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