]> git.saurik.com Git - redis.git/blob - src/redis.c
BITOP command 10x speed improvement.
[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 #include "slowlog.h"
32 #include "bio.h"
33
34 #include <time.h>
35 #include <signal.h>
36 #include <sys/wait.h>
37 #include <errno.h>
38 #include <assert.h>
39 #include <ctype.h>
40 #include <stdarg.h>
41 #include <arpa/inet.h>
42 #include <sys/stat.h>
43 #include <fcntl.h>
44 #include <sys/time.h>
45 #include <sys/resource.h>
46 #include <sys/uio.h>
47 #include <limits.h>
48 #include <float.h>
49 #include <math.h>
50 #include <sys/resource.h>
51 #include <sys/utsname.h>
52
53 /* Our shared "common" objects */
54
55 struct sharedObjectsStruct shared;
56
57 /* Global vars that are actually used as constants. The following double
58 * values are used for double on-disk serialization, and are initialized
59 * at runtime to avoid strange compiler optimizations. */
60
61 double R_Zero, R_PosInf, R_NegInf, R_Nan;
62
63 /*================================= Globals ================================= */
64
65 /* Global vars */
66 struct redisServer server; /* server global state */
67 struct redisCommand *commandTable;
68
69 /* Our command table.
70 *
71 * Every entry is composed of the following fields:
72 *
73 * name: a string representing the command name.
74 * function: pointer to the C function implementing the command.
75 * arity: number of arguments, it is possible to use -N to say >= N
76 * sflags: command flags as string. See below for a table of flags.
77 * flags: flags as bitmask. Computed by Redis using the 'sflags' field.
78 * get_keys_proc: an optional function to get key arguments from a command.
79 * This is only used when the following three fields are not
80 * enough to specify what arguments are keys.
81 * first_key_index: first argument that is a key
82 * last_key_index: last argument that is a key
83 * key_step: step to get all the keys from first to last argument. For instance
84 * in MSET the step is two since arguments are key,val,key,val,...
85 * microseconds: microseconds of total execution time for this command.
86 * calls: total number of calls of this command.
87 *
88 * The flags, microseconds and calls fields are computed by Redis and should
89 * always be set to zero.
90 *
91 * Command flags are expressed using strings where every character represents
92 * a flag. Later the populateCommandTable() function will take care of
93 * populating the real 'flags' field using this characters.
94 *
95 * This is the meaning of the flags:
96 *
97 * w: write command (may modify the key space).
98 * r: read command (will never modify the key space).
99 * m: may increase memory usage once called. Don't allow if out of memory.
100 * a: admin command, like SAVE or SHUTDOWN.
101 * p: Pub/Sub related command.
102 * f: force replication of this command, regarless of server.dirty.
103 * s: command not allowed in scripts.
104 * R: random command. Command is not deterministic, that is, the same command
105 * with the same arguments, with the same key space, may have different
106 * results. For instance SPOP and RANDOMKEY are two random commands.
107 * S: Sort command output array if called from script, so that the output
108 * is deterministic.
109 */
110 struct redisCommand redisCommandTable[] = {
111 {"get",getCommand,2,"r",0,NULL,1,1,1,0,0},
112 {"set",setCommand,3,"wm",0,noPreloadGetKeys,1,1,1,0,0},
113 {"setnx",setnxCommand,3,"wm",0,noPreloadGetKeys,1,1,1,0,0},
114 {"setex",setexCommand,4,"wm",0,noPreloadGetKeys,1,1,1,0,0},
115 {"psetex",psetexCommand,4,"wm",0,noPreloadGetKeys,1,1,1,0,0},
116 {"append",appendCommand,3,"wm",0,NULL,1,1,1,0,0},
117 {"strlen",strlenCommand,2,"r",0,NULL,1,1,1,0,0},
118 {"del",delCommand,-2,"w",0,noPreloadGetKeys,1,-1,1,0,0},
119 {"exists",existsCommand,2,"r",0,NULL,1,1,1,0,0},
120 {"setbit",setbitCommand,4,"wm",0,NULL,1,1,1,0,0},
121 {"getbit",getbitCommand,3,"r",0,NULL,1,1,1,0,0},
122 {"setrange",setrangeCommand,4,"wm",0,NULL,1,1,1,0,0},
123 {"getrange",getrangeCommand,4,"r",0,NULL,1,1,1,0,0},
124 {"substr",getrangeCommand,4,"r",0,NULL,1,1,1,0,0},
125 {"incr",incrCommand,2,"wm",0,NULL,1,1,1,0,0},
126 {"decr",decrCommand,2,"wm",0,NULL,1,1,1,0,0},
127 {"mget",mgetCommand,-2,"r",0,NULL,1,-1,1,0,0},
128 {"rpush",rpushCommand,-3,"wm",0,NULL,1,1,1,0,0},
129 {"lpush",lpushCommand,-3,"wm",0,NULL,1,1,1,0,0},
130 {"rpushx",rpushxCommand,3,"wm",0,NULL,1,1,1,0,0},
131 {"lpushx",lpushxCommand,3,"wm",0,NULL,1,1,1,0,0},
132 {"linsert",linsertCommand,5,"wm",0,NULL,1,1,1,0,0},
133 {"rpop",rpopCommand,2,"w",0,NULL,1,1,1,0,0},
134 {"lpop",lpopCommand,2,"w",0,NULL,1,1,1,0,0},
135 {"brpop",brpopCommand,-3,"ws",0,NULL,1,1,1,0,0},
136 {"brpoplpush",brpoplpushCommand,4,"wms",0,NULL,1,2,1,0,0},
137 {"blpop",blpopCommand,-3,"ws",0,NULL,1,-2,1,0,0},
138 {"llen",llenCommand,2,"r",0,NULL,1,1,1,0,0},
139 {"lindex",lindexCommand,3,"r",0,NULL,1,1,1,0,0},
140 {"lset",lsetCommand,4,"wm",0,NULL,1,1,1,0,0},
141 {"lrange",lrangeCommand,4,"r",0,NULL,1,1,1,0,0},
142 {"ltrim",ltrimCommand,4,"w",0,NULL,1,1,1,0,0},
143 {"lrem",lremCommand,4,"w",0,NULL,1,1,1,0,0},
144 {"rpoplpush",rpoplpushCommand,3,"wm",0,NULL,1,2,1,0,0},
145 {"sadd",saddCommand,-3,"wm",0,NULL,1,1,1,0,0},
146 {"srem",sremCommand,-3,"w",0,NULL,1,1,1,0,0},
147 {"smove",smoveCommand,4,"w",0,NULL,1,2,1,0,0},
148 {"sismember",sismemberCommand,3,"r",0,NULL,1,1,1,0,0},
149 {"scard",scardCommand,2,"r",0,NULL,1,1,1,0,0},
150 {"spop",spopCommand,2,"wRs",0,NULL,1,1,1,0,0},
151 {"srandmember",srandmemberCommand,2,"rR",0,NULL,1,1,1,0,0},
152 {"sinter",sinterCommand,-2,"rS",0,NULL,1,-1,1,0,0},
153 {"sinterstore",sinterstoreCommand,-3,"wm",0,NULL,1,-1,1,0,0},
154 {"sunion",sunionCommand,-2,"rS",0,NULL,1,-1,1,0,0},
155 {"sunionstore",sunionstoreCommand,-3,"wm",0,NULL,1,-1,1,0,0},
156 {"sdiff",sdiffCommand,-2,"rS",0,NULL,1,-1,1,0,0},
157 {"sdiffstore",sdiffstoreCommand,-3,"wm",0,NULL,1,-1,1,0,0},
158 {"smembers",sinterCommand,2,"rS",0,NULL,1,1,1,0,0},
159 {"zadd",zaddCommand,-4,"wm",0,NULL,1,1,1,0,0},
160 {"zincrby",zincrbyCommand,4,"wm",0,NULL,1,1,1,0,0},
161 {"zrem",zremCommand,-3,"w",0,NULL,1,1,1,0,0},
162 {"zremrangebyscore",zremrangebyscoreCommand,4,"w",0,NULL,1,1,1,0,0},
163 {"zremrangebyrank",zremrangebyrankCommand,4,"w",0,NULL,1,1,1,0,0},
164 {"zunionstore",zunionstoreCommand,-4,"wm",0,zunionInterGetKeys,0,0,0,0,0},
165 {"zinterstore",zinterstoreCommand,-4,"wm",0,zunionInterGetKeys,0,0,0,0,0},
166 {"zrange",zrangeCommand,-4,"r",0,NULL,1,1,1,0,0},
167 {"zrangebyscore",zrangebyscoreCommand,-4,"r",0,NULL,1,1,1,0,0},
168 {"zrevrangebyscore",zrevrangebyscoreCommand,-4,"r",0,NULL,1,1,1,0,0},
169 {"zcount",zcountCommand,4,"r",0,NULL,1,1,1,0,0},
170 {"zrevrange",zrevrangeCommand,-4,"r",0,NULL,1,1,1,0,0},
171 {"zcard",zcardCommand,2,"r",0,NULL,1,1,1,0,0},
172 {"zscore",zscoreCommand,3,"r",0,NULL,1,1,1,0,0},
173 {"zrank",zrankCommand,3,"r",0,NULL,1,1,1,0,0},
174 {"zrevrank",zrevrankCommand,3,"r",0,NULL,1,1,1,0,0},
175 {"hset",hsetCommand,4,"wm",0,NULL,1,1,1,0,0},
176 {"hsetnx",hsetnxCommand,4,"wm",0,NULL,1,1,1,0,0},
177 {"hget",hgetCommand,3,"r",0,NULL,1,1,1,0,0},
178 {"hmset",hmsetCommand,-4,"wm",0,NULL,1,1,1,0,0},
179 {"hmget",hmgetCommand,-3,"r",0,NULL,1,1,1,0,0},
180 {"hincrby",hincrbyCommand,4,"wm",0,NULL,1,1,1,0,0},
181 {"hincrbyfloat",hincrbyfloatCommand,4,"wm",0,NULL,1,1,1,0,0},
182 {"hdel",hdelCommand,-3,"w",0,NULL,1,1,1,0,0},
183 {"hlen",hlenCommand,2,"r",0,NULL,1,1,1,0,0},
184 {"hkeys",hkeysCommand,2,"rS",0,NULL,1,1,1,0,0},
185 {"hvals",hvalsCommand,2,"rS",0,NULL,1,1,1,0,0},
186 {"hgetall",hgetallCommand,2,"r",0,NULL,1,1,1,0,0},
187 {"hexists",hexistsCommand,3,"r",0,NULL,1,1,1,0,0},
188 {"incrby",incrbyCommand,3,"wm",0,NULL,1,1,1,0,0},
189 {"decrby",decrbyCommand,3,"wm",0,NULL,1,1,1,0,0},
190 {"incrbyfloat",incrbyfloatCommand,3,"wm",0,NULL,1,1,1,0,0},
191 {"getset",getsetCommand,3,"wm",0,NULL,1,1,1,0,0},
192 {"mset",msetCommand,-3,"wm",0,NULL,1,-1,2,0,0},
193 {"msetnx",msetnxCommand,-3,"wm",0,NULL,1,-1,2,0,0},
194 {"randomkey",randomkeyCommand,1,"rR",0,NULL,0,0,0,0,0},
195 {"select",selectCommand,2,"r",0,NULL,0,0,0,0,0},
196 {"move",moveCommand,3,"w",0,NULL,1,1,1,0,0},
197 {"rename",renameCommand,3,"w",0,renameGetKeys,1,2,1,0,0},
198 {"renamenx",renamenxCommand,3,"w",0,renameGetKeys,1,2,1,0,0},
199 {"expire",expireCommand,3,"w",0,NULL,1,1,1,0,0},
200 {"expireat",expireatCommand,3,"w",0,NULL,1,1,1,0,0},
201 {"pexpire",pexpireCommand,3,"w",0,NULL,1,1,1,0,0},
202 {"pexpireat",pexpireatCommand,3,"w",0,NULL,1,1,1,0,0},
203 {"keys",keysCommand,2,"rS",0,NULL,0,0,0,0,0},
204 {"dbsize",dbsizeCommand,1,"r",0,NULL,0,0,0,0,0},
205 {"auth",authCommand,2,"rs",0,NULL,0,0,0,0,0},
206 {"ping",pingCommand,1,"r",0,NULL,0,0,0,0,0},
207 {"echo",echoCommand,2,"r",0,NULL,0,0,0,0,0},
208 {"save",saveCommand,1,"ars",0,NULL,0,0,0,0,0},
209 {"bgsave",bgsaveCommand,1,"ar",0,NULL,0,0,0,0,0},
210 {"bgrewriteaof",bgrewriteaofCommand,1,"ar",0,NULL,0,0,0,0,0},
211 {"shutdown",shutdownCommand,-1,"ar",0,NULL,0,0,0,0,0},
212 {"lastsave",lastsaveCommand,1,"r",0,NULL,0,0,0,0,0},
213 {"type",typeCommand,2,"r",0,NULL,1,1,1,0,0},
214 {"multi",multiCommand,1,"rs",0,NULL,0,0,0,0,0},
215 {"exec",execCommand,1,"s",0,NULL,0,0,0,0,0},
216 {"discard",discardCommand,1,"rs",0,NULL,0,0,0,0,0},
217 {"sync",syncCommand,1,"ars",0,NULL,0,0,0,0,0},
218 {"flushdb",flushdbCommand,1,"w",0,NULL,0,0,0,0,0},
219 {"flushall",flushallCommand,1,"w",0,NULL,0,0,0,0,0},
220 {"sort",sortCommand,-2,"wmS",0,NULL,1,1,1,0,0},
221 {"info",infoCommand,-1,"r",0,NULL,0,0,0,0,0},
222 {"monitor",monitorCommand,1,"ars",0,NULL,0,0,0,0,0},
223 {"ttl",ttlCommand,2,"r",0,NULL,1,1,1,0,0},
224 {"pttl",pttlCommand,2,"r",0,NULL,1,1,1,0,0},
225 {"persist",persistCommand,2,"w",0,NULL,1,1,1,0,0},
226 {"slaveof",slaveofCommand,3,"as",0,NULL,0,0,0,0,0},
227 {"debug",debugCommand,-2,"as",0,NULL,0,0,0,0,0},
228 {"config",configCommand,-2,"ar",0,NULL,0,0,0,0,0},
229 {"subscribe",subscribeCommand,-2,"rps",0,NULL,0,0,0,0,0},
230 {"unsubscribe",unsubscribeCommand,-1,"rps",0,NULL,0,0,0,0,0},
231 {"psubscribe",psubscribeCommand,-2,"rps",0,NULL,0,0,0,0,0},
232 {"punsubscribe",punsubscribeCommand,-1,"rps",0,NULL,0,0,0,0,0},
233 {"publish",publishCommand,3,"pf",0,NULL,0,0,0,0,0},
234 {"watch",watchCommand,-2,"rs",0,noPreloadGetKeys,1,-1,1,0,0},
235 {"unwatch",unwatchCommand,1,"rs",0,NULL,0,0,0,0,0},
236 {"restore",restoreCommand,4,"awm",0,NULL,1,1,1,0,0},
237 {"migrate",migrateCommand,6,"aw",0,NULL,0,0,0,0,0},
238 {"dump",dumpCommand,2,"ar",0,NULL,1,1,1,0,0},
239 {"object",objectCommand,-2,"r",0,NULL,2,2,2,0,0},
240 {"client",clientCommand,-2,"ar",0,NULL,0,0,0,0,0},
241 {"eval",evalCommand,-3,"s",0,zunionInterGetKeys,0,0,0,0,0},
242 {"evalsha",evalShaCommand,-3,"s",0,zunionInterGetKeys,0,0,0,0,0},
243 {"slowlog",slowlogCommand,-2,"r",0,NULL,0,0,0,0,0},
244 {"script",scriptCommand,-2,"ras",0,NULL,0,0,0,0,0},
245 {"time",timeCommand,1,"rR",0,NULL,0,0,0,0,0},
246 {"bitop",bitopCommand,-4,"wm",0,NULL,2,-1,1,0,0},
247 {"bitcount",bitcountCommand,-2,"r",0,NULL,1,1,1,0,0}
248 };
249
250 /*============================ Utility functions ============================ */
251
252 /* Low level logging. To use only for very big messages, otherwise
253 * redisLog() is to prefer. */
254 void redisLogRaw(int level, const char *msg) {
255 const int syslogLevelMap[] = { LOG_DEBUG, LOG_INFO, LOG_NOTICE, LOG_WARNING };
256 const char *c = ".-*#";
257 FILE *fp;
258 char buf[64];
259 int rawmode = (level & REDIS_LOG_RAW);
260
261 level &= 0xff; /* clear flags */
262 if (level < server.verbosity) return;
263
264 fp = (server.logfile == NULL) ? stdout : fopen(server.logfile,"a");
265 if (!fp) return;
266
267 if (rawmode) {
268 fprintf(fp,"%s",msg);
269 } else {
270 int off;
271 struct timeval tv;
272
273 gettimeofday(&tv,NULL);
274 off = strftime(buf,sizeof(buf),"%d %b %H:%M:%S.",localtime(&tv.tv_sec));
275 snprintf(buf+off,sizeof(buf)-off,"%03d",(int)tv.tv_usec/1000);
276 fprintf(fp,"[%d] %s %c %s\n",(int)getpid(),buf,c[level],msg);
277 }
278 fflush(fp);
279
280 if (server.logfile) fclose(fp);
281
282 if (server.syslog_enabled) syslog(syslogLevelMap[level], "%s", msg);
283 }
284
285 /* Like redisLogRaw() but with printf-alike support. This is the funciton that
286 * is used across the code. The raw version is only used in order to dump
287 * the INFO output on crash. */
288 void redisLog(int level, const char *fmt, ...) {
289 va_list ap;
290 char msg[REDIS_MAX_LOGMSG_LEN];
291
292 if ((level&0xff) < server.verbosity) return;
293
294 va_start(ap, fmt);
295 vsnprintf(msg, sizeof(msg), fmt, ap);
296 va_end(ap);
297
298 redisLogRaw(level,msg);
299 }
300
301 /* Log a fixed message without printf-alike capabilities, in a way that is
302 * safe to call from a signal handler.
303 *
304 * We actually use this only for signals that are not fatal from the point
305 * of view of Redis. Signals that are going to kill the server anyway and
306 * where we need printf-alike features are served by redisLog(). */
307 void redisLogFromHandler(int level, const char *msg) {
308 int fd;
309 char buf[64];
310
311 if ((level&0xff) < server.verbosity ||
312 (server.logfile == NULL && server.daemonize)) return;
313 fd = server.logfile ?
314 open(server.logfile, O_APPEND|O_CREAT|O_WRONLY, 0644) :
315 STDOUT_FILENO;
316 if (fd == -1) return;
317 ll2string(buf,sizeof(buf),getpid());
318 if (write(fd,"[",1) == -1) goto err;
319 if (write(fd,buf,strlen(buf)) == -1) goto err;
320 if (write(fd," | signal handler] (",20) == -1) goto err;
321 ll2string(buf,sizeof(buf),time(NULL));
322 if (write(fd,buf,strlen(buf)) == -1) goto err;
323 if (write(fd,") ",2) == -1) goto err;
324 if (write(fd,msg,strlen(msg)) == -1) goto err;
325 if (write(fd,"\n",1) == -1) goto err;
326 err:
327 if (server.logfile) close(fd);
328 }
329
330 /* Redis generally does not try to recover from out of memory conditions
331 * when allocating objects or strings, it is not clear if it will be possible
332 * to report this condition to the client since the networking layer itself
333 * is based on heap allocation for send buffers, so we simply abort.
334 * At least the code will be simpler to read... */
335 void oom(const char *msg) {
336 redisLog(REDIS_WARNING, "%s: Out of memory\n",msg);
337 sleep(1);
338 abort();
339 }
340
341 /* Return the UNIX time in microseconds */
342 long long ustime(void) {
343 struct timeval tv;
344 long long ust;
345
346 gettimeofday(&tv, NULL);
347 ust = ((long long)tv.tv_sec)*1000000;
348 ust += tv.tv_usec;
349 return ust;
350 }
351
352 /* Return the UNIX time in milliseconds */
353 long long mstime(void) {
354 return ustime()/1000;
355 }
356
357 /* After an RDB dump or AOF rewrite we exit from children using _exit() instead of
358 * exit(), because the latter may interact with the same file objects used by
359 * the parent process. However if we are testing the coverage normal exit() is
360 * used in order to obtain the right coverage information. */
361 void exitFromChild(int retcode) {
362 #ifdef COVERAGE_TEST
363 exit(retcode);
364 #else
365 _exit(retcode);
366 #endif
367 }
368
369 /*====================== Hash table type implementation ==================== */
370
371 /* This is an hash table type that uses the SDS dynamic strings libary as
372 * keys and radis objects as values (objects can hold SDS strings,
373 * lists, sets). */
374
375 void dictVanillaFree(void *privdata, void *val)
376 {
377 DICT_NOTUSED(privdata);
378 zfree(val);
379 }
380
381 void dictListDestructor(void *privdata, void *val)
382 {
383 DICT_NOTUSED(privdata);
384 listRelease((list*)val);
385 }
386
387 int dictSdsKeyCompare(void *privdata, const void *key1,
388 const void *key2)
389 {
390 int l1,l2;
391 DICT_NOTUSED(privdata);
392
393 l1 = sdslen((sds)key1);
394 l2 = sdslen((sds)key2);
395 if (l1 != l2) return 0;
396 return memcmp(key1, key2, l1) == 0;
397 }
398
399 /* A case insensitive version used for the command lookup table. */
400 int dictSdsKeyCaseCompare(void *privdata, const void *key1,
401 const void *key2)
402 {
403 DICT_NOTUSED(privdata);
404
405 return strcasecmp(key1, key2) == 0;
406 }
407
408 void dictRedisObjectDestructor(void *privdata, void *val)
409 {
410 DICT_NOTUSED(privdata);
411
412 if (val == NULL) return; /* Values of swapped out keys as set to NULL */
413 decrRefCount(val);
414 }
415
416 void dictSdsDestructor(void *privdata, void *val)
417 {
418 DICT_NOTUSED(privdata);
419
420 sdsfree(val);
421 }
422
423 int dictObjKeyCompare(void *privdata, const void *key1,
424 const void *key2)
425 {
426 const robj *o1 = key1, *o2 = key2;
427 return dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
428 }
429
430 unsigned int dictObjHash(const void *key) {
431 const robj *o = key;
432 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
433 }
434
435 unsigned int dictSdsHash(const void *key) {
436 return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
437 }
438
439 unsigned int dictSdsCaseHash(const void *key) {
440 return dictGenCaseHashFunction((unsigned char*)key, sdslen((char*)key));
441 }
442
443 int dictEncObjKeyCompare(void *privdata, const void *key1,
444 const void *key2)
445 {
446 robj *o1 = (robj*) key1, *o2 = (robj*) key2;
447 int cmp;
448
449 if (o1->encoding == REDIS_ENCODING_INT &&
450 o2->encoding == REDIS_ENCODING_INT)
451 return o1->ptr == o2->ptr;
452
453 o1 = getDecodedObject(o1);
454 o2 = getDecodedObject(o2);
455 cmp = dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
456 decrRefCount(o1);
457 decrRefCount(o2);
458 return cmp;
459 }
460
461 unsigned int dictEncObjHash(const void *key) {
462 robj *o = (robj*) key;
463
464 if (o->encoding == REDIS_ENCODING_RAW) {
465 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
466 } else {
467 if (o->encoding == REDIS_ENCODING_INT) {
468 char buf[32];
469 int len;
470
471 len = ll2string(buf,32,(long)o->ptr);
472 return dictGenHashFunction((unsigned char*)buf, len);
473 } else {
474 unsigned int hash;
475
476 o = getDecodedObject(o);
477 hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
478 decrRefCount(o);
479 return hash;
480 }
481 }
482 }
483
484 /* Sets type hash table */
485 dictType setDictType = {
486 dictEncObjHash, /* hash function */
487 NULL, /* key dup */
488 NULL, /* val dup */
489 dictEncObjKeyCompare, /* key compare */
490 dictRedisObjectDestructor, /* key destructor */
491 NULL /* val destructor */
492 };
493
494 /* Sorted sets hash (note: a skiplist is used in addition to the hash table) */
495 dictType zsetDictType = {
496 dictEncObjHash, /* hash function */
497 NULL, /* key dup */
498 NULL, /* val dup */
499 dictEncObjKeyCompare, /* key compare */
500 dictRedisObjectDestructor, /* key destructor */
501 NULL /* val destructor */
502 };
503
504 /* Db->dict, keys are sds strings, vals are Redis objects. */
505 dictType dbDictType = {
506 dictSdsHash, /* hash function */
507 NULL, /* key dup */
508 NULL, /* val dup */
509 dictSdsKeyCompare, /* key compare */
510 dictSdsDestructor, /* key destructor */
511 dictRedisObjectDestructor /* val destructor */
512 };
513
514 /* Db->expires */
515 dictType keyptrDictType = {
516 dictSdsHash, /* hash function */
517 NULL, /* key dup */
518 NULL, /* val dup */
519 dictSdsKeyCompare, /* key compare */
520 NULL, /* key destructor */
521 NULL /* val destructor */
522 };
523
524 /* Command table. sds string -> command struct pointer. */
525 dictType commandTableDictType = {
526 dictSdsCaseHash, /* hash function */
527 NULL, /* key dup */
528 NULL, /* val dup */
529 dictSdsKeyCaseCompare, /* key compare */
530 dictSdsDestructor, /* key destructor */
531 NULL /* val destructor */
532 };
533
534 /* Hash type hash table (note that small hashes are represented with zimpaps) */
535 dictType hashDictType = {
536 dictEncObjHash, /* hash function */
537 NULL, /* key dup */
538 NULL, /* val dup */
539 dictEncObjKeyCompare, /* key compare */
540 dictRedisObjectDestructor, /* key destructor */
541 dictRedisObjectDestructor /* val destructor */
542 };
543
544 /* Keylist hash table type has unencoded redis objects as keys and
545 * lists as values. It's used for blocking operations (BLPOP) and to
546 * map swapped keys to a list of clients waiting for this keys to be loaded. */
547 dictType keylistDictType = {
548 dictObjHash, /* hash function */
549 NULL, /* key dup */
550 NULL, /* val dup */
551 dictObjKeyCompare, /* key compare */
552 dictRedisObjectDestructor, /* key destructor */
553 dictListDestructor /* val destructor */
554 };
555
556 int htNeedsResize(dict *dict) {
557 long long size, used;
558
559 size = dictSlots(dict);
560 used = dictSize(dict);
561 return (size && used && size > DICT_HT_INITIAL_SIZE &&
562 (used*100/size < REDIS_HT_MINFILL));
563 }
564
565 /* If the percentage of used slots in the HT reaches REDIS_HT_MINFILL
566 * we resize the hash table to save memory */
567 void tryResizeHashTables(void) {
568 int j;
569
570 for (j = 0; j < server.dbnum; j++) {
571 if (htNeedsResize(server.db[j].dict))
572 dictResize(server.db[j].dict);
573 if (htNeedsResize(server.db[j].expires))
574 dictResize(server.db[j].expires);
575 }
576 }
577
578 /* Our hash table implementation performs rehashing incrementally while
579 * we write/read from the hash table. Still if the server is idle, the hash
580 * table will use two tables for a long time. So we try to use 1 millisecond
581 * of CPU time at every serverCron() loop in order to rehash some key. */
582 void incrementallyRehash(void) {
583 int j;
584
585 for (j = 0; j < server.dbnum; j++) {
586 /* Keys dictionary */
587 if (dictIsRehashing(server.db[j].dict)) {
588 dictRehashMilliseconds(server.db[j].dict,1);
589 break; /* already used our millisecond for this loop... */
590 }
591 /* Expires */
592 if (dictIsRehashing(server.db[j].expires)) {
593 dictRehashMilliseconds(server.db[j].expires,1);
594 break; /* already used our millisecond for this loop... */
595 }
596 }
597 }
598
599 /* This function is called once a background process of some kind terminates,
600 * as we want to avoid resizing the hash tables when there is a child in order
601 * to play well with copy-on-write (otherwise when a resize happens lots of
602 * memory pages are copied). The goal of this function is to update the ability
603 * for dict.c to resize the hash tables accordingly to the fact we have o not
604 * running childs. */
605 void updateDictResizePolicy(void) {
606 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1)
607 dictEnableResize();
608 else
609 dictDisableResize();
610 }
611
612 /* ======================= Cron: called every 100 ms ======================== */
613
614 /* Try to expire a few timed out keys. The algorithm used is adaptive and
615 * will use few CPU cycles if there are few expiring keys, otherwise
616 * it will get more aggressive to avoid that too much memory is used by
617 * keys that can be removed from the keyspace. */
618 void activeExpireCycle(void) {
619 int j, iteration = 0;
620 long long start = ustime(), timelimit;
621
622 /* We can use at max REDIS_EXPIRELOOKUPS_TIME_PERC percentage of CPU time
623 * per iteration. Since this function gets called with a frequency of
624 * REDIS_HZ times per second, the following is the max amount of
625 * microseconds we can spend in this function. */
626 timelimit = 1000000*REDIS_EXPIRELOOKUPS_TIME_PERC/REDIS_HZ/100;
627 if (timelimit <= 0) timelimit = 1;
628
629 for (j = 0; j < server.dbnum; j++) {
630 int expired;
631 redisDb *db = server.db+j;
632
633 /* Continue to expire if at the end of the cycle more than 25%
634 * of the keys were expired. */
635 do {
636 unsigned long num = dictSize(db->expires);
637 unsigned long slots = dictSlots(db->expires);
638 long long now = mstime();
639
640 /* When there are less than 1% filled slots getting random
641 * keys is expensive, so stop here waiting for better times...
642 * The dictionary will be resized asap. */
643 if (num && slots > DICT_HT_INITIAL_SIZE &&
644 (num*100/slots < 1)) break;
645
646 /* The main collection cycle. Sample random keys among keys
647 * with an expire set, checking for expired ones. */
648 expired = 0;
649 if (num > REDIS_EXPIRELOOKUPS_PER_CRON)
650 num = REDIS_EXPIRELOOKUPS_PER_CRON;
651 while (num--) {
652 dictEntry *de;
653 long long t;
654
655 if ((de = dictGetRandomKey(db->expires)) == NULL) break;
656 t = dictGetSignedIntegerVal(de);
657 if (now > t) {
658 sds key = dictGetKey(de);
659 robj *keyobj = createStringObject(key,sdslen(key));
660
661 propagateExpire(db,keyobj);
662 dbDelete(db,keyobj);
663 decrRefCount(keyobj);
664 expired++;
665 server.stat_expiredkeys++;
666 }
667 }
668 /* We can't block forever here even if there are many keys to
669 * expire. So after a given amount of milliseconds return to the
670 * caller waiting for the other active expire cycle. */
671 iteration++;
672 if ((iteration & 0xf) == 0 && /* check once every 16 cycles. */
673 (ustime()-start) > timelimit) return;
674 } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4);
675 }
676 }
677
678 void updateLRUClock(void) {
679 server.lruclock = (server.unixtime/REDIS_LRU_CLOCK_RESOLUTION) &
680 REDIS_LRU_CLOCK_MAX;
681 }
682
683
684 /* Add a sample to the operations per second array of samples. */
685 void trackOperationsPerSecond(void) {
686 long long t = mstime() - server.ops_sec_last_sample_time;
687 long long ops = server.stat_numcommands - server.ops_sec_last_sample_ops;
688 long long ops_sec;
689
690 ops_sec = t > 0 ? (ops*1000/t) : 0;
691
692 server.ops_sec_samples[server.ops_sec_idx] = ops_sec;
693 server.ops_sec_idx = (server.ops_sec_idx+1) % REDIS_OPS_SEC_SAMPLES;
694 server.ops_sec_last_sample_time = mstime();
695 server.ops_sec_last_sample_ops = server.stat_numcommands;
696 }
697
698 /* Return the mean of all the samples. */
699 long long getOperationsPerSecond(void) {
700 int j;
701 long long sum = 0;
702
703 for (j = 0; j < REDIS_OPS_SEC_SAMPLES; j++)
704 sum += server.ops_sec_samples[j];
705 return sum / REDIS_OPS_SEC_SAMPLES;
706 }
707
708 /* Check for timeouts. Returns non-zero if the client was terminated */
709 int clientsCronHandleTimeout(redisClient *c) {
710 time_t now = server.unixtime;
711
712 if (server.maxidletime &&
713 !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */
714 !(c->flags & REDIS_MASTER) && /* no timeout for masters */
715 !(c->flags & REDIS_BLOCKED) && /* no timeout for BLPOP */
716 dictSize(c->pubsub_channels) == 0 && /* no timeout for pubsub */
717 listLength(c->pubsub_patterns) == 0 &&
718 (now - c->lastinteraction > server.maxidletime))
719 {
720 redisLog(REDIS_VERBOSE,"Closing idle client");
721 freeClient(c);
722 return 1;
723 } else if (c->flags & REDIS_BLOCKED) {
724 if (c->bpop.timeout != 0 && c->bpop.timeout < now) {
725 addReply(c,shared.nullmultibulk);
726 unblockClientWaitingData(c);
727 }
728 }
729 return 0;
730 }
731
732 /* The client query buffer is an sds.c string that can end with a lot of
733 * free space not used, this function reclaims space if needed.
734 *
735 * The funciton always returns 0 as it never terminates the client. */
736 int clientsCronResizeQueryBuffer(redisClient *c) {
737 size_t querybuf_size = sdsAllocSize(c->querybuf);
738 time_t idletime = server.unixtime - c->lastinteraction;
739
740 /* There are two conditions to resize the query buffer:
741 * 1) Query buffer is > BIG_ARG and too big for latest peak.
742 * 2) Client is inactive and the buffer is bigger than 1k. */
743 if (((querybuf_size > REDIS_MBULK_BIG_ARG) &&
744 (querybuf_size/(c->querybuf_peak+1)) > 2) ||
745 (querybuf_size > 1024 && idletime > 2))
746 {
747 /* Only resize the query buffer if it is actually wasting space. */
748 if (sdsavail(c->querybuf) > 1024) {
749 c->querybuf = sdsRemoveFreeSpace(c->querybuf);
750 }
751 }
752 /* Reset the peak again to capture the peak memory usage in the next
753 * cycle. */
754 c->querybuf_peak = 0;
755 return 0;
756 }
757
758 void clientsCron(void) {
759 /* Make sure to process at least 1/(REDIS_HZ*10) of clients per call.
760 * Since this function is called REDIS_HZ times per second we are sure that
761 * in the worst case we process all the clients in 10 seconds.
762 * In normal conditions (a reasonable number of clients) we process
763 * all the clients in a shorter time. */
764 int numclients = listLength(server.clients);
765 int iterations = numclients/(REDIS_HZ*10);
766
767 if (iterations < 50)
768 iterations = (numclients < 50) ? numclients : 50;
769 while(listLength(server.clients) && iterations--) {
770 redisClient *c;
771 listNode *head;
772
773 /* Rotate the list, take the current head, process.
774 * This way if the client must be removed from the list it's the
775 * first element and we don't incur into O(N) computation. */
776 listRotate(server.clients);
777 head = listFirst(server.clients);
778 c = listNodeValue(head);
779 /* The following functions do different service checks on the client.
780 * The protocol is that they return non-zero if the client was
781 * terminated. */
782 if (clientsCronHandleTimeout(c)) continue;
783 if (clientsCronResizeQueryBuffer(c)) continue;
784 }
785 }
786
787 /* This is our timer interrupt, called REDIS_HZ times per second.
788 * Here is where we do a number of things that need to be done asynchronously.
789 * For instance:
790 *
791 * - Active expired keys collection (it is also performed in a lazy way on
792 * lookup).
793 * - Software watchdong.
794 * - Update some statistic.
795 * - Incremental rehashing of the DBs hash tables.
796 * - Triggering BGSAVE / AOF rewrite, and handling of terminated children.
797 * - Clients timeout of differnet kinds.
798 * - Replication reconnection.
799 * - Many more...
800 *
801 * Everything directly called here will be called REDIS_HZ times per second,
802 * so in order to throttle execution of things we want to do less frequently
803 * a macro is used: run_with_period(milliseconds) { .... }
804 */
805
806 /* Using the following macro you can run code inside serverCron() with the
807 * specified period, specified in milliseconds.
808 * The actual resolution depends on REDIS_HZ. */
809 #define run_with_period(_ms_) if (!(loops % ((_ms_)/(1000/REDIS_HZ))))
810
811 int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
812 int j, loops = server.cronloops;
813 REDIS_NOTUSED(eventLoop);
814 REDIS_NOTUSED(id);
815 REDIS_NOTUSED(clientData);
816
817 /* Software watchdog: deliver the SIGALRM that will reach the signal
818 * handler if we don't return here fast enough. */
819 if (server.watchdog_period) watchdogScheduleSignal(server.watchdog_period);
820
821 /* We take a cached value of the unix time in the global state because
822 * with virtual memory and aging there is to store the current time
823 * in objects at every object access, and accuracy is not needed.
824 * To access a global var is faster than calling time(NULL) */
825 server.unixtime = time(NULL);
826
827 run_with_period(100) trackOperationsPerSecond();
828
829 /* We have just 22 bits per object for LRU information.
830 * So we use an (eventually wrapping) LRU clock with 10 seconds resolution.
831 * 2^22 bits with 10 seconds resoluton is more or less 1.5 years.
832 *
833 * Note that even if this will wrap after 1.5 years it's not a problem,
834 * everything will still work but just some object will appear younger
835 * to Redis. But for this to happen a given object should never be touched
836 * for 1.5 years.
837 *
838 * Note that you can change the resolution altering the
839 * REDIS_LRU_CLOCK_RESOLUTION define.
840 */
841 updateLRUClock();
842
843 /* Record the max memory used since the server was started. */
844 if (zmalloc_used_memory() > server.stat_peak_memory)
845 server.stat_peak_memory = zmalloc_used_memory();
846
847 /* We received a SIGTERM, shutting down here in a safe way, as it is
848 * not ok doing so inside the signal handler. */
849 if (server.shutdown_asap) {
850 if (prepareForShutdown(0) == REDIS_OK) exit(0);
851 redisLog(REDIS_WARNING,"SIGTERM received but errors trying to shut down the server, check the logs for more information");
852 }
853
854 /* Show some info about non-empty databases */
855 run_with_period(5000) {
856 for (j = 0; j < server.dbnum; j++) {
857 long long size, used, vkeys;
858
859 size = dictSlots(server.db[j].dict);
860 used = dictSize(server.db[j].dict);
861 vkeys = dictSize(server.db[j].expires);
862 if (used || vkeys) {
863 redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
864 /* dictPrintStats(server.dict); */
865 }
866 }
867 }
868
869 /* We don't want to resize the hash tables while a bacground saving
870 * is in progress: the saving child is created using fork() that is
871 * implemented with a copy-on-write semantic in most modern systems, so
872 * if we resize the HT while there is the saving child at work actually
873 * a lot of memory movements in the parent will cause a lot of pages
874 * copied. */
875 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1) {
876 tryResizeHashTables();
877 if (server.activerehashing) incrementallyRehash();
878 }
879
880 /* Show information about connected clients */
881 run_with_period(5000) {
882 redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use",
883 listLength(server.clients)-listLength(server.slaves),
884 listLength(server.slaves),
885 zmalloc_used_memory());
886 }
887
888 /* We need to do a few operations on clients asynchronously. */
889 clientsCron();
890
891 /* Start a scheduled AOF rewrite if this was requested by the user while
892 * a BGSAVE was in progress. */
893 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1 &&
894 server.aof_rewrite_scheduled)
895 {
896 rewriteAppendOnlyFileBackground();
897 }
898
899 /* Check if a background saving or AOF rewrite in progress terminated. */
900 if (server.rdb_child_pid != -1 || server.aof_child_pid != -1) {
901 int statloc;
902 pid_t pid;
903
904 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
905 int exitcode = WEXITSTATUS(statloc);
906 int bysignal = 0;
907
908 if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
909
910 if (pid == server.rdb_child_pid) {
911 backgroundSaveDoneHandler(exitcode,bysignal);
912 } else {
913 backgroundRewriteDoneHandler(exitcode,bysignal);
914 }
915 updateDictResizePolicy();
916 }
917 } else {
918 /* If there is not a background saving/rewrite in progress check if
919 * we have to save/rewrite now */
920 for (j = 0; j < server.saveparamslen; j++) {
921 struct saveparam *sp = server.saveparams+j;
922
923 if (server.dirty >= sp->changes &&
924 server.unixtime-server.lastsave > sp->seconds) {
925 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
926 sp->changes, sp->seconds);
927 rdbSaveBackground(server.rdb_filename);
928 break;
929 }
930 }
931
932 /* Trigger an AOF rewrite if needed */
933 if (server.rdb_child_pid == -1 &&
934 server.aof_child_pid == -1 &&
935 server.aof_rewrite_perc &&
936 server.aof_current_size > server.aof_rewrite_min_size)
937 {
938 long long base = server.aof_rewrite_base_size ?
939 server.aof_rewrite_base_size : 1;
940 long long growth = (server.aof_current_size*100/base) - 100;
941 if (growth >= server.aof_rewrite_perc) {
942 redisLog(REDIS_NOTICE,"Starting automatic rewriting of AOF on %lld%% growth",growth);
943 rewriteAppendOnlyFileBackground();
944 }
945 }
946 }
947
948
949 /* If we postponed an AOF buffer flush, let's try to do it every time the
950 * cron function is called. */
951 if (server.aof_flush_postponed_start) flushAppendOnlyFile(0);
952
953 /* Expire a few keys per cycle, only if this is a master.
954 * On slaves we wait for DEL operations synthesized by the master
955 * in order to guarantee a strict consistency. */
956 if (server.masterhost == NULL) activeExpireCycle();
957
958 /* Close clients that need to be closed asynchronous */
959 freeClientsInAsyncFreeQueue();
960
961 /* Replication cron function -- used to reconnect to master and
962 * to detect transfer failures. */
963 run_with_period(1000) replicationCron();
964
965 server.cronloops++;
966 return 1000/REDIS_HZ;
967 }
968
969 /* This function gets called every time Redis is entering the
970 * main loop of the event driven library, that is, before to sleep
971 * for ready file descriptors. */
972 void beforeSleep(struct aeEventLoop *eventLoop) {
973 REDIS_NOTUSED(eventLoop);
974 listNode *ln;
975 redisClient *c;
976
977 /* Try to process pending commands for clients that were just unblocked. */
978 while (listLength(server.unblocked_clients)) {
979 ln = listFirst(server.unblocked_clients);
980 redisAssert(ln != NULL);
981 c = ln->value;
982 listDelNode(server.unblocked_clients,ln);
983 c->flags &= ~REDIS_UNBLOCKED;
984
985 /* Process remaining data in the input buffer. */
986 if (c->querybuf && sdslen(c->querybuf) > 0) {
987 server.current_client = c;
988 processInputBuffer(c);
989 server.current_client = NULL;
990 }
991 }
992
993 /* Write the AOF buffer on disk */
994 flushAppendOnlyFile(0);
995 }
996
997 /* =========================== Server initialization ======================== */
998
999 void createSharedObjects(void) {
1000 int j;
1001
1002 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
1003 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
1004 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
1005 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
1006 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
1007 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
1008 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
1009 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
1010 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
1011 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
1012 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
1013 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
1014 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
1015 "-ERR Operation against a key holding the wrong kind of value\r\n"));
1016 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
1017 "-ERR no such key\r\n"));
1018 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
1019 "-ERR syntax error\r\n"));
1020 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
1021 "-ERR source and destination objects are the same\r\n"));
1022 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
1023 "-ERR index out of range\r\n"));
1024 shared.noscripterr = createObject(REDIS_STRING,sdsnew(
1025 "-NOSCRIPT No matching script. Please use EVAL.\r\n"));
1026 shared.loadingerr = createObject(REDIS_STRING,sdsnew(
1027 "-LOADING Redis is loading the dataset in memory\r\n"));
1028 shared.slowscripterr = createObject(REDIS_STRING,sdsnew(
1029 "-BUSY Redis is busy running a script. You can only call SCRIPT KILL or SHUTDOWN NOSAVE.\r\n"));
1030 shared.masterdownerr = createObject(REDIS_STRING,sdsnew(
1031 "-MASTERDOWN Link with MASTER is down and slave-serve-stale-data is set to 'no'.\r\n"));
1032 shared.bgsaveerr = createObject(REDIS_STRING,sdsnew(
1033 "-MISCONF Redis is configured to save RDB snapshots, but is currently not able to persist on disk. Commands that may modify the data set are disabled. Please check Redis logs for details about the error.\r\n"));
1034 shared.roslaveerr = createObject(REDIS_STRING,sdsnew(
1035 "-READONLY You can't write against a read only slave.\r\n"));
1036 shared.oomerr = createObject(REDIS_STRING,sdsnew(
1037 "-OOM command not allowed when used memory > 'maxmemory'.\r\n"));
1038 shared.space = createObject(REDIS_STRING,sdsnew(" "));
1039 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
1040 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
1041
1042 for (j = 0; j < REDIS_SHARED_SELECT_CMDS; j++) {
1043 shared.select[j] = createObject(REDIS_STRING,
1044 sdscatprintf(sdsempty(),"select %d\r\n", j));
1045 }
1046 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
1047 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
1048 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
1049 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
1050 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
1051 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
1052 shared.del = createStringObject("DEL",3);
1053 shared.rpop = createStringObject("RPOP",4);
1054 shared.lpop = createStringObject("LPOP",4);
1055 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
1056 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
1057 shared.integers[j]->encoding = REDIS_ENCODING_INT;
1058 }
1059 for (j = 0; j < REDIS_SHARED_BULKHDR_LEN; j++) {
1060 shared.mbulkhdr[j] = createObject(REDIS_STRING,
1061 sdscatprintf(sdsempty(),"*%d\r\n",j));
1062 shared.bulkhdr[j] = createObject(REDIS_STRING,
1063 sdscatprintf(sdsempty(),"$%d\r\n",j));
1064 }
1065 }
1066
1067 void initServerConfig() {
1068 getRandomHexChars(server.runid,REDIS_RUN_ID_SIZE);
1069 server.runid[REDIS_RUN_ID_SIZE] = '\0';
1070 server.arch_bits = (sizeof(long) == 8) ? 64 : 32;
1071 server.port = REDIS_SERVERPORT;
1072 server.bindaddr = NULL;
1073 server.unixsocket = NULL;
1074 server.unixsocketperm = 0;
1075 server.ipfd = -1;
1076 server.sofd = -1;
1077 server.dbnum = REDIS_DEFAULT_DBNUM;
1078 server.verbosity = REDIS_NOTICE;
1079 server.maxidletime = REDIS_MAXIDLETIME;
1080 server.client_max_querybuf_len = REDIS_MAX_QUERYBUF_LEN;
1081 server.saveparams = NULL;
1082 server.loading = 0;
1083 server.logfile = NULL; /* NULL = log on standard output */
1084 server.syslog_enabled = 0;
1085 server.syslog_ident = zstrdup("redis");
1086 server.syslog_facility = LOG_LOCAL0;
1087 server.daemonize = 0;
1088 server.aof_state = REDIS_AOF_OFF;
1089 server.aof_fsync = AOF_FSYNC_EVERYSEC;
1090 server.aof_no_fsync_on_rewrite = 0;
1091 server.aof_rewrite_perc = REDIS_AOF_REWRITE_PERC;
1092 server.aof_rewrite_min_size = REDIS_AOF_REWRITE_MIN_SIZE;
1093 server.aof_rewrite_base_size = 0;
1094 server.aof_rewrite_scheduled = 0;
1095 server.aof_last_fsync = time(NULL);
1096 server.aof_delayed_fsync = 0;
1097 server.aof_fd = -1;
1098 server.aof_selected_db = -1; /* Make sure the first time will not match */
1099 server.aof_flush_postponed_start = 0;
1100 server.pidfile = zstrdup("/var/run/redis.pid");
1101 server.rdb_filename = zstrdup("dump.rdb");
1102 server.aof_filename = zstrdup("appendonly.aof");
1103 server.requirepass = NULL;
1104 server.rdb_compression = 1;
1105 server.rdb_checksum = 1;
1106 server.activerehashing = 1;
1107 server.maxclients = REDIS_MAX_CLIENTS;
1108 server.bpop_blocked_clients = 0;
1109 server.maxmemory = 0;
1110 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
1111 server.maxmemory_samples = 3;
1112 server.hash_max_ziplist_entries = REDIS_HASH_MAX_ZIPLIST_ENTRIES;
1113 server.hash_max_ziplist_value = REDIS_HASH_MAX_ZIPLIST_VALUE;
1114 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
1115 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
1116 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
1117 server.zset_max_ziplist_entries = REDIS_ZSET_MAX_ZIPLIST_ENTRIES;
1118 server.zset_max_ziplist_value = REDIS_ZSET_MAX_ZIPLIST_VALUE;
1119 server.shutdown_asap = 0;
1120 server.repl_ping_slave_period = REDIS_REPL_PING_SLAVE_PERIOD;
1121 server.repl_timeout = REDIS_REPL_TIMEOUT;
1122 server.lua_caller = NULL;
1123 server.lua_time_limit = REDIS_LUA_TIME_LIMIT;
1124 server.lua_client = NULL;
1125 server.lua_timedout = 0;
1126
1127 updateLRUClock();
1128 resetServerSaveParams();
1129
1130 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
1131 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
1132 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
1133 /* Replication related */
1134 server.masterauth = NULL;
1135 server.masterhost = NULL;
1136 server.masterport = 6379;
1137 server.master = NULL;
1138 server.repl_state = REDIS_REPL_NONE;
1139 server.repl_syncio_timeout = REDIS_REPL_SYNCIO_TIMEOUT;
1140 server.repl_serve_stale_data = 1;
1141 server.repl_slave_ro = 1;
1142 server.repl_down_since = time(NULL);
1143
1144 /* Client output buffer limits */
1145 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].hard_limit_bytes = 0;
1146 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].soft_limit_bytes = 0;
1147 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].soft_limit_seconds = 0;
1148 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].hard_limit_bytes = 1024*1024*256;
1149 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].soft_limit_bytes = 1024*1024*64;
1150 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].soft_limit_seconds = 60;
1151 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].hard_limit_bytes = 1024*1024*32;
1152 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].soft_limit_bytes = 1024*1024*8;
1153 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].soft_limit_seconds = 60;
1154
1155 /* Double constants initialization */
1156 R_Zero = 0.0;
1157 R_PosInf = 1.0/R_Zero;
1158 R_NegInf = -1.0/R_Zero;
1159 R_Nan = R_Zero/R_Zero;
1160
1161 /* Command table -- we intiialize it here as it is part of the
1162 * initial configuration, since command names may be changed via
1163 * redis.conf using the rename-command directive. */
1164 server.commands = dictCreate(&commandTableDictType,NULL);
1165 populateCommandTable();
1166 server.delCommand = lookupCommandByCString("del");
1167 server.multiCommand = lookupCommandByCString("multi");
1168 server.lpushCommand = lookupCommandByCString("lpush");
1169
1170 /* Slow log */
1171 server.slowlog_log_slower_than = REDIS_SLOWLOG_LOG_SLOWER_THAN;
1172 server.slowlog_max_len = REDIS_SLOWLOG_MAX_LEN;
1173
1174 /* Debugging */
1175 server.assert_failed = "<no assertion failed>";
1176 server.assert_file = "<no file>";
1177 server.assert_line = 0;
1178 server.bug_report_start = 0;
1179 server.watchdog_period = 0;
1180 }
1181
1182 /* This function will try to raise the max number of open files accordingly to
1183 * the configured max number of clients. It will also account for 32 additional
1184 * file descriptors as we need a few more for persistence, listening
1185 * sockets, log files and so forth.
1186 *
1187 * If it will not be possible to set the limit accordingly to the configured
1188 * max number of clients, the function will do the reverse setting
1189 * server.maxclients to the value that we can actually handle. */
1190 void adjustOpenFilesLimit(void) {
1191 rlim_t maxfiles = server.maxclients+32;
1192 struct rlimit limit;
1193
1194 if (getrlimit(RLIMIT_NOFILE,&limit) == -1) {
1195 redisLog(REDIS_WARNING,"Unable to obtain the current NOFILE limit (%s), assuming 1024 and setting the max clients configuration accordingly.",
1196 strerror(errno));
1197 server.maxclients = 1024-32;
1198 } else {
1199 rlim_t oldlimit = limit.rlim_cur;
1200
1201 /* Set the max number of files if the current limit is not enough
1202 * for our needs. */
1203 if (oldlimit < maxfiles) {
1204 rlim_t f;
1205
1206 f = maxfiles;
1207 while(f > oldlimit) {
1208 limit.rlim_cur = f;
1209 limit.rlim_max = f;
1210 if (setrlimit(RLIMIT_NOFILE,&limit) != -1) break;
1211 f -= 128;
1212 }
1213 if (f < oldlimit) f = oldlimit;
1214 if (f != maxfiles) {
1215 server.maxclients = f-32;
1216 redisLog(REDIS_WARNING,"Unable to set the max number of files limit to %d (%s), setting the max clients configuration to %d.",
1217 (int) maxfiles, strerror(errno), (int) server.maxclients);
1218 } else {
1219 redisLog(REDIS_NOTICE,"Max number of open files set to %d",
1220 (int) maxfiles);
1221 }
1222 }
1223 }
1224 }
1225
1226 void initServer() {
1227 int j;
1228
1229 signal(SIGHUP, SIG_IGN);
1230 signal(SIGPIPE, SIG_IGN);
1231 setupSignalHandlers();
1232
1233 if (server.syslog_enabled) {
1234 openlog(server.syslog_ident, LOG_PID | LOG_NDELAY | LOG_NOWAIT,
1235 server.syslog_facility);
1236 }
1237
1238 server.current_client = NULL;
1239 server.clients = listCreate();
1240 server.clients_to_close = listCreate();
1241 server.slaves = listCreate();
1242 server.monitors = listCreate();
1243 server.unblocked_clients = listCreate();
1244
1245 createSharedObjects();
1246 adjustOpenFilesLimit();
1247 server.el = aeCreateEventLoop(server.maxclients+1024);
1248 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
1249
1250 if (server.port != 0) {
1251 server.ipfd = anetTcpServer(server.neterr,server.port,server.bindaddr);
1252 if (server.ipfd == ANET_ERR) {
1253 redisLog(REDIS_WARNING, "Opening port %d: %s",
1254 server.port, server.neterr);
1255 exit(1);
1256 }
1257 }
1258 if (server.unixsocket != NULL) {
1259 unlink(server.unixsocket); /* don't care if this fails */
1260 server.sofd = anetUnixServer(server.neterr,server.unixsocket,server.unixsocketperm);
1261 if (server.sofd == ANET_ERR) {
1262 redisLog(REDIS_WARNING, "Opening socket: %s", server.neterr);
1263 exit(1);
1264 }
1265 }
1266 if (server.ipfd < 0 && server.sofd < 0) {
1267 redisLog(REDIS_WARNING, "Configured to not listen anywhere, exiting.");
1268 exit(1);
1269 }
1270 for (j = 0; j < server.dbnum; j++) {
1271 server.db[j].dict = dictCreate(&dbDictType,NULL);
1272 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
1273 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
1274 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
1275 server.db[j].id = j;
1276 }
1277 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
1278 server.pubsub_patterns = listCreate();
1279 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
1280 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
1281 server.cronloops = 0;
1282 server.rdb_child_pid = -1;
1283 server.aof_child_pid = -1;
1284 aofRewriteBufferReset();
1285 server.aof_buf = sdsempty();
1286 server.lastsave = time(NULL);
1287 server.dirty = 0;
1288 server.stat_numcommands = 0;
1289 server.stat_numconnections = 0;
1290 server.stat_expiredkeys = 0;
1291 server.stat_evictedkeys = 0;
1292 server.stat_starttime = time(NULL);
1293 server.stat_keyspace_misses = 0;
1294 server.stat_keyspace_hits = 0;
1295 server.stat_peak_memory = 0;
1296 server.stat_fork_time = 0;
1297 server.stat_rejected_conn = 0;
1298 memset(server.ops_sec_samples,0,sizeof(server.ops_sec_samples));
1299 server.ops_sec_idx = 0;
1300 server.ops_sec_last_sample_time = mstime();
1301 server.ops_sec_last_sample_ops = 0;
1302 server.unixtime = time(NULL);
1303 server.lastbgsave_status = REDIS_OK;
1304 server.stop_writes_on_bgsave_err = 1;
1305 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
1306 if (server.ipfd > 0 && aeCreateFileEvent(server.el,server.ipfd,AE_READABLE,
1307 acceptTcpHandler,NULL) == AE_ERR) oom("creating file event");
1308 if (server.sofd > 0 && aeCreateFileEvent(server.el,server.sofd,AE_READABLE,
1309 acceptUnixHandler,NULL) == AE_ERR) oom("creating file event");
1310
1311 if (server.aof_state == REDIS_AOF_ON) {
1312 server.aof_fd = open(server.aof_filename,
1313 O_WRONLY|O_APPEND|O_CREAT,0644);
1314 if (server.aof_fd == -1) {
1315 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
1316 strerror(errno));
1317 exit(1);
1318 }
1319 }
1320
1321 /* 32 bit instances are limited to 4GB of address space, so if there is
1322 * no explicit limit in the user provided configuration we set a limit
1323 * at 3.5GB using maxmemory with 'noeviction' policy'. This saves
1324 * useless crashes of the Redis instance. */
1325 if (server.arch_bits == 32 && server.maxmemory == 0) {
1326 redisLog(REDIS_WARNING,"Warning: 32 bit instance detected but no memory limit set. Setting 3.5 GB maxmemory limit with 'noeviction' policy now.");
1327 server.maxmemory = 3584LL*(1024*1024); /* 3584 MB = 3.5 GB */
1328 server.maxmemory_policy = REDIS_MAXMEMORY_NO_EVICTION;
1329 }
1330
1331 scriptingInit();
1332 slowlogInit();
1333 bioInit();
1334 }
1335
1336 /* Populates the Redis Command Table starting from the hard coded list
1337 * we have on top of redis.c file. */
1338 void populateCommandTable(void) {
1339 int j;
1340 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1341
1342 for (j = 0; j < numcommands; j++) {
1343 struct redisCommand *c = redisCommandTable+j;
1344 char *f = c->sflags;
1345 int retval;
1346
1347 while(*f != '\0') {
1348 switch(*f) {
1349 case 'w': c->flags |= REDIS_CMD_WRITE; break;
1350 case 'r': c->flags |= REDIS_CMD_READONLY; break;
1351 case 'm': c->flags |= REDIS_CMD_DENYOOM; break;
1352 case 'a': c->flags |= REDIS_CMD_ADMIN; break;
1353 case 'p': c->flags |= REDIS_CMD_PUBSUB; break;
1354 case 'f': c->flags |= REDIS_CMD_FORCE_REPLICATION; break;
1355 case 's': c->flags |= REDIS_CMD_NOSCRIPT; break;
1356 case 'R': c->flags |= REDIS_CMD_RANDOM; break;
1357 case 'S': c->flags |= REDIS_CMD_SORT_FOR_SCRIPT; break;
1358 default: redisPanic("Unsupported command flag"); break;
1359 }
1360 f++;
1361 }
1362
1363 retval = dictAdd(server.commands, sdsnew(c->name), c);
1364 assert(retval == DICT_OK);
1365 }
1366 }
1367
1368 void resetCommandTableStats(void) {
1369 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1370 int j;
1371
1372 for (j = 0; j < numcommands; j++) {
1373 struct redisCommand *c = redisCommandTable+j;
1374
1375 c->microseconds = 0;
1376 c->calls = 0;
1377 }
1378 }
1379
1380 /* ========================== Redis OP Array API ============================ */
1381
1382 void redisOpArrayInit(redisOpArray *oa) {
1383 oa->ops = NULL;
1384 oa->numops = 0;
1385 }
1386
1387 int redisOpArrayAppend(redisOpArray *oa, struct redisCommand *cmd, int dbid,
1388 robj **argv, int argc, int target)
1389 {
1390 redisOp *op;
1391
1392 oa->ops = zrealloc(oa->ops,sizeof(redisOp)*(oa->numops+1));
1393 op = oa->ops+oa->numops;
1394 op->cmd = cmd;
1395 op->dbid = dbid;
1396 op->argv = argv;
1397 op->argc = argc;
1398 op->target = target;
1399 oa->numops++;
1400 return oa->numops;
1401 }
1402
1403 void redisOpArrayFree(redisOpArray *oa) {
1404 while(oa->numops) {
1405 int j;
1406 redisOp *op;
1407
1408 oa->numops--;
1409 op = oa->ops+oa->numops;
1410 for (j = 0; j < op->argc; j++)
1411 decrRefCount(op->argv[j]);
1412 zfree(op->argv);
1413 }
1414 zfree(oa->ops);
1415 }
1416
1417 /* ====================== Commands lookup and execution ===================== */
1418
1419 struct redisCommand *lookupCommand(sds name) {
1420 return dictFetchValue(server.commands, name);
1421 }
1422
1423 struct redisCommand *lookupCommandByCString(char *s) {
1424 struct redisCommand *cmd;
1425 sds name = sdsnew(s);
1426
1427 cmd = dictFetchValue(server.commands, name);
1428 sdsfree(name);
1429 return cmd;
1430 }
1431
1432 /* Propagate the specified command (in the context of the specified database id)
1433 * to AOF, Slaves and Monitors.
1434 *
1435 * flags are an xor between:
1436 * + REDIS_PROPAGATE_NONE (no propagation of command at all)
1437 * + REDIS_PROPAGATE_AOF (propagate into the AOF file if is enabled)
1438 * + REDIS_PROPAGATE_REPL (propagate into the replication link)
1439 */
1440 void propagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
1441 int flags)
1442 {
1443 if (server.aof_state != REDIS_AOF_OFF && flags & REDIS_PROPAGATE_AOF)
1444 feedAppendOnlyFile(cmd,dbid,argv,argc);
1445 if (flags & REDIS_PROPAGATE_REPL && listLength(server.slaves))
1446 replicationFeedSlaves(server.slaves,dbid,argv,argc);
1447 }
1448
1449 /* Used inside commands to schedule the propagation of additional commands
1450 * after the current command is propagated to AOF / Replication. */
1451 void alsoPropagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
1452 int target)
1453 {
1454 redisOpArrayAppend(&server.also_propagate,cmd,dbid,argv,argc,target);
1455 }
1456
1457 /* Call() is the core of Redis execution of a command */
1458 void call(redisClient *c, int flags) {
1459 long long dirty, start = ustime(), duration;
1460
1461 /* Sent the command to clients in MONITOR mode, only if the commands are
1462 * not geneated from reading an AOF. */
1463 if (listLength(server.monitors) && !server.loading)
1464 replicationFeedMonitors(c,server.monitors,c->db->id,c->argv,c->argc);
1465
1466 /* Call the command. */
1467 redisOpArrayInit(&server.also_propagate);
1468 dirty = server.dirty;
1469 c->cmd->proc(c);
1470 dirty = server.dirty-dirty;
1471 duration = ustime()-start;
1472
1473 /* When EVAL is called loading the AOF we don't want commands called
1474 * from Lua to go into the slowlog or to populate statistics. */
1475 if (server.loading && c->flags & REDIS_LUA_CLIENT)
1476 flags &= ~(REDIS_CALL_SLOWLOG | REDIS_CALL_STATS);
1477
1478 /* Log the command into the Slow log if needed, and populate the
1479 * per-command statistics that we show in INFO commandstats. */
1480 if (flags & REDIS_CALL_SLOWLOG)
1481 slowlogPushEntryIfNeeded(c->argv,c->argc,duration);
1482 if (flags & REDIS_CALL_STATS) {
1483 c->cmd->microseconds += duration;
1484 c->cmd->calls++;
1485 }
1486
1487 /* Propagate the command into the AOF and replication link */
1488 if (flags & REDIS_CALL_PROPAGATE) {
1489 int flags = REDIS_PROPAGATE_NONE;
1490
1491 if (c->cmd->flags & REDIS_CMD_FORCE_REPLICATION)
1492 flags |= REDIS_PROPAGATE_REPL;
1493 if (dirty)
1494 flags |= (REDIS_PROPAGATE_REPL | REDIS_PROPAGATE_AOF);
1495 if (flags != REDIS_PROPAGATE_NONE)
1496 propagate(c->cmd,c->db->id,c->argv,c->argc,flags);
1497 }
1498 /* Commands such as LPUSH or BRPOPLPUSH may propagate an additional
1499 * PUSH command. */
1500 if (server.also_propagate.numops) {
1501 int j;
1502 redisOp *rop;
1503
1504 for (j = 0; j < server.also_propagate.numops; j++) {
1505 rop = &server.also_propagate.ops[j];
1506 propagate(rop->cmd, rop->dbid, rop->argv, rop->argc, rop->target);
1507 }
1508 redisOpArrayFree(&server.also_propagate);
1509 }
1510 server.stat_numcommands++;
1511 }
1512
1513 /* If this function gets called we already read a whole
1514 * command, argments are in the client argv/argc fields.
1515 * processCommand() execute the command or prepare the
1516 * server for a bulk read from the client.
1517 *
1518 * If 1 is returned the client is still alive and valid and
1519 * and other operations can be performed by the caller. Otherwise
1520 * if 0 is returned the client was destroied (i.e. after QUIT). */
1521 int processCommand(redisClient *c) {
1522 /* The QUIT command is handled separately. Normal command procs will
1523 * go through checking for replication and QUIT will cause trouble
1524 * when FORCE_REPLICATION is enabled and would be implemented in
1525 * a regular command proc. */
1526 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
1527 addReply(c,shared.ok);
1528 c->flags |= REDIS_CLOSE_AFTER_REPLY;
1529 return REDIS_ERR;
1530 }
1531
1532 /* Now lookup the command and check ASAP about trivial error conditions
1533 * such as wrong arity, bad command name and so forth. */
1534 c->cmd = c->lastcmd = lookupCommand(c->argv[0]->ptr);
1535 if (!c->cmd) {
1536 addReplyErrorFormat(c,"unknown command '%s'",
1537 (char*)c->argv[0]->ptr);
1538 return REDIS_OK;
1539 } else if ((c->cmd->arity > 0 && c->cmd->arity != c->argc) ||
1540 (c->argc < -c->cmd->arity)) {
1541 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
1542 c->cmd->name);
1543 return REDIS_OK;
1544 }
1545
1546 /* Check if the user is authenticated */
1547 if (server.requirepass && !c->authenticated && c->cmd->proc != authCommand)
1548 {
1549 addReplyError(c,"operation not permitted");
1550 return REDIS_OK;
1551 }
1552
1553 /* Handle the maxmemory directive.
1554 *
1555 * First we try to free some memory if possible (if there are volatile
1556 * keys in the dataset). If there are not the only thing we can do
1557 * is returning an error. */
1558 if (server.maxmemory) {
1559 int retval = freeMemoryIfNeeded();
1560 if ((c->cmd->flags & REDIS_CMD_DENYOOM) && retval == REDIS_ERR) {
1561 addReply(c, shared.oomerr);
1562 return REDIS_OK;
1563 }
1564 }
1565
1566 /* Don't accept write commands if there are problems persisting on disk. */
1567 if (server.stop_writes_on_bgsave_err &&
1568 server.saveparamslen > 0
1569 && server.lastbgsave_status == REDIS_ERR &&
1570 c->cmd->flags & REDIS_CMD_WRITE)
1571 {
1572 addReply(c, shared.bgsaveerr);
1573 return REDIS_OK;
1574 }
1575
1576 /* Don't accept wirte commands if this is a read only slave. But
1577 * accept write commands if this is our master. */
1578 if (server.masterhost && server.repl_slave_ro &&
1579 !(c->flags & REDIS_MASTER) &&
1580 c->cmd->flags & REDIS_CMD_WRITE)
1581 {
1582 addReply(c, shared.roslaveerr);
1583 return REDIS_OK;
1584 }
1585
1586 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
1587 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
1588 &&
1589 c->cmd->proc != subscribeCommand &&
1590 c->cmd->proc != unsubscribeCommand &&
1591 c->cmd->proc != psubscribeCommand &&
1592 c->cmd->proc != punsubscribeCommand) {
1593 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
1594 return REDIS_OK;
1595 }
1596
1597 /* Only allow INFO and SLAVEOF when slave-serve-stale-data is no and
1598 * we are a slave with a broken link with master. */
1599 if (server.masterhost && server.repl_state != REDIS_REPL_CONNECTED &&
1600 server.repl_serve_stale_data == 0 &&
1601 c->cmd->proc != infoCommand && c->cmd->proc != slaveofCommand)
1602 {
1603 addReply(c, shared.masterdownerr);
1604 return REDIS_OK;
1605 }
1606
1607 /* Loading DB? Return an error if the command is not INFO */
1608 if (server.loading && c->cmd->proc != infoCommand) {
1609 addReply(c, shared.loadingerr);
1610 return REDIS_OK;
1611 }
1612
1613 /* Lua script too slow? Only allow SHUTDOWN NOSAVE and SCRIPT KILL. */
1614 if (server.lua_timedout &&
1615 !(c->cmd->proc == shutdownCommand &&
1616 c->argc == 2 &&
1617 tolower(((char*)c->argv[1]->ptr)[0]) == 'n') &&
1618 !(c->cmd->proc == scriptCommand &&
1619 c->argc == 2 &&
1620 tolower(((char*)c->argv[1]->ptr)[0]) == 'k'))
1621 {
1622 addReply(c, shared.slowscripterr);
1623 return REDIS_OK;
1624 }
1625
1626 /* Exec the command */
1627 if (c->flags & REDIS_MULTI &&
1628 c->cmd->proc != execCommand && c->cmd->proc != discardCommand &&
1629 c->cmd->proc != multiCommand && c->cmd->proc != watchCommand)
1630 {
1631 queueMultiCommand(c);
1632 addReply(c,shared.queued);
1633 } else {
1634 call(c,REDIS_CALL_FULL);
1635 }
1636 return REDIS_OK;
1637 }
1638
1639 /*================================== Shutdown =============================== */
1640
1641 int prepareForShutdown(int flags) {
1642 int save = flags & REDIS_SHUTDOWN_SAVE;
1643 int nosave = flags & REDIS_SHUTDOWN_NOSAVE;
1644
1645 redisLog(REDIS_WARNING,"User requested shutdown...");
1646 /* Kill the saving child if there is a background saving in progress.
1647 We want to avoid race conditions, for instance our saving child may
1648 overwrite the synchronous saving did by SHUTDOWN. */
1649 if (server.rdb_child_pid != -1) {
1650 redisLog(REDIS_WARNING,"There is a child saving an .rdb. Killing it!");
1651 kill(server.rdb_child_pid,SIGKILL);
1652 rdbRemoveTempFile(server.rdb_child_pid);
1653 }
1654 if (server.aof_state != REDIS_AOF_OFF) {
1655 /* Kill the AOF saving child as the AOF we already have may be longer
1656 * but contains the full dataset anyway. */
1657 if (server.aof_child_pid != -1) {
1658 redisLog(REDIS_WARNING,
1659 "There is a child rewriting the AOF. Killing it!");
1660 kill(server.aof_child_pid,SIGKILL);
1661 }
1662 /* Append only file: fsync() the AOF and exit */
1663 redisLog(REDIS_NOTICE,"Calling fsync() on the AOF file.");
1664 aof_fsync(server.aof_fd);
1665 }
1666 if ((server.saveparamslen > 0 && !nosave) || save) {
1667 redisLog(REDIS_NOTICE,"Saving the final RDB snapshot before exiting.");
1668 /* Snapshotting. Perform a SYNC SAVE and exit */
1669 if (rdbSave(server.rdb_filename) != REDIS_OK) {
1670 /* Ooops.. error saving! The best we can do is to continue
1671 * operating. Note that if there was a background saving process,
1672 * in the next cron() Redis will be notified that the background
1673 * saving aborted, handling special stuff like slaves pending for
1674 * synchronization... */
1675 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit.");
1676 return REDIS_ERR;
1677 }
1678 }
1679 if (server.daemonize) {
1680 redisLog(REDIS_NOTICE,"Removing the pid file.");
1681 unlink(server.pidfile);
1682 }
1683 /* Close the listening sockets. Apparently this allows faster restarts. */
1684 if (server.ipfd != -1) close(server.ipfd);
1685 if (server.sofd != -1) close(server.sofd);
1686 if (server.unixsocket) {
1687 redisLog(REDIS_NOTICE,"Removing the unix socket file.");
1688 unlink(server.unixsocket); /* don't care if this fails */
1689 }
1690
1691 redisLog(REDIS_WARNING,"Redis is now ready to exit, bye bye...");
1692 return REDIS_OK;
1693 }
1694
1695 /*================================== Commands =============================== */
1696
1697 void authCommand(redisClient *c) {
1698 if (!server.requirepass) {
1699 addReplyError(c,"Client sent AUTH, but no password is set");
1700 } else if (!strcmp(c->argv[1]->ptr, server.requirepass)) {
1701 c->authenticated = 1;
1702 addReply(c,shared.ok);
1703 } else {
1704 c->authenticated = 0;
1705 addReplyError(c,"invalid password");
1706 }
1707 }
1708
1709 void pingCommand(redisClient *c) {
1710 addReply(c,shared.pong);
1711 }
1712
1713 void echoCommand(redisClient *c) {
1714 addReplyBulk(c,c->argv[1]);
1715 }
1716
1717 void timeCommand(redisClient *c) {
1718 struct timeval tv;
1719
1720 /* gettimeofday() can only fail if &tv is a bad addresss so we
1721 * don't check for errors. */
1722 gettimeofday(&tv,NULL);
1723 addReplyMultiBulkLen(c,2);
1724 addReplyBulkLongLong(c,tv.tv_sec);
1725 addReplyBulkLongLong(c,tv.tv_usec);
1726 }
1727
1728 /* Convert an amount of bytes into a human readable string in the form
1729 * of 100B, 2G, 100M, 4K, and so forth. */
1730 void bytesToHuman(char *s, unsigned long long n) {
1731 double d;
1732
1733 if (n < 1024) {
1734 /* Bytes */
1735 sprintf(s,"%lluB",n);
1736 return;
1737 } else if (n < (1024*1024)) {
1738 d = (double)n/(1024);
1739 sprintf(s,"%.2fK",d);
1740 } else if (n < (1024LL*1024*1024)) {
1741 d = (double)n/(1024*1024);
1742 sprintf(s,"%.2fM",d);
1743 } else if (n < (1024LL*1024*1024*1024)) {
1744 d = (double)n/(1024LL*1024*1024);
1745 sprintf(s,"%.2fG",d);
1746 }
1747 }
1748
1749 /* Create the string returned by the INFO command. This is decoupled
1750 * by the INFO command itself as we need to report the same information
1751 * on memory corruption problems. */
1752 sds genRedisInfoString(char *section) {
1753 sds info = sdsempty();
1754 time_t uptime = server.unixtime-server.stat_starttime;
1755 int j, numcommands;
1756 struct rusage self_ru, c_ru;
1757 unsigned long lol, bib;
1758 int allsections = 0, defsections = 0;
1759 int sections = 0;
1760
1761 if (section) {
1762 allsections = strcasecmp(section,"all") == 0;
1763 defsections = strcasecmp(section,"default") == 0;
1764 }
1765
1766 getrusage(RUSAGE_SELF, &self_ru);
1767 getrusage(RUSAGE_CHILDREN, &c_ru);
1768 getClientsMaxBuffers(&lol,&bib);
1769
1770 /* Server */
1771 if (allsections || defsections || !strcasecmp(section,"server")) {
1772 struct utsname name;
1773
1774 if (sections++) info = sdscat(info,"\r\n");
1775 uname(&name);
1776 info = sdscatprintf(info,
1777 "# Server\r\n"
1778 "redis_version:%s\r\n"
1779 "redis_git_sha1:%s\r\n"
1780 "redis_git_dirty:%d\r\n"
1781 "os:%s %s %s\r\n"
1782 "arch_bits:%d\r\n"
1783 "multiplexing_api:%s\r\n"
1784 "gcc_version:%d.%d.%d\r\n"
1785 "process_id:%ld\r\n"
1786 "run_id:%s\r\n"
1787 "tcp_port:%d\r\n"
1788 "uptime_in_seconds:%ld\r\n"
1789 "uptime_in_days:%ld\r\n"
1790 "lru_clock:%ld\r\n",
1791 REDIS_VERSION,
1792 redisGitSHA1(),
1793 strtol(redisGitDirty(),NULL,10) > 0,
1794 name.sysname, name.release, name.machine,
1795 server.arch_bits,
1796 aeGetApiName(),
1797 #ifdef __GNUC__
1798 __GNUC__,__GNUC_MINOR__,__GNUC_PATCHLEVEL__,
1799 #else
1800 0,0,0,
1801 #endif
1802 (long) getpid(),
1803 server.runid,
1804 server.port,
1805 uptime,
1806 uptime/(3600*24),
1807 (unsigned long) server.lruclock);
1808 }
1809
1810 /* Clients */
1811 if (allsections || defsections || !strcasecmp(section,"clients")) {
1812 if (sections++) info = sdscat(info,"\r\n");
1813 info = sdscatprintf(info,
1814 "# Clients\r\n"
1815 "connected_clients:%lu\r\n"
1816 "client_longest_output_list:%lu\r\n"
1817 "client_biggest_input_buf:%lu\r\n"
1818 "blocked_clients:%d\r\n",
1819 listLength(server.clients)-listLength(server.slaves),
1820 lol, bib,
1821 server.bpop_blocked_clients);
1822 }
1823
1824 /* Memory */
1825 if (allsections || defsections || !strcasecmp(section,"memory")) {
1826 char hmem[64];
1827 char peak_hmem[64];
1828
1829 bytesToHuman(hmem,zmalloc_used_memory());
1830 bytesToHuman(peak_hmem,server.stat_peak_memory);
1831 if (sections++) info = sdscat(info,"\r\n");
1832 info = sdscatprintf(info,
1833 "# Memory\r\n"
1834 "used_memory:%zu\r\n"
1835 "used_memory_human:%s\r\n"
1836 "used_memory_rss:%zu\r\n"
1837 "used_memory_peak:%zu\r\n"
1838 "used_memory_peak_human:%s\r\n"
1839 "used_memory_lua:%lld\r\n"
1840 "mem_fragmentation_ratio:%.2f\r\n"
1841 "mem_allocator:%s\r\n",
1842 zmalloc_used_memory(),
1843 hmem,
1844 zmalloc_get_rss(),
1845 server.stat_peak_memory,
1846 peak_hmem,
1847 ((long long)lua_gc(server.lua,LUA_GCCOUNT,0))*1024LL,
1848 zmalloc_get_fragmentation_ratio(),
1849 ZMALLOC_LIB
1850 );
1851 }
1852
1853 /* Persistence */
1854 if (allsections || defsections || !strcasecmp(section,"persistence")) {
1855 if (sections++) info = sdscat(info,"\r\n");
1856 info = sdscatprintf(info,
1857 "# Persistence\r\n"
1858 "loading:%d\r\n"
1859 "aof_enabled:%d\r\n"
1860 "changes_since_last_save:%lld\r\n"
1861 "bgsave_in_progress:%d\r\n"
1862 "last_save_time:%ld\r\n"
1863 "last_bgsave_status:%s\r\n"
1864 "bgrewriteaof_in_progress:%d\r\n"
1865 "bgrewriteaof_scheduled:%d\r\n",
1866 server.loading,
1867 server.aof_state != REDIS_AOF_OFF,
1868 server.dirty,
1869 server.rdb_child_pid != -1,
1870 server.lastsave,
1871 server.lastbgsave_status == REDIS_OK ? "ok" : "err",
1872 server.aof_child_pid != -1,
1873 server.aof_rewrite_scheduled);
1874
1875 if (server.aof_state != REDIS_AOF_OFF) {
1876 info = sdscatprintf(info,
1877 "aof_current_size:%lld\r\n"
1878 "aof_base_size:%lld\r\n"
1879 "aof_pending_rewrite:%d\r\n"
1880 "aof_buffer_length:%zu\r\n"
1881 "aof_rewrite_buffer_length:%zu\r\n"
1882 "aof_pending_bio_fsync:%llu\r\n"
1883 "aof_delayed_fsync:%lu\r\n",
1884 (long long) server.aof_current_size,
1885 (long long) server.aof_rewrite_base_size,
1886 server.aof_rewrite_scheduled,
1887 sdslen(server.aof_buf),
1888 aofRewriteBufferSize(),
1889 bioPendingJobsOfType(REDIS_BIO_AOF_FSYNC),
1890 server.aof_delayed_fsync);
1891 }
1892
1893 if (server.loading) {
1894 double perc;
1895 time_t eta, elapsed;
1896 off_t remaining_bytes = server.loading_total_bytes-
1897 server.loading_loaded_bytes;
1898
1899 perc = ((double)server.loading_loaded_bytes /
1900 server.loading_total_bytes) * 100;
1901
1902 elapsed = server.unixtime-server.loading_start_time;
1903 if (elapsed == 0) {
1904 eta = 1; /* A fake 1 second figure if we don't have
1905 enough info */
1906 } else {
1907 eta = (elapsed*remaining_bytes)/server.loading_loaded_bytes;
1908 }
1909
1910 info = sdscatprintf(info,
1911 "loading_start_time:%ld\r\n"
1912 "loading_total_bytes:%llu\r\n"
1913 "loading_loaded_bytes:%llu\r\n"
1914 "loading_loaded_perc:%.2f\r\n"
1915 "loading_eta_seconds:%ld\r\n"
1916 ,(unsigned long) server.loading_start_time,
1917 (unsigned long long) server.loading_total_bytes,
1918 (unsigned long long) server.loading_loaded_bytes,
1919 perc,
1920 eta
1921 );
1922 }
1923 }
1924
1925 /* Stats */
1926 if (allsections || defsections || !strcasecmp(section,"stats")) {
1927 if (sections++) info = sdscat(info,"\r\n");
1928 info = sdscatprintf(info,
1929 "# Stats\r\n"
1930 "total_connections_received:%lld\r\n"
1931 "total_commands_processed:%lld\r\n"
1932 "instantaneous_ops_per_sec:%lld\r\n"
1933 "rejected_connections:%lld\r\n"
1934 "expired_keys:%lld\r\n"
1935 "evicted_keys:%lld\r\n"
1936 "keyspace_hits:%lld\r\n"
1937 "keyspace_misses:%lld\r\n"
1938 "pubsub_channels:%ld\r\n"
1939 "pubsub_patterns:%lu\r\n"
1940 "latest_fork_usec:%lld\r\n",
1941 server.stat_numconnections,
1942 server.stat_numcommands,
1943 getOperationsPerSecond(),
1944 server.stat_rejected_conn,
1945 server.stat_expiredkeys,
1946 server.stat_evictedkeys,
1947 server.stat_keyspace_hits,
1948 server.stat_keyspace_misses,
1949 dictSize(server.pubsub_channels),
1950 listLength(server.pubsub_patterns),
1951 server.stat_fork_time);
1952 }
1953
1954 /* Replication */
1955 if (allsections || defsections || !strcasecmp(section,"replication")) {
1956 if (sections++) info = sdscat(info,"\r\n");
1957 info = sdscatprintf(info,
1958 "# Replication\r\n"
1959 "role:%s\r\n",
1960 server.masterhost == NULL ? "master" : "slave");
1961 if (server.masterhost) {
1962 info = sdscatprintf(info,
1963 "master_host:%s\r\n"
1964 "master_port:%d\r\n"
1965 "master_link_status:%s\r\n"
1966 "master_last_io_seconds_ago:%d\r\n"
1967 "master_sync_in_progress:%d\r\n"
1968 ,server.masterhost,
1969 server.masterport,
1970 (server.repl_state == REDIS_REPL_CONNECTED) ?
1971 "up" : "down",
1972 server.master ?
1973 ((int)(server.unixtime-server.master->lastinteraction)) : -1,
1974 server.repl_state == REDIS_REPL_TRANSFER
1975 );
1976
1977 if (server.repl_state == REDIS_REPL_TRANSFER) {
1978 info = sdscatprintf(info,
1979 "master_sync_left_bytes:%ld\r\n"
1980 "master_sync_last_io_seconds_ago:%d\r\n"
1981 ,(long)server.repl_transfer_left,
1982 (int)(server.unixtime-server.repl_transfer_lastio)
1983 );
1984 }
1985
1986 if (server.repl_state != REDIS_REPL_CONNECTED) {
1987 info = sdscatprintf(info,
1988 "master_link_down_since_seconds:%ld\r\n",
1989 (long)server.unixtime-server.repl_down_since);
1990 }
1991 }
1992 info = sdscatprintf(info,
1993 "connected_slaves:%lu\r\n",
1994 listLength(server.slaves));
1995 if (listLength(server.slaves)) {
1996 int slaveid = 0;
1997 listNode *ln;
1998 listIter li;
1999
2000 listRewind(server.slaves,&li);
2001 while((ln = listNext(&li))) {
2002 redisClient *slave = listNodeValue(ln);
2003 char *state = NULL;
2004 char ip[32];
2005 int port;
2006
2007 if (anetPeerToString(slave->fd,ip,&port) == -1) continue;
2008 switch(slave->replstate) {
2009 case REDIS_REPL_WAIT_BGSAVE_START:
2010 case REDIS_REPL_WAIT_BGSAVE_END:
2011 state = "wait_bgsave";
2012 break;
2013 case REDIS_REPL_SEND_BULK:
2014 state = "send_bulk";
2015 break;
2016 case REDIS_REPL_ONLINE:
2017 state = "online";
2018 break;
2019 }
2020 if (state == NULL) continue;
2021 info = sdscatprintf(info,"slave%d:%s,%d,%s\r\n",
2022 slaveid,ip,port,state);
2023 slaveid++;
2024 }
2025 }
2026 }
2027
2028 /* CPU */
2029 if (allsections || defsections || !strcasecmp(section,"cpu")) {
2030 if (sections++) info = sdscat(info,"\r\n");
2031 info = sdscatprintf(info,
2032 "# CPU\r\n"
2033 "used_cpu_sys:%.2f\r\n"
2034 "used_cpu_user:%.2f\r\n"
2035 "used_cpu_sys_children:%.2f\r\n"
2036 "used_cpu_user_children:%.2f\r\n",
2037 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
2038 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
2039 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000,
2040 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000);
2041 }
2042
2043 /* cmdtime */
2044 if (allsections || !strcasecmp(section,"commandstats")) {
2045 if (sections++) info = sdscat(info,"\r\n");
2046 info = sdscatprintf(info, "# Commandstats\r\n");
2047 numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
2048 for (j = 0; j < numcommands; j++) {
2049 struct redisCommand *c = redisCommandTable+j;
2050
2051 if (!c->calls) continue;
2052 info = sdscatprintf(info,
2053 "cmdstat_%s:calls=%lld,usec=%lld,usec_per_call=%.2f\r\n",
2054 c->name, c->calls, c->microseconds,
2055 (c->calls == 0) ? 0 : ((float)c->microseconds/c->calls));
2056 }
2057 }
2058
2059 /* Key space */
2060 if (allsections || defsections || !strcasecmp(section,"keyspace")) {
2061 if (sections++) info = sdscat(info,"\r\n");
2062 info = sdscatprintf(info, "# Keyspace\r\n");
2063 for (j = 0; j < server.dbnum; j++) {
2064 long long keys, vkeys;
2065
2066 keys = dictSize(server.db[j].dict);
2067 vkeys = dictSize(server.db[j].expires);
2068 if (keys || vkeys) {
2069 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
2070 j, keys, vkeys);
2071 }
2072 }
2073 }
2074 return info;
2075 }
2076
2077 void infoCommand(redisClient *c) {
2078 char *section = c->argc == 2 ? c->argv[1]->ptr : "default";
2079
2080 if (c->argc > 2) {
2081 addReply(c,shared.syntaxerr);
2082 return;
2083 }
2084 sds info = genRedisInfoString(section);
2085 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
2086 (unsigned long)sdslen(info)));
2087 addReplySds(c,info);
2088 addReply(c,shared.crlf);
2089 }
2090
2091 void monitorCommand(redisClient *c) {
2092 /* ignore MONITOR if aleady slave or in monitor mode */
2093 if (c->flags & REDIS_SLAVE) return;
2094
2095 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
2096 c->slaveseldb = 0;
2097 listAddNodeTail(server.monitors,c);
2098 addReply(c,shared.ok);
2099 }
2100
2101 /* ============================ Maxmemory directive ======================== */
2102
2103 /* This function gets called when 'maxmemory' is set on the config file to limit
2104 * the max memory used by the server, before processing a command.
2105 *
2106 * The goal of the function is to free enough memory to keep Redis under the
2107 * configured memory limit.
2108 *
2109 * The function starts calculating how many bytes should be freed to keep
2110 * Redis under the limit, and enters a loop selecting the best keys to
2111 * evict accordingly to the configured policy.
2112 *
2113 * If all the bytes needed to return back under the limit were freed the
2114 * function returns REDIS_OK, otherwise REDIS_ERR is returned, and the caller
2115 * should block the execution of commands that will result in more memory
2116 * used by the server.
2117 */
2118 int freeMemoryIfNeeded(void) {
2119 size_t mem_used, mem_tofree, mem_freed;
2120 int slaves = listLength(server.slaves);
2121
2122 /* Remove the size of slaves output buffers and AOF buffer from the
2123 * count of used memory. */
2124 mem_used = zmalloc_used_memory();
2125 if (slaves) {
2126 listIter li;
2127 listNode *ln;
2128
2129 listRewind(server.slaves,&li);
2130 while((ln = listNext(&li))) {
2131 redisClient *slave = listNodeValue(ln);
2132 unsigned long obuf_bytes = getClientOutputBufferMemoryUsage(slave);
2133 if (obuf_bytes > mem_used)
2134 mem_used = 0;
2135 else
2136 mem_used -= obuf_bytes;
2137 }
2138 }
2139 if (server.aof_state != REDIS_AOF_OFF) {
2140 mem_used -= sdslen(server.aof_buf);
2141 mem_used -= aofRewriteBufferSize();
2142 }
2143
2144 /* Check if we are over the memory limit. */
2145 if (mem_used <= server.maxmemory) return REDIS_OK;
2146
2147 if (server.maxmemory_policy == REDIS_MAXMEMORY_NO_EVICTION)
2148 return REDIS_ERR; /* We need to free memory, but policy forbids. */
2149
2150 /* Compute how much memory we need to free. */
2151 mem_tofree = mem_used - server.maxmemory;
2152 mem_freed = 0;
2153 while (mem_freed < mem_tofree) {
2154 int j, k, keys_freed = 0;
2155
2156 for (j = 0; j < server.dbnum; j++) {
2157 long bestval = 0; /* just to prevent warning */
2158 sds bestkey = NULL;
2159 struct dictEntry *de;
2160 redisDb *db = server.db+j;
2161 dict *dict;
2162
2163 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
2164 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
2165 {
2166 dict = server.db[j].dict;
2167 } else {
2168 dict = server.db[j].expires;
2169 }
2170 if (dictSize(dict) == 0) continue;
2171
2172 /* volatile-random and allkeys-random policy */
2173 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
2174 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
2175 {
2176 de = dictGetRandomKey(dict);
2177 bestkey = dictGetKey(de);
2178 }
2179
2180 /* volatile-lru and allkeys-lru policy */
2181 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
2182 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
2183 {
2184 for (k = 0; k < server.maxmemory_samples; k++) {
2185 sds thiskey;
2186 long thisval;
2187 robj *o;
2188
2189 de = dictGetRandomKey(dict);
2190 thiskey = dictGetKey(de);
2191 /* When policy is volatile-lru we need an additonal lookup
2192 * to locate the real key, as dict is set to db->expires. */
2193 if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
2194 de = dictFind(db->dict, thiskey);
2195 o = dictGetVal(de);
2196 thisval = estimateObjectIdleTime(o);
2197
2198 /* Higher idle time is better candidate for deletion */
2199 if (bestkey == NULL || thisval > bestval) {
2200 bestkey = thiskey;
2201 bestval = thisval;
2202 }
2203 }
2204 }
2205
2206 /* volatile-ttl */
2207 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
2208 for (k = 0; k < server.maxmemory_samples; k++) {
2209 sds thiskey;
2210 long thisval;
2211
2212 de = dictGetRandomKey(dict);
2213 thiskey = dictGetKey(de);
2214 thisval = (long) dictGetVal(de);
2215
2216 /* Expire sooner (minor expire unix timestamp) is better
2217 * candidate for deletion */
2218 if (bestkey == NULL || thisval < bestval) {
2219 bestkey = thiskey;
2220 bestval = thisval;
2221 }
2222 }
2223 }
2224
2225 /* Finally remove the selected key. */
2226 if (bestkey) {
2227 long long delta;
2228
2229 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
2230 propagateExpire(db,keyobj);
2231 /* We compute the amount of memory freed by dbDelete() alone.
2232 * It is possible that actually the memory needed to propagate
2233 * the DEL in AOF and replication link is greater than the one
2234 * we are freeing removing the key, but we can't account for
2235 * that otherwise we would never exit the loop.
2236 *
2237 * AOF and Output buffer memory will be freed eventually so
2238 * we only care about memory used by the key space. */
2239 delta = (long long) zmalloc_used_memory();
2240 dbDelete(db,keyobj);
2241 delta -= (long long) zmalloc_used_memory();
2242 mem_freed += delta;
2243 server.stat_evictedkeys++;
2244 decrRefCount(keyobj);
2245 keys_freed++;
2246
2247 /* When the memory to free starts to be big enough, we may
2248 * start spending so much time here that is impossible to
2249 * deliver data to the slaves fast enough, so we force the
2250 * transmission here inside the loop. */
2251 if (slaves) flushSlavesOutputBuffers();
2252 }
2253 }
2254 if (!keys_freed) return REDIS_ERR; /* nothing to free... */
2255 }
2256 return REDIS_OK;
2257 }
2258
2259 /* =================================== Main! ================================ */
2260
2261 #ifdef __linux__
2262 int linuxOvercommitMemoryValue(void) {
2263 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
2264 char buf[64];
2265
2266 if (!fp) return -1;
2267 if (fgets(buf,64,fp) == NULL) {
2268 fclose(fp);
2269 return -1;
2270 }
2271 fclose(fp);
2272
2273 return atoi(buf);
2274 }
2275
2276 void linuxOvercommitMemoryWarning(void) {
2277 if (linuxOvercommitMemoryValue() == 0) {
2278 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.");
2279 }
2280 }
2281 #endif /* __linux__ */
2282
2283 void createPidFile(void) {
2284 /* Try to write the pid file in a best-effort way. */
2285 FILE *fp = fopen(server.pidfile,"w");
2286 if (fp) {
2287 fprintf(fp,"%d\n",(int)getpid());
2288 fclose(fp);
2289 }
2290 }
2291
2292 void daemonize(void) {
2293 int fd;
2294
2295 if (fork() != 0) exit(0); /* parent exits */
2296 setsid(); /* create a new session */
2297
2298 /* Every output goes to /dev/null. If Redis is daemonized but
2299 * the 'logfile' is set to 'stdout' in the configuration file
2300 * it will not log at all. */
2301 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
2302 dup2(fd, STDIN_FILENO);
2303 dup2(fd, STDOUT_FILENO);
2304 dup2(fd, STDERR_FILENO);
2305 if (fd > STDERR_FILENO) close(fd);
2306 }
2307 }
2308
2309 void version() {
2310 printf("Redis server v=%s sha=%s:%d malloc=%s bits=%d\n",
2311 REDIS_VERSION,
2312 redisGitSHA1(),
2313 atoi(redisGitDirty()) > 0,
2314 ZMALLOC_LIB,
2315 sizeof(long) == 4 ? 32 : 64);
2316 exit(0);
2317 }
2318
2319 void usage() {
2320 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf] [options]\n");
2321 fprintf(stderr," ./redis-server - (read config from stdin)\n");
2322 fprintf(stderr," ./redis-server -v or --version\n");
2323 fprintf(stderr," ./redis-server -h or --help\n");
2324 fprintf(stderr," ./redis-server --test-memory <megabytes>\n\n");
2325 fprintf(stderr,"Examples:\n");
2326 fprintf(stderr," ./redis-server (run the server with default conf)\n");
2327 fprintf(stderr," ./redis-server /etc/redis/6379.conf\n");
2328 fprintf(stderr," ./redis-server --port 7777\n");
2329 fprintf(stderr," ./redis-server --port 7777 --slaveof 127.0.0.1 8888\n");
2330 fprintf(stderr," ./redis-server /etc/myredis.conf --loglevel verbose\n");
2331 exit(1);
2332 }
2333
2334 void redisAsciiArt(void) {
2335 #include "asciilogo.h"
2336 char *buf = zmalloc(1024*16);
2337
2338 snprintf(buf,1024*16,ascii_logo,
2339 REDIS_VERSION,
2340 redisGitSHA1(),
2341 strtol(redisGitDirty(),NULL,10) > 0,
2342 (sizeof(long) == 8) ? "64" : "32",
2343 "stand alone",
2344 server.port,
2345 (long) getpid()
2346 );
2347 redisLogRaw(REDIS_NOTICE|REDIS_LOG_RAW,buf);
2348 zfree(buf);
2349 }
2350
2351 static void sigtermHandler(int sig) {
2352 REDIS_NOTUSED(sig);
2353
2354 redisLogFromHandler(REDIS_WARNING,"Received SIGTERM, scheduling shutdown...");
2355 server.shutdown_asap = 1;
2356 }
2357
2358 void setupSignalHandlers(void) {
2359 struct sigaction act;
2360
2361 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction is used.
2362 * Otherwise, sa_handler is used. */
2363 sigemptyset(&act.sa_mask);
2364 act.sa_flags = 0;
2365 act.sa_handler = sigtermHandler;
2366 sigaction(SIGTERM, &act, NULL);
2367
2368 #ifdef HAVE_BACKTRACE
2369 sigemptyset(&act.sa_mask);
2370 act.sa_flags = SA_NODEFER | SA_RESETHAND | SA_SIGINFO;
2371 act.sa_sigaction = sigsegvHandler;
2372 sigaction(SIGSEGV, &act, NULL);
2373 sigaction(SIGBUS, &act, NULL);
2374 sigaction(SIGFPE, &act, NULL);
2375 sigaction(SIGILL, &act, NULL);
2376 #endif
2377 return;
2378 }
2379
2380 void memtest(size_t megabytes, int passes);
2381
2382 int main(int argc, char **argv) {
2383 long long start;
2384 struct timeval tv;
2385
2386 /* We need to initialize our libraries, and the server configuration. */
2387 zmalloc_enable_thread_safeness();
2388 srand(time(NULL)^getpid());
2389 gettimeofday(&tv,NULL);
2390 dictSetHashFunctionSeed(tv.tv_sec^tv.tv_usec^getpid());
2391 initServerConfig();
2392
2393 if (argc >= 2) {
2394 int j = 1; /* First option to parse in argv[] */
2395 sds options = sdsempty();
2396 char *configfile = NULL;
2397
2398 /* Handle special options --help and --version */
2399 if (strcmp(argv[1], "-v") == 0 ||
2400 strcmp(argv[1], "--version") == 0) version();
2401 if (strcmp(argv[1], "--help") == 0 ||
2402 strcmp(argv[1], "-h") == 0) usage();
2403 if (strcmp(argv[1], "--test-memory") == 0) {
2404 if (argc == 3) {
2405 memtest(atoi(argv[2]),50);
2406 exit(0);
2407 } else {
2408 fprintf(stderr,"Please specify the amount of memory to test in megabytes.\n");
2409 fprintf(stderr,"Example: ./redis-server --test-memory 4096\n\n");
2410 exit(1);
2411 }
2412 }
2413
2414 /* First argument is the config file name? */
2415 if (argv[j][0] != '-' || argv[j][1] != '-')
2416 configfile = argv[j++];
2417 /* All the other options are parsed and conceptually appended to the
2418 * configuration file. For instance --port 6380 will generate the
2419 * string "port 6380\n" to be parsed after the actual file name
2420 * is parsed, if any. */
2421 while(j != argc) {
2422 if (argv[j][0] == '-' && argv[j][1] == '-') {
2423 /* Option name */
2424 if (sdslen(options)) options = sdscat(options,"\n");
2425 options = sdscat(options,argv[j]+2);
2426 options = sdscat(options," ");
2427 } else {
2428 /* Option argument */
2429 options = sdscatrepr(options,argv[j],strlen(argv[j]));
2430 options = sdscat(options," ");
2431 }
2432 j++;
2433 }
2434 resetServerSaveParams();
2435 loadServerConfig(configfile,options);
2436 sdsfree(options);
2437 } else {
2438 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'");
2439 }
2440 if (server.daemonize) daemonize();
2441 initServer();
2442 if (server.daemonize) createPidFile();
2443 redisAsciiArt();
2444 redisLog(REDIS_WARNING,"Server started, Redis version " REDIS_VERSION);
2445 #ifdef __linux__
2446 linuxOvercommitMemoryWarning();
2447 #endif
2448 start = ustime();
2449 if (server.aof_state == REDIS_AOF_ON) {
2450 if (loadAppendOnlyFile(server.aof_filename) == REDIS_OK)
2451 redisLog(REDIS_NOTICE,"DB loaded from append only file: %.3f seconds",(float)(ustime()-start)/1000000);
2452 } else {
2453 if (rdbLoad(server.rdb_filename) == REDIS_OK) {
2454 redisLog(REDIS_NOTICE,"DB loaded from disk: %.3f seconds",
2455 (float)(ustime()-start)/1000000);
2456 } else if (errno != ENOENT) {
2457 redisLog(REDIS_WARNING,"Fatal error loading the DB. Exiting.");
2458 exit(1);
2459 }
2460 }
2461 if (server.ipfd > 0)
2462 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
2463 if (server.sofd > 0)
2464 redisLog(REDIS_NOTICE,"The server is now ready to accept connections at %s", server.unixsocket);
2465 aeSetBeforeSleepProc(server.el,beforeSleep);
2466 aeMain(server.el);
2467 aeDeleteEventLoop(server.el);
2468 return 0;
2469 }
2470
2471 /* The End */