Line data Source code
1 : /*-------------------------------------------------------------------------
2 : *
3 : * mac.c
4 : * PostgreSQL type definitions for 6 byte, EUI-48, MAC addresses.
5 : *
6 : * Portions Copyright (c) 1998-2024, PostgreSQL Global Development Group
7 : *
8 : * IDENTIFICATION
9 : * src/backend/utils/adt/mac.c
10 : *
11 : *-------------------------------------------------------------------------
12 : */
13 :
14 : #include "postgres.h"
15 :
16 : #include "common/hashfn.h"
17 : #include "lib/hyperloglog.h"
18 : #include "libpq/pqformat.h"
19 : #include "port/pg_bswap.h"
20 : #include "utils/fmgrprotos.h"
21 : #include "utils/guc.h"
22 : #include "utils/inet.h"
23 : #include "utils/sortsupport.h"
24 :
25 :
26 : /*
27 : * Utility macros used for sorting and comparing:
28 : */
29 :
30 : #define hibits(addr) \
31 : ((unsigned long)(((addr)->a<<16)|((addr)->b<<8)|((addr)->c)))
32 :
33 : #define lobits(addr) \
34 : ((unsigned long)(((addr)->d<<16)|((addr)->e<<8)|((addr)->f)))
35 :
36 : /* sortsupport for macaddr */
37 : typedef struct
38 : {
39 : int64 input_count; /* number of non-null values seen */
40 : bool estimating; /* true if estimating cardinality */
41 :
42 : hyperLogLogState abbr_card; /* cardinality estimator */
43 : } macaddr_sortsupport_state;
44 :
45 : static int macaddr_cmp_internal(macaddr *a1, macaddr *a2);
46 : static int macaddr_fast_cmp(Datum x, Datum y, SortSupport ssup);
47 : static bool macaddr_abbrev_abort(int memtupcount, SortSupport ssup);
48 : static Datum macaddr_abbrev_convert(Datum original, SortSupport ssup);
49 :
50 : /*
51 : * MAC address reader. Accepts several common notations.
52 : */
53 :
54 : Datum
55 3728 : macaddr_in(PG_FUNCTION_ARGS)
56 : {
57 3728 : char *str = PG_GETARG_CSTRING(0);
58 3728 : Node *escontext = fcinfo->context;
59 : macaddr *result;
60 : int a,
61 : b,
62 : c,
63 : d,
64 : e,
65 : f;
66 : char junk[2];
67 : int count;
68 :
69 : /* %1s matches iff there is trailing non-whitespace garbage */
70 :
71 3728 : count = sscanf(str, "%x:%x:%x:%x:%x:%x%1s",
72 : &a, &b, &c, &d, &e, &f, junk);
73 3728 : if (count != 6)
74 72 : count = sscanf(str, "%x-%x-%x-%x-%x-%x%1s",
75 : &a, &b, &c, &d, &e, &f, junk);
76 3728 : if (count != 6)
77 66 : count = sscanf(str, "%2x%2x%2x:%2x%2x%2x%1s",
78 : &a, &b, &c, &d, &e, &f, junk);
79 3728 : if (count != 6)
80 60 : count = sscanf(str, "%2x%2x%2x-%2x%2x%2x%1s",
81 : &a, &b, &c, &d, &e, &f, junk);
82 3728 : if (count != 6)
83 54 : count = sscanf(str, "%2x%2x.%2x%2x.%2x%2x%1s",
84 : &a, &b, &c, &d, &e, &f, junk);
85 3728 : if (count != 6)
86 48 : count = sscanf(str, "%2x%2x-%2x%2x-%2x%2x%1s",
87 : &a, &b, &c, &d, &e, &f, junk);
88 3728 : if (count != 6)
89 42 : count = sscanf(str, "%2x%2x%2x%2x%2x%2x%1s",
90 : &a, &b, &c, &d, &e, &f, junk);
91 3728 : if (count != 6)
92 36 : ereturn(escontext, (Datum) 0,
93 : (errcode(ERRCODE_INVALID_TEXT_REPRESENTATION),
94 : errmsg("invalid input syntax for type %s: \"%s\"", "macaddr",
95 : str)));
96 :
97 3692 : if ((a < 0) || (a > 255) || (b < 0) || (b > 255) ||
98 3692 : (c < 0) || (c > 255) || (d < 0) || (d > 255) ||
99 3692 : (e < 0) || (e > 255) || (f < 0) || (f > 255))
100 0 : ereturn(escontext, (Datum) 0,
101 : (errcode(ERRCODE_NUMERIC_VALUE_OUT_OF_RANGE),
102 : errmsg("invalid octet value in \"macaddr\" value: \"%s\"", str)));
103 :
104 3692 : result = (macaddr *) palloc(sizeof(macaddr));
105 :
106 3692 : result->a = a;
107 3692 : result->b = b;
108 3692 : result->c = c;
109 3692 : result->d = d;
110 3692 : result->e = e;
111 3692 : result->f = f;
112 :
113 3692 : PG_RETURN_MACADDR_P(result);
114 : }
115 :
116 : /*
117 : * MAC address output function. Fixed format.
118 : */
119 :
120 : Datum
121 3208 : macaddr_out(PG_FUNCTION_ARGS)
122 : {
123 3208 : macaddr *addr = PG_GETARG_MACADDR_P(0);
124 : char *result;
125 :
126 3208 : result = (char *) palloc(32);
127 :
128 3208 : snprintf(result, 32, "%02x:%02x:%02x:%02x:%02x:%02x",
129 3208 : addr->a, addr->b, addr->c, addr->d, addr->e, addr->f);
130 :
131 3208 : PG_RETURN_CSTRING(result);
132 : }
133 :
134 : /*
135 : * macaddr_recv - converts external binary format to macaddr
136 : *
137 : * The external representation is just the six bytes, MSB first.
138 : */
139 : Datum
140 0 : macaddr_recv(PG_FUNCTION_ARGS)
141 : {
142 0 : StringInfo buf = (StringInfo) PG_GETARG_POINTER(0);
143 : macaddr *addr;
144 :
145 0 : addr = (macaddr *) palloc(sizeof(macaddr));
146 :
147 0 : addr->a = pq_getmsgbyte(buf);
148 0 : addr->b = pq_getmsgbyte(buf);
149 0 : addr->c = pq_getmsgbyte(buf);
150 0 : addr->d = pq_getmsgbyte(buf);
151 0 : addr->e = pq_getmsgbyte(buf);
152 0 : addr->f = pq_getmsgbyte(buf);
153 :
154 0 : PG_RETURN_MACADDR_P(addr);
155 : }
156 :
157 : /*
158 : * macaddr_send - converts macaddr to binary format
159 : */
160 : Datum
161 0 : macaddr_send(PG_FUNCTION_ARGS)
162 : {
163 0 : macaddr *addr = PG_GETARG_MACADDR_P(0);
164 : StringInfoData buf;
165 :
166 0 : pq_begintypsend(&buf);
167 0 : pq_sendbyte(&buf, addr->a);
168 0 : pq_sendbyte(&buf, addr->b);
169 0 : pq_sendbyte(&buf, addr->c);
170 0 : pq_sendbyte(&buf, addr->d);
171 0 : pq_sendbyte(&buf, addr->e);
172 0 : pq_sendbyte(&buf, addr->f);
173 0 : PG_RETURN_BYTEA_P(pq_endtypsend(&buf));
174 : }
175 :
176 :
177 : /*
178 : * Comparison function for sorting:
179 : */
180 :
181 : static int
182 61484 : macaddr_cmp_internal(macaddr *a1, macaddr *a2)
183 : {
184 61484 : if (hibits(a1) < hibits(a2))
185 25120 : return -1;
186 36364 : else if (hibits(a1) > hibits(a2))
187 24822 : return 1;
188 11542 : else if (lobits(a1) < lobits(a2))
189 58 : return -1;
190 11484 : else if (lobits(a1) > lobits(a2))
191 50 : return 1;
192 : else
193 11434 : return 0;
194 : }
195 :
196 : Datum
197 13278 : macaddr_cmp(PG_FUNCTION_ARGS)
198 : {
199 13278 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
200 13278 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
201 :
202 13278 : PG_RETURN_INT32(macaddr_cmp_internal(a1, a2));
203 : }
204 :
205 : /*
206 : * Boolean comparisons.
207 : */
208 :
209 : Datum
210 20014 : macaddr_lt(PG_FUNCTION_ARGS)
211 : {
212 20014 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
213 20014 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
214 :
215 20014 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) < 0);
216 : }
217 :
218 : Datum
219 6236 : macaddr_le(PG_FUNCTION_ARGS)
220 : {
221 6236 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
222 6236 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
223 :
224 6236 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) <= 0);
225 : }
226 :
227 : Datum
228 7660 : macaddr_eq(PG_FUNCTION_ARGS)
229 : {
230 7660 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
231 7660 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
232 :
233 7660 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) == 0);
234 : }
235 :
236 : Datum
237 4916 : macaddr_ge(PG_FUNCTION_ARGS)
238 : {
239 4916 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
240 4916 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
241 :
242 4916 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) >= 0);
243 : }
244 :
245 : Datum
246 9032 : macaddr_gt(PG_FUNCTION_ARGS)
247 : {
248 9032 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
249 9032 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
250 :
251 9032 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) > 0);
252 : }
253 :
254 : Datum
255 24 : macaddr_ne(PG_FUNCTION_ARGS)
256 : {
257 24 : macaddr *a1 = PG_GETARG_MACADDR_P(0);
258 24 : macaddr *a2 = PG_GETARG_MACADDR_P(1);
259 :
260 24 : PG_RETURN_BOOL(macaddr_cmp_internal(a1, a2) != 0);
261 : }
262 :
263 : /*
264 : * Support function for hash indexes on macaddr.
265 : */
266 : Datum
267 2334 : hashmacaddr(PG_FUNCTION_ARGS)
268 : {
269 2334 : macaddr *key = PG_GETARG_MACADDR_P(0);
270 :
271 2334 : return hash_any((unsigned char *) key, sizeof(macaddr));
272 : }
273 :
274 : Datum
275 60 : hashmacaddrextended(PG_FUNCTION_ARGS)
276 : {
277 60 : macaddr *key = PG_GETARG_MACADDR_P(0);
278 :
279 60 : return hash_any_extended((unsigned char *) key, sizeof(macaddr),
280 60 : PG_GETARG_INT64(1));
281 : }
282 :
283 : /*
284 : * Arithmetic functions: bitwise NOT, AND, OR.
285 : */
286 : Datum
287 72 : macaddr_not(PG_FUNCTION_ARGS)
288 : {
289 72 : macaddr *addr = PG_GETARG_MACADDR_P(0);
290 : macaddr *result;
291 :
292 72 : result = (macaddr *) palloc(sizeof(macaddr));
293 72 : result->a = ~addr->a;
294 72 : result->b = ~addr->b;
295 72 : result->c = ~addr->c;
296 72 : result->d = ~addr->d;
297 72 : result->e = ~addr->e;
298 72 : result->f = ~addr->f;
299 72 : PG_RETURN_MACADDR_P(result);
300 : }
301 :
302 : Datum
303 72 : macaddr_and(PG_FUNCTION_ARGS)
304 : {
305 72 : macaddr *addr1 = PG_GETARG_MACADDR_P(0);
306 72 : macaddr *addr2 = PG_GETARG_MACADDR_P(1);
307 : macaddr *result;
308 :
309 72 : result = (macaddr *) palloc(sizeof(macaddr));
310 72 : result->a = addr1->a & addr2->a;
311 72 : result->b = addr1->b & addr2->b;
312 72 : result->c = addr1->c & addr2->c;
313 72 : result->d = addr1->d & addr2->d;
314 72 : result->e = addr1->e & addr2->e;
315 72 : result->f = addr1->f & addr2->f;
316 72 : PG_RETURN_MACADDR_P(result);
317 : }
318 :
319 : Datum
320 72 : macaddr_or(PG_FUNCTION_ARGS)
321 : {
322 72 : macaddr *addr1 = PG_GETARG_MACADDR_P(0);
323 72 : macaddr *addr2 = PG_GETARG_MACADDR_P(1);
324 : macaddr *result;
325 :
326 72 : result = (macaddr *) palloc(sizeof(macaddr));
327 72 : result->a = addr1->a | addr2->a;
328 72 : result->b = addr1->b | addr2->b;
329 72 : result->c = addr1->c | addr2->c;
330 72 : result->d = addr1->d | addr2->d;
331 72 : result->e = addr1->e | addr2->e;
332 72 : result->f = addr1->f | addr2->f;
333 72 : PG_RETURN_MACADDR_P(result);
334 : }
335 :
336 : /*
337 : * Truncation function to allow comparing mac manufacturers.
338 : * From suggestion by Alex Pilosov <alex@pilosoft.com>
339 : */
340 : Datum
341 72 : macaddr_trunc(PG_FUNCTION_ARGS)
342 : {
343 72 : macaddr *addr = PG_GETARG_MACADDR_P(0);
344 : macaddr *result;
345 :
346 72 : result = (macaddr *) palloc(sizeof(macaddr));
347 :
348 72 : result->a = addr->a;
349 72 : result->b = addr->b;
350 72 : result->c = addr->c;
351 72 : result->d = 0;
352 72 : result->e = 0;
353 72 : result->f = 0;
354 :
355 72 : PG_RETURN_MACADDR_P(result);
356 : }
357 :
358 : /*
359 : * SortSupport strategy function. Populates a SortSupport struct with the
360 : * information necessary to use comparison by abbreviated keys.
361 : */
362 : Datum
363 20 : macaddr_sortsupport(PG_FUNCTION_ARGS)
364 : {
365 20 : SortSupport ssup = (SortSupport) PG_GETARG_POINTER(0);
366 :
367 20 : ssup->comparator = macaddr_fast_cmp;
368 20 : ssup->ssup_extra = NULL;
369 :
370 20 : if (ssup->abbreviate)
371 : {
372 : macaddr_sortsupport_state *uss;
373 : MemoryContext oldcontext;
374 :
375 20 : oldcontext = MemoryContextSwitchTo(ssup->ssup_cxt);
376 :
377 20 : uss = palloc(sizeof(macaddr_sortsupport_state));
378 20 : uss->input_count = 0;
379 20 : uss->estimating = true;
380 20 : initHyperLogLog(&uss->abbr_card, 10);
381 :
382 20 : ssup->ssup_extra = uss;
383 :
384 20 : ssup->comparator = ssup_datum_unsigned_cmp;
385 20 : ssup->abbrev_converter = macaddr_abbrev_convert;
386 20 : ssup->abbrev_abort = macaddr_abbrev_abort;
387 20 : ssup->abbrev_full_comparator = macaddr_fast_cmp;
388 :
389 20 : MemoryContextSwitchTo(oldcontext);
390 : }
391 :
392 20 : PG_RETURN_VOID();
393 : }
394 :
395 : /*
396 : * SortSupport "traditional" comparison function. Pulls two MAC addresses from
397 : * the heap and runs a standard comparison on them.
398 : */
399 : static int
400 324 : macaddr_fast_cmp(Datum x, Datum y, SortSupport ssup)
401 : {
402 324 : macaddr *arg1 = DatumGetMacaddrP(x);
403 324 : macaddr *arg2 = DatumGetMacaddrP(y);
404 :
405 324 : return macaddr_cmp_internal(arg1, arg2);
406 : }
407 :
408 : /*
409 : * Callback for estimating effectiveness of abbreviated key optimization.
410 : *
411 : * We pay no attention to the cardinality of the non-abbreviated data, because
412 : * there is no equality fast-path within authoritative macaddr comparator.
413 : */
414 : static bool
415 12 : macaddr_abbrev_abort(int memtupcount, SortSupport ssup)
416 : {
417 12 : macaddr_sortsupport_state *uss = ssup->ssup_extra;
418 : double abbr_card;
419 :
420 12 : if (memtupcount < 10000 || uss->input_count < 10000 || !uss->estimating)
421 12 : return false;
422 :
423 0 : abbr_card = estimateHyperLogLog(&uss->abbr_card);
424 :
425 : /*
426 : * If we have >100k distinct values, then even if we were sorting many
427 : * billion rows we'd likely still break even, and the penalty of undoing
428 : * that many rows of abbrevs would probably not be worth it. At this point
429 : * we stop counting because we know that we're now fully committed.
430 : */
431 0 : if (abbr_card > 100000.0)
432 : {
433 0 : if (trace_sort)
434 0 : elog(LOG,
435 : "macaddr_abbrev: estimation ends at cardinality %f"
436 : " after " INT64_FORMAT " values (%d rows)",
437 : abbr_card, uss->input_count, memtupcount);
438 0 : uss->estimating = false;
439 0 : return false;
440 : }
441 :
442 : /*
443 : * Target minimum cardinality is 1 per ~2k of non-null inputs. 0.5 row
444 : * fudge factor allows us to abort earlier on genuinely pathological data
445 : * where we've had exactly one abbreviated value in the first 2k
446 : * (non-null) rows.
447 : */
448 0 : if (abbr_card < uss->input_count / 2000.0 + 0.5)
449 : {
450 0 : if (trace_sort)
451 0 : elog(LOG,
452 : "macaddr_abbrev: aborting abbreviation at cardinality %f"
453 : " below threshold %f after " INT64_FORMAT " values (%d rows)",
454 : abbr_card, uss->input_count / 2000.0 + 0.5, uss->input_count,
455 : memtupcount);
456 0 : return true;
457 : }
458 :
459 0 : if (trace_sort)
460 0 : elog(LOG,
461 : "macaddr_abbrev: cardinality %f after " INT64_FORMAT
462 : " values (%d rows)", abbr_card, uss->input_count, memtupcount);
463 :
464 0 : return false;
465 : }
466 :
467 : /*
468 : * SortSupport conversion routine. Converts original macaddr representation
469 : * to abbreviated key representation.
470 : *
471 : * Packs the bytes of a 6-byte MAC address into a Datum and treats it as an
472 : * unsigned integer for purposes of comparison. On a 64-bit machine, there
473 : * will be two zeroed bytes of padding. The integer is converted to native
474 : * endianness to facilitate easy comparison.
475 : */
476 : static Datum
477 168 : macaddr_abbrev_convert(Datum original, SortSupport ssup)
478 : {
479 168 : macaddr_sortsupport_state *uss = ssup->ssup_extra;
480 168 : macaddr *authoritative = DatumGetMacaddrP(original);
481 : Datum res;
482 :
483 : /*
484 : * On a 64-bit machine, zero out the 8-byte datum and copy the 6 bytes of
485 : * the MAC address in. There will be two bytes of zero padding on the end
486 : * of the least significant bits.
487 : */
488 : #if SIZEOF_DATUM == 8
489 : memset(&res, 0, SIZEOF_DATUM);
490 168 : memcpy(&res, authoritative, sizeof(macaddr));
491 : #else /* SIZEOF_DATUM != 8 */
492 : memcpy(&res, authoritative, SIZEOF_DATUM);
493 : #endif
494 168 : uss->input_count += 1;
495 :
496 : /*
497 : * Cardinality estimation. The estimate uses uint32, so on a 64-bit
498 : * architecture, XOR the two 32-bit halves together to produce slightly
499 : * more entropy. The two zeroed bytes won't have any practical impact on
500 : * this operation.
501 : */
502 168 : if (uss->estimating)
503 : {
504 : uint32 tmp;
505 :
506 : #if SIZEOF_DATUM == 8
507 168 : tmp = (uint32) res ^ (uint32) ((uint64) res >> 32);
508 : #else /* SIZEOF_DATUM != 8 */
509 : tmp = (uint32) res;
510 : #endif
511 :
512 168 : addHyperLogLog(&uss->abbr_card, DatumGetUInt32(hash_uint32(tmp)));
513 : }
514 :
515 : /*
516 : * Byteswap on little-endian machines.
517 : *
518 : * This is needed so that ssup_datum_unsigned_cmp() (an unsigned integer
519 : * 3-way comparator) works correctly on all platforms. Without this, the
520 : * comparator would have to call memcmp() with a pair of pointers to the
521 : * first byte of each abbreviated key, which is slower.
522 : */
523 168 : res = DatumBigEndianToNative(res);
524 :
525 168 : return res;
526 : }
|
