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 --- or would be,
9 * except for major brain damage in the design of the old COPY OUT protocol.
10 * Unfortunately, COPY OUT was designed to commandeer the communication
11 * channel (it just transfers data without wrapping it into messages).
12 * No other messages can be sent while COPY OUT is in progress; and if the
13 * copy is aborted by an ereport(ERROR), we need to close out the copy so that
14 * the frontend gets back into sync. Therefore, these routines have to be
15 * aware of COPY OUT state. (New COPY-OUT is message-based and does *not*
16 * set the DoingCopyOut flag.)
17 *
18 * NOTE: generally, it's a bad idea to emit outgoing messages directly with
19 * pq_putbytes(), especially if the message would require multiple calls
20 * to send. Instead, use the routines in pqformat.c to construct the message
21 * in a buffer and then emit it in one call to pq_putmessage. This ensures
22 * that the channel will not be clogged by an incomplete message if execution
23 * is aborted by ereport(ERROR) partway through the message. The only
24 * non-libpq code that should call pq_putbytes directly is old-style COPY OUT.
25 *
26 * At one time, libpq was shared between frontend and backend, but now
27 * the backend's "backend/libpq" is quite separate from "interfaces/libpq".
28 * All that remains is similarities of names to trap the unwary...
29 *
30 * Portions Copyright (c) 1996-2018, PostgreSQL Global Development Group
31 * Portions Copyright (c) 1994, Regents of the University of California
32 *
33 * src/backend/libpq/pqcomm.c
34 *
35 *-------------------------------------------------------------------------
36 */
37
38 /*------------------------
39 * INTERFACE ROUTINES
40 *
41 * setup/teardown:
42 * StreamServerPort - Open postmaster's server port
43 * StreamConnection - Create new connection with client
44 * StreamClose - Close a client/backend connection
45 * TouchSocketFiles - Protect socket files against /tmp cleaners
46 * pq_init - initialize libpq at backend startup
47 * pq_comm_reset - reset libpq during error recovery
48 * pq_close - shutdown libpq at backend exit
49 *
50 * low-level I/O:
51 * pq_getbytes - get a known number of bytes from connection
52 * pq_getstring - get a null terminated string from connection
53 * pq_getmessage - get a message with length word from connection
54 * pq_getbyte - get next byte from connection
55 * pq_peekbyte - peek at next byte from connection
56 * pq_putbytes - send bytes to connection (not flushed until pq_flush)
57 * pq_flush - flush pending output
58 * pq_flush_if_writable - flush pending output if writable without blocking
59 * pq_getbyte_if_available - get a byte if available without blocking
60 *
61 * message-level I/O (and old-style-COPY-OUT cruft):
62 * pq_putmessage - send a normal message (suppressed in COPY OUT mode)
63 * pq_putmessage_noblock - buffer a normal message (suppressed in COPY OUT)
64 * pq_startcopyout - inform libpq that a COPY OUT transfer is beginning
65 * pq_endcopyout - end a COPY OUT transfer
66 *
67 *------------------------
68 */
69 #include "postgres.h"
70
71 #include <signal.h>
72 #include <fcntl.h>
73 #include <grp.h>
74 #include <unistd.h>
75 #include <sys/file.h>
76 #include <sys/socket.h>
77 #include <sys/stat.h>
78 #include <sys/time.h>
79 #include <netdb.h>
80 #include <netinet/in.h>
81 #ifdef HAVE_NETINET_TCP_H
82 #include <netinet/tcp.h>
83 #endif
84 #ifdef HAVE_UTIME_H
85 #include <utime.h>
86 #endif
87 #ifdef _MSC_VER /* mstcpip.h is missing on mingw */
88 #include <mstcpip.h>
89 #endif
90
91 #include "common/ip.h"
92 #include "libpq/libpq.h"
93 #include "miscadmin.h"
94 #include "port/pg_bswap.h"
95 #include "storage/ipc.h"
96 #include "utils/guc.h"
97 #include "utils/memutils.h"
98
99 /*
100 * Cope with the various platform-specific ways to spell TCP keepalive socket
101 * options. This doesn't cover Windows, which as usual does its own thing.
102 */
103 #if defined(TCP_KEEPIDLE)
104 /* TCP_KEEPIDLE is the name of this option on Linux and *BSD */
105 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPIDLE
106 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPIDLE"
107 #elif defined(TCP_KEEPALIVE_THRESHOLD)
108 /* TCP_KEEPALIVE_THRESHOLD is the name of this option on Solaris >= 11 */
109 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE_THRESHOLD
110 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE_THRESHOLD"
111 #elif defined(TCP_KEEPALIVE) && defined(__darwin__)
112 /* TCP_KEEPALIVE is the name of this option on macOS */
113 /* Caution: Solaris has this symbol but it means something different */
114 #define PG_TCP_KEEPALIVE_IDLE TCP_KEEPALIVE
115 #define PG_TCP_KEEPALIVE_IDLE_STR "TCP_KEEPALIVE"
116 #endif
117
118 /*
119 * Configuration options
120 */
121 int Unix_socket_permissions;
122 char *Unix_socket_group;
123
124 /* Where the Unix socket files are (list of palloc'd strings) */
125 static List *sock_paths = NIL;
126
127 /*
128 * Buffers for low-level I/O.
129 *
130 * The receive buffer is fixed size. Send buffer is usually 8k, but can be
131 * enlarged by pq_putmessage_noblock() if the message doesn't fit otherwise.
132 */
133
134 #define PQ_SEND_BUFFER_SIZE 8192
135 #define PQ_RECV_BUFFER_SIZE 8192
136
137 static char *PqSendBuffer;
138 static int PqSendBufferSize; /* Size send buffer */
139 static int PqSendPointer; /* Next index to store a byte in PqSendBuffer */
140 static int PqSendStart; /* Next index to send a byte in PqSendBuffer */
141
142 static char PqRecvBuffer[PQ_RECV_BUFFER_SIZE];
143 static int PqRecvPointer; /* Next index to read a byte from PqRecvBuffer */
144 static int PqRecvLength; /* End of data available in PqRecvBuffer */
145
146 /*
147 * Message status
148 */
149 static bool PqCommBusy; /* busy sending data to the client */
150 static bool PqCommReadingMsg; /* in the middle of reading a message */
151 static bool DoingCopyOut; /* in old-protocol COPY OUT processing */
152
153
154 /* Internal functions */
155 static void socket_comm_reset(void);
156 static void socket_close(int code, Datum arg);
157 static void socket_set_nonblocking(bool nonblocking);
158 static int socket_flush(void);
159 static int socket_flush_if_writable(void);
160 static bool socket_is_send_pending(void);
161 static int socket_putmessage(char msgtype, const char *s, size_t len);
162 static void socket_putmessage_noblock(char msgtype, const char *s, size_t len);
163 static void socket_startcopyout(void);
164 static void socket_endcopyout(bool errorAbort);
165 static int internal_putbytes(const char *s, size_t len);
166 static int internal_flush(void);
167
168 #ifdef HAVE_UNIX_SOCKETS
169 static int Lock_AF_UNIX(char *unixSocketDir, char *unixSocketPath);
170 static int Setup_AF_UNIX(char *sock_path);
171 #endif /* HAVE_UNIX_SOCKETS */
172
173 static PQcommMethods PqCommSocketMethods = {
174 socket_comm_reset,
175 socket_flush,
176 socket_flush_if_writable,
177 socket_is_send_pending,
178 socket_putmessage,
179 socket_putmessage_noblock,
180 socket_startcopyout,
181 socket_endcopyout
182 };
183
184 PQcommMethods *PqCommMethods = &PqCommSocketMethods;
185
186 WaitEventSet *FeBeWaitSet;
187
188
189 /* --------------------------------
190 * pq_init - initialize libpq at backend startup
191 * --------------------------------
192 */
193 void
pq_init(void)194 pq_init(void)
195 {
196 /* initialize state variables */
197 PqSendBufferSize = PQ_SEND_BUFFER_SIZE;
198 PqSendBuffer = MemoryContextAlloc(TopMemoryContext, PqSendBufferSize);
199 PqSendPointer = PqSendStart = PqRecvPointer = PqRecvLength = 0;
200 PqCommBusy = false;
201 PqCommReadingMsg = false;
202 DoingCopyOut = false;
203
204 /* set up process-exit hook to close the socket */
205 on_proc_exit(socket_close, 0);
206
207 /*
208 * In backends (as soon as forked) we operate the underlying socket in
209 * nonblocking mode and use latches to implement blocking semantics if
210 * needed. That allows us to provide safely interruptible reads and
211 * writes.
212 *
213 * Use COMMERROR on failure, because ERROR would try to send the error to
214 * the client, which might require changing the mode again, leading to
215 * infinite recursion.
216 */
217 #ifndef WIN32
218 if (!pg_set_noblock(MyProcPort->sock))
219 ereport(COMMERROR,
220 (errmsg("could not set socket to nonblocking mode: %m")));
221 #endif
222
223 FeBeWaitSet = CreateWaitEventSet(TopMemoryContext, 3);
224 AddWaitEventToSet(FeBeWaitSet, WL_SOCKET_WRITEABLE, MyProcPort->sock,
225 NULL, NULL);
226 AddWaitEventToSet(FeBeWaitSet, WL_LATCH_SET, -1, MyLatch, NULL);
227 AddWaitEventToSet(FeBeWaitSet, WL_POSTMASTER_DEATH, -1, NULL, NULL);
228 }
229
230 /* --------------------------------
231 * socket_comm_reset - reset libpq during error recovery
232 *
233 * This is called from error recovery at the outer idle loop. It's
234 * just to get us out of trouble if we somehow manage to elog() from
235 * inside a pqcomm.c routine (which ideally will never happen, but...)
236 * --------------------------------
237 */
238 static void
socket_comm_reset(void)239 socket_comm_reset(void)
240 {
241 /* Do not throw away pending data, but do reset the busy flag */
242 PqCommBusy = false;
243 /* We can abort any old-style COPY OUT, too */
244 pq_endcopyout(true);
245 }
246
247 /* --------------------------------
248 * socket_close - shutdown libpq at backend exit
249 *
250 * This is the one pg_on_exit_callback in place during BackendInitialize().
251 * That function's unusual signal handling constrains that this callback be
252 * safe to run at any instant.
253 * --------------------------------
254 */
255 static void
socket_close(int code,Datum arg)256 socket_close(int code, Datum arg)
257 {
258 /* Nothing to do in a standalone backend, where MyProcPort is NULL. */
259 if (MyProcPort != NULL)
260 {
261 #if defined(ENABLE_GSS) || defined(ENABLE_SSPI)
262 #ifdef ENABLE_GSS
263 OM_uint32 min_s;
264
265 /*
266 * Shutdown GSSAPI layer. This section does nothing when interrupting
267 * BackendInitialize(), because pg_GSS_recvauth() makes first use of
268 * "ctx" and "cred".
269 */
270 if (MyProcPort->gss->ctx != GSS_C_NO_CONTEXT)
271 gss_delete_sec_context(&min_s, &MyProcPort->gss->ctx, NULL);
272
273 if (MyProcPort->gss->cred != GSS_C_NO_CREDENTIAL)
274 gss_release_cred(&min_s, &MyProcPort->gss->cred);
275 #endif /* ENABLE_GSS */
276
277 /*
278 * GSS and SSPI share the port->gss struct. Since nowhere else does a
279 * postmaster child free this, doing so is safe when interrupting
280 * BackendInitialize().
281 */
282 free(MyProcPort->gss);
283 #endif /* ENABLE_GSS || ENABLE_SSPI */
284
285 /*
286 * Cleanly shut down SSL layer. Nowhere else does a postmaster child
287 * call this, so this is safe when interrupting BackendInitialize().
288 */
289 secure_close(MyProcPort);
290
291 /*
292 * Formerly we did an explicit close() here, but it seems better to
293 * leave the socket open until the process dies. This allows clients
294 * to perform a "synchronous close" if they care --- wait till the
295 * transport layer reports connection closure, and you can be sure the
296 * backend has exited.
297 *
298 * We do set sock to PGINVALID_SOCKET to prevent any further I/O,
299 * though.
300 */
301 MyProcPort->sock = PGINVALID_SOCKET;
302 }
303 }
304
305
306
307 /*
308 * Streams -- wrapper around Unix socket system calls
309 *
310 *
311 * Stream functions are used for vanilla TCP connection protocol.
312 */
313
314
315 /*
316 * StreamServerPort -- open a "listening" port to accept connections.
317 *
318 * family should be AF_UNIX or AF_UNSPEC; portNumber is the port number.
319 * For AF_UNIX ports, hostName should be NULL and unixSocketDir must be
320 * specified. For TCP ports, hostName is either NULL for all interfaces or
321 * the interface to listen on, and unixSocketDir is ignored (can be NULL).
322 *
323 * Successfully opened sockets are added to the ListenSocket[] array (of
324 * length MaxListen), at the first position that isn't PGINVALID_SOCKET.
325 *
326 * RETURNS: STATUS_OK or STATUS_ERROR
327 */
328
329 int
StreamServerPort(int family,char * hostName,unsigned short portNumber,char * unixSocketDir,pgsocket ListenSocket[],int MaxListen)330 StreamServerPort(int family, char *hostName, unsigned short portNumber,
331 char *unixSocketDir,
332 pgsocket ListenSocket[], int MaxListen)
333 {
334 pgsocket fd;
335 int err;
336 int maxconn;
337 int ret;
338 char portNumberStr[32];
339 const char *familyDesc;
340 char familyDescBuf[64];
341 const char *addrDesc;
342 char addrBuf[NI_MAXHOST];
343 char *service;
344 struct addrinfo *addrs = NULL,
345 *addr;
346 struct addrinfo hint;
347 int listen_index = 0;
348 int added = 0;
349
350 #ifdef HAVE_UNIX_SOCKETS
351 char unixSocketPath[MAXPGPATH];
352 #endif
353 #if !defined(WIN32) || defined(IPV6_V6ONLY)
354 int one = 1;
355 #endif
356
357 /* Initialize hint structure */
358 MemSet(&hint, 0, sizeof(hint));
359 hint.ai_family = family;
360 hint.ai_flags = AI_PASSIVE;
361 hint.ai_socktype = SOCK_STREAM;
362
363 #ifdef HAVE_UNIX_SOCKETS
364 if (family == AF_UNIX)
365 {
366 /*
367 * Create unixSocketPath from portNumber and unixSocketDir and lock
368 * that file path
369 */
370 UNIXSOCK_PATH(unixSocketPath, portNumber, unixSocketDir);
371 if (strlen(unixSocketPath) >= UNIXSOCK_PATH_BUFLEN)
372 {
373 ereport(LOG,
374 (errmsg("Unix-domain socket path \"%s\" is too long (maximum %d bytes)",
375 unixSocketPath,
376 (int) (UNIXSOCK_PATH_BUFLEN - 1))));
377 return STATUS_ERROR;
378 }
379 if (Lock_AF_UNIX(unixSocketDir, unixSocketPath) != STATUS_OK)
380 return STATUS_ERROR;
381 service = unixSocketPath;
382 }
383 else
384 #endif /* HAVE_UNIX_SOCKETS */
385 {
386 snprintf(portNumberStr, sizeof(portNumberStr), "%d", portNumber);
387 service = portNumberStr;
388 }
389
390 ret = pg_getaddrinfo_all(hostName, service, &hint, &addrs);
391 if (ret || !addrs)
392 {
393 if (hostName)
394 ereport(LOG,
395 (errmsg("could not translate host name \"%s\", service \"%s\" to address: %s",
396 hostName, service, gai_strerror(ret))));
397 else
398 ereport(LOG,
399 (errmsg("could not translate service \"%s\" to address: %s",
400 service, gai_strerror(ret))));
401 if (addrs)
402 pg_freeaddrinfo_all(hint.ai_family, addrs);
403 return STATUS_ERROR;
404 }
405
406 for (addr = addrs; addr; addr = addr->ai_next)
407 {
408 if (!IS_AF_UNIX(family) && IS_AF_UNIX(addr->ai_family))
409 {
410 /*
411 * Only set up a unix domain socket when they really asked for it.
412 * The service/port is different in that case.
413 */
414 continue;
415 }
416
417 /* See if there is still room to add 1 more socket. */
418 for (; listen_index < MaxListen; listen_index++)
419 {
420 if (ListenSocket[listen_index] == PGINVALID_SOCKET)
421 break;
422 }
423 if (listen_index >= MaxListen)
424 {
425 ereport(LOG,
426 (errmsg("could not bind to all requested addresses: MAXLISTEN (%d) exceeded",
427 MaxListen)));
428 break;
429 }
430
431 /* set up address family name for log messages */
432 switch (addr->ai_family)
433 {
434 case AF_INET:
435 familyDesc = _("IPv4");
436 break;
437 #ifdef HAVE_IPV6
438 case AF_INET6:
439 familyDesc = _("IPv6");
440 break;
441 #endif
442 #ifdef HAVE_UNIX_SOCKETS
443 case AF_UNIX:
444 familyDesc = _("Unix");
445 break;
446 #endif
447 default:
448 snprintf(familyDescBuf, sizeof(familyDescBuf),
449 _("unrecognized address family %d"),
450 addr->ai_family);
451 familyDesc = familyDescBuf;
452 break;
453 }
454
455 /* set up text form of address for log messages */
456 #ifdef HAVE_UNIX_SOCKETS
457 if (addr->ai_family == AF_UNIX)
458 addrDesc = unixSocketPath;
459 else
460 #endif
461 {
462 pg_getnameinfo_all((const struct sockaddr_storage *) addr->ai_addr,
463 addr->ai_addrlen,
464 addrBuf, sizeof(addrBuf),
465 NULL, 0,
466 NI_NUMERICHOST);
467 addrDesc = addrBuf;
468 }
469
470 if ((fd = socket(addr->ai_family, SOCK_STREAM, 0)) == PGINVALID_SOCKET)
471 {
472 ereport(LOG,
473 (errcode_for_socket_access(),
474 /* translator: first %s is IPv4, IPv6, or Unix */
475 errmsg("could not create %s socket for address \"%s\": %m",
476 familyDesc, addrDesc)));
477 continue;
478 }
479
480 #ifndef WIN32
481
482 /*
483 * Without the SO_REUSEADDR flag, a new postmaster can't be started
484 * right away after a stop or crash, giving "address already in use"
485 * error on TCP ports.
486 *
487 * On win32, however, this behavior only happens if the
488 * SO_EXLUSIVEADDRUSE is set. With SO_REUSEADDR, win32 allows multiple
489 * servers to listen on the same address, resulting in unpredictable
490 * behavior. With no flags at all, win32 behaves as Unix with
491 * SO_REUSEADDR.
492 */
493 if (!IS_AF_UNIX(addr->ai_family))
494 {
495 if ((setsockopt(fd, SOL_SOCKET, SO_REUSEADDR,
496 (char *) &one, sizeof(one))) == -1)
497 {
498 ereport(LOG,
499 (errcode_for_socket_access(),
500 /* translator: first %s is IPv4, IPv6, or Unix */
501 errmsg("setsockopt(SO_REUSEADDR) failed for %s address \"%s\": %m",
502 familyDesc, addrDesc)));
503 closesocket(fd);
504 continue;
505 }
506 }
507 #endif
508
509 #ifdef IPV6_V6ONLY
510 if (addr->ai_family == AF_INET6)
511 {
512 if (setsockopt(fd, IPPROTO_IPV6, IPV6_V6ONLY,
513 (char *) &one, sizeof(one)) == -1)
514 {
515 ereport(LOG,
516 (errcode_for_socket_access(),
517 /* translator: first %s is IPv4, IPv6, or Unix */
518 errmsg("setsockopt(IPV6_V6ONLY) failed for %s address \"%s\": %m",
519 familyDesc, addrDesc)));
520 closesocket(fd);
521 continue;
522 }
523 }
524 #endif
525
526 /*
527 * Note: This might fail on some OS's, like Linux older than
528 * 2.4.21-pre3, that don't have the IPV6_V6ONLY socket option, and map
529 * ipv4 addresses to ipv6. It will show ::ffff:ipv4 for all ipv4
530 * connections.
531 */
532 err = bind(fd, addr->ai_addr, addr->ai_addrlen);
533 if (err < 0)
534 {
535 ereport(LOG,
536 (errcode_for_socket_access(),
537 /* translator: first %s is IPv4, IPv6, or Unix */
538 errmsg("could not bind %s address \"%s\": %m",
539 familyDesc, addrDesc),
540 (IS_AF_UNIX(addr->ai_family)) ?
541 errhint("Is another postmaster already running on port %d?"
542 " If not, remove socket file \"%s\" and retry.",
543 (int) portNumber, service) :
544 errhint("Is another postmaster already running on port %d?"
545 " If not, wait a few seconds and retry.",
546 (int) portNumber)));
547 closesocket(fd);
548 continue;
549 }
550
551 #ifdef HAVE_UNIX_SOCKETS
552 if (addr->ai_family == AF_UNIX)
553 {
554 if (Setup_AF_UNIX(service) != STATUS_OK)
555 {
556 closesocket(fd);
557 break;
558 }
559 }
560 #endif
561
562 /*
563 * Select appropriate accept-queue length limit. PG_SOMAXCONN is only
564 * intended to provide a clamp on the request on platforms where an
565 * overly large request provokes a kernel error (are there any?).
566 */
567 maxconn = MaxBackends * 2;
568 if (maxconn > PG_SOMAXCONN)
569 maxconn = PG_SOMAXCONN;
570
571 err = listen(fd, maxconn);
572 if (err < 0)
573 {
574 ereport(LOG,
575 (errcode_for_socket_access(),
576 /* translator: first %s is IPv4, IPv6, or Unix */
577 errmsg("could not listen on %s address \"%s\": %m",
578 familyDesc, addrDesc)));
579 closesocket(fd);
580 continue;
581 }
582
583 #ifdef HAVE_UNIX_SOCKETS
584 if (addr->ai_family == AF_UNIX)
585 ereport(LOG,
586 (errmsg("listening on Unix socket \"%s\"",
587 addrDesc)));
588 else
589 #endif
590 ereport(LOG,
591 /* translator: first %s is IPv4 or IPv6 */
592 (errmsg("listening on %s address \"%s\", port %d",
593 familyDesc, addrDesc, (int) portNumber)));
594
595 ListenSocket[listen_index] = fd;
596 added++;
597 }
598
599 pg_freeaddrinfo_all(hint.ai_family, addrs);
600
601 if (!added)
602 return STATUS_ERROR;
603
604 return STATUS_OK;
605 }
606
607
608 #ifdef HAVE_UNIX_SOCKETS
609
610 /*
611 * Lock_AF_UNIX -- configure unix socket file path
612 */
613 static int
Lock_AF_UNIX(char * unixSocketDir,char * unixSocketPath)614 Lock_AF_UNIX(char *unixSocketDir, char *unixSocketPath)
615 {
616 /*
617 * Grab an interlock file associated with the socket file.
618 *
619 * Note: there are two reasons for using a socket lock file, rather than
620 * trying to interlock directly on the socket itself. First, it's a lot
621 * more portable, and second, it lets us remove any pre-existing socket
622 * file without race conditions.
623 */
624 CreateSocketLockFile(unixSocketPath, true, unixSocketDir);
625
626 /*
627 * Once we have the interlock, we can safely delete any pre-existing
628 * socket file to avoid failure at bind() time.
629 */
630 (void) unlink(unixSocketPath);
631
632 /*
633 * Remember socket file pathnames for later maintenance.
634 */
635 sock_paths = lappend(sock_paths, pstrdup(unixSocketPath));
636
637 return STATUS_OK;
638 }
639
640
641 /*
642 * Setup_AF_UNIX -- configure unix socket permissions
643 */
644 static int
Setup_AF_UNIX(char * sock_path)645 Setup_AF_UNIX(char *sock_path)
646 {
647 /*
648 * Fix socket ownership/permission if requested. Note we must do this
649 * before we listen() to avoid a window where unwanted connections could
650 * get accepted.
651 */
652 Assert(Unix_socket_group);
653 if (Unix_socket_group[0] != '\0')
654 {
655 #ifdef WIN32
656 elog(WARNING, "configuration item unix_socket_group is not supported on this platform");
657 #else
658 char *endptr;
659 unsigned long val;
660 gid_t gid;
661
662 val = strtoul(Unix_socket_group, &endptr, 10);
663 if (*endptr == '\0')
664 { /* numeric group id */
665 gid = val;
666 }
667 else
668 { /* convert group name to id */
669 struct group *gr;
670
671 gr = getgrnam(Unix_socket_group);
672 if (!gr)
673 {
674 ereport(LOG,
675 (errmsg("group \"%s\" does not exist",
676 Unix_socket_group)));
677 return STATUS_ERROR;
678 }
679 gid = gr->gr_gid;
680 }
681 if (chown(sock_path, -1, gid) == -1)
682 {
683 ereport(LOG,
684 (errcode_for_file_access(),
685 errmsg("could not set group of file \"%s\": %m",
686 sock_path)));
687 return STATUS_ERROR;
688 }
689 #endif
690 }
691
692 if (chmod(sock_path, Unix_socket_permissions) == -1)
693 {
694 ereport(LOG,
695 (errcode_for_file_access(),
696 errmsg("could not set permissions of file \"%s\": %m",
697 sock_path)));
698 return STATUS_ERROR;
699 }
700 return STATUS_OK;
701 }
702 #endif /* HAVE_UNIX_SOCKETS */
703
704
705 /*
706 * StreamConnection -- create a new connection with client using
707 * server port. Set port->sock to the FD of the new connection.
708 *
709 * ASSUME: that this doesn't need to be non-blocking because
710 * the Postmaster uses select() to tell when the server master
711 * socket is ready for accept().
712 *
713 * RETURNS: STATUS_OK or STATUS_ERROR
714 */
715 int
StreamConnection(pgsocket server_fd,Port * port)716 StreamConnection(pgsocket server_fd, Port *port)
717 {
718 /* accept connection and fill in the client (remote) address */
719 port->raddr.salen = sizeof(port->raddr.addr);
720 if ((port->sock = accept(server_fd,
721 (struct sockaddr *) &port->raddr.addr,
722 &port->raddr.salen)) == PGINVALID_SOCKET)
723 {
724 ereport(LOG,
725 (errcode_for_socket_access(),
726 errmsg("could not accept new connection: %m")));
727
728 /*
729 * If accept() fails then postmaster.c will still see the server
730 * socket as read-ready, and will immediately try again. To avoid
731 * uselessly sucking lots of CPU, delay a bit before trying again.
732 * (The most likely reason for failure is being out of kernel file
733 * table slots; we can do little except hope some will get freed up.)
734 */
735 pg_usleep(100000L); /* wait 0.1 sec */
736 return STATUS_ERROR;
737 }
738
739 /* fill in the server (local) address */
740 port->laddr.salen = sizeof(port->laddr.addr);
741 if (getsockname(port->sock,
742 (struct sockaddr *) &port->laddr.addr,
743 &port->laddr.salen) < 0)
744 {
745 elog(LOG, "getsockname() failed: %m");
746 return STATUS_ERROR;
747 }
748
749 /* select NODELAY and KEEPALIVE options if it's a TCP connection */
750 if (!IS_AF_UNIX(port->laddr.addr.ss_family))
751 {
752 int on;
753 #ifdef WIN32
754 int oldopt;
755 int optlen;
756 int newopt;
757 #endif
758
759 #ifdef TCP_NODELAY
760 on = 1;
761 if (setsockopt(port->sock, IPPROTO_TCP, TCP_NODELAY,
762 (char *) &on, sizeof(on)) < 0)
763 {
764 elog(LOG, "setsockopt(%s) failed: %m", "TCP_NODELAY");
765 return STATUS_ERROR;
766 }
767 #endif
768 on = 1;
769 if (setsockopt(port->sock, SOL_SOCKET, SO_KEEPALIVE,
770 (char *) &on, sizeof(on)) < 0)
771 {
772 elog(LOG, "setsockopt(%s) failed: %m", "SO_KEEPALIVE");
773 return STATUS_ERROR;
774 }
775
776 #ifdef WIN32
777
778 /*
779 * This is a Win32 socket optimization. The OS send buffer should be
780 * large enough to send the whole Postgres send buffer in one go, or
781 * performance suffers. The Postgres send buffer can be enlarged if a
782 * very large message needs to be sent, but we won't attempt to
783 * enlarge the OS buffer if that happens, so somewhat arbitrarily
784 * ensure that the OS buffer is at least PQ_SEND_BUFFER_SIZE * 4.
785 * (That's 32kB with the current default).
786 *
787 * The default OS buffer size used to be 8kB in earlier Windows
788 * versions, but was raised to 64kB in Windows 2012. So it shouldn't
789 * be necessary to change it in later versions anymore. Changing it
790 * unnecessarily can even reduce performance, because setting
791 * SO_SNDBUF in the application disables the "dynamic send buffering"
792 * feature that was introduced in Windows 7. So before fiddling with
793 * SO_SNDBUF, check if the current buffer size is already large enough
794 * and only increase it if necessary.
795 *
796 * See https://support.microsoft.com/kb/823764/EN-US/ and
797 * https://msdn.microsoft.com/en-us/library/bb736549%28v=vs.85%29.aspx
798 */
799 optlen = sizeof(oldopt);
800 if (getsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &oldopt,
801 &optlen) < 0)
802 {
803 elog(LOG, "getsockopt(%s) failed: %m", "SO_SNDBUF");
804 return STATUS_ERROR;
805 }
806 newopt = PQ_SEND_BUFFER_SIZE * 4;
807 if (oldopt < newopt)
808 {
809 if (setsockopt(port->sock, SOL_SOCKET, SO_SNDBUF, (char *) &newopt,
810 sizeof(newopt)) < 0)
811 {
812 elog(LOG, "setsockopt(%s) failed: %m", "SO_SNDBUF");
813 return STATUS_ERROR;
814 }
815 }
816 #endif
817
818 /*
819 * Also apply the current keepalive parameters. If we fail to set a
820 * parameter, don't error out, because these aren't universally
821 * supported. (Note: you might think we need to reset the GUC
822 * variables to 0 in such a case, but it's not necessary because the
823 * show hooks for these variables report the truth anyway.)
824 */
825 (void) pq_setkeepalivesidle(tcp_keepalives_idle, port);
826 (void) pq_setkeepalivesinterval(tcp_keepalives_interval, port);
827 (void) pq_setkeepalivescount(tcp_keepalives_count, port);
828 }
829
830 return STATUS_OK;
831 }
832
833 /*
834 * StreamClose -- close a client/backend connection
835 *
836 * NOTE: this is NOT used to terminate a session; it is just used to release
837 * the file descriptor in a process that should no longer have the socket
838 * open. (For example, the postmaster calls this after passing ownership
839 * of the connection to a child process.) It is expected that someone else
840 * still has the socket open. So, we only want to close the descriptor,
841 * we do NOT want to send anything to the far end.
842 */
843 void
StreamClose(pgsocket sock)844 StreamClose(pgsocket sock)
845 {
846 closesocket(sock);
847 }
848
849 /*
850 * TouchSocketFiles -- mark socket files as recently accessed
851 *
852 * This routine should be called every so often to ensure that the socket
853 * files have a recent mod date (ordinary operations on sockets usually won't
854 * change the mod date). That saves them from being removed by
855 * overenthusiastic /tmp-directory-cleaner daemons. (Another reason we should
856 * never have put the socket file in /tmp...)
857 */
858 void
TouchSocketFiles(void)859 TouchSocketFiles(void)
860 {
861 ListCell *l;
862
863 /* Loop through all created sockets... */
864 foreach(l, sock_paths)
865 {
866 char *sock_path = (char *) lfirst(l);
867
868 /*
869 * utime() is POSIX standard, utimes() is a common alternative. If we
870 * have neither, there's no way to affect the mod or access time of
871 * the socket :-(
872 *
873 * In either path, we ignore errors; there's no point in complaining.
874 */
875 #ifdef HAVE_UTIME
876 utime(sock_path, NULL);
877 #else /* !HAVE_UTIME */
878 #ifdef HAVE_UTIMES
879 utimes(sock_path, NULL);
880 #endif /* HAVE_UTIMES */
881 #endif /* HAVE_UTIME */
882 }
883 }
884
885 /*
886 * RemoveSocketFiles -- unlink socket files at postmaster shutdown
887 */
888 void
RemoveSocketFiles(void)889 RemoveSocketFiles(void)
890 {
891 ListCell *l;
892
893 /* Loop through all created sockets... */
894 foreach(l, sock_paths)
895 {
896 char *sock_path = (char *) lfirst(l);
897
898 /* Ignore any error. */
899 (void) unlink(sock_path);
900 }
901 /* Since we're about to exit, no need to reclaim storage */
902 sock_paths = NIL;
903 }
904
905
906 /* --------------------------------
907 * Low-level I/O routines begin here.
908 *
909 * These routines communicate with a frontend client across a connection
910 * already established by the preceding routines.
911 * --------------------------------
912 */
913
914 /* --------------------------------
915 * socket_set_nonblocking - set socket blocking/non-blocking
916 *
917 * Sets the socket non-blocking if nonblocking is true, or sets it
918 * blocking otherwise.
919 * --------------------------------
920 */
921 static void
socket_set_nonblocking(bool nonblocking)922 socket_set_nonblocking(bool nonblocking)
923 {
924 if (MyProcPort == NULL)
925 ereport(ERROR,
926 (errcode(ERRCODE_CONNECTION_DOES_NOT_EXIST),
927 errmsg("there is no client connection")));
928
929 MyProcPort->noblock = nonblocking;
930 }
931
932 /* --------------------------------
933 * pq_recvbuf - load some bytes into the input buffer
934 *
935 * returns 0 if OK, EOF if trouble
936 * --------------------------------
937 */
938 static int
pq_recvbuf(void)939 pq_recvbuf(void)
940 {
941 if (PqRecvPointer > 0)
942 {
943 if (PqRecvLength > PqRecvPointer)
944 {
945 /* still some unread data, left-justify it in the buffer */
946 memmove(PqRecvBuffer, PqRecvBuffer + PqRecvPointer,
947 PqRecvLength - PqRecvPointer);
948 PqRecvLength -= PqRecvPointer;
949 PqRecvPointer = 0;
950 }
951 else
952 PqRecvLength = PqRecvPointer = 0;
953 }
954
955 /* Ensure that we're in blocking mode */
956 socket_set_nonblocking(false);
957
958 /* Can fill buffer from PqRecvLength and upwards */
959 for (;;)
960 {
961 int r;
962
963 r = secure_read(MyProcPort, PqRecvBuffer + PqRecvLength,
964 PQ_RECV_BUFFER_SIZE - PqRecvLength);
965
966 if (r < 0)
967 {
968 if (errno == EINTR)
969 continue; /* Ok if interrupted */
970
971 /*
972 * Careful: an ereport() that tries to write to the client would
973 * cause recursion to here, leading to stack overflow and core
974 * dump! This message must go *only* to the postmaster log.
975 */
976 ereport(COMMERROR,
977 (errcode_for_socket_access(),
978 errmsg("could not receive data from client: %m")));
979 return EOF;
980 }
981 if (r == 0)
982 {
983 /*
984 * EOF detected. We used to write a log message here, but it's
985 * better to expect the ultimate caller to do that.
986 */
987 return EOF;
988 }
989 /* r contains number of bytes read, so just incr length */
990 PqRecvLength += r;
991 return 0;
992 }
993 }
994
995 /* --------------------------------
996 * pq_getbyte - get a single byte from connection, or return EOF
997 * --------------------------------
998 */
999 int
pq_getbyte(void)1000 pq_getbyte(void)
1001 {
1002 Assert(PqCommReadingMsg);
1003
1004 while (PqRecvPointer >= PqRecvLength)
1005 {
1006 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1007 return EOF; /* Failed to recv data */
1008 }
1009 return (unsigned char) PqRecvBuffer[PqRecvPointer++];
1010 }
1011
1012 /* --------------------------------
1013 * pq_peekbyte - peek at next byte from connection
1014 *
1015 * Same as pq_getbyte() except we don't advance the pointer.
1016 * --------------------------------
1017 */
1018 int
pq_peekbyte(void)1019 pq_peekbyte(void)
1020 {
1021 Assert(PqCommReadingMsg);
1022
1023 while (PqRecvPointer >= PqRecvLength)
1024 {
1025 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1026 return EOF; /* Failed to recv data */
1027 }
1028 return (unsigned char) PqRecvBuffer[PqRecvPointer];
1029 }
1030
1031 /* --------------------------------
1032 * pq_getbyte_if_available - get a single byte from connection,
1033 * if available
1034 *
1035 * The received byte is stored in *c. Returns 1 if a byte was read,
1036 * 0 if no data was available, or EOF if trouble.
1037 * --------------------------------
1038 */
1039 int
pq_getbyte_if_available(unsigned char * c)1040 pq_getbyte_if_available(unsigned char *c)
1041 {
1042 int r;
1043
1044 Assert(PqCommReadingMsg);
1045
1046 if (PqRecvPointer < PqRecvLength)
1047 {
1048 *c = PqRecvBuffer[PqRecvPointer++];
1049 return 1;
1050 }
1051
1052 /* Put the socket into non-blocking mode */
1053 socket_set_nonblocking(true);
1054
1055 r = secure_read(MyProcPort, c, 1);
1056 if (r < 0)
1057 {
1058 /*
1059 * Ok if no data available without blocking or interrupted (though
1060 * EINTR really shouldn't happen with a non-blocking socket). Report
1061 * other errors.
1062 */
1063 if (errno == EAGAIN || errno == EWOULDBLOCK || errno == EINTR)
1064 r = 0;
1065 else
1066 {
1067 /*
1068 * Careful: an ereport() that tries to write to the client would
1069 * cause recursion to here, leading to stack overflow and core
1070 * dump! This message must go *only* to the postmaster log.
1071 */
1072 ereport(COMMERROR,
1073 (errcode_for_socket_access(),
1074 errmsg("could not receive data from client: %m")));
1075 r = EOF;
1076 }
1077 }
1078 else if (r == 0)
1079 {
1080 /* EOF detected */
1081 r = EOF;
1082 }
1083
1084 return r;
1085 }
1086
1087 /* --------------------------------
1088 * pq_getbytes - get a known number of bytes from connection
1089 *
1090 * returns 0 if OK, EOF if trouble
1091 * --------------------------------
1092 */
1093 int
pq_getbytes(char * s,size_t len)1094 pq_getbytes(char *s, size_t len)
1095 {
1096 size_t amount;
1097
1098 Assert(PqCommReadingMsg);
1099
1100 while (len > 0)
1101 {
1102 while (PqRecvPointer >= PqRecvLength)
1103 {
1104 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1105 return EOF; /* Failed to recv data */
1106 }
1107 amount = PqRecvLength - PqRecvPointer;
1108 if (amount > len)
1109 amount = len;
1110 memcpy(s, PqRecvBuffer + PqRecvPointer, amount);
1111 PqRecvPointer += amount;
1112 s += amount;
1113 len -= amount;
1114 }
1115 return 0;
1116 }
1117
1118 /* --------------------------------
1119 * pq_discardbytes - throw away a known number of bytes
1120 *
1121 * same as pq_getbytes except we do not copy the data to anyplace.
1122 * this is used for resynchronizing after read errors.
1123 *
1124 * returns 0 if OK, EOF if trouble
1125 * --------------------------------
1126 */
1127 static int
pq_discardbytes(size_t len)1128 pq_discardbytes(size_t len)
1129 {
1130 size_t amount;
1131
1132 Assert(PqCommReadingMsg);
1133
1134 while (len > 0)
1135 {
1136 while (PqRecvPointer >= PqRecvLength)
1137 {
1138 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1139 return EOF; /* Failed to recv data */
1140 }
1141 amount = PqRecvLength - PqRecvPointer;
1142 if (amount > len)
1143 amount = len;
1144 PqRecvPointer += amount;
1145 len -= amount;
1146 }
1147 return 0;
1148 }
1149
1150 /* --------------------------------
1151 * pq_getstring - get a null terminated string from connection
1152 *
1153 * The return value is placed in an expansible StringInfo, which has
1154 * already been initialized by the caller.
1155 *
1156 * This is used only for dealing with old-protocol clients. The idea
1157 * is to produce a StringInfo that looks the same as we would get from
1158 * pq_getmessage() with a newer client; we will then process it with
1159 * pq_getmsgstring. Therefore, no character set conversion is done here,
1160 * even though this is presumably useful only for text.
1161 *
1162 * returns 0 if OK, EOF if trouble
1163 * --------------------------------
1164 */
1165 int
pq_getstring(StringInfo s)1166 pq_getstring(StringInfo s)
1167 {
1168 int i;
1169
1170 Assert(PqCommReadingMsg);
1171
1172 resetStringInfo(s);
1173
1174 /* Read until we get the terminating '\0' */
1175 for (;;)
1176 {
1177 while (PqRecvPointer >= PqRecvLength)
1178 {
1179 if (pq_recvbuf()) /* If nothing in buffer, then recv some */
1180 return EOF; /* Failed to recv data */
1181 }
1182
1183 for (i = PqRecvPointer; i < PqRecvLength; i++)
1184 {
1185 if (PqRecvBuffer[i] == '\0')
1186 {
1187 /* include the '\0' in the copy */
1188 appendBinaryStringInfo(s, PqRecvBuffer + PqRecvPointer,
1189 i - PqRecvPointer + 1);
1190 PqRecvPointer = i + 1; /* advance past \0 */
1191 return 0;
1192 }
1193 }
1194
1195 /* If we're here we haven't got the \0 in the buffer yet. */
1196 appendBinaryStringInfo(s, PqRecvBuffer + PqRecvPointer,
1197 PqRecvLength - PqRecvPointer);
1198 PqRecvPointer = PqRecvLength;
1199 }
1200 }
1201
1202 /* --------------------------------
1203 * pq_buffer_has_data - is any buffered data available to read?
1204 *
1205 * This will *not* attempt to read more data.
1206 * --------------------------------
1207 */
1208 bool
pq_buffer_has_data(void)1209 pq_buffer_has_data(void)
1210 {
1211 return (PqRecvPointer < PqRecvLength);
1212 }
1213
1214
1215 /* --------------------------------
1216 * pq_startmsgread - begin reading a message from the client.
1217 *
1218 * This must be called before any of the pq_get* functions.
1219 * --------------------------------
1220 */
1221 void
pq_startmsgread(void)1222 pq_startmsgread(void)
1223 {
1224 /*
1225 * There shouldn't be a read active already, but let's check just to be
1226 * sure.
1227 */
1228 if (PqCommReadingMsg)
1229 ereport(FATAL,
1230 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1231 errmsg("terminating connection because protocol synchronization was lost")));
1232
1233 PqCommReadingMsg = true;
1234 }
1235
1236
1237 /* --------------------------------
1238 * pq_endmsgread - finish reading message.
1239 *
1240 * This must be called after reading a V2 protocol message with
1241 * pq_getstring() and friends, to indicate that we have read the whole
1242 * message. In V3 protocol, pq_getmessage() does this implicitly.
1243 * --------------------------------
1244 */
1245 void
pq_endmsgread(void)1246 pq_endmsgread(void)
1247 {
1248 Assert(PqCommReadingMsg);
1249
1250 PqCommReadingMsg = false;
1251 }
1252
1253 /* --------------------------------
1254 * pq_is_reading_msg - are we currently reading a message?
1255 *
1256 * This is used in error recovery at the outer idle loop to detect if we have
1257 * lost protocol sync, and need to terminate the connection. pq_startmsgread()
1258 * will check for that too, but it's nicer to detect it earlier.
1259 * --------------------------------
1260 */
1261 bool
pq_is_reading_msg(void)1262 pq_is_reading_msg(void)
1263 {
1264 return PqCommReadingMsg;
1265 }
1266
1267 /* --------------------------------
1268 * pq_getmessage - get a message with length word from connection
1269 *
1270 * The return value is placed in an expansible StringInfo, which has
1271 * already been initialized by the caller.
1272 * Only the message body is placed in the StringInfo; the length word
1273 * is removed. Also, s->cursor is initialized to zero for convenience
1274 * in scanning the message contents.
1275 *
1276 * If maxlen is not zero, it is an upper limit on the length of the
1277 * message we are willing to accept. We abort the connection (by
1278 * returning EOF) if client tries to send more than that.
1279 *
1280 * returns 0 if OK, EOF if trouble
1281 * --------------------------------
1282 */
1283 int
pq_getmessage(StringInfo s,int maxlen)1284 pq_getmessage(StringInfo s, int maxlen)
1285 {
1286 int32 len;
1287
1288 Assert(PqCommReadingMsg);
1289
1290 resetStringInfo(s);
1291
1292 /* Read message length word */
1293 if (pq_getbytes((char *) &len, 4) == EOF)
1294 {
1295 ereport(COMMERROR,
1296 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1297 errmsg("unexpected EOF within message length word")));
1298 return EOF;
1299 }
1300
1301 len = pg_ntoh32(len);
1302
1303 if (len < 4 ||
1304 (maxlen > 0 && len > maxlen))
1305 {
1306 ereport(COMMERROR,
1307 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1308 errmsg("invalid message length")));
1309 return EOF;
1310 }
1311
1312 len -= 4; /* discount length itself */
1313
1314 if (len > 0)
1315 {
1316 /*
1317 * Allocate space for message. If we run out of room (ridiculously
1318 * large message), we will elog(ERROR), but we want to discard the
1319 * message body so as not to lose communication sync.
1320 */
1321 PG_TRY();
1322 {
1323 enlargeStringInfo(s, len);
1324 }
1325 PG_CATCH();
1326 {
1327 if (pq_discardbytes(len) == EOF)
1328 ereport(COMMERROR,
1329 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1330 errmsg("incomplete message from client")));
1331
1332 /* we discarded the rest of the message so we're back in sync. */
1333 PqCommReadingMsg = false;
1334 PG_RE_THROW();
1335 }
1336 PG_END_TRY();
1337
1338 /* And grab the message */
1339 if (pq_getbytes(s->data, len) == EOF)
1340 {
1341 ereport(COMMERROR,
1342 (errcode(ERRCODE_PROTOCOL_VIOLATION),
1343 errmsg("incomplete message from client")));
1344 return EOF;
1345 }
1346 s->len = len;
1347 /* Place a trailing null per StringInfo convention */
1348 s->data[len] = '\0';
1349 }
1350
1351 /* finished reading the message. */
1352 PqCommReadingMsg = false;
1353
1354 return 0;
1355 }
1356
1357
1358 /* --------------------------------
1359 * pq_putbytes - send bytes to connection (not flushed until pq_flush)
1360 *
1361 * returns 0 if OK, EOF if trouble
1362 * --------------------------------
1363 */
1364 int
pq_putbytes(const char * s,size_t len)1365 pq_putbytes(const char *s, size_t len)
1366 {
1367 int res;
1368
1369 /* Should only be called by old-style COPY OUT */
1370 Assert(DoingCopyOut);
1371 /* No-op if reentrant call */
1372 if (PqCommBusy)
1373 return 0;
1374 PqCommBusy = true;
1375 res = internal_putbytes(s, len);
1376 PqCommBusy = false;
1377 return res;
1378 }
1379
1380 static int
internal_putbytes(const char * s,size_t len)1381 internal_putbytes(const char *s, size_t len)
1382 {
1383 size_t amount;
1384
1385 while (len > 0)
1386 {
1387 /* If buffer is full, then flush it out */
1388 if (PqSendPointer >= PqSendBufferSize)
1389 {
1390 socket_set_nonblocking(false);
1391 if (internal_flush())
1392 return EOF;
1393 }
1394 amount = PqSendBufferSize - PqSendPointer;
1395 if (amount > len)
1396 amount = len;
1397 memcpy(PqSendBuffer + PqSendPointer, s, amount);
1398 PqSendPointer += amount;
1399 s += amount;
1400 len -= amount;
1401 }
1402 return 0;
1403 }
1404
1405 /* --------------------------------
1406 * socket_flush - flush pending output
1407 *
1408 * returns 0 if OK, EOF if trouble
1409 * --------------------------------
1410 */
1411 static int
socket_flush(void)1412 socket_flush(void)
1413 {
1414 int res;
1415
1416 /* No-op if reentrant call */
1417 if (PqCommBusy)
1418 return 0;
1419 PqCommBusy = true;
1420 socket_set_nonblocking(false);
1421 res = internal_flush();
1422 PqCommBusy = false;
1423 return res;
1424 }
1425
1426 /* --------------------------------
1427 * internal_flush - flush pending output
1428 *
1429 * Returns 0 if OK (meaning everything was sent, or operation would block
1430 * and the socket is in non-blocking mode), or EOF if trouble.
1431 * --------------------------------
1432 */
1433 static int
internal_flush(void)1434 internal_flush(void)
1435 {
1436 static int last_reported_send_errno = 0;
1437
1438 char *bufptr = PqSendBuffer + PqSendStart;
1439 char *bufend = PqSendBuffer + PqSendPointer;
1440
1441 while (bufptr < bufend)
1442 {
1443 int r;
1444
1445 r = secure_write(MyProcPort, bufptr, bufend - bufptr);
1446
1447 if (r <= 0)
1448 {
1449 if (errno == EINTR)
1450 continue; /* Ok if we were interrupted */
1451
1452 /*
1453 * Ok if no data writable without blocking, and the socket is in
1454 * non-blocking mode.
1455 */
1456 if (errno == EAGAIN ||
1457 errno == EWOULDBLOCK)
1458 {
1459 return 0;
1460 }
1461
1462 /*
1463 * Careful: an ereport() that tries to write to the client would
1464 * cause recursion to here, leading to stack overflow and core
1465 * dump! This message must go *only* to the postmaster log.
1466 *
1467 * If a client disconnects while we're in the midst of output, we
1468 * might write quite a bit of data before we get to a safe query
1469 * abort point. So, suppress duplicate log messages.
1470 */
1471 if (errno != last_reported_send_errno)
1472 {
1473 last_reported_send_errno = errno;
1474 ereport(COMMERROR,
1475 (errcode_for_socket_access(),
1476 errmsg("could not send data to client: %m")));
1477 }
1478
1479 /*
1480 * We drop the buffered data anyway so that processing can
1481 * continue, even though we'll probably quit soon. We also set a
1482 * flag that'll cause the next CHECK_FOR_INTERRUPTS to terminate
1483 * the connection.
1484 */
1485 PqSendStart = PqSendPointer = 0;
1486 ClientConnectionLost = 1;
1487 InterruptPending = 1;
1488 return EOF;
1489 }
1490
1491 last_reported_send_errno = 0; /* reset after any successful send */
1492 bufptr += r;
1493 PqSendStart += r;
1494 }
1495
1496 PqSendStart = PqSendPointer = 0;
1497 return 0;
1498 }
1499
1500 /* --------------------------------
1501 * pq_flush_if_writable - flush pending output if writable without blocking
1502 *
1503 * Returns 0 if OK, or EOF if trouble.
1504 * --------------------------------
1505 */
1506 static int
socket_flush_if_writable(void)1507 socket_flush_if_writable(void)
1508 {
1509 int res;
1510
1511 /* Quick exit if nothing to do */
1512 if (PqSendPointer == PqSendStart)
1513 return 0;
1514
1515 /* No-op if reentrant call */
1516 if (PqCommBusy)
1517 return 0;
1518
1519 /* Temporarily put the socket into non-blocking mode */
1520 socket_set_nonblocking(true);
1521
1522 PqCommBusy = true;
1523 res = internal_flush();
1524 PqCommBusy = false;
1525 return res;
1526 }
1527
1528 /* --------------------------------
1529 * socket_is_send_pending - is there any pending data in the output buffer?
1530 * --------------------------------
1531 */
1532 static bool
socket_is_send_pending(void)1533 socket_is_send_pending(void)
1534 {
1535 return (PqSendStart < PqSendPointer);
1536 }
1537
1538 /* --------------------------------
1539 * Message-level I/O routines begin here.
1540 *
1541 * These routines understand about the old-style COPY OUT protocol.
1542 * --------------------------------
1543 */
1544
1545
1546 /* --------------------------------
1547 * socket_putmessage - send a normal message (suppressed in COPY OUT mode)
1548 *
1549 * If msgtype is not '\0', it is a message type code to place before
1550 * the message body. If msgtype is '\0', then the message has no type
1551 * code (this is only valid in pre-3.0 protocols).
1552 *
1553 * len is the length of the message body data at *s. In protocol 3.0
1554 * and later, a message length word (equal to len+4 because it counts
1555 * itself too) is inserted by this routine.
1556 *
1557 * All normal messages are suppressed while old-style COPY OUT is in
1558 * progress. (In practice only a few notice messages might get emitted
1559 * then; dropping them is annoying, but at least they will still appear
1560 * in the postmaster log.)
1561 *
1562 * We also suppress messages generated while pqcomm.c is busy. This
1563 * avoids any possibility of messages being inserted within other
1564 * messages. The only known trouble case arises if SIGQUIT occurs
1565 * during a pqcomm.c routine --- quickdie() will try to send a warning
1566 * message, and the most reasonable approach seems to be to drop it.
1567 *
1568 * returns 0 if OK, EOF if trouble
1569 * --------------------------------
1570 */
1571 static int
socket_putmessage(char msgtype,const char * s,size_t len)1572 socket_putmessage(char msgtype, const char *s, size_t len)
1573 {
1574 if (DoingCopyOut || PqCommBusy)
1575 return 0;
1576 PqCommBusy = true;
1577 if (msgtype)
1578 if (internal_putbytes(&msgtype, 1))
1579 goto fail;
1580 if (PG_PROTOCOL_MAJOR(FrontendProtocol) >= 3)
1581 {
1582 uint32 n32;
1583
1584 n32 = pg_hton32((uint32) (len + 4));
1585 if (internal_putbytes((char *) &n32, 4))
1586 goto fail;
1587 }
1588 if (internal_putbytes(s, len))
1589 goto fail;
1590 PqCommBusy = false;
1591 return 0;
1592
1593 fail:
1594 PqCommBusy = false;
1595 return EOF;
1596 }
1597
1598 /* --------------------------------
1599 * pq_putmessage_noblock - like pq_putmessage, but never blocks
1600 *
1601 * If the output buffer is too small to hold the message, the buffer
1602 * is enlarged.
1603 */
1604 static void
socket_putmessage_noblock(char msgtype,const char * s,size_t len)1605 socket_putmessage_noblock(char msgtype, const char *s, size_t len)
1606 {
1607 int res PG_USED_FOR_ASSERTS_ONLY;
1608 int required;
1609
1610 /*
1611 * Ensure we have enough space in the output buffer for the message header
1612 * as well as the message itself.
1613 */
1614 required = PqSendPointer + 1 + 4 + len;
1615 if (required > PqSendBufferSize)
1616 {
1617 PqSendBuffer = repalloc(PqSendBuffer, required);
1618 PqSendBufferSize = required;
1619 }
1620 res = pq_putmessage(msgtype, s, len);
1621 Assert(res == 0); /* should not fail when the message fits in
1622 * buffer */
1623 }
1624
1625
1626 /* --------------------------------
1627 * socket_startcopyout - inform libpq that an old-style COPY OUT transfer
1628 * is beginning
1629 * --------------------------------
1630 */
1631 static void
socket_startcopyout(void)1632 socket_startcopyout(void)
1633 {
1634 DoingCopyOut = true;
1635 }
1636
1637 /* --------------------------------
1638 * socket_endcopyout - end an old-style COPY OUT transfer
1639 *
1640 * If errorAbort is indicated, we are aborting a COPY OUT due to an error,
1641 * and must send a terminator line. Since a partial data line might have
1642 * been emitted, send a couple of newlines first (the first one could
1643 * get absorbed by a backslash...) Note that old-style COPY OUT does
1644 * not allow binary transfers, so a textual terminator is always correct.
1645 * --------------------------------
1646 */
1647 static void
socket_endcopyout(bool errorAbort)1648 socket_endcopyout(bool errorAbort)
1649 {
1650 if (!DoingCopyOut)
1651 return;
1652 if (errorAbort)
1653 pq_putbytes("\n\n\\.\n", 5);
1654 /* in non-error case, copy.c will have emitted the terminator line */
1655 DoingCopyOut = false;
1656 }
1657
1658 /*
1659 * Support for TCP Keepalive parameters
1660 */
1661
1662 /*
1663 * On Windows, we need to set both idle and interval at the same time.
1664 * We also cannot reset them to the default (setting to zero will
1665 * actually set them to zero, not default), therefore we fallback to
1666 * the out-of-the-box default instead.
1667 */
1668 #if defined(WIN32) && defined(SIO_KEEPALIVE_VALS)
1669 static int
pq_setkeepaliveswin32(Port * port,int idle,int interval)1670 pq_setkeepaliveswin32(Port *port, int idle, int interval)
1671 {
1672 struct tcp_keepalive ka;
1673 DWORD retsize;
1674
1675 if (idle <= 0)
1676 idle = 2 * 60 * 60; /* default = 2 hours */
1677 if (interval <= 0)
1678 interval = 1; /* default = 1 second */
1679
1680 ka.onoff = 1;
1681 ka.keepalivetime = idle * 1000;
1682 ka.keepaliveinterval = interval * 1000;
1683
1684 if (WSAIoctl(port->sock,
1685 SIO_KEEPALIVE_VALS,
1686 (LPVOID) &ka,
1687 sizeof(ka),
1688 NULL,
1689 0,
1690 &retsize,
1691 NULL,
1692 NULL)
1693 != 0)
1694 {
1695 elog(LOG, "WSAIoctl(SIO_KEEPALIVE_VALS) failed: %ui",
1696 WSAGetLastError());
1697 return STATUS_ERROR;
1698 }
1699 if (port->keepalives_idle != idle)
1700 port->keepalives_idle = idle;
1701 if (port->keepalives_interval != interval)
1702 port->keepalives_interval = interval;
1703 return STATUS_OK;
1704 }
1705 #endif
1706
1707 int
pq_getkeepalivesidle(Port * port)1708 pq_getkeepalivesidle(Port *port)
1709 {
1710 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1711 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1712 return 0;
1713
1714 if (port->keepalives_idle != 0)
1715 return port->keepalives_idle;
1716
1717 if (port->default_keepalives_idle == 0)
1718 {
1719 #ifndef WIN32
1720 ACCEPT_TYPE_ARG3 size = sizeof(port->default_keepalives_idle);
1721
1722 if (getsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1723 (char *) &port->default_keepalives_idle,
1724 &size) < 0)
1725 {
1726 elog(LOG, "getsockopt(%s) failed: %m", PG_TCP_KEEPALIVE_IDLE_STR);
1727 port->default_keepalives_idle = -1; /* don't know */
1728 }
1729 #else /* WIN32 */
1730 /* We can't get the defaults on Windows, so return "don't know" */
1731 port->default_keepalives_idle = -1;
1732 #endif /* WIN32 */
1733 }
1734
1735 return port->default_keepalives_idle;
1736 #else
1737 return 0;
1738 #endif
1739 }
1740
1741 int
pq_setkeepalivesidle(int idle,Port * port)1742 pq_setkeepalivesidle(int idle, Port *port)
1743 {
1744 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1745 return STATUS_OK;
1746
1747 /* check SIO_KEEPALIVE_VALS here, not just WIN32, as some toolchains lack it */
1748 #if defined(PG_TCP_KEEPALIVE_IDLE) || defined(SIO_KEEPALIVE_VALS)
1749 if (idle == port->keepalives_idle)
1750 return STATUS_OK;
1751
1752 #ifndef WIN32
1753 if (port->default_keepalives_idle <= 0)
1754 {
1755 if (pq_getkeepalivesidle(port) < 0)
1756 {
1757 if (idle == 0)
1758 return STATUS_OK; /* default is set but unknown */
1759 else
1760 return STATUS_ERROR;
1761 }
1762 }
1763
1764 if (idle == 0)
1765 idle = port->default_keepalives_idle;
1766
1767 if (setsockopt(port->sock, IPPROTO_TCP, PG_TCP_KEEPALIVE_IDLE,
1768 (char *) &idle, sizeof(idle)) < 0)
1769 {
1770 elog(LOG, "setsockopt(%s) failed: %m", PG_TCP_KEEPALIVE_IDLE_STR);
1771 return STATUS_ERROR;
1772 }
1773
1774 port->keepalives_idle = idle;
1775 #else /* WIN32 */
1776 return pq_setkeepaliveswin32(port, idle, port->keepalives_interval);
1777 #endif
1778 #else
1779 if (idle != 0)
1780 {
1781 elog(LOG, "setting the keepalive idle time is not supported");
1782 return STATUS_ERROR;
1783 }
1784 #endif
1785
1786 return STATUS_OK;
1787 }
1788
1789 int
pq_getkeepalivesinterval(Port * port)1790 pq_getkeepalivesinterval(Port *port)
1791 {
1792 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1793 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1794 return 0;
1795
1796 if (port->keepalives_interval != 0)
1797 return port->keepalives_interval;
1798
1799 if (port->default_keepalives_interval == 0)
1800 {
1801 #ifndef WIN32
1802 ACCEPT_TYPE_ARG3 size = sizeof(port->default_keepalives_interval);
1803
1804 if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1805 (char *) &port->default_keepalives_interval,
1806 &size) < 0)
1807 {
1808 elog(LOG, "getsockopt(%s) failed: %m", "TCP_KEEPINTVL");
1809 port->default_keepalives_interval = -1; /* don't know */
1810 }
1811 #else
1812 /* We can't get the defaults on Windows, so return "don't know" */
1813 port->default_keepalives_interval = -1;
1814 #endif /* WIN32 */
1815 }
1816
1817 return port->default_keepalives_interval;
1818 #else
1819 return 0;
1820 #endif
1821 }
1822
1823 int
pq_setkeepalivesinterval(int interval,Port * port)1824 pq_setkeepalivesinterval(int interval, Port *port)
1825 {
1826 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1827 return STATUS_OK;
1828
1829 #if defined(TCP_KEEPINTVL) || defined(SIO_KEEPALIVE_VALS)
1830 if (interval == port->keepalives_interval)
1831 return STATUS_OK;
1832
1833 #ifndef WIN32
1834 if (port->default_keepalives_interval <= 0)
1835 {
1836 if (pq_getkeepalivesinterval(port) < 0)
1837 {
1838 if (interval == 0)
1839 return STATUS_OK; /* default is set but unknown */
1840 else
1841 return STATUS_ERROR;
1842 }
1843 }
1844
1845 if (interval == 0)
1846 interval = port->default_keepalives_interval;
1847
1848 if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPINTVL,
1849 (char *) &interval, sizeof(interval)) < 0)
1850 {
1851 elog(LOG, "setsockopt(%s) failed: %m", "TCP_KEEPINTVL");
1852 return STATUS_ERROR;
1853 }
1854
1855 port->keepalives_interval = interval;
1856 #else /* WIN32 */
1857 return pq_setkeepaliveswin32(port, port->keepalives_idle, interval);
1858 #endif
1859 #else
1860 if (interval != 0)
1861 {
1862 elog(LOG, "setsockopt(%s) not supported", "TCP_KEEPINTVL");
1863 return STATUS_ERROR;
1864 }
1865 #endif
1866
1867 return STATUS_OK;
1868 }
1869
1870 int
pq_getkeepalivescount(Port * port)1871 pq_getkeepalivescount(Port *port)
1872 {
1873 #ifdef TCP_KEEPCNT
1874 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1875 return 0;
1876
1877 if (port->keepalives_count != 0)
1878 return port->keepalives_count;
1879
1880 if (port->default_keepalives_count == 0)
1881 {
1882 ACCEPT_TYPE_ARG3 size = sizeof(port->default_keepalives_count);
1883
1884 if (getsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1885 (char *) &port->default_keepalives_count,
1886 &size) < 0)
1887 {
1888 elog(LOG, "getsockopt(%s) failed: %m", "TCP_KEEPCNT");
1889 port->default_keepalives_count = -1; /* don't know */
1890 }
1891 }
1892
1893 return port->default_keepalives_count;
1894 #else
1895 return 0;
1896 #endif
1897 }
1898
1899 int
pq_setkeepalivescount(int count,Port * port)1900 pq_setkeepalivescount(int count, Port *port)
1901 {
1902 if (port == NULL || IS_AF_UNIX(port->laddr.addr.ss_family))
1903 return STATUS_OK;
1904
1905 #ifdef TCP_KEEPCNT
1906 if (count == port->keepalives_count)
1907 return STATUS_OK;
1908
1909 if (port->default_keepalives_count <= 0)
1910 {
1911 if (pq_getkeepalivescount(port) < 0)
1912 {
1913 if (count == 0)
1914 return STATUS_OK; /* default is set but unknown */
1915 else
1916 return STATUS_ERROR;
1917 }
1918 }
1919
1920 if (count == 0)
1921 count = port->default_keepalives_count;
1922
1923 if (setsockopt(port->sock, IPPROTO_TCP, TCP_KEEPCNT,
1924 (char *) &count, sizeof(count)) < 0)
1925 {
1926 elog(LOG, "setsockopt(%s) failed: %m", "TCP_KEEPCNT");
1927 return STATUS_ERROR;
1928 }
1929
1930 port->keepalives_count = count;
1931 #else
1932 if (count != 0)
1933 {
1934 elog(LOG, "setsockopt(%s) not supported", "TCP_KEEPCNT");
1935 return STATUS_ERROR;
1936 }
1937 #endif
1938
1939 return STATUS_OK;
1940 }
1941