]> git.saurik.com Git - redis.git/blob - src/redis.c
version bumped to 2.1.7 after merging with aaslave branch for non blocking slaves
[redis.git] / src / redis.c
1 /*
2 * Copyright (c) 2009-2010, Salvatore Sanfilippo <antirez at gmail dot com>
3 * All rights reserved.
4 *
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions are met:
7 *
8 * * Redistributions of source code must retain the above copyright notice,
9 * this list of conditions and the following disclaimer.
10 * * Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * * Neither the name of Redis nor the names of its contributors may be used
14 * to endorse or promote products derived from this software without
15 * specific prior written permission.
16 *
17 * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
18 * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR CONTRIBUTORS BE
21 * LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
22 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
23 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
24 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
25 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
26 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
27 * POSSIBILITY OF SUCH DAMAGE.
28 */
29
30 #include "redis.h"
31
32 #ifdef HAVE_BACKTRACE
33 #include <execinfo.h>
34 #include <ucontext.h>
35 #endif /* HAVE_BACKTRACE */
36
37 #include <time.h>
38 #include <signal.h>
39 #include <sys/wait.h>
40 #include <errno.h>
41 #include <assert.h>
42 #include <ctype.h>
43 #include <stdarg.h>
44 #include <arpa/inet.h>
45 #include <sys/stat.h>
46 #include <fcntl.h>
47 #include <sys/time.h>
48 #include <sys/resource.h>
49 #include <sys/uio.h>
50 #include <limits.h>
51 #include <float.h>
52 #include <math.h>
53 #include <pthread.h>
54 #include <sys/resource.h>
55
56 /* Our shared "common" objects */
57
58 struct sharedObjectsStruct shared;
59
60 /* Global vars that are actally used as constants. The following double
61 * values are used for double on-disk serialization, and are initialized
62 * at runtime to avoid strange compiler optimizations. */
63
64 double R_Zero, R_PosInf, R_NegInf, R_Nan;
65
66 /*================================= Globals ================================= */
67
68 /* Global vars */
69 struct redisServer server; /* server global state */
70 struct redisCommand *commandTable;
71 struct redisCommand readonlyCommandTable[] = {
72 {"get",getCommand,2,0,NULL,1,1,1},
73 {"set",setCommand,3,REDIS_CMD_DENYOOM,NULL,0,0,0},
74 {"setnx",setnxCommand,3,REDIS_CMD_DENYOOM,NULL,0,0,0},
75 {"setex",setexCommand,4,REDIS_CMD_DENYOOM,NULL,0,0,0},
76 {"append",appendCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
77 {"substr",substrCommand,4,0,NULL,1,1,1},
78 {"strlen",strlenCommand,2,0,NULL,1,1,1},
79 {"del",delCommand,-2,0,NULL,0,0,0},
80 {"exists",existsCommand,2,0,NULL,1,1,1},
81 {"incr",incrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1},
82 {"decr",decrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1},
83 {"mget",mgetCommand,-2,0,NULL,1,-1,1},
84 {"rpush",rpushCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
85 {"lpush",lpushCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
86 {"rpushx",rpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
87 {"lpushx",lpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
88 {"linsert",linsertCommand,5,REDIS_CMD_DENYOOM,NULL,1,1,1},
89 {"rpop",rpopCommand,2,0,NULL,1,1,1},
90 {"lpop",lpopCommand,2,0,NULL,1,1,1},
91 {"brpop",brpopCommand,-3,0,NULL,1,1,1},
92 {"blpop",blpopCommand,-3,0,NULL,1,1,1},
93 {"llen",llenCommand,2,0,NULL,1,1,1},
94 {"lindex",lindexCommand,3,0,NULL,1,1,1},
95 {"lset",lsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
96 {"lrange",lrangeCommand,4,0,NULL,1,1,1},
97 {"ltrim",ltrimCommand,4,0,NULL,1,1,1},
98 {"lrem",lremCommand,4,0,NULL,1,1,1},
99 {"rpoplpush",rpoplpushcommand,3,REDIS_CMD_DENYOOM,NULL,1,2,1},
100 {"sadd",saddCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
101 {"srem",sremCommand,3,0,NULL,1,1,1},
102 {"smove",smoveCommand,4,0,NULL,1,2,1},
103 {"sismember",sismemberCommand,3,0,NULL,1,1,1},
104 {"scard",scardCommand,2,0,NULL,1,1,1},
105 {"spop",spopCommand,2,0,NULL,1,1,1},
106 {"srandmember",srandmemberCommand,2,0,NULL,1,1,1},
107 {"sinter",sinterCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
108 {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
109 {"sunion",sunionCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
110 {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
111 {"sdiff",sdiffCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1},
112 {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1},
113 {"smembers",sinterCommand,2,0,NULL,1,1,1},
114 {"zadd",zaddCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
115 {"zincrby",zincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
116 {"zrem",zremCommand,3,0,NULL,1,1,1},
117 {"zremrangebyscore",zremrangebyscoreCommand,4,0,NULL,1,1,1},
118 {"zremrangebyrank",zremrangebyrankCommand,4,0,NULL,1,1,1},
119 {"zunionstore",zunionstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0},
120 {"zinterstore",zinterstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterBlockClientOnSwappedKeys,0,0,0},
121 {"zrange",zrangeCommand,-4,0,NULL,1,1,1},
122 {"zrangebyscore",zrangebyscoreCommand,-4,0,NULL,1,1,1},
123 {"zrevrangebyscore",zrevrangebyscoreCommand,-4,0,NULL,1,1,1},
124 {"zcount",zcountCommand,4,0,NULL,1,1,1},
125 {"zrevrange",zrevrangeCommand,-4,0,NULL,1,1,1},
126 {"zcard",zcardCommand,2,0,NULL,1,1,1},
127 {"zscore",zscoreCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
128 {"zrank",zrankCommand,3,0,NULL,1,1,1},
129 {"zrevrank",zrevrankCommand,3,0,NULL,1,1,1},
130 {"hset",hsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
131 {"hsetnx",hsetnxCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
132 {"hget",hgetCommand,3,0,NULL,1,1,1},
133 {"hmset",hmsetCommand,-4,REDIS_CMD_DENYOOM,NULL,1,1,1},
134 {"hmget",hmgetCommand,-3,0,NULL,1,1,1},
135 {"hincrby",hincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1},
136 {"hdel",hdelCommand,3,0,NULL,1,1,1},
137 {"hlen",hlenCommand,2,0,NULL,1,1,1},
138 {"hkeys",hkeysCommand,2,0,NULL,1,1,1},
139 {"hvals",hvalsCommand,2,0,NULL,1,1,1},
140 {"hgetall",hgetallCommand,2,0,NULL,1,1,1},
141 {"hexists",hexistsCommand,3,0,NULL,1,1,1},
142 {"incrby",incrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
143 {"decrby",decrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
144 {"getset",getsetCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1},
145 {"mset",msetCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2},
146 {"msetnx",msetnxCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2},
147 {"randomkey",randomkeyCommand,1,0,NULL,0,0,0},
148 {"select",selectCommand,2,0,NULL,0,0,0},
149 {"move",moveCommand,3,0,NULL,1,1,1},
150 {"rename",renameCommand,3,0,NULL,1,1,1},
151 {"renamenx",renamenxCommand,3,0,NULL,1,1,1},
152 {"expire",expireCommand,3,0,NULL,0,0,0},
153 {"expireat",expireatCommand,3,0,NULL,0,0,0},
154 {"keys",keysCommand,2,0,NULL,0,0,0},
155 {"dbsize",dbsizeCommand,1,0,NULL,0,0,0},
156 {"auth",authCommand,2,0,NULL,0,0,0},
157 {"ping",pingCommand,1,0,NULL,0,0,0},
158 {"echo",echoCommand,2,0,NULL,0,0,0},
159 {"save",saveCommand,1,0,NULL,0,0,0},
160 {"bgsave",bgsaveCommand,1,0,NULL,0,0,0},
161 {"bgrewriteaof",bgrewriteaofCommand,1,0,NULL,0,0,0},
162 {"shutdown",shutdownCommand,1,0,NULL,0,0,0},
163 {"lastsave",lastsaveCommand,1,0,NULL,0,0,0},
164 {"type",typeCommand,2,0,NULL,1,1,1},
165 {"multi",multiCommand,1,0,NULL,0,0,0},
166 {"exec",execCommand,1,REDIS_CMD_DENYOOM,execBlockClientOnSwappedKeys,0,0,0},
167 {"discard",discardCommand,1,0,NULL,0,0,0},
168 {"sync",syncCommand,1,0,NULL,0,0,0},
169 {"flushdb",flushdbCommand,1,0,NULL,0,0,0},
170 {"flushall",flushallCommand,1,0,NULL,0,0,0},
171 {"sort",sortCommand,-2,REDIS_CMD_DENYOOM,NULL,1,1,1},
172 {"info",infoCommand,1,0,NULL,0,0,0},
173 {"monitor",monitorCommand,1,0,NULL,0,0,0},
174 {"ttl",ttlCommand,2,0,NULL,1,1,1},
175 {"persist",persistCommand,2,0,NULL,1,1,1},
176 {"slaveof",slaveofCommand,3,0,NULL,0,0,0},
177 {"debug",debugCommand,-2,0,NULL,0,0,0},
178 {"config",configCommand,-2,0,NULL,0,0,0},
179 {"subscribe",subscribeCommand,-2,0,NULL,0,0,0},
180 {"unsubscribe",unsubscribeCommand,-1,0,NULL,0,0,0},
181 {"psubscribe",psubscribeCommand,-2,0,NULL,0,0,0},
182 {"punsubscribe",punsubscribeCommand,-1,0,NULL,0,0,0},
183 {"publish",publishCommand,3,REDIS_CMD_FORCE_REPLICATION,NULL,0,0,0},
184 {"watch",watchCommand,-2,0,NULL,0,0,0},
185 {"unwatch",unwatchCommand,1,0,NULL,0,0,0}
186 };
187
188 /*============================ Utility functions ============================ */
189
190 void redisLog(int level, const char *fmt, ...) {
191 va_list ap;
192 FILE *fp;
193 char *c = ".-*#";
194 char buf[64];
195 time_t now;
196
197 if (level < server.verbosity) return;
198
199 fp = (server.logfile == NULL) ? stdout : fopen(server.logfile,"a");
200 if (!fp) return;
201
202 va_start(ap, fmt);
203 now = time(NULL);
204 strftime(buf,64,"%d %b %H:%M:%S",localtime(&now));
205 fprintf(fp,"[%d] %s %c ",(int)getpid(),buf,c[level]);
206 vfprintf(fp, fmt, ap);
207 fprintf(fp,"\n");
208 fflush(fp);
209 va_end(ap);
210
211 if (server.logfile) fclose(fp);
212 }
213
214 /* Redis generally does not try to recover from out of memory conditions
215 * when allocating objects or strings, it is not clear if it will be possible
216 * to report this condition to the client since the networking layer itself
217 * is based on heap allocation for send buffers, so we simply abort.
218 * At least the code will be simpler to read... */
219 void oom(const char *msg) {
220 redisLog(REDIS_WARNING, "%s: Out of memory\n",msg);
221 sleep(1);
222 abort();
223 }
224
225 /*====================== Hash table type implementation ==================== */
226
227 /* This is an hash table type that uses the SDS dynamic strings libary as
228 * keys and radis objects as values (objects can hold SDS strings,
229 * lists, sets). */
230
231 void dictVanillaFree(void *privdata, void *val)
232 {
233 DICT_NOTUSED(privdata);
234 zfree(val);
235 }
236
237 void dictListDestructor(void *privdata, void *val)
238 {
239 DICT_NOTUSED(privdata);
240 listRelease((list*)val);
241 }
242
243 int dictSdsKeyCompare(void *privdata, const void *key1,
244 const void *key2)
245 {
246 int l1,l2;
247 DICT_NOTUSED(privdata);
248
249 l1 = sdslen((sds)key1);
250 l2 = sdslen((sds)key2);
251 if (l1 != l2) return 0;
252 return memcmp(key1, key2, l1) == 0;
253 }
254
255 /* A case insensitive version used for the command lookup table. */
256 int dictSdsKeyCaseCompare(void *privdata, const void *key1,
257 const void *key2)
258 {
259 DICT_NOTUSED(privdata);
260
261 return strcasecmp(key1, key2) == 0;
262 }
263
264 void dictRedisObjectDestructor(void *privdata, void *val)
265 {
266 DICT_NOTUSED(privdata);
267
268 if (val == NULL) return; /* Values of swapped out keys as set to NULL */
269 decrRefCount(val);
270 }
271
272 void dictSdsDestructor(void *privdata, void *val)
273 {
274 DICT_NOTUSED(privdata);
275
276 sdsfree(val);
277 }
278
279 int dictObjKeyCompare(void *privdata, const void *key1,
280 const void *key2)
281 {
282 const robj *o1 = key1, *o2 = key2;
283 return dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
284 }
285
286 unsigned int dictObjHash(const void *key) {
287 const robj *o = key;
288 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
289 }
290
291 unsigned int dictSdsHash(const void *key) {
292 return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
293 }
294
295 unsigned int dictSdsCaseHash(const void *key) {
296 return dictGenCaseHashFunction((unsigned char*)key, sdslen((char*)key));
297 }
298
299 int dictEncObjKeyCompare(void *privdata, const void *key1,
300 const void *key2)
301 {
302 robj *o1 = (robj*) key1, *o2 = (robj*) key2;
303 int cmp;
304
305 if (o1->encoding == REDIS_ENCODING_INT &&
306 o2->encoding == REDIS_ENCODING_INT)
307 return o1->ptr == o2->ptr;
308
309 o1 = getDecodedObject(o1);
310 o2 = getDecodedObject(o2);
311 cmp = dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
312 decrRefCount(o1);
313 decrRefCount(o2);
314 return cmp;
315 }
316
317 unsigned int dictEncObjHash(const void *key) {
318 robj *o = (robj*) key;
319
320 if (o->encoding == REDIS_ENCODING_RAW) {
321 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
322 } else {
323 if (o->encoding == REDIS_ENCODING_INT) {
324 char buf[32];
325 int len;
326
327 len = ll2string(buf,32,(long)o->ptr);
328 return dictGenHashFunction((unsigned char*)buf, len);
329 } else {
330 unsigned int hash;
331
332 o = getDecodedObject(o);
333 hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
334 decrRefCount(o);
335 return hash;
336 }
337 }
338 }
339
340 /* Sets type */
341 dictType setDictType = {
342 dictEncObjHash, /* hash function */
343 NULL, /* key dup */
344 NULL, /* val dup */
345 dictEncObjKeyCompare, /* key compare */
346 dictRedisObjectDestructor, /* key destructor */
347 NULL /* val destructor */
348 };
349
350 /* Sorted sets hash (note: a skiplist is used in addition to the hash table) */
351 dictType zsetDictType = {
352 dictEncObjHash, /* hash function */
353 NULL, /* key dup */
354 NULL, /* val dup */
355 dictEncObjKeyCompare, /* key compare */
356 dictRedisObjectDestructor, /* key destructor */
357 NULL /* val destructor */
358 };
359
360 /* Db->dict, keys are sds strings, vals are Redis objects. */
361 dictType dbDictType = {
362 dictSdsHash, /* hash function */
363 NULL, /* key dup */
364 NULL, /* val dup */
365 dictSdsKeyCompare, /* key compare */
366 dictSdsDestructor, /* key destructor */
367 dictRedisObjectDestructor /* val destructor */
368 };
369
370 /* Db->expires */
371 dictType keyptrDictType = {
372 dictSdsHash, /* hash function */
373 NULL, /* key dup */
374 NULL, /* val dup */
375 dictSdsKeyCompare, /* key compare */
376 NULL, /* key destructor */
377 NULL /* val destructor */
378 };
379
380 /* Command table. sds string -> command struct pointer. */
381 dictType commandTableDictType = {
382 dictSdsCaseHash, /* hash function */
383 NULL, /* key dup */
384 NULL, /* val dup */
385 dictSdsKeyCaseCompare, /* key compare */
386 dictSdsDestructor, /* key destructor */
387 NULL /* val destructor */
388 };
389
390 /* Hash type hash table (note that small hashes are represented with zimpaps) */
391 dictType hashDictType = {
392 dictEncObjHash, /* hash function */
393 NULL, /* key dup */
394 NULL, /* val dup */
395 dictEncObjKeyCompare, /* key compare */
396 dictRedisObjectDestructor, /* key destructor */
397 dictRedisObjectDestructor /* val destructor */
398 };
399
400 /* Keylist hash table type has unencoded redis objects as keys and
401 * lists as values. It's used for blocking operations (BLPOP) and to
402 * map swapped keys to a list of clients waiting for this keys to be loaded. */
403 dictType keylistDictType = {
404 dictObjHash, /* hash function */
405 NULL, /* key dup */
406 NULL, /* val dup */
407 dictObjKeyCompare, /* key compare */
408 dictRedisObjectDestructor, /* key destructor */
409 dictListDestructor /* val destructor */
410 };
411
412 int htNeedsResize(dict *dict) {
413 long long size, used;
414
415 size = dictSlots(dict);
416 used = dictSize(dict);
417 return (size && used && size > DICT_HT_INITIAL_SIZE &&
418 (used*100/size < REDIS_HT_MINFILL));
419 }
420
421 /* If the percentage of used slots in the HT reaches REDIS_HT_MINFILL
422 * we resize the hash table to save memory */
423 void tryResizeHashTables(void) {
424 int j;
425
426 for (j = 0; j < server.dbnum; j++) {
427 if (htNeedsResize(server.db[j].dict))
428 dictResize(server.db[j].dict);
429 if (htNeedsResize(server.db[j].expires))
430 dictResize(server.db[j].expires);
431 }
432 }
433
434 /* Our hash table implementation performs rehashing incrementally while
435 * we write/read from the hash table. Still if the server is idle, the hash
436 * table will use two tables for a long time. So we try to use 1 millisecond
437 * of CPU time at every serverCron() loop in order to rehash some key. */
438 void incrementallyRehash(void) {
439 int j;
440
441 for (j = 0; j < server.dbnum; j++) {
442 if (dictIsRehashing(server.db[j].dict)) {
443 dictRehashMilliseconds(server.db[j].dict,1);
444 break; /* already used our millisecond for this loop... */
445 }
446 }
447 }
448
449 /* This function is called once a background process of some kind terminates,
450 * as we want to avoid resizing the hash tables when there is a child in order
451 * to play well with copy-on-write (otherwise when a resize happens lots of
452 * memory pages are copied). The goal of this function is to update the ability
453 * for dict.c to resize the hash tables accordingly to the fact we have o not
454 * running childs. */
455 void updateDictResizePolicy(void) {
456 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
457 dictEnableResize();
458 else
459 dictDisableResize();
460 }
461
462 /* ======================= Cron: called every 100 ms ======================== */
463
464 /* Try to expire a few timed out keys. The algorithm used is adaptive and
465 * will use few CPU cycles if there are few expiring keys, otherwise
466 * it will get more aggressive to avoid that too much memory is used by
467 * keys that can be removed from the keyspace. */
468 void activeExpireCycle(void) {
469 int j;
470
471 for (j = 0; j < server.dbnum; j++) {
472 int expired;
473 redisDb *db = server.db+j;
474
475 /* Continue to expire if at the end of the cycle more than 25%
476 * of the keys were expired. */
477 do {
478 long num = dictSize(db->expires);
479 time_t now = time(NULL);
480
481 expired = 0;
482 if (num > REDIS_EXPIRELOOKUPS_PER_CRON)
483 num = REDIS_EXPIRELOOKUPS_PER_CRON;
484 while (num--) {
485 dictEntry *de;
486 time_t t;
487
488 if ((de = dictGetRandomKey(db->expires)) == NULL) break;
489 t = (time_t) dictGetEntryVal(de);
490 if (now > t) {
491 sds key = dictGetEntryKey(de);
492 robj *keyobj = createStringObject(key,sdslen(key));
493
494 propagateExpire(db,keyobj);
495 dbDelete(db,keyobj);
496 decrRefCount(keyobj);
497 expired++;
498 server.stat_expiredkeys++;
499 }
500 }
501 } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4);
502 }
503 }
504
505 void updateLRUClock(void) {
506 server.lruclock = (time(NULL)/REDIS_LRU_CLOCK_RESOLUTION) &
507 REDIS_LRU_CLOCK_MAX;
508 }
509
510 int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
511 int j, loops = server.cronloops++;
512 REDIS_NOTUSED(eventLoop);
513 REDIS_NOTUSED(id);
514 REDIS_NOTUSED(clientData);
515
516 /* We take a cached value of the unix time in the global state because
517 * with virtual memory and aging there is to store the current time
518 * in objects at every object access, and accuracy is not needed.
519 * To access a global var is faster than calling time(NULL) */
520 server.unixtime = time(NULL);
521 /* We have just 22 bits per object for LRU information.
522 * So we use an (eventually wrapping) LRU clock with 10 seconds resolution.
523 * 2^22 bits with 10 seconds resoluton is more or less 1.5 years.
524 *
525 * Note that even if this will wrap after 1.5 years it's not a problem,
526 * everything will still work but just some object will appear younger
527 * to Redis. But for this to happen a given object should never be touched
528 * for 1.5 years.
529 *
530 * Note that you can change the resolution altering the
531 * REDIS_LRU_CLOCK_RESOLUTION define.
532 */
533 updateLRUClock();
534
535 /* We received a SIGTERM, shutting down here in a safe way, as it is
536 * not ok doing so inside the signal handler. */
537 if (server.shutdown_asap) {
538 if (prepareForShutdown() == REDIS_OK) exit(0);
539 redisLog(REDIS_WARNING,"SIGTERM received but errors trying to shut down the server, check the logs for more information");
540 }
541
542 /* Show some info about non-empty databases */
543 for (j = 0; j < server.dbnum; j++) {
544 long long size, used, vkeys;
545
546 size = dictSlots(server.db[j].dict);
547 used = dictSize(server.db[j].dict);
548 vkeys = dictSize(server.db[j].expires);
549 if (!(loops % 50) && (used || vkeys)) {
550 redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
551 /* dictPrintStats(server.dict); */
552 }
553 }
554
555 /* We don't want to resize the hash tables while a bacground saving
556 * is in progress: the saving child is created using fork() that is
557 * implemented with a copy-on-write semantic in most modern systems, so
558 * if we resize the HT while there is the saving child at work actually
559 * a lot of memory movements in the parent will cause a lot of pages
560 * copied. */
561 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1) {
562 if (!(loops % 10)) tryResizeHashTables();
563 if (server.activerehashing) incrementallyRehash();
564 }
565
566 /* Show information about connected clients */
567 if (!(loops % 50)) {
568 redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use",
569 listLength(server.clients)-listLength(server.slaves),
570 listLength(server.slaves),
571 zmalloc_used_memory());
572 }
573
574 /* Close connections of timedout clients */
575 if ((server.maxidletime && !(loops % 100)) || server.blpop_blocked_clients)
576 closeTimedoutClients();
577
578 /* Check if a background saving or AOF rewrite in progress terminated */
579 if (server.bgsavechildpid != -1 || server.bgrewritechildpid != -1) {
580 int statloc;
581 pid_t pid;
582
583 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
584 if (pid == server.bgsavechildpid) {
585 backgroundSaveDoneHandler(statloc);
586 } else {
587 backgroundRewriteDoneHandler(statloc);
588 }
589 updateDictResizePolicy();
590 }
591 } else {
592 /* If there is not a background saving in progress check if
593 * we have to save now */
594 time_t now = time(NULL);
595 for (j = 0; j < server.saveparamslen; j++) {
596 struct saveparam *sp = server.saveparams+j;
597
598 if (server.dirty >= sp->changes &&
599 now-server.lastsave > sp->seconds) {
600 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
601 sp->changes, sp->seconds);
602 rdbSaveBackground(server.dbfilename);
603 break;
604 }
605 }
606 }
607
608 /* Expire a few keys per cycle, only if this is a master.
609 * On slaves we wait for DEL operations synthesized by the master
610 * in order to guarantee a strict consistency. */
611 if (server.masterhost == NULL) activeExpireCycle();
612
613 /* Swap a few keys on disk if we are over the memory limit and VM
614 * is enbled. Try to free objects from the free list first. */
615 if (vmCanSwapOut()) {
616 while (server.vm_enabled && zmalloc_used_memory() >
617 server.vm_max_memory)
618 {
619 int retval = (server.vm_max_threads == 0) ?
620 vmSwapOneObjectBlocking() :
621 vmSwapOneObjectThreaded();
622 if (retval == REDIS_ERR && !(loops % 300) &&
623 zmalloc_used_memory() >
624 (server.vm_max_memory+server.vm_max_memory/10))
625 {
626 redisLog(REDIS_WARNING,"WARNING: vm-max-memory limit exceeded by more than 10%% but unable to swap more objects out!");
627 }
628 /* Note that when using threade I/O we free just one object,
629 * because anyway when the I/O thread in charge to swap this
630 * object out will finish, the handler of completed jobs
631 * will try to swap more objects if we are still out of memory. */
632 if (retval == REDIS_ERR || server.vm_max_threads > 0) break;
633 }
634 }
635
636 /* Replication cron function -- used to reconnect to master and
637 * to detect transfer failures. */
638 if (!(loops % 10)) replicationCron();
639
640 return 100;
641 }
642
643 /* This function gets called every time Redis is entering the
644 * main loop of the event driven library, that is, before to sleep
645 * for ready file descriptors. */
646 void beforeSleep(struct aeEventLoop *eventLoop) {
647 REDIS_NOTUSED(eventLoop);
648
649 /* Awake clients that got all the swapped keys they requested */
650 if (server.vm_enabled && listLength(server.io_ready_clients)) {
651 listIter li;
652 listNode *ln;
653
654 listRewind(server.io_ready_clients,&li);
655 while((ln = listNext(&li))) {
656 redisClient *c = ln->value;
657 struct redisCommand *cmd;
658
659 /* Resume the client. */
660 listDelNode(server.io_ready_clients,ln);
661 c->flags &= (~REDIS_IO_WAIT);
662 server.vm_blocked_clients--;
663 aeCreateFileEvent(server.el, c->fd, AE_READABLE,
664 readQueryFromClient, c);
665 cmd = lookupCommand(c->argv[0]->ptr);
666 redisAssert(cmd != NULL);
667 call(c,cmd);
668 resetClient(c);
669 /* There may be more data to process in the input buffer. */
670 if (c->querybuf && sdslen(c->querybuf) > 0)
671 processInputBuffer(c);
672 }
673 }
674 /* Write the AOF buffer on disk */
675 flushAppendOnlyFile();
676 }
677
678 /* =========================== Server initialization ======================== */
679
680 void createSharedObjects(void) {
681 int j;
682
683 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
684 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
685 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
686 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
687 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
688 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
689 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
690 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
691 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
692 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
693 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
694 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
695 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
696 "-ERR Operation against a key holding the wrong kind of value\r\n"));
697 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
698 "-ERR no such key\r\n"));
699 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
700 "-ERR syntax error\r\n"));
701 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
702 "-ERR source and destination objects are the same\r\n"));
703 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
704 "-ERR index out of range\r\n"));
705 shared.space = createObject(REDIS_STRING,sdsnew(" "));
706 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
707 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
708 shared.select0 = createStringObject("select 0\r\n",10);
709 shared.select1 = createStringObject("select 1\r\n",10);
710 shared.select2 = createStringObject("select 2\r\n",10);
711 shared.select3 = createStringObject("select 3\r\n",10);
712 shared.select4 = createStringObject("select 4\r\n",10);
713 shared.select5 = createStringObject("select 5\r\n",10);
714 shared.select6 = createStringObject("select 6\r\n",10);
715 shared.select7 = createStringObject("select 7\r\n",10);
716 shared.select8 = createStringObject("select 8\r\n",10);
717 shared.select9 = createStringObject("select 9\r\n",10);
718 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
719 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
720 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
721 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
722 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
723 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
724 shared.mbulk3 = createStringObject("*3\r\n",4);
725 shared.mbulk4 = createStringObject("*4\r\n",4);
726 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
727 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
728 shared.integers[j]->encoding = REDIS_ENCODING_INT;
729 }
730 }
731
732 void initServerConfig() {
733 server.port = REDIS_SERVERPORT;
734 server.bindaddr = NULL;
735 server.unixsocket = NULL;
736 server.ipfd = -1;
737 server.sofd = -1;
738 server.dbnum = REDIS_DEFAULT_DBNUM;
739 server.verbosity = REDIS_VERBOSE;
740 server.maxidletime = REDIS_MAXIDLETIME;
741 server.saveparams = NULL;
742 server.logfile = NULL; /* NULL = log on standard output */
743 server.glueoutputbuf = 1;
744 server.daemonize = 0;
745 server.appendonly = 0;
746 server.appendfsync = APPENDFSYNC_EVERYSEC;
747 server.no_appendfsync_on_rewrite = 0;
748 server.lastfsync = time(NULL);
749 server.appendfd = -1;
750 server.appendseldb = -1; /* Make sure the first time will not match */
751 server.pidfile = zstrdup("/var/run/redis.pid");
752 server.dbfilename = zstrdup("dump.rdb");
753 server.appendfilename = zstrdup("appendonly.aof");
754 server.requirepass = NULL;
755 server.rdbcompression = 1;
756 server.activerehashing = 1;
757 server.maxclients = 0;
758 server.blpop_blocked_clients = 0;
759 server.maxmemory = 0;
760 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
761 server.maxmemory_samples = 3;
762 server.vm_enabled = 0;
763 server.vm_swap_file = zstrdup("/tmp/redis-%p.vm");
764 server.vm_page_size = 256; /* 256 bytes per page */
765 server.vm_pages = 1024*1024*100; /* 104 millions of pages */
766 server.vm_max_memory = 1024LL*1024*1024*1; /* 1 GB of RAM */
767 server.vm_max_threads = 4;
768 server.vm_blocked_clients = 0;
769 server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES;
770 server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE;
771 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
772 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
773 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
774 server.shutdown_asap = 0;
775
776 updateLRUClock();
777 resetServerSaveParams();
778
779 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
780 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
781 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
782 /* Replication related */
783 server.isslave = 0;
784 server.masterauth = NULL;
785 server.masterhost = NULL;
786 server.masterport = 6379;
787 server.master = NULL;
788 server.replstate = REDIS_REPL_NONE;
789 server.repl_serve_stale_data = 1;
790
791 /* Double constants initialization */
792 R_Zero = 0.0;
793 R_PosInf = 1.0/R_Zero;
794 R_NegInf = -1.0/R_Zero;
795 R_Nan = R_Zero/R_Zero;
796
797 /* Command table -- we intiialize it here as it is part of the
798 * initial configuration, since command names may be changed via
799 * redis.conf using the rename-command directive. */
800 server.commands = dictCreate(&commandTableDictType,NULL);
801 populateCommandTable();
802 server.delCommand = lookupCommandByCString("del");
803 server.multiCommand = lookupCommandByCString("multi");
804 }
805
806 void initServer() {
807 int j;
808
809 signal(SIGHUP, SIG_IGN);
810 signal(SIGPIPE, SIG_IGN);
811 setupSigSegvAction();
812
813 server.mainthread = pthread_self();
814 server.devnull = fopen("/dev/null","w");
815 if (server.devnull == NULL) {
816 redisLog(REDIS_WARNING, "Can't open /dev/null: %s", server.neterr);
817 exit(1);
818 }
819 server.clients = listCreate();
820 server.slaves = listCreate();
821 server.monitors = listCreate();
822 createSharedObjects();
823 server.el = aeCreateEventLoop();
824 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
825 server.ipfd = anetTcpServer(server.neterr,server.port,server.bindaddr);
826 if (server.ipfd == ANET_ERR) {
827 redisLog(REDIS_WARNING, "Opening port: %s", server.neterr);
828 exit(1);
829 }
830 if (server.unixsocket != NULL) {
831 unlink(server.unixsocket); /* don't care if this fails */
832 server.sofd = anetUnixServer(server.neterr,server.unixsocket);
833 if (server.sofd == ANET_ERR) {
834 redisLog(REDIS_WARNING, "Opening socket: %s", server.neterr);
835 exit(1);
836 }
837 }
838 if (server.ipfd < 0 && server.sofd < 0) {
839 redisLog(REDIS_WARNING, "Configured to not listen anywhere, exiting.");
840 exit(1);
841 }
842 for (j = 0; j < server.dbnum; j++) {
843 server.db[j].dict = dictCreate(&dbDictType,NULL);
844 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
845 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
846 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
847 if (server.vm_enabled)
848 server.db[j].io_keys = dictCreate(&keylistDictType,NULL);
849 server.db[j].id = j;
850 }
851 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
852 server.pubsub_patterns = listCreate();
853 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
854 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
855 server.cronloops = 0;
856 server.bgsavechildpid = -1;
857 server.bgrewritechildpid = -1;
858 server.bgrewritebuf = sdsempty();
859 server.aofbuf = sdsempty();
860 server.lastsave = time(NULL);
861 server.dirty = 0;
862 server.stat_numcommands = 0;
863 server.stat_numconnections = 0;
864 server.stat_expiredkeys = 0;
865 server.stat_starttime = time(NULL);
866 server.stat_keyspace_misses = 0;
867 server.stat_keyspace_hits = 0;
868 server.unixtime = time(NULL);
869 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
870 if (server.ipfd > 0 && aeCreateFileEvent(server.el,server.ipfd,AE_READABLE,
871 acceptTcpHandler,NULL) == AE_ERR) oom("creating file event");
872 if (server.sofd > 0 && aeCreateFileEvent(server.el,server.sofd,AE_READABLE,
873 acceptUnixHandler,NULL) == AE_ERR) oom("creating file event");
874
875 if (server.appendonly) {
876 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
877 if (server.appendfd == -1) {
878 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
879 strerror(errno));
880 exit(1);
881 }
882 }
883
884 if (server.vm_enabled) vmInit();
885 }
886
887 /* Populates the Redis Command Table starting from the hard coded list
888 * we have on top of redis.c file. */
889 void populateCommandTable(void) {
890 int j;
891 int numcommands = sizeof(readonlyCommandTable)/sizeof(struct redisCommand);
892
893 for (j = 0; j < numcommands; j++) {
894 struct redisCommand *c = readonlyCommandTable+j;
895 int retval;
896
897 retval = dictAdd(server.commands, sdsnew(c->name), c);
898 assert(retval == DICT_OK);
899 }
900 }
901
902 /* ====================== Commands lookup and execution ===================== */
903
904 struct redisCommand *lookupCommand(sds name) {
905 return dictFetchValue(server.commands, name);
906 }
907
908 struct redisCommand *lookupCommandByCString(char *s) {
909 struct redisCommand *cmd;
910 sds name = sdsnew(s);
911
912 cmd = dictFetchValue(server.commands, name);
913 sdsfree(name);
914 return cmd;
915 }
916
917 /* Call() is the core of Redis execution of a command */
918 void call(redisClient *c, struct redisCommand *cmd) {
919 long long dirty;
920
921 dirty = server.dirty;
922 cmd->proc(c);
923 dirty = server.dirty-dirty;
924
925 if (server.appendonly && dirty)
926 feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc);
927 if ((dirty || cmd->flags & REDIS_CMD_FORCE_REPLICATION) &&
928 listLength(server.slaves))
929 replicationFeedSlaves(server.slaves,c->db->id,c->argv,c->argc);
930 if (listLength(server.monitors))
931 replicationFeedMonitors(server.monitors,c->db->id,c->argv,c->argc);
932 server.stat_numcommands++;
933 }
934
935 /* If this function gets called we already read a whole
936 * command, argments are in the client argv/argc fields.
937 * processCommand() execute the command or prepare the
938 * server for a bulk read from the client.
939 *
940 * If 1 is returned the client is still alive and valid and
941 * and other operations can be performed by the caller. Otherwise
942 * if 0 is returned the client was destroied (i.e. after QUIT). */
943 int processCommand(redisClient *c) {
944 struct redisCommand *cmd;
945
946 /* The QUIT command is handled separately. Normal command procs will
947 * go through checking for replication and QUIT will cause trouble
948 * when FORCE_REPLICATION is enabled and would be implemented in
949 * a regular command proc. */
950 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
951 addReply(c,shared.ok);
952 c->flags |= REDIS_CLOSE_AFTER_REPLY;
953 return REDIS_ERR;
954 }
955
956 /* Now lookup the command and check ASAP about trivial error conditions
957 * such wrong arity, bad command name and so forth. */
958 cmd = lookupCommand(c->argv[0]->ptr);
959 if (!cmd) {
960 addReplyErrorFormat(c,"unknown command '%s'",
961 (char*)c->argv[0]->ptr);
962 return REDIS_OK;
963 } else if ((cmd->arity > 0 && cmd->arity != c->argc) ||
964 (c->argc < -cmd->arity)) {
965 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
966 cmd->name);
967 return REDIS_OK;
968 }
969
970 /* Check if the user is authenticated */
971 if (server.requirepass && !c->authenticated && cmd->proc != authCommand) {
972 addReplyError(c,"operation not permitted");
973 return REDIS_OK;
974 }
975
976 /* Handle the maxmemory directive.
977 *
978 * First we try to free some memory if possible (if there are volatile
979 * keys in the dataset). If there are not the only thing we can do
980 * is returning an error. */
981 if (server.maxmemory) freeMemoryIfNeeded();
982 if (server.maxmemory && (cmd->flags & REDIS_CMD_DENYOOM) &&
983 zmalloc_used_memory() > server.maxmemory)
984 {
985 addReplyError(c,"command not allowed when used memory > 'maxmemory'");
986 return REDIS_OK;
987 }
988
989 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
990 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
991 &&
992 cmd->proc != subscribeCommand && cmd->proc != unsubscribeCommand &&
993 cmd->proc != psubscribeCommand && cmd->proc != punsubscribeCommand) {
994 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
995 return REDIS_OK;
996 }
997
998 /* Only allow INFO and SLAVEOF when slave-serve-stale-data is no and
999 * we are a slave with a broken link with master. */
1000 if (server.masterhost && server.replstate != REDIS_REPL_CONNECTED &&
1001 server.repl_serve_stale_data == 0 &&
1002 cmd->proc != infoCommand && cmd->proc != slaveofCommand)
1003 {
1004 addReplyError(c,
1005 "link with MASTER is down and slave-serve-stale-data is set to no");
1006 return REDIS_OK;
1007 }
1008
1009 /* Exec the command */
1010 if (c->flags & REDIS_MULTI &&
1011 cmd->proc != execCommand && cmd->proc != discardCommand &&
1012 cmd->proc != multiCommand && cmd->proc != watchCommand)
1013 {
1014 queueMultiCommand(c,cmd);
1015 addReply(c,shared.queued);
1016 } else {
1017 if (server.vm_enabled && server.vm_max_threads > 0 &&
1018 blockClientOnSwappedKeys(c,cmd)) return REDIS_ERR;
1019 call(c,cmd);
1020 }
1021 return REDIS_OK;
1022 }
1023
1024 /*================================== Shutdown =============================== */
1025
1026 int prepareForShutdown() {
1027 redisLog(REDIS_WARNING,"User requested shutdown, saving DB...");
1028 /* Kill the saving child if there is a background saving in progress.
1029 We want to avoid race conditions, for instance our saving child may
1030 overwrite the synchronous saving did by SHUTDOWN. */
1031 if (server.bgsavechildpid != -1) {
1032 redisLog(REDIS_WARNING,"There is a live saving child. Killing it!");
1033 kill(server.bgsavechildpid,SIGKILL);
1034 rdbRemoveTempFile(server.bgsavechildpid);
1035 }
1036 if (server.appendonly) {
1037 /* Append only file: fsync() the AOF and exit */
1038 aof_fsync(server.appendfd);
1039 if (server.vm_enabled) unlink(server.vm_swap_file);
1040 } else if (server.saveparamslen > 0) {
1041 /* Snapshotting. Perform a SYNC SAVE and exit */
1042 if (rdbSave(server.dbfilename) != REDIS_OK) {
1043 /* Ooops.. error saving! The best we can do is to continue
1044 * operating. Note that if there was a background saving process,
1045 * in the next cron() Redis will be notified that the background
1046 * saving aborted, handling special stuff like slaves pending for
1047 * synchronization... */
1048 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit");
1049 return REDIS_ERR;
1050 }
1051 } else {
1052 redisLog(REDIS_WARNING,"Not saving DB.");
1053 }
1054 if (server.daemonize) unlink(server.pidfile);
1055 redisLog(REDIS_WARNING,"Server exit now, bye bye...");
1056 return REDIS_OK;
1057 }
1058
1059 /*================================== Commands =============================== */
1060
1061 void authCommand(redisClient *c) {
1062 if (!server.requirepass || !strcmp(c->argv[1]->ptr, server.requirepass)) {
1063 c->authenticated = 1;
1064 addReply(c,shared.ok);
1065 } else {
1066 c->authenticated = 0;
1067 addReplyError(c,"invalid password");
1068 }
1069 }
1070
1071 void pingCommand(redisClient *c) {
1072 addReply(c,shared.pong);
1073 }
1074
1075 void echoCommand(redisClient *c) {
1076 addReplyBulk(c,c->argv[1]);
1077 }
1078
1079 /* Convert an amount of bytes into a human readable string in the form
1080 * of 100B, 2G, 100M, 4K, and so forth. */
1081 void bytesToHuman(char *s, unsigned long long n) {
1082 double d;
1083
1084 if (n < 1024) {
1085 /* Bytes */
1086 sprintf(s,"%lluB",n);
1087 return;
1088 } else if (n < (1024*1024)) {
1089 d = (double)n/(1024);
1090 sprintf(s,"%.2fK",d);
1091 } else if (n < (1024LL*1024*1024)) {
1092 d = (double)n/(1024*1024);
1093 sprintf(s,"%.2fM",d);
1094 } else if (n < (1024LL*1024*1024*1024)) {
1095 d = (double)n/(1024LL*1024*1024);
1096 sprintf(s,"%.2fG",d);
1097 }
1098 }
1099
1100 /* Create the string returned by the INFO command. This is decoupled
1101 * by the INFO command itself as we need to report the same information
1102 * on memory corruption problems. */
1103 sds genRedisInfoString(void) {
1104 sds info;
1105 time_t uptime = time(NULL)-server.stat_starttime;
1106 int j;
1107 char hmem[64];
1108 struct rusage self_ru, c_ru;
1109
1110 getrusage(RUSAGE_SELF, &self_ru);
1111 getrusage(RUSAGE_CHILDREN, &c_ru);
1112
1113 bytesToHuman(hmem,zmalloc_used_memory());
1114 info = sdscatprintf(sdsempty(),
1115 "redis_version:%s\r\n"
1116 "redis_git_sha1:%s\r\n"
1117 "redis_git_dirty:%d\r\n"
1118 "arch_bits:%s\r\n"
1119 "multiplexing_api:%s\r\n"
1120 "process_id:%ld\r\n"
1121 "uptime_in_seconds:%ld\r\n"
1122 "uptime_in_days:%ld\r\n"
1123 "lru_clock:%ld\r\n"
1124 "used_cpu_sys:%.2f\r\n"
1125 "used_cpu_user:%.2f\r\n"
1126 "used_cpu_sys_childrens:%.2f\r\n"
1127 "used_cpu_user_childrens:%.2f\r\n"
1128 "connected_clients:%d\r\n"
1129 "connected_slaves:%d\r\n"
1130 "blocked_clients:%d\r\n"
1131 "used_memory:%zu\r\n"
1132 "used_memory_human:%s\r\n"
1133 "used_memory_rss:%zu\r\n"
1134 "mem_fragmentation_ratio:%.2f\r\n"
1135 "use_tcmalloc:%d\r\n"
1136 "changes_since_last_save:%lld\r\n"
1137 "bgsave_in_progress:%d\r\n"
1138 "last_save_time:%ld\r\n"
1139 "bgrewriteaof_in_progress:%d\r\n"
1140 "total_connections_received:%lld\r\n"
1141 "total_commands_processed:%lld\r\n"
1142 "expired_keys:%lld\r\n"
1143 "keyspace_hits:%lld\r\n"
1144 "keyspace_misses:%lld\r\n"
1145 "hash_max_zipmap_entries:%zu\r\n"
1146 "hash_max_zipmap_value:%zu\r\n"
1147 "pubsub_channels:%ld\r\n"
1148 "pubsub_patterns:%u\r\n"
1149 "vm_enabled:%d\r\n"
1150 "role:%s\r\n"
1151 ,REDIS_VERSION,
1152 redisGitSHA1(),
1153 strtol(redisGitDirty(),NULL,10) > 0,
1154 (sizeof(long) == 8) ? "64" : "32",
1155 aeGetApiName(),
1156 (long) getpid(),
1157 uptime,
1158 uptime/(3600*24),
1159 (unsigned long) server.lruclock,
1160 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
1161 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
1162 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000,
1163 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000,
1164 listLength(server.clients)-listLength(server.slaves),
1165 listLength(server.slaves),
1166 server.blpop_blocked_clients,
1167 zmalloc_used_memory(),
1168 hmem,
1169 zmalloc_get_rss(),
1170 zmalloc_get_fragmentation_ratio(),
1171 #ifdef USE_TCMALLOC
1172 1,
1173 #else
1174 0,
1175 #endif
1176 server.dirty,
1177 server.bgsavechildpid != -1,
1178 server.lastsave,
1179 server.bgrewritechildpid != -1,
1180 server.stat_numconnections,
1181 server.stat_numcommands,
1182 server.stat_expiredkeys,
1183 server.stat_keyspace_hits,
1184 server.stat_keyspace_misses,
1185 server.hash_max_zipmap_entries,
1186 server.hash_max_zipmap_value,
1187 dictSize(server.pubsub_channels),
1188 listLength(server.pubsub_patterns),
1189 server.vm_enabled != 0,
1190 server.masterhost == NULL ? "master" : "slave"
1191 );
1192 if (server.masterhost) {
1193 info = sdscatprintf(info,
1194 "master_host:%s\r\n"
1195 "master_port:%d\r\n"
1196 "master_link_status:%s\r\n"
1197 "master_last_io_seconds_ago:%d\r\n"
1198 "master_sync_in_progress:%d\r\n"
1199 ,server.masterhost,
1200 server.masterport,
1201 (server.replstate == REDIS_REPL_CONNECTED) ?
1202 "up" : "down",
1203 server.master ? ((int)(time(NULL)-server.master->lastinteraction)) : -1,
1204 server.replstate == REDIS_REPL_TRANSFER
1205 );
1206
1207 if (server.replstate == REDIS_REPL_TRANSFER) {
1208 info = sdscatprintf(info,
1209 "master_sync_left_bytes:%ld\r\n"
1210 "master_sync_last_io_seconds_ago:%d\r\n"
1211 ,(long)server.repl_transfer_left,
1212 (int)(time(NULL)-server.repl_transfer_lastio)
1213 );
1214 }
1215 }
1216 if (server.vm_enabled) {
1217 lockThreadedIO();
1218 info = sdscatprintf(info,
1219 "vm_conf_max_memory:%llu\r\n"
1220 "vm_conf_page_size:%llu\r\n"
1221 "vm_conf_pages:%llu\r\n"
1222 "vm_stats_used_pages:%llu\r\n"
1223 "vm_stats_swapped_objects:%llu\r\n"
1224 "vm_stats_swappin_count:%llu\r\n"
1225 "vm_stats_swappout_count:%llu\r\n"
1226 "vm_stats_io_newjobs_len:%lu\r\n"
1227 "vm_stats_io_processing_len:%lu\r\n"
1228 "vm_stats_io_processed_len:%lu\r\n"
1229 "vm_stats_io_active_threads:%lu\r\n"
1230 "vm_stats_blocked_clients:%lu\r\n"
1231 ,(unsigned long long) server.vm_max_memory,
1232 (unsigned long long) server.vm_page_size,
1233 (unsigned long long) server.vm_pages,
1234 (unsigned long long) server.vm_stats_used_pages,
1235 (unsigned long long) server.vm_stats_swapped_objects,
1236 (unsigned long long) server.vm_stats_swapins,
1237 (unsigned long long) server.vm_stats_swapouts,
1238 (unsigned long) listLength(server.io_newjobs),
1239 (unsigned long) listLength(server.io_processing),
1240 (unsigned long) listLength(server.io_processed),
1241 (unsigned long) server.io_active_threads,
1242 (unsigned long) server.vm_blocked_clients
1243 );
1244 unlockThreadedIO();
1245 }
1246 for (j = 0; j < server.dbnum; j++) {
1247 long long keys, vkeys;
1248
1249 keys = dictSize(server.db[j].dict);
1250 vkeys = dictSize(server.db[j].expires);
1251 if (keys || vkeys) {
1252 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
1253 j, keys, vkeys);
1254 }
1255 }
1256 return info;
1257 }
1258
1259 void infoCommand(redisClient *c) {
1260 sds info = genRedisInfoString();
1261 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
1262 (unsigned long)sdslen(info)));
1263 addReplySds(c,info);
1264 addReply(c,shared.crlf);
1265 }
1266
1267 void monitorCommand(redisClient *c) {
1268 /* ignore MONITOR if aleady slave or in monitor mode */
1269 if (c->flags & REDIS_SLAVE) return;
1270
1271 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
1272 c->slaveseldb = 0;
1273 listAddNodeTail(server.monitors,c);
1274 addReply(c,shared.ok);
1275 }
1276
1277 /* ============================ Maxmemory directive ======================== */
1278
1279 /* This function gets called when 'maxmemory' is set on the config file to limit
1280 * the max memory used by the server, and we are out of memory.
1281 * This function will try to, in order:
1282 *
1283 * - Free objects from the free list
1284 * - Try to remove keys with an EXPIRE set
1285 *
1286 * It is not possible to free enough memory to reach used-memory < maxmemory
1287 * the server will start refusing commands that will enlarge even more the
1288 * memory usage.
1289 */
1290 void freeMemoryIfNeeded(void) {
1291 /* Remove keys accordingly to the active policy as long as we are
1292 * over the memory limit. */
1293 while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) {
1294 int j, k, freed = 0;
1295
1296 for (j = 0; j < server.dbnum; j++) {
1297 long bestval = 0; /* just to prevent warning */
1298 sds bestkey = NULL;
1299 struct dictEntry *de;
1300 redisDb *db = server.db+j;
1301 dict *dict;
1302
1303 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1304 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
1305 {
1306 dict = server.db[j].dict;
1307 } else {
1308 dict = server.db[j].expires;
1309 }
1310 if (dictSize(dict) == 0) continue;
1311
1312 /* volatile-random and allkeys-random policy */
1313 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
1314 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
1315 {
1316 de = dictGetRandomKey(dict);
1317 bestkey = dictGetEntryKey(de);
1318 }
1319
1320 /* volatile-lru and allkeys-lru policy */
1321 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1322 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1323 {
1324 for (k = 0; k < server.maxmemory_samples; k++) {
1325 sds thiskey;
1326 long thisval;
1327 robj *o;
1328
1329 de = dictGetRandomKey(dict);
1330 thiskey = dictGetEntryKey(de);
1331 o = dictGetEntryVal(de);
1332 thisval = estimateObjectIdleTime(o);
1333
1334 /* Higher idle time is better candidate for deletion */
1335 if (bestkey == NULL || thisval > bestval) {
1336 bestkey = thiskey;
1337 bestval = thisval;
1338 }
1339 }
1340 }
1341
1342 /* volatile-ttl */
1343 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
1344 for (k = 0; k < server.maxmemory_samples; k++) {
1345 sds thiskey;
1346 long thisval;
1347
1348 de = dictGetRandomKey(dict);
1349 thiskey = dictGetEntryKey(de);
1350 thisval = (long) dictGetEntryVal(de);
1351
1352 /* Expire sooner (minor expire unix timestamp) is better
1353 * candidate for deletion */
1354 if (bestkey == NULL || thisval < bestval) {
1355 bestkey = thiskey;
1356 bestval = thisval;
1357 }
1358 }
1359 }
1360
1361 /* Finally remove the selected key. */
1362 if (bestkey) {
1363 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
1364 dbDelete(db,keyobj);
1365 server.stat_expiredkeys++;
1366 decrRefCount(keyobj);
1367 freed++;
1368 }
1369 }
1370 if (!freed) return; /* nothing to free... */
1371 }
1372
1373 while(0) {
1374 int j, k, freed = 0;
1375 for (j = 0; j < server.dbnum; j++) {
1376 int minttl = -1;
1377 sds minkey = NULL;
1378 robj *keyobj = NULL;
1379 struct dictEntry *de;
1380
1381 if (dictSize(server.db[j].expires)) {
1382 freed = 1;
1383 /* From a sample of three keys drop the one nearest to
1384 * the natural expire */
1385 for (k = 0; k < 3; k++) {
1386 time_t t;
1387
1388 de = dictGetRandomKey(server.db[j].expires);
1389 t = (time_t) dictGetEntryVal(de);
1390 if (minttl == -1 || t < minttl) {
1391 minkey = dictGetEntryKey(de);
1392 minttl = t;
1393 }
1394 }
1395 keyobj = createStringObject(minkey,sdslen(minkey));
1396 dbDelete(server.db+j,keyobj);
1397 server.stat_expiredkeys++;
1398 decrRefCount(keyobj);
1399 }
1400 }
1401 if (!freed) return; /* nothing to free... */
1402 }
1403 }
1404
1405 /* =================================== Main! ================================ */
1406
1407 #ifdef __linux__
1408 int linuxOvercommitMemoryValue(void) {
1409 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
1410 char buf[64];
1411
1412 if (!fp) return -1;
1413 if (fgets(buf,64,fp) == NULL) {
1414 fclose(fp);
1415 return -1;
1416 }
1417 fclose(fp);
1418
1419 return atoi(buf);
1420 }
1421
1422 void linuxOvercommitMemoryWarning(void) {
1423 if (linuxOvercommitMemoryValue() == 0) {
1424 redisLog(REDIS_WARNING,"WARNING overcommit_memory is set to 0! Background save may fail under low memory condition. To fix this issue add 'vm.overcommit_memory = 1' to /etc/sysctl.conf and then reboot or run the command 'sysctl vm.overcommit_memory=1' for this to take effect.");
1425 }
1426 }
1427 #endif /* __linux__ */
1428
1429 void createPidFile(void) {
1430 /* Try to write the pid file in a best-effort way. */
1431 FILE *fp = fopen(server.pidfile,"w");
1432 if (fp) {
1433 fprintf(fp,"%d\n",getpid());
1434 fclose(fp);
1435 }
1436 }
1437
1438 void daemonize(void) {
1439 int fd;
1440
1441 if (fork() != 0) exit(0); /* parent exits */
1442 setsid(); /* create a new session */
1443
1444 /* Every output goes to /dev/null. If Redis is daemonized but
1445 * the 'logfile' is set to 'stdout' in the configuration file
1446 * it will not log at all. */
1447 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
1448 dup2(fd, STDIN_FILENO);
1449 dup2(fd, STDOUT_FILENO);
1450 dup2(fd, STDERR_FILENO);
1451 if (fd > STDERR_FILENO) close(fd);
1452 }
1453 }
1454
1455 void version() {
1456 printf("Redis server version %s (%s:%d)\n", REDIS_VERSION,
1457 redisGitSHA1(), atoi(redisGitDirty()) > 0);
1458 exit(0);
1459 }
1460
1461 void usage() {
1462 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n");
1463 fprintf(stderr," ./redis-server - (read config from stdin)\n");
1464 exit(1);
1465 }
1466
1467 int main(int argc, char **argv) {
1468 time_t start;
1469
1470 initServerConfig();
1471 if (argc == 2) {
1472 if (strcmp(argv[1], "-v") == 0 ||
1473 strcmp(argv[1], "--version") == 0) version();
1474 if (strcmp(argv[1], "--help") == 0) usage();
1475 resetServerSaveParams();
1476 loadServerConfig(argv[1]);
1477 } else if ((argc > 2)) {
1478 usage();
1479 } else {
1480 redisLog(REDIS_WARNING,"Warning: no config file specified, using the default config. In order to specify a config file use 'redis-server /path/to/redis.conf'");
1481 }
1482 if (server.daemonize) daemonize();
1483 initServer();
1484 if (server.daemonize) createPidFile();
1485 redisLog(REDIS_NOTICE,"Server started, Redis version " REDIS_VERSION);
1486 #ifdef __linux__
1487 linuxOvercommitMemoryWarning();
1488 #endif
1489 start = time(NULL);
1490 if (server.appendonly) {
1491 if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK)
1492 redisLog(REDIS_NOTICE,"DB loaded from append only file: %ld seconds",time(NULL)-start);
1493 } else {
1494 if (rdbLoad(server.dbfilename) == REDIS_OK)
1495 redisLog(REDIS_NOTICE,"DB loaded from disk: %ld seconds",time(NULL)-start);
1496 }
1497 if (server.ipfd > 0)
1498 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
1499 if (server.sofd > 0)
1500 redisLog(REDIS_NOTICE,"The server is now ready to accept connections at %s", server.unixsocket);
1501 aeSetBeforeSleepProc(server.el,beforeSleep);
1502 aeMain(server.el);
1503 aeDeleteEventLoop(server.el);
1504 return 0;
1505 }
1506
1507 /* ============================= Backtrace support ========================= */
1508
1509 #ifdef HAVE_BACKTRACE
1510 void *getMcontextEip(ucontext_t *uc) {
1511 #if defined(__FreeBSD__)
1512 return (void*) uc->uc_mcontext.mc_eip;
1513 #elif defined(__dietlibc__)
1514 return (void*) uc->uc_mcontext.eip;
1515 #elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
1516 #if __x86_64__
1517 return (void*) uc->uc_mcontext->__ss.__rip;
1518 #else
1519 return (void*) uc->uc_mcontext->__ss.__eip;
1520 #endif
1521 #elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6)
1522 #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__)
1523 return (void*) uc->uc_mcontext->__ss.__rip;
1524 #else
1525 return (void*) uc->uc_mcontext->__ss.__eip;
1526 #endif
1527 #elif defined(__i386__)
1528 return (void*) uc->uc_mcontext.gregs[14]; /* Linux 32 */
1529 #elif defined(__X86_64__) || defined(__x86_64__)
1530 return (void*) uc->uc_mcontext.gregs[16]; /* Linux 64 */
1531 #elif defined(__ia64__) /* Linux IA64 */
1532 return (void*) uc->uc_mcontext.sc_ip;
1533 #else
1534 return NULL;
1535 #endif
1536 }
1537
1538 void segvHandler(int sig, siginfo_t *info, void *secret) {
1539 void *trace[100];
1540 char **messages = NULL;
1541 int i, trace_size = 0;
1542 ucontext_t *uc = (ucontext_t*) secret;
1543 sds infostring;
1544 struct sigaction act;
1545 REDIS_NOTUSED(info);
1546
1547 redisLog(REDIS_WARNING,
1548 "======= Ooops! Redis %s got signal: -%d- =======", REDIS_VERSION, sig);
1549 infostring = genRedisInfoString();
1550 redisLog(REDIS_WARNING, "%s",infostring);
1551 /* It's not safe to sdsfree() the returned string under memory
1552 * corruption conditions. Let it leak as we are going to abort */
1553
1554 trace_size = backtrace(trace, 100);
1555 /* overwrite sigaction with caller's address */
1556 if (getMcontextEip(uc) != NULL) {
1557 trace[1] = getMcontextEip(uc);
1558 }
1559 messages = backtrace_symbols(trace, trace_size);
1560
1561 for (i=1; i<trace_size; ++i)
1562 redisLog(REDIS_WARNING,"%s", messages[i]);
1563
1564 /* free(messages); Don't call free() with possibly corrupted memory. */
1565 if (server.daemonize) unlink(server.pidfile);
1566
1567 /* Make sure we exit with the right signal at the end. So for instance
1568 * the core will be dumped if enabled. */
1569 sigemptyset (&act.sa_mask);
1570 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1571 * is used. Otherwise, sa_handler is used */
1572 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1573 act.sa_handler = SIG_DFL;
1574 sigaction (sig, &act, NULL);
1575 kill(getpid(),sig);
1576 }
1577
1578 void sigtermHandler(int sig) {
1579 REDIS_NOTUSED(sig);
1580
1581 redisLog(REDIS_WARNING,"SIGTERM received, scheduling shutting down...");
1582 server.shutdown_asap = 1;
1583 }
1584
1585 void setupSigSegvAction(void) {
1586 struct sigaction act;
1587
1588 sigemptyset (&act.sa_mask);
1589 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1590 * is used. Otherwise, sa_handler is used */
1591 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
1592 act.sa_sigaction = segvHandler;
1593 sigaction (SIGSEGV, &act, NULL);
1594 sigaction (SIGBUS, &act, NULL);
1595 sigaction (SIGFPE, &act, NULL);
1596 sigaction (SIGILL, &act, NULL);
1597 sigaction (SIGBUS, &act, NULL);
1598
1599 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1600 act.sa_handler = sigtermHandler;
1601 sigaction (SIGTERM, &act, NULL);
1602 return;
1603 }
1604
1605 #else /* HAVE_BACKTRACE */
1606 void setupSigSegvAction(void) {
1607 }
1608 #endif /* HAVE_BACKTRACE */
1609
1610 /* The End */