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_OBJFREELIST_MAX   1000000 /* Max number of objects to cache */
  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
  52 /* If more then REDIS_WRITEV_THRESHOLD write packets are pending use writev */
  53 #define REDIS_WRITEV_THRESHOLD      3
  54 /* Max number of iovecs used for each writev call */
  55 #define REDIS_WRITEV_IOVEC_COUNT    256
  56
  57 /* Hash table parameters */
  58 #define REDIS_HT_MINFILL        10      /* Minimal hash table fill 10% */
  59
  60 /* Command flags:
  61  *   REDIS_CMD_DENYOOM:
  62  *     Commands marked with this flag will return an error when 'maxmemory' is
  63  *     set and the server is using more than 'maxmemory' bytes of memory.
  64  *     In short: commands with this flag are denied on low memory conditions.
  65  *   REDIS_CMD_FORCE_REPLICATION:
  66  *     Force replication even if dirty is 0. */
  67 #define REDIS_CMD_DENYOOM 4
  68 #define REDIS_CMD_FORCE_REPLICATION 8
  69
  70 /* Object types */
  71 #define REDIS_STRING 0
  72 #define REDIS_LIST 1
  73 #define REDIS_SET 2
  74 #define REDIS_ZSET 3
  75 #define REDIS_HASH 4
  76 #define REDIS_VMPOINTER 8
  77
  78 /* Objects encoding. Some kind of objects like Strings and Hashes can be
  79  * internally represented in multiple ways. The 'encoding' field of the object
  80  * is set to one of this fields for this object. */
  81 #define REDIS_ENCODING_RAW 0     /* Raw representation */
  82 #define REDIS_ENCODING_INT 1     /* Encoded as integer */
  83 #define REDIS_ENCODING_HT 2      /* Encoded as hash table */
  84 #define REDIS_ENCODING_ZIPMAP 3  /* Encoded as zipmap */
  85 #define REDIS_ENCODING_LINKEDLIST 4 /* Encoded as regular linked list */
  86 #define REDIS_ENCODING_ZIPLIST 5 /* Encoded as ziplist */
  87 #define REDIS_ENCODING_INTSET 6  /* Encoded as intset */
  88
  89 /* Object types only used for dumping to disk */
  90 #define REDIS_EXPIRETIME 253
  91 #define REDIS_SELECTDB 254
  92 #define REDIS_EOF 255
  93
  94 /* Defines related to the dump file format. To store 32 bits lengths for short
  95  * keys requires a lot of space, so we check the most significant 2 bits of
  96  * the first byte to interpreter the length:
  97  *
  98  * 00|000000 => if the two MSB are 00 the len is the 6 bits of this byte
  99  * 01|000000 00000000 =>  01, the len is 14 byes, 6 bits + 8 bits of next byte
 100  * 10|000000 [32 bit integer] => if it's 01, a full 32 bit len will follow
 101  * 11|000000 this means: specially encoded object will follow. The six bits
 102  *           number specify the kind of object that follows.
 103  *           See the REDIS_RDB_ENC_* defines.
 104  *
 105  * Lenghts up to 63 are stored using a single byte, most DB keys, and may
 106  * values, will fit inside. */
 107 #define REDIS_RDB_6BITLEN 0
 108 #define REDIS_RDB_14BITLEN 1
 109 #define REDIS_RDB_32BITLEN 2
 110 #define REDIS_RDB_ENCVAL 3
 111 #define REDIS_RDB_LENERR UINT_MAX
 112
 113 /* When a length of a string object stored on disk has the first two bits
 114  * set, the remaining two bits specify a special encoding for the object
 115  * accordingly to the following defines: */
 116 #define REDIS_RDB_ENC_INT8 0        /* 8 bit signed integer */
 117 #define REDIS_RDB_ENC_INT16 1       /* 16 bit signed integer */
 118 #define REDIS_RDB_ENC_INT32 2       /* 32 bit signed integer */
 119 #define REDIS_RDB_ENC_LZF 3         /* string compressed with FASTLZ */
 120
 121 /* Virtual memory object->where field. */
 122 #define REDIS_VM_MEMORY 0       /* The object is on memory */
 123 #define REDIS_VM_SWAPPED 1      /* The object is on disk */
 124 #define REDIS_VM_SWAPPING 2     /* Redis is swapping this object on disk */
 125 #define REDIS_VM_LOADING 3      /* Redis is loading this object from disk */
 126
 127 /* Virtual memory static configuration stuff.
 128  * Check vmFindContiguousPages() to know more about this magic numbers. */
 129 #define REDIS_VM_MAX_NEAR_PAGES 65536
 130 #define REDIS_VM_MAX_RANDOM_JUMP 4096
 131 #define REDIS_VM_MAX_THREADS 32
 132 #define REDIS_THREAD_STACK_SIZE (1024*1024*4)
 133 /* The following is the *percentage* of completed I/O jobs to process when the
 134  * handelr is called. While Virtual Memory I/O operations are performed by
 135  * threads, this operations must be processed by the main thread when completed
 136  * in order to take effect. */
 137 #define REDIS_MAX_COMPLETED_JOBS_PROCESSED 1
 138
 139 /* Client flags */
 140 #define REDIS_SLAVE 1       /* This client is a slave server */
 141 #define REDIS_MASTER 2      /* This client is a master server */
 142 #define REDIS_MONITOR 4     /* This client is a slave monitor, see MONITOR */
 143 #define REDIS_MULTI 8       /* This client is in a MULTI context */
 144 #define REDIS_BLOCKED 16    /* The client is waiting in a blocking operation */
 145 #define REDIS_IO_WAIT 32    /* The client is waiting for Virtual Memory I/O */
 146 #define REDIS_DIRTY_CAS 64  /* Watched keys modified. EXEC will fail. */
 147 #define REDIS_CLOSE_AFTER_REPLY 128 /* Close after writing entire reply. */
 148
 149 /* Client request types */
 150 #define REDIS_REQ_INLINE 1
 151 #define REDIS_REQ_MULTIBULK 2
 152
 153 /* Slave replication state - slave side */
 154 #define REDIS_REPL_NONE 0   /* No active replication */
 155 #define REDIS_REPL_CONNECT 1    /* Must connect to master */
 156 #define REDIS_REPL_CONNECTED 2  /* 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
 213 /* We can print the stacktrace, so our assert is defined this way: */
 214 #define redisAssert(_e) ((_e)?(void)0 : (_redisAssert(#_e,__FILE__,__LINE__),_exit(1)))
 215 #define redisPanic(_e) _redisPanic(#_e,__FILE__,__LINE__),_exit(1)
 216 void _redisAssert(char *estr, char *file, int line);
 217 void _redisPanic(char *msg, char *file, int line);
 218
 219 /*-----------------------------------------------------------------------------
 220  * Data types
 221  *----------------------------------------------------------------------------*/
 222
 223 /* A redis object, that is a type able to hold a string / list / set */
 224
 225 /* The actual Redis Object */
 226 #define REDIS_LRU_CLOCK_MAX ((1<<21)-1) /* Max value of obj->lru */
 227 #define REDIS_LRU_CLOCK_RESOLUTION 10 /* LRU clock resolution in seconds */
 228 typedef struct redisObject {
 229     unsigned type:4;
 230     unsigned storage:2;     /* REDIS_VM_MEMORY or REDIS_VM_SWAPPING */
 231     unsigned encoding:4;
 232     unsigned lru:22;        /* lru time (relative to server.lruclock) */
 233     int refcount;
 234     void *ptr;
 235     /* VM fields are only allocated if VM is active, otherwise the
 236      * object allocation function will just allocate
 237      * sizeof(redisObjct) minus sizeof(redisObjectVM), so using
 238      * Redis without VM active will not have any overhead. */
 239 } robj;
 240
 241 /* The VM pointer structure - identifies an object in the swap file.
 242  *
 243  * This object is stored in place of the value
 244  * object in the main key->value hash table representing a database.
 245  * Note that the first fields (type, storage) are the same as the redisObject
 246  * structure so that vmPointer strucuters can be accessed even when casted
 247  * as redisObject structures.
 248  *
 249  * This is useful as we don't know if a value object is or not on disk, but we
 250  * are always able to read obj->storage to check this. For vmPointer
 251  * structures "type" is set to REDIS_VMPOINTER (even if without this field
 252  * is still possible to check the kind of object from the value of 'storage').*/
 253 typedef struct vmPointer {
 254     unsigned type:4;
 255     unsigned storage:2; /* REDIS_VM_SWAPPED or REDIS_VM_LOADING */
 256     unsigned notused:26;
 257     unsigned int vtype; /* type of the object stored in the swap file */
 258     off_t page;         /* the page at witch the object is stored on disk */
 259     off_t usedpages;    /* number of pages used on disk */
 260 } vmpointer;
 261
 262 /* Macro used to initalize a Redis object allocated on the stack.
 263  * Note that this macro is taken near the structure definition to make sure
 264  * we'll update it when the structure is changed, to avoid bugs like
 265  * bug #85 introduced exactly in this way. */
 266 #define initStaticStringObject(_var,_ptr) do { \
 267     _var.refcount = 1; \
 268     _var.type = REDIS_STRING; \
 269     _var.encoding = REDIS_ENCODING_RAW; \
 270     _var.ptr = _ptr; \
 271     _var.storage = REDIS_VM_MEMORY; \
 272 } while(0);
 273
 274 typedef struct redisDb {
 275     dict *dict;                 /* The keyspace for this DB */
 276     dict *expires;              /* Timeout of keys with a timeout set */
 277     dict *blocking_keys;        /* Keys with clients waiting for data (BLPOP) */
 278     dict *io_keys;              /* Keys with clients waiting for VM I/O */
 279     dict *watched_keys;         /* WATCHED keys for MULTI/EXEC CAS */
 280     int id;
 281 } redisDb;
 282
 283 /* Client MULTI/EXEC state */
 284 typedef struct multiCmd {
 285     robj **argv;
 286     int argc;
 287     struct redisCommand *cmd;
 288 } multiCmd;
 289
 290 typedef struct multiState {
 291     multiCmd *commands;     /* Array of MULTI commands */
 292     int count;              /* Total number of MULTI commands */
 293 } multiState;
 294
 295 /* With multiplexing we need to take per-clinet state.
 296  * Clients are taken in a liked list. */
 297 typedef struct redisClient {
 298     int fd;
 299     redisDb *db;
 300     int dictid;
 301     sds querybuf;
 302     int argc;
 303     robj **argv;
 304     int reqtype;
 305     int multibulklen;       /* number of multi bulk arguments left to read */
 306     long bulklen;           /* length of bulk argument in multi bulk request */
 307     list *reply;
 308     int sentlen;
 309     time_t lastinteraction; /* time of the last interaction, used for timeout */
 310     int flags;              /* REDIS_SLAVE | REDIS_MONITOR | REDIS_MULTI ... */
 311     int slaveseldb;         /* slave selected db, if this client is a slave */
 312     int authenticated;      /* when requirepass is non-NULL */
 313     int replstate;          /* replication state if this is a slave */
 314     int repldbfd;           /* replication DB file descriptor */
 315     long repldboff;         /* replication DB file offset */
 316     off_t repldbsize;       /* replication DB file size */
 317     multiState mstate;      /* MULTI/EXEC state */
 318     robj **blocking_keys;   /* The key we are waiting to terminate a blocking
 319                              * operation such as BLPOP. Otherwise NULL. */
 320     int blocking_keys_num;  /* Number of blocking keys */
 321     time_t blockingto;      /* Blocking operation timeout. If UNIX current time
 322                              * is >= blockingto then the operation timed out. */
 323     list *io_keys;          /* Keys this client is waiting to be loaded from the
 324                              * swap file in order to continue. */
 325     list *watched_keys;     /* Keys WATCHED for MULTI/EXEC CAS */
 326     dict *pubsub_channels;  /* channels a client is interested in (SUBSCRIBE) */
 327     list *pubsub_patterns;  /* patterns a client is interested in (SUBSCRIBE) */
 328
 329     /* Response buffer */
 330     int bufpos;
 331     char buf[REDIS_REPLY_CHUNK_BYTES];
 332 } redisClient;
 333
 334 struct saveparam {
 335     time_t seconds;
 336     int changes;
 337 };
 338
 339 struct sharedObjectsStruct {
 340     robj *crlf, *ok, *err, *emptybulk, *czero, *cone, *cnegone, *pong, *space,
 341     *colon, *nullbulk, *nullmultibulk, *queued,
 342     *emptymultibulk, *wrongtypeerr, *nokeyerr, *syntaxerr, *sameobjecterr,
 343     *outofrangeerr, *plus,
 344     *select0, *select1, *select2, *select3, *select4,
 345     *select5, *select6, *select7, *select8, *select9,
 346     *messagebulk, *pmessagebulk, *subscribebulk, *unsubscribebulk, *mbulk3,
 347     *mbulk4, *psubscribebulk, *punsubscribebulk,
 348     *integers[REDIS_SHARED_INTEGERS];
 349 };
 350
 351 /* Global server state structure */
 352 struct redisServer {
 353     pthread_t mainthread;
 354     int port;
 355     int fd;
 356     redisDb *db;
 357     long long dirty;            /* changes to DB from the last save */
 358     long long dirty_before_bgsave; /* used to restore dirty on failed BGSAVE */
 359     list *clients;
 360     list *slaves, *monitors;
 361     char neterr[ANET_ERR_LEN];
 362     aeEventLoop *el;
 363     int cronloops;              /* number of times the cron function run */
 364     list *objfreelist;          /* A list of freed objects to avoid malloc() */
 365     time_t lastsave;                /* Unix time of last save succeeede */
 366     /* Fields used only for stats */
 367     time_t stat_starttime;          /* server start time */
 368     long long stat_numcommands;     /* number of processed commands */
 369     long long stat_numconnections;  /* number of connections received */
 370     long long stat_expiredkeys;     /* number of expired keys */
 371     long long stat_keyspace_hits;   /* number of successful lookups of keys */
 372     long long stat_keyspace_misses; /* number of failed lookups of keys */
 373     /* Configuration */
 374     int verbosity;
 375     int glueoutputbuf;
 376     int maxidletime;
 377     int dbnum;
 378     int daemonize;
 379     int appendonly;
 380     int appendfsync;
 381     int no_appendfsync_on_rewrite;
 382     int shutdown_asap;
 383     time_t lastfsync;
 384     int appendfd;
 385     int appendseldb;
 386     char *pidfile;
 387     pid_t bgsavechildpid;
 388     pid_t bgrewritechildpid;
 389     sds bgrewritebuf; /* buffer taken by parent during oppend only rewrite */
 390     sds aofbuf;       /* AOF buffer, written before entering the event loop */
 391     struct saveparam *saveparams;
 392     int saveparamslen;
 393     char *logfile;
 394     char *bindaddr;
 395     char *dbfilename;
 396     char *appendfilename;
 397     char *requirepass;
 398     int rdbcompression;
 399     int activerehashing;
 400     /* Replication related */
 401     int isslave;
 402     char *masterauth;
 403     char *masterhost;
 404     int masterport;
 405     redisClient *master;    /* client that is master for this slave */
 406     int replstate;
 407     unsigned int maxclients;
 408     unsigned long long maxmemory;
 409     int maxmemory_policy;
 410     int maxmemory_samples;
 411     unsigned int blpop_blocked_clients;
 412     unsigned int vm_blocked_clients;
 413     /* Sort parameters - qsort_r() is only available under BSD so we
 414      * have to take this state global, in order to pass it to sortCompare() */
 415     int sort_desc;
 416     int sort_alpha;
 417     int sort_bypattern;
 418     /* Virtual memory configuration */
 419     int vm_enabled;
 420     char *vm_swap_file;
 421     off_t vm_page_size;
 422     off_t vm_pages;
 423     unsigned long long vm_max_memory;
 424     /* Zip structure config */
 425     size_t hash_max_zipmap_entries;
 426     size_t hash_max_zipmap_value;
 427     size_t list_max_ziplist_entries;
 428     size_t list_max_ziplist_value;
 429     size_t set_max_intset_entries;
 430     /* Virtual memory state */
 431     FILE *vm_fp;
 432     int vm_fd;
 433     off_t vm_next_page; /* Next probably empty page */
 434     off_t vm_near_pages; /* Number of pages allocated sequentially */
 435     unsigned char *vm_bitmap; /* Bitmap of free/used pages */
 436     time_t unixtime;    /* Unix time sampled every second. */
 437     /* Virtual memory I/O threads stuff */
 438     /* An I/O thread process an element taken from the io_jobs queue and
 439      * put the result of the operation in the io_done list. While the
 440      * job is being processed, it's put on io_processing queue. */
 441     list *io_newjobs; /* List of VM I/O jobs yet to be processed */
 442     list *io_processing; /* List of VM I/O jobs being processed */
 443     list *io_processed; /* List of VM I/O jobs already processed */
 444     list *io_ready_clients; /* Clients ready to be unblocked. All keys loaded */
 445     pthread_mutex_t io_mutex; /* lock to access io_jobs/io_done/io_thread_job */
 446     pthread_mutex_t obj_freelist_mutex; /* safe redis objects creation/free */
 447     pthread_mutex_t io_swapfile_mutex; /* So we can lseek + write */
 448     pthread_attr_t io_threads_attr; /* attributes for threads creation */
 449     int io_active_threads; /* Number of running I/O threads */
 450     int vm_max_threads; /* Max number of I/O threads running at the same time */
 451     /* Our main thread is blocked on the event loop, locking for sockets ready
 452      * to be read or written, so when a threaded I/O operation is ready to be
 453      * processed by the main thread, the I/O thread will use a unix pipe to
 454      * awake the main thread. The followings are the two pipe FDs. */
 455     int io_ready_pipe_read;
 456     int io_ready_pipe_write;
 457     /* Virtual memory stats */
 458     unsigned long long vm_stats_used_pages;
 459     unsigned long long vm_stats_swapped_objects;
 460     unsigned long long vm_stats_swapouts;
 461     unsigned long long vm_stats_swapins;
 462     /* Pubsub */
 463     dict *pubsub_channels; /* Map channels to list of subscribed clients */
 464     list *pubsub_patterns; /* A list of pubsub_patterns */
 465     /* Misc */
 466     FILE *devnull;
 467     unsigned lruclock:22;        /* clock incrementing every minute, for LRU */
 468     unsigned lruclock_padding:10;
 469 };
 470
 471 typedef struct pubsubPattern {
 472     redisClient *client;
 473     robj *pattern;
 474 } pubsubPattern;
 475
 476 typedef void redisCommandProc(redisClient *c);
 477 typedef void redisVmPreloadProc(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 478 struct redisCommand {
 479     char *name;
 480     redisCommandProc *proc;
 481     int arity;
 482     int flags;
 483     /* Use a function to determine which keys need to be loaded
 484      * in the background prior to executing this command. Takes precedence
 485      * over vm_firstkey and others, ignored when NULL */
 486     redisVmPreloadProc *vm_preload_proc;
 487     /* What keys should be loaded in background when calling this command? */
 488     int vm_firstkey; /* The first argument that's a key (0 = no keys) */
 489     int vm_lastkey;  /* THe last argument that's a key */
 490     int vm_keystep;  /* The step between first and last key */
 491 };
 492
 493 struct redisFunctionSym {
 494     char *name;
 495     unsigned long pointer;
 496 };
 497
 498 typedef struct _redisSortObject {
 499     robj *obj;
 500     union {
 501         double score;
 502         robj *cmpobj;
 503     } u;
 504 } redisSortObject;
 505
 506 typedef struct _redisSortOperation {
 507     int type;
 508     robj *pattern;
 509 } redisSortOperation;
 510
 511 /* ZSETs use a specialized version of Skiplists */
 512 typedef struct zskiplistNode {
 513     robj *obj;
 514     double score;
 515     struct zskiplistNode *backward;
 516     struct zskiplistLevel {
 517         struct zskiplistNode *forward;
 518         unsigned int span;
 519     } level[];
 520 } zskiplistNode;
 521
 522 typedef struct zskiplist {
 523     struct zskiplistNode *header, *tail;
 524     unsigned long length;
 525     int level;
 526 } zskiplist;
 527
 528 typedef struct zset {
 529     dict *dict;
 530     zskiplist *zsl;
 531 } zset;
 532
 533 /* VM threaded I/O request message */
 534 #define REDIS_IOJOB_LOAD 0          /* Load from disk to memory */
 535 #define REDIS_IOJOB_PREPARE_SWAP 1  /* Compute needed pages */
 536 #define REDIS_IOJOB_DO_SWAP 2       /* Swap from memory to disk */
 537 typedef struct iojob {
 538     int type;   /* Request type, REDIS_IOJOB_* */
 539     redisDb *db;/* Redis database */
 540     robj *key;  /* This I/O request is about swapping this key */
 541     robj *id;   /* Unique identifier of this job:
 542                    this is the object to swap for REDIS_IOREQ_*_SWAP, or the
 543                    vmpointer objct for REDIS_IOREQ_LOAD. */
 544     robj *val;  /* the value to swap for REDIS_IOREQ_*_SWAP, otherwise this
 545                  * field is populated by the I/O thread for REDIS_IOREQ_LOAD. */
 546     off_t page; /* Swap page where to read/write the object */
 547     off_t pages; /* Swap pages needed to save object. PREPARE_SWAP return val */
 548     int canceled; /* True if this command was canceled by blocking side of VM */
 549     pthread_t thread; /* ID of the thread processing this entry */
 550 } iojob;
 551
 552 /* Structure to hold list iteration abstraction. */
 553 typedef struct {
 554     robj *subject;
 555     unsigned char encoding;
 556     unsigned char direction; /* Iteration direction */
 557     unsigned char *zi;
 558     listNode *ln;
 559 } listTypeIterator;
 560
 561 /* Structure for an entry while iterating over a list. */
 562 typedef struct {
 563     listTypeIterator *li;
 564     unsigned char *zi;  /* Entry in ziplist */
 565     listNode *ln;       /* Entry in linked list */
 566 } listTypeEntry;
 567
 568 /* Structure to hold set iteration abstraction. */
 569 typedef struct {
 570     robj *subject;
 571     int encoding;
 572     int ii; /* intset iterator */
 573     dictIterator *di;
 574 } setTypeIterator;
 575
 576 /* Structure to hold hash iteration abstration. Note that iteration over
 577  * hashes involves both fields and values. Because it is possible that
 578  * not both are required, store pointers in the iterator to avoid
 579  * unnecessary memory allocation for fields/values. */
 580 typedef struct {
 581     int encoding;
 582     unsigned char *zi;
 583     unsigned char *zk, *zv;
 584     unsigned int zklen, zvlen;
 585
 586     dictIterator *di;
 587     dictEntry *de;
 588 } hashTypeIterator;
 589
 590 #define REDIS_HASH_KEY 1
 591 #define REDIS_HASH_VALUE 2
 592
 593 /*-----------------------------------------------------------------------------
 594  * Extern declarations
 595  *----------------------------------------------------------------------------*/
 596
 597 extern struct redisServer server;
 598 extern struct sharedObjectsStruct shared;
 599 extern dictType setDictType;
 600 extern dictType zsetDictType;
 601 extern double R_Zero, R_PosInf, R_NegInf, R_Nan;
 602 dictType hashDictType;
 603
 604 /*-----------------------------------------------------------------------------
 605  * Functions prototypes
 606  *----------------------------------------------------------------------------*/
 607
 608 /* networking.c -- Networking and Client related operations */
 609 redisClient *createClient(int fd);
 610 void closeTimedoutClients(void);
 611 void freeClient(redisClient *c);
 612 void resetClient(redisClient *c);
 613 void sendReplyToClient(aeEventLoop *el, int fd, void *privdata, int mask);
 614 void sendReplyToClientWritev(aeEventLoop *el, int fd, void *privdata, int mask);
 615 void addReply(redisClient *c, robj *obj);
 616 void *addDeferredMultiBulkLength(redisClient *c);
 617 void setDeferredMultiBulkLength(redisClient *c, void *node, long length);
 618 void addReplySds(redisClient *c, sds s);
 619 void processInputBuffer(redisClient *c);
 620 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 621 void readQueryFromClient(aeEventLoop *el, int fd, void *privdata, int mask);
 622 void addReplyBulk(redisClient *c, robj *obj);
 623 void addReplyBulkCString(redisClient *c, char *s);
 624 void acceptHandler(aeEventLoop *el, int fd, void *privdata, int mask);
 625 void addReply(redisClient *c, robj *obj);
 626 void addReplySds(redisClient *c, sds s);
 627 void addReplyError(redisClient *c, char *err);
 628 void addReplyStatus(redisClient *c, char *status);
 629 void addReplyDouble(redisClient *c, double d);
 630 void addReplyLongLong(redisClient *c, long long ll);
 631 void addReplyMultiBulkLen(redisClient *c, long length);
 632 void *dupClientReplyValue(void *o);
 633
 634 #ifdef __GNUC__
 635 void addReplyErrorFormat(redisClient *c, const char *fmt, ...)
 636     __attribute__((format(printf, 2, 3)));
 637 void addReplyStatusFormat(redisClient *c, const char *fmt, ...)
 638     __attribute__((format(printf, 2, 3)));
 639 #else
 640 void addReplyErrorFormat(redisClient *c, const char *fmt, ...);
 641 void addReplyStatusFormat(redisClient *c, const char *fmt, ...);
 642 #endif
 643
 644 /* List data type */
 645 void listTypeTryConversion(robj *subject, robj *value);
 646 void listTypePush(robj *subject, robj *value, int where);
 647 robj *listTypePop(robj *subject, int where);
 648 unsigned long listTypeLength(robj *subject);
 649 listTypeIterator *listTypeInitIterator(robj *subject, int index, unsigned char direction);
 650 void listTypeReleaseIterator(listTypeIterator *li);
 651 int listTypeNext(listTypeIterator *li, listTypeEntry *entry);
 652 robj *listTypeGet(listTypeEntry *entry);
 653 void listTypeInsert(listTypeEntry *entry, robj *value, int where);
 654 int listTypeEqual(listTypeEntry *entry, robj *o);
 655 void listTypeDelete(listTypeEntry *entry);
 656 void listTypeConvert(robj *subject, int enc);
 657 void unblockClientWaitingData(redisClient *c);
 658 int handleClientsWaitingListPush(redisClient *c, robj *key, robj *ele);
 659 void popGenericCommand(redisClient *c, int where);
 660
 661 /* MULTI/EXEC/WATCH... */
 662 void unwatchAllKeys(redisClient *c);
 663 void initClientMultiState(redisClient *c);
 664 void freeClientMultiState(redisClient *c);
 665 void queueMultiCommand(redisClient *c, struct redisCommand *cmd);
 666 void touchWatchedKey(redisDb *db, robj *key);
 667 void touchWatchedKeysOnFlush(int dbid);
 668
 669 /* Redis object implementation */
 670 void decrRefCount(void *o);
 671 void incrRefCount(robj *o);
 672 void freeStringObject(robj *o);
 673 void freeListObject(robj *o);
 674 void freeSetObject(robj *o);
 675 void freeZsetObject(robj *o);
 676 void freeHashObject(robj *o);
 677 robj *createObject(int type, void *ptr);
 678 robj *createStringObject(char *ptr, size_t len);
 679 robj *dupStringObject(robj *o);
 680 robj *tryObjectEncoding(robj *o);
 681 robj *getDecodedObject(robj *o);
 682 size_t stringObjectLen(robj *o);
 683 int tryFreeOneObjectFromFreelist(void);
 684 robj *createStringObjectFromLongLong(long long value);
 685 robj *createListObject(void);
 686 robj *createZiplistObject(void);
 687 robj *createSetObject(void);
 688 robj *createIntsetObject(void);
 689 robj *createHashObject(void);
 690 robj *createZsetObject(void);
 691 int getLongFromObjectOrReply(redisClient *c, robj *o, long *target, const char *msg);
 692 int checkType(redisClient *c, robj *o, int type);
 693 int getLongLongFromObjectOrReply(redisClient *c, robj *o, long long *target, const char *msg);
 694 int getDoubleFromObjectOrReply(redisClient *c, robj *o, double *target, const char *msg);
 695 int getLongLongFromObject(robj *o, long long *target);
 696 char *strEncoding(int encoding);
 697 int compareStringObjects(robj *a, robj *b);
 698 int equalStringObjects(robj *a, robj *b);
 699 unsigned long estimateObjectIdleTime(robj *o);
 700
 701 /* Synchronous I/O with timeout */
 702 int syncWrite(int fd, char *ptr, ssize_t size, int timeout);
 703 int syncRead(int fd, char *ptr, ssize_t size, int timeout);
 704 int syncReadLine(int fd, char *ptr, ssize_t size, int timeout);
 705 int fwriteBulkString(FILE *fp, char *s, unsigned long len);
 706 int fwriteBulkDouble(FILE *fp, double d);
 707 int fwriteBulkLongLong(FILE *fp, long long l);
 708 int fwriteBulkObject(FILE *fp, robj *obj);
 709
 710 /* Replication */
 711 void replicationFeedSlaves(list *slaves, int dictid, robj **argv, int argc);
 712 void replicationFeedMonitors(list *monitors, int dictid, robj **argv, int argc);
 713 int syncWithMaster(void);
 714 void updateSlavesWaitingBgsave(int bgsaveerr);
 715
 716 /* RDB persistence */
 717 int rdbLoad(char *filename);
 718 int rdbSaveBackground(char *filename);
 719 void rdbRemoveTempFile(pid_t childpid);
 720 int rdbSave(char *filename);
 721 int rdbSaveObject(FILE *fp, robj *o);
 722 off_t rdbSavedObjectPages(robj *o, FILE *fp);
 723 off_t rdbSavedObjectLen(robj *o, FILE *fp);
 724 robj *rdbLoadObject(int type, FILE *fp);
 725 void backgroundSaveDoneHandler(int statloc);
 726
 727 /* AOF persistence */
 728 void flushAppendOnlyFile(void);
 729 void feedAppendOnlyFile(struct redisCommand *cmd, int dictid, robj **argv, int argc);
 730 void aofRemoveTempFile(pid_t childpid);
 731 int rewriteAppendOnlyFileBackground(void);
 732 int loadAppendOnlyFile(char *filename);
 733 void stopAppendOnly(void);
 734 int startAppendOnly(void);
 735 void backgroundRewriteDoneHandler(int statloc);
 736
 737 /* Sorted sets data type */
 738 zskiplist *zslCreate(void);
 739 void zslFree(zskiplist *zsl);
 740 zskiplistNode *zslInsert(zskiplist *zsl, double score, robj *obj);
 741
 742 /* Core functions */
 743 void freeMemoryIfNeeded(void);
 744 int processCommand(redisClient *c);
 745 void setupSigSegvAction(void);
 746 struct redisCommand *lookupCommand(char *name);
 747 void call(redisClient *c, struct redisCommand *cmd);
 748 int prepareForShutdown();
 749 void redisLog(int level, const char *fmt, ...);
 750 void usage();
 751 void updateDictResizePolicy(void);
 752 int htNeedsResize(dict *dict);
 753 void oom(const char *msg);
 754
 755 /* Virtual Memory */
 756 void vmInit(void);
 757 void vmMarkPagesFree(off_t page, off_t count);
 758 robj *vmLoadObject(robj *o);
 759 robj *vmPreviewObject(robj *o);
 760 int vmSwapOneObjectBlocking(void);
 761 int vmSwapOneObjectThreaded(void);
 762 int vmCanSwapOut(void);
 763 void vmThreadedIOCompletedJob(aeEventLoop *el, int fd, void *privdata, int mask);
 764 void vmCancelThreadedIOJob(robj *o);
 765 void lockThreadedIO(void);
 766 void unlockThreadedIO(void);
 767 int vmSwapObjectThreaded(robj *key, robj *val, redisDb *db);
 768 void freeIOJob(iojob *j);
 769 void queueIOJob(iojob *j);
 770 int vmWriteObjectOnSwap(robj *o, off_t page);
 771 robj *vmReadObjectFromSwap(off_t page, int type);
 772 void waitEmptyIOJobsQueue(void);
 773 void vmReopenSwapFile(void);
 774 int vmFreePage(off_t page);
 775 void zunionInterBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 776 void execBlockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd, int argc, robj **argv);
 777 int blockClientOnSwappedKeys(redisClient *c, struct redisCommand *cmd);
 778 int dontWaitForSwappedKey(redisClient *c, robj *key);
 779 void handleClientsBlockedOnSwappedKey(redisDb *db, robj *key);
 780 vmpointer *vmSwapObjectBlocking(robj *val);
 781
 782 /* Set data type */
 783 robj *setTypeCreate(robj *value);
 784 int setTypeAdd(robj *subject, robj *value);
 785 int setTypeRemove(robj *subject, robj *value);
 786 int setTypeIsMember(robj *subject, robj *value);
 787 setTypeIterator *setTypeInitIterator(robj *subject);
 788 void setTypeReleaseIterator(setTypeIterator *si);
 789 robj *setTypeNext(setTypeIterator *si);
 790 robj *setTypeRandomElement(robj *subject);
 791 unsigned long setTypeSize(robj *subject);
 792 void setTypeConvert(robj *subject, int enc);
 793
 794 /* Hash data type */
 795 void convertToRealHash(robj *o);
 796 void hashTypeTryConversion(robj *subject, robj **argv, int start, int end);
 797 void hashTypeTryObjectEncoding(robj *subject, robj **o1, robj **o2);
 798 robj *hashTypeGet(robj *o, robj *key);
 799 int hashTypeExists(robj *o, robj *key);
 800 int hashTypeSet(robj *o, robj *key, robj *value);
 801 int hashTypeDelete(robj *o, robj *key);
 802 unsigned long hashTypeLength(robj *o);
 803 hashTypeIterator *hashTypeInitIterator(robj *subject);
 804 void hashTypeReleaseIterator(hashTypeIterator *hi);
 805 int hashTypeNext(hashTypeIterator *hi);
 806 robj *hashTypeCurrent(hashTypeIterator *hi, int what);
 807 robj *hashTypeLookupWriteOrCreate(redisClient *c, robj *key);
 808
 809 /* Pub / Sub */
 810 int pubsubUnsubscribeAllChannels(redisClient *c, int notify);
 811 int pubsubUnsubscribeAllPatterns(redisClient *c, int notify);
 812 void freePubsubPattern(void *p);
 813 int listMatchPubsubPattern(void *a, void *b);
 814
 815 /* Utility functions */
 816 int stringmatchlen(const char *pattern, int patternLen,
 817         const char *string, int stringLen, int nocase);
 818 int stringmatch(const char *pattern, const char *string, int nocase);
 819 long long memtoll(const char *p, int *err);
 820 int ll2string(char *s, size_t len, long long value);
 821 int isStringRepresentableAsLong(sds s, long *longval);
 822 int isStringRepresentableAsLongLong(sds s, long long *longval);
 823 int isObjectRepresentableAsLongLong(robj *o, long long *llongval);
 824
 825 /* Configuration */
 826 void loadServerConfig(char *filename);
 827 void appendServerSaveParams(time_t seconds, int changes);
 828 void resetServerSaveParams();
 829
 830 /* db.c -- Keyspace access API */
 831 int removeExpire(redisDb *db, robj *key);
 832 void propagateExpire(redisDb *db, robj *key);
 833 int expireIfNeeded(redisDb *db, robj *key);
 834 time_t getExpire(redisDb *db, robj *key);
 835 void setExpire(redisDb *db, robj *key, time_t when);
 836 robj *lookupKey(redisDb *db, robj *key);
 837 robj *lookupKeyRead(redisDb *db, robj *key);
 838 robj *lookupKeyWrite(redisDb *db, robj *key);
 839 robj *lookupKeyReadOrReply(redisClient *c, robj *key, robj *reply);
 840 robj *lookupKeyWriteOrReply(redisClient *c, robj *key, robj *reply);
 841 int dbAdd(redisDb *db, robj *key, robj *val);
 842 int dbReplace(redisDb *db, robj *key, robj *val);
 843 int dbExists(redisDb *db, robj *key);
 844 robj *dbRandomKey(redisDb *db);
 845 int dbDelete(redisDb *db, robj *key);
 846 long long emptyDb();
 847 int selectDb(redisClient *c, int id);
 848
 849 /* Git SHA1 */
 850 char *redisGitSHA1(void);
 851 char *redisGitDirty(void);
 852
 853 /* Commands prototypes */
 854 void authCommand(redisClient *c);
 855 void pingCommand(redisClient *c);
 856 void echoCommand(redisClient *c);
 857 void setCommand(redisClient *c);
 858 void setnxCommand(redisClient *c);
 859 void setexCommand(redisClient *c);
 860 void getCommand(redisClient *c);
 861 void delCommand(redisClient *c);
 862 void existsCommand(redisClient *c);
 863 void incrCommand(redisClient *c);
 864 void decrCommand(redisClient *c);
 865 void incrbyCommand(redisClient *c);
 866 void decrbyCommand(redisClient *c);
 867 void selectCommand(redisClient *c);
 868 void randomkeyCommand(redisClient *c);
 869 void keysCommand(redisClient *c);
 870 void dbsizeCommand(redisClient *c);
 871 void lastsaveCommand(redisClient *c);
 872 void saveCommand(redisClient *c);
 873 void bgsaveCommand(redisClient *c);
 874 void bgrewriteaofCommand(redisClient *c);
 875 void shutdownCommand(redisClient *c);
 876 void moveCommand(redisClient *c);
 877 void renameCommand(redisClient *c);
 878 void renamenxCommand(redisClient *c);
 879 void lpushCommand(redisClient *c);
 880 void rpushCommand(redisClient *c);
 881 void lpushxCommand(redisClient *c);
 882 void rpushxCommand(redisClient *c);
 883 void linsertCommand(redisClient *c);
 884 void lpopCommand(redisClient *c);
 885 void rpopCommand(redisClient *c);
 886 void llenCommand(redisClient *c);
 887 void lindexCommand(redisClient *c);
 888 void lrangeCommand(redisClient *c);
 889 void ltrimCommand(redisClient *c);
 890 void typeCommand(redisClient *c);
 891 void lsetCommand(redisClient *c);
 892 void saddCommand(redisClient *c);
 893 void sremCommand(redisClient *c);
 894 void smoveCommand(redisClient *c);
 895 void sismemberCommand(redisClient *c);
 896 void scardCommand(redisClient *c);
 897 void spopCommand(redisClient *c);
 898 void srandmemberCommand(redisClient *c);
 899 void sinterCommand(redisClient *c);
 900 void sinterstoreCommand(redisClient *c);
 901 void sunionCommand(redisClient *c);
 902 void sunionstoreCommand(redisClient *c);
 903 void sdiffCommand(redisClient *c);
 904 void sdiffstoreCommand(redisClient *c);
 905 void syncCommand(redisClient *c);
 906 void flushdbCommand(redisClient *c);
 907 void flushallCommand(redisClient *c);
 908 void sortCommand(redisClient *c);
 909 void lremCommand(redisClient *c);
 910 void rpoplpushcommand(redisClient *c);
 911 void infoCommand(redisClient *c);
 912 void mgetCommand(redisClient *c);
 913 void monitorCommand(redisClient *c);
 914 void expireCommand(redisClient *c);
 915 void expireatCommand(redisClient *c);
 916 void getsetCommand(redisClient *c);
 917 void ttlCommand(redisClient *c);
 918 void persistCommand(redisClient *c);
 919 void slaveofCommand(redisClient *c);
 920 void debugCommand(redisClient *c);
 921 void msetCommand(redisClient *c);
 922 void msetnxCommand(redisClient *c);
 923 void zaddCommand(redisClient *c);
 924 void zincrbyCommand(redisClient *c);
 925 void zrangeCommand(redisClient *c);
 926 void zrangebyscoreCommand(redisClient *c);
 927 void zrevrangebyscoreCommand(redisClient *c);
 928 void zcountCommand(redisClient *c);
 929 void zrevrangeCommand(redisClient *c);
 930 void zcardCommand(redisClient *c);
 931 void zremCommand(redisClient *c);
 932 void zscoreCommand(redisClient *c);
 933 void zremrangebyscoreCommand(redisClient *c);
 934 void multiCommand(redisClient *c);
 935 void execCommand(redisClient *c);
 936 void discardCommand(redisClient *c);
 937 void blpopCommand(redisClient *c);
 938 void brpopCommand(redisClient *c);
 939 void appendCommand(redisClient *c);
 940 void substrCommand(redisClient *c);
 941 void strlenCommand(redisClient *c);
 942 void zrankCommand(redisClient *c);
 943 void zrevrankCommand(redisClient *c);
 944 void hsetCommand(redisClient *c);
 945 void hsetnxCommand(redisClient *c);
 946 void hgetCommand(redisClient *c);
 947 void hmsetCommand(redisClient *c);
 948 void hmgetCommand(redisClient *c);
 949 void hdelCommand(redisClient *c);
 950 void hlenCommand(redisClient *c);
 951 void zremrangebyrankCommand(redisClient *c);
 952 void zunionstoreCommand(redisClient *c);
 953 void zinterstoreCommand(redisClient *c);
 954 void hkeysCommand(redisClient *c);
 955 void hvalsCommand(redisClient *c);
 956 void hgetallCommand(redisClient *c);
 957 void hexistsCommand(redisClient *c);
 958 void configCommand(redisClient *c);
 959 void hincrbyCommand(redisClient *c);
 960 void subscribeCommand(redisClient *c);
 961 void unsubscribeCommand(redisClient *c);
 962 void psubscribeCommand(redisClient *c);
 963 void punsubscribeCommand(redisClient *c);
 964 void publishCommand(redisClient *c);
 965 void watchCommand(redisClient *c);
 966 void unwatchCommand(redisClient *c);
 967
 968 #if defined(__GNUC__)
 969 void *calloc(size_t count, size_t size) __attribute__ ((deprecated));
 970 void free(void *ptr) __attribute__ ((deprecated));
 971 void *malloc(size_t size) __attribute__ ((deprecated));
 972 void *realloc(void *ptr, size_t size) __attribute__ ((deprecated));
 973 #endif
 974
 975 #endif