db2/db_int.h

   1 /*-
   2  * See the file LICENSE for redistribution information.
   3  *
   4  * Copyright (c) 1996, 1997
   5  *      Sleepycat Software.  All rights reserved.
   6  *
   7  *      @(#)db_int.h.src        10.30 (Sleepycat) 9/23/97
   8  */
   9
  10 #ifndef _DB_INTERNAL_H_
  11 #define _DB_INTERNAL_H_
  12
  13 #include "db.h"                         /* Standard DB include file. */
  14 #include "queue.h"
  15 #include "os_ext.h"
  16
  17 /*******************************************************
  18  * General purpose constants and macros.
  19  *******************************************************/
  20 #define UINT32_T_MAX    0xffffffff      /* Maximum 32 bit unsigned. */
  21 #define UINT16_T_MAX        0xffff      /* Maximum 16 bit unsigned. */
  22
  23 #define DB_MIN_PGSIZE   0x000200        /* Minimum page size. */
  24 #define DB_MAX_PGSIZE   0x010000        /* Maximum page size. */
  25
  26 #define DB_MINCACHE     10              /* Minimum cached pages */
  27
  28 /*
  29  * Aligning items to particular sizes or in pages or memory.  ALIGNP is a
  30  * separate macro, as we've had to cast the pointer to different integral
  31  * types on different architectures.
  32  *
  33  * We cast pointers into unsigned longs when manipulating them because C89
  34  * guarantees that u_long is the largest available integral type and further,
  35  * to never generate overflows.  However, neither C89 or C9X  requires that
  36  * any integer type be large enough to hold a pointer, although C9X created
  37  * the intptr_t type, which is guaranteed to hold a pointer but may or may
  38  * not exist.  At some point in the future, we should test for intptr_t and
  39  * use it where available.
  40  */
  41 #undef  ALIGNTYPE
  42 #define ALIGNTYPE               u_long
  43 #undef  ALIGNP
  44 #define ALIGNP(value, bound)    ALIGN((ALIGNTYPE)value, bound)
  45 #undef  ALIGN
  46 #define ALIGN(value, bound)     (((value) + (bound) - 1) & ~((bound) - 1))
  47
  48 /*
  49  * There are several on-page structures that are declared to have a number of
  50  * fields followed by a variable length array of items.  The structure size
  51  * without including the variable length array or the address of the first of
  52  * those elements can be found using SSZ.
  53  *
  54  * This macro can also be used to find the offset of a structure element in a
  55  * structure.  This is used in various places to copy structure elements from
  56  * unaligned memory references, e.g., pointers into a packed page.
  57  *
  58  * There are two versions because compilers object if you take the address of
  59  * an array.
  60  */
  61 #undef  SSZ
  62 #define SSZ(name, field)        ((int)&(((name *)0)->field))
  63
  64 #undef  SSZA
  65 #define SSZA(name, field)       ((int)&(((name *)0)->field[0]))
  66
  67 /* Free and free-string macros that overwrite memory during debugging. */
  68 #ifdef DEBUG
  69 #undef  FREE
  70 #define FREE(p, len) {                                                  \
  71         memset(p, 0xff, len);                                           \
  72         free(p);                                                        \
  73 }
  74 #undef  FREES
  75 #define FREES(p) {                                                      \
  76         FREE(p, strlen(p));                                             \
  77 }
  78 #else
  79 #undef  FREE
  80 #define FREE(p, len) {                                                  \
  81         free(p);                                                        \
  82 }
  83 #undef  FREES
  84 #define FREES(p) {                                                      \
  85         free(p);                                                        \
  86 }
  87 #endif
  88
  89 /* Structure used to print flag values. */
  90 typedef struct __fn {
  91         u_int32_t   mask;               /* Flag value. */
  92         const char *name;               /* Flag name. */
  93 } FN;
  94
  95 /* Set, clear and test flags. */
  96 #define F_SET(p, f)     (p)->flags |= (f)
  97 #define F_CLR(p, f)     (p)->flags &= ~(f)
  98 #define F_ISSET(p, f)   ((p)->flags & (f))
  99 #define LF_SET(f)       (flags |= (f))
 100 #define LF_CLR(f)       (flags &= ~(f))
 101 #define LF_ISSET(f)     (flags & (f))
 102
 103 /* Display separator string. */
 104 #undef  DB_LINE
 105 #define DB_LINE "=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-="
 106
 107 /*******************************************************
 108  * Files.
 109  *******************************************************/
 110 #ifndef MAXPATHLEN              /* Maximum path length. */
 111 #ifdef PATH_MAX
 112 #define MAXPATHLEN      PATH_MAX
 113 #else
 114 #define MAXPATHLEN      1024
 115 #endif
 116 #endif
 117
 118 #define PATH_DOT        "."     /* Current working directory. */
 119 #define PATH_SEPARATOR  "/"     /* Path separator character. */
 120
 121 #ifndef S_IRUSR                 /* UNIX specific file permissions. */
 122 #define S_IRUSR 0000400         /* R for owner */
 123 #define S_IWUSR 0000200         /* W for owner */
 124 #define S_IRGRP 0000040         /* R for group */
 125 #define S_IWGRP 0000020         /* W for group */
 126 #define S_IROTH 0000004         /* R for other */
 127 #define S_IWOTH 0000002         /* W for other */
 128 #endif
 129
 130 #ifndef S_ISDIR                 /* UNIX specific: directory test. */
 131 #define S_ISDIR(m)      ((m & 0170000) == 0040000)
 132 #endif
 133
 134 /*******************************************************
 135  * Mutex support.
 136  *******************************************************/
 137 typedef unsigned char tsl_t;
 138
 139
 140
 141 /*
 142  * !!!
 143  * Various systems require different alignments for mutexes (the worst we've
 144  * seen so far is 16-bytes on some HP architectures).  The mutex (tsl_t) must
 145  * be first in the db_mutex_t structure, which must itself be first in the
 146  * region.  This ensures the alignment is as returned by mmap(2), which should
 147  * be sufficient.  All other mutex users must ensure proper alignment locally.
 148  */
 149 #define MUTEX_ALIGNMENT 1
 150
 151 /*
 152  * The offset of a mutex in memory.
 153  */
 154 #define MUTEX_LOCK_OFFSET(a, b) ((off_t)((u_int8_t *)b - (u_int8_t *)a))
 155
 156 typedef struct _db_mutex_t {
 157 #ifdef HAVE_SPINLOCKS
 158         tsl_t   tsl_resource;           /* Resource test and set. */
 159 #ifdef DEBUG
 160         u_long  pid;                    /* Lock holder: 0 or process pid. */
 161 #endif
 162 #else
 163         off_t   off;                    /* Backing file offset. */
 164         u_long  pid;                    /* Lock holder: 0 or process pid. */
 165 #endif
 166 #ifdef MUTEX_STATISTICS
 167         u_long  mutex_set_wait;         /* Blocking mutex: required waiting. */
 168         u_long  mutex_set_nowait;       /* Blocking mutex: without waiting. */
 169 #endif
 170 } db_mutex_t;
 171
 172 #include "mutex_ext.h"
 173
 174 /*******************************************************
 175  * Access methods.
 176  *******************************************************/
 177 /* Lock/unlock a DB thread. */
 178 #define DB_THREAD_LOCK(dbp)                                             \
 179         (F_ISSET(dbp, DB_AM_THREAD) ?                                   \
 180             __db_mutex_lock((db_mutex_t *)(dbp)->mutexp,  -1,           \
 181                 (dbp)->dbenv == NULL ? NULL : (dbp)->dbenv->db_yield) : 0)
 182 #define DB_THREAD_UNLOCK(dbp)                                           \
 183         (F_ISSET(dbp, DB_AM_THREAD) ?                                   \
 184             __db_mutex_unlock((db_mutex_t *)(dbp)->mutexp,  -1) : 0)
 185
 186 /* Btree/recno local statistics structure. */
 187 struct __db_bt_lstat;   typedef struct __db_bt_lstat DB_BTREE_LSTAT;
 188 struct __db_bt_lstat {
 189         u_int32_t bt_freed;             /* Pages freed for reuse. */
 190         u_int32_t bt_pfxsaved;          /* Bytes saved by prefix compression. */
 191         u_int32_t bt_split;             /* Total number of splits. */
 192         u_int32_t bt_rootsplit;         /* Root page splits. */
 193         u_int32_t bt_fastsplit;         /* Fast splits. */
 194         u_int32_t bt_added;             /* Items added. */
 195         u_int32_t bt_deleted;           /* Items deleted. */
 196         u_int32_t bt_get;               /* Items retrieved. */
 197         u_int32_t bt_cache_hit;         /* Hits in fast-insert code. */
 198         u_int32_t bt_cache_miss;        /* Misses in fast-insert code. */
 199 };
 200
 201 /*******************************************************
 202  * Environment.
 203  *******************************************************/
 204 /* Type passed to __db_appname(). */
 205 typedef enum {
 206         DB_APP_NONE=0,                  /* No type (region). */
 207         DB_APP_DATA,                    /* Data file. */
 208         DB_APP_LOG,                     /* Log file. */
 209         DB_APP_TMP                      /* Temporary file. */
 210 } APPNAME;
 211
 212 /*******************************************************
 213  * Regions.
 214  *******************************************************/
 215 /*
 216  * The shared memory regions share an initial structure so that the general
 217  * region code can handle races between the region being deleted and other
 218  * processes waiting on the region mutex.
 219  *
 220  * !!!
 221  * Note, the mutex must be the first entry in the region; see comment above.
 222  */
 223 typedef struct _rlayout {
 224         db_mutex_t lock;                /* Region mutex. */
 225         u_int32_t  refcnt;              /* Region reference count. */
 226         size_t     size;                /* Region length. */
 227         int        majver;              /* Major version number. */
 228         int        minver;              /* Minor version number. */
 229         int        patch;               /* Patch version number. */
 230
 231 #define DB_R_DELETED    0x01            /* Region was deleted. */
 232         u_int32_t  flags;
 233 } RLAYOUT;
 234
 235 /*******************************************************
 236  * Mpool.
 237  *******************************************************/
 238 /*
 239  * File types for DB access methods.  Negative numbers are reserved to DB.
 240  */
 241 #define DB_FTYPE_BTREE          -1      /* Btree. */
 242 #define DB_FTYPE_HASH           -2      /* Hash. */
 243
 244 /* Structure used as the DB pgin/pgout pgcookie. */
 245 typedef struct __dbpginfo {
 246         size_t  db_pagesize;            /* Underlying page size. */
 247         int     needswap;               /* If swapping required. */
 248 } DB_PGINFO;
 249
 250 /*******************************************************
 251  * Log.
 252  *******************************************************/
 253 /* Initialize an LSN to 'zero'. */
 254 #define ZERO_LSN(LSN) {                                                 \
 255         (LSN).file = 0;                                                 \
 256         (LSN).offset = 0;                                               \
 257 }
 258
 259 /* Return 1 if LSN is a 'zero' lsn, otherwise return 0. */
 260 #define IS_ZERO_LSN(LSN)        ((LSN).file == 0)
 261
 262 /* Test if we need to log a change. */
 263 #define DB_LOGGING(dbp) \
 264         (F_ISSET(dbp, DB_AM_LOGGING) && !F_ISSET(dbp, DB_AM_RECOVER))
 265
 266 #ifdef DEBUG
 267 /*
 268  * Debugging macro to log operations.
 269  *      If DEBUG_WOP is defined, log operations that modify the database.
 270  *      If DEBUG_ROP is defined, log operations that read the database.
 271  *
 272  * D dbp
 273  * T txn
 274  * O operation (string)
 275  * K key
 276  * A data
 277  * F flags
 278  */
 279 #define LOG_OP(D, T, O, K, A, F) {                                      \
 280         DB_LSN _lsn;                                                    \
 281         DBT _op;                                                        \
 282         if (DB_LOGGING((D))) {                                          \
 283                 memset(&_op, 0, sizeof(_op));                           \
 284                 _op.data = O;                                           \
 285                 _op.size = strlen(O) + 1;                               \
 286                 (void)__db_debug_log((D)->dbenv->lg_info,               \
 287                     T, &_lsn, 0, &_op, (D)->log_fileid, K, A, F);       \
 288         }                                                               \
 289 }
 290 #ifdef DEBUG_ROP
 291 #define DEBUG_LREAD(D, T, O, K, A, F)   LOG_OP(D, T, O, K, A, F)
 292 #else
 293 #define DEBUG_LREAD(D, T, O, K, A, F)
 294 #endif
 295 #ifdef DEBUG_WOP
 296 #define DEBUG_LWRITE(D, T, O, K, A, F)  LOG_OP(D, T, O, K, A, F)
 297 #else
 298 #define DEBUG_LWRITE(D, T, O, K, A, F)
 299 #endif
 300 #else
 301 #define DEBUG_LREAD(D, T, O, K, A, F)
 302 #define DEBUG_LWRITE(D, T, O, K, A, F)
 303 #endif /* DEBUG */
 304
 305 /*******************************************************
 306  * Transactions and recovery.
 307  *******************************************************/
 308 /*
 309  * Out of band value for a lock.  The locks are returned to callers as offsets
 310  * into the lock regions.  Since the RLAYOUT structure begins all regions, an
 311  * offset of 0 is guaranteed not to be a valid lock.
 312  */
 313 #define LOCK_INVALID    0
 314
 315 /* The structure allocated for every transaction. */
 316 struct __db_txn {
 317         DB_TXNMGR       *mgrp;          /* Pointer to transaction manager. */
 318         DB_TXN          *parent;        /* Pointer to transaction's parent. */
 319         DB_LSN          last_lsn;       /* Lsn of last log write. */
 320         u_int32_t       txnid;          /* Unique transaction id. */
 321         size_t          off;            /* Detail structure within region. */
 322         TAILQ_ENTRY(__db_txn) links;
 323 };
 324 #endif /* !_DB_INTERNAL_H_ */