */
1449 348119 : socket_set_nonblocking(true);
1450 :
1451 348119 : PqCommBusy = true;
1452 348119 : res = internal_flush();
1453 348119 : PqCommBusy = false;
1454 348119 : return res;
1455 : }
1456 :
1457 : /* --------------------------------
1458 : * socket_is_send_pending - is there any pending data in the output buffer?
1459 : * --------------------------------
1460 : */
1461 : static bool
1462 1922404 : socket_is_send_pending(void)
1463 : {
1464 1922404 : return (PqSendStart < PqSendPointer);
1465 : }
1466 :
1467 : /* --------------------------------
1468 : * Message-level I/O routines begin here.
1469 : * --------------------------------
1470 : */
1471 :
1472 :
1473 : /* --------------------------------
1474 : * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1475 : *
1476 : * msgtype is a message type code to place before the message body.
1477 : *
1478 : * len is the length of the message body data at *s. A message length
1479 : * word (equal to len+4 because it counts itself too) is inserted by this
1480 : * routine.
1481 : *
1482 : * We suppress messages generated while pqcomm.c is busy. This
1483 : * avoids any possibility of messages being inserted within other
1484 : * messages. The only known trouble case arises if SIGQUIT occurs
1485 : * during a pqcomm.c routine --- quickdie() will try to send a warning
1486 : * message, and the most reasonable approach seems to be to drop it.
1487 : *
1488 : * returns 0 if OK, EOF if trouble
1489 : * --------------------------------
1490 : */
1491 : static int
1492 8426933 : socket_putmessage(char msgtype, const char *s, size_t len)
1493 : {
1494 : uint32 n32;
1495 :
1496 : Assert(msgtype != 0);
1497 :
1498 8426933 : if (PqCommBusy)
1499 0 : return 0;
1500 8426933 : PqCommBusy = true;
1501 8426933 : if (internal_putbytes(&msgtype, 1))
1502 0 : goto fail;
1503 :
1504 8426933 : n32 = pg_hton32((uint32) (len + 4));
1505 8426933 : if (internal_putbytes(&n32, 4))
1506 0 : goto fail;
1507 :
1508 8426933 : if (internal_putbytes(s, len))
1509 12 : goto fail;
1510 8426921 : PqCommBusy = false;
1511 8426921 : return 0;
1512 :
1513 12 : fail:
1514 12 : PqCommBusy = false;
1515 12 : return EOF;
1516 : }
1517 :
1518 : /* --------------------------------
1519 : * pq_putmessage_noblock - like pq_putmessage, but never blocks
1520 : *
1521 : * If the output buffer is too small to hold the message, the buffer
1522 : * is enlarged.
1523 : */
1524 : static void
1525 309103 : socket_putmessage_noblock(char msgtype, const char *s, size_t len)
1526 : {
1527 : int res PG_USED_FOR_ASSERTS_ONLY;
1528 : int required;
1529 :
1530 : /*
1531 : * Ensure we have enough space in the output buffer for the message header
1532 : * as well as the message itself.
1533 : */
1534 309103 : required = PqSendPointer + 1 + 4 + len;
1535 309103 : if (required > PqSendBufferSize)
1536 : {
1537 542 : PqSendBuffer = repalloc(PqSendBuffer, required);
1538 542 : PqSendBufferSize = required;
1539 : }
1540 309103 : res = pq_putmessage(msgtype, s, len);
1541 : Assert(res == 0); /* should not fail when the message fits in
1542 : * buffer */
1543 309103 : }
1544 :
1545 : /* --------------------------------
1546 : * pq_putmessage_v2 - send a message in protocol version 2
1547 : *
1548 : * msgtype is a message type code to place before the message body.
1549 : *
1550 : * We no longer support protocol version 2, but we have kept this
1551 : * function so that if a client tries to connect with protocol version 2,
1552 : * as a courtesy we can still send the "unsupported protocol version"
1553 : * error to the client in the old format.
1554 : *
1555 : * Like in pq_putmessage(), we suppress messages generated while
1556 : * pqcomm.c is busy.
1557 : *
1558 : * returns 0 if OK, EOF if trouble
1559 : * --------------------------------
1560 : */
1561 : int
1562 0 : pq_putmessage_v2(char msgtype, const char *s, size_t len)
1563 : {
1564 : Assert(msgtype != 0);
1565 :
1566 0 : if (PqCommBusy)
1567 0 : return 0;
1568 0 : PqCommBusy = true;
1569 0 : if (internal_putbytes(&msgtype, 1))
1570 0 : goto fail;
1571 :
1572 0 : if (internal_putbytes(s, len))
1573 0 : goto fail;
1574 0 : PqCommBusy = false;
1575 0 : return 0;
1576 :
1577 0 : fail:
1578 0 : PqCommBusy = false;
1579 0 : return EOF;
1580 : }
1581 :
1582 : /*
1583 : * Support for TCP Keepalive parameters
1584 : */
1585 :
1586 : /*
1587 : * On Windows, we need to set both idle and interval at the same time.
1588 : * We also cannot reset them to the default (setting to zero will
1589 : * actually set them to zero, not default), therefore we fallback to
1590 : * the out-of-the-box default instead.
1591 : */
1592 : #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1593 : static int
1594 : pq_setkeepaliveswin32(Port *port, int idle, int interval)
1595 : {
1596 : struct tcp_keepalive ka;
1597 : DWORD retsize;
1598 :
1599 : if (idle <= 0)
1600 : idle = 2 * 60 * 60; /* default = 2 hours */
1601 : if (interval <= 0)
1602 : interval = 1; /* default = 1 second */
1603 :
1604 : ka.onoff = 1;
1605 : ka.keepalivetime = idle * 1000;
1606 : ka.keepaliveinterval = interval * 1000;
1607 :
1608 : if (WSAIoctl(port->sock,
1609 : SIO_KEEPALIVE_VALS,
1610 : (LPVOID) &ka,
1611 : sizeof(ka),
1612 : NULL,
1613 : 0,
1614 : &retsize,
1615 : NULL,
1616 : NULL)
1617 : != 0)
1618 : {
1619 : ereport(LOG,
1620 : (errmsg("%s(%s) failed: error code %d",
1621 : "WSAIoctl", "SIO_KEEPALIVE_VALS", WSAGetLastError())));
1622 : return STATUS_ERROR;
1623 : }
1624 : if (port->keepalives_idle != idle)
1625 : port->keepalives_idle = idle;
1626 : if (port->keepalives_interval != interval)
1627 : port->keepalives_interval = interval;
1628 : return STATUS_OK;
1629 : }
1630 : #endif
1631 :
1632 : int
1633 1797 : pq_getkeepalivesidle(Port *port)
1634 : {
1635 : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1636 1797 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1637 1797 : return 0;
1638 :
1639 0 : if (port->keepalives_idle != 0)
1640 0 : return port->keepalives_idle;
1641 :
1642 0 : if (port->default_keepalives_idle == 0)
1643 : {
1644 : #ifndef WIN32
1645 0 : socklen_t size = sizeof(port->default_keepalives_idle);
1646 :
1647 0 : if (getsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1648 0 : (char *) &port->default_keepalives_idle,
1649 : &size) < 0)
1650 : {
1651 0 : ereport(LOG,
1652 : (errmsg("%s(%s) failed: %m", "getsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
1653 0 : port->default_keepalives_idle = -1; /* don't know */
1654 : }
1655 : #else /* WIN32 */
1656 : /* We can't get the defaults on Windows, so return "don't know" */
1657 : port->default_keepalives_idle = -1;
1658 : #endif /* WIN32 */
1659 : }
1660 :
1661 0 : return port->default_keepalives_idle;
1662 : #else
1663 : return 0;
1664 : #endif
1665 : }
1666 :
1667 : int
1668 1556 : pq_setkeepalivesidle(int idle, Port *port)
1669 : {
1670 1556 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1671 1221 : return STATUS_OK;
1672 :
1673 : /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1674 : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1675 335 : if (idle == port->keepalives_idle)
1676 335 : return STATUS_OK;
1677 :
1678 : #ifndef WIN32
1679 0 : if (port->default_keepalives_idle <= 0)
1680 : {
1681 0 : if (pq_getkeepalivesidle(port) < 0)
1682 : {
1683 0 : if (idle == 0)
1684 0 : return STATUS_OK; /* default is set but unknown */
1685 : else
1686 0 : return STATUS_ERROR;
1687 : }
1688 : }
1689 :
1690 0 : if (idle == 0)
1691 0 : idle = port->default_keepalives_idle;
1692 :
1693 0 : if (setsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1694 : (char *) &idle, sizeof(idle)) < 0)
1695 : {
1696 0 : ereport(LOG,
1697 : (errmsg("%s(%s) failed: %m", "setsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
1698 0 : return STATUS_ERROR;
1699 : }
1700 :
1701 0 : port->keepalives_idle = idle;
1702 : #else /* WIN32 */
1703 : return pq_setkeepaliveswin32(port, idle, port->keepalives_interval);
1704 : #endif
1705 : #else
1706 : if (idle != 0)
1707 : {
1708 : ereport(LOG,
1709 : (errmsg("setting the keepalive idle time is not supported")));
1710 : return STATUS_ERROR;
1711 : }
1712 : #endif
1713 :
1714 0 : return STATUS_OK;
1715 : }
1716 :
1717 : int
1718 1797 : pq_getkeepalivesinterval(Port *port)
1719 : {
1720 : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1721 1797 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1722 1797 : return 0;
1723 :
1724 0 : if (port->keepalives_interval != 0)
1725 0 : return port->keepalives_interval;
1726 :
1727 0 : if (port->default_keepalives_interval == 0)
1728 : {
1729 : #ifndef WIN32
1730 0 : socklen_t size = sizeof(port->default_keepalives_interval);
1731 :
1732 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1733 0 : (char *) &port->default_keepalives_interval,
1734 : &size) < 0)
1735 : {
1736 0 : ereport(LOG,
1737 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPINTVL")));
1738 0 : port->default_keepalives_interval = -1; /* don't know */
1739 : }
1740 : #else
1741 : /* We can't get the defaults on Windows, so return "don't know" */
1742 : port->default_keepalives_interval = -1;
1743 : #endif /* WIN32 */
1744 : }
1745 :
1746 0 : return port->default_keepalives_interval;
1747 : #else
1748 : return 0;
1749 : #endif
1750 : }
1751 :
1752 : int
1753 1556 : pq_setkeepalivesinterval(int interval, Port *port)
1754 : {
1755 1556 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1756 1221 : return STATUS_OK;
1757 :
1758 : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1759 335 : if (interval == port->keepalives_interval)
1760 335 : return STATUS_OK;
1761 :
1762 : #ifndef WIN32
1763 0 : if (port->default_keepalives_interval <= 0)
1764 : {
1765 0 : if (pq_getkeepalivesinterval(port) < 0)
1766 : {
1767 0 : if (interval == 0)
1768 0 : return STATUS_OK; /* default is set but unknown */
1769 : else
1770 0 : return STATUS_ERROR;
1771 : }
1772 : }
1773 :
1774 0 : if (interval == 0)
1775 0 : interval = port->default_keepalives_interval;
1776 :
1777 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1778 : (char *) &interval, sizeof(interval)) < 0)
1779 : {
1780 0 : ereport(LOG,
1781 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPINTVL")));
1782 0 : return STATUS_ERROR;
1783 : }
1784 :
1785 0 : port->keepalives_interval = interval;
1786 : #else /* WIN32 */
1787 : return pq_setkeepaliveswin32(port, port->keepalives_idle, interval);
1788 : #endif
1789 : #else
1790 : if (interval != 0)
1791 : {
1792 : ereport(LOG,
1793 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPINTVL")));
1794 : return STATUS_ERROR;
1795 : }
1796 : #endif
1797 :
1798 0 : return STATUS_OK;
1799 : }
1800 :
1801 : int
1802 1797 : pq_getkeepalivescount(Port *port)
1803 : {
1804 : #ifdef TCP_KEEPCNT
1805 1797 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1806 1797 : return 0;
1807 :
1808 0 : if (port->keepalives_count != 0)
1809 0 : return port->keepalives_count;
1810 :
1811 0 : if (port->default_keepalives_count == 0)
1812 : {
1813 0 : socklen_t size = sizeof(port->default_keepalives_count);
1814 :
1815 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1816 0 : (char *) &port->default_keepalives_count,
1817 : &size) < 0)
1818 : {
1819 0 : ereport(LOG,
1820 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPCNT")));
1821 0 : port->default_keepalives_count = -1; /* don't know */
1822 : }
1823 : }
1824 :
1825 0 : return port->default_keepalives_count;
1826 : #else
1827 : return 0;
1828 : #endif
1829 : }
1830 :
1831 : int
1832 1556 : pq_setkeepalivescount(int count, Port *port)
1833 : {
1834 1556 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1835 1221 : return STATUS_OK;
1836 :
1837 : #ifdef TCP_KEEPCNT
1838 335 : if (count == port->keepalives_count)
1839 335 : return STATUS_OK;
1840 :
1841 0 : if (port->default_keepalives_count <= 0)
1842 : {
1843 0 : if (pq_getkeepalivescount(port) < 0)
1844 : {
1845 0 : if (count == 0)
1846 0 : return STATUS_OK; /* default is set but unknown */
1847 : else
1848 0 : return STATUS_ERROR;
1849 : }
1850 : }
1851 :
1852 0 : if (count == 0)
1853 0 : count = port->default_keepalives_count;
1854 :
1855 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1856 : (char *) &count, sizeof(count)) < 0)
1857 : {
1858 0 : ereport(LOG,
1859 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPCNT")));
1860 0 : return STATUS_ERROR;
1861 : }
1862 :
1863 0 : port->keepalives_count = count;
1864 : #else
1865 : if (count != 0)
1866 : {
1867 : ereport(LOG,
1868 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPCNT")));
1869 : return STATUS_ERROR;
1870 : }
1871 : #endif
1872 :
1873 0 : return STATUS_OK;
1874 : }
1875 :
1876 : int
1877 1797 : pq_gettcpusertimeout(Port *port)
1878 : {
1879 : #ifdef TCP_USER_TIMEOUT
1880 1797 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1881 1797 : return 0;
1882 :
1883 0 : if (port->tcp_user_timeout != 0)
1884 0 : return port->tcp_user_timeout;
1885 :
1886 0 : if (port->default_tcp_user_timeout == 0)
1887 : {
1888 0 : socklen_t size = sizeof(port->default_tcp_user_timeout);
1889 :
1890 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1891 0 : (char *) &port->default_tcp_user_timeout,
1892 : &size) < 0)
1893 : {
1894 0 : ereport(LOG,
1895 : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_USER_TIMEOUT")));
1896 0 : port->default_tcp_user_timeout = -1; /* don't know */
1897 : }
1898 : }
1899 :
1900 0 : return port->default_tcp_user_timeout;
1901 : #else
1902 : return 0;
1903 : #endif
1904 : }
1905 :
1906 : int
1907 1556 : pq_settcpusertimeout(int timeout, Port *port)
1908 : {
1909 1556 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1910 1221 : return STATUS_OK;
1911 :
1912 : #ifdef TCP_USER_TIMEOUT
1913 335 : if (timeout == port->tcp_user_timeout)
1914 335 : return STATUS_OK;
1915 :
1916 0 : if (port->default_tcp_user_timeout <= 0)
1917 : {
1918 0 : if (pq_gettcpusertimeout(port) < 0)
1919 : {
1920 0 : if (timeout == 0)
1921 0 : return STATUS_OK; /* default is set but unknown */
1922 : else
1923 0 : return STATUS_ERROR;
1924 : }
1925 : }
1926 :
1927 0 : if (timeout == 0)
1928 0 : timeout = port->default_tcp_user_timeout;
1929 :
1930 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1931 : (char *) &timeout, sizeof(timeout)) < 0)
1932 : {
1933 0 : ereport(LOG,
1934 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_USER_TIMEOUT")));
1935 0 : return STATUS_ERROR;
1936 : }
1937 :
1938 0 : port->tcp_user_timeout = timeout;
1939 : #else
1940 : if (timeout != 0)
1941 : {
1942 : ereport(LOG,
1943 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_USER_TIMEOUT")));
1944 : return STATUS_ERROR;
1945 : }
1946 : #endif
1947 :
1948 0 : return STATUS_OK;
1949 : }
1950 :
1951 : /*
1952 : * GUC assign_hook for tcp_keepalives_idle
1953 : */
1954 : void
1955 1221 : assign_tcp_keepalives_idle(int newval, void *extra)
1956 : {
1957 : /*
1958 : * The kernel API provides no way to test a value without setting it; and
1959 : * once we set it we might fail to unset it. So there seems little point
1960 : * in fully implementing the check-then-assign GUC API for these
1961 : * variables. Instead we just do the assignment on demand.
1962 : * pq_setkeepalivesidle reports any problems via ereport(LOG).
1963 : *
1964 : * This approach means that the GUC value might have little to do with the
1965 : * actual kernel value, so we use a show_hook that retrieves the kernel
1966 : * value rather than trusting GUC's copy.
1967 : */
1968 1221 : (void) pq_setkeepalivesidle(newval, MyProcPort);
1969 1221 : }
1970 :
1971 : /*
1972 : * GUC show_hook for tcp_keepalives_idle
1973 : */
1974 : const char *
1975 1797 : show_tcp_keepalives_idle(void)
1976 : {
1977 : /* See comments in assign_tcp_keepalives_idle */
1978 : static char nbuf[16];
1979 :
1980 1797 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesidle(MyProcPort));
1981 1797 : return nbuf;
1982 : }
1983 :
1984 : /*
1985 : * GUC assign_hook for tcp_keepalives_interval
1986 : */
1987 : void
1988 1221 : assign_tcp_keepalives_interval(int newval, void *extra)
1989 : {
1990 : /* See comments in assign_tcp_keepalives_idle */
1991 1221 : (void) pq_setkeepalivesinterval(newval, MyProcPort);
1992 1221 : }
1993 :
1994 : /*
1995 : * GUC show_hook for tcp_keepalives_interval
1996 : */
1997 : const char *
1998 1797 : show_tcp_keepalives_interval(void)
1999 : {
2000 : /* See comments in assign_tcp_keepalives_idle */
2001 : static char nbuf[16];
2002 :
2003 1797 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesinterval(MyProcPort));
2004 1797 : return nbuf;
2005 : }
2006 :
2007 : /*
2008 : * GUC assign_hook for tcp_keepalives_count
2009 : */
2010 : void
2011 1221 : assign_tcp_keepalives_count(int newval, void *extra)
2012 : {
2013 : /* See comments in assign_tcp_keepalives_idle */
2014 1221 : (void) pq_setkeepalivescount(newval, MyProcPort);
2015 1221 : }
2016 :
2017 : /*
2018 : * GUC show_hook for tcp_keepalives_count
2019 : */
2020 : const char *
2021 1797 : show_tcp_keepalives_count(void)
2022 : {
2023 : /* See comments in assign_tcp_keepalives_idle */
2024 : static char nbuf[16];
2025 :
2026 1797 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivescount(MyProcPort));
2027 1797 : return nbuf;
2028 : }
2029 :
2030 : /*
2031 : * GUC assign_hook for tcp_user_timeout
2032 : */
2033 : void
2034 1221 : assign_tcp_user_timeout(int newval, void *extra)
2035 : {
2036 : /* See comments in assign_tcp_keepalives_idle */
2037 1221 : (void) pq_settcpusertimeout(newval, MyProcPort);
2038 1221 : }
2039 :
2040 : /*
2041 : * GUC show_hook for tcp_user_timeout
2042 : */
2043 : const char *
2044 1797 : show_tcp_user_timeout(void)
2045 : {
2046 : /* See comments in assign_tcp_keepalives_idle */
2047 : static char nbuf[16];
2048 :
2049 1797 : snprintf(nbuf, sizeof(nbuf), "%d", pq_gettcpusertimeout(MyProcPort));
2050 1797 : return nbuf;
2051 : }
2052 :
2053 : /*
2054 : * Check if the client is still connected.
2055 : */
2056 : bool
2057 0 : pq_check_connection(void)
2058 : {
2059 : WaitEvent events[FeBeWaitSetNEvents];
2060 : int rc;
2061 :
2062 : /*
2063 : * It's OK to modify the socket event filter without restoring, because
2064 : * all FeBeWaitSet socket wait sites do the same.
2065 : */
2066 0 : ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetSocketPos, WL_SOCKET_CLOSED, NULL);
2067 :
2068 0 : retry:
2069 0 : rc = WaitEventSetWait(FeBeWaitSet, 0, events, lengthof(events), 0);
2070 0 : for (int i = 0; i < rc; ++i)
2071 : {
2072 0 : if (events[i].events & WL_SOCKET_CLOSED)
2073 0 : return false;
2074 0 : if (events[i].events & WL_LATCH_SET)
2075 : {
2076 : /*
2077 : * A latch event might be preventing other events from being
2078 : * reported. Reset it and poll again. No need to restore it
2079 : * because no code should expect latches to survive across
2080 : * CHECK_FOR_INTERRUPTS().
2081 : */
2082 0 : ResetLatch(MyLatch);
2083 0 : goto retry;
2084 : }
2085 : }
2086 :
2087 0 : return true;
2088 : }
|
