]> git.saurik.com Git - redis.git/blob - src/redis.h
b0491e400329e4c6892518640c6ed9a57687e242
[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 /* Object types only used for persistence in .rdb files */
74 #define REDIS_HASH_ZIPMAP 9
75
76 /* Objects encoding. Some kind of objects like Strings and Hashes can be
77 * internally represented in multiple ways. The 'encoding' field of the object
78 * is set to one of this fields for this object. */
79 #define REDIS_ENCODING_RAW 0 /* Raw representation */
80 #define REDIS_ENCODING_INT 1 /* Encoded as integer */
81 #define REDIS_ENCODING_HT 2 /* Encoded as hash table */
82 #define REDIS_ENCODING_ZIPMAP 3 /* Encoded as zipmap */
83 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
84 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
85 #define REDIS_ENCODING_INTSET 6 /* Encoded as intset */
86
87 /* Object types only used for dumping to disk */
88 #define REDIS_EXPIRETIME 253
89 #define REDIS_SELECTDB 254
90 #define REDIS_EOF 255
91
92 /* Defines related to the dump file format. To store 32 bits lengths for short
93 * keys requires a lot of space, so we check the most significant 2 bits of
94 * the first byte to interpreter the length:
95 *
96 * 00|000000 => if the two MSB are 00 the len is the 6 bits of this byte
97 * 01|000000 00000000 => 01, the len is 14 byes, 6 bits + 8 bits of next byte
98 * 10|000000 [32 bit integer] => if it's 01, a full 32 bit len will follow
99 * 11|000000 this means: specially encoded object will follow. The six bits
100 * number specify the kind of object that follows.
101 * See the REDIS_RDB_ENC_* defines.
102 *
103 * Lenghts up to 63 are stored using a single byte, most DB keys, and may
104 * values, will fit inside. */
105 #define REDIS_RDB_6BITLEN 0
106 #define REDIS_RDB_14BITLEN 1
107 #define REDIS_RDB_32BITLEN 2
108 #define REDIS_RDB_ENCVAL 3
109 #define REDIS_RDB_LENERR UINT_MAX
110
111 /* When a length of a string object stored on disk has the first two bits
112 * set, the remaining two bits specify a special encoding for the object
113 * accordingly to the following defines: */
114 #define REDIS_RDB_ENC_INT8 0 /* 8 bit signed integer */
115 #define REDIS_RDB_ENC_INT16 1 /* 16 bit signed integer */
116 #define REDIS_RDB_ENC_INT32 2 /* 32 bit signed integer */
117 #define REDIS_RDB_ENC_LZF 3 /* string compressed with FASTLZ */
118
119 /* Scheduled IO opeations flags. */
120 #define REDIS_IO_LOAD 1
121 #define REDIS_IO_SAVE 2
122 #define REDIS_IO_LOADINPROG 4
123 #define REDIS_IO_SAVEINPROG 8
124
125 /* Generic IO flags */
126 #define REDIS_IO_ONLYLOADS 1
127 #define REDIS_IO_ASAP 2
128
129 #define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1
130 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
131
132 /* Client flags */
133 #define REDIS_SLAVE 1 /* This client is a slave server */
134 #define REDIS_MASTER 2 /* This client is a master server */
135 #define REDIS_MONITOR 4 /* This client is a slave monitor, see MONITOR */
136 #define REDIS_MULTI 8 /* This client is in a MULTI context */
137 #define REDIS_BLOCKED 16 /* The client is waiting in a blocking operation */
138 #define REDIS_IO_WAIT 32 /* The client is waiting for Virtual Memory I/O */
139 #define REDIS_DIRTY_CAS 64 /* Watched keys modified. EXEC will fail. */
140 #define REDIS_CLOSE_AFTER_REPLY 128 /* Close after writing entire reply. */
141 #define REDIS_UNBLOCKED 256 /* This client was unblocked and is stored in
142 server.unblocked_clients */
143
144 /* Client request types */
145 #define REDIS_REQ_INLINE 1
146 #define REDIS_REQ_MULTIBULK 2
147
148 /* Slave replication state - slave side */
149 #define REDIS_REPL_NONE 0 /* No active replication */
150 #define REDIS_REPL_CONNECT 1 /* Must connect to master */
151 #define REDIS_REPL_TRANSFER 2 /* Receiving .rdb from master */
152 #define REDIS_REPL_CONNECTED 3 /* Connected to master */
153
154 /* Slave replication state - from the point of view of master
155 * Note that in SEND_BULK and ONLINE state the slave receives new updates
156 * in its output queue. In the WAIT_BGSAVE state instead the server is waiting
157 * to start the next background saving in order to send updates to it. */
158 #define REDIS_REPL_WAIT_BGSAVE_START 3 /* master waits bgsave to start feeding it */
159 #define REDIS_REPL_WAIT_BGSAVE_END 4 /* master waits bgsave to start bulk DB transmission */
160 #define REDIS_REPL_SEND_BULK 5 /* master is sending the bulk DB */
161 #define REDIS_REPL_ONLINE 6 /* bulk DB already transmitted, receive updates */
162
163 /* List related stuff */
164 #define REDIS_HEAD 0
165 #define REDIS_TAIL 1
166
167 /* Sort operations */
168 #define REDIS_SORT_GET 0
169 #define REDIS_SORT_ASC 1
170 #define REDIS_SORT_DESC 2
171 #define REDIS_SORTKEY_MAX 1024
172
173 /* Log levels */
174 #define REDIS_DEBUG 0
175 #define REDIS_VERBOSE 1
176 #define REDIS_NOTICE 2
177 #define REDIS_WARNING 3
178
179 /* Anti-warning macro... */
180 #define REDIS_NOTUSED(V) ((void) V)
181
182 #define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^32 elements */
183 #define ZSKIPLIST_P 0.25 /* Skiplist P = 1/4 */
184
185 /* Append only defines */
186 #define APPENDFSYNC_NO 0
187 #define APPENDFSYNC_ALWAYS 1
188 #define APPENDFSYNC_EVERYSEC 2
189
190 /* Zip structure related defaults */
191 #define REDIS_HASH_MAX_ZIPMAP_ENTRIES 512
192 #define REDIS_HASH_MAX_ZIPMAP_VALUE 64
193 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 512
194 #define REDIS_LIST_MAX_ZIPLIST_VALUE 64
195 #define REDIS_SET_MAX_INTSET_ENTRIES 512
196
197 /* Sets operations codes */
198 #define REDIS_OP_UNION 0
199 #define REDIS_OP_DIFF 1
200 #define REDIS_OP_INTER 2
201
202 /* Redis maxmemory strategies */
203 #define REDIS_MAXMEMORY_VOLATILE_LRU 0
204 #define REDIS_MAXMEMORY_VOLATILE_TTL 1
205 #define REDIS_MAXMEMORY_VOLATILE_RANDOM 2
206 #define REDIS_MAXMEMORY_ALLKEYS_LRU 3
207 #define REDIS_MAXMEMORY_ALLKEYS_RANDOM 4
208 #define REDIS_MAXMEMORY_NO_EVICTION 5
209
210 /* Diskstore background saving thread states */
211 #define REDIS_BGSAVE_THREAD_UNACTIVE 0
212 #define REDIS_BGSAVE_THREAD_ACTIVE 1
213 #define REDIS_BGSAVE_THREAD_DONE_OK 2
214 #define REDIS_BGSAVE_THREAD_DONE_ERR 3
215
216 /* We can print the stacktrace, so our assert is defined this way: */
217 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
218 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
219 void _redisAssert(char *estr, char *file, int line);
220 void _redisPanic(char *msg, char *file, int line);
221
222 /*-----------------------------------------------------------------------------
223 * Data types
224 *----------------------------------------------------------------------------*/
225
226 /* A redis object, that is a type able to hold a string / list / set */
227
228 /* The actual Redis Object */
229 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
230 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
231 typedef struct redisObject {
232 unsigned type:4;
233 unsigned notused:2; /* Not used */
234 unsigned encoding:4;
235 unsigned lru:22; /* lru time (relative to server.lruclock) */
236 int refcount;
237 void *ptr;
238 /* VM fields are only allocated if VM is active, otherwise the
239 * object allocation function will just allocate
240 * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
241 * Redis without VM active will not have any overhead. */
242 } robj;
243
244 /* The VM pointer structure - identifies an object in the swap file.
245 *
246 * This object is stored in place of the value
247 * object in the main key->value hash table representing a database.
248 * Note that the first fields (type, storage) are the same as the redisObject
249 * structure so that vmPointer strucuters can be accessed even when casted
250 * as redisObject structures.
251 *
252 * This is useful as we don't know if a value object is or not on disk, but we
253 * are always able to read obj->storage to check this. For vmPointer
254 * structures "type" is set to REDIS_VMPOINTER (even if without this field
255 * is still possible to check the kind of object from the value of 'storage').*/
256 typedef struct vmPointer {
257 unsigned type:4;
258 unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
259 unsigned notused:26;
260 unsigned int vtype; /* type of the object stored in the swap file */
261 off_t page; /* the page at witch the object is stored on disk */
262 off_t usedpages; /* number of pages used on disk */
263 } vmpointer;
264
265 /* Macro used to initalize a Redis object allocated on the stack.
266 * Note that this macro is taken near the structure definition to make sure
267 * we'll update it when the structure is changed, to avoid bugs like
268 * bug #85 introduced exactly in this way. */
269 #define initStaticStringObject(_var,_ptr) do { \
270 _var.refcount = 1; \
271 _var.type = REDIS_STRING; \
272 _var.encoding = REDIS_ENCODING_RAW; \
273 _var.ptr = _ptr; \
274 } while(0);
275
276 typedef struct redisDb {
277 dict *dict; /* The keyspace for this DB */
278 dict *expires; /* Timeout of keys with a timeout set */
279 dict *blocking_keys; /* Keys with clients waiting for data (BLPOP) */
280 dict *io_keys; /* Keys with clients waiting for DS I/O */
281 dict *io_negcache; /* Negative caching for disk store */
282 dict *io_queued; /* Queued IO operations hash table */
283 dict *watched_keys; /* WATCHED keys for MULTI/EXEC CAS */
284 int id;
285 } redisDb;
286
287 /* Client MULTI/EXEC state */
288 typedef struct multiCmd {
289 robj **argv;
290 int argc;
291 struct redisCommand *cmd;
292 } multiCmd;
293
294 typedef struct multiState {
295 multiCmd *commands; /* Array of MULTI commands */
296 int count; /* Total number of MULTI commands */
297 } multiState;
298
299 typedef struct blockingState {
300 robj **keys; /* The key we are waiting to terminate a blocking
301 * operation such as BLPOP. Otherwise NULL. */
302 int count; /* Number of blocking keys */
303 time_t timeout; /* Blocking operation timeout. If UNIX current time
304 * is >= timeout then the operation timed out. */
305 robj *target; /* The key that should receive the element,
306 * for BRPOPLPUSH. */
307 } blockingState;
308
309 /* With multiplexing we need to take per-clinet state.
310 * Clients are taken in a liked list. */
311 typedef struct redisClient {
312 int fd;
313 redisDb *db;
314 int dictid;
315 sds querybuf;
316 int argc;
317 robj **argv;
318 int reqtype;
319 int multibulklen; /* number of multi bulk arguments left to read */
320 long bulklen; /* length of bulk argument in multi bulk request */
321 list *reply;
322 int sentlen;
323 time_t lastinteraction; /* time of the last interaction, used for timeout */
324 int flags; /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
325 int slaveseldb; /* slave selected db, if this client is a slave */
326 int authenticated; /* when requirepass is non-NULL */
327 int replstate; /* replication state if this is a slave */
328 int repldbfd; /* replication DB file descriptor */
329 long repldboff; /* replication DB file offset */
330 off_t repldbsize; /* replication DB file size */
331 multiState mstate; /* MULTI/EXEC state */
332 blockingState bpop; /* blocking state */
333 list *io_keys; /* Keys this client is waiting to be loaded from the
334 * swap file in order to continue. */
335 list *watched_keys; /* Keys WATCHED for MULTI/EXEC CAS */
336 dict *pubsub_channels; /* channels a client is interested in (SUBSCRIBE) */
337 list *pubsub_patterns; /* patterns a client is interested in (SUBSCRIBE) */
338
339 /* Response buffer */
340 int bufpos;
341 char buf[REDIS_REPLY_CHUNK_BYTES];
342 } redisClient;
343
344 struct saveparam {
345 time_t seconds;
346 int changes;
347 };
348
349 struct sharedObjectsStruct {
350 robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
351 *colon, *nullbulk, *nullmultibulk, *queued,
352 *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
353 *outofrangeerr, *loadingerr, *plus,
354 *select0, *select1, *select2, *select3, *select4,
355 *select5, *select6, *select7, *select8, *select9,
356 *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
357 *mbulk4, *psubscribebulk, *punsubscribebulk,
358 *integers[REDIS_SHARED_INTEGERS];
359 };
360
361 /* Global server state structure */
362 struct redisServer {
363 /* General */
364 pthread_t mainthread;
365 redisDb *db;
366 dict *commands; /* Command table hahs table */
367 aeEventLoop *el;
368 /* Networking */
369 int port;
370 char *bindaddr;
371 char *unixsocket;
372 int ipfd;
373 int sofd;
374 list *clients;
375 list *slaves, *monitors;
376 char neterr[ANET_ERR_LEN];
377 /* RDB / AOF loading information */
378 int loading;
379 off_t loading_total_bytes;
380 off_t loading_loaded_bytes;
381 time_t loading_start_time;
382 /* Fast pointers to often looked up command */
383 struct redisCommand *delCommand, *multiCommand;
384 int cronloops; /* number of times the cron function run */
385 time_t lastsave; /* Unix time of last save succeeede */
386 /* Fields used only for stats */
387 time_t stat_starttime; /* server start time */
388 long long stat_numcommands; /* number of processed commands */
389 long long stat_numconnections; /* number of connections received */
390 long long stat_expiredkeys; /* number of expired keys */
391 long long stat_evictedkeys; /* number of evicted keys (maxmemory) */
392 long long stat_keyspace_hits; /* number of successful lookups of keys */
393 long long stat_keyspace_misses; /* number of failed lookups of keys */
394 /* Configuration */
395 int verbosity;
396 int maxidletime;
397 int dbnum;
398 int daemonize;
399 int appendonly;
400 int appendfsync;
401 int no_appendfsync_on_rewrite;
402 int shutdown_asap;
403 int activerehashing;
404 char *requirepass;
405 /* Persistence */
406 long long dirty; /* changes to DB from the last save */
407 long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
408 time_t lastfsync;
409 int appendfd;
410 int appendseldb;
411 char *pidfile;
412 pid_t bgsavechildpid;
413 pid_t bgrewritechildpid;
414 int bgsavethread_state;
415 pthread_mutex_t bgsavethread_mutex;
416 pthread_t bgsavethread;
417 sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
418 sds aofbuf; /* AOF buffer, written before entering the event loop */
419 struct saveparam *saveparams;
420 int saveparamslen;
421 char *dbfilename;
422 int rdbcompression;
423 char *appendfilename;
424 /* Logging */
425 char *logfile;
426 int syslog_enabled;
427 char *syslog_ident;
428 int syslog_facility;
429 /* Replication related */
430 int isslave;
431 /* Slave specific fields */
432 char *masterauth;
433 char *masterhost;
434 int masterport;
435 redisClient *master; /* client that is master for this slave */
436 int replstate; /* replication status if the instance is a slave */
437 off_t repl_transfer_left; /* bytes left reading .rdb */
438 int repl_transfer_s; /* slave -> master SYNC socket */
439 int repl_transfer_fd; /* slave -> master SYNC temp file descriptor */
440 char *repl_transfer_tmpfile; /* slave-> master SYNC temp file name */
441 time_t repl_transfer_lastio; /* unix time of the latest read, for timeout */
442 int repl_serve_stale_data; /* Serve stale data when link is down? */
443 /* Limits */
444 unsigned int maxclients;
445 unsigned long long maxmemory;
446 int maxmemory_policy;
447 int maxmemory_samples;
448 /* Blocked clients */
449 unsigned int bpop_blocked_clients;
450 unsigned int cache_blocked_clients;
451 list *unblocked_clients; /* list of clients to unblock before next loop */
452 list *cache_io_queue; /* IO operations queue */
453 int cache_flush_delay; /* seconds to wait before flushing keys */
454 /* Sort parameters - qsort_r() is only available under BSD so we
455 * have to take this state global, in order to pass it to sortCompare() */
456 int sort_desc;
457 int sort_alpha;
458 int sort_bypattern;
459 /* Virtual memory configuration */
460 int ds_enabled; /* backend disk in redis.conf */
461 char *ds_path; /* location of the disk store on disk */
462 unsigned long long cache_max_memory;
463 /* Zip structure config */
464 size_t hash_max_zipmap_entries;
465 size_t hash_max_zipmap_value;
466 size_t list_max_ziplist_entries;
467 size_t list_max_ziplist_value;
468 size_t set_max_intset_entries;
469 time_t unixtime; /* Unix time sampled every second. */
470 /* Virtual memory I/O threads stuff */
471 /* An I/O thread process an element taken from the io_jobs queue and
472 * put the result of the operation in the io_done list. While the
473 * job is being processed, it's put on io_processing queue. */
474 list *io_newjobs; /* List of VM I/O jobs yet to be processed */
475 list *io_processing; /* List of VM I/O jobs being processed */
476 list *io_processed; /* List of VM I/O jobs already processed */
477 list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
478 pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
479 pthread_cond_t io_condvar; /* I/O threads conditional variable */
480 pthread_attr_t io_threads_attr; /* attributes for threads creation */
481 int io_active_threads; /* Number of running I/O threads */
482 int vm_max_threads; /* Max number of I/O threads running at the same time */
483 /* Our main thread is blocked on the event loop, locking for sockets ready
484 * to be read or written, so when a threaded I/O operation is ready to be
485 * processed by the main thread, the I/O thread will use a unix pipe to
486 * awake the main thread. The followings are the two pipe FDs. */
487 int io_ready_pipe_read;
488 int io_ready_pipe_write;
489 /* Virtual memory stats */
490 unsigned long long vm_stats_used_pages;
491 unsigned long long vm_stats_swapped_objects;
492 unsigned long long vm_stats_swapouts;
493 unsigned long long vm_stats_swapins;
494 /* Pubsub */
495 dict *pubsub_channels; /* Map channels to list of subscribed clients */
496 list *pubsub_patterns; /* A list of pubsub_patterns */
497 /* Misc */
498 unsigned lruclock:22; /* clock incrementing every minute, for LRU */
499 unsigned lruclock_padding:10;
500 };
501
502 typedef struct pubsubPattern {
503 redisClient *client;
504 robj *pattern;
505 } pubsubPattern;
506
507 typedef void redisCommandProc(redisClient *c);
508 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
509 struct redisCommand {
510 char *name;
511 redisCommandProc *proc;
512 int arity;
513 int flags;
514 /* Use a function to determine which keys need to be loaded
515 * in the background prior to executing this command. Takes precedence
516 * over vm_firstkey and others, ignored when NULL */
517 redisVmPreloadProc *vm_preload_proc;
518 /* What keys should be loaded in background when calling this command? */
519 int vm_firstkey; /* The first argument that's a key (0 = no keys) */
520 int vm_lastkey; /* THe last argument that's a key */
521 int vm_keystep; /* The step between first and last key */
522 long long microseconds, calls;
523 };
524
525 struct redisFunctionSym {
526 char *name;
527 unsigned long pointer;
528 };
529
530 typedef struct _redisSortObject {
531 robj *obj;
532 union {
533 double score;
534 robj *cmpobj;
535 } u;
536 } redisSortObject;
537
538 typedef struct _redisSortOperation {
539 int type;
540 robj *pattern;
541 } redisSortOperation;
542
543 /* ZSETs use a specialized version of Skiplists */
544 typedef struct zskiplistNode {
545 robj *obj;
546 double score;
547 struct zskiplistNode *backward;
548 struct zskiplistLevel {
549 struct zskiplistNode *forward;
550 unsigned int span;
551 } level[];
552 } zskiplistNode;
553
554 typedef struct zskiplist {
555 struct zskiplistNode *header, *tail;
556 unsigned long length;
557 int level;
558 } zskiplist;
559
560 typedef struct zset {
561 dict *dict;
562 zskiplist *zsl;
563 } zset;
564
565 /* DIsk store threaded I/O request message */
566 #define REDIS_IOJOB_LOAD 0
567 #define REDIS_IOJOB_SAVE 1
568
569 typedef struct iojob {
570 int type; /* Request type, REDIS_IOJOB_* */
571 redisDb *db;/* Redis database */
572 robj *key; /* This I/O request is about this key */
573 robj *val; /* the value to swap for REDIS_IOJOB_SAVE, otherwise this
574 * field is populated by the I/O thread for REDIS_IOJOB_LOAD. */
575 time_t expire; /* Expire time for this key on REDIS_IOJOB_LOAD */
576 } iojob;
577
578 /* IO operations scheduled -- check dscache.c for more info */
579 typedef struct ioop {
580 int type;
581 redisDb *db;
582 robj *key;
583 time_t ctime; /* This is the creation time of the entry. */
584 } ioop;
585
586 /* Structure to hold list iteration abstraction. */
587 typedef struct {
588 robj *subject;
589 unsigned char encoding;
590 unsigned char direction; /* Iteration direction */
591 unsigned char *zi;
592 listNode *ln;
593 } listTypeIterator;
594
595 /* Structure for an entry while iterating over a list. */
596 typedef struct {
597 listTypeIterator *li;
598 unsigned char *zi; /* Entry in ziplist */
599 listNode *ln; /* Entry in linked list */
600 } listTypeEntry;
601
602 /* Structure to hold set iteration abstraction. */
603 typedef struct {
604 robj *subject;
605 int encoding;
606 int ii; /* intset iterator */
607 dictIterator *di;
608 } setTypeIterator;
609
610 /* Structure to hold hash iteration abstration. Note that iteration over
611 * hashes involves both fields and values. Because it is possible that
612 * not both are required, store pointers in the iterator to avoid
613 * unnecessary memory allocation for fields/values. */
614 typedef struct {
615 int encoding;
616 unsigned char *zi;
617 unsigned char *zk, *zv;
618 unsigned int zklen, zvlen;
619
620 dictIterator *di;
621 dictEntry *de;
622 } hashTypeIterator;
623
624 #define REDIS_HASH_KEY 1
625 #define REDIS_HASH_VALUE 2
626
627 /*-----------------------------------------------------------------------------
628 * Extern declarations
629 *----------------------------------------------------------------------------*/
630
631 extern struct redisServer server;
632 extern struct sharedObjectsStruct shared;
633 extern dictType setDictType;
634 extern dictType zsetDictType;
635 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
636 dictType hashDictType;
637
638 /*-----------------------------------------------------------------------------
639 * Functions prototypes
640 *----------------------------------------------------------------------------*/
641
642 /* Utils */
643 long long ustime(void);
644
645 /* networking.c -- Networking and Client related operations */
646 redisClient *createClient(int fd);
647 void closeTimedoutClients(void);
648 void freeClient(redisClient *c);
649 void resetClient(redisClient *c);
650 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
651 void addReply(redisClient *c, robj *obj);
652 void *addDeferredMultiBulkLength(redisClient *c);
653 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
654 void addReplySds(redisClient *c, sds s);
655 void processInputBuffer(redisClient *c);
656 void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask);
657 void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask);
658 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
659 void addReplyBulk(redisClient *c, robj *obj);
660 void addReplyBulkCString(redisClient *c, char *s);
661 void addReplyBulkCBuffer(redisClient *c, void *p, size_t len);
662 void addReplyBulkLongLong(redisClient *c, long long ll);
663 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
664 void addReply(redisClient *c, robj *obj);
665 void addReplySds(redisClient *c, sds s);
666 void addReplyError(redisClient *c, char *err);
667 void addReplyStatus(redisClient *c, char *status);
668 void addReplyDouble(redisClient *c, double d);
669 void addReplyLongLong(redisClient *c, long long ll);
670 void addReplyMultiBulkLen(redisClient *c, long length);
671 void *dupClientReplyValue(void *o);
672 void getClientsMaxBuffers(unsigned long *longest_output_list,
673 unsigned long *biggest_input_buffer);
674
675 #ifdef __GNUC__
676 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
677 __attribute__((format(printf, 2, 3)));
678 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
679 __attribute__((format(printf, 2, 3)));
680 #else
681 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
682 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
683 #endif
684
685 /* List data type */
686 void listTypeTryConversion(robj *subject, robj *value);
687 void listTypePush(robj *subject, robj *value, int where);
688 robj *listTypePop(robj *subject, int where);
689 unsigned long listTypeLength(robj *subject);
690 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
691 void listTypeReleaseIterator(listTypeIterator *li);
692 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
693 robj *listTypeGet(listTypeEntry *entry);
694 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
695 int listTypeEqual(listTypeEntry *entry, robj *o);
696 void listTypeDelete(listTypeEntry *entry);
697 void listTypeConvert(robj *subject, int enc);
698 void unblockClientWaitingData(redisClient *c);
699 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
700 void popGenericCommand(redisClient *c, int where);
701
702 /* MULTI/EXEC/WATCH... */
703 void unwatchAllKeys(redisClient *c);
704 void initClientMultiState(redisClient *c);
705 void freeClientMultiState(redisClient *c);
706 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
707 void touchWatchedKey(redisDb *db, robj *key);
708 void touchWatchedKeysOnFlush(int dbid);
709
710 /* Redis object implementation */
711 void decrRefCount(void *o);
712 void incrRefCount(robj *o);
713 void freeStringObject(robj *o);
714 void freeListObject(robj *o);
715 void freeSetObject(robj *o);
716 void freeZsetObject(robj *o);
717 void freeHashObject(robj *o);
718 robj *createObject(int type, void *ptr);
719 robj *createStringObject(char *ptr, size_t len);
720 robj *dupStringObject(robj *o);
721 robj *tryObjectEncoding(robj *o);
722 robj *getDecodedObject(robj *o);
723 size_t stringObjectLen(robj *o);
724 robj *createStringObjectFromLongLong(long long value);
725 robj *createListObject(void);
726 robj *createZiplistObject(void);
727 robj *createSetObject(void);
728 robj *createIntsetObject(void);
729 robj *createHashObject(void);
730 robj *createZsetObject(void);
731 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
732 int checkType(redisClient *c, robj *o, int type);
733 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
734 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
735 int getLongLongFromObject(robj *o, long long *target);
736 char *strEncoding(int encoding);
737 int compareStringObjects(robj *a, robj *b);
738 int equalStringObjects(robj *a, robj *b);
739 unsigned long estimateObjectIdleTime(robj *o);
740
741 /* Synchronous I/O with timeout */
742 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
743 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
744 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
745 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
746 int fwriteBulkDouble(FILE *fp, double d);
747 int fwriteBulkLongLong(FILE *fp, long long l);
748 int fwriteBulkObject(FILE *fp, robj *obj);
749
750 /* Replication */
751 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
752 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
753 int syncWithMaster(void);
754 void updateSlavesWaitingBgsave(int bgsaveerr);
755 void replicationCron(void);
756
757 /* Generic persistence functions */
758 void startLoading(FILE *fp);
759 void loadingProgress(off_t pos);
760 void stopLoading(void);
761
762 /* RDB persistence */
763 int rdbLoad(char *filename);
764 int rdbSaveBackground(char *filename);
765 void rdbRemoveTempFile(pid_t childpid);
766 int rdbSave(char *filename);
767 int rdbSaveObject(FILE *fp, robj *o);
768 off_t rdbSavedObjectLen(robj *o);
769 off_t rdbSavedObjectPages(robj *o);
770 robj *rdbLoadObject(int type, FILE *fp);
771 void backgroundSaveDoneHandler(int exitcode, int bysignal);
772 int rdbSaveKeyValuePair(FILE *fp, robj *key, robj *val, time_t expireitme, time_t now);
773 int rdbLoadType(FILE *fp);
774 time_t rdbLoadTime(FILE *fp);
775 robj *rdbLoadStringObject(FILE *fp);
776 int rdbSaveType(FILE *fp, unsigned char type);
777 int rdbSaveLen(FILE *fp, uint32_t len);
778
779 /* AOF persistence */
780 void flushAppendOnlyFile(void);
781 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
782 void aofRemoveTempFile(pid_t childpid);
783 int rewriteAppendOnlyFileBackground(void);
784 int loadAppendOnlyFile(char *filename);
785 void stopAppendOnly(void);
786 int startAppendOnly(void);
787 void backgroundRewriteDoneHandler(int exitcode, int bysignal);
788
789 /* Sorted sets data type */
790 zskiplist *zslCreate(void);
791 void zslFree(zskiplist *zsl);
792 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
793
794 /* Core functions */
795 void freeMemoryIfNeeded(void);
796 int processCommand(redisClient *c);
797 void setupSigSegvAction(void);
798 struct redisCommand *lookupCommand(sds name);
799 struct redisCommand *lookupCommandByCString(char *s);
800 void call(redisClient *c, struct redisCommand *cmd);
801 int prepareForShutdown();
802 void redisLog(int level, const char *fmt, ...);
803 void usage();
804 void updateDictResizePolicy(void);
805 int htNeedsResize(dict *dict);
806 void oom(const char *msg);
807 void populateCommandTable(void);
808 void resetCommandTableStats(void);
809
810 /* Disk store */
811 int dsOpen(void);
812 int dsClose(void);
813 int dsSet(redisDb *db, robj *key, robj *val, time_t expire);
814 robj *dsGet(redisDb *db, robj *key, time_t *expire);
815 int dsDel(redisDb *db, robj *key);
816 int dsExists(redisDb *db, robj *key);
817 void dsFlushDb(int dbid);
818 int dsRdbSaveBackground(char *filename);
819 int dsRdbSave(char *filename);
820
821 /* Disk Store Cache */
822 void dsInit(void);
823 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
824 void lockThreadedIO(void);
825 void unlockThreadedIO(void);
826 void freeIOJob(iojob *j);
827 void queueIOJob(iojob *j);
828 void waitEmptyIOJobsQueue(void);
829 void processAllPendingIOJobs(void);
830 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
831 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
832 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
833 int dontWaitForSwappedKey(redisClient *c, robj *key);
834 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
835 int cacheFreeOneEntry(void);
836 void cacheScheduleIOAddFlag(redisDb *db, robj *key, long flag);
837 void cacheScheduleIODelFlag(redisDb *db, robj *key, long flag);
838 int cacheScheduleIOGetFlags(redisDb *db, robj *key);
839 void cacheScheduleIO(redisDb *db, robj *key, int type);
840 void cacheCron(void);
841 int cacheKeyMayExist(redisDb *db, robj *key);
842 void cacheSetKeyMayExist(redisDb *db, robj *key);
843 void cacheSetKeyDoesNotExist(redisDb *db, robj *key);
844 void cacheForcePointInTime(void);
845
846 /* Set data type */
847 robj *setTypeCreate(robj *value);
848 int setTypeAdd(robj *subject, robj *value);
849 int setTypeRemove(robj *subject, robj *value);
850 int setTypeIsMember(robj *subject, robj *value);
851 setTypeIterator *setTypeInitIterator(robj *subject);
852 void setTypeReleaseIterator(setTypeIterator *si);
853 int setTypeNext(setTypeIterator *si, robj **objele, int64_t *llele);
854 robj *setTypeNextObject(setTypeIterator *si);
855 int setTypeRandomElement(robj *setobj, robj **objele, int64_t *llele);
856 unsigned long setTypeSize(robj *subject);
857 void setTypeConvert(robj *subject, int enc);
858
859 /* Hash data type */
860 void convertToRealHash(robj *o);
861 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
862 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
863 int hashTypeGet(robj *o, robj *key, robj **objval, unsigned char **v, unsigned int *vlen);
864 robj *hashTypeGetObject(robj *o, robj *key);
865 int hashTypeExists(robj *o, robj *key);
866 int hashTypeSet(robj *o, robj *key, robj *value);
867 int hashTypeDelete(robj *o, robj *key);
868 unsigned long hashTypeLength(robj *o);
869 hashTypeIterator *hashTypeInitIterator(robj *subject);
870 void hashTypeReleaseIterator(hashTypeIterator *hi);
871 int hashTypeNext(hashTypeIterator *hi);
872 int hashTypeCurrent(hashTypeIterator *hi, int what, robj **objval, unsigned char **v, unsigned int *vlen);
873 robj *hashTypeCurrentObject(hashTypeIterator *hi, int what);
874 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
875
876 /* Pub / Sub */
877 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
878 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
879 void freePubsubPattern(void *p);
880 int listMatchPubsubPattern(void *a, void *b);
881
882 /* Utility functions */
883 int stringmatchlen(const char *pattern, int patternLen,
884 const char *string, int stringLen, int nocase);
885 int stringmatch(const char *pattern, const char *string, int nocase);
886 long long memtoll(const char *p, int *err);
887 int ll2string(char *s, size_t len, long long value);
888 int isStringRepresentableAsLong(sds s, long *longval);
889 int isStringRepresentableAsLongLong(sds s, long long *longval);
890 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
891
892 /* Configuration */
893 void loadServerConfig(char *filename);
894 void appendServerSaveParams(time_t seconds, int changes);
895 void resetServerSaveParams();
896
897 /* db.c -- Keyspace access API */
898 int removeExpire(redisDb *db, robj *key);
899 void propagateExpire(redisDb *db, robj *key);
900 int expireIfNeeded(redisDb *db, robj *key);
901 time_t getExpire(redisDb *db, robj *key);
902 void setExpire(redisDb *db, robj *key, time_t when);
903 robj *lookupKey(redisDb *db, robj *key);
904 robj *lookupKeyRead(redisDb *db, robj *key);
905 robj *lookupKeyWrite(redisDb *db, robj *key);
906 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
907 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
908 int dbAdd(redisDb *db, robj *key, robj *val);
909 int dbReplace(redisDb *db, robj *key, robj *val);
910 int dbExists(redisDb *db, robj *key);
911 robj *dbRandomKey(redisDb *db);
912 int dbDelete(redisDb *db, robj *key);
913 long long emptyDb();
914 int selectDb(redisClient *c, int id);
915 void signalModifiedKey(redisDb *db, robj *key);
916 void signalFlushedDb(int dbid);
917
918 /* Git SHA1 */
919 char *redisGitSHA1(void);
920 char *redisGitDirty(void);
921
922 /* Commands prototypes */
923 void authCommand(redisClient *c);
924 void pingCommand(redisClient *c);
925 void echoCommand(redisClient *c);
926 void setCommand(redisClient *c);
927 void setnxCommand(redisClient *c);
928 void setexCommand(redisClient *c);
929 void getCommand(redisClient *c);
930 void delCommand(redisClient *c);
931 void existsCommand(redisClient *c);
932 void setbitCommand(redisClient *c);
933 void getbitCommand(redisClient *c);
934 void setrangeCommand(redisClient *c);
935 void getrangeCommand(redisClient *c);
936 void incrCommand(redisClient *c);
937 void decrCommand(redisClient *c);
938 void incrbyCommand(redisClient *c);
939 void decrbyCommand(redisClient *c);
940 void selectCommand(redisClient *c);
941 void randomkeyCommand(redisClient *c);
942 void keysCommand(redisClient *c);
943 void dbsizeCommand(redisClient *c);
944 void lastsaveCommand(redisClient *c);
945 void saveCommand(redisClient *c);
946 void bgsaveCommand(redisClient *c);
947 void bgrewriteaofCommand(redisClient *c);
948 void shutdownCommand(redisClient *c);
949 void moveCommand(redisClient *c);
950 void renameCommand(redisClient *c);
951 void renamenxCommand(redisClient *c);
952 void lpushCommand(redisClient *c);
953 void rpushCommand(redisClient *c);
954 void lpushxCommand(redisClient *c);
955 void rpushxCommand(redisClient *c);
956 void linsertCommand(redisClient *c);
957 void lpopCommand(redisClient *c);
958 void rpopCommand(redisClient *c);
959 void llenCommand(redisClient *c);
960 void lindexCommand(redisClient *c);
961 void lrangeCommand(redisClient *c);
962 void ltrimCommand(redisClient *c);
963 void typeCommand(redisClient *c);
964 void lsetCommand(redisClient *c);
965 void saddCommand(redisClient *c);
966 void sremCommand(redisClient *c);
967 void smoveCommand(redisClient *c);
968 void sismemberCommand(redisClient *c);
969 void scardCommand(redisClient *c);
970 void spopCommand(redisClient *c);
971 void srandmemberCommand(redisClient *c);
972 void sinterCommand(redisClient *c);
973 void sinterstoreCommand(redisClient *c);
974 void sunionCommand(redisClient *c);
975 void sunionstoreCommand(redisClient *c);
976 void sdiffCommand(redisClient *c);
977 void sdiffstoreCommand(redisClient *c);
978 void syncCommand(redisClient *c);
979 void flushdbCommand(redisClient *c);
980 void flushallCommand(redisClient *c);
981 void sortCommand(redisClient *c);
982 void lremCommand(redisClient *c);
983 void rpoplpushCommand(redisClient *c);
984 void infoCommand(redisClient *c);
985 void mgetCommand(redisClient *c);
986 void monitorCommand(redisClient *c);
987 void expireCommand(redisClient *c);
988 void expireatCommand(redisClient *c);
989 void getsetCommand(redisClient *c);
990 void ttlCommand(redisClient *c);
991 void persistCommand(redisClient *c);
992 void slaveofCommand(redisClient *c);
993 void debugCommand(redisClient *c);
994 void msetCommand(redisClient *c);
995 void msetnxCommand(redisClient *c);
996 void zaddCommand(redisClient *c);
997 void zincrbyCommand(redisClient *c);
998 void zrangeCommand(redisClient *c);
999 void zrangebyscoreCommand(redisClient *c);
1000 void zrevrangebyscoreCommand(redisClient *c);
1001 void zcountCommand(redisClient *c);
1002 void zrevrangeCommand(redisClient *c);
1003 void zcardCommand(redisClient *c);
1004 void zremCommand(redisClient *c);
1005 void zscoreCommand(redisClient *c);
1006 void zremrangebyscoreCommand(redisClient *c);
1007 void multiCommand(redisClient *c);
1008 void execCommand(redisClient *c);
1009 void discardCommand(redisClient *c);
1010 void blpopCommand(redisClient *c);
1011 void brpopCommand(redisClient *c);
1012 void brpoplpushCommand(redisClient *c);
1013 void appendCommand(redisClient *c);
1014 void strlenCommand(redisClient *c);
1015 void zrankCommand(redisClient *c);
1016 void zrevrankCommand(redisClient *c);
1017 void hsetCommand(redisClient *c);
1018 void hsetnxCommand(redisClient *c);
1019 void hgetCommand(redisClient *c);
1020 void hmsetCommand(redisClient *c);
1021 void hmgetCommand(redisClient *c);
1022 void hdelCommand(redisClient *c);
1023 void hlenCommand(redisClient *c);
1024 void zremrangebyrankCommand(redisClient *c);
1025 void zunionstoreCommand(redisClient *c);
1026 void zinterstoreCommand(redisClient *c);
1027 void hkeysCommand(redisClient *c);
1028 void hvalsCommand(redisClient *c);
1029 void hgetallCommand(redisClient *c);
1030 void hexistsCommand(redisClient *c);
1031 void configCommand(redisClient *c);
1032 void hincrbyCommand(redisClient *c);
1033 void subscribeCommand(redisClient *c);
1034 void unsubscribeCommand(redisClient *c);
1035 void psubscribeCommand(redisClient *c);
1036 void punsubscribeCommand(redisClient *c);
1037 void publishCommand(redisClient *c);
1038 void watchCommand(redisClient *c);
1039 void unwatchCommand(redisClient *c);
1040
1041 #if defined(__GNUC__)
1042 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
1043 void free(void *ptr) __attribute__ ((deprecated));
1044 void *malloc(size_t size) __attribute__ ((deprecated));
1045 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
1046 #endif
1047
1048 #endif