install latest threads2 code

[btree] / threads2j.c

// btree version threads2j linux futex concurrency version
// 24 JAN 2014

// author: karl malbrain, malbrain@cal.berkeley.edu

/*
This work, including the source code, documentation
and related data, is placed into the public domain.

The orginal author is Karl Malbrain.

THIS SOFTWARE IS PROVIDED AS-IS WITHOUT WARRANTY
OF ANY KIND, NOT EVEN THE IMPLIED WARRANTY OF
MERCHANTABILITY. THE AUTHOR OF THIS SOFTWARE,
ASSUMES _NO_ RESPONSIBILITY FOR ANY CONSEQUENCE
RESULTING FROM THE USE, MODIFICATION, OR
REDISTRIBUTION OF THIS SOFTWARE.
*/

// Please see the project home page for documentation
// code.google.com/p/high-concurrency-btree

#define _FILE_OFFSET_BITS 64
#define _LARGEFILE64_SOURCE

#ifdef linux
#define _GNU_SOURCE
#include <linux/futex.h>
#define SYS_futex 202
#endif

#ifdef unix
#include <unistd.h>
#include <stdio.h>
#include <stdlib.h>
#include <fcntl.h>
#include <sys/time.h>
#include <sys/mman.h>
#include <errno.h>
#include <pthread.h>
#include <limits.h>
#else
#define WIN32_LEAN_AND_MEAN
#include <windows.h>
#include <stdio.h>
#include <stdlib.h>
#include <time.h>
#include <fcntl.h>
#include <process.h>
#include <intrin.h>
#endif

#include <memory.h>
#include <string.h>

typedef unsigned long long      uid;

#ifndef unix
typedef unsigned long long      off64_t;
typedef unsigned short          ushort;
typedef unsigned int            uint;
#endif

#define BT_ro 0x6f72    // ro
#define BT_rw 0x7772    // rw

#define BT_latchtable   128                                     // number of latch manager slots

#define BT_maxbits              24                                      // maximum page size in bits
#define BT_minbits              9                                       // minimum page size in bits
#define BT_minpage              (1 << BT_minbits)       // minimum page size
#define BT_maxpage              (1 << BT_maxbits)       // maximum page size

/*
There are five lock types for each node in three independent sets: 
1. (set 1) AccessIntent: Sharable. Going to Read the node. Incompatible with NodeDelete. 
2. (set 1) NodeDelete: Exclusive. About to release the node. Incompatible with AccessIntent. 
3. (set 2) ReadLock: Sharable. Read the node. Incompatible with WriteLock. 
4. (set 2) WriteLock: Exclusive. Modify the node. Incompatible with ReadLock and other WriteLocks. 
5. (set 3) ParentModification: Exclusive. Change the node's parent keys. Incompatible with another ParentModification. 
*/

typedef enum{
        BtLockAccess,
        BtLockDelete,
        BtLockRead,
        BtLockWrite,
        BtLockParent
} BtLock;

//      mode & definition for latch implementation

enum {
        Mutex = 1 << 0,         // the mutex bit
        Write = 1 << 1,         // the writers bit
        Share = 1 << 2,         // reader count
        PendRd = 1 << 12,       // reader contended count
        PendWr = 1 << 22        // writer contended count
} LockMode;

enum {
        QueRd = 1,      // reader queue
        QueWr = 2       // writer queue
} RWQueue;

// share is count of read accessors
// grant write lock when share == 0

typedef struct {
        volatile uint mutex:1;          // 1 = busy
        volatile uint write:1;          // 1 = exclusive
        volatile uint share:10;         // count of readers holding locks
        volatile uint readwait:10;      // count of readers waiting
        volatile uint writewait:10;     // count of writers waiting
} BtLatch;

//      Define the length of the page and key pointers

#define BtId 6

//      Page key slot definition.

//      If BT_maxbits is 15 or less, you can save 4 bytes
//      for each key stored by making the first two uints
//      into ushorts.  You can also save 4 bytes by removing
//      the tod field from the key.

//      Keys are marked dead, but remain on the page until
//      it cleanup is called. The fence key (highest key) for
//      the page is always present, even after cleanup.

typedef struct {
        uint off:BT_maxbits;            // page offset for key start
        uint dead:1;                            // set for deleted key
        uint tod;                                       // time-stamp for key
        unsigned char id[BtId];         // id associated with key
} BtSlot;

//      The key structure occupies space at the upper end of
//      each page.  It's a length byte followed by the value
//      bytes.

typedef struct {
        unsigned char len;
        unsigned char key[1];
} *BtKey;

//      The first part of an index page.
//      It is immediately followed
//      by the BtSlot array of keys.

typedef struct Page {
        uint cnt;                                       // count of keys in page
        uint act;                                       // count of active keys
        uint min;                                       // next key offset
        unsigned char bits;                     // page size in bits
        unsigned char lvl:6;            // level of page
        unsigned char kill:1;           // page is being deleted
        unsigned char dirty:1;          // page has deleted keys
        unsigned char right[BtId];      // page number to right
        BtSlot table[0];                        // array of key slots
} *BtPage;

//  hash table entries

typedef struct {
        BtLatch latch[1];
        volatile ushort slot;           // Latch table entry at head of chain
} BtHashEntry;

//      latch manager table structure

typedef struct {
        BtLatch readwr[1];                      // read/write page lock
        BtLatch access[1];                      // Access Intent/Page delete
        BtLatch parent[1];                      // adoption of foster children
        BtLatch busy[1];                        // slot is being moved between chains
        volatile ushort next;           // next entry in hash table chain
        volatile ushort prev;           // prev entry in hash table chain
        volatile ushort pin;            // number of outstanding locks
        volatile ushort hash;           // hash slot entry is under
        volatile uid page_no;           // latch set page number
} BtLatchSet;

//      The memory mapping pool table buffer manager entry

typedef struct {
        unsigned long long int lru;     // number of times accessed
        uid  basepage;                          // mapped base page number
        char *map;                                      // mapped memory pointer
        ushort slot;                            // slot index in this array
        ushort pin;                                     // mapped page pin counter
        void *hashprev;                         // previous pool entry for the same hash idx
        void *hashnext;                         // next pool entry for the same hash idx
#ifndef unix
        HANDLE hmap;                            // Windows memory mapping handle
#endif
} BtPool;

//      structure for latch manager on ALLOC_page

typedef struct {
        struct Page alloc[2];           // next & free page_nos in right ptr
        BtLatch lock[1];                        // allocation area lite latch
        ushort latchdeployed;           // highest number of latch entries deployed
        ushort nlatchpage;                      // number of latch pages at BT_latch
        ushort latchtotal;                      // number of page latch entries
        ushort latchhash;                       // number of latch hash table slots
        ushort latchvictim;                     // next latch entry to examine
        BtHashEntry table[0];           // the hash table
} BtLatchMgr;

//      The object structure for Btree access

typedef struct {
        uint page_size;                         // page size    
        uint page_bits;                         // page size in bits    
        uint seg_bits;                          // seg size in pages in bits
        uint mode;                                      // read-write mode
#ifdef unix
        char *pooladvise;                       // bit maps for pool page advisements
        int idx;
#else
        HANDLE idx;
#endif
        ushort poolcnt;                         // highest page pool node in use
        ushort poolmax;                         // highest page pool node allocated
        ushort poolmask;                        // total number of pages in mmap segment - 1
        ushort evicted;                         // last evicted hash table slot
        ushort hashsize;                        // size of Hash Table for pool entries
        ushort *hash;                           // pool index for hash entries
        BtLatch *latch;                         // latches for pool hash slots
        BtLatchMgr *latchmgr;           // mapped latch page from allocation page
        BtLatchSet *latchsets;          // mapped latch set from latch pages
        BtPool *pool;                           // memory pool page segments
#ifndef unix
        HANDLE halloc;                          // allocation and latch table handle
#endif
} BtMgr;

typedef struct {
        BtMgr *mgr;                     // buffer manager for thread
        BtPage cursor;          // cached frame for start/next (never mapped)
        BtPage frame;           // spare frame for the page split (never mapped)
        BtPage zero;            // page of zeroes to extend the file (never mapped)
        BtPage page;            // current page mapped from file
        uid page_no;            // current page number  
        uid cursor_page;        // current cursor page number   
        BtLatchSet *set;        // current page latchset
        BtPool *pool;           // current page pool
        unsigned char *mem;     // frame, cursor, page memory buffer
        int found;                      // last delete or insert was found
        int err;                        // last error
} BtDb;

typedef enum {
        BTERR_ok = 0,
        BTERR_struct,
        BTERR_ovflw,
        BTERR_lock,
        BTERR_map,
        BTERR_wrt,
        BTERR_hash
} BTERR;

// B-Tree functions
extern void bt_close (BtDb *bt);
extern BtDb *bt_open (BtMgr *mgr);
extern BTERR  bt_insertkey (BtDb *bt, unsigned char *key, uint len, uint lvl, uid id, uint tod);
extern BTERR  bt_deletekey (BtDb *bt, unsigned char *key, uint len, uint lvl);
extern uid bt_findkey    (BtDb *bt, unsigned char *key, uint len);
extern uint bt_startkey  (BtDb *bt, unsigned char *key, uint len);
extern uint bt_nextkey   (BtDb *bt, uint slot);

//      manager functions
extern BtMgr *bt_mgr (char *name, uint mode, uint bits, uint poolsize, uint segsize, uint hashsize);
void bt_mgrclose (BtMgr *mgr);

//  Helper functions to return slot values

extern BtKey bt_key (BtDb *bt, uint slot);
extern uid bt_uid (BtDb *bt, uint slot);
extern uint bt_tod (BtDb *bt, uint slot);

//  BTree page number constants
#define ALLOC_page              0       // allocation & lock manager hash table
#define ROOT_page               1       // root of the btree
#define LEAF_page               2       // first page of leaves
#define LATCH_page              3       // pages for lock manager

//      Number of levels to create in a new BTree

#define MIN_lvl                 2

//  The page is allocated from low and hi ends.
//  The key offsets and row-id's are allocated
//  from the bottom, while the text of the key
//  is allocated from the top.  When the two
//  areas meet, the page is split into two.

//  A key consists of a length byte, two bytes of
//  index number (0 - 65534), and up to 253 bytes
//  of key value.  Duplicate keys are discarded.
//  Associated with each key is a 48 bit row-id.

//  The b-tree root is always located at page 1.
//      The first leaf page of level zero is always
//      located on page 2.

//      The b-tree pages are linked with next
//      pointers to facilitate enumerators,
//      and provide for concurrency.

//      When to root page fills, it is split in two and
//      the tree height is raised by a new root at page
//      one with two keys.

//      Deleted keys are marked with a dead bit until
//      page cleanup The fence key for a node is always
//      present, even after deletion and cleanup.

//  Groups of pages called segments from the btree are optionally
//  cached with a memory mapped pool. A hash table is used to keep
//  track of the cached segments.  This behaviour is controlled
//  by the cache block size parameter to bt_open.

//  To achieve maximum concurrency one page is locked at a time
//  as the tree is traversed to find leaf key in question. The right
//  page numbers are used in cases where the page is being split,
//      or consolidated.

//  Page 0 is dedicated to lock for new page extensions,
//      and chains empty pages together for reuse.

//      The ParentModification lock on a node is obtained to prevent resplitting
//      or deleting a node before its fence is posted into its upper level.

//      Empty pages are chained together through the ALLOC page and reused.

//      Access macros to address slot and key values from the page

#define slotptr(page, slot) (page->table + slot-1)
#define keyptr(page, slot) ((BtKey)((unsigned char*)(page) + slotptr(page, slot)->off))

void bt_putid(unsigned char *dest, uid id)
{
int i = BtId;

        while( i-- )
                dest[i] = (unsigned char)id, id >>= 8;
}

uid bt_getid(unsigned char *src)
{
uid id = 0;
int i;

        for( i = 0; i < BtId; i++ )
                id <<= 8, id |= *src++; 

        return id;
}

//      Latch Manager

int sys_futex(void *addr1, int op, int val1, struct timespec *timeout, void *addr2, int val3)
{
        return syscall(SYS_futex, addr1, op, val1, timeout, addr2, val3);
}

//      wait until write lock mode is clear
//      and add 1 to the share count

void bt_spinreadlock(BtLatch *latch, int private)
{
uint prev;

  if( private )
        private = FUTEX_PRIVATE_FLAG;

  while( 1 ) {
        //      obtain latch mutex
        if( __sync_fetch_and_or((uint *)latch, Mutex) & Mutex ) {
                sched_yield();
                continue;
        }

        //  wait for writers to clear
        //      increment read waiters and wait

        if( latch->write || latch->writewait ) {
                __sync_fetch_and_add ((uint *)latch, PendRd);
                prev = __sync_fetch_and_and ((uint *)latch, ~Mutex) & ~Mutex;
                sys_futex( (uint *)latch, FUTEX_WAIT_BITSET | private, prev, NULL, NULL, QueRd );
                __sync_fetch_and_sub ((uint *)latch, PendRd);
                continue;
        }
        
        // increment reader lock count
        // and release latch mutex

        __sync_fetch_and_add ((uint *)latch, Share);
        __sync_fetch_and_and ((uint *)latch, ~Mutex);
        return;
  }
}

//      wait for other read and write latches to relinquish

void bt_spinwritelock(BtLatch *latch, int private)
{
uint prev;

  if( private )
        private = FUTEX_PRIVATE_FLAG;

  while( 1 ) {
        //      obtain latch mutex
        if( __sync_fetch_and_or((uint *)latch, Mutex) & Mutex ) {
                sched_yield();
                continue;
        }

        //      wait for write and reader count to clear

        if( latch->write || latch->share ) {
                __sync_fetch_and_add ((uint *)latch, PendWr);
                prev = __sync_fetch_and_and ((uint *)latch, ~Mutex) & ~Mutex;
                sys_futex( (uint *)latch, FUTEX_WAIT_BITSET | private, prev, NULL, NULL, QueWr );
                __sync_fetch_and_sub ((uint *)latch, PendWr);
                continue;
        }
        
        //      take write mutex
        //      release latch mutex

        __sync_fetch_and_or ((uint *)latch, Write);
        __sync_fetch_and_and ((uint *)latch, ~Mutex);
        return;
  }
}

//      try to obtain write lock

//      return 1 if obtained,
//              0 otherwise

int bt_spinwritetry(BtLatch *latch)
{
int ans;

        //      try for mutex,
        //      abandon request if not taken

        if( __sync_fetch_and_or((uint *)latch, Mutex) & Mutex )
                return 0;

        //      see if write mode is available

        if( !latch->write && !latch->share ) {
                __sync_fetch_and_or ((uint *)latch, Write);
                ans = 1;
        } else
                ans = 0;

        // release latch mutex

        __sync_fetch_and_and ((uint *)latch, ~Mutex);
        return ans;
}

//      clear write lock

void bt_spinreleasewrite(BtLatch *latch, int private)
{
  if( private )
        private = FUTEX_PRIVATE_FLAG;

        //      obtain latch mutex

        while( __sync_fetch_and_or((uint *)latch, Mutex) & Mutex )
                sched_yield();

        __sync_fetch_and_and ((uint *)latch, ~Write);

        // favor writers

        if( latch->writewait )
          if( sys_futex( (uint *)latch, FUTEX_WAKE_BITSET | private, 1, NULL, NULL, QueWr ) )
                goto wakexit;

        if( latch->readwait )
                sys_futex( (uint *)latch, FUTEX_WAKE_BITSET | private, INT_MAX, NULL, NULL, QueRd );

        // release latch mutex

wakexit:
        __sync_fetch_and_and ((uint *)latch, ~Mutex);
}

//      decrement reader count

void bt_spinreleaseread(BtLatch *latch, int private)
{
  if( private )
        private = FUTEX_PRIVATE_FLAG;

        //      obtain latch mutex

        while( __sync_fetch_and_or((uint *)latch, Mutex) & Mutex )
                sched_yield();

        __sync_fetch_and_sub ((uint *)latch, Share);

        // wake waiting writers

        if( !latch->share && latch->writewait )
                sys_futex( (uint *)latch, FUTEX_WAKE_BITSET | private, 1, NULL, NULL, QueWr );

        // release latch mutex

        __sync_fetch_and_and ((uint *)latch, ~Mutex);
}

//      link latch table entry into latch hash table

void bt_latchlink (BtDb *bt, ushort hashidx, ushort victim, uid page_no)
{
BtLatchSet *set = bt->mgr->latchsets + victim;

        if( set->next = bt->mgr->latchmgr->table[hashidx].slot )
                bt->mgr->latchsets[set->next].prev = victim;

        bt->mgr->latchmgr->table[hashidx].slot = victim;
        set->page_no = page_no;
        set->hash = hashidx;
        set->prev = 0;
}

//      release latch pin

void bt_unpinlatch (BtLatchSet *set)
{
#ifdef unix
        __sync_fetch_and_add(&set->pin, -1);
#else
        _InterlockedDecrement16 (&set->pin);
#endif
}

//      find existing latchset or inspire new one
//      return with latchset pinned

BtLatchSet *bt_pinlatch (BtDb *bt, uid page_no)
{
ushort hashidx = page_no % bt->mgr->latchmgr->latchhash;
ushort slot, avail = 0, victim, idx;
BtLatchSet *set;

        //  obtain read lock on hash table entry

        bt_spinreadlock(bt->mgr->latchmgr->table[hashidx].latch, 0);

        if( slot = bt->mgr->latchmgr->table[hashidx].slot ) do
        {
                set = bt->mgr->latchsets + slot;
                if( page_no == set->page_no )
                        break;
        } while( slot = set->next );

        if( slot ) {
#ifdef unix
                __sync_fetch_and_add(&set->pin, 1);
#else
                _InterlockedIncrement16 (&set->pin);
#endif
        }

    bt_spinreleaseread (bt->mgr->latchmgr->table[hashidx].latch, 0);

        if( slot )
                return set;

  //  try again, this time with write lock

  bt_spinwritelock(bt->mgr->latchmgr->table[hashidx].latch, 0);

  if( slot = bt->mgr->latchmgr->table[hashidx].slot ) do
  {
        set = bt->mgr->latchsets + slot;
        if( page_no == set->page_no )
                break;
        if( !set->pin && !avail )
                avail = slot;
  } while( slot = set->next );

  //  found our entry, or take over an unpinned one

  if( slot || (slot = avail) ) {
        set = bt->mgr->latchsets + slot;
#ifdef unix
        __sync_fetch_and_add(&set->pin, 1);
#else
        _InterlockedIncrement16 (&set->pin);
#endif
        set->page_no = page_no;
        bt_spinreleasewrite(bt->mgr->latchmgr->table[hashidx].latch, 0);
        return set;
  }

        //  see if there are any unused entries
#ifdef unix
        victim = __sync_fetch_and_add (&bt->mgr->latchmgr->latchdeployed, 1) + 1;
#else
        victim = _InterlockedIncrement16 (&bt->mgr->latchmgr->latchdeployed);
#endif

        if( victim < bt->mgr->latchmgr->latchtotal ) {
                set = bt->mgr->latchsets + victim;
#ifdef unix
                __sync_fetch_and_add(&set->pin, 1);
#else
                _InterlockedIncrement16 (&set->pin);
#endif
                bt_latchlink (bt, hashidx, victim, page_no);
                bt_spinreleasewrite (bt->mgr->latchmgr->table[hashidx].latch, 0);
                return set;
        }

#ifdef unix
        victim = __sync_fetch_and_add (&bt->mgr->latchmgr->latchdeployed, -1);
#else
        victim = _InterlockedDecrement16 (&bt->mgr->latchmgr->latchdeployed);
#endif
  //  find and reuse previous lock entry

  while( 1 ) {
#ifdef unix
        victim = __sync_fetch_and_add(&bt->mgr->latchmgr->latchvictim, 1);
#else
        victim = _InterlockedIncrement16 (&bt->mgr->latchmgr->latchvictim) - 1;
#endif
        //      we don't use slot zero

        if( victim %= bt->mgr->latchmgr->latchtotal )
                set = bt->mgr->latchsets + victim;
        else
                continue;

        //      take control of our slot
        //      from other threads

        if( set->pin || !bt_spinwritetry (set->busy) )
                continue;

        idx = set->hash;

        // try to get write lock on hash chain
        //      skip entry if not obtained
        //      or has outstanding locks

        if( !bt_spinwritetry (bt->mgr->latchmgr->table[idx].latch) ) {
                bt_spinreleasewrite (set->busy, 0);
                continue;
        }

        if( set->pin ) {
                bt_spinreleasewrite (set->busy, 0);
                bt_spinreleasewrite (bt->mgr->latchmgr->table[idx].latch, 0);
                continue;
        }

        //  unlink our available victim from its hash chain

        if( set->prev )
                bt->mgr->latchsets[set->prev].next = set->next;
        else
                bt->mgr->latchmgr->table[idx].slot = set->next;

        if( set->next )
                bt->mgr->latchsets[set->next].prev = set->prev;

        bt_spinreleasewrite (bt->mgr->latchmgr->table[idx].latch, 0);
#ifdef unix
        __sync_fetch_and_add(&set->pin, 1);
#else
        _InterlockedIncrement16 (&set->pin);
#endif
        bt_latchlink (bt, hashidx, victim, page_no);
        bt_spinreleasewrite (bt->mgr->latchmgr->table[hashidx].latch, 0);
        bt_spinreleasewrite (set->busy, 0);
        return set;
  }
}

void bt_mgrclose (BtMgr *mgr)
{
BtPool *pool;
uint slot;

        // release mapped pages
        //      note that slot zero is never used

        for( slot = 1; slot < mgr->poolmax; slot++ ) {
                pool = mgr->pool + slot;
                if( pool->slot )
#ifdef unix
                        munmap (pool->map, (mgr->poolmask+1) << mgr->page_bits);
#else
                {
                        FlushViewOfFile(pool->map, 0);
                        UnmapViewOfFile(pool->map);
                        CloseHandle(pool->hmap);
                }
#endif
        }

#ifdef unix
        munmap (mgr->latchsets, mgr->latchmgr->nlatchpage * mgr->page_size);
        munmap (mgr->latchmgr, mgr->page_size);
#else
        FlushViewOfFile(mgr->latchmgr, 0);
        UnmapViewOfFile(mgr->latchmgr);
        CloseHandle(mgr->halloc);
#endif
#ifdef unix
        close (mgr->idx);
        free (mgr->pool);
        free (mgr->hash);
        free (mgr->latch);
        free (mgr->pooladvise);
        free (mgr);
#else
        FlushFileBuffers(mgr->idx);
        CloseHandle(mgr->idx);
        GlobalFree (mgr->pool);
        GlobalFree (mgr->hash);
        GlobalFree (mgr->latch);
        GlobalFree (mgr);
#endif
}

//      close and release memory

void bt_close (BtDb *bt)
{
#ifdef unix
        if ( bt->mem )
                free (bt->mem);
#else
        if ( bt->mem)
                VirtualFree (bt->mem, 0, MEM_RELEASE);
#endif
        free (bt);
}

//  open/create new btree buffer manager

//      call with file_name, BT_openmode, bits in page size (e.g. 16),
//              size of mapped page pool (e.g. 8192)

BtMgr *bt_mgr (char *name, uint mode, uint bits, uint poolmax, uint segsize, uint hashsize)
{
uint lvl, attr, cacheblk, last, slot, idx;
uint nlatchpage, latchhash;
BtLatchMgr *latchmgr;
off64_t size;
uint amt[1];
BtMgr* mgr;
BtKey key;
int flag;
#ifndef unix
SYSTEM_INFO sysinfo[1];
#endif

        // determine sanity of page size and buffer pool

        if( bits > BT_maxbits )
                bits = BT_maxbits;
        else if( bits < BT_minbits )
                bits = BT_minbits;

        if( !poolmax )
                return NULL;    // must have buffer pool

#ifdef unix
        mgr = calloc (1, sizeof(BtMgr));
        mgr->idx = open ((char*)name, O_RDWR | O_CREAT, 0666);

        if( mgr->idx == -1 )
                return free(mgr), NULL;
        
        cacheblk = 4096;        // minimum mmap segment size for unix

#else
        mgr = GlobalAlloc (GMEM_FIXED|GMEM_ZEROINIT, sizeof(BtMgr));
        attr = FILE_ATTRIBUTE_NORMAL;
        mgr->idx = CreateFile(name, GENERIC_READ| GENERIC_WRITE, FILE_SHARE_READ|FILE_SHARE_WRITE, NULL, OPEN_ALWAYS, attr, NULL);

        if( mgr->idx == INVALID_HANDLE_VALUE )
                return GlobalFree(mgr), NULL;

        // normalize cacheblk to multiple of sysinfo->dwAllocationGranularity
        GetSystemInfo(sysinfo);
        cacheblk = sysinfo->dwAllocationGranularity;
#endif

#ifdef unix
        latchmgr = malloc (BT_maxpage);
        *amt = 0;

        // read minimum page size to get root info

        if( size = lseek (mgr->idx, 0L, 2) ) {
                if( pread(mgr->idx, latchmgr, BT_minpage, 0) == BT_minpage )
                        bits = latchmgr->alloc->bits;
                else
                        return free(mgr), free(latchmgr), NULL;
        } else if( mode == BT_ro )
                return free(latchmgr), free (mgr), NULL;
#else
        latchmgr = VirtualAlloc(NULL, BT_maxpage, MEM_COMMIT, PAGE_READWRITE);
        size = GetFileSize(mgr->idx, amt);

        if( size || *amt ) {
                if( !ReadFile(mgr->idx, (char *)latchmgr, BT_minpage, amt, NULL) )
                        return bt_mgrclose (mgr), NULL;
                bits = latchmgr->alloc->bits;
        } else if( mode == BT_ro )
                return bt_mgrclose (mgr), NULL;
#endif

        mgr->page_size = 1 << bits;
        mgr->page_bits = bits;

        mgr->poolmax = poolmax;
        mgr->mode = mode;

        if( cacheblk < mgr->page_size )
                cacheblk = mgr->page_size;

        //  mask for partial memmaps

        mgr->poolmask = (cacheblk >> bits) - 1;

        //      see if requested size of pages per memmap is greater

        if( (1 << segsize) > mgr->poolmask )
                mgr->poolmask = (1 << segsize) - 1;

        mgr->seg_bits = 0;

        while( (1 << mgr->seg_bits) <= mgr->poolmask )
                mgr->seg_bits++;

        mgr->hashsize = hashsize;

#ifdef unix
        mgr->pool = calloc (poolmax, sizeof(BtPool));
        mgr->hash = calloc (hashsize, sizeof(ushort));
        mgr->latch = calloc (hashsize, sizeof(BtLatch));
        mgr->pooladvise = calloc (poolmax, (mgr->poolmask + 8) / 8);
#else
        mgr->pool = GlobalAlloc (GMEM_FIXED|GMEM_ZEROINIT, poolmax * sizeof(BtPool));
        mgr->hash = GlobalAlloc (GMEM_FIXED|GMEM_ZEROINIT, hashsize * sizeof(ushort));
        mgr->latch = GlobalAlloc (GMEM_FIXED|GMEM_ZEROINIT, hashsize * sizeof(BtLatch));
#endif

        if( size || *amt )
                goto mgrlatch;

        // initialize an empty b-tree with latch page, root page, page of leaves
        // and page(s) of latches

        memset (latchmgr, 0, 1 << bits);
        nlatchpage = BT_latchtable / (mgr->page_size / sizeof(BtLatchSet)) + 1; 
        bt_putid(latchmgr->alloc->right, MIN_lvl+1+nlatchpage);
        latchmgr->alloc->bits = mgr->page_bits;

        latchmgr->nlatchpage = nlatchpage;
        latchmgr->latchtotal = nlatchpage * (mgr->page_size / sizeof(BtLatchSet));

        //  initialize latch manager

        latchhash = (mgr->page_size - sizeof(BtLatchMgr)) / sizeof(BtHashEntry);

        //      size of hash table = total number of latchsets

        if( latchhash > latchmgr->latchtotal )
                latchhash = latchmgr->latchtotal;

        latchmgr->latchhash = latchhash;

#ifdef unix
        if( write (mgr->idx, latchmgr, mgr->page_size) < mgr->page_size )
                return bt_mgrclose (mgr), NULL;
#else
        if( !WriteFile (mgr->idx, (char *)latchmgr, mgr->page_size, amt, NULL) )
                return bt_mgrclose (mgr), NULL;

        if( *amt < mgr->page_size )
                return bt_mgrclose (mgr), NULL;
#endif

        memset (latchmgr, 0, 1 << bits);
        latchmgr->alloc->bits = mgr->page_bits;

        for( lvl=MIN_lvl; lvl--; ) {
                slotptr(latchmgr->alloc, 1)->off = mgr->page_size - 3;
                bt_putid(slotptr(latchmgr->alloc, 1)->id, lvl ? MIN_lvl - lvl + 1 : 0);         // next(lower) page number
                key = keyptr(latchmgr->alloc, 1);
                key->len = 2;                   // create stopper key
                key->key[0] = 0xff;
                key->key[1] = 0xff;
                latchmgr->alloc->min = mgr->page_size - 3;
                latchmgr->alloc->lvl = lvl;
                latchmgr->alloc->cnt = 1;
                latchmgr->alloc->act = 1;
#ifdef unix
                if( write (mgr->idx, latchmgr, mgr->page_size) < mgr->page_size )
                        return bt_mgrclose (mgr), NULL;
#else
                if( !WriteFile (mgr->idx, (char *)latchmgr, mgr->page_size, amt, NULL) )
                        return bt_mgrclose (mgr), NULL;

                if( *amt < mgr->page_size )
                        return bt_mgrclose (mgr), NULL;
#endif
        }

        // clear out latch manager locks
        //      and rest of pages to round out segment

        memset(latchmgr, 0, mgr->page_size);
        last = MIN_lvl + 1;

        while( last <= ((MIN_lvl + 1 + nlatchpage) | mgr->poolmask) ) {
#ifdef unix
                pwrite(mgr->idx, latchmgr, mgr->page_size, last << mgr->page_bits);
#else
                SetFilePointer (mgr->idx, last << mgr->page_bits, NULL, FILE_BEGIN);
                if( !WriteFile (mgr->idx, (char *)latchmgr, mgr->page_size, amt, NULL) )
                        return bt_mgrclose (mgr), NULL;
                if( *amt < mgr->page_size )
                        return bt_mgrclose (mgr), NULL;
#endif
                last++;
        }

mgrlatch:
#ifdef unix
        flag = PROT_READ | PROT_WRITE;
        mgr->latchmgr = mmap (0, mgr->page_size, flag, MAP_SHARED, mgr->idx, ALLOC_page * mgr->page_size);
        if( mgr->latchmgr == MAP_FAILED )
                return bt_mgrclose (mgr), NULL;
        mgr->latchsets = (BtLatchSet *)mmap (0, mgr->latchmgr->nlatchpage * mgr->page_size, flag, MAP_SHARED, mgr->idx, LATCH_page * mgr->page_size);
        if( mgr->latchsets == MAP_FAILED )
                return bt_mgrclose (mgr), NULL;
#else
        flag = PAGE_READWRITE;
        mgr->halloc = CreateFileMapping(mgr->idx, NULL, flag, 0, (BT_latchtable / (mgr->page_size / sizeof(BtLatchSet)) + 1 + LATCH_page) * mgr->page_size, NULL);
        if( !mgr->halloc )
                return bt_mgrclose (mgr), NULL;

        flag = FILE_MAP_WRITE;
        mgr->latchmgr = MapViewOfFile(mgr->halloc, flag, 0, 0, (BT_latchtable / (mgr->page_size / sizeof(BtLatchSet)) + 1 + LATCH_page) * mgr->page_size);
        if( !mgr->latchmgr )
                return GetLastError(), bt_mgrclose (mgr), NULL;

        mgr->latchsets = (void *)((char *)mgr->latchmgr + LATCH_page * mgr->page_size);
#endif

#ifdef unix
        free (latchmgr);
#else
        VirtualFree (latchmgr, 0, MEM_RELEASE);
#endif
        return mgr;
}

//      open BTree access method
//      based on buffer manager

BtDb *bt_open (BtMgr *mgr)
{
BtDb *bt = malloc (sizeof(*bt));

        memset (bt, 0, sizeof(*bt));
        bt->mgr = mgr;
#ifdef unix
        bt->mem = malloc (3 *mgr->page_size);
#else
        bt->mem = VirtualAlloc(NULL, 3 * mgr->page_size, MEM_COMMIT, PAGE_READWRITE);
#endif
        bt->frame = (BtPage)bt->mem;
        bt->zero = (BtPage)(bt->mem + 1 * mgr->page_size);
        bt->cursor = (BtPage)(bt->mem + 2 * mgr->page_size);

        memset (bt->zero, 0, mgr->page_size);
        return bt;
}

//  compare two keys, returning > 0, = 0, or < 0
//  as the comparison value

int keycmp (BtKey key1, unsigned char *key2, uint len2)
{
uint len1 = key1->len;
int ans;

        if( ans = memcmp (key1->key, key2, len1 > len2 ? len2 : len1) )
                return ans;

        if( len1 > len2 )
                return 1;
        if( len1 < len2 )
                return -1;

        return 0;
}

//      Buffer Pool mgr

// find segment in pool
// must be called with hashslot idx locked
//      return NULL if not there
//      otherwise return node

BtPool *bt_findpool(BtDb *bt, uid page_no, uint idx)
{
BtPool *pool;
uint slot;

        // compute start of hash chain in pool

        if( slot = bt->mgr->hash[idx] ) 
                pool = bt->mgr->pool + slot;
        else
                return NULL;

        page_no &= ~bt->mgr->poolmask;

        while( pool->basepage != page_no )
          if( pool = pool->hashnext )
                continue;
          else
                return NULL;

        return pool;
}

// add segment to hash table

void bt_linkhash(BtDb *bt, BtPool *pool, uid page_no, int idx)
{
BtPool *node;
uint slot;

        pool->hashprev = pool->hashnext = NULL;
        pool->basepage = page_no & ~bt->mgr->poolmask;
        pool->lru = 1;

        if( slot = bt->mgr->hash[idx] ) {
                node = bt->mgr->pool + slot;
                pool->hashnext = node;
                node->hashprev = pool;
        }

        bt->mgr->hash[idx] = pool->slot;
}

//      find best segment to evict from buffer pool

BtPool *bt_findlru (BtDb *bt, uint hashslot)
{
unsigned long long int target = ~0LL;
BtPool *pool = NULL, *node;

        if( !hashslot )
                return NULL;

        node = bt->mgr->pool + hashslot;

        //  scan pool entries under hash table slot

        do {
          if( node->pin )
                continue;
          if( node->lru > target )
                continue;
          target = node->lru;
          pool = node;
        } while( node = node->hashnext );

        return pool;
}

//  map new buffer pool segment to virtual memory

BTERR bt_mapsegment(BtDb *bt, BtPool *pool, uid page_no)
{
off64_t off = (page_no & ~bt->mgr->poolmask) << bt->mgr->page_bits;
off64_t limit = off + ((bt->mgr->poolmask+1) << bt->mgr->page_bits);
int flag;

#ifdef unix
        flag = PROT_READ | ( bt->mgr->mode == BT_ro ? 0 : PROT_WRITE );
        pool->map = mmap (0, (bt->mgr->poolmask+1) << bt->mgr->page_bits, flag, MAP_SHARED, bt->mgr->idx, off);
        if( pool->map == MAP_FAILED )
                return bt->err = BTERR_map;

        // clear out madvise issued bits
        memset (bt->mgr->pooladvise + pool->slot * ((bt->mgr->poolmask + 8) / 8), 0, (bt->mgr->poolmask + 8)/8);
#else
        flag = ( bt->mgr->mode == BT_ro ? PAGE_READONLY : PAGE_READWRITE );
        pool->hmap = CreateFileMapping(bt->mgr->idx, NULL, flag, (DWORD)(limit >> 32), (DWORD)limit, NULL);
        if( !pool->hmap )
                return bt->err = BTERR_map;

        flag = ( bt->mgr->mode == BT_ro ? FILE_MAP_READ : FILE_MAP_WRITE );
        pool->map = MapViewOfFile(pool->hmap, flag, (DWORD)(off >> 32), (DWORD)off, (bt->mgr->poolmask+1) << bt->mgr->page_bits);
        if( !pool->map )
                return bt->err = BTERR_map;
#endif
        return bt->err = 0;
}

//      calculate page within pool

BtPage bt_page (BtDb *bt, BtPool *pool, uid page_no)
{
uint subpage = (uint)(page_no & bt->mgr->poolmask); // page within mapping
BtPage page;

        page = (BtPage)(pool->map + (subpage << bt->mgr->page_bits));
#ifdef unix
        {
        uint idx = subpage / 8;
        uint bit = subpage % 8;

                if( ~((bt->mgr->pooladvise + pool->slot * ((bt->mgr->poolmask + 8)/8))[idx] >> bit) & 1 ) {
                  madvise (page, bt->mgr->page_size, MADV_WILLNEED);
                  (bt->mgr->pooladvise + pool->slot * ((bt->mgr->poolmask + 8)/8))[idx] |= 1 << bit;
                }
        }
#endif
        return page;
}

//  release pool pin

void bt_unpinpool (BtPool *pool)
{
#ifdef unix
        __sync_fetch_and_add(&pool->pin, -1);
#else
        _InterlockedDecrement16 (&pool->pin);
#endif
}

//      find or place requested page in segment-pool
//      return pool table entry, incrementing pin

BtPool *bt_pinpool(BtDb *bt, uid page_no)
{
BtPool *pool, *node, *next;
uint slot, idx, victim;

        //      lock hash table chain

        idx = (uint)(page_no >> bt->mgr->seg_bits) % bt->mgr->hashsize;
        bt_spinreadlock (&bt->mgr->latch[idx], 1);

        //      look up in hash table

        if( pool = bt_findpool(bt, page_no, idx) ) {
#ifdef unix
                __sync_fetch_and_add(&pool->pin, 1);
#else
                _InterlockedIncrement16 (&pool->pin);
#endif
                bt_spinreleaseread (&bt->mgr->latch[idx], 1);
                pool->lru++;
                return pool;
        }

        //      upgrade to write lock

        bt_spinreleaseread (&bt->mgr->latch[idx], 1);
        bt_spinwritelock (&bt->mgr->latch[idx], 1);

        // try to find page in pool with write lock

        if( pool = bt_findpool(bt, page_no, idx) ) {
#ifdef unix
                __sync_fetch_and_add(&pool->pin, 1);
#else
                _InterlockedIncrement16 (&pool->pin);
#endif
                bt_spinreleasewrite (&bt->mgr->latch[idx], 1);
                pool->lru++;
                return pool;
        }

        // allocate a new pool node
        // and add to hash table

#ifdef unix
        slot = __sync_fetch_and_add(&bt->mgr->poolcnt, 1);
#else
        slot = _InterlockedIncrement16 (&bt->mgr->poolcnt) - 1;
#endif

        if( ++slot < bt->mgr->poolmax ) {
                pool = bt->mgr->pool + slot;
                pool->slot = slot;

                if( bt_mapsegment(bt, pool, page_no) )
                        return NULL;

                bt_linkhash(bt, pool, page_no, idx);
#ifdef unix
                __sync_fetch_and_add(&pool->pin, 1);
#else
                _InterlockedIncrement16 (&pool->pin);
#endif
                bt_spinreleasewrite (&bt->mgr->latch[idx], 1);
                return pool;
        }

        // pool table is full
        //      find best pool entry to evict

#ifdef unix
        __sync_fetch_and_add(&bt->mgr->poolcnt, -1);
#else
        _InterlockedDecrement16 (&bt->mgr->poolcnt);
#endif

        while( 1 ) {
#ifdef unix
                victim = __sync_fetch_and_add(&bt->mgr->evicted, 1);
#else
                victim = _InterlockedIncrement16 (&bt->mgr->evicted) - 1;
#endif
                victim %= bt->mgr->hashsize;

                // try to get write lock
                //      skip entry if not obtained

                if( !bt_spinwritetry (&bt->mgr->latch[victim]) )
                        continue;

                //  if pool entry is empty
                //      or any pages are pinned
                //      skip this entry

                if( !(pool = bt_findlru(bt, bt->mgr->hash[victim])) ) {
                        bt_spinreleasewrite (&bt->mgr->latch[victim], 1);
                        continue;
                }

                // unlink victim pool node from hash table

                if( node = pool->hashprev )
                        node->hashnext = pool->hashnext;
                else if( node = pool->hashnext )
                        bt->mgr->hash[victim] = node->slot;
                else
                        bt->mgr->hash[victim] = 0;

                if( node = pool->hashnext )
                        node->hashprev = pool->hashprev;

                bt_spinreleasewrite (&bt->mgr->latch[victim], 1);

                //      remove old file mapping
#ifdef unix
                munmap (pool->map, (bt->mgr->poolmask+1) << bt->mgr->page_bits);
#else
                FlushViewOfFile(pool->map, 0);
                UnmapViewOfFile(pool->map);
                CloseHandle(pool->hmap);
#endif
                pool->map = NULL;

                //  create new pool mapping
                //  and link into hash table

                if( bt_mapsegment(bt, pool, page_no) )
                        return NULL;

                bt_linkhash(bt, pool, page_no, idx);
#ifdef unix
                __sync_fetch_and_add(&pool->pin, 1);
#else
                _InterlockedIncrement16 (&pool->pin);
#endif
                bt_spinreleasewrite (&bt->mgr->latch[idx], 1);
                return pool;
        }
}

// place write, read, or parent lock on requested page_no.

void bt_lockpage(BtLock mode, BtLatchSet *set)
{
        switch( mode ) {
        case BtLockRead:
                bt_spinreadlock (set->readwr, 0);
                break;
        case BtLockWrite:
                bt_spinwritelock (set->readwr, 0);
                break;
        case BtLockAccess:
                bt_spinreadlock (set->access, 0);
                break;
        case BtLockDelete:
                bt_spinwritelock (set->access, 0);
                break;
        case BtLockParent:
                bt_spinwritelock (set->parent, 0);
                break;
        }
}

// remove write, read, or parent lock on requested page

void bt_unlockpage(BtLock mode, BtLatchSet *set)
{
        switch( mode ) {
        case BtLockRead:
                bt_spinreleaseread (set->readwr, 0);
                break;
        case BtLockWrite:
                bt_spinreleasewrite (set->readwr, 0);
                break;
        case BtLockAccess:
                bt_spinreleaseread (set->access, 0);
                break;
        case BtLockDelete:
                bt_spinreleasewrite (set->access, 0);
                break;
        case BtLockParent:
                bt_spinreleasewrite (set->parent, 0);
                break;
        }
}

//      allocate a new page and write page into it

uid bt_newpage(BtDb *bt, BtPage page)
{
BtLatchSet *set;
BtPool *pool;
uid new_page;
BtPage pmap;
int reuse;

        //      lock allocation page

        bt_spinwritelock(bt->mgr->latchmgr->lock, 0);

        // use empty chain first
        // else allocate empty page

        if( new_page = bt_getid(bt->mgr->latchmgr->alloc[1].right) ) {
                if( pool = bt_pinpool (bt, new_page) )
                        pmap = bt_page (bt, pool, new_page);
                else
                        return 0;
                bt_putid(bt->mgr->latchmgr->alloc[1].right, bt_getid(pmap->right));
                bt_unpinpool (pool);
                reuse = 1;
        } else {
                new_page = bt_getid(bt->mgr->latchmgr->alloc->right);
                bt_putid(bt->mgr->latchmgr->alloc->right, new_page+1);
                reuse = 0;
        }
#ifdef unix
        if ( pwrite(bt->mgr->idx, page, bt->mgr->page_size, new_page << bt->mgr->page_bits) < bt->mgr->page_size )
                return bt->err = BTERR_wrt, 0;

        // if writing first page of pool block, zero last page in the block

        if ( !reuse && bt->mgr->poolmask > 0 && (new_page & bt->mgr->poolmask) == 0 )
        {
                // use zero buffer to write zeros
                memset(bt->zero, 0, bt->mgr->page_size);
                if ( pwrite(bt->mgr->idx,bt->zero, bt->mgr->page_size, (new_page | bt->mgr->poolmask) << bt->mgr->page_bits) < bt->mgr->page_size )
                        return bt->err = BTERR_wrt, 0;
        }
#else
        //      bring new page into pool and copy page.
        //      this will extend the file into the new pages.

        if( pool = bt_pinpool (bt, new_page) )
                pmap = bt_page (bt, pool, new_page);
        else
                return 0;

        memcpy(pmap, page, bt->mgr->page_size);
        bt_unpinpool (pool);
#endif
        // unlock allocation latch and return new page no

        bt_spinreleasewrite(bt->mgr->latchmgr->lock, 0);
        return new_page;
}

//  find slot in page for given key at a given level

int bt_findslot (BtDb *bt, unsigned char *key, uint len)
{
uint diff, higher = bt->page->cnt, low = 1, slot;
uint good = 0;

        //      make stopper key an infinite fence value

        if( bt_getid (bt->page->right) )
                higher++;
        else
                good++;

        //      low is the next candidate, higher is already
        //      tested as .ge. the given key, loop ends when they meet

        while( diff = higher - low ) {
                slot = low + ( diff >> 1 );
                if( keycmp (keyptr(bt->page, slot), key, len) < 0 )
                        low = slot + 1;
                else
                        higher = slot, good++;
        }

        //      return zero if key is on right link page

        return good ? higher : 0;
}

//  find and load page at given level for given key
//      leave page rd or wr locked as requested

int bt_loadpage (BtDb *bt, unsigned char *key, uint len, uint lvl, uint lock)
{
uid page_no = ROOT_page, prevpage = 0;
BtLatchSet *set, *prevset;
uint drill = 0xff, slot;
uint mode, prevmode;
BtPool *prevpool;

  //  start at root of btree and drill down

  bt->set = NULL;

  do {
        // determine lock mode of drill level
        mode = (lock == BtLockWrite) && (drill == lvl) ? BtLockWrite : BtLockRead; 

        bt->set = bt_pinlatch (bt, page_no);
        bt->page_no = page_no;

        // pin page contents

        if( bt->pool = bt_pinpool (bt, page_no) )
                bt->page = bt_page (bt, bt->pool, page_no);
        else
                return 0;

        // obtain access lock using lock chaining with Access mode

        if( page_no > ROOT_page )
          bt_lockpage(BtLockAccess, bt->set);

        //      release & unpin parent page

        if( prevpage ) {
          bt_unlockpage(prevmode, prevset);
          bt_unpinlatch (prevset);
          bt_unpinpool (prevpool);
          prevpage = 0;
        }

        // obtain read lock using lock chaining

        bt_lockpage(mode, bt->set);

        if( page_no > ROOT_page )
          bt_unlockpage(BtLockAccess, bt->set);

        // re-read and re-lock root after determining actual level of root

        if( bt->page->lvl != drill) {
                if ( bt->page_no != ROOT_page )
                        return bt->err = BTERR_struct, 0;
                        
                drill = bt->page->lvl;

                if( lock == BtLockWrite && drill == lvl ) {
                  bt_unlockpage(mode, bt->set);
                  bt_unpinlatch (bt->set);
                  bt_unpinpool (bt->pool);
                  continue;
                }
        }

        //  find key on page at this level
        //  and descend to requested level

        if( !bt->page->kill && (slot = bt_findslot (bt, key, len)) ) {
          if( drill == lvl )
                return slot;

          while( slotptr(bt->page, slot)->dead )
                if( slot++ < bt->page->cnt )
                        continue;
                else {
                        page_no = bt_getid(bt->page->right);
                        goto slideright;
                }

          page_no = bt_getid(slotptr(bt->page, slot)->id);
          drill--;
        }

        //  or slide right into next page
        //  (slide left from deleted page)

        else
                page_no = bt_getid(bt->page->right);

        //  continue down / right using overlapping locks
        //  to protect pages being killed or split.

slideright:
        prevpage = bt->page_no;
        prevpool = bt->pool;
        prevset = bt->set;
        prevmode = mode;
  } while( page_no );

  // return error on end of right chain

  bt->err = BTERR_struct;
  return 0;     // return error
}

//  find and delete key on page by marking delete flag bit
//  when page becomes empty, delete it

BTERR bt_deletekey (BtDb *bt, unsigned char *key, uint len, uint lvl)
{
unsigned char lowerkey[256], higherkey[256];
BtLatchSet *rset, *set;
BtPool *pool, *rpool;
uid page_no, right;
uint slot, tod;
BtPage rpage;
BtKey ptr;

        if( slot = bt_loadpage (bt, key, len, lvl, BtLockWrite) )
                ptr = keyptr(bt->page, slot);
        else
                return bt->err;

        // if key is found delete it, otherwise ignore request

        if( bt->found = !keycmp (ptr, key, len) )
                if( bt->found = slotptr(bt->page, slot)->dead == 0 ) {
                        slotptr(bt->page,slot)->dead = 1;
                        if( slot < bt->page->cnt )
                                bt->page->dirty = 1;
                        bt->page->act--;
                }

        // return if page is not empty, or it has no right sibling

        right = bt_getid(bt->page->right);
        page_no = bt->page_no;
        pool = bt->pool;
        set = bt->set;

        if( !right || bt->page->act ) {
                bt_unlockpage(BtLockWrite, set);
                bt_unpinlatch (set);
                bt_unpinpool (pool);
                return bt->err;
        }

        // obtain Parent lock over write lock

        bt_lockpage(BtLockParent, set);

        // keep copy of key to delete

        ptr = keyptr(bt->page, bt->page->cnt);
        memcpy(lowerkey, ptr, ptr->len + 1);

        // lock and map right page

        if( rpool = bt_pinpool (bt, right) )
                rpage = bt_page (bt, rpool, right);
        else
                return bt->err;

        rset = bt_pinlatch (bt, right);
        bt_lockpage(BtLockWrite, rset);

        // pull contents of next page into current empty page 

        memcpy (bt->page, rpage, bt->mgr->page_size);

        //      keep copy of key to update

        ptr = keyptr(rpage, rpage->cnt);
        memcpy(higherkey, ptr, ptr->len + 1);

        //  Mark right page as deleted and point it to left page
        //      until we can post updates at higher level.

        bt_putid(rpage->right, page_no);
        rpage->kill = 1;
        rpage->cnt = 0;

        bt_unlockpage(BtLockWrite, rset);
        bt_unlockpage(BtLockWrite, set);

        //  delete old lower key to consolidated node

        if( bt_deletekey (bt, lowerkey + 1, *lowerkey, lvl + 1) )
                return bt->err;

        //  redirect higher key directly to consolidated node

        tod = (uint)time(NULL);

        if( bt_insertkey (bt, higherkey+1, *higherkey, lvl + 1, page_no, tod) )
                return bt->err;

        //      add killed right block to free chain
        //      lock latch mgr

        bt_spinwritelock(bt->mgr->latchmgr->lock, 0);

        //      store free chain in allocation page second right
        bt_putid(rpage->right, bt_getid(bt->mgr->latchmgr->alloc[1].right));
        bt_putid(bt->mgr->latchmgr->alloc[1].right, right);

        // unlock latch mgr and unpin right page

        bt_spinreleasewrite(bt->mgr->latchmgr->lock, 0);
        bt_unpinlatch (rset);
        bt_unpinpool (rpool);

        //      remove ParentModify lock

        bt_unlockpage(BtLockParent, set);
        bt_unpinlatch (set);
        bt_unpinpool (pool);
        return 0;
}

//      find key in leaf level and return row-id

uid bt_findkey (BtDb *bt, unsigned char *key, uint len)
{
uint  slot;
BtKey ptr;
uid id;

        if( slot = bt_loadpage (bt, key, len, 0, BtLockRead) )
                ptr = keyptr(bt->page, slot);
        else
                return 0;

        // if key exists, return row-id
        //      otherwise return 0

        if( ptr->len == len && !memcmp (ptr->key, key, len) )
                id = bt_getid(slotptr(bt->page,slot)->id);
        else
                id = 0;

        bt_unlockpage (BtLockRead, bt->set);
        bt_unpinlatch (bt->set);
        bt_unpinpool (bt->pool);
        return id;
}

//      check page for space available,
//      clean if necessary and return
//      =0 - page needs splitting
//      >0 - go ahead at returned slot

uint bt_cleanpage(BtDb *bt, uint amt, uint slot)
{
uint nxt = bt->mgr->page_size;
BtPage page = bt->page;
uint cnt = 0, idx = 0;
uint max = page->cnt;
uint newslot;
BtKey key;

        if( page->min >= (max+1) * sizeof(BtSlot) + sizeof(*page) + amt + 1 )
                return slot;

        //      skip cleanup if nothing to reclaim

        if( !page->dirty )
                return 0;

        memcpy (bt->frame, page, bt->mgr->page_size);

        // skip page info and set rest of page to zero

        memset (page+1, 0, bt->mgr->page_size - sizeof(*page));
        page->dirty = 0;
        page->act = 0;

        // always leave fence key in list

        while( cnt++ < max ) {
                if( cnt == slot )
                        newslot = idx + 1;
                else if( cnt < max && slotptr(bt->frame,cnt)->dead )
                        continue;

                // copy key
                key = keyptr(bt->frame, cnt);
                nxt -= key->len + 1;
                memcpy ((unsigned char *)page + nxt, key, key->len + 1);

                // copy slot
                memcpy(slotptr(page, ++idx)->id, slotptr(bt->frame, cnt)->id, BtId);
                if( !(slotptr(page, idx)->dead = slotptr(bt->frame, cnt)->dead) )
                        page->act++;
                slotptr(page, idx)->tod = slotptr(bt->frame, cnt)->tod;
                slotptr(page, idx)->off = nxt;
        }
        page->min = nxt;
        page->cnt = idx;

        if( page->min >= (idx+1) * sizeof(BtSlot) + sizeof(*page) + amt + 1 )
                return newslot;

        return 0;
}

// split the root and raise the height of the btree

BTERR bt_splitroot(BtDb *bt,  unsigned char *newkey, unsigned char *oldkey, uid page_no2)
{
uint nxt = bt->mgr->page_size;
BtPage root = bt->page;
uid new_page;

        //  Obtain an empty page to use, and copy the current
        //  root contents into it which is the lower half of
        //      the old root.

        if( !(new_page = bt_newpage(bt, root)) )
                return bt->err;

        // preserve the page info at the bottom
        // and set rest to zero

        memset(root+1, 0, bt->mgr->page_size - sizeof(*root));

        // insert first key on newroot page

        nxt -= *newkey + 1;
        memcpy ((unsigned char *)root + nxt, newkey, *newkey + 1);
        bt_putid(slotptr(root, 1)->id, new_page);
        slotptr(root, 1)->off = nxt;
        
        // insert second key on newroot page
        // and increase the root height

        nxt -= *oldkey + 1;
        memcpy ((unsigned char *)root + nxt, oldkey, *oldkey + 1);
        bt_putid(slotptr(root, 2)->id, page_no2);
        slotptr(root, 2)->off = nxt;

        bt_putid(root->right, 0);
        root->min = nxt;                // reset lowest used offset and key count
        root->cnt = 2;
        root->act = 2;
        root->lvl++;

        // release and unpin root (bt->page)

        bt_unlockpage(BtLockWrite, bt->set);
        bt_unpinlatch (bt->set);
        bt_unpinpool (bt->pool);
        return 0;
}

//  split already locked full node
//      return unlocked.

BTERR bt_splitpage (BtDb *bt)
{
uint cnt = 0, idx = 0, max, nxt = bt->mgr->page_size;
unsigned char oldkey[256], lowerkey[256];
uid page_no = bt->page_no, right;
BtLatchSet *nset, *set = bt->set;
BtPool *pool = bt->pool;
BtPage page = bt->page;
uint lvl = page->lvl;
uid new_page;
BtKey key;
uint tod;

        //  split higher half of keys to bt->frame
        //      the last key (fence key) might be dead

        tod = (uint)time(NULL);

        memset (bt->frame, 0, bt->mgr->page_size);
        max = (int)page->cnt;
        cnt = max / 2;
        idx = 0;

        while( cnt++ < max ) {
                key = keyptr(page, cnt);
                nxt -= key->len + 1;
                memcpy ((unsigned char *)bt->frame + nxt, key, key->len + 1);
                memcpy(slotptr(bt->frame,++idx)->id, slotptr(page,cnt)->id, BtId);
                if( !(slotptr(bt->frame, idx)->dead = slotptr(page, cnt)->dead) )
                        bt->frame->act++;
                slotptr(bt->frame, idx)->tod = slotptr(page, cnt)->tod;
                slotptr(bt->frame, idx)->off = nxt;
        }

        // remember existing fence key for new page to the right

        memcpy (oldkey, key, key->len + 1);

        bt->frame->bits = bt->mgr->page_bits;
        bt->frame->min = nxt;
        bt->frame->cnt = idx;
        bt->frame->lvl = lvl;

        // link right node

        if( page_no > ROOT_page ) {
                right = bt_getid (page->right);
                bt_putid(bt->frame->right, right);
        }

        //      get new free page and write frame to it.

        if( !(new_page = bt_newpage(bt, bt->frame)) )
                return bt->err;

        //      update lower keys to continue in old page

        memcpy (bt->frame, page, bt->mgr->page_size);
        memset (page+1, 0, bt->mgr->page_size - sizeof(*page));
        nxt = bt->mgr->page_size;
        page->act = 0;
        cnt = 0;
        idx = 0;

        //  assemble page of smaller keys
        //      (they're all active keys)

        while( cnt++ < max / 2 ) {
                key = keyptr(bt->frame, cnt);
                nxt -= key->len + 1;
                memcpy ((unsigned char *)page + nxt, key, key->len + 1);
                memcpy(slotptr(page,++idx)->id, slotptr(bt->frame,cnt)->id, BtId);
                slotptr(page, idx)->tod = slotptr(bt->frame, cnt)->tod;
                slotptr(page, idx)->off = nxt;
                page->act++;
        }

        // remember fence key for old page

        memcpy(lowerkey, key, key->len + 1);
        bt_putid(page->right, new_page);
        page->min = nxt;
        page->cnt = idx;

        // if current page is the root page, split it

        if( page_no == ROOT_page )
                return bt_splitroot (bt, lowerkey, oldkey, new_page);

        // obtain Parent/Write locks
        // for left and right node pages

        nset = bt_pinlatch (bt, new_page);

        bt_lockpage (BtLockParent, nset);
        bt_lockpage (BtLockParent, set);

        //  release wr lock on left page
        //  (keep the SMO in sequence)

        bt_unlockpage (BtLockWrite, set);

        // insert new fence for reformulated left block

        if( bt_insertkey (bt, lowerkey+1, *lowerkey, lvl + 1, page_no, tod) )
                return bt->err;

        // fix old fence for newly allocated right block page

        if( bt_insertkey (bt, oldkey+1, *oldkey, lvl + 1, new_page, tod) )
                return bt->err;

        // release Parent locks

        bt_unlockpage (BtLockParent, nset);
        bt_unlockpage (BtLockParent, set);
        bt_unpinlatch (nset);
        bt_unpinlatch (set);
        bt_unpinpool (pool);
        return 0;
}

//  Insert new key into the btree at requested level.
//  Level zero pages are leaf pages. Page is unlocked at exit.

BTERR bt_insertkey (BtDb *bt, unsigned char *key, uint len, uint lvl, uid id, uint tod)
{
uint slot, idx;
BtPage page;
BtKey ptr;

  while( 1 ) {
        if( slot = bt_loadpage (bt, key, len, lvl, BtLockWrite) )
                ptr = keyptr(bt->page, slot);
        else
        {
                if ( !bt->err )
                        bt->err = BTERR_ovflw;
                return bt->err;
        }

        // if key already exists, update id and return

        page = bt->page;

        if( bt->found = !keycmp (ptr, key, len) ) {
                slotptr(page, slot)->dead = 0;
                slotptr(page, slot)->tod = tod;
                bt_putid(slotptr(page,slot)->id, id);
                bt_unlockpage(BtLockWrite, bt->set);
                bt_unpinlatch(bt->set);
                bt_unpinpool (bt->pool);
                return bt->err;
        }

        // check if page has enough space

        if( slot = bt_cleanpage (bt, len, slot) )
                break;

        if( bt_splitpage (bt) )
                return bt->err;
  }

  // calculate next available slot and copy key into page

  page->min -= len + 1; // reset lowest used offset
  ((unsigned char *)page)[page->min] = len;
  memcpy ((unsigned char *)page + page->min +1, key, len );

  for( idx = slot; idx < page->cnt; idx++ )
        if( slotptr(page, idx)->dead )
                break;

  // now insert key into array before slot
  // preserving the fence slot

  if( idx == page->cnt )
        idx++, page->cnt++;

  page->act++;

  while( idx > slot )
        *slotptr(page, idx) = *slotptr(page, idx -1), idx--;

  bt_putid(slotptr(page,slot)->id, id);
  slotptr(page, slot)->off = page->min;
  slotptr(page, slot)->tod = tod;
  slotptr(page, slot)->dead = 0;

  bt_unlockpage (BtLockWrite, bt->set);
  bt_unpinlatch (bt->set);
  bt_unpinpool (bt->pool);
  return 0;
}

//  cache page of keys into cursor and return starting slot for given key

uint bt_startkey (BtDb *bt, unsigned char *key, uint len)
{
uint slot;

        // cache page for retrieval
        if( slot = bt_loadpage (bt, key, len, 0, BtLockRead) )
                memcpy (bt->cursor, bt->page, bt->mgr->page_size);
        bt->cursor_page = bt->page_no;
        bt_unlockpage(BtLockRead, bt->set);
        bt_unpinlatch (bt->set);
        bt_unpinpool (bt->pool);
        return slot;
}

//  return next slot for cursor page
//  or slide cursor right into next page

uint bt_nextkey (BtDb *bt, uint slot)
{
BtPool *pool;
BtPage page;
uid right;

  do {
        right = bt_getid(bt->cursor->right);
        while( slot++ < bt->cursor->cnt )
          if( slotptr(bt->cursor,slot)->dead )
                continue;
          else if( right || (slot < bt->cursor->cnt))
                return slot;
          else
                break;

        if( !right )
                break;

        bt->cursor_page = right;

        if( pool = bt_pinpool (bt, right) )
                page = bt_page (bt, pool, right);
        else
                return 0;

        bt->set = bt_pinlatch (bt, right);
    bt_lockpage(BtLockRead, bt->set);

        memcpy (bt->cursor, page, bt->mgr->page_size);

        bt_unlockpage(BtLockRead, bt->set);
        bt_unpinlatch (bt->set);
        bt_unpinpool (pool);
        slot = 0;
  } while( 1 );

  return bt->err = 0;
}

BtKey bt_key(BtDb *bt, uint slot)
{
        return keyptr(bt->cursor, slot);
}

uid bt_uid(BtDb *bt, uint slot)
{
        return bt_getid(slotptr(bt->cursor,slot)->id);
}

uint bt_tod(BtDb *bt, uint slot)
{
        return slotptr(bt->cursor,slot)->tod;
}

#ifdef STANDALONE

void bt_latchaudit (BtDb *bt)
{
ushort idx, hashidx;
BtLatchSet *set;
BtPool *pool;
BtPage page;
uid page_no;

#ifdef unix
        for( idx = 1; idx < bt->mgr->latchmgr->latchdeployed; idx++ ) {
                set = bt->mgr->latchsets + idx;
                if( *(ushort *)set->readwr || *(ushort *)set->access || *(ushort *)set->parent ) {
                        fprintf(stderr, "latchset %d locked for page %6x\n", idx, set->page_no);
                        *(ushort *)set->readwr = 0;
                        *(ushort *)set->access = 0;
                        *(ushort *)set->parent = 0;
                }
                if( set->pin ) {
                        fprintf(stderr, "latchset %d pinned\n", idx);
                        set->pin = 0;
                }
        }

        for( hashidx = 0; hashidx < bt->mgr->latchmgr->latchhash; hashidx++ ) {
          if( *(uint *)bt->mgr->latchmgr->table[hashidx].latch )
                fprintf(stderr, "latchmgr locked\n");
          if( idx = bt->mgr->latchmgr->table[hashidx].slot ) do {
                set = bt->mgr->latchsets + idx;
                if( *(uint *)set->readwr || *(ushort *)set->access || *(ushort *)set->parent )
                        fprintf(stderr, "latchset %d locked\n", idx);
                if( set->hash != hashidx )
                        fprintf(stderr, "latchset %d wrong hashidx\n", idx);
                if( set->pin )
                        fprintf(stderr, "latchset %d pinned\n", idx);
          } while( idx = set->next );
        }
        page_no = bt_getid(bt->mgr->latchmgr->alloc[1].right);

        while( page_no ) {
                fprintf(stderr, "free: %.6x\n", (uint)page_no);
                pool = bt_pinpool (bt, page_no);
                page = bt_page (bt, pool, page_no);
            page_no = bt_getid(page->right);
                bt_unpinpool (pool);
        }
#endif
}

typedef struct {
        char type, idx;
        char *infile;
        BtMgr *mgr;
        int num;
} ThreadArg;

//  standalone program to index file of keys
//  then list them onto std-out

#ifdef unix
void *index_file (void *arg)
#else
uint __stdcall index_file (void *arg)
#endif
{
int line = 0, found = 0, cnt = 0;
uid next, page_no = LEAF_page;  // start on first page of leaves
unsigned char key[256];
ThreadArg *args = arg;
int ch, len = 0, slot;
time_t tod[1];
BtPool *pool;
BtPage page;
BtKey ptr;
BtDb *bt;
FILE *in;

        bt = bt_open (args->mgr);
        time (tod);

        switch(args->type | 0x20)
        {
        case 'a':
                fprintf(stderr, "started latch mgr audit\n");
                bt_latchaudit (bt);
                fprintf(stderr, "finished latch mgr audit\n");
                break;

        case 'w':
                fprintf(stderr, "started indexing for %s\n", args->infile);
                if( in = fopen (args->infile, "rb") )
                  while( ch = getc(in), ch != EOF )
                        if( ch == '\n' )
                        {
                          line++;

                          if( args->num == 1 )
                                sprintf((char *)key+len, "%.9d", 1000000000 - line), len += 9;

                          else if( args->num )
                                sprintf((char *)key+len, "%.9d", line + args->idx * args->num), len += 9;

                          if( bt_insertkey (bt, key, len, 0, line, *tod) )
                                fprintf(stderr, "Error %d Line: %d\n", bt->err, line), exit(0);
                          len = 0;
                        }
                        else if( len < 255 )
                                key[len++] = ch;
                fprintf(stderr, "finished %s for %d keys\n", args->infile, line);
                break;

        case 'd':
                fprintf(stderr, "started deleting keys for %s\n", args->infile);
                if( in = fopen (args->infile, "rb") )
                  while( ch = getc(in), ch != EOF )
                        if( ch == '\n' )
                        {
                          line++;
                          if( args->num == 1 )
                                sprintf((char *)key+len, "%.9d", 1000000000 - line), len += 9;

                          else if( args->num )
                                sprintf((char *)key+len, "%.9d", line + args->idx * args->num), len += 9;

                          if( bt_deletekey (bt, key, len, 0) )
                                fprintf(stderr, "Error %d Line: %d\n", bt->err, line), exit(0);
                          len = 0;
                        }
                        else if( len < 255 )
                                key[len++] = ch;
                fprintf(stderr, "finished %s for keys, %d \n", args->infile, line);
                break;

        case 'f':
                fprintf(stderr, "started finding keys for %s\n", args->infile);
                if( in = fopen (args->infile, "rb") )
                  while( ch = getc(in), ch != EOF )
                        if( ch == '\n' )
                        {
                          line++;
                          if( args->num == 1 )
                                sprintf((char *)key+len, "%.9d", 1000000000 - line), len += 9;

                          else if( args->num )
                                sprintf((char *)key+len, "%.9d", line + args->idx * args->num), len += 9;

                          if( bt_findkey (bt, key, len) )
                                found++;
                          else if( bt->err )
                                fprintf(stderr, "Error %d Syserr %d Line: %d\n", bt->err, errno, line), exit(0);
                          len = 0;
                        }
                        else if( len < 255 )
                                key[len++] = ch;
                fprintf(stderr, "finished %s for %d keys, found %d\n", args->infile, line, found);
                break;

        case 's':
                len = key[0] = 0;

                fprintf(stderr, "started reading\n");

                if( slot = bt_startkey (bt, key, len) )
                  slot--;
                else
                  fprintf(stderr, "Error %d in StartKey. Syserror: %d\n", bt->err, errno), exit(0);

                while( slot = bt_nextkey (bt, slot) ) {
                        ptr = bt_key(bt, slot);
                        fwrite (ptr->key, ptr->len, 1, stdout);
                        fputc ('\n', stdout);
                }

                break;

        case 'c':
                fprintf(stderr, "started reading\n");

                do {
                        if( bt->pool = bt_pinpool (bt, page_no) )
                                page = bt_page (bt, bt->pool, page_no);
                        else
                                break;
                        bt->set = bt_pinlatch (bt, page_no);
                        bt_lockpage (BtLockRead, bt->set);
                        cnt += page->act;
                        next = bt_getid (page->right);
                        bt_unlockpage (BtLockRead, bt->set);
                        bt_unpinlatch (bt->set);
                        bt_unpinpool (bt->pool);
                } while( page_no = next );

                cnt--;  // remove stopper key
                fprintf(stderr, " Total keys read %d\n", cnt);
                break;
        }

        bt_close (bt);
#ifdef unix
        return NULL;
#else
        return 0;
#endif
}

typedef struct timeval timer;

int main (int argc, char **argv)
{
int idx, cnt, len, slot, err;
int segsize, bits = 16;
#ifdef unix
pthread_t *threads;
timer start, stop;
#else
time_t start[1], stop[1];
HANDLE *threads;
#endif
double real_time;
ThreadArg *args;
uint poolsize = 0;
int num = 0;
char key[1];
BtMgr *mgr;
BtKey ptr;
BtDb *bt;

        if( argc < 3 ) {
                fprintf (stderr, "Usage: %s idx_file Read/Write/Scan/Delete/Find [page_bits mapped_segments seg_bits line_numbers src_file1 src_file2 ... ]\n", argv[0]);
                fprintf (stderr, "  where page_bits is the page size in bits\n");
                fprintf (stderr, "  mapped_segments is the number of mmap segments in buffer pool\n");
                fprintf (stderr, "  seg_bits is the size of individual segments in buffer pool in pages in bits\n");
                fprintf (stderr, "  line_numbers = 1 to append line numbers to keys\n");
                fprintf (stderr, "  src_file1 thru src_filen are files of keys separated by newline\n");
                exit(0);
        }

#ifdef unix
        gettimeofday(&start, NULL);
#else
        time(start);
#endif

        if( argc > 3 )
                bits = atoi(argv[3]);

        if( argc > 4 )
                poolsize = atoi(argv[4]);

        if( !poolsize )
                fprintf (stderr, "Warning: no mapped_pool\n");

        if( poolsize > 65535 )
                fprintf (stderr, "Warning: mapped_pool > 65535 segments\n");

        if( argc > 5 )
                segsize = atoi(argv[5]);
        else
                segsize = 4;    // 16 pages per mmap segment

        if( argc > 6 )
                num = atoi(argv[6]);

        cnt = argc - 7;
#ifdef unix
        threads = malloc (cnt * sizeof(pthread_t));
#else
        threads = GlobalAlloc (GMEM_FIXED|GMEM_ZEROINIT, cnt * sizeof(HANDLE));
#endif
        args = malloc (cnt * sizeof(ThreadArg));

        mgr = bt_mgr ((argv[1]), BT_rw, bits, poolsize, segsize, poolsize / 8);

        if( !mgr ) {
                fprintf(stderr, "Index Open Error %s\n", argv[1]);
                exit (1);
        }

        //      fire off threads

        for( idx = 0; idx < cnt; idx++ ) {
                args[idx].infile = argv[idx + 7];
                args[idx].type = argv[2][0];
                args[idx].mgr = mgr;
                args[idx].num = num;
                args[idx].idx = idx;
#ifdef unix
                if( err = pthread_create (threads + idx, NULL, index_file, args + idx) )
                        fprintf(stderr, "Error creating thread %d\n", err);
#else
                threads[idx] = (HANDLE)_beginthreadex(NULL, 65536, index_file, args + idx, 0, NULL);
#endif
        }

        //      wait for termination

#ifdef unix
        for( idx = 0; idx < cnt; idx++ )
                pthread_join (threads[idx], NULL);
        gettimeofday(&stop, NULL);
        real_time = 1000.0 * ( stop.tv_sec - start.tv_sec ) + 0.001 * (stop.tv_usec - start.tv_usec );
#else
        WaitForMultipleObjects (cnt, threads, TRUE, INFINITE);

        for( idx = 0; idx < cnt; idx++ )
                CloseHandle(threads[idx]);

        time (stop);
        real_time = 1000 * (*stop - *start);
#endif
        fprintf(stderr, " Time to complete: %.2f seconds\n", real_time/1000);
        bt_mgrclose (mgr);
}

#endif  //STANDALONE