]> git.saurik.com Git - redis.git/blob - src/redis.h
Persistence code for encoded sorted sets
[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 /* Object types only used for persistence in .rdb files */
75 #define REDIS_HASH_ZIPMAP 9
76 #define REDIS_LIST_ZIPLIST 10
77 #define REDIS_SET_INTSET 11
78 #define REDIS_ZSET_ZIPLIST 12
79
80 /* Objects encoding. Some kind of objects like Strings and Hashes can be
81 * internally represented in multiple ways. The 'encoding' field of the object
82 * is set to one of this fields for this object. */
83 #define REDIS_ENCODING_RAW 0 /* Raw representation */
84 #define REDIS_ENCODING_INT 1 /* Encoded as integer */
85 #define REDIS_ENCODING_HT 2 /* Encoded as hash table */
86 #define REDIS_ENCODING_ZIPMAP 3 /* Encoded as zipmap */
87 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
88 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
89 #define REDIS_ENCODING_INTSET 6 /* Encoded as intset */
90
91 /* Object types only used for dumping to disk */
92 #define REDIS_EXPIRETIME 253
93 #define REDIS_SELECTDB 254
94 #define REDIS_EOF 255
95
96 /* Defines related to the dump file format. To store 32 bits lengths for short
97 * keys requires a lot of space, so we check the most significant 2 bits of
98 * the first byte to interpreter the length:
99 *
100 * 00|000000 => if the two MSB are 00 the len is the 6 bits of this byte
101 * 01|000000 00000000 => 01, the len is 14 byes, 6 bits + 8 bits of next byte
102 * 10|000000 [32 bit integer] => if it's 01, a full 32 bit len will follow
103 * 11|000000 this means: specially encoded object will follow. The six bits
104 * number specify the kind of object that follows.
105 * See the REDIS_RDB_ENC_* defines.
106 *
107 * Lenghts up to 63 are stored using a single byte, most DB keys, and may
108 * values, will fit inside. */
109 #define REDIS_RDB_6BITLEN 0
110 #define REDIS_RDB_14BITLEN 1
111 #define REDIS_RDB_32BITLEN 2
112 #define REDIS_RDB_ENCVAL 3
113 #define REDIS_RDB_LENERR UINT_MAX
114
115 /* When a length of a string object stored on disk has the first two bits
116 * set, the remaining two bits specify a special encoding for the object
117 * accordingly to the following defines: */
118 #define REDIS_RDB_ENC_INT8 0 /* 8 bit signed integer */
119 #define REDIS_RDB_ENC_INT16 1 /* 16 bit signed integer */
120 #define REDIS_RDB_ENC_INT32 2 /* 32 bit signed integer */
121 #define REDIS_RDB_ENC_LZF 3 /* string compressed with FASTLZ */
122
123 /* Scheduled IO opeations flags. */
124 #define REDIS_IO_LOAD 1
125 #define REDIS_IO_SAVE 2
126 #define REDIS_IO_LOADINPROG 4
127 #define REDIS_IO_SAVEINPROG 8
128
129 /* Generic IO flags */
130 #define REDIS_IO_ONLYLOADS 1
131 #define REDIS_IO_ASAP 2
132
133 #define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1
134 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
135
136 /* Client flags */
137 #define REDIS_SLAVE 1 /* This client is a slave server */
138 #define REDIS_MASTER 2 /* This client is a master server */
139 #define REDIS_MONITOR 4 /* This client is a slave monitor, see MONITOR */
140 #define REDIS_MULTI 8 /* This client is in a MULTI context */
141 #define REDIS_BLOCKED 16 /* The client is waiting in a blocking operation */
142 #define REDIS_IO_WAIT 32 /* The client is waiting for Virtual Memory I/O */
143 #define REDIS_DIRTY_CAS 64 /* Watched keys modified. EXEC will fail. */
144 #define REDIS_CLOSE_AFTER_REPLY 128 /* Close after writing entire reply. */
145 #define REDIS_UNBLOCKED 256 /* This client was unblocked and is stored in
146 server.unblocked_clients */
147
148 /* Client request types */
149 #define REDIS_REQ_INLINE 1
150 #define REDIS_REQ_MULTIBULK 2
151
152 /* Slave replication state - slave side */
153 #define REDIS_REPL_NONE 0 /* No active replication */
154 #define REDIS_REPL_CONNECT 1 /* Must connect to master */
155 #define REDIS_REPL_TRANSFER 2 /* Receiving .rdb from master */
156 #define REDIS_REPL_CONNECTED 3 /* Connected to master */
157
158 /* Slave replication state - from the point of view of master
159 * Note that in SEND_BULK and ONLINE state the slave receives new updates
160 * in its output queue. In the WAIT_BGSAVE state instead the server is waiting
161 * to start the next background saving in order to send updates to it. */
162 #define REDIS_REPL_WAIT_BGSAVE_START 3 /* master waits bgsave to start feeding it */
163 #define REDIS_REPL_WAIT_BGSAVE_END 4 /* master waits bgsave to start bulk DB transmission */
164 #define REDIS_REPL_SEND_BULK 5 /* master is sending the bulk DB */
165 #define REDIS_REPL_ONLINE 6 /* bulk DB already transmitted, receive updates */
166
167 /* List related stuff */
168 #define REDIS_HEAD 0
169 #define REDIS_TAIL 1
170
171 /* Sort operations */
172 #define REDIS_SORT_GET 0
173 #define REDIS_SORT_ASC 1
174 #define REDIS_SORT_DESC 2
175 #define REDIS_SORTKEY_MAX 1024
176
177 /* Log levels */
178 #define REDIS_DEBUG 0
179 #define REDIS_VERBOSE 1
180 #define REDIS_NOTICE 2
181 #define REDIS_WARNING 3
182
183 /* Anti-warning macro... */
184 #define REDIS_NOTUSED(V) ((void) V)
185
186 #define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^32 elements */
187 #define ZSKIPLIST_P 0.25 /* Skiplist P = 1/4 */
188
189 /* Append only defines */
190 #define APPENDFSYNC_NO 0
191 #define APPENDFSYNC_ALWAYS 1
192 #define APPENDFSYNC_EVERYSEC 2
193
194 /* Zip structure related defaults */
195 #define REDIS_HASH_MAX_ZIPMAP_ENTRIES 512
196 #define REDIS_HASH_MAX_ZIPMAP_VALUE 64
197 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 512
198 #define REDIS_LIST_MAX_ZIPLIST_VALUE 64
199 #define REDIS_SET_MAX_INTSET_ENTRIES 512
200
201 /* Sets operations codes */
202 #define REDIS_OP_UNION 0
203 #define REDIS_OP_DIFF 1
204 #define REDIS_OP_INTER 2
205
206 /* Redis maxmemory strategies */
207 #define REDIS_MAXMEMORY_VOLATILE_LRU 0
208 #define REDIS_MAXMEMORY_VOLATILE_TTL 1
209 #define REDIS_MAXMEMORY_VOLATILE_RANDOM 2
210 #define REDIS_MAXMEMORY_ALLKEYS_LRU 3
211 #define REDIS_MAXMEMORY_ALLKEYS_RANDOM 4
212 #define REDIS_MAXMEMORY_NO_EVICTION 5
213
214 /* Diskstore background saving thread states */
215 #define REDIS_BGSAVE_THREAD_UNACTIVE 0
216 #define REDIS_BGSAVE_THREAD_ACTIVE 1
217 #define REDIS_BGSAVE_THREAD_DONE_OK 2
218 #define REDIS_BGSAVE_THREAD_DONE_ERR 3
219
220 /* We can print the stacktrace, so our assert is defined this way: */
221 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
222 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
223 void _redisAssert(char *estr, char *file, int line);
224 void _redisPanic(char *msg, char *file, int line);
225
226 /*-----------------------------------------------------------------------------
227 * Data types
228 *----------------------------------------------------------------------------*/
229
230 /* A redis object, that is a type able to hold a string / list / set */
231
232 /* The actual Redis Object */
233 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
234 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
235 typedef struct redisObject {
236 unsigned type:4;
237 unsigned notused:2; /* Not used */
238 unsigned encoding:4;
239 unsigned lru:22; /* lru time (relative to server.lruclock) */
240 int refcount;
241 void *ptr;
242 /* VM fields are only allocated if VM is active, otherwise the
243 * object allocation function will just allocate
244 * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
245 * Redis without VM active will not have any overhead. */
246 } robj;
247
248 /* The VM pointer structure - identifies an object in the swap file.
249 *
250 * This object is stored in place of the value
251 * object in the main key->value hash table representing a database.
252 * Note that the first fields (type, storage) are the same as the redisObject
253 * structure so that vmPointer strucuters can be accessed even when casted
254 * as redisObject structures.
255 *
256 * This is useful as we don't know if a value object is or not on disk, but we
257 * are always able to read obj->storage to check this. For vmPointer
258 * structures "type" is set to REDIS_VMPOINTER (even if without this field
259 * is still possible to check the kind of object from the value of 'storage').*/
260 typedef struct vmPointer {
261 unsigned type:4;
262 unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
263 unsigned notused:26;
264 unsigned int vtype; /* type of the object stored in the swap file */
265 off_t page; /* the page at witch the object is stored on disk */
266 off_t usedpages; /* number of pages used on disk */
267 } vmpointer;
268
269 /* Macro used to initalize a Redis object allocated on the stack.
270 * Note that this macro is taken near the structure definition to make sure
271 * we'll update it when the structure is changed, to avoid bugs like
272 * bug #85 introduced exactly in this way. */
273 #define initStaticStringObject(_var,_ptr) do { \
274 _var.refcount = 1; \
275 _var.type = REDIS_STRING; \
276 _var.encoding = REDIS_ENCODING_RAW; \
277 _var.ptr = _ptr; \
278 } while(0);
279
280 typedef struct redisDb {
281 dict *dict; /* The keyspace for this DB */
282 dict *expires; /* Timeout of keys with a timeout set */
283 dict *blocking_keys; /* Keys with clients waiting for data (BLPOP) */
284 dict *io_keys; /* Keys with clients waiting for DS I/O */
285 dict *io_negcache; /* Negative caching for disk store */
286 dict *io_queued; /* Queued IO operations hash table */
287 dict *watched_keys; /* WATCHED keys for MULTI/EXEC CAS */
288 int id;
289 } redisDb;
290
291 /* Client MULTI/EXEC state */
292 typedef struct multiCmd {
293 robj **argv;
294 int argc;
295 struct redisCommand *cmd;
296 } multiCmd;
297
298 typedef struct multiState {
299 multiCmd *commands; /* Array of MULTI commands */
300 int count; /* Total number of MULTI commands */
301 } multiState;
302
303 typedef struct blockingState {
304 robj **keys; /* The key we are waiting to terminate a blocking
305 * operation such as BLPOP. Otherwise NULL. */
306 int count; /* Number of blocking keys */
307 time_t timeout; /* Blocking operation timeout. If UNIX current time
308 * is >= timeout then the operation timed out. */
309 robj *target; /* The key that should receive the element,
310 * for BRPOPLPUSH. */
311 } blockingState;
312
313 /* With multiplexing we need to take per-clinet state.
314 * Clients are taken in a liked list. */
315 typedef struct redisClient {
316 int fd;
317 redisDb *db;
318 int dictid;
319 sds querybuf;
320 int argc;
321 robj **argv;
322 int reqtype;
323 int multibulklen; /* number of multi bulk arguments left to read */
324 long bulklen; /* length of bulk argument in multi bulk request */
325 list *reply;
326 int sentlen;
327 time_t lastinteraction; /* time of the last interaction, used for timeout */
328 int flags; /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
329 int slaveseldb; /* slave selected db, if this client is a slave */
330 int authenticated; /* when requirepass is non-NULL */
331 int replstate; /* replication state if this is a slave */
332 int repldbfd; /* replication DB file descriptor */
333 long repldboff; /* replication DB file offset */
334 off_t repldbsize; /* replication DB file size */
335 multiState mstate; /* MULTI/EXEC state */
336 blockingState bpop; /* blocking state */
337 list *io_keys; /* Keys this client is waiting to be loaded from the
338 * swap file in order to continue. */
339 list *watched_keys; /* Keys WATCHED for MULTI/EXEC CAS */
340 dict *pubsub_channels; /* channels a client is interested in (SUBSCRIBE) */
341 list *pubsub_patterns; /* patterns a client is interested in (SUBSCRIBE) */
342
343 /* Response buffer */
344 int bufpos;
345 char buf[REDIS_REPLY_CHUNK_BYTES];
346 } redisClient;
347
348 struct saveparam {
349 time_t seconds;
350 int changes;
351 };
352
353 struct sharedObjectsStruct {
354 robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
355 *colon, *nullbulk, *nullmultibulk, *queued,
356 *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
357 *outofrangeerr, *loadingerr, *plus,
358 *select0, *select1, *select2, *select3, *select4,
359 *select5, *select6, *select7, *select8, *select9,
360 *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
361 *mbulk4, *psubscribebulk, *punsubscribebulk,
362 *integers[REDIS_SHARED_INTEGERS];
363 };
364
365 /* Global server state structure */
366 struct redisServer {
367 /* General */
368 pthread_t mainthread;
369 redisDb *db;
370 dict *commands; /* Command table hahs table */
371 aeEventLoop *el;
372 /* Networking */
373 int port;
374 char *bindaddr;
375 char *unixsocket;
376 int ipfd;
377 int sofd;
378 list *clients;
379 list *slaves, *monitors;
380 char neterr[ANET_ERR_LEN];
381 /* RDB / AOF loading information */
382 int loading;
383 off_t loading_total_bytes;
384 off_t loading_loaded_bytes;
385 time_t loading_start_time;
386 /* Fast pointers to often looked up command */
387 struct redisCommand *delCommand, *multiCommand;
388 int cronloops; /* number of times the cron function run */
389 time_t lastsave; /* Unix time of last save succeeede */
390 /* Fields used only for stats */
391 time_t stat_starttime; /* server start time */
392 long long stat_numcommands; /* number of processed commands */
393 long long stat_numconnections; /* number of connections received */
394 long long stat_expiredkeys; /* number of expired keys */
395 long long stat_evictedkeys; /* number of evicted keys (maxmemory) */
396 long long stat_keyspace_hits; /* number of successful lookups of keys */
397 long long stat_keyspace_misses; /* number of failed lookups of keys */
398 /* Configuration */
399 int verbosity;
400 int maxidletime;
401 int dbnum;
402 int daemonize;
403 int appendonly;
404 int appendfsync;
405 int no_appendfsync_on_rewrite;
406 int shutdown_asap;
407 int activerehashing;
408 char *requirepass;
409 /* Persistence */
410 long long dirty; /* changes to DB from the last save */
411 long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
412 time_t lastfsync;
413 int appendfd;
414 int appendseldb;
415 char *pidfile;
416 pid_t bgsavechildpid;
417 pid_t bgrewritechildpid;
418 int bgsavethread_state;
419 pthread_mutex_t bgsavethread_mutex;
420 pthread_t bgsavethread;
421 sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
422 sds aofbuf; /* AOF buffer, written before entering the event loop */
423 struct saveparam *saveparams;
424 int saveparamslen;
425 char *dbfilename;
426 int rdbcompression;
427 char *appendfilename;
428 /* Logging */
429 char *logfile;
430 int syslog_enabled;
431 char *syslog_ident;
432 int syslog_facility;
433 /* Replication related */
434 int isslave;
435 /* Slave specific fields */
436 char *masterauth;
437 char *masterhost;
438 int masterport;
439 redisClient *master; /* client that is master for this slave */
440 int replstate; /* replication status if the instance is a slave */
441 off_t repl_transfer_left; /* bytes left reading .rdb */
442 int repl_transfer_s; /* slave -> master SYNC socket */
443 int repl_transfer_fd; /* slave -> master SYNC temp file descriptor */
444 char *repl_transfer_tmpfile; /* slave-> master SYNC temp file name */
445 time_t repl_transfer_lastio; /* unix time of the latest read, for timeout */
446 int repl_serve_stale_data; /* Serve stale data when link is down? */
447 /* Limits */
448 unsigned int maxclients;
449 unsigned long long maxmemory;
450 int maxmemory_policy;
451 int maxmemory_samples;
452 /* Blocked clients */
453 unsigned int bpop_blocked_clients;
454 unsigned int cache_blocked_clients;
455 list *unblocked_clients; /* list of clients to unblock before next loop */
456 list *cache_io_queue; /* IO operations queue */
457 int cache_flush_delay; /* seconds to wait before flushing keys */
458 /* Sort parameters - qsort_r() is only available under BSD so we
459 * have to take this state global, in order to pass it to sortCompare() */
460 int sort_desc;
461 int sort_alpha;
462 int sort_bypattern;
463 /* Virtual memory configuration */
464 int ds_enabled; /* backend disk in redis.conf */
465 char *ds_path; /* location of the disk store on disk */
466 unsigned long long cache_max_memory;
467 /* Zip structure config */
468 size_t hash_max_zipmap_entries;
469 size_t hash_max_zipmap_value;
470 size_t list_max_ziplist_entries;
471 size_t list_max_ziplist_value;
472 size_t set_max_intset_entries;
473 time_t unixtime; /* Unix time sampled every second. */
474 /* Virtual memory I/O threads stuff */
475 /* An I/O thread process an element taken from the io_jobs queue and
476 * put the result of the operation in the io_done list. While the
477 * job is being processed, it's put on io_processing queue. */
478 list *io_newjobs; /* List of VM I/O jobs yet to be processed */
479 list *io_processing; /* List of VM I/O jobs being processed */
480 list *io_processed; /* List of VM I/O jobs already processed */
481 list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
482 pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
483 pthread_cond_t io_condvar; /* I/O threads conditional variable */
484 pthread_attr_t io_threads_attr; /* attributes for threads creation */
485 int io_active_threads; /* Number of running I/O threads */
486 int vm_max_threads; /* Max number of I/O threads running at the same time */
487 /* Our main thread is blocked on the event loop, locking for sockets ready
488 * to be read or written, so when a threaded I/O operation is ready to be
489 * processed by the main thread, the I/O thread will use a unix pipe to
490 * awake the main thread. The followings are the two pipe FDs. */
491 int io_ready_pipe_read;
492 int io_ready_pipe_write;
493 /* Virtual memory stats */
494 unsigned long long vm_stats_used_pages;
495 unsigned long long vm_stats_swapped_objects;
496 unsigned long long vm_stats_swapouts;
497 unsigned long long vm_stats_swapins;
498 /* Pubsub */
499 dict *pubsub_channels; /* Map channels to list of subscribed clients */
500 list *pubsub_patterns; /* A list of pubsub_patterns */
501 /* Misc */
502 unsigned lruclock:22; /* clock incrementing every minute, for LRU */
503 unsigned lruclock_padding:10;
504 };
505
506 typedef struct pubsubPattern {
507 redisClient *client;
508 robj *pattern;
509 } pubsubPattern;
510
511 typedef void redisCommandProc(redisClient *c);
512 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
513 struct redisCommand {
514 char *name;
515 redisCommandProc *proc;
516 int arity;
517 int flags;
518 /* Use a function to determine which keys need to be loaded
519 * in the background prior to executing this command. Takes precedence
520 * over vm_firstkey and others, ignored when NULL */
521 redisVmPreloadProc *vm_preload_proc;
522 /* What keys should be loaded in background when calling this command? */
523 int vm_firstkey; /* The first argument that's a key (0 = no keys) */
524 int vm_lastkey; /* THe last argument that's a key */
525 int vm_keystep; /* The step between first and last key */
526 long long microseconds, calls;
527 };
528
529 struct redisFunctionSym {
530 char *name;
531 unsigned long pointer;
532 };
533
534 typedef struct _redisSortObject {
535 robj *obj;
536 union {
537 double score;
538 robj *cmpobj;
539 } u;
540 } redisSortObject;
541
542 typedef struct _redisSortOperation {
543 int type;
544 robj *pattern;
545 } redisSortOperation;
546
547 /* ZSETs use a specialized version of Skiplists */
548 typedef struct zskiplistNode {
549 robj *obj;
550 double score;
551 struct zskiplistNode *backward;
552 struct zskiplistLevel {
553 struct zskiplistNode *forward;
554 unsigned int span;
555 } level[];
556 } zskiplistNode;
557
558 typedef struct zskiplist {
559 struct zskiplistNode *header, *tail;
560 unsigned long length;
561 int level;
562 } zskiplist;
563
564 typedef struct zset {
565 dict *dict;
566 zskiplist *zsl;
567 } zset;
568
569 /* DIsk store threaded I/O request message */
570 #define REDIS_IOJOB_LOAD 0
571 #define REDIS_IOJOB_SAVE 1
572
573 typedef struct iojob {
574 int type; /* Request type, REDIS_IOJOB_* */
575 redisDb *db;/* Redis database */
576 robj *key; /* This I/O request is about this key */
577 robj *val; /* the value to swap for REDIS_IOJOB_SAVE, otherwise this
578 * field is populated by the I/O thread for REDIS_IOJOB_LOAD. */
579 time_t expire; /* Expire time for this key on REDIS_IOJOB_LOAD */
580 } iojob;
581
582 /* IO operations scheduled -- check dscache.c for more info */
583 typedef struct ioop {
584 int type;
585 redisDb *db;
586 robj *key;
587 time_t ctime; /* This is the creation time of the entry. */
588 } ioop;
589
590 /* Structure to hold list iteration abstraction. */
591 typedef struct {
592 robj *subject;
593 unsigned char encoding;
594 unsigned char direction; /* Iteration direction */
595 unsigned char *zi;
596 listNode *ln;
597 } listTypeIterator;
598
599 /* Structure for an entry while iterating over a list. */
600 typedef struct {
601 listTypeIterator *li;
602 unsigned char *zi; /* Entry in ziplist */
603 listNode *ln; /* Entry in linked list */
604 } listTypeEntry;
605
606 /* Structure to hold set iteration abstraction. */
607 typedef struct {
608 robj *subject;
609 int encoding;
610 int ii; /* intset iterator */
611 dictIterator *di;
612 } setTypeIterator;
613
614 /* Structure to hold hash iteration abstration. Note that iteration over
615 * hashes involves both fields and values. Because it is possible that
616 * not both are required, store pointers in the iterator to avoid
617 * unnecessary memory allocation for fields/values. */
618 typedef struct {
619 int encoding;
620 unsigned char *zi;
621 unsigned char *zk, *zv;
622 unsigned int zklen, zvlen;
623
624 dictIterator *di;
625 dictEntry *de;
626 } hashTypeIterator;
627
628 #define REDIS_HASH_KEY 1
629 #define REDIS_HASH_VALUE 2
630
631 /*-----------------------------------------------------------------------------
632 * Extern declarations
633 *----------------------------------------------------------------------------*/
634
635 extern struct redisServer server;
636 extern struct sharedObjectsStruct shared;
637 extern dictType setDictType;
638 extern dictType zsetDictType;
639 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
640 dictType hashDictType;
641
642 /*-----------------------------------------------------------------------------
643 * Functions prototypes
644 *----------------------------------------------------------------------------*/
645
646 /* Utils */
647 long long ustime(void);
648
649 /* networking.c -- Networking and Client related operations */
650 redisClient *createClient(int fd);
651 void closeTimedoutClients(void);
652 void freeClient(redisClient *c);
653 void resetClient(redisClient *c);
654 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
655 void addReply(redisClient *c, robj *obj);
656 void *addDeferredMultiBulkLength(redisClient *c);
657 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
658 void addReplySds(redisClient *c, sds s);
659 void processInputBuffer(redisClient *c);
660 void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask);
661 void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask);
662 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
663 void addReplyBulk(redisClient *c, robj *obj);
664 void addReplyBulkCString(redisClient *c, char *s);
665 void addReplyBulkCBuffer(redisClient *c, void *p, size_t len);
666 void addReplyBulkLongLong(redisClient *c, long long ll);
667 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
668 void addReply(redisClient *c, robj *obj);
669 void addReplySds(redisClient *c, sds s);
670 void addReplyError(redisClient *c, char *err);
671 void addReplyStatus(redisClient *c, char *status);
672 void addReplyDouble(redisClient *c, double d);
673 void addReplyLongLong(redisClient *c, long long ll);
674 void addReplyMultiBulkLen(redisClient *c, long length);
675 void *dupClientReplyValue(void *o);
676 void getClientsMaxBuffers(unsigned long *longest_output_list,
677 unsigned long *biggest_input_buffer);
678
679 #ifdef __GNUC__
680 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
681 __attribute__((format(printf, 2, 3)));
682 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
683 __attribute__((format(printf, 2, 3)));
684 #else
685 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
686 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
687 #endif
688
689 /* List data type */
690 void listTypeTryConversion(robj *subject, robj *value);
691 void listTypePush(robj *subject, robj *value, int where);
692 robj *listTypePop(robj *subject, int where);
693 unsigned long listTypeLength(robj *subject);
694 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
695 void listTypeReleaseIterator(listTypeIterator *li);
696 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
697 robj *listTypeGet(listTypeEntry *entry);
698 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
699 int listTypeEqual(listTypeEntry *entry, robj *o);
700 void listTypeDelete(listTypeEntry *entry);
701 void listTypeConvert(robj *subject, int enc);
702 void unblockClientWaitingData(redisClient *c);
703 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
704 void popGenericCommand(redisClient *c, int where);
705
706 /* MULTI/EXEC/WATCH... */
707 void unwatchAllKeys(redisClient *c);
708 void initClientMultiState(redisClient *c);
709 void freeClientMultiState(redisClient *c);
710 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
711 void touchWatchedKey(redisDb *db, robj *key);
712 void touchWatchedKeysOnFlush(int dbid);
713
714 /* Redis object implementation */
715 void decrRefCount(void *o);
716 void incrRefCount(robj *o);
717 void freeStringObject(robj *o);
718 void freeListObject(robj *o);
719 void freeSetObject(robj *o);
720 void freeZsetObject(robj *o);
721 void freeHashObject(robj *o);
722 robj *createObject(int type, void *ptr);
723 robj *createStringObject(char *ptr, size_t len);
724 robj *dupStringObject(robj *o);
725 robj *tryObjectEncoding(robj *o);
726 robj *getDecodedObject(robj *o);
727 size_t stringObjectLen(robj *o);
728 robj *createStringObjectFromLongLong(long long value);
729 robj *createListObject(void);
730 robj *createZiplistObject(void);
731 robj *createSetObject(void);
732 robj *createIntsetObject(void);
733 robj *createHashObject(void);
734 robj *createZsetObject(void);
735 robj *createZsetZiplistObject(void);
736 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
737 int checkType(redisClient *c, robj *o, int type);
738 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
739 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
740 int getLongLongFromObject(robj *o, long long *target);
741 char *strEncoding(int encoding);
742 int compareStringObjects(robj *a, robj *b);
743 int equalStringObjects(robj *a, robj *b);
744 unsigned long estimateObjectIdleTime(robj *o);
745
746 /* Synchronous I/O with timeout */
747 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
748 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
749 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
750 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
751 int fwriteBulkDouble(FILE *fp, double d);
752 int fwriteBulkLongLong(FILE *fp, long long l);
753 int fwriteBulkObject(FILE *fp, robj *obj);
754
755 /* Replication */
756 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
757 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
758 int syncWithMaster(void);
759 void updateSlavesWaitingBgsave(int bgsaveerr);
760 void replicationCron(void);
761
762 /* Generic persistence functions */
763 void startLoading(FILE *fp);
764 void loadingProgress(off_t pos);
765 void stopLoading(void);
766
767 /* RDB persistence */
768 int rdbLoad(char *filename);
769 int rdbSaveBackground(char *filename);
770 void rdbRemoveTempFile(pid_t childpid);
771 int rdbSave(char *filename);
772 int rdbSaveObject(FILE *fp, robj *o);
773 off_t rdbSavedObjectLen(robj *o);
774 off_t rdbSavedObjectPages(robj *o);
775 robj *rdbLoadObject(int type, FILE *fp);
776 void backgroundSaveDoneHandler(int exitcode, int bysignal);
777 int rdbSaveKeyValuePair(FILE *fp, robj *key, robj *val, time_t expireitme, time_t now);
778 int rdbLoadType(FILE *fp);
779 time_t rdbLoadTime(FILE *fp);
780 robj *rdbLoadStringObject(FILE *fp);
781 int rdbSaveType(FILE *fp, unsigned char type);
782 int rdbSaveLen(FILE *fp, uint32_t len);
783
784 /* AOF persistence */
785 void flushAppendOnlyFile(void);
786 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
787 void aofRemoveTempFile(pid_t childpid);
788 int rewriteAppendOnlyFileBackground(void);
789 int loadAppendOnlyFile(char *filename);
790 void stopAppendOnly(void);
791 int startAppendOnly(void);
792 void backgroundRewriteDoneHandler(int exitcode, int bysignal);
793
794 /* Sorted sets data type */
795 zskiplist *zslCreate(void);
796 void zslFree(zskiplist *zsl);
797 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
798
799 /* Core functions */
800 void freeMemoryIfNeeded(void);
801 int processCommand(redisClient *c);
802 void setupSignalHandlers(void);
803 struct redisCommand *lookupCommand(sds name);
804 struct redisCommand *lookupCommandByCString(char *s);
805 void call(redisClient *c, struct redisCommand *cmd);
806 int prepareForShutdown();
807 void redisLog(int level, const char *fmt, ...);
808 void usage();
809 void updateDictResizePolicy(void);
810 int htNeedsResize(dict *dict);
811 void oom(const char *msg);
812 void populateCommandTable(void);
813 void resetCommandTableStats(void);
814
815 /* Disk store */
816 int dsOpen(void);
817 int dsClose(void);
818 int dsSet(redisDb *db, robj *key, robj *val, time_t expire);
819 robj *dsGet(redisDb *db, robj *key, time_t *expire);
820 int dsDel(redisDb *db, robj *key);
821 int dsExists(redisDb *db, robj *key);
822 void dsFlushDb(int dbid);
823 int dsRdbSaveBackground(char *filename);
824 int dsRdbSave(char *filename);
825
826 /* Disk Store Cache */
827 void dsInit(void);
828 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
829 void lockThreadedIO(void);
830 void unlockThreadedIO(void);
831 void freeIOJob(iojob *j);
832 void queueIOJob(iojob *j);
833 void waitEmptyIOJobsQueue(void);
834 void processAllPendingIOJobs(void);
835 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
836 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
837 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
838 int dontWaitForSwappedKey(redisClient *c, robj *key);
839 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
840 int cacheFreeOneEntry(void);
841 void cacheScheduleIOAddFlag(redisDb *db, robj *key, long flag);
842 void cacheScheduleIODelFlag(redisDb *db, robj *key, long flag);
843 int cacheScheduleIOGetFlags(redisDb *db, robj *key);
844 void cacheScheduleIO(redisDb *db, robj *key, int type);
845 void cacheCron(void);
846 int cacheKeyMayExist(redisDb *db, robj *key);
847 void cacheSetKeyMayExist(redisDb *db, robj *key);
848 void cacheSetKeyDoesNotExist(redisDb *db, robj *key);
849 void cacheForcePointInTime(void);
850
851 /* Set data type */
852 robj *setTypeCreate(robj *value);
853 int setTypeAdd(robj *subject, robj *value);
854 int setTypeRemove(robj *subject, robj *value);
855 int setTypeIsMember(robj *subject, robj *value);
856 setTypeIterator *setTypeInitIterator(robj *subject);
857 void setTypeReleaseIterator(setTypeIterator *si);
858 int setTypeNext(setTypeIterator *si, robj **objele, int64_t *llele);
859 robj *setTypeNextObject(setTypeIterator *si);
860 int setTypeRandomElement(robj *setobj, robj **objele, int64_t *llele);
861 unsigned long setTypeSize(robj *subject);
862 void setTypeConvert(robj *subject, int enc);
863
864 /* Hash data type */
865 void convertToRealHash(robj *o);
866 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
867 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
868 int hashTypeGet(robj *o, robj *key, robj **objval, unsigned char **v, unsigned int *vlen);
869 robj *hashTypeGetObject(robj *o, robj *key);
870 int hashTypeExists(robj *o, robj *key);
871 int hashTypeSet(robj *o, robj *key, robj *value);
872 int hashTypeDelete(robj *o, robj *key);
873 unsigned long hashTypeLength(robj *o);
874 hashTypeIterator *hashTypeInitIterator(robj *subject);
875 void hashTypeReleaseIterator(hashTypeIterator *hi);
876 int hashTypeNext(hashTypeIterator *hi);
877 int hashTypeCurrent(hashTypeIterator *hi, int what, robj **objval, unsigned char **v, unsigned int *vlen);
878 robj *hashTypeCurrentObject(hashTypeIterator *hi, int what);
879 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
880
881 /* Pub / Sub */
882 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
883 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
884 void freePubsubPattern(void *p);
885 int listMatchPubsubPattern(void *a, void *b);
886
887 /* Utility functions */
888 int stringmatchlen(const char *pattern, int patternLen,
889 const char *string, int stringLen, int nocase);
890 int stringmatch(const char *pattern, const char *string, int nocase);
891 long long memtoll(const char *p, int *err);
892 int ll2string(char *s, size_t len, long long value);
893 int d2string(char *s, size_t len, double value);
894 int isStringRepresentableAsLong(sds s, long *longval);
895 int isStringRepresentableAsLongLong(sds s, long long *longval);
896 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
897
898 /* Configuration */
899 void loadServerConfig(char *filename);
900 void appendServerSaveParams(time_t seconds, int changes);
901 void resetServerSaveParams();
902
903 /* db.c -- Keyspace access API */
904 int removeExpire(redisDb *db, robj *key);
905 void propagateExpire(redisDb *db, robj *key);
906 int expireIfNeeded(redisDb *db, robj *key);
907 time_t getExpire(redisDb *db, robj *key);
908 void setExpire(redisDb *db, robj *key, time_t when);
909 robj *lookupKey(redisDb *db, robj *key);
910 robj *lookupKeyRead(redisDb *db, robj *key);
911 robj *lookupKeyWrite(redisDb *db, robj *key);
912 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
913 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
914 int dbAdd(redisDb *db, robj *key, robj *val);
915 int dbReplace(redisDb *db, robj *key, robj *val);
916 int dbExists(redisDb *db, robj *key);
917 robj *dbRandomKey(redisDb *db);
918 int dbDelete(redisDb *db, robj *key);
919 long long emptyDb();
920 int selectDb(redisClient *c, int id);
921 void signalModifiedKey(redisDb *db, robj *key);
922 void signalFlushedDb(int dbid);
923
924 /* Git SHA1 */
925 char *redisGitSHA1(void);
926 char *redisGitDirty(void);
927
928 /* Commands prototypes */
929 void authCommand(redisClient *c);
930 void pingCommand(redisClient *c);
931 void echoCommand(redisClient *c);
932 void setCommand(redisClient *c);
933 void setnxCommand(redisClient *c);
934 void setexCommand(redisClient *c);
935 void getCommand(redisClient *c);
936 void delCommand(redisClient *c);
937 void existsCommand(redisClient *c);
938 void setbitCommand(redisClient *c);
939 void getbitCommand(redisClient *c);
940 void setrangeCommand(redisClient *c);
941 void getrangeCommand(redisClient *c);
942 void incrCommand(redisClient *c);
943 void decrCommand(redisClient *c);
944 void incrbyCommand(redisClient *c);
945 void decrbyCommand(redisClient *c);
946 void selectCommand(redisClient *c);
947 void randomkeyCommand(redisClient *c);
948 void keysCommand(redisClient *c);
949 void dbsizeCommand(redisClient *c);
950 void lastsaveCommand(redisClient *c);
951 void saveCommand(redisClient *c);
952 void bgsaveCommand(redisClient *c);
953 void bgrewriteaofCommand(redisClient *c);
954 void shutdownCommand(redisClient *c);
955 void moveCommand(redisClient *c);
956 void renameCommand(redisClient *c);
957 void renamenxCommand(redisClient *c);
958 void lpushCommand(redisClient *c);
959 void rpushCommand(redisClient *c);
960 void lpushxCommand(redisClient *c);
961 void rpushxCommand(redisClient *c);
962 void linsertCommand(redisClient *c);
963 void lpopCommand(redisClient *c);
964 void rpopCommand(redisClient *c);
965 void llenCommand(redisClient *c);
966 void lindexCommand(redisClient *c);
967 void lrangeCommand(redisClient *c);
968 void ltrimCommand(redisClient *c);
969 void typeCommand(redisClient *c);
970 void lsetCommand(redisClient *c);
971 void saddCommand(redisClient *c);
972 void sremCommand(redisClient *c);
973 void smoveCommand(redisClient *c);
974 void sismemberCommand(redisClient *c);
975 void scardCommand(redisClient *c);
976 void spopCommand(redisClient *c);
977 void srandmemberCommand(redisClient *c);
978 void sinterCommand(redisClient *c);
979 void sinterstoreCommand(redisClient *c);
980 void sunionCommand(redisClient *c);
981 void sunionstoreCommand(redisClient *c);
982 void sdiffCommand(redisClient *c);
983 void sdiffstoreCommand(redisClient *c);
984 void syncCommand(redisClient *c);
985 void flushdbCommand(redisClient *c);
986 void flushallCommand(redisClient *c);
987 void sortCommand(redisClient *c);
988 void lremCommand(redisClient *c);
989 void rpoplpushCommand(redisClient *c);
990 void infoCommand(redisClient *c);
991 void mgetCommand(redisClient *c);
992 void monitorCommand(redisClient *c);
993 void expireCommand(redisClient *c);
994 void expireatCommand(redisClient *c);
995 void getsetCommand(redisClient *c);
996 void ttlCommand(redisClient *c);
997 void persistCommand(redisClient *c);
998 void slaveofCommand(redisClient *c);
999 void debugCommand(redisClient *c);
1000 void msetCommand(redisClient *c);
1001 void msetnxCommand(redisClient *c);
1002 void zaddCommand(redisClient *c);
1003 void zincrbyCommand(redisClient *c);
1004 void zrangeCommand(redisClient *c);
1005 void zrangebyscoreCommand(redisClient *c);
1006 void zrevrangebyscoreCommand(redisClient *c);
1007 void zcountCommand(redisClient *c);
1008 void zrevrangeCommand(redisClient *c);
1009 void zcardCommand(redisClient *c);
1010 void zremCommand(redisClient *c);
1011 void zscoreCommand(redisClient *c);
1012 void zremrangebyscoreCommand(redisClient *c);
1013 void multiCommand(redisClient *c);
1014 void execCommand(redisClient *c);
1015 void discardCommand(redisClient *c);
1016 void blpopCommand(redisClient *c);
1017 void brpopCommand(redisClient *c);
1018 void brpoplpushCommand(redisClient *c);
1019 void appendCommand(redisClient *c);
1020 void strlenCommand(redisClient *c);
1021 void zrankCommand(redisClient *c);
1022 void zrevrankCommand(redisClient *c);
1023 void hsetCommand(redisClient *c);
1024 void hsetnxCommand(redisClient *c);
1025 void hgetCommand(redisClient *c);
1026 void hmsetCommand(redisClient *c);
1027 void hmgetCommand(redisClient *c);
1028 void hdelCommand(redisClient *c);
1029 void hlenCommand(redisClient *c);
1030 void zremrangebyrankCommand(redisClient *c);
1031 void zunionstoreCommand(redisClient *c);
1032 void zinterstoreCommand(redisClient *c);
1033 void hkeysCommand(redisClient *c);
1034 void hvalsCommand(redisClient *c);
1035 void hgetallCommand(redisClient *c);
1036 void hexistsCommand(redisClient *c);
1037 void configCommand(redisClient *c);
1038 void hincrbyCommand(redisClient *c);
1039 void subscribeCommand(redisClient *c);
1040 void unsubscribeCommand(redisClient *c);
1041 void psubscribeCommand(redisClient *c);
1042 void punsubscribeCommand(redisClient *c);
1043 void publishCommand(redisClient *c);
1044 void watchCommand(redisClient *c);
1045 void unwatchCommand(redisClient *c);
1046
1047 #if defined(__GNUC__)
1048 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
1049 void free(void *ptr) __attribute__ ((deprecated));
1050 void *malloc(size_t size) __attribute__ ((deprecated));
1051 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
1052 #endif
1053
1054 #endif