]> git.saurik.com Git - redis.git/blob - src/redis.c
fa0603cb36f651c0885ec63ef2f49814f71d03f5
[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 {"replconf",replconfCommand,-1,"ars",0,NULL,0,0,0,0,0},
219 {"flushdb",flushdbCommand,1,"w",0,NULL,0,0,0,0,0},
220 {"flushall",flushallCommand,1,"w",0,NULL,0,0,0,0,0},
221 {"sort",sortCommand,-2,"wmS",0,NULL,1,1,1,0,0},
222 {"info",infoCommand,-1,"r",0,NULL,0,0,0,0,0},
223 {"monitor",monitorCommand,1,"ars",0,NULL,0,0,0,0,0},
224 {"ttl",ttlCommand,2,"r",0,NULL,1,1,1,0,0},
225 {"pttl",pttlCommand,2,"r",0,NULL,1,1,1,0,0},
226 {"persist",persistCommand,2,"w",0,NULL,1,1,1,0,0},
227 {"slaveof",slaveofCommand,3,"as",0,NULL,0,0,0,0,0},
228 {"debug",debugCommand,-2,"as",0,NULL,0,0,0,0,0},
229 {"config",configCommand,-2,"ar",0,NULL,0,0,0,0,0},
230 {"subscribe",subscribeCommand,-2,"rps",0,NULL,0,0,0,0,0},
231 {"unsubscribe",unsubscribeCommand,-1,"rps",0,NULL,0,0,0,0,0},
232 {"psubscribe",psubscribeCommand,-2,"rps",0,NULL,0,0,0,0,0},
233 {"punsubscribe",punsubscribeCommand,-1,"rps",0,NULL,0,0,0,0,0},
234 {"publish",publishCommand,3,"pf",0,NULL,0,0,0,0,0},
235 {"watch",watchCommand,-2,"rs",0,noPreloadGetKeys,1,-1,1,0,0},
236 {"unwatch",unwatchCommand,1,"rs",0,NULL,0,0,0,0,0},
237 {"cluster",clusterCommand,-2,"ar",0,NULL,0,0,0,0,0},
238 {"restore",restoreCommand,4,"awm",0,NULL,1,1,1,0,0},
239 {"migrate",migrateCommand,6,"aw",0,NULL,0,0,0,0,0},
240 {"asking",askingCommand,1,"r",0,NULL,0,0,0,0,0},
241 {"dump",dumpCommand,2,"ar",0,NULL,1,1,1,0,0},
242 {"object",objectCommand,-2,"r",0,NULL,2,2,2,0,0},
243 {"client",clientCommand,-2,"ar",0,NULL,0,0,0,0,0},
244 {"eval",evalCommand,-3,"s",0,zunionInterGetKeys,0,0,0,0,0},
245 {"evalsha",evalShaCommand,-3,"s",0,zunionInterGetKeys,0,0,0,0,0},
246 {"slowlog",slowlogCommand,-2,"r",0,NULL,0,0,0,0,0},
247 {"script",scriptCommand,-2,"ras",0,NULL,0,0,0,0,0},
248 {"time",timeCommand,1,"rR",0,NULL,0,0,0,0,0},
249 {"bitop",bitopCommand,-4,"wm",0,NULL,2,-1,1,0,0},
250 {"bitcount",bitcountCommand,-2,"r",0,NULL,1,1,1,0,0}
251 };
252
253 /*============================ Utility functions ============================ */
254
255 /* Low level logging. To use only for very big messages, otherwise
256 * redisLog() is to prefer. */
257 void redisLogRaw(int level, const char *msg) {
258 const int syslogLevelMap[] = { LOG_DEBUG, LOG_INFO, LOG_NOTICE, LOG_WARNING };
259 const char *c = ".-*#";
260 FILE *fp;
261 char buf[64];
262 int rawmode = (level & REDIS_LOG_RAW);
263
264 level &= 0xff; /* clear flags */
265 if (level < server.verbosity) return;
266
267 fp = (server.logfile == NULL) ? stdout : fopen(server.logfile,"a");
268 if (!fp) return;
269
270 if (rawmode) {
271 fprintf(fp,"%s",msg);
272 } else {
273 int off;
274 struct timeval tv;
275
276 gettimeofday(&tv,NULL);
277 off = strftime(buf,sizeof(buf),"%d %b %H:%M:%S.",localtime(&tv.tv_sec));
278 snprintf(buf+off,sizeof(buf)-off,"%03d",(int)tv.tv_usec/1000);
279 fprintf(fp,"[%d] %s %c %s\n",(int)getpid(),buf,c[level],msg);
280 }
281 fflush(fp);
282
283 if (server.logfile) fclose(fp);
284
285 if (server.syslog_enabled) syslog(syslogLevelMap[level], "%s", msg);
286 }
287
288 /* Like redisLogRaw() but with printf-alike support. This is the funciton that
289 * is used across the code. The raw version is only used in order to dump
290 * the INFO output on crash. */
291 void redisLog(int level, const char *fmt, ...) {
292 va_list ap;
293 char msg[REDIS_MAX_LOGMSG_LEN];
294
295 if ((level&0xff) < server.verbosity) return;
296
297 va_start(ap, fmt);
298 vsnprintf(msg, sizeof(msg), fmt, ap);
299 va_end(ap);
300
301 redisLogRaw(level,msg);
302 }
303
304 /* Log a fixed message without printf-alike capabilities, in a way that is
305 * safe to call from a signal handler.
306 *
307 * We actually use this only for signals that are not fatal from the point
308 * of view of Redis. Signals that are going to kill the server anyway and
309 * where we need printf-alike features are served by redisLog(). */
310 void redisLogFromHandler(int level, const char *msg) {
311 int fd;
312 char buf[64];
313
314 if ((level&0xff) < server.verbosity ||
315 (server.logfile == NULL && server.daemonize)) return;
316 fd = server.logfile ?
317 open(server.logfile, O_APPEND|O_CREAT|O_WRONLY, 0644) :
318 STDOUT_FILENO;
319 if (fd == -1) return;
320 ll2string(buf,sizeof(buf),getpid());
321 if (write(fd,"[",1) == -1) goto err;
322 if (write(fd,buf,strlen(buf)) == -1) goto err;
323 if (write(fd," | signal handler] (",20) == -1) goto err;
324 ll2string(buf,sizeof(buf),time(NULL));
325 if (write(fd,buf,strlen(buf)) == -1) goto err;
326 if (write(fd,") ",2) == -1) goto err;
327 if (write(fd,msg,strlen(msg)) == -1) goto err;
328 if (write(fd,"\n",1) == -1) goto err;
329 err:
330 if (server.logfile) close(fd);
331 }
332
333 /* Redis generally does not try to recover from out of memory conditions
334 * when allocating objects or strings, it is not clear if it will be possible
335 * to report this condition to the client since the networking layer itself
336 * is based on heap allocation for send buffers, so we simply abort.
337 * At least the code will be simpler to read... */
338 void oom(const char *msg) {
339 redisLog(REDIS_WARNING, "%s: Out of memory\n",msg);
340 sleep(1);
341 abort();
342 }
343
344 /* Return the UNIX time in microseconds */
345 long long ustime(void) {
346 struct timeval tv;
347 long long ust;
348
349 gettimeofday(&tv, NULL);
350 ust = ((long long)tv.tv_sec)*1000000;
351 ust += tv.tv_usec;
352 return ust;
353 }
354
355 /* Return the UNIX time in milliseconds */
356 long long mstime(void) {
357 return ustime()/1000;
358 }
359
360 /* After an RDB dump or AOF rewrite we exit from children using _exit() instead of
361 * exit(), because the latter may interact with the same file objects used by
362 * the parent process. However if we are testing the coverage normal exit() is
363 * used in order to obtain the right coverage information. */
364 void exitFromChild(int retcode) {
365 #ifdef COVERAGE_TEST
366 exit(retcode);
367 #else
368 _exit(retcode);
369 #endif
370 }
371
372 /*====================== Hash table type implementation ==================== */
373
374 /* This is an hash table type that uses the SDS dynamic strings libary as
375 * keys and radis objects as values (objects can hold SDS strings,
376 * lists, sets). */
377
378 void dictVanillaFree(void *privdata, void *val)
379 {
380 DICT_NOTUSED(privdata);
381 zfree(val);
382 }
383
384 void dictListDestructor(void *privdata, void *val)
385 {
386 DICT_NOTUSED(privdata);
387 listRelease((list*)val);
388 }
389
390 int dictSdsKeyCompare(void *privdata, const void *key1,
391 const void *key2)
392 {
393 int l1,l2;
394 DICT_NOTUSED(privdata);
395
396 l1 = sdslen((sds)key1);
397 l2 = sdslen((sds)key2);
398 if (l1 != l2) return 0;
399 return memcmp(key1, key2, l1) == 0;
400 }
401
402 /* A case insensitive version used for the command lookup table. */
403 int dictSdsKeyCaseCompare(void *privdata, const void *key1,
404 const void *key2)
405 {
406 DICT_NOTUSED(privdata);
407
408 return strcasecmp(key1, key2) == 0;
409 }
410
411 void dictRedisObjectDestructor(void *privdata, void *val)
412 {
413 DICT_NOTUSED(privdata);
414
415 if (val == NULL) return; /* Values of swapped out keys as set to NULL */
416 decrRefCount(val);
417 }
418
419 void dictSdsDestructor(void *privdata, void *val)
420 {
421 DICT_NOTUSED(privdata);
422
423 sdsfree(val);
424 }
425
426 int dictObjKeyCompare(void *privdata, const void *key1,
427 const void *key2)
428 {
429 const robj *o1 = key1, *o2 = key2;
430 return dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
431 }
432
433 unsigned int dictObjHash(const void *key) {
434 const robj *o = key;
435 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
436 }
437
438 unsigned int dictSdsHash(const void *key) {
439 return dictGenHashFunction((unsigned char*)key, sdslen((char*)key));
440 }
441
442 unsigned int dictSdsCaseHash(const void *key) {
443 return dictGenCaseHashFunction((unsigned char*)key, sdslen((char*)key));
444 }
445
446 int dictEncObjKeyCompare(void *privdata, const void *key1,
447 const void *key2)
448 {
449 robj *o1 = (robj*) key1, *o2 = (robj*) key2;
450 int cmp;
451
452 if (o1->encoding == REDIS_ENCODING_INT &&
453 o2->encoding == REDIS_ENCODING_INT)
454 return o1->ptr == o2->ptr;
455
456 o1 = getDecodedObject(o1);
457 o2 = getDecodedObject(o2);
458 cmp = dictSdsKeyCompare(privdata,o1->ptr,o2->ptr);
459 decrRefCount(o1);
460 decrRefCount(o2);
461 return cmp;
462 }
463
464 unsigned int dictEncObjHash(const void *key) {
465 robj *o = (robj*) key;
466
467 if (o->encoding == REDIS_ENCODING_RAW) {
468 return dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
469 } else {
470 if (o->encoding == REDIS_ENCODING_INT) {
471 char buf[32];
472 int len;
473
474 len = ll2string(buf,32,(long)o->ptr);
475 return dictGenHashFunction((unsigned char*)buf, len);
476 } else {
477 unsigned int hash;
478
479 o = getDecodedObject(o);
480 hash = dictGenHashFunction(o->ptr, sdslen((sds)o->ptr));
481 decrRefCount(o);
482 return hash;
483 }
484 }
485 }
486
487 /* Sets type hash table */
488 dictType setDictType = {
489 dictEncObjHash, /* hash function */
490 NULL, /* key dup */
491 NULL, /* val dup */
492 dictEncObjKeyCompare, /* key compare */
493 dictRedisObjectDestructor, /* key destructor */
494 NULL /* val destructor */
495 };
496
497 /* Sorted sets hash (note: a skiplist is used in addition to the hash table) */
498 dictType zsetDictType = {
499 dictEncObjHash, /* hash function */
500 NULL, /* key dup */
501 NULL, /* val dup */
502 dictEncObjKeyCompare, /* key compare */
503 dictRedisObjectDestructor, /* key destructor */
504 NULL /* val destructor */
505 };
506
507 /* Db->dict, keys are sds strings, vals are Redis objects. */
508 dictType dbDictType = {
509 dictSdsHash, /* hash function */
510 NULL, /* key dup */
511 NULL, /* val dup */
512 dictSdsKeyCompare, /* key compare */
513 dictSdsDestructor, /* key destructor */
514 dictRedisObjectDestructor /* val destructor */
515 };
516
517 /* Db->expires */
518 dictType keyptrDictType = {
519 dictSdsHash, /* hash function */
520 NULL, /* key dup */
521 NULL, /* val dup */
522 dictSdsKeyCompare, /* key compare */
523 NULL, /* key destructor */
524 NULL /* val destructor */
525 };
526
527 /* Command table. sds string -> command struct pointer. */
528 dictType commandTableDictType = {
529 dictSdsCaseHash, /* hash function */
530 NULL, /* key dup */
531 NULL, /* val dup */
532 dictSdsKeyCaseCompare, /* key compare */
533 dictSdsDestructor, /* key destructor */
534 NULL /* val destructor */
535 };
536
537 /* Hash type hash table (note that small hashes are represented with zimpaps) */
538 dictType hashDictType = {
539 dictEncObjHash, /* hash function */
540 NULL, /* key dup */
541 NULL, /* val dup */
542 dictEncObjKeyCompare, /* key compare */
543 dictRedisObjectDestructor, /* key destructor */
544 dictRedisObjectDestructor /* val destructor */
545 };
546
547 /* Keylist hash table type has unencoded redis objects as keys and
548 * lists as values. It's used for blocking operations (BLPOP) and to
549 * map swapped keys to a list of clients waiting for this keys to be loaded. */
550 dictType keylistDictType = {
551 dictObjHash, /* hash function */
552 NULL, /* key dup */
553 NULL, /* val dup */
554 dictObjKeyCompare, /* key compare */
555 dictRedisObjectDestructor, /* key destructor */
556 dictListDestructor /* val destructor */
557 };
558
559 /* Cluster nodes hash table, mapping nodes addresses 1.2.3.4:6379 to
560 * clusterNode structures. */
561 dictType clusterNodesDictType = {
562 dictSdsHash, /* hash function */
563 NULL, /* key dup */
564 NULL, /* val dup */
565 dictSdsKeyCompare, /* key compare */
566 dictSdsDestructor, /* key destructor */
567 NULL /* val destructor */
568 };
569
570 int htNeedsResize(dict *dict) {
571 long long size, used;
572
573 size = dictSlots(dict);
574 used = dictSize(dict);
575 return (size && used && size > DICT_HT_INITIAL_SIZE &&
576 (used*100/size < REDIS_HT_MINFILL));
577 }
578
579 /* If the percentage of used slots in the HT reaches REDIS_HT_MINFILL
580 * we resize the hash table to save memory */
581 void tryResizeHashTables(void) {
582 int j;
583
584 for (j = 0; j < server.dbnum; j++) {
585 if (htNeedsResize(server.db[j].dict))
586 dictResize(server.db[j].dict);
587 if (htNeedsResize(server.db[j].expires))
588 dictResize(server.db[j].expires);
589 }
590 }
591
592 /* Our hash table implementation performs rehashing incrementally while
593 * we write/read from the hash table. Still if the server is idle, the hash
594 * table will use two tables for a long time. So we try to use 1 millisecond
595 * of CPU time at every serverCron() loop in order to rehash some key. */
596 void incrementallyRehash(void) {
597 int j;
598
599 for (j = 0; j < server.dbnum; j++) {
600 /* Keys dictionary */
601 if (dictIsRehashing(server.db[j].dict)) {
602 dictRehashMilliseconds(server.db[j].dict,1);
603 break; /* already used our millisecond for this loop... */
604 }
605 /* Expires */
606 if (dictIsRehashing(server.db[j].expires)) {
607 dictRehashMilliseconds(server.db[j].expires,1);
608 break; /* already used our millisecond for this loop... */
609 }
610 }
611 }
612
613 /* This function is called once a background process of some kind terminates,
614 * as we want to avoid resizing the hash tables when there is a child in order
615 * to play well with copy-on-write (otherwise when a resize happens lots of
616 * memory pages are copied). The goal of this function is to update the ability
617 * for dict.c to resize the hash tables accordingly to the fact we have o not
618 * running childs. */
619 void updateDictResizePolicy(void) {
620 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1)
621 dictEnableResize();
622 else
623 dictDisableResize();
624 }
625
626 /* ======================= Cron: called every 100 ms ======================== */
627
628 /* Try to expire a few timed out keys. The algorithm used is adaptive and
629 * will use few CPU cycles if there are few expiring keys, otherwise
630 * it will get more aggressive to avoid that too much memory is used by
631 * keys that can be removed from the keyspace. */
632 void activeExpireCycle(void) {
633 int j, iteration = 0;
634 long long start = ustime(), timelimit;
635
636 /* We can use at max REDIS_EXPIRELOOKUPS_TIME_PERC percentage of CPU time
637 * per iteration. Since this function gets called with a frequency of
638 * REDIS_HZ times per second, the following is the max amount of
639 * microseconds we can spend in this function. */
640 timelimit = 1000000*REDIS_EXPIRELOOKUPS_TIME_PERC/REDIS_HZ/100;
641 if (timelimit <= 0) timelimit = 1;
642
643 for (j = 0; j < server.dbnum; j++) {
644 int expired;
645 redisDb *db = server.db+j;
646
647 /* Continue to expire if at the end of the cycle more than 25%
648 * of the keys were expired. */
649 do {
650 unsigned long num = dictSize(db->expires);
651 unsigned long slots = dictSlots(db->expires);
652 long long now = mstime();
653
654 /* When there are less than 1% filled slots getting random
655 * keys is expensive, so stop here waiting for better times...
656 * The dictionary will be resized asap. */
657 if (num && slots > DICT_HT_INITIAL_SIZE &&
658 (num*100/slots < 1)) break;
659
660 /* The main collection cycle. Sample random keys among keys
661 * with an expire set, checking for expired ones. */
662 expired = 0;
663 if (num > REDIS_EXPIRELOOKUPS_PER_CRON)
664 num = REDIS_EXPIRELOOKUPS_PER_CRON;
665 while (num--) {
666 dictEntry *de;
667 long long t;
668
669 if ((de = dictGetRandomKey(db->expires)) == NULL) break;
670 t = dictGetSignedIntegerVal(de);
671 if (now > t) {
672 sds key = dictGetKey(de);
673 robj *keyobj = createStringObject(key,sdslen(key));
674
675 propagateExpire(db,keyobj);
676 dbDelete(db,keyobj);
677 decrRefCount(keyobj);
678 expired++;
679 server.stat_expiredkeys++;
680 }
681 }
682 /* We can't block forever here even if there are many keys to
683 * expire. So after a given amount of milliseconds return to the
684 * caller waiting for the other active expire cycle. */
685 iteration++;
686 if ((iteration & 0xf) == 0 && /* check once every 16 cycles. */
687 (ustime()-start) > timelimit) return;
688 } while (expired > REDIS_EXPIRELOOKUPS_PER_CRON/4);
689 }
690 }
691
692 void updateLRUClock(void) {
693 server.lruclock = (server.unixtime/REDIS_LRU_CLOCK_RESOLUTION) &
694 REDIS_LRU_CLOCK_MAX;
695 }
696
697
698 /* Add a sample to the operations per second array of samples. */
699 void trackOperationsPerSecond(void) {
700 long long t = mstime() - server.ops_sec_last_sample_time;
701 long long ops = server.stat_numcommands - server.ops_sec_last_sample_ops;
702 long long ops_sec;
703
704 ops_sec = t > 0 ? (ops*1000/t) : 0;
705
706 server.ops_sec_samples[server.ops_sec_idx] = ops_sec;
707 server.ops_sec_idx = (server.ops_sec_idx+1) % REDIS_OPS_SEC_SAMPLES;
708 server.ops_sec_last_sample_time = mstime();
709 server.ops_sec_last_sample_ops = server.stat_numcommands;
710 }
711
712 /* Return the mean of all the samples. */
713 long long getOperationsPerSecond(void) {
714 int j;
715 long long sum = 0;
716
717 for (j = 0; j < REDIS_OPS_SEC_SAMPLES; j++)
718 sum += server.ops_sec_samples[j];
719 return sum / REDIS_OPS_SEC_SAMPLES;
720 }
721
722 /* Check for timeouts. Returns non-zero if the client was terminated */
723 int clientsCronHandleTimeout(redisClient *c) {
724 time_t now = server.unixtime;
725
726 if (server.maxidletime &&
727 !(c->flags & REDIS_SLAVE) && /* no timeout for slaves */
728 !(c->flags & REDIS_MASTER) && /* no timeout for masters */
729 !(c->flags & REDIS_BLOCKED) && /* no timeout for BLPOP */
730 dictSize(c->pubsub_channels) == 0 && /* no timeout for pubsub */
731 listLength(c->pubsub_patterns) == 0 &&
732 (now - c->lastinteraction > server.maxidletime))
733 {
734 redisLog(REDIS_VERBOSE,"Closing idle client");
735 freeClient(c);
736 return 1;
737 } else if (c->flags & REDIS_BLOCKED) {
738 if (c->bpop.timeout != 0 && c->bpop.timeout < now) {
739 addReply(c,shared.nullmultibulk);
740 unblockClientWaitingData(c);
741 }
742 }
743 return 0;
744 }
745
746 /* The client query buffer is an sds.c string that can end with a lot of
747 * free space not used, this function reclaims space if needed.
748 *
749 * The funciton always returns 0 as it never terminates the client. */
750 int clientsCronResizeQueryBuffer(redisClient *c) {
751 size_t querybuf_size = sdsAllocSize(c->querybuf);
752 time_t idletime = server.unixtime - c->lastinteraction;
753
754 /* There are two conditions to resize the query buffer:
755 * 1) Query buffer is > BIG_ARG and too big for latest peak.
756 * 2) Client is inactive and the buffer is bigger than 1k. */
757 if (((querybuf_size > REDIS_MBULK_BIG_ARG) &&
758 (querybuf_size/(c->querybuf_peak+1)) > 2) ||
759 (querybuf_size > 1024 && idletime > 2))
760 {
761 /* Only resize the query buffer if it is actually wasting space. */
762 if (sdsavail(c->querybuf) > 1024) {
763 c->querybuf = sdsRemoveFreeSpace(c->querybuf);
764 }
765 }
766 /* Reset the peak again to capture the peak memory usage in the next
767 * cycle. */
768 c->querybuf_peak = 0;
769 return 0;
770 }
771
772 void clientsCron(void) {
773 /* Make sure to process at least 1/(REDIS_HZ*10) of clients per call.
774 * Since this function is called REDIS_HZ times per second we are sure that
775 * in the worst case we process all the clients in 10 seconds.
776 * In normal conditions (a reasonable number of clients) we process
777 * all the clients in a shorter time. */
778 int numclients = listLength(server.clients);
779 int iterations = numclients/(REDIS_HZ*10);
780
781 if (iterations < 50)
782 iterations = (numclients < 50) ? numclients : 50;
783 while(listLength(server.clients) && iterations--) {
784 redisClient *c;
785 listNode *head;
786
787 /* Rotate the list, take the current head, process.
788 * This way if the client must be removed from the list it's the
789 * first element and we don't incur into O(N) computation. */
790 listRotate(server.clients);
791 head = listFirst(server.clients);
792 c = listNodeValue(head);
793 /* The following functions do different service checks on the client.
794 * The protocol is that they return non-zero if the client was
795 * terminated. */
796 if (clientsCronHandleTimeout(c)) continue;
797 if (clientsCronResizeQueryBuffer(c)) continue;
798 }
799 }
800
801 /* This is our timer interrupt, called REDIS_HZ times per second.
802 * Here is where we do a number of things that need to be done asynchronously.
803 * For instance:
804 *
805 * - Active expired keys collection (it is also performed in a lazy way on
806 * lookup).
807 * - Software watchdong.
808 * - Update some statistic.
809 * - Incremental rehashing of the DBs hash tables.
810 * - Triggering BGSAVE / AOF rewrite, and handling of terminated children.
811 * - Clients timeout of differnet kinds.
812 * - Replication reconnection.
813 * - Many more...
814 *
815 * Everything directly called here will be called REDIS_HZ times per second,
816 * so in order to throttle execution of things we want to do less frequently
817 * a macro is used: run_with_period(milliseconds) { .... }
818 */
819
820 /* Using the following macro you can run code inside serverCron() with the
821 * specified period, specified in milliseconds.
822 * The actual resolution depends on REDIS_HZ. */
823 #define run_with_period(_ms_) if (!(loops % ((_ms_)/(1000/REDIS_HZ))))
824
825 int serverCron(struct aeEventLoop *eventLoop, long long id, void *clientData) {
826 int j, loops = server.cronloops;
827 REDIS_NOTUSED(eventLoop);
828 REDIS_NOTUSED(id);
829 REDIS_NOTUSED(clientData);
830
831 /* Software watchdog: deliver the SIGALRM that will reach the signal
832 * handler if we don't return here fast enough. */
833 if (server.watchdog_period) watchdogScheduleSignal(server.watchdog_period);
834
835 /* We take a cached value of the unix time in the global state because
836 * with virtual memory and aging there is to store the current time
837 * in objects at every object access, and accuracy is not needed.
838 * To access a global var is faster than calling time(NULL) */
839 server.unixtime = time(NULL);
840
841 run_with_period(100) trackOperationsPerSecond();
842
843 /* We have just 22 bits per object for LRU information.
844 * So we use an (eventually wrapping) LRU clock with 10 seconds resolution.
845 * 2^22 bits with 10 seconds resoluton is more or less 1.5 years.
846 *
847 * Note that even if this will wrap after 1.5 years it's not a problem,
848 * everything will still work but just some object will appear younger
849 * to Redis. But for this to happen a given object should never be touched
850 * for 1.5 years.
851 *
852 * Note that you can change the resolution altering the
853 * REDIS_LRU_CLOCK_RESOLUTION define.
854 */
855 updateLRUClock();
856
857 /* Record the max memory used since the server was started. */
858 if (zmalloc_used_memory() > server.stat_peak_memory)
859 server.stat_peak_memory = zmalloc_used_memory();
860
861 /* We received a SIGTERM, shutting down here in a safe way, as it is
862 * not ok doing so inside the signal handler. */
863 if (server.shutdown_asap) {
864 if (prepareForShutdown(0) == REDIS_OK) exit(0);
865 redisLog(REDIS_WARNING,"SIGTERM received but errors trying to shut down the server, check the logs for more information");
866 }
867
868 /* Show some info about non-empty databases */
869 run_with_period(5000) {
870 for (j = 0; j < server.dbnum; j++) {
871 long long size, used, vkeys;
872
873 size = dictSlots(server.db[j].dict);
874 used = dictSize(server.db[j].dict);
875 vkeys = dictSize(server.db[j].expires);
876 if (used || vkeys) {
877 redisLog(REDIS_VERBOSE,"DB %d: %lld keys (%lld volatile) in %lld slots HT.",j,used,vkeys,size);
878 /* dictPrintStats(server.dict); */
879 }
880 }
881 }
882
883 /* We don't want to resize the hash tables while a bacground saving
884 * is in progress: the saving child is created using fork() that is
885 * implemented with a copy-on-write semantic in most modern systems, so
886 * if we resize the HT while there is the saving child at work actually
887 * a lot of memory movements in the parent will cause a lot of pages
888 * copied. */
889 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1) {
890 tryResizeHashTables();
891 if (server.activerehashing) incrementallyRehash();
892 }
893
894 /* Show information about connected clients */
895 run_with_period(5000) {
896 redisLog(REDIS_VERBOSE,"%d clients connected (%d slaves), %zu bytes in use",
897 listLength(server.clients)-listLength(server.slaves),
898 listLength(server.slaves),
899 zmalloc_used_memory());
900 }
901
902 /* We need to do a few operations on clients asynchronously. */
903 clientsCron();
904
905 /* Start a scheduled AOF rewrite if this was requested by the user while
906 * a BGSAVE was in progress. */
907 if (server.rdb_child_pid == -1 && server.aof_child_pid == -1 &&
908 server.aof_rewrite_scheduled)
909 {
910 rewriteAppendOnlyFileBackground();
911 }
912
913 /* Check if a background saving or AOF rewrite in progress terminated. */
914 if (server.rdb_child_pid != -1 || server.aof_child_pid != -1) {
915 int statloc;
916 pid_t pid;
917
918 if ((pid = wait3(&statloc,WNOHANG,NULL)) != 0) {
919 int exitcode = WEXITSTATUS(statloc);
920 int bysignal = 0;
921
922 if (WIFSIGNALED(statloc)) bysignal = WTERMSIG(statloc);
923
924 if (pid == server.rdb_child_pid) {
925 backgroundSaveDoneHandler(exitcode,bysignal);
926 } else {
927 backgroundRewriteDoneHandler(exitcode,bysignal);
928 }
929 updateDictResizePolicy();
930 }
931 } else {
932 /* If there is not a background saving/rewrite in progress check if
933 * we have to save/rewrite now */
934 for (j = 0; j < server.saveparamslen; j++) {
935 struct saveparam *sp = server.saveparams+j;
936
937 if (server.dirty >= sp->changes &&
938 server.unixtime-server.lastsave > sp->seconds) {
939 redisLog(REDIS_NOTICE,"%d changes in %d seconds. Saving...",
940 sp->changes, sp->seconds);
941 rdbSaveBackground(server.rdb_filename);
942 break;
943 }
944 }
945
946 /* Trigger an AOF rewrite if needed */
947 if (server.rdb_child_pid == -1 &&
948 server.aof_child_pid == -1 &&
949 server.aof_rewrite_perc &&
950 server.aof_current_size > server.aof_rewrite_min_size)
951 {
952 long long base = server.aof_rewrite_base_size ?
953 server.aof_rewrite_base_size : 1;
954 long long growth = (server.aof_current_size*100/base) - 100;
955 if (growth >= server.aof_rewrite_perc) {
956 redisLog(REDIS_NOTICE,"Starting automatic rewriting of AOF on %lld%% growth",growth);
957 rewriteAppendOnlyFileBackground();
958 }
959 }
960 }
961
962
963 /* If we postponed an AOF buffer flush, let's try to do it every time the
964 * cron function is called. */
965 if (server.aof_flush_postponed_start) flushAppendOnlyFile(0);
966
967 /* Expire a few keys per cycle, only if this is a master.
968 * On slaves we wait for DEL operations synthesized by the master
969 * in order to guarantee a strict consistency. */
970 if (server.masterhost == NULL) activeExpireCycle();
971
972 /* Close clients that need to be closed asynchronous */
973 freeClientsInAsyncFreeQueue();
974
975 /* Replication cron function -- used to reconnect to master and
976 * to detect transfer failures. */
977 run_with_period(1000) replicationCron();
978
979 /* Run other sub-systems specific cron jobs */
980 run_with_period(1000) {
981 if (server.cluster_enabled) clusterCron();
982 }
983
984 server.cronloops++;
985 return 1000/REDIS_HZ;
986 }
987
988 /* This function gets called every time Redis is entering the
989 * main loop of the event driven library, that is, before to sleep
990 * for ready file descriptors. */
991 void beforeSleep(struct aeEventLoop *eventLoop) {
992 REDIS_NOTUSED(eventLoop);
993 listNode *ln;
994 redisClient *c;
995
996 /* Try to process pending commands for clients that were just unblocked. */
997 while (listLength(server.unblocked_clients)) {
998 ln = listFirst(server.unblocked_clients);
999 redisAssert(ln != NULL);
1000 c = ln->value;
1001 listDelNode(server.unblocked_clients,ln);
1002 c->flags &= ~REDIS_UNBLOCKED;
1003
1004 /* Process remaining data in the input buffer. */
1005 if (c->querybuf && sdslen(c->querybuf) > 0) {
1006 server.current_client = c;
1007 processInputBuffer(c);
1008 server.current_client = NULL;
1009 }
1010 }
1011
1012 /* Write the AOF buffer on disk */
1013 flushAppendOnlyFile(0);
1014 }
1015
1016 /* =========================== Server initialization ======================== */
1017
1018 void createSharedObjects(void) {
1019 int j;
1020
1021 shared.crlf = createObject(REDIS_STRING,sdsnew("\r\n"));
1022 shared.ok = createObject(REDIS_STRING,sdsnew("+OK\r\n"));
1023 shared.err = createObject(REDIS_STRING,sdsnew("-ERR\r\n"));
1024 shared.emptybulk = createObject(REDIS_STRING,sdsnew("$0\r\n\r\n"));
1025 shared.czero = createObject(REDIS_STRING,sdsnew(":0\r\n"));
1026 shared.cone = createObject(REDIS_STRING,sdsnew(":1\r\n"));
1027 shared.cnegone = createObject(REDIS_STRING,sdsnew(":-1\r\n"));
1028 shared.nullbulk = createObject(REDIS_STRING,sdsnew("$-1\r\n"));
1029 shared.nullmultibulk = createObject(REDIS_STRING,sdsnew("*-1\r\n"));
1030 shared.emptymultibulk = createObject(REDIS_STRING,sdsnew("*0\r\n"));
1031 shared.pong = createObject(REDIS_STRING,sdsnew("+PONG\r\n"));
1032 shared.queued = createObject(REDIS_STRING,sdsnew("+QUEUED\r\n"));
1033 shared.wrongtypeerr = createObject(REDIS_STRING,sdsnew(
1034 "-ERR Operation against a key holding the wrong kind of value\r\n"));
1035 shared.nokeyerr = createObject(REDIS_STRING,sdsnew(
1036 "-ERR no such key\r\n"));
1037 shared.syntaxerr = createObject(REDIS_STRING,sdsnew(
1038 "-ERR syntax error\r\n"));
1039 shared.sameobjecterr = createObject(REDIS_STRING,sdsnew(
1040 "-ERR source and destination objects are the same\r\n"));
1041 shared.outofrangeerr = createObject(REDIS_STRING,sdsnew(
1042 "-ERR index out of range\r\n"));
1043 shared.noscripterr = createObject(REDIS_STRING,sdsnew(
1044 "-NOSCRIPT No matching script. Please use EVAL.\r\n"));
1045 shared.loadingerr = createObject(REDIS_STRING,sdsnew(
1046 "-LOADING Redis is loading the dataset in memory\r\n"));
1047 shared.slowscripterr = createObject(REDIS_STRING,sdsnew(
1048 "-BUSY Redis is busy running a script. You can only call SCRIPT KILL or SHUTDOWN NOSAVE.\r\n"));
1049 shared.masterdownerr = createObject(REDIS_STRING,sdsnew(
1050 "-MASTERDOWN Link with MASTER is down and slave-serve-stale-data is set to 'no'.\r\n"));
1051 shared.bgsaveerr = createObject(REDIS_STRING,sdsnew(
1052 "-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"));
1053 shared.roslaveerr = createObject(REDIS_STRING,sdsnew(
1054 "-READONLY You can't write against a read only slave.\r\n"));
1055 shared.oomerr = createObject(REDIS_STRING,sdsnew(
1056 "-OOM command not allowed when used memory > 'maxmemory'.\r\n"));
1057 shared.space = createObject(REDIS_STRING,sdsnew(" "));
1058 shared.colon = createObject(REDIS_STRING,sdsnew(":"));
1059 shared.plus = createObject(REDIS_STRING,sdsnew("+"));
1060
1061 for (j = 0; j < REDIS_SHARED_SELECT_CMDS; j++) {
1062 shared.select[j] = createObject(REDIS_STRING,
1063 sdscatprintf(sdsempty(),"select %d\r\n", j));
1064 }
1065 shared.messagebulk = createStringObject("$7\r\nmessage\r\n",13);
1066 shared.pmessagebulk = createStringObject("$8\r\npmessage\r\n",14);
1067 shared.subscribebulk = createStringObject("$9\r\nsubscribe\r\n",15);
1068 shared.unsubscribebulk = createStringObject("$11\r\nunsubscribe\r\n",18);
1069 shared.psubscribebulk = createStringObject("$10\r\npsubscribe\r\n",17);
1070 shared.punsubscribebulk = createStringObject("$12\r\npunsubscribe\r\n",19);
1071 shared.del = createStringObject("DEL",3);
1072 shared.rpop = createStringObject("RPOP",4);
1073 shared.lpop = createStringObject("LPOP",4);
1074 for (j = 0; j < REDIS_SHARED_INTEGERS; j++) {
1075 shared.integers[j] = createObject(REDIS_STRING,(void*)(long)j);
1076 shared.integers[j]->encoding = REDIS_ENCODING_INT;
1077 }
1078 for (j = 0; j < REDIS_SHARED_BULKHDR_LEN; j++) {
1079 shared.mbulkhdr[j] = createObject(REDIS_STRING,
1080 sdscatprintf(sdsempty(),"*%d\r\n",j));
1081 shared.bulkhdr[j] = createObject(REDIS_STRING,
1082 sdscatprintf(sdsempty(),"$%d\r\n",j));
1083 }
1084 }
1085
1086 void initServerConfig() {
1087 getRandomHexChars(server.runid,REDIS_RUN_ID_SIZE);
1088 server.runid[REDIS_RUN_ID_SIZE] = '\0';
1089 server.arch_bits = (sizeof(long) == 8) ? 64 : 32;
1090 server.port = REDIS_SERVERPORT;
1091 server.bindaddr = NULL;
1092 server.unixsocket = NULL;
1093 server.unixsocketperm = 0;
1094 server.ipfd = -1;
1095 server.sofd = -1;
1096 server.dbnum = REDIS_DEFAULT_DBNUM;
1097 server.verbosity = REDIS_NOTICE;
1098 server.maxidletime = REDIS_MAXIDLETIME;
1099 server.client_max_querybuf_len = REDIS_MAX_QUERYBUF_LEN;
1100 server.saveparams = NULL;
1101 server.loading = 0;
1102 server.logfile = NULL; /* NULL = log on standard output */
1103 server.syslog_enabled = 0;
1104 server.syslog_ident = zstrdup("redis");
1105 server.syslog_facility = LOG_LOCAL0;
1106 server.daemonize = 0;
1107 server.aof_state = REDIS_AOF_OFF;
1108 server.aof_fsync = AOF_FSYNC_EVERYSEC;
1109 server.aof_no_fsync_on_rewrite = 0;
1110 server.aof_rewrite_perc = REDIS_AOF_REWRITE_PERC;
1111 server.aof_rewrite_min_size = REDIS_AOF_REWRITE_MIN_SIZE;
1112 server.aof_rewrite_base_size = 0;
1113 server.aof_rewrite_scheduled = 0;
1114 server.aof_last_fsync = time(NULL);
1115 server.aof_rewrite_time_last = -1;
1116 server.aof_rewrite_time_start = -1;
1117 server.aof_delayed_fsync = 0;
1118 server.aof_fd = -1;
1119 server.aof_selected_db = -1; /* Make sure the first time will not match */
1120 server.aof_flush_postponed_start = 0;
1121 server.pidfile = zstrdup("/var/run/redis.pid");
1122 server.rdb_filename = zstrdup("dump.rdb");
1123 server.aof_filename = zstrdup("appendonly.aof");
1124 server.requirepass = NULL;
1125 server.rdb_compression = 1;
1126 server.rdb_checksum = 1;
1127 server.activerehashing = 1;
1128 server.maxclients = REDIS_MAX_CLIENTS;
1129 server.bpop_blocked_clients = 0;
1130 server.maxmemory = 0;
1131 server.maxmemory_policy = REDIS_MAXMEMORY_VOLATILE_LRU;
1132 server.maxmemory_samples = 3;
1133 server.hash_max_ziplist_entries = REDIS_HASH_MAX_ZIPLIST_ENTRIES;
1134 server.hash_max_ziplist_value = REDIS_HASH_MAX_ZIPLIST_VALUE;
1135 server.list_max_ziplist_entries = REDIS_LIST_MAX_ZIPLIST_ENTRIES;
1136 server.list_max_ziplist_value = REDIS_LIST_MAX_ZIPLIST_VALUE;
1137 server.set_max_intset_entries = REDIS_SET_MAX_INTSET_ENTRIES;
1138 server.zset_max_ziplist_entries = REDIS_ZSET_MAX_ZIPLIST_ENTRIES;
1139 server.zset_max_ziplist_value = REDIS_ZSET_MAX_ZIPLIST_VALUE;
1140 server.shutdown_asap = 0;
1141 server.repl_ping_slave_period = REDIS_REPL_PING_SLAVE_PERIOD;
1142 server.repl_timeout = REDIS_REPL_TIMEOUT;
1143 server.cluster_enabled = 0;
1144 server.cluster.configfile = zstrdup("nodes.conf");
1145 server.lua_caller = NULL;
1146 server.lua_time_limit = REDIS_LUA_TIME_LIMIT;
1147 server.lua_client = NULL;
1148 server.lua_timedout = 0;
1149
1150 updateLRUClock();
1151 resetServerSaveParams();
1152
1153 appendServerSaveParams(60*60,1); /* save after 1 hour and 1 change */
1154 appendServerSaveParams(300,100); /* save after 5 minutes and 100 changes */
1155 appendServerSaveParams(60,10000); /* save after 1 minute and 10000 changes */
1156 /* Replication related */
1157 server.masterauth = NULL;
1158 server.masterhost = NULL;
1159 server.masterport = 6379;
1160 server.master = NULL;
1161 server.repl_state = REDIS_REPL_NONE;
1162 server.repl_syncio_timeout = REDIS_REPL_SYNCIO_TIMEOUT;
1163 server.repl_serve_stale_data = 1;
1164 server.repl_slave_ro = 1;
1165 server.repl_down_since = time(NULL);
1166
1167 /* Client output buffer limits */
1168 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].hard_limit_bytes = 0;
1169 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].soft_limit_bytes = 0;
1170 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_NORMAL].soft_limit_seconds = 0;
1171 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].hard_limit_bytes = 1024*1024*256;
1172 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].soft_limit_bytes = 1024*1024*64;
1173 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_SLAVE].soft_limit_seconds = 60;
1174 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].hard_limit_bytes = 1024*1024*32;
1175 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].soft_limit_bytes = 1024*1024*8;
1176 server.client_obuf_limits[REDIS_CLIENT_LIMIT_CLASS_PUBSUB].soft_limit_seconds = 60;
1177
1178 /* Double constants initialization */
1179 R_Zero = 0.0;
1180 R_PosInf = 1.0/R_Zero;
1181 R_NegInf = -1.0/R_Zero;
1182 R_Nan = R_Zero/R_Zero;
1183
1184 /* Command table -- we intiialize it here as it is part of the
1185 * initial configuration, since command names may be changed via
1186 * redis.conf using the rename-command directive. */
1187 server.commands = dictCreate(&commandTableDictType,NULL);
1188 populateCommandTable();
1189 server.delCommand = lookupCommandByCString("del");
1190 server.multiCommand = lookupCommandByCString("multi");
1191 server.lpushCommand = lookupCommandByCString("lpush");
1192
1193 /* Slow log */
1194 server.slowlog_log_slower_than = REDIS_SLOWLOG_LOG_SLOWER_THAN;
1195 server.slowlog_max_len = REDIS_SLOWLOG_MAX_LEN;
1196
1197 /* Debugging */
1198 server.assert_failed = "<no assertion failed>";
1199 server.assert_file = "<no file>";
1200 server.assert_line = 0;
1201 server.bug_report_start = 0;
1202 server.watchdog_period = 0;
1203 }
1204
1205 /* This function will try to raise the max number of open files accordingly to
1206 * the configured max number of clients. It will also account for 32 additional
1207 * file descriptors as we need a few more for persistence, listening
1208 * sockets, log files and so forth.
1209 *
1210 * If it will not be possible to set the limit accordingly to the configured
1211 * max number of clients, the function will do the reverse setting
1212 * server.maxclients to the value that we can actually handle. */
1213 void adjustOpenFilesLimit(void) {
1214 rlim_t maxfiles = server.maxclients+32;
1215 struct rlimit limit;
1216
1217 if (getrlimit(RLIMIT_NOFILE,&limit) == -1) {
1218 redisLog(REDIS_WARNING,"Unable to obtain the current NOFILE limit (%s), assuming 1024 and setting the max clients configuration accordingly.",
1219 strerror(errno));
1220 server.maxclients = 1024-32;
1221 } else {
1222 rlim_t oldlimit = limit.rlim_cur;
1223
1224 /* Set the max number of files if the current limit is not enough
1225 * for our needs. */
1226 if (oldlimit < maxfiles) {
1227 rlim_t f;
1228
1229 f = maxfiles;
1230 while(f > oldlimit) {
1231 limit.rlim_cur = f;
1232 limit.rlim_max = f;
1233 if (setrlimit(RLIMIT_NOFILE,&limit) != -1) break;
1234 f -= 128;
1235 }
1236 if (f < oldlimit) f = oldlimit;
1237 if (f != maxfiles) {
1238 server.maxclients = f-32;
1239 redisLog(REDIS_WARNING,"Unable to set the max number of files limit to %d (%s), setting the max clients configuration to %d.",
1240 (int) maxfiles, strerror(errno), (int) server.maxclients);
1241 } else {
1242 redisLog(REDIS_NOTICE,"Max number of open files set to %d",
1243 (int) maxfiles);
1244 }
1245 }
1246 }
1247 }
1248
1249 void initServer() {
1250 int j;
1251
1252 signal(SIGHUP, SIG_IGN);
1253 signal(SIGPIPE, SIG_IGN);
1254 setupSignalHandlers();
1255
1256 if (server.syslog_enabled) {
1257 openlog(server.syslog_ident, LOG_PID | LOG_NDELAY | LOG_NOWAIT,
1258 server.syslog_facility);
1259 }
1260
1261 server.current_client = NULL;
1262 server.clients = listCreate();
1263 server.clients_to_close = listCreate();
1264 server.slaves = listCreate();
1265 server.monitors = listCreate();
1266 server.unblocked_clients = listCreate();
1267
1268 createSharedObjects();
1269 adjustOpenFilesLimit();
1270 server.el = aeCreateEventLoop(server.maxclients+1024);
1271 server.db = zmalloc(sizeof(redisDb)*server.dbnum);
1272
1273 if (server.port != 0) {
1274 server.ipfd = anetTcpServer(server.neterr,server.port,server.bindaddr);
1275 if (server.ipfd == ANET_ERR) {
1276 redisLog(REDIS_WARNING, "Opening port %d: %s",
1277 server.port, server.neterr);
1278 exit(1);
1279 }
1280 }
1281 if (server.unixsocket != NULL) {
1282 unlink(server.unixsocket); /* don't care if this fails */
1283 server.sofd = anetUnixServer(server.neterr,server.unixsocket,server.unixsocketperm);
1284 if (server.sofd == ANET_ERR) {
1285 redisLog(REDIS_WARNING, "Opening socket: %s", server.neterr);
1286 exit(1);
1287 }
1288 }
1289 if (server.ipfd < 0 && server.sofd < 0) {
1290 redisLog(REDIS_WARNING, "Configured to not listen anywhere, exiting.");
1291 exit(1);
1292 }
1293 for (j = 0; j < server.dbnum; j++) {
1294 server.db[j].dict = dictCreate(&dbDictType,NULL);
1295 server.db[j].expires = dictCreate(&keyptrDictType,NULL);
1296 server.db[j].blocking_keys = dictCreate(&keylistDictType,NULL);
1297 server.db[j].watched_keys = dictCreate(&keylistDictType,NULL);
1298 server.db[j].id = j;
1299 }
1300 server.pubsub_channels = dictCreate(&keylistDictType,NULL);
1301 server.pubsub_patterns = listCreate();
1302 listSetFreeMethod(server.pubsub_patterns,freePubsubPattern);
1303 listSetMatchMethod(server.pubsub_patterns,listMatchPubsubPattern);
1304 server.cronloops = 0;
1305 server.rdb_child_pid = -1;
1306 server.aof_child_pid = -1;
1307 aofRewriteBufferReset();
1308 server.aof_buf = sdsempty();
1309 server.lastsave = time(NULL);
1310 server.rdb_save_time_last = -1;
1311 server.rdb_save_time_start = -1;
1312 server.dirty = 0;
1313 server.stat_numcommands = 0;
1314 server.stat_numconnections = 0;
1315 server.stat_expiredkeys = 0;
1316 server.stat_evictedkeys = 0;
1317 server.stat_starttime = time(NULL);
1318 server.stat_keyspace_misses = 0;
1319 server.stat_keyspace_hits = 0;
1320 server.stat_peak_memory = 0;
1321 server.stat_fork_time = 0;
1322 server.stat_rejected_conn = 0;
1323 memset(server.ops_sec_samples,0,sizeof(server.ops_sec_samples));
1324 server.ops_sec_idx = 0;
1325 server.ops_sec_last_sample_time = mstime();
1326 server.ops_sec_last_sample_ops = 0;
1327 server.unixtime = time(NULL);
1328 server.lastbgsave_status = REDIS_OK;
1329 server.stop_writes_on_bgsave_err = 1;
1330 aeCreateTimeEvent(server.el, 1, serverCron, NULL, NULL);
1331 if (server.ipfd > 0 && aeCreateFileEvent(server.el,server.ipfd,AE_READABLE,
1332 acceptTcpHandler,NULL) == AE_ERR) oom("creating file event");
1333 if (server.sofd > 0 && aeCreateFileEvent(server.el,server.sofd,AE_READABLE,
1334 acceptUnixHandler,NULL) == AE_ERR) oom("creating file event");
1335
1336 if (server.aof_state == REDIS_AOF_ON) {
1337 server.aof_fd = open(server.aof_filename,
1338 O_WRONLY|O_APPEND|O_CREAT,0644);
1339 if (server.aof_fd == -1) {
1340 redisLog(REDIS_WARNING, "Can't open the append-only file: %s",
1341 strerror(errno));
1342 exit(1);
1343 }
1344 }
1345
1346 /* 32 bit instances are limited to 4GB of address space, so if there is
1347 * no explicit limit in the user provided configuration we set a limit
1348 * at 3.5GB using maxmemory with 'noeviction' policy'. This saves
1349 * useless crashes of the Redis instance. */
1350 if (server.arch_bits == 32 && server.maxmemory == 0) {
1351 redisLog(REDIS_WARNING,"Warning: 32 bit instance detected but no memory limit set. Setting 3.5 GB maxmemory limit with 'noeviction' policy now.");
1352 server.maxmemory = 3584LL*(1024*1024); /* 3584 MB = 3.5 GB */
1353 server.maxmemory_policy = REDIS_MAXMEMORY_NO_EVICTION;
1354 }
1355
1356 if (server.cluster_enabled) clusterInit();
1357 scriptingInit();
1358 slowlogInit();
1359 bioInit();
1360 }
1361
1362 /* Populates the Redis Command Table starting from the hard coded list
1363 * we have on top of redis.c file. */
1364 void populateCommandTable(void) {
1365 int j;
1366 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1367
1368 for (j = 0; j < numcommands; j++) {
1369 struct redisCommand *c = redisCommandTable+j;
1370 char *f = c->sflags;
1371 int retval;
1372
1373 while(*f != '\0') {
1374 switch(*f) {
1375 case 'w': c->flags |= REDIS_CMD_WRITE; break;
1376 case 'r': c->flags |= REDIS_CMD_READONLY; break;
1377 case 'm': c->flags |= REDIS_CMD_DENYOOM; break;
1378 case 'a': c->flags |= REDIS_CMD_ADMIN; break;
1379 case 'p': c->flags |= REDIS_CMD_PUBSUB; break;
1380 case 'f': c->flags |= REDIS_CMD_FORCE_REPLICATION; break;
1381 case 's': c->flags |= REDIS_CMD_NOSCRIPT; break;
1382 case 'R': c->flags |= REDIS_CMD_RANDOM; break;
1383 case 'S': c->flags |= REDIS_CMD_SORT_FOR_SCRIPT; break;
1384 default: redisPanic("Unsupported command flag"); break;
1385 }
1386 f++;
1387 }
1388
1389 retval = dictAdd(server.commands, sdsnew(c->name), c);
1390 assert(retval == DICT_OK);
1391 }
1392 }
1393
1394 void resetCommandTableStats(void) {
1395 int numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
1396 int j;
1397
1398 for (j = 0; j < numcommands; j++) {
1399 struct redisCommand *c = redisCommandTable+j;
1400
1401 c->microseconds = 0;
1402 c->calls = 0;
1403 }
1404 }
1405
1406 /* ========================== Redis OP Array API ============================ */
1407
1408 void redisOpArrayInit(redisOpArray *oa) {
1409 oa->ops = NULL;
1410 oa->numops = 0;
1411 }
1412
1413 int redisOpArrayAppend(redisOpArray *oa, struct redisCommand *cmd, int dbid,
1414 robj **argv, int argc, int target)
1415 {
1416 redisOp *op;
1417
1418 oa->ops = zrealloc(oa->ops,sizeof(redisOp)*(oa->numops+1));
1419 op = oa->ops+oa->numops;
1420 op->cmd = cmd;
1421 op->dbid = dbid;
1422 op->argv = argv;
1423 op->argc = argc;
1424 op->target = target;
1425 oa->numops++;
1426 return oa->numops;
1427 }
1428
1429 void redisOpArrayFree(redisOpArray *oa) {
1430 while(oa->numops) {
1431 int j;
1432 redisOp *op;
1433
1434 oa->numops--;
1435 op = oa->ops+oa->numops;
1436 for (j = 0; j < op->argc; j++)
1437 decrRefCount(op->argv[j]);
1438 zfree(op->argv);
1439 }
1440 zfree(oa->ops);
1441 }
1442
1443 /* ====================== Commands lookup and execution ===================== */
1444
1445 struct redisCommand *lookupCommand(sds name) {
1446 return dictFetchValue(server.commands, name);
1447 }
1448
1449 struct redisCommand *lookupCommandByCString(char *s) {
1450 struct redisCommand *cmd;
1451 sds name = sdsnew(s);
1452
1453 cmd = dictFetchValue(server.commands, name);
1454 sdsfree(name);
1455 return cmd;
1456 }
1457
1458 /* Propagate the specified command (in the context of the specified database id)
1459 * to AOF, Slaves and Monitors.
1460 *
1461 * flags are an xor between:
1462 * + REDIS_PROPAGATE_NONE (no propagation of command at all)
1463 * + REDIS_PROPAGATE_AOF (propagate into the AOF file if is enabled)
1464 * + REDIS_PROPAGATE_REPL (propagate into the replication link)
1465 */
1466 void propagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
1467 int flags)
1468 {
1469 if (server.aof_state != REDIS_AOF_OFF && flags & REDIS_PROPAGATE_AOF)
1470 feedAppendOnlyFile(cmd,dbid,argv,argc);
1471 if (flags & REDIS_PROPAGATE_REPL && listLength(server.slaves))
1472 replicationFeedSlaves(server.slaves,dbid,argv,argc);
1473 }
1474
1475 /* Used inside commands to schedule the propagation of additional commands
1476 * after the current command is propagated to AOF / Replication. */
1477 void alsoPropagate(struct redisCommand *cmd, int dbid, robj **argv, int argc,
1478 int target)
1479 {
1480 redisOpArrayAppend(&server.also_propagate,cmd,dbid,argv,argc,target);
1481 }
1482
1483 /* Call() is the core of Redis execution of a command */
1484 void call(redisClient *c, int flags) {
1485 long long dirty, start = ustime(), duration;
1486
1487 /* Sent the command to clients in MONITOR mode, only if the commands are
1488 * not geneated from reading an AOF. */
1489 if (listLength(server.monitors) && !server.loading)
1490 replicationFeedMonitors(c,server.monitors,c->db->id,c->argv,c->argc);
1491
1492 /* Call the command. */
1493 redisOpArrayInit(&server.also_propagate);
1494 dirty = server.dirty;
1495 c->cmd->proc(c);
1496 dirty = server.dirty-dirty;
1497 duration = ustime()-start;
1498
1499 /* When EVAL is called loading the AOF we don't want commands called
1500 * from Lua to go into the slowlog or to populate statistics. */
1501 if (server.loading && c->flags & REDIS_LUA_CLIENT)
1502 flags &= ~(REDIS_CALL_SLOWLOG | REDIS_CALL_STATS);
1503
1504 /* Log the command into the Slow log if needed, and populate the
1505 * per-command statistics that we show in INFO commandstats. */
1506 if (flags & REDIS_CALL_SLOWLOG)
1507 slowlogPushEntryIfNeeded(c->argv,c->argc,duration);
1508 if (flags & REDIS_CALL_STATS) {
1509 c->cmd->microseconds += duration;
1510 c->cmd->calls++;
1511 }
1512
1513 /* Propagate the command into the AOF and replication link */
1514 if (flags & REDIS_CALL_PROPAGATE) {
1515 int flags = REDIS_PROPAGATE_NONE;
1516
1517 if (c->cmd->flags & REDIS_CMD_FORCE_REPLICATION)
1518 flags |= REDIS_PROPAGATE_REPL;
1519 if (dirty)
1520 flags |= (REDIS_PROPAGATE_REPL | REDIS_PROPAGATE_AOF);
1521 if (flags != REDIS_PROPAGATE_NONE)
1522 propagate(c->cmd,c->db->id,c->argv,c->argc,flags);
1523 }
1524 /* Commands such as LPUSH or BRPOPLPUSH may propagate an additional
1525 * PUSH command. */
1526 if (server.also_propagate.numops) {
1527 int j;
1528 redisOp *rop;
1529
1530 for (j = 0; j < server.also_propagate.numops; j++) {
1531 rop = &server.also_propagate.ops[j];
1532 propagate(rop->cmd, rop->dbid, rop->argv, rop->argc, rop->target);
1533 }
1534 redisOpArrayFree(&server.also_propagate);
1535 }
1536 server.stat_numcommands++;
1537 }
1538
1539 /* If this function gets called we already read a whole
1540 * command, argments are in the client argv/argc fields.
1541 * processCommand() execute the command or prepare the
1542 * server for a bulk read from the client.
1543 *
1544 * If 1 is returned the client is still alive and valid and
1545 * and other operations can be performed by the caller. Otherwise
1546 * if 0 is returned the client was destroied (i.e. after QUIT). */
1547 int processCommand(redisClient *c) {
1548 /* The QUIT command is handled separately. Normal command procs will
1549 * go through checking for replication and QUIT will cause trouble
1550 * when FORCE_REPLICATION is enabled and would be implemented in
1551 * a regular command proc. */
1552 if (!strcasecmp(c->argv[0]->ptr,"quit")) {
1553 addReply(c,shared.ok);
1554 c->flags |= REDIS_CLOSE_AFTER_REPLY;
1555 return REDIS_ERR;
1556 }
1557
1558 /* Now lookup the command and check ASAP about trivial error conditions
1559 * such as wrong arity, bad command name and so forth. */
1560 c->cmd = c->lastcmd = lookupCommand(c->argv[0]->ptr);
1561 if (!c->cmd) {
1562 addReplyErrorFormat(c,"unknown command '%s'",
1563 (char*)c->argv[0]->ptr);
1564 return REDIS_OK;
1565 } else if ((c->cmd->arity > 0 && c->cmd->arity != c->argc) ||
1566 (c->argc < -c->cmd->arity)) {
1567 addReplyErrorFormat(c,"wrong number of arguments for '%s' command",
1568 c->cmd->name);
1569 return REDIS_OK;
1570 }
1571
1572 /* Check if the user is authenticated */
1573 if (server.requirepass && !c->authenticated && c->cmd->proc != authCommand)
1574 {
1575 addReplyError(c,"operation not permitted");
1576 return REDIS_OK;
1577 }
1578
1579 /* If cluster is enabled, redirect here */
1580 if (server.cluster_enabled &&
1581 !(c->cmd->getkeys_proc == NULL && c->cmd->firstkey == 0)) {
1582 int hashslot;
1583
1584 if (server.cluster.state != REDIS_CLUSTER_OK) {
1585 addReplyError(c,"The cluster is down. Check with CLUSTER INFO for more information");
1586 return REDIS_OK;
1587 } else {
1588 int ask;
1589 clusterNode *n = getNodeByQuery(c,c->cmd,c->argv,c->argc,&hashslot,&ask);
1590 if (n == NULL) {
1591 addReplyError(c,"Multi keys request invalid in cluster");
1592 return REDIS_OK;
1593 } else if (n != server.cluster.myself) {
1594 addReplySds(c,sdscatprintf(sdsempty(),
1595 "-%s %d %s:%d\r\n", ask ? "ASK" : "MOVED",
1596 hashslot,n->ip,n->port));
1597 return REDIS_OK;
1598 }
1599 }
1600 }
1601
1602 /* Handle the maxmemory directive.
1603 *
1604 * First we try to free some memory if possible (if there are volatile
1605 * keys in the dataset). If there are not the only thing we can do
1606 * is returning an error. */
1607 if (server.maxmemory) {
1608 int retval = freeMemoryIfNeeded();
1609 if ((c->cmd->flags & REDIS_CMD_DENYOOM) && retval == REDIS_ERR) {
1610 addReply(c, shared.oomerr);
1611 return REDIS_OK;
1612 }
1613 }
1614
1615 /* Don't accept write commands if there are problems persisting on disk. */
1616 if (server.stop_writes_on_bgsave_err &&
1617 server.saveparamslen > 0
1618 && server.lastbgsave_status == REDIS_ERR &&
1619 c->cmd->flags & REDIS_CMD_WRITE)
1620 {
1621 addReply(c, shared.bgsaveerr);
1622 return REDIS_OK;
1623 }
1624
1625 /* Don't accept wirte commands if this is a read only slave. But
1626 * accept write commands if this is our master. */
1627 if (server.masterhost && server.repl_slave_ro &&
1628 !(c->flags & REDIS_MASTER) &&
1629 c->cmd->flags & REDIS_CMD_WRITE)
1630 {
1631 addReply(c, shared.roslaveerr);
1632 return REDIS_OK;
1633 }
1634
1635 /* Only allow SUBSCRIBE and UNSUBSCRIBE in the context of Pub/Sub */
1636 if ((dictSize(c->pubsub_channels) > 0 || listLength(c->pubsub_patterns) > 0)
1637 &&
1638 c->cmd->proc != subscribeCommand &&
1639 c->cmd->proc != unsubscribeCommand &&
1640 c->cmd->proc != psubscribeCommand &&
1641 c->cmd->proc != punsubscribeCommand) {
1642 addReplyError(c,"only (P)SUBSCRIBE / (P)UNSUBSCRIBE / QUIT allowed in this context");
1643 return REDIS_OK;
1644 }
1645
1646 /* Only allow INFO and SLAVEOF when slave-serve-stale-data is no and
1647 * we are a slave with a broken link with master. */
1648 if (server.masterhost && server.repl_state != REDIS_REPL_CONNECTED &&
1649 server.repl_serve_stale_data == 0 &&
1650 c->cmd->proc != infoCommand && c->cmd->proc != slaveofCommand)
1651 {
1652 addReply(c, shared.masterdownerr);
1653 return REDIS_OK;
1654 }
1655
1656 /* Loading DB? Return an error if the command is not INFO */
1657 if (server.loading && c->cmd->proc != infoCommand) {
1658 addReply(c, shared.loadingerr);
1659 return REDIS_OK;
1660 }
1661
1662 /* Lua script too slow? Only allow SHUTDOWN NOSAVE and SCRIPT KILL. */
1663 if (server.lua_timedout &&
1664 !(c->cmd->proc == shutdownCommand &&
1665 c->argc == 2 &&
1666 tolower(((char*)c->argv[1]->ptr)[0]) == 'n') &&
1667 !(c->cmd->proc == scriptCommand &&
1668 c->argc == 2 &&
1669 tolower(((char*)c->argv[1]->ptr)[0]) == 'k'))
1670 {
1671 addReply(c, shared.slowscripterr);
1672 return REDIS_OK;
1673 }
1674
1675 /* Exec the command */
1676 if (c->flags & REDIS_MULTI &&
1677 c->cmd->proc != execCommand && c->cmd->proc != discardCommand &&
1678 c->cmd->proc != multiCommand && c->cmd->proc != watchCommand)
1679 {
1680 queueMultiCommand(c);
1681 addReply(c,shared.queued);
1682 } else {
1683 call(c,REDIS_CALL_FULL);
1684 }
1685 return REDIS_OK;
1686 }
1687
1688 /*================================== Shutdown =============================== */
1689
1690 int prepareForShutdown(int flags) {
1691 int save = flags & REDIS_SHUTDOWN_SAVE;
1692 int nosave = flags & REDIS_SHUTDOWN_NOSAVE;
1693
1694 redisLog(REDIS_WARNING,"User requested shutdown...");
1695 /* Kill the saving child if there is a background saving in progress.
1696 We want to avoid race conditions, for instance our saving child may
1697 overwrite the synchronous saving did by SHUTDOWN. */
1698 if (server.rdb_child_pid != -1) {
1699 redisLog(REDIS_WARNING,"There is a child saving an .rdb. Killing it!");
1700 kill(server.rdb_child_pid,SIGKILL);
1701 rdbRemoveTempFile(server.rdb_child_pid);
1702 }
1703 if (server.aof_state != REDIS_AOF_OFF) {
1704 /* Kill the AOF saving child as the AOF we already have may be longer
1705 * but contains the full dataset anyway. */
1706 if (server.aof_child_pid != -1) {
1707 redisLog(REDIS_WARNING,
1708 "There is a child rewriting the AOF. Killing it!");
1709 kill(server.aof_child_pid,SIGKILL);
1710 }
1711 /* Append only file: fsync() the AOF and exit */
1712 redisLog(REDIS_NOTICE,"Calling fsync() on the AOF file.");
1713 aof_fsync(server.aof_fd);
1714 }
1715 if ((server.saveparamslen > 0 && !nosave) || save) {
1716 redisLog(REDIS_NOTICE,"Saving the final RDB snapshot before exiting.");
1717 /* Snapshotting. Perform a SYNC SAVE and exit */
1718 if (rdbSave(server.rdb_filename) != REDIS_OK) {
1719 /* Ooops.. error saving! The best we can do is to continue
1720 * operating. Note that if there was a background saving process,
1721 * in the next cron() Redis will be notified that the background
1722 * saving aborted, handling special stuff like slaves pending for
1723 * synchronization... */
1724 redisLog(REDIS_WARNING,"Error trying to save the DB, can't exit.");
1725 return REDIS_ERR;
1726 }
1727 }
1728 if (server.daemonize) {
1729 redisLog(REDIS_NOTICE,"Removing the pid file.");
1730 unlink(server.pidfile);
1731 }
1732 /* Close the listening sockets. Apparently this allows faster restarts. */
1733 if (server.ipfd != -1) close(server.ipfd);
1734 if (server.sofd != -1) close(server.sofd);
1735 if (server.unixsocket) {
1736 redisLog(REDIS_NOTICE,"Removing the unix socket file.");
1737 unlink(server.unixsocket); /* don't care if this fails */
1738 }
1739
1740 redisLog(REDIS_WARNING,"Redis is now ready to exit, bye bye...");
1741 return REDIS_OK;
1742 }
1743
1744 /*================================== Commands =============================== */
1745
1746 /* Return zero if strings are the same, non-zero if they are not.
1747 * The comparison is performed in a way that prevents an attacker to obtain
1748 * information about the nature of the strings just monitoring the execution
1749 * time of the function.
1750 *
1751 * Note that limiting the comparison length to strings up to 512 bytes we
1752 * can avoid leaking any information about the password length and any
1753 * possible branch misprediction related leak.
1754 */
1755 int time_independent_strcmp(char *a, char *b) {
1756 char bufa[REDIS_AUTHPASS_MAX_LEN], bufb[REDIS_AUTHPASS_MAX_LEN];
1757 /* The above two strlen perform len(a) + len(b) operations where either
1758 * a or b are fixed (our password) length, and the difference is only
1759 * relative to the length of the user provided string, so no information
1760 * leak is possible in the following two lines of code. */
1761 int alen = strlen(a);
1762 int blen = strlen(b);
1763 int j;
1764 int diff = 0;
1765
1766 /* We can't compare strings longer than our static buffers.
1767 * Note that this will never pass the first test in practical circumstances
1768 * so there is no info leak. */
1769 if (alen > sizeof(bufa) || blen > sizeof(bufb)) return 1;
1770
1771 memset(bufa,0,sizeof(bufa)); /* Constant time. */
1772 memset(bufb,0,sizeof(bufb)); /* Constant time. */
1773 /* Again the time of the following two copies is proportional to
1774 * len(a) + len(b) so no info is leaked. */
1775 memcpy(bufa,a,alen);
1776 memcpy(bufb,b,blen);
1777
1778 /* Always compare all the chars in the two buffers without
1779 * conditional expressions. */
1780 for (j = 0; j < sizeof(bufa); j++) {
1781 diff |= (bufa[j] ^ bufb[j]);
1782 }
1783 /* Length must be equal as well. */
1784 diff |= alen ^ blen;
1785 return diff; /* If zero strings are the same. */
1786 }
1787
1788 void authCommand(redisClient *c) {
1789 if (!server.requirepass) {
1790 addReplyError(c,"Client sent AUTH, but no password is set");
1791 } else if (!time_independent_strcmp(c->argv[1]->ptr, server.requirepass)) {
1792 c->authenticated = 1;
1793 addReply(c,shared.ok);
1794 } else {
1795 c->authenticated = 0;
1796 addReplyError(c,"invalid password");
1797 }
1798 }
1799
1800 void pingCommand(redisClient *c) {
1801 addReply(c,shared.pong);
1802 }
1803
1804 void echoCommand(redisClient *c) {
1805 addReplyBulk(c,c->argv[1]);
1806 }
1807
1808 void timeCommand(redisClient *c) {
1809 struct timeval tv;
1810
1811 /* gettimeofday() can only fail if &tv is a bad addresss so we
1812 * don't check for errors. */
1813 gettimeofday(&tv,NULL);
1814 addReplyMultiBulkLen(c,2);
1815 addReplyBulkLongLong(c,tv.tv_sec);
1816 addReplyBulkLongLong(c,tv.tv_usec);
1817 }
1818
1819 /* Convert an amount of bytes into a human readable string in the form
1820 * of 100B, 2G, 100M, 4K, and so forth. */
1821 void bytesToHuman(char *s, unsigned long long n) {
1822 double d;
1823
1824 if (n < 1024) {
1825 /* Bytes */
1826 sprintf(s,"%lluB",n);
1827 return;
1828 } else if (n < (1024*1024)) {
1829 d = (double)n/(1024);
1830 sprintf(s,"%.2fK",d);
1831 } else if (n < (1024LL*1024*1024)) {
1832 d = (double)n/(1024*1024);
1833 sprintf(s,"%.2fM",d);
1834 } else if (n < (1024LL*1024*1024*1024)) {
1835 d = (double)n/(1024LL*1024*1024);
1836 sprintf(s,"%.2fG",d);
1837 }
1838 }
1839
1840 /* Create the string returned by the INFO command. This is decoupled
1841 * by the INFO command itself as we need to report the same information
1842 * on memory corruption problems. */
1843 sds genRedisInfoString(char *section) {
1844 sds info = sdsempty();
1845 time_t uptime = server.unixtime-server.stat_starttime;
1846 int j, numcommands;
1847 struct rusage self_ru, c_ru;
1848 unsigned long lol, bib;
1849 int allsections = 0, defsections = 0;
1850 int sections = 0;
1851
1852 if (section) {
1853 allsections = strcasecmp(section,"all") == 0;
1854 defsections = strcasecmp(section,"default") == 0;
1855 }
1856
1857 getrusage(RUSAGE_SELF, &self_ru);
1858 getrusage(RUSAGE_CHILDREN, &c_ru);
1859 getClientsMaxBuffers(&lol,&bib);
1860
1861 /* Server */
1862 if (allsections || defsections || !strcasecmp(section,"server")) {
1863 struct utsname name;
1864
1865 if (sections++) info = sdscat(info,"\r\n");
1866 uname(&name);
1867 info = sdscatprintf(info,
1868 "# Server\r\n"
1869 "redis_version:%s\r\n"
1870 "redis_git_sha1:%s\r\n"
1871 "redis_git_dirty:%d\r\n"
1872 "os:%s %s %s\r\n"
1873 "arch_bits:%d\r\n"
1874 "multiplexing_api:%s\r\n"
1875 "gcc_version:%d.%d.%d\r\n"
1876 "process_id:%ld\r\n"
1877 "run_id:%s\r\n"
1878 "tcp_port:%d\r\n"
1879 "uptime_in_seconds:%ld\r\n"
1880 "uptime_in_days:%ld\r\n"
1881 "lru_clock:%ld\r\n",
1882 REDIS_VERSION,
1883 redisGitSHA1(),
1884 strtol(redisGitDirty(),NULL,10) > 0,
1885 name.sysname, name.release, name.machine,
1886 server.arch_bits,
1887 aeGetApiName(),
1888 #ifdef __GNUC__
1889 __GNUC__,__GNUC_MINOR__,__GNUC_PATCHLEVEL__,
1890 #else
1891 0,0,0,
1892 #endif
1893 (long) getpid(),
1894 server.runid,
1895 server.port,
1896 uptime,
1897 uptime/(3600*24),
1898 (unsigned long) server.lruclock);
1899 }
1900
1901 /* Clients */
1902 if (allsections || defsections || !strcasecmp(section,"clients")) {
1903 if (sections++) info = sdscat(info,"\r\n");
1904 info = sdscatprintf(info,
1905 "# Clients\r\n"
1906 "connected_clients:%lu\r\n"
1907 "client_longest_output_list:%lu\r\n"
1908 "client_biggest_input_buf:%lu\r\n"
1909 "blocked_clients:%d\r\n",
1910 listLength(server.clients)-listLength(server.slaves),
1911 lol, bib,
1912 server.bpop_blocked_clients);
1913 }
1914
1915 /* Memory */
1916 if (allsections || defsections || !strcasecmp(section,"memory")) {
1917 char hmem[64];
1918 char peak_hmem[64];
1919
1920 bytesToHuman(hmem,zmalloc_used_memory());
1921 bytesToHuman(peak_hmem,server.stat_peak_memory);
1922 if (sections++) info = sdscat(info,"\r\n");
1923 info = sdscatprintf(info,
1924 "# Memory\r\n"
1925 "used_memory:%zu\r\n"
1926 "used_memory_human:%s\r\n"
1927 "used_memory_rss:%zu\r\n"
1928 "used_memory_peak:%zu\r\n"
1929 "used_memory_peak_human:%s\r\n"
1930 "used_memory_lua:%lld\r\n"
1931 "mem_fragmentation_ratio:%.2f\r\n"
1932 "mem_allocator:%s\r\n",
1933 zmalloc_used_memory(),
1934 hmem,
1935 zmalloc_get_rss(),
1936 server.stat_peak_memory,
1937 peak_hmem,
1938 ((long long)lua_gc(server.lua,LUA_GCCOUNT,0))*1024LL,
1939 zmalloc_get_fragmentation_ratio(),
1940 ZMALLOC_LIB
1941 );
1942 }
1943
1944 /* Persistence */
1945 if (allsections || defsections || !strcasecmp(section,"persistence")) {
1946 if (sections++) info = sdscat(info,"\r\n");
1947 info = sdscatprintf(info,
1948 "# Persistence\r\n"
1949 "loading:%d\r\n"
1950 "rdb_changes_since_last_save:%lld\r\n"
1951 "rdb_bgsave_in_progress:%d\r\n"
1952 "rdb_last_save_time:%ld\r\n"
1953 "rdb_last_bgsave_status:%s\r\n"
1954 "rdb_last_bgsave_time_sec:%ld\r\n"
1955 "rdb_current_bgsave_time_sec:%ld\r\n"
1956 "aof_enabled:%d\r\n"
1957 "aof_rewrite_in_progress:%d\r\n"
1958 "aof_rewrite_scheduled:%d\r\n"
1959 "aof_last_rewrite_time_sec:%ld\r\n"
1960 "aof_current_rewrite_time_sec:%ld\r\n",
1961 server.loading,
1962 server.dirty,
1963 server.rdb_child_pid != -1,
1964 server.lastsave,
1965 server.lastbgsave_status == REDIS_OK ? "ok" : "err",
1966 server.rdb_save_time_last,
1967 (server.rdb_child_pid == -1) ?
1968 -1 : time(NULL)-server.rdb_save_time_start,
1969 server.aof_state != REDIS_AOF_OFF,
1970 server.aof_child_pid != -1,
1971 server.aof_rewrite_scheduled,
1972 server.aof_rewrite_time_last,
1973 (server.aof_child_pid == -1) ?
1974 -1 : time(NULL)-server.aof_rewrite_time_start);
1975
1976 if (server.aof_state != REDIS_AOF_OFF) {
1977 info = sdscatprintf(info,
1978 "aof_current_size:%lld\r\n"
1979 "aof_base_size:%lld\r\n"
1980 "aof_pending_rewrite:%d\r\n"
1981 "aof_buffer_length:%zu\r\n"
1982 "aof_rewrite_buffer_length:%zu\r\n"
1983 "aof_pending_bio_fsync:%llu\r\n"
1984 "aof_delayed_fsync:%lu\r\n",
1985 (long long) server.aof_current_size,
1986 (long long) server.aof_rewrite_base_size,
1987 server.aof_rewrite_scheduled,
1988 sdslen(server.aof_buf),
1989 aofRewriteBufferSize(),
1990 bioPendingJobsOfType(REDIS_BIO_AOF_FSYNC),
1991 server.aof_delayed_fsync);
1992 }
1993
1994 if (server.loading) {
1995 double perc;
1996 time_t eta, elapsed;
1997 off_t remaining_bytes = server.loading_total_bytes-
1998 server.loading_loaded_bytes;
1999
2000 perc = ((double)server.loading_loaded_bytes /
2001 server.loading_total_bytes) * 100;
2002
2003 elapsed = server.unixtime-server.loading_start_time;
2004 if (elapsed == 0) {
2005 eta = 1; /* A fake 1 second figure if we don't have
2006 enough info */
2007 } else {
2008 eta = (elapsed*remaining_bytes)/server.loading_loaded_bytes;
2009 }
2010
2011 info = sdscatprintf(info,
2012 "loading_start_time:%ld\r\n"
2013 "loading_total_bytes:%llu\r\n"
2014 "loading_loaded_bytes:%llu\r\n"
2015 "loading_loaded_perc:%.2f\r\n"
2016 "loading_eta_seconds:%ld\r\n"
2017 ,(unsigned long) server.loading_start_time,
2018 (unsigned long long) server.loading_total_bytes,
2019 (unsigned long long) server.loading_loaded_bytes,
2020 perc,
2021 eta
2022 );
2023 }
2024 }
2025
2026 /* Stats */
2027 if (allsections || defsections || !strcasecmp(section,"stats")) {
2028 if (sections++) info = sdscat(info,"\r\n");
2029 info = sdscatprintf(info,
2030 "# Stats\r\n"
2031 "total_connections_received:%lld\r\n"
2032 "total_commands_processed:%lld\r\n"
2033 "instantaneous_ops_per_sec:%lld\r\n"
2034 "rejected_connections:%lld\r\n"
2035 "expired_keys:%lld\r\n"
2036 "evicted_keys:%lld\r\n"
2037 "keyspace_hits:%lld\r\n"
2038 "keyspace_misses:%lld\r\n"
2039 "pubsub_channels:%ld\r\n"
2040 "pubsub_patterns:%lu\r\n"
2041 "latest_fork_usec:%lld\r\n",
2042 server.stat_numconnections,
2043 server.stat_numcommands,
2044 getOperationsPerSecond(),
2045 server.stat_rejected_conn,
2046 server.stat_expiredkeys,
2047 server.stat_evictedkeys,
2048 server.stat_keyspace_hits,
2049 server.stat_keyspace_misses,
2050 dictSize(server.pubsub_channels),
2051 listLength(server.pubsub_patterns),
2052 server.stat_fork_time);
2053 }
2054
2055 /* Replication */
2056 if (allsections || defsections || !strcasecmp(section,"replication")) {
2057 if (sections++) info = sdscat(info,"\r\n");
2058 info = sdscatprintf(info,
2059 "# Replication\r\n"
2060 "role:%s\r\n",
2061 server.masterhost == NULL ? "master" : "slave");
2062 if (server.masterhost) {
2063 info = sdscatprintf(info,
2064 "master_host:%s\r\n"
2065 "master_port:%d\r\n"
2066 "master_link_status:%s\r\n"
2067 "master_last_io_seconds_ago:%d\r\n"
2068 "master_sync_in_progress:%d\r\n"
2069 ,server.masterhost,
2070 server.masterport,
2071 (server.repl_state == REDIS_REPL_CONNECTED) ?
2072 "up" : "down",
2073 server.master ?
2074 ((int)(server.unixtime-server.master->lastinteraction)) : -1,
2075 server.repl_state == REDIS_REPL_TRANSFER
2076 );
2077
2078 if (server.repl_state == REDIS_REPL_TRANSFER) {
2079 info = sdscatprintf(info,
2080 "master_sync_left_bytes:%ld\r\n"
2081 "master_sync_last_io_seconds_ago:%d\r\n"
2082 ,(long)server.repl_transfer_left,
2083 (int)(server.unixtime-server.repl_transfer_lastio)
2084 );
2085 }
2086
2087 if (server.repl_state != REDIS_REPL_CONNECTED) {
2088 info = sdscatprintf(info,
2089 "master_link_down_since_seconds:%ld\r\n",
2090 (long)server.unixtime-server.repl_down_since);
2091 }
2092 }
2093 info = sdscatprintf(info,
2094 "connected_slaves:%lu\r\n",
2095 listLength(server.slaves));
2096 if (listLength(server.slaves)) {
2097 int slaveid = 0;
2098 listNode *ln;
2099 listIter li;
2100
2101 listRewind(server.slaves,&li);
2102 while((ln = listNext(&li))) {
2103 redisClient *slave = listNodeValue(ln);
2104 char *state = NULL;
2105 char ip[32];
2106 int port;
2107
2108 if (anetPeerToString(slave->fd,ip,&port) == -1) continue;
2109 switch(slave->replstate) {
2110 case REDIS_REPL_WAIT_BGSAVE_START:
2111 case REDIS_REPL_WAIT_BGSAVE_END:
2112 state = "wait_bgsave";
2113 break;
2114 case REDIS_REPL_SEND_BULK:
2115 state = "send_bulk";
2116 break;
2117 case REDIS_REPL_ONLINE:
2118 state = "online";
2119 break;
2120 }
2121 if (state == NULL) continue;
2122 info = sdscatprintf(info,"slave%d:%s,%d,%s\r\n",
2123 slaveid,ip,slave->slave_listening_port,state);
2124 slaveid++;
2125 }
2126 }
2127 }
2128
2129 /* CPU */
2130 if (allsections || defsections || !strcasecmp(section,"cpu")) {
2131 if (sections++) info = sdscat(info,"\r\n");
2132 info = sdscatprintf(info,
2133 "# CPU\r\n"
2134 "used_cpu_sys:%.2f\r\n"
2135 "used_cpu_user:%.2f\r\n"
2136 "used_cpu_sys_children:%.2f\r\n"
2137 "used_cpu_user_children:%.2f\r\n",
2138 (float)self_ru.ru_stime.tv_sec+(float)self_ru.ru_stime.tv_usec/1000000,
2139 (float)self_ru.ru_utime.tv_sec+(float)self_ru.ru_utime.tv_usec/1000000,
2140 (float)c_ru.ru_stime.tv_sec+(float)c_ru.ru_stime.tv_usec/1000000,
2141 (float)c_ru.ru_utime.tv_sec+(float)c_ru.ru_utime.tv_usec/1000000);
2142 }
2143
2144 /* cmdtime */
2145 if (allsections || !strcasecmp(section,"commandstats")) {
2146 if (sections++) info = sdscat(info,"\r\n");
2147 info = sdscatprintf(info, "# Commandstats\r\n");
2148 numcommands = sizeof(redisCommandTable)/sizeof(struct redisCommand);
2149 for (j = 0; j < numcommands; j++) {
2150 struct redisCommand *c = redisCommandTable+j;
2151
2152 if (!c->calls) continue;
2153 info = sdscatprintf(info,
2154 "cmdstat_%s:calls=%lld,usec=%lld,usec_per_call=%.2f\r\n",
2155 c->name, c->calls, c->microseconds,
2156 (c->calls == 0) ? 0 : ((float)c->microseconds/c->calls));
2157 }
2158 }
2159
2160 /* Cluster */
2161 if (allsections || defsections || !strcasecmp(section,"cluster")) {
2162 if (sections++) info = sdscat(info,"\r\n");
2163 info = sdscatprintf(info,
2164 "# Cluster\r\n"
2165 "cluster_enabled:%d\r\n",
2166 server.cluster_enabled);
2167 }
2168
2169 /* Key space */
2170 if (allsections || defsections || !strcasecmp(section,"keyspace")) {
2171 if (sections++) info = sdscat(info,"\r\n");
2172 info = sdscatprintf(info, "# Keyspace\r\n");
2173 for (j = 0; j < server.dbnum; j++) {
2174 long long keys, vkeys;
2175
2176 keys = dictSize(server.db[j].dict);
2177 vkeys = dictSize(server.db[j].expires);
2178 if (keys || vkeys) {
2179 info = sdscatprintf(info, "db%d:keys=%lld,expires=%lld\r\n",
2180 j, keys, vkeys);
2181 }
2182 }
2183 }
2184 return info;
2185 }
2186
2187 void infoCommand(redisClient *c) {
2188 char *section = c->argc == 2 ? c->argv[1]->ptr : "default";
2189
2190 if (c->argc > 2) {
2191 addReply(c,shared.syntaxerr);
2192 return;
2193 }
2194 sds info = genRedisInfoString(section);
2195 addReplySds(c,sdscatprintf(sdsempty(),"$%lu\r\n",
2196 (unsigned long)sdslen(info)));
2197 addReplySds(c,info);
2198 addReply(c,shared.crlf);
2199 }
2200
2201 void monitorCommand(redisClient *c) {
2202 /* ignore MONITOR if aleady slave or in monitor mode */
2203 if (c->flags & REDIS_SLAVE) return;
2204
2205 c->flags |= (REDIS_SLAVE|REDIS_MONITOR);
2206 c->slaveseldb = 0;
2207 listAddNodeTail(server.monitors,c);
2208 addReply(c,shared.ok);
2209 }
2210
2211 /* ============================ Maxmemory directive ======================== */
2212
2213 /* This function gets called when 'maxmemory' is set on the config file to limit
2214 * the max memory used by the server, before processing a command.
2215 *
2216 * The goal of the function is to free enough memory to keep Redis under the
2217 * configured memory limit.
2218 *
2219 * The function starts calculating how many bytes should be freed to keep
2220 * Redis under the limit, and enters a loop selecting the best keys to
2221 * evict accordingly to the configured policy.
2222 *
2223 * If all the bytes needed to return back under the limit were freed the
2224 * function returns REDIS_OK, otherwise REDIS_ERR is returned, and the caller
2225 * should block the execution of commands that will result in more memory
2226 * used by the server.
2227 */
2228 int freeMemoryIfNeeded(void) {
2229 size_t mem_used, mem_tofree, mem_freed;
2230 int slaves = listLength(server.slaves);
2231
2232 /* Remove the size of slaves output buffers and AOF buffer from the
2233 * count of used memory. */
2234 mem_used = zmalloc_used_memory();
2235 if (slaves) {
2236 listIter li;
2237 listNode *ln;
2238
2239 listRewind(server.slaves,&li);
2240 while((ln = listNext(&li))) {
2241 redisClient *slave = listNodeValue(ln);
2242 unsigned long obuf_bytes = getClientOutputBufferMemoryUsage(slave);
2243 if (obuf_bytes > mem_used)
2244 mem_used = 0;
2245 else
2246 mem_used -= obuf_bytes;
2247 }
2248 }
2249 if (server.aof_state != REDIS_AOF_OFF) {
2250 mem_used -= sdslen(server.aof_buf);
2251 mem_used -= aofRewriteBufferSize();
2252 }
2253
2254 /* Check if we are over the memory limit. */
2255 if (mem_used <= server.maxmemory) return REDIS_OK;
2256
2257 if (server.maxmemory_policy == REDIS_MAXMEMORY_NO_EVICTION)
2258 return REDIS_ERR; /* We need to free memory, but policy forbids. */
2259
2260 /* Compute how much memory we need to free. */
2261 mem_tofree = mem_used - server.maxmemory;
2262 mem_freed = 0;
2263 while (mem_freed < mem_tofree) {
2264 int j, k, keys_freed = 0;
2265
2266 for (j = 0; j < server.dbnum; j++) {
2267 long bestval = 0; /* just to prevent warning */
2268 sds bestkey = NULL;
2269 struct dictEntry *de;
2270 redisDb *db = server.db+j;
2271 dict *dict;
2272
2273 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
2274 server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM)
2275 {
2276 dict = server.db[j].dict;
2277 } else {
2278 dict = server.db[j].expires;
2279 }
2280 if (dictSize(dict) == 0) continue;
2281
2282 /* volatile-random and allkeys-random policy */
2283 if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_RANDOM ||
2284 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_RANDOM)
2285 {
2286 de = dictGetRandomKey(dict);
2287 bestkey = dictGetKey(de);
2288 }
2289
2290 /* volatile-lru and allkeys-lru policy */
2291 else if (server.maxmemory_policy == REDIS_MAXMEMORY_ALLKEYS_LRU ||
2292 server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
2293 {
2294 for (k = 0; k < server.maxmemory_samples; k++) {
2295 sds thiskey;
2296 long thisval;
2297 robj *o;
2298
2299 de = dictGetRandomKey(dict);
2300 thiskey = dictGetKey(de);
2301 /* When policy is volatile-lru we need an additonal lookup
2302 * to locate the real key, as dict is set to db->expires. */
2303 if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_LRU)
2304 de = dictFind(db->dict, thiskey);
2305 o = dictGetVal(de);
2306 thisval = estimateObjectIdleTime(o);
2307
2308 /* Higher idle time is better candidate for deletion */
2309 if (bestkey == NULL || thisval > bestval) {
2310 bestkey = thiskey;
2311 bestval = thisval;
2312 }
2313 }
2314 }
2315
2316 /* volatile-ttl */
2317 else if (server.maxmemory_policy == REDIS_MAXMEMORY_VOLATILE_TTL) {
2318 for (k = 0; k < server.maxmemory_samples; k++) {
2319 sds thiskey;
2320 long thisval;
2321
2322 de = dictGetRandomKey(dict);
2323 thiskey = dictGetKey(de);
2324 thisval = (long) dictGetVal(de);
2325
2326 /* Expire sooner (minor expire unix timestamp) is better
2327 * candidate for deletion */
2328 if (bestkey == NULL || thisval < bestval) {
2329 bestkey = thiskey;
2330 bestval = thisval;
2331 }
2332 }
2333 }
2334
2335 /* Finally remove the selected key. */
2336 if (bestkey) {
2337 long long delta;
2338
2339 robj *keyobj = createStringObject(bestkey,sdslen(bestkey));
2340 propagateExpire(db,keyobj);
2341 /* We compute the amount of memory freed by dbDelete() alone.
2342 * It is possible that actually the memory needed to propagate
2343 * the DEL in AOF and replication link is greater than the one
2344 * we are freeing removing the key, but we can't account for
2345 * that otherwise we would never exit the loop.
2346 *
2347 * AOF and Output buffer memory will be freed eventually so
2348 * we only care about memory used by the key space. */
2349 delta = (long long) zmalloc_used_memory();
2350 dbDelete(db,keyobj);
2351 delta -= (long long) zmalloc_used_memory();
2352 mem_freed += delta;
2353 server.stat_evictedkeys++;
2354 decrRefCount(keyobj);
2355 keys_freed++;
2356
2357 /* When the memory to free starts to be big enough, we may
2358 * start spending so much time here that is impossible to
2359 * deliver data to the slaves fast enough, so we force the
2360 * transmission here inside the loop. */
2361 if (slaves) flushSlavesOutputBuffers();
2362 }
2363 }
2364 if (!keys_freed) return REDIS_ERR; /* nothing to free... */
2365 }
2366 return REDIS_OK;
2367 }
2368
2369 /* =================================== Main! ================================ */
2370
2371 #ifdef __linux__
2372 int linuxOvercommitMemoryValue(void) {
2373 FILE *fp = fopen("/proc/sys/vm/overcommit_memory","r");
2374 char buf[64];
2375
2376 if (!fp) return -1;
2377 if (fgets(buf,64,fp) == NULL) {
2378 fclose(fp);
2379 return -1;
2380 }
2381 fclose(fp);
2382
2383 return atoi(buf);
2384 }
2385
2386 void linuxOvercommitMemoryWarning(void) {
2387 if (linuxOvercommitMemoryValue() == 0) {
2388 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.");
2389 }
2390 }
2391 #endif /* __linux__ */
2392
2393 void createPidFile(void) {
2394 /* Try to write the pid file in a best-effort way. */
2395 FILE *fp = fopen(server.pidfile,"w");
2396 if (fp) {
2397 fprintf(fp,"%d\n",(int)getpid());
2398 fclose(fp);
2399 }
2400 }
2401
2402 void daemonize(void) {
2403 int fd;
2404
2405 if (fork() != 0) exit(0); /* parent exits */
2406 setsid(); /* create a new session */
2407
2408 /* Every output goes to /dev/null. If Redis is daemonized but
2409 * the 'logfile' is set to 'stdout' in the configuration file
2410 * it will not log at all. */
2411 if ((fd = open("/dev/null", O_RDWR, 0)) != -1) {
2412 dup2(fd, STDIN_FILENO);
2413 dup2(fd, STDOUT_FILENO);
2414 dup2(fd, STDERR_FILENO);
2415 if (fd > STDERR_FILENO) close(fd);
2416 }
2417 }
2418
2419 void version() {
2420 printf("Redis server v=%s sha=%s:%d malloc=%s bits=%d\n",
2421 REDIS_VERSION,
2422 redisGitSHA1(),
2423 atoi(redisGitDirty()) > 0,
2424 ZMALLOC_LIB,
2425 sizeof(long) == 4 ? 32 : 64);
2426 exit(0);
2427 }
2428
2429 void usage() {
2430 fprintf(stderr,"Usage: ./redis-server [/path/to/redis.conf] [options]\n");
2431 fprintf(stderr," ./redis-server - (read config from stdin)\n");
2432 fprintf(stderr," ./redis-server -v or --version\n");
2433 fprintf(stderr," ./redis-server -h or --help\n");
2434 fprintf(stderr," ./redis-server --test-memory <megabytes>\n\n");
2435 fprintf(stderr,"Examples:\n");
2436 fprintf(stderr," ./redis-server (run the server with default conf)\n");
2437 fprintf(stderr," ./redis-server /etc/redis/6379.conf\n");
2438 fprintf(stderr," ./redis-server --port 7777\n");
2439 fprintf(stderr," ./redis-server --port 7777 --slaveof 127.0.0.1 8888\n");
2440 fprintf(stderr," ./redis-server /etc/myredis.conf --loglevel verbose\n");
2441 exit(1);
2442 }
2443
2444 void redisAsciiArt(void) {
2445 #include "asciilogo.h"
2446 char *buf = zmalloc(1024*16);
2447
2448 snprintf(buf,1024*16,ascii_logo,
2449 REDIS_VERSION,
2450 redisGitSHA1(),
2451 strtol(redisGitDirty(),NULL,10) > 0,
2452 (sizeof(long) == 8) ? "64" : "32",
2453 server.cluster_enabled ? "cluster" : "stand alone",
2454 server.port,
2455 (long) getpid()
2456 );
2457 redisLogRaw(REDIS_NOTICE|REDIS_LOG_RAW,buf);
2458 zfree(buf);
2459 }
2460
2461 static void sigtermHandler(int sig) {
2462 REDIS_NOTUSED(sig);
2463
2464 redisLogFromHandler(REDIS_WARNING,"Received SIGTERM, scheduling shutdown...");
2465 server.shutdown_asap = 1;
2466 }
2467
2468 void setupSignalHandlers(void) {
2469 struct sigaction act;
2470
2471 /* When the SA_SIGINFO flag is set in sa_flags then sa_sigaction is used.
2472 * Otherwise, sa_handler is used. */
2473 sigemptyset(&act.sa_mask);
2474 act.sa_flags = 0;
2475 act.sa_handler = sigtermHandler;
2476 sigaction(SIGTERM, &act, NULL);
2477
2478 #ifdef HAVE_BACKTRACE
2479 sigemptyset(&act.sa_mask);
2480 act.sa_flags = SA_NODEFER | SA_RESETHAND | SA_SIGINFO;
2481 act.sa_sigaction = sigsegvHandler;
2482 sigaction(SIGSEGV, &act, NULL);
2483 sigaction(SIGBUS, &act, NULL);
2484 sigaction(SIGFPE, &act, NULL);
2485 sigaction(SIGILL, &act, NULL);
2486 #endif
2487 return;
2488 }
2489
2490 void memtest(size_t megabytes, int passes);
2491
2492 int main(int argc, char **argv) {
2493 long long start;
2494 struct timeval tv;
2495
2496 /* We need to initialize our libraries, and the server configuration. */
2497 zmalloc_enable_thread_safeness();
2498 srand(time(NULL)^getpid());
2499 gettimeofday(&tv,NULL);
2500 dictSetHashFunctionSeed(tv.tv_sec^tv.tv_usec^getpid());
2501 initServerConfig();
2502
2503 if (argc >= 2) {
2504 int j = 1; /* First option to parse in argv[] */
2505 sds options = sdsempty();
2506 char *configfile = NULL;
2507
2508 /* Handle special options --help and --version */
2509 if (strcmp(argv[1], "-v") == 0 ||
2510 strcmp(argv[1], "--version") == 0) version();
2511 if (strcmp(argv[1], "--help") == 0 ||
2512 strcmp(argv[1], "-h") == 0) usage();
2513 if (strcmp(argv[1], "--test-memory") == 0) {
2514 if (argc == 3) {
2515 memtest(atoi(argv[2]),50);
2516 exit(0);
2517 } else {
2518 fprintf(stderr,"Please specify the amount of memory to test in megabytes.\n");
2519 fprintf(stderr,"Example: ./redis-server --test-memory 4096\n\n");
2520 exit(1);
2521 }
2522 }
2523
2524 /* First argument is the config file name? */
2525 if (argv[j][0] != '-' || argv[j][1] != '-')
2526 configfile = argv[j++];
2527 /* All the other options are parsed and conceptually appended to the
2528 * configuration file. For instance --port 6380 will generate the
2529 * string "port 6380\n" to be parsed after the actual file name
2530 * is parsed, if any. */
2531 while(j != argc) {
2532 if (argv[j][0] == '-' && argv[j][1] == '-') {
2533 /* Option name */
2534 if (sdslen(options)) options = sdscat(options,"\n");
2535 options = sdscat(options,argv[j]+2);
2536 options = sdscat(options," ");
2537 } else {
2538 /* Option argument */
2539 options = sdscatrepr(options,argv[j],strlen(argv[j]));
2540 options = sdscat(options," ");
2541 }
2542 j++;
2543 }
2544 resetServerSaveParams();
2545 loadServerConfig(configfile,options);
2546 sdsfree(options);
2547 } else {
2548 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'");
2549 }
2550 if (server.daemonize) daemonize();
2551 initServer();
2552 if (server.daemonize) createPidFile();
2553 redisAsciiArt();
2554 redisLog(REDIS_WARNING,"Server started, Redis version " REDIS_VERSION);
2555 #ifdef __linux__
2556 linuxOvercommitMemoryWarning();
2557 #endif
2558 start = ustime();
2559 if (server.aof_state == REDIS_AOF_ON) {
2560 if (loadAppendOnlyFile(server.aof_filename) == REDIS_OK)
2561 redisLog(REDIS_NOTICE,"DB loaded from append only file: %.3f seconds",(float)(ustime()-start)/1000000);
2562 } else {
2563 if (rdbLoad(server.rdb_filename) == REDIS_OK) {
2564 redisLog(REDIS_NOTICE,"DB loaded from disk: %.3f seconds",
2565 (float)(ustime()-start)/1000000);
2566 } else if (errno != ENOENT) {
2567 redisLog(REDIS_WARNING,"Fatal error loading the DB. Exiting.");
2568 exit(1);
2569 }
2570 }
2571 if (server.ipfd > 0)
2572 redisLog(REDIS_NOTICE,"The server is now ready to accept connections on port %d", server.port);
2573 if (server.sofd > 0)
2574 redisLog(REDIS_NOTICE,"The server is now ready to accept connections at %s", server.unixsocket);
2575 aeSetBeforeSleepProc(server.el,beforeSleep);
2576 aeMain(server.el);
2577 aeDeleteEventLoop(server.el);
2578 return 0;
2579 }
2580
2581 /* The End */