]> git.saurik.com Git - redis.git/blob - src/redis.h
redisServer structure fields reordered for clarity
[redis.git] / src / redis.h
1 #ifndef __REDIS_H
2 #define __REDIS_H
3
4 #include "fmacros.h"
5 #include "config.h"
6
7 #if defined(__sun)
8 #include "solarisfixes.h"
9 #endif
10
11 #include <stdio.h>
12 #include <stdlib.h>
13 #include <string.h>
14 #include <time.h>
15 #include <limits.h>
16 #include <unistd.h>
17 #include <errno.h>
18 #include <inttypes.h>
19 #include <pthread.h>
20 #include <syslog.h>
21
22 #include "ae.h" /* Event driven programming library */
23 #include "sds.h" /* Dynamic safe strings */
24 #include "dict.h" /* Hash tables */
25 #include "adlist.h" /* Linked lists */
26 #include "zmalloc.h" /* total memory usage aware version of malloc/free */
27 #include "anet.h" /* Networking the easy way */
28 #include "zipmap.h" /* Compact string -> string data structure */
29 #include "ziplist.h" /* Compact list data structure */
30 #include "intset.h" /* Compact integer set structure */
31 #include "version.h"
32
33 /* Error codes */
34 #define REDIS_OK 0
35 #define REDIS_ERR -1
36
37 /* Static server configuration */
38 #define REDIS_SERVERPORT 6379 /* TCP port */
39 #define REDIS_MAXIDLETIME (60*5) /* default client timeout */
40 #define REDIS_IOBUF_LEN 1024
41 #define REDIS_LOADBUF_LEN 1024
42 #define REDIS_STATIC_ARGS 8
43 #define REDIS_DEFAULT_DBNUM 16
44 #define REDIS_CONFIGLINE_MAX 1024
45 #define REDIS_MAX_SYNC_TIME 60 /* Slave can't take more to sync */
46 #define REDIS_EXPIRELOOKUPS_PER_CRON 10 /* lookup 10 expires per loop */
47 #define REDIS_MAX_WRITE_PER_EVENT (1024*64)
48 #define REDIS_REQUEST_MAX_SIZE (1024*1024*256) /* max bytes in inline command */
49 #define REDIS_SHARED_INTEGERS 10000
50 #define REDIS_REPLY_CHUNK_BYTES (5*1500) /* 5 TCP packets with default MTU */
51 #define REDIS_MAX_LOGMSG_LEN 1024 /* Default maximum length of syslog messages */
52
53 /* Hash table parameters */
54 #define REDIS_HT_MINFILL 10 /* Minimal hash table fill 10% */
55
56 /* Command flags:
57 * REDIS_CMD_DENYOOM:
58 * Commands marked with this flag will return an error when 'maxmemory' is
59 * set and the server is using more than 'maxmemory' bytes of memory.
60 * In short: commands with this flag are denied on low memory conditions.
61 * REDIS_CMD_FORCE_REPLICATION:
62 * Force replication even if dirty is 0. */
63 #define REDIS_CMD_DENYOOM 4
64 #define REDIS_CMD_FORCE_REPLICATION 8
65
66 /* Object types */
67 #define REDIS_STRING 0
68 #define REDIS_LIST 1
69 #define REDIS_SET 2
70 #define REDIS_ZSET 3
71 #define REDIS_HASH 4
72 #define REDIS_VMPOINTER 8
73
74 /* Objects encoding. Some kind of objects like Strings and Hashes can be
75 * internally represented in multiple ways. The 'encoding' field of the object
76 * is set to one of this fields for this object. */
77 #define REDIS_ENCODING_RAW 0 /* Raw representation */
78 #define REDIS_ENCODING_INT 1 /* Encoded as integer */
79 #define REDIS_ENCODING_HT 2 /* Encoded as hash table */
80 #define REDIS_ENCODING_ZIPMAP 3 /* Encoded as zipmap */
81 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
82 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
83 #define REDIS_ENCODING_INTSET 6 /* Encoded as intset */
84
85 /* Object types only used for dumping to disk */
86 #define REDIS_EXPIRETIME 253
87 #define REDIS_SELECTDB 254
88 #define REDIS_EOF 255
89
90 /* Defines related to the dump file format. To store 32 bits lengths for short
91 * keys requires a lot of space, so we check the most significant 2 bits of
92 * the first byte to interpreter the length:
93 *
94 * 00|000000 => if the two MSB are 00 the len is the 6 bits of this byte
95 * 01|000000 00000000 => 01, the len is 14 byes, 6 bits + 8 bits of next byte
96 * 10|000000 [32 bit integer] => if it's 01, a full 32 bit len will follow
97 * 11|000000 this means: specially encoded object will follow. The six bits
98 * number specify the kind of object that follows.
99 * See the REDIS_RDB_ENC_* defines.
100 *
101 * Lenghts up to 63 are stored using a single byte, most DB keys, and may
102 * values, will fit inside. */
103 #define REDIS_RDB_6BITLEN 0
104 #define REDIS_RDB_14BITLEN 1
105 #define REDIS_RDB_32BITLEN 2
106 #define REDIS_RDB_ENCVAL 3
107 #define REDIS_RDB_LENERR UINT_MAX
108
109 /* When a length of a string object stored on disk has the first two bits
110 * set, the remaining two bits specify a special encoding for the object
111 * accordingly to the following defines: */
112 #define REDIS_RDB_ENC_INT8 0 /* 8 bit signed integer */
113 #define REDIS_RDB_ENC_INT16 1 /* 16 bit signed integer */
114 #define REDIS_RDB_ENC_INT32 2 /* 32 bit signed integer */
115 #define REDIS_RDB_ENC_LZF 3 /* string compressed with FASTLZ */
116
117 /* Scheduled IO opeations flags. */
118 #define REDIS_IO_LOAD 1
119 #define REDIS_IO_SAVE 2
120 #define REDIS_IO_LOADINPROG 4
121 #define REDIS_IO_SAVEINPROG 8
122
123 /* Generic IO flags */
124 #define REDIS_IO_ONLYLOADS 1
125 #define REDIS_IO_ASAP 2
126
127 #define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1
128 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
129
130 /* Client flags */
131 #define REDIS_SLAVE 1 /* This client is a slave server */
132 #define REDIS_MASTER 2 /* This client is a master server */
133 #define REDIS_MONITOR 4 /* This client is a slave monitor, see MONITOR */
134 #define REDIS_MULTI 8 /* This client is in a MULTI context */
135 #define REDIS_BLOCKED 16 /* The client is waiting in a blocking operation */
136 #define REDIS_IO_WAIT 32 /* The client is waiting for Virtual Memory I/O */
137 #define REDIS_DIRTY_CAS 64 /* Watched keys modified. EXEC will fail. */
138 #define REDIS_CLOSE_AFTER_REPLY 128 /* Close after writing entire reply. */
139
140 /* Client request types */
141 #define REDIS_REQ_INLINE 1
142 #define REDIS_REQ_MULTIBULK 2
143
144 /* Slave replication state - slave side */
145 #define REDIS_REPL_NONE 0 /* No active replication */
146 #define REDIS_REPL_CONNECT 1 /* Must connect to master */
147 #define REDIS_REPL_TRANSFER 2 /* Receiving .rdb from master */
148 #define REDIS_REPL_CONNECTED 3 /* Connected to master */
149
150 /* Slave replication state - from the point of view of master
151 * Note that in SEND_BULK and ONLINE state the slave receives new updates
152 * in its output queue. In the WAIT_BGSAVE state instead the server is waiting
153 * to start the next background saving in order to send updates to it. */
154 #define REDIS_REPL_WAIT_BGSAVE_START 3 /* master waits bgsave to start feeding it */
155 #define REDIS_REPL_WAIT_BGSAVE_END 4 /* master waits bgsave to start bulk DB transmission */
156 #define REDIS_REPL_SEND_BULK 5 /* master is sending the bulk DB */
157 #define REDIS_REPL_ONLINE 6 /* bulk DB already transmitted, receive updates */
158
159 /* List related stuff */
160 #define REDIS_HEAD 0
161 #define REDIS_TAIL 1
162
163 /* Sort operations */
164 #define REDIS_SORT_GET 0
165 #define REDIS_SORT_ASC 1
166 #define REDIS_SORT_DESC 2
167 #define REDIS_SORTKEY_MAX 1024
168
169 /* Log levels */
170 #define REDIS_DEBUG 0
171 #define REDIS_VERBOSE 1
172 #define REDIS_NOTICE 2
173 #define REDIS_WARNING 3
174
175 /* Anti-warning macro... */
176 #define REDIS_NOTUSED(V) ((void) V)
177
178 #define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^32 elements */
179 #define ZSKIPLIST_P 0.25 /* Skiplist P = 1/4 */
180
181 /* Append only defines */
182 #define APPENDFSYNC_NO 0
183 #define APPENDFSYNC_ALWAYS 1
184 #define APPENDFSYNC_EVERYSEC 2
185
186 /* Zip structure related defaults */
187 #define REDIS_HASH_MAX_ZIPMAP_ENTRIES 64
188 #define REDIS_HASH_MAX_ZIPMAP_VALUE 512
189 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 512
190 #define REDIS_LIST_MAX_ZIPLIST_VALUE 64
191 #define REDIS_SET_MAX_INTSET_ENTRIES 512
192
193 /* Sets operations codes */
194 #define REDIS_OP_UNION 0
195 #define REDIS_OP_DIFF 1
196 #define REDIS_OP_INTER 2
197
198 /* Redis maxmemory strategies */
199 #define REDIS_MAXMEMORY_VOLATILE_LRU 0
200 #define REDIS_MAXMEMORY_VOLATILE_TTL 1
201 #define REDIS_MAXMEMORY_VOLATILE_RANDOM 2
202 #define REDIS_MAXMEMORY_ALLKEYS_LRU 3
203 #define REDIS_MAXMEMORY_ALLKEYS_RANDOM 4
204 #define REDIS_MAXMEMORY_NO_EVICTION 5
205
206 /* Diskstore background saving thread states */
207 #define REDIS_BGSAVE_THREAD_UNACTIVE 0
208 #define REDIS_BGSAVE_THREAD_ACTIVE 1
209 #define REDIS_BGSAVE_THREAD_DONE_OK 2
210 #define REDIS_BGSAVE_THREAD_DONE_ERR 3
211
212 /* We can print the stacktrace, so our assert is defined this way: */
213 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
214 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
215 void _redisAssert(char *estr, char *file, int line);
216 void _redisPanic(char *msg, char *file, int line);
217
218 /*-----------------------------------------------------------------------------
219 * Data types
220 *----------------------------------------------------------------------------*/
221
222 /* A redis object, that is a type able to hold a string / list / set */
223
224 /* The actual Redis Object */
225 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
226 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
227 typedef struct redisObject {
228 unsigned type:4;
229 unsigned notused:2; /* Not used */
230 unsigned encoding:4;
231 unsigned lru:22; /* lru time (relative to server.lruclock) */
232 int refcount;
233 void *ptr;
234 /* VM fields are only allocated if VM is active, otherwise the
235 * object allocation function will just allocate
236 * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
237 * Redis without VM active will not have any overhead. */
238 } robj;
239
240 /* The VM pointer structure - identifies an object in the swap file.
241 *
242 * This object is stored in place of the value
243 * object in the main key->value hash table representing a database.
244 * Note that the first fields (type, storage) are the same as the redisObject
245 * structure so that vmPointer strucuters can be accessed even when casted
246 * as redisObject structures.
247 *
248 * This is useful as we don't know if a value object is or not on disk, but we
249 * are always able to read obj->storage to check this. For vmPointer
250 * structures "type" is set to REDIS_VMPOINTER (even if without this field
251 * is still possible to check the kind of object from the value of 'storage').*/
252 typedef struct vmPointer {
253 unsigned type:4;
254 unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
255 unsigned notused:26;
256 unsigned int vtype; /* type of the object stored in the swap file */
257 off_t page; /* the page at witch the object is stored on disk */
258 off_t usedpages; /* number of pages used on disk */
259 } vmpointer;
260
261 /* Macro used to initalize a Redis object allocated on the stack.
262 * Note that this macro is taken near the structure definition to make sure
263 * we'll update it when the structure is changed, to avoid bugs like
264 * bug #85 introduced exactly in this way. */
265 #define initStaticStringObject(_var,_ptr) do { \
266 _var.refcount = 1; \
267 _var.type = REDIS_STRING; \
268 _var.encoding = REDIS_ENCODING_RAW; \
269 _var.ptr = _ptr; \
270 } while(0);
271
272 typedef struct redisDb {
273 dict *dict; /* The keyspace for this DB */
274 dict *expires; /* Timeout of keys with a timeout set */
275 dict *blocking_keys; /* Keys with clients waiting for data (BLPOP) */
276 dict *io_keys; /* Keys with clients waiting for DS I/O */
277 dict *io_negcache; /* Negative caching for disk store */
278 dict *io_queued; /* Queued IO operations hash table */
279 dict *watched_keys; /* WATCHED keys for MULTI/EXEC CAS */
280 int id;
281 } redisDb;
282
283 /* Client MULTI/EXEC state */
284 typedef struct multiCmd {
285 robj **argv;
286 int argc;
287 struct redisCommand *cmd;
288 } multiCmd;
289
290 typedef struct multiState {
291 multiCmd *commands; /* Array of MULTI commands */
292 int count; /* Total number of MULTI commands */
293 } multiState;
294
295 typedef struct blockingState {
296 robj **keys; /* The key we are waiting to terminate a blocking
297 * operation such as BLPOP. Otherwise NULL. */
298 int count; /* Number of blocking keys */
299 time_t timeout; /* Blocking operation timeout. If UNIX current time
300 * is >= timeout then the operation timed out. */
301 robj *target; /* The key that should receive the element,
302 * for BRPOPLPUSH. */
303 } blockingState;
304
305 /* With multiplexing we need to take per-clinet state.
306 * Clients are taken in a liked list. */
307 typedef struct redisClient {
308 int fd;
309 redisDb *db;
310 int dictid;
311 sds querybuf;
312 int argc;
313 robj **argv;
314 int reqtype;
315 int multibulklen; /* number of multi bulk arguments left to read */
316 long bulklen; /* length of bulk argument in multi bulk request */
317 list *reply;
318 int sentlen;
319 time_t lastinteraction; /* time of the last interaction, used for timeout */
320 int flags; /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
321 int slaveseldb; /* slave selected db, if this client is a slave */
322 int authenticated; /* when requirepass is non-NULL */
323 int replstate; /* replication state if this is a slave */
324 int repldbfd; /* replication DB file descriptor */
325 long repldboff; /* replication DB file offset */
326 off_t repldbsize; /* replication DB file size */
327 multiState mstate; /* MULTI/EXEC state */
328 blockingState bpop; /* blocking state */
329 list *io_keys; /* Keys this client is waiting to be loaded from the
330 * swap file in order to continue. */
331 list *watched_keys; /* Keys WATCHED for MULTI/EXEC CAS */
332 dict *pubsub_channels; /* channels a client is interested in (SUBSCRIBE) */
333 list *pubsub_patterns; /* patterns a client is interested in (SUBSCRIBE) */
334
335 /* Response buffer */
336 int bufpos;
337 char buf[REDIS_REPLY_CHUNK_BYTES];
338 } redisClient;
339
340 struct saveparam {
341 time_t seconds;
342 int changes;
343 };
344
345 struct sharedObjectsStruct {
346 robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
347 *colon, *nullbulk, *nullmultibulk, *queued,
348 *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
349 *outofrangeerr, *loadingerr, *plus,
350 *select0, *select1, *select2, *select3, *select4,
351 *select5, *select6, *select7, *select8, *select9,
352 *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
353 *mbulk4, *psubscribebulk, *punsubscribebulk,
354 *integers[REDIS_SHARED_INTEGERS];
355 };
356
357 /* Global server state structure */
358 struct redisServer {
359 /* General */
360 pthread_t mainthread;
361 redisDb *db;
362 dict *commands; /* Command table hahs table */
363 aeEventLoop *el;
364 /* Networking */
365 int port;
366 char *bindaddr;
367 char *unixsocket;
368 int ipfd;
369 int sofd;
370 list *clients;
371 list *slaves, *monitors;
372 char neterr[ANET_ERR_LEN];
373 /* RDB / AOF loading information */
374 int loading;
375 off_t loading_total_bytes;
376 off_t loading_loaded_bytes;
377 time_t loading_start_time;
378 /* Fast pointers to often looked up command */
379 struct redisCommand *delCommand, *multiCommand;
380 int cronloops; /* number of times the cron function run */
381 time_t lastsave; /* Unix time of last save succeeede */
382 /* Fields used only for stats */
383 time_t stat_starttime; /* server start time */
384 long long stat_numcommands; /* number of processed commands */
385 long long stat_numconnections; /* number of connections received */
386 long long stat_expiredkeys; /* number of expired keys */
387 long long stat_evictedkeys; /* number of evicted keys (maxmemory) */
388 long long stat_keyspace_hits; /* number of successful lookups of keys */
389 long long stat_keyspace_misses; /* number of failed lookups of keys */
390 /* Configuration */
391 int verbosity;
392 int maxidletime;
393 int dbnum;
394 int daemonize;
395 int appendonly;
396 int appendfsync;
397 int no_appendfsync_on_rewrite;
398 int shutdown_asap;
399 int activerehashing;
400 char *requirepass;
401 /* Persistence */
402 long long dirty; /* changes to DB from the last save */
403 long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
404 time_t lastfsync;
405 int appendfd;
406 int appendseldb;
407 char *pidfile;
408 pid_t bgsavechildpid;
409 pid_t bgrewritechildpid;
410 int bgsavethread_state;
411 pthread_mutex_t bgsavethread_mutex;
412 pthread_t bgsavethread;
413 sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
414 sds aofbuf; /* AOF buffer, written before entering the event loop */
415 struct saveparam *saveparams;
416 int saveparamslen;
417 char *dbfilename;
418 int rdbcompression;
419 char *appendfilename;
420 /* Logging */
421 char *logfile;
422 int syslog_enabled;
423 char *syslog_ident;
424 int syslog_facility;
425 /* Replication related */
426 int isslave;
427 /* Slave specific fields */
428 char *masterauth;
429 char *masterhost;
430 int masterport;
431 redisClient *master; /* client that is master for this slave */
432 int replstate; /* replication status if the instance is a slave */
433 off_t repl_transfer_left; /* bytes left reading .rdb */
434 int repl_transfer_s; /* slave -> master SYNC socket */
435 int repl_transfer_fd; /* slave -> master SYNC temp file descriptor */
436 char *repl_transfer_tmpfile; /* slave-> master SYNC temp file name */
437 time_t repl_transfer_lastio; /* unix time of the latest read, for timeout */
438 int repl_serve_stale_data; /* Serve stale data when link is down? */
439 /* Limits */
440 unsigned int maxclients;
441 unsigned long long maxmemory;
442 int maxmemory_policy;
443 int maxmemory_samples;
444 /* Blocked clients */
445 unsigned int bpop_blocked_clients;
446 unsigned int cache_blocked_clients;
447 list *unblocked_clients; /* list of clients to unblock before next loop */
448 list *cache_io_queue; /* IO operations queue */
449 int cache_flush_delay; /* seconds to wait before flushing keys */
450 /* Sort parameters - qsort_r() is only available under BSD so we
451 * have to take this state global, in order to pass it to sortCompare() */
452 int sort_desc;
453 int sort_alpha;
454 int sort_bypattern;
455 /* Virtual memory configuration */
456 int ds_enabled; /* backend disk in redis.conf */
457 char *ds_path; /* location of the disk store on disk */
458 unsigned long long cache_max_memory;
459 /* Zip structure config */
460 size_t hash_max_zipmap_entries;
461 size_t hash_max_zipmap_value;
462 size_t list_max_ziplist_entries;
463 size_t list_max_ziplist_value;
464 size_t set_max_intset_entries;
465 time_t unixtime; /* Unix time sampled every second. */
466 /* Virtual memory I/O threads stuff */
467 /* An I/O thread process an element taken from the io_jobs queue and
468 * put the result of the operation in the io_done list. While the
469 * job is being processed, it's put on io_processing queue. */
470 list *io_newjobs; /* List of VM I/O jobs yet to be processed */
471 list *io_processing; /* List of VM I/O jobs being processed */
472 list *io_processed; /* List of VM I/O jobs already processed */
473 list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
474 pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
475 pthread_cond_t io_condvar; /* I/O threads conditional variable */
476 pthread_attr_t io_threads_attr; /* attributes for threads creation */
477 int io_active_threads; /* Number of running I/O threads */
478 int vm_max_threads; /* Max number of I/O threads running at the same time */
479 /* Our main thread is blocked on the event loop, locking for sockets ready
480 * to be read or written, so when a threaded I/O operation is ready to be
481 * processed by the main thread, the I/O thread will use a unix pipe to
482 * awake the main thread. The followings are the two pipe FDs. */
483 int io_ready_pipe_read;
484 int io_ready_pipe_write;
485 /* Virtual memory stats */
486 unsigned long long vm_stats_used_pages;
487 unsigned long long vm_stats_swapped_objects;
488 unsigned long long vm_stats_swapouts;
489 unsigned long long vm_stats_swapins;
490 /* Pubsub */
491 dict *pubsub_channels; /* Map channels to list of subscribed clients */
492 list *pubsub_patterns; /* A list of pubsub_patterns */
493 /* Misc */
494 unsigned lruclock:22; /* clock incrementing every minute, for LRU */
495 unsigned lruclock_padding:10;
496 };
497
498 typedef struct pubsubPattern {
499 redisClient *client;
500 robj *pattern;
501 } pubsubPattern;
502
503 typedef void redisCommandProc(redisClient *c);
504 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
505 struct redisCommand {
506 char *name;
507 redisCommandProc *proc;
508 int arity;
509 int flags;
510 /* Use a function to determine which keys need to be loaded
511 * in the background prior to executing this command. Takes precedence
512 * over vm_firstkey and others, ignored when NULL */
513 redisVmPreloadProc *vm_preload_proc;
514 /* What keys should be loaded in background when calling this command? */
515 int vm_firstkey; /* The first argument that's a key (0 = no keys) */
516 int vm_lastkey; /* THe last argument that's a key */
517 int vm_keystep; /* The step between first and last key */
518 };
519
520 struct redisFunctionSym {
521 char *name;
522 unsigned long pointer;
523 };
524
525 typedef struct _redisSortObject {
526 robj *obj;
527 union {
528 double score;
529 robj *cmpobj;
530 } u;
531 } redisSortObject;
532
533 typedef struct _redisSortOperation {
534 int type;
535 robj *pattern;
536 } redisSortOperation;
537
538 /* ZSETs use a specialized version of Skiplists */
539 typedef struct zskiplistNode {
540 robj *obj;
541 double score;
542 struct zskiplistNode *backward;
543 struct zskiplistLevel {
544 struct zskiplistNode *forward;
545 unsigned int span;
546 } level[];
547 } zskiplistNode;
548
549 typedef struct zskiplist {
550 struct zskiplistNode *header, *tail;
551 unsigned long length;
552 int level;
553 } zskiplist;
554
555 typedef struct zset {
556 dict *dict;
557 zskiplist *zsl;
558 } zset;
559
560 /* DIsk store threaded I/O request message */
561 #define REDIS_IOJOB_LOAD 0
562 #define REDIS_IOJOB_SAVE 1
563
564 typedef struct iojob {
565 int type; /* Request type, REDIS_IOJOB_* */
566 redisDb *db;/* Redis database */
567 robj *key; /* This I/O request is about this key */
568 robj *val; /* the value to swap for REDIS_IOJOB_SAVE, otherwise this
569 * field is populated by the I/O thread for REDIS_IOJOB_LOAD. */
570 time_t expire; /* Expire time for this key on REDIS_IOJOB_LOAD */
571 } iojob;
572
573 /* IO operations scheduled -- check dscache.c for more info */
574 typedef struct ioop {
575 int type;
576 redisDb *db;
577 robj *key;
578 time_t ctime; /* This is the creation time of the entry. */
579 } ioop;
580
581 /* Structure to hold list iteration abstraction. */
582 typedef struct {
583 robj *subject;
584 unsigned char encoding;
585 unsigned char direction; /* Iteration direction */
586 unsigned char *zi;
587 listNode *ln;
588 } listTypeIterator;
589
590 /* Structure for an entry while iterating over a list. */
591 typedef struct {
592 listTypeIterator *li;
593 unsigned char *zi; /* Entry in ziplist */
594 listNode *ln; /* Entry in linked list */
595 } listTypeEntry;
596
597 /* Structure to hold set iteration abstraction. */
598 typedef struct {
599 robj *subject;
600 int encoding;
601 int ii; /* intset iterator */
602 dictIterator *di;
603 } setTypeIterator;
604
605 /* Structure to hold hash iteration abstration. Note that iteration over
606 * hashes involves both fields and values. Because it is possible that
607 * not both are required, store pointers in the iterator to avoid
608 * unnecessary memory allocation for fields/values. */
609 typedef struct {
610 int encoding;
611 unsigned char *zi;
612 unsigned char *zk, *zv;
613 unsigned int zklen, zvlen;
614
615 dictIterator *di;
616 dictEntry *de;
617 } hashTypeIterator;
618
619 #define REDIS_HASH_KEY 1
620 #define REDIS_HASH_VALUE 2
621
622 /*-----------------------------------------------------------------------------
623 * Extern declarations
624 *----------------------------------------------------------------------------*/
625
626 extern struct redisServer server;
627 extern struct sharedObjectsStruct shared;
628 extern dictType setDictType;
629 extern dictType zsetDictType;
630 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
631 dictType hashDictType;
632
633 /*-----------------------------------------------------------------------------
634 * Functions prototypes
635 *----------------------------------------------------------------------------*/
636
637 /* networking.c -- Networking and Client related operations */
638 redisClient *createClient(int fd);
639 void closeTimedoutClients(void);
640 void freeClient(redisClient *c);
641 void resetClient(redisClient *c);
642 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
643 void addReply(redisClient *c, robj *obj);
644 void *addDeferredMultiBulkLength(redisClient *c);
645 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
646 void addReplySds(redisClient *c, sds s);
647 void processInputBuffer(redisClient *c);
648 void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask);
649 void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask);
650 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
651 void addReplyBulk(redisClient *c, robj *obj);
652 void addReplyBulkCString(redisClient *c, char *s);
653 void addReplyBulkCBuffer(redisClient *c, void *p, size_t len);
654 void addReplyBulkLongLong(redisClient *c, long long ll);
655 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
656 void addReply(redisClient *c, robj *obj);
657 void addReplySds(redisClient *c, sds s);
658 void addReplyError(redisClient *c, char *err);
659 void addReplyStatus(redisClient *c, char *status);
660 void addReplyDouble(redisClient *c, double d);
661 void addReplyLongLong(redisClient *c, long long ll);
662 void addReplyMultiBulkLen(redisClient *c, long length);
663 void *dupClientReplyValue(void *o);
664
665 #ifdef __GNUC__
666 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
667 __attribute__((format(printf, 2, 3)));
668 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
669 __attribute__((format(printf, 2, 3)));
670 #else
671 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
672 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
673 #endif
674
675 /* List data type */
676 void listTypeTryConversion(robj *subject, robj *value);
677 void listTypePush(robj *subject, robj *value, int where);
678 robj *listTypePop(robj *subject, int where);
679 unsigned long listTypeLength(robj *subject);
680 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
681 void listTypeReleaseIterator(listTypeIterator *li);
682 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
683 robj *listTypeGet(listTypeEntry *entry);
684 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
685 int listTypeEqual(listTypeEntry *entry, robj *o);
686 void listTypeDelete(listTypeEntry *entry);
687 void listTypeConvert(robj *subject, int enc);
688 void unblockClientWaitingData(redisClient *c);
689 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
690 void popGenericCommand(redisClient *c, int where);
691
692 /* MULTI/EXEC/WATCH... */
693 void unwatchAllKeys(redisClient *c);
694 void initClientMultiState(redisClient *c);
695 void freeClientMultiState(redisClient *c);
696 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
697 void touchWatchedKey(redisDb *db, robj *key);
698 void touchWatchedKeysOnFlush(int dbid);
699
700 /* Redis object implementation */
701 void decrRefCount(void *o);
702 void incrRefCount(robj *o);
703 void freeStringObject(robj *o);
704 void freeListObject(robj *o);
705 void freeSetObject(robj *o);
706 void freeZsetObject(robj *o);
707 void freeHashObject(robj *o);
708 robj *createObject(int type, void *ptr);
709 robj *createStringObject(char *ptr, size_t len);
710 robj *dupStringObject(robj *o);
711 robj *tryObjectEncoding(robj *o);
712 robj *getDecodedObject(robj *o);
713 size_t stringObjectLen(robj *o);
714 robj *createStringObjectFromLongLong(long long value);
715 robj *createListObject(void);
716 robj *createZiplistObject(void);
717 robj *createSetObject(void);
718 robj *createIntsetObject(void);
719 robj *createHashObject(void);
720 robj *createZsetObject(void);
721 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
722 int checkType(redisClient *c, robj *o, int type);
723 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
724 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
725 int getLongLongFromObject(robj *o, long long *target);
726 char *strEncoding(int encoding);
727 int compareStringObjects(robj *a, robj *b);
728 int equalStringObjects(robj *a, robj *b);
729 unsigned long estimateObjectIdleTime(robj *o);
730
731 /* Synchronous I/O with timeout */
732 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
733 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
734 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
735 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
736 int fwriteBulkDouble(FILE *fp, double d);
737 int fwriteBulkLongLong(FILE *fp, long long l);
738 int fwriteBulkObject(FILE *fp, robj *obj);
739
740 /* Replication */
741 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
742 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
743 int syncWithMaster(void);
744 void updateSlavesWaitingBgsave(int bgsaveerr);
745 void replicationCron(void);
746
747 /* Generic persistence functions */
748 void startLoading(FILE *fp);
749 void loadingProgress(off_t pos);
750 void stopLoading(void);
751
752 /* RDB persistence */
753 int rdbLoad(char *filename);
754 int rdbSaveBackground(char *filename);
755 void rdbRemoveTempFile(pid_t childpid);
756 int rdbSave(char *filename);
757 int rdbSaveObject(FILE *fp, robj *o);
758 off_t rdbSavedObjectLen(robj *o);
759 off_t rdbSavedObjectPages(robj *o);
760 robj *rdbLoadObject(int type, FILE *fp);
761 void backgroundSaveDoneHandler(int exitcode, int bysignal);
762 int rdbSaveKeyValuePair(FILE *fp, redisDb *db, robj *key, robj *val, time_t now);
763 int rdbLoadType(FILE *fp);
764 time_t rdbLoadTime(FILE *fp);
765 robj *rdbLoadStringObject(FILE *fp);
766 int rdbSaveType(FILE *fp, unsigned char type);
767 int rdbSaveLen(FILE *fp, uint32_t len);
768
769 /* AOF persistence */
770 void flushAppendOnlyFile(void);
771 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
772 void aofRemoveTempFile(pid_t childpid);
773 int rewriteAppendOnlyFileBackground(void);
774 int loadAppendOnlyFile(char *filename);
775 void stopAppendOnly(void);
776 int startAppendOnly(void);
777 void backgroundRewriteDoneHandler(int exitcode, int bysignal);
778
779 /* Sorted sets data type */
780 zskiplist *zslCreate(void);
781 void zslFree(zskiplist *zsl);
782 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
783
784 /* Core functions */
785 void freeMemoryIfNeeded(void);
786 int processCommand(redisClient *c);
787 void setupSigSegvAction(void);
788 struct redisCommand *lookupCommand(sds name);
789 struct redisCommand *lookupCommandByCString(char *s);
790 void call(redisClient *c, struct redisCommand *cmd);
791 int prepareForShutdown();
792 void redisLog(int level, const char *fmt, ...);
793 void usage();
794 void updateDictResizePolicy(void);
795 int htNeedsResize(dict *dict);
796 void oom(const char *msg);
797 void populateCommandTable(void);
798
799 /* Disk store */
800 int dsOpen(void);
801 int dsClose(void);
802 int dsSet(redisDb *db, robj *key, robj *val);
803 robj *dsGet(redisDb *db, robj *key, time_t *expire);
804 int dsDel(redisDb *db, robj *key);
805 int dsExists(redisDb *db, robj *key);
806 void dsFlushDb(int dbid);
807 int dsRdbSaveBackground(char *filename);
808 int dsRdbSave(char *filename);
809
810 /* Disk Store Cache */
811 void dsInit(void);
812 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
813 void lockThreadedIO(void);
814 void unlockThreadedIO(void);
815 void freeIOJob(iojob *j);
816 void queueIOJob(iojob *j);
817 void waitEmptyIOJobsQueue(void);
818 void processAllPendingIOJobs(void);
819 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
820 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
821 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
822 int dontWaitForSwappedKey(redisClient *c, robj *key);
823 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
824 int cacheFreeOneEntry(void);
825 void cacheScheduleIOAddFlag(redisDb *db, robj *key, long flag);
826 void cacheScheduleIODelFlag(redisDb *db, robj *key, long flag);
827 int cacheScheduleIOGetFlags(redisDb *db, robj *key);
828 void cacheScheduleIO(redisDb *db, robj *key, int type);
829 void cacheCron(void);
830 int cacheKeyMayExist(redisDb *db, robj *key);
831 void cacheSetKeyMayExist(redisDb *db, robj *key);
832 void cacheSetKeyDoesNotExist(redisDb *db, robj *key);
833 void cacheForcePointInTime(void);
834
835 /* Set data type */
836 robj *setTypeCreate(robj *value);
837 int setTypeAdd(robj *subject, robj *value);
838 int setTypeRemove(robj *subject, robj *value);
839 int setTypeIsMember(robj *subject, robj *value);
840 setTypeIterator *setTypeInitIterator(robj *subject);
841 void setTypeReleaseIterator(setTypeIterator *si);
842 int setTypeNext(setTypeIterator *si, robj **objele, int64_t *llele);
843 robj *setTypeNextObject(setTypeIterator *si);
844 int setTypeRandomElement(robj *setobj, robj **objele, int64_t *llele);
845 unsigned long setTypeSize(robj *subject);
846 void setTypeConvert(robj *subject, int enc);
847
848 /* Hash data type */
849 void convertToRealHash(robj *o);
850 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
851 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
852 int hashTypeGet(robj *o, robj *key, robj **objval, unsigned char **v, unsigned int *vlen);
853 robj *hashTypeGetObject(robj *o, robj *key);
854 int hashTypeExists(robj *o, robj *key);
855 int hashTypeSet(robj *o, robj *key, robj *value);
856 int hashTypeDelete(robj *o, robj *key);
857 unsigned long hashTypeLength(robj *o);
858 hashTypeIterator *hashTypeInitIterator(robj *subject);
859 void hashTypeReleaseIterator(hashTypeIterator *hi);
860 int hashTypeNext(hashTypeIterator *hi);
861 int hashTypeCurrent(hashTypeIterator *hi, int what, robj **objval, unsigned char **v, unsigned int *vlen);
862 robj *hashTypeCurrentObject(hashTypeIterator *hi, int what);
863 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
864
865 /* Pub / Sub */
866 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
867 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
868 void freePubsubPattern(void *p);
869 int listMatchPubsubPattern(void *a, void *b);
870
871 /* Utility functions */
872 int stringmatchlen(const char *pattern, int patternLen,
873 const char *string, int stringLen, int nocase);
874 int stringmatch(const char *pattern, const char *string, int nocase);
875 long long memtoll(const char *p, int *err);
876 int ll2string(char *s, size_t len, long long value);
877 int isStringRepresentableAsLong(sds s, long *longval);
878 int isStringRepresentableAsLongLong(sds s, long long *longval);
879 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
880
881 /* Configuration */
882 void loadServerConfig(char *filename);
883 void appendServerSaveParams(time_t seconds, int changes);
884 void resetServerSaveParams();
885
886 /* db.c -- Keyspace access API */
887 int removeExpire(redisDb *db, robj *key);
888 void propagateExpire(redisDb *db, robj *key);
889 int expireIfNeeded(redisDb *db, robj *key);
890 time_t getExpire(redisDb *db, robj *key);
891 void setExpire(redisDb *db, robj *key, time_t when);
892 robj *lookupKey(redisDb *db, robj *key);
893 robj *lookupKeyRead(redisDb *db, robj *key);
894 robj *lookupKeyWrite(redisDb *db, robj *key);
895 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
896 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
897 int dbAdd(redisDb *db, robj *key, robj *val);
898 int dbReplace(redisDb *db, robj *key, robj *val);
899 int dbExists(redisDb *db, robj *key);
900 robj *dbRandomKey(redisDb *db);
901 int dbDelete(redisDb *db, robj *key);
902 long long emptyDb();
903 int selectDb(redisClient *c, int id);
904 void signalModifiedKey(redisDb *db, robj *key);
905 void signalFlushedDb(int dbid);
906
907 /* Git SHA1 */
908 char *redisGitSHA1(void);
909 char *redisGitDirty(void);
910
911 /* Commands prototypes */
912 void authCommand(redisClient *c);
913 void pingCommand(redisClient *c);
914 void echoCommand(redisClient *c);
915 void setCommand(redisClient *c);
916 void setnxCommand(redisClient *c);
917 void setexCommand(redisClient *c);
918 void getCommand(redisClient *c);
919 void delCommand(redisClient *c);
920 void existsCommand(redisClient *c);
921 void setbitCommand(redisClient *c);
922 void getbitCommand(redisClient *c);
923 void setrangeCommand(redisClient *c);
924 void getrangeCommand(redisClient *c);
925 void incrCommand(redisClient *c);
926 void decrCommand(redisClient *c);
927 void incrbyCommand(redisClient *c);
928 void decrbyCommand(redisClient *c);
929 void selectCommand(redisClient *c);
930 void randomkeyCommand(redisClient *c);
931 void keysCommand(redisClient *c);
932 void dbsizeCommand(redisClient *c);
933 void lastsaveCommand(redisClient *c);
934 void saveCommand(redisClient *c);
935 void bgsaveCommand(redisClient *c);
936 void bgrewriteaofCommand(redisClient *c);
937 void shutdownCommand(redisClient *c);
938 void moveCommand(redisClient *c);
939 void renameCommand(redisClient *c);
940 void renamenxCommand(redisClient *c);
941 void lpushCommand(redisClient *c);
942 void rpushCommand(redisClient *c);
943 void lpushxCommand(redisClient *c);
944 void rpushxCommand(redisClient *c);
945 void linsertCommand(redisClient *c);
946 void lpopCommand(redisClient *c);
947 void rpopCommand(redisClient *c);
948 void llenCommand(redisClient *c);
949 void lindexCommand(redisClient *c);
950 void lrangeCommand(redisClient *c);
951 void ltrimCommand(redisClient *c);
952 void typeCommand(redisClient *c);
953 void lsetCommand(redisClient *c);
954 void saddCommand(redisClient *c);
955 void sremCommand(redisClient *c);
956 void smoveCommand(redisClient *c);
957 void sismemberCommand(redisClient *c);
958 void scardCommand(redisClient *c);
959 void spopCommand(redisClient *c);
960 void srandmemberCommand(redisClient *c);
961 void sinterCommand(redisClient *c);
962 void sinterstoreCommand(redisClient *c);
963 void sunionCommand(redisClient *c);
964 void sunionstoreCommand(redisClient *c);
965 void sdiffCommand(redisClient *c);
966 void sdiffstoreCommand(redisClient *c);
967 void syncCommand(redisClient *c);
968 void flushdbCommand(redisClient *c);
969 void flushallCommand(redisClient *c);
970 void sortCommand(redisClient *c);
971 void lremCommand(redisClient *c);
972 void rpoplpushCommand(redisClient *c);
973 void infoCommand(redisClient *c);
974 void mgetCommand(redisClient *c);
975 void monitorCommand(redisClient *c);
976 void expireCommand(redisClient *c);
977 void expireatCommand(redisClient *c);
978 void getsetCommand(redisClient *c);
979 void ttlCommand(redisClient *c);
980 void persistCommand(redisClient *c);
981 void slaveofCommand(redisClient *c);
982 void debugCommand(redisClient *c);
983 void msetCommand(redisClient *c);
984 void msetnxCommand(redisClient *c);
985 void zaddCommand(redisClient *c);
986 void zincrbyCommand(redisClient *c);
987 void zrangeCommand(redisClient *c);
988 void zrangebyscoreCommand(redisClient *c);
989 void zrevrangebyscoreCommand(redisClient *c);
990 void zcountCommand(redisClient *c);
991 void zrevrangeCommand(redisClient *c);
992 void zcardCommand(redisClient *c);
993 void zremCommand(redisClient *c);
994 void zscoreCommand(redisClient *c);
995 void zremrangebyscoreCommand(redisClient *c);
996 void multiCommand(redisClient *c);
997 void execCommand(redisClient *c);
998 void discardCommand(redisClient *c);
999 void blpopCommand(redisClient *c);
1000 void brpopCommand(redisClient *c);
1001 void brpoplpushCommand(redisClient *c);
1002 void appendCommand(redisClient *c);
1003 void strlenCommand(redisClient *c);
1004 void zrankCommand(redisClient *c);
1005 void zrevrankCommand(redisClient *c);
1006 void hsetCommand(redisClient *c);
1007 void hsetnxCommand(redisClient *c);
1008 void hgetCommand(redisClient *c);
1009 void hmsetCommand(redisClient *c);
1010 void hmgetCommand(redisClient *c);
1011 void hdelCommand(redisClient *c);
1012 void hlenCommand(redisClient *c);
1013 void zremrangebyrankCommand(redisClient *c);
1014 void zunionstoreCommand(redisClient *c);
1015 void zinterstoreCommand(redisClient *c);
1016 void hkeysCommand(redisClient *c);
1017 void hvalsCommand(redisClient *c);
1018 void hgetallCommand(redisClient *c);
1019 void hexistsCommand(redisClient *c);
1020 void configCommand(redisClient *c);
1021 void hincrbyCommand(redisClient *c);
1022 void subscribeCommand(redisClient *c);
1023 void unsubscribeCommand(redisClient *c);
1024 void psubscribeCommand(redisClient *c);
1025 void punsubscribeCommand(redisClient *c);
1026 void publishCommand(redisClient *c);
1027 void watchCommand(redisClient *c);
1028 void unwatchCommand(redisClient *c);
1029
1030 #if defined(__GNUC__)
1031 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
1032 void free(void *ptr) __attribute__ ((deprecated));
1033 void *malloc(size_t size) __attribute__ ((deprecated));
1034 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
1035 #endif
1036
1037 #endif