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
  21 #include "ae.h"     /* Event driven programming library */
  22 #include "sds.h"    /* Dynamic safe strings */
  23 #include "dict.h"   /* Hash tables */
  24 #include "adlist.h" /* Linked lists */
  25 #include "zmalloc.h" /* total memory usage aware version of malloc/free */
  26 #include "anet.h"   /* Networking the easy way */
  27 #include "zipmap.h" /* Compact string -> string data structure */
  28 #include "ziplist.h" /* Compact list data structure */
  29 #include "intset.h" /* Compact integer set structure */
  30 #include "version.h"
  31
  32 /* Error codes */
  33 #define REDIS_OK                0
  34 #define REDIS_ERR               -1
  35
  36 /* Static server configuration */
  37 #define REDIS_SERVERPORT        6379    /* TCP port */
  38 #define REDIS_MAXIDLETIME       (60*5)  /* default client timeout */
  39 #define REDIS_IOBUF_LEN         1024
  40 #define REDIS_LOADBUF_LEN       1024
  41 #define REDIS_STATIC_ARGS       8
  42 #define REDIS_DEFAULT_DBNUM     16
  43 #define REDIS_CONFIGLINE_MAX    1024
  44 #define REDIS_MAX_SYNC_TIME     60      /* Slave can't take more to sync */
  45 #define REDIS_EXPIRELOOKUPS_PER_CRON    10 /* lookup 10 expires per loop */
  46 #define REDIS_MAX_WRITE_PER_EVENT (1024*64)
  47 #define REDIS_REQUEST_MAX_SIZE (1024*1024*256) /* max bytes in inline command */
  48 #define REDIS_SHARED_INTEGERS 10000
  49 #define REDIS_REPLY_CHUNK_BYTES (5*1500) /* 5 TCP packets with default MTU */
  50
  51 /* If more then REDIS_WRITEV_THRESHOLD write packets are pending use writev */
  52 #define REDIS_WRITEV_THRESHOLD      3
  53 /* Max number of iovecs used for each writev call */
  54 #define REDIS_WRITEV_IOVEC_COUNT    256
  55
  56 /* Hash table parameters */
  57 #define REDIS_HT_MINFILL        10      /* Minimal hash table fill 10% */
  58
  59 /* Command flags:
  60  *   REDIS_CMD_DENYOOM:
  61  *     Commands marked with this flag will return an error when 'maxmemory' is
  62  *     set and the server is using more than 'maxmemory' bytes of memory.
  63  *     In short: commands with this flag are denied on low memory conditions.
  64  *   REDIS_CMD_FORCE_REPLICATION:
  65  *     Force replication even if dirty is 0. */
  66 #define REDIS_CMD_DENYOOM 4
  67 #define REDIS_CMD_FORCE_REPLICATION 8
  68
  69 /* Object types */
  70 #define REDIS_STRING 0
  71 #define REDIS_LIST 1
  72 #define REDIS_SET 2
  73 #define REDIS_ZSET 3
  74 #define REDIS_HASH 4
  75 #define REDIS_VMPOINTER 8
  76
  77 /* Objects encoding. Some kind of objects like Strings and Hashes can be
  78  * internally represented in multiple ways. The 'encoding' field of the object
  79  * is set to one of this fields for this object. */
  80 #define REDIS_ENCODING_RAW 0     /* Raw representation */
  81 #define REDIS_ENCODING_INT 1     /* Encoded as integer */
  82 #define REDIS_ENCODING_HT 2      /* Encoded as hash table */
  83 #define REDIS_ENCODING_ZIPMAP 3  /* Encoded as zipmap */
  84 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
  85 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
  86 #define REDIS_ENCODING_INTSET 6  /* Encoded as intset */
  87
  88 /* Object types only used for dumping to disk */
  89 #define REDIS_EXPIRETIME 253
  90 #define REDIS_SELECTDB 254
  91 #define REDIS_EOF 255
  92
  93 /* Defines related to the dump file format. To store 32 bits lengths for short
  94  * keys requires a lot of space, so we check the most significant 2 bits of
  95  * the first byte to interpreter the length:
  96  *
  97  * 00|000000 => if the two MSB are 00 the len is the 6 bits of this byte
  98  * 01|000000 00000000 =>  01, the len is 14 byes, 6 bits + 8 bits of next byte
  99  * 10|000000 [32 bit integer] => if it's 01, a full 32 bit len will follow
 100  * 11|000000 this means: specially encoded object will follow. The six bits
 101  *           number specify the kind of object that follows.
 102  *           See the REDIS_RDB_ENC_* defines.
 103  *
 104  * Lenghts up to 63 are stored using a single byte, most DB keys, and may
 105  * values, will fit inside. */
 106 #define REDIS_RDB_6BITLEN 0
 107 #define REDIS_RDB_14BITLEN 1
 108 #define REDIS_RDB_32BITLEN 2
 109 #define REDIS_RDB_ENCVAL 3
 110 #define REDIS_RDB_LENERR UINT_MAX
 111
 112 /* When a length of a string object stored on disk has the first two bits
 113  * set, the remaining two bits specify a special encoding for the object
 114  * accordingly to the following defines: */
 115 #define REDIS_RDB_ENC_INT8 0        /* 8 bit signed integer */
 116 #define REDIS_RDB_ENC_INT16 1       /* 16 bit signed integer */
 117 #define REDIS_RDB_ENC_INT32 2       /* 32 bit signed integer */
 118 #define REDIS_RDB_ENC_LZF 3         /* string compressed with FASTLZ */
 119
 120 /* Virtual memory object->where field. */
 121 #define REDIS_VM_MEMORY 0       /* The object is on memory */
 122 #define REDIS_VM_SWAPPED 1      /* The object is on disk */
 123 #define REDIS_VM_SWAPPING 2     /* Redis is swapping this object on disk */
 124 #define REDIS_VM_LOADING 3      /* Redis is loading this object from disk */
 125
 126 /* Virtual memory static configuration stuff.
 127  * Check vmFindContiguousPages() to know more about this magic numbers. */
 128 #define REDIS_VM_MAX_NEAR_PAGES 65536
 129 #define REDIS_VM_MAX_RANDOM_JUMP 4096
 130 #define REDIS_VM_MAX_THREADS 32
 131 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
 132 /* The following is the *percentage* of completed I/O jobs to process when the
 133  * handelr is called. While Virtual Memory I/O operations are performed by
 134  * threads, this operations must be processed by the main thread when completed
 135  * in order to take effect. */
 136 #define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1
 137
 138 /* Client flags */
 139 #define REDIS_SLAVE 1       /* This client is a slave server */
 140 #define REDIS_MASTER 2      /* This client is a master server */
 141 #define REDIS_MONITOR 4     /* This client is a slave monitor, see MONITOR */
 142 #define REDIS_MULTI 8       /* This client is in a MULTI context */
 143 #define REDIS_BLOCKED 16    /* The client is waiting in a blocking operation */
 144 #define REDIS_IO_WAIT 32    /* The client is waiting for Virtual Memory I/O */
 145 #define REDIS_DIRTY_CAS 64  /* Watched keys modified. EXEC will fail. */
 146 #define REDIS_CLOSE_AFTER_REPLY 128 /* Close after writing entire reply. */
 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 64
 196 #define REDIS_HASH_MAX_ZIPMAP_VALUE 512
 197 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 1024
 198 #define REDIS_LIST_MAX_ZIPLIST_VALUE 32
 199 #define REDIS_SET_MAX_INTSET_ENTRIES 4096
 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 /* We can print the stacktrace, so our assert is defined this way: */
 215 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
 216 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
 217 void _redisAssert(char *estr, char *file, int line);
 218 void _redisPanic(char *msg, char *file, int line);
 219
 220 /*-----------------------------------------------------------------------------
 221  * Data types
 222  *----------------------------------------------------------------------------*/
 223
 224 /* A redis object, that is a type able to hold a string / list / set */
 225
 226 /* The actual Redis Object */
 227 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
 228 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
 229 typedef struct redisObject {
 230     unsigned type:4;
 231     unsigned storage:2;     /* REDIS_VM_MEMORY or REDIS_VM_SWAPPING */
 232     unsigned encoding:4;
 233     unsigned lru:22;        /* lru time (relative to server.lruclock) */
 234     int refcount;
 235     void *ptr;
 236     /* VM fields are only allocated if VM is active, otherwise the
 237      * object allocation function will just allocate
 238      * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
 239      * Redis without VM active will not have any overhead. */
 240 } robj;
 241
 242 /* The VM pointer structure - identifies an object in the swap file.
 243  *
 244  * This object is stored in place of the value
 245  * object in the main key->value hash table representing a database.
 246  * Note that the first fields (type, storage) are the same as the redisObject
 247  * structure so that vmPointer strucuters can be accessed even when casted
 248  * as redisObject structures.
 249  *
 250  * This is useful as we don't know if a value object is or not on disk, but we
 251  * are always able to read obj->storage to check this. For vmPointer
 252  * structures "type" is set to REDIS_VMPOINTER (even if without this field
 253  * is still possible to check the kind of object from the value of 'storage').*/
 254 typedef struct vmPointer {
 255     unsigned type:4;
 256     unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
 257     unsigned notused:26;
 258     unsigned int vtype; /* type of the object stored in the swap file */
 259     off_t page;         /* the page at witch the object is stored on disk */
 260     off_t usedpages;    /* number of pages used on disk */
 261 } vmpointer;
 262
 263 /* Macro used to initalize a Redis object allocated on the stack.
 264  * Note that this macro is taken near the structure definition to make sure
 265  * we'll update it when the structure is changed, to avoid bugs like
 266  * bug #85 introduced exactly in this way. */
 267 #define initStaticStringObject(_var,_ptr) do { \
 268     _var.refcount = 1; \
 269     _var.type = REDIS_STRING; \
 270     _var.encoding = REDIS_ENCODING_RAW; \
 271     _var.ptr = _ptr; \
 272     _var.storage = REDIS_VM_MEMORY; \
 273 } while(0);
 274
 275 typedef struct redisDb {
 276     dict *dict;                 /* The keyspace for this DB */
 277     dict *expires;              /* Timeout of keys with a timeout set */
 278     dict *blocking_keys;        /* Keys with clients waiting for data (BLPOP) */
 279     dict *io_keys;              /* Keys with clients waiting for VM I/O */
 280     dict *watched_keys;         /* WATCHED keys for MULTI/EXEC CAS */
 281     int id;
 282 } redisDb;
 283
 284 /* Client MULTI/EXEC state */
 285 typedef struct multiCmd {
 286     robj **argv;
 287     int argc;
 288     struct redisCommand *cmd;
 289 } multiCmd;
 290
 291 typedef struct multiState {
 292     multiCmd *commands;     /* Array of MULTI commands */
 293     int count;              /* Total number of MULTI commands */
 294 } multiState;
 295
 296 /* With multiplexing we need to take per-clinet state.
 297  * Clients are taken in a liked list. */
 298 typedef struct redisClient {
 299     int fd;
 300     redisDb *db;
 301     int dictid;
 302     sds querybuf;
 303     int argc;
 304     robj **argv;
 305     int reqtype;
 306     int multibulklen;       /* number of multi bulk arguments left to read */
 307     long bulklen;           /* length of bulk argument in multi bulk request */
 308     list *reply;
 309     int sentlen;
 310     time_t lastinteraction; /* time of the last interaction, used for timeout */
 311     int flags;              /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
 312     int slaveseldb;         /* slave selected db, if this client is a slave */
 313     int authenticated;      /* when requirepass is non-NULL */
 314     int replstate;          /* replication state if this is a slave */
 315     int repldbfd;           /* replication DB file descriptor */
 316     long repldboff;         /* replication DB file offset */
 317     off_t repldbsize;       /* replication DB file size */
 318     multiState mstate;      /* MULTI/EXEC state */
 319     robj **blocking_keys;   /* The key we are waiting to terminate a blocking
 320                              * operation such as BLPOP. Otherwise NULL. */
 321     int blocking_keys_num;  /* Number of blocking keys */
 322     time_t blockingto;      /* Blocking operation timeout. If UNIX current time
 323                              * is >= blockingto then the operation timed out. */
 324     list *io_keys;          /* Keys this client is waiting to be loaded from the
 325                              * swap file in order to continue. */
 326     list *watched_keys;     /* Keys WATCHED for MULTI/EXEC CAS */
 327     dict *pubsub_channels;  /* channels a client is interested in (SUBSCRIBE) */
 328     list *pubsub_patterns;  /* patterns a client is interested in (SUBSCRIBE) */
 329
 330     /* Response buffer */
 331     int bufpos;
 332     char buf[REDIS_REPLY_CHUNK_BYTES];
 333 } redisClient;
 334
 335 struct saveparam {
 336     time_t seconds;
 337     int changes;
 338 };
 339
 340 struct sharedObjectsStruct {
 341     robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
 342     *colon, *nullbulk, *nullmultibulk, *queued,
 343     *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
 344     *outofrangeerr, *loadingerr, *plus,
 345     *select0, *select1, *select2, *select3, *select4,
 346     *select5, *select6, *select7, *select8, *select9,
 347     *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
 348     *mbulk4, *psubscribebulk, *punsubscribebulk,
 349     *integers[REDIS_SHARED_INTEGERS];
 350 };
 351
 352 /* Global server state structure */
 353 struct redisServer {
 354     pthread_t mainthread;
 355     int port;
 356     char *bindaddr;
 357     char *unixsocket;
 358     int ipfd;
 359     int sofd;
 360     redisDb *db;
 361     long long dirty;            /* changes to DB from the last save */
 362     long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
 363     list *clients;
 364     dict *commands;             /* Command table hahs table */
 365     /* RDB / AOF loading information */
 366     int loading;
 367     off_t loading_total_bytes;
 368     off_t loading_loaded_bytes;
 369     time_t loading_start_time;
 370     /* Fast pointers to often looked up command */
 371     struct redisCommand *delCommand, *multiCommand;
 372     list *slaves, *monitors;
 373     char neterr[ANET_ERR_LEN];
 374     aeEventLoop *el;
 375     int cronloops;              /* number of times the cron function run */
 376     time_t lastsave;                /* Unix time of last save succeeede */
 377     /* Fields used only for stats */
 378     time_t stat_starttime;          /* server start time */
 379     long long stat_numcommands;     /* number of processed commands */
 380     long long stat_numconnections;  /* number of connections received */
 381     long long stat_expiredkeys;     /* number of expired keys */
 382     long long stat_keyspace_hits;   /* number of successful lookups of keys */
 383     long long stat_keyspace_misses; /* number of failed lookups of keys */
 384     /* Configuration */
 385     int verbosity;
 386     int glueoutputbuf;
 387     int maxidletime;
 388     int dbnum;
 389     int daemonize;
 390     int appendonly;
 391     int appendfsync;
 392     int no_appendfsync_on_rewrite;
 393     int shutdown_asap;
 394     time_t lastfsync;
 395     int appendfd;
 396     int appendseldb;
 397     char *pidfile;
 398     pid_t bgsavechildpid;
 399     pid_t bgrewritechildpid;
 400     sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
 401     sds aofbuf;       /* AOF buffer, written before entering the event loop */
 402     struct saveparam *saveparams;
 403     int saveparamslen;
 404     char *logfile;
 405     char *dbfilename;
 406     char *appendfilename;
 407     char *requirepass;
 408     int rdbcompression;
 409     int activerehashing;
 410     /* Replication related */
 411     int isslave;
 412     /* Slave specific fields */
 413     char *masterauth;
 414     char *masterhost;
 415     int masterport;
 416     redisClient *master;    /* client that is master for this slave */
 417     int replstate;          /* replication status if the instance is a slave */
 418     off_t repl_transfer_left;  /* bytes left reading .rdb  */
 419     int repl_transfer_s;    /* slave -> master SYNC socket */
 420     int repl_transfer_fd;   /* slave -> master SYNC temp file descriptor */
 421     char *repl_transfer_tmpfile; /* slave-> master SYNC temp file name */
 422     time_t repl_transfer_lastio; /* unix time of the latest read, for timeout */
 423     int repl_serve_stale_data; /* Serve stale data when link is down? */
 424     /* Limits */
 425     unsigned int maxclients;
 426     unsigned long long maxmemory;
 427     int maxmemory_policy;
 428     int maxmemory_samples;
 429     /* Blocked clients */
 430     unsigned int blpop_blocked_clients;
 431     unsigned int vm_blocked_clients;
 432     /* Sort parameters - qsort_r() is only available under BSD so we
 433      * have to take this state global, in order to pass it to sortCompare() */
 434     int sort_desc;
 435     int sort_alpha;
 436     int sort_bypattern;
 437     /* Virtual memory configuration */
 438     int vm_enabled;
 439     char *vm_swap_file;
 440     off_t vm_page_size;
 441     off_t vm_pages;
 442     unsigned long long vm_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     /* Virtual memory state */
 450     FILE *vm_fp;
 451     int vm_fd;
 452     off_t vm_next_page; /* Next probably empty page */
 453     off_t vm_near_pages; /* Number of pages allocated sequentially */
 454     unsigned char *vm_bitmap; /* Bitmap of free/used pages */
 455     time_t unixtime;    /* Unix time sampled every second. */
 456     /* Virtual memory I/O threads stuff */
 457     /* An I/O thread process an element taken from the io_jobs queue and
 458      * put the result of the operation in the io_done list. While the
 459      * job is being processed, it's put on io_processing queue. */
 460     list *io_newjobs; /* List of VM I/O jobs yet to be processed */
 461     list *io_processing; /* List of VM I/O jobs being processed */
 462     list *io_processed; /* List of VM I/O jobs already processed */
 463     list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
 464     pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
 465     pthread_mutex_t io_swapfile_mutex; /* So we can lseek + write */
 466     pthread_attr_t io_threads_attr; /* attributes for threads creation */
 467     int io_active_threads; /* Number of running I/O threads */
 468     int vm_max_threads; /* Max number of I/O threads running at the same time */
 469     /* Our main thread is blocked on the event loop, locking for sockets ready
 470      * to be read or written, so when a threaded I/O operation is ready to be
 471      * processed by the main thread, the I/O thread will use a unix pipe to
 472      * awake the main thread. The followings are the two pipe FDs. */
 473     int io_ready_pipe_read;
 474     int io_ready_pipe_write;
 475     /* Virtual memory stats */
 476     unsigned long long vm_stats_used_pages;
 477     unsigned long long vm_stats_swapped_objects;
 478     unsigned long long vm_stats_swapouts;
 479     unsigned long long vm_stats_swapins;
 480     /* Pubsub */
 481     dict *pubsub_channels; /* Map channels to list of subscribed clients */
 482     list *pubsub_patterns; /* A list of pubsub_patterns */
 483     /* Misc */
 484     unsigned lruclock:22;        /* clock incrementing every minute, for LRU */
 485     unsigned lruclock_padding:10;
 486 };
 487
 488 typedef struct pubsubPattern {
 489     redisClient *client;
 490     robj *pattern;
 491 } pubsubPattern;
 492
 493 typedef void redisCommandProc(redisClient *c);
 494 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 495 struct redisCommand {
 496     char *name;
 497     redisCommandProc *proc;
 498     int arity;
 499     int flags;
 500     /* Use a function to determine which keys need to be loaded
 501      * in the background prior to executing this command. Takes precedence
 502      * over vm_firstkey and others, ignored when NULL */
 503     redisVmPreloadProc *vm_preload_proc;
 504     /* What keys should be loaded in background when calling this command? */
 505     int vm_firstkey; /* The first argument that's a key (0 = no keys) */
 506     int vm_lastkey;  /* THe last argument that's a key */
 507     int vm_keystep;  /* The step between first and last key */
 508 };
 509
 510 struct redisFunctionSym {
 511     char *name;
 512     unsigned long pointer;
 513 };
 514
 515 typedef struct _redisSortObject {
 516     robj *obj;
 517     union {
 518         double score;
 519         robj *cmpobj;
 520     } u;
 521 } redisSortObject;
 522
 523 typedef struct _redisSortOperation {
 524     int type;
 525     robj *pattern;
 526 } redisSortOperation;
 527
 528 /* ZSETs use a specialized version of Skiplists */
 529 typedef struct zskiplistNode {
 530     robj *obj;
 531     double score;
 532     struct zskiplistNode *backward;
 533     struct zskiplistLevel {
 534         struct zskiplistNode *forward;
 535         unsigned int span;
 536     } level[];
 537 } zskiplistNode;
 538
 539 typedef struct zskiplist {
 540     struct zskiplistNode *header, *tail;
 541     unsigned long length;
 542     int level;
 543 } zskiplist;
 544
 545 typedef struct zset {
 546     dict *dict;
 547     zskiplist *zsl;
 548 } zset;
 549
 550 /* VM threaded I/O request message */
 551 #define REDIS_IOJOB_LOAD 0          /* Load from disk to memory */
 552 #define REDIS_IOJOB_PREPARE_SWAP 1  /* Compute needed pages */
 553 #define REDIS_IOJOB_DO_SWAP 2       /* Swap from memory to disk */
 554 typedef struct iojob {
 555     int type;   /* Request type, REDIS_IOJOB_* */
 556     redisDb *db;/* Redis database */
 557     robj *key;  /* This I/O request is about swapping this key */
 558     robj *id;   /* Unique identifier of this job:
 559                    this is the object to swap for REDIS_IOREQ_*_SWAP, or the
 560                    vmpointer objct for REDIS_IOREQ_LOAD. */
 561     robj *val;  /* the value to swap for REDIS_IOREQ_*_SWAP, otherwise this
 562                  * field is populated by the I/O thread for REDIS_IOREQ_LOAD. */
 563     off_t page; /* Swap page where to read/write the object */
 564     off_t pages; /* Swap pages needed to save object. PREPARE_SWAP return val */
 565     int canceled; /* True if this command was canceled by blocking side of VM */
 566     pthread_t thread; /* ID of the thread processing this entry */
 567 } iojob;
 568
 569 /* Structure to hold list iteration abstraction. */
 570 typedef struct {
 571     robj *subject;
 572     unsigned char encoding;
 573     unsigned char direction; /* Iteration direction */
 574     unsigned char *zi;
 575     listNode *ln;
 576 } listTypeIterator;
 577
 578 /* Structure for an entry while iterating over a list. */
 579 typedef struct {
 580     listTypeIterator *li;
 581     unsigned char *zi;  /* Entry in ziplist */
 582     listNode *ln;       /* Entry in linked list */
 583 } listTypeEntry;
 584
 585 /* Structure to hold set iteration abstraction. */
 586 typedef struct {
 587     robj *subject;
 588     int encoding;
 589     int ii; /* intset iterator */
 590     dictIterator *di;
 591 } setTypeIterator;
 592
 593 /* Structure to hold hash iteration abstration. Note that iteration over
 594  * hashes involves both fields and values. Because it is possible that
 595  * not both are required, store pointers in the iterator to avoid
 596  * unnecessary memory allocation for fields/values. */
 597 typedef struct {
 598     int encoding;
 599     unsigned char *zi;
 600     unsigned char *zk, *zv;
 601     unsigned int zklen, zvlen;
 602
 603     dictIterator *di;
 604     dictEntry *de;
 605 } hashTypeIterator;
 606
 607 #define REDIS_HASH_KEY 1
 608 #define REDIS_HASH_VALUE 2
 609
 610 /*-----------------------------------------------------------------------------
 611  * Extern declarations
 612  *----------------------------------------------------------------------------*/
 613
 614 extern struct redisServer server;
 615 extern struct sharedObjectsStruct shared;
 616 extern dictType setDictType;
 617 extern dictType zsetDictType;
 618 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
 619 dictType hashDictType;
 620
 621 /*-----------------------------------------------------------------------------
 622  * Functions prototypes
 623  *----------------------------------------------------------------------------*/
 624
 625 /* networking.c -- Networking and Client related operations */
 626 redisClient *createClient(int fd);
 627 void closeTimedoutClients(void);
 628 void freeClient(redisClient *c);
 629 void resetClient(redisClient *c);
 630 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
 631 void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int mask);
 632 void addReply(redisClient *c, robj *obj);
 633 void *addDeferredMultiBulkLength(redisClient *c);
 634 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
 635 void addReplySds(redisClient *c, sds s);
 636 void processInputBuffer(redisClient *c);
 637 void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 638 void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 639 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
 640 void addReplyBulk(redisClient *c, robj *obj);
 641 void addReplyBulkCString(redisClient *c, char *s);
 642 void addReplyBulkCBuffer(redisClient *c, void *p, size_t len);
 643 void addReplyBulkLongLong(redisClient *c, long long ll);
 644 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 645 void addReply(redisClient *c, robj *obj);
 646 void addReplySds(redisClient *c, sds s);
 647 void addReplyError(redisClient *c, char *err);
 648 void addReplyStatus(redisClient *c, char *status);
 649 void addReplyDouble(redisClient *c, double d);
 650 void addReplyLongLong(redisClient *c, long long ll);
 651 void addReplyMultiBulkLen(redisClient *c, long length);
 652 void *dupClientReplyValue(void *o);
 653
 654 #ifdef __GNUC__
 655 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
 656     __attribute__((format(printf, 2, 3)));
 657 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
 658     __attribute__((format(printf, 2, 3)));
 659 #else
 660 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
 661 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
 662 #endif
 663
 664 /* List data type */
 665 void listTypeTryConversion(robj *subject, robj *value);
 666 void listTypePush(robj *subject, robj *value, int where);
 667 robj *listTypePop(robj *subject, int where);
 668 unsigned long listTypeLength(robj *subject);
 669 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
 670 void listTypeReleaseIterator(listTypeIterator *li);
 671 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
 672 robj *listTypeGet(listTypeEntry *entry);
 673 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
 674 int listTypeEqual(listTypeEntry *entry, robj *o);
 675 void listTypeDelete(listTypeEntry *entry);
 676 void listTypeConvert(robj *subject, int enc);
 677 void unblockClientWaitingData(redisClient *c);
 678 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
 679 void popGenericCommand(redisClient *c, int where);
 680
 681 /* MULTI/EXEC/WATCH... */
 682 void unwatchAllKeys(redisClient *c);
 683 void initClientMultiState(redisClient *c);
 684 void freeClientMultiState(redisClient *c);
 685 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
 686 void touchWatchedKey(redisDb *db, robj *key);
 687 void touchWatchedKeysOnFlush(int dbid);
 688
 689 /* Redis object implementation */
 690 void decrRefCount(void *o);
 691 void incrRefCount(robj *o);
 692 void freeStringObject(robj *o);
 693 void freeListObject(robj *o);
 694 void freeSetObject(robj *o);
 695 void freeZsetObject(robj *o);
 696 void freeHashObject(robj *o);
 697 robj *createObject(int type, void *ptr);
 698 robj *createStringObject(char *ptr, size_t len);
 699 robj *dupStringObject(robj *o);
 700 robj *tryObjectEncoding(robj *o);
 701 robj *getDecodedObject(robj *o);
 702 size_t stringObjectLen(robj *o);
 703 robj *createStringObjectFromLongLong(long long value);
 704 robj *createListObject(void);
 705 robj *createZiplistObject(void);
 706 robj *createSetObject(void);
 707 robj *createIntsetObject(void);
 708 robj *createHashObject(void);
 709 robj *createZsetObject(void);
 710 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
 711 int checkType(redisClient *c, robj *o, int type);
 712 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
 713 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
 714 int getLongLongFromObject(robj *o, long long *target);
 715 char *strEncoding(int encoding);
 716 int compareStringObjects(robj *a, robj *b);
 717 int equalStringObjects(robj *a, robj *b);
 718 unsigned long estimateObjectIdleTime(robj *o);
 719
 720 /* Synchronous I/O with timeout */
 721 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
 722 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
 723 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
 724 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
 725 int fwriteBulkDouble(FILE *fp, double d);
 726 int fwriteBulkLongLong(FILE *fp, long long l);
 727 int fwriteBulkObject(FILE *fp, robj *obj);
 728
 729 /* Replication */
 730 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
 731 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
 732 int syncWithMaster(void);
 733 void updateSlavesWaitingBgsave(int bgsaveerr);
 734 void replicationCron(void);
 735
 736 /* Generic persistence functions */
 737 void startLoading(FILE *fp);
 738 void loadingProgress(off_t pos);
 739 void stopLoading(void);
 740
 741 /* RDB persistence */
 742 int rdbLoad(char *filename);
 743 int rdbSaveBackground(char *filename);
 744 void rdbRemoveTempFile(pid_t childpid);
 745 int rdbSave(char *filename);
 746 int rdbSaveObject(FILE *fp, robj *o);
 747 off_t rdbSavedObjectLen(robj *o);
 748 off_t rdbSavedObjectPages(robj *o);
 749 robj *rdbLoadObject(int type, FILE *fp);
 750 void backgroundSaveDoneHandler(int statloc);
 751
 752 /* AOF persistence */
 753 void flushAppendOnlyFile(void);
 754 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
 755 void aofRemoveTempFile(pid_t childpid);
 756 int rewriteAppendOnlyFileBackground(void);
 757 int loadAppendOnlyFile(char *filename);
 758 void stopAppendOnly(void);
 759 int startAppendOnly(void);
 760 void backgroundRewriteDoneHandler(int statloc);
 761
 762 /* Sorted sets data type */
 763 zskiplist *zslCreate(void);
 764 void zslFree(zskiplist *zsl);
 765 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
 766
 767 /* Core functions */
 768 void freeMemoryIfNeeded(void);
 769 int processCommand(redisClient *c);
 770 void setupSigSegvAction(void);
 771 struct redisCommand *lookupCommand(sds name);
 772 struct redisCommand *lookupCommandByCString(char *s);
 773 void call(redisClient *c, struct redisCommand *cmd);
 774 int prepareForShutdown();
 775 void redisLog(int level, const char *fmt, ...);
 776 void usage();
 777 void updateDictResizePolicy(void);
 778 int htNeedsResize(dict *dict);
 779 void oom(const char *msg);
 780 void populateCommandTable(void);
 781
 782 /* Virtual Memory */
 783 void vmInit(void);
 784 void vmMarkPagesFree(off_t page, off_t count);
 785 robj *vmLoadObject(robj *o);
 786 robj *vmPreviewObject(robj *o);
 787 int vmSwapOneObjectBlocking(void);
 788 int vmSwapOneObjectThreaded(void);
 789 int vmCanSwapOut(void);
 790 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
 791 void vmCancelThreadedIOJob(robj *o);
 792 void lockThreadedIO(void);
 793 void unlockThreadedIO(void);
 794 int vmSwapObjectThreaded(robj *key, robj *val, redisDb *db);
 795 void freeIOJob(iojob *j);
 796 void queueIOJob(iojob *j);
 797 int vmWriteObjectOnSwap(robj *o, off_t page);
 798 robj *vmReadObjectFromSwap(off_t page, int type);
 799 void waitEmptyIOJobsQueue(void);
 800 void vmReopenSwapFile(void);
 801 int vmFreePage(off_t page);
 802 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 803 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 804 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
 805 int dontWaitForSwappedKey(redisClient *c, robj *key);
 806 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
 807 vmpointer *vmSwapObjectBlocking(robj *val);
 808
 809 /* Set data type */
 810 robj *setTypeCreate(robj *value);
 811 int setTypeAdd(robj *subject, robj *value);
 812 int setTypeRemove(robj *subject, robj *value);
 813 int setTypeIsMember(robj *subject, robj *value);
 814 setTypeIterator *setTypeInitIterator(robj *subject);
 815 void setTypeReleaseIterator(setTypeIterator *si);
 816 robj *setTypeNext(setTypeIterator *si);
 817 int setTypeRandomElement(robj *setobj, robj **objele, long long *llele);
 818 unsigned long setTypeSize(robj *subject);
 819 void setTypeConvert(robj *subject, int enc);
 820
 821 /* Hash data type */
 822 void convertToRealHash(robj *o);
 823 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
 824 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
 825 robj *hashTypeGet(robj *o, robj *key);
 826 int hashTypeExists(robj *o, robj *key);
 827 int hashTypeSet(robj *o, robj *key, robj *value);
 828 int hashTypeDelete(robj *o, robj *key);
 829 unsigned long hashTypeLength(robj *o);
 830 hashTypeIterator *hashTypeInitIterator(robj *subject);
 831 void hashTypeReleaseIterator(hashTypeIterator *hi);
 832 int hashTypeNext(hashTypeIterator *hi);
 833 robj *hashTypeCurrent(hashTypeIterator *hi, int what);
 834 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
 835
 836 /* Pub / Sub */
 837 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
 838 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
 839 void freePubsubPattern(void *p);
 840 int listMatchPubsubPattern(void *a, void *b);
 841
 842 /* Utility functions */
 843 int stringmatchlen(const char *pattern, int patternLen,
 844         const char *string, int stringLen, int nocase);
 845 int stringmatch(const char *pattern, const char *string, int nocase);
 846 long long memtoll(const char *p, int *err);
 847 int ll2string(char *s, size_t len, long long value);
 848 int isStringRepresentableAsLong(sds s, long *longval);
 849 int isStringRepresentableAsLongLong(sds s, long long *longval);
 850 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
 851
 852 /* Configuration */
 853 void loadServerConfig(char *filename);
 854 void appendServerSaveParams(time_t seconds, int changes);
 855 void resetServerSaveParams();
 856
 857 /* db.c -- Keyspace access API */
 858 int removeExpire(redisDb *db, robj *key);
 859 void propagateExpire(redisDb *db, robj *key);
 860 int expireIfNeeded(redisDb *db, robj *key);
 861 time_t getExpire(redisDb *db, robj *key);
 862 void setExpire(redisDb *db, robj *key, time_t when);
 863 robj *lookupKey(redisDb *db, robj *key);
 864 robj *lookupKeyRead(redisDb *db, robj *key);
 865 robj *lookupKeyWrite(redisDb *db, robj *key);
 866 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
 867 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
 868 int dbAdd(redisDb *db, robj *key, robj *val);
 869 int dbReplace(redisDb *db, robj *key, robj *val);
 870 int dbExists(redisDb *db, robj *key);
 871 robj *dbRandomKey(redisDb *db);
 872 int dbDelete(redisDb *db, robj *key);
 873 long long emptyDb();
 874 int selectDb(redisClient *c, int id);
 875
 876 /* Git SHA1 */
 877 char *redisGitSHA1(void);
 878 char *redisGitDirty(void);
 879
 880 /* Commands prototypes */
 881 void authCommand(redisClient *c);
 882 void pingCommand(redisClient *c);
 883 void echoCommand(redisClient *c);
 884 void setCommand(redisClient *c);
 885 void setnxCommand(redisClient *c);
 886 void setexCommand(redisClient *c);
 887 void getCommand(redisClient *c);
 888 void delCommand(redisClient *c);
 889 void existsCommand(redisClient *c);
 890 void setbitCommand(redisClient *c);
 891 void getbitCommand(redisClient *c);
 892 void setrangeCommand(redisClient *c);
 893 void incrCommand(redisClient *c);
 894 void decrCommand(redisClient *c);
 895 void incrbyCommand(redisClient *c);
 896 void decrbyCommand(redisClient *c);
 897 void selectCommand(redisClient *c);
 898 void randomkeyCommand(redisClient *c);
 899 void keysCommand(redisClient *c);
 900 void dbsizeCommand(redisClient *c);
 901 void lastsaveCommand(redisClient *c);
 902 void saveCommand(redisClient *c);
 903 void bgsaveCommand(redisClient *c);
 904 void bgrewriteaofCommand(redisClient *c);
 905 void shutdownCommand(redisClient *c);
 906 void moveCommand(redisClient *c);
 907 void renameCommand(redisClient *c);
 908 void renamenxCommand(redisClient *c);
 909 void lpushCommand(redisClient *c);
 910 void rpushCommand(redisClient *c);
 911 void lpushxCommand(redisClient *c);
 912 void rpushxCommand(redisClient *c);
 913 void linsertCommand(redisClient *c);
 914 void lpopCommand(redisClient *c);
 915 void rpopCommand(redisClient *c);
 916 void llenCommand(redisClient *c);
 917 void lindexCommand(redisClient *c);
 918 void lrangeCommand(redisClient *c);
 919 void ltrimCommand(redisClient *c);
 920 void typeCommand(redisClient *c);
 921 void lsetCommand(redisClient *c);
 922 void saddCommand(redisClient *c);
 923 void sremCommand(redisClient *c);
 924 void smoveCommand(redisClient *c);
 925 void sismemberCommand(redisClient *c);
 926 void scardCommand(redisClient *c);
 927 void spopCommand(redisClient *c);
 928 void srandmemberCommand(redisClient *c);
 929 void sinterCommand(redisClient *c);
 930 void sinterstoreCommand(redisClient *c);
 931 void sunionCommand(redisClient *c);
 932 void sunionstoreCommand(redisClient *c);
 933 void sdiffCommand(redisClient *c);
 934 void sdiffstoreCommand(redisClient *c);
 935 void syncCommand(redisClient *c);
 936 void flushdbCommand(redisClient *c);
 937 void flushallCommand(redisClient *c);
 938 void sortCommand(redisClient *c);
 939 void lremCommand(redisClient *c);
 940 void rpoplpushcommand(redisClient *c);
 941 void infoCommand(redisClient *c);
 942 void mgetCommand(redisClient *c);
 943 void monitorCommand(redisClient *c);
 944 void expireCommand(redisClient *c);
 945 void expireatCommand(redisClient *c);
 946 void getsetCommand(redisClient *c);
 947 void ttlCommand(redisClient *c);
 948 void persistCommand(redisClient *c);
 949 void slaveofCommand(redisClient *c);
 950 void debugCommand(redisClient *c);
 951 void msetCommand(redisClient *c);
 952 void msetnxCommand(redisClient *c);
 953 void zaddCommand(redisClient *c);
 954 void zincrbyCommand(redisClient *c);
 955 void zrangeCommand(redisClient *c);
 956 void zrangebyscoreCommand(redisClient *c);
 957 void zrevrangebyscoreCommand(redisClient *c);
 958 void zcountCommand(redisClient *c);
 959 void zrevrangeCommand(redisClient *c);
 960 void zcardCommand(redisClient *c);
 961 void zremCommand(redisClient *c);
 962 void zscoreCommand(redisClient *c);
 963 void zremrangebyscoreCommand(redisClient *c);
 964 void multiCommand(redisClient *c);
 965 void execCommand(redisClient *c);
 966 void discardCommand(redisClient *c);
 967 void blpopCommand(redisClient *c);
 968 void brpopCommand(redisClient *c);
 969 void appendCommand(redisClient *c);
 970 void substrCommand(redisClient *c);
 971 void strlenCommand(redisClient *c);
 972 void zrankCommand(redisClient *c);
 973 void zrevrankCommand(redisClient *c);
 974 void hsetCommand(redisClient *c);
 975 void hsetnxCommand(redisClient *c);
 976 void hgetCommand(redisClient *c);
 977 void hmsetCommand(redisClient *c);
 978 void hmgetCommand(redisClient *c);
 979 void hdelCommand(redisClient *c);
 980 void hlenCommand(redisClient *c);
 981 void zremrangebyrankCommand(redisClient *c);
 982 void zunionstoreCommand(redisClient *c);
 983 void zinterstoreCommand(redisClient *c);
 984 void hkeysCommand(redisClient *c);
 985 void hvalsCommand(redisClient *c);
 986 void hgetallCommand(redisClient *c);
 987 void hexistsCommand(redisClient *c);
 988 void configCommand(redisClient *c);
 989 void hincrbyCommand(redisClient *c);
 990 void subscribeCommand(redisClient *c);
 991 void unsubscribeCommand(redisClient *c);
 992 void psubscribeCommand(redisClient *c);
 993 void punsubscribeCommand(redisClient *c);
 994 void publishCommand(redisClient *c);
 995 void watchCommand(redisClient *c);
 996 void unwatchCommand(redisClient *c);
 997
 998 #if defined(__GNUC__)
 999 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
1000 void free(void *ptr) __attribute__ ((deprecated));
1001 void *malloc(size_t size) __attribute__ ((deprecated));
1002 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
1003 #endif
1004
1005 #endif