Age Owner TLA 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) 1996-2023, 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 : * StreamServerPort - Open postmaster's server port
33 : * StreamConnection - Create new connection with client
34 : * StreamClose - Close a client/backend connection
35 : * TouchSocketFiles - Protect socket files against /tmp cleaners
36 : * pq_init - initialize libpq at backend startup
37 : * socket_comm_reset - reset libpq during error recovery
38 : * socket_close - shutdown libpq at backend exit
39 : *
40 : * low-level I/O:
41 : * pq_getbytes - get a known number of bytes from connection
42 : * pq_getmessage - get a message with length word from connection
43 : * pq_getbyte - get next byte from connection
44 : * pq_peekbyte - peek at next byte from connection
45 : * pq_flush - flush pending output
46 : * pq_flush_if_writable - flush pending output if writable without blocking
47 : * pq_getbyte_if_available - get a byte if available without blocking
48 : *
49 : * message-level I/O
50 : * pq_putmessage - send a normal message (suppressed in COPY OUT mode)
51 : * pq_putmessage_noblock - buffer a normal message (suppressed in COPY OUT)
52 : *
53 : *------------------------
54 : */
55 : #include "postgres.h"
56 :
57 : #ifdef HAVE_POLL_H
58 : #include <poll.h>
59 : #endif
60 : #include <signal.h>
61 : #include <fcntl.h>
62 : #include <grp.h>
63 : #include <unistd.h>
64 : #include <sys/file.h>
65 : #include <sys/socket.h>
66 : #include <sys/stat.h>
67 : #include <sys/time.h>
68 : #include <netdb.h>
69 : #include <netinet/in.h>
70 : #include <netinet/tcp.h>
71 : #include <utime.h>
72 : #ifdef WIN32
73 : #include <mstcpip.h>
74 : #endif
75 :
76 : #include "common/ip.h"
77 : #include "libpq/libpq.h"
78 : #include "miscadmin.h"
79 : #include "port/pg_bswap.h"
80 : #include "storage/ipc.h"
81 : #include "utils/guc_hooks.h"
82 : #include "utils/memutils.h"
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 int PqSendPointer; /* Next index to store a byte in PqSendBuffer */
125 : static int 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 int internal_putbytes(const char *s, size_t len);
148 : static int internal_flush(void);
149 :
150 : static int Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath);
151 : static int Setup_AF_UNIX(const char *sock_path);
152 :
153 : static const PQcommMethods PqCommSocketMethods = {
154 : socket_comm_reset,
155 : socket_flush,
156 : socket_flush_if_writable,
157 : socket_is_send_pending,
158 : socket_putmessage,
159 : socket_putmessage_noblock
160 : };
161 :
162 : const PQcommMethods *PqCommMethods = &PqCommSocketMethods;
163 :
164 : WaitEventSet *FeBeWaitSet;
165 :
166 :
167 : /* --------------------------------
8750 tgl 168 ECB : * pq_init - initialize libpq at backend startup
169 : * --------------------------------
170 : */
171 : void
8750 tgl 172 GIC 8662 : pq_init(void)
173 : {
697 tgl 174 ECB : int socket_pos PG_USED_FOR_ASSERTS_ONLY;
175 : int latch_pos PG_USED_FOR_ASSERTS_ONLY;
769 tmunro 176 :
2807 tgl 177 : /* initialize state variables */
4393 heikki.linnakangas 178 CBC 8662 : PqSendBufferSize = PQ_SEND_BUFFER_SIZE;
4393 heikki.linnakangas 179 GIC 8662 : PqSendBuffer = MemoryContextAlloc(TopMemoryContext, PqSendBufferSize);
180 8662 : PqSendPointer = PqSendStart = PqRecvPointer = PqRecvLength = 0;
6769 tgl 181 CBC 8662 : PqCommBusy = false;
2988 heikki.linnakangas 182 GIC 8662 : PqCommReadingMsg = false;
183 :
184 : /* set up process-exit hook to close the socket */
3082 rhaas 185 8662 : on_proc_exit(socket_close, 0);
186 :
187 : /*
188 : * In backends (as soon as forked) we operate the underlying socket in
189 : * nonblocking mode and use latches to implement blocking semantics if
190 : * needed. That allows us to provide safely interruptible reads and
191 : * writes.
192 : *
193 : * Use COMMERROR on failure, because ERROR would try to send the error to
2987 andres 194 ECB : * the client, which might require changing the mode again, leading to
2987 andres 195 EUB : * infinite recursion.
196 : */
197 : #ifndef WIN32
2987 andres 198 GIC 8662 : if (!pg_set_noblock(MyProcPort->sock))
2987 andres 199 UIC 0 : ereport(COMMERROR,
200 : (errmsg("could not set socket to nonblocking mode: %m")));
201 : #endif
37 tmunro 202 ECB :
203 : #ifndef WIN32
204 :
205 : /* Don't give the socket to any subprograms we execute. */
37 tmunro 206 GNC 8662 : if (fcntl(MyProcPort->sock, F_SETFD, FD_CLOEXEC) < 0)
37 tmunro 207 UNC 0 : elog(FATAL, "fcntl(F_SETFD) failed on socket: %m");
208 : #endif
209 :
419 tmunro 210 GBC 8662 : FeBeWaitSet = CreateWaitEventSet(TopMemoryContext, FeBeWaitSetNEvents);
769 tmunro 211 GIC 8662 : socket_pos = AddWaitEventToSet(FeBeWaitSet, WL_SOCKET_WRITEABLE,
212 8662 : MyProcPort->sock, NULL, NULL);
769 tmunro 213 CBC 8662 : latch_pos = AddWaitEventToSet(FeBeWaitSet, WL_LATCH_SET, PGINVALID_SOCKET,
769 tmunro 214 ECB : MyLatch, NULL);
769 tmunro 215 CBC 8662 : AddWaitEventToSet(FeBeWaitSet, WL_POSTMASTER_DEATH, PGINVALID_SOCKET,
2575 andres 216 ECB : NULL, NULL);
217 :
769 tmunro 218 : /*
219 : * The event positions match the order we added them, but let's sanity
220 : * check them to be sure.
221 : */
769 tmunro 222 GIC 8662 : Assert(socket_pos == FeBeWaitSetSocketPos);
223 8662 : Assert(latch_pos == FeBeWaitSetLatchPos);
9770 scrappy 224 8662 : }
9770 scrappy 225 ECB :
6769 tgl 226 : /* --------------------------------
3082 rhaas 227 : * socket_comm_reset - reset libpq during error recovery
228 : *
229 : * This is called from error recovery at the outer idle loop. It's
230 : * just to get us out of trouble if we somehow manage to elog() from
231 : * inside a pqcomm.c routine (which ideally will never happen, but...)
232 : * --------------------------------
233 : */
234 : static void
3082 rhaas 235 GIC 17741 : socket_comm_reset(void)
236 : {
237 : /* Do not throw away pending data, but do reset the busy flag */
6769 tgl 238 CBC 17741 : PqCommBusy = false;
6769 tgl 239 GIC 17741 : }
240 :
9770 scrappy 241 ECB : /* --------------------------------
3082 rhaas 242 : * socket_close - shutdown libpq at backend exit
243 : *
244 : * This is the one pg_on_exit_callback in place during BackendInitialize().
245 : * That function's unusual signal handling constrains that this callback be
246 : * safe to run at any instant.
247 : * --------------------------------
248 : */
249 : static void
3082 rhaas 250 GIC 8662 : socket_close(int code, Datum arg)
251 : {
252 : /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
8661 tgl 253 CBC 8662 : if (MyProcPort != NULL)
254 : {
255 : #ifdef ENABLE_GSS
2883 noah 256 ECB : /*
257 : * Shutdown GSSAPI layer. This section does nothing when interrupting
258 : * BackendInitialize(), because pg_GSS_recvauth() makes first use of
259 : * "ctx" and "cred".
260 : *
261 : * Note that we don't bother to free MyProcPort->gss, since we're
262 : * about to exit anyway.
263 : */
832 tgl 264 GIC 8662 : if (MyProcPort->gss)
265 : {
266 : OM_uint32 min_s;
5752 magnus 267 ECB :
832 tgl 268 GIC 23 : if (MyProcPort->gss->ctx != GSS_C_NO_CONTEXT)
269 21 : gss_delete_sec_context(&min_s, &MyProcPort->gss->ctx, NULL);
270 :
832 tgl 271 CBC 23 : if (MyProcPort->gss->cred != GSS_C_NO_CREDENTIAL)
832 tgl 272 LBC 0 : gss_release_cred(&min_s, &MyProcPort->gss->cred);
273 : }
832 tgl 274 ECB : #endif /* ENABLE_GSS */
5752 magnus 275 EUB :
276 : /*
277 : * Cleanly shut down SSL layer. Nowhere else does a postmaster child
278 : * call this, so this is safe when interrupting BackendInitialize().
279 : */
7604 bruce 280 GIC 8662 : secure_close(MyProcPort);
281 :
282 : /*
383 tgl 283 ECB : * Formerly we did an explicit close() here, but it seems better to
284 : * leave the socket open until the process dies. This allows clients
285 : * to perform a "synchronous close" if they care --- wait till the
286 : * transport layer reports connection closure, and you can be sure the
287 : * backend has exited.
288 : *
289 : * We do set sock to PGINVALID_SOCKET to prevent any further I/O,
290 : * though.
291 : */
4837 magnus 292 GIC 8662 : MyProcPort->sock = PGINVALID_SOCKET;
293 : }
9770 scrappy 294 8662 : }
9770 scrappy 295 ECB :
296 :
297 :
298 : /*
299 : * Streams -- wrapper around Unix socket system calls
300 : *
301 : *
302 : * Stream functions are used for vanilla TCP connection protocol.
303 : */
304 :
305 :
306 : /*
307 : * StreamServerPort -- open a "listening" port to accept connections.
308 : *
309 : * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
310 : * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
311 : * specified. For TCP ports, hostName is either NULL for all interfaces or
312 : * the interface to listen on, and unixSocketDir is ignored (can be NULL).
313 : *
314 : * Successfully opened sockets are added to the ListenSocket[] array (of
315 : * length MaxListen), at the first position that isn't PGINVALID_SOCKET.
316 : *
317 : * RETURNS: STATUS_OK or STATUS_ERROR
318 : */
319 :
320 : int
1164 peter 321 GIC 624 : StreamServerPort(int family, const char *hostName, unsigned short portNumber,
322 : const char *unixSocketDir,
323 : pgsocket ListenSocket[], int MaxListen)
9770 scrappy 324 ECB : {
325 : pgsocket fd;
326 : int err;
327 : int maxconn;
328 : int ret;
329 : char portNumberStr[32];
330 : const char *familyDesc;
331 : char familyDescBuf[64];
332 : const char *addrDesc;
333 : char addrBuf[NI_MAXHOST];
334 : char *service;
7188 bruce 335 GIC 624 : struct addrinfo *addrs = NULL,
336 : *addr;
337 : struct addrinfo hint;
7241 bruce 338 CBC 624 : int listen_index = 0;
7241 bruce 339 GIC 624 : int added = 0;
3894 tgl 340 ECB : char unixSocketPath[MAXPGPATH];
341 : #if !defined(WIN32) || defined(IPV6_V6ONLY)
5738 magnus 342 CBC 624 : int one = 1;
343 : #endif
344 :
345 : /* Initialize hint structure */
7398 bruce 346 4368 : MemSet(&hint, 0, sizeof(hint));
347 624 : hint.ai_family = family;
7200 tgl 348 624 : hint.ai_flags = AI_PASSIVE;
7398 bruce 349 624 : hint.ai_socktype = SOCK_STREAM;
350 :
7429 bruce 351 GIC 624 : if (family == AF_UNIX)
352 : {
353 : /*
354 : * Create unixSocketPath from portNumber and unixSocketDir and lock
355 : * that file path
3894 tgl 356 ECB : */
3894 tgl 357 CBC 590 : UNIXSOCK_PATH(unixSocketPath, portNumber, unixSocketDir);
3783 tgl 358 GIC 590 : if (strlen(unixSocketPath) >= UNIXSOCK_PATH_BUFLEN)
3783 tgl 359 EUB : {
3783 tgl 360 UIC 0 : ereport(LOG,
361 : (errmsg("Unix-domain socket path \"%s\" is too long (maximum %d bytes)",
362 : unixSocketPath,
3783 tgl 363 EUB : (int) (UNIXSOCK_PATH_BUFLEN - 1))));
3783 tgl 364 UIC 0 : return STATUS_ERROR;
3783 tgl 365 ECB : }
3894 tgl 366 GBC 590 : if (Lock_AF_UNIX(unixSocketDir, unixSocketPath) != STATUS_OK)
7398 bruce 367 LBC 0 : return STATUS_ERROR;
3894 tgl 368 GIC 590 : service = unixSocketPath;
369 : }
370 : else
7398 bruce 371 ECB : {
7398 bruce 372 GIC 34 : snprintf(portNumberStr, sizeof(portNumberStr), "%d", portNumber);
373 34 : service = portNumberStr;
7398 bruce 374 ECB : }
7289 375 :
6383 tgl 376 GIC 624 : ret = pg_getaddrinfo_all(hostName, service, &hint, &addrs);
7200 tgl 377 GBC 624 : if (ret || !addrs)
7398 bruce 378 EUB : {
7200 tgl 379 UIC 0 : if (hostName)
380 0 : ereport(LOG,
381 : (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
7200 tgl 382 EUB : hostName, service, gai_strerror(ret))));
383 : else
7200 tgl 384 UIC 0 : ereport(LOG,
2118 tgl 385 EUB : (errmsg("could not translate service \"%s\" to address: %s",
386 : service, gai_strerror(ret))));
6892 bruce 387 UBC 0 : if (addrs)
6383 tgl 388 UIC 0 : pg_freeaddrinfo_all(hint.ai_family, addrs);
7398 bruce 389 0 : return STATUS_ERROR;
7398 bruce 390 ECB : }
391 :
7241 bruce 392 CBC 1248 : for (addr = addrs; addr; addr = addr->ai_next)
393 : {
418 peter 394 GIC 624 : if (family != AF_UNIX && addr->ai_family == AF_UNIX)
395 : {
396 : /*
397 : * Only set up a unix domain socket when they really asked for it.
6385 bruce 398 EUB : * The service/port is different in that case.
399 : */
7241 bruce 400 UIC 0 : continue;
401 : }
7429 bruce 402 ECB :
403 : /* See if there is still room to add 1 more socket. */
7241 bruce 404 CBC 655 : for (; listen_index < MaxListen; listen_index++)
7241 bruce 405 ECB : {
4837 magnus 406 GIC 655 : if (ListenSocket[listen_index] == PGINVALID_SOCKET)
7241 bruce 407 CBC 624 : break;
408 : }
7200 tgl 409 GBC 624 : if (listen_index >= MaxListen)
410 : {
6661 tgl 411 UIC 0 : ereport(LOG,
6661 tgl 412 EUB : (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
413 : MaxListen)));
7241 bruce 414 UIC 0 : break;
415 : }
7200 tgl 416 ECB :
417 : /* set up address family name for log messages */
7180 tgl 418 CBC 624 : switch (addr->ai_family)
7180 tgl 419 ECB : {
7180 tgl 420 CBC 34 : case AF_INET:
6620 bruce 421 GBC 34 : familyDesc = _("IPv4");
7180 tgl 422 34 : break;
7180 tgl 423 LBC 0 : case AF_INET6:
6620 bruce 424 0 : familyDesc = _("IPv6");
7180 tgl 425 0 : break;
7180 tgl 426 GBC 590 : case AF_UNIX:
6620 bruce 427 GIC 590 : familyDesc = _("Unix");
7180 tgl 428 GBC 590 : break;
7180 tgl 429 UIC 0 : default:
430 0 : snprintf(familyDescBuf, sizeof(familyDescBuf),
6620 bruce 431 0 : _("unrecognized address family %d"),
7180 tgl 432 ECB : addr->ai_family);
7180 tgl 433 LBC 0 : familyDesc = familyDescBuf;
7180 tgl 434 UIC 0 : break;
435 : }
7180 tgl 436 ECB :
2221 437 : /* set up text form of address for log messages */
2221 tgl 438 GIC 624 : if (addr->ai_family == AF_UNIX)
439 590 : addrDesc = unixSocketPath;
2221 tgl 440 ECB : else
441 : {
2221 tgl 442 CBC 34 : pg_getnameinfo_all((const struct sockaddr_storage *) addr->ai_addr,
2221 tgl 443 GIC 34 : addr->ai_addrlen,
2221 tgl 444 EUB : addrBuf, sizeof(addrBuf),
445 : NULL, 0,
446 : NI_NUMERICHOST);
2221 tgl 447 GIC 34 : addrDesc = addrBuf;
448 : }
2221 tgl 449 EUB :
3280 bruce 450 GIC 624 : if ((fd = socket(addr->ai_family, SOCK_STREAM, 0)) == PGINVALID_SOCKET)
451 : {
7201 tgl 452 UIC 0 : ereport(LOG,
453 : (errcode_for_socket_access(),
454 : /* translator: first %s is IPv4, IPv6, or Unix */
455 : errmsg("could not create %s socket for address \"%s\": %m",
456 : familyDesc, addrDesc)));
7241 bruce 457 0 : continue;
458 : }
459 :
460 : #ifndef WIN32
461 :
462 : /*
463 : * Without the SO_REUSEADDR flag, a new postmaster can't be started
464 : * right away after a stop or crash, giving "address already in use"
5624 bruce 465 ECB : * error on TCP ports.
466 : *
467 : * On win32, however, this behavior only happens if the
468 : * SO_EXCLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows
469 : * multiple servers to listen on the same address, resulting in
1335 michael 470 EUB : * unpredictable behavior. With no flags at all, win32 behaves as Unix
471 : * with SO_REUSEADDR.
472 : */
418 peter 473 GIC 624 : if (addr->ai_family != AF_UNIX)
474 : {
7241 bruce 475 34 : if ((setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
7188 bruce 476 EUB : (char *) &one, sizeof(one))) == -1)
7241 477 : {
7201 tgl 478 UIC 0 : ereport(LOG,
479 : (errcode_for_socket_access(),
480 : /* translator: third %s is IPv4, IPv6, or Unix */
481 : errmsg("%s(%s) failed for %s address \"%s\": %m",
482 : "setsockopt", "SO_REUSEADDR",
2221 tgl 483 ECB : familyDesc, addrDesc)));
7241 bruce 484 UIC 0 : closesocket(fd);
7241 bruce 485 UBC 0 : continue;
486 : }
487 : }
5788 magnus 488 EUB : #endif
489 :
490 : #ifdef IPV6_V6ONLY
7241 bruce 491 GIC 624 : if (addr->ai_family == AF_INET6)
492 : {
7241 bruce 493 UIC 0 : if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY,
7188 bruce 494 EUB : (char *) &one, sizeof(one)) == -1)
7241 495 : {
7201 tgl 496 UIC 0 : ereport(LOG,
497 : (errcode_for_socket_access(),
498 : /* translator: third %s is IPv4, IPv6, or Unix */
499 : errmsg("%s(%s) failed for %s address \"%s\": %m",
500 : "setsockopt", "IPV6_V6ONLY",
501 : familyDesc, addrDesc)));
7241 bruce 502 0 : closesocket(fd);
503 0 : continue;
504 : }
505 : }
7241 bruce 506 ECB : #endif
507 :
7241 bruce 508 EUB : /*
7188 509 : * Note: This might fail on some OS's, like Linux older than
510 : * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
3260 511 : * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
512 : * connections.
513 : */
7241 bruce 514 GIC 624 : err = bind(fd, addr->ai_addr, addr->ai_addrlen);
515 624 : if (err < 0)
7241 bruce 516 UIC 0 : {
865 peter 517 0 : int saved_errno = errno;
518 :
7201 tgl 519 0 : ereport(LOG,
520 : (errcode_for_socket_access(),
521 : /* translator: first %s is IPv4, IPv6, or Unix */
522 : errmsg("could not bind %s address \"%s\": %m",
2221 tgl 523 EUB : familyDesc, addrDesc),
865 peter 524 : saved_errno == EADDRINUSE ?
525 : (addr->ai_family == AF_UNIX ?
526 : errhint("Is another postmaster already running on port %d?",
865 peter 527 ECB : (int) portNumber) :
528 : errhint("Is another postmaster already running on port %d?"
529 : " If not, wait a few seconds and retry.",
530 : (int) portNumber)) : 0));
7241 bruce 531 UBC 0 : closesocket(fd);
532 0 : continue;
533 : }
534 :
7241 bruce 535 GIC 624 : if (addr->ai_family == AF_UNIX)
536 : {
3894 tgl 537 590 : if (Setup_AF_UNIX(service) != STATUS_OK)
538 : {
7241 bruce 539 UIC 0 : closesocket(fd);
7241 bruce 540 LBC 0 : break;
541 : }
7429 bruce 542 ECB : }
543 :
7241 bruce 544 EUB : /*
545 : * Select appropriate accept-queue length limit. It seems reasonable
546 : * to use a value similar to the maximum number of child processes
547 : * that the postmaster will permit.
548 : */
229 tgl 549 GNC 624 : maxconn = MaxConnections * 2;
550 :
7241 bruce 551 CBC 624 : err = listen(fd, maxconn);
552 624 : if (err < 0)
553 : {
7201 tgl 554 UIC 0 : ereport(LOG,
555 : (errcode_for_socket_access(),
2221 tgl 556 ECB : /* translator: first %s is IPv4, IPv6, or Unix */
557 : errmsg("could not listen on %s address \"%s\": %m",
558 : familyDesc, addrDesc)));
7241 bruce 559 UIC 0 : closesocket(fd);
560 0 : continue;
7241 bruce 561 ECB : }
2221 tgl 562 :
2217 tgl 563 GIC 624 : if (addr->ai_family == AF_UNIX)
2217 tgl 564 CBC 590 : ereport(LOG,
565 : (errmsg("listening on Unix socket \"%s\"",
2217 tgl 566 ECB : addrDesc)));
2217 tgl 567 EUB : else
2217 tgl 568 CBC 34 : ereport(LOG,
569 : /* translator: first %s is IPv4 or IPv6 */
570 : (errmsg("listening on %s address \"%s\", port %d",
571 : familyDesc, addrDesc, (int) portNumber)));
572 :
7241 bruce 573 GIC 624 : ListenSocket[listen_index] = fd;
574 624 : added++;
575 : }
7942 tgl 576 ECB :
6383 tgl 577 GIC 624 : pg_freeaddrinfo_all(hint.ai_family, addrs);
578 :
7241 bruce 579 CBC 624 : if (!added)
7429 bruce 580 UBC 0 : return STATUS_ERROR;
581 :
7398 bruce 582 GIC 624 : return STATUS_OK;
583 : }
584 :
585 :
586 : /*
587 : * Lock_AF_UNIX -- configure unix socket file path
7398 bruce 588 ECB : */
589 : static int
1164 peter 590 GIC 590 : Lock_AF_UNIX(const char *unixSocketDir, const char *unixSocketPath)
591 : {
592 : /* no lock file for abstract sockets */
865 593 590 : if (unixSocketPath[0] == '@')
865 peter 594 LBC 0 : return STATUS_OK;
595 :
596 : /*
597 : * Grab an interlock file associated with the socket file.
598 : *
4609 tgl 599 ECB : * Note: there are two reasons for using a socket lock file, rather than
600 : * trying to interlock directly on the socket itself. First, it's a lot
601 : * more portable, and second, it lets us remove any pre-existing socket
602 : * file without race conditions.
603 : */
3894 tgl 604 GIC 590 : CreateSocketLockFile(unixSocketPath, true, unixSocketDir);
605 :
606 : /*
607 : * Once we have the interlock, we can safely delete any pre-existing
608 : * socket file to avoid failure at bind() time.
7398 bruce 609 ECB : */
2807 tgl 610 GIC 590 : (void) unlink(unixSocketPath);
611 :
3894 tgl 612 ECB : /*
2807 tgl 613 EUB : * Remember socket file pathnames for later maintenance.
614 : */
3894 tgl 615 GIC 590 : sock_paths = lappend(sock_paths, pstrdup(unixSocketPath));
616 :
7398 bruce 617 590 : return STATUS_OK;
618 : }
619 :
7398 bruce 620 ECB :
621 : /*
622 : * Setup_AF_UNIX -- configure unix socket permissions
623 : */
624 : static int
1164 peter 625 GIC 590 : Setup_AF_UNIX(const char *sock_path)
626 : {
627 : /* no file system permissions for abstract sockets */
865 628 590 : if (sock_path[0] == '@')
865 peter 629 UIC 0 : return STATUS_OK;
865 peter 630 EUB :
7398 bruce 631 : /*
632 : * Fix socket ownership/permission if requested. Note we must do this
6385 633 : * before we listen() to avoid a window where unwanted connections could
634 : * get accepted.
635 : */
7398 bruce 636 GIC 590 : Assert(Unix_socket_group);
637 590 : if (Unix_socket_group[0] != '\0')
638 : {
7269 bruce 639 EUB : #ifdef WIN32
7201 tgl 640 : elog(WARNING, "configuration item unix_socket_group is not supported on this platform");
641 : #else
7398 bruce 642 : char *endptr;
643 : unsigned long val;
644 : gid_t gid;
645 :
7398 bruce 646 UIC 0 : val = strtoul(Unix_socket_group, &endptr, 10);
7398 bruce 647 UBC 0 : if (*endptr == '\0')
648 : { /* numeric group id */
649 0 : gid = val;
650 : }
7398 bruce 651 EUB : else
652 : { /* convert group name to id */
653 : struct group *gr;
654 :
7398 bruce 655 UBC 0 : gr = getgrnam(Unix_socket_group);
7398 bruce 656 UIC 0 : if (!gr)
657 : {
7201 tgl 658 0 : ereport(LOG,
659 : (errmsg("group \"%s\" does not exist",
7201 tgl 660 ECB : Unix_socket_group)));
7398 bruce 661 UIC 0 : return STATUS_ERROR;
7398 bruce 662 EUB : }
7398 bruce 663 UIC 0 : gid = gr->gr_gid;
664 : }
665 0 : if (chown(sock_path, -1, gid) == -1)
7398 bruce 666 EUB : {
7201 tgl 667 UIC 0 : ereport(LOG,
7201 tgl 668 ECB : (errcode_for_file_access(),
669 : errmsg("could not set group of file \"%s\": %m",
670 : sock_path)));
7398 bruce 671 UIC 0 : return STATUS_ERROR;
672 : }
673 : #endif
674 : }
675 :
7398 bruce 676 GIC 590 : if (chmod(sock_path, Unix_socket_permissions) == -1)
677 : {
7201 tgl 678 UIC 0 : ereport(LOG,
679 : (errcode_for_file_access(),
680 : errmsg("could not set permissions of file \"%s\": %m",
681 : sock_path)));
7398 bruce 682 LBC 0 : return STATUS_ERROR;
683 : }
7429 bruce 684 GIC 590 : return STATUS_OK;
9770 scrappy 685 ECB : }
7398 bruce 686 :
687 :
688 : /*
9770 scrappy 689 EUB : * StreamConnection -- create a new connection with client using
690 : * server port. Set port->sock to the FD of the new connection.
691 : *
692 : * ASSUME: that this doesn't need to be non-blocking because
693 : * the Postmaster waits for the socket to be ready to accept().
694 : *
695 : * RETURNS: STATUS_OK or STATUS_ERROR
696 : */
697 : int
4837 magnus 698 GIC 8835 : StreamConnection(pgsocket server_fd, Port *port)
9770 scrappy 699 EUB : {
7201 tgl 700 : /* accept connection and fill in the client (remote) address */
7241 bruce 701 GIC 8835 : port->raddr.salen = sizeof(port->raddr.addr);
9345 702 8835 : if ((port->sock = accept(server_fd,
2118 tgl 703 8835 : (struct sockaddr *) &port->raddr.addr,
3280 bruce 704 ECB : &port->raddr.salen)) == PGINVALID_SOCKET)
9345 705 : {
7201 tgl 706 LBC 0 : ereport(LOG,
707 : (errcode_for_socket_access(),
708 : errmsg("could not accept new connection: %m")));
5624 bruce 709 EUB :
710 : /*
5899 tgl 711 : * If accept() fails then postmaster.c will still see the server
712 : * socket as read-ready, and will immediately try again. To avoid
713 : * uselessly sucking lots of CPU, delay a bit before trying again.
714 : * (The most likely reason for failure is being out of kernel file
5899 tgl 715 ECB : * table slots; we can do little except hope some will get freed up.)
716 : */
5899 tgl 717 UIC 0 : pg_usleep(100000L); /* wait 0.1 sec */
8986 bruce 718 0 : return STATUS_ERROR;
719 : }
720 :
721 : /* fill in the server (local) address */
7241 bruce 722 GIC 8835 : port->laddr.salen = sizeof(port->laddr.addr);
7201 tgl 723 8835 : if (getsockname(port->sock,
2118 724 8835 : (struct sockaddr *) &port->laddr.addr,
7241 bruce 725 ECB : &port->laddr.salen) < 0)
9345 726 : {
856 peter 727 UIC 0 : ereport(LOG,
728 : (errmsg("%s() failed: %m", "getsockname")));
8986 bruce 729 UBC 0 : return STATUS_ERROR;
730 : }
8848 tgl 731 EUB :
732 : /* select NODELAY and KEEPALIVE options if it's a TCP connection */
418 peter 733 GIC 8835 : if (port->laddr.addr.ss_family != AF_UNIX)
9345 bruce 734 ECB : {
7201 tgl 735 : int on;
736 : #ifdef WIN32
737 : int oldopt;
2834 heikki.linnakangas 738 EUB : int optlen;
739 : int newopt;
740 : #endif
741 :
742 : #ifdef TCP_NODELAY
7201 tgl 743 GIC 200 : on = 1;
8358 744 200 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_NODELAY,
745 : (char *) &on, sizeof(on)) < 0)
746 : {
856 peter 747 UIC 0 : ereport(LOG,
748 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_NODELAY")));
8359 ishii 749 0 : return STATUS_ERROR;
750 : }
751 : #endif
7201 tgl 752 GIC 200 : on = 1;
8359 ishii 753 200 : if (setsockopt(port->sock, SOL_SOCKET, SO_KEEPALIVE,
754 : (char *) &on, sizeof(on)) < 0)
755 : {
856 peter 756 UIC 0 : ereport(LOG,
757 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_KEEPALIVE")));
8986 bruce 758 0 : return STATUS_ERROR;
759 : }
760 :
761 : #ifdef WIN32
762 :
763 : /*
764 : * This is a Win32 socket optimization. The OS send buffer should be
765 : * large enough to send the whole Postgres send buffer in one go, or
766 : * performance suffers. The Postgres send buffer can be enlarged if a
767 : * very large message needs to be sent, but we won't attempt to
768 : * enlarge the OS buffer if that happens, so somewhat arbitrarily
769 : * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
770 : * (That's 32kB with the current default).
771 : *
772 : * The default OS buffer size used to be 8kB in earlier Windows
773 : * versions, but was raised to 64kB in Windows 2012. So it shouldn't
774 : * be necessary to change it in later versions anymore. Changing it
775 : * unnecessarily can even reduce performance, because setting
776 : * SO_SNDBUF in the application disables the "dynamic send buffering"
777 : * feature that was introduced in Windows 7. So before fiddling with
778 : * SO_SNDBUF, check if the current buffer size is already large enough
779 : * and only increase it if necessary.
780 : *
781 : * See https://support.microsoft.com/kb/823764/EN-US/ and
782 : * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
783 : */
784 : optlen = sizeof(oldopt);
785 : if (getsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &oldopt,
786 : &optlen) < 0)
787 : {
788 : ereport(LOG,
789 : (errmsg("%s(%s) failed: %m", "getsockopt", "SO_SNDBUF")));
790 : return STATUS_ERROR;
791 : }
792 : newopt = PQ_SEND_BUFFER_SIZE * 4;
793 : if (oldopt < newopt)
2834 heikki.linnakangas 794 ECB : {
795 : if (setsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &newopt,
796 : sizeof(newopt)) < 0)
797 : {
798 : ereport(LOG,
799 : (errmsg("%s(%s) failed: %m", "setsockopt", "SO_SNDBUF")));
800 : return STATUS_ERROR;
801 : }
802 : }
803 : #endif
804 :
805 : /*
806 : * Also apply the current keepalive parameters. If we fail to set a
807 : * parameter, don't error out, because these aren't universally
808 : * supported. (Note: you might think we need to reset the GUC
809 : * variables to 0 in such a case, but it's not necessary because the
810 : * show hooks for these variables report the truth anyway.)
811 : */
6418 tgl 812 GIC 200 : (void) pq_setkeepalivesidle(tcp_keepalives_idle, port);
813 200 : (void) pq_setkeepalivesinterval(tcp_keepalives_interval, port);
6418 tgl 814 CBC 200 : (void) pq_setkeepalivescount(tcp_keepalives_count, port);
1464 michael 815 GIC 200 : (void) pq_settcpusertimeout(tcp_user_timeout, port);
9489 vadim4o 816 ECB : }
9345 bruce 817 :
8986 bruce 818 GIC 8835 : return STATUS_OK;
819 : }
820 :
821 : /*
822 : * StreamClose -- close a client/backend connection
823 : *
824 : * NOTE: this is NOT used to terminate a session; it is just used to release
825 : * the file descriptor in a process that should no longer have the socket
826 : * open. (For example, the postmaster calls this after passing ownership
827 : * of the connection to a child process.) It is expected that someone else
828 : * still has the socket open. So, we only want to close the descriptor,
7255 tgl 829 EUB : * we do NOT want to send anything to the far end.
830 : */
831 : void
4837 magnus 832 GIC 22543 : StreamClose(pgsocket sock)
833 : {
7289 bruce 834 GBC 22543 : closesocket(sock);
9770 scrappy 835 GIC 22543 : }
9770 scrappy 836 EUB :
837 : /*
838 : * TouchSocketFiles -- mark socket files as recently accessed
7379 tgl 839 : *
840 : * This routine should be called every so often to ensure that the socket
3894 841 : * files have a recent mod date (ordinary operations on sockets usually won't
842 : * change the mod date). That saves them from being removed by
843 : * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
844 : * never have put the socket file in /tmp...)
845 : */
846 : void
3894 tgl 847 LBC 0 : TouchSocketFiles(void)
848 : {
849 : ListCell *l;
850 :
851 : /* Loop through all created sockets... */
852 0 : foreach(l, sock_paths)
853 : {
854 0 : char *sock_path = (char *) lfirst(l);
855 :
856 : /* Ignore errors; there's no point in complaining */
1143 857 0 : (void) utime(sock_path, NULL);
858 : }
7379 tgl 859 UIC 0 : }
7379 tgl 860 ECB :
2807 861 : /*
862 : * RemoveSocketFiles -- unlink socket files at postmaster shutdown
863 : */
864 : void
2807 tgl 865 GIC 593 : RemoveSocketFiles(void)
866 : {
867 : ListCell *l;
868 :
869 : /* Loop through all created sockets... */
870 1183 : foreach(l, sock_paths)
871 : {
872 590 : char *sock_path = (char *) lfirst(l);
873 :
874 : /* Ignore any error. */
875 590 : (void) unlink(sock_path);
876 : }
877 : /* Since we're about to exit, no need to reclaim storage */
878 593 : sock_paths = NIL;
879 593 : }
2807 tgl 880 ECB :
881 :
8750 882 : /* --------------------------------
8750 tgl 883 EUB : * Low-level I/O routines begin here.
884 : *
885 : * These routines communicate with a frontend client across a connection
886 : * already established by the preceding routines.
8750 tgl 887 ECB : * --------------------------------
888 : */
889 :
890 : /* --------------------------------
891 : * socket_set_nonblocking - set socket blocking/non-blocking
892 : *
893 : * Sets the socket non-blocking if nonblocking is true, or sets it
894 : * blocking otherwise.
895 : * --------------------------------
896 : */
4393 heikki.linnakangas 897 : static void
3082 rhaas 898 GIC 2193246 : socket_set_nonblocking(bool nonblocking)
4393 heikki.linnakangas 899 ECB : {
3082 rhaas 900 GIC 2193246 : if (MyProcPort == NULL)
3082 rhaas 901 LBC 0 : ereport(ERROR,
902 : (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST),
903 : errmsg("there is no client connection")));
3082 rhaas 904 EUB :
4393 heikki.linnakangas 905 GBC 2193246 : MyProcPort->noblock = nonblocking;
906 2193246 : }
8750 tgl 907 EUB :
908 : /* --------------------------------
909 : * pq_recvbuf - load some bytes into the input buffer
8750 tgl 910 ECB : *
911 : * returns 0 if OK, EOF if trouble
912 : * --------------------------------
913 : */
914 : static int
8750 tgl 915 GIC 297406 : pq_recvbuf(void)
916 : {
917 297406 : if (PqRecvPointer > 0)
8750 tgl 918 EUB : {
8750 tgl 919 GIC 288744 : if (PqRecvLength > PqRecvPointer)
920 : {
8750 tgl 921 ECB : /* still some unread data, left-justify it in the buffer */
8720 bruce 922 LBC 0 : memmove(PqRecvBuffer, PqRecvBuffer + PqRecvPointer,
8720 bruce 923 UIC 0 : PqRecvLength - PqRecvPointer);
8750 tgl 924 LBC 0 : PqRecvLength -= PqRecvPointer;
8750 tgl 925 UIC 0 : PqRecvPointer = 0;
8750 tgl 926 ECB : }
8750 tgl 927 EUB : else
8750 tgl 928 GIC 288744 : PqRecvLength = PqRecvPointer = 0;
929 : }
930 :
931 : /* Ensure that we're in blocking mode */
3082 rhaas 932 297406 : socket_set_nonblocking(false);
933 :
8750 tgl 934 ECB : /* Can fill buffer from PqRecvLength and upwards */
935 : for (;;)
8750 tgl 936 UIC 0 : {
8595 bruce 937 ECB : int r;
938 :
7604 bruce 939 CBC 594776 : r = secure_read(MyProcPort, PqRecvBuffer + PqRecvLength,
4393 heikki.linnakangas 940 GIC 297406 : PQ_RECV_BUFFER_SIZE - PqRecvLength);
941 :
8750 tgl 942 297370 : if (r < 0)
943 : {
944 1 : if (errno == EINTR)
8750 tgl 945 LBC 0 : continue; /* Ok if interrupted */
946 :
947 : /*
6385 bruce 948 ECB : * Careful: an ereport() that tries to write to the client would
949 : * cause recursion to here, leading to stack overflow and core
950 : * dump! This message must go *only* to the postmaster log.
951 : */
7201 tgl 952 GIC 1 : ereport(COMMERROR,
953 : (errcode_for_socket_access(),
954 : errmsg("could not receive data from client: %m")));
8750 955 1 : return EOF;
956 : }
957 297369 : if (r == 0)
8750 tgl 958 ECB : {
959 : /*
6385 bruce 960 : * EOF detected. We used to write a log message here, but it's
961 : * better to expect the ultimate caller to do that.
7295 tgl 962 : */
8750 tgl 963 GIC 62 : return EOF;
8750 tgl 964 ECB : }
965 : /* r contains number of bytes read, so just incr length */
8750 tgl 966 GIC 297307 : PqRecvLength += r;
8750 tgl 967 CBC 297307 : return 0;
968 : }
969 : }
970 :
971 : /* --------------------------------
972 : * pq_getbyte - get a single byte from connection, or return EOF
973 : * --------------------------------
974 : */
975 : int
8750 tgl 976 GIC 486559 : pq_getbyte(void)
8750 tgl 977 EUB : {
2988 heikki.linnakangas 978 GIC 486559 : Assert(PqCommReadingMsg);
2988 heikki.linnakangas 979 EUB :
8750 tgl 980 GIC 737364 : while (PqRecvPointer >= PqRecvLength)
8750 tgl 981 EUB : {
8750 tgl 982 GIC 250891 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
8750 tgl 983 GBC 50 : return EOF; /* Failed to recv data */
8750 tgl 984 EUB : }
6406 tgl 985 GIC 486473 : return (unsigned char) PqRecvBuffer[PqRecvPointer++];
8750 tgl 986 EUB : }
987 :
988 : /* --------------------------------
989 : * pq_peekbyte - peek at next byte from connection
990 : *
991 : * Same as pq_getbyte() except we don't advance the pointer.
992 : * --------------------------------
993 : */
994 : int
8750 tgl 995 UIC 0 : pq_peekbyte(void)
996 : {
2984 noah 997 0 : Assert(PqCommReadingMsg);
2984 noah 998 ECB :
8750 tgl 999 UIC 0 : while (PqRecvPointer >= PqRecvLength)
1000 : {
1001 0 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
8750 tgl 1002 LBC 0 : return EOF; /* Failed to recv data */
1003 : }
6406 1004 0 : return (unsigned char) PqRecvBuffer[PqRecvPointer];
1005 : }
8977 vadim4o 1006 ECB :
4832 heikki.linnakangas 1007 : /* --------------------------------
1008 : * pq_getbyte_if_available - get a single byte from connection,
1009 : * if available
1010 : *
4798 1011 : * The received byte is stored in *c. Returns 1 if a byte was read,
1012 : * 0 if no data was available, or EOF if trouble.
4832 1013 : * --------------------------------
1014 : */
1015 : int
4832 heikki.linnakangas 1016 GIC 1045101 : pq_getbyte_if_available(unsigned char *c)
1017 : {
1018 : int r;
1019 :
2988 1020 1045101 : Assert(PqCommReadingMsg);
2988 heikki.linnakangas 1021 ECB :
4832 heikki.linnakangas 1022 CBC 1045101 : if (PqRecvPointer < PqRecvLength)
1023 : {
4832 heikki.linnakangas 1024 GIC 55270 : *c = PqRecvBuffer[PqRecvPointer++];
1025 55270 : return 1;
1026 : }
1027 :
1028 : /* Put the socket into non-blocking mode */
3082 rhaas 1029 989831 : socket_set_nonblocking(true);
4393 heikki.linnakangas 1030 ECB :
4393 heikki.linnakangas 1031 GIC 989831 : r = secure_read(MyProcPort, c, 1);
1032 989831 : if (r < 0)
4832 heikki.linnakangas 1033 ECB : {
1034 : /*
1035 : * Ok if no data available without blocking or interrupted (though
4382 bruce 1036 : * EINTR really shouldn't happen with a non-blocking socket). Report
1037 : * other errors.
1038 : */
4393 heikki.linnakangas 1039 CBC 953409 : if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
4393 heikki.linnakangas 1040 GIC 953408 : r = 0;
1041 : else
4798 heikki.linnakangas 1042 ECB : {
1043 : /*
1044 : * Careful: an ereport() that tries to write to the client would
1045 : * cause recursion to here, leading to stack overflow and core
1046 : * dump! This message must go *only* to the postmaster log.
1047 : */
4393 heikki.linnakangas 1048 GIC 1 : ereport(COMMERROR,
1049 : (errcode_for_socket_access(),
1050 : errmsg("could not receive data from client: %m")));
4798 1051 1 : r = EOF;
4798 heikki.linnakangas 1052 ECB : }
1053 : }
4393 heikki.linnakangas 1054 GIC 36422 : else if (r == 0)
1055 : {
4393 heikki.linnakangas 1056 ECB : /* EOF detected */
4393 heikki.linnakangas 1057 GIC 14 : r = EOF;
4832 heikki.linnakangas 1058 ECB : }
1059 :
4832 heikki.linnakangas 1060 CBC 989831 : return r;
1061 : }
4832 heikki.linnakangas 1062 ECB :
8750 tgl 1063 : /* --------------------------------
1064 : * pq_getbytes - get a known number of bytes from connection
1065 : *
1066 : * returns 0 if OK, EOF if trouble
1067 : * --------------------------------
8853 scrappy 1068 : */
8750 tgl 1069 : int
8750 tgl 1070 CBC 1163683 : pq_getbytes(char *s, size_t len)
8853 scrappy 1071 ECB : {
1072 : size_t amount;
8750 tgl 1073 :
2988 heikki.linnakangas 1074 GIC 1163683 : Assert(PqCommReadingMsg);
1075 :
8750 tgl 1076 2328616 : while (len > 0)
1077 : {
1078 1211448 : while (PqRecvPointer >= PqRecvLength)
1079 : {
1080 46515 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1081 13 : return EOF; /* Failed to recv data */
1082 : }
1083 1164933 : amount = PqRecvLength - PqRecvPointer;
1084 1164933 : if (amount > len)
1085 867657 : amount = len;
8750 tgl 1086 GBC 1164933 : memcpy(s, PqRecvBuffer + PqRecvPointer, amount);
8750 tgl 1087 GIC 1164933 : PqRecvPointer += amount;
1088 1164933 : s += amount;
1089 1164933 : len -= amount;
8750 tgl 1090 EUB : }
8750 tgl 1091 GIC 1163670 : return 0;
8750 tgl 1092 EUB : }
1093 :
6747 1094 : /* --------------------------------
1095 : * pq_discardbytes - throw away a known number of bytes
1096 : *
1097 : * same as pq_getbytes except we do not copy the data to anyplace.
1098 : * this is used for resynchronizing after read errors.
1099 : *
1100 : * returns 0 if OK, EOF if trouble
1101 : * --------------------------------
1102 : */
1103 : static int
6747 tgl 1104 UIC 0 : pq_discardbytes(size_t len)
6747 tgl 1105 EUB : {
1106 : size_t amount;
1107 :
2988 heikki.linnakangas 1108 UIC 0 : Assert(PqCommReadingMsg);
1109 :
6747 tgl 1110 0 : while (len > 0)
1111 : {
1112 0 : while (PqRecvPointer >= PqRecvLength)
1113 : {
1114 0 : if (pq_recvbuf()) /* If nothing in buffer, then recv some */
6747 tgl 1115 LBC 0 : return EOF; /* Failed to recv data */
1116 : }
1117 0 : amount = PqRecvLength - PqRecvPointer;
6747 tgl 1118 UIC 0 : if (amount > len)
1119 0 : amount = len;
1120 0 : PqRecvPointer += amount;
1121 0 : len -= amount;
1122 : }
1123 0 : return 0;
1124 : }
1125 :
1126 : /* --------------------------------
1127 : * pq_buffer_has_data - is any buffered data available to read?
517 tgl 1128 ECB : *
1129 : * This will *not* attempt to read more data.
1130 : * --------------------------------
1131 : */
1132 : bool
517 tgl 1133 GIC 175 : pq_buffer_has_data(void)
517 tgl 1134 ECB : {
517 tgl 1135 GBC 175 : return (PqRecvPointer < PqRecvLength);
1136 : }
1137 :
1138 :
2988 heikki.linnakangas 1139 ECB : /* --------------------------------
2878 bruce 1140 : * pq_startmsgread - begin reading a message from the client.
1141 : *
1142 : * This must be called before any of the pq_get* functions.
1143 : * --------------------------------
1144 : */
1145 : void
2988 heikki.linnakangas 1146 GIC 1540497 : pq_startmsgread(void)
1147 : {
1148 : /*
1149 : * There shouldn't be a read active already, but let's check just to be
1150 : * sure.
1151 : */
2988 heikki.linnakangas 1152 CBC 1540497 : if (PqCommReadingMsg)
2988 heikki.linnakangas 1153 UIC 0 : ereport(FATAL,
2988 heikki.linnakangas 1154 ECB : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1155 : errmsg("terminating connection because protocol synchronization was lost")));
1156 :
2988 heikki.linnakangas 1157 CBC 1540497 : PqCommReadingMsg = true;
2988 heikki.linnakangas 1158 GIC 1540497 : }
1159 :
1160 :
1161 : /* --------------------------------
1162 : * pq_endmsgread - finish reading message.
1163 : *
1164 : * This must be called after reading a message with pq_getbytes()
1165 : * and friends, to indicate that we have read the whole message.
1166 : * pq_getmessage() does this implicitly.
1167 : * --------------------------------
2988 heikki.linnakangas 1168 ECB : */
1169 : void
2988 heikki.linnakangas 1170 CBC 962232 : pq_endmsgread(void)
1171 : {
2988 heikki.linnakangas 1172 GIC 962232 : Assert(PqCommReadingMsg);
1173 :
1174 962232 : PqCommReadingMsg = false;
1175 962232 : }
1176 :
1177 : /* --------------------------------
1178 : * pq_is_reading_msg - are we currently reading a message?
1179 : *
1180 : * This is used in error recovery at the outer idle loop to detect if we have
1181 : * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1182 : * will check for that too, but it's nicer to detect it earlier.
1183 : * --------------------------------
1184 : */
1185 : bool
1186 17741 : pq_is_reading_msg(void)
1187 : {
1188 17741 : return PqCommReadingMsg;
1189 : }
2988 heikki.linnakangas 1190 ECB :
1191 : /* --------------------------------
1192 : * pq_getmessage - get a message with length word from connection
1193 : *
7295 tgl 1194 : * The return value is placed in an expansible StringInfo, which has
1195 : * already been initialized by the caller.
1196 : * Only the message body is placed in the StringInfo; the length word
1197 : * is removed. Also, s->cursor is initialized to zero for convenience
1198 : * in scanning the message contents.
1199 : *
1200 : * maxlen is the upper limit on the length of the
7295 tgl 1201 EUB : * message we are willing to accept. We abort the connection (by
1202 : * returning EOF) if client tries to send more than that.
1203 : *
1204 : * returns 0 if OK, EOF if trouble
1205 : * --------------------------------
1206 : */
7295 tgl 1207 ECB : int
7295 tgl 1208 GIC 578151 : pq_getmessage(StringInfo s, int maxlen)
7295 tgl 1209 ECB : {
1210 : int32 len;
7295 tgl 1211 EUB :
2988 heikki.linnakangas 1212 GIC 578151 : Assert(PqCommReadingMsg);
1213 :
5881 neilc 1214 GBC 578151 : resetStringInfo(s);
1215 :
1216 : /* Read message length word */
7295 tgl 1217 CBC 578151 : if (pq_getbytes((char *) &len, 4) == EOF)
1218 : {
7201 tgl 1219 LBC 0 : ereport(COMMERROR,
1220 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1221 : errmsg("unexpected EOF within message length word")));
7295 tgl 1222 UIC 0 : return EOF;
1223 : }
1224 :
2016 andres 1225 GIC 578151 : len = pg_ntoh32(len);
7295 tgl 1226 ECB :
711 tgl 1227 GIC 578151 : if (len < 4 || len > maxlen)
7295 tgl 1228 ECB : {
7201 tgl 1229 UIC 0 : ereport(COMMERROR,
7201 tgl 1230 EUB : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1231 : errmsg("invalid message length")));
7295 tgl 1232 UBC 0 : return EOF;
7295 tgl 1233 EUB : }
1234 :
6747 tgl 1235 GIC 578151 : len -= 4; /* discount length itself */
1236 :
7295 1237 578151 : if (len > 0)
7295 tgl 1238 EUB : {
6747 1239 : /*
1240 : * Allocate space for message. If we run out of room (ridiculously
6747 tgl 1241 ECB : * large message), we will elog(ERROR), but we want to discard the
1242 : * message body so as not to lose communication sync.
1243 : */
6747 tgl 1244 CBC 559047 : PG_TRY();
1245 : {
6747 tgl 1246 GBC 559047 : enlargeStringInfo(s, len);
1247 : }
6747 tgl 1248 UIC 0 : PG_CATCH();
6747 tgl 1249 EUB : {
6747 tgl 1250 UIC 0 : if (pq_discardbytes(len) == EOF)
6747 tgl 1251 LBC 0 : ereport(COMMERROR,
1252 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
6747 tgl 1253 ECB : errmsg("incomplete message from client")));
1254 :
1255 : /* we discarded the rest of the message so we're back in sync. */
2988 heikki.linnakangas 1256 UIC 0 : PqCommReadingMsg = false;
6747 tgl 1257 LBC 0 : PG_RE_THROW();
1258 : }
6747 tgl 1259 CBC 559047 : PG_END_TRY();
1260 :
1261 : /* And grab the message */
7295 tgl 1262 GIC 559047 : if (pq_getbytes(s->data, len) == EOF)
1263 : {
7201 tgl 1264 LBC 0 : ereport(COMMERROR,
1265 : (errcode(ERRCODE_PROTOCOL_VIOLATION),
1266 : errmsg("incomplete message from client")));
7522 tgl 1267 UIC 0 : return EOF;
7295 tgl 1268 ECB : }
7295 tgl 1269 GIC 559047 : s->len = len;
1270 : /* Place a trailing null per StringInfo convention */
7295 tgl 1271 CBC 559047 : s->data[len] = '\0';
1272 : }
7295 tgl 1273 ECB :
2988 heikki.linnakangas 1274 : /* finished reading the message. */
2988 heikki.linnakangas 1275 CBC 578151 : PqCommReadingMsg = false;
1276 :
7295 tgl 1277 578151 : return 0;
8750 tgl 1278 ECB : }
1279 :
1280 :
6769 1281 : static int
6769 tgl 1282 CBC 20368017 : internal_putbytes(const char *s, size_t len)
8750 tgl 1283 ECB : {
1284 : size_t amount;
1285 :
8750 tgl 1286 GIC 41120142 : while (len > 0)
1287 : {
1288 : /* If buffer is full, then flush it out */
4393 heikki.linnakangas 1289 20752130 : if (PqSendPointer >= PqSendBufferSize)
1290 : {
3082 rhaas 1291 408483 : socket_set_nonblocking(false);
6769 tgl 1292 408483 : if (internal_flush())
8750 1293 5 : return EOF;
1294 : }
4393 heikki.linnakangas 1295 CBC 20752125 : amount = PqSendBufferSize - PqSendPointer;
8750 tgl 1296 GIC 20752125 : if (amount > len)
1297 20334961 : amount = len;
1298 20752125 : memcpy(PqSendBuffer + PqSendPointer, s, amount);
1299 20752125 : PqSendPointer += amount;
8750 tgl 1300 CBC 20752125 : s += amount;
8750 tgl 1301 GBC 20752125 : len -= amount;
8750 tgl 1302 ECB : }
8750 tgl 1303 CBC 20368012 : return 0;
8750 tgl 1304 ECB : }
1305 :
1306 : /* --------------------------------
1307 : * socket_flush - flush pending output
1308 : *
1309 : * returns 0 if OK, EOF if trouble
1310 : * --------------------------------
1311 : */
1312 : static int
3082 rhaas 1313 GIC 279645 : socket_flush(void)
1314 : {
1315 : int res;
1316 :
6769 tgl 1317 ECB : /* No-op if reentrant call */
6769 tgl 1318 GIC 279645 : if (PqCommBusy)
6769 tgl 1319 UIC 0 : return 0;
6769 tgl 1320 GIC 279645 : PqCommBusy = true;
3082 rhaas 1321 CBC 279645 : socket_set_nonblocking(false);
6769 tgl 1322 279645 : res = internal_flush();
6769 tgl 1323 GIC 279645 : PqCommBusy = false;
6769 tgl 1324 CBC 279645 : return res;
1325 : }
1326 :
1327 : /* --------------------------------
4393 heikki.linnakangas 1328 ECB : * internal_flush - flush pending output
1329 : *
1330 : * Returns 0 if OK (meaning everything was sent, or operation would block
1331 : * and the socket is in non-blocking mode), or EOF if trouble.
1332 : * --------------------------------
4393 heikki.linnakangas 1333 EUB : */
1334 : static int
6769 tgl 1335 GIC 906009 : internal_flush(void)
1336 : {
1337 : static int last_reported_send_errno = 0;
1338 :
4393 heikki.linnakangas 1339 CBC 906009 : char *bufptr = PqSendBuffer + PqSendStart;
6406 tgl 1340 906009 : char *bufend = PqSendBuffer + PqSendPointer;
1341 :
8750 1342 1810814 : while (bufptr < bufend)
1343 : {
1344 : int r;
1345 :
7604 bruce 1346 GIC 909157 : r = secure_write(MyProcPort, bufptr, bufend - bufptr);
1347 :
8750 tgl 1348 909157 : if (r <= 0)
1349 : {
1350 4352 : if (errno == EINTR)
8750 tgl 1351 UIC 0 : continue; /* Ok if we were interrupted */
1352 :
1353 : /*
4382 bruce 1354 ECB : * Ok if no data writable without blocking, and the socket is in
1355 : * non-blocking mode.
4393 heikki.linnakangas 1356 : */
4393 heikki.linnakangas 1357 CBC 4352 : if (errno == EAGAIN ||
4393 heikki.linnakangas 1358 GIC 36 : errno == EWOULDBLOCK)
1359 : {
1360 4316 : return 0;
1361 : }
1362 :
1363 : /*
1364 : * Careful: an ereport() that tries to write to the client would
1365 : * cause recursion to here, leading to stack overflow and core
1366 : * dump! This message must go *only* to the postmaster log.
1367 : *
7522 bruce 1368 ECB : * If a client disconnects while we're in the midst of output, we
6385 1369 : * might write quite a bit of data before we get to a safe query
1370 : * abort point. So, suppress duplicate log messages.
8750 tgl 1371 : */
7818 tgl 1372 GIC 36 : if (errno != last_reported_send_errno)
1373 : {
7818 tgl 1374 CBC 36 : last_reported_send_errno = errno;
7201 1375 36 : ereport(COMMERROR,
7201 tgl 1376 ECB : (errcode_for_socket_access(),
1377 : errmsg("could not send data to client: %m")));
1378 : }
8720 bruce 1379 :
1380 : /*
1381 : * We drop the buffered data anyway so that processing can
1382 : * continue, even though we'll probably quit soon. We also set a
1383 : * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1384 : * the connection.
1385 : */
4393 heikki.linnakangas 1386 GIC 36 : PqSendStart = PqSendPointer = 0;
4139 1387 36 : ClientConnectionLost = 1;
1388 36 : InterruptPending = 1;
8750 tgl 1389 36 : return EOF;
8750 tgl 1390 ECB : }
1391 :
7522 bruce 1392 GIC 904805 : last_reported_send_errno = 0; /* reset after any successful send */
8750 tgl 1393 904805 : bufptr += r;
4393 heikki.linnakangas 1394 904805 : PqSendStart += r;
8750 tgl 1395 ECB : }
7818 1396 :
4393 heikki.linnakangas 1397 GIC 901657 : PqSendStart = PqSendPointer = 0;
8750 tgl 1398 901657 : return 0;
8750 tgl 1399 ECB : }
8750 tgl 1400 EUB :
1401 : /* --------------------------------
1402 : * pq_flush_if_writable - flush pending output if writable without blocking
4393 heikki.linnakangas 1403 ECB : *
1404 : * Returns 0 if OK, or EOF if trouble.
1405 : * --------------------------------
1406 : */
3082 rhaas 1407 : static int
3082 rhaas 1408 CBC 1136342 : socket_flush_if_writable(void)
1409 : {
1410 : int res;
1411 :
1412 : /* Quick exit if nothing to do */
4393 heikki.linnakangas 1413 GIC 1136342 : if (PqSendPointer == PqSendStart)
1414 918461 : return 0;
1415 :
4393 heikki.linnakangas 1416 ECB : /* No-op if reentrant call */
4393 heikki.linnakangas 1417 GIC 217881 : if (PqCommBusy)
4393 heikki.linnakangas 1418 LBC 0 : return 0;
1419 :
1420 : /* Temporarily put the socket into non-blocking mode */
3082 rhaas 1421 GIC 217881 : socket_set_nonblocking(true);
1422 :
4393 heikki.linnakangas 1423 217881 : PqCommBusy = true;
1424 217881 : res = internal_flush();
1425 217881 : PqCommBusy = false;
1426 217881 : return res;
1427 : }
1428 :
1429 : /* --------------------------------
1430 : * socket_is_send_pending - is there any pending data in the output buffer?
1431 : * --------------------------------
1432 : */
1433 : static bool
3082 rhaas 1434 2198296 : socket_is_send_pending(void)
1435 : {
4393 heikki.linnakangas 1436 2198296 : return (PqSendStart < PqSendPointer);
1437 : }
1438 :
1439 : /* --------------------------------
1440 : * Message-level I/O routines begin here.
1441 : * --------------------------------
1442 : */
1443 :
1444 :
1445 : /* --------------------------------
3082 rhaas 1446 ECB : * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1447 : *
1448 : * msgtype is a message type code to place before the message body.
1449 : *
766 heikki.linnakangas 1450 : * len is the length of the message body data at *s. A message length
1451 : * word (equal to len+4 because it counts itself too) is inserted by this
1452 : * routine.
8750 tgl 1453 EUB : *
766 heikki.linnakangas 1454 ECB : * We suppress messages generated while pqcomm.c is busy. This
6769 tgl 1455 : * avoids any possibility of messages being inserted within other
6769 tgl 1456 EUB : * messages. The only known trouble case arises if SIGQUIT occurs
1457 : * during a pqcomm.c routine --- quickdie() will try to send a warning
6769 tgl 1458 ECB : * message, and the most reasonable approach seems to be to drop it.
1459 : *
8750 tgl 1460 EUB : * returns 0 if OK, EOF if trouble
1461 : * --------------------------------
8750 tgl 1462 ECB : */
3082 rhaas 1463 : static int
3082 rhaas 1464 CBC 6789339 : socket_putmessage(char msgtype, const char *s, size_t len)
8750 tgl 1465 ECB : {
1466 : uint32 n32;
766 heikki.linnakangas 1467 :
766 heikki.linnakangas 1468 CBC 6789339 : Assert(msgtype != 0);
766 heikki.linnakangas 1469 ECB :
766 heikki.linnakangas 1470 GIC 6789339 : if (PqCommBusy)
8750 tgl 1471 UIC 0 : return 0;
6769 tgl 1472 GIC 6789339 : PqCommBusy = true;
766 heikki.linnakangas 1473 6789339 : if (internal_putbytes(&msgtype, 1))
766 heikki.linnakangas 1474 UIC 0 : goto fail;
1475 :
766 heikki.linnakangas 1476 GIC 6789339 : n32 = pg_hton32((uint32) (len + 4));
1477 6789339 : if (internal_putbytes((char *) &n32, 4))
766 heikki.linnakangas 1478 UIC 0 : goto fail;
7292 tgl 1479 ECB :
6769 tgl 1480 GIC 6789339 : if (internal_putbytes(s, len))
1481 5 : goto fail;
1482 6789334 : PqCommBusy = false;
1483 6789334 : return 0;
1484 :
1485 5 : fail:
1486 5 : PqCommBusy = false;
1487 5 : return EOF;
8750 tgl 1488 ECB : }
1489 :
1490 : /* --------------------------------
4393 heikki.linnakangas 1491 : * pq_putmessage_noblock - like pq_putmessage, but never blocks
1492 : *
1493 : * If the output buffer is too small to hold the message, the buffer
1494 : * is enlarged.
1495 : */
1496 : static void
3082 rhaas 1497 CBC 213466 : socket_putmessage_noblock(char msgtype, const char *s, size_t len)
1498 : {
1499 : int res PG_USED_FOR_ASSERTS_ONLY;
1500 : int required;
1501 :
1502 : /*
1503 : * Ensure we have enough space in the output buffer for the message header
1504 : * as well as the message itself.
1505 : */
4393 heikki.linnakangas 1506 GIC 213466 : required = PqSendPointer + 1 + 4 + len;
1507 213466 : if (required > PqSendBufferSize)
1508 : {
1509 331 : PqSendBuffer = repalloc(PqSendBuffer, required);
1510 331 : PqSendBufferSize = required;
1511 : }
1512 213466 : res = pq_putmessage(msgtype, s, len);
4382 bruce 1513 213466 : Assert(res == 0); /* should not fail when the message fits in
1514 : * buffer */
4393 heikki.linnakangas 1515 213466 : }
4393 heikki.linnakangas 1516 EUB :
1517 : /* --------------------------------
766 1518 : * pq_putmessage_v2 - send a message in protocol version 2
1519 : *
1520 : * msgtype is a message type code to place before the message body.
8750 tgl 1521 : *
766 heikki.linnakangas 1522 : * We no longer support protocol version 2, but we have kept this
1523 : * function so that if a client tries to connect with protocol version 2,
1524 : * as a courtesy we can still send the "unsupported protocol version"
1525 : * error to the client in the old format.
1526 : *
1527 : * Like in pq_putmessage(), we suppress messages generated while
1528 : * pqcomm.c is busy.
1529 : *
1530 : * returns 0 if OK, EOF if trouble
8750 tgl 1531 : * --------------------------------
1532 : */
766 heikki.linnakangas 1533 : int
766 heikki.linnakangas 1534 UIC 0 : pq_putmessage_v2(char msgtype, const char *s, size_t len)
1535 : {
1536 0 : Assert(msgtype != 0);
1537 :
1538 0 : if (PqCommBusy)
1539 0 : return 0;
1540 0 : PqCommBusy = true;
1541 0 : if (internal_putbytes(&msgtype, 1))
1542 0 : goto fail;
1543 :
1544 0 : if (internal_putbytes(s, len))
1545 0 : goto fail;
1546 0 : PqCommBusy = false;
1547 0 : return 0;
1548 :
1549 0 : fail:
1550 0 : PqCommBusy = false;
1551 0 : return EOF;
1552 : }
1553 :
1554 : /*
1555 : * Support for TCP Keepalive parameters
1556 : */
1557 :
1558 : /*
1559 : * On Windows, we need to set both idle and interval at the same time.
1560 : * We also cannot reset them to the default (setting to zero will
1561 : * actually set them to zero, not default), therefore we fallback to
1562 : * the out-of-the-box default instead.
1563 : */
1564 : #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1565 : static int
1566 : pq_setkeepaliveswin32(Port *port, int idle, int interval)
1567 : {
1568 : struct tcp_keepalive ka;
1569 : DWORD retsize;
1570 :
1571 : if (idle <= 0)
1572 : idle = 2 * 60 * 60; /* default = 2 hours */
1573 : if (interval <= 0)
1574 : interval = 1; /* default = 1 second */
1575 :
1576 : ka.onoff = 1;
1577 : ka.keepalivetime = idle * 1000;
1578 : ka.keepaliveinterval = interval * 1000;
1579 :
1580 : if (WSAIoctl(port->sock,
1581 : SIO_KEEPALIVE_VALS,
1582 : (LPVOID) &ka,
1583 : sizeof(ka),
1584 : NULL,
1585 : 0,
1586 : &retsize,
4658 magnus 1587 ECB : NULL,
1588 : NULL)
1589 : != 0)
1590 : {
856 peter 1591 : ereport(LOG,
1592 : (errmsg("%s(%s) failed: error code %d",
716 peter 1593 EUB : "WSAIoctl", "SIO_KEEPALIVE_VALS", WSAGetLastError())));
4658 magnus 1594 : return STATUS_ERROR;
1595 : }
1596 : if (port->keepalives_idle != idle)
1597 : port->keepalives_idle = idle;
1598 : if (port->keepalives_interval != interval)
1599 : port->keepalives_interval = interval;
1600 : return STATUS_OK;
1601 : }
1602 : #endif
1603 :
1604 : int
6462 bruce 1605 GBC 1088 : pq_getkeepalivesidle(Port *port)
1606 : {
2111 tgl 1607 EUB : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
418 peter 1608 GIC 1088 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1609 1088 : return 0;
1610 :
6462 bruce 1611 UIC 0 : if (port->keepalives_idle != 0)
1612 0 : return port->keepalives_idle;
1613 :
1614 0 : if (port->default_keepalives_idle == 0)
6462 bruce 1615 EUB : {
1616 : #ifndef WIN32
516 peter 1617 UIC 0 : socklen_t size = sizeof(port->default_keepalives_idle);
1618 :
2111 tgl 1619 0 : if (getsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
2112 1620 0 : (char *) &port->default_keepalives_idle,
1621 : &size) < 0)
2112 tgl 1622 ECB : {
856 peter 1623 UIC 0 : ereport(LOG,
716 peter 1624 ECB : (errmsg("%s(%s) failed: %m", "getsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
2112 tgl 1625 LBC 0 : port->default_keepalives_idle = -1; /* don't know */
1626 : }
1627 : #else /* WIN32 */
1628 : /* We can't get the defaults on Windows, so return "don't know" */
4658 magnus 1629 ECB : port->default_keepalives_idle = -1;
2118 tgl 1630 : #endif /* WIN32 */
1631 : }
1632 :
6462 bruce 1633 UBC 0 : return port->default_keepalives_idle;
1634 : #else
6462 bruce 1635 EUB : return 0;
1636 : #endif
1637 : }
6418 tgl 1638 :
1639 : int
6462 bruce 1640 GBC 2057 : pq_setkeepalivesidle(int idle, Port *port)
1641 : {
418 peter 1642 GIC 2057 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1643 1857 : return STATUS_OK;
6462 bruce 1644 EUB :
2112 tgl 1645 : /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1646 : #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
6462 bruce 1647 GBC 200 : if (idle == port->keepalives_idle)
6462 bruce 1648 GIC 200 : return STATUS_OK;
1649 :
4658 magnus 1650 EUB : #ifndef WIN32
6418 tgl 1651 UIC 0 : if (port->default_keepalives_idle <= 0)
6462 bruce 1652 EUB : {
6462 bruce 1653 UIC 0 : if (pq_getkeepalivesidle(port) < 0)
1654 : {
6418 tgl 1655 UBC 0 : if (idle == 0)
2118 tgl 1656 UIC 0 : return STATUS_OK; /* default is set but unknown */
1657 : else
6418 1658 0 : return STATUS_ERROR;
1659 : }
1660 : }
1661 :
6462 bruce 1662 0 : if (idle == 0)
1663 0 : idle = port->default_keepalives_idle;
1664 :
2111 tgl 1665 0 : if (setsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1666 : (char *) &idle, sizeof(idle)) < 0)
1667 : {
856 peter 1668 UBC 0 : ereport(LOG,
1669 : (errmsg("%s(%s) failed: %m", "setsockopt", PG_TCP_KEEPALIVE_IDLE_STR)));
2112 tgl 1670 UIC 0 : return STATUS_ERROR;
1671 : }
6462 bruce 1672 ECB :
6462 bruce 1673 UIC 0 : port->keepalives_idle = idle;
1674 : #else /* WIN32 */
4658 magnus 1675 ECB : return pq_setkeepaliveswin32(port, idle, port->keepalives_interval);
1676 : #endif
1677 : #else
6462 bruce 1678 EUB : if (idle != 0)
1679 : {
1680 : ereport(LOG,
856 peter 1681 : (errmsg("setting the keepalive idle time is not supported")));
1682 : return STATUS_ERROR;
1683 : }
6462 bruce 1684 : #endif
1685 :
6462 bruce 1686 UBC 0 : return STATUS_OK;
6462 bruce 1687 EUB : }
1688 :
1689 : int
6462 bruce 1690 GBC 1088 : pq_getkeepalivesinterval(Port *port)
1691 : {
4658 magnus 1692 EUB : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
418 peter 1693 GIC 1088 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1694 1088 : return 0;
1695 :
6462 bruce 1696 UIC 0 : if (port->keepalives_interval != 0)
1697 0 : return port->keepalives_interval;
1698 :
1699 0 : if (port->default_keepalives_interval == 0)
6462 bruce 1700 EUB : {
1701 : #ifndef WIN32
516 peter 1702 UIC 0 : socklen_t size = sizeof(port->default_keepalives_interval);
1703 :
6462 tgl 1704 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
6418 1705 0 : (char *) &port->default_keepalives_interval,
1706 : &size) < 0)
6462 bruce 1707 ECB : {
856 peter 1708 UIC 0 : ereport(LOG,
716 peter 1709 ECB : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPINTVL")));
2118 tgl 1710 LBC 0 : port->default_keepalives_interval = -1; /* don't know */
1711 : }
1712 : #else
4658 magnus 1713 ECB : /* We can't get the defaults on Windows, so return "don't know" */
1714 : port->default_keepalives_interval = -1;
1715 : #endif /* WIN32 */
1716 : }
6462 bruce 1717 EUB :
6462 bruce 1718 UIC 0 : return port->default_keepalives_interval;
6462 bruce 1719 EUB : #else
1720 : return 0;
1721 : #endif
1722 : }
1723 :
1724 : int
6462 bruce 1725 GIC 2057 : pq_setkeepalivesinterval(int interval, Port *port)
1726 : {
418 peter 1727 2057 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1728 GBC 1857 : return STATUS_OK;
6462 bruce 1729 EUB :
1730 : #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
6462 bruce 1731 GBC 200 : if (interval == port->keepalives_interval)
6462 bruce 1732 GIC 200 : return STATUS_OK;
1733 :
4658 magnus 1734 EUB : #ifndef WIN32
6418 tgl 1735 UIC 0 : if (port->default_keepalives_interval <= 0)
6418 tgl 1736 EUB : {
6462 bruce 1737 UIC 0 : if (pq_getkeepalivesinterval(port) < 0)
1738 : {
6418 tgl 1739 UBC 0 : if (interval == 0)
2118 tgl 1740 UIC 0 : return STATUS_OK; /* default is set but unknown */
1741 : else
6418 1742 0 : return STATUS_ERROR;
1743 : }
1744 : }
1745 :
6462 bruce 1746 0 : if (interval == 0)
1747 0 : interval = port->default_keepalives_interval;
1748 :
tgl 1749 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1750 : (char *) &interval, sizeof(interval)) < 0)
1751 : {
856 peter 1752 UBC 0 : ereport(LOG,
1753 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPINTVL")));
6462 bruce 1754 UIC 0 : return STATUS_ERROR;
1755 : }
6462 bruce 1756 ECB :
6462 bruce 1757 UIC 0 : port->keepalives_interval = interval;
1758 : #else /* WIN32 */
4658 magnus 1759 ECB : return pq_setkeepaliveswin32(port, port->keepalives_idle, interval);
1760 : #endif
1761 : #else
6462 bruce 1762 EUB : if (interval != 0)
1763 : {
1764 : ereport(LOG,
716 peter 1765 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPINTVL")));
1766 : return STATUS_ERROR;
6418 tgl 1767 : }
1768 : #endif
6462 bruce 1769 :
6462 bruce 1770 UBC 0 : return STATUS_OK;
1771 : }
1772 :
6462 bruce 1773 EUB : int
6462 bruce 1774 GIC 1088 : pq_getkeepalivescount(Port *port)
6462 bruce 1775 EUB : {
1776 : #ifdef TCP_KEEPCNT
418 peter 1777 GIC 1088 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1778 1088 : return 0;
6462 bruce 1779 EUB :
6462 bruce 1780 UIC 0 : if (port->keepalives_count != 0)
1781 0 : return port->keepalives_count;
1782 :
1783 0 : if (port->default_keepalives_count == 0)
1784 : {
516 peter 1785 0 : socklen_t size = sizeof(port->default_keepalives_count);
6418 tgl 1786 ECB :
6462 tgl 1787 UIC 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
6418 tgl 1788 LBC 0 : (char *) &port->default_keepalives_count,
6462 bruce 1789 ECB : &size) < 0)
1790 : {
856 peter 1791 UIC 0 : ereport(LOG,
716 peter 1792 ECB : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_KEEPCNT")));
2118 tgl 1793 LBC 0 : port->default_keepalives_count = -1; /* don't know */
1794 : }
6462 bruce 1795 EUB : }
1796 :
6462 bruce 1797 UBC 0 : return port->default_keepalives_count;
1798 : #else
6462 bruce 1799 EUB : return 0;
1800 : #endif
1801 : }
6418 tgl 1802 :
1803 : int
6462 bruce 1804 GIC 2057 : pq_setkeepalivescount(int count, Port *port)
1805 : {
418 peter 1806 GBC 2057 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
6462 bruce 1807 1857 : return STATUS_OK;
1808 :
4775 bruce 1809 EUB : #ifdef TCP_KEEPCNT
6462 bruce 1810 GIC 200 : if (count == port->keepalives_count)
1811 200 : return STATUS_OK;
6462 bruce 1812 EUB :
6418 tgl 1813 UIC 0 : if (port->default_keepalives_count <= 0)
6418 tgl 1814 EUB : {
6462 bruce 1815 UIC 0 : if (pq_getkeepalivescount(port) < 0)
1816 : {
6418 tgl 1817 UBC 0 : if (count == 0)
2118 tgl 1818 UIC 0 : return STATUS_OK; /* default is set but unknown */
1819 : else
6418 1820 0 : return STATUS_ERROR;
1821 : }
1822 : }
1823 :
6462 bruce 1824 0 : if (count == 0)
1825 0 : count = port->default_keepalives_count;
1826 :
6462 tgl 1827 UBC 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1828 : (char *) &count, sizeof(count)) < 0)
1829 : {
856 peter 1830 UIC 0 : ereport(LOG,
716 peter 1831 ECB : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_KEEPCNT")));
6462 bruce 1832 UIC 0 : return STATUS_ERROR;
1833 : }
6462 bruce 1834 ECB :
6462 bruce 1835 LBC 0 : port->keepalives_count = count;
1836 : #else
6462 bruce 1837 EUB : if (count != 0)
1838 : {
1839 : ereport(LOG,
716 peter 1840 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_KEEPCNT")));
1841 : return STATUS_ERROR;
6462 bruce 1842 : }
1843 : #endif
1844 :
6462 bruce 1845 UBC 0 : return STATUS_OK;
1846 : }
1847 :
1464 michael 1848 EUB : int
1464 michael 1849 GIC 1088 : pq_gettcpusertimeout(Port *port)
1464 michael 1850 EUB : {
1851 : #ifdef TCP_USER_TIMEOUT
418 peter 1852 GIC 1088 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1464 michael 1853 1088 : return 0;
1464 michael 1854 EUB :
1464 michael 1855 UIC 0 : if (port->tcp_user_timeout != 0)
1856 0 : return port->tcp_user_timeout;
1857 :
1858 0 : if (port->default_tcp_user_timeout == 0)
1859 : {
516 peter 1860 0 : socklen_t size = sizeof(port->default_tcp_user_timeout);
1464 michael 1861 ECB :
1464 michael 1862 UIC 0 : if (getsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1464 michael 1863 LBC 0 : (char *) &port->default_tcp_user_timeout,
1464 michael 1864 ECB : &size) < 0)
1865 : {
856 peter 1866 UIC 0 : ereport(LOG,
716 peter 1867 ECB : (errmsg("%s(%s) failed: %m", "getsockopt", "TCP_USER_TIMEOUT")));
1464 michael 1868 LBC 0 : port->default_tcp_user_timeout = -1; /* don't know */
1869 : }
1464 michael 1870 EUB : }
1871 :
1464 michael 1872 UBC 0 : return port->default_tcp_user_timeout;
1873 : #else
1464 michael 1874 EUB : return 0;
1875 : #endif
1876 : }
1877 :
1878 : int
1464 michael 1879 GIC 2057 : pq_settcpusertimeout(int timeout, Port *port)
1880 : {
418 peter 1881 GBC 2057 : if (port == NULL || port->laddr.addr.ss_family == AF_UNIX)
1464 michael 1882 1857 : return STATUS_OK;
1883 :
1464 michael 1884 EUB : #ifdef TCP_USER_TIMEOUT
1464 michael 1885 GIC 200 : if (timeout == port->tcp_user_timeout)
1886 200 : return STATUS_OK;
1464 michael 1887 EUB :
1464 michael 1888 UIC 0 : if (port->default_tcp_user_timeout <= 0)
1464 michael 1889 EUB : {
1464 michael 1890 UIC 0 : if (pq_gettcpusertimeout(port) < 0)
1891 : {
1464 michael 1892 UBC 0 : if (timeout == 0)
1464 michael 1893 UIC 0 : return STATUS_OK; /* default is set but unknown */
1894 : else
1895 0 : return STATUS_ERROR;
1896 : }
1897 : }
1898 :
1899 0 : if (timeout == 0)
1900 0 : timeout = port->default_tcp_user_timeout;
1901 :
1464 michael 1902 UBC 0 : if (setsockopt(port->sock, IPPROTO_TCP, TCP_USER_TIMEOUT,
1903 : (char *) &timeout, sizeof(timeout)) < 0)
1904 : {
856 peter 1905 UIC 0 : ereport(LOG,
1906 : (errmsg("%s(%s) failed: %m", "setsockopt", "TCP_USER_TIMEOUT")));
1464 michael 1907 0 : return STATUS_ERROR;
1908 : }
1464 michael 1909 ECB :
1464 michael 1910 UIC 0 : port->tcp_user_timeout = timeout;
1911 : #else
1912 : if (timeout != 0)
1913 : {
1914 : ereport(LOG,
1915 : (errmsg("%s(%s) not supported", "setsockopt", "TCP_USER_TIMEOUT")));
1916 : return STATUS_ERROR;
1917 : }
1918 : #endif
1919 :
1920 0 : return STATUS_OK;
1921 : }
736 tmunro 1922 ECB :
1923 : /*
1924 : * GUC assign_hook for tcp_keepalives_idle
1925 : */
1926 : void
208 tgl 1927 GNC 1857 : assign_tcp_keepalives_idle(int newval, void *extra)
1928 : {
1929 : /*
1930 : * The kernel API provides no way to test a value without setting it; and
1931 : * once we set it we might fail to unset it. So there seems little point
1932 : * in fully implementing the check-then-assign GUC API for these
1933 : * variables. Instead we just do the assignment on demand.
1934 : * pq_setkeepalivesidle reports any problems via ereport(LOG).
1935 : *
1936 : * This approach means that the GUC value might have little to do with the
1937 : * actual kernel value, so we use a show_hook that retrieves the kernel
1938 : * value rather than trusting GUC's copy.
1939 : */
1940 1857 : (void) pq_setkeepalivesidle(newval, MyProcPort);
1941 1857 : }
1942 :
1943 : /*
1944 : * GUC show_hook for tcp_keepalives_idle
1945 : */
1946 : const char *
1947 1088 : show_tcp_keepalives_idle(void)
1948 : {
1949 : /* See comments in assign_tcp_keepalives_idle */
1950 : static char nbuf[16];
1951 :
1952 1088 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesidle(MyProcPort));
1953 1088 : return nbuf;
1954 : }
1955 :
1956 : /*
1957 : * GUC assign_hook for tcp_keepalives_interval
1958 : */
1959 : void
1960 1857 : assign_tcp_keepalives_interval(int newval, void *extra)
1961 : {
1962 : /* See comments in assign_tcp_keepalives_idle */
1963 1857 : (void) pq_setkeepalivesinterval(newval, MyProcPort);
1964 1857 : }
1965 :
1966 : /*
1967 : * GUC show_hook for tcp_keepalives_interval
1968 : */
1969 : const char *
1970 1088 : show_tcp_keepalives_interval(void)
1971 : {
1972 : /* See comments in assign_tcp_keepalives_idle */
1973 : static char nbuf[16];
1974 :
1975 1088 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivesinterval(MyProcPort));
1976 1088 : return nbuf;
1977 : }
1978 :
1979 : /*
1980 : * GUC assign_hook for tcp_keepalives_count
1981 : */
1982 : void
1983 1857 : assign_tcp_keepalives_count(int newval, void *extra)
1984 : {
1985 : /* See comments in assign_tcp_keepalives_idle */
1986 1857 : (void) pq_setkeepalivescount(newval, MyProcPort);
1987 1857 : }
1988 :
1989 : /*
1990 : * GUC show_hook for tcp_keepalives_count
1991 : */
1992 : const char *
1993 1088 : show_tcp_keepalives_count(void)
1994 : {
1995 : /* See comments in assign_tcp_keepalives_idle */
1996 : static char nbuf[16];
1997 :
1998 1088 : snprintf(nbuf, sizeof(nbuf), "%d", pq_getkeepalivescount(MyProcPort));
1999 1088 : return nbuf;
2000 : }
2001 :
2002 : /*
2003 : * GUC assign_hook for tcp_user_timeout
2004 : */
2005 : void
2006 1857 : assign_tcp_user_timeout(int newval, void *extra)
2007 : {
2008 : /* See comments in assign_tcp_keepalives_idle */
2009 1857 : (void) pq_settcpusertimeout(newval, MyProcPort);
2010 1857 : }
2011 :
2012 : /*
2013 : * GUC show_hook for tcp_user_timeout
2014 : */
2015 : const char *
2016 1088 : show_tcp_user_timeout(void)
2017 : {
2018 : /* See comments in assign_tcp_keepalives_idle */
2019 : static char nbuf[16];
2020 :
2021 1088 : snprintf(nbuf, sizeof(nbuf), "%d", pq_gettcpusertimeout(MyProcPort));
2022 1088 : return nbuf;
2023 : }
2024 :
736 tmunro 2025 ECB : /*
2026 : * Check if the client is still connected.
2027 : */
2028 : bool
736 tmunro 2029 UIC 0 : pq_check_connection(void)
2030 : {
419 tmunro 2031 ECB : WaitEvent events[FeBeWaitSetNEvents];
2032 : int rc;
2033 :
2034 : /*
2035 : * It's OK to modify the socket event filter without restoring, because
2036 : * all FeBeWaitSet socket wait sites do the same.
2037 : */
419 tmunro 2038 UIC 0 : ModifyWaitEvent(FeBeWaitSet, FeBeWaitSetSocketPos, WL_SOCKET_CLOSED, NULL);
2039 :
2040 0 : retry:
2041 0 : rc = WaitEventSetWait(FeBeWaitSet, 0, events, lengthof(events), 0);
2042 0 : for (int i = 0; i < rc; ++i)
2043 : {
419 tmunro 2044 LBC 0 : if (events[i].events & WL_SOCKET_CLOSED)
419 tmunro 2045 UIC 0 : return false;
2046 0 : if (events[i].events & WL_LATCH_SET)
419 tmunro 2047 ECB : {
2048 : /*
2049 : * A latch event might be preventing other events from being
2050 : * reported. Reset it and poll again. No need to restore it
2051 : * because no code should expect latches to survive across
2052 : * CHECK_FOR_INTERRUPTS().
2053 : */
332 tgl 2054 LBC 0 : ResetLatch(MyLatch);
332 tgl 2055 UIC 0 : goto retry;
2056 : }
2057 : }
2058 :
736 tmunro 2059 LBC 0 : return true;
736 tmunro 2060 ECB : }
|
