| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* copyto.c |
| 4 |
* COPY <table> TO file/program/client |
| 5 |
* |
| 6 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 7 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 8 |
* |
| 9 |
* |
| 10 |
* IDENTIFICATION |
| 11 |
* src/backend/commands/copyto.c |
| 12 |
* |
| 13 |
*------------------------------------------------------------------------- |
| 14 |
*/ |
| 15 |
#include "postgres.h" |
| 16 |
|
| 17 |
#include <ctype.h> |
| 18 |
#include <unistd.h> |
| 19 |
#include <sys/stat.h> |
| 20 |
|
| 21 |
#include "access/tableam.h" |
| 22 |
#include "commands/copyapi.h" |
| 23 |
#include "commands/progress.h" |
| 24 |
#include "executor/execdesc.h" |
| 25 |
#include "executor/executor.h" |
| 26 |
#include "executor/tuptable.h" |
| 27 |
#include "libpq/libpq.h" |
| 28 |
#include "libpq/pqformat.h" |
| 29 |
#include "mb/pg_wchar.h" |
| 30 |
#include "miscadmin.h" |
| 31 |
#include "pgstat.h" |
| 32 |
#include "storage/fd.h" |
| 33 |
#include "tcop/tcopprot.h" |
| 34 |
#include "utils/lsyscache.h" |
| 35 |
#include "utils/memutils.h" |
| 36 |
#include "utils/rel.h" |
| 37 |
#include "utils/snapmgr.h" |
| 38 |
|
| 39 |
/* |
| 40 |
* Represents the different dest cases we need to worry about at |
| 41 |
* the bottom level |
| 42 |
*/ |
| 43 |
typedef enum CopyDest |
| 44 |
{ |
| 45 |
COPY_FILE, /* to file (or a piped program) */ |
| 46 |
COPY_FRONTEND, /* to frontend */ |
| 47 |
COPY_CALLBACK, /* to callback function */ |
| 48 |
} CopyDest; |
| 49 |
|
| 50 |
/* |
| 51 |
* This struct contains all the state variables used throughout a COPY TO |
| 52 |
* operation. |
| 53 |
* |
| 54 |
* Multi-byte encodings: all supported client-side encodings encode multi-byte |
| 55 |
* characters by having the first byte's high bit set. Subsequent bytes of the |
| 56 |
* character can have the high bit not set. When scanning data in such an |
| 57 |
* encoding to look for a match to a single-byte (ie ASCII) character, we must |
| 58 |
* use the full pg_encoding_mblen() machinery to skip over multibyte |
| 59 |
* characters, else we might find a false match to a trailing byte. In |
| 60 |
* supported server encodings, there is no possibility of a false match, and |
| 61 |
* it's faster to make useless comparisons to trailing bytes than it is to |
| 62 |
* invoke pg_encoding_mblen() to skip over them. encoding_embeds_ascii is true |
| 63 |
* when we have to do it the hard way. |
| 64 |
*/ |
| 65 |
typedef struct CopyToStateData |
| 66 |
{ |
| 67 |
/* format-specific routines */ |
| 68 |
const CopyToRoutine *routine; |
| 69 |
|
| 70 |
/* low-level state data */ |
| 71 |
CopyDest copy_dest; /* type of copy source/destination */ |
| 72 |
FILE *copy_file; /* used if copy_dest == COPY_FILE */ |
| 73 |
StringInfo fe_msgbuf; /* used for all dests during COPY TO */ |
| 74 |
|
| 75 |
int file_encoding; /* file or remote side's character encoding */ |
| 76 |
bool need_transcoding; /* file encoding diff from server? */ |
| 77 |
bool encoding_embeds_ascii; /* ASCII can be non-first byte? */ |
| 78 |
|
| 79 |
/* parameters from the COPY command */ |
| 80 |
Relation rel; /* relation to copy to */ |
| 81 |
QueryDesc *queryDesc; /* executable query to copy from */ |
| 82 |
List *attnumlist; /* integer list of attnums to copy */ |
| 83 |
char *filename; /* filename, or NULL for STDOUT */ |
| 84 |
bool is_program; /* is 'filename' a program to popen? */ |
| 85 |
copy_data_dest_cb data_dest_cb; /* function for writing data */ |
| 86 |
|
| 87 |
CopyFormatOptions opts; |
| 88 |
Node *whereClause; /* WHERE condition (or NULL) */ |
| 89 |
|
| 90 |
/* |
| 91 |
* Working state |
| 92 |
*/ |
| 93 |
MemoryContext copycontext; /* per-copy execution context */ |
| 94 |
|
| 95 |
FmgrInfo *out_functions; /* lookup info for output functions */ |
| 96 |
MemoryContext rowcontext; /* per-row evaluation context */ |
| 97 |
uint64 bytes_processed; /* number of bytes processed so far */ |
| 98 |
} CopyToStateData; |
| 99 |
|
| 100 |
/* DestReceiver for COPY (query) TO */ |
| 101 |
typedef struct |
| 102 |
{ |
| 103 |
DestReceiver pub; /* publicly-known function pointers */ |
| 104 |
CopyToState cstate; /* CopyToStateData for the command */ |
| 105 |
uint64 processed; /* # of tuples processed */ |
| 106 |
} DR_copy; |
| 107 |
|
| 108 |
/* NOTE: there's a copy of this in copyfromparse.c */ |
| 109 |
static const char BinarySignature[11] = "PGCOPY\n\377\r\n\0"; |
| 110 |
|
| 111 |
|
| 112 |
/* non-export function prototypes */ |
| 113 |
static void EndCopy(CopyToState cstate); |
| 114 |
static void ClosePipeToProgram(CopyToState cstate); |
| 115 |
static void CopyOneRowTo(CopyToState cstate, TupleTableSlot *slot); |
| 116 |
static void CopyAttributeOutText(CopyToState cstate, const char *string); |
| 117 |
static void CopyAttributeOutCSV(CopyToState cstate, const char *string, |
| 118 |
bool use_quote); |
| 119 |
|
| 120 |
/* built-in format-specific routines */ |
| 121 |
static void CopyToTextLikeStart(CopyToState cstate, TupleDesc tupDesc); |
| 122 |
static void CopyToTextLikeOutFunc(CopyToState cstate, Oid atttypid, FmgrInfo *finfo); |
| 123 |
static void CopyToTextOneRow(CopyToState cstate, TupleTableSlot *slot); |
| 124 |
static void CopyToCSVOneRow(CopyToState cstate, TupleTableSlot *slot); |
| 125 |
static void CopyToTextLikeOneRow(CopyToState cstate, TupleTableSlot *slot, |
| 126 |
bool is_csv); |
| 127 |
static void CopyToTextLikeEnd(CopyToState cstate); |
| 128 |
static void CopyToBinaryStart(CopyToState cstate, TupleDesc tupDesc); |
| 129 |
static void CopyToBinaryOutFunc(CopyToState cstate, Oid atttypid, FmgrInfo *finfo); |
| 130 |
static void CopyToBinaryOneRow(CopyToState cstate, TupleTableSlot *slot); |
| 131 |
static void CopyToBinaryEnd(CopyToState cstate); |
| 132 |
|
| 133 |
/* Low-level communications functions */ |
| 134 |
static void SendCopyBegin(CopyToState cstate); |
| 135 |
static void SendCopyEnd(CopyToState cstate); |
| 136 |
static void CopySendData(CopyToState cstate, const void *databuf, int datasize); |
| 137 |
static void CopySendString(CopyToState cstate, const char *str); |
| 138 |
static void CopySendChar(CopyToState cstate, char c); |
| 139 |
static void CopySendEndOfRow(CopyToState cstate); |
| 140 |
static void CopySendTextLikeEndOfRow(CopyToState cstate); |
| 141 |
static void CopySendInt32(CopyToState cstate, int32 val); |
| 142 |
static void CopySendInt16(CopyToState cstate, int16 val); |
| 143 |
|
| 144 |
/* |
| 145 |
* COPY TO routines for built-in formats. |
| 146 |
* |
| 147 |
* CSV and text formats share the same TextLike routines except for the |
| 148 |
* one-row callback. |
| 149 |
*/ |
| 150 |
|
| 151 |
/* text format */ |
| 152 |
static const CopyToRoutine CopyToRoutineText = { |
| 153 |
.CopyToStart = CopyToTextLikeStart, |
| 154 |
.CopyToOutFunc = CopyToTextLikeOutFunc, |
| 155 |
.CopyToOneRow = CopyToTextOneRow, |
| 156 |
.CopyToEnd = CopyToTextLikeEnd, |
| 157 |
}; |
| 158 |
|
| 159 |
/* CSV format */ |
| 160 |
static const CopyToRoutine CopyToRoutineCSV = { |
| 161 |
.CopyToStart = CopyToTextLikeStart, |
| 162 |
.CopyToOutFunc = CopyToTextLikeOutFunc, |
| 163 |
.CopyToOneRow = CopyToCSVOneRow, |
| 164 |
.CopyToEnd = CopyToTextLikeEnd, |
| 165 |
}; |
| 166 |
|
| 167 |
/* binary format */ |
| 168 |
static const CopyToRoutine CopyToRoutineBinary = { |
| 169 |
.CopyToStart = CopyToBinaryStart, |
| 170 |
.CopyToOutFunc = CopyToBinaryOutFunc, |
| 171 |
.CopyToOneRow = CopyToBinaryOneRow, |
| 172 |
.CopyToEnd = CopyToBinaryEnd, |
| 173 |
}; |
| 174 |
|
| 175 |
/* Return a COPY TO routine for the given options */ |
| 176 |
static const CopyToRoutine * |
| 177 |
CopyToGetRoutine(const CopyFormatOptions *opts) |
| 178 |
{ |
| 179 |
if (opts->csv_mode) |
| 180 |
return &CopyToRoutineCSV; |
| 181 |
else if (opts->binary) |
| 182 |
return &CopyToRoutineBinary; |
| 183 |
|
| 184 |
/* default is text */ |
| 185 |
return &CopyToRoutineText; |
| 186 |
} |
| 187 |
|
| 188 |
/* Implementation of the start callback for text and CSV formats */ |
| 189 |
static void |
| 190 |
CopyToTextLikeStart(CopyToState cstate, TupleDesc tupDesc) |
| 191 |
{ |
| 192 |
/* |
| 193 |
* For non-binary copy, we need to convert null_print to file encoding, |
| 194 |
* because it will be sent directly with CopySendString. |
| 195 |
*/ |
| 196 |
if (cstate->need_transcoding) |
| 197 |
cstate->opts.null_print_client = pg_server_to_any(cstate->opts.null_print, |
| 198 |
cstate->opts.null_print_len, |
| 199 |
cstate->file_encoding); |
| 200 |
|
| 201 |
/* if a header has been requested send the line */ |
| 202 |
if (cstate->opts.header_line) |
| 203 |
{ |
| 204 |
ListCell *cur; |
| 205 |
bool hdr_delim = false; |
| 206 |
|
| 207 |
foreach(cur, cstate->attnumlist) |
| 208 |
{ |
| 209 |
int attnum = lfirst_int(cur); |
| 210 |
char *colname; |
| 211 |
|
| 212 |
if (hdr_delim) |
| 213 |
CopySendChar(cstate, cstate->opts.delim[0]); |
| 214 |
hdr_delim = true; |
| 215 |
|
| 216 |
colname = NameStr(TupleDescAttr(tupDesc, attnum - 1)->attname); |
| 217 |
|
| 218 |
if (cstate->opts.csv_mode) |
| 219 |
CopyAttributeOutCSV(cstate, colname, false); |
| 220 |
else |
| 221 |
CopyAttributeOutText(cstate, colname); |
| 222 |
} |
| 223 |
|
| 224 |
CopySendTextLikeEndOfRow(cstate); |
| 225 |
} |
| 226 |
} |
| 227 |
|
| 228 |
/* |
| 229 |
* Implementation of the outfunc callback for text and CSV formats. Assign |
| 230 |
* the output function data to the given *finfo. |
| 231 |
*/ |
| 232 |
static void |
| 233 |
CopyToTextLikeOutFunc(CopyToState cstate, Oid atttypid, FmgrInfo *finfo) |
| 234 |
{ |
| 235 |
Oid func_oid; |
| 236 |
bool is_varlena; |
| 237 |
|
| 238 |
/* Set output function for an attribute */ |
| 239 |
getTypeOutputInfo(atttypid, &func_oid, &is_varlena); |
| 240 |
fmgr_info(func_oid, finfo); |
| 241 |
} |
| 242 |
|
| 243 |
/* Implementation of the per-row callback for text format */ |
| 244 |
static void |
| 245 |
CopyToTextOneRow(CopyToState cstate, TupleTableSlot *slot) |
| 246 |
{ |
| 247 |
CopyToTextLikeOneRow(cstate, slot, false); |
| 248 |
} |
| 249 |
|
| 250 |
/* Implementation of the per-row callback for CSV format */ |
| 251 |
static void |
| 252 |
CopyToCSVOneRow(CopyToState cstate, TupleTableSlot *slot) |
| 253 |
{ |
| 254 |
CopyToTextLikeOneRow(cstate, slot, true); |
| 255 |
} |
| 256 |
|
| 257 |
/* |
| 258 |
* Workhorse for CopyToTextOneRow() and CopyToCSVOneRow(). |
| 259 |
* |
| 260 |
* We use pg_always_inline to reduce function call overhead |
| 261 |
* and to help compilers to optimize away the 'is_csv' condition. |
| 262 |
*/ |
| 263 |
static pg_always_inline void |
| 264 |
CopyToTextLikeOneRow(CopyToState cstate, |
| 265 |
TupleTableSlot *slot, |
| 266 |
bool is_csv) |
| 267 |
{ |
| 268 |
bool need_delim = false; |
| 269 |
FmgrInfo *out_functions = cstate->out_functions; |
| 270 |
|
| 271 |
foreach_int(attnum, cstate->attnumlist) |
| 272 |
{ |
| 273 |
Datum value = slot->tts_values[attnum - 1]; |
| 274 |
bool isnull = slot->tts_isnull[attnum - 1]; |
| 275 |
|
| 276 |
if (need_delim) |
| 277 |
CopySendChar(cstate, cstate->opts.delim[0]); |
| 278 |
need_delim = true; |
| 279 |
|
| 280 |
if (isnull) |
| 281 |
{ |
| 282 |
CopySendString(cstate, cstate->opts.null_print_client); |
| 283 |
} |
| 284 |
else |
| 285 |
{ |
| 286 |
char *string; |
| 287 |
|
| 288 |
string = OutputFunctionCall(&out_functions[attnum - 1], |
| 289 |
value); |
| 290 |
|
| 291 |
if (is_csv) |
| 292 |
CopyAttributeOutCSV(cstate, string, |
| 293 |
cstate->opts.force_quote_flags[attnum - 1]); |
| 294 |
else |
| 295 |
CopyAttributeOutText(cstate, string); |
| 296 |
} |
| 297 |
} |
| 298 |
|
| 299 |
CopySendTextLikeEndOfRow(cstate); |
| 300 |
} |
| 301 |
|
| 302 |
/* Implementation of the end callback for text and CSV formats */ |
| 303 |
static void |
| 304 |
CopyToTextLikeEnd(CopyToState cstate) |
| 305 |
{ |
| 306 |
/* Nothing to do here */ |
| 307 |
} |
| 308 |
|
| 309 |
/* |
| 310 |
* Implementation of the start callback for binary format. Send a header |
| 311 |
* for a binary copy. |
| 312 |
*/ |
| 313 |
static void |
| 314 |
CopyToBinaryStart(CopyToState cstate, TupleDesc tupDesc) |
| 315 |
{ |
| 316 |
int32 tmp; |
| 317 |
|
| 318 |
/* Signature */ |
| 319 |
CopySendData(cstate, BinarySignature, 11); |
| 320 |
/* Flags field */ |
| 321 |
tmp = 0; |
| 322 |
CopySendInt32(cstate, tmp); |
| 323 |
/* No header extension */ |
| 324 |
tmp = 0; |
| 325 |
CopySendInt32(cstate, tmp); |
| 326 |
} |
| 327 |
|
| 328 |
/* |
| 329 |
* Implementation of the outfunc callback for binary format. Assign |
| 330 |
* the binary output function to the given *finfo. |
| 331 |
*/ |
| 332 |
static void |
| 333 |
CopyToBinaryOutFunc(CopyToState cstate, Oid atttypid, FmgrInfo *finfo) |
| 334 |
{ |
| 335 |
Oid func_oid; |
| 336 |
bool is_varlena; |
| 337 |
|
| 338 |
/* Set output function for an attribute */ |
| 339 |
getTypeBinaryOutputInfo(atttypid, &func_oid, &is_varlena); |
| 340 |
fmgr_info(func_oid, finfo); |
| 341 |
} |
| 342 |
|
| 343 |
/* Implementation of the per-row callback for binary format */ |
| 344 |
static void |
| 345 |
CopyToBinaryOneRow(CopyToState cstate, TupleTableSlot *slot) |
| 346 |
{ |
| 347 |
FmgrInfo *out_functions = cstate->out_functions; |
| 348 |
|
| 349 |
/* Binary per-tuple header */ |
| 350 |
CopySendInt16(cstate, list_length(cstate->attnumlist)); |
| 351 |
|
| 352 |
foreach_int(attnum, cstate->attnumlist) |
| 353 |
{ |
| 354 |
Datum value = slot->tts_values[attnum - 1]; |
| 355 |
bool isnull = slot->tts_isnull[attnum - 1]; |
| 356 |
|
| 357 |
if (isnull) |
| 358 |
{ |
| 359 |
CopySendInt32(cstate, -1); |
| 360 |
} |
| 361 |
else |
| 362 |
{ |
| 363 |
bytea *outputbytes; |
| 364 |
|
| 365 |
outputbytes = SendFunctionCall(&out_functions[attnum - 1], |
| 366 |
value); |
| 367 |
CopySendInt32(cstate, VARSIZE(outputbytes) - VARHDRSZ); |
| 368 |
CopySendData(cstate, VARDATA(outputbytes), |
| 369 |
VARSIZE(outputbytes) - VARHDRSZ); |
| 370 |
} |
| 371 |
} |
| 372 |
|
| 373 |
CopySendEndOfRow(cstate); |
| 374 |
} |
| 375 |
|
| 376 |
/* Implementation of the end callback for binary format */ |
| 377 |
static void |
| 378 |
CopyToBinaryEnd(CopyToState cstate) |
| 379 |
{ |
| 380 |
/* Generate trailer for a binary copy */ |
| 381 |
CopySendInt16(cstate, -1); |
| 382 |
/* Need to flush out the trailer */ |
| 383 |
CopySendEndOfRow(cstate); |
| 384 |
} |
| 385 |
|
| 386 |
/* |
| 387 |
* Send copy start/stop messages for frontend copies. These have changed |
| 388 |
* in past protocol redesigns. |
| 389 |
*/ |
| 390 |
static void |
| 391 |
SendCopyBegin(CopyToState cstate) |
| 392 |
{ |
| 393 |
StringInfoData buf; |
| 394 |
int natts = list_length(cstate->attnumlist); |
| 395 |
int16 format = (cstate->opts.binary ? 1 : 0); |
| 396 |
int i; |
| 397 |
|
| 398 |
pq_beginmessage(&buf, PqMsg_CopyOutResponse); |
| 399 |
pq_sendbyte(&buf, format); /* overall format */ |
| 400 |
pq_sendint16(&buf, natts); |
| 401 |
for (i = 0; i < natts; i++) |
| 402 |
pq_sendint16(&buf, format); /* per-column formats */ |
| 403 |
pq_endmessage(&buf); |
| 404 |
cstate->copy_dest = COPY_FRONTEND; |
| 405 |
} |
| 406 |
|
| 407 |
static void |
| 408 |
SendCopyEnd(CopyToState cstate) |
| 409 |
{ |
| 410 |
/* Shouldn't have any unsent data */ |
| 411 |
Assert(cstate->fe_msgbuf->len == 0); |
| 412 |
/* Send Copy Done message */ |
| 413 |
pq_putemptymessage(PqMsg_CopyDone); |
| 414 |
} |
| 415 |
|
| 416 |
/*---------- |
| 417 |
* CopySendData sends output data to the destination (file or frontend) |
| 418 |
* CopySendString does the same for null-terminated strings |
| 419 |
* CopySendChar does the same for single characters |
| 420 |
* CopySendEndOfRow does the appropriate thing at end of each data row |
| 421 |
* (data is not actually flushed except by CopySendEndOfRow) |
| 422 |
* |
| 423 |
* NB: no data conversion is applied by these functions |
| 424 |
*---------- |
| 425 |
*/ |
| 426 |
static void |
| 427 |
CopySendData(CopyToState cstate, const void *databuf, int datasize) |
| 428 |
{ |
| 429 |
appendBinaryStringInfo(cstate->fe_msgbuf, databuf, datasize); |
| 430 |
} |
| 431 |
|
| 432 |
static void |
| 433 |
CopySendString(CopyToState cstate, const char *str) |
| 434 |
{ |
| 435 |
appendBinaryStringInfo(cstate->fe_msgbuf, str, strlen(str)); |
| 436 |
} |
| 437 |
|
| 438 |
static void |
| 439 |
CopySendChar(CopyToState cstate, char c) |
| 440 |
{ |
| 441 |
appendStringInfoCharMacro(cstate->fe_msgbuf, c); |
| 442 |
} |
| 443 |
|
| 444 |
static void |
| 445 |
CopySendEndOfRow(CopyToState cstate) |
| 446 |
{ |
| 447 |
StringInfo fe_msgbuf = cstate->fe_msgbuf; |
| 448 |
|
| 449 |
switch (cstate->copy_dest) |
| 450 |
{ |
| 451 |
case COPY_FILE: |
| 452 |
if (fwrite(fe_msgbuf->data, fe_msgbuf->len, 1, |
| 453 |
cstate->copy_file) != 1 || |
| 454 |
ferror(cstate->copy_file)) |
| 455 |
{ |
| 456 |
if (cstate->is_program) |
| 457 |
{ |
| 458 |
if (errno == EPIPE) |
| 459 |
{ |
| 460 |
/* |
| 461 |
* The pipe will be closed automatically on error at |
| 462 |
* the end of transaction, but we might get a better |
| 463 |
* error message from the subprocess' exit code than |
| 464 |
* just "Broken Pipe" |
| 465 |
*/ |
| 466 |
ClosePipeToProgram(cstate); |
| 467 |
|
| 468 |
/* |
| 469 |
* If ClosePipeToProgram() didn't throw an error, the |
| 470 |
* program terminated normally, but closed the pipe |
| 471 |
* first. Restore errno, and throw an error. |
| 472 |
*/ |
| 473 |
errno = EPIPE; |
| 474 |
} |
| 475 |
ereport(ERROR, |
| 476 |
(errcode_for_file_access(), |
| 477 |
errmsg("could not write to COPY program: %m"))); |
| 478 |
} |
| 479 |
else |
| 480 |
ereport(ERROR, |
| 481 |
(errcode_for_file_access(), |
| 482 |
errmsg("could not write to COPY file: %m"))); |
| 483 |
} |
| 484 |
break; |
| 485 |
case COPY_FRONTEND: |
| 486 |
/* Dump the accumulated row as one CopyData message */ |
| 487 |
(void) pq_putmessage(PqMsg_CopyData, fe_msgbuf->data, fe_msgbuf->len); |
| 488 |
break; |
| 489 |
case COPY_CALLBACK: |
| 490 |
cstate->data_dest_cb(fe_msgbuf->data, fe_msgbuf->len); |
| 491 |
break; |
| 492 |
} |
| 493 |
|
| 494 |
/* Update the progress */ |
| 495 |
cstate->bytes_processed += fe_msgbuf->len; |
| 496 |
pgstat_progress_update_param(PROGRESS_COPY_BYTES_PROCESSED, cstate->bytes_processed); |
| 497 |
|
| 498 |
resetStringInfo(fe_msgbuf); |
| 499 |
} |
| 500 |
|
| 501 |
/* |
| 502 |
* Wrapper function of CopySendEndOfRow for text and CSV formats. Sends the |
| 503 |
* line termination and do common appropriate things for the end of row. |
| 504 |
*/ |
| 505 |
static inline void |
| 506 |
CopySendTextLikeEndOfRow(CopyToState cstate) |
| 507 |
{ |
| 508 |
switch (cstate->copy_dest) |
| 509 |
{ |
| 510 |
case COPY_FILE: |
| 511 |
/* Default line termination depends on platform */ |
| 512 |
#ifndef WIN32 |
| 513 |
CopySendChar(cstate, '\n'); |
| 514 |
#else |
| 515 |
CopySendString(cstate, "\r\n"); |
| 516 |
#endif |
| 517 |
break; |
| 518 |
case COPY_FRONTEND: |
| 519 |
/* The FE/BE protocol uses \n as newline for all platforms */ |
| 520 |
CopySendChar(cstate, '\n'); |
| 521 |
break; |
| 522 |
default: |
| 523 |
break; |
| 524 |
} |
| 525 |
|
| 526 |
/* Now take the actions related to the end of a row */ |
| 527 |
CopySendEndOfRow(cstate); |
| 528 |
} |
| 529 |
|
| 530 |
/* |
| 531 |
* These functions do apply some data conversion |
| 532 |
*/ |
| 533 |
|
| 534 |
/* |
| 535 |
* CopySendInt32 sends an int32 in network byte order |
| 536 |
*/ |
| 537 |
static inline void |
| 538 |
CopySendInt32(CopyToState cstate, int32 val) |
| 539 |
{ |
| 540 |
uint32 buf; |
| 541 |
|
| 542 |
buf = pg_hton32((uint32) val); |
| 543 |
CopySendData(cstate, &buf, sizeof(buf)); |
| 544 |
} |
| 545 |
|
| 546 |
/* |
| 547 |
* CopySendInt16 sends an int16 in network byte order |
| 548 |
*/ |
| 549 |
static inline void |
| 550 |
CopySendInt16(CopyToState cstate, int16 val) |
| 551 |
{ |
| 552 |
uint16 buf; |
| 553 |
|
| 554 |
buf = pg_hton16((uint16) val); |
| 555 |
CopySendData(cstate, &buf, sizeof(buf)); |
| 556 |
} |
| 557 |
|
| 558 |
/* |
| 559 |
* Closes the pipe to an external program, checking the pclose() return code. |
| 560 |
*/ |
| 561 |
static void |
| 562 |
ClosePipeToProgram(CopyToState cstate) |
| 563 |
{ |
| 564 |
int pclose_rc; |
| 565 |
|
| 566 |
Assert(cstate->is_program); |
| 567 |
|
| 568 |
pclose_rc = ClosePipeStream(cstate->copy_file); |
| 569 |
if (pclose_rc == -1) |
| 570 |
ereport(ERROR, |
| 571 |
(errcode_for_file_access(), |
| 572 |
errmsg("could not close pipe to external command: %m"))); |
| 573 |
else if (pclose_rc != 0) |
| 574 |
{ |
| 575 |
ereport(ERROR, |
| 576 |
(errcode(ERRCODE_EXTERNAL_ROUTINE_EXCEPTION), |
| 577 |
errmsg("program \"%s\" failed", |
| 578 |
cstate->filename), |
| 579 |
errdetail_internal("%s", wait_result_to_str(pclose_rc)))); |
| 580 |
} |
| 581 |
} |
| 582 |
|
| 583 |
/* |
| 584 |
* Release resources allocated in a cstate for COPY TO/FROM. |
| 585 |
*/ |
| 586 |
static void |
| 587 |
EndCopy(CopyToState cstate) |
| 588 |
{ |
| 589 |
if (cstate->is_program) |
| 590 |
{ |
| 591 |
ClosePipeToProgram(cstate); |
| 592 |
} |
| 593 |
else |
| 594 |
{ |
| 595 |
if (cstate->filename != NULL && FreeFile(cstate->copy_file)) |
| 596 |
ereport(ERROR, |
| 597 |
(errcode_for_file_access(), |
| 598 |
errmsg("could not close file \"%s\": %m", |
| 599 |
cstate->filename))); |
| 600 |
} |
| 601 |
|
| 602 |
pgstat_progress_end_command(); |
| 603 |
|
| 604 |
MemoryContextDelete(cstate->copycontext); |
| 605 |
pfree(cstate); |
| 606 |
} |
| 607 |
|
| 608 |
/* |
| 609 |
* Setup CopyToState to read tuples from a table or a query for COPY TO. |
| 610 |
* |
| 611 |
* 'rel': Relation to be copied |
| 612 |
* 'raw_query': Query whose results are to be copied |
| 613 |
* 'queryRelId': OID of base relation to convert to a query (for RLS) |
| 614 |
* 'filename': Name of server-local file to write, NULL for STDOUT |
| 615 |
* 'is_program': true if 'filename' is program to execute |
| 616 |
* 'data_dest_cb': Callback that processes the output data |
| 617 |
* 'attnamelist': List of char *, columns to include. NIL selects all cols. |
| 618 |
* 'options': List of DefElem. See copy_opt_item in gram.y for selections. |
| 619 |
* |
| 620 |
* Returns a CopyToState, to be passed to DoCopyTo() and related functions. |
| 621 |
*/ |
| 622 |
CopyToState |
| 623 |
BeginCopyTo(ParseState *pstate, |
| 624 |
Relation rel, |
| 625 |
RawStmt *raw_query, |
| 626 |
Oid queryRelId, |
| 627 |
const char *filename, |
| 628 |
bool is_program, |
| 629 |
copy_data_dest_cb data_dest_cb, |
| 630 |
List *attnamelist, |
| 631 |
List *options) |
| 632 |
{ |
| 633 |
CopyToState cstate; |
| 634 |
bool pipe = (filename == NULL && data_dest_cb == NULL); |
| 635 |
TupleDesc tupDesc; |
| 636 |
int num_phys_attrs; |
| 637 |
MemoryContext oldcontext; |
| 638 |
const int progress_cols[] = { |
| 639 |
PROGRESS_COPY_COMMAND, |
| 640 |
PROGRESS_COPY_TYPE |
| 641 |
}; |
| 642 |
int64 progress_vals[] = { |
| 643 |
PROGRESS_COPY_COMMAND_TO, |
| 644 |
0 |
| 645 |
}; |
| 646 |
|
| 647 |
if (rel != NULL && rel->rd_rel->relkind != RELKIND_RELATION) |
| 648 |
{ |
| 649 |
if (rel->rd_rel->relkind == RELKIND_VIEW) |
| 650 |
ereport(ERROR, |
| 651 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 652 |
errmsg("cannot copy from view \"%s\"", |
| 653 |
RelationGetRelationName(rel)), |
| 654 |
errhint("Try the COPY (SELECT ...) TO variant."))); |
| 655 |
else if (rel->rd_rel->relkind == RELKIND_MATVIEW) |
| 656 |
{ |
| 657 |
if (!RelationIsPopulated(rel)) |
| 658 |
ereport(ERROR, |
| 659 |
errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 660 |
errmsg("cannot copy from unpopulated materialized view \"%s\"", |
| 661 |
RelationGetRelationName(rel)), |
| 662 |
errhint("Use the REFRESH MATERIALIZED VIEW command.")); |
| 663 |
} |
| 664 |
else if (rel->rd_rel->relkind == RELKIND_FOREIGN_TABLE) |
| 665 |
ereport(ERROR, |
| 666 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 667 |
errmsg("cannot copy from foreign table \"%s\"", |
| 668 |
RelationGetRelationName(rel)), |
| 669 |
errhint("Try the COPY (SELECT ...) TO variant."))); |
| 670 |
else if (rel->rd_rel->relkind == RELKIND_SEQUENCE) |
| 671 |
ereport(ERROR, |
| 672 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 673 |
errmsg("cannot copy from sequence \"%s\"", |
| 674 |
RelationGetRelationName(rel)))); |
| 675 |
else if (rel->rd_rel->relkind == RELKIND_PARTITIONED_TABLE) |
| 676 |
ereport(ERROR, |
| 677 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 678 |
errmsg("cannot copy from partitioned table \"%s\"", |
| 679 |
RelationGetRelationName(rel)), |
| 680 |
errhint("Try the COPY (SELECT ...) TO variant."))); |
| 681 |
else |
| 682 |
ereport(ERROR, |
| 683 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 684 |
errmsg("cannot copy from non-table relation \"%s\"", |
| 685 |
RelationGetRelationName(rel)))); |
| 686 |
} |
| 687 |
|
| 688 |
|
| 689 |
/* Allocate workspace and zero all fields */ |
| 690 |
cstate = (CopyToStateData *) palloc0(sizeof(CopyToStateData)); |
| 691 |
|
| 692 |
/* |
| 693 |
* We allocate everything used by a cstate in a new memory context. This |
| 694 |
* avoids memory leaks during repeated use of COPY in a query. |
| 695 |
*/ |
| 696 |
cstate->copycontext = AllocSetContextCreate(CurrentMemoryContext, |
| 697 |
"COPY", |
| 698 |
ALLOCSET_DEFAULT_SIZES); |
| 699 |
|
| 700 |
oldcontext = MemoryContextSwitchTo(cstate->copycontext); |
| 701 |
|
| 702 |
/* Extract options from the statement node tree */ |
| 703 |
ProcessCopyOptions(pstate, &cstate->opts, false /* is_from */ , options); |
| 704 |
|
| 705 |
/* Set format routine */ |
| 706 |
cstate->routine = CopyToGetRoutine(&cstate->opts); |
| 707 |
|
| 708 |
/* Process the source/target relation or query */ |
| 709 |
if (rel) |
| 710 |
{ |
| 711 |
Assert(!raw_query); |
| 712 |
|
| 713 |
cstate->rel = rel; |
| 714 |
|
| 715 |
tupDesc = RelationGetDescr(cstate->rel); |
| 716 |
} |
| 717 |
else |
| 718 |
{ |
| 719 |
List *rewritten; |
| 720 |
Query *query; |
| 721 |
PlannedStmt *plan; |
| 722 |
DestReceiver *dest; |
| 723 |
|
| 724 |
cstate->rel = NULL; |
| 725 |
|
| 726 |
/* |
| 727 |
* Run parse analysis and rewrite. Note this also acquires sufficient |
| 728 |
* locks on the source table(s). |
| 729 |
*/ |
| 730 |
rewritten = pg_analyze_and_rewrite_fixedparams(raw_query, |
| 731 |
pstate->p_sourcetext, NULL, 0, |
| 732 |
NULL); |
| 733 |
|
| 734 |
/* check that we got back something we can work with */ |
| 735 |
if (rewritten == NIL) |
| 736 |
{ |
| 737 |
ereport(ERROR, |
| 738 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 739 |
errmsg("DO INSTEAD NOTHING rules are not supported for COPY"))); |
| 740 |
} |
| 741 |
else if (list_length(rewritten) > 1) |
| 742 |
{ |
| 743 |
ListCell *lc; |
| 744 |
|
| 745 |
/* examine queries to determine which error message to issue */ |
| 746 |
foreach(lc, rewritten) |
| 747 |
{ |
| 748 |
Query *q = lfirst_node(Query, lc); |
| 749 |
|
| 750 |
if (q->querySource == QSRC_QUAL_INSTEAD_RULE) |
| 751 |
ereport(ERROR, |
| 752 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 753 |
errmsg("conditional DO INSTEAD rules are not supported for COPY"))); |
| 754 |
if (q->querySource == QSRC_NON_INSTEAD_RULE) |
| 755 |
ereport(ERROR, |
| 756 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 757 |
errmsg("DO ALSO rules are not supported for COPY"))); |
| 758 |
} |
| 759 |
|
| 760 |
ereport(ERROR, |
| 761 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 762 |
errmsg("multi-statement DO INSTEAD rules are not supported for COPY"))); |
| 763 |
} |
| 764 |
|
| 765 |
query = linitial_node(Query, rewritten); |
| 766 |
|
| 767 |
/* The grammar allows SELECT INTO, but we don't support that */ |
| 768 |
if (query->utilityStmt != NULL && |
| 769 |
IsA(query->utilityStmt, CreateTableAsStmt)) |
| 770 |
ereport(ERROR, |
| 771 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 772 |
errmsg("COPY (SELECT INTO) is not supported"))); |
| 773 |
|
| 774 |
/* The only other utility command we could see is NOTIFY */ |
| 775 |
if (query->utilityStmt != NULL) |
| 776 |
ereport(ERROR, |
| 777 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 778 |
errmsg("COPY query must not be a utility command"))); |
| 779 |
|
| 780 |
/* |
| 781 |
* Similarly the grammar doesn't enforce the presence of a RETURNING |
| 782 |
* clause, but this is required here. |
| 783 |
*/ |
| 784 |
if (query->commandType != CMD_SELECT && |
| 785 |
query->returningList == NIL) |
| 786 |
{ |
| 787 |
Assert(query->commandType == CMD_INSERT || |
| 788 |
query->commandType == CMD_UPDATE || |
| 789 |
query->commandType == CMD_DELETE || |
| 790 |
query->commandType == CMD_MERGE); |
| 791 |
|
| 792 |
ereport(ERROR, |
| 793 |
(errcode(ERRCODE_FEATURE_NOT_SUPPORTED), |
| 794 |
errmsg("COPY query must have a RETURNING clause"))); |
| 795 |
} |
| 796 |
|
| 797 |
/* plan the query */ |
| 798 |
plan = pg_plan_query(query, pstate->p_sourcetext, |
| 799 |
CURSOR_OPT_PARALLEL_OK, NULL); |
| 800 |
|
| 801 |
/* |
| 802 |
* With row-level security and a user using "COPY relation TO", we |
| 803 |
* have to convert the "COPY relation TO" to a query-based COPY (eg: |
| 804 |
* "COPY (SELECT * FROM ONLY relation) TO"), to allow the rewriter to |
| 805 |
* add in any RLS clauses. |
| 806 |
* |
| 807 |
* When this happens, we are passed in the relid of the originally |
| 808 |
* found relation (which we have locked). As the planner will look up |
| 809 |
* the relation again, we double-check here to make sure it found the |
| 810 |
* same one that we have locked. |
| 811 |
*/ |
| 812 |
if (queryRelId != InvalidOid) |
| 813 |
{ |
| 814 |
/* |
| 815 |
* Note that with RLS involved there may be multiple relations, |
| 816 |
* and while the one we need is almost certainly first, we don't |
| 817 |
* make any guarantees of that in the planner, so check the whole |
| 818 |
* list and make sure we find the original relation. |
| 819 |
*/ |
| 820 |
if (!list_member_oid(plan->relationOids, queryRelId)) |
| 821 |
ereport(ERROR, |
| 822 |
(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE), |
| 823 |
errmsg("relation referenced by COPY statement has changed"))); |
| 824 |
} |
| 825 |
|
| 826 |
/* |
| 827 |
* Use a snapshot with an updated command ID to ensure this query sees |
| 828 |
* results of any previously executed queries. |
| 829 |
*/ |
| 830 |
PushCopiedSnapshot(GetActiveSnapshot()); |
| 831 |
UpdateActiveSnapshotCommandId(); |
| 832 |
|
| 833 |
/* Create dest receiver for COPY OUT */ |
| 834 |
dest = CreateDestReceiver(DestCopyOut); |
| 835 |
((DR_copy *) dest)->cstate = cstate; |
| 836 |
|
| 837 |
/* Create a QueryDesc requesting no output */ |
| 838 |
cstate->queryDesc = CreateQueryDesc(plan, pstate->p_sourcetext, |
| 839 |
GetActiveSnapshot(), |
| 840 |
InvalidSnapshot, |
| 841 |
dest, NULL, NULL, 0); |
| 842 |
|
| 843 |
/* |
| 844 |
* Call ExecutorStart to prepare the plan for execution. |
| 845 |
* |
| 846 |
* ExecutorStart computes a result tupdesc for us |
| 847 |
*/ |
| 848 |
ExecutorStart(cstate->queryDesc, 0); |
| 849 |
|
| 850 |
tupDesc = cstate->queryDesc->tupDesc; |
| 851 |
} |
| 852 |
|
| 853 |
/* Generate or convert list of attributes to process */ |
| 854 |
cstate->attnumlist = CopyGetAttnums(tupDesc, cstate->rel, attnamelist); |
| 855 |
|
| 856 |
num_phys_attrs = tupDesc->natts; |
| 857 |
|
| 858 |
/* Convert FORCE_QUOTE name list to per-column flags, check validity */ |
| 859 |
cstate->opts.force_quote_flags = (bool *) palloc0(num_phys_attrs * sizeof(bool)); |
| 860 |
if (cstate->opts.force_quote_all) |
| 861 |
{ |
| 862 |
MemSet(cstate->opts.force_quote_flags, true, num_phys_attrs * sizeof(bool)); |
| 863 |
} |
| 864 |
else if (cstate->opts.force_quote) |
| 865 |
{ |
| 866 |
List *attnums; |
| 867 |
ListCell *cur; |
| 868 |
|
| 869 |
attnums = CopyGetAttnums(tupDesc, cstate->rel, cstate->opts.force_quote); |
| 870 |
|
| 871 |
foreach(cur, attnums) |
| 872 |
{ |
| 873 |
int attnum = lfirst_int(cur); |
| 874 |
Form_pg_attribute attr = TupleDescAttr(tupDesc, attnum - 1); |
| 875 |
|
| 876 |
if (!list_member_int(cstate->attnumlist, attnum)) |
| 877 |
ereport(ERROR, |
| 878 |
(errcode(ERRCODE_INVALID_COLUMN_REFERENCE), |
| 879 |
/*- translator: %s is the name of a COPY option, e.g. FORCE_NOT_NULL */ |
| 880 |
errmsg("%s column \"%s\" not referenced by COPY", |
| 881 |
"FORCE_QUOTE", NameStr(attr->attname)))); |
| 882 |
cstate->opts.force_quote_flags[attnum - 1] = true; |
| 883 |
} |
| 884 |
} |
| 885 |
|
| 886 |
/* Use client encoding when ENCODING option is not specified. */ |
| 887 |
if (cstate->opts.file_encoding < 0) |
| 888 |
cstate->file_encoding = pg_get_client_encoding(); |
| 889 |
else |
| 890 |
cstate->file_encoding = cstate->opts.file_encoding; |
| 891 |
|
| 892 |
/* |
| 893 |
* Set up encoding conversion info if the file and server encodings differ |
| 894 |
* (see also pg_server_to_any). |
| 895 |
*/ |
| 896 |
if (cstate->file_encoding == GetDatabaseEncoding() || |
| 897 |
cstate->file_encoding == PG_SQL_ASCII) |
| 898 |
cstate->need_transcoding = false; |
| 899 |
else |
| 900 |
cstate->need_transcoding = true; |
| 901 |
|
| 902 |
/* See Multibyte encoding comment above */ |
| 903 |
cstate->encoding_embeds_ascii = PG_ENCODING_IS_CLIENT_ONLY(cstate->file_encoding); |
| 904 |
|
| 905 |
cstate->copy_dest = COPY_FILE; /* default */ |
| 906 |
|
| 907 |
if (data_dest_cb) |
| 908 |
{ |
| 909 |
progress_vals[1] = PROGRESS_COPY_TYPE_CALLBACK; |
| 910 |
cstate->copy_dest = COPY_CALLBACK; |
| 911 |
cstate->data_dest_cb = data_dest_cb; |
| 912 |
} |
| 913 |
else if (pipe) |
| 914 |
{ |
| 915 |
progress_vals[1] = PROGRESS_COPY_TYPE_PIPE; |
| 916 |
|
| 917 |
Assert(!is_program); /* the grammar does not allow this */ |
| 918 |
if (whereToSendOutput != DestRemote) |
| 919 |
cstate->copy_file = stdout; |
| 920 |
} |
| 921 |
else |
| 922 |
{ |
| 923 |
cstate->filename = pstrdup(filename); |
| 924 |
cstate->is_program = is_program; |
| 925 |
|
| 926 |
if (is_program) |
| 927 |
{ |
| 928 |
progress_vals[1] = PROGRESS_COPY_TYPE_PROGRAM; |
| 929 |
cstate->copy_file = OpenPipeStream(cstate->filename, PG_BINARY_W); |
| 930 |
if (cstate->copy_file == NULL) |
| 931 |
ereport(ERROR, |
| 932 |
(errcode_for_file_access(), |
| 933 |
errmsg("could not execute command \"%s\": %m", |
| 934 |
cstate->filename))); |
| 935 |
} |
| 936 |
else |
| 937 |
{ |
| 938 |
mode_t oumask; /* Pre-existing umask value */ |
| 939 |
struct stat st; |
| 940 |
|
| 941 |
progress_vals[1] = PROGRESS_COPY_TYPE_FILE; |
| 942 |
|
| 943 |
/* |
| 944 |
* Prevent write to relative path ... too easy to shoot oneself in |
| 945 |
* the foot by overwriting a database file ... |
| 946 |
*/ |
| 947 |
if (!is_absolute_path(filename)) |
| 948 |
ereport(ERROR, |
| 949 |
(errcode(ERRCODE_INVALID_NAME), |
| 950 |
errmsg("relative path not allowed for COPY to file"))); |
| 951 |
|
| 952 |
oumask = umask(S_IWGRP | S_IWOTH); |
| 953 |
PG_TRY(); |
| 954 |
{ |
| 955 |
cstate->copy_file = AllocateFile(cstate->filename, PG_BINARY_W); |
| 956 |
} |
| 957 |
PG_FINALLY(); |
| 958 |
{ |
| 959 |
umask(oumask); |
| 960 |
} |
| 961 |
PG_END_TRY(); |
| 962 |
if (cstate->copy_file == NULL) |
| 963 |
{ |
| 964 |
/* copy errno because ereport subfunctions might change it */ |
| 965 |
int save_errno = errno; |
| 966 |
|
| 967 |
ereport(ERROR, |
| 968 |
(errcode_for_file_access(), |
| 969 |
errmsg("could not open file \"%s\" for writing: %m", |
| 970 |
cstate->filename), |
| 971 |
(save_errno == ENOENT || save_errno == EACCES) ? |
| 972 |
errhint("COPY TO instructs the PostgreSQL server process to write a file. " |
| 973 |
"You may want a client-side facility such as psql's \\copy.") : 0)); |
| 974 |
} |
| 975 |
|
| 976 |
if (fstat(fileno(cstate->copy_file), &st)) |
| 977 |
ereport(ERROR, |
| 978 |
(errcode_for_file_access(), |
| 979 |
errmsg("could not stat file \"%s\": %m", |
| 980 |
cstate->filename))); |
| 981 |
|
| 982 |
if (S_ISDIR(st.st_mode)) |
| 983 |
ereport(ERROR, |
| 984 |
(errcode(ERRCODE_WRONG_OBJECT_TYPE), |
| 985 |
errmsg("\"%s\" is a directory", cstate->filename))); |
| 986 |
} |
| 987 |
} |
| 988 |
|
| 989 |
/* initialize progress */ |
| 990 |
pgstat_progress_start_command(PROGRESS_COMMAND_COPY, |
| 991 |
cstate->rel ? RelationGetRelid(cstate->rel) : InvalidOid); |
| 992 |
pgstat_progress_update_multi_param(2, progress_cols, progress_vals); |
| 993 |
|
| 994 |
cstate->bytes_processed = 0; |
| 995 |
|
| 996 |
MemoryContextSwitchTo(oldcontext); |
| 997 |
|
| 998 |
return cstate; |
| 999 |
} |
| 1000 |
|
| 1001 |
/* |
| 1002 |
* Clean up storage and release resources for COPY TO. |
| 1003 |
*/ |
| 1004 |
void |
| 1005 |
EndCopyTo(CopyToState cstate) |
| 1006 |
{ |
| 1007 |
if (cstate->queryDesc != NULL) |
| 1008 |
{ |
| 1009 |
/* Close down the query and free resources. */ |
| 1010 |
ExecutorFinish(cstate->queryDesc); |
| 1011 |
ExecutorEnd(cstate->queryDesc); |
| 1012 |
FreeQueryDesc(cstate->queryDesc); |
| 1013 |
PopActiveSnapshot(); |
| 1014 |
} |
| 1015 |
|
| 1016 |
/* Clean up storage */ |
| 1017 |
EndCopy(cstate); |
| 1018 |
} |
| 1019 |
|
| 1020 |
/* |
| 1021 |
* Copy from relation or query TO file. |
| 1022 |
* |
| 1023 |
* Returns the number of rows processed. |
| 1024 |
*/ |
| 1025 |
uint64 |
| 1026 |
DoCopyTo(CopyToState cstate) |
| 1027 |
{ |
| 1028 |
bool pipe = (cstate->filename == NULL && cstate->data_dest_cb == NULL); |
| 1029 |
bool fe_copy = (pipe && whereToSendOutput == DestRemote); |
| 1030 |
TupleDesc tupDesc; |
| 1031 |
int num_phys_attrs; |
| 1032 |
ListCell *cur; |
| 1033 |
uint64 processed; |
| 1034 |
|
| 1035 |
if (fe_copy) |
| 1036 |
SendCopyBegin(cstate); |
| 1037 |
|
| 1038 |
if (cstate->rel) |
| 1039 |
tupDesc = RelationGetDescr(cstate->rel); |
| 1040 |
else |
| 1041 |
tupDesc = cstate->queryDesc->tupDesc; |
| 1042 |
num_phys_attrs = tupDesc->natts; |
| 1043 |
cstate->opts.null_print_client = cstate->opts.null_print; /* default */ |
| 1044 |
|
| 1045 |
/* We use fe_msgbuf as a per-row buffer regardless of copy_dest */ |
| 1046 |
cstate->fe_msgbuf = makeStringInfo(); |
| 1047 |
|
| 1048 |
/* Get info about the columns we need to process. */ |
| 1049 |
cstate->out_functions = (FmgrInfo *) palloc(num_phys_attrs * sizeof(FmgrInfo)); |
| 1050 |
foreach(cur, cstate->attnumlist) |
| 1051 |
{ |
| 1052 |
int attnum = lfirst_int(cur); |
| 1053 |
Form_pg_attribute attr = TupleDescAttr(tupDesc, attnum - 1); |
| 1054 |
|
| 1055 |
cstate->routine->CopyToOutFunc(cstate, attr->atttypid, |
| 1056 |
&cstate->out_functions[attnum - 1]); |
| 1057 |
} |
| 1058 |
|
| 1059 |
/* |
| 1060 |
* Create a temporary memory context that we can reset once per row to |
| 1061 |
* recover palloc'd memory. This avoids any problems with leaks inside |
| 1062 |
* datatype output routines, and should be faster than retail pfree's |
| 1063 |
* anyway. (We don't need a whole econtext as CopyFrom does.) |
| 1064 |
*/ |
| 1065 |
cstate->rowcontext = AllocSetContextCreate(CurrentMemoryContext, |
| 1066 |
"COPY TO", |
| 1067 |
ALLOCSET_DEFAULT_SIZES); |
| 1068 |
|
| 1069 |
cstate->routine->CopyToStart(cstate, tupDesc); |
| 1070 |
|
| 1071 |
if (cstate->rel) |
| 1072 |
{ |
| 1073 |
TupleTableSlot *slot; |
| 1074 |
TableScanDesc scandesc; |
| 1075 |
|
| 1076 |
scandesc = table_beginscan(cstate->rel, GetActiveSnapshot(), 0, NULL); |
| 1077 |
slot = table_slot_create(cstate->rel, NULL); |
| 1078 |
|
| 1079 |
processed = 0; |
| 1080 |
while (table_scan_getnextslot(scandesc, ForwardScanDirection, slot)) |
| 1081 |
{ |
| 1082 |
CHECK_FOR_INTERRUPTS(); |
| 1083 |
|
| 1084 |
/* Deconstruct the tuple ... */ |
| 1085 |
slot_getallattrs(slot); |
| 1086 |
|
| 1087 |
/* Format and send the data */ |
| 1088 |
CopyOneRowTo(cstate, slot); |
| 1089 |
|
| 1090 |
/* |
| 1091 |
* Increment the number of processed tuples, and report the |
| 1092 |
* progress. |
| 1093 |
*/ |
| 1094 |
pgstat_progress_update_param(PROGRESS_COPY_TUPLES_PROCESSED, |
| 1095 |
++processed); |
| 1096 |
} |
| 1097 |
|
| 1098 |
ExecDropSingleTupleTableSlot(slot); |
| 1099 |
table_endscan(scandesc); |
| 1100 |
} |
| 1101 |
else |
| 1102 |
{ |
| 1103 |
/* run the plan --- the dest receiver will send tuples */ |
| 1104 |
ExecutorRun(cstate->queryDesc, ForwardScanDirection, 0); |
| 1105 |
processed = ((DR_copy *) cstate->queryDesc->dest)->processed; |
| 1106 |
} |
| 1107 |
|
| 1108 |
cstate->routine->CopyToEnd(cstate); |
| 1109 |
|
| 1110 |
MemoryContextDelete(cstate->rowcontext); |
| 1111 |
|
| 1112 |
if (fe_copy) |
| 1113 |
SendCopyEnd(cstate); |
| 1114 |
|
| 1115 |
return processed; |
| 1116 |
} |
| 1117 |
|
| 1118 |
/* |
| 1119 |
* Emit one row during DoCopyTo(). |
| 1120 |
*/ |
| 1121 |
static inline void |
| 1122 |
CopyOneRowTo(CopyToState cstate, TupleTableSlot *slot) |
| 1123 |
{ |
| 1124 |
MemoryContext oldcontext; |
| 1125 |
|
| 1126 |
MemoryContextReset(cstate->rowcontext); |
| 1127 |
oldcontext = MemoryContextSwitchTo(cstate->rowcontext); |
| 1128 |
|
| 1129 |
/* Make sure the tuple is fully deconstructed */ |
| 1130 |
slot_getallattrs(slot); |
| 1131 |
|
| 1132 |
cstate->routine->CopyToOneRow(cstate, slot); |
| 1133 |
|
| 1134 |
MemoryContextSwitchTo(oldcontext); |
| 1135 |
} |
| 1136 |
|
| 1137 |
/* |
| 1138 |
* Send text representation of one attribute, with conversion and escaping |
| 1139 |
*/ |
| 1140 |
#define DUMPSOFAR() \ |
| 1141 |
do { \ |
| 1142 |
if (ptr > start) \ |
| 1143 |
CopySendData(cstate, start, ptr - start); \ |
| 1144 |
} while (0) |
| 1145 |
|
| 1146 |
static void |
| 1147 |
CopyAttributeOutText(CopyToState cstate, const char *string) |
| 1148 |
{ |
| 1149 |
const char *ptr; |
| 1150 |
const char *start; |
| 1151 |
char c; |
| 1152 |
char delimc = cstate->opts.delim[0]; |
| 1153 |
|
| 1154 |
if (cstate->need_transcoding) |
| 1155 |
ptr = pg_server_to_any(string, strlen(string), cstate->file_encoding); |
| 1156 |
else |
| 1157 |
ptr = string; |
| 1158 |
|
| 1159 |
/* |
| 1160 |
* We have to grovel through the string searching for control characters |
| 1161 |
* and instances of the delimiter character. In most cases, though, these |
| 1162 |
* are infrequent. To avoid overhead from calling CopySendData once per |
| 1163 |
* character, we dump out all characters between escaped characters in a |
| 1164 |
* single call. The loop invariant is that the data from "start" to "ptr" |
| 1165 |
* can be sent literally, but hasn't yet been. |
| 1166 |
* |
| 1167 |
* We can skip pg_encoding_mblen() overhead when encoding is safe, because |
| 1168 |
* in valid backend encodings, extra bytes of a multibyte character never |
| 1169 |
* look like ASCII. This loop is sufficiently performance-critical that |
| 1170 |
* it's worth making two copies of it to get the IS_HIGHBIT_SET() test out |
| 1171 |
* of the normal safe-encoding path. |
| 1172 |
*/ |
| 1173 |
if (cstate->encoding_embeds_ascii) |
| 1174 |
{ |
| 1175 |
start = ptr; |
| 1176 |
while ((c = *ptr) != '\0') |
| 1177 |
{ |
| 1178 |
if ((unsigned char) c < (unsigned char) 0x20) |
| 1179 |
{ |
| 1180 |
/* |
| 1181 |
* \r and \n must be escaped, the others are traditional. We |
| 1182 |
* prefer to dump these using the C-like notation, rather than |
| 1183 |
* a backslash and the literal character, because it makes the |
| 1184 |
* dump file a bit more proof against Microsoftish data |
| 1185 |
* mangling. |
| 1186 |
*/ |
| 1187 |
switch (c) |
| 1188 |
{ |
| 1189 |
case '\b': |
| 1190 |
c = 'b'; |
| 1191 |
break; |
| 1192 |
case '\f': |
| 1193 |
c = 'f'; |
| 1194 |
break; |
| 1195 |
case '\n': |
| 1196 |
c = 'n'; |
| 1197 |
break; |
| 1198 |
case '\r': |
| 1199 |
c = 'r'; |
| 1200 |
break; |
| 1201 |
case '\t': |
| 1202 |
c = 't'; |
| 1203 |
break; |
| 1204 |
case '\v': |
| 1205 |
c = 'v'; |
| 1206 |
break; |
| 1207 |
default: |
| 1208 |
/* If it's the delimiter, must backslash it */ |
| 1209 |
if (c == delimc) |
| 1210 |
break; |
| 1211 |
/* All ASCII control chars are length 1 */ |
| 1212 |
ptr++; |
| 1213 |
continue; /* fall to end of loop */ |
| 1214 |
} |
| 1215 |
/* if we get here, we need to convert the control char */ |
| 1216 |
DUMPSOFAR(); |
| 1217 |
CopySendChar(cstate, '\\'); |
| 1218 |
CopySendChar(cstate, c); |
| 1219 |
start = ++ptr; /* do not include char in next run */ |
| 1220 |
} |
| 1221 |
else if (c == '\\' || c == delimc) |
| 1222 |
{ |
| 1223 |
DUMPSOFAR(); |
| 1224 |
CopySendChar(cstate, '\\'); |
| 1225 |
start = ptr++; /* we include char in next run */ |
| 1226 |
} |
| 1227 |
else if (IS_HIGHBIT_SET(c)) |
| 1228 |
ptr += pg_encoding_mblen(cstate->file_encoding, ptr); |
| 1229 |
else |
| 1230 |
ptr++; |
| 1231 |
} |
| 1232 |
} |
| 1233 |
else |
| 1234 |
{ |
| 1235 |
start = ptr; |
| 1236 |
while ((c = *ptr) != '\0') |
| 1237 |
{ |
| 1238 |
if ((unsigned char) c < (unsigned char) 0x20) |
| 1239 |
{ |
| 1240 |
/* |
| 1241 |
* \r and \n must be escaped, the others are traditional. We |
| 1242 |
* prefer to dump these using the C-like notation, rather than |
| 1243 |
* a backslash and the literal character, because it makes the |
| 1244 |
* dump file a bit more proof against Microsoftish data |
| 1245 |
* mangling. |
| 1246 |
*/ |
| 1247 |
switch (c) |
| 1248 |
{ |
| 1249 |
case '\b': |
| 1250 |
c = 'b'; |
| 1251 |
break; |
| 1252 |
case '\f': |
| 1253 |
c = 'f'; |
| 1254 |
break; |
| 1255 |
case '\n': |
| 1256 |
c = 'n'; |
| 1257 |
break; |
| 1258 |
case '\r': |
| 1259 |
c = 'r'; |
| 1260 |
break; |
| 1261 |
case '\t': |
| 1262 |
c = 't'; |
| 1263 |
break; |
| 1264 |
case '\v': |
| 1265 |
c = 'v'; |
| 1266 |
break; |
| 1267 |
default: |
| 1268 |
/* If it's the delimiter, must backslash it */ |
| 1269 |
if (c == delimc) |
| 1270 |
break; |
| 1271 |
/* All ASCII control chars are length 1 */ |
| 1272 |
ptr++; |
| 1273 |
continue; /* fall to end of loop */ |
| 1274 |
} |
| 1275 |
/* if we get here, we need to convert the control char */ |
| 1276 |
DUMPSOFAR(); |
| 1277 |
CopySendChar(cstate, '\\'); |
| 1278 |
CopySendChar(cstate, c); |
| 1279 |
start = ++ptr; /* do not include char in next run */ |
| 1280 |
} |
| 1281 |
else if (c == '\\' || c == delimc) |
| 1282 |
{ |
| 1283 |
DUMPSOFAR(); |
| 1284 |
CopySendChar(cstate, '\\'); |
| 1285 |
start = ptr++; /* we include char in next run */ |
| 1286 |
} |
| 1287 |
else |
| 1288 |
ptr++; |
| 1289 |
} |
| 1290 |
} |
| 1291 |
|
| 1292 |
DUMPSOFAR(); |
| 1293 |
} |
| 1294 |
|
| 1295 |
/* |
| 1296 |
* Send text representation of one attribute, with conversion and |
| 1297 |
* CSV-style escaping |
| 1298 |
*/ |
| 1299 |
static void |
| 1300 |
CopyAttributeOutCSV(CopyToState cstate, const char *string, |
| 1301 |
bool use_quote) |
| 1302 |
{ |
| 1303 |
const char *ptr; |
| 1304 |
const char *start; |
| 1305 |
char c; |
| 1306 |
char delimc = cstate->opts.delim[0]; |
| 1307 |
char quotec = cstate->opts.quote[0]; |
| 1308 |
char escapec = cstate->opts.escape[0]; |
| 1309 |
bool single_attr = (list_length(cstate->attnumlist) == 1); |
| 1310 |
|
| 1311 |
/* force quoting if it matches null_print (before conversion!) */ |
| 1312 |
if (!use_quote && strcmp(string, cstate->opts.null_print) == 0) |
| 1313 |
use_quote = true; |
| 1314 |
|
| 1315 |
if (cstate->need_transcoding) |
| 1316 |
ptr = pg_server_to_any(string, strlen(string), cstate->file_encoding); |
| 1317 |
else |
| 1318 |
ptr = string; |
| 1319 |
|
| 1320 |
/* |
| 1321 |
* Make a preliminary pass to discover if it needs quoting |
| 1322 |
*/ |
| 1323 |
if (!use_quote) |
| 1324 |
{ |
| 1325 |
/* |
| 1326 |
* Quote '\.' if it appears alone on a line, so that it will not be |
| 1327 |
* interpreted as an end-of-data marker. (PG 18 and up will not |
| 1328 |
* interpret '\.' in CSV that way, except in embedded-in-SQL data; but |
| 1329 |
* we want the data to be loadable by older versions too. Also, this |
| 1330 |
* avoids breaking clients that are still using PQgetline().) |
| 1331 |
*/ |
| 1332 |
if (single_attr && strcmp(ptr, "\\.") == 0) |
| 1333 |
use_quote = true; |
| 1334 |
else |
| 1335 |
{ |
| 1336 |
const char *tptr = ptr; |
| 1337 |
|
| 1338 |
while ((c = *tptr) != '\0') |
| 1339 |
{ |
| 1340 |
if (c == delimc || c == quotec || c == '\n' || c == '\r') |
| 1341 |
{ |
| 1342 |
use_quote = true; |
| 1343 |
break; |
| 1344 |
} |
| 1345 |
if (IS_HIGHBIT_SET(c) && cstate->encoding_embeds_ascii) |
| 1346 |
tptr += pg_encoding_mblen(cstate->file_encoding, tptr); |
| 1347 |
else |
| 1348 |
tptr++; |
| 1349 |
} |
| 1350 |
} |
| 1351 |
} |
| 1352 |
|
| 1353 |
if (use_quote) |
| 1354 |
{ |
| 1355 |
CopySendChar(cstate, quotec); |
| 1356 |
|
| 1357 |
/* |
| 1358 |
* We adopt the same optimization strategy as in CopyAttributeOutText |
| 1359 |
*/ |
| 1360 |
start = ptr; |
| 1361 |
while ((c = *ptr) != '\0') |
| 1362 |
{ |
| 1363 |
if (c == quotec || c == escapec) |
| 1364 |
{ |
| 1365 |
DUMPSOFAR(); |
| 1366 |
CopySendChar(cstate, escapec); |
| 1367 |
start = ptr; /* we include char in next run */ |
| 1368 |
} |
| 1369 |
if (IS_HIGHBIT_SET(c) && cstate->encoding_embeds_ascii) |
| 1370 |
ptr += pg_encoding_mblen(cstate->file_encoding, ptr); |
| 1371 |
else |
| 1372 |
ptr++; |
| 1373 |
} |
| 1374 |
DUMPSOFAR(); |
| 1375 |
|
| 1376 |
CopySendChar(cstate, quotec); |
| 1377 |
} |
| 1378 |
else |
| 1379 |
{ |
| 1380 |
/* If it doesn't need quoting, we can just dump it as-is */ |
| 1381 |
CopySendString(cstate, ptr); |
| 1382 |
} |
| 1383 |
} |
| 1384 |
|
| 1385 |
/* |
| 1386 |
* copy_dest_startup --- executor startup |
| 1387 |
*/ |
| 1388 |
static void |
| 1389 |
copy_dest_startup(DestReceiver *self, int operation, TupleDesc typeinfo) |
| 1390 |
{ |
| 1391 |
/* no-op */ |
| 1392 |
} |
| 1393 |
|
| 1394 |
/* |
| 1395 |
* copy_dest_receive --- receive one tuple |
| 1396 |
*/ |
| 1397 |
static bool |
| 1398 |
copy_dest_receive(TupleTableSlot *slot, DestReceiver *self) |
| 1399 |
{ |
| 1400 |
DR_copy *myState = (DR_copy *) self; |
| 1401 |
CopyToState cstate = myState->cstate; |
| 1402 |
|
| 1403 |
/* Send the data */ |
| 1404 |
CopyOneRowTo(cstate, slot); |
| 1405 |
|
| 1406 |
/* Increment the number of processed tuples, and report the progress */ |
| 1407 |
pgstat_progress_update_param(PROGRESS_COPY_TUPLES_PROCESSED, |
| 1408 |
++myState->processed); |
| 1409 |
|
| 1410 |
return true; |
| 1411 |
} |
| 1412 |
|
| 1413 |
/* |
| 1414 |
* copy_dest_shutdown --- executor end |
| 1415 |
*/ |
| 1416 |
static void |
| 1417 |
copy_dest_shutdown(DestReceiver *self) |
| 1418 |
{ |
| 1419 |
/* no-op */ |
| 1420 |
} |
| 1421 |
|
| 1422 |
/* |
| 1423 |
* copy_dest_destroy --- release DestReceiver object |
| 1424 |
*/ |
| 1425 |
static void |
| 1426 |
copy_dest_destroy(DestReceiver *self) |
| 1427 |
{ |
| 1428 |
pfree(self); |
| 1429 |
} |
| 1430 |
|
| 1431 |
/* |
| 1432 |
* CreateCopyDestReceiver -- create a suitable DestReceiver object |
| 1433 |
*/ |
| 1434 |
DestReceiver * |
| 1435 |
CreateCopyDestReceiver(void) |
| 1436 |
{ |
| 1437 |
DR_copy *self = (DR_copy *) palloc(sizeof(DR_copy)); |
| 1438 |
|
| 1439 |
self->pub.receiveSlot = copy_dest_receive; |
| 1440 |
self->pub.rStartup = copy_dest_startup; |
| 1441 |
self->pub.rShutdown = copy_dest_shutdown; |
| 1442 |
self->pub.rDestroy = copy_dest_destroy; |
| 1443 |
self->pub.mydest = DestCopyOut; |
| 1444 |
|
| 1445 |
self->cstate = NULL; /* will be set later */ |
| 1446 |
self->processed = 0; |
| 1447 |
|
| 1448 |
return (DestReceiver *) self; |
| 1449 |
} |