]> git.saurik.com Git - redis.git/blob - src/rdb.c
Sentinel: when leader in wait-start, sense another leader as race.
[redis.git] / src / rdb.c
1 #include "redis.h"
2 #include "lzf.h" /* LZF compression library */
3 #include "zipmap.h"
4 #include "endianconv.h"
5
6 #include <math.h>
7 #include <sys/types.h>
8 #include <sys/time.h>
9 #include <sys/resource.h>
10 #include <sys/wait.h>
11 #include <arpa/inet.h>
12 #include <sys/stat.h>
13
14 static int rdbWriteRaw(rio *rdb, void *p, size_t len) {
15 if (rdb && rioWrite(rdb,p,len) == 0)
16 return -1;
17 return len;
18 }
19
20 int rdbSaveType(rio *rdb, unsigned char type) {
21 return rdbWriteRaw(rdb,&type,1);
22 }
23
24 /* Load a "type" in RDB format, that is a one byte unsigned integer.
25 * This function is not only used to load object types, but also special
26 * "types" like the end-of-file type, the EXPIRE type, and so forth. */
27 int rdbLoadType(rio *rdb) {
28 unsigned char type;
29 if (rioRead(rdb,&type,1) == 0) return -1;
30 return type;
31 }
32
33 time_t rdbLoadTime(rio *rdb) {
34 int32_t t32;
35 if (rioRead(rdb,&t32,4) == 0) return -1;
36 return (time_t)t32;
37 }
38
39 int rdbSaveMillisecondTime(rio *rdb, long long t) {
40 int64_t t64 = (int64_t) t;
41 return rdbWriteRaw(rdb,&t64,8);
42 }
43
44 long long rdbLoadMillisecondTime(rio *rdb) {
45 int64_t t64;
46 if (rioRead(rdb,&t64,8) == 0) return -1;
47 return (long long)t64;
48 }
49
50 /* Saves an encoded length. The first two bits in the first byte are used to
51 * hold the encoding type. See the REDIS_RDB_* definitions for more information
52 * on the types of encoding. */
53 int rdbSaveLen(rio *rdb, uint32_t len) {
54 unsigned char buf[2];
55 size_t nwritten;
56
57 if (len < (1<<6)) {
58 /* Save a 6 bit len */
59 buf[0] = (len&0xFF)|(REDIS_RDB_6BITLEN<<6);
60 if (rdbWriteRaw(rdb,buf,1) == -1) return -1;
61 nwritten = 1;
62 } else if (len < (1<<14)) {
63 /* Save a 14 bit len */
64 buf[0] = ((len>>8)&0xFF)|(REDIS_RDB_14BITLEN<<6);
65 buf[1] = len&0xFF;
66 if (rdbWriteRaw(rdb,buf,2) == -1) return -1;
67 nwritten = 2;
68 } else {
69 /* Save a 32 bit len */
70 buf[0] = (REDIS_RDB_32BITLEN<<6);
71 if (rdbWriteRaw(rdb,buf,1) == -1) return -1;
72 len = htonl(len);
73 if (rdbWriteRaw(rdb,&len,4) == -4) return -1;
74 nwritten = 1+4;
75 }
76 return nwritten;
77 }
78
79 /* Load an encoded length. The "isencoded" argument is set to 1 if the length
80 * is not actually a length but an "encoding type". See the REDIS_RDB_ENC_*
81 * definitions in rdb.h for more information. */
82 uint32_t rdbLoadLen(rio *rdb, int *isencoded) {
83 unsigned char buf[2];
84 uint32_t len;
85 int type;
86
87 if (isencoded) *isencoded = 0;
88 if (rioRead(rdb,buf,1) == 0) return REDIS_RDB_LENERR;
89 type = (buf[0]&0xC0)>>6;
90 if (type == REDIS_RDB_ENCVAL) {
91 /* Read a 6 bit encoding type. */
92 if (isencoded) *isencoded = 1;
93 return buf[0]&0x3F;
94 } else if (type == REDIS_RDB_6BITLEN) {
95 /* Read a 6 bit len. */
96 return buf[0]&0x3F;
97 } else if (type == REDIS_RDB_14BITLEN) {
98 /* Read a 14 bit len. */
99 if (rioRead(rdb,buf+1,1) == 0) return REDIS_RDB_LENERR;
100 return ((buf[0]&0x3F)<<8)|buf[1];
101 } else {
102 /* Read a 32 bit len. */
103 if (rioRead(rdb,&len,4) == 0) return REDIS_RDB_LENERR;
104 return ntohl(len);
105 }
106 }
107
108 /* Encodes the "value" argument as integer when it fits in the supported ranges
109 * for encoded types. If the function successfully encodes the integer, the
110 * representation is stored in the buffer pointer to by "enc" and the string
111 * length is returned. Otherwise 0 is returned. */
112 int rdbEncodeInteger(long long value, unsigned char *enc) {
113 if (value >= -(1<<7) && value <= (1<<7)-1) {
114 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT8;
115 enc[1] = value&0xFF;
116 return 2;
117 } else if (value >= -(1<<15) && value <= (1<<15)-1) {
118 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT16;
119 enc[1] = value&0xFF;
120 enc[2] = (value>>8)&0xFF;
121 return 3;
122 } else if (value >= -((long long)1<<31) && value <= ((long long)1<<31)-1) {
123 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT32;
124 enc[1] = value&0xFF;
125 enc[2] = (value>>8)&0xFF;
126 enc[3] = (value>>16)&0xFF;
127 enc[4] = (value>>24)&0xFF;
128 return 5;
129 } else {
130 return 0;
131 }
132 }
133
134 /* Loads an integer-encoded object with the specified encoding type "enctype".
135 * If the "encode" argument is set the function may return an integer-encoded
136 * string object, otherwise it always returns a raw string object. */
137 robj *rdbLoadIntegerObject(rio *rdb, int enctype, int encode) {
138 unsigned char enc[4];
139 long long val;
140
141 if (enctype == REDIS_RDB_ENC_INT8) {
142 if (rioRead(rdb,enc,1) == 0) return NULL;
143 val = (signed char)enc[0];
144 } else if (enctype == REDIS_RDB_ENC_INT16) {
145 uint16_t v;
146 if (rioRead(rdb,enc,2) == 0) return NULL;
147 v = enc[0]|(enc[1]<<8);
148 val = (int16_t)v;
149 } else if (enctype == REDIS_RDB_ENC_INT32) {
150 uint32_t v;
151 if (rioRead(rdb,enc,4) == 0) return NULL;
152 v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
153 val = (int32_t)v;
154 } else {
155 val = 0; /* anti-warning */
156 redisPanic("Unknown RDB integer encoding type");
157 }
158 if (encode)
159 return createStringObjectFromLongLong(val);
160 else
161 return createObject(REDIS_STRING,sdsfromlonglong(val));
162 }
163
164 /* String objects in the form "2391" "-100" without any space and with a
165 * range of values that can fit in an 8, 16 or 32 bit signed value can be
166 * encoded as integers to save space */
167 int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) {
168 long long value;
169 char *endptr, buf[32];
170
171 /* Check if it's possible to encode this value as a number */
172 value = strtoll(s, &endptr, 10);
173 if (endptr[0] != '\0') return 0;
174 ll2string(buf,32,value);
175
176 /* If the number converted back into a string is not identical
177 * then it's not possible to encode the string as integer */
178 if (strlen(buf) != len || memcmp(buf,s,len)) return 0;
179
180 return rdbEncodeInteger(value,enc);
181 }
182
183 int rdbSaveLzfStringObject(rio *rdb, unsigned char *s, size_t len) {
184 size_t comprlen, outlen;
185 unsigned char byte;
186 int n, nwritten = 0;
187 void *out;
188
189 /* We require at least four bytes compression for this to be worth it */
190 if (len <= 4) return 0;
191 outlen = len-4;
192 if ((out = zmalloc(outlen+1)) == NULL) return 0;
193 comprlen = lzf_compress(s, len, out, outlen);
194 if (comprlen == 0) {
195 zfree(out);
196 return 0;
197 }
198 /* Data compressed! Let's save it on disk */
199 byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF;
200 if ((n = rdbWriteRaw(rdb,&byte,1)) == -1) goto writeerr;
201 nwritten += n;
202
203 if ((n = rdbSaveLen(rdb,comprlen)) == -1) goto writeerr;
204 nwritten += n;
205
206 if ((n = rdbSaveLen(rdb,len)) == -1) goto writeerr;
207 nwritten += n;
208
209 if ((n = rdbWriteRaw(rdb,out,comprlen)) == -1) goto writeerr;
210 nwritten += n;
211
212 zfree(out);
213 return nwritten;
214
215 writeerr:
216 zfree(out);
217 return -1;
218 }
219
220 robj *rdbLoadLzfStringObject(rio *rdb) {
221 unsigned int len, clen;
222 unsigned char *c = NULL;
223 sds val = NULL;
224
225 if ((clen = rdbLoadLen(rdb,NULL)) == REDIS_RDB_LENERR) return NULL;
226 if ((len = rdbLoadLen(rdb,NULL)) == REDIS_RDB_LENERR) return NULL;
227 if ((c = zmalloc(clen)) == NULL) goto err;
228 if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
229 if (rioRead(rdb,c,clen) == 0) goto err;
230 if (lzf_decompress(c,clen,val,len) == 0) goto err;
231 zfree(c);
232 return createObject(REDIS_STRING,val);
233 err:
234 zfree(c);
235 sdsfree(val);
236 return NULL;
237 }
238
239 /* Save a string objet as [len][data] on disk. If the object is a string
240 * representation of an integer value we try to save it in a special form */
241 int rdbSaveRawString(rio *rdb, unsigned char *s, size_t len) {
242 int enclen;
243 int n, nwritten = 0;
244
245 /* Try integer encoding */
246 if (len <= 11) {
247 unsigned char buf[5];
248 if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) {
249 if (rdbWriteRaw(rdb,buf,enclen) == -1) return -1;
250 return enclen;
251 }
252 }
253
254 /* Try LZF compression - under 20 bytes it's unable to compress even
255 * aaaaaaaaaaaaaaaaaa so skip it */
256 if (server.rdb_compression && len > 20) {
257 n = rdbSaveLzfStringObject(rdb,s,len);
258 if (n == -1) return -1;
259 if (n > 0) return n;
260 /* Return value of 0 means data can't be compressed, save the old way */
261 }
262
263 /* Store verbatim */
264 if ((n = rdbSaveLen(rdb,len)) == -1) return -1;
265 nwritten += n;
266 if (len > 0) {
267 if (rdbWriteRaw(rdb,s,len) == -1) return -1;
268 nwritten += len;
269 }
270 return nwritten;
271 }
272
273 /* Save a long long value as either an encoded string or a string. */
274 int rdbSaveLongLongAsStringObject(rio *rdb, long long value) {
275 unsigned char buf[32];
276 int n, nwritten = 0;
277 int enclen = rdbEncodeInteger(value,buf);
278 if (enclen > 0) {
279 return rdbWriteRaw(rdb,buf,enclen);
280 } else {
281 /* Encode as string */
282 enclen = ll2string((char*)buf,32,value);
283 redisAssert(enclen < 32);
284 if ((n = rdbSaveLen(rdb,enclen)) == -1) return -1;
285 nwritten += n;
286 if ((n = rdbWriteRaw(rdb,buf,enclen)) == -1) return -1;
287 nwritten += n;
288 }
289 return nwritten;
290 }
291
292 /* Like rdbSaveStringObjectRaw() but handle encoded objects */
293 int rdbSaveStringObject(rio *rdb, robj *obj) {
294 /* Avoid to decode the object, then encode it again, if the
295 * object is alrady integer encoded. */
296 if (obj->encoding == REDIS_ENCODING_INT) {
297 return rdbSaveLongLongAsStringObject(rdb,(long)obj->ptr);
298 } else {
299 redisAssertWithInfo(NULL,obj,obj->encoding == REDIS_ENCODING_RAW);
300 return rdbSaveRawString(rdb,obj->ptr,sdslen(obj->ptr));
301 }
302 }
303
304 robj *rdbGenericLoadStringObject(rio *rdb, int encode) {
305 int isencoded;
306 uint32_t len;
307 sds val;
308
309 len = rdbLoadLen(rdb,&isencoded);
310 if (isencoded) {
311 switch(len) {
312 case REDIS_RDB_ENC_INT8:
313 case REDIS_RDB_ENC_INT16:
314 case REDIS_RDB_ENC_INT32:
315 return rdbLoadIntegerObject(rdb,len,encode);
316 case REDIS_RDB_ENC_LZF:
317 return rdbLoadLzfStringObject(rdb);
318 default:
319 redisPanic("Unknown RDB encoding type");
320 }
321 }
322
323 if (len == REDIS_RDB_LENERR) return NULL;
324 val = sdsnewlen(NULL,len);
325 if (len && rioRead(rdb,val,len) == 0) {
326 sdsfree(val);
327 return NULL;
328 }
329 return createObject(REDIS_STRING,val);
330 }
331
332 robj *rdbLoadStringObject(rio *rdb) {
333 return rdbGenericLoadStringObject(rdb,0);
334 }
335
336 robj *rdbLoadEncodedStringObject(rio *rdb) {
337 return rdbGenericLoadStringObject(rdb,1);
338 }
339
340 /* Save a double value. Doubles are saved as strings prefixed by an unsigned
341 * 8 bit integer specifing the length of the representation.
342 * This 8 bit integer has special values in order to specify the following
343 * conditions:
344 * 253: not a number
345 * 254: + inf
346 * 255: - inf
347 */
348 int rdbSaveDoubleValue(rio *rdb, double val) {
349 unsigned char buf[128];
350 int len;
351
352 if (isnan(val)) {
353 buf[0] = 253;
354 len = 1;
355 } else if (!isfinite(val)) {
356 len = 1;
357 buf[0] = (val < 0) ? 255 : 254;
358 } else {
359 #if (DBL_MANT_DIG >= 52) && (LLONG_MAX == 0x7fffffffffffffffLL)
360 /* Check if the float is in a safe range to be casted into a
361 * long long. We are assuming that long long is 64 bit here.
362 * Also we are assuming that there are no implementations around where
363 * double has precision < 52 bit.
364 *
365 * Under this assumptions we test if a double is inside an interval
366 * where casting to long long is safe. Then using two castings we
367 * make sure the decimal part is zero. If all this is true we use
368 * integer printing function that is much faster. */
369 double min = -4503599627370495; /* (2^52)-1 */
370 double max = 4503599627370496; /* -(2^52) */
371 if (val > min && val < max && val == ((double)((long long)val)))
372 ll2string((char*)buf+1,sizeof(buf),(long long)val);
373 else
374 #endif
375 snprintf((char*)buf+1,sizeof(buf)-1,"%.17g",val);
376 buf[0] = strlen((char*)buf+1);
377 len = buf[0]+1;
378 }
379 return rdbWriteRaw(rdb,buf,len);
380 }
381
382 /* For information about double serialization check rdbSaveDoubleValue() */
383 int rdbLoadDoubleValue(rio *rdb, double *val) {
384 char buf[128];
385 unsigned char len;
386
387 if (rioRead(rdb,&len,1) == 0) return -1;
388 switch(len) {
389 case 255: *val = R_NegInf; return 0;
390 case 254: *val = R_PosInf; return 0;
391 case 253: *val = R_Nan; return 0;
392 default:
393 if (rioRead(rdb,buf,len) == 0) return -1;
394 buf[len] = '\0';
395 sscanf(buf, "%lg", val);
396 return 0;
397 }
398 }
399
400 /* Save the object type of object "o". */
401 int rdbSaveObjectType(rio *rdb, robj *o) {
402 switch (o->type) {
403 case REDIS_STRING:
404 return rdbSaveType(rdb,REDIS_RDB_TYPE_STRING);
405 case REDIS_LIST:
406 if (o->encoding == REDIS_ENCODING_ZIPLIST)
407 return rdbSaveType(rdb,REDIS_RDB_TYPE_LIST_ZIPLIST);
408 else if (o->encoding == REDIS_ENCODING_LINKEDLIST)
409 return rdbSaveType(rdb,REDIS_RDB_TYPE_LIST);
410 else
411 redisPanic("Unknown list encoding");
412 case REDIS_SET:
413 if (o->encoding == REDIS_ENCODING_INTSET)
414 return rdbSaveType(rdb,REDIS_RDB_TYPE_SET_INTSET);
415 else if (o->encoding == REDIS_ENCODING_HT)
416 return rdbSaveType(rdb,REDIS_RDB_TYPE_SET);
417 else
418 redisPanic("Unknown set encoding");
419 case REDIS_ZSET:
420 if (o->encoding == REDIS_ENCODING_ZIPLIST)
421 return rdbSaveType(rdb,REDIS_RDB_TYPE_ZSET_ZIPLIST);
422 else if (o->encoding == REDIS_ENCODING_SKIPLIST)
423 return rdbSaveType(rdb,REDIS_RDB_TYPE_ZSET);
424 else
425 redisPanic("Unknown sorted set encoding");
426 case REDIS_HASH:
427 if (o->encoding == REDIS_ENCODING_ZIPLIST)
428 return rdbSaveType(rdb,REDIS_RDB_TYPE_HASH_ZIPLIST);
429 else if (o->encoding == REDIS_ENCODING_HT)
430 return rdbSaveType(rdb,REDIS_RDB_TYPE_HASH);
431 else
432 redisPanic("Unknown hash encoding");
433 default:
434 redisPanic("Unknown object type");
435 }
436 return -1; /* avoid warning */
437 }
438
439 /* Use rdbLoadType() to load a TYPE in RDB format, but returns -1 if the
440 * type is not specifically a valid Object Type. */
441 int rdbLoadObjectType(rio *rdb) {
442 int type;
443 if ((type = rdbLoadType(rdb)) == -1) return -1;
444 if (!rdbIsObjectType(type)) return -1;
445 return type;
446 }
447
448 /* Save a Redis object. Returns -1 on error, 0 on success. */
449 int rdbSaveObject(rio *rdb, robj *o) {
450 int n, nwritten = 0;
451
452 if (o->type == REDIS_STRING) {
453 /* Save a string value */
454 if ((n = rdbSaveStringObject(rdb,o)) == -1) return -1;
455 nwritten += n;
456 } else if (o->type == REDIS_LIST) {
457 /* Save a list value */
458 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
459 size_t l = ziplistBlobLen((unsigned char*)o->ptr);
460
461 if ((n = rdbSaveRawString(rdb,o->ptr,l)) == -1) return -1;
462 nwritten += n;
463 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
464 list *list = o->ptr;
465 listIter li;
466 listNode *ln;
467
468 if ((n = rdbSaveLen(rdb,listLength(list))) == -1) return -1;
469 nwritten += n;
470
471 listRewind(list,&li);
472 while((ln = listNext(&li))) {
473 robj *eleobj = listNodeValue(ln);
474 if ((n = rdbSaveStringObject(rdb,eleobj)) == -1) return -1;
475 nwritten += n;
476 }
477 } else {
478 redisPanic("Unknown list encoding");
479 }
480 } else if (o->type == REDIS_SET) {
481 /* Save a set value */
482 if (o->encoding == REDIS_ENCODING_HT) {
483 dict *set = o->ptr;
484 dictIterator *di = dictGetIterator(set);
485 dictEntry *de;
486
487 if ((n = rdbSaveLen(rdb,dictSize(set))) == -1) return -1;
488 nwritten += n;
489
490 while((de = dictNext(di)) != NULL) {
491 robj *eleobj = dictGetKey(de);
492 if ((n = rdbSaveStringObject(rdb,eleobj)) == -1) return -1;
493 nwritten += n;
494 }
495 dictReleaseIterator(di);
496 } else if (o->encoding == REDIS_ENCODING_INTSET) {
497 size_t l = intsetBlobLen((intset*)o->ptr);
498
499 if ((n = rdbSaveRawString(rdb,o->ptr,l)) == -1) return -1;
500 nwritten += n;
501 } else {
502 redisPanic("Unknown set encoding");
503 }
504 } else if (o->type == REDIS_ZSET) {
505 /* Save a sorted set value */
506 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
507 size_t l = ziplistBlobLen((unsigned char*)o->ptr);
508
509 if ((n = rdbSaveRawString(rdb,o->ptr,l)) == -1) return -1;
510 nwritten += n;
511 } else if (o->encoding == REDIS_ENCODING_SKIPLIST) {
512 zset *zs = o->ptr;
513 dictIterator *di = dictGetIterator(zs->dict);
514 dictEntry *de;
515
516 if ((n = rdbSaveLen(rdb,dictSize(zs->dict))) == -1) return -1;
517 nwritten += n;
518
519 while((de = dictNext(di)) != NULL) {
520 robj *eleobj = dictGetKey(de);
521 double *score = dictGetVal(de);
522
523 if ((n = rdbSaveStringObject(rdb,eleobj)) == -1) return -1;
524 nwritten += n;
525 if ((n = rdbSaveDoubleValue(rdb,*score)) == -1) return -1;
526 nwritten += n;
527 }
528 dictReleaseIterator(di);
529 } else {
530 redisPanic("Unknown sorted set encoding");
531 }
532 } else if (o->type == REDIS_HASH) {
533 /* Save a hash value */
534 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
535 size_t l = ziplistBlobLen((unsigned char*)o->ptr);
536
537 if ((n = rdbSaveRawString(rdb,o->ptr,l)) == -1) return -1;
538 nwritten += n;
539
540 } else if (o->encoding == REDIS_ENCODING_HT) {
541 dictIterator *di = dictGetIterator(o->ptr);
542 dictEntry *de;
543
544 if ((n = rdbSaveLen(rdb,dictSize((dict*)o->ptr))) == -1) return -1;
545 nwritten += n;
546
547 while((de = dictNext(di)) != NULL) {
548 robj *key = dictGetKey(de);
549 robj *val = dictGetVal(de);
550
551 if ((n = rdbSaveStringObject(rdb,key)) == -1) return -1;
552 nwritten += n;
553 if ((n = rdbSaveStringObject(rdb,val)) == -1) return -1;
554 nwritten += n;
555 }
556 dictReleaseIterator(di);
557
558 } else {
559 redisPanic("Unknown hash encoding");
560 }
561
562 } else {
563 redisPanic("Unknown object type");
564 }
565 return nwritten;
566 }
567
568 /* Return the length the object will have on disk if saved with
569 * the rdbSaveObject() function. Currently we use a trick to get
570 * this length with very little changes to the code. In the future
571 * we could switch to a faster solution. */
572 off_t rdbSavedObjectLen(robj *o) {
573 int len = rdbSaveObject(NULL,o);
574 redisAssertWithInfo(NULL,o,len != -1);
575 return len;
576 }
577
578 /* Save a key-value pair, with expire time, type, key, value.
579 * On error -1 is returned.
580 * On success if the key was actaully saved 1 is returned, otherwise 0
581 * is returned (the key was already expired). */
582 int rdbSaveKeyValuePair(rio *rdb, robj *key, robj *val,
583 long long expiretime, long long now)
584 {
585 /* Save the expire time */
586 if (expiretime != -1) {
587 /* If this key is already expired skip it */
588 if (expiretime < now) return 0;
589 if (rdbSaveType(rdb,REDIS_RDB_OPCODE_EXPIRETIME_MS) == -1) return -1;
590 if (rdbSaveMillisecondTime(rdb,expiretime) == -1) return -1;
591 }
592
593 /* Save type, key, value */
594 if (rdbSaveObjectType(rdb,val) == -1) return -1;
595 if (rdbSaveStringObject(rdb,key) == -1) return -1;
596 if (rdbSaveObject(rdb,val) == -1) return -1;
597 return 1;
598 }
599
600 /* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
601 int rdbSave(char *filename) {
602 dictIterator *di = NULL;
603 dictEntry *de;
604 char tmpfile[256];
605 char magic[10];
606 int j;
607 long long now = mstime();
608 FILE *fp;
609 rio rdb;
610 uint64_t cksum;
611
612 snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
613 fp = fopen(tmpfile,"w");
614 if (!fp) {
615 redisLog(REDIS_WARNING, "Failed opening .rdb for saving: %s",
616 strerror(errno));
617 return REDIS_ERR;
618 }
619
620 rioInitWithFile(&rdb,fp);
621 if (server.rdb_checksum)
622 rdb.update_cksum = rioGenericUpdateChecksum;
623 snprintf(magic,sizeof(magic),"REDIS%04d",REDIS_RDB_VERSION);
624 if (rdbWriteRaw(&rdb,magic,9) == -1) goto werr;
625
626 for (j = 0; j < server.dbnum; j++) {
627 redisDb *db = server.db+j;
628 dict *d = db->dict;
629 if (dictSize(d) == 0) continue;
630 di = dictGetSafeIterator(d);
631 if (!di) {
632 fclose(fp);
633 return REDIS_ERR;
634 }
635
636 /* Write the SELECT DB opcode */
637 if (rdbSaveType(&rdb,REDIS_RDB_OPCODE_SELECTDB) == -1) goto werr;
638 if (rdbSaveLen(&rdb,j) == -1) goto werr;
639
640 /* Iterate this DB writing every entry */
641 while((de = dictNext(di)) != NULL) {
642 sds keystr = dictGetKey(de);
643 robj key, *o = dictGetVal(de);
644 long long expire;
645
646 initStaticStringObject(key,keystr);
647 expire = getExpire(db,&key);
648 if (rdbSaveKeyValuePair(&rdb,&key,o,expire,now) == -1) goto werr;
649 }
650 dictReleaseIterator(di);
651 }
652 di = NULL; /* So that we don't release it again on error. */
653
654 /* EOF opcode */
655 if (rdbSaveType(&rdb,REDIS_RDB_OPCODE_EOF) == -1) goto werr;
656
657 /* CRC64 checksum. It will be zero if checksum computation is disabled, the
658 * loading code skips the check in this case. */
659 cksum = rdb.cksum;
660 memrev64ifbe(&cksum);
661 rioWrite(&rdb,&cksum,8);
662
663 /* Make sure data will not remain on the OS's output buffers */
664 fflush(fp);
665 fsync(fileno(fp));
666 fclose(fp);
667
668 /* Use RENAME to make sure the DB file is changed atomically only
669 * if the generate DB file is ok. */
670 if (rename(tmpfile,filename) == -1) {
671 redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
672 unlink(tmpfile);
673 return REDIS_ERR;
674 }
675 redisLog(REDIS_NOTICE,"DB saved on disk");
676 server.dirty = 0;
677 server.lastsave = time(NULL);
678 server.lastbgsave_status = REDIS_OK;
679 return REDIS_OK;
680
681 werr:
682 fclose(fp);
683 unlink(tmpfile);
684 redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
685 if (di) dictReleaseIterator(di);
686 return REDIS_ERR;
687 }
688
689 int rdbSaveBackground(char *filename) {
690 pid_t childpid;
691 long long start;
692
693 if (server.rdb_child_pid != -1) return REDIS_ERR;
694
695 server.dirty_before_bgsave = server.dirty;
696
697 start = ustime();
698 if ((childpid = fork()) == 0) {
699 int retval;
700
701 /* Child */
702 if (server.ipfd > 0) close(server.ipfd);
703 if (server.sofd > 0) close(server.sofd);
704 retval = rdbSave(filename);
705 exitFromChild((retval == REDIS_OK) ? 0 : 1);
706 } else {
707 /* Parent */
708 server.stat_fork_time = ustime()-start;
709 if (childpid == -1) {
710 redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
711 strerror(errno));
712 return REDIS_ERR;
713 }
714 redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
715 server.rdb_save_time_start = time(NULL);
716 server.rdb_child_pid = childpid;
717 updateDictResizePolicy();
718 return REDIS_OK;
719 }
720 return REDIS_OK; /* unreached */
721 }
722
723 void rdbRemoveTempFile(pid_t childpid) {
724 char tmpfile[256];
725
726 snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
727 unlink(tmpfile);
728 }
729
730 /* Load a Redis object of the specified type from the specified file.
731 * On success a newly allocated object is returned, otherwise NULL. */
732 robj *rdbLoadObject(int rdbtype, rio *rdb) {
733 robj *o, *ele, *dec;
734 size_t len;
735 unsigned int i;
736
737 redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",rdbtype,rioTell(rdb));
738 if (rdbtype == REDIS_RDB_TYPE_STRING) {
739 /* Read string value */
740 if ((o = rdbLoadEncodedStringObject(rdb)) == NULL) return NULL;
741 o = tryObjectEncoding(o);
742 } else if (rdbtype == REDIS_RDB_TYPE_LIST) {
743 /* Read list value */
744 if ((len = rdbLoadLen(rdb,NULL)) == REDIS_RDB_LENERR) return NULL;
745
746 /* Use a real list when there are too many entries */
747 if (len > server.list_max_ziplist_entries) {
748 o = createListObject();
749 } else {
750 o = createZiplistObject();
751 }
752
753 /* Load every single element of the list */
754 while(len--) {
755 if ((ele = rdbLoadEncodedStringObject(rdb)) == NULL) return NULL;
756
757 /* If we are using a ziplist and the value is too big, convert
758 * the object to a real list. */
759 if (o->encoding == REDIS_ENCODING_ZIPLIST &&
760 ele->encoding == REDIS_ENCODING_RAW &&
761 sdslen(ele->ptr) > server.list_max_ziplist_value)
762 listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
763
764 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
765 dec = getDecodedObject(ele);
766 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
767 decrRefCount(dec);
768 decrRefCount(ele);
769 } else {
770 ele = tryObjectEncoding(ele);
771 listAddNodeTail(o->ptr,ele);
772 }
773 }
774 } else if (rdbtype == REDIS_RDB_TYPE_SET) {
775 /* Read list/set value */
776 if ((len = rdbLoadLen(rdb,NULL)) == REDIS_RDB_LENERR) return NULL;
777
778 /* Use a regular set when there are too many entries. */
779 if (len > server.set_max_intset_entries) {
780 o = createSetObject();
781 /* It's faster to expand the dict to the right size asap in order
782 * to avoid rehashing */
783 if (len > DICT_HT_INITIAL_SIZE)
784 dictExpand(o->ptr,len);
785 } else {
786 o = createIntsetObject();
787 }
788
789 /* Load every single element of the list/set */
790 for (i = 0; i < len; i++) {
791 long long llval;
792 if ((ele = rdbLoadEncodedStringObject(rdb)) == NULL) return NULL;
793 ele = tryObjectEncoding(ele);
794
795 if (o->encoding == REDIS_ENCODING_INTSET) {
796 /* Fetch integer value from element */
797 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
798 o->ptr = intsetAdd(o->ptr,llval,NULL);
799 } else {
800 setTypeConvert(o,REDIS_ENCODING_HT);
801 dictExpand(o->ptr,len);
802 }
803 }
804
805 /* This will also be called when the set was just converted
806 * to regular hashtable encoded set */
807 if (o->encoding == REDIS_ENCODING_HT) {
808 dictAdd((dict*)o->ptr,ele,NULL);
809 } else {
810 decrRefCount(ele);
811 }
812 }
813 } else if (rdbtype == REDIS_RDB_TYPE_ZSET) {
814 /* Read list/set value */
815 size_t zsetlen;
816 size_t maxelelen = 0;
817 zset *zs;
818
819 if ((zsetlen = rdbLoadLen(rdb,NULL)) == REDIS_RDB_LENERR) return NULL;
820 o = createZsetObject();
821 zs = o->ptr;
822
823 /* Load every single element of the list/set */
824 while(zsetlen--) {
825 robj *ele;
826 double score;
827 zskiplistNode *znode;
828
829 if ((ele = rdbLoadEncodedStringObject(rdb)) == NULL) return NULL;
830 ele = tryObjectEncoding(ele);
831 if (rdbLoadDoubleValue(rdb,&score) == -1) return NULL;
832
833 /* Don't care about integer-encoded strings. */
834 if (ele->encoding == REDIS_ENCODING_RAW &&
835 sdslen(ele->ptr) > maxelelen)
836 maxelelen = sdslen(ele->ptr);
837
838 znode = zslInsert(zs->zsl,score,ele);
839 dictAdd(zs->dict,ele,&znode->score);
840 incrRefCount(ele); /* added to skiplist */
841 }
842
843 /* Convert *after* loading, since sorted sets are not stored ordered. */
844 if (zsetLength(o) <= server.zset_max_ziplist_entries &&
845 maxelelen <= server.zset_max_ziplist_value)
846 zsetConvert(o,REDIS_ENCODING_ZIPLIST);
847 } else if (rdbtype == REDIS_RDB_TYPE_HASH) {
848 size_t len;
849 int ret;
850
851 len = rdbLoadLen(rdb, NULL);
852 if (len == REDIS_RDB_LENERR) return NULL;
853
854 o = createHashObject();
855
856 /* Too many entries? Use an hash table. */
857 if (len > server.hash_max_ziplist_entries)
858 hashTypeConvert(o, REDIS_ENCODING_HT);
859
860 /* Load every field and value into the ziplist */
861 while (o->encoding == REDIS_ENCODING_ZIPLIST && len > 0) {
862 robj *field, *value;
863
864 len--;
865 /* Load raw strings */
866 field = rdbLoadStringObject(rdb);
867 if (field == NULL) return NULL;
868 redisAssert(field->encoding == REDIS_ENCODING_RAW);
869 value = rdbLoadStringObject(rdb);
870 if (value == NULL) return NULL;
871 redisAssert(field->encoding == REDIS_ENCODING_RAW);
872
873 /* Add pair to ziplist */
874 o->ptr = ziplistPush(o->ptr, field->ptr, sdslen(field->ptr), ZIPLIST_TAIL);
875 o->ptr = ziplistPush(o->ptr, value->ptr, sdslen(value->ptr), ZIPLIST_TAIL);
876 /* Convert to hash table if size threshold is exceeded */
877 if (sdslen(field->ptr) > server.hash_max_ziplist_value ||
878 sdslen(value->ptr) > server.hash_max_ziplist_value)
879 {
880 decrRefCount(field);
881 decrRefCount(value);
882 hashTypeConvert(o, REDIS_ENCODING_HT);
883 break;
884 }
885 decrRefCount(field);
886 decrRefCount(value);
887 }
888
889 /* Load remaining fields and values into the hash table */
890 while (o->encoding == REDIS_ENCODING_HT && len > 0) {
891 robj *field, *value;
892
893 len--;
894 /* Load encoded strings */
895 field = rdbLoadEncodedStringObject(rdb);
896 if (field == NULL) return NULL;
897 value = rdbLoadEncodedStringObject(rdb);
898 if (value == NULL) return NULL;
899
900 field = tryObjectEncoding(field);
901 value = tryObjectEncoding(value);
902
903 /* Add pair to hash table */
904 ret = dictAdd((dict*)o->ptr, field, value);
905 redisAssert(ret == REDIS_OK);
906 }
907
908 /* All pairs should be read by now */
909 redisAssert(len == 0);
910
911 } else if (rdbtype == REDIS_RDB_TYPE_HASH_ZIPMAP ||
912 rdbtype == REDIS_RDB_TYPE_LIST_ZIPLIST ||
913 rdbtype == REDIS_RDB_TYPE_SET_INTSET ||
914 rdbtype == REDIS_RDB_TYPE_ZSET_ZIPLIST ||
915 rdbtype == REDIS_RDB_TYPE_HASH_ZIPLIST)
916 {
917 robj *aux = rdbLoadStringObject(rdb);
918
919 if (aux == NULL) return NULL;
920 o = createObject(REDIS_STRING,NULL); /* string is just placeholder */
921 o->ptr = zmalloc(sdslen(aux->ptr));
922 memcpy(o->ptr,aux->ptr,sdslen(aux->ptr));
923 decrRefCount(aux);
924
925 /* Fix the object encoding, and make sure to convert the encoded
926 * data type into the base type if accordingly to the current
927 * configuration there are too many elements in the encoded data
928 * type. Note that we only check the length and not max element
929 * size as this is an O(N) scan. Eventually everything will get
930 * converted. */
931 switch(rdbtype) {
932 case REDIS_RDB_TYPE_HASH_ZIPMAP:
933 /* Convert to ziplist encoded hash. This must be deprecated
934 * when loading dumps created by Redis 2.4 gets deprecated. */
935 {
936 unsigned char *zl = ziplistNew();
937 unsigned char *zi = zipmapRewind(o->ptr);
938 unsigned char *fstr, *vstr;
939 unsigned int flen, vlen;
940 unsigned int maxlen = 0;
941
942 while ((zi = zipmapNext(zi, &fstr, &flen, &vstr, &vlen)) != NULL) {
943 if (flen > maxlen) maxlen = flen;
944 if (vlen > maxlen) maxlen = vlen;
945 zl = ziplistPush(zl, fstr, flen, ZIPLIST_TAIL);
946 zl = ziplistPush(zl, vstr, vlen, ZIPLIST_TAIL);
947 }
948
949 zfree(o->ptr);
950 o->ptr = zl;
951 o->type = REDIS_HASH;
952 o->encoding = REDIS_ENCODING_ZIPLIST;
953
954 if (hashTypeLength(o) > server.hash_max_ziplist_entries ||
955 maxlen > server.hash_max_ziplist_value)
956 {
957 hashTypeConvert(o, REDIS_ENCODING_HT);
958 }
959 }
960 break;
961 case REDIS_RDB_TYPE_LIST_ZIPLIST:
962 o->type = REDIS_LIST;
963 o->encoding = REDIS_ENCODING_ZIPLIST;
964 if (ziplistLen(o->ptr) > server.list_max_ziplist_entries)
965 listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
966 break;
967 case REDIS_RDB_TYPE_SET_INTSET:
968 o->type = REDIS_SET;
969 o->encoding = REDIS_ENCODING_INTSET;
970 if (intsetLen(o->ptr) > server.set_max_intset_entries)
971 setTypeConvert(o,REDIS_ENCODING_HT);
972 break;
973 case REDIS_RDB_TYPE_ZSET_ZIPLIST:
974 o->type = REDIS_ZSET;
975 o->encoding = REDIS_ENCODING_ZIPLIST;
976 if (zsetLength(o) > server.zset_max_ziplist_entries)
977 zsetConvert(o,REDIS_ENCODING_SKIPLIST);
978 break;
979 case REDIS_RDB_TYPE_HASH_ZIPLIST:
980 o->type = REDIS_HASH;
981 o->encoding = REDIS_ENCODING_ZIPLIST;
982 if (hashTypeLength(o) > server.hash_max_ziplist_entries)
983 hashTypeConvert(o, REDIS_ENCODING_HT);
984 break;
985 default:
986 redisPanic("Unknown encoding");
987 break;
988 }
989 } else {
990 redisPanic("Unknown object type");
991 }
992 return o;
993 }
994
995 /* Mark that we are loading in the global state and setup the fields
996 * needed to provide loading stats. */
997 void startLoading(FILE *fp) {
998 struct stat sb;
999
1000 /* Load the DB */
1001 server.loading = 1;
1002 server.loading_start_time = time(NULL);
1003 if (fstat(fileno(fp), &sb) == -1) {
1004 server.loading_total_bytes = 1; /* just to avoid division by zero */
1005 } else {
1006 server.loading_total_bytes = sb.st_size;
1007 }
1008 }
1009
1010 /* Refresh the loading progress info */
1011 void loadingProgress(off_t pos) {
1012 server.loading_loaded_bytes = pos;
1013 }
1014
1015 /* Loading finished */
1016 void stopLoading(void) {
1017 server.loading = 0;
1018 }
1019
1020 int rdbLoad(char *filename) {
1021 uint32_t dbid;
1022 int type, rdbver;
1023 redisDb *db = server.db+0;
1024 char buf[1024];
1025 long long expiretime, now = mstime();
1026 long loops = 0;
1027 FILE *fp;
1028 rio rdb;
1029
1030 fp = fopen(filename,"r");
1031 if (!fp) {
1032 errno = ENOENT;
1033 return REDIS_ERR;
1034 }
1035 rioInitWithFile(&rdb,fp);
1036 if (server.rdb_checksum)
1037 rdb.update_cksum = rioGenericUpdateChecksum;
1038 if (rioRead(&rdb,buf,9) == 0) goto eoferr;
1039 buf[9] = '\0';
1040 if (memcmp(buf,"REDIS",5) != 0) {
1041 fclose(fp);
1042 redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
1043 errno = EINVAL;
1044 return REDIS_ERR;
1045 }
1046 rdbver = atoi(buf+5);
1047 if (rdbver < 1 || rdbver > REDIS_RDB_VERSION) {
1048 fclose(fp);
1049 redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
1050 errno = EINVAL;
1051 return REDIS_ERR;
1052 }
1053
1054 startLoading(fp);
1055 while(1) {
1056 robj *key, *val;
1057 expiretime = -1;
1058
1059 /* Serve the clients from time to time */
1060 if (!(loops++ % 1000)) {
1061 loadingProgress(rioTell(&rdb));
1062 aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
1063 }
1064
1065 /* Read type. */
1066 if ((type = rdbLoadType(&rdb)) == -1) goto eoferr;
1067 if (type == REDIS_RDB_OPCODE_EXPIRETIME) {
1068 if ((expiretime = rdbLoadTime(&rdb)) == -1) goto eoferr;
1069 /* We read the time so we need to read the object type again. */
1070 if ((type = rdbLoadType(&rdb)) == -1) goto eoferr;
1071 /* the EXPIRETIME opcode specifies time in seconds, so convert
1072 * into milliesconds. */
1073 expiretime *= 1000;
1074 } else if (type == REDIS_RDB_OPCODE_EXPIRETIME_MS) {
1075 /* Milliseconds precision expire times introduced with RDB
1076 * version 3. */
1077 if ((expiretime = rdbLoadMillisecondTime(&rdb)) == -1) goto eoferr;
1078 /* We read the time so we need to read the object type again. */
1079 if ((type = rdbLoadType(&rdb)) == -1) goto eoferr;
1080 }
1081
1082 if (type == REDIS_RDB_OPCODE_EOF)
1083 break;
1084
1085 /* Handle SELECT DB opcode as a special case */
1086 if (type == REDIS_RDB_OPCODE_SELECTDB) {
1087 if ((dbid = rdbLoadLen(&rdb,NULL)) == REDIS_RDB_LENERR)
1088 goto eoferr;
1089 if (dbid >= (unsigned)server.dbnum) {
1090 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
1091 exit(1);
1092 }
1093 db = server.db+dbid;
1094 continue;
1095 }
1096 /* Read key */
1097 if ((key = rdbLoadStringObject(&rdb)) == NULL) goto eoferr;
1098 /* Read value */
1099 if ((val = rdbLoadObject(type,&rdb)) == NULL) goto eoferr;
1100 /* Check if the key already expired. This function is used when loading
1101 * an RDB file from disk, either at startup, or when an RDB was
1102 * received from the master. In the latter case, the master is
1103 * responsible for key expiry. If we would expire keys here, the
1104 * snapshot taken by the master may not be reflected on the slave. */
1105 if (server.masterhost == NULL && expiretime != -1 && expiretime < now) {
1106 decrRefCount(key);
1107 decrRefCount(val);
1108 continue;
1109 }
1110 /* Add the new object in the hash table */
1111 dbAdd(db,key,val);
1112
1113 /* Set the expire time if needed */
1114 if (expiretime != -1) setExpire(db,key,expiretime);
1115
1116 decrRefCount(key);
1117 }
1118 /* Verify the checksum if RDB version is >= 5 */
1119 if (rdbver >= 5 && server.rdb_checksum) {
1120 uint64_t cksum, expected = rdb.cksum;
1121
1122 if (rioRead(&rdb,&cksum,8) == 0) goto eoferr;
1123 memrev64ifbe(&cksum);
1124 if (cksum == 0) {
1125 redisLog(REDIS_WARNING,"RDB file was saved with checksum disabled: no check performed.");
1126 } else if (cksum != expected) {
1127 redisLog(REDIS_WARNING,"Wrong RDB checksum. Aborting now.");
1128 exit(1);
1129 }
1130 }
1131
1132 fclose(fp);
1133 stopLoading();
1134 return REDIS_OK;
1135
1136 eoferr: /* unexpected end of file is handled here with a fatal exit */
1137 redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
1138 exit(1);
1139 return REDIS_ERR; /* Just to avoid warning */
1140 }
1141
1142 /* A background saving child (BGSAVE) terminated its work. Handle this. */
1143 void backgroundSaveDoneHandler(int exitcode, int bysignal) {
1144 if (!bysignal && exitcode == 0) {
1145 redisLog(REDIS_NOTICE,
1146 "Background saving terminated with success");
1147 server.dirty = server.dirty - server.dirty_before_bgsave;
1148 server.lastsave = time(NULL);
1149 server.lastbgsave_status = REDIS_OK;
1150 } else if (!bysignal && exitcode != 0) {
1151 redisLog(REDIS_WARNING, "Background saving error");
1152 server.lastbgsave_status = REDIS_ERR;
1153 } else {
1154 redisLog(REDIS_WARNING,
1155 "Background saving terminated by signal %d", bysignal);
1156 rdbRemoveTempFile(server.rdb_child_pid);
1157 server.lastbgsave_status = REDIS_ERR;
1158 }
1159 server.rdb_child_pid = -1;
1160 server.rdb_save_time_last = time(NULL)-server.rdb_save_time_start;
1161 server.rdb_save_time_start = -1;
1162 /* Possibly there are slaves waiting for a BGSAVE in order to be served
1163 * (the first stage of SYNC is a bulk transfer of dump.rdb) */
1164 updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
1165 }
1166
1167 void saveCommand(redisClient *c) {
1168 if (server.rdb_child_pid != -1) {
1169 addReplyError(c,"Background save already in progress");
1170 return;
1171 }
1172 if (rdbSave(server.rdb_filename) == REDIS_OK) {
1173 addReply(c,shared.ok);
1174 } else {
1175 addReply(c,shared.err);
1176 }
1177 }
1178
1179 void bgsaveCommand(redisClient *c) {
1180 if (server.rdb_child_pid != -1) {
1181 addReplyError(c,"Background save already in progress");
1182 } else if (server.aof_child_pid != -1) {
1183 addReplyError(c,"Can't BGSAVE while AOF log rewriting is in progress");
1184 } else if (rdbSaveBackground(server.rdb_filename) == REDIS_OK) {
1185 addReplyStatus(c,"Background saving started");
1186 } else {
1187 addReply(c,shared.err);
1188 }
1189 }