]> git.saurik.com Git - redis.git/blob - src/redis.c
9243d55418dc52b263ab476addfee6557fab0c64
[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 redisCommandTable[] = {
72 {"get",getCommand,2,0,NULL,1,1,1,0,0},
73 {"set",setCommand,3,REDIS_CMD_DENYOOM,noPreloadGetKeys,1,1,1,0,0},
74 {"setnx",setnxCommand,3,REDIS_CMD_DENYOOM,noPreloadGetKeys,1,1,1,0,0},
75 {"setex",setexCommand,4,REDIS_CMD_DENYOOM,noPreloadGetKeys,2,2,1,0,0},
76 {"append",appendCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
77 {"strlen",strlenCommand,2,0,NULL,1,1,1,0,0},
78 {"del",delCommand,-2,0,noPreloadGetKeys,1,-1,1,0,0},
79 {"exists",existsCommand,2,0,NULL,1,1,1,0,0},
80 {"setbit",setbitCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
81 {"getbit",getbitCommand,3,0,NULL,1,1,1,0,0},
82 {"setrange",setrangeCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
83 {"getrange",getrangeCommand,4,0,NULL,1,1,1,0,0},
84 {"substr",getrangeCommand,4,0,NULL,1,1,1,0,0},
85 {"incr",incrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
86 {"decr",decrCommand,2,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
87 {"mget",mgetCommand,-2,0,NULL,1,-1,1,0,0},
88 {"rpush",rpushCommand,-3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
89 {"lpush",lpushCommand,-3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
90 {"rpushx",rpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
91 {"lpushx",lpushxCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
92 {"linsert",linsertCommand,5,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
93 {"rpop",rpopCommand,2,0,NULL,1,1,1,0,0},
94 {"lpop",lpopCommand,2,0,NULL,1,1,1,0,0},
95 {"brpop",brpopCommand,-3,0,NULL,1,1,1,0,0},
96 {"brpoplpush",brpoplpushCommand,4,REDIS_CMD_DENYOOM,NULL,1,2,1,0,0},
97 {"blpop",blpopCommand,-3,0,NULL,1,-2,1,0,0},
98 {"llen",llenCommand,2,0,NULL,1,1,1,0,0},
99 {"lindex",lindexCommand,3,0,NULL,1,1,1,0,0},
100 {"lset",lsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
101 {"lrange",lrangeCommand,4,0,NULL,1,1,1,0,0},
102 {"ltrim",ltrimCommand,4,0,NULL,1,1,1,0,0},
103 {"lrem",lremCommand,4,0,NULL,1,1,1,0,0},
104 {"rpoplpush",rpoplpushCommand,3,REDIS_CMD_DENYOOM,NULL,1,2,1,0,0},
105 {"sadd",saddCommand,-3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
106 {"srem",sremCommand,-3,0,NULL,1,1,1,0,0},
107 {"smove",smoveCommand,4,0,NULL,1,2,1,0,0},
108 {"sismember",sismemberCommand,3,0,NULL,1,1,1,0,0},
109 {"scard",scardCommand,2,0,NULL,1,1,1,0,0},
110 {"spop",spopCommand,2,0,NULL,1,1,1,0,0},
111 {"srandmember",srandmemberCommand,2,0,NULL,1,1,1,0,0},
112 {"sinter",sinterCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1,0,0},
113 {"sinterstore",sinterstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1,0,0},
114 {"sunion",sunionCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1,0,0},
115 {"sunionstore",sunionstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1,0,0},
116 {"sdiff",sdiffCommand,-2,REDIS_CMD_DENYOOM,NULL,1,-1,1,0,0},
117 {"sdiffstore",sdiffstoreCommand,-3,REDIS_CMD_DENYOOM,NULL,2,-1,1,0,0},
118 {"smembers",sinterCommand,2,0,NULL,1,1,1,0,0},
119 {"zadd",zaddCommand,-4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
120 {"zincrby",zincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
121 {"zrem",zremCommand,-3,0,NULL,1,1,1,0,0},
122 {"zremrangebyscore",zremrangebyscoreCommand,4,0,NULL,1,1,1,0,0},
123 {"zremrangebyrank",zremrangebyrankCommand,4,0,NULL,1,1,1,0,0},
124 {"zunionstore",zunionstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterGetKeys,0,0,0,0,0},
125 {"zinterstore",zinterstoreCommand,-4,REDIS_CMD_DENYOOM,zunionInterGetKeys,0,0,0,0,0},
126 {"zrange",zrangeCommand,-4,0,NULL,1,1,1,0,0},
127 {"zrangebyscore",zrangebyscoreCommand,-4,0,NULL,1,1,1,0,0},
128 {"zrevrangebyscore",zrevrangebyscoreCommand,-4,0,NULL,1,1,1,0,0},
129 {"zcount",zcountCommand,4,0,NULL,1,1,1,0,0},
130 {"zrevrange",zrevrangeCommand,-4,0,NULL,1,1,1,0,0},
131 {"zcard",zcardCommand,2,0,NULL,1,1,1,0,0},
132 {"zscore",zscoreCommand,3,0,NULL,1,1,1,0,0},
133 {"zrank",zrankCommand,3,0,NULL,1,1,1,0,0},
134 {"zrevrank",zrevrankCommand,3,0,NULL,1,1,1,0,0},
135 {"hset",hsetCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
136 {"hsetnx",hsetnxCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
137 {"hget",hgetCommand,3,0,NULL,1,1,1,0,0},
138 {"hmset",hmsetCommand,-4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
139 {"hmget",hmgetCommand,-3,0,NULL,1,1,1,0,0},
140 {"hincrby",hincrbyCommand,4,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
141 {"hdel",hdelCommand,-3,0,NULL,1,1,1,0,0},
142 {"hlen",hlenCommand,2,0,NULL,1,1,1,0,0},
143 {"hkeys",hkeysCommand,2,0,NULL,1,1,1,0,0},
144 {"hvals",hvalsCommand,2,0,NULL,1,1,1,0,0},
145 {"hgetall",hgetallCommand,2,0,NULL,1,1,1,0,0},
146 {"hexists",hexistsCommand,3,0,NULL,1,1,1,0,0},
147 {"incrby",incrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
148 {"decrby",decrbyCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
149 {"getset",getsetCommand,3,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
150 {"mset",msetCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2,0,0},
151 {"msetnx",msetnxCommand,-3,REDIS_CMD_DENYOOM,NULL,1,-1,2,0,0},
152 {"randomkey",randomkeyCommand,1,0,NULL,0,0,0,0,0},
153 {"select",selectCommand,2,0,NULL,0,0,0,0,0},
154 {"move",moveCommand,3,0,NULL,1,1,1,0,0},
155 {"rename",renameCommand,3,0,renameGetKeys,1,2,1,0,0},
156 {"renamenx",renamenxCommand,3,0,renameGetKeys,1,2,1,0,0},
157 {"expire",expireCommand,3,0,NULL,1,1,1,0,0},
158 {"expireat",expireatCommand,3,0,NULL,1,1,1,0,0},
159 {"keys",keysCommand,2,0,NULL,0,0,0,0,0},
160 {"dbsize",dbsizeCommand,1,0,NULL,0,0,0,0,0},
161 {"auth",authCommand,2,0,NULL,0,0,0,0,0},
162 {"ping",pingCommand,1,0,NULL,0,0,0,0,0},
163 {"echo",echoCommand,2,0,NULL,0,0,0,0,0},
164 {"save",saveCommand,1,0,NULL,0,0,0,0,0},
165 {"bgsave",bgsaveCommand,1,0,NULL,0,0,0,0,0},
166 {"bgrewriteaof",bgrewriteaofCommand,1,0,NULL,0,0,0,0,0},
167 {"shutdown",shutdownCommand,1,0,NULL,0,0,0,0,0},
168 {"lastsave",lastsaveCommand,1,0,NULL,0,0,0,0,0},
169 {"type",typeCommand,2,0,NULL,1,1,1,0,0},
170 {"multi",multiCommand,1,0,NULL,0,0,0,0,0},
171 {"exec",execCommand,1,REDIS_CMD_DENYOOM,NULL,0,0,0,0,0},
172 {"discard",discardCommand,1,0,NULL,0,0,0,0,0},
173 {"sync",syncCommand,1,0,NULL,0,0,0,0,0},
174 {"flushdb",flushdbCommand,1,0,NULL,0,0,0,0,0},
175 {"flushall",flushallCommand,1,0,NULL,0,0,0,0,0},
176 {"sort",sortCommand,-2,REDIS_CMD_DENYOOM,NULL,1,1,1,0,0},
177 {"info",infoCommand,-1,0,NULL,0,0,0,0,0},
178 {"monitor",monitorCommand,1,0,NULL,0,0,0,0,0},
179 {"ttl",ttlCommand,2,0,NULL,1,1,1,0,0},
180 {"persist",persistCommand,2,0,NULL,1,1,1,0,0},
181 {"slaveof",slaveofCommand,3,0,NULL,0,0,0,0,0},
182 {"debug",debugCommand,-2,0,NULL,0,0,0,0,0},
183 {"config",configCommand,-2,0,NULL,0,0,0,0,0},
184 {"subscribe",subscribeCommand,-2,0,NULL,0,0,0,0,0},
185 {"unsubscribe",unsubscribeCommand,-1,0,NULL,0,0,0,0,0},
186 {"psubscribe",psubscribeCommand,-2,0,NULL,0,0,0,0,0},
187 {"punsubscribe",punsubscribeCommand,-1,0,NULL,0,0,0,0,0},
188 {"publish",publishCommand,3,REDIS_CMD_FORCE_REPLICATION,NULL,0,0,0,0,0},
189 {"watch",watchCommand,-2,0,noPreloadGetKeys,1,-1,1,0,0},
190 {"unwatch",unwatchCommand,1,0,NULL,0,0,0,0,0},
191 {"cluster",clusterCommand,-2,0,NULL,0,0,0,0,0},
192 {"restore",restoreCommand,4,0,NULL,0,0,0,0,0},
193 {"migrate",migrateCommand,6,0,NULL,0,0,0,0,0},
194 {"dump",dumpCommand,2,0,NULL,0,0,0,0,0},
195 {"object",objectCommand,-2,0,NULL,0,0,0,0,0},
196 {"client",clientCommand,-2,0,NULL,0,0,0,0,0}
197 };
198
199 /*============================ Utility functions ============================ */
200
201 /* Low level logging. To use only for very big messages, otherwise
202 * redisLog() is to prefer. */
203 void redisLogRaw(int level, const char *msg) {
204 const int syslogLevelMap[] = { LOG_DEBUG, LOG_INFO, LOG_NOTICE, LOG_WARNING };
205 const char *c = ".-*#";
206 time_t now = time(NULL);
207 FILE *fp;
208 char buf[64];
209 int rawmode = (level & REDIS_LOG_RAW);
210
211 level &= 0xff; /* clear flags */
212 if (level < server.verbosity) return;
213
214 fp = (server.logfile == NULL) ? stdout : fopen(server.logfile,"a");
215 if (!fp) return;
216
217 if (rawmode) {
218 fprintf(fp,"%s",msg);
219 } else {
220 strftime(buf,sizeof(buf),"%d %b %H:%M:%S",localtime(&now));
221 fprintf(fp,"[%d] %s %c %s\n",(int)getpid(),buf,c[level],msg);
222 }
223 fflush(fp);
224
225 if (server.logfile) fclose(fp);
226
227 if (server.syslog_enabled) syslog(syslogLevelMap[level], "%s", msg);
228 }
229
230 /* Like redisLogRaw() but with printf-alike support. This is the funciton that
231 * is used across the code. The raw version is only used in order to dump
232 * the INFO output on crash. */
233 void redisLog(int level, const char *fmt, ...) {
234 va_list ap;
235 char msg[REDIS_MAX_LOGMSG_LEN];
236
237 if ((level&0xff) < server.verbosity) return;
238
239 va_start(ap, fmt);
240 vsnprintf(msg, sizeof(msg), fmt, ap);
241 va_end(ap);
242
243 redisLogRaw(level,msg);
244 }
245
246 /* Redis generally does not try to recover from out of memory conditions
247 * when allocating objects or strings, it is not clear if it will be possible
248 * to report this condition to the client since the networking layer itself
249 * is based on heap allocation for send buffers, so we simply abort.
250 * At least the code will be simpler to read... */
251 void oom(const char *msg) {
252 redisLog(REDIS_WARNING, "%s: Out of memory\n",msg);
253 sleep(1);
254 abort();
255 }
256
257 /* Return the UNIX time in microseconds */
258 long long ustime(void) {
259 struct timeval tv;
260 long long ust;
261
262 gettimeofday(&tv, NULL);
263 ust = ((long long)tv.tv_sec)*1000000;
264 ust += tv.tv_usec;
265 return ust;
266 }
267
268 /*====================== Hash table type implementation ==================== */
269
270 /* This is an hash table type that uses the SDS dynamic strings libary as
271 * keys and radis objects as values (objects can hold SDS strings,
272 * lists, sets). */
273
274 void dictVanillaFree(void *privdata, void *val)
275 {
276 DICT_NOTUSED(privdata);
277 zfree(val);
278 }
279
280 void dictListDestructor(void *privdata, void *val)
281 {
282 DICT_NOTUSED(privdata);
283 listRelease((list*)val);
284 }
285
286 int dictSdsKeyCompare(void *privdata, const void *key1,
287 const void *key2)
288 {
289 int l1,l2;
290 DICT_NOTUSED(privdata);
291
292 l1 = sdslen((sds)key1);
293 l2 = sdslen((sds)key2);
294 if (l1 != l2) return 0;
295 return memcmp(key1, key2, l1) == 0;
296 }
297
298 /* A case insensitive version used for the command lookup table. */
299 int dictSdsKeyCaseCompare(void *privdata, const void *key1,
300 const void *key2)
301 {
302 DICT_NOTUSED(privdata);
303
304 return strcasecmp(key1, key2) == 0;
305 }
306
307 void dictRedisObjectDestructor(void *privdata, void *val)
308 {
309 DICT_NOTUSED(privdata);
310
311 if (val == NULL) return; /* Values of swapped out keys as set to NULL */
312 decrRefCount(val);
313 }
314
315 void dictSdsDestructor(void *privdata, void *val)
316 {
317 DICT_NOTUSED(privdata);
318
319 sdsfree(val);
320 }
321
322 int dictObjKeyCompare(void *privdata, const void *key1,
323 const void *key2)
324 {
325 const robj *o1 = key1, *o2 = key2;
326 return dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
327 }
328
329 unsigned int dictObjHash(const void *key) {
330 const robj *o = key;
331 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
332 }
333
334 unsigned int dictSdsHash(const void *key) {
335 return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
336 }
337
338 unsigned int dictSdsCaseHash(const void *key) {
339 return dictGenCaseHashFunction((unsigned char*)key, sdslen((char*)key));
340 }
341
342 int dictEncObjKeyCompare(void *privdata, const void *key1,
343 const void *key2)
344 {
345 robj *o1 = (robj*) key1, *o2 = (robj*) key2;
346 int cmp;
347
348 if (o1->encoding == REDIS_ENCODING_INT &&
349 o2->encoding == REDIS_ENCODING_INT)
350 return o1->ptr == o2->ptr;
351
352 o1 = getDecodedObject(o1);
353 o2 = getDecodedObject(o2);
354 cmp = dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
355 decrRefCount(o1);
356 decrRefCount(o2);
357 return cmp;
358 }
359
360 unsigned int dictEncObjHash(const void *key) {
361 robj *o = (robj*) key;
362
363 if (o->encoding == REDIS_ENCODING_RAW) {
364 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
365 } else {
366 if (o->encoding == REDIS_ENCODING_INT) {
367 char buf[32];
368 int len;
369
370 len = ll2string(buf,32,(long)o->ptr);
371 return dictGenHashFunction((unsigned char*)buf, len);
372 } else {
373 unsigned int hash;
374
375 o = getDecodedObject(o);
376 hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
377 decrRefCount(o);
378 return hash;
379 }
380 }
381 }
382
383 /* Sets type and diskstore negative caching hash table */
384 dictType setDictType = {
385 dictEncObjHash, /* hash function */
386 NULL, /* key dup */
387 NULL, /* val dup */
388 dictEncObjKeyCompare, /* key compare */
389 dictRedisObjectDestructor, /* key destructor */
390 NULL /* val destructor */
391 };
392
393 /* Sorted sets hash (note: a skiplist is used in addition to the hash table) */
394 dictType zsetDictType = {
395 dictEncObjHash, /* hash function */
396 NULL, /* key dup */
397 NULL, /* val dup */
398 dictEncObjKeyCompare, /* key compare */
399 dictRedisObjectDestructor, /* key destructor */
400 NULL /* val destructor */
401 };
402
403 /* Db->dict, keys are sds strings, vals are Redis objects. */
404 dictType dbDictType = {
405 dictSdsHash, /* hash function */
406 NULL, /* key dup */
407 NULL, /* val dup */
408 dictSdsKeyCompare, /* key compare */
409 dictSdsDestructor, /* key destructor */
410 dictRedisObjectDestructor /* val destructor */
411 };
412
413 /* Db->expires */
414 dictType keyptrDictType = {
415 dictSdsHash, /* hash function */
416 NULL, /* key dup */
417 NULL, /* val dup */
418 dictSdsKeyCompare, /* key compare */
419 NULL, /* key destructor */
420 NULL /* val destructor */
421 };
422
423 /* Command table. sds string -> command struct pointer. */
424 dictType commandTableDictType = {
425 dictSdsCaseHash, /* hash function */
426 NULL, /* key dup */
427 NULL, /* val dup */
428 dictSdsKeyCaseCompare, /* key compare */
429 dictSdsDestructor, /* key destructor */
430 NULL /* val destructor */
431 };
432
433 /* Hash type hash table (note that small hashes are represented with zimpaps) */
434 dictType hashDictType = {
435 dictEncObjHash, /* hash function */
436 NULL, /* key dup */
437 NULL, /* val dup */
438 dictEncObjKeyCompare, /* key compare */
439 dictRedisObjectDestructor, /* key destructor */
440 dictRedisObjectDestructor /* val destructor */
441 };
442
443 /* Keylist hash table type has unencoded redis objects as keys and
444 * lists as values. It's used for blocking operations (BLPOP) and to
445 * map swapped keys to a list of clients waiting for this keys to be loaded. */
446 dictType keylistDictType = {
447 dictObjHash, /* hash function */
448 NULL, /* key dup */
449 NULL, /* val dup */
450 dictObjKeyCompare, /* key compare */
451 dictRedisObjectDestructor, /* key destructor */
452 dictListDestructor /* val destructor */
453 };
454
455 /* Cluster nodes hash table, mapping nodes addresses 1.2.3.4:6379 to
456 * clusterNode structures. */
457 dictType clusterNodesDictType = {
458 dictSdsHash, /* hash function */
459 NULL, /* key dup */
460 NULL, /* val dup */
461 dictSdsKeyCompare, /* key compare */
462 dictSdsDestructor, /* key destructor */
463 NULL /* val destructor */
464 };
465
466 int htNeedsResize(dict *dict) {
467 long long size, used;
468
469 size = dictSlots(dict);
470 used = dictSize(dict);
471 return (size && used && size > DICT_HT_INITIAL_SIZE &&
472 (used*100/size < REDIS_HT_MINFILL));
473 }
474
475 /* If the percentage of used slots in the HT reaches REDIS_HT_MINFILL
476 * we resize the hash table to save memory */
477 void tryResizeHashTables(void) {
478 int j;
479
480 for (j = 0; j < server.dbnum; j++) {
481 if (htNeedsResize(server.db[j].dict))
482 dictResize(server.db[j].dict);
483 if (htNeedsResize(server.db[j].expires))
484 dictResize(server.db[j].expires);
485 }
486 }
487
488 /* Our hash table implementation performs rehashing incrementally while
489 * we write/read from the hash table. Still if the server is idle, the hash
490 * table will use two tables for a long time. So we try to use 1 millisecond
491 * of CPU time at every serverCron() loop in order to rehash some key. */
492 void incrementallyRehash(void) {
493 int j;
494
495 for (j = 0; j < server.dbnum; j++) {
496 if (dictIsRehashing(server.db[j].dict)) {
497 dictRehashMilliseconds(server.db[j].dict,1);
498 break; /* already used our millisecond for this loop... */
499 }
500 }
501 }
502
503 /* This function is called once a background process of some kind terminates,
504 * as we want to avoid resizing the hash tables when there is a child in order
505 * to play well with copy-on-write (otherwise when a resize happens lots of
506 * memory pages are copied). The goal of this function is to update the ability
507 * for dict.c to resize the hash tables accordingly to the fact we have o not
508 * running childs. */
509 void updateDictResizePolicy(void) {
510 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1)
511 dictEnableResize();
512 else
513 dictDisableResize();
514 }
515
516 /* ======================= Cron: called every 100 ms ======================== */
517
518 /* Try to expire a few timed out keys. The algorithm used is adaptive and
519 * will use few CPU cycles if there are few expiring keys, otherwise
520 * it will get more aggressive to avoid that too much memory is used by
521 * keys that can be removed from the keyspace. */
522 void activeExpireCycle(void) {
523 int j;
524
525 for (j = 0; j < server.dbnum; j++) {
526 int expired;
527 redisDb *db = server.db+j;
528
529 /* Continue to expire if at the end of the cycle more than 25%
530 * of the keys were expired. */
531 do {
532 long num = dictSize(db->expires);
533 time_t now = time(NULL);
534
535 expired = 0;
536 if (num > REDIS_EXPIRELOOKUPS_PER_CRON)
537 num = REDIS_EXPIRELOOKUPS_PER_CRON;
538 while (num--) {
539 dictEntry *de;
540 time_t t;
541
542 if ((de = dictGetRandomKey(db->expires)) == NULL) break;
543 t = (time_t) dictGetEntryVal(de);
544 if (now > t) {
545 sds key = dictGetEntryKey(de);
546 robj *keyobj = createStringObject(key,sdslen(key));
547
548 propagateExpire(db,keyobj);
549 dbDelete(db,keyobj);
550 decrRefCount(keyobj);
551 expired++;
552 server.stat_expiredkeys++;
553 }
554 }
555 } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4);
556 }
557 }
558
559 void updateLRUClock(void) {
560 server.lruclock = (time(NULL)/REDIS_LRU_CLOCK_RESOLUTION) &
561 REDIS_LRU_CLOCK_MAX;
562 }
563
564 int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
565 int j, loops = server.cronloops;
566 REDIS_NOTUSED(eventLoop);
567 REDIS_NOTUSED(id);
568 REDIS_NOTUSED(clientData);
569
570 /* We take a cached value of the unix time in the global state because
571 * with virtual memory and aging there is to store the current time
572 * in objects at every object access, and accuracy is not needed.
573 * To access a global var is faster than calling time(NULL) */
574 server.unixtime = time(NULL);
575 /* We have just 22 bits per object for LRU information.
576 * So we use an (eventually wrapping) LRU clock with 10 seconds resolution.
577 * 2^22 bits with 10 seconds resoluton is more or less 1.5 years.
578 *
579 * Note that even if this will wrap after 1.5 years it's not a problem,
580 * everything will still work but just some object will appear younger
581 * to Redis. But for this to happen a given object should never be touched
582 * for 1.5 years.
583 *
584 * Note that you can change the resolution altering the
585 * REDIS_LRU_CLOCK_RESOLUTION define.
586 */
587 updateLRUClock();
588
589 /* Record the max memory used since the server was started. */
590 if (zmalloc_used_memory() > server.stat_peak_memory)
591 server.stat_peak_memory = zmalloc_used_memory();
592
593 /* We received a SIGTERM, shutting down here in a safe way, as it is
594 * not ok doing so inside the signal handler. */
595 if (server.shutdown_asap) {
596 if (prepareForShutdown() == REDIS_OK) exit(0);
597 redisLog(REDIS_WARNING,"SIGTERM received but errors trying to shut down the server, check the logs for more information");
598 }
599
600 /* Show some info about non-empty databases */
601 for (j = 0; j < server.dbnum; j++) {
602 long long size, used, vkeys;
603
604 size = dictSlots(server.db[j].dict);
605 used = dictSize(server.db[j].dict);
606 vkeys = dictSize(server.db[j].expires);
607 if (!(loops % 50) && (used || vkeys)) {
608 redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
609 /* dictPrintStats(server.dict); */
610 }
611 }
612
613 /* We don't want to resize the hash tables while a bacground saving
614 * is in progress: the saving child is created using fork() that is
615 * implemented with a copy-on-write semantic in most modern systems, so
616 * if we resize the HT while there is the saving child at work actually
617 * a lot of memory movements in the parent will cause a lot of pages
618 * copied. */
619 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1) {
620 if (!(loops % 10)) tryResizeHashTables();
621 if (server.activerehashing) incrementallyRehash();
622 }
623
624 /* Show information about connected clients */
625 if (!(loops % 50)) {
626 redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use",
627 listLength(server.clients)-listLength(server.slaves),
628 listLength(server.slaves),
629 zmalloc_used_memory());
630 }
631
632 /* Close connections of timedout clients */
633 if ((server.maxidletime && !(loops % 100)) || server.bpop_blocked_clients)
634 closeTimedoutClients();
635
636 /* Start a scheduled AOF rewrite if this was requested by the user while
637 * a BGSAVE was in progress. */
638 if (server.bgsavechildpid == -1 && server.bgrewritechildpid == -1 &&
639 server.aofrewrite_scheduled)
640 {
641 rewriteAppendOnlyFileBackground();
642 }
643
644 /* Check if a background saving or AOF rewrite in progress terminated. */
645 if (server.bgsavechildpid != -1 || server.bgrewritechildpid != -1) {
646 int statloc;
647 pid_t pid;
648
649 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
650 int exitcode = WEXITSTATUS(statloc);
651 int bysignal = 0;
652
653 if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
654
655 if (pid == server.bgsavechildpid) {
656 backgroundSaveDoneHandler(exitcode,bysignal);
657 } else {
658 backgroundRewriteDoneHandler(exitcode,bysignal);
659 }
660 updateDictResizePolicy();
661 }
662 } else if (server.bgsavethread != (pthread_t) -1) {
663 if (server.bgsavethread != (pthread_t) -1) {
664 int state;
665
666 pthread_mutex_lock(&server.bgsavethread_mutex);
667 state = server.bgsavethread_state;
668 pthread_mutex_unlock(&server.bgsavethread_mutex);
669
670 if (state == REDIS_BGSAVE_THREAD_DONE_OK ||
671 state == REDIS_BGSAVE_THREAD_DONE_ERR)
672 {
673 backgroundSaveDoneHandler(
674 (state == REDIS_BGSAVE_THREAD_DONE_OK) ? 0 : 1, 0);
675 }
676 }
677 } else if (!server.ds_enabled) {
678 time_t now = time(NULL);
679
680 /* If there is not a background saving/rewrite in progress check if
681 * we have to save/rewrite now */
682 for (j = 0; j < server.saveparamslen; j++) {
683 struct saveparam *sp = server.saveparams+j;
684
685 if (server.dirty >= sp->changes &&
686 now-server.lastsave > sp->seconds) {
687 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
688 sp->changes, sp->seconds);
689 rdbSaveBackground(server.dbfilename);
690 break;
691 }
692 }
693
694 /* Trigger an AOF rewrite if needed */
695 if (server.auto_aofrewrite_perc &&
696 server.appendonly_current_size > server.auto_aofrewrite_min_size)
697 {
698 int growth = (server.appendonly_current_size*100/
699 server.auto_aofrewrite_base_size);
700 if (growth >= server.auto_aofrewrite_perc) {
701 redisLog(REDIS_NOTICE,"Starting automatic rewriting of AOF on %d growth",growth);
702 rewriteAppendOnlyFileBackground();
703 }
704 }
705 }
706
707 /* Expire a few keys per cycle, only if this is a master.
708 * On slaves we wait for DEL operations synthesized by the master
709 * in order to guarantee a strict consistency. */
710 if (server.masterhost == NULL) activeExpireCycle();
711
712 /* Remove a few cached objects from memory if we are over the
713 * configured memory limit */
714 if (server.ds_enabled) cacheCron();
715
716 /* Replication cron function -- used to reconnect to master and
717 * to detect transfer failures. */
718 if (!(loops % 10)) replicationCron();
719
720 /* Run other sub-systems specific cron jobs */
721 if (server.cluster_enabled && !(loops % 10)) clusterCron();
722
723 server.cronloops++;
724 return 100;
725 }
726
727 /* This function gets called every time Redis is entering the
728 * main loop of the event driven library, that is, before to sleep
729 * for ready file descriptors. */
730 void beforeSleep(struct aeEventLoop *eventLoop) {
731 REDIS_NOTUSED(eventLoop);
732 listNode *ln;
733 redisClient *c;
734
735 /* Awake clients that got all the on disk keys they requested */
736 if (server.ds_enabled && listLength(server.io_ready_clients)) {
737 listIter li;
738
739 listRewind(server.io_ready_clients,&li);
740 while((ln = listNext(&li))) {
741 c = ln->value;
742 struct redisCommand *cmd;
743
744 /* Resume the client. */
745 listDelNode(server.io_ready_clients,ln);
746 c->flags &= (~REDIS_IO_WAIT);
747 server.cache_blocked_clients--;
748 aeCreateFileEvent(server.el, c->fd, AE_READABLE,
749 readQueryFromClient, c);
750 cmd = lookupCommand(c->argv[0]->ptr);
751 redisAssert(cmd != NULL);
752 call(c,cmd);
753 resetClient(c);
754 /* There may be more data to process in the input buffer. */
755 if (c->querybuf && sdslen(c->querybuf) > 0)
756 processInputBuffer(c);
757 }
758 }
759
760 /* Try to process pending commands for clients that were just unblocked. */
761 while (listLength(server.unblocked_clients)) {
762 ln = listFirst(server.unblocked_clients);
763 redisAssert(ln != NULL);
764 c = ln->value;
765 listDelNode(server.unblocked_clients,ln);
766 c->flags &= ~REDIS_UNBLOCKED;
767
768 /* Process remaining data in the input buffer. */
769 if (c->querybuf && sdslen(c->querybuf) > 0)
770 processInputBuffer(c);
771 }
772
773 /* Write the AOF buffer on disk */
774 flushAppendOnlyFile();
775 }
776
777 /* =========================== Server initialization ======================== */
778
779 void createSharedObjects(void) {
780 int j;
781
782 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
783 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
784 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
785 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
786 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
787 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
788 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
789 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
790 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
791 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
792 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
793 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
794 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
795 "-ERR Operation against a key holding the wrong kind of value\r\n"));
796 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
797 "-ERR no such key\r\n"));
798 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
799 "-ERR syntax error\r\n"));
800 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
801 "-ERR source and destination objects are the same\r\n"));
802 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
803 "-ERR index out of range\r\n"));
804 shared.loadingerr = createObject(REDIS_STRING,sdsnew(
805 "-LOADING Redis is loading the dataset in memory\r\n"));
806 shared.space = createObject(REDIS_STRING,sdsnew(" "));
807 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
808 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
809 shared.select0 = createStringObject("select 0\r\n",10);
810 shared.select1 = createStringObject("select 1\r\n",10);
811 shared.select2 = createStringObject("select 2\r\n",10);
812 shared.select3 = createStringObject("select 3\r\n",10);
813 shared.select4 = createStringObject("select 4\r\n",10);
814 shared.select5 = createStringObject("select 5\r\n",10);
815 shared.select6 = createStringObject("select 6\r\n",10);
816 shared.select7 = createStringObject("select 7\r\n",10);
817 shared.select8 = createStringObject("select 8\r\n",10);
818 shared.select9 = createStringObject("select 9\r\n",10);
819 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
820 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
821 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
822 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
823 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
824 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
825 shared.mbulk3 = createStringObject("*3\r\n",4);
826 shared.mbulk4 = createStringObject("*4\r\n",4);
827 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
828 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
829 shared.integers[j]->encoding = REDIS_ENCODING_INT;
830 }
831 }
832
833 void initServerConfig() {
834 server.port = REDIS_SERVERPORT;
835 server.bindaddr = NULL;
836 server.unixsocket = NULL;
837 server.ipfd = -1;
838 server.sofd = -1;
839 server.dbnum = REDIS_DEFAULT_DBNUM;
840 server.verbosity = REDIS_VERBOSE;
841 server.maxidletime = REDIS_MAXIDLETIME;
842 server.saveparams = NULL;
843 server.loading = 0;
844 server.logfile = NULL; /* NULL = log on standard output */
845 server.syslog_enabled = 0;
846 server.syslog_ident = zstrdup("redis");
847 server.syslog_facility = LOG_LOCAL0;
848 server.daemonize = 0;
849 server.appendonly = 0;
850 server.appendfsync = APPENDFSYNC_EVERYSEC;
851 server.no_appendfsync_on_rewrite = 0;
852 server.auto_aofrewrite_perc = REDIS_AUTO_AOFREWRITE_PERC;
853 server.auto_aofrewrite_min_size = REDIS_AUTO_AOFREWRITE_MIN_SIZE;
854 server.auto_aofrewrite_base_size = 0;
855 server.aofrewrite_scheduled = 0;
856 server.lastfsync = time(NULL);
857 server.appendfd = -1;
858 server.appendseldb = -1; /* Make sure the first time will not match */
859 server.pidfile = zstrdup("/var/run/redis.pid");
860 server.dbfilename = zstrdup("dump.rdb");
861 server.appendfilename = zstrdup("appendonly.aof");
862 server.requirepass = NULL;
863 server.rdbcompression = 1;
864 server.activerehashing = 1;
865 server.maxclients = 0;
866 server.bpop_blocked_clients = 0;
867 server.maxmemory = 0;
868 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
869 server.maxmemory_samples = 3;
870 server.ds_enabled = 0;
871 server.ds_path = sdsnew("/tmp/redis.ds");
872 server.cache_max_memory = 64LL*1024*1024; /* 64 MB of RAM */
873 server.cache_blocked_clients = 0;
874 server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES;
875 server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE;
876 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
877 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
878 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
879 server.zset_max_ziplist_entries = REDIS_ZSET_MAX_ZIPLIST_ENTRIES;
880 server.zset_max_ziplist_value = REDIS_ZSET_MAX_ZIPLIST_VALUE;
881 server.shutdown_asap = 0;
882 server.cache_flush_delay = 0;
883 server.cluster_enabled = 0;
884 server.cluster.configfile = zstrdup("nodes.conf");
885
886 updateLRUClock();
887 resetServerSaveParams();
888
889 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
890 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
891 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
892 /* Replication related */
893 server.isslave = 0;
894 server.masterauth = NULL;
895 server.masterhost = NULL;
896 server.masterport = 6379;
897 server.master = NULL;
898 server.replstate = REDIS_REPL_NONE;
899 server.repl_syncio_timeout = REDIS_REPL_SYNCIO_TIMEOUT;
900 server.repl_serve_stale_data = 1;
901
902 /* Double constants initialization */
903 R_Zero = 0.0;
904 R_PosInf = 1.0/R_Zero;
905 R_NegInf = -1.0/R_Zero;
906 R_Nan = R_Zero/R_Zero;
907
908 /* Command table -- we intiialize it here as it is part of the
909 * initial configuration, since command names may be changed via
910 * redis.conf using the rename-command directive. */
911 server.commands = dictCreate(&commandTableDictType,NULL);
912 populateCommandTable();
913 server.delCommand = lookupCommandByCString("del");
914 server.multiCommand = lookupCommandByCString("multi");
915 }
916
917 void initServer() {
918 int j;
919
920 signal(SIGHUP, SIG_IGN);
921 signal(SIGPIPE, SIG_IGN);
922 setupSignalHandlers();
923
924 if (server.syslog_enabled) {
925 openlog(server.syslog_ident, LOG_PID | LOG_NDELAY | LOG_NOWAIT,
926 server.syslog_facility);
927 }
928
929 server.mainthread = pthread_self();
930 server.clients = listCreate();
931 server.slaves = listCreate();
932 server.monitors = listCreate();
933 server.unblocked_clients = listCreate();
934 server.cache_io_queue = listCreate();
935
936 createSharedObjects();
937 server.el = aeCreateEventLoop();
938 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
939
940 if (server.port != 0) {
941 server.ipfd = anetTcpServer(server.neterr,server.port,server.bindaddr);
942 if (server.ipfd == ANET_ERR) {
943 redisLog(REDIS_WARNING, "Opening port: %s", server.neterr);
944 exit(1);
945 }
946 }
947 if (server.unixsocket != NULL) {
948 unlink(server.unixsocket); /* don't care if this fails */
949 server.sofd = anetUnixServer(server.neterr,server.unixsocket);
950 if (server.sofd == ANET_ERR) {
951 redisLog(REDIS_WARNING, "Opening socket: %s", server.neterr);
952 exit(1);
953 }
954 }
955 if (server.ipfd < 0 && server.sofd < 0) {
956 redisLog(REDIS_WARNING, "Configured to not listen anywhere, exiting.");
957 exit(1);
958 }
959 for (j = 0; j < server.dbnum; j++) {
960 server.db[j].dict = dictCreate(&dbDictType,NULL);
961 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
962 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
963 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
964 if (server.ds_enabled) {
965 server.db[j].io_keys = dictCreate(&keylistDictType,NULL);
966 server.db[j].io_negcache = dictCreate(&setDictType,NULL);
967 server.db[j].io_queued = dictCreate(&setDictType,NULL);
968 }
969 server.db[j].id = j;
970 }
971 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
972 server.pubsub_patterns = listCreate();
973 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
974 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
975 server.cronloops = 0;
976 server.bgsavechildpid = -1;
977 server.bgrewritechildpid = -1;
978 server.bgsavethread_state = REDIS_BGSAVE_THREAD_UNACTIVE;
979 server.bgsavethread = (pthread_t) -1;
980 server.bgrewritebuf = sdsempty();
981 server.aofbuf = sdsempty();
982 server.lastsave = time(NULL);
983 server.dirty = 0;
984 server.stat_numcommands = 0;
985 server.stat_numconnections = 0;
986 server.stat_expiredkeys = 0;
987 server.stat_evictedkeys = 0;
988 server.stat_starttime = time(NULL);
989 server.stat_keyspace_misses = 0;
990 server.stat_keyspace_hits = 0;
991 server.stat_peak_memory = 0;
992 server.stat_fork_time = 0;
993 server.unixtime = time(NULL);
994 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
995 if (server.ipfd > 0 && aeCreateFileEvent(server.el,server.ipfd,AE_READABLE,
996 acceptTcpHandler,NULL) == AE_ERR) oom("creating file event");
997 if (server.sofd > 0 && aeCreateFileEvent(server.el,server.sofd,AE_READABLE,
998 acceptUnixHandler,NULL) == AE_ERR) oom("creating file event");
999
1000 if (server.appendonly) {
1001 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
1002 if (server.appendfd == -1) {
1003 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
1004 strerror(errno));
1005 exit(1);
1006 }
1007 }
1008
1009 if (server.ds_enabled) dsInit();
1010 if (server.cluster_enabled) clusterInit();
1011 srand(time(NULL)^getpid());
1012 }
1013
1014 /* Populates the Redis Command Table starting from the hard coded list
1015 * we have on top of redis.c file. */
1016 void populateCommandTable(void) {
1017 int j;
1018 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1019
1020 for (j = 0; j < numcommands; j++) {
1021 struct redisCommand *c = redisCommandTable+j;
1022 int retval;
1023
1024 retval = dictAdd(server.commands, sdsnew(c->name), c);
1025 assert(retval == DICT_OK);
1026 }
1027 }
1028
1029 void resetCommandTableStats(void) {
1030 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1031 int j;
1032
1033 for (j = 0; j < numcommands; j++) {
1034 struct redisCommand *c = redisCommandTable+j;
1035
1036 c->microseconds = 0;
1037 c->calls = 0;
1038 }
1039 }
1040
1041 /* ====================== Commands lookup and execution ===================== */
1042
1043 struct redisCommand *lookupCommand(sds name) {
1044 return dictFetchValue(server.commands, name);
1045 }
1046
1047 struct redisCommand *lookupCommandByCString(char *s) {
1048 struct redisCommand *cmd;
1049 sds name = sdsnew(s);
1050
1051 cmd = dictFetchValue(server.commands, name);
1052 sdsfree(name);
1053 return cmd;
1054 }
1055
1056 /* Call() is the core of Redis execution of a command */
1057 void call(redisClient *c, struct redisCommand *cmd) {
1058 long long dirty, start = ustime();
1059
1060 dirty = server.dirty;
1061 cmd->proc(c);
1062 dirty = server.dirty-dirty;
1063 cmd->microseconds += ustime()-start;
1064 cmd->calls++;
1065
1066 if (server.appendonly && dirty)
1067 feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc);
1068 if ((dirty || cmd->flags & REDIS_CMD_FORCE_REPLICATION) &&
1069 listLength(server.slaves))
1070 replicationFeedSlaves(server.slaves,c->db->id,c->argv,c->argc);
1071 if (listLength(server.monitors))
1072 replicationFeedMonitors(server.monitors,c->db->id,c->argv,c->argc);
1073 server.stat_numcommands++;
1074 }
1075
1076 /* If this function gets called we already read a whole
1077 * command, argments are in the client argv/argc fields.
1078 * processCommand() execute the command or prepare the
1079 * server for a bulk read from the client.
1080 *
1081 * If 1 is returned the client is still alive and valid and
1082 * and other operations can be performed by the caller. Otherwise
1083 * if 0 is returned the client was destroied (i.e. after QUIT). */
1084 int processCommand(redisClient *c) {
1085 struct redisCommand *cmd;
1086
1087 /* The QUIT command is handled separately. Normal command procs will
1088 * go through checking for replication and QUIT will cause trouble
1089 * when FORCE_REPLICATION is enabled and would be implemented in
1090 * a regular command proc. */
1091 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
1092 addReply(c,shared.ok);
1093 c->flags |= REDIS_CLOSE_AFTER_REPLY;
1094 return REDIS_ERR;
1095 }
1096
1097 /* Now lookup the command and check ASAP about trivial error conditions
1098 * such wrong arity, bad command name and so forth. */
1099 cmd = lookupCommand(c->argv[0]->ptr);
1100 if (!cmd) {
1101 addReplyErrorFormat(c,"unknown command '%s'",
1102 (char*)c->argv[0]->ptr);
1103 return REDIS_OK;
1104 } else if ((cmd->arity > 0 && cmd->arity != c->argc) ||
1105 (c->argc < -cmd->arity)) {
1106 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
1107 cmd->name);
1108 return REDIS_OK;
1109 }
1110
1111 /* Check if the user is authenticated */
1112 if (server.requirepass && !c->authenticated && cmd->proc != authCommand) {
1113 addReplyError(c,"operation not permitted");
1114 return REDIS_OK;
1115 }
1116
1117 /* If cluster is enabled, redirect here */
1118 if (server.cluster_enabled &&
1119 !(cmd->getkeys_proc == NULL && cmd->firstkey == 0)) {
1120 int hashslot;
1121
1122 if (server.cluster.state != REDIS_CLUSTER_OK) {
1123 addReplyError(c,"The cluster is down. Check with CLUSTER INFO for more information");
1124 return REDIS_OK;
1125 } else {
1126 int ask;
1127 clusterNode *n = getNodeByQuery(c,cmd,c->argv,c->argc,&hashslot,&ask);
1128 if (n == NULL) {
1129 addReplyError(c,"Multi keys request invalid in cluster");
1130 return REDIS_OK;
1131 } else if (n != server.cluster.myself) {
1132 addReplySds(c,sdscatprintf(sdsempty(),
1133 "-%s %d %s:%d\r\n", ask ? "ASK" : "MOVED",
1134 hashslot,n->ip,n->port));
1135 return REDIS_OK;
1136 }
1137 }
1138 }
1139
1140 /* Handle the maxmemory directive.
1141 *
1142 * First we try to free some memory if possible (if there are volatile
1143 * keys in the dataset). If there are not the only thing we can do
1144 * is returning an error. */
1145 if (server.maxmemory) freeMemoryIfNeeded();
1146 if (server.maxmemory && (cmd->flags & REDIS_CMD_DENYOOM) &&
1147 zmalloc_used_memory() > server.maxmemory)
1148 {
1149 addReplyError(c,"command not allowed when used memory > 'maxmemory'");
1150 return REDIS_OK;
1151 }
1152
1153 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
1154 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
1155 &&
1156 cmd->proc != subscribeCommand && cmd->proc != unsubscribeCommand &&
1157 cmd->proc != psubscribeCommand && cmd->proc != punsubscribeCommand) {
1158 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
1159 return REDIS_OK;
1160 }
1161
1162 /* Only allow INFO and SLAVEOF when slave-serve-stale-data is no and
1163 * we are a slave with a broken link with master. */
1164 if (server.masterhost && server.replstate != REDIS_REPL_CONNECTED &&
1165 server.repl_serve_stale_data == 0 &&
1166 cmd->proc != infoCommand && cmd->proc != slaveofCommand)
1167 {
1168 addReplyError(c,
1169 "link with MASTER is down and slave-serve-stale-data is set to no");
1170 return REDIS_OK;
1171 }
1172
1173 /* Loading DB? Return an error if the command is not INFO */
1174 if (server.loading && cmd->proc != infoCommand) {
1175 addReply(c, shared.loadingerr);
1176 return REDIS_OK;
1177 }
1178
1179 /* Exec the command */
1180 if (c->flags & REDIS_MULTI &&
1181 cmd->proc != execCommand && cmd->proc != discardCommand &&
1182 cmd->proc != multiCommand && cmd->proc != watchCommand)
1183 {
1184 queueMultiCommand(c,cmd);
1185 addReply(c,shared.queued);
1186 } else {
1187 if (server.ds_enabled && blockClientOnSwappedKeys(c,cmd))
1188 return REDIS_ERR;
1189 call(c,cmd);
1190 }
1191 return REDIS_OK;
1192 }
1193
1194 /*================================== Shutdown =============================== */
1195
1196 int prepareForShutdown() {
1197 redisLog(REDIS_WARNING,"User requested shutdown, saving DB...");
1198 /* Kill the saving child if there is a background saving in progress.
1199 We want to avoid race conditions, for instance our saving child may
1200 overwrite the synchronous saving did by SHUTDOWN. */
1201 if (server.bgsavechildpid != -1) {
1202 redisLog(REDIS_WARNING,"There is a live saving child. Killing it!");
1203 kill(server.bgsavechildpid,SIGKILL);
1204 rdbRemoveTempFile(server.bgsavechildpid);
1205 }
1206 if (server.ds_enabled) {
1207 /* FIXME: flush all objects on disk */
1208 } else if (server.appendonly) {
1209 /* Append only file: fsync() the AOF and exit */
1210 aof_fsync(server.appendfd);
1211 } else if (server.saveparamslen > 0) {
1212 /* Snapshotting. Perform a SYNC SAVE and exit */
1213 if (rdbSave(server.dbfilename) != REDIS_OK) {
1214 /* Ooops.. error saving! The best we can do is to continue
1215 * operating. Note that if there was a background saving process,
1216 * in the next cron() Redis will be notified that the background
1217 * saving aborted, handling special stuff like slaves pending for
1218 * synchronization... */
1219 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit");
1220 return REDIS_ERR;
1221 }
1222 } else {
1223 redisLog(REDIS_WARNING,"Not saving DB.");
1224 }
1225 if (server.daemonize) unlink(server.pidfile);
1226 redisLog(REDIS_WARNING,"Server exit now, bye bye...");
1227 return REDIS_OK;
1228 }
1229
1230 /*================================== Commands =============================== */
1231
1232 void authCommand(redisClient *c) {
1233 if (!server.requirepass || !strcmp(c->argv[1]->ptr, server.requirepass)) {
1234 c->authenticated = 1;
1235 addReply(c,shared.ok);
1236 } else {
1237 c->authenticated = 0;
1238 addReplyError(c,"invalid password");
1239 }
1240 }
1241
1242 void pingCommand(redisClient *c) {
1243 addReply(c,shared.pong);
1244 }
1245
1246 void echoCommand(redisClient *c) {
1247 addReplyBulk(c,c->argv[1]);
1248 }
1249
1250 /* Convert an amount of bytes into a human readable string in the form
1251 * of 100B, 2G, 100M, 4K, and so forth. */
1252 void bytesToHuman(char *s, unsigned long long n) {
1253 double d;
1254
1255 if (n < 1024) {
1256 /* Bytes */
1257 sprintf(s,"%lluB",n);
1258 return;
1259 } else if (n < (1024*1024)) {
1260 d = (double)n/(1024);
1261 sprintf(s,"%.2fK",d);
1262 } else if (n < (1024LL*1024*1024)) {
1263 d = (double)n/(1024*1024);
1264 sprintf(s,"%.2fM",d);
1265 } else if (n < (1024LL*1024*1024*1024)) {
1266 d = (double)n/(1024LL*1024*1024);
1267 sprintf(s,"%.2fG",d);
1268 }
1269 }
1270
1271 /* Create the string returned by the INFO command. This is decoupled
1272 * by the INFO command itself as we need to report the same information
1273 * on memory corruption problems. */
1274 sds genRedisInfoString(char *section) {
1275 sds info = sdsempty();
1276 time_t uptime = time(NULL)-server.stat_starttime;
1277 int j, numcommands;
1278 struct rusage self_ru, c_ru;
1279 unsigned long lol, bib;
1280 int allsections = 0, defsections = 0;
1281 int sections = 0;
1282
1283 if (section) {
1284 allsections = strcasecmp(section,"all") == 0;
1285 defsections = strcasecmp(section,"default") == 0;
1286 }
1287
1288 getrusage(RUSAGE_SELF, &self_ru);
1289 getrusage(RUSAGE_CHILDREN, &c_ru);
1290 getClientsMaxBuffers(&lol,&bib);
1291
1292 /* Server */
1293 if (allsections || defsections || !strcasecmp(section,"server")) {
1294 if (sections++) info = sdscat(info,"\r\n");
1295 info = sdscatprintf(info,
1296 "# Server\r\n"
1297 "redis_version:%s\r\n"
1298 "redis_git_sha1:%s\r\n"
1299 "redis_git_dirty:%d\r\n"
1300 "arch_bits:%s\r\n"
1301 "multiplexing_api:%s\r\n"
1302 "process_id:%ld\r\n"
1303 "tcp_port:%d\r\n"
1304 "uptime_in_seconds:%ld\r\n"
1305 "uptime_in_days:%ld\r\n"
1306 "lru_clock:%ld\r\n",
1307 REDIS_VERSION,
1308 redisGitSHA1(),
1309 strtol(redisGitDirty(),NULL,10) > 0,
1310 (sizeof(long) == 8) ? "64" : "32",
1311 aeGetApiName(),
1312 (long) getpid(),
1313 server.port,
1314 uptime,
1315 uptime/(3600*24),
1316 (unsigned long) server.lruclock);
1317 }
1318
1319 /* Clients */
1320 if (allsections || defsections || !strcasecmp(section,"clients")) {
1321 if (sections++) info = sdscat(info,"\r\n");
1322 info = sdscatprintf(info,
1323 "# Clients\r\n"
1324 "connected_clients:%d\r\n"
1325 "client_longest_output_list:%lu\r\n"
1326 "client_biggest_input_buf:%lu\r\n"
1327 "blocked_clients:%d\r\n",
1328 listLength(server.clients)-listLength(server.slaves),
1329 lol, bib,
1330 server.bpop_blocked_clients);
1331 }
1332
1333 /* Memory */
1334 if (allsections || defsections || !strcasecmp(section,"memory")) {
1335 char hmem[64];
1336 char peak_hmem[64];
1337
1338 bytesToHuman(hmem,zmalloc_used_memory());
1339 bytesToHuman(peak_hmem,server.stat_peak_memory);
1340 if (sections++) info = sdscat(info,"\r\n");
1341 info = sdscatprintf(info,
1342 "# Memory\r\n"
1343 "used_memory:%zu\r\n"
1344 "used_memory_human:%s\r\n"
1345 "used_memory_rss:%zu\r\n"
1346 "used_memory_peak:%zu\r\n"
1347 "used_memory_peak_human:%s\r\n"
1348 "mem_fragmentation_ratio:%.2f\r\n"
1349 "mem_allocator:%s\r\n",
1350 zmalloc_used_memory(),
1351 hmem,
1352 zmalloc_get_rss(),
1353 server.stat_peak_memory,
1354 peak_hmem,
1355 zmalloc_get_fragmentation_ratio(),
1356 REDIS_MALLOC
1357 );
1358 }
1359
1360 /* Allocation statistics */
1361 if (allsections || !strcasecmp(section,"allocstats")) {
1362 if (sections++) info = sdscat(info,"\r\n");
1363 info = sdscat(info, "# Allocstats\r\nallocation_stats:");
1364 for (j = 0; j <= ZMALLOC_MAX_ALLOC_STAT; j++) {
1365 size_t count = zmalloc_allocations_for_size(j);
1366 if (count) {
1367 if (info[sdslen(info)-1] != ':') info = sdscatlen(info,",",1);
1368 info = sdscatprintf(info,"%s%d=%zu",
1369 (j == ZMALLOC_MAX_ALLOC_STAT) ? ">=" : "",
1370 j,count);
1371 }
1372 }
1373 info = sdscat(info,"\r\n");
1374 }
1375
1376 /* Persistence */
1377 if (allsections || defsections || !strcasecmp(section,"persistence")) {
1378 if (sections++) info = sdscat(info,"\r\n");
1379 info = sdscatprintf(info,
1380 "# Persistence\r\n"
1381 "loading:%d\r\n"
1382 "aof_enabled:%d\r\n"
1383 "changes_since_last_save:%lld\r\n"
1384 "bgsave_in_progress:%d\r\n"
1385 "last_save_time:%ld\r\n"
1386 "bgrewriteaof_in_progress:%d\r\n",
1387 server.loading,
1388 server.appendonly,
1389 server.dirty,
1390 server.bgsavechildpid != -1 ||
1391 server.bgsavethread != (pthread_t) -1,
1392 server.lastsave,
1393 server.bgrewritechildpid != -1);
1394
1395 if (server.loading) {
1396 double perc;
1397 time_t eta, elapsed;
1398 off_t remaining_bytes = server.loading_total_bytes-
1399 server.loading_loaded_bytes;
1400
1401 perc = ((double)server.loading_loaded_bytes /
1402 server.loading_total_bytes) * 100;
1403
1404 elapsed = time(NULL)-server.loading_start_time;
1405 if (elapsed == 0) {
1406 eta = 1; /* A fake 1 second figure if we don't have
1407 enough info */
1408 } else {
1409 eta = (elapsed*remaining_bytes)/server.loading_loaded_bytes;
1410 }
1411
1412 info = sdscatprintf(info,
1413 "loading_start_time:%ld\r\n"
1414 "loading_total_bytes:%llu\r\n"
1415 "loading_loaded_bytes:%llu\r\n"
1416 "loading_loaded_perc:%.2f\r\n"
1417 "loading_eta_seconds:%ld\r\n"
1418 ,(unsigned long) server.loading_start_time,
1419 (unsigned long long) server.loading_total_bytes,
1420 (unsigned long long) server.loading_loaded_bytes,
1421 perc,
1422 eta
1423 );
1424 }
1425 }
1426
1427 /* Diskstore */
1428 if (allsections || defsections || !strcasecmp(section,"diskstore")) {
1429 if (sections++) info = sdscat(info,"\r\n");
1430 info = sdscatprintf(info,
1431 "# Diskstore\r\n"
1432 "ds_enabled:%d\r\n",
1433 server.ds_enabled != 0);
1434 if (server.ds_enabled) {
1435 lockThreadedIO();
1436 info = sdscatprintf(info,
1437 "cache_max_memory:%llu\r\n"
1438 "cache_blocked_clients:%lu\r\n"
1439 "cache_io_queue_len:%lu\r\n"
1440 "cache_io_jobs_new:%lu\r\n"
1441 "cache_io_jobs_processing:%lu\r\n"
1442 "cache_io_jobs_processed:%lu\r\n"
1443 "cache_io_ready_clients:%lu\r\n"
1444 ,(unsigned long long) server.cache_max_memory,
1445 (unsigned long) server.cache_blocked_clients,
1446 (unsigned long) listLength(server.cache_io_queue),
1447 (unsigned long) listLength(server.io_newjobs),
1448 (unsigned long) listLength(server.io_processing),
1449 (unsigned long) listLength(server.io_processed),
1450 (unsigned long) listLength(server.io_ready_clients)
1451 );
1452 unlockThreadedIO();
1453 }
1454 }
1455
1456 /* Stats */
1457 if (allsections || defsections || !strcasecmp(section,"stats")) {
1458 if (sections++) info = sdscat(info,"\r\n");
1459 info = sdscatprintf(info,
1460 "# Stats\r\n"
1461 "total_connections_received:%lld\r\n"
1462 "total_commands_processed:%lld\r\n"
1463 "expired_keys:%lld\r\n"
1464 "evicted_keys:%lld\r\n"
1465 "keyspace_hits:%lld\r\n"
1466 "keyspace_misses:%lld\r\n"
1467 "pubsub_channels:%ld\r\n"
1468 "pubsub_patterns:%u\r\n"
1469 "latest_fork_usec:%lld\r\n",
1470 server.stat_numconnections,
1471 server.stat_numcommands,
1472 server.stat_expiredkeys,
1473 server.stat_evictedkeys,
1474 server.stat_keyspace_hits,
1475 server.stat_keyspace_misses,
1476 dictSize(server.pubsub_channels),
1477 listLength(server.pubsub_patterns),
1478 server.stat_fork_time);
1479 }
1480
1481 /* Replication */
1482 if (allsections || defsections || !strcasecmp(section,"replication")) {
1483 if (sections++) info = sdscat(info,"\r\n");
1484 info = sdscatprintf(info,
1485 "# Replication\r\n"
1486 "role:%s\r\n",
1487 server.masterhost == NULL ? "master" : "slave");
1488 if (server.masterhost) {
1489 info = sdscatprintf(info,
1490 "master_host:%s\r\n"
1491 "master_port:%d\r\n"
1492 "master_link_status:%s\r\n"
1493 "master_last_io_seconds_ago:%d\r\n"
1494 "master_sync_in_progress:%d\r\n"
1495 ,server.masterhost,
1496 server.masterport,
1497 (server.replstate == REDIS_REPL_CONNECTED) ?
1498 "up" : "down",
1499 server.master ?
1500 ((int)(time(NULL)-server.master->lastinteraction)) : -1,
1501 server.replstate == REDIS_REPL_TRANSFER
1502 );
1503
1504 if (server.replstate == REDIS_REPL_TRANSFER) {
1505 info = sdscatprintf(info,
1506 "master_sync_left_bytes:%ld\r\n"
1507 "master_sync_last_io_seconds_ago:%d\r\n"
1508 ,(long)server.repl_transfer_left,
1509 (int)(time(NULL)-server.repl_transfer_lastio)
1510 );
1511 }
1512 }
1513 info = sdscatprintf(info,
1514 "connected_slaves:%d\r\n",
1515 listLength(server.slaves));
1516 }
1517
1518 /* CPU */
1519 if (allsections || defsections || !strcasecmp(section,"cpu")) {
1520 if (sections++) info = sdscat(info,"\r\n");
1521 info = sdscatprintf(info,
1522 "# CPU\r\n"
1523 "used_cpu_sys:%.2f\r\n"
1524 "used_cpu_user:%.2f\r\n"
1525 "used_cpu_sys_childrens:%.2f\r\n"
1526 "used_cpu_user_childrens:%.2f\r\n",
1527 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
1528 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
1529 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000,
1530 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000);
1531 }
1532
1533 /* cmdtime */
1534 if (allsections || !strcasecmp(section,"commandstats")) {
1535 if (sections++) info = sdscat(info,"\r\n");
1536 info = sdscatprintf(info, "# Commandstats\r\n");
1537 numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1538 for (j = 0; j < numcommands; j++) {
1539 struct redisCommand *c = redisCommandTable+j;
1540
1541 if (!c->calls) continue;
1542 info = sdscatprintf(info,
1543 "cmdstat_%s:calls=%lld,usec=%lld,usec_per_call=%.2f\r\n",
1544 c->name, c->calls, c->microseconds,
1545 (c->calls == 0) ? 0 : ((float)c->microseconds/c->calls));
1546 }
1547 }
1548
1549 /* Clusetr */
1550 if (allsections || defsections || !strcasecmp(section,"cluster")) {
1551 if (sections++) info = sdscat(info,"\r\n");
1552 info = sdscatprintf(info,
1553 "# Cluster\r\n"
1554 "cluster_enabled:%d\r\n",
1555 server.cluster_enabled);
1556 }
1557
1558 /* Key space */
1559 if (allsections || defsections || !strcasecmp(section,"keyspace")) {
1560 if (sections++) info = sdscat(info,"\r\n");
1561 info = sdscatprintf(info, "# Keyspace\r\n");
1562 for (j = 0; j < server.dbnum; j++) {
1563 long long keys, vkeys;
1564
1565 keys = dictSize(server.db[j].dict);
1566 vkeys = dictSize(server.db[j].expires);
1567 if (keys || vkeys) {
1568 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
1569 j, keys, vkeys);
1570 }
1571 }
1572 }
1573 return info;
1574 }
1575
1576 void infoCommand(redisClient *c) {
1577 char *section = c->argc == 2 ? c->argv[1]->ptr : "default";
1578
1579 if (c->argc > 2) {
1580 addReply(c,shared.syntaxerr);
1581 return;
1582 }
1583 sds info = genRedisInfoString(section);
1584 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
1585 (unsigned long)sdslen(info)));
1586 addReplySds(c,info);
1587 addReply(c,shared.crlf);
1588 }
1589
1590 void monitorCommand(redisClient *c) {
1591 /* ignore MONITOR if aleady slave or in monitor mode */
1592 if (c->flags & REDIS_SLAVE) return;
1593
1594 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
1595 c->slaveseldb = 0;
1596 listAddNodeTail(server.monitors,c);
1597 addReply(c,shared.ok);
1598 }
1599
1600 /* ============================ Maxmemory directive ======================== */
1601
1602 /* This function gets called when 'maxmemory' is set on the config file to limit
1603 * the max memory used by the server, and we are out of memory.
1604 * This function will try to, in order:
1605 *
1606 * - Free objects from the free list
1607 * - Try to remove keys with an EXPIRE set
1608 *
1609 * It is not possible to free enough memory to reach used-memory < maxmemory
1610 * the server will start refusing commands that will enlarge even more the
1611 * memory usage.
1612 */
1613 void freeMemoryIfNeeded(void) {
1614 /* Remove keys accordingly to the active policy as long as we are
1615 * over the memory limit. */
1616 if (server.maxmemory_policy == REDIS_MAXMEMORY_NO_EVICTION) return;
1617
1618 while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) {
1619 int j, k, freed = 0;
1620
1621 for (j = 0; j < server.dbnum; j++) {
1622 long bestval = 0; /* just to prevent warning */
1623 sds bestkey = NULL;
1624 struct dictEntry *de;
1625 redisDb *db = server.db+j;
1626 dict *dict;
1627
1628 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1629 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
1630 {
1631 dict = server.db[j].dict;
1632 } else {
1633 dict = server.db[j].expires;
1634 }
1635 if (dictSize(dict) == 0) continue;
1636
1637 /* volatile-random and allkeys-random policy */
1638 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
1639 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
1640 {
1641 de = dictGetRandomKey(dict);
1642 bestkey = dictGetEntryKey(de);
1643 }
1644
1645 /* volatile-lru and allkeys-lru policy */
1646 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1647 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1648 {
1649 for (k = 0; k < server.maxmemory_samples; k++) {
1650 sds thiskey;
1651 long thisval;
1652 robj *o;
1653
1654 de = dictGetRandomKey(dict);
1655 thiskey = dictGetEntryKey(de);
1656 /* When policy is volatile-lru we need an additonal lookup
1657 * to locate the real key, as dict is set to db->expires. */
1658 if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1659 de = dictFind(db->dict, thiskey);
1660 o = dictGetEntryVal(de);
1661 thisval = estimateObjectIdleTime(o);
1662
1663 /* Higher idle time is better candidate for deletion */
1664 if (bestkey == NULL || thisval > bestval) {
1665 bestkey = thiskey;
1666 bestval = thisval;
1667 }
1668 }
1669 }
1670
1671 /* volatile-ttl */
1672 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
1673 for (k = 0; k < server.maxmemory_samples; k++) {
1674 sds thiskey;
1675 long thisval;
1676
1677 de = dictGetRandomKey(dict);
1678 thiskey = dictGetEntryKey(de);
1679 thisval = (long) dictGetEntryVal(de);
1680
1681 /* Expire sooner (minor expire unix timestamp) is better
1682 * candidate for deletion */
1683 if (bestkey == NULL || thisval < bestval) {
1684 bestkey = thiskey;
1685 bestval = thisval;
1686 }
1687 }
1688 }
1689
1690 /* Finally remove the selected key. */
1691 if (bestkey) {
1692 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
1693 propagateExpire(db,keyobj);
1694 dbDelete(db,keyobj);
1695 server.stat_evictedkeys++;
1696 decrRefCount(keyobj);
1697 freed++;
1698 }
1699 }
1700 if (!freed) return; /* nothing to free... */
1701 }
1702 }
1703
1704 /* =================================== Main! ================================ */
1705
1706 #ifdef __linux__
1707 int linuxOvercommitMemoryValue(void) {
1708 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
1709 char buf[64];
1710
1711 if (!fp) return -1;
1712 if (fgets(buf,64,fp) == NULL) {
1713 fclose(fp);
1714 return -1;
1715 }
1716 fclose(fp);
1717
1718 return atoi(buf);
1719 }
1720
1721 void linuxOvercommitMemoryWarning(void) {
1722 if (linuxOvercommitMemoryValue() == 0) {
1723 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.");
1724 }
1725 }
1726 #endif /* __linux__ */
1727
1728 void createPidFile(void) {
1729 /* Try to write the pid file in a best-effort way. */
1730 FILE *fp = fopen(server.pidfile,"w");
1731 if (fp) {
1732 fprintf(fp,"%d\n",(int)getpid());
1733 fclose(fp);
1734 }
1735 }
1736
1737 void daemonize(void) {
1738 int fd;
1739
1740 if (fork() != 0) exit(0); /* parent exits */
1741 setsid(); /* create a new session */
1742
1743 /* Every output goes to /dev/null. If Redis is daemonized but
1744 * the 'logfile' is set to 'stdout' in the configuration file
1745 * it will not log at all. */
1746 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
1747 dup2(fd, STDIN_FILENO);
1748 dup2(fd, STDOUT_FILENO);
1749 dup2(fd, STDERR_FILENO);
1750 if (fd > STDERR_FILENO) close(fd);
1751 }
1752 }
1753
1754 void version() {
1755 printf("Redis server version %s (%s:%d)\n", REDIS_VERSION,
1756 redisGitSHA1(), atoi(redisGitDirty()) > 0);
1757 exit(0);
1758 }
1759
1760 void usage() {
1761 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n");
1762 fprintf(stderr," ./redis-server - (read config from stdin)\n");
1763 exit(1);
1764 }
1765
1766 void redisAsciiArt(void) {
1767 #include "asciilogo.h"
1768 char *buf = zmalloc(1024*16);
1769
1770 snprintf(buf,1024*16,ascii_logo,
1771 REDIS_VERSION,
1772 redisGitSHA1(),
1773 strtol(redisGitDirty(),NULL,10) > 0,
1774 (sizeof(long) == 8) ? "64" : "32",
1775 server.cluster_enabled ? "cluster" : "stand alone",
1776 server.port,
1777 (long) getpid()
1778 );
1779 redisLogRaw(REDIS_NOTICE|REDIS_LOG_RAW,buf);
1780 zfree(buf);
1781 }
1782
1783 int main(int argc, char **argv) {
1784 long long start;
1785
1786 initServerConfig();
1787 if (argc == 2) {
1788 if (strcmp(argv[1], "-v") == 0 ||
1789 strcmp(argv[1], "--version") == 0) version();
1790 if (strcmp(argv[1], "--help") == 0) usage();
1791 resetServerSaveParams();
1792 loadServerConfig(argv[1]);
1793 } else if ((argc > 2)) {
1794 usage();
1795 } else {
1796 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'");
1797 }
1798 if (server.daemonize) daemonize();
1799 initServer();
1800 if (server.daemonize) createPidFile();
1801 redisAsciiArt();
1802 redisLog(REDIS_NOTICE,"Server started, Redis version " REDIS_VERSION);
1803 #ifdef __linux__
1804 linuxOvercommitMemoryWarning();
1805 #endif
1806 start = ustime();
1807 if (server.ds_enabled) {
1808 redisLog(REDIS_NOTICE,"DB not loaded (running with disk back end)");
1809 } else if (server.appendonly) {
1810 if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK)
1811 redisLog(REDIS_NOTICE,"DB loaded from append only file: %.3f seconds",(float)(ustime()-start)/1000000);
1812 } else {
1813 if (rdbLoad(server.dbfilename) == REDIS_OK)
1814 redisLog(REDIS_NOTICE,"DB loaded from disk: %.3f seconds",(float)(ustime()-start)/1000000);
1815 }
1816 if (server.ipfd > 0)
1817 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
1818 if (server.sofd > 0)
1819 redisLog(REDIS_NOTICE,"The server is now ready to accept connections at %s", server.unixsocket);
1820 aeSetBeforeSleepProc(server.el,beforeSleep);
1821 aeMain(server.el);
1822 aeDeleteEventLoop(server.el);
1823 return 0;
1824 }
1825
1826 #ifdef HAVE_BACKTRACE
1827 static void *getMcontextEip(ucontext_t *uc) {
1828 #if defined(__FreeBSD__)
1829 return (void*) uc->uc_mcontext.mc_eip;
1830 #elif defined(__dietlibc__)
1831 return (void*) uc->uc_mcontext.eip;
1832 #elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
1833 #if __x86_64__
1834 return (void*) uc->uc_mcontext->__ss.__rip;
1835 #else
1836 return (void*) uc->uc_mcontext->__ss.__eip;
1837 #endif
1838 #elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6)
1839 #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__)
1840 return (void*) uc->uc_mcontext->__ss.__rip;
1841 #else
1842 return (void*) uc->uc_mcontext->__ss.__eip;
1843 #endif
1844 #elif defined(__i386__)
1845 return (void*) uc->uc_mcontext.gregs[14]; /* Linux 32 */
1846 #elif defined(__X86_64__) || defined(__x86_64__)
1847 return (void*) uc->uc_mcontext.gregs[16]; /* Linux 64 */
1848 #elif defined(__ia64__) /* Linux IA64 */
1849 return (void*) uc->uc_mcontext.sc_ip;
1850 #else
1851 return NULL;
1852 #endif
1853 }
1854
1855 static void sigsegvHandler(int sig, siginfo_t *info, void *secret) {
1856 void *trace[100];
1857 char **messages = NULL;
1858 int i, trace_size = 0;
1859 ucontext_t *uc = (ucontext_t*) secret;
1860 sds infostring;
1861 struct sigaction act;
1862 REDIS_NOTUSED(info);
1863
1864 redisLog(REDIS_WARNING,
1865 "======= Ooops! Redis %s got signal: -%d- =======", REDIS_VERSION, sig);
1866 infostring = genRedisInfoString("all");
1867 redisLogRaw(REDIS_WARNING, infostring);
1868 /* It's not safe to sdsfree() the returned string under memory
1869 * corruption conditions. Let it leak as we are going to abort */
1870
1871 trace_size = backtrace(trace, 100);
1872 /* overwrite sigaction with caller's address */
1873 if (getMcontextEip(uc) != NULL) {
1874 trace[1] = getMcontextEip(uc);
1875 }
1876 messages = backtrace_symbols(trace, trace_size);
1877
1878 for (i=1; i<trace_size; ++i)
1879 redisLog(REDIS_WARNING,"%s", messages[i]);
1880
1881 /* free(messages); Don't call free() with possibly corrupted memory. */
1882 if (server.daemonize) unlink(server.pidfile);
1883
1884 /* Make sure we exit with the right signal at the end. So for instance
1885 * the core will be dumped if enabled. */
1886 sigemptyset (&act.sa_mask);
1887 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1888 * is used. Otherwise, sa_handler is used */
1889 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1890 act.sa_handler = SIG_DFL;
1891 sigaction (sig, &act, NULL);
1892 kill(getpid(),sig);
1893 }
1894 #endif /* HAVE_BACKTRACE */
1895
1896 static void sigtermHandler(int sig) {
1897 REDIS_NOTUSED(sig);
1898
1899 redisLog(REDIS_WARNING,"Received SIGTERM, scheduling shutdown...");
1900 server.shutdown_asap = 1;
1901 }
1902
1903 void setupSignalHandlers(void) {
1904 struct sigaction act;
1905
1906 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction is used.
1907 * Otherwise, sa_handler is used. */
1908 sigemptyset(&act.sa_mask);
1909 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1910 act.sa_handler = sigtermHandler;
1911 sigaction(SIGTERM, &act, NULL);
1912
1913 #ifdef HAVE_BACKTRACE
1914 sigemptyset(&act.sa_mask);
1915 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
1916 act.sa_sigaction = sigsegvHandler;
1917 sigaction(SIGSEGV, &act, NULL);
1918 sigaction(SIGBUS, &act, NULL);
1919 sigaction(SIGFPE, &act, NULL);
1920 sigaction(SIGILL, &act, NULL);
1921 #endif
1922 return;
1923 }
1924
1925 /* The End */