Line data Source code
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_attribute_always_inline bool CopyFromTextLikeOneRow(CopyFromState cstate,
151 : ExprContext *econtext,
152 : Datum *values,
153 : bool *nulls,
154 : bool is_csv);
155 : static pg_attribute_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 2228 : ReceiveCopyBegin(CopyFromState cstate)
171 : {
172 : StringInfoData buf;
173 2228 : int natts = list_length(cstate->attnumlist);
174 2228 : int16 format = (cstate->opts.binary ? 1 : 0);
175 : int i;
176 :
177 2228 : pq_beginmessage(&buf, PqMsg_CopyInResponse);
178 2228 : pq_sendbyte(&buf, format); /* overall format */
179 2228 : pq_sendint16(&buf, natts);
180 9742 : for (i = 0; i < natts; i++)
181 7514 : pq_sendint16(&buf, format); /* per-column formats */
182 2228 : pq_endmessage(&buf);
183 2228 : cstate->copy_src = COPY_FRONTEND;
184 2228 : cstate->fe_msgbuf = makeStringInfo();
185 : /* We *must* flush here to ensure FE knows it can send. */
186 2228 : pq_flush();
187 2228 : }
188 :
189 : void
190 14 : ReceiveCopyBinaryHeader(CopyFromState cstate)
191 : {
192 : char readSig[11];
193 : int32 tmp;
194 :
195 : /* Signature */
196 14 : if (CopyReadBinaryData(cstate, readSig, 11) != 11 ||
197 14 : memcmp(readSig, BinarySignature, 11) != 0)
198 0 : ereport(ERROR,
199 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
200 : errmsg("COPY file signature not recognized")));
201 : /* Flags field */
202 14 : if (!CopyGetInt32(cstate, &tmp))
203 0 : ereport(ERROR,
204 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
205 : errmsg("invalid COPY file header (missing flags)")));
206 14 : if ((tmp & (1 << 16)) != 0)
207 0 : ereport(ERROR,
208 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
209 : errmsg("invalid COPY file header (WITH OIDS)")));
210 14 : tmp &= ~(1 << 16);
211 14 : if ((tmp >> 16) != 0)
212 0 : ereport(ERROR,
213 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
214 : errmsg("unrecognized critical flags in COPY file header")));
215 : /* Header extension length */
216 14 : if (!CopyGetInt32(cstate, &tmp) ||
217 14 : tmp < 0)
218 0 : ereport(ERROR,
219 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
220 : errmsg("invalid COPY file header (missing length)")));
221 : /* Skip extension header, if present */
222 14 : while (tmp-- > 0)
223 : {
224 0 : if (CopyReadBinaryData(cstate, readSig, 1) != 1)
225 0 : ereport(ERROR,
226 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
227 : errmsg("invalid COPY file header (wrong length)")));
228 : }
229 14 : }
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 440024 : CopyGetData(CopyFromState cstate, void *databuf, int minread, int maxread)
246 : {
247 440024 : int bytesread = 0;
248 :
249 440024 : switch (cstate->copy_src)
250 : {
251 1092 : case COPY_FILE:
252 1092 : bytesread = fread(databuf, 1, maxread, cstate->copy_file);
253 1092 : if (ferror(cstate->copy_file))
254 0 : ereport(ERROR,
255 : (errcode_for_file_access(),
256 : errmsg("could not read from COPY file: %m")));
257 1092 : if (bytesread == 0)
258 428 : cstate->raw_reached_eof = true;
259 1092 : break;
260 410970 : case COPY_FRONTEND:
261 819184 : while (maxread > 0 && bytesread < minread && !cstate->raw_reached_eof)
262 : {
263 : int avail;
264 :
265 818454 : while (cstate->fe_msgbuf->cursor >= cstate->fe_msgbuf->len)
266 : {
267 : /* Try to receive another message */
268 : int mtype;
269 : int maxmsglen;
270 :
271 410240 : readmessage:
272 410240 : HOLD_CANCEL_INTERRUPTS();
273 410240 : pq_startmsgread();
274 410240 : mtype = pq_getbyte();
275 410240 : if (mtype == EOF)
276 0 : 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 408214 : case PqMsg_CopyData:
283 408214 : maxmsglen = PQ_LARGE_MESSAGE_LIMIT;
284 408214 : break;
285 2024 : case PqMsg_CopyDone:
286 : case PqMsg_CopyFail:
287 : case PqMsg_Flush:
288 : case PqMsg_Sync:
289 2024 : maxmsglen = PQ_SMALL_MESSAGE_LIMIT;
290 2024 : break;
291 2 : default:
292 2 : 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 410238 : if (pq_getmessage(cstate->fe_msgbuf, maxmsglen))
301 0 : ereport(ERROR,
302 : (errcode(ERRCODE_CONNECTION_FAILURE),
303 : errmsg("unexpected EOF on client connection with an open transaction")));
304 410238 : RESUME_CANCEL_INTERRUPTS();
305 : /* ... and process it */
306 : switch (mtype)
307 : {
308 408214 : case PqMsg_CopyData:
309 408214 : break;
310 2024 : case PqMsg_CopyDone:
311 : /* COPY IN correctly terminated by frontend */
312 2024 : cstate->raw_reached_eof = true;
313 2024 : return bytesread;
314 0 : case PqMsg_CopyFail:
315 0 : ereport(ERROR,
316 : (errcode(ERRCODE_QUERY_CANCELED),
317 : errmsg("COPY from stdin failed: %s",
318 : pq_getmsgstring(cstate->fe_msgbuf))));
319 : break;
320 0 : 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 0 : goto readmessage;
330 408214 : default:
331 : Assert(false); /* NOT REACHED */
332 : }
333 : }
334 408214 : avail = cstate->fe_msgbuf->len - cstate->fe_msgbuf->cursor;
335 408214 : if (avail > maxread)
336 0 : avail = maxread;
337 408214 : pq_copymsgbytes(cstate->fe_msgbuf, databuf, avail);
338 408214 : databuf = (void *) ((char *) databuf + avail);
339 408214 : maxread -= avail;
340 408214 : bytesread += avail;
341 : }
342 408944 : break;
343 27962 : case COPY_CALLBACK:
344 27962 : bytesread = cstate->data_source_cb(databuf, minread, maxread);
345 27962 : break;
346 : }
347 :
348 437998 : 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
362 186 : CopyGetInt32(CopyFromState cstate, int32 *val)
363 : {
364 : uint32 buf;
365 :
366 186 : if (CopyReadBinaryData(cstate, (char *) &buf, sizeof(buf)) != sizeof(buf))
367 : {
368 0 : *val = 0; /* suppress compiler warning */
369 0 : return false;
370 : }
371 186 : *val = (int32) pg_ntoh32(buf);
372 186 : return true;
373 : }
374 :
375 : /*
376 : * CopyGetInt16 reads an int16 that appears in network byte order
377 : */
378 : static inline bool
379 42 : CopyGetInt16(CopyFromState cstate, int16 *val)
380 : {
381 : uint16 buf;
382 :
383 42 : if (CopyReadBinaryData(cstate, (char *) &buf, sizeof(buf)) != sizeof(buf))
384 : {
385 0 : *val = 0; /* suppress compiler warning */
386 0 : return false;
387 : }
388 42 : *val = (int16) pg_ntoh16(buf);
389 42 : 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 875914 : 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 875914 : 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 875830 : int preverifiedlen = cstate->input_buf_len;
416 875830 : int unverifiedlen = cstate->raw_buf_len - cstate->input_buf_len;
417 : int nverified;
418 :
419 875830 : if (unverifiedlen == 0)
420 : {
421 : /*
422 : * If no more raw data is coming, report the EOF to the caller.
423 : */
424 439414 : if (cstate->raw_reached_eof)
425 1498 : cstate->input_reached_eof = true;
426 439414 : return;
427 : }
428 :
429 : /*
430 : * Verify the new data, including any residual unverified bytes from
431 : * previous round.
432 : */
433 436416 : nverified = pg_encoding_verifymbstr(cstate->file_encoding,
434 436416 : cstate->raw_buf + preverifiedlen,
435 : unverifiedlen);
436 436416 : 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 0 : if (cstate->raw_reached_eof || unverifiedlen >= pg_encoding_max_length(cstate->file_encoding))
448 0 : cstate->input_reached_error = true;
449 0 : return;
450 : }
451 436416 : 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 84 : if (RAW_BUF_BYTES(cstate) == 0)
466 : {
467 : /*
468 : * If no more raw data is coming, report the EOF to the caller.
469 : */
470 48 : if (cstate->raw_reached_eof)
471 12 : cstate->input_reached_eof = true;
472 48 : return;
473 : }
474 :
475 : /*
476 : * First, copy down any unprocessed data.
477 : */
478 36 : nbytes = INPUT_BUF_BYTES(cstate);
479 36 : if (nbytes > 0 && cstate->input_buf_index > 0)
480 0 : memmove(cstate->input_buf, cstate->input_buf + cstate->input_buf_index,
481 : nbytes);
482 36 : cstate->input_buf_index = 0;
483 36 : cstate->input_buf_len = nbytes;
484 36 : cstate->input_buf[nbytes] = '\0';
485 :
486 36 : src = (unsigned char *) cstate->raw_buf + cstate->raw_buf_index;
487 36 : srclen = cstate->raw_buf_len - cstate->raw_buf_index;
488 36 : dst = (unsigned char *) cstate->input_buf + cstate->input_buf_len;
489 36 : 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 36 : convertedlen = pg_do_encoding_conversion_buf(cstate->conversion_proc,
505 : cstate->file_encoding,
506 : GetDatabaseEncoding(),
507 : src, srclen,
508 : dst, dstlen,
509 : true);
510 36 : 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 24 : if (cstate->raw_reached_eof || srclen >= MAX_CONVERSION_INPUT_LENGTH)
521 12 : cstate->input_reached_error = true;
522 24 : return;
523 : }
524 12 : cstate->raw_buf_index += convertedlen;
525 12 : cstate->input_buf_len += strlen((char *) dst);
526 : }
527 : }
528 :
529 : /*
530 : * Report an encoding or conversion error.
531 : */
532 : static void
533 12 : CopyConversionError(CopyFromState cstate)
534 : {
535 : Assert(cstate->raw_buf_len > 0);
536 : Assert(cstate->input_reached_error);
537 :
538 12 : 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 0 : report_invalid_encoding(cstate->file_encoding,
545 0 : cstate->raw_buf + cstate->input_buf_len,
546 0 : 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 12 : src = (unsigned char *) cstate->raw_buf + cstate->raw_buf_index;
564 12 : srclen = cstate->raw_buf_len - cstate->raw_buf_index;
565 12 : dst = (unsigned char *) cstate->input_buf + cstate->input_buf_len;
566 12 : dstlen = INPUT_BUF_SIZE - cstate->input_buf_len + 1;
567 :
568 12 : (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 0 : 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 438000 : 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 438000 : 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 438000 : nbytes = RAW_BUF_BYTES(cstate);
610 438000 : if (nbytes > 0 && cstate->raw_buf_index > 0)
611 36 : memmove(cstate->raw_buf, cstate->raw_buf + cstate->raw_buf_index,
612 : nbytes);
613 438000 : cstate->raw_buf_len -= cstate->raw_buf_index;
614 438000 : 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 438000 : if (cstate->raw_buf == cstate->input_buf)
621 : {
622 437916 : cstate->input_buf_len -= cstate->input_buf_index;
623 437916 : cstate->input_buf_index = 0;
624 : }
625 :
626 : /* Load more data */
627 438000 : inbytes = CopyGetData(cstate, cstate->raw_buf + cstate->raw_buf_len,
628 438000 : 1, RAW_BUF_SIZE - cstate->raw_buf_len);
629 437998 : nbytes += inbytes;
630 437998 : cstate->raw_buf[nbytes] = '\0';
631 437998 : cstate->raw_buf_len = nbytes;
632 :
633 437998 : cstate->bytes_processed += inbytes;
634 437998 : pgstat_progress_update_param(PROGRESS_COPY_BYTES_PROCESSED, cstate->bytes_processed);
635 :
636 437998 : if (inbytes == 0)
637 1534 : cstate->raw_reached_eof = true;
638 437998 : }
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 437952 : CopyLoadInputBuf(CopyFromState cstate)
651 : {
652 437952 : 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 437952 : if (cstate->raw_buf == cstate->input_buf)
660 : {
661 : Assert(!cstate->need_transcoding);
662 : Assert(cstate->input_buf_index >= cstate->raw_buf_index);
663 437916 : 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 875914 : CopyConvertBuf(cstate);
670 :
671 : /* If we now have some more input bytes ready, return them */
672 875914 : if (INPUT_BUF_BYTES(cstate) > nbytes)
673 436428 : 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 439486 : if (cstate->input_reached_error)
681 12 : CopyConversionError(cstate);
682 :
683 : /* no more input, and everything has been converted */
684 439474 : if (cstate->input_reached_eof)
685 1510 : break;
686 :
687 : /* Try to load more raw data */
688 : Assert(!cstate->raw_reached_eof);
689 437964 : 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
701 382 : CopyReadBinaryData(CopyFromState cstate, char *dest, int nbytes)
702 : {
703 382 : int copied_bytes = 0;
704 :
705 382 : if (RAW_BUF_BYTES(cstate) >= nbytes)
706 : {
707 : /* Enough bytes are present in the buffer. */
708 346 : memcpy(dest, cstate->raw_buf + cstate->raw_buf_index, nbytes);
709 346 : cstate->raw_buf_index += nbytes;
710 346 : 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 36 : if (RAW_BUF_BYTES(cstate) == 0)
724 : {
725 36 : CopyLoadRawBuf(cstate);
726 36 : if (cstate->raw_reached_eof)
727 12 : break; /* EOF */
728 : }
729 :
730 : /* Transfer some bytes. */
731 24 : copy_bytes = Min(nbytes - copied_bytes, RAW_BUF_BYTES(cstate));
732 24 : memcpy(dest, cstate->raw_buf + cstate->raw_buf_index, copy_bytes);
733 24 : cstate->raw_buf_index += copy_bytes;
734 24 : dest += copy_bytes;
735 24 : copied_bytes += copy_bytes;
736 24 : } while (copied_bytes < nbytes);
737 : }
738 :
739 382 : 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 0 : NextCopyFromRawFields(CopyFromState cstate, char ***fields, int *nfields)
748 : {
749 0 : return NextCopyFromRawFieldsInternal(cstate, fields, nfields,
750 0 : 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_attribute_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_attribute_always_inline bool
771 2105534 : NextCopyFromRawFieldsInternal(CopyFromState cstate, char ***fields, int *nfields, bool is_csv)
772 : {
773 : int fldct;
774 2105534 : bool done = false;
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 2105534 : if (cstate->cur_lineno == 0 && cstate->opts.header_line != COPY_HEADER_FALSE)
781 : {
782 : ListCell *cur;
783 : TupleDesc tupDesc;
784 138 : int lines_to_skip = cstate->opts.header_line;
785 :
786 : /* If set to "match", one header line is skipped */
787 138 : if (cstate->opts.header_line == COPY_HEADER_MATCH)
788 76 : lines_to_skip = 1;
789 :
790 138 : tupDesc = RelationGetDescr(cstate->rel);
791 :
792 318 : for (int i = 0; i < lines_to_skip; i++)
793 : {
794 186 : cstate->cur_lineno++;
795 186 : if ((done = CopyReadLine(cstate, is_csv)))
796 6 : break;
797 : }
798 :
799 138 : if (cstate->opts.header_line == COPY_HEADER_MATCH)
800 : {
801 : int fldnum;
802 :
803 76 : if (is_csv)
804 10 : fldct = CopyReadAttributesCSV(cstate);
805 : else
806 66 : fldct = CopyReadAttributesText(cstate);
807 :
808 76 : if (fldct != list_length(cstate->attnumlist))
809 24 : ereport(ERROR,
810 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
811 : errmsg("wrong number of fields in header line: got %d, expected %d",
812 : fldct, list_length(cstate->attnumlist))));
813 :
814 52 : fldnum = 0;
815 158 : foreach(cur, cstate->attnumlist)
816 : {
817 126 : int attnum = lfirst_int(cur);
818 : char *colName;
819 126 : Form_pg_attribute attr = TupleDescAttr(tupDesc, attnum - 1);
820 :
821 : Assert(fldnum < cstate->max_fields);
822 :
823 126 : colName = cstate->raw_fields[fldnum++];
824 126 : if (colName == NULL)
825 6 : ereport(ERROR,
826 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
827 : errmsg("column name mismatch in header line field %d: got null value (\"%s\"), expected \"%s\"",
828 : fldnum, cstate->opts.null_print, NameStr(attr->attname))));
829 :
830 120 : if (namestrcmp(&attr->attname, colName) != 0)
831 : {
832 14 : ereport(ERROR,
833 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
834 : errmsg("column name mismatch in header line field %d: got \"%s\", expected \"%s\"",
835 : fldnum, colName, NameStr(attr->attname))));
836 : }
837 : }
838 : }
839 :
840 94 : if (done)
841 6 : return false;
842 : }
843 :
844 2105484 : cstate->cur_lineno++;
845 :
846 : /* Actually read the line into memory here */
847 2105484 : done = CopyReadLine(cstate, is_csv);
848 :
849 : /*
850 : * EOF at start of line means we're done. If we see EOF after some
851 : * characters, we act as though it was newline followed by EOF, ie,
852 : * process the line and then exit loop on next iteration.
853 : */
854 2105458 : if (done && cstate->line_buf.len == 0)
855 2798 : return false;
856 :
857 : /* Parse the line into de-escaped field values */
858 2102660 : if (is_csv)
859 476 : fldct = CopyReadAttributesCSV(cstate);
860 : else
861 2102184 : fldct = CopyReadAttributesText(cstate);
862 :
863 2102648 : *fields = cstate->raw_fields;
864 2102648 : *nfields = fldct;
865 2102648 : return true;
866 : }
867 :
868 : /*
869 : * Read next tuple from file for COPY FROM. Return false if no more tuples.
870 : *
871 : * 'econtext' is used to evaluate default expression for each column that is
872 : * either not read from the file or is using the DEFAULT option of COPY FROM.
873 : * It can be NULL when no default values are used, i.e. when all columns are
874 : * read from the file, and DEFAULT option is unset.
875 : *
876 : * 'values' and 'nulls' arrays must be the same length as columns of the
877 : * relation passed to BeginCopyFrom. This function fills the arrays.
878 : */
879 : bool
880 2105576 : NextCopyFrom(CopyFromState cstate, ExprContext *econtext,
881 : Datum *values, bool *nulls)
882 : {
883 : TupleDesc tupDesc;
884 : AttrNumber num_phys_attrs,
885 2105576 : num_defaults = cstate->num_defaults;
886 : int i;
887 2105576 : int *defmap = cstate->defmap;
888 2105576 : ExprState **defexprs = cstate->defexprs;
889 :
890 2105576 : tupDesc = RelationGetDescr(cstate->rel);
891 2105576 : num_phys_attrs = tupDesc->natts;
892 :
893 : /* Initialize all values for row to NULL */
894 9237092 : MemSet(values, 0, num_phys_attrs * sizeof(Datum));
895 2105576 : MemSet(nulls, true, num_phys_attrs * sizeof(bool));
896 2347744 : MemSet(cstate->defaults, false, num_phys_attrs * sizeof(bool));
897 :
898 : /* Get one row from source */
899 2105576 : if (!cstate->routine->CopyFromOneRow(cstate, econtext, values, nulls))
900 2816 : return false;
901 :
902 : /*
903 : * Now compute and insert any defaults available for the columns not
904 : * provided by the input data. Anything not processed here or above will
905 : * remain NULL.
906 : */
907 2163132 : for (i = 0; i < num_defaults; i++)
908 : {
909 : /*
910 : * The caller must supply econtext and have switched into the
911 : * per-tuple memory context in it.
912 : */
913 : Assert(econtext != NULL);
914 : Assert(CurrentMemoryContext == econtext->ecxt_per_tuple_memory);
915 :
916 60530 : values[defmap[i]] = ExecEvalExpr(defexprs[defmap[i]], econtext,
917 60530 : &nulls[defmap[i]]);
918 : }
919 :
920 2102602 : return true;
921 : }
922 :
923 : /* Implementation of the per-row callback for text format */
924 : bool
925 2104830 : CopyFromTextOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values,
926 : bool *nulls)
927 : {
928 2104830 : return CopyFromTextLikeOneRow(cstate, econtext, values, nulls, false);
929 : }
930 :
931 : /* Implementation of the per-row callback for CSV format */
932 : bool
933 704 : CopyFromCSVOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values,
934 : bool *nulls)
935 : {
936 704 : return CopyFromTextLikeOneRow(cstate, econtext, values, nulls, true);
937 : }
938 :
939 : /*
940 : * Workhorse for CopyFromTextOneRow() and CopyFromCSVOneRow().
941 : *
942 : * We use pg_attribute_always_inline to reduce function call overhead
943 : * and to help compilers to optimize away the 'is_csv' condition.
944 : */
945 : static pg_attribute_always_inline bool
946 2105534 : CopyFromTextLikeOneRow(CopyFromState cstate, ExprContext *econtext,
947 : Datum *values, bool *nulls, bool is_csv)
948 : {
949 : TupleDesc tupDesc;
950 : AttrNumber attr_count;
951 2105534 : FmgrInfo *in_functions = cstate->in_functions;
952 2105534 : Oid *typioparams = cstate->typioparams;
953 2105534 : ExprState **defexprs = cstate->defexprs;
954 : char **field_strings;
955 : ListCell *cur;
956 : int fldct;
957 : int fieldno;
958 : char *string;
959 :
960 2105534 : tupDesc = RelationGetDescr(cstate->rel);
961 2105534 : attr_count = list_length(cstate->attnumlist);
962 :
963 : /* read raw fields in the next line */
964 2105534 : if (!NextCopyFromRawFieldsInternal(cstate, &field_strings, &fldct, is_csv))
965 2804 : return false;
966 :
967 : /* check for overflowing fields */
968 2102648 : if (attr_count > 0 && fldct > attr_count)
969 18 : ereport(ERROR,
970 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
971 : errmsg("extra data after last expected column")));
972 :
973 2102630 : fieldno = 0;
974 :
975 : /* Loop to read the user attributes on the line. */
976 9092494 : foreach(cur, cstate->attnumlist)
977 : {
978 6990048 : int attnum = lfirst_int(cur);
979 6990048 : int m = attnum - 1;
980 6990048 : Form_pg_attribute att = TupleDescAttr(tupDesc, m);
981 :
982 6990048 : if (fieldno >= fldct)
983 18 : ereport(ERROR,
984 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
985 : errmsg("missing data for column \"%s\"",
986 : NameStr(att->attname))));
987 6990030 : string = field_strings[fieldno++];
988 :
989 6990030 : if (cstate->convert_select_flags &&
990 20 : !cstate->convert_select_flags[m])
991 : {
992 : /* ignore input field, leaving column as NULL */
993 10 : continue;
994 : }
995 :
996 6990020 : if (is_csv)
997 : {
998 974 : if (string == NULL &&
999 44 : cstate->opts.force_notnull_flags[m])
1000 : {
1001 : /*
1002 : * FORCE_NOT_NULL option is set and column is NULL - convert
1003 : * it to the NULL string.
1004 : */
1005 28 : string = cstate->opts.null_print;
1006 : }
1007 946 : else if (string != NULL && cstate->opts.force_null_flags[m]
1008 50 : && strcmp(string, cstate->opts.null_print) == 0)
1009 : {
1010 : /*
1011 : * FORCE_NULL option is set and column matches the NULL
1012 : * string. It must have been quoted, or otherwise the string
1013 : * would already have been set to NULL. Convert it to NULL as
1014 : * specified.
1015 : */
1016 26 : string = NULL;
1017 : }
1018 : }
1019 :
1020 6990020 : cstate->cur_attname = NameStr(att->attname);
1021 6990020 : cstate->cur_attval = string;
1022 :
1023 6990020 : if (string != NULL)
1024 6856632 : nulls[m] = false;
1025 :
1026 6990020 : if (cstate->defaults[m])
1027 : {
1028 : /* We must have switched into the per-tuple memory context */
1029 : Assert(econtext != NULL);
1030 : Assert(CurrentMemoryContext == econtext->ecxt_per_tuple_memory);
1031 :
1032 60 : values[m] = ExecEvalExpr(defexprs[m], econtext, &nulls[m]);
1033 : }
1034 :
1035 : /*
1036 : * If ON_ERROR is specified with IGNORE, skip rows with soft errors
1037 : */
1038 6989922 : else if (!InputFunctionCallSafe(&in_functions[m],
1039 : string,
1040 6989960 : typioparams[m],
1041 : att->atttypmod,
1042 6989960 : (Node *) cstate->escontext,
1043 6989960 : &values[m]))
1044 : {
1045 : Assert(cstate->opts.on_error != COPY_ON_ERROR_STOP);
1046 :
1047 128 : cstate->num_errors++;
1048 :
1049 128 : if (cstate->opts.log_verbosity == COPY_LOG_VERBOSITY_VERBOSE)
1050 : {
1051 : /*
1052 : * Since we emit line number and column info in the below
1053 : * notice message, we suppress error context information other
1054 : * than the relation name.
1055 : */
1056 : Assert(!cstate->relname_only);
1057 42 : cstate->relname_only = true;
1058 :
1059 42 : if (cstate->cur_attval)
1060 : {
1061 : char *attval;
1062 :
1063 36 : attval = CopyLimitPrintoutLength(cstate->cur_attval);
1064 36 : ereport(NOTICE,
1065 : errmsg("skipping row due to data type incompatibility at line %" PRIu64 " for column \"%s\": \"%s\"",
1066 : cstate->cur_lineno,
1067 : cstate->cur_attname,
1068 : attval));
1069 36 : pfree(attval);
1070 : }
1071 : else
1072 6 : ereport(NOTICE,
1073 : errmsg("skipping row due to data type incompatibility at line %" PRIu64 " for column \"%s\": null input",
1074 : cstate->cur_lineno,
1075 : cstate->cur_attname));
1076 :
1077 : /* reset relname_only */
1078 42 : cstate->relname_only = false;
1079 : }
1080 :
1081 128 : return true;
1082 : }
1083 :
1084 6989854 : cstate->cur_attname = NULL;
1085 6989854 : cstate->cur_attval = NULL;
1086 : }
1087 :
1088 : Assert(fieldno == attr_count);
1089 :
1090 2102446 : return true;
1091 : }
1092 :
1093 : /* Implementation of the per-row callback for binary format */
1094 : bool
1095 42 : CopyFromBinaryOneRow(CopyFromState cstate, ExprContext *econtext, Datum *values,
1096 : bool *nulls)
1097 : {
1098 : TupleDesc tupDesc;
1099 : AttrNumber attr_count;
1100 42 : FmgrInfo *in_functions = cstate->in_functions;
1101 42 : Oid *typioparams = cstate->typioparams;
1102 : int16 fld_count;
1103 : ListCell *cur;
1104 :
1105 42 : tupDesc = RelationGetDescr(cstate->rel);
1106 42 : attr_count = list_length(cstate->attnumlist);
1107 :
1108 42 : cstate->cur_lineno++;
1109 :
1110 42 : if (!CopyGetInt16(cstate, &fld_count))
1111 : {
1112 : /* EOF detected (end of file, or protocol-level EOF) */
1113 0 : return false;
1114 : }
1115 :
1116 42 : if (fld_count == -1)
1117 : {
1118 : /*
1119 : * Received EOF marker. Wait for the protocol-level EOF, and complain
1120 : * if it doesn't come immediately. In COPY FROM STDIN, this ensures
1121 : * that we correctly handle CopyFail, if client chooses to send that
1122 : * now. When copying from file, we could ignore the rest of the file
1123 : * like in text mode, but we choose to be consistent with the COPY
1124 : * FROM STDIN case.
1125 : */
1126 : char dummy;
1127 :
1128 12 : if (CopyReadBinaryData(cstate, &dummy, 1) > 0)
1129 0 : ereport(ERROR,
1130 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1131 : errmsg("received copy data after EOF marker")));
1132 12 : return false;
1133 : }
1134 :
1135 30 : if (fld_count != attr_count)
1136 0 : ereport(ERROR,
1137 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1138 : errmsg("row field count is %d, expected %d",
1139 : (int) fld_count, attr_count)));
1140 :
1141 186 : foreach(cur, cstate->attnumlist)
1142 : {
1143 158 : int attnum = lfirst_int(cur);
1144 158 : int m = attnum - 1;
1145 158 : Form_pg_attribute att = TupleDescAttr(tupDesc, m);
1146 :
1147 158 : cstate->cur_attname = NameStr(att->attname);
1148 314 : values[m] = CopyReadBinaryAttribute(cstate,
1149 158 : &in_functions[m],
1150 158 : typioparams[m],
1151 : att->atttypmod,
1152 : &nulls[m]);
1153 156 : cstate->cur_attname = NULL;
1154 : }
1155 :
1156 28 : return true;
1157 : }
1158 :
1159 : /*
1160 : * Read the next input line and stash it in line_buf.
1161 : *
1162 : * Result is true if read was terminated by EOF, false if terminated
1163 : * by newline. The terminating newline or EOF marker is not included
1164 : * in the final value of line_buf.
1165 : */
1166 : static bool
1167 2105670 : CopyReadLine(CopyFromState cstate, bool is_csv)
1168 : {
1169 : bool result;
1170 :
1171 2105670 : resetStringInfo(&cstate->line_buf);
1172 2105670 : cstate->line_buf_valid = false;
1173 :
1174 : /* Parse data and transfer into line_buf */
1175 2105670 : result = CopyReadLineText(cstate, is_csv);
1176 :
1177 2105644 : if (result)
1178 : {
1179 : /*
1180 : * Reached EOF. In protocol version 3, we should ignore anything
1181 : * after \. up to the protocol end of copy data. (XXX maybe better
1182 : * not to treat \. as special?)
1183 : */
1184 2804 : if (cstate->copy_src == COPY_FRONTEND)
1185 : {
1186 : int inbytes;
1187 :
1188 : do
1189 : {
1190 2024 : inbytes = CopyGetData(cstate, cstate->input_buf,
1191 : 1, INPUT_BUF_SIZE);
1192 2024 : } while (inbytes > 0);
1193 2024 : cstate->input_buf_index = 0;
1194 2024 : cstate->input_buf_len = 0;
1195 2024 : cstate->raw_buf_index = 0;
1196 2024 : cstate->raw_buf_len = 0;
1197 : }
1198 : }
1199 : else
1200 : {
1201 : /*
1202 : * If we didn't hit EOF, then we must have transferred the EOL marker
1203 : * to line_buf along with the data. Get rid of it.
1204 : */
1205 2102840 : switch (cstate->eol_type)
1206 : {
1207 2102840 : case EOL_NL:
1208 : Assert(cstate->line_buf.len >= 1);
1209 : Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\n');
1210 2102840 : cstate->line_buf.len--;
1211 2102840 : cstate->line_buf.data[cstate->line_buf.len] = '\0';
1212 2102840 : break;
1213 0 : case EOL_CR:
1214 : Assert(cstate->line_buf.len >= 1);
1215 : Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\r');
1216 0 : cstate->line_buf.len--;
1217 0 : cstate->line_buf.data[cstate->line_buf.len] = '\0';
1218 0 : break;
1219 0 : case EOL_CRNL:
1220 : Assert(cstate->line_buf.len >= 2);
1221 : Assert(cstate->line_buf.data[cstate->line_buf.len - 2] == '\r');
1222 : Assert(cstate->line_buf.data[cstate->line_buf.len - 1] == '\n');
1223 0 : cstate->line_buf.len -= 2;
1224 0 : cstate->line_buf.data[cstate->line_buf.len] = '\0';
1225 0 : break;
1226 0 : case EOL_UNKNOWN:
1227 : /* shouldn't get here */
1228 : Assert(false);
1229 0 : break;
1230 : }
1231 : }
1232 :
1233 : /* Now it's safe to use the buffer in error messages */
1234 2105644 : cstate->line_buf_valid = true;
1235 :
1236 2105644 : return result;
1237 : }
1238 :
1239 : /*
1240 : * CopyReadLineText - inner loop of CopyReadLine for text mode
1241 : */
1242 : static bool
1243 2105670 : CopyReadLineText(CopyFromState cstate, bool is_csv)
1244 : {
1245 : char *copy_input_buf;
1246 : int input_buf_ptr;
1247 : int copy_buf_len;
1248 2105670 : bool need_data = false;
1249 2105670 : bool hit_eof = false;
1250 2105670 : bool result = false;
1251 :
1252 : /* CSV variables */
1253 2105670 : bool in_quote = false,
1254 2105670 : last_was_esc = false;
1255 2105670 : char quotec = '\0';
1256 2105670 : char escapec = '\0';
1257 :
1258 2105670 : if (is_csv)
1259 : {
1260 810 : quotec = cstate->opts.quote[0];
1261 810 : escapec = cstate->opts.escape[0];
1262 : /* ignore special escape processing if it's the same as quotec */
1263 810 : if (quotec == escapec)
1264 616 : escapec = '\0';
1265 : }
1266 :
1267 : /*
1268 : * The objective of this loop is to transfer the entire next input line
1269 : * into line_buf. Hence, we only care for detecting newlines (\r and/or
1270 : * \n) and the end-of-copy marker (\.).
1271 : *
1272 : * In CSV mode, \r and \n inside a quoted field are just part of the data
1273 : * value and are put in line_buf. We keep just enough state to know if we
1274 : * are currently in a quoted field or not.
1275 : *
1276 : * The input has already been converted to the database encoding. All
1277 : * supported server encodings have the property that all bytes in a
1278 : * multi-byte sequence have the high bit set, so a multibyte character
1279 : * cannot contain any newline or escape characters embedded in the
1280 : * multibyte sequence. Therefore, we can process the input byte-by-byte,
1281 : * regardless of the encoding.
1282 : *
1283 : * For speed, we try to move data from input_buf to line_buf in chunks
1284 : * rather than one character at a time. input_buf_ptr points to the next
1285 : * character to examine; any characters from input_buf_index to
1286 : * input_buf_ptr have been determined to be part of the line, but not yet
1287 : * transferred to line_buf.
1288 : *
1289 : * For a little extra speed within the loop, we copy input_buf and
1290 : * input_buf_len into local variables.
1291 : */
1292 2105670 : copy_input_buf = cstate->input_buf;
1293 2105670 : input_buf_ptr = cstate->input_buf_index;
1294 2105670 : copy_buf_len = cstate->input_buf_len;
1295 :
1296 : for (;;)
1297 49000198 : {
1298 : int prev_raw_ptr;
1299 : char c;
1300 :
1301 : /*
1302 : * Load more data if needed.
1303 : *
1304 : * TODO: We could just force four bytes of read-ahead and avoid the
1305 : * many calls to IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(). That was
1306 : * unsafe with the old v2 COPY protocol, but we don't support that
1307 : * anymore.
1308 : */
1309 51105868 : if (input_buf_ptr >= copy_buf_len || need_data)
1310 : {
1311 437952 : REFILL_LINEBUF;
1312 :
1313 437952 : CopyLoadInputBuf(cstate);
1314 : /* update our local variables */
1315 437938 : hit_eof = cstate->input_reached_eof;
1316 437938 : input_buf_ptr = cstate->input_buf_index;
1317 437938 : copy_buf_len = cstate->input_buf_len;
1318 :
1319 : /*
1320 : * If we are completely out of data, break out of the loop,
1321 : * reporting EOF.
1322 : */
1323 437938 : if (INPUT_BUF_BYTES(cstate) <= 0)
1324 : {
1325 1510 : result = true;
1326 1510 : break;
1327 : }
1328 436428 : need_data = false;
1329 : }
1330 :
1331 : /* OK to fetch a character */
1332 51104344 : prev_raw_ptr = input_buf_ptr;
1333 51104344 : c = copy_input_buf[input_buf_ptr++];
1334 :
1335 51104344 : if (is_csv)
1336 : {
1337 : /*
1338 : * If character is '\r', we may need to look ahead below. Force
1339 : * fetch of the next character if we don't already have it. We
1340 : * need to do this before changing CSV state, in case '\r' is also
1341 : * the quote or escape character.
1342 : */
1343 6552 : if (c == '\r')
1344 : {
1345 36 : IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0);
1346 : }
1347 :
1348 : /*
1349 : * Dealing with quotes and escapes here is mildly tricky. If the
1350 : * quote char is also the escape char, there's no problem - we
1351 : * just use the char as a toggle. If they are different, we need
1352 : * to ensure that we only take account of an escape inside a
1353 : * quoted field and immediately preceding a quote char, and not
1354 : * the second in an escape-escape sequence.
1355 : */
1356 6552 : if (in_quote && c == escapec)
1357 48 : last_was_esc = !last_was_esc;
1358 6552 : if (c == quotec && !last_was_esc)
1359 508 : in_quote = !in_quote;
1360 6552 : if (c != escapec)
1361 6498 : last_was_esc = false;
1362 :
1363 : /*
1364 : * Updating the line count for embedded CR and/or LF chars is
1365 : * necessarily a little fragile - this test is probably about the
1366 : * best we can do. (XXX it's arguable whether we should do this
1367 : * at all --- is cur_lineno a physical or logical count?)
1368 : */
1369 6552 : if (in_quote && c == (cstate->eol_type == EOL_NL ? '\n' : '\r'))
1370 36 : cstate->cur_lineno++;
1371 : }
1372 :
1373 : /* Process \r */
1374 51104344 : if (c == '\r' && (!is_csv || !in_quote))
1375 : {
1376 : /* Check for \r\n on first line, _and_ handle \r\n. */
1377 0 : if (cstate->eol_type == EOL_UNKNOWN ||
1378 0 : cstate->eol_type == EOL_CRNL)
1379 : {
1380 : /*
1381 : * If need more data, go back to loop top to load it.
1382 : *
1383 : * Note that if we are at EOF, c will wind up as '\0' because
1384 : * of the guaranteed pad of input_buf.
1385 : */
1386 0 : IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0);
1387 :
1388 : /* get next char */
1389 0 : c = copy_input_buf[input_buf_ptr];
1390 :
1391 0 : if (c == '\n')
1392 : {
1393 0 : input_buf_ptr++; /* eat newline */
1394 0 : cstate->eol_type = EOL_CRNL; /* in case not set yet */
1395 : }
1396 : else
1397 : {
1398 : /* found \r, but no \n */
1399 0 : if (cstate->eol_type == EOL_CRNL)
1400 0 : ereport(ERROR,
1401 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1402 : !is_csv ?
1403 : errmsg("literal carriage return found in data") :
1404 : errmsg("unquoted carriage return found in data"),
1405 : !is_csv ?
1406 : errhint("Use \"\\r\" to represent carriage return.") :
1407 : errhint("Use quoted CSV field to represent carriage return.")));
1408 :
1409 : /*
1410 : * if we got here, it is the first line and we didn't find
1411 : * \n, so don't consume the peeked character
1412 : */
1413 0 : cstate->eol_type = EOL_CR;
1414 : }
1415 : }
1416 0 : else if (cstate->eol_type == EOL_NL)
1417 0 : ereport(ERROR,
1418 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1419 : !is_csv ?
1420 : errmsg("literal carriage return found in data") :
1421 : errmsg("unquoted carriage return found in data"),
1422 : !is_csv ?
1423 : errhint("Use \"\\r\" to represent carriage return.") :
1424 : errhint("Use quoted CSV field to represent carriage return.")));
1425 : /* If reach here, we have found the line terminator */
1426 0 : break;
1427 : }
1428 :
1429 : /* Process \n */
1430 51104344 : if (c == '\n' && (!is_csv || !in_quote))
1431 : {
1432 2102840 : if (cstate->eol_type == EOL_CR || cstate->eol_type == EOL_CRNL)
1433 0 : ereport(ERROR,
1434 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1435 : !is_csv ?
1436 : errmsg("literal newline found in data") :
1437 : errmsg("unquoted newline found in data"),
1438 : !is_csv ?
1439 : errhint("Use \"\\n\" to represent newline.") :
1440 : errhint("Use quoted CSV field to represent newline.")));
1441 2102840 : cstate->eol_type = EOL_NL; /* in case not set yet */
1442 : /* If reach here, we have found the line terminator */
1443 2102840 : break;
1444 : }
1445 :
1446 : /*
1447 : * Process backslash, except in CSV mode where backslash is a normal
1448 : * character.
1449 : */
1450 49001504 : if (c == '\\' && !is_csv)
1451 : {
1452 : char c2;
1453 :
1454 139600 : IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0);
1455 139564 : IF_NEED_REFILL_AND_EOF_BREAK(0);
1456 :
1457 : /* -----
1458 : * get next character
1459 : * Note: we do not change c so if it isn't \., we can fall
1460 : * through and continue processing.
1461 : * -----
1462 : */
1463 139564 : c2 = copy_input_buf[input_buf_ptr];
1464 :
1465 139564 : if (c2 == '.')
1466 : {
1467 1306 : input_buf_ptr++; /* consume the '.' */
1468 1306 : if (cstate->eol_type == EOL_CRNL)
1469 : {
1470 : /* Get the next character */
1471 0 : IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0);
1472 : /* if hit_eof, c2 will become '\0' */
1473 0 : c2 = copy_input_buf[input_buf_ptr++];
1474 :
1475 0 : if (c2 == '\n')
1476 0 : ereport(ERROR,
1477 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1478 : errmsg("end-of-copy marker does not match previous newline style")));
1479 0 : else if (c2 != '\r')
1480 0 : ereport(ERROR,
1481 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1482 : errmsg("end-of-copy marker is not alone on its line")));
1483 : }
1484 :
1485 : /* Get the next character */
1486 1306 : IF_NEED_REFILL_AND_NOT_EOF_CONTINUE(0);
1487 : /* if hit_eof, c2 will become '\0' */
1488 1306 : c2 = copy_input_buf[input_buf_ptr++];
1489 :
1490 1306 : if (c2 != '\r' && c2 != '\n')
1491 6 : ereport(ERROR,
1492 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1493 : errmsg("end-of-copy marker is not alone on its line")));
1494 :
1495 1300 : if ((cstate->eol_type == EOL_NL && c2 != '\n') ||
1496 1300 : (cstate->eol_type == EOL_CRNL && c2 != '\n') ||
1497 1300 : (cstate->eol_type == EOL_CR && c2 != '\r'))
1498 0 : ereport(ERROR,
1499 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1500 : errmsg("end-of-copy marker does not match previous newline style")));
1501 :
1502 : /*
1503 : * If there is any data on this line before the \., complain.
1504 : */
1505 1300 : if (cstate->line_buf.len > 0 ||
1506 1300 : prev_raw_ptr > cstate->input_buf_index)
1507 6 : ereport(ERROR,
1508 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1509 : errmsg("end-of-copy marker is not alone on its line")));
1510 :
1511 : /*
1512 : * Discard the \. and newline, then report EOF.
1513 : */
1514 1294 : cstate->input_buf_index = input_buf_ptr;
1515 1294 : result = true; /* report EOF */
1516 1294 : break;
1517 : }
1518 : else
1519 : {
1520 : /*
1521 : * If we are here, it means we found a backslash followed by
1522 : * something other than a period. In non-CSV mode, anything
1523 : * after a backslash is special, so we skip over that second
1524 : * character too. If we didn't do that \\. would be
1525 : * considered an eof-of copy, while in non-CSV mode it is a
1526 : * literal backslash followed by a period.
1527 : */
1528 138258 : input_buf_ptr++;
1529 : }
1530 : }
1531 : } /* end of outer loop */
1532 :
1533 : /*
1534 : * Transfer any still-uncopied data to line_buf.
1535 : */
1536 2105644 : REFILL_LINEBUF;
1537 :
1538 2105644 : return result;
1539 : }
1540 :
1541 : /*
1542 : * Return decimal value for a hexadecimal digit
1543 : */
1544 : static int
1545 0 : GetDecimalFromHex(char hex)
1546 : {
1547 0 : if (isdigit((unsigned char) hex))
1548 0 : return hex - '0';
1549 : else
1550 0 : return pg_ascii_tolower((unsigned char) hex) - 'a' + 10;
1551 : }
1552 :
1553 : /*
1554 : * Parse the current line into separate attributes (fields),
1555 : * performing de-escaping as needed.
1556 : *
1557 : * The input is in line_buf. We use attribute_buf to hold the result
1558 : * strings. cstate->raw_fields[k] is set to point to the k'th attribute
1559 : * string, or NULL when the input matches the null marker string.
1560 : * This array is expanded as necessary.
1561 : *
1562 : * (Note that the caller cannot check for nulls since the returned
1563 : * string would be the post-de-escaping equivalent, which may look
1564 : * the same as some valid data string.)
1565 : *
1566 : * delim is the column delimiter string (must be just one byte for now).
1567 : * null_print is the null marker string. Note that this is compared to
1568 : * the pre-de-escaped input string.
1569 : *
1570 : * The return value is the number of fields actually read.
1571 : */
1572 : static int
1573 2102250 : CopyReadAttributesText(CopyFromState cstate)
1574 : {
1575 2102250 : char delimc = cstate->opts.delim[0];
1576 : int fieldno;
1577 : char *output_ptr;
1578 : char *cur_ptr;
1579 : char *line_end_ptr;
1580 :
1581 : /*
1582 : * We need a special case for zero-column tables: check that the input
1583 : * line is empty, and return.
1584 : */
1585 2102250 : if (cstate->max_fields <= 0)
1586 : {
1587 10 : if (cstate->line_buf.len != 0)
1588 0 : ereport(ERROR,
1589 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1590 : errmsg("extra data after last expected column")));
1591 10 : return 0;
1592 : }
1593 :
1594 2102240 : resetStringInfo(&cstate->attribute_buf);
1595 :
1596 : /*
1597 : * The de-escaped attributes will certainly not be longer than the input
1598 : * data line, so we can just force attribute_buf to be large enough and
1599 : * then transfer data without any checks for enough space. We need to do
1600 : * it this way because enlarging attribute_buf mid-stream would invalidate
1601 : * pointers already stored into cstate->raw_fields[].
1602 : */
1603 2102240 : if (cstate->attribute_buf.maxlen <= cstate->line_buf.len)
1604 30 : enlargeStringInfo(&cstate->attribute_buf, cstate->line_buf.len);
1605 2102240 : output_ptr = cstate->attribute_buf.data;
1606 :
1607 : /* set pointer variables for loop */
1608 2102240 : cur_ptr = cstate->line_buf.data;
1609 2102240 : line_end_ptr = cstate->line_buf.data + cstate->line_buf.len;
1610 :
1611 : /* Outer loop iterates over fields */
1612 2102240 : fieldno = 0;
1613 : for (;;)
1614 4887184 : {
1615 6989424 : bool found_delim = false;
1616 : char *start_ptr;
1617 : char *end_ptr;
1618 : int input_len;
1619 6989424 : bool saw_non_ascii = false;
1620 :
1621 : /* Make sure there is enough space for the next value */
1622 6989424 : if (fieldno >= cstate->max_fields)
1623 : {
1624 36 : cstate->max_fields *= 2;
1625 36 : cstate->raw_fields =
1626 36 : repalloc(cstate->raw_fields, cstate->max_fields * sizeof(char *));
1627 : }
1628 :
1629 : /* Remember start of field on both input and output sides */
1630 6989424 : start_ptr = cur_ptr;
1631 6989424 : cstate->raw_fields[fieldno] = output_ptr;
1632 :
1633 : /*
1634 : * Scan data for field.
1635 : *
1636 : * Note that in this loop, we are scanning to locate the end of field
1637 : * and also speculatively performing de-escaping. Once we find the
1638 : * end-of-field, we can match the raw field contents against the null
1639 : * marker string. Only after that comparison fails do we know that
1640 : * de-escaping is actually the right thing to do; therefore we *must
1641 : * not* throw any syntax errors before we've done the null-marker
1642 : * check.
1643 : */
1644 : for (;;)
1645 44106500 : {
1646 : char c;
1647 :
1648 51095924 : end_ptr = cur_ptr;
1649 51095924 : if (cur_ptr >= line_end_ptr)
1650 2102234 : break;
1651 48993690 : c = *cur_ptr++;
1652 48993690 : if (c == delimc)
1653 : {
1654 4887190 : found_delim = true;
1655 4887190 : break;
1656 : }
1657 44106500 : if (c == '\\')
1658 : {
1659 138258 : if (cur_ptr >= line_end_ptr)
1660 0 : break;
1661 138258 : c = *cur_ptr++;
1662 138258 : switch (c)
1663 : {
1664 12 : case '0':
1665 : case '1':
1666 : case '2':
1667 : case '3':
1668 : case '4':
1669 : case '5':
1670 : case '6':
1671 : case '7':
1672 : {
1673 : /* handle \013 */
1674 : int val;
1675 :
1676 12 : val = OCTVALUE(c);
1677 12 : if (cur_ptr < line_end_ptr)
1678 : {
1679 6 : c = *cur_ptr;
1680 6 : if (ISOCTAL(c))
1681 : {
1682 0 : cur_ptr++;
1683 0 : val = (val << 3) + OCTVALUE(c);
1684 0 : if (cur_ptr < line_end_ptr)
1685 : {
1686 0 : c = *cur_ptr;
1687 0 : if (ISOCTAL(c))
1688 : {
1689 0 : cur_ptr++;
1690 0 : val = (val << 3) + OCTVALUE(c);
1691 : }
1692 : }
1693 : }
1694 : }
1695 12 : c = val & 0377;
1696 12 : if (c == '\0' || IS_HIGHBIT_SET(c))
1697 12 : saw_non_ascii = true;
1698 : }
1699 12 : break;
1700 12 : case 'x':
1701 : /* Handle \x3F */
1702 12 : if (cur_ptr < line_end_ptr)
1703 : {
1704 6 : char hexchar = *cur_ptr;
1705 :
1706 6 : if (isxdigit((unsigned char) hexchar))
1707 : {
1708 0 : int val = GetDecimalFromHex(hexchar);
1709 :
1710 0 : cur_ptr++;
1711 0 : if (cur_ptr < line_end_ptr)
1712 : {
1713 0 : hexchar = *cur_ptr;
1714 0 : if (isxdigit((unsigned char) hexchar))
1715 : {
1716 0 : cur_ptr++;
1717 0 : val = (val << 4) + GetDecimalFromHex(hexchar);
1718 : }
1719 : }
1720 0 : c = val & 0xff;
1721 0 : if (c == '\0' || IS_HIGHBIT_SET(c))
1722 0 : saw_non_ascii = true;
1723 : }
1724 : }
1725 12 : break;
1726 0 : case 'b':
1727 0 : c = '\b';
1728 0 : break;
1729 0 : case 'f':
1730 0 : c = '\f';
1731 0 : break;
1732 4182 : case 'n':
1733 4182 : c = '\n';
1734 4182 : break;
1735 0 : case 'r':
1736 0 : c = '\r';
1737 0 : break;
1738 0 : case 't':
1739 0 : c = '\t';
1740 0 : break;
1741 0 : case 'v':
1742 0 : c = '\v';
1743 0 : break;
1744 :
1745 : /*
1746 : * in all other cases, take the char after '\'
1747 : * literally
1748 : */
1749 : }
1750 : }
1751 :
1752 : /* Add c to output string */
1753 44106500 : *output_ptr++ = c;
1754 : }
1755 :
1756 : /* Check whether raw input matched null marker */
1757 6989424 : input_len = end_ptr - start_ptr;
1758 6989424 : if (input_len == cstate->opts.null_print_len &&
1759 786144 : strncmp(start_ptr, cstate->opts.null_print, input_len) == 0)
1760 133352 : cstate->raw_fields[fieldno] = NULL;
1761 : /* Check whether raw input matched default marker */
1762 6856072 : else if (fieldno < list_length(cstate->attnumlist) &&
1763 6856030 : cstate->opts.default_print &&
1764 114 : input_len == cstate->opts.default_print_len &&
1765 30 : strncmp(start_ptr, cstate->opts.default_print, input_len) == 0)
1766 24 : {
1767 : /* fieldno is 0-indexed and attnum is 1-indexed */
1768 30 : int m = list_nth_int(cstate->attnumlist, fieldno) - 1;
1769 :
1770 30 : if (cstate->defexprs[m] != NULL)
1771 : {
1772 : /* defaults contain entries for all physical attributes */
1773 24 : cstate->defaults[m] = true;
1774 : }
1775 : else
1776 : {
1777 6 : TupleDesc tupDesc = RelationGetDescr(cstate->rel);
1778 6 : Form_pg_attribute att = TupleDescAttr(tupDesc, m);
1779 :
1780 6 : ereport(ERROR,
1781 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1782 : errmsg("unexpected default marker in COPY data"),
1783 : errdetail("Column \"%s\" has no default value.",
1784 : NameStr(att->attname))));
1785 : }
1786 : }
1787 : else
1788 : {
1789 : /*
1790 : * At this point we know the field is supposed to contain data.
1791 : *
1792 : * If we de-escaped any non-7-bit-ASCII chars, make sure the
1793 : * resulting string is valid data for the db encoding.
1794 : */
1795 6856042 : if (saw_non_ascii)
1796 : {
1797 0 : char *fld = cstate->raw_fields[fieldno];
1798 :
1799 0 : pg_verifymbstr(fld, output_ptr - fld, false);
1800 : }
1801 : }
1802 :
1803 : /* Terminate attribute value in output area */
1804 6989418 : *output_ptr++ = '\0';
1805 :
1806 6989418 : fieldno++;
1807 : /* Done if we hit EOL instead of a delim */
1808 6989418 : if (!found_delim)
1809 2102234 : break;
1810 : }
1811 :
1812 : /* Clean up state of attribute_buf */
1813 2102234 : output_ptr--;
1814 : Assert(*output_ptr == '\0');
1815 2102234 : cstate->attribute_buf.len = (output_ptr - cstate->attribute_buf.data);
1816 :
1817 2102234 : return fieldno;
1818 : }
1819 :
1820 : /*
1821 : * Parse the current line into separate attributes (fields),
1822 : * performing de-escaping as needed. This has exactly the same API as
1823 : * CopyReadAttributesText, except we parse the fields according to
1824 : * "standard" (i.e. common) CSV usage.
1825 : */
1826 : static int
1827 486 : CopyReadAttributesCSV(CopyFromState cstate)
1828 : {
1829 486 : char delimc = cstate->opts.delim[0];
1830 486 : char quotec = cstate->opts.quote[0];
1831 486 : char escapec = cstate->opts.escape[0];
1832 : int fieldno;
1833 : char *output_ptr;
1834 : char *cur_ptr;
1835 : char *line_end_ptr;
1836 :
1837 : /*
1838 : * We need a special case for zero-column tables: check that the input
1839 : * line is empty, and return.
1840 : */
1841 486 : if (cstate->max_fields <= 0)
1842 : {
1843 0 : if (cstate->line_buf.len != 0)
1844 0 : ereport(ERROR,
1845 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1846 : errmsg("extra data after last expected column")));
1847 0 : return 0;
1848 : }
1849 :
1850 486 : resetStringInfo(&cstate->attribute_buf);
1851 :
1852 : /*
1853 : * The de-escaped attributes will certainly not be longer than the input
1854 : * data line, so we can just force attribute_buf to be large enough and
1855 : * then transfer data without any checks for enough space. We need to do
1856 : * it this way because enlarging attribute_buf mid-stream would invalidate
1857 : * pointers already stored into cstate->raw_fields[].
1858 : */
1859 486 : if (cstate->attribute_buf.maxlen <= cstate->line_buf.len)
1860 0 : enlargeStringInfo(&cstate->attribute_buf, cstate->line_buf.len);
1861 486 : output_ptr = cstate->attribute_buf.data;
1862 :
1863 : /* set pointer variables for loop */
1864 486 : cur_ptr = cstate->line_buf.data;
1865 486 : line_end_ptr = cstate->line_buf.data + cstate->line_buf.len;
1866 :
1867 : /* Outer loop iterates over fields */
1868 486 : fieldno = 0;
1869 : for (;;)
1870 530 : {
1871 1016 : bool found_delim = false;
1872 1016 : bool saw_quote = false;
1873 : char *start_ptr;
1874 : char *end_ptr;
1875 : int input_len;
1876 :
1877 : /* Make sure there is enough space for the next value */
1878 1016 : if (fieldno >= cstate->max_fields)
1879 : {
1880 0 : cstate->max_fields *= 2;
1881 0 : cstate->raw_fields =
1882 0 : repalloc(cstate->raw_fields, cstate->max_fields * sizeof(char *));
1883 : }
1884 :
1885 : /* Remember start of field on both input and output sides */
1886 1016 : start_ptr = cur_ptr;
1887 1016 : cstate->raw_fields[fieldno] = output_ptr;
1888 :
1889 : /*
1890 : * Scan data for field,
1891 : *
1892 : * The loop starts in "not quote" mode and then toggles between that
1893 : * and "in quote" mode. The loop exits normally if it is in "not
1894 : * quote" mode and a delimiter or line end is seen.
1895 : */
1896 : for (;;)
1897 222 : {
1898 : char c;
1899 :
1900 : /* Not in quote */
1901 : for (;;)
1902 : {
1903 3254 : end_ptr = cur_ptr;
1904 3254 : if (cur_ptr >= line_end_ptr)
1905 480 : goto endfield;
1906 2774 : c = *cur_ptr++;
1907 : /* unquoted field delimiter */
1908 2774 : if (c == delimc)
1909 : {
1910 536 : found_delim = true;
1911 536 : goto endfield;
1912 : }
1913 : /* start of quoted field (or part of field) */
1914 2238 : if (c == quotec)
1915 : {
1916 222 : saw_quote = true;
1917 222 : break;
1918 : }
1919 : /* Add c to output string */
1920 2016 : *output_ptr++ = c;
1921 : }
1922 :
1923 : /* In quote */
1924 : for (;;)
1925 : {
1926 1390 : end_ptr = cur_ptr;
1927 1390 : if (cur_ptr >= line_end_ptr)
1928 0 : ereport(ERROR,
1929 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1930 : errmsg("unterminated CSV quoted field")));
1931 :
1932 1390 : c = *cur_ptr++;
1933 :
1934 : /* escape within a quoted field */
1935 1390 : if (c == escapec)
1936 : {
1937 : /*
1938 : * peek at the next char if available, and escape it if it
1939 : * is an escape char or a quote char
1940 : */
1941 118 : if (cur_ptr < line_end_ptr)
1942 : {
1943 72 : char nextc = *cur_ptr;
1944 :
1945 72 : if (nextc == escapec || nextc == quotec)
1946 : {
1947 24 : *output_ptr++ = nextc;
1948 24 : cur_ptr++;
1949 24 : continue;
1950 : }
1951 : }
1952 : }
1953 :
1954 : /*
1955 : * end of quoted field. Must do this test after testing for
1956 : * escape in case quote char and escape char are the same
1957 : * (which is the common case).
1958 : */
1959 1366 : if (c == quotec)
1960 222 : break;
1961 :
1962 : /* Add c to output string */
1963 1144 : *output_ptr++ = c;
1964 : }
1965 : }
1966 1016 : endfield:
1967 :
1968 : /* Terminate attribute value in output area */
1969 1016 : *output_ptr++ = '\0';
1970 :
1971 : /* Check whether raw input matched null marker */
1972 1016 : input_len = end_ptr - start_ptr;
1973 1016 : if (!saw_quote && input_len == cstate->opts.null_print_len &&
1974 44 : strncmp(start_ptr, cstate->opts.null_print, input_len) == 0)
1975 44 : cstate->raw_fields[fieldno] = NULL;
1976 : /* Check whether raw input matched default marker */
1977 972 : else if (fieldno < list_length(cstate->attnumlist) &&
1978 972 : cstate->opts.default_print &&
1979 150 : input_len == cstate->opts.default_print_len &&
1980 42 : strncmp(start_ptr, cstate->opts.default_print, input_len) == 0)
1981 : {
1982 : /* fieldno is 0-index and attnum is 1-index */
1983 42 : int m = list_nth_int(cstate->attnumlist, fieldno) - 1;
1984 :
1985 42 : if (cstate->defexprs[m] != NULL)
1986 : {
1987 : /* defaults contain entries for all physical attributes */
1988 36 : cstate->defaults[m] = true;
1989 : }
1990 : else
1991 : {
1992 6 : TupleDesc tupDesc = RelationGetDescr(cstate->rel);
1993 6 : Form_pg_attribute att = TupleDescAttr(tupDesc, m);
1994 :
1995 6 : ereport(ERROR,
1996 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
1997 : errmsg("unexpected default marker in COPY data"),
1998 : errdetail("Column \"%s\" has no default value.",
1999 : NameStr(att->attname))));
2000 : }
2001 : }
2002 :
2003 1010 : fieldno++;
2004 : /* Done if we hit EOL instead of a delim */
2005 1010 : if (!found_delim)
2006 480 : break;
2007 : }
2008 :
2009 : /* Clean up state of attribute_buf */
2010 480 : output_ptr--;
2011 : Assert(*output_ptr == '\0');
2012 480 : cstate->attribute_buf.len = (output_ptr - cstate->attribute_buf.data);
2013 :
2014 480 : return fieldno;
2015 : }
2016 :
2017 :
2018 : /*
2019 : * Read a binary attribute
2020 : */
2021 : static Datum
2022 158 : CopyReadBinaryAttribute(CopyFromState cstate, FmgrInfo *flinfo,
2023 : Oid typioparam, int32 typmod,
2024 : bool *isnull)
2025 : {
2026 : int32 fld_size;
2027 : Datum result;
2028 :
2029 158 : if (!CopyGetInt32(cstate, &fld_size))
2030 0 : ereport(ERROR,
2031 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
2032 : errmsg("unexpected EOF in COPY data")));
2033 158 : if (fld_size == -1)
2034 : {
2035 30 : *isnull = true;
2036 30 : return ReceiveFunctionCall(flinfo, NULL, typioparam, typmod);
2037 : }
2038 128 : if (fld_size < 0)
2039 0 : ereport(ERROR,
2040 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
2041 : errmsg("invalid field size")));
2042 :
2043 : /* reset attribute_buf to empty, and load raw data in it */
2044 128 : resetStringInfo(&cstate->attribute_buf);
2045 :
2046 128 : enlargeStringInfo(&cstate->attribute_buf, fld_size);
2047 128 : if (CopyReadBinaryData(cstate, cstate->attribute_buf.data,
2048 128 : fld_size) != fld_size)
2049 0 : ereport(ERROR,
2050 : (errcode(ERRCODE_BAD_COPY_FILE_FORMAT),
2051 : errmsg("unexpected EOF in COPY data")));
2052 :
2053 128 : cstate->attribute_buf.len = fld_size;
2054 128 : cstate->attribute_buf.data[fld_size] = '\0';
2055 :
2056 : /* Call the column type's binary input converter */
2057 128 : result = ReceiveFunctionCall(flinfo, &cstate->attribute_buf,
2058 : typioparam, typmod);
2059 :
2060 : /* Trouble if it didn't eat the whole buffer */
2061 128 : if (cstate->attribute_buf.cursor != cstate->attribute_buf.len)
2062 2 : ereport(ERROR,
2063 : (errcode(ERRCODE_INVALID_BINARY_REPRESENTATION),
2064 : errmsg("incorrect binary data format")));
2065 :
2066 126 : *isnull = false;
2067 126 : return result;
2068 : }
|
