Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * datum.c
4 : * POSTGRES Datum (abstract data type) manipulation routines.
5 : *
6 : * Portions Copyright (c) 1996-2019, PostgreSQL Global Development Group
7 : * Portions Copyright (c) 1994, Regents of the University of California
8 : *
9 : *
10 : * IDENTIFICATION
11 : * src/backend/utils/adt/datum.c
12 : *
13 : *-------------------------------------------------------------------------
14 : */
15 :
16 : /*
17 : * In the implementation of these routines we assume the following:
18 : *
19 : * A) if a type is "byVal" then all the information is stored in the
20 : * Datum itself (i.e. no pointers involved!). In this case the
21 : * length of the type is always greater than zero and not more than
22 : * "sizeof(Datum)"
23 : *
24 : * B) if a type is not "byVal" and it has a fixed length (typlen > 0),
25 : * then the "Datum" always contains a pointer to a stream of bytes.
26 : * The number of significant bytes are always equal to the typlen.
27 : *
28 : * C) if a type is not "byVal" and has typlen == -1,
29 : * then the "Datum" always points to a "struct varlena".
30 : * This varlena structure has information about the actual length of this
31 : * particular instance of the type and about its value.
32 : *
33 : * D) if a type is not "byVal" and has typlen == -2,
34 : * then the "Datum" always points to a null-terminated C string.
35 : *
36 : * Note that we do not treat "toasted" datums specially; therefore what
37 : * will be copied or compared is the compressed data or toast reference.
38 : * An exception is made for datumCopy() of an expanded object, however,
39 : * because most callers expect to get a simple contiguous (and pfree'able)
40 : * result from datumCopy(). See also datumTransfer().
41 : */
42 :
43 : #include "postgres.h"
44 :
45 : #include "access/detoast.h"
46 : #include "fmgr.h"
47 : #include "utils/datum.h"
48 : #include "utils/expandeddatum.h"
49 :
50 :
51 : /*-------------------------------------------------------------------------
52 : * datumGetSize
53 : *
54 : * Find the "real" size of a datum, given the datum value,
55 : * whether it is a "by value", and the declared type length.
56 : * (For TOAST pointer datums, this is the size of the pointer datum.)
57 : *
58 : * This is essentially an out-of-line version of the att_addlength_datum()
59 : * macro in access/tupmacs.h. We do a tad more error checking though.
60 : *-------------------------------------------------------------------------
61 : */
62 : Size
63 5985598 : datumGetSize(Datum value, bool typByVal, int typLen)
64 : {
65 : Size size;
66 :
67 5985598 : if (typByVal)
68 : {
69 : /* Pass-by-value types are always fixed-length */
70 : Assert(typLen > 0 && typLen <= sizeof(Datum));
71 1013466 : size = (Size) typLen;
72 : }
73 : else
74 : {
75 4972132 : if (typLen > 0)
76 : {
77 : /* Fixed-length pass-by-ref type */
78 3451028 : size = (Size) typLen;
79 : }
80 1521104 : else if (typLen == -1)
81 : {
82 : /* It is a varlena datatype */
83 1405938 : struct varlena *s = (struct varlena *) DatumGetPointer(value);
84 :
85 1405938 : if (!PointerIsValid(s))
86 0 : ereport(ERROR,
87 : (errcode(ERRCODE_DATA_EXCEPTION),
88 : errmsg("invalid Datum pointer")));
89 :
90 1405938 : size = (Size) VARSIZE_ANY(s);
91 : }
92 115166 : else if (typLen == -2)
93 : {
94 : /* It is a cstring datatype */
95 115166 : char *s = (char *) DatumGetPointer(value);
96 :
97 115166 : if (!PointerIsValid(s))
98 0 : ereport(ERROR,
99 : (errcode(ERRCODE_DATA_EXCEPTION),
100 : errmsg("invalid Datum pointer")));
101 :
102 115166 : size = (Size) (strlen(s) + 1);
103 : }
104 : else
105 : {
106 0 : elog(ERROR, "invalid typLen: %d", typLen);
107 : size = 0; /* keep compiler quiet */
108 : }
109 : }
110 :
111 5985598 : return size;
112 : }
113 :
114 : /*-------------------------------------------------------------------------
115 : * datumCopy
116 : *
117 : * Make a copy of a non-NULL datum.
118 : *
119 : * If the datatype is pass-by-reference, memory is obtained with palloc().
120 : *
121 : * If the value is a reference to an expanded object, we flatten into memory
122 : * obtained with palloc(). We need to copy because one of the main uses of
123 : * this function is to copy a datum out of a transient memory context that's
124 : * about to be destroyed, and the expanded object is probably in a child
125 : * context that will also go away. Moreover, many callers assume that the
126 : * result is a single pfree-able chunk.
127 : *-------------------------------------------------------------------------
128 : */
129 : Datum
130 15782252 : datumCopy(Datum value, bool typByVal, int typLen)
131 : {
132 : Datum res;
133 :
134 15782252 : if (typByVal)
135 9013888 : res = value;
136 6768364 : else if (typLen == -1)
137 : {
138 : /* It is a varlena datatype */
139 3831154 : struct varlena *vl = (struct varlena *) DatumGetPointer(value);
140 :
141 3831154 : if (VARATT_IS_EXTERNAL_EXPANDED(vl))
142 452 : {
143 : /* Flatten into the caller's memory context */
144 452 : ExpandedObjectHeader *eoh = DatumGetEOHP(value);
145 : Size resultsize;
146 : char *resultptr;
147 :
148 452 : resultsize = EOH_get_flat_size(eoh);
149 452 : resultptr = (char *) palloc(resultsize);
150 452 : EOH_flatten_into(eoh, (void *) resultptr, resultsize);
151 452 : res = PointerGetDatum(resultptr);
152 : }
153 : else
154 : {
155 : /* Otherwise, just copy the varlena datum verbatim */
156 : Size realSize;
157 : char *resultptr;
158 :
159 3830702 : realSize = (Size) VARSIZE_ANY(vl);
160 3830702 : resultptr = (char *) palloc(realSize);
161 3830702 : memcpy(resultptr, vl, realSize);
162 3830702 : res = PointerGetDatum(resultptr);
163 : }
164 : }
165 : else
166 : {
167 : /* Pass by reference, but not varlena, so not toasted */
168 : Size realSize;
169 : char *resultptr;
170 :
171 2937210 : realSize = datumGetSize(value, typByVal, typLen);
172 :
173 2937210 : resultptr = (char *) palloc(realSize);
174 2937210 : memcpy(resultptr, DatumGetPointer(value), realSize);
175 2937210 : res = PointerGetDatum(resultptr);
176 : }
177 15782252 : return res;
178 : }
179 :
180 : /*-------------------------------------------------------------------------
181 : * datumTransfer
182 : *
183 : * Transfer a non-NULL datum into the current memory context.
184 : *
185 : * This is equivalent to datumCopy() except when the datum is a read-write
186 : * pointer to an expanded object. In that case we merely reparent the object
187 : * into the current context, and return its standard R/W pointer (in case the
188 : * given one is a transient pointer of shorter lifespan).
189 : *-------------------------------------------------------------------------
190 : */
191 : Datum
192 24402 : datumTransfer(Datum value, bool typByVal, int typLen)
193 : {
194 47616 : if (!typByVal && typLen == -1 &&
195 26752 : VARATT_IS_EXTERNAL_EXPANDED_RW(DatumGetPointer(value)))
196 3534 : value = TransferExpandedObject(value, CurrentMemoryContext);
197 : else
198 20868 : value = datumCopy(value, typByVal, typLen);
199 24402 : return value;
200 : }
201 :
202 : /*-------------------------------------------------------------------------
203 : * datumIsEqual
204 : *
205 : * Return true if two datums are equal, false otherwise
206 : *
207 : * NOTE: XXX!
208 : * We just compare the bytes of the two values, one by one.
209 : * This routine will return false if there are 2 different
210 : * representations of the same value (something along the lines
211 : * of say the representation of zero in one's complement arithmetic).
212 : * Also, it will probably not give the answer you want if either
213 : * datum has been "toasted".
214 : *
215 : * Do not try to make this any smarter than it currently is with respect
216 : * to "toasted" datums, because some of the callers could be working in the
217 : * context of an aborted transaction.
218 : *-------------------------------------------------------------------------
219 : */
220 : bool
221 3110460 : datumIsEqual(Datum value1, Datum value2, bool typByVal, int typLen)
222 : {
223 : bool res;
224 :
225 3110460 : if (typByVal)
226 : {
227 : /*
228 : * just compare the two datums. NOTE: just comparing "len" bytes will
229 : * not do the work, because we do not know how these bytes are aligned
230 : * inside the "Datum". We assume instead that any given datatype is
231 : * consistent about how it fills extraneous bits in the Datum.
232 : */
233 2398862 : res = (value1 == value2);
234 : }
235 : else
236 : {
237 : Size size1,
238 : size2;
239 : char *s1,
240 : *s2;
241 :
242 : /*
243 : * Compare the bytes pointed by the pointers stored in the datums.
244 : */
245 711598 : size1 = datumGetSize(value1, typByVal, typLen);
246 711598 : size2 = datumGetSize(value2, typByVal, typLen);
247 711598 : if (size1 != size2)
248 14620 : return false;
249 696978 : s1 = (char *) DatumGetPointer(value1);
250 696978 : s2 = (char *) DatumGetPointer(value2);
251 696978 : res = (memcmp(s1, s2, size1) == 0);
252 : }
253 3095840 : return res;
254 : }
255 :
256 : /*-------------------------------------------------------------------------
257 : * datum_image_eq
258 : *
259 : * Compares two datums for identical contents, based on byte images. Return
260 : * true if the two datums are equal, false otherwise.
261 : *-------------------------------------------------------------------------
262 : */
263 : bool
264 1104 : datum_image_eq(Datum value1, Datum value2, bool typByVal, int typLen)
265 : {
266 1104 : bool result = true;
267 :
268 1104 : if (typLen == -1)
269 : {
270 : Size len1,
271 : len2;
272 :
273 86 : len1 = toast_raw_datum_size(value1);
274 86 : len2 = toast_raw_datum_size(value2);
275 : /* No need to de-toast if lengths don't match. */
276 86 : if (len1 != len2)
277 0 : result = false;
278 : else
279 : {
280 : struct varlena *arg1val;
281 : struct varlena *arg2val;
282 :
283 86 : arg1val = PG_DETOAST_DATUM_PACKED(value1);
284 86 : arg2val = PG_DETOAST_DATUM_PACKED(value2);
285 :
286 172 : result = (memcmp(VARDATA_ANY(arg1val),
287 86 : VARDATA_ANY(arg2val),
288 : len1 - VARHDRSZ) == 0);
289 :
290 : /* Only free memory if it's a copy made here. */
291 86 : if ((Pointer) arg1val != (Pointer) value1)
292 0 : pfree(arg1val);
293 86 : if ((Pointer) arg2val != (Pointer) value2)
294 0 : pfree(arg2val);
295 : }
296 : }
297 1018 : else if (typByVal)
298 : {
299 998 : result = (value1 == value2);
300 : }
301 : else
302 : {
303 20 : result = (memcmp(DatumGetPointer(value1),
304 : DatumGetPointer(value2),
305 : typLen) == 0);
306 : }
307 :
308 1104 : return result;
309 : }
310 :
311 : /*-------------------------------------------------------------------------
312 : * datumEstimateSpace
313 : *
314 : * Compute the amount of space that datumSerialize will require for a
315 : * particular Datum.
316 : *-------------------------------------------------------------------------
317 : */
318 : Size
319 168 : datumEstimateSpace(Datum value, bool isnull, bool typByVal, int typLen)
320 : {
321 168 : Size sz = sizeof(int);
322 :
323 168 : if (!isnull)
324 : {
325 : /* no need to use add_size, can't overflow */
326 168 : if (typByVal)
327 160 : sz += sizeof(Datum);
328 16 : else if (typLen == -1 &&
329 12 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
330 : {
331 : /* Expanded objects need to be flattened, see comment below */
332 4 : sz += EOH_get_flat_size(DatumGetEOHP(value));
333 : }
334 : else
335 4 : sz += datumGetSize(value, typByVal, typLen);
336 : }
337 :
338 168 : return sz;
339 : }
340 :
341 : /*-------------------------------------------------------------------------
342 : * datumSerialize
343 : *
344 : * Serialize a possibly-NULL datum into caller-provided storage.
345 : *
346 : * Note: "expanded" objects are flattened so as to produce a self-contained
347 : * representation, but other sorts of toast pointers are transferred as-is.
348 : * This is because the intended use of this function is to pass the value
349 : * to another process within the same database server. The other process
350 : * could not access an "expanded" object within this process's memory, but
351 : * we assume it can dereference the same TOAST pointers this one can.
352 : *
353 : * The format is as follows: first, we write a 4-byte header word, which
354 : * is either the length of a pass-by-reference datum, -1 for a
355 : * pass-by-value datum, or -2 for a NULL. If the value is NULL, nothing
356 : * further is written. If it is pass-by-value, sizeof(Datum) bytes
357 : * follow. Otherwise, the number of bytes indicated by the header word
358 : * follow. The caller is responsible for ensuring that there is enough
359 : * storage to store the number of bytes that will be written; use
360 : * datumEstimateSpace() to find out how many will be needed.
361 : * *start_address is updated to point to the byte immediately following
362 : * those written.
363 : *-------------------------------------------------------------------------
364 : */
365 : void
366 168 : datumSerialize(Datum value, bool isnull, bool typByVal, int typLen,
367 : char **start_address)
368 : {
369 168 : ExpandedObjectHeader *eoh = NULL;
370 : int header;
371 :
372 : /* Write header word. */
373 168 : if (isnull)
374 0 : header = -2;
375 168 : else if (typByVal)
376 160 : header = -1;
377 16 : else if (typLen == -1 &&
378 12 : VARATT_IS_EXTERNAL_EXPANDED(DatumGetPointer(value)))
379 : {
380 4 : eoh = DatumGetEOHP(value);
381 4 : header = EOH_get_flat_size(eoh);
382 : }
383 : else
384 4 : header = datumGetSize(value, typByVal, typLen);
385 168 : memcpy(*start_address, &header, sizeof(int));
386 168 : *start_address += sizeof(int);
387 :
388 : /* If not null, write payload bytes. */
389 168 : if (!isnull)
390 : {
391 168 : if (typByVal)
392 : {
393 160 : memcpy(*start_address, &value, sizeof(Datum));
394 160 : *start_address += sizeof(Datum);
395 : }
396 8 : else if (eoh)
397 : {
398 : char *tmp;
399 :
400 : /*
401 : * EOH_flatten_into expects the target address to be maxaligned,
402 : * so we can't store directly to *start_address.
403 : */
404 4 : tmp = (char *) palloc(header);
405 4 : EOH_flatten_into(eoh, (void *) tmp, header);
406 4 : memcpy(*start_address, tmp, header);
407 4 : *start_address += header;
408 :
409 : /* be tidy. */
410 4 : pfree(tmp);
411 : }
412 : else
413 : {
414 4 : memcpy(*start_address, DatumGetPointer(value), header);
415 4 : *start_address += header;
416 : }
417 : }
418 168 : }
419 :
420 : /*-------------------------------------------------------------------------
421 : * datumRestore
422 : *
423 : * Restore a possibly-NULL datum previously serialized by datumSerialize.
424 : * *start_address is updated according to the number of bytes consumed.
425 : *-------------------------------------------------------------------------
426 : */
427 : Datum
428 360 : datumRestore(char **start_address, bool *isnull)
429 : {
430 : int header;
431 : void *d;
432 :
433 : /* Read header word. */
434 360 : memcpy(&header, *start_address, sizeof(int));
435 360 : *start_address += sizeof(int);
436 :
437 : /* If this datum is NULL, we can stop here. */
438 360 : if (header == -2)
439 : {
440 0 : *isnull = true;
441 0 : return (Datum) 0;
442 : }
443 :
444 : /* OK, datum is not null. */
445 360 : *isnull = false;
446 :
447 : /* If this datum is pass-by-value, sizeof(Datum) bytes follow. */
448 360 : if (header == -1)
449 : {
450 : Datum val;
451 :
452 336 : memcpy(&val, *start_address, sizeof(Datum));
453 336 : *start_address += sizeof(Datum);
454 336 : return val;
455 : }
456 :
457 : /* Pass-by-reference case; copy indicated number of bytes. */
458 : Assert(header > 0);
459 24 : d = palloc(header);
460 24 : memcpy(d, *start_address, header);
461 24 : *start_address += header;
462 24 : return PointerGetDatum(d);
463 : }
|
