]> git.saurik.com Git - redis.git/blob - src/redis.c
Merge branch 'unstable'
[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 /* Check if a background saving or AOF rewrite in progress terminated. */
637 if (server.bgsavechildpid != -1 || server.bgrewritechildpid != -1) {
638 int statloc;
639 pid_t pid;
640
641 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
642 int exitcode = WEXITSTATUS(statloc);
643 int bysignal = 0;
644
645 if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
646
647 if (pid == server.bgsavechildpid) {
648 backgroundSaveDoneHandler(exitcode,bysignal);
649 } else {
650 backgroundRewriteDoneHandler(exitcode,bysignal);
651 }
652 updateDictResizePolicy();
653 }
654 } else if (server.bgsavethread != (pthread_t) -1) {
655 if (server.bgsavethread != (pthread_t) -1) {
656 int state;
657
658 pthread_mutex_lock(&server.bgsavethread_mutex);
659 state = server.bgsavethread_state;
660 pthread_mutex_unlock(&server.bgsavethread_mutex);
661
662 if (state == REDIS_BGSAVE_THREAD_DONE_OK ||
663 state == REDIS_BGSAVE_THREAD_DONE_ERR)
664 {
665 backgroundSaveDoneHandler(
666 (state == REDIS_BGSAVE_THREAD_DONE_OK) ? 0 : 1, 0);
667 }
668 }
669 } else if (!server.ds_enabled) {
670 /* If there is not a background saving in progress check if
671 * we have to save now */
672 time_t now = time(NULL);
673 for (j = 0; j < server.saveparamslen; j++) {
674 struct saveparam *sp = server.saveparams+j;
675
676 if (server.dirty >= sp->changes &&
677 now-server.lastsave > sp->seconds) {
678 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
679 sp->changes, sp->seconds);
680 rdbSaveBackground(server.dbfilename);
681 break;
682 }
683 }
684 }
685
686 /* Expire a few keys per cycle, only if this is a master.
687 * On slaves we wait for DEL operations synthesized by the master
688 * in order to guarantee a strict consistency. */
689 if (server.masterhost == NULL) activeExpireCycle();
690
691 /* Remove a few cached objects from memory if we are over the
692 * configured memory limit */
693 if (server.ds_enabled) cacheCron();
694
695 /* Replication cron function -- used to reconnect to master and
696 * to detect transfer failures. */
697 if (!(loops % 10)) replicationCron();
698
699 /* Run other sub-systems specific cron jobs */
700 if (server.cluster_enabled && !(loops % 10)) clusterCron();
701
702 server.cronloops++;
703 return 100;
704 }
705
706 /* This function gets called every time Redis is entering the
707 * main loop of the event driven library, that is, before to sleep
708 * for ready file descriptors. */
709 void beforeSleep(struct aeEventLoop *eventLoop) {
710 REDIS_NOTUSED(eventLoop);
711 listNode *ln;
712 redisClient *c;
713
714 /* Awake clients that got all the on disk keys they requested */
715 if (server.ds_enabled && listLength(server.io_ready_clients)) {
716 listIter li;
717
718 listRewind(server.io_ready_clients,&li);
719 while((ln = listNext(&li))) {
720 c = ln->value;
721 struct redisCommand *cmd;
722
723 /* Resume the client. */
724 listDelNode(server.io_ready_clients,ln);
725 c->flags &= (~REDIS_IO_WAIT);
726 server.cache_blocked_clients--;
727 aeCreateFileEvent(server.el, c->fd, AE_READABLE,
728 readQueryFromClient, c);
729 cmd = lookupCommand(c->argv[0]->ptr);
730 redisAssert(cmd != NULL);
731 call(c,cmd);
732 resetClient(c);
733 /* There may be more data to process in the input buffer. */
734 if (c->querybuf && sdslen(c->querybuf) > 0)
735 processInputBuffer(c);
736 }
737 }
738
739 /* Try to process pending commands for clients that were just unblocked. */
740 while (listLength(server.unblocked_clients)) {
741 ln = listFirst(server.unblocked_clients);
742 redisAssert(ln != NULL);
743 c = ln->value;
744 listDelNode(server.unblocked_clients,ln);
745 c->flags &= ~REDIS_UNBLOCKED;
746
747 /* Process remaining data in the input buffer. */
748 if (c->querybuf && sdslen(c->querybuf) > 0)
749 processInputBuffer(c);
750 }
751
752 /* Write the AOF buffer on disk */
753 flushAppendOnlyFile();
754 }
755
756 /* =========================== Server initialization ======================== */
757
758 void createSharedObjects(void) {
759 int j;
760
761 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
762 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
763 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
764 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
765 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
766 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
767 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
768 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
769 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
770 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
771 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
772 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
773 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
774 "-ERR Operation against a key holding the wrong kind of value\r\n"));
775 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
776 "-ERR no such key\r\n"));
777 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
778 "-ERR syntax error\r\n"));
779 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
780 "-ERR source and destination objects are the same\r\n"));
781 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
782 "-ERR index out of range\r\n"));
783 shared.loadingerr = createObject(REDIS_STRING,sdsnew(
784 "-LOADING Redis is loading the dataset in memory\r\n"));
785 shared.space = createObject(REDIS_STRING,sdsnew(" "));
786 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
787 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
788 shared.select0 = createStringObject("select 0\r\n",10);
789 shared.select1 = createStringObject("select 1\r\n",10);
790 shared.select2 = createStringObject("select 2\r\n",10);
791 shared.select3 = createStringObject("select 3\r\n",10);
792 shared.select4 = createStringObject("select 4\r\n",10);
793 shared.select5 = createStringObject("select 5\r\n",10);
794 shared.select6 = createStringObject("select 6\r\n",10);
795 shared.select7 = createStringObject("select 7\r\n",10);
796 shared.select8 = createStringObject("select 8\r\n",10);
797 shared.select9 = createStringObject("select 9\r\n",10);
798 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
799 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
800 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
801 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
802 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
803 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
804 shared.mbulk3 = createStringObject("*3\r\n",4);
805 shared.mbulk4 = createStringObject("*4\r\n",4);
806 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
807 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
808 shared.integers[j]->encoding = REDIS_ENCODING_INT;
809 }
810 }
811
812 void initServerConfig() {
813 server.port = REDIS_SERVERPORT;
814 server.bindaddr = NULL;
815 server.unixsocket = NULL;
816 server.ipfd = -1;
817 server.sofd = -1;
818 server.dbnum = REDIS_DEFAULT_DBNUM;
819 server.verbosity = REDIS_VERBOSE;
820 server.maxidletime = REDIS_MAXIDLETIME;
821 server.saveparams = NULL;
822 server.loading = 0;
823 server.logfile = NULL; /* NULL = log on standard output */
824 server.syslog_enabled = 0;
825 server.syslog_ident = zstrdup("redis");
826 server.syslog_facility = LOG_LOCAL0;
827 server.daemonize = 0;
828 server.appendonly = 0;
829 server.appendfsync = APPENDFSYNC_EVERYSEC;
830 server.no_appendfsync_on_rewrite = 0;
831 server.lastfsync = time(NULL);
832 server.appendfd = -1;
833 server.appendseldb = -1; /* Make sure the first time will not match */
834 server.pidfile = zstrdup("/var/run/redis.pid");
835 server.dbfilename = zstrdup("dump.rdb");
836 server.appendfilename = zstrdup("appendonly.aof");
837 server.requirepass = NULL;
838 server.rdbcompression = 1;
839 server.activerehashing = 1;
840 server.maxclients = 0;
841 server.bpop_blocked_clients = 0;
842 server.maxmemory = 0;
843 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
844 server.maxmemory_samples = 3;
845 server.ds_enabled = 0;
846 server.ds_path = sdsnew("/tmp/redis.ds");
847 server.cache_max_memory = 64LL*1024*1024; /* 64 MB of RAM */
848 server.cache_blocked_clients = 0;
849 server.hash_max_zipmap_entries = REDIS_HASH_MAX_ZIPMAP_ENTRIES;
850 server.hash_max_zipmap_value = REDIS_HASH_MAX_ZIPMAP_VALUE;
851 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
852 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
853 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
854 server.zset_max_ziplist_entries = REDIS_ZSET_MAX_ZIPLIST_ENTRIES;
855 server.zset_max_ziplist_value = REDIS_ZSET_MAX_ZIPLIST_VALUE;
856 server.shutdown_asap = 0;
857 server.cache_flush_delay = 0;
858 server.cluster_enabled = 0;
859 server.cluster.configfile = zstrdup("nodes.conf");
860
861 updateLRUClock();
862 resetServerSaveParams();
863
864 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
865 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
866 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
867 /* Replication related */
868 server.isslave = 0;
869 server.masterauth = NULL;
870 server.masterhost = NULL;
871 server.masterport = 6379;
872 server.master = NULL;
873 server.replstate = REDIS_REPL_NONE;
874 server.repl_serve_stale_data = 1;
875
876 /* Double constants initialization */
877 R_Zero = 0.0;
878 R_PosInf = 1.0/R_Zero;
879 R_NegInf = -1.0/R_Zero;
880 R_Nan = R_Zero/R_Zero;
881
882 /* Command table -- we intiialize it here as it is part of the
883 * initial configuration, since command names may be changed via
884 * redis.conf using the rename-command directive. */
885 server.commands = dictCreate(&commandTableDictType,NULL);
886 populateCommandTable();
887 server.delCommand = lookupCommandByCString("del");
888 server.multiCommand = lookupCommandByCString("multi");
889 }
890
891 void initServer() {
892 int j;
893
894 signal(SIGHUP, SIG_IGN);
895 signal(SIGPIPE, SIG_IGN);
896 setupSignalHandlers();
897
898 if (server.syslog_enabled) {
899 openlog(server.syslog_ident, LOG_PID | LOG_NDELAY | LOG_NOWAIT,
900 server.syslog_facility);
901 }
902
903 server.mainthread = pthread_self();
904 server.clients = listCreate();
905 server.slaves = listCreate();
906 server.monitors = listCreate();
907 server.unblocked_clients = listCreate();
908 server.cache_io_queue = listCreate();
909
910 createSharedObjects();
911 server.el = aeCreateEventLoop();
912 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
913
914 if (server.port != 0) {
915 server.ipfd = anetTcpServer(server.neterr,server.port,server.bindaddr);
916 if (server.ipfd == ANET_ERR) {
917 redisLog(REDIS_WARNING, "Opening port: %s", server.neterr);
918 exit(1);
919 }
920 }
921 if (server.unixsocket != NULL) {
922 unlink(server.unixsocket); /* don't care if this fails */
923 server.sofd = anetUnixServer(server.neterr,server.unixsocket);
924 if (server.sofd == ANET_ERR) {
925 redisLog(REDIS_WARNING, "Opening socket: %s", server.neterr);
926 exit(1);
927 }
928 }
929 if (server.ipfd < 0 && server.sofd < 0) {
930 redisLog(REDIS_WARNING, "Configured to not listen anywhere, exiting.");
931 exit(1);
932 }
933 for (j = 0; j < server.dbnum; j++) {
934 server.db[j].dict = dictCreate(&dbDictType,NULL);
935 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
936 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
937 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
938 if (server.ds_enabled) {
939 server.db[j].io_keys = dictCreate(&keylistDictType,NULL);
940 server.db[j].io_negcache = dictCreate(&setDictType,NULL);
941 server.db[j].io_queued = dictCreate(&setDictType,NULL);
942 }
943 server.db[j].id = j;
944 }
945 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
946 server.pubsub_patterns = listCreate();
947 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
948 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
949 server.cronloops = 0;
950 server.bgsavechildpid = -1;
951 server.bgrewritechildpid = -1;
952 server.bgsavethread_state = REDIS_BGSAVE_THREAD_UNACTIVE;
953 server.bgsavethread = (pthread_t) -1;
954 server.bgrewritebuf = sdsempty();
955 server.aofbuf = sdsempty();
956 server.lastsave = time(NULL);
957 server.dirty = 0;
958 server.stat_numcommands = 0;
959 server.stat_numconnections = 0;
960 server.stat_expiredkeys = 0;
961 server.stat_evictedkeys = 0;
962 server.stat_starttime = time(NULL);
963 server.stat_keyspace_misses = 0;
964 server.stat_keyspace_hits = 0;
965 server.stat_peak_memory = 0;
966 server.unixtime = time(NULL);
967 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
968 if (server.ipfd > 0 && aeCreateFileEvent(server.el,server.ipfd,AE_READABLE,
969 acceptTcpHandler,NULL) == AE_ERR) oom("creating file event");
970 if (server.sofd > 0 && aeCreateFileEvent(server.el,server.sofd,AE_READABLE,
971 acceptUnixHandler,NULL) == AE_ERR) oom("creating file event");
972
973 if (server.appendonly) {
974 server.appendfd = open(server.appendfilename,O_WRONLY|O_APPEND|O_CREAT,0644);
975 if (server.appendfd == -1) {
976 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
977 strerror(errno));
978 exit(1);
979 }
980 }
981
982 if (server.ds_enabled) dsInit();
983 if (server.cluster_enabled) clusterInit();
984 srand(time(NULL)^getpid());
985 }
986
987 /* Populates the Redis Command Table starting from the hard coded list
988 * we have on top of redis.c file. */
989 void populateCommandTable(void) {
990 int j;
991 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
992
993 for (j = 0; j < numcommands; j++) {
994 struct redisCommand *c = redisCommandTable+j;
995 int retval;
996
997 retval = dictAdd(server.commands, sdsnew(c->name), c);
998 assert(retval == DICT_OK);
999 }
1000 }
1001
1002 void resetCommandTableStats(void) {
1003 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1004 int j;
1005
1006 for (j = 0; j < numcommands; j++) {
1007 struct redisCommand *c = redisCommandTable+j;
1008
1009 c->microseconds = 0;
1010 c->calls = 0;
1011 }
1012 }
1013
1014 /* ====================== Commands lookup and execution ===================== */
1015
1016 struct redisCommand *lookupCommand(sds name) {
1017 return dictFetchValue(server.commands, name);
1018 }
1019
1020 struct redisCommand *lookupCommandByCString(char *s) {
1021 struct redisCommand *cmd;
1022 sds name = sdsnew(s);
1023
1024 cmd = dictFetchValue(server.commands, name);
1025 sdsfree(name);
1026 return cmd;
1027 }
1028
1029 /* Call() is the core of Redis execution of a command */
1030 void call(redisClient *c, struct redisCommand *cmd) {
1031 long long dirty, start = ustime();
1032
1033 dirty = server.dirty;
1034 cmd->proc(c);
1035 dirty = server.dirty-dirty;
1036 cmd->microseconds += ustime()-start;
1037 cmd->calls++;
1038
1039 if (server.appendonly && dirty)
1040 feedAppendOnlyFile(cmd,c->db->id,c->argv,c->argc);
1041 if ((dirty || cmd->flags & REDIS_CMD_FORCE_REPLICATION) &&
1042 listLength(server.slaves))
1043 replicationFeedSlaves(server.slaves,c->db->id,c->argv,c->argc);
1044 if (listLength(server.monitors))
1045 replicationFeedMonitors(server.monitors,c->db->id,c->argv,c->argc);
1046 server.stat_numcommands++;
1047 }
1048
1049 /* If this function gets called we already read a whole
1050 * command, argments are in the client argv/argc fields.
1051 * processCommand() execute the command or prepare the
1052 * server for a bulk read from the client.
1053 *
1054 * If 1 is returned the client is still alive and valid and
1055 * and other operations can be performed by the caller. Otherwise
1056 * if 0 is returned the client was destroied (i.e. after QUIT). */
1057 int processCommand(redisClient *c) {
1058 struct redisCommand *cmd;
1059
1060 /* The QUIT command is handled separately. Normal command procs will
1061 * go through checking for replication and QUIT will cause trouble
1062 * when FORCE_REPLICATION is enabled and would be implemented in
1063 * a regular command proc. */
1064 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
1065 addReply(c,shared.ok);
1066 c->flags |= REDIS_CLOSE_AFTER_REPLY;
1067 return REDIS_ERR;
1068 }
1069
1070 /* Now lookup the command and check ASAP about trivial error conditions
1071 * such wrong arity, bad command name and so forth. */
1072 cmd = lookupCommand(c->argv[0]->ptr);
1073 if (!cmd) {
1074 addReplyErrorFormat(c,"unknown command '%s'",
1075 (char*)c->argv[0]->ptr);
1076 return REDIS_OK;
1077 } else if ((cmd->arity > 0 && cmd->arity != c->argc) ||
1078 (c->argc < -cmd->arity)) {
1079 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
1080 cmd->name);
1081 return REDIS_OK;
1082 }
1083
1084 /* Check if the user is authenticated */
1085 if (server.requirepass && !c->authenticated && cmd->proc != authCommand) {
1086 addReplyError(c,"operation not permitted");
1087 return REDIS_OK;
1088 }
1089
1090 /* If cluster is enabled, redirect here */
1091 if (server.cluster_enabled &&
1092 !(cmd->getkeys_proc == NULL && cmd->firstkey == 0)) {
1093 int hashslot;
1094
1095 if (server.cluster.state != REDIS_CLUSTER_OK) {
1096 addReplyError(c,"The cluster is down. Check with CLUSTER INFO for more information");
1097 return REDIS_OK;
1098 } else {
1099 int ask;
1100 clusterNode *n = getNodeByQuery(c,cmd,c->argv,c->argc,&hashslot,&ask);
1101 if (n == NULL) {
1102 addReplyError(c,"Multi keys request invalid in cluster");
1103 return REDIS_OK;
1104 } else if (n != server.cluster.myself) {
1105 addReplySds(c,sdscatprintf(sdsempty(),
1106 "-%s %d %s:%d\r\n", ask ? "ASK" : "MOVED",
1107 hashslot,n->ip,n->port));
1108 return REDIS_OK;
1109 }
1110 }
1111 }
1112
1113 /* Handle the maxmemory directive.
1114 *
1115 * First we try to free some memory if possible (if there are volatile
1116 * keys in the dataset). If there are not the only thing we can do
1117 * is returning an error. */
1118 if (server.maxmemory) freeMemoryIfNeeded();
1119 if (server.maxmemory && (cmd->flags & REDIS_CMD_DENYOOM) &&
1120 zmalloc_used_memory() > server.maxmemory)
1121 {
1122 addReplyError(c,"command not allowed when used memory > 'maxmemory'");
1123 return REDIS_OK;
1124 }
1125
1126 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
1127 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
1128 &&
1129 cmd->proc != subscribeCommand && cmd->proc != unsubscribeCommand &&
1130 cmd->proc != psubscribeCommand && cmd->proc != punsubscribeCommand) {
1131 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
1132 return REDIS_OK;
1133 }
1134
1135 /* Only allow INFO and SLAVEOF when slave-serve-stale-data is no and
1136 * we are a slave with a broken link with master. */
1137 if (server.masterhost && server.replstate != REDIS_REPL_CONNECTED &&
1138 server.repl_serve_stale_data == 0 &&
1139 cmd->proc != infoCommand && cmd->proc != slaveofCommand)
1140 {
1141 addReplyError(c,
1142 "link with MASTER is down and slave-serve-stale-data is set to no");
1143 return REDIS_OK;
1144 }
1145
1146 /* Loading DB? Return an error if the command is not INFO */
1147 if (server.loading && cmd->proc != infoCommand) {
1148 addReply(c, shared.loadingerr);
1149 return REDIS_OK;
1150 }
1151
1152 /* Exec the command */
1153 if (c->flags & REDIS_MULTI &&
1154 cmd->proc != execCommand && cmd->proc != discardCommand &&
1155 cmd->proc != multiCommand && cmd->proc != watchCommand)
1156 {
1157 queueMultiCommand(c,cmd);
1158 addReply(c,shared.queued);
1159 } else {
1160 if (server.ds_enabled && blockClientOnSwappedKeys(c,cmd))
1161 return REDIS_ERR;
1162 call(c,cmd);
1163 }
1164 return REDIS_OK;
1165 }
1166
1167 /*================================== Shutdown =============================== */
1168
1169 int prepareForShutdown() {
1170 redisLog(REDIS_WARNING,"User requested shutdown, saving DB...");
1171 /* Kill the saving child if there is a background saving in progress.
1172 We want to avoid race conditions, for instance our saving child may
1173 overwrite the synchronous saving did by SHUTDOWN. */
1174 if (server.bgsavechildpid != -1) {
1175 redisLog(REDIS_WARNING,"There is a live saving child. Killing it!");
1176 kill(server.bgsavechildpid,SIGKILL);
1177 rdbRemoveTempFile(server.bgsavechildpid);
1178 }
1179 if (server.ds_enabled) {
1180 /* FIXME: flush all objects on disk */
1181 } else if (server.appendonly) {
1182 /* Append only file: fsync() the AOF and exit */
1183 aof_fsync(server.appendfd);
1184 } else if (server.saveparamslen > 0) {
1185 /* Snapshotting. Perform a SYNC SAVE and exit */
1186 if (rdbSave(server.dbfilename) != REDIS_OK) {
1187 /* Ooops.. error saving! The best we can do is to continue
1188 * operating. Note that if there was a background saving process,
1189 * in the next cron() Redis will be notified that the background
1190 * saving aborted, handling special stuff like slaves pending for
1191 * synchronization... */
1192 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit");
1193 return REDIS_ERR;
1194 }
1195 } else {
1196 redisLog(REDIS_WARNING,"Not saving DB.");
1197 }
1198 if (server.daemonize) unlink(server.pidfile);
1199 redisLog(REDIS_WARNING,"Server exit now, bye bye...");
1200 return REDIS_OK;
1201 }
1202
1203 /*================================== Commands =============================== */
1204
1205 void authCommand(redisClient *c) {
1206 if (!server.requirepass || !strcmp(c->argv[1]->ptr, server.requirepass)) {
1207 c->authenticated = 1;
1208 addReply(c,shared.ok);
1209 } else {
1210 c->authenticated = 0;
1211 addReplyError(c,"invalid password");
1212 }
1213 }
1214
1215 void pingCommand(redisClient *c) {
1216 addReply(c,shared.pong);
1217 }
1218
1219 void echoCommand(redisClient *c) {
1220 addReplyBulk(c,c->argv[1]);
1221 }
1222
1223 /* Convert an amount of bytes into a human readable string in the form
1224 * of 100B, 2G, 100M, 4K, and so forth. */
1225 void bytesToHuman(char *s, unsigned long long n) {
1226 double d;
1227
1228 if (n < 1024) {
1229 /* Bytes */
1230 sprintf(s,"%lluB",n);
1231 return;
1232 } else if (n < (1024*1024)) {
1233 d = (double)n/(1024);
1234 sprintf(s,"%.2fK",d);
1235 } else if (n < (1024LL*1024*1024)) {
1236 d = (double)n/(1024*1024);
1237 sprintf(s,"%.2fM",d);
1238 } else if (n < (1024LL*1024*1024*1024)) {
1239 d = (double)n/(1024LL*1024*1024);
1240 sprintf(s,"%.2fG",d);
1241 }
1242 }
1243
1244 /* Create the string returned by the INFO command. This is decoupled
1245 * by the INFO command itself as we need to report the same information
1246 * on memory corruption problems. */
1247 sds genRedisInfoString(char *section) {
1248 sds info = sdsempty();
1249 time_t uptime = time(NULL)-server.stat_starttime;
1250 int j, numcommands;
1251 struct rusage self_ru, c_ru;
1252 unsigned long lol, bib;
1253 int allsections = 0, defsections = 0;
1254 int sections = 0;
1255
1256 if (section) {
1257 allsections = strcasecmp(section,"all") == 0;
1258 defsections = strcasecmp(section,"default") == 0;
1259 }
1260
1261 getrusage(RUSAGE_SELF, &self_ru);
1262 getrusage(RUSAGE_CHILDREN, &c_ru);
1263 getClientsMaxBuffers(&lol,&bib);
1264
1265 /* Server */
1266 if (allsections || defsections || !strcasecmp(section,"server")) {
1267 if (sections++) info = sdscat(info,"\r\n");
1268 info = sdscatprintf(info,
1269 "# Server\r\n"
1270 "redis_version:%s\r\n"
1271 "redis_git_sha1:%s\r\n"
1272 "redis_git_dirty:%d\r\n"
1273 "arch_bits:%s\r\n"
1274 "multiplexing_api:%s\r\n"
1275 "process_id:%ld\r\n"
1276 "tcp_port:%d\r\n"
1277 "uptime_in_seconds:%ld\r\n"
1278 "uptime_in_days:%ld\r\n"
1279 "lru_clock:%ld\r\n",
1280 REDIS_VERSION,
1281 redisGitSHA1(),
1282 strtol(redisGitDirty(),NULL,10) > 0,
1283 (sizeof(long) == 8) ? "64" : "32",
1284 aeGetApiName(),
1285 (long) getpid(),
1286 server.port,
1287 uptime,
1288 uptime/(3600*24),
1289 (unsigned long) server.lruclock);
1290 }
1291
1292 /* Clients */
1293 if (allsections || defsections || !strcasecmp(section,"clients")) {
1294 if (sections++) info = sdscat(info,"\r\n");
1295 info = sdscatprintf(info,
1296 "# Clients\r\n"
1297 "connected_clients:%d\r\n"
1298 "client_longest_output_list:%lu\r\n"
1299 "client_biggest_input_buf:%lu\r\n"
1300 "blocked_clients:%d\r\n",
1301 listLength(server.clients)-listLength(server.slaves),
1302 lol, bib,
1303 server.bpop_blocked_clients);
1304 }
1305
1306 /* Memory */
1307 if (allsections || defsections || !strcasecmp(section,"memory")) {
1308 char hmem[64];
1309 char peak_hmem[64];
1310
1311 bytesToHuman(hmem,zmalloc_used_memory());
1312 bytesToHuman(peak_hmem,server.stat_peak_memory);
1313 if (sections++) info = sdscat(info,"\r\n");
1314 info = sdscatprintf(info,
1315 "# Memory\r\n"
1316 "used_memory:%zu\r\n"
1317 "used_memory_human:%s\r\n"
1318 "used_memory_rss:%zu\r\n"
1319 "used_memory_peak:%zu\r\n"
1320 "used_memory_peak_human:%s\r\n"
1321 "mem_fragmentation_ratio:%.2f\r\n"
1322 "mem_allocator:%s\r\n",
1323 zmalloc_used_memory(),
1324 hmem,
1325 zmalloc_get_rss(),
1326 server.stat_peak_memory,
1327 peak_hmem,
1328 zmalloc_get_fragmentation_ratio(),
1329 REDIS_MALLOC
1330 );
1331 }
1332
1333 /* Allocation statistics */
1334 if (allsections || !strcasecmp(section,"allocstats")) {
1335 if (sections++) info = sdscat(info,"\r\n");
1336 info = sdscat(info, "# Allocstats\r\nallocation_stats:");
1337 for (j = 0; j <= ZMALLOC_MAX_ALLOC_STAT; j++) {
1338 size_t count = zmalloc_allocations_for_size(j);
1339 if (count) {
1340 if (info[sdslen(info)-1] != ':') info = sdscatlen(info,",",1);
1341 info = sdscatprintf(info,"%s%d=%zu",
1342 (j == ZMALLOC_MAX_ALLOC_STAT) ? ">=" : "",
1343 j,count);
1344 }
1345 }
1346 info = sdscat(info,"\r\n");
1347 }
1348
1349 /* Persistence */
1350 if (allsections || defsections || !strcasecmp(section,"persistence")) {
1351 if (sections++) info = sdscat(info,"\r\n");
1352 info = sdscatprintf(info,
1353 "# Persistence\r\n"
1354 "loading:%d\r\n"
1355 "aof_enabled:%d\r\n"
1356 "changes_since_last_save:%lld\r\n"
1357 "bgsave_in_progress:%d\r\n"
1358 "last_save_time:%ld\r\n"
1359 "bgrewriteaof_in_progress:%d\r\n",
1360 server.loading,
1361 server.appendonly,
1362 server.dirty,
1363 server.bgsavechildpid != -1 ||
1364 server.bgsavethread != (pthread_t) -1,
1365 server.lastsave,
1366 server.bgrewritechildpid != -1);
1367
1368 if (server.loading) {
1369 double perc;
1370 time_t eta, elapsed;
1371 off_t remaining_bytes = server.loading_total_bytes-
1372 server.loading_loaded_bytes;
1373
1374 perc = ((double)server.loading_loaded_bytes /
1375 server.loading_total_bytes) * 100;
1376
1377 elapsed = time(NULL)-server.loading_start_time;
1378 if (elapsed == 0) {
1379 eta = 1; /* A fake 1 second figure if we don't have
1380 enough info */
1381 } else {
1382 eta = (elapsed*remaining_bytes)/server.loading_loaded_bytes;
1383 }
1384
1385 info = sdscatprintf(info,
1386 "loading_start_time:%ld\r\n"
1387 "loading_total_bytes:%llu\r\n"
1388 "loading_loaded_bytes:%llu\r\n"
1389 "loading_loaded_perc:%.2f\r\n"
1390 "loading_eta_seconds:%ld\r\n"
1391 ,(unsigned long) server.loading_start_time,
1392 (unsigned long long) server.loading_total_bytes,
1393 (unsigned long long) server.loading_loaded_bytes,
1394 perc,
1395 eta
1396 );
1397 }
1398 }
1399
1400 /* Diskstore */
1401 if (allsections || defsections || !strcasecmp(section,"diskstore")) {
1402 if (sections++) info = sdscat(info,"\r\n");
1403 info = sdscatprintf(info,
1404 "# Diskstore\r\n"
1405 "ds_enabled:%d\r\n",
1406 server.ds_enabled != 0);
1407 if (server.ds_enabled) {
1408 lockThreadedIO();
1409 info = sdscatprintf(info,
1410 "cache_max_memory:%llu\r\n"
1411 "cache_blocked_clients:%lu\r\n"
1412 "cache_io_queue_len:%lu\r\n"
1413 "cache_io_jobs_new:%lu\r\n"
1414 "cache_io_jobs_processing:%lu\r\n"
1415 "cache_io_jobs_processed:%lu\r\n"
1416 "cache_io_ready_clients:%lu\r\n"
1417 ,(unsigned long long) server.cache_max_memory,
1418 (unsigned long) server.cache_blocked_clients,
1419 (unsigned long) listLength(server.cache_io_queue),
1420 (unsigned long) listLength(server.io_newjobs),
1421 (unsigned long) listLength(server.io_processing),
1422 (unsigned long) listLength(server.io_processed),
1423 (unsigned long) listLength(server.io_ready_clients)
1424 );
1425 unlockThreadedIO();
1426 }
1427 }
1428
1429 /* Stats */
1430 if (allsections || defsections || !strcasecmp(section,"stats")) {
1431 if (sections++) info = sdscat(info,"\r\n");
1432 info = sdscatprintf(info,
1433 "# Stats\r\n"
1434 "total_connections_received:%lld\r\n"
1435 "total_commands_processed:%lld\r\n"
1436 "expired_keys:%lld\r\n"
1437 "evicted_keys:%lld\r\n"
1438 "keyspace_hits:%lld\r\n"
1439 "keyspace_misses:%lld\r\n"
1440 "pubsub_channels:%ld\r\n"
1441 "pubsub_patterns:%u\r\n",
1442 server.stat_numconnections,
1443 server.stat_numcommands,
1444 server.stat_expiredkeys,
1445 server.stat_evictedkeys,
1446 server.stat_keyspace_hits,
1447 server.stat_keyspace_misses,
1448 dictSize(server.pubsub_channels),
1449 listLength(server.pubsub_patterns));
1450 }
1451
1452 /* Replication */
1453 if (allsections || defsections || !strcasecmp(section,"replication")) {
1454 if (sections++) info = sdscat(info,"\r\n");
1455 info = sdscatprintf(info,
1456 "# Replication\r\n"
1457 "role:%s\r\n",
1458 server.masterhost == NULL ? "master" : "slave");
1459 if (server.masterhost) {
1460 info = sdscatprintf(info,
1461 "master_host:%s\r\n"
1462 "master_port:%d\r\n"
1463 "master_link_status:%s\r\n"
1464 "master_last_io_seconds_ago:%d\r\n"
1465 "master_sync_in_progress:%d\r\n"
1466 ,server.masterhost,
1467 server.masterport,
1468 (server.replstate == REDIS_REPL_CONNECTED) ?
1469 "up" : "down",
1470 server.master ?
1471 ((int)(time(NULL)-server.master->lastinteraction)) : -1,
1472 server.replstate == REDIS_REPL_TRANSFER
1473 );
1474
1475 if (server.replstate == REDIS_REPL_TRANSFER) {
1476 info = sdscatprintf(info,
1477 "master_sync_left_bytes:%ld\r\n"
1478 "master_sync_last_io_seconds_ago:%d\r\n"
1479 ,(long)server.repl_transfer_left,
1480 (int)(time(NULL)-server.repl_transfer_lastio)
1481 );
1482 }
1483 }
1484 info = sdscatprintf(info,
1485 "connected_slaves:%d\r\n",
1486 listLength(server.slaves));
1487 }
1488
1489 /* CPU */
1490 if (allsections || defsections || !strcasecmp(section,"cpu")) {
1491 if (sections++) info = sdscat(info,"\r\n");
1492 info = sdscatprintf(info,
1493 "# CPU\r\n"
1494 "used_cpu_sys:%.2f\r\n"
1495 "used_cpu_user:%.2f\r\n"
1496 "used_cpu_sys_childrens:%.2f\r\n"
1497 "used_cpu_user_childrens:%.2f\r\n",
1498 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
1499 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
1500 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000,
1501 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000);
1502 }
1503
1504 /* cmdtime */
1505 if (allsections || !strcasecmp(section,"commandstats")) {
1506 if (sections++) info = sdscat(info,"\r\n");
1507 info = sdscatprintf(info, "# Commandstats\r\n");
1508 numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1509 for (j = 0; j < numcommands; j++) {
1510 struct redisCommand *c = redisCommandTable+j;
1511
1512 if (!c->calls) continue;
1513 info = sdscatprintf(info,
1514 "cmdstat_%s:calls=%lld,usec=%lld,usec_per_call=%.2f\r\n",
1515 c->name, c->calls, c->microseconds,
1516 (c->calls == 0) ? 0 : ((float)c->microseconds/c->calls));
1517 }
1518 }
1519
1520 /* Clusetr */
1521 if (allsections || defsections || !strcasecmp(section,"cluster")) {
1522 if (sections++) info = sdscat(info,"\r\n");
1523 info = sdscatprintf(info,
1524 "# Cluster\r\n"
1525 "cluster_enabled:%d\r\n",
1526 server.cluster_enabled);
1527 }
1528
1529 /* Key space */
1530 if (allsections || defsections || !strcasecmp(section,"keyspace")) {
1531 if (sections++) info = sdscat(info,"\r\n");
1532 info = sdscatprintf(info, "# Keyspace\r\n");
1533 for (j = 0; j < server.dbnum; j++) {
1534 long long keys, vkeys;
1535
1536 keys = dictSize(server.db[j].dict);
1537 vkeys = dictSize(server.db[j].expires);
1538 if (keys || vkeys) {
1539 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
1540 j, keys, vkeys);
1541 }
1542 }
1543 }
1544 return info;
1545 }
1546
1547 void infoCommand(redisClient *c) {
1548 char *section = c->argc == 2 ? c->argv[1]->ptr : "default";
1549
1550 if (c->argc > 2) {
1551 addReply(c,shared.syntaxerr);
1552 return;
1553 }
1554 sds info = genRedisInfoString(section);
1555 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
1556 (unsigned long)sdslen(info)));
1557 addReplySds(c,info);
1558 addReply(c,shared.crlf);
1559 }
1560
1561 void monitorCommand(redisClient *c) {
1562 /* ignore MONITOR if aleady slave or in monitor mode */
1563 if (c->flags & REDIS_SLAVE) return;
1564
1565 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
1566 c->slaveseldb = 0;
1567 listAddNodeTail(server.monitors,c);
1568 addReply(c,shared.ok);
1569 }
1570
1571 /* ============================ Maxmemory directive ======================== */
1572
1573 /* This function gets called when 'maxmemory' is set on the config file to limit
1574 * the max memory used by the server, and we are out of memory.
1575 * This function will try to, in order:
1576 *
1577 * - Free objects from the free list
1578 * - Try to remove keys with an EXPIRE set
1579 *
1580 * It is not possible to free enough memory to reach used-memory < maxmemory
1581 * the server will start refusing commands that will enlarge even more the
1582 * memory usage.
1583 */
1584 void freeMemoryIfNeeded(void) {
1585 /* Remove keys accordingly to the active policy as long as we are
1586 * over the memory limit. */
1587 if (server.maxmemory_policy == REDIS_MAXMEMORY_NO_EVICTION) return;
1588
1589 while (server.maxmemory && zmalloc_used_memory() > server.maxmemory) {
1590 int j, k, freed = 0;
1591
1592 for (j = 0; j < server.dbnum; j++) {
1593 long bestval = 0; /* just to prevent warning */
1594 sds bestkey = NULL;
1595 struct dictEntry *de;
1596 redisDb *db = server.db+j;
1597 dict *dict;
1598
1599 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1600 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
1601 {
1602 dict = server.db[j].dict;
1603 } else {
1604 dict = server.db[j].expires;
1605 }
1606 if (dictSize(dict) == 0) continue;
1607
1608 /* volatile-random and allkeys-random policy */
1609 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
1610 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
1611 {
1612 de = dictGetRandomKey(dict);
1613 bestkey = dictGetEntryKey(de);
1614 }
1615
1616 /* volatile-lru and allkeys-lru policy */
1617 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
1618 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1619 {
1620 for (k = 0; k < server.maxmemory_samples; k++) {
1621 sds thiskey;
1622 long thisval;
1623 robj *o;
1624
1625 de = dictGetRandomKey(dict);
1626 thiskey = dictGetEntryKey(de);
1627 /* When policy is volatile-lru we need an additonal lookup
1628 * to locate the real key, as dict is set to db->expires. */
1629 if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
1630 de = dictFind(db->dict, thiskey);
1631 o = dictGetEntryVal(de);
1632 thisval = estimateObjectIdleTime(o);
1633
1634 /* Higher idle time is better candidate for deletion */
1635 if (bestkey == NULL || thisval > bestval) {
1636 bestkey = thiskey;
1637 bestval = thisval;
1638 }
1639 }
1640 }
1641
1642 /* volatile-ttl */
1643 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
1644 for (k = 0; k < server.maxmemory_samples; k++) {
1645 sds thiskey;
1646 long thisval;
1647
1648 de = dictGetRandomKey(dict);
1649 thiskey = dictGetEntryKey(de);
1650 thisval = (long) dictGetEntryVal(de);
1651
1652 /* Expire sooner (minor expire unix timestamp) is better
1653 * candidate for deletion */
1654 if (bestkey == NULL || thisval < bestval) {
1655 bestkey = thiskey;
1656 bestval = thisval;
1657 }
1658 }
1659 }
1660
1661 /* Finally remove the selected key. */
1662 if (bestkey) {
1663 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
1664 propagateExpire(db,keyobj);
1665 dbDelete(db,keyobj);
1666 server.stat_evictedkeys++;
1667 decrRefCount(keyobj);
1668 freed++;
1669 }
1670 }
1671 if (!freed) return; /* nothing to free... */
1672 }
1673 }
1674
1675 /* =================================== Main! ================================ */
1676
1677 #ifdef __linux__
1678 int linuxOvercommitMemoryValue(void) {
1679 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
1680 char buf[64];
1681
1682 if (!fp) return -1;
1683 if (fgets(buf,64,fp) == NULL) {
1684 fclose(fp);
1685 return -1;
1686 }
1687 fclose(fp);
1688
1689 return atoi(buf);
1690 }
1691
1692 void linuxOvercommitMemoryWarning(void) {
1693 if (linuxOvercommitMemoryValue() == 0) {
1694 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.");
1695 }
1696 }
1697 #endif /* __linux__ */
1698
1699 void createPidFile(void) {
1700 /* Try to write the pid file in a best-effort way. */
1701 FILE *fp = fopen(server.pidfile,"w");
1702 if (fp) {
1703 fprintf(fp,"%d\n",(int)getpid());
1704 fclose(fp);
1705 }
1706 }
1707
1708 void daemonize(void) {
1709 int fd;
1710
1711 if (fork() != 0) exit(0); /* parent exits */
1712 setsid(); /* create a new session */
1713
1714 /* Every output goes to /dev/null. If Redis is daemonized but
1715 * the 'logfile' is set to 'stdout' in the configuration file
1716 * it will not log at all. */
1717 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
1718 dup2(fd, STDIN_FILENO);
1719 dup2(fd, STDOUT_FILENO);
1720 dup2(fd, STDERR_FILENO);
1721 if (fd > STDERR_FILENO) close(fd);
1722 }
1723 }
1724
1725 void version() {
1726 printf("Redis server version %s (%s:%d)\n", REDIS_VERSION,
1727 redisGitSHA1(), atoi(redisGitDirty()) > 0);
1728 exit(0);
1729 }
1730
1731 void usage() {
1732 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf]\n");
1733 fprintf(stderr," ./redis-server - (read config from stdin)\n");
1734 exit(1);
1735 }
1736
1737 void redisAsciiArt(void) {
1738 #include "asciilogo.h"
1739 char *buf = zmalloc(1024*16);
1740
1741 snprintf(buf,1024*16,ascii_logo,
1742 REDIS_VERSION,
1743 redisGitSHA1(),
1744 strtol(redisGitDirty(),NULL,10) > 0,
1745 (sizeof(long) == 8) ? "64" : "32",
1746 server.cluster_enabled ? "cluster" : "stand alone",
1747 server.port,
1748 (long) getpid()
1749 );
1750 redisLogRaw(REDIS_NOTICE|REDIS_LOG_RAW,buf);
1751 zfree(buf);
1752 }
1753
1754 int main(int argc, char **argv) {
1755 long long start;
1756
1757 initServerConfig();
1758 if (argc == 2) {
1759 if (strcmp(argv[1], "-v") == 0 ||
1760 strcmp(argv[1], "--version") == 0) version();
1761 if (strcmp(argv[1], "--help") == 0) usage();
1762 resetServerSaveParams();
1763 loadServerConfig(argv[1]);
1764 } else if ((argc > 2)) {
1765 usage();
1766 } else {
1767 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'");
1768 }
1769 if (server.daemonize) daemonize();
1770 initServer();
1771 if (server.daemonize) createPidFile();
1772 redisAsciiArt();
1773 redisLog(REDIS_NOTICE,"Server started, Redis version " REDIS_VERSION);
1774 #ifdef __linux__
1775 linuxOvercommitMemoryWarning();
1776 #endif
1777 start = ustime();
1778 if (server.ds_enabled) {
1779 redisLog(REDIS_NOTICE,"DB not loaded (running with disk back end)");
1780 } else if (server.appendonly) {
1781 if (loadAppendOnlyFile(server.appendfilename) == REDIS_OK)
1782 redisLog(REDIS_NOTICE,"DB loaded from append only file: %.3f seconds",(float)(ustime()-start)/1000000);
1783 } else {
1784 if (rdbLoad(server.dbfilename) == REDIS_OK)
1785 redisLog(REDIS_NOTICE,"DB loaded from disk: %.3f seconds",(float)(ustime()-start)/1000000);
1786 }
1787 if (server.ipfd > 0)
1788 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
1789 if (server.sofd > 0)
1790 redisLog(REDIS_NOTICE,"The server is now ready to accept connections at %s", server.unixsocket);
1791 aeSetBeforeSleepProc(server.el,beforeSleep);
1792 aeMain(server.el);
1793 aeDeleteEventLoop(server.el);
1794 return 0;
1795 }
1796
1797 #ifdef HAVE_BACKTRACE
1798 static void *getMcontextEip(ucontext_t *uc) {
1799 #if defined(__FreeBSD__)
1800 return (void*) uc->uc_mcontext.mc_eip;
1801 #elif defined(__dietlibc__)
1802 return (void*) uc->uc_mcontext.eip;
1803 #elif defined(__APPLE__) && !defined(MAC_OS_X_VERSION_10_6)
1804 #if __x86_64__
1805 return (void*) uc->uc_mcontext->__ss.__rip;
1806 #else
1807 return (void*) uc->uc_mcontext->__ss.__eip;
1808 #endif
1809 #elif defined(__APPLE__) && defined(MAC_OS_X_VERSION_10_6)
1810 #if defined(_STRUCT_X86_THREAD_STATE64) && !defined(__i386__)
1811 return (void*) uc->uc_mcontext->__ss.__rip;
1812 #else
1813 return (void*) uc->uc_mcontext->__ss.__eip;
1814 #endif
1815 #elif defined(__i386__)
1816 return (void*) uc->uc_mcontext.gregs[14]; /* Linux 32 */
1817 #elif defined(__X86_64__) || defined(__x86_64__)
1818 return (void*) uc->uc_mcontext.gregs[16]; /* Linux 64 */
1819 #elif defined(__ia64__) /* Linux IA64 */
1820 return (void*) uc->uc_mcontext.sc_ip;
1821 #else
1822 return NULL;
1823 #endif
1824 }
1825
1826 static void sigsegvHandler(int sig, siginfo_t *info, void *secret) {
1827 void *trace[100];
1828 char **messages = NULL;
1829 int i, trace_size = 0;
1830 ucontext_t *uc = (ucontext_t*) secret;
1831 sds infostring;
1832 struct sigaction act;
1833 REDIS_NOTUSED(info);
1834
1835 redisLog(REDIS_WARNING,
1836 "======= Ooops! Redis %s got signal: -%d- =======", REDIS_VERSION, sig);
1837 infostring = genRedisInfoString("all");
1838 redisLogRaw(REDIS_WARNING, infostring);
1839 /* It's not safe to sdsfree() the returned string under memory
1840 * corruption conditions. Let it leak as we are going to abort */
1841
1842 trace_size = backtrace(trace, 100);
1843 /* overwrite sigaction with caller's address */
1844 if (getMcontextEip(uc) != NULL) {
1845 trace[1] = getMcontextEip(uc);
1846 }
1847 messages = backtrace_symbols(trace, trace_size);
1848
1849 for (i=1; i<trace_size; ++i)
1850 redisLog(REDIS_WARNING,"%s", messages[i]);
1851
1852 /* free(messages); Don't call free() with possibly corrupted memory. */
1853 if (server.daemonize) unlink(server.pidfile);
1854
1855 /* Make sure we exit with the right signal at the end. So for instance
1856 * the core will be dumped if enabled. */
1857 sigemptyset (&act.sa_mask);
1858 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction
1859 * is used. Otherwise, sa_handler is used */
1860 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1861 act.sa_handler = SIG_DFL;
1862 sigaction (sig, &act, NULL);
1863 kill(getpid(),sig);
1864 }
1865 #endif /* HAVE_BACKTRACE */
1866
1867 static void sigtermHandler(int sig) {
1868 REDIS_NOTUSED(sig);
1869
1870 redisLog(REDIS_WARNING,"Received SIGTERM, scheduling shutdown...");
1871 server.shutdown_asap = 1;
1872 }
1873
1874 void setupSignalHandlers(void) {
1875 struct sigaction act;
1876
1877 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction is used.
1878 * Otherwise, sa_handler is used. */
1879 sigemptyset(&act.sa_mask);
1880 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND;
1881 act.sa_handler = sigtermHandler;
1882 sigaction(SIGTERM, &act, NULL);
1883
1884 #ifdef HAVE_BACKTRACE
1885 sigemptyset(&act.sa_mask);
1886 act.sa_flags = SA_NODEFER | SA_ONSTACK | SA_RESETHAND | SA_SIGINFO;
1887 act.sa_sigaction = sigsegvHandler;
1888 sigaction(SIGSEGV, &act, NULL);
1889 sigaction(SIGBUS, &act, NULL);
1890 sigaction(SIGFPE, &act, NULL);
1891 sigaction(SIGILL, &act, NULL);
1892 #endif
1893 return;
1894 }
1895
1896 /* The End */