]> git.saurik.com Git - redis.git/blame - src/rdb.c
version set to 2.3.0
[redis.git] / src / rdb.c
CommitLineData
e2641e09 1#include "redis.h"
2#include "lzf.h" /* LZF compression library */
3
4#include <math.h>
3688d7f3 5#include <sys/types.h>
6#include <sys/time.h>
7#include <sys/resource.h>
8#include <sys/wait.h>
9#include <arpa/inet.h>
97e7f8ae 10#include <sys/stat.h>
e2641e09 11
9a68cf91
PN
12/* Convenience wrapper around fwrite, that returns the number of bytes written
13 * to the file instead of the number of objects (see fwrite(3)) and -1 in the
14 * case of an error. It also supports a NULL *fp to skip writing altogether
15 * instead of writing to /dev/null. */
16static int rdbWriteRaw(FILE *fp, void *p, size_t len) {
17 if (fp != NULL && fwrite(p,len,1,fp) == 0) return -1;
18 return len;
19}
20
e2641e09 21int rdbSaveType(FILE *fp, unsigned char type) {
9a68cf91 22 return rdbWriteRaw(fp,&type,1);
e2641e09 23}
24
25int rdbSaveTime(FILE *fp, time_t t) {
26 int32_t t32 = (int32_t) t;
9a68cf91 27 return rdbWriteRaw(fp,&t32,4);
e2641e09 28}
29
30/* check rdbLoadLen() comments for more info */
31int rdbSaveLen(FILE *fp, uint32_t len) {
32 unsigned char buf[2];
8a623a98 33 int nwritten;
e2641e09 34
35 if (len < (1<<6)) {
36 /* Save a 6 bit len */
37 buf[0] = (len&0xFF)|(REDIS_RDB_6BITLEN<<6);
9a68cf91 38 if (rdbWriteRaw(fp,buf,1) == -1) return -1;
8a623a98 39 nwritten = 1;
e2641e09 40 } else if (len < (1<<14)) {
41 /* Save a 14 bit len */
42 buf[0] = ((len>>8)&0xFF)|(REDIS_RDB_14BITLEN<<6);
43 buf[1] = len&0xFF;
9a68cf91 44 if (rdbWriteRaw(fp,buf,2) == -1) return -1;
8a623a98 45 nwritten = 2;
e2641e09 46 } else {
47 /* Save a 32 bit len */
48 buf[0] = (REDIS_RDB_32BITLEN<<6);
9a68cf91 49 if (rdbWriteRaw(fp,buf,1) == -1) return -1;
e2641e09 50 len = htonl(len);
9a68cf91 51 if (rdbWriteRaw(fp,&len,4) == -1) return -1;
8a623a98 52 nwritten = 1+4;
e2641e09 53 }
8a623a98 54 return nwritten;
e2641e09 55}
56
57/* Encode 'value' as an integer if possible (if integer will fit the
58 * supported range). If the function sucessful encoded the integer
59 * then the (up to 5 bytes) encoded representation is written in the
60 * string pointed by 'enc' and the length is returned. Otherwise
61 * 0 is returned. */
62int rdbEncodeInteger(long long value, unsigned char *enc) {
63 /* Finally check if it fits in our ranges */
64 if (value >= -(1<<7) && value <= (1<<7)-1) {
65 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT8;
66 enc[1] = value&0xFF;
67 return 2;
68 } else if (value >= -(1<<15) && value <= (1<<15)-1) {
69 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT16;
70 enc[1] = value&0xFF;
71 enc[2] = (value>>8)&0xFF;
72 return 3;
73 } else if (value >= -((long long)1<<31) && value <= ((long long)1<<31)-1) {
74 enc[0] = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_INT32;
75 enc[1] = value&0xFF;
76 enc[2] = (value>>8)&0xFF;
77 enc[3] = (value>>16)&0xFF;
78 enc[4] = (value>>24)&0xFF;
79 return 5;
80 } else {
81 return 0;
82 }
83}
84
85/* String objects in the form "2391" "-100" without any space and with a
86 * range of values that can fit in an 8, 16 or 32 bit signed value can be
87 * encoded as integers to save space */
88int rdbTryIntegerEncoding(char *s, size_t len, unsigned char *enc) {
89 long long value;
90 char *endptr, buf[32];
91
92 /* Check if it's possible to encode this value as a number */
93 value = strtoll(s, &endptr, 10);
94 if (endptr[0] != '\0') return 0;
95 ll2string(buf,32,value);
96
97 /* If the number converted back into a string is not identical
98 * then it's not possible to encode the string as integer */
99 if (strlen(buf) != len || memcmp(buf,s,len)) return 0;
100
101 return rdbEncodeInteger(value,enc);
102}
103
104int rdbSaveLzfStringObject(FILE *fp, unsigned char *s, size_t len) {
105 size_t comprlen, outlen;
106 unsigned char byte;
8a623a98 107 int n, nwritten = 0;
e2641e09 108 void *out;
109
110 /* We require at least four bytes compression for this to be worth it */
111 if (len <= 4) return 0;
112 outlen = len-4;
113 if ((out = zmalloc(outlen+1)) == NULL) return 0;
114 comprlen = lzf_compress(s, len, out, outlen);
115 if (comprlen == 0) {
116 zfree(out);
117 return 0;
118 }
119 /* Data compressed! Let's save it on disk */
120 byte = (REDIS_RDB_ENCVAL<<6)|REDIS_RDB_ENC_LZF;
9a68cf91
PN
121 if ((n = rdbWriteRaw(fp,&byte,1)) == -1) goto writeerr;
122 nwritten += n;
8a623a98
PN
123
124 if ((n = rdbSaveLen(fp,comprlen)) == -1) goto writeerr;
125 nwritten += n;
126
127 if ((n = rdbSaveLen(fp,len)) == -1) goto writeerr;
128 nwritten += n;
129
9a68cf91
PN
130 if ((n = rdbWriteRaw(fp,out,comprlen)) == -1) goto writeerr;
131 nwritten += n;
8a623a98 132
e2641e09 133 zfree(out);
8a623a98 134 return nwritten;
e2641e09 135
136writeerr:
137 zfree(out);
138 return -1;
139}
140
141/* Save a string objet as [len][data] on disk. If the object is a string
142 * representation of an integer value we try to safe it in a special form */
143int rdbSaveRawString(FILE *fp, unsigned char *s, size_t len) {
144 int enclen;
8a623a98 145 int n, nwritten = 0;
e2641e09 146
147 /* Try integer encoding */
148 if (len <= 11) {
149 unsigned char buf[5];
150 if ((enclen = rdbTryIntegerEncoding((char*)s,len,buf)) > 0) {
9a68cf91 151 if (rdbWriteRaw(fp,buf,enclen) == -1) return -1;
8a623a98 152 return enclen;
e2641e09 153 }
154 }
155
156 /* Try LZF compression - under 20 bytes it's unable to compress even
157 * aaaaaaaaaaaaaaaaaa so skip it */
158 if (server.rdbcompression && len > 20) {
8a623a98
PN
159 n = rdbSaveLzfStringObject(fp,s,len);
160 if (n == -1) return -1;
161 if (n > 0) return n;
162 /* Return value of 0 means data can't be compressed, save the old way */
e2641e09 163 }
164
165 /* Store verbatim */
8a623a98
PN
166 if ((n = rdbSaveLen(fp,len)) == -1) return -1;
167 nwritten += n;
168 if (len > 0) {
9a68cf91 169 if (rdbWriteRaw(fp,s,len) == -1) return -1;
8a623a98
PN
170 nwritten += len;
171 }
172 return nwritten;
e2641e09 173}
174
175/* Save a long long value as either an encoded string or a string. */
176int rdbSaveLongLongAsStringObject(FILE *fp, long long value) {
177 unsigned char buf[32];
8a623a98 178 int n, nwritten = 0;
e2641e09 179 int enclen = rdbEncodeInteger(value,buf);
180 if (enclen > 0) {
9a68cf91 181 return rdbWriteRaw(fp,buf,enclen);
e2641e09 182 } else {
183 /* Encode as string */
184 enclen = ll2string((char*)buf,32,value);
185 redisAssert(enclen < 32);
8a623a98
PN
186 if ((n = rdbSaveLen(fp,enclen)) == -1) return -1;
187 nwritten += n;
9a68cf91
PN
188 if ((n = rdbWriteRaw(fp,buf,enclen)) == -1) return -1;
189 nwritten += n;
e2641e09 190 }
8a623a98 191 return nwritten;
e2641e09 192}
193
194/* Like rdbSaveStringObjectRaw() but handle encoded objects */
195int rdbSaveStringObject(FILE *fp, robj *obj) {
196 /* Avoid to decode the object, then encode it again, if the
197 * object is alrady integer encoded. */
198 if (obj->encoding == REDIS_ENCODING_INT) {
199 return rdbSaveLongLongAsStringObject(fp,(long)obj->ptr);
200 } else {
201 redisAssert(obj->encoding == REDIS_ENCODING_RAW);
202 return rdbSaveRawString(fp,obj->ptr,sdslen(obj->ptr));
203 }
204}
205
206/* Save a double value. Doubles are saved as strings prefixed by an unsigned
207 * 8 bit integer specifing the length of the representation.
208 * This 8 bit integer has special values in order to specify the following
209 * conditions:
210 * 253: not a number
211 * 254: + inf
212 * 255: - inf
213 */
214int rdbSaveDoubleValue(FILE *fp, double val) {
215 unsigned char buf[128];
216 int len;
217
218 if (isnan(val)) {
219 buf[0] = 253;
220 len = 1;
221 } else if (!isfinite(val)) {
222 len = 1;
223 buf[0] = (val < 0) ? 255 : 254;
224 } else {
225#if (DBL_MANT_DIG >= 52) && (LLONG_MAX == 0x7fffffffffffffffLL)
226 /* Check if the float is in a safe range to be casted into a
227 * long long. We are assuming that long long is 64 bit here.
228 * Also we are assuming that there are no implementations around where
229 * double has precision < 52 bit.
230 *
231 * Under this assumptions we test if a double is inside an interval
232 * where casting to long long is safe. Then using two castings we
233 * make sure the decimal part is zero. If all this is true we use
234 * integer printing function that is much faster. */
235 double min = -4503599627370495; /* (2^52)-1 */
236 double max = 4503599627370496; /* -(2^52) */
237 if (val > min && val < max && val == ((double)((long long)val)))
238 ll2string((char*)buf+1,sizeof(buf),(long long)val);
239 else
240#endif
241 snprintf((char*)buf+1,sizeof(buf)-1,"%.17g",val);
242 buf[0] = strlen((char*)buf+1);
243 len = buf[0]+1;
244 }
9a68cf91 245 return rdbWriteRaw(fp,buf,len);
e2641e09 246}
247
248/* Save a Redis object. */
249int rdbSaveObject(FILE *fp, robj *o) {
8a623a98
PN
250 int n, nwritten = 0;
251
e2641e09 252 if (o->type == REDIS_STRING) {
253 /* Save a string value */
8a623a98
PN
254 if ((n = rdbSaveStringObject(fp,o)) == -1) return -1;
255 nwritten += n;
e2641e09 256 } else if (o->type == REDIS_LIST) {
257 /* Save a list value */
258 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
259 unsigned char *p;
260 unsigned char *vstr;
261 unsigned int vlen;
262 long long vlong;
263
8a623a98
PN
264 if ((n = rdbSaveLen(fp,ziplistLen(o->ptr))) == -1) return -1;
265 nwritten += n;
266
e2641e09 267 p = ziplistIndex(o->ptr,0);
268 while(ziplistGet(p,&vstr,&vlen,&vlong)) {
269 if (vstr) {
8a623a98 270 if ((n = rdbSaveRawString(fp,vstr,vlen)) == -1)
e2641e09 271 return -1;
8a623a98 272 nwritten += n;
e2641e09 273 } else {
8a623a98 274 if ((n = rdbSaveLongLongAsStringObject(fp,vlong)) == -1)
e2641e09 275 return -1;
8a623a98 276 nwritten += n;
e2641e09 277 }
278 p = ziplistNext(o->ptr,p);
279 }
280 } else if (o->encoding == REDIS_ENCODING_LINKEDLIST) {
281 list *list = o->ptr;
282 listIter li;
283 listNode *ln;
284
8a623a98
PN
285 if ((n = rdbSaveLen(fp,listLength(list))) == -1) return -1;
286 nwritten += n;
287
e2641e09 288 listRewind(list,&li);
289 while((ln = listNext(&li))) {
290 robj *eleobj = listNodeValue(ln);
8a623a98
PN
291 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
292 nwritten += n;
e2641e09 293 }
294 } else {
295 redisPanic("Unknown list encoding");
296 }
297 } else if (o->type == REDIS_SET) {
298 /* Save a set value */
96ffb2fe
PN
299 if (o->encoding == REDIS_ENCODING_HT) {
300 dict *set = o->ptr;
301 dictIterator *di = dictGetIterator(set);
302 dictEntry *de;
e2641e09 303
8a623a98
PN
304 if ((n = rdbSaveLen(fp,dictSize(set))) == -1) return -1;
305 nwritten += n;
306
96ffb2fe
PN
307 while((de = dictNext(di)) != NULL) {
308 robj *eleobj = dictGetEntryKey(de);
8a623a98
PN
309 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
310 nwritten += n;
96ffb2fe
PN
311 }
312 dictReleaseIterator(di);
313 } else if (o->encoding == REDIS_ENCODING_INTSET) {
314 intset *is = o->ptr;
23c64fe5 315 int64_t llval;
96ffb2fe
PN
316 int i = 0;
317
8a623a98
PN
318 if ((n = rdbSaveLen(fp,intsetLen(is))) == -1) return -1;
319 nwritten += n;
320
96ffb2fe 321 while(intsetGet(is,i++,&llval)) {
8a623a98
PN
322 if ((n = rdbSaveLongLongAsStringObject(fp,llval)) == -1) return -1;
323 nwritten += n;
96ffb2fe
PN
324 }
325 } else {
326 redisPanic("Unknown set encoding");
e2641e09 327 }
e2641e09 328 } else if (o->type == REDIS_ZSET) {
329 /* Save a set value */
330 zset *zs = o->ptr;
331 dictIterator *di = dictGetIterator(zs->dict);
332 dictEntry *de;
333
8a623a98
PN
334 if ((n = rdbSaveLen(fp,dictSize(zs->dict))) == -1) return -1;
335 nwritten += n;
336
e2641e09 337 while((de = dictNext(di)) != NULL) {
338 robj *eleobj = dictGetEntryKey(de);
339 double *score = dictGetEntryVal(de);
340
8a623a98
PN
341 if ((n = rdbSaveStringObject(fp,eleobj)) == -1) return -1;
342 nwritten += n;
343 if ((n = rdbSaveDoubleValue(fp,*score)) == -1) return -1;
344 nwritten += n;
e2641e09 345 }
346 dictReleaseIterator(di);
347 } else if (o->type == REDIS_HASH) {
348 /* Save a hash value */
349 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
350 unsigned char *p = zipmapRewind(o->ptr);
351 unsigned int count = zipmapLen(o->ptr);
352 unsigned char *key, *val;
353 unsigned int klen, vlen;
354
8a623a98
PN
355 if ((n = rdbSaveLen(fp,count)) == -1) return -1;
356 nwritten += n;
357
e2641e09 358 while((p = zipmapNext(p,&key,&klen,&val,&vlen)) != NULL) {
8a623a98
PN
359 if ((n = rdbSaveRawString(fp,key,klen)) == -1) return -1;
360 nwritten += n;
361 if ((n = rdbSaveRawString(fp,val,vlen)) == -1) return -1;
362 nwritten += n;
e2641e09 363 }
364 } else {
365 dictIterator *di = dictGetIterator(o->ptr);
366 dictEntry *de;
367
8a623a98
PN
368 if ((n = rdbSaveLen(fp,dictSize((dict*)o->ptr))) == -1) return -1;
369 nwritten += n;
370
e2641e09 371 while((de = dictNext(di)) != NULL) {
372 robj *key = dictGetEntryKey(de);
373 robj *val = dictGetEntryVal(de);
374
8a623a98
PN
375 if ((n = rdbSaveStringObject(fp,key)) == -1) return -1;
376 nwritten += n;
377 if ((n = rdbSaveStringObject(fp,val)) == -1) return -1;
378 nwritten += n;
e2641e09 379 }
380 dictReleaseIterator(di);
381 }
382 } else {
383 redisPanic("Unknown object type");
384 }
8a623a98 385 return nwritten;
e2641e09 386}
387
388/* Return the length the object will have on disk if saved with
389 * the rdbSaveObject() function. Currently we use a trick to get
390 * this length with very little changes to the code. In the future
391 * we could switch to a faster solution. */
bd70a5f5
PN
392off_t rdbSavedObjectLen(robj *o) {
393 int len = rdbSaveObject(NULL,o);
394 redisAssert(len != -1);
395 return len;
e2641e09 396}
397
e2641e09 398/* Save the DB on disk. Return REDIS_ERR on error, REDIS_OK on success */
399int rdbSave(char *filename) {
400 dictIterator *di = NULL;
401 dictEntry *de;
402 FILE *fp;
403 char tmpfile[256];
404 int j;
405 time_t now = time(NULL);
406
16d77878 407 /* FIXME: implement .rdb save for disk store properly */
408 redisAssert(server.ds_enabled == 0);
e2641e09 409
410 snprintf(tmpfile,256,"temp-%d.rdb", (int) getpid());
411 fp = fopen(tmpfile,"w");
412 if (!fp) {
413 redisLog(REDIS_WARNING, "Failed saving the DB: %s", strerror(errno));
414 return REDIS_ERR;
415 }
416 if (fwrite("REDIS0001",9,1,fp) == 0) goto werr;
417 for (j = 0; j < server.dbnum; j++) {
418 redisDb *db = server.db+j;
419 dict *d = db->dict;
420 if (dictSize(d) == 0) continue;
421 di = dictGetIterator(d);
422 if (!di) {
423 fclose(fp);
424 return REDIS_ERR;
425 }
426
427 /* Write the SELECT DB opcode */
428 if (rdbSaveType(fp,REDIS_SELECTDB) == -1) goto werr;
429 if (rdbSaveLen(fp,j) == -1) goto werr;
430
431 /* Iterate this DB writing every entry */
432 while((de = dictNext(di)) != NULL) {
433 sds keystr = dictGetEntryKey(de);
434 robj key, *o = dictGetEntryVal(de);
435 time_t expiretime;
436
437 initStaticStringObject(key,keystr);
438 expiretime = getExpire(db,&key);
439
440 /* Save the expire time */
441 if (expiretime != -1) {
442 /* If this key is already expired skip it */
443 if (expiretime < now) continue;
444 if (rdbSaveType(fp,REDIS_EXPIRETIME) == -1) goto werr;
445 if (rdbSaveTime(fp,expiretime) == -1) goto werr;
446 }
16d77878 447 /* Save type, key, value */
448 if (rdbSaveType(fp,o->type) == -1) goto werr;
449 if (rdbSaveStringObject(fp,&key) == -1) goto werr;
450 if (rdbSaveObject(fp,o) == -1) goto werr;
e2641e09 451 }
452 dictReleaseIterator(di);
453 }
454 /* EOF opcode */
455 if (rdbSaveType(fp,REDIS_EOF) == -1) goto werr;
456
457 /* Make sure data will not remain on the OS's output buffers */
458 fflush(fp);
459 fsync(fileno(fp));
460 fclose(fp);
461
462 /* Use RENAME to make sure the DB file is changed atomically only
463 * if the generate DB file is ok. */
464 if (rename(tmpfile,filename) == -1) {
465 redisLog(REDIS_WARNING,"Error moving temp DB file on the final destination: %s", strerror(errno));
466 unlink(tmpfile);
467 return REDIS_ERR;
468 }
469 redisLog(REDIS_NOTICE,"DB saved on disk");
470 server.dirty = 0;
471 server.lastsave = time(NULL);
472 return REDIS_OK;
473
474werr:
475 fclose(fp);
476 unlink(tmpfile);
477 redisLog(REDIS_WARNING,"Write error saving DB on disk: %s", strerror(errno));
478 if (di) dictReleaseIterator(di);
479 return REDIS_ERR;
480}
481
482int rdbSaveBackground(char *filename) {
483 pid_t childpid;
484
485 if (server.bgsavechildpid != -1) return REDIS_ERR;
16d77878 486 redisAssert(server.ds_enabled == 0);
2f6b31c3 487 server.dirty_before_bgsave = server.dirty;
e2641e09 488 if ((childpid = fork()) == 0) {
489 /* Child */
a5639e7d
PN
490 if (server.ipfd > 0) close(server.ipfd);
491 if (server.sofd > 0) close(server.sofd);
e2641e09 492 if (rdbSave(filename) == REDIS_OK) {
493 _exit(0);
494 } else {
495 _exit(1);
496 }
497 } else {
498 /* Parent */
499 if (childpid == -1) {
500 redisLog(REDIS_WARNING,"Can't save in background: fork: %s",
501 strerror(errno));
502 return REDIS_ERR;
503 }
504 redisLog(REDIS_NOTICE,"Background saving started by pid %d",childpid);
505 server.bgsavechildpid = childpid;
506 updateDictResizePolicy();
507 return REDIS_OK;
508 }
509 return REDIS_OK; /* unreached */
510}
511
512void rdbRemoveTempFile(pid_t childpid) {
513 char tmpfile[256];
514
515 snprintf(tmpfile,256,"temp-%d.rdb", (int) childpid);
516 unlink(tmpfile);
517}
518
519int rdbLoadType(FILE *fp) {
520 unsigned char type;
521 if (fread(&type,1,1,fp) == 0) return -1;
522 return type;
523}
524
525time_t rdbLoadTime(FILE *fp) {
526 int32_t t32;
527 if (fread(&t32,4,1,fp) == 0) return -1;
528 return (time_t) t32;
529}
530
531/* Load an encoded length from the DB, see the REDIS_RDB_* defines on the top
532 * of this file for a description of how this are stored on disk.
533 *
534 * isencoded is set to 1 if the readed length is not actually a length but
535 * an "encoding type", check the above comments for more info */
536uint32_t rdbLoadLen(FILE *fp, int *isencoded) {
537 unsigned char buf[2];
538 uint32_t len;
539 int type;
540
541 if (isencoded) *isencoded = 0;
542 if (fread(buf,1,1,fp) == 0) return REDIS_RDB_LENERR;
543 type = (buf[0]&0xC0)>>6;
544 if (type == REDIS_RDB_6BITLEN) {
545 /* Read a 6 bit len */
546 return buf[0]&0x3F;
547 } else if (type == REDIS_RDB_ENCVAL) {
548 /* Read a 6 bit len encoding type */
549 if (isencoded) *isencoded = 1;
550 return buf[0]&0x3F;
551 } else if (type == REDIS_RDB_14BITLEN) {
552 /* Read a 14 bit len */
553 if (fread(buf+1,1,1,fp) == 0) return REDIS_RDB_LENERR;
554 return ((buf[0]&0x3F)<<8)|buf[1];
555 } else {
556 /* Read a 32 bit len */
557 if (fread(&len,4,1,fp) == 0) return REDIS_RDB_LENERR;
558 return ntohl(len);
559 }
560}
561
562/* Load an integer-encoded object from file 'fp', with the specified
563 * encoding type 'enctype'. If encode is true the function may return
564 * an integer-encoded object as reply, otherwise the returned object
565 * will always be encoded as a raw string. */
566robj *rdbLoadIntegerObject(FILE *fp, int enctype, int encode) {
567 unsigned char enc[4];
568 long long val;
569
570 if (enctype == REDIS_RDB_ENC_INT8) {
571 if (fread(enc,1,1,fp) == 0) return NULL;
572 val = (signed char)enc[0];
573 } else if (enctype == REDIS_RDB_ENC_INT16) {
574 uint16_t v;
575 if (fread(enc,2,1,fp) == 0) return NULL;
576 v = enc[0]|(enc[1]<<8);
577 val = (int16_t)v;
578 } else if (enctype == REDIS_RDB_ENC_INT32) {
579 uint32_t v;
580 if (fread(enc,4,1,fp) == 0) return NULL;
581 v = enc[0]|(enc[1]<<8)|(enc[2]<<16)|(enc[3]<<24);
582 val = (int32_t)v;
583 } else {
584 val = 0; /* anti-warning */
585 redisPanic("Unknown RDB integer encoding type");
586 }
587 if (encode)
588 return createStringObjectFromLongLong(val);
589 else
590 return createObject(REDIS_STRING,sdsfromlonglong(val));
591}
592
593robj *rdbLoadLzfStringObject(FILE*fp) {
594 unsigned int len, clen;
595 unsigned char *c = NULL;
596 sds val = NULL;
597
598 if ((clen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
599 if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
600 if ((c = zmalloc(clen)) == NULL) goto err;
601 if ((val = sdsnewlen(NULL,len)) == NULL) goto err;
602 if (fread(c,clen,1,fp) == 0) goto err;
603 if (lzf_decompress(c,clen,val,len) == 0) goto err;
604 zfree(c);
605 return createObject(REDIS_STRING,val);
606err:
607 zfree(c);
608 sdsfree(val);
609 return NULL;
610}
611
612robj *rdbGenericLoadStringObject(FILE*fp, int encode) {
613 int isencoded;
614 uint32_t len;
615 sds val;
616
617 len = rdbLoadLen(fp,&isencoded);
618 if (isencoded) {
619 switch(len) {
620 case REDIS_RDB_ENC_INT8:
621 case REDIS_RDB_ENC_INT16:
622 case REDIS_RDB_ENC_INT32:
623 return rdbLoadIntegerObject(fp,len,encode);
624 case REDIS_RDB_ENC_LZF:
625 return rdbLoadLzfStringObject(fp);
626 default:
627 redisPanic("Unknown RDB encoding type");
628 }
629 }
630
631 if (len == REDIS_RDB_LENERR) return NULL;
632 val = sdsnewlen(NULL,len);
633 if (len && fread(val,len,1,fp) == 0) {
634 sdsfree(val);
635 return NULL;
636 }
637 return createObject(REDIS_STRING,val);
638}
639
640robj *rdbLoadStringObject(FILE *fp) {
641 return rdbGenericLoadStringObject(fp,0);
642}
643
644robj *rdbLoadEncodedStringObject(FILE *fp) {
645 return rdbGenericLoadStringObject(fp,1);
646}
647
648/* For information about double serialization check rdbSaveDoubleValue() */
649int rdbLoadDoubleValue(FILE *fp, double *val) {
650 char buf[128];
651 unsigned char len;
652
653 if (fread(&len,1,1,fp) == 0) return -1;
654 switch(len) {
655 case 255: *val = R_NegInf; return 0;
656 case 254: *val = R_PosInf; return 0;
657 case 253: *val = R_Nan; return 0;
658 default:
659 if (fread(buf,len,1,fp) == 0) return -1;
660 buf[len] = '\0';
661 sscanf(buf, "%lg", val);
662 return 0;
663 }
664}
665
666/* Load a Redis object of the specified type from the specified file.
667 * On success a newly allocated object is returned, otherwise NULL. */
668robj *rdbLoadObject(int type, FILE *fp) {
669 robj *o, *ele, *dec;
670 size_t len;
96ffb2fe 671 unsigned int i;
e2641e09 672
673 redisLog(REDIS_DEBUG,"LOADING OBJECT %d (at %d)\n",type,ftell(fp));
674 if (type == REDIS_STRING) {
675 /* Read string value */
676 if ((o = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
677 o = tryObjectEncoding(o);
678 } else if (type == REDIS_LIST) {
679 /* Read list value */
680 if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
681
682 /* Use a real list when there are too many entries */
683 if (len > server.list_max_ziplist_entries) {
684 o = createListObject();
685 } else {
686 o = createZiplistObject();
687 }
688
689 /* Load every single element of the list */
690 while(len--) {
691 if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
692
693 /* If we are using a ziplist and the value is too big, convert
694 * the object to a real list. */
695 if (o->encoding == REDIS_ENCODING_ZIPLIST &&
696 ele->encoding == REDIS_ENCODING_RAW &&
697 sdslen(ele->ptr) > server.list_max_ziplist_value)
698 listTypeConvert(o,REDIS_ENCODING_LINKEDLIST);
699
700 if (o->encoding == REDIS_ENCODING_ZIPLIST) {
701 dec = getDecodedObject(ele);
702 o->ptr = ziplistPush(o->ptr,dec->ptr,sdslen(dec->ptr),REDIS_TAIL);
703 decrRefCount(dec);
704 decrRefCount(ele);
705 } else {
706 ele = tryObjectEncoding(ele);
707 listAddNodeTail(o->ptr,ele);
708 }
709 }
710 } else if (type == REDIS_SET) {
711 /* Read list/set value */
712 if ((len = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
96ffb2fe
PN
713
714 /* Use a regular set when there are too many entries. */
715 if (len > server.set_max_intset_entries) {
716 o = createSetObject();
717 /* It's faster to expand the dict to the right size asap in order
718 * to avoid rehashing */
719 if (len > DICT_HT_INITIAL_SIZE)
720 dictExpand(o->ptr,len);
721 } else {
722 o = createIntsetObject();
723 }
724
e2641e09 725 /* Load every single element of the list/set */
96ffb2fe
PN
726 for (i = 0; i < len; i++) {
727 long long llval;
e2641e09 728 if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
729 ele = tryObjectEncoding(ele);
96ffb2fe
PN
730
731 if (o->encoding == REDIS_ENCODING_INTSET) {
732 /* Fetch integer value from element */
2df84b72 733 if (isObjectRepresentableAsLongLong(ele,&llval) == REDIS_OK) {
96ffb2fe
PN
734 o->ptr = intsetAdd(o->ptr,llval,NULL);
735 } else {
736 setTypeConvert(o,REDIS_ENCODING_HT);
737 dictExpand(o->ptr,len);
738 }
739 }
740
741 /* This will also be called when the set was just converted
742 * to regular hashtable encoded set */
743 if (o->encoding == REDIS_ENCODING_HT) {
744 dictAdd((dict*)o->ptr,ele,NULL);
bad7d097 745 } else {
746 decrRefCount(ele);
96ffb2fe 747 }
e2641e09 748 }
749 } else if (type == REDIS_ZSET) {
750 /* Read list/set value */
751 size_t zsetlen;
752 zset *zs;
753
754 if ((zsetlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
755 o = createZsetObject();
756 zs = o->ptr;
757 /* Load every single element of the list/set */
758 while(zsetlen--) {
759 robj *ele;
56e52b69
PN
760 double score;
761 zskiplistNode *znode;
e2641e09 762
763 if ((ele = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
764 ele = tryObjectEncoding(ele);
56e52b69
PN
765 if (rdbLoadDoubleValue(fp,&score) == -1) return NULL;
766 znode = zslInsert(zs->zsl,score,ele);
767 dictAdd(zs->dict,ele,&znode->score);
e2641e09 768 incrRefCount(ele); /* added to skiplist */
769 }
770 } else if (type == REDIS_HASH) {
771 size_t hashlen;
772
773 if ((hashlen = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR) return NULL;
774 o = createHashObject();
775 /* Too many entries? Use an hash table. */
776 if (hashlen > server.hash_max_zipmap_entries)
777 convertToRealHash(o);
778 /* Load every key/value, then set it into the zipmap or hash
779 * table, as needed. */
780 while(hashlen--) {
781 robj *key, *val;
782
783 if ((key = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
784 if ((val = rdbLoadEncodedStringObject(fp)) == NULL) return NULL;
785 /* If we are using a zipmap and there are too big values
786 * the object is converted to real hash table encoding. */
787 if (o->encoding != REDIS_ENCODING_HT &&
788 ((key->encoding == REDIS_ENCODING_RAW &&
789 sdslen(key->ptr) > server.hash_max_zipmap_value) ||
790 (val->encoding == REDIS_ENCODING_RAW &&
791 sdslen(val->ptr) > server.hash_max_zipmap_value)))
792 {
793 convertToRealHash(o);
794 }
795
796 if (o->encoding == REDIS_ENCODING_ZIPMAP) {
797 unsigned char *zm = o->ptr;
798 robj *deckey, *decval;
799
800 /* We need raw string objects to add them to the zipmap */
801 deckey = getDecodedObject(key);
802 decval = getDecodedObject(val);
803 zm = zipmapSet(zm,deckey->ptr,sdslen(deckey->ptr),
804 decval->ptr,sdslen(decval->ptr),NULL);
805 o->ptr = zm;
806 decrRefCount(deckey);
807 decrRefCount(decval);
808 decrRefCount(key);
809 decrRefCount(val);
810 } else {
811 key = tryObjectEncoding(key);
812 val = tryObjectEncoding(val);
813 dictAdd((dict*)o->ptr,key,val);
814 }
815 }
816 } else {
817 redisPanic("Unknown object type");
818 }
819 return o;
820}
821
97e7f8ae 822/* Mark that we are loading in the global state and setup the fields
823 * needed to provide loading stats. */
824void startLoading(FILE *fp) {
825 struct stat sb;
826
827 /* Load the DB */
828 server.loading = 1;
829 server.loading_start_time = time(NULL);
830 if (fstat(fileno(fp), &sb) == -1) {
831 server.loading_total_bytes = 1; /* just to avoid division by zero */
832 } else {
833 server.loading_total_bytes = sb.st_size;
834 }
835}
836
837/* Refresh the loading progress info */
838void loadingProgress(off_t pos) {
839 server.loading_loaded_bytes = pos;
840}
841
842/* Loading finished */
843void stopLoading(void) {
844 server.loading = 0;
845}
846
e2641e09 847int rdbLoad(char *filename) {
848 FILE *fp;
849 uint32_t dbid;
850 int type, retval, rdbver;
851 int swap_all_values = 0;
852 redisDb *db = server.db+0;
853 char buf[1024];
854 time_t expiretime, now = time(NULL);
97e7f8ae 855 long loops = 0;
e2641e09 856
857 fp = fopen(filename,"r");
858 if (!fp) return REDIS_ERR;
859 if (fread(buf,9,1,fp) == 0) goto eoferr;
860 buf[9] = '\0';
861 if (memcmp(buf,"REDIS",5) != 0) {
862 fclose(fp);
863 redisLog(REDIS_WARNING,"Wrong signature trying to load DB from file");
864 return REDIS_ERR;
865 }
866 rdbver = atoi(buf+5);
867 if (rdbver != 1) {
868 fclose(fp);
869 redisLog(REDIS_WARNING,"Can't handle RDB format version %d",rdbver);
870 return REDIS_ERR;
871 }
97e7f8ae 872
873 startLoading(fp);
e2641e09 874 while(1) {
875 robj *key, *val;
e2641e09 876 expiretime = -1;
97e7f8ae 877
878 /* Serve the clients from time to time */
879 if (!(loops++ % 1000)) {
880 loadingProgress(ftello(fp));
881 aeProcessEvents(server.el, AE_FILE_EVENTS|AE_DONT_WAIT);
882 }
883
e2641e09 884 /* Read type. */
885 if ((type = rdbLoadType(fp)) == -1) goto eoferr;
886 if (type == REDIS_EXPIRETIME) {
887 if ((expiretime = rdbLoadTime(fp)) == -1) goto eoferr;
888 /* We read the time so we need to read the object type again */
889 if ((type = rdbLoadType(fp)) == -1) goto eoferr;
890 }
891 if (type == REDIS_EOF) break;
892 /* Handle SELECT DB opcode as a special case */
893 if (type == REDIS_SELECTDB) {
894 if ((dbid = rdbLoadLen(fp,NULL)) == REDIS_RDB_LENERR)
895 goto eoferr;
896 if (dbid >= (unsigned)server.dbnum) {
897 redisLog(REDIS_WARNING,"FATAL: Data file was created with a Redis server configured to handle more than %d databases. Exiting\n", server.dbnum);
898 exit(1);
899 }
900 db = server.db+dbid;
901 continue;
902 }
903 /* Read key */
904 if ((key = rdbLoadStringObject(fp)) == NULL) goto eoferr;
905 /* Read value */
906 if ((val = rdbLoadObject(type,fp)) == NULL) goto eoferr;
907 /* Check if the key already expired */
908 if (expiretime != -1 && expiretime < now) {
909 decrRefCount(key);
910 decrRefCount(val);
911 continue;
912 }
913 /* Add the new object in the hash table */
914 retval = dbAdd(db,key,val);
915 if (retval == REDIS_ERR) {
916 redisLog(REDIS_WARNING,"Loading DB, duplicated key (%s) found! Unrecoverable error, exiting now.", key->ptr);
917 exit(1);
918 }
919 /* Set the expire time if needed */
920 if (expiretime != -1) setExpire(db,key,expiretime);
921
922 /* Handle swapping while loading big datasets when VM is on */
923
924 /* If we detecter we are hopeless about fitting something in memory
925 * we just swap every new key on disk. Directly...
926 * Note that's important to check for this condition before resorting
927 * to random sampling, otherwise we may try to swap already
928 * swapped keys. */
929 if (swap_all_values) {
930 dictEntry *de = dictFind(db->dict,key->ptr);
931
932 /* de may be NULL since the key already expired */
933 if (de) {
934 vmpointer *vp;
935 val = dictGetEntryVal(de);
936
937 if (val->refcount == 1 &&
938 (vp = vmSwapObjectBlocking(val)) != NULL)
939 dictGetEntryVal(de) = vp;
940 }
941 decrRefCount(key);
942 continue;
943 }
944 decrRefCount(key);
e2641e09 945 }
946 fclose(fp);
97e7f8ae 947 stopLoading();
e2641e09 948 return REDIS_OK;
949
950eoferr: /* unexpected end of file is handled here with a fatal exit */
951 redisLog(REDIS_WARNING,"Short read or OOM loading DB. Unrecoverable error, aborting now.");
952 exit(1);
953 return REDIS_ERR; /* Just to avoid warning */
954}
955
956/* A background saving child (BGSAVE) terminated its work. Handle this. */
957void backgroundSaveDoneHandler(int statloc) {
958 int exitcode = WEXITSTATUS(statloc);
959 int bysignal = WIFSIGNALED(statloc);
960
961 if (!bysignal && exitcode == 0) {
962 redisLog(REDIS_NOTICE,
963 "Background saving terminated with success");
2f6b31c3 964 server.dirty = server.dirty - server.dirty_before_bgsave;
e2641e09 965 server.lastsave = time(NULL);
966 } else if (!bysignal && exitcode != 0) {
967 redisLog(REDIS_WARNING, "Background saving error");
968 } else {
969 redisLog(REDIS_WARNING,
970 "Background saving terminated by signal %d", WTERMSIG(statloc));
971 rdbRemoveTempFile(server.bgsavechildpid);
972 }
973 server.bgsavechildpid = -1;
974 /* Possibly there are slaves waiting for a BGSAVE in order to be served
975 * (the first stage of SYNC is a bulk transfer of dump.rdb) */
976 updateSlavesWaitingBgsave(exitcode == 0 ? REDIS_OK : REDIS_ERR);
977}