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