| 1 |
/*------------------------------------------------------------------------- |
| 2 |
* |
| 3 |
* copyfromparse.c |
| 4 |
* Parse CSV/text/binary format for COPY FROM. |
| 5 |
* |
| 6 |
* This file contains routines to parse the text, CSV and binary input |
| 7 |
* formats. The main entry point is NextCopyFrom(), which parses the |
| 8 |
* next input line and returns it as Datums. |
| 9 |
* |
| 10 |
* In text/CSV mode, the parsing happens in multiple stages: |
| 11 |
* |
| 12 |
* [data source] --> raw_buf --> input_buf --> line_buf --> attribute_buf |
| 13 |
* 1. 2. 3. 4. |
| 14 |
* |
| 15 |
* 1. CopyLoadRawBuf() reads raw data from the input file or client, and |
| 16 |
* places it into 'raw_buf'. |
| 17 |
* |
| 18 |
* 2. CopyConvertBuf() calls the encoding conversion function to convert |
| 19 |
* the data in 'raw_buf' from client to server encoding, placing the |
| 20 |
* converted result in 'input_buf'. |
| 21 |
* |
| 22 |
* 3. CopyReadLine() parses the data in 'input_buf', one line at a time. |
| 23 |
* It is responsible for finding the next newline marker, taking quote and |
| 24 |
* escape characters into account according to the COPY options. The line |
| 25 |
* is copied into 'line_buf', with quotes and escape characters still |
| 26 |
* intact. |
| 27 |
* |
| 28 |
* 4. CopyReadAttributesText/CSV() function takes the input line from |
| 29 |
* 'line_buf', and splits it into fields, unescaping the data as required. |
| 30 |
* The fields are stored in 'attribute_buf', and 'raw_fields' array holds |
| 31 |
* pointers to each field. |
| 32 |
* |
| 33 |
* If encoding conversion is not required, a shortcut is taken in step 2 to |
| 34 |
* avoid copying the data unnecessarily. The 'input_buf' pointer is set to |
| 35 |
* point directly to 'raw_buf', so that CopyLoadRawBuf() loads the raw data |
| 36 |
* directly into 'input_buf'. CopyConvertBuf() then merely validates that |
| 37 |
* the data is valid in the current encoding. |
| 38 |
* |
| 39 |
* In binary mode, the pipeline is much simpler. Input is loaded into |
| 40 |
* 'raw_buf', and encoding conversion is done in the datatype-specific |
| 41 |
* receive functions, if required. 'input_buf' and 'line_buf' are not used, |
| 42 |
* but 'attribute_buf' is used as a temporary buffer to hold one attribute's |
| 43 |
* data when it's passed the receive function. |
| 44 |
* |
| 45 |
* 'raw_buf' is always 64 kB in size (RAW_BUF_SIZE). 'input_buf' is also |
| 46 |
* 64 kB (INPUT_BUF_SIZE), if encoding conversion is required. 'line_buf' |
| 47 |
* and 'attribute_buf' are expanded on demand, to hold the longest line |
| 48 |
* encountered so far. |
| 49 |
* |
| 50 |
* Portions Copyright (c) 1996-2025, PostgreSQL Global Development Group |
| 51 |
* Portions Copyright (c) 1994, Regents of the University of California |
| 52 |
* |
| 53 |
* |
| 54 |
* IDENTIFICATION |
| 55 |
* src/backend/commands/copyfromparse.c |
| 56 |
* |
| 57 |
*------------------------------------------------------------------------- |
| 58 |
*/ |
| 59 |
#include "postgres.h" |
| 60 |
|
| 61 |
#include <ctype.h> |
| 62 |
#include <unistd.h> |
| 63 |
#include <sys/stat.h> |
| 64 |
|
| 65 |
#include "commands/copyapi.h" |
| 66 |
#include "commands/copyfrom_internal.h" |
| 67 |
#include "commands/progress.h" |
| 68 |
#include "executor/executor.h" |
| 69 |
#include "libpq/libpq.h" |
| 70 |
#include "libpq/pqformat.h" |
| 71 |
#include "mb/pg_wchar.h" |
| 72 |
#include "miscadmin.h" |
| 73 |
#include "pgstat.h" |
| 74 |
#include "port/pg_bswap.h" |
| 75 |
#include "utils/builtins.h" |
| 76 |
#include "utils/rel.h" |
| 77 |
|
| 78 |
#define ISOCTAL(c) (((c) >= '0') && ((c) <= '7')) |
| 79 |
#define OCTVALUE(c) ((c) - '0') |
| 80 |
|
| 81 |
/* |
| 82 |
* These macros centralize code used to process line_buf and input_buf buffers. |
| 83 |
* They are macros because they often do continue/break control and to avoid |
| 84 |
* function call overhead in tight COPY loops. |
| 85 |
* |
| 86 |
* We must use "if (1)" because the usual "do {...} while(0)" wrapper would |
| 87 |
* prevent the continue/break processing from working. We end the "if (1)" |
| 88 |
* with "else ((void) 0)" to ensure the "if" does not unintentionally match |
| 89 |
* any "else" in the calling code, and to avoid any compiler warnings about |
| 90 |
* empty statements. See http://www.cit.gu.edu.au/~anthony/info/C/C.macros. |
| 91 |
*/ |
| 92 |
|
| 93 |
/* |
| 94 |
* This keeps the character read at the top of the loop in the buffer |
| 95 |
* even if there is more than one read-ahead. |
| 96 |
*/ |
| 97 |
#define IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(extralen) \ |
| 98 |
if (1) \ |
| 99 |
{ \ |
| 100 |
if (input_buf_ptr + (extralen) >= copy_buf_len && !hit_eof) \ |
| 101 |
{ \ |
| 102 |
input_buf_ptr = prev_raw_ptr; /* undo fetch */ \ |
| 103 |
need_data = true; \ |
| 104 |
continue; \ |
| 105 |
} \ |
| 106 |
} else ((void) 0) |
| 107 |
|
| 108 |
/* This consumes the remainder of the buffer and breaks */ |
| 109 |
#define IF_NEED_REFILL_AND_EOF_BREAK(extralen) \ |
| 110 |
if (1) \ |
| 111 |
{ \ |
| 112 |
if (input_buf_ptr + (extralen) >= copy_buf_len && hit_eof) \ |
| 113 |
{ \ |
| 114 |
if (extralen) \ |
| 115 |
input_buf_ptr = copy_buf_len; /* consume the partial character */ \ |
| 116 |
/* backslash just before EOF, treat as data char */ \ |
| 117 |
result = true; \ |
| 118 |
break; \ |
| 119 |
} \ |
| 120 |
} else ((void) 0) |
| 121 |
|
| 122 |
/* |
| 123 |
* Transfer any approved data to line_buf; must do this to be sure |
| 124 |
* there is some room in input_buf. |
| 125 |
*/ |
| 126 |
#define REFILL_LINEBUF \ |
| 127 |
if (1) \ |
| 128 |
{ \ |
| 129 |
if (input_buf_ptr > cstate->input_buf_index) \ |
| 130 |
{ \ |
| 131 |
appendBinaryStringInfo(&cstate->line_buf, \ |
| 132 |
cstate->input_buf + cstate->input_buf_index, \ |
| 133 |
input_buf_ptr - cstate->input_buf_index); \ |
| 134 |
cstate->input_buf_index = input_buf_ptr; \ |
| 135 |
} \ |
| 136 |
} else ((void) 0) |
| 137 |
|
| 138 |
/* NOTE: there's a copy of this in copyto.c */ |
| 139 |
static const char BinarySignature[11] = "PGCOPY\n\377\r\n\0"; |
| 140 |
|
| 141 |
|
| 142 |
/* non-export function prototypes */ |
| 143 |
static bool CopyReadLine(CopyFromState cstate, bool is_csv); |
| 144 |
static bool CopyReadLineText(CopyFromState cstate, bool is_csv); |
| 145 |
static int CopyReadAttributesText(CopyFromState cstate); |
| 146 |
static int CopyReadAttributesCSV(CopyFromState cstate); |
| 147 |
static Datum CopyReadBinaryAttribute(CopyFromState cstate, FmgrInfo *flinfo, |
| 148 |
Oid typioparam, int32 typmod, |
| 149 |
bool *isnull); |
| 150 |
static pg_always_inline bool CopyFromTextLikeOneRow(CopyFromState cstate, |
| 151 |
ExprContext *econtext, |
| 152 |
Datum *values, |
| 153 |
bool *nulls, |
| 154 |
bool is_csv); |
| 155 |
static pg_always_inline bool NextCopyFromRawFieldsInternal(CopyFromState cstate, |
| 156 |
char ***fields, |
| 157 |
int *nfields, |
| 158 |
bool is_csv); |
| 159 |
|
| 160 |
|
| 161 |
/* Low-level communications functions */ |
| 162 |
static int CopyGetData(CopyFromState cstate, void *databuf, |
| 163 |
int minread, int maxread); |
| 164 |
static inline bool CopyGetInt32(CopyFromState cstate, int32 *val); |
| 165 |
static inline bool CopyGetInt16(CopyFromState cstate, int16 *val); |
| 166 |
static void CopyLoadInputBuf(CopyFromState cstate); |
| 167 |
static int CopyReadBinaryData(CopyFromState cstate, char *dest, int nbytes); |
| 168 |
|
| 169 |
void |
| 170 |
ReceiveCopyBegin(CopyFromState cstate) |
| 171 |
{ |
| 172 |
StringInfoData buf; |
| 173 |
int natts = list_length(cstate->attnumlist); |
| 174 |
int16 format = (cstate->opts.binary ? 1 : 0); |
| 175 |
int i; |
| 176 |
|
| 177 |
pq_beginmessage(&buf, PqMsg_CopyInResponse); |
| 178 |
pq_sendbyte(&buf, format); /* overall format */ |
| 179 |
pq_sendint16(&buf, natts); |
| 180 |
for (i = 0; i < natts; i++) |
| 181 |
pq_sendint16(&buf, format); /* per-column formats */ |
| 182 |
pq_endmessage(&buf); |
| 183 |
cstate->copy_src = COPY_FRONTEND; |
| 184 |
cstate->fe_msgbuf = makeStringInfo(); |
| 185 |
/* We *must* flush here to ensure FE knows it can send. */ |
| 186 |
pq_flush(); |
| 187 |
} |
| 188 |
|
| 189 |
void |
| 190 |
ReceiveCopyBinaryHeader(CopyFromState cstate) |
| 191 |
{ |
| 192 |
char readSig[11]; |
| 193 |
int32 tmp; |
| 194 |
|
| 195 |
/* Signature */ |
| 196 |
if (CopyReadBinaryData(cstate, readSig, 11) != 11 || |
| 197 |
memcmp(readSig, BinarySignature, 11) != 0) |
| 198 |
ereport(ERROR, |
| 199 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 200 |
errmsg("COPY file signature not recognized"))); |
| 201 |
/* Flags field */ |
| 202 |
if (!CopyGetInt32(cstate, &tmp)) |
| 203 |
ereport(ERROR, |
| 204 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 205 |
errmsg("invalid COPY file header (missing flags)"))); |
| 206 |
if ((tmp & (1 << 16)) != 0) |
| 207 |
ereport(ERROR, |
| 208 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 209 |
errmsg("invalid COPY file header (WITH OIDS)"))); |
| 210 |
tmp &= ~(1 << 16); |
| 211 |
if ((tmp >> 16) != 0) |
| 212 |
ereport(ERROR, |
| 213 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 214 |
errmsg("unrecognized critical flags in COPY file header"))); |
| 215 |
/* Header extension length */ |
| 216 |
if (!CopyGetInt32(cstate, &tmp) || |
| 217 |
tmp < 0) |
| 218 |
ereport(ERROR, |
| 219 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 220 |
errmsg("invalid COPY file header (missing length)"))); |
| 221 |
/* Skip extension header, if present */ |
| 222 |
while (tmp-- > 0) |
| 223 |
{ |
| 224 |
if (CopyReadBinaryData(cstate, readSig, 1) != 1) |
| 225 |
ereport(ERROR, |
| 226 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 227 |
errmsg("invalid COPY file header (wrong length)"))); |
| 228 |
} |
| 229 |
} |
| 230 |
|
| 231 |
/* |
| 232 |
* CopyGetData reads data from the source (file or frontend) |
| 233 |
* |
| 234 |
* We attempt to read at least minread, and at most maxread, bytes from |
| 235 |
* the source. The actual number of bytes read is returned; if this is |
| 236 |
* less than minread, EOF was detected. |
| 237 |
* |
| 238 |
* Note: when copying from the frontend, we expect a proper EOF mark per |
| 239 |
* protocol; if the frontend simply drops the connection, we raise error. |
| 240 |
* It seems unwise to allow the COPY IN to complete normally in that case. |
| 241 |
* |
| 242 |
* NB: no data conversion is applied here. |
| 243 |
*/ |
| 244 |
static int |
| 245 |
CopyGetData(CopyFromState cstate, void *databuf, int minread, int maxread) |
| 246 |
{ |
| 247 |
int bytesread = 0; |
| 248 |
|
| 249 |
switch (cstate->copy_src) |
| 250 |
{ |
| 251 |
case COPY_FILE: |
| 252 |
bytesread = fread(databuf, 1, maxread, cstate->copy_file); |
| 253 |
if (ferror(cstate->copy_file)) |
| 254 |
ereport(ERROR, |
| 255 |
(errcode_for_file_access(), |
| 256 |
errmsg("could not read from COPY file: %m"))); |
| 257 |
if (bytesread == 0) |
| 258 |
cstate->raw_reached_eof = true; |
| 259 |
break; |
| 260 |
case COPY_FRONTEND: |
| 261 |
while (maxread > 0 && bytesread < minread && !cstate->raw_reached_eof) |
| 262 |
{ |
| 263 |
int avail; |
| 264 |
|
| 265 |
while (cstate->fe_msgbuf->cursor >= cstate->fe_msgbuf->len) |
| 266 |
{ |
| 267 |
/* Try to receive another message */ |
| 268 |
int mtype; |
| 269 |
int maxmsglen; |
| 270 |
|
| 271 |
readmessage: |
| 272 |
HOLD_CANCEL_INTERRUPTS(); |
| 273 |
pq_startmsgread(); |
| 274 |
mtype = pq_getbyte(); |
| 275 |
if (mtype == EOF) |
| 276 |
ereport(ERROR, |
| 277 |
(errcode(ERRCODE_CONNECTION_FAILURE), |
| 278 |
errmsg("unexpected EOF on client connection with an open transaction"))); |
| 279 |
/* Validate message type and set packet size limit */ |
| 280 |
switch (mtype) |
| 281 |
{ |
| 282 |
case PqMsg_CopyData: |
| 283 |
maxmsglen = PQ_LARGE_MESSAGE_LIMIT; |
| 284 |
break; |
| 285 |
case PqMsg_CopyDone: |
| 286 |
case PqMsg_CopyFail: |
| 287 |
case PqMsg_Flush: |
| 288 |
case PqMsg_Sync: |
| 289 |
maxmsglen = PQ_SMALL_MESSAGE_LIMIT; |
| 290 |
break; |
| 291 |
default: |
| 292 |
ereport(ERROR, |
| 293 |
(errcode(ERRCODE_PROTOCOL_VIOLATION), |
| 294 |
errmsg("unexpected message type 0x%02X during COPY from stdin", |
| 295 |
mtype))); |
| 296 |
maxmsglen = 0; /* keep compiler quiet */ |
| 297 |
break; |
| 298 |
} |
| 299 |
/* Now collect the message body */ |
| 300 |
if (pq_getmessage(cstate->fe_msgbuf, maxmsglen)) |
| 301 |
ereport(ERROR, |
| 302 |
(errcode(ERRCODE_CONNECTION_FAILURE), |
| 303 |
errmsg("unexpected EOF on client connection with an open transaction"))); |
| 304 |
RESUME_CANCEL_INTERRUPTS(); |
| 305 |
/* ... and process it */ |
| 306 |
switch (mtype) |
| 307 |
{ |
| 308 |
case PqMsg_CopyData: |
| 309 |
break; |
| 310 |
case PqMsg_CopyDone: |
| 311 |
/* COPY IN correctly terminated by frontend */ |
| 312 |
cstate->raw_reached_eof = true; |
| 313 |
return bytesread; |
| 314 |
case PqMsg_CopyFail: |
| 315 |
ereport(ERROR, |
| 316 |
(errcode(ERRCODE_QUERY_CANCELED), |
| 317 |
errmsg("COPY from stdin failed: %s", |
| 318 |
pq_getmsgstring(cstate->fe_msgbuf)))); |
| 319 |
break; |
| 320 |
case PqMsg_Flush: |
| 321 |
case PqMsg_Sync: |
| 322 |
|
| 323 |
/* |
| 324 |
* Ignore Flush/Sync for the convenience of client |
| 325 |
* libraries (such as libpq) that may send those |
| 326 |
* without noticing that the command they just |
| 327 |
* sent was COPY. |
| 328 |
*/ |
| 329 |
goto readmessage; |
| 330 |
default: |
| 331 |
Assert(false); /* NOT REACHED */ |
| 332 |
} |
| 333 |
} |
| 334 |
avail = cstate->fe_msgbuf->len - cstate->fe_msgbuf->cursor; |
| 335 |
if (avail > maxread) |
| 336 |
avail = maxread; |
| 337 |
pq_copymsgbytes(cstate->fe_msgbuf, databuf, avail); |
| 338 |
databuf = (void *) ((char *) databuf + avail); |
| 339 |
maxread -= avail; |
| 340 |
bytesread += avail; |
| 341 |
} |
| 342 |
break; |
| 343 |
case COPY_CALLBACK: |
| 344 |
bytesread = cstate->data_source_cb(databuf, minread, maxread); |
| 345 |
break; |
| 346 |
} |
| 347 |
|
| 348 |
return bytesread; |
| 349 |
} |
| 350 |
|
| 351 |
|
| 352 |
/* |
| 353 |
* These functions do apply some data conversion |
| 354 |
*/ |
| 355 |
|
| 356 |
/* |
| 357 |
* CopyGetInt32 reads an int32 that appears in network byte order |
| 358 |
* |
| 359 |
* Returns true if OK, false if EOF |
| 360 |
*/ |
| 361 |
static inline bool |
| 362 |
CopyGetInt32(CopyFromState cstate, int32 *val) |
| 363 |
{ |
| 364 |
uint32 buf; |
| 365 |
|
| 366 |
if (CopyReadBinaryData(cstate, (char *) &buf, sizeof(buf)) != sizeof(buf)) |
| 367 |
{ |
| 368 |
*val = 0; /* suppress compiler warning */ |
| 369 |
return false; |
| 370 |
} |
| 371 |
*val = (int32) pg_ntoh32(buf); |
| 372 |
return true; |
| 373 |
} |
| 374 |
|
| 375 |
/* |
| 376 |
* CopyGetInt16 reads an int16 that appears in network byte order |
| 377 |
*/ |
| 378 |
static inline bool |
| 379 |
CopyGetInt16(CopyFromState cstate, int16 *val) |
| 380 |
{ |
| 381 |
uint16 buf; |
| 382 |
|
| 383 |
if (CopyReadBinaryData(cstate, (char *) &buf, sizeof(buf)) != sizeof(buf)) |
| 384 |
{ |
| 385 |
*val = 0; /* suppress compiler warning */ |
| 386 |
return false; |
| 387 |
} |
| 388 |
*val = (int16) pg_ntoh16(buf); |
| 389 |
return true; |
| 390 |
} |
| 391 |
|
| 392 |
|
| 393 |
/* |
| 394 |
* Perform encoding conversion on data in 'raw_buf', writing the converted |
| 395 |
* data into 'input_buf'. |
| 396 |
* |
| 397 |
* On entry, there must be some data to convert in 'raw_buf'. |
| 398 |
*/ |
| 399 |
static void |
| 400 |
CopyConvertBuf(CopyFromState cstate) |
| 401 |
{ |
| 402 |
/* |
| 403 |
* If the file and server encoding are the same, no encoding conversion is |
| 404 |
* required. However, we still need to verify that the input is valid for |
| 405 |
* the encoding. |
| 406 |
*/ |
| 407 |
if (!cstate->need_transcoding) |
| 408 |
{ |
| 409 |
/* |
| 410 |
* When conversion is not required, input_buf and raw_buf are the |
| 411 |
* same. raw_buf_len is the total number of bytes in the buffer, and |
| 412 |
* input_buf_len tracks how many of those bytes have already been |
| 413 |
* verified. |
| 414 |
*/ |
| 415 |
int preverifiedlen = cstate->input_buf_len; |
| 416 |
int unverifiedlen = cstate->raw_buf_len - cstate->input_buf_len; |
| 417 |
int nverified; |
| 418 |
|
| 419 |
if (unverifiedlen == 0) |
| 420 |
{ |
| 421 |
/* |
| 422 |
* If no more raw data is coming, report the EOF to the caller. |
| 423 |
*/ |
| 424 |
if (cstate->raw_reached_eof) |
| 425 |
cstate->input_reached_eof = true; |
| 426 |
return; |
| 427 |
} |
| 428 |
|
| 429 |
/* |
| 430 |
* Verify the new data, including any residual unverified bytes from |
| 431 |
* previous round. |
| 432 |
*/ |
| 433 |
nverified = pg_encoding_verifymbstr(cstate->file_encoding, |
| 434 |
cstate->raw_buf + preverifiedlen, |
| 435 |
unverifiedlen); |
| 436 |
if (nverified == 0) |
| 437 |
{ |
| 438 |
/* |
| 439 |
* Could not verify anything. |
| 440 |
* |
| 441 |
* If there is no more raw input data coming, it means that there |
| 442 |
* was an incomplete multi-byte sequence at the end. Also, if |
| 443 |
* there's "enough" input left, we should be able to verify at |
| 444 |
* least one character, and a failure to do so means that we've |
| 445 |
* hit an invalid byte sequence. |
| 446 |
*/ |
| 447 |
if (cstate->raw_reached_eof || unverifiedlen >= pg_encoding_max_length(cstate->file_encoding)) |
| 448 |
cstate->input_reached_error = true; |
| 449 |
return; |
| 450 |
} |
| 451 |
cstate->input_buf_len += nverified; |
| 452 |
} |
| 453 |
else |
| 454 |
{ |
| 455 |
/* |
| 456 |
* Encoding conversion is needed. |
| 457 |
*/ |
| 458 |
int nbytes; |
| 459 |
unsigned char *src; |
| 460 |
int srclen; |
| 461 |
unsigned char *dst; |
| 462 |
int dstlen; |
| 463 |
int convertedlen; |
| 464 |
|
| 465 |
if (RAW_BUF_BYTES(cstate) == 0) |
| 466 |
{ |
| 467 |
/* |
| 468 |
* If no more raw data is coming, report the EOF to the caller. |
| 469 |
*/ |
| 470 |
if (cstate->raw_reached_eof) |
| 471 |
cstate->input_reached_eof = true; |
| 472 |
return; |
| 473 |
} |
| 474 |
|
| 475 |
/* |
| 476 |
* First, copy down any unprocessed data. |
| 477 |
*/ |
| 478 |
nbytes = INPUT_BUF_BYTES(cstate); |
| 479 |
if (nbytes > 0 && cstate->input_buf_index > 0) |
| 480 |
memmove(cstate->input_buf, cstate->input_buf + cstate->input_buf_index, |
| 481 |
nbytes); |
| 482 |
cstate->input_buf_index = 0; |
| 483 |
cstate->input_buf_len = nbytes; |
| 484 |
cstate->input_buf[nbytes] = '\0'; |
| 485 |
|
| 486 |
src = (unsigned char *) cstate->raw_buf + cstate->raw_buf_index; |
| 487 |
srclen = cstate->raw_buf_len - cstate->raw_buf_index; |
| 488 |
dst = (unsigned char *) cstate->input_buf + cstate->input_buf_len; |
| 489 |
dstlen = INPUT_BUF_SIZE - cstate->input_buf_len + 1; |
| 490 |
|
| 491 |
/* |
| 492 |
* Do the conversion. This might stop short, if there is an invalid |
| 493 |
* byte sequence in the input. We'll convert as much as we can in |
| 494 |
* that case. |
| 495 |
* |
| 496 |
* Note: Even if we hit an invalid byte sequence, we don't report the |
| 497 |
* error until all the valid bytes have been consumed. The input |
| 498 |
* might contain an end-of-input marker (\.), and we don't want to |
| 499 |
* report an error if the invalid byte sequence is after the |
| 500 |
* end-of-input marker. We might unnecessarily convert some data |
| 501 |
* after the end-of-input marker as long as it's valid for the |
| 502 |
* encoding, but that's harmless. |
| 503 |
*/ |
| 504 |
convertedlen = pg_do_encoding_conversion_buf(cstate->conversion_proc, |
| 505 |
cstate->file_encoding, |
| 506 |
GetDatabaseEncoding(), |
| 507 |
src, srclen, |
| 508 |
dst, dstlen, |
| 509 |
true); |
| 510 |
if (convertedlen == 0) |
| 511 |
{ |
| 512 |
/* |
| 513 |
* Could not convert anything. If there is no more raw input data |
| 514 |
* coming, it means that there was an incomplete multi-byte |
| 515 |
* sequence at the end. Also, if there is plenty of input left, |
| 516 |
* we should be able to convert at least one character, so a |
| 517 |
* failure to do so must mean that we've hit a byte sequence |
| 518 |
* that's invalid. |
| 519 |
*/ |
| 520 |
if (cstate->raw_reached_eof || srclen >= MAX_CONVERSION_INPUT_LENGTH) |
| 521 |
cstate->input_reached_error = true; |
| 522 |
return; |
| 523 |
} |
| 524 |
cstate->raw_buf_index += convertedlen; |
| 525 |
cstate->input_buf_len += strlen((char *) dst); |
| 526 |
} |
| 527 |
} |
| 528 |
|
| 529 |
/* |
| 530 |
* Report an encoding or conversion error. |
| 531 |
*/ |
| 532 |
static void |
| 533 |
CopyConversionError(CopyFromState cstate) |
| 534 |
{ |
| 535 |
Assert(cstate->raw_buf_len > 0); |
| 536 |
Assert(cstate->input_reached_error); |
| 537 |
|
| 538 |
if (!cstate->need_transcoding) |
| 539 |
{ |
| 540 |
/* |
| 541 |
* Everything up to input_buf_len was successfully verified, and |
| 542 |
* input_buf_len points to the invalid or incomplete character. |
| 543 |
*/ |
| 544 |
report_invalid_encoding(cstate->file_encoding, |
| 545 |
cstate->raw_buf + cstate->input_buf_len, |
| 546 |
cstate->raw_buf_len - cstate->input_buf_len); |
| 547 |
} |
| 548 |
else |
| 549 |
{ |
| 550 |
/* |
| 551 |
* raw_buf_index points to the invalid or untranslatable character. We |
| 552 |
* let the conversion routine report the error, because it can provide |
| 553 |
* a more specific error message than we could here. An earlier call |
| 554 |
* to the conversion routine in CopyConvertBuf() detected that there |
| 555 |
* is an error, now we call the conversion routine again with |
| 556 |
* noError=false, to have it throw the error. |
| 557 |
*/ |
| 558 |
unsigned char *src; |
| 559 |
int srclen; |
| 560 |
unsigned char *dst; |
| 561 |
int dstlen; |
| 562 |
|
| 563 |
src = (unsigned char *) cstate->raw_buf + cstate->raw_buf_index; |
| 564 |
srclen = cstate->raw_buf_len - cstate->raw_buf_index; |
| 565 |
dst = (unsigned char *) cstate->input_buf + cstate->input_buf_len; |
| 566 |
dstlen = INPUT_BUF_SIZE - cstate->input_buf_len + 1; |
| 567 |
|
| 568 |
(void) pg_do_encoding_conversion_buf(cstate->conversion_proc, |
| 569 |
cstate->file_encoding, |
| 570 |
GetDatabaseEncoding(), |
| 571 |
src, srclen, |
| 572 |
dst, dstlen, |
| 573 |
false); |
| 574 |
|
| 575 |
/* |
| 576 |
* The conversion routine should have reported an error, so this |
| 577 |
* should not be reached. |
| 578 |
*/ |
| 579 |
elog(ERROR, "encoding conversion failed without error"); |
| 580 |
} |
| 581 |
} |
| 582 |
|
| 583 |
/* |
| 584 |
* Load more data from data source to raw_buf. |
| 585 |
* |
| 586 |
* If RAW_BUF_BYTES(cstate) > 0, the unprocessed bytes are moved to the |
| 587 |
* beginning of the buffer, and we load new data after that. |
| 588 |
*/ |
| 589 |
static void |
| 590 |
CopyLoadRawBuf(CopyFromState cstate) |
| 591 |
{ |
| 592 |
int nbytes; |
| 593 |
int inbytes; |
| 594 |
|
| 595 |
/* |
| 596 |
* In text mode, if encoding conversion is not required, raw_buf and |
| 597 |
* input_buf point to the same buffer. Their len/index better agree, too. |
| 598 |
*/ |
| 599 |
if (cstate->raw_buf == cstate->input_buf) |
| 600 |
{ |
| 601 |
Assert(!cstate->need_transcoding); |
| 602 |
Assert(cstate->raw_buf_index == cstate->input_buf_index); |
| 603 |
Assert(cstate->input_buf_len <= cstate->raw_buf_len); |
| 604 |
} |
| 605 |
|
| 606 |
/* |
| 607 |
* Copy down the unprocessed data if any. |
| 608 |
*/ |
| 609 |
nbytes = RAW_BUF_BYTES(cstate); |
| 610 |
if (nbytes > 0 && cstate->raw_buf_index > 0) |
| 611 |
memmove(cstate->raw_buf, cstate->raw_buf + cstate->raw_buf_index, |
| 612 |
nbytes); |
| 613 |
cstate->raw_buf_len -= cstate->raw_buf_index; |
| 614 |
cstate->raw_buf_index = 0; |
| 615 |
|
| 616 |
/* |
| 617 |
* If raw_buf and input_buf are in fact the same buffer, adjust the |
| 618 |
* input_buf variables, too. |
| 619 |
*/ |
| 620 |
if (cstate->raw_buf == cstate->input_buf) |
| 621 |
{ |
| 622 |
cstate->input_buf_len -= cstate->input_buf_index; |
| 623 |
cstate->input_buf_index = 0; |
| 624 |
} |
| 625 |
|
| 626 |
/* Load more data */ |
| 627 |
inbytes = CopyGetData(cstate, cstate->raw_buf + cstate->raw_buf_len, |
| 628 |
1, RAW_BUF_SIZE - cstate->raw_buf_len); |
| 629 |
nbytes += inbytes; |
| 630 |
cstate->raw_buf[nbytes] = '\0'; |
| 631 |
cstate->raw_buf_len = nbytes; |
| 632 |
|
| 633 |
cstate->bytes_processed += inbytes; |
| 634 |
pgstat_progress_update_param(PROGRESS_COPY_BYTES_PROCESSED, cstate->bytes_processed); |
| 635 |
|
| 636 |
if (inbytes == 0) |
| 637 |
cstate->raw_reached_eof = true; |
| 638 |
} |
| 639 |
|
| 640 |
/* |
| 641 |
* CopyLoadInputBuf loads some more data into input_buf |
| 642 |
* |
| 643 |
* On return, at least one more input character is loaded into |
| 644 |
* input_buf, or input_reached_eof is set. |
| 645 |
* |
| 646 |
* If INPUT_BUF_BYTES(cstate) > 0, the unprocessed bytes are moved to the start |
| 647 |
* of the buffer and then we load more data after that. |
| 648 |
*/ |
| 649 |
static void |
| 650 |
CopyLoadInputBuf(CopyFromState cstate) |
| 651 |
{ |
| 652 |
int nbytes = INPUT_BUF_BYTES(cstate); |
| 653 |
|
| 654 |
/* |
| 655 |
* The caller has updated input_buf_index to indicate how much of the |
| 656 |
* input has been consumed and isn't needed anymore. If input_buf is the |
| 657 |
* same physical area as raw_buf, update raw_buf_index accordingly. |
| 658 |
*/ |
| 659 |
if (cstate->raw_buf == cstate->input_buf) |
| 660 |
{ |
| 661 |
Assert(!cstate->need_transcoding); |
| 662 |
Assert(cstate->input_buf_index >= cstate->raw_buf_index); |
| 663 |
cstate->raw_buf_index = cstate->input_buf_index; |
| 664 |
} |
| 665 |
|
| 666 |
for (;;) |
| 667 |
{ |
| 668 |
/* If we now have some unconverted data, try to convert it */ |
| 669 |
CopyConvertBuf(cstate); |
| 670 |
|
| 671 |
/* If we now have some more input bytes ready, return them */ |
| 672 |
if (INPUT_BUF_BYTES(cstate) > nbytes) |
| 673 |
return; |
| 674 |
|
| 675 |
/* |
| 676 |
* If we reached an invalid byte sequence, or we're at an incomplete |
| 677 |
* multi-byte character but there is no more raw input data, report |
| 678 |
* conversion error. |
| 679 |
*/ |
| 680 |
if (cstate->input_reached_error) |
| 681 |
CopyConversionError(cstate); |
| 682 |
|
| 683 |
/* no more input, and everything has been converted */ |
| 684 |
if (cstate->input_reached_eof) |
| 685 |
break; |
| 686 |
|
| 687 |
/* Try to load more raw data */ |
| 688 |
Assert(!cstate->raw_reached_eof); |
| 689 |
CopyLoadRawBuf(cstate); |
| 690 |
} |
| 691 |
} |
| 692 |
|
| 693 |
/* |
| 694 |
* CopyReadBinaryData |
| 695 |
* |
| 696 |
* Reads up to 'nbytes' bytes from cstate->copy_file via cstate->raw_buf |
| 697 |
* and writes them to 'dest'. Returns the number of bytes read (which |
| 698 |
* would be less than 'nbytes' only if we reach EOF). |
| 699 |
*/ |
| 700 |
static int |
| 701 |
CopyReadBinaryData(CopyFromState cstate, char *dest, int nbytes) |
| 702 |
{ |
| 703 |
int copied_bytes = 0; |
| 704 |
|
| 705 |
if (RAW_BUF_BYTES(cstate) >= nbytes) |
| 706 |
{ |
| 707 |
/* Enough bytes are present in the buffer. */ |
| 708 |
memcpy(dest, cstate->raw_buf + cstate->raw_buf_index, nbytes); |
| 709 |
cstate->raw_buf_index += nbytes; |
| 710 |
copied_bytes = nbytes; |
| 711 |
} |
| 712 |
else |
| 713 |
{ |
| 714 |
/* |
| 715 |
* Not enough bytes in the buffer, so must read from the file. Need |
| 716 |
* to loop since 'nbytes' could be larger than the buffer size. |
| 717 |
*/ |
| 718 |
do |
| 719 |
{ |
| 720 |
int copy_bytes; |
| 721 |
|
| 722 |
/* Load more data if buffer is empty. */ |
| 723 |
if (RAW_BUF_BYTES(cstate) == 0) |
| 724 |
{ |
| 725 |
CopyLoadRawBuf(cstate); |
| 726 |
if (cstate->raw_reached_eof) |
| 727 |
break; /* EOF */ |
| 728 |
} |
| 729 |
|
| 730 |
/* Transfer some bytes. */ |
| 731 |
copy_bytes = Min(nbytes - copied_bytes, RAW_BUF_BYTES(cstate)); |
| 732 |
memcpy(dest, cstate->raw_buf + cstate->raw_buf_index, copy_bytes); |
| 733 |
cstate->raw_buf_index += copy_bytes; |
| 734 |
dest += copy_bytes; |
| 735 |
copied_bytes += copy_bytes; |
| 736 |
} while (copied_bytes < nbytes); |
| 737 |
} |
| 738 |
|
| 739 |
return copied_bytes; |
| 740 |
} |
| 741 |
|
| 742 |
/* |
| 743 |
* This function is exposed for use by extensions that read raw fields in the |
| 744 |
* next line. See NextCopyFromRawFieldsInternal() for details. |
| 745 |
*/ |
| 746 |
bool |
| 747 |
NextCopyFromRawFields(CopyFromState cstate, char ***fields, int *nfields) |
| 748 |
{ |
| 749 |
return NextCopyFromRawFieldsInternal(cstate, fields, nfields, |
| 750 |
cstate->opts.csv_mode); |
| 751 |
} |
| 752 |
|
| 753 |
/* |
| 754 |
* Workhorse for NextCopyFromRawFields(). |
| 755 |
* |
| 756 |
* Read raw fields in the next line for COPY FROM in text or csv mode. Return |
| 757 |
* false if no more lines. |
| 758 |
* |
| 759 |
* An internal temporary buffer is returned via 'fields'. It is valid until |
| 760 |
* the next call of the function. Since the function returns all raw fields |
| 761 |
* in the input file, 'nfields' could be different from the number of columns |
| 762 |
* in the relation. |
| 763 |
* |
| 764 |
* NOTE: force_not_null option are not applied to the returned fields. |
| 765 |
* |
| 766 |
* We use pg_always_inline to reduce function call overhead |
| 767 |
* and to help compilers to optimize away the 'is_csv' condition when called |
| 768 |
* by internal functions such as CopyFromTextLikeOneRow(). |
| 769 |
*/ |
| 770 |
static pg_always_inline bool |
| 771 |
NextCopyFromRawFieldsInternal(CopyFromState cstate, char ***fields, int *nfields, bool is_csv) |
| 772 |
{ |
| 773 |
int fldct; |
| 774 |
bool done; |
| 775 |
|
| 776 |
/* only available for text or csv input */ |
| 777 |
Assert(!cstate->opts.binary); |
| 778 |
|
| 779 |
/* on input check that the header line is correct if needed */ |
| 780 |
if (cstate->cur_lineno == 0 && cstate->opts.header_line) |
| 781 |
{ |
| 782 |
ListCell *cur; |
| 783 |
TupleDesc tupDesc; |
| 784 |
|
| 785 |
tupDesc = RelationGetDescr(cstate->rel); |
| 786 |
|
| 787 |
cstate->cur_lineno++; |
| 788 |
done = CopyReadLine(cstate, is_csv); |
| 789 |
|
| 790 |
if (cstate->opts.header_line == COPY_HEADER_MATCH) |
| 791 |
{ |
| 792 |
int fldnum; |
| 793 |
|
| 794 |
if (is_csv) |
| 795 |
fldct = CopyReadAttributesCSV(cstate); |
| 796 |
else |
| 797 |
fldct = CopyReadAttributesText(cstate); |
| 798 |
|
| 799 |
if (fldct != list_length(cstate->attnumlist)) |
| 800 |
ereport(ERROR, |
| 801 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 802 |
errmsg("wrong number of fields in header line: got %d, expected %d", |
| 803 |
fldct, list_length(cstate->attnumlist)))); |
| 804 |
|
| 805 |
fldnum = 0; |
| 806 |
foreach(cur, cstate->attnumlist) |
| 807 |
{ |
| 808 |
int attnum = lfirst_int(cur); |
| 809 |
char *colName; |
| 810 |
Form_pg_attribute attr = TupleDescAttr(tupDesc, attnum - 1); |
| 811 |
|
| 812 |
Assert(fldnum < cstate->max_fields); |
| 813 |
|
| 814 |
colName = cstate->raw_fields[fldnum++]; |
| 815 |
if (colName == NULL) |
| 816 |
ereport(ERROR, |
| 817 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 818 |
errmsg("column name mismatch in header line field %d: got null value (\"%s\"), expected \"%s\"", |
| 819 |
fldnum, cstate->opts.null_print, NameStr(attr->attname)))); |
| 820 |
|
| 821 |
if (namestrcmp(&attr->attname, colName) != 0) |
| 822 |
{ |
| 823 |
ereport(ERROR, |
| 824 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 825 |
errmsg("column name mismatch in header line field %d: got \"%s\", expected \"%s\"", |
| 826 |
fldnum, colName, NameStr(attr->attname)))); |
| 827 |
} |
| 828 |
} |
| 829 |
} |
| 830 |
|
| 831 |
if (done) |
| 832 |
return false; |
| 833 |
} |
| 834 |
|
| 835 |
cstate->cur_lineno++; |
| 836 |
|
| 837 |
/* Actually read the line into memory here */ |
| 838 |
done = CopyReadLine(cstate, is_csv); |
| 839 |
|
| 840 |
/* |
| 841 |
* EOF at start of line means we're done. If we see EOF after some |
| 842 |
* characters, we act as though it was newline followed by EOF, ie, |
| 843 |
* process the line and then exit loop on next iteration. |
| 844 |
*/ |
| 845 |
if (done && cstate->line_buf.len == 0) |
| 846 |
return false; |
| 847 |
|
| 848 |
/* Parse the line into de-escaped field values */ |
| 849 |
if (is_csv) |
| 850 |
fldct = CopyReadAttributesCSV(cstate); |
| 851 |
else |
| 852 |
fldct = CopyReadAttributesText(cstate); |
| 853 |
|
| 854 |
*fields = cstate->raw_fields; |
| 855 |
*nfields = fldct; |
| 856 |
return true; |
| 857 |
} |
| 858 |
|
| 859 |
/* |
| 860 |
* Read next tuple from file for COPY FROM. Return false if no more tuples. |
| 861 |
* |
| 862 |
* 'econtext' is used to evaluate default expression for each column that is |
| 863 |
* either not read from the file or is using the DEFAULT option of COPY FROM. |
| 864 |
* It can be NULL when no default values are used, i.e. when all columns are |
| 865 |
* read from the file, and DEFAULT option is unset. |
| 866 |
* |
| 867 |
* 'values' and 'nulls' arrays must be the same length as columns of the |
| 868 |
* relation passed to BeginCopyFrom. This function fills the arrays. |
| 869 |
*/ |
| 870 |
bool |
| 871 |
NextCopyFrom(CopyFromState cstate, ExprContext *econtext, |
| 872 |
Datum *values, bool *nulls) |
| 873 |
{ |
| 874 |
TupleDesc tupDesc; |
| 875 |
AttrNumber num_phys_attrs, |
| 876 |
num_defaults = cstate->num_defaults; |
| 877 |
int i; |
| 878 |
int *defmap = cstate->defmap; |
| 879 |
ExprState **defexprs = cstate->defexprs; |
| 880 |
|
| 881 |
tupDesc = RelationGetDescr(cstate->rel); |
| 882 |
num_phys_attrs = tupDesc->natts; |
| 883 |
|
| 884 |
/* Initialize all values for row to NULL */ |
| 885 |
MemSet(values, 0, num_phys_attrs * sizeof(Datum)); |
| 886 |
MemSet(nulls, true, num_phys_attrs * sizeof(bool)); |
| 887 |
MemSet(cstate->defaults, false, num_phys_attrs * sizeof(bool)); |
| 888 |
|
| 889 |
/* Get one row from source */ |
| 890 |
if (!cstate->routine->CopyFromOneRow(cstate, econtext, values, nulls)) |
| 891 |
return false; |
| 892 |
|
| 893 |
/* |
| 894 |
* Now compute and insert any defaults available for the columns not |
| 895 |
* provided by the input data. Anything not processed here or above will |
| 896 |
* remain NULL. |
| 897 |
*/ |
| 898 |
for (i = 0; i < num_defaults; i++) |
| 899 |
{ |
| 900 |
/* |
| 901 |
* The caller must supply econtext and have switched into the |
| 902 |
* per-tuple memory context in it. |
| 903 |
*/ |
| 904 |
Assert(econtext != NULL); |
| 905 |
Assert(CurrentMemoryContext == econtext->ecxt_per_tuple_memory); |
| 906 |
|
| 907 |
values[defmap[i]] = ExecEvalExpr(defexprs[defmap[i]], econtext, |
| 908 |
&nulls[defmap[i]]); |
| 909 |
} |
| 910 |
|
| 911 |
return true; |
| 912 |
} |
| 913 |
|
| 914 |
/* Implementation of the per-row callback for text format */ |
| 915 |
bool |
| 916 |
CopyFromTextOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values, |
| 917 |
bool *nulls) |
| 918 |
{ |
| 919 |
return CopyFromTextLikeOneRow(cstate, econtext, values, nulls, false); |
| 920 |
} |
| 921 |
|
| 922 |
/* Implementation of the per-row callback for CSV format */ |
| 923 |
bool |
| 924 |
CopyFromCSVOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values, |
| 925 |
bool *nulls) |
| 926 |
{ |
| 927 |
return CopyFromTextLikeOneRow(cstate, econtext, values, nulls, true); |
| 928 |
} |
| 929 |
|
| 930 |
/* |
| 931 |
* Workhorse for CopyFromTextOneRow() and CopyFromCSVOneRow(). |
| 932 |
* |
| 933 |
* We use pg_always_inline to reduce function call overhead |
| 934 |
* and to help compilers to optimize away the 'is_csv' condition. |
| 935 |
*/ |
| 936 |
static pg_always_inline bool |
| 937 |
CopyFromTextLikeOneRow(CopyFromState cstate, ExprContext *econtext, |
| 938 |
Datum *values, bool *nulls, bool is_csv) |
| 939 |
{ |
| 940 |
TupleDesc tupDesc; |
| 941 |
AttrNumber attr_count; |
| 942 |
FmgrInfo *in_functions = cstate->in_functions; |
| 943 |
Oid *typioparams = cstate->typioparams; |
| 944 |
ExprState **defexprs = cstate->defexprs; |
| 945 |
char **field_strings; |
| 946 |
ListCell *cur; |
| 947 |
int fldct; |
| 948 |
int fieldno; |
| 949 |
char *string; |
| 950 |
|
| 951 |
tupDesc = RelationGetDescr(cstate->rel); |
| 952 |
attr_count = list_length(cstate->attnumlist); |
| 953 |
|
| 954 |
/* read raw fields in the next line */ |
| 955 |
if (!NextCopyFromRawFieldsInternal(cstate, &field_strings, &fldct, is_csv)) |
| 956 |
return false; |
| 957 |
|
| 958 |
/* check for overflowing fields */ |
| 959 |
if (attr_count > 0 && fldct > attr_count) |
| 960 |
ereport(ERROR, |
| 961 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 962 |
errmsg("extra data after last expected column"))); |
| 963 |
|
| 964 |
fieldno = 0; |
| 965 |
|
| 966 |
/* Loop to read the user attributes on the line. */ |
| 967 |
foreach(cur, cstate->attnumlist) |
| 968 |
{ |
| 969 |
int attnum = lfirst_int(cur); |
| 970 |
int m = attnum - 1; |
| 971 |
Form_pg_attribute att = TupleDescAttr(tupDesc, m); |
| 972 |
|
| 973 |
if (fieldno >= fldct) |
| 974 |
ereport(ERROR, |
| 975 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 976 |
errmsg("missing data for column \"%s\"", |
| 977 |
NameStr(att->attname)))); |
| 978 |
string = field_strings[fieldno++]; |
| 979 |
|
| 980 |
if (cstate->convert_select_flags && |
| 981 |
!cstate->convert_select_flags[m]) |
| 982 |
{ |
| 983 |
/* ignore input field, leaving column as NULL */ |
| 984 |
continue; |
| 985 |
} |
| 986 |
|
| 987 |
if (is_csv) |
| 988 |
{ |
| 989 |
if (string == NULL && |
| 990 |
cstate->opts.force_notnull_flags[m]) |
| 991 |
{ |
| 992 |
/* |
| 993 |
* FORCE_NOT_NULL option is set and column is NULL - convert |
| 994 |
* it to the NULL string. |
| 995 |
*/ |
| 996 |
string = cstate->opts.null_print; |
| 997 |
} |
| 998 |
else if (string != NULL && cstate->opts.force_null_flags[m] |
| 999 |
&& strcmp(string, cstate->opts.null_print) == 0) |
| 1000 |
{ |
| 1001 |
/* |
| 1002 |
* FORCE_NULL option is set and column matches the NULL |
| 1003 |
* string. It must have been quoted, or otherwise the string |
| 1004 |
* would already have been set to NULL. Convert it to NULL as |
| 1005 |
* specified. |
| 1006 |
*/ |
| 1007 |
string = NULL; |
| 1008 |
} |
| 1009 |
} |
| 1010 |
|
| 1011 |
cstate->cur_attname = NameStr(att->attname); |
| 1012 |
cstate->cur_attval = string; |
| 1013 |
|
| 1014 |
if (string != NULL) |
| 1015 |
nulls[m] = false; |
| 1016 |
|
| 1017 |
if (cstate->defaults[m]) |
| 1018 |
{ |
| 1019 |
/* We must have switched into the per-tuple memory context */ |
| 1020 |
Assert(econtext != NULL); |
| 1021 |
Assert(CurrentMemoryContext == econtext->ecxt_per_tuple_memory); |
| 1022 |
|
| 1023 |
values[m] = ExecEvalExpr(defexprs[m], econtext, &nulls[m]); |
| 1024 |
} |
| 1025 |
|
| 1026 |
/* |
| 1027 |
* If ON_ERROR is specified with IGNORE, skip rows with soft errors |
| 1028 |
*/ |
| 1029 |
else if (!InputFunctionCallSafe(&in_functions[m], |
| 1030 |
string, |
| 1031 |
typioparams[m], |
| 1032 |
att->atttypmod, |
| 1033 |
(Node *) cstate->escontext, |
| 1034 |
&values[m])) |
| 1035 |
{ |
| 1036 |
Assert(cstate->opts.on_error != COPY_ON_ERROR_STOP); |
| 1037 |
|
| 1038 |
cstate->num_errors++; |
| 1039 |
|
| 1040 |
if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE) |
| 1041 |
{ |
| 1042 |
/* |
| 1043 |
* Since we emit line number and column info in the below |
| 1044 |
* notice message, we suppress error context information other |
| 1045 |
* than the relation name. |
| 1046 |
*/ |
| 1047 |
Assert(!cstate->relname_only); |
| 1048 |
cstate->relname_only = true; |
| 1049 |
|
| 1050 |
if (cstate->cur_attval) |
| 1051 |
{ |
| 1052 |
char *attval; |
| 1053 |
|
| 1054 |
attval = CopyLimitPrintoutLength(cstate->cur_attval); |
| 1055 |
ereport(NOTICE, |
| 1056 |
errmsg("skipping row due to data type incompatibility at line %" PRIu64 " for column \"%s\": \"%s\"", |
| 1057 |
cstate->cur_lineno, |
| 1058 |
cstate->cur_attname, |
| 1059 |
attval)); |
| 1060 |
pfree(attval); |
| 1061 |
} |
| 1062 |
else |
| 1063 |
ereport(NOTICE, |
| 1064 |
errmsg("skipping row due to data type incompatibility at line %" PRIu64 " for column \"%s\": null input", |
| 1065 |
cstate->cur_lineno, |
| 1066 |
cstate->cur_attname)); |
| 1067 |
|
| 1068 |
/* reset relname_only */ |
| 1069 |
cstate->relname_only = false; |
| 1070 |
} |
| 1071 |
|
| 1072 |
return true; |
| 1073 |
} |
| 1074 |
|
| 1075 |
cstate->cur_attname = NULL; |
| 1076 |
cstate->cur_attval = NULL; |
| 1077 |
} |
| 1078 |
|
| 1079 |
Assert(fieldno == attr_count); |
| 1080 |
|
| 1081 |
return true; |
| 1082 |
} |
| 1083 |
|
| 1084 |
/* Implementation of the per-row callback for binary format */ |
| 1085 |
bool |
| 1086 |
CopyFromBinaryOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values, |
| 1087 |
bool *nulls) |
| 1088 |
{ |
| 1089 |
TupleDesc tupDesc; |
| 1090 |
AttrNumber attr_count; |
| 1091 |
FmgrInfo *in_functions = cstate->in_functions; |
| 1092 |
Oid *typioparams = cstate->typioparams; |
| 1093 |
int16 fld_count; |
| 1094 |
ListCell *cur; |
| 1095 |
|
| 1096 |
tupDesc = RelationGetDescr(cstate->rel); |
| 1097 |
attr_count = list_length(cstate->attnumlist); |
| 1098 |
|
| 1099 |
cstate->cur_lineno++; |
| 1100 |
|
| 1101 |
if (!CopyGetInt16(cstate, &fld_count)) |
| 1102 |
{ |
| 1103 |
/* EOF detected (end of file, or protocol-level EOF) */ |
| 1104 |
return false; |
| 1105 |
} |
| 1106 |
|
| 1107 |
if (fld_count == -1) |
| 1108 |
{ |
| 1109 |
/* |
| 1110 |
* Received EOF marker. Wait for the protocol-level EOF, and complain |
| 1111 |
* if it doesn't come immediately. In COPY FROM STDIN, this ensures |
| 1112 |
* that we correctly handle CopyFail, if client chooses to send that |
| 1113 |
* now. When copying from file, we could ignore the rest of the file |
| 1114 |
* like in text mode, but we choose to be consistent with the COPY |
| 1115 |
* FROM STDIN case. |
| 1116 |
*/ |
| 1117 |
char dummy; |
| 1118 |
|
| 1119 |
if (CopyReadBinaryData(cstate, &dummy, 1) > 0) |
| 1120 |
ereport(ERROR, |
| 1121 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1122 |
errmsg("received copy data after EOF marker"))); |
| 1123 |
return false; |
| 1124 |
} |
| 1125 |
|
| 1126 |
if (fld_count != attr_count) |
| 1127 |
ereport(ERROR, |
| 1128 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1129 |
errmsg("row field count is %d, expected %d", |
| 1130 |
(int) fld_count, attr_count))); |
| 1131 |
|
| 1132 |
foreach(cur, cstate->attnumlist) |
| 1133 |
{ |
| 1134 |
int attnum = lfirst_int(cur); |
| 1135 |
int m = attnum - 1; |
| 1136 |
Form_pg_attribute att = TupleDescAttr(tupDesc, m); |
| 1137 |
|
| 1138 |
cstate->cur_attname = NameStr(att->attname); |
| 1139 |
values[m] = CopyReadBinaryAttribute(cstate, |
| 1140 |
&in_functions[m], |
| 1141 |
typioparams[m], |
| 1142 |
att->atttypmod, |
| 1143 |
&nulls[m]); |
| 1144 |
cstate->cur_attname = NULL; |
| 1145 |
} |
| 1146 |
|
| 1147 |
return true; |
| 1148 |
} |
| 1149 |
|
| 1150 |
/* |
| 1151 |
* Read the next input line and stash it in line_buf. |
| 1152 |
* |
| 1153 |
* Result is true if read was terminated by EOF, false if terminated |
| 1154 |
* by newline. The terminating newline or EOF marker is not included |
| 1155 |
* in the final value of line_buf. |
| 1156 |
*/ |
| 1157 |
static bool |
| 1158 |
CopyReadLine(CopyFromState cstate, bool is_csv) |
| 1159 |
{ |
| 1160 |
bool result; |
| 1161 |
|
| 1162 |
resetStringInfo(&cstate->line_buf); |
| 1163 |
cstate->line_buf_valid = false; |
| 1164 |
|
| 1165 |
/* Parse data and transfer into line_buf */ |
| 1166 |
result = CopyReadLineText(cstate, is_csv); |
| 1167 |
|
| 1168 |
if (result) |
| 1169 |
{ |
| 1170 |
/* |
| 1171 |
* Reached EOF. In protocol version 3, we should ignore anything |
| 1172 |
* after \. up to the protocol end of copy data. (XXX maybe better |
| 1173 |
* not to treat \. as special?) |
| 1174 |
*/ |
| 1175 |
if (cstate->copy_src == COPY_FRONTEND) |
| 1176 |
{ |
| 1177 |
int inbytes; |
| 1178 |
|
| 1179 |
do |
| 1180 |
{ |
| 1181 |
inbytes = CopyGetData(cstate, cstate->input_buf, |
| 1182 |
1, INPUT_BUF_SIZE); |
| 1183 |
} while (inbytes > 0); |
| 1184 |
cstate->input_buf_index = 0; |
| 1185 |
cstate->input_buf_len = 0; |
| 1186 |
cstate->raw_buf_index = 0; |
| 1187 |
cstate->raw_buf_len = 0; |
| 1188 |
} |
| 1189 |
} |
| 1190 |
else |
| 1191 |
{ |
| 1192 |
/* |
| 1193 |
* If we didn't hit EOF, then we must have transferred the EOL marker |
| 1194 |
* to line_buf along with the data. Get rid of it. |
| 1195 |
*/ |
| 1196 |
switch (cstate->eol_type) |
| 1197 |
{ |
| 1198 |
case EOL_NL: |
| 1199 |
Assert(cstate->line_buf.len >= 1); |
| 1200 |
Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\n'); |
| 1201 |
cstate->line_buf.len--; |
| 1202 |
cstate->line_buf.data[cstate->line_buf.len] = '\0'; |
| 1203 |
break; |
| 1204 |
case EOL_CR: |
| 1205 |
Assert(cstate->line_buf.len >= 1); |
| 1206 |
Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\r'); |
| 1207 |
cstate->line_buf.len--; |
| 1208 |
cstate->line_buf.data[cstate->line_buf.len] = '\0'; |
| 1209 |
break; |
| 1210 |
case EOL_CRNL: |
| 1211 |
Assert(cstate->line_buf.len >= 2); |
| 1212 |
Assert(cstate->line_buf.data[cstate->line_buf.len - 2] == '\r'); |
| 1213 |
Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\n'); |
| 1214 |
cstate->line_buf.len -= 2; |
| 1215 |
cstate->line_buf.data[cstate->line_buf.len] = '\0'; |
| 1216 |
break; |
| 1217 |
case EOL_UNKNOWN: |
| 1218 |
/* shouldn't get here */ |
| 1219 |
Assert(false); |
| 1220 |
break; |
| 1221 |
} |
| 1222 |
} |
| 1223 |
|
| 1224 |
/* Now it's safe to use the buffer in error messages */ |
| 1225 |
cstate->line_buf_valid = true; |
| 1226 |
|
| 1227 |
return result; |
| 1228 |
} |
| 1229 |
|
| 1230 |
/* |
| 1231 |
* CopyReadLineText - inner loop of CopyReadLine for text mode |
| 1232 |
*/ |
| 1233 |
static bool |
| 1234 |
CopyReadLineText(CopyFromState cstate, bool is_csv) |
| 1235 |
{ |
| 1236 |
char *copy_input_buf; |
| 1237 |
int input_buf_ptr; |
| 1238 |
int copy_buf_len; |
| 1239 |
bool need_data = false; |
| 1240 |
bool hit_eof = false; |
| 1241 |
bool result = false; |
| 1242 |
|
| 1243 |
/* CSV variables */ |
| 1244 |
bool in_quote = false, |
| 1245 |
last_was_esc = false; |
| 1246 |
char quotec = '\0'; |
| 1247 |
char escapec = '\0'; |
| 1248 |
|
| 1249 |
if (is_csv) |
| 1250 |
{ |
| 1251 |
quotec = cstate->opts.quote[0]; |
| 1252 |
escapec = cstate->opts.escape[0]; |
| 1253 |
/* ignore special escape processing if it's the same as quotec */ |
| 1254 |
if (quotec == escapec) |
| 1255 |
escapec = '\0'; |
| 1256 |
} |
| 1257 |
|
| 1258 |
/* |
| 1259 |
* The objective of this loop is to transfer the entire next input line |
| 1260 |
* into line_buf. Hence, we only care for detecting newlines (\r and/or |
| 1261 |
* \n) and the end-of-copy marker (\.). |
| 1262 |
* |
| 1263 |
* In CSV mode, \r and \n inside a quoted field are just part of the data |
| 1264 |
* value and are put in line_buf. We keep just enough state to know if we |
| 1265 |
* are currently in a quoted field or not. |
| 1266 |
* |
| 1267 |
* The input has already been converted to the database encoding. All |
| 1268 |
* supported server encodings have the property that all bytes in a |
| 1269 |
* multi-byte sequence have the high bit set, so a multibyte character |
| 1270 |
* cannot contain any newline or escape characters embedded in the |
| 1271 |
* multibyte sequence. Therefore, we can process the input byte-by-byte, |
| 1272 |
* regardless of the encoding. |
| 1273 |
* |
| 1274 |
* For speed, we try to move data from input_buf to line_buf in chunks |
| 1275 |
* rather than one character at a time. input_buf_ptr points to the next |
| 1276 |
* character to examine; any characters from input_buf_index to |
| 1277 |
* input_buf_ptr have been determined to be part of the line, but not yet |
| 1278 |
* transferred to line_buf. |
| 1279 |
* |
| 1280 |
* For a little extra speed within the loop, we copy input_buf and |
| 1281 |
* input_buf_len into local variables. |
| 1282 |
*/ |
| 1283 |
copy_input_buf = cstate->input_buf; |
| 1284 |
input_buf_ptr = cstate->input_buf_index; |
| 1285 |
copy_buf_len = cstate->input_buf_len; |
| 1286 |
|
| 1287 |
for (;;) |
| 1288 |
{ |
| 1289 |
int prev_raw_ptr; |
| 1290 |
char c; |
| 1291 |
|
| 1292 |
/* |
| 1293 |
* Load more data if needed. |
| 1294 |
* |
| 1295 |
* TODO: We could just force four bytes of read-ahead and avoid the |
| 1296 |
* many calls to IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(). That was |
| 1297 |
* unsafe with the old v2 COPY protocol, but we don't support that |
| 1298 |
* anymore. |
| 1299 |
*/ |
| 1300 |
if (input_buf_ptr >= copy_buf_len || need_data) |
| 1301 |
{ |
| 1302 |
REFILL_LINEBUF; |
| 1303 |
|
| 1304 |
CopyLoadInputBuf(cstate); |
| 1305 |
/* update our local variables */ |
| 1306 |
hit_eof = cstate->input_reached_eof; |
| 1307 |
input_buf_ptr = cstate->input_buf_index; |
| 1308 |
copy_buf_len = cstate->input_buf_len; |
| 1309 |
|
| 1310 |
/* |
| 1311 |
* If we are completely out of data, break out of the loop, |
| 1312 |
* reporting EOF. |
| 1313 |
*/ |
| 1314 |
if (INPUT_BUF_BYTES(cstate) <= 0) |
| 1315 |
{ |
| 1316 |
result = true; |
| 1317 |
break; |
| 1318 |
} |
| 1319 |
need_data = false; |
| 1320 |
} |
| 1321 |
|
| 1322 |
/* OK to fetch a character */ |
| 1323 |
prev_raw_ptr = input_buf_ptr; |
| 1324 |
c = copy_input_buf[input_buf_ptr++]; |
| 1325 |
|
| 1326 |
if (is_csv) |
| 1327 |
{ |
| 1328 |
/* |
| 1329 |
* If character is '\r', we may need to look ahead below. Force |
| 1330 |
* fetch of the next character if we don't already have it. We |
| 1331 |
* need to do this before changing CSV state, in case '\r' is also |
| 1332 |
* the quote or escape character. |
| 1333 |
*/ |
| 1334 |
if (c == '\r') |
| 1335 |
{ |
| 1336 |
IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0); |
| 1337 |
} |
| 1338 |
|
| 1339 |
/* |
| 1340 |
* Dealing with quotes and escapes here is mildly tricky. If the |
| 1341 |
* quote char is also the escape char, there's no problem - we |
| 1342 |
* just use the char as a toggle. If they are different, we need |
| 1343 |
* to ensure that we only take account of an escape inside a |
| 1344 |
* quoted field and immediately preceding a quote char, and not |
| 1345 |
* the second in an escape-escape sequence. |
| 1346 |
*/ |
| 1347 |
if (in_quote && c == escapec) |
| 1348 |
last_was_esc = !last_was_esc; |
| 1349 |
if (c == quotec && !last_was_esc) |
| 1350 |
in_quote = !in_quote; |
| 1351 |
if (c != escapec) |
| 1352 |
last_was_esc = false; |
| 1353 |
|
| 1354 |
/* |
| 1355 |
* Updating the line count for embedded CR and/or LF chars is |
| 1356 |
* necessarily a little fragile - this test is probably about the |
| 1357 |
* best we can do. (XXX it's arguable whether we should do this |
| 1358 |
* at all --- is cur_lineno a physical or logical count?) |
| 1359 |
*/ |
| 1360 |
if (in_quote && c == (cstate->eol_type == EOL_NL ? '\n' : '\r')) |
| 1361 |
cstate->cur_lineno++; |
| 1362 |
} |
| 1363 |
|
| 1364 |
/* Process \r */ |
| 1365 |
if (c == '\r' && (!is_csv || !in_quote)) |
| 1366 |
{ |
| 1367 |
/* Check for \r\n on first line, _and_ handle \r\n. */ |
| 1368 |
if (cstate->eol_type == EOL_UNKNOWN || |
| 1369 |
cstate->eol_type == EOL_CRNL) |
| 1370 |
{ |
| 1371 |
/* |
| 1372 |
* If need more data, go back to loop top to load it. |
| 1373 |
* |
| 1374 |
* Note that if we are at EOF, c will wind up as '\0' because |
| 1375 |
* of the guaranteed pad of input_buf. |
| 1376 |
*/ |
| 1377 |
IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0); |
| 1378 |
|
| 1379 |
/* get next char */ |
| 1380 |
c = copy_input_buf[input_buf_ptr]; |
| 1381 |
|
| 1382 |
if (c == '\n') |
| 1383 |
{ |
| 1384 |
input_buf_ptr++; /* eat newline */ |
| 1385 |
cstate->eol_type = EOL_CRNL; /* in case not set yet */ |
| 1386 |
} |
| 1387 |
else |
| 1388 |
{ |
| 1389 |
/* found \r, but no \n */ |
| 1390 |
if (cstate->eol_type == EOL_CRNL) |
| 1391 |
ereport(ERROR, |
| 1392 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1393 |
!is_csv ? |
| 1394 |
errmsg("literal carriage return found in data") : |
| 1395 |
errmsg("unquoted carriage return found in data"), |
| 1396 |
!is_csv ? |
| 1397 |
errhint("Use \"\\r\" to represent carriage return.") : |
| 1398 |
errhint("Use quoted CSV field to represent carriage return."))); |
| 1399 |
|
| 1400 |
/* |
| 1401 |
* if we got here, it is the first line and we didn't find |
| 1402 |
* \n, so don't consume the peeked character |
| 1403 |
*/ |
| 1404 |
cstate->eol_type = EOL_CR; |
| 1405 |
} |
| 1406 |
} |
| 1407 |
else if (cstate->eol_type == EOL_NL) |
| 1408 |
ereport(ERROR, |
| 1409 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1410 |
!is_csv ? |
| 1411 |
errmsg("literal carriage return found in data") : |
| 1412 |
errmsg("unquoted carriage return found in data"), |
| 1413 |
!is_csv ? |
| 1414 |
errhint("Use \"\\r\" to represent carriage return.") : |
| 1415 |
errhint("Use quoted CSV field to represent carriage return."))); |
| 1416 |
/* If reach here, we have found the line terminator */ |
| 1417 |
break; |
| 1418 |
} |
| 1419 |
|
| 1420 |
/* Process \n */ |
| 1421 |
if (c == '\n' && (!is_csv || !in_quote)) |
| 1422 |
{ |
| 1423 |
if (cstate->eol_type == EOL_CR || cstate->eol_type == EOL_CRNL) |
| 1424 |
ereport(ERROR, |
| 1425 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1426 |
!is_csv ? |
| 1427 |
errmsg("literal newline found in data") : |
| 1428 |
errmsg("unquoted newline found in data"), |
| 1429 |
!is_csv ? |
| 1430 |
errhint("Use \"\\n\" to represent newline.") : |
| 1431 |
errhint("Use quoted CSV field to represent newline."))); |
| 1432 |
cstate->eol_type = EOL_NL; /* in case not set yet */ |
| 1433 |
/* If reach here, we have found the line terminator */ |
| 1434 |
break; |
| 1435 |
} |
| 1436 |
|
| 1437 |
/* |
| 1438 |
* Process backslash, except in CSV mode where backslash is a normal |
| 1439 |
* character. |
| 1440 |
*/ |
| 1441 |
if (c == '\\' && !is_csv) |
| 1442 |
{ |
| 1443 |
char c2; |
| 1444 |
|
| 1445 |
IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0); |
| 1446 |
IF_NEED_REFILL_AND_EOF_BREAK(0); |
| 1447 |
|
| 1448 |
/* ----- |
| 1449 |
* get next character |
| 1450 |
* Note: we do not change c so if it isn't \., we can fall |
| 1451 |
* through and continue processing. |
| 1452 |
* ----- |
| 1453 |
*/ |
| 1454 |
c2 = copy_input_buf[input_buf_ptr]; |
| 1455 |
|
| 1456 |
if (c2 == '.') |
| 1457 |
{ |
| 1458 |
input_buf_ptr++; /* consume the '.' */ |
| 1459 |
if (cstate->eol_type == EOL_CRNL) |
| 1460 |
{ |
| 1461 |
/* Get the next character */ |
| 1462 |
IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0); |
| 1463 |
/* if hit_eof, c2 will become '\0' */ |
| 1464 |
c2 = copy_input_buf[input_buf_ptr++]; |
| 1465 |
|
| 1466 |
if (c2 == '\n') |
| 1467 |
ereport(ERROR, |
| 1468 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1469 |
errmsg("end-of-copy marker does not match previous newline style"))); |
| 1470 |
else if (c2 != '\r') |
| 1471 |
ereport(ERROR, |
| 1472 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1473 |
errmsg("end-of-copy marker is not alone on its line"))); |
| 1474 |
} |
| 1475 |
|
| 1476 |
/* Get the next character */ |
| 1477 |
IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0); |
| 1478 |
/* if hit_eof, c2 will become '\0' */ |
| 1479 |
c2 = copy_input_buf[input_buf_ptr++]; |
| 1480 |
|
| 1481 |
if (c2 != '\r' && c2 != '\n') |
| 1482 |
ereport(ERROR, |
| 1483 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1484 |
errmsg("end-of-copy marker is not alone on its line"))); |
| 1485 |
|
| 1486 |
if ((cstate->eol_type == EOL_NL && c2 != '\n') || |
| 1487 |
(cstate->eol_type == EOL_CRNL && c2 != '\n') || |
| 1488 |
(cstate->eol_type == EOL_CR && c2 != '\r')) |
| 1489 |
ereport(ERROR, |
| 1490 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1491 |
errmsg("end-of-copy marker does not match previous newline style"))); |
| 1492 |
|
| 1493 |
/* |
| 1494 |
* If there is any data on this line before the \., complain. |
| 1495 |
*/ |
| 1496 |
if (cstate->line_buf.len > 0 || |
| 1497 |
prev_raw_ptr > cstate->input_buf_index) |
| 1498 |
ereport(ERROR, |
| 1499 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1500 |
errmsg("end-of-copy marker is not alone on its line"))); |
| 1501 |
|
| 1502 |
/* |
| 1503 |
* Discard the \. and newline, then report EOF. |
| 1504 |
*/ |
| 1505 |
cstate->input_buf_index = input_buf_ptr; |
| 1506 |
result = true; /* report EOF */ |
| 1507 |
break; |
| 1508 |
} |
| 1509 |
else |
| 1510 |
{ |
| 1511 |
/* |
| 1512 |
* If we are here, it means we found a backslash followed by |
| 1513 |
* something other than a period. In non-CSV mode, anything |
| 1514 |
* after a backslash is special, so we skip over that second |
| 1515 |
* character too. If we didn't do that \\. would be |
| 1516 |
* considered an eof-of copy, while in non-CSV mode it is a |
| 1517 |
* literal backslash followed by a period. |
| 1518 |
*/ |
| 1519 |
input_buf_ptr++; |
| 1520 |
} |
| 1521 |
} |
| 1522 |
} /* end of outer loop */ |
| 1523 |
|
| 1524 |
/* |
| 1525 |
* Transfer any still-uncopied data to line_buf. |
| 1526 |
*/ |
| 1527 |
REFILL_LINEBUF; |
| 1528 |
|
| 1529 |
return result; |
| 1530 |
} |
| 1531 |
|
| 1532 |
/* |
| 1533 |
* Return decimal value for a hexadecimal digit |
| 1534 |
*/ |
| 1535 |
static int |
| 1536 |
GetDecimalFromHex(char hex) |
| 1537 |
{ |
| 1538 |
if (isdigit((unsigned char) hex)) |
| 1539 |
return hex - '0'; |
| 1540 |
else |
| 1541 |
return tolower((unsigned char) hex) - 'a' + 10; |
| 1542 |
} |
| 1543 |
|
| 1544 |
/* |
| 1545 |
* Parse the current line into separate attributes (fields), |
| 1546 |
* performing de-escaping as needed. |
| 1547 |
* |
| 1548 |
* The input is in line_buf. We use attribute_buf to hold the result |
| 1549 |
* strings. cstate->raw_fields[k] is set to point to the k'th attribute |
| 1550 |
* string, or NULL when the input matches the null marker string. |
| 1551 |
* This array is expanded as necessary. |
| 1552 |
* |
| 1553 |
* (Note that the caller cannot check for nulls since the returned |
| 1554 |
* string would be the post-de-escaping equivalent, which may look |
| 1555 |
* the same as some valid data string.) |
| 1556 |
* |
| 1557 |
* delim is the column delimiter string (must be just one byte for now). |
| 1558 |
* null_print is the null marker string. Note that this is compared to |
| 1559 |
* the pre-de-escaped input string. |
| 1560 |
* |
| 1561 |
* The return value is the number of fields actually read. |
| 1562 |
*/ |
| 1563 |
static int |
| 1564 |
CopyReadAttributesText(CopyFromState cstate) |
| 1565 |
{ |
| 1566 |
char delimc = cstate->opts.delim[0]; |
| 1567 |
int fieldno; |
| 1568 |
char *output_ptr; |
| 1569 |
char *cur_ptr; |
| 1570 |
char *line_end_ptr; |
| 1571 |
|
| 1572 |
/* |
| 1573 |
* We need a special case for zero-column tables: check that the input |
| 1574 |
* line is empty, and return. |
| 1575 |
*/ |
| 1576 |
if (cstate->max_fields <= 0) |
| 1577 |
{ |
| 1578 |
if (cstate->line_buf.len != 0) |
| 1579 |
ereport(ERROR, |
| 1580 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1581 |
errmsg("extra data after last expected column"))); |
| 1582 |
return 0; |
| 1583 |
} |
| 1584 |
|
| 1585 |
resetStringInfo(&cstate->attribute_buf); |
| 1586 |
|
| 1587 |
/* |
| 1588 |
* The de-escaped attributes will certainly not be longer than the input |
| 1589 |
* data line, so we can just force attribute_buf to be large enough and |
| 1590 |
* then transfer data without any checks for enough space. We need to do |
| 1591 |
* it this way because enlarging attribute_buf mid-stream would invalidate |
| 1592 |
* pointers already stored into cstate->raw_fields[]. |
| 1593 |
*/ |
| 1594 |
if (cstate->attribute_buf.maxlen <= cstate->line_buf.len) |
| 1595 |
enlargeStringInfo(&cstate->attribute_buf, cstate->line_buf.len); |
| 1596 |
output_ptr = cstate->attribute_buf.data; |
| 1597 |
|
| 1598 |
/* set pointer variables for loop */ |
| 1599 |
cur_ptr = cstate->line_buf.data; |
| 1600 |
line_end_ptr = cstate->line_buf.data + cstate->line_buf.len; |
| 1601 |
|
| 1602 |
/* Outer loop iterates over fields */ |
| 1603 |
fieldno = 0; |
| 1604 |
for (;;) |
| 1605 |
{ |
| 1606 |
bool found_delim = false; |
| 1607 |
char *start_ptr; |
| 1608 |
char *end_ptr; |
| 1609 |
int input_len; |
| 1610 |
bool saw_non_ascii = false; |
| 1611 |
|
| 1612 |
/* Make sure there is enough space for the next value */ |
| 1613 |
if (fieldno >= cstate->max_fields) |
| 1614 |
{ |
| 1615 |
cstate->max_fields *= 2; |
| 1616 |
cstate->raw_fields = |
| 1617 |
repalloc(cstate->raw_fields, cstate->max_fields * sizeof(char *)); |
| 1618 |
} |
| 1619 |
|
| 1620 |
/* Remember start of field on both input and output sides */ |
| 1621 |
start_ptr = cur_ptr; |
| 1622 |
cstate->raw_fields[fieldno] = output_ptr; |
| 1623 |
|
| 1624 |
/* |
| 1625 |
* Scan data for field. |
| 1626 |
* |
| 1627 |
* Note that in this loop, we are scanning to locate the end of field |
| 1628 |
* and also speculatively performing de-escaping. Once we find the |
| 1629 |
* end-of-field, we can match the raw field contents against the null |
| 1630 |
* marker string. Only after that comparison fails do we know that |
| 1631 |
* de-escaping is actually the right thing to do; therefore we *must |
| 1632 |
* not* throw any syntax errors before we've done the null-marker |
| 1633 |
* check. |
| 1634 |
*/ |
| 1635 |
for (;;) |
| 1636 |
{ |
| 1637 |
char c; |
| 1638 |
|
| 1639 |
end_ptr = cur_ptr; |
| 1640 |
if (cur_ptr >= line_end_ptr) |
| 1641 |
break; |
| 1642 |
c = *cur_ptr++; |
| 1643 |
if (c == delimc) |
| 1644 |
{ |
| 1645 |
found_delim = true; |
| 1646 |
break; |
| 1647 |
} |
| 1648 |
if (c == '\\') |
| 1649 |
{ |
| 1650 |
if (cur_ptr >= line_end_ptr) |
| 1651 |
break; |
| 1652 |
c = *cur_ptr++; |
| 1653 |
switch (c) |
| 1654 |
{ |
| 1655 |
case '0': |
| 1656 |
case '1': |
| 1657 |
case '2': |
| 1658 |
case '3': |
| 1659 |
case '4': |
| 1660 |
case '5': |
| 1661 |
case '6': |
| 1662 |
case '7': |
| 1663 |
{ |
| 1664 |
/* handle \013 */ |
| 1665 |
int val; |
| 1666 |
|
| 1667 |
val = OCTVALUE(c); |
| 1668 |
if (cur_ptr < line_end_ptr) |
| 1669 |
{ |
| 1670 |
c = *cur_ptr; |
| 1671 |
if (ISOCTAL(c)) |
| 1672 |
{ |
| 1673 |
cur_ptr++; |
| 1674 |
val = (val << 3) + OCTVALUE(c); |
| 1675 |
if (cur_ptr < line_end_ptr) |
| 1676 |
{ |
| 1677 |
c = *cur_ptr; |
| 1678 |
if (ISOCTAL(c)) |
| 1679 |
{ |
| 1680 |
cur_ptr++; |
| 1681 |
val = (val << 3) + OCTVALUE(c); |
| 1682 |
} |
| 1683 |
} |
| 1684 |
} |
| 1685 |
} |
| 1686 |
c = val & 0377; |
| 1687 |
if (c == '\0' || IS_HIGHBIT_SET(c)) |
| 1688 |
saw_non_ascii = true; |
| 1689 |
} |
| 1690 |
break; |
| 1691 |
case 'x': |
| 1692 |
/* Handle \x3F */ |
| 1693 |
if (cur_ptr < line_end_ptr) |
| 1694 |
{ |
| 1695 |
char hexchar = *cur_ptr; |
| 1696 |
|
| 1697 |
if (isxdigit((unsigned char) hexchar)) |
| 1698 |
{ |
| 1699 |
int val = GetDecimalFromHex(hexchar); |
| 1700 |
|
| 1701 |
cur_ptr++; |
| 1702 |
if (cur_ptr < line_end_ptr) |
| 1703 |
{ |
| 1704 |
hexchar = *cur_ptr; |
| 1705 |
if (isxdigit((unsigned char) hexchar)) |
| 1706 |
{ |
| 1707 |
cur_ptr++; |
| 1708 |
val = (val << 4) + GetDecimalFromHex(hexchar); |
| 1709 |
} |
| 1710 |
} |
| 1711 |
c = val & 0xff; |
| 1712 |
if (c == '\0' || IS_HIGHBIT_SET(c)) |
| 1713 |
saw_non_ascii = true; |
| 1714 |
} |
| 1715 |
} |
| 1716 |
break; |
| 1717 |
case 'b': |
| 1718 |
c = '\b'; |
| 1719 |
break; |
| 1720 |
case 'f': |
| 1721 |
c = '\f'; |
| 1722 |
break; |
| 1723 |
case 'n': |
| 1724 |
c = '\n'; |
| 1725 |
break; |
| 1726 |
case 'r': |
| 1727 |
c = '\r'; |
| 1728 |
break; |
| 1729 |
case 't': |
| 1730 |
c = '\t'; |
| 1731 |
break; |
| 1732 |
case 'v': |
| 1733 |
c = '\v'; |
| 1734 |
break; |
| 1735 |
|
| 1736 |
/* |
| 1737 |
* in all other cases, take the char after '\' |
| 1738 |
* literally |
| 1739 |
*/ |
| 1740 |
} |
| 1741 |
} |
| 1742 |
|
| 1743 |
/* Add c to output string */ |
| 1744 |
*output_ptr++ = c; |
| 1745 |
} |
| 1746 |
|
| 1747 |
/* Check whether raw input matched null marker */ |
| 1748 |
input_len = end_ptr - start_ptr; |
| 1749 |
if (input_len == cstate->opts.null_print_len && |
| 1750 |
strncmp(start_ptr, cstate->opts.null_print, input_len) == 0) |
| 1751 |
cstate->raw_fields[fieldno] = NULL; |
| 1752 |
/* Check whether raw input matched default marker */ |
| 1753 |
else if (fieldno < list_length(cstate->attnumlist) && |
| 1754 |
cstate->opts.default_print && |
| 1755 |
input_len == cstate->opts.default_print_len && |
| 1756 |
strncmp(start_ptr, cstate->opts.default_print, input_len) == 0) |
| 1757 |
{ |
| 1758 |
/* fieldno is 0-indexed and attnum is 1-indexed */ |
| 1759 |
int m = list_nth_int(cstate->attnumlist, fieldno) - 1; |
| 1760 |
|
| 1761 |
if (cstate->defexprs[m] != NULL) |
| 1762 |
{ |
| 1763 |
/* defaults contain entries for all physical attributes */ |
| 1764 |
cstate->defaults[m] = true; |
| 1765 |
} |
| 1766 |
else |
| 1767 |
{ |
| 1768 |
TupleDesc tupDesc = RelationGetDescr(cstate->rel); |
| 1769 |
Form_pg_attribute att = TupleDescAttr(tupDesc, m); |
| 1770 |
|
| 1771 |
ereport(ERROR, |
| 1772 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1773 |
errmsg("unexpected default marker in COPY data"), |
| 1774 |
errdetail("Column \"%s\" has no default value.", |
| 1775 |
NameStr(att->attname)))); |
| 1776 |
} |
| 1777 |
} |
| 1778 |
else |
| 1779 |
{ |
| 1780 |
/* |
| 1781 |
* At this point we know the field is supposed to contain data. |
| 1782 |
* |
| 1783 |
* If we de-escaped any non-7-bit-ASCII chars, make sure the |
| 1784 |
* resulting string is valid data for the db encoding. |
| 1785 |
*/ |
| 1786 |
if (saw_non_ascii) |
| 1787 |
{ |
| 1788 |
char *fld = cstate->raw_fields[fieldno]; |
| 1789 |
|
| 1790 |
pg_verifymbstr(fld, output_ptr - fld, false); |
| 1791 |
} |
| 1792 |
} |
| 1793 |
|
| 1794 |
/* Terminate attribute value in output area */ |
| 1795 |
*output_ptr++ = '\0'; |
| 1796 |
|
| 1797 |
fieldno++; |
| 1798 |
/* Done if we hit EOL instead of a delim */ |
| 1799 |
if (!found_delim) |
| 1800 |
break; |
| 1801 |
} |
| 1802 |
|
| 1803 |
/* Clean up state of attribute_buf */ |
| 1804 |
output_ptr--; |
| 1805 |
Assert(*output_ptr == '\0'); |
| 1806 |
cstate->attribute_buf.len = (output_ptr - cstate->attribute_buf.data); |
| 1807 |
|
| 1808 |
return fieldno; |
| 1809 |
} |
| 1810 |
|
| 1811 |
/* |
| 1812 |
* Parse the current line into separate attributes (fields), |
| 1813 |
* performing de-escaping as needed. This has exactly the same API as |
| 1814 |
* CopyReadAttributesText, except we parse the fields according to |
| 1815 |
* "standard" (i.e. common) CSV usage. |
| 1816 |
*/ |
| 1817 |
static int |
| 1818 |
CopyReadAttributesCSV(CopyFromState cstate) |
| 1819 |
{ |
| 1820 |
char delimc = cstate->opts.delim[0]; |
| 1821 |
char quotec = cstate->opts.quote[0]; |
| 1822 |
char escapec = cstate->opts.escape[0]; |
| 1823 |
int fieldno; |
| 1824 |
char *output_ptr; |
| 1825 |
char *cur_ptr; |
| 1826 |
char *line_end_ptr; |
| 1827 |
|
| 1828 |
/* |
| 1829 |
* We need a special case for zero-column tables: check that the input |
| 1830 |
* line is empty, and return. |
| 1831 |
*/ |
| 1832 |
if (cstate->max_fields <= 0) |
| 1833 |
{ |
| 1834 |
if (cstate->line_buf.len != 0) |
| 1835 |
ereport(ERROR, |
| 1836 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1837 |
errmsg("extra data after last expected column"))); |
| 1838 |
return 0; |
| 1839 |
} |
| 1840 |
|
| 1841 |
resetStringInfo(&cstate->attribute_buf); |
| 1842 |
|
| 1843 |
/* |
| 1844 |
* The de-escaped attributes will certainly not be longer than the input |
| 1845 |
* data line, so we can just force attribute_buf to be large enough and |
| 1846 |
* then transfer data without any checks for enough space. We need to do |
| 1847 |
* it this way because enlarging attribute_buf mid-stream would invalidate |
| 1848 |
* pointers already stored into cstate->raw_fields[]. |
| 1849 |
*/ |
| 1850 |
if (cstate->attribute_buf.maxlen <= cstate->line_buf.len) |
| 1851 |
enlargeStringInfo(&cstate->attribute_buf, cstate->line_buf.len); |
| 1852 |
output_ptr = cstate->attribute_buf.data; |
| 1853 |
|
| 1854 |
/* set pointer variables for loop */ |
| 1855 |
cur_ptr = cstate->line_buf.data; |
| 1856 |
line_end_ptr = cstate->line_buf.data + cstate->line_buf.len; |
| 1857 |
|
| 1858 |
/* Outer loop iterates over fields */ |
| 1859 |
fieldno = 0; |
| 1860 |
for (;;) |
| 1861 |
{ |
| 1862 |
bool found_delim = false; |
| 1863 |
bool saw_quote = false; |
| 1864 |
char *start_ptr; |
| 1865 |
char *end_ptr; |
| 1866 |
int input_len; |
| 1867 |
|
| 1868 |
/* Make sure there is enough space for the next value */ |
| 1869 |
if (fieldno >= cstate->max_fields) |
| 1870 |
{ |
| 1871 |
cstate->max_fields *= 2; |
| 1872 |
cstate->raw_fields = |
| 1873 |
repalloc(cstate->raw_fields, cstate->max_fields * sizeof(char *)); |
| 1874 |
} |
| 1875 |
|
| 1876 |
/* Remember start of field on both input and output sides */ |
| 1877 |
start_ptr = cur_ptr; |
| 1878 |
cstate->raw_fields[fieldno] = output_ptr; |
| 1879 |
|
| 1880 |
/* |
| 1881 |
* Scan data for field, |
| 1882 |
* |
| 1883 |
* The loop starts in "not quote" mode and then toggles between that |
| 1884 |
* and "in quote" mode. The loop exits normally if it is in "not |
| 1885 |
* quote" mode and a delimiter or line end is seen. |
| 1886 |
*/ |
| 1887 |
for (;;) |
| 1888 |
{ |
| 1889 |
char c; |
| 1890 |
|
| 1891 |
/* Not in quote */ |
| 1892 |
for (;;) |
| 1893 |
{ |
| 1894 |
end_ptr = cur_ptr; |
| 1895 |
if (cur_ptr >= line_end_ptr) |
| 1896 |
goto endfield; |
| 1897 |
c = *cur_ptr++; |
| 1898 |
/* unquoted field delimiter */ |
| 1899 |
if (c == delimc) |
| 1900 |
{ |
| 1901 |
found_delim = true; |
| 1902 |
goto endfield; |
| 1903 |
} |
| 1904 |
/* start of quoted field (or part of field) */ |
| 1905 |
if (c == quotec) |
| 1906 |
{ |
| 1907 |
saw_quote = true; |
| 1908 |
break; |
| 1909 |
} |
| 1910 |
/* Add c to output string */ |
| 1911 |
*output_ptr++ = c; |
| 1912 |
} |
| 1913 |
|
| 1914 |
/* In quote */ |
| 1915 |
for (;;) |
| 1916 |
{ |
| 1917 |
end_ptr = cur_ptr; |
| 1918 |
if (cur_ptr >= line_end_ptr) |
| 1919 |
ereport(ERROR, |
| 1920 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1921 |
errmsg("unterminated CSV quoted field"))); |
| 1922 |
|
| 1923 |
c = *cur_ptr++; |
| 1924 |
|
| 1925 |
/* escape within a quoted field */ |
| 1926 |
if (c == escapec) |
| 1927 |
{ |
| 1928 |
/* |
| 1929 |
* peek at the next char if available, and escape it if it |
| 1930 |
* is an escape char or a quote char |
| 1931 |
*/ |
| 1932 |
if (cur_ptr < line_end_ptr) |
| 1933 |
{ |
| 1934 |
char nextc = *cur_ptr; |
| 1935 |
|
| 1936 |
if (nextc == escapec || nextc == quotec) |
| 1937 |
{ |
| 1938 |
*output_ptr++ = nextc; |
| 1939 |
cur_ptr++; |
| 1940 |
continue; |
| 1941 |
} |
| 1942 |
} |
| 1943 |
} |
| 1944 |
|
| 1945 |
/* |
| 1946 |
* end of quoted field. Must do this test after testing for |
| 1947 |
* escape in case quote char and escape char are the same |
| 1948 |
* (which is the common case). |
| 1949 |
*/ |
| 1950 |
if (c == quotec) |
| 1951 |
break; |
| 1952 |
|
| 1953 |
/* Add c to output string */ |
| 1954 |
*output_ptr++ = c; |
| 1955 |
} |
| 1956 |
} |
| 1957 |
endfield: |
| 1958 |
|
| 1959 |
/* Terminate attribute value in output area */ |
| 1960 |
*output_ptr++ = '\0'; |
| 1961 |
|
| 1962 |
/* Check whether raw input matched null marker */ |
| 1963 |
input_len = end_ptr - start_ptr; |
| 1964 |
if (!saw_quote && input_len == cstate->opts.null_print_len && |
| 1965 |
strncmp(start_ptr, cstate->opts.null_print, input_len) == 0) |
| 1966 |
cstate->raw_fields[fieldno] = NULL; |
| 1967 |
/* Check whether raw input matched default marker */ |
| 1968 |
else if (fieldno < list_length(cstate->attnumlist) && |
| 1969 |
cstate->opts.default_print && |
| 1970 |
input_len == cstate->opts.default_print_len && |
| 1971 |
strncmp(start_ptr, cstate->opts.default_print, input_len) == 0) |
| 1972 |
{ |
| 1973 |
/* fieldno is 0-index and attnum is 1-index */ |
| 1974 |
int m = list_nth_int(cstate->attnumlist, fieldno) - 1; |
| 1975 |
|
| 1976 |
if (cstate->defexprs[m] != NULL) |
| 1977 |
{ |
| 1978 |
/* defaults contain entries for all physical attributes */ |
| 1979 |
cstate->defaults[m] = true; |
| 1980 |
} |
| 1981 |
else |
| 1982 |
{ |
| 1983 |
TupleDesc tupDesc = RelationGetDescr(cstate->rel); |
| 1984 |
Form_pg_attribute att = TupleDescAttr(tupDesc, m); |
| 1985 |
|
| 1986 |
ereport(ERROR, |
| 1987 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 1988 |
errmsg("unexpected default marker in COPY data"), |
| 1989 |
errdetail("Column \"%s\" has no default value.", |
| 1990 |
NameStr(att->attname)))); |
| 1991 |
} |
| 1992 |
} |
| 1993 |
|
| 1994 |
fieldno++; |
| 1995 |
/* Done if we hit EOL instead of a delim */ |
| 1996 |
if (!found_delim) |
| 1997 |
break; |
| 1998 |
} |
| 1999 |
|
| 2000 |
/* Clean up state of attribute_buf */ |
| 2001 |
output_ptr--; |
| 2002 |
Assert(*output_ptr == '\0'); |
| 2003 |
cstate->attribute_buf.len = (output_ptr - cstate->attribute_buf.data); |
| 2004 |
|
| 2005 |
return fieldno; |
| 2006 |
} |
| 2007 |
|
| 2008 |
|
| 2009 |
/* |
| 2010 |
* Read a binary attribute |
| 2011 |
*/ |
| 2012 |
static Datum |
| 2013 |
CopyReadBinaryAttribute(CopyFromState cstate, FmgrInfo *flinfo, |
| 2014 |
Oid typioparam, int32 typmod, |
| 2015 |
bool *isnull) |
| 2016 |
{ |
| 2017 |
int32 fld_size; |
| 2018 |
Datum result; |
| 2019 |
|
| 2020 |
if (!CopyGetInt32(cstate, &fld_size)) |
| 2021 |
ereport(ERROR, |
| 2022 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 2023 |
errmsg("unexpected EOF in COPY data"))); |
| 2024 |
if (fld_size == -1) |
| 2025 |
{ |
| 2026 |
*isnull = true; |
| 2027 |
return ReceiveFunctionCall(flinfo, NULL, typioparam, typmod); |
| 2028 |
} |
| 2029 |
if (fld_size < 0) |
| 2030 |
ereport(ERROR, |
| 2031 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 2032 |
errmsg("invalid field size"))); |
| 2033 |
|
| 2034 |
/* reset attribute_buf to empty, and load raw data in it */ |
| 2035 |
resetStringInfo(&cstate->attribute_buf); |
| 2036 |
|
| 2037 |
enlargeStringInfo(&cstate->attribute_buf, fld_size); |
| 2038 |
if (CopyReadBinaryData(cstate, cstate->attribute_buf.data, |
| 2039 |
fld_size) != fld_size) |
| 2040 |
ereport(ERROR, |
| 2041 |
(errcode(ERRCODE_BAD_COPY_FILE_FORMAT), |
| 2042 |
errmsg("unexpected EOF in COPY data"))); |
| 2043 |
|
| 2044 |
cstate->attribute_buf.len = fld_size; |
| 2045 |
cstate->attribute_buf.data[fld_size] = '\0'; |
| 2046 |
|
| 2047 |
/* Call the column type's binary input converter */ |
| 2048 |
result = ReceiveFunctionCall(flinfo, &cstate->attribute_buf, |
| 2049 |
typioparam, typmod); |
| 2050 |
|
| 2051 |
/* Trouble if it didn't eat the whole buffer */ |
| 2052 |
if (cstate->attribute_buf.cursor != cstate->attribute_buf.len) |
| 2053 |
ereport(ERROR, |
| 2054 |
(errcode(ERRCODE_INVALID_BINARY_REPRESENTATION), |
| 2055 |
errmsg("incorrect binary data format"))); |
| 2056 |
|
| 2057 |
*isnull = false; |
| 2058 |
return result; |
| 2059 |
} |