/* gdbmopen.c - Open the dbm file and initialize data structures for use. */

/* This file is part of GDBM, the GNU data base manager.
   Copyright (C) 1990-2021 Free Software Foundation, Inc.

   GDBM is free software; you can redistribute it and/or modify
   it under the terms of the GNU General Public License as published by
   the Free Software Foundation; either version 3, or (at your option)
   any later version.

   GDBM is distributed in the hope that it will be useful,
   but WITHOUT ANY WARRANTY; without even the implied warranty of
   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
   GNU General Public License for more details.

   You should have received a copy of the GNU General Public License
   along with GDBM. If not, see <http://www.gnu.org/licenses/>.   */

/* Include system configuration before all else. */
#include "autoconf.h"

#include "gdbmdefs.h"
#include <stddef.h>

static void
compute_directory_size (blksize_t block_size,
			int *ret_dir_size, int *ret_dir_bits)
{
  /* Create the initial hash table directory.  */
  int dir_size = 8 * sizeof (off_t);
  int dir_bits = 3;

  if (block_size > INT_MAX / 2)
    block_size = INT_MAX / 2;
  while (dir_size < block_size && dir_bits < GDBM_HASH_BITS - 3)
    {
      dir_size <<= 1;
      dir_bits++;
    }

  *ret_dir_size = dir_size;
  *ret_dir_bits = dir_bits;
}

static inline int
bucket_element_count (size_t bucket_size)
{
  return (bucket_size - sizeof (hash_bucket)) / sizeof (bucket_element) + 1;
}

static void
gdbm_header_avail (gdbm_file_header *hdr,
		   avail_block **avail_ptr, size_t *avail_size,
		   gdbm_ext_header **exhdr)
{
  switch (hdr->header_magic)
    { 
    case GDBM_OMAGIC:
    case GDBM_MAGIC: 
      *exhdr = NULL;
      *avail_ptr = &((gdbm_file_standard_header*)hdr)->avail;
      *avail_size = (hdr->block_size -
		     offsetof (gdbm_file_standard_header, avail));
      break;
      
    case GDBM_NUMSYNC_MAGIC:
      *exhdr = &((gdbm_file_extended_header*)hdr)->ext;
      *avail_ptr = &((gdbm_file_extended_header*)hdr)->avail;
      *avail_size = (hdr->block_size -
		     offsetof (gdbm_file_extended_header, avail));
      break;
    }
}

static int
validate_header_std (gdbm_file_header const *hdr, struct stat const *st)
{
  int result = GDBM_NO_ERROR;
  int dir_size, dir_bits;
  
  if (!(hdr->block_size > 0
	&& hdr->block_size > sizeof (gdbm_file_header)
	&& hdr->block_size - sizeof (gdbm_file_header) >=
	sizeof(avail_block)))
    {
      return GDBM_BLOCK_SIZE_ERROR;
    }

  /* Technically speaking, the condition below should read
         hdr->next_block != st->st_size
     However, gdbm versions prior to commit 4e819c98 could leave
     hdr->next_block pointing beyond current end of file. To ensure
     backward compatibility with these versions, the condition has been
     slackened to this: */
  if (hdr->next_block < st->st_size)
    result = GDBM_NEED_RECOVERY;

  /* Make sure dir and dir + dir_size fall within the file boundary */
  if (!(hdr->dir > 0
	&& hdr->dir < st->st_size
	&& hdr->dir_size > 0
	&& hdr->dir + hdr->dir_size < st->st_size))
    return GDBM_BAD_HEADER;

  compute_directory_size (hdr->block_size, &dir_size, &dir_bits);
  if (!(hdr->dir_size >= dir_size))
    return GDBM_BAD_HEADER;
  compute_directory_size (hdr->dir_size, &dir_size, &dir_bits);
  if (hdr->dir_bits != dir_bits)
    return GDBM_BAD_HEADER;
  
  if (!(hdr->bucket_size > sizeof (hash_bucket)))
    return GDBM_BAD_HEADER;

  if (hdr->bucket_elems != bucket_element_count (hdr->bucket_size))
    return GDBM_BAD_HEADER;

  return result;
}

static int
validate_header_numsync (gdbm_file_header const *hdr, struct stat const *st)
{
  int result = GDBM_NO_ERROR;
  int dir_size, dir_bits;
  
  if (!(hdr->block_size > 0
	&& hdr->block_size > (sizeof (gdbm_file_header) + sizeof (gdbm_ext_header))
	&& hdr->block_size - (sizeof (gdbm_file_header) + sizeof (gdbm_ext_header)) >=
	sizeof(avail_block)))
    {
      return GDBM_BLOCK_SIZE_ERROR;
    }

  /* Technically speaking, the condition below should read
         hdr->next_block != st->st_size
     However, gdbm versions prior to commit 4e819c98 could leave
     hdr->next_block pointing beyond current end of file. To ensure
     backward compatibility with these versions, the condition has been
     slackened to this: */
  if (hdr->next_block < st->st_size)
    result = GDBM_NEED_RECOVERY;

  /* Make sure dir and dir + dir_size fall within the file boundary */
  if (!(hdr->dir > 0
	&& hdr->dir < st->st_size
	&& hdr->dir_size > 0
	&& hdr->dir + hdr->dir_size < st->st_size))
    return GDBM_BAD_HEADER;

  compute_directory_size (hdr->block_size, &dir_size, &dir_bits);
  if (!(hdr->dir_size >= dir_size))
    return GDBM_BAD_HEADER;
  compute_directory_size (hdr->dir_size, &dir_size, &dir_bits);
  if (hdr->dir_bits != dir_bits)
    return GDBM_BAD_HEADER;
  
  if (!(hdr->bucket_size > sizeof (hash_bucket)))
    return GDBM_BAD_HEADER;

  if (hdr->bucket_elems != bucket_element_count (hdr->bucket_size))
    return GDBM_BAD_HEADER;

  return result;
}

static int
validate_header (gdbm_file_header const *hdr, struct stat const *st)
{
  /* Is the magic number good? */
  switch (hdr->header_magic)
    {
    case GDBM_OMAGIC:
    case GDBM_MAGIC:
      return validate_header_std (hdr, st);
      
    case GDBM_NUMSYNC_MAGIC:
      return validate_header_numsync (hdr, st);

    default:
      switch (hdr->header_magic)
	{
	case GDBM_OMAGIC_SWAP:
	case GDBM_MAGIC32_SWAP:
	case GDBM_MAGIC64_SWAP:
	case GDBM_NUMSYNC_MAGIC32_SWAP:
	case GDBM_NUMSYNC_MAGIC64_SWAP:
	  return GDBM_BYTE_SWAPPED;

	case GDBM_MAGIC32:
	case GDBM_MAGIC64:
	case GDBM_NUMSYNC_MAGIC32:
	case GDBM_NUMSYNC_MAGIC64:
	  return GDBM_BAD_FILE_OFFSET;

	default:
	  return GDBM_BAD_MAGIC_NUMBER;
	}
    }
}

int
_gdbm_validate_header (GDBM_FILE dbf)
{
  struct stat file_stat;
  int rc;
  
  if (fstat (dbf->desc, &file_stat))
    return GDBM_FILE_STAT_ERROR;

  rc = validate_header (dbf->header, &file_stat);
  if (rc == 0)
    {
      if (gdbm_avail_block_validate (dbf, dbf->avail, dbf->avail_size))
	rc = GDBM_BAD_AVAIL;
    }
  return rc;
}

/* Do we have ftruncate? */
static inline int
_gdbm_ftruncate (GDBM_FILE dbf)
{
#if HAVE_FTRUNCATE
  return ftruncate (dbf->desc, 0);
#else
  int fd;
  fd = open (dbf->name, O_RDWR|O_TRUNC, mode);
  if (fd == -1)
    return -1;
  return close (fd);
#endif
}

GDBM_FILE 
gdbm_fd_open (int fd, const char *file_name, int block_size,
	      int flags, void (*fatal_func) (const char *))
{
  GDBM_FILE dbf;		/* The record to return. */
  struct stat file_stat;	/* Space for the stat information. */
  off_t       file_pos;		/* Used with seeks. */
  int 	      index;		/* Used as a loop index. */
  
  /* Initialize the gdbm_errno variable. */
  gdbm_set_errno (NULL, GDBM_NO_ERROR, FALSE);

  /* Get the status of the file. */
  if (fstat (fd, &file_stat))
    {
      if (flags & GDBM_CLOERROR)
	SAVE_ERRNO (close (fd));
      GDBM_SET_ERRNO2 (NULL, GDBM_FILE_STAT_ERROR, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }
  
  /* Allocate new info structure. */
  dbf = calloc (1, sizeof (*dbf));
  if (dbf == NULL)
    {
      if (flags & GDBM_CLOERROR)
	SAVE_ERRNO (close (fd));
      GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }

  dbf->desc = fd;
  
  /* Initialize some fields for known values.  This is done so gdbm_close
     will work if called before allocating some structures. */
  dbf->dir  = NULL;
  dbf->bucket = NULL;
  dbf->header = NULL;

  /* Initialize cache */
  dbf->cache_tree = _gdbm_cache_tree_alloc ();
  _gdbm_cache_init (dbf, DEFAULT_CACHESIZE);

  dbf->file_size = -1;

  dbf->memory_mapping = FALSE;
  dbf->mapped_size_max = SIZE_T_MAX;
  dbf->mapped_region = NULL;
  dbf->mapped_size = 0;
  dbf->mapped_pos = 0;
  dbf->mapped_off = 0;

  /* Save name of file. */
  dbf->name = strdup (file_name);
  if (dbf->name == NULL)
    {
      if (flags & GDBM_CLOERROR)
	close (fd);
      _gdbm_cache_free (dbf);
      free (dbf);
      GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }

  /* Initialize the fatal error routine. */
  dbf->fatal_err = fatal_func;

  dbf->fast_write = TRUE;	/* Default to setting fast_write. */
  dbf->file_locking = TRUE;	/* Default to doing file locking. */
  dbf->central_free = FALSE;	/* Default to not using central_free. */
  dbf->coalesce_blocks = FALSE; /* Default to not coalesce blocks. */

  dbf->need_recovery = FALSE;
  dbf->last_error = GDBM_NO_ERROR;
  dbf->last_syserror = 0;
  dbf->last_errstr = NULL;

  _gdbmsync_init (dbf);
  
  /* GDBM_FAST used to determine whether or not we set fast_write. */
  if (flags & GDBM_SYNC)
    {
      /* If GDBM_SYNC has been requested, don't do fast_write. */
      dbf->fast_write = FALSE;
    }
  if (flags & GDBM_NOLOCK)
    {
      dbf->file_locking = FALSE;
    }

  dbf->cloexec = !!(flags & GDBM_CLOEXEC);
  
  /* Zero-length file can't be a reader... */
  if (((flags & GDBM_OPENMASK) == GDBM_READER) && (file_stat.st_size == 0))
    {
      if (!(flags & GDBM_CLOERROR))
	dbf->desc = -1;
      gdbm_close (dbf);
      GDBM_SET_ERRNO2 (NULL, GDBM_EMPTY_DATABASE, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }

  /* Record the kind of user. */
  dbf->read_write = (flags & GDBM_OPENMASK);

  /* Lock the file in the appropriate way. */
  if (dbf->file_locking)
    {
      if (_gdbm_lock_file (dbf) == -1)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
          GDBM_SET_ERRNO2 (NULL,
			   (flags & GDBM_OPENMASK) == GDBM_READER
			     ? GDBM_CANT_BE_READER : GDBM_CANT_BE_WRITER,
			   FALSE,
			   GDBM_DEBUG_OPEN);
	  return NULL;
	}
    }

  /* If we do have a write lock and it was a GDBM_NEWDB, it is 
     now time to truncate the file. */
  if ((flags & GDBM_OPENMASK) == GDBM_NEWDB && file_stat.st_size != 0)
    {
      if (_gdbm_ftruncate (dbf))
	{
	  GDBM_SET_ERRNO2 (dbf, GDBM_FILE_TRUNCATE_ERROR, FALSE,
			   GDBM_DEBUG_OPEN);
	}
      else if (fstat (dbf->desc, &file_stat))
	{
	  GDBM_SET_ERRNO2 (dbf, GDBM_FILE_STAT_ERROR, FALSE, GDBM_DEBUG_OPEN);
	}

      if (gdbm_last_errno (dbf))
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
    }

  /* Decide if this is a new file or an old file. */
  if (file_stat.st_size == 0)
    {
      /* This is a new file.  Create an empty database.  */
      int dir_size, dir_bits;
      
      /* Start with the blocksize. */
      if (block_size < GDBM_MIN_BLOCK_SIZE)
	{
	  block_size = STATBLKSIZE (file_stat);
	  flags &= ~GDBM_BSEXACT;
	}
      compute_directory_size (block_size, &dir_size, &dir_bits);
      GDBM_DEBUG (GDBM_DEBUG_OPEN, "%s: computed dir_size=%d, dir_bits=%d",
		  dbf->name, dir_size, dir_bits);
      /* Check for correct block_size. */
      if (dir_size != block_size)
	{
	  if (flags & GDBM_BSEXACT)
	    {
	      if (!(flags & GDBM_CLOERROR))
		dbf->desc = -1;
	      gdbm_close (dbf);
	      GDBM_SET_ERRNO2 (NULL, GDBM_BLOCK_SIZE_ERROR, FALSE,
			       GDBM_DEBUG_OPEN);
	      return NULL;
	    }
	  else
	    block_size = dir_size;
	}
      GDBM_DEBUG (GDBM_DEBUG_OPEN, "%s: block_size=%d", dbf->name, block_size);
      
      /* Get space for the file header. It will be written to disk, so
         make sure there's no garbage in it. */
      dbf->header = calloc (1, block_size);
      if (dbf->header == NULL)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}

      /* Set the magic number and the block_size. */
      if (flags & GDBM_NUMSYNC)
	dbf->header->header_magic = GDBM_NUMSYNC_MAGIC;
      else
	dbf->header->header_magic = GDBM_MAGIC;

      /*
       * Set block size.  It must be initialized for gdbm_header_avail to work.
       */
      dbf->header->block_size = block_size;
      gdbm_header_avail (dbf->header, &dbf->avail, &dbf->avail_size, &dbf->xheader);
      dbf->header->dir_size = dir_size;
      dbf->header->dir_bits = dir_bits;

      /* Allocate the space for the directory. */
      dbf->dir = (off_t *) malloc (dbf->header->dir_size);
      if (dbf->dir == NULL)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}
      dbf->header->dir = dbf->header->block_size;

      /* Create the first and only hash bucket. */
      dbf->header->bucket_elems = bucket_element_count (dbf->header->block_size);
      dbf->header->bucket_size  = dbf->header->block_size;
      dbf->bucket = calloc (1, dbf->header->bucket_size);
      if (dbf->bucket == NULL)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}
      _gdbm_new_bucket (dbf, dbf->bucket, 0);
      dbf->bucket->av_count = 1;
      dbf->bucket->bucket_avail[0].av_adr = 3*dbf->header->block_size;
      dbf->bucket->bucket_avail[0].av_size = dbf->header->block_size;

      /* Set table entries to point to hash buckets. */
      for (index = 0; index < GDBM_DIR_COUNT (dbf); index++)
	dbf->dir[index] = 2*dbf->header->block_size;

      /* Initialize the active avail block. */
      dbf->avail->size = (dbf->avail_size - offsetof(avail_block, av_table))
	                  / sizeof (avail_elem);
      dbf->avail->count = 0;
      dbf->avail->next_block = 0;
      
      dbf->header->next_block  = 4*dbf->header->block_size;

      /* Write initial configuration to the file. */
      /* Block 0 is the file header and active avail block. */
      if (_gdbm_full_write (dbf, dbf->header, dbf->header->block_size))
	{
	  GDBM_DEBUG (GDBM_DEBUG_OPEN|GDBM_DEBUG_ERR,
		      "%s: error writing header: %s",
		      dbf->name, gdbm_db_strerror (dbf));
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}

      /* Block 1 is the initial bucket directory. */
      if (_gdbm_full_write (dbf, dbf->dir, dbf->header->dir_size))
	{
	  GDBM_DEBUG (GDBM_DEBUG_OPEN|GDBM_DEBUG_ERR,
		      "%s: error writing directory: %s",
		      dbf->name, gdbm_db_strerror (dbf));	  
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}

      /* Block 2 is the only bucket. */
      if (_gdbm_full_write (dbf, dbf->bucket, dbf->header->bucket_size))
	{
	  GDBM_DEBUG (GDBM_DEBUG_OPEN|GDBM_DEBUG_ERR,
		      "%s: error writing bucket: %s",
		      dbf->name, gdbm_db_strerror (dbf));	  
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
      
      if (_gdbm_file_extend (dbf, dbf->header->next_block))
	{
	  GDBM_DEBUG (GDBM_DEBUG_OPEN|GDBM_DEBUG_ERR,
		      "%s: error extending file: %s",
		      dbf->name, gdbm_db_strerror (dbf));	  
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
	  
      /* Wait for initial configuration to be written to disk. */
      gdbm_file_sync (dbf);

      free (dbf->bucket);
    }
  else
    {
      /* This is an old database.  Read in the information from the file
	 header and initialize the hash directory. */

      gdbm_file_header partial_header;  /* For the first part of it. */
      int rc;
      
      /* Read the partial file header. */
      if (_gdbm_full_read (dbf, &partial_header, sizeof (partial_header)))
	{
	  GDBM_DEBUG (GDBM_DEBUG_ERR|GDBM_DEBUG_OPEN,
		      "%s: error reading partial header: %s",
		      dbf->name, gdbm_db_strerror (dbf));
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}

      /* Is the header valid? */
      rc = validate_header (&partial_header, &file_stat);
      if (rc == GDBM_NEED_RECOVERY)
	{
	  dbf->need_recovery = 1;
	}
      else if (rc != GDBM_NO_ERROR)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, rc, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}
      
      /* It is a good database, read the entire header. */
      dbf->header = malloc (partial_header.block_size);
      if (dbf->header == NULL)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}
      
      memcpy (dbf->header, &partial_header, sizeof (partial_header));
      if (_gdbm_full_read (dbf, dbf->header + 1,
			   dbf->header->block_size - sizeof (gdbm_file_header)))
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
      gdbm_header_avail (dbf->header, &dbf->avail, &dbf->avail_size, &dbf->xheader);

      if (((dbf->header->block_size -
	    (GDBM_HEADER_AVAIL_OFFSET (dbf) +
	     sizeof (avail_block))) / sizeof (avail_elem) + 1) != dbf->avail->size)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, GDBM_BAD_HEADER, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}
      
      if (gdbm_avail_block_validate (dbf, dbf->avail, dbf->avail_size))
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
      
      /* Allocate space for the hash table directory.  */
      dbf->dir = malloc (dbf->header->dir_size);
      if (dbf->dir == NULL)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  gdbm_close (dbf);
	  GDBM_SET_ERRNO2 (NULL, GDBM_MALLOC_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}

      /* Read the hash table directory. */
      file_pos = gdbm_file_seek (dbf, dbf->header->dir, SEEK_SET);
      if (file_pos != dbf->header->dir)
	{
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  GDBM_SET_ERRNO2 (NULL, GDBM_FILE_SEEK_ERROR, FALSE, GDBM_DEBUG_OPEN);
	  return NULL;
	}

      if (_gdbm_full_read (dbf, dbf->dir, dbf->header->dir_size))
	{
	  GDBM_DEBUG (GDBM_DEBUG_ERR|GDBM_DEBUG_OPEN,
		      "%s: error reading dir: %s",
		      dbf->name, gdbm_db_strerror (dbf));
	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}

    }

#if HAVE_MMAP
  if (!(flags & GDBM_NOMMAP))
    {
      dbf->mmap_preread = (flags & GDBM_PREREAD) != 0;
      if (_gdbm_mapped_init (dbf) == 0)
	dbf->memory_mapping = TRUE;
      else
	{
	  /* gdbm_errno should already be set. */
	  GDBM_DEBUG (GDBM_DEBUG_ERR|GDBM_DEBUG_OPEN,
		      "%s: _gdbm_mapped_init failed: %s",
		      dbf->name, strerror (errno));

	  if (!(flags & GDBM_CLOERROR))
	    dbf->desc = -1;
	  SAVE_ERRNO (gdbm_close (dbf));
	  return NULL;
	}
    }
#endif

  /* Finish initializing dbf. */
  dbf->bucket = NULL;
  dbf->bucket_dir = 0;
  dbf->header_changed = FALSE;
  dbf->directory_changed = FALSE;
  dbf->bucket_changed = FALSE;
  dbf->second_changed = FALSE;

  if (flags & GDBM_XVERIFY)
    {
      gdbm_avail_verify (dbf);
    }
  
  GDBM_DEBUG (GDBM_DEBUG_ALL, "%s: opened %s", dbf->name,
	      dbf->need_recovery ? "for recovery" : "successfully");

  /* Everything is fine, return the pointer to the file
     information structure.  */
  return dbf;
}
  
/* Initialize dbm system.  FILE is a pointer to the file name.  If the file
   has a size of zero bytes, a file initialization procedure is performed,
   setting up the initial structure in the file.  BLOCK_SIZE is used during
   initialization to determine the size of various constructs.  If the value
   is less than GDBM_MIN_BLOCK_SIZE, the file system blocksize is used,
   otherwise the value of BLOCK_SIZE is used.  BLOCK_SIZE is ignored if the
   file has previously initialized.  If FLAGS is set to GDBM_READ the user
   wants to just read the database and any call to dbm_store or dbm_delete
   will fail. Many readers can access the database at the same time.  If FLAGS
   is set to GDBM_WRITE, the user wants both read and write access to the
   database and requires exclusive access.  If FLAGS is GDBM_WRCREAT, the user
   wants both read and write access to the database and if the database does
   not exist, create a new one.  If FLAGS is GDBM_NEWDB, the user want a
   new database created, regardless of whether one existed, and wants read
   and write access to the new database.  Any error detected will cause a 
   return value of null and an approprate value will be in gdbm_errno.  If
   no errors occur, a pointer to the "gdbm file descriptor" will be
   returned. */
   

GDBM_FILE 
gdbm_open (const char *file, int block_size, int flags, int mode,
     	   void (*fatal_func) (const char *))
{
  int fd;
  /* additional bits for open(2) flags */
  int fbits = 0;

  switch (flags & GDBM_OPENMASK)
    {
    case GDBM_READER:
      fbits = O_RDONLY;
      break;

    case GDBM_WRITER:
      fbits = O_RDWR;
      break;

    case GDBM_WRCREAT:
    case GDBM_NEWDB:
      fbits = O_RDWR|O_CREAT;
      break;

    default:
      errno = EINVAL;
      GDBM_SET_ERRNO2 (NULL, GDBM_FILE_OPEN_ERROR, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }
  if (flags & GDBM_CLOEXEC)
    fbits |= O_CLOEXEC;
  
  fd = open (file, fbits, mode);
  if (fd < 0)
    {
      GDBM_SET_ERRNO2 (NULL, GDBM_FILE_OPEN_ERROR, FALSE, GDBM_DEBUG_OPEN);
      return NULL;
    }
  return gdbm_fd_open (fd, file, block_size, flags | GDBM_CLOERROR,
		       fatal_func);
}

int
_gdbm_file_size (GDBM_FILE dbf, off_t *psize)
{
  if (dbf->file_size == -1)
    {
      struct stat sb;
      if (fstat (dbf->desc, &sb))
	{
	  GDBM_SET_ERRNO (dbf, GDBM_FILE_STAT_ERROR, FALSE);
	  return -1;
	}
      dbf->file_size = sb.st_size;
    }
  *psize = dbf->file_size;
  return 0;
}

/*
 * Convert from numsync to the standard GDBM format.  This is pretty
 * straightforward: the avail block gets expanded by
 * sizeof(gdbm_ext_header), so we only need to shift the avail block
 * up this number of bytes, change the avail and avail_size pointers
 * and update the magic number.
 */
static int
_gdbm_convert_from_numsync (GDBM_FILE dbf)
{
  avail_block *old_avail = dbf->avail;

  /* Change the magic number */
  dbf->header->header_magic = GDBM_MAGIC;
  /* Update avail pointer and size */
  gdbm_header_avail (dbf->header, &dbf->avail, &dbf->avail_size, &dbf->xheader);
  
  /* Move data up */
  memmove (dbf->avail, old_avail, dbf->avail_size - sizeof (gdbm_ext_header));

  /* Fix up the avail table size */
  dbf->avail->size = (dbf->avail_size - offsetof (avail_block, av_table))
	                  / sizeof (avail_elem);

  dbf->header_changed = TRUE;
  
  return 0;
}

/*
 * Convert the database from the standard to extended (numsync) format.
 * The avail block shrinks by sizeof(gdbm_ext_header) bytes.  The av_table
 * entries that don't fit into the new size need to be returned to the
 * avail pool using the _gdbm_free call.
 */
static int
_gdbm_convert_to_numsync (GDBM_FILE dbf)
{
  avail_block *old_avail = dbf->avail;
  size_t old_avail_size = dbf->avail->size;
  size_t n; /* Number of elements to return to the pool */
  int rc;
  avail_elem *av = NULL;
  
  /* Change the magic number */
  dbf->header->header_magic = GDBM_NUMSYNC_MAGIC;
  /* Update avail pointer and size */
  gdbm_header_avail (dbf->header, &dbf->avail, &dbf->avail_size, &dbf->xheader);
  /*
   * Compute new av_table size.
   * NOTE: Don't try to modify dbf->avail until the final move, otherwise
   * the available block would end up clobbered.  All modifications are
   * applied to old_avail.
   */
  old_avail->size = (dbf->avail_size - offsetof (avail_block, av_table))
	                  / sizeof (avail_elem);
  /* Compute the number of avail elements that don't fit in the new table. */
  n = old_avail_size - old_avail->size;
  if (n > 0)
    {
      /* Stash them away */
      av = calloc (n, sizeof (av[0]));
      if (!av)
	{
	  GDBM_SET_ERRNO (dbf, GDBM_MALLOC_ERROR, FALSE);
	  return -1;
	}
      n = 0;
      while (old_avail->count > old_avail->size)
	{
	  old_avail->count--;
	  av[n++] = old_avail->av_table[old_avail->count];
	}
    }

  /*
   * Move the modified avail block into its new place.  From now on,
   * old_avail may not be use.  The database header is in consistent
   * state and all modifications should be applied to it directly.
   */
  memmove (dbf->avail, old_avail, dbf->avail_size);

  /* Initialize the extended header */
  memset (dbf->xheader, 0, sizeof (dbf->xheader[0]));

  rc = 0; /* Assume success */
  
  if (av)
    {
      /* Return stashed av_table elements to the available pool. */
      /* _gdbm_free needs a non-NULL bucket, so get one: */
      if (!dbf->bucket)
	rc = _gdbm_get_bucket (dbf, 0);
      if (rc == 0)
	{
	  size_t i;
	  
	  for (i = 0; i < n; i++)
	    {
	      rc = _gdbm_free (dbf, av[i].av_adr, av[i].av_size);
	      if (rc)
		break;
	    }
	}
      free (av);
    }

  dbf->header_changed = TRUE;
  
  return rc;
}

int
gdbm_convert (GDBM_FILE dbf, int flag)
{
  int rc;
  
  /* Return immediately if the database needs recovery */	
  GDBM_ASSERT_CONSISTENCY (dbf, -1);
  
  /* First check to make sure this guy is a writer. */
  if (dbf->read_write == GDBM_READER)
    {
      GDBM_SET_ERRNO2 (dbf, GDBM_READER_CANT_STORE, FALSE,
		       GDBM_DEBUG_STORE);
      return -1;
    }

  switch (flag)
    {
    case 0:
    case GDBM_NUMSYNC:
      break;

    default:
      GDBM_SET_ERRNO2 (dbf, GDBM_MALFORMED_DATA, FALSE,
		       GDBM_DEBUG_STORE);
      return -1;
    }

  rc = 0;
  switch (dbf->header->header_magic)
    {
    case GDBM_OMAGIC:
    case GDBM_MAGIC:
      if (flag == GDBM_NUMSYNC)
	rc = _gdbm_convert_to_numsync (dbf);
      break;
      
    case GDBM_NUMSYNC_MAGIC:
      if (flag == 0)
	rc = _gdbm_convert_from_numsync (dbf);	
    }

  if (rc == 0)
    rc = _gdbm_end_update (dbf);
  
  return 0;
}