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_CONNECTED 2  /* Connected to master */
 156
 157 /* Slave replication state - from the point of view of master
 158  * Note that in SEND_BULK and ONLINE state the slave receives new updates
 159  * in its output queue. In the WAIT_BGSAVE state instead the server is waiting
 160  * to start the next background saving in order to send updates to it. */
 161 #define REDIS_REPL_WAIT_BGSAVE_START 3 /* master waits bgsave to start feeding it */
 162 #define REDIS_REPL_WAIT_BGSAVE_END 4 /* master waits bgsave to start bulk DB transmission */
 163 #define REDIS_REPL_SEND_BULK 5 /* master is sending the bulk DB */
 164 #define REDIS_REPL_ONLINE 6 /* bulk DB already transmitted, receive updates */
 165
 166 /* List related stuff */
 167 #define REDIS_HEAD 0
 168 #define REDIS_TAIL 1
 169
 170 /* Sort operations */
 171 #define REDIS_SORT_GET 0
 172 #define REDIS_SORT_ASC 1
 173 #define REDIS_SORT_DESC 2
 174 #define REDIS_SORTKEY_MAX 1024
 175
 176 /* Log levels */
 177 #define REDIS_DEBUG 0
 178 #define REDIS_VERBOSE 1
 179 #define REDIS_NOTICE 2
 180 #define REDIS_WARNING 3
 181
 182 /* Anti-warning macro... */
 183 #define REDIS_NOTUSED(V) ((void) V)
 184
 185 #define ZSKIPLIST_MAXLEVEL 32 /* Should be enough for 2^32 elements */
 186 #define ZSKIPLIST_P 0.25      /* Skiplist P = 1/4 */
 187
 188 /* Append only defines */
 189 #define APPENDFSYNC_NO 0
 190 #define APPENDFSYNC_ALWAYS 1
 191 #define APPENDFSYNC_EVERYSEC 2
 192
 193 /* Zip structure related defaults */
 194 #define REDIS_HASH_MAX_ZIPMAP_ENTRIES 64
 195 #define REDIS_HASH_MAX_ZIPMAP_VALUE 512
 196 #define REDIS_LIST_MAX_ZIPLIST_ENTRIES 1024
 197 #define REDIS_LIST_MAX_ZIPLIST_VALUE 32
 198 #define REDIS_SET_MAX_INTSET_ENTRIES 4096
 199
 200 /* Sets operations codes */
 201 #define REDIS_OP_UNION 0
 202 #define REDIS_OP_DIFF 1
 203 #define REDIS_OP_INTER 2
 204
 205 /* Redis maxmemory strategies */
 206 #define REDIS_MAXMEMORY_VOLATILE_LRU 0
 207 #define REDIS_MAXMEMORY_VOLATILE_TTL 1
 208 #define REDIS_MAXMEMORY_VOLATILE_RANDOM 2
 209 #define REDIS_MAXMEMORY_ALLKEYS_LRU 3
 210 #define REDIS_MAXMEMORY_ALLKEYS_RANDOM 4
 211
 212 /* We can print the stacktrace, so our assert is defined this way: */
 213 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
 214 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
 215 void _redisAssert(char *estr, char *file, int line);
 216 void _redisPanic(char *msg, char *file, int line);
 217
 218 /*-----------------------------------------------------------------------------
 219  * Data types
 220  *----------------------------------------------------------------------------*/
 221
 222 /* A redis object, that is a type able to hold a string / list / set */
 223
 224 /* The actual Redis Object */
 225 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
 226 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
 227 typedef struct redisObject {
 228     unsigned type:4;
 229     unsigned storage:2;     /* REDIS_VM_MEMORY or REDIS_VM_SWAPPING */
 230     unsigned encoding:4;
 231     unsigned lru:22;        /* lru time (relative to server.lruclock) */
 232     int refcount;
 233     void *ptr;
 234     /* VM fields are only allocated if VM is active, otherwise the
 235      * object allocation function will just allocate
 236      * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
 237      * Redis without VM active will not have any overhead. */
 238 } robj;
 239
 240 /* The VM pointer structure - identifies an object in the swap file.
 241  *
 242  * This object is stored in place of the value
 243  * object in the main key->value hash table representing a database.
 244  * Note that the first fields (type, storage) are the same as the redisObject
 245  * structure so that vmPointer strucuters can be accessed even when casted
 246  * as redisObject structures.
 247  *
 248  * This is useful as we don't know if a value object is or not on disk, but we
 249  * are always able to read obj->storage to check this. For vmPointer
 250  * structures "type" is set to REDIS_VMPOINTER (even if without this field
 251  * is still possible to check the kind of object from the value of 'storage').*/
 252 typedef struct vmPointer {
 253     unsigned type:4;
 254     unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
 255     unsigned notused:26;
 256     unsigned int vtype; /* type of the object stored in the swap file */
 257     off_t page;         /* the page at witch the object is stored on disk */
 258     off_t usedpages;    /* number of pages used on disk */
 259 } vmpointer;
 260
 261 /* Macro used to initalize a Redis object allocated on the stack.
 262  * Note that this macro is taken near the structure definition to make sure
 263  * we'll update it when the structure is changed, to avoid bugs like
 264  * bug #85 introduced exactly in this way. */
 265 #define initStaticStringObject(_var,_ptr) do { \
 266     _var.refcount = 1; \
 267     _var.type = REDIS_STRING; \
 268     _var.encoding = REDIS_ENCODING_RAW; \
 269     _var.ptr = _ptr; \
 270     _var.storage = REDIS_VM_MEMORY; \
 271 } while(0);
 272
 273 typedef struct redisDb {
 274     dict *dict;                 /* The keyspace for this DB */
 275     dict *expires;              /* Timeout of keys with a timeout set */
 276     dict *blocking_keys;        /* Keys with clients waiting for data (BLPOP) */
 277     dict *io_keys;              /* Keys with clients waiting for VM I/O */
 278     dict *watched_keys;         /* WATCHED keys for MULTI/EXEC CAS */
 279     int id;
 280 } redisDb;
 281
 282 /* Client MULTI/EXEC state */
 283 typedef struct multiCmd {
 284     robj **argv;
 285     int argc;
 286     struct redisCommand *cmd;
 287 } multiCmd;
 288
 289 typedef struct multiState {
 290     multiCmd *commands;     /* Array of MULTI commands */
 291     int count;              /* Total number of MULTI commands */
 292 } multiState;
 293
 294 /* With multiplexing we need to take per-clinet state.
 295  * Clients are taken in a liked list. */
 296 typedef struct redisClient {
 297     int fd;
 298     redisDb *db;
 299     int dictid;
 300     sds querybuf;
 301     int argc;
 302     robj **argv;
 303     int reqtype;
 304     int multibulklen;       /* number of multi bulk arguments left to read */
 305     long bulklen;           /* length of bulk argument in multi bulk request */
 306     list *reply;
 307     int sentlen;
 308     time_t lastinteraction; /* time of the last interaction, used for timeout */
 309     int flags;              /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
 310     int slaveseldb;         /* slave selected db, if this client is a slave */
 311     int authenticated;      /* when requirepass is non-NULL */
 312     int replstate;          /* replication state if this is a slave */
 313     int repldbfd;           /* replication DB file descriptor */
 314     long repldboff;         /* replication DB file offset */
 315     off_t repldbsize;       /* replication DB file size */
 316     multiState mstate;      /* MULTI/EXEC state */
 317     robj **blocking_keys;   /* The key we are waiting to terminate a blocking
 318                              * operation such as BLPOP. Otherwise NULL. */
 319     int blocking_keys_num;  /* Number of blocking keys */
 320     time_t blockingto;      /* Blocking operation timeout. If UNIX current time
 321                              * is >= blockingto then the operation timed out. */
 322     list *io_keys;          /* Keys this client is waiting to be loaded from the
 323                              * swap file in order to continue. */
 324     list *watched_keys;     /* Keys WATCHED for MULTI/EXEC CAS */
 325     dict *pubsub_channels;  /* channels a client is interested in (SUBSCRIBE) */
 326     list *pubsub_patterns;  /* patterns a client is interested in (SUBSCRIBE) */
 327
 328     /* Response buffer */
 329     int bufpos;
 330     char buf[REDIS_REPLY_CHUNK_BYTES];
 331 } redisClient;
 332
 333 struct saveparam {
 334     time_t seconds;
 335     int changes;
 336 };
 337
 338 struct sharedObjectsStruct {
 339     robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
 340     *colon, *nullbulk, *nullmultibulk, *queued,
 341     *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
 342     *outofrangeerr, *plus,
 343     *select0, *select1, *select2, *select3, *select4,
 344     *select5, *select6, *select7, *select8, *select9,
 345     *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
 346     *mbulk4, *psubscribebulk, *punsubscribebulk,
 347     *integers[REDIS_SHARED_INTEGERS];
 348 };
 349
 350 /* Global server state structure */
 351 struct redisServer {
 352     pthread_t mainthread;
 353     int port;
 354     char *bindaddr;
 355     char *unixsocket;
 356     int ipfd;
 357     int sofd;
 358     redisDb *db;
 359     long long dirty;            /* changes to DB from the last save */
 360     long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
 361     list *clients;
 362     dict *commands;             /* Command table hahs table */
 363     struct redisCommand *delCommand, *multiCommand; /* often lookedup cmds */
 364     list *slaves, *monitors;
 365     char neterr[ANET_ERR_LEN];
 366     aeEventLoop *el;
 367     int cronloops;              /* number of times the cron function run */
 368     time_t lastsave;                /* Unix time of last save succeeede */
 369     /* Fields used only for stats */
 370     time_t stat_starttime;          /* server start time */
 371     long long stat_numcommands;     /* number of processed commands */
 372     long long stat_numconnections;  /* number of connections received */
 373     long long stat_expiredkeys;     /* number of expired keys */
 374     long long stat_keyspace_hits;   /* number of successful lookups of keys */
 375     long long stat_keyspace_misses; /* number of failed lookups of keys */
 376     /* Configuration */
 377     int verbosity;
 378     int glueoutputbuf;
 379     int maxidletime;
 380     int dbnum;
 381     int daemonize;
 382     int appendonly;
 383     int appendfsync;
 384     int no_appendfsync_on_rewrite;
 385     int shutdown_asap;
 386     time_t lastfsync;
 387     int appendfd;
 388     int appendseldb;
 389     char *pidfile;
 390     pid_t bgsavechildpid;
 391     pid_t bgrewritechildpid;
 392     sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
 393     sds aofbuf;       /* AOF buffer, written before entering the event loop */
 394     struct saveparam *saveparams;
 395     int saveparamslen;
 396     char *logfile;
 397     char *dbfilename;
 398     char *appendfilename;
 399     char *requirepass;
 400     int rdbcompression;
 401     int activerehashing;
 402     /* Replication related */
 403     int isslave;
 404     char *masterauth;
 405     char *masterhost;
 406     int masterport;
 407     redisClient *master;    /* client that is master for this slave */
 408     int replstate;
 409     unsigned int maxclients;
 410     unsigned long long maxmemory;
 411     int maxmemory_policy;
 412     int maxmemory_samples;
 413     unsigned int blpop_blocked_clients;
 414     unsigned int vm_blocked_clients;
 415     /* Sort parameters - qsort_r() is only available under BSD so we
 416      * have to take this state global, in order to pass it to sortCompare() */
 417     int sort_desc;
 418     int sort_alpha;
 419     int sort_bypattern;
 420     /* Virtual memory configuration */
 421     int vm_enabled;
 422     char *vm_swap_file;
 423     off_t vm_page_size;
 424     off_t vm_pages;
 425     unsigned long long vm_max_memory;
 426     /* Zip structure config */
 427     size_t hash_max_zipmap_entries;
 428     size_t hash_max_zipmap_value;
 429     size_t list_max_ziplist_entries;
 430     size_t list_max_ziplist_value;
 431     size_t set_max_intset_entries;
 432     /* Virtual memory state */
 433     FILE *vm_fp;
 434     int vm_fd;
 435     off_t vm_next_page; /* Next probably empty page */
 436     off_t vm_near_pages; /* Number of pages allocated sequentially */
 437     unsigned char *vm_bitmap; /* Bitmap of free/used pages */
 438     time_t unixtime;    /* Unix time sampled every second. */
 439     /* Virtual memory I/O threads stuff */
 440     /* An I/O thread process an element taken from the io_jobs queue and
 441      * put the result of the operation in the io_done list. While the
 442      * job is being processed, it's put on io_processing queue. */
 443     list *io_newjobs; /* List of VM I/O jobs yet to be processed */
 444     list *io_processing; /* List of VM I/O jobs being processed */
 445     list *io_processed; /* List of VM I/O jobs already processed */
 446     list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
 447     pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
 448     pthread_mutex_t io_swapfile_mutex; /* So we can lseek + write */
 449     pthread_attr_t io_threads_attr; /* attributes for threads creation */
 450     int io_active_threads; /* Number of running I/O threads */
 451     int vm_max_threads; /* Max number of I/O threads running at the same time */
 452     /* Our main thread is blocked on the event loop, locking for sockets ready
 453      * to be read or written, so when a threaded I/O operation is ready to be
 454      * processed by the main thread, the I/O thread will use a unix pipe to
 455      * awake the main thread. The followings are the two pipe FDs. */
 456     int io_ready_pipe_read;
 457     int io_ready_pipe_write;
 458     /* Virtual memory stats */
 459     unsigned long long vm_stats_used_pages;
 460     unsigned long long vm_stats_swapped_objects;
 461     unsigned long long vm_stats_swapouts;
 462     unsigned long long vm_stats_swapins;
 463     /* Pubsub */
 464     dict *pubsub_channels; /* Map channels to list of subscribed clients */
 465     list *pubsub_patterns; /* A list of pubsub_patterns */
 466     /* Misc */
 467     FILE *devnull;
 468     unsigned lruclock:22;        /* clock incrementing every minute, for LRU */
 469     unsigned lruclock_padding:10;
 470 };
 471
 472 typedef struct pubsubPattern {
 473     redisClient *client;
 474     robj *pattern;
 475 } pubsubPattern;
 476
 477 typedef void redisCommandProc(redisClient *c);
 478 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 479 struct redisCommand {
 480     char *name;
 481     redisCommandProc *proc;
 482     int arity;
 483     int flags;
 484     /* Use a function to determine which keys need to be loaded
 485      * in the background prior to executing this command. Takes precedence
 486      * over vm_firstkey and others, ignored when NULL */
 487     redisVmPreloadProc *vm_preload_proc;
 488     /* What keys should be loaded in background when calling this command? */
 489     int vm_firstkey; /* The first argument that's a key (0 = no keys) */
 490     int vm_lastkey;  /* THe last argument that's a key */
 491     int vm_keystep;  /* The step between first and last key */
 492 };
 493
 494 struct redisFunctionSym {
 495     char *name;
 496     unsigned long pointer;
 497 };
 498
 499 typedef struct _redisSortObject {
 500     robj *obj;
 501     union {
 502         double score;
 503         robj *cmpobj;
 504     } u;
 505 } redisSortObject;
 506
 507 typedef struct _redisSortOperation {
 508     int type;
 509     robj *pattern;
 510 } redisSortOperation;
 511
 512 /* ZSETs use a specialized version of Skiplists */
 513 typedef struct zskiplistNode {
 514     robj *obj;
 515     double score;
 516     struct zskiplistNode *backward;
 517     struct zskiplistLevel {
 518         struct zskiplistNode *forward;
 519         unsigned int span;
 520     } level[];
 521 } zskiplistNode;
 522
 523 typedef struct zskiplist {
 524     struct zskiplistNode *header, *tail;
 525     unsigned long length;
 526     int level;
 527 } zskiplist;
 528
 529 typedef struct zset {
 530     dict *dict;
 531     zskiplist *zsl;
 532 } zset;
 533
 534 /* VM threaded I/O request message */
 535 #define REDIS_IOJOB_LOAD 0          /* Load from disk to memory */
 536 #define REDIS_IOJOB_PREPARE_SWAP 1  /* Compute needed pages */
 537 #define REDIS_IOJOB_DO_SWAP 2       /* Swap from memory to disk */
 538 typedef struct iojob {
 539     int type;   /* Request type, REDIS_IOJOB_* */
 540     redisDb *db;/* Redis database */
 541     robj *key;  /* This I/O request is about swapping this key */
 542     robj *id;   /* Unique identifier of this job:
 543                    this is the object to swap for REDIS_IOREQ_*_SWAP, or the
 544                    vmpointer objct for REDIS_IOREQ_LOAD. */
 545     robj *val;  /* the value to swap for REDIS_IOREQ_*_SWAP, otherwise this
 546                  * field is populated by the I/O thread for REDIS_IOREQ_LOAD. */
 547     off_t page; /* Swap page where to read/write the object */
 548     off_t pages; /* Swap pages needed to save object. PREPARE_SWAP return val */
 549     int canceled; /* True if this command was canceled by blocking side of VM */
 550     pthread_t thread; /* ID of the thread processing this entry */
 551 } iojob;
 552
 553 /* Structure to hold list iteration abstraction. */
 554 typedef struct {
 555     robj *subject;
 556     unsigned char encoding;
 557     unsigned char direction; /* Iteration direction */
 558     unsigned char *zi;
 559     listNode *ln;
 560 } listTypeIterator;
 561
 562 /* Structure for an entry while iterating over a list. */
 563 typedef struct {
 564     listTypeIterator *li;
 565     unsigned char *zi;  /* Entry in ziplist */
 566     listNode *ln;       /* Entry in linked list */
 567 } listTypeEntry;
 568
 569 /* Structure to hold set iteration abstraction. */
 570 typedef struct {
 571     robj *subject;
 572     int encoding;
 573     int ii; /* intset iterator */
 574     dictIterator *di;
 575 } setTypeIterator;
 576
 577 /* Structure to hold hash iteration abstration. Note that iteration over
 578  * hashes involves both fields and values. Because it is possible that
 579  * not both are required, store pointers in the iterator to avoid
 580  * unnecessary memory allocation for fields/values. */
 581 typedef struct {
 582     int encoding;
 583     unsigned char *zi;
 584     unsigned char *zk, *zv;
 585     unsigned int zklen, zvlen;
 586
 587     dictIterator *di;
 588     dictEntry *de;
 589 } hashTypeIterator;
 590
 591 #define REDIS_HASH_KEY 1
 592 #define REDIS_HASH_VALUE 2
 593
 594 /*-----------------------------------------------------------------------------
 595  * Extern declarations
 596  *----------------------------------------------------------------------------*/
 597
 598 extern struct redisServer server;
 599 extern struct sharedObjectsStruct shared;
 600 extern dictType setDictType;
 601 extern dictType zsetDictType;
 602 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
 603 dictType hashDictType;
 604
 605 /*-----------------------------------------------------------------------------
 606  * Functions prototypes
 607  *----------------------------------------------------------------------------*/
 608
 609 /* networking.c -- Networking and Client related operations */
 610 redisClient *createClient(int fd);
 611 void closeTimedoutClients(void);
 612 void freeClient(redisClient *c);
 613 void resetClient(redisClient *c);
 614 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
 615 void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int mask);
 616 void addReply(redisClient *c, robj *obj);
 617 void *addDeferredMultiBulkLength(redisClient *c);
 618 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
 619 void addReplySds(redisClient *c, sds s);
 620 void processInputBuffer(redisClient *c);
 621 void acceptTcpHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 622 void acceptUnixHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 623 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
 624 void addReplyBulk(redisClient *c, robj *obj);
 625 void addReplyBulkCString(redisClient *c, char *s);
 626 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 627 void addReply(redisClient *c, robj *obj);
 628 void addReplySds(redisClient *c, sds s);
 629 void addReplyError(redisClient *c, char *err);
 630 void addReplyStatus(redisClient *c, char *status);
 631 void addReplyDouble(redisClient *c, double d);
 632 void addReplyLongLong(redisClient *c, long long ll);
 633 void addReplyMultiBulkLen(redisClient *c, long length);
 634 void *dupClientReplyValue(void *o);
 635
 636 #ifdef __GNUC__
 637 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
 638     __attribute__((format(printf, 2, 3)));
 639 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
 640     __attribute__((format(printf, 2, 3)));
 641 #else
 642 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
 643 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
 644 #endif
 645
 646 /* List data type */
 647 void listTypeTryConversion(robj *subject, robj *value);
 648 void listTypePush(robj *subject, robj *value, int where);
 649 robj *listTypePop(robj *subject, int where);
 650 unsigned long listTypeLength(robj *subject);
 651 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
 652 void listTypeReleaseIterator(listTypeIterator *li);
 653 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
 654 robj *listTypeGet(listTypeEntry *entry);
 655 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
 656 int listTypeEqual(listTypeEntry *entry, robj *o);
 657 void listTypeDelete(listTypeEntry *entry);
 658 void listTypeConvert(robj *subject, int enc);
 659 void unblockClientWaitingData(redisClient *c);
 660 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
 661 void popGenericCommand(redisClient *c, int where);
 662
 663 /* MULTI/EXEC/WATCH... */
 664 void unwatchAllKeys(redisClient *c);
 665 void initClientMultiState(redisClient *c);
 666 void freeClientMultiState(redisClient *c);
 667 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
 668 void touchWatchedKey(redisDb *db, robj *key);
 669 void touchWatchedKeysOnFlush(int dbid);
 670
 671 /* Redis object implementation */
 672 void decrRefCount(void *o);
 673 void incrRefCount(robj *o);
 674 void freeStringObject(robj *o);
 675 void freeListObject(robj *o);
 676 void freeSetObject(robj *o);
 677 void freeZsetObject(robj *o);
 678 void freeHashObject(robj *o);
 679 robj *createObject(int type, void *ptr);
 680 robj *createStringObject(char *ptr, size_t len);
 681 robj *dupStringObject(robj *o);
 682 robj *tryObjectEncoding(robj *o);
 683 robj *getDecodedObject(robj *o);
 684 size_t stringObjectLen(robj *o);
 685 robj *createStringObjectFromLongLong(long long value);
 686 robj *createListObject(void);
 687 robj *createZiplistObject(void);
 688 robj *createSetObject(void);
 689 robj *createIntsetObject(void);
 690 robj *createHashObject(void);
 691 robj *createZsetObject(void);
 692 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
 693 int checkType(redisClient *c, robj *o, int type);
 694 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
 695 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
 696 int getLongLongFromObject(robj *o, long long *target);
 697 char *strEncoding(int encoding);
 698 int compareStringObjects(robj *a, robj *b);
 699 int equalStringObjects(robj *a, robj *b);
 700 unsigned long estimateObjectIdleTime(robj *o);
 701
 702 /* Synchronous I/O with timeout */
 703 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
 704 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
 705 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
 706 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
 707 int fwriteBulkDouble(FILE *fp, double d);
 708 int fwriteBulkLongLong(FILE *fp, long long l);
 709 int fwriteBulkObject(FILE *fp, robj *obj);
 710
 711 /* Replication */
 712 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
 713 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
 714 int syncWithMaster(void);
 715 void updateSlavesWaitingBgsave(int bgsaveerr);
 716
 717 /* RDB persistence */
 718 int rdbLoad(char *filename);
 719 int rdbSaveBackground(char *filename);
 720 void rdbRemoveTempFile(pid_t childpid);
 721 int rdbSave(char *filename);
 722 int rdbSaveObject(FILE *fp, robj *o);
 723 off_t rdbSavedObjectPages(robj *o, FILE *fp);
 724 off_t rdbSavedObjectLen(robj *o, FILE *fp);
 725 robj *rdbLoadObject(int type, FILE *fp);
 726 void backgroundSaveDoneHandler(int statloc);
 727
 728 /* AOF persistence */
 729 void flushAppendOnlyFile(void);
 730 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
 731 void aofRemoveTempFile(pid_t childpid);
 732 int rewriteAppendOnlyFileBackground(void);
 733 int loadAppendOnlyFile(char *filename);
 734 void stopAppendOnly(void);
 735 int startAppendOnly(void);
 736 void backgroundRewriteDoneHandler(int statloc);
 737
 738 /* Sorted sets data type */
 739 zskiplist *zslCreate(void);
 740 void zslFree(zskiplist *zsl);
 741 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
 742
 743 /* Core functions */
 744 void freeMemoryIfNeeded(void);
 745 int processCommand(redisClient *c);
 746 void setupSigSegvAction(void);
 747 struct redisCommand *lookupCommand(sds name);
 748 struct redisCommand *lookupCommandByCString(char *s);
 749 void call(redisClient *c, struct redisCommand *cmd);
 750 int prepareForShutdown();
 751 void redisLog(int level, const char *fmt, ...);
 752 void usage();
 753 void updateDictResizePolicy(void);
 754 int htNeedsResize(dict *dict);
 755 void oom(const char *msg);
 756 void populateCommandTable(void);
 757
 758 /* Virtual Memory */
 759 void vmInit(void);
 760 void vmMarkPagesFree(off_t page, off_t count);
 761 robj *vmLoadObject(robj *o);
 762 robj *vmPreviewObject(robj *o);
 763 int vmSwapOneObjectBlocking(void);
 764 int vmSwapOneObjectThreaded(void);
 765 int vmCanSwapOut(void);
 766 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
 767 void vmCancelThreadedIOJob(robj *o);
 768 void lockThreadedIO(void);
 769 void unlockThreadedIO(void);
 770 int vmSwapObjectThreaded(robj *key, robj *val, redisDb *db);
 771 void freeIOJob(iojob *j);
 772 void queueIOJob(iojob *j);
 773 int vmWriteObjectOnSwap(robj *o, off_t page);
 774 robj *vmReadObjectFromSwap(off_t page, int type);
 775 void waitEmptyIOJobsQueue(void);
 776 void vmReopenSwapFile(void);
 777 int vmFreePage(off_t page);
 778 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 779 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 780 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
 781 int dontWaitForSwappedKey(redisClient *c, robj *key);
 782 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
 783 vmpointer *vmSwapObjectBlocking(robj *val);
 784
 785 /* Set data type */
 786 robj *setTypeCreate(robj *value);
 787 int setTypeAdd(robj *subject, robj *value);
 788 int setTypeRemove(robj *subject, robj *value);
 789 int setTypeIsMember(robj *subject, robj *value);
 790 setTypeIterator *setTypeInitIterator(robj *subject);
 791 void setTypeReleaseIterator(setTypeIterator *si);
 792 robj *setTypeNext(setTypeIterator *si);
 793 robj *setTypeRandomElement(robj *subject);
 794 unsigned long setTypeSize(robj *subject);
 795 void setTypeConvert(robj *subject, int enc);
 796
 797 /* Hash data type */
 798 void convertToRealHash(robj *o);
 799 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
 800 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
 801 robj *hashTypeGet(robj *o, robj *key);
 802 int hashTypeExists(robj *o, robj *key);
 803 int hashTypeSet(robj *o, robj *key, robj *value);
 804 int hashTypeDelete(robj *o, robj *key);
 805 unsigned long hashTypeLength(robj *o);
 806 hashTypeIterator *hashTypeInitIterator(robj *subject);
 807 void hashTypeReleaseIterator(hashTypeIterator *hi);
 808 int hashTypeNext(hashTypeIterator *hi);
 809 robj *hashTypeCurrent(hashTypeIterator *hi, int what);
 810 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
 811
 812 /* Pub / Sub */
 813 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
 814 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
 815 void freePubsubPattern(void *p);
 816 int listMatchPubsubPattern(void *a, void *b);
 817
 818 /* Utility functions */
 819 int stringmatchlen(const char *pattern, int patternLen,
 820         const char *string, int stringLen, int nocase);
 821 int stringmatch(const char *pattern, const char *string, int nocase);
 822 long long memtoll(const char *p, int *err);
 823 int ll2string(char *s, size_t len, long long value);
 824 int isStringRepresentableAsLong(sds s, long *longval);
 825 int isStringRepresentableAsLongLong(sds s, long long *longval);
 826 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
 827
 828 /* Configuration */
 829 void loadServerConfig(char *filename);
 830 void appendServerSaveParams(time_t seconds, int changes);
 831 void resetServerSaveParams();
 832
 833 /* db.c -- Keyspace access API */
 834 int removeExpire(redisDb *db, robj *key);
 835 void propagateExpire(redisDb *db, robj *key);
 836 int expireIfNeeded(redisDb *db, robj *key);
 837 time_t getExpire(redisDb *db, robj *key);
 838 void setExpire(redisDb *db, robj *key, time_t when);
 839 robj *lookupKey(redisDb *db, robj *key);
 840 robj *lookupKeyRead(redisDb *db, robj *key);
 841 robj *lookupKeyWrite(redisDb *db, robj *key);
 842 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
 843 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
 844 int dbAdd(redisDb *db, robj *key, robj *val);
 845 int dbReplace(redisDb *db, robj *key, robj *val);
 846 int dbExists(redisDb *db, robj *key);
 847 robj *dbRandomKey(redisDb *db);
 848 int dbDelete(redisDb *db, robj *key);
 849 long long emptyDb();
 850 int selectDb(redisClient *c, int id);
 851
 852 /* Git SHA1 */
 853 char *redisGitSHA1(void);
 854 char *redisGitDirty(void);
 855
 856 /* Commands prototypes */
 857 void authCommand(redisClient *c);
 858 void pingCommand(redisClient *c);
 859 void echoCommand(redisClient *c);
 860 void setCommand(redisClient *c);
 861 void setnxCommand(redisClient *c);
 862 void setexCommand(redisClient *c);
 863 void getCommand(redisClient *c);
 864 void delCommand(redisClient *c);
 865 void existsCommand(redisClient *c);
 866 void incrCommand(redisClient *c);
 867 void decrCommand(redisClient *c);
 868 void incrbyCommand(redisClient *c);
 869 void decrbyCommand(redisClient *c);
 870 void selectCommand(redisClient *c);
 871 void randomkeyCommand(redisClient *c);
 872 void keysCommand(redisClient *c);
 873 void dbsizeCommand(redisClient *c);
 874 void lastsaveCommand(redisClient *c);
 875 void saveCommand(redisClient *c);
 876 void bgsaveCommand(redisClient *c);
 877 void bgrewriteaofCommand(redisClient *c);
 878 void shutdownCommand(redisClient *c);
 879 void moveCommand(redisClient *c);
 880 void renameCommand(redisClient *c);
 881 void renamenxCommand(redisClient *c);
 882 void lpushCommand(redisClient *c);
 883 void rpushCommand(redisClient *c);
 884 void lpushxCommand(redisClient *c);
 885 void rpushxCommand(redisClient *c);
 886 void linsertCommand(redisClient *c);
 887 void lpopCommand(redisClient *c);
 888 void rpopCommand(redisClient *c);
 889 void llenCommand(redisClient *c);
 890 void lindexCommand(redisClient *c);
 891 void lrangeCommand(redisClient *c);
 892 void ltrimCommand(redisClient *c);
 893 void typeCommand(redisClient *c);
 894 void lsetCommand(redisClient *c);
 895 void saddCommand(redisClient *c);
 896 void sremCommand(redisClient *c);
 897 void smoveCommand(redisClient *c);
 898 void sismemberCommand(redisClient *c);
 899 void scardCommand(redisClient *c);
 900 void spopCommand(redisClient *c);
 901 void srandmemberCommand(redisClient *c);
 902 void sinterCommand(redisClient *c);
 903 void sinterstoreCommand(redisClient *c);
 904 void sunionCommand(redisClient *c);
 905 void sunionstoreCommand(redisClient *c);
 906 void sdiffCommand(redisClient *c);
 907 void sdiffstoreCommand(redisClient *c);
 908 void syncCommand(redisClient *c);
 909 void flushdbCommand(redisClient *c);
 910 void flushallCommand(redisClient *c);
 911 void sortCommand(redisClient *c);
 912 void lremCommand(redisClient *c);
 913 void rpoplpushcommand(redisClient *c);
 914 void infoCommand(redisClient *c);
 915 void mgetCommand(redisClient *c);
 916 void monitorCommand(redisClient *c);
 917 void expireCommand(redisClient *c);
 918 void expireatCommand(redisClient *c);
 919 void getsetCommand(redisClient *c);
 920 void ttlCommand(redisClient *c);
 921 void persistCommand(redisClient *c);
 922 void slaveofCommand(redisClient *c);
 923 void debugCommand(redisClient *c);
 924 void msetCommand(redisClient *c);
 925 void msetnxCommand(redisClient *c);
 926 void zaddCommand(redisClient *c);
 927 void zincrbyCommand(redisClient *c);
 928 void zrangeCommand(redisClient *c);
 929 void zrangebyscoreCommand(redisClient *c);
 930 void zrevrangebyscoreCommand(redisClient *c);
 931 void zcountCommand(redisClient *c);
 932 void zrevrangeCommand(redisClient *c);
 933 void zcardCommand(redisClient *c);
 934 void zremCommand(redisClient *c);
 935 void zscoreCommand(redisClient *c);
 936 void zremrangebyscoreCommand(redisClient *c);
 937 void multiCommand(redisClient *c);
 938 void execCommand(redisClient *c);
 939 void discardCommand(redisClient *c);
 940 void blpopCommand(redisClient *c);
 941 void brpopCommand(redisClient *c);
 942 void appendCommand(redisClient *c);
 943 void substrCommand(redisClient *c);
 944 void strlenCommand(redisClient *c);
 945 void zrankCommand(redisClient *c);
 946 void zrevrankCommand(redisClient *c);
 947 void hsetCommand(redisClient *c);
 948 void hsetnxCommand(redisClient *c);
 949 void hgetCommand(redisClient *c);
 950 void hmsetCommand(redisClient *c);
 951 void hmgetCommand(redisClient *c);
 952 void hdelCommand(redisClient *c);
 953 void hlenCommand(redisClient *c);
 954 void zremrangebyrankCommand(redisClient *c);
 955 void zunionstoreCommand(redisClient *c);
 956 void zinterstoreCommand(redisClient *c);
 957 void hkeysCommand(redisClient *c);
 958 void hvalsCommand(redisClient *c);
 959 void hgetallCommand(redisClient *c);
 960 void hexistsCommand(redisClient *c);
 961 void configCommand(redisClient *c);
 962 void hincrbyCommand(redisClient *c);
 963 void subscribeCommand(redisClient *c);
 964 void unsubscribeCommand(redisClient *c);
 965 void psubscribeCommand(redisClient *c);
 966 void punsubscribeCommand(redisClient *c);
 967 void publishCommand(redisClient *c);
 968 void watchCommand(redisClient *c);
 969 void unwatchCommand(redisClient *c);
 970
 971 #if defined(__GNUC__)
 972 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
 973 void free(void *ptr) __attribute__ ((deprecated));
 974 void *malloc(size_t size) __attribute__ ((deprecated));
 975 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
 976 #endif
 977
 978 #endif