-/*
- * Copyright (C) 1995, Index Data I/S
- * All rights reserved.
- * Sebastian Hammer, Adam Dickmeiss
- *
- * $Log: cfile.c,v $
- * Revision 1.3 1995-12-01 11:37:22 adam
- * Cached/commit files implemented as meta-files.
- *
- * Revision 1.2 1995/11/30 17:00:49 adam
- * Several bug fixes. Commit system runs now.
- *
- * Revision 1.1 1995/11/30 08:33:11 adam
- * Started work on commit facility.
- *
- */
+/* $Id: cfile.c,v 1.32 2004-12-08 12:23:08 adam Exp $
+ Copyright (C) 1995,1996,1997,1998,1999,2000,2001,2002
+ Index Data Aps
+
+This file is part of the Zebra server.
+
+Zebra 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 2, or (at your option) any later
+version.
+
+Zebra 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 Zebra; see the file LICENSE.zebra. If not, write to the
+Free Software Foundation, 59 Temple Place - Suite 330, Boston, MA
+02111-1307, USA.
+*/
#include <assert.h>
#include <stdlib.h>
#include <string.h>
-#include <alexutil.h>
-#include <mfile.h>
+#include <zebrautl.h>
+#include "mfile.h"
#include "cfile.h"
static int write_head (CFile cf)
{
- int left = cf->head.hash_size * sizeof(int);
+ int left = cf->head.hash_size * sizeof(zint);
int bno = 1;
const char *tab = (char*) cf->array;
- while (left >= HASH_BSIZE)
+ if (!tab)
+ return 0;
+ while (left >= (int) HASH_BSIZE)
{
mf_write (cf->hash_mf, bno++, 0, 0, tab);
tab += HASH_BSIZE;
static int read_head (CFile cf)
{
- int left = cf->head.hash_size * sizeof(int);
+ int left = cf->head.hash_size * sizeof(zint);
int bno = 1;
char *tab = (char*) cf->array;
- while (left >= HASH_BSIZE)
+ if (!tab)
+ return 0;
+ while (left >= (int) HASH_BSIZE)
{
mf_read (cf->hash_mf, bno++, 0, 0, tab);
tab += HASH_BSIZE;
}
-CFile cf_open (MFile mf, const char *fname, int block_size, int wflag,
- int *firstp)
+CFile cf_open (MFile mf, MFile_area area, const char *fname,
+ int block_size, int wflag, int *firstp)
{
- char path[256];
+ char path[1024];
int i;
- CFile cf = xmalloc (sizeof(*cf));
+ CFile cf = (CFile) xmalloc (sizeof(*cf));
int hash_bytes;
cf->rmf = mf;
- sprintf (path, "%s.b", fname);
- if (!(cf->block_mf = mf_open (0, path, block_size, wflag)))
+ yaz_log (YLOG_DEBUG, "cf: open %s %s", cf->rmf->name, wflag ? "rdwr" : "rd");
+ sprintf (path, "%s-b", fname);
+ if (!(cf->block_mf = mf_open (area, path, block_size, wflag)))
{
- logf (LOG_FATAL|LOG_ERRNO, "Failed to open %s", path);
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "Failed to open %s", path);
exit (1);
}
- sprintf (path, "%s.h", fname);
- if (!(cf->hash_mf = mf_open (0, path, HASH_BSIZE, wflag)))
+ sprintf (path, "%s-i", fname);
+ if (!(cf->hash_mf = mf_open (area, path, HASH_BSIZE, wflag)))
{
- logf (LOG_FATAL|LOG_ERRNO, "Failed to open %s", path);
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "Failed to open %s", path);
exit (1);
}
- if (!mf_read (cf->hash_mf, 0, 0, sizeof(cf->head), &cf->head))
+ assert (firstp);
+ if (!mf_read (cf->hash_mf, 0, 0, sizeof(cf->head), &cf->head) ||
+ !cf->head.state)
{
*firstp = 1;
+ cf->head.state = 1;
cf->head.block_size = block_size;
- cf->head.hash_size = 401;
- hash_bytes = cf->head.hash_size * sizeof(int);
- cf->head.next_bucket =
+ cf->head.hash_size = 199;
+ hash_bytes = cf->head.hash_size * sizeof(zint);
+ cf->head.flat_bucket = cf->head.next_bucket = cf->head.first_bucket =
(hash_bytes+sizeof(cf->head))/HASH_BSIZE + 2;
cf->head.next_block = 1;
if (wflag)
mf_write (cf->hash_mf, 0, 0, sizeof(cf->head), &cf->head);
- cf->array = xmalloc (hash_bytes);
+ cf->array = (zint *) xmalloc (hash_bytes);
for (i = 0; i<cf->head.hash_size; i++)
cf->array[i] = 0;
if (wflag)
{
*firstp = 0;
assert (cf->head.block_size == block_size);
- assert (cf->head.hash_size > 2 && cf->head.hash_size < 200000);
- hash_bytes = cf->head.hash_size * sizeof(int);
+ assert (cf->head.hash_size > 2);
+ hash_bytes = cf->head.hash_size * sizeof(zint);
assert (cf->head.next_bucket > 0);
- cf->array = xmalloc (hash_bytes);
+ assert (cf->head.next_block > 0);
+ if (cf->head.state == 1)
+ cf->array = (zint *) xmalloc (hash_bytes);
+ else
+ cf->array = NULL;
read_head (cf);
}
- cf->parray = xmalloc (cf->head.hash_size * sizeof(*cf->parray));
- for (i = 0; i<cf->head.hash_size; i++)
- cf->parray[i] = NULL;
+ if (cf->head.state == 1)
+ {
+ cf->parray = (struct CFile_hash_bucket **)
+ xmalloc (cf->head.hash_size * sizeof(*cf->parray));
+ for (i = 0; i<cf->head.hash_size; i++)
+ cf->parray[i] = NULL;
+ }
+ else
+ cf->parray = NULL;
cf->bucket_lru_front = cf->bucket_lru_back = NULL;
cf->bucket_in_memory = 0;
- cf->max_bucket_in_memory = 200;
+ cf->max_bucket_in_memory = 100;
cf->dirty = 0;
- cf->iobuf = xmalloc (cf->head.block_size);
+ cf->iobuf = (char *) xmalloc (cf->head.block_size);
memset (cf->iobuf, 0, cf->head.block_size);
+ cf->no_hits = 0;
+ cf->no_miss = 0;
+ zebra_mutex_init (&cf->mutex);
return cf;
}
-static int cf_hash (CFile cf, int no)
+static int cf_hash (CFile cf, zint no)
{
- return (no>>3) % cf->head.hash_size;
+ return (int) (((no >> 3) % cf->head.hash_size));
}
static void release_bucket (CFile cf, struct CFile_hash_bucket *p)
}
}
-static struct CFile_hash_bucket *alloc_bucket (CFile cf, int block_no, int hno)
+static struct CFile_hash_bucket *alloc_bucket (CFile cf, zint block_no, int hno)
{
struct CFile_hash_bucket *p, **pp;
flush_bucket (cf, 1);
assert (cf->bucket_in_memory < cf->max_bucket_in_memory);
++(cf->bucket_in_memory);
- p = xmalloc (sizeof(*p));
+ p = (struct CFile_hash_bucket *) xmalloc (sizeof(*p));
p->lru_next = NULL;
p->lru_prev = cf->bucket_lru_front;
return p;
}
-static struct CFile_hash_bucket *get_bucket (CFile cf, int block_no, int hno)
+static struct CFile_hash_bucket *get_bucket (CFile cf, zint block_no, int hno)
{
struct CFile_hash_bucket *p;
p = alloc_bucket (cf, block_no, hno);
if (!mf_read (cf->hash_mf, block_no, 0, 0, &p->ph))
{
- logf (LOG_FATAL|LOG_ERRNO, "read get_bucket");
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "read get_bucket");
exit (1);
}
assert (p->ph.this_bucket == block_no);
return p;
}
-static struct CFile_hash_bucket *new_bucket (CFile cf, int *block_no, int hno)
+static struct CFile_hash_bucket *new_bucket (CFile cf, zint *block_nop, int hno)
{
struct CFile_hash_bucket *p;
int i;
+ zint block_no;
- *block_no = cf->head.next_bucket++;
- p = alloc_bucket (cf, *block_no, hno);
+ block_no = *block_nop = cf->head.next_bucket++;
+ p = alloc_bucket (cf, block_no, hno);
for (i = 0; i<HASH_BUCKET; i++)
{
p->ph.no[i] = 0;
}
p->ph.next_bucket = 0;
- p->ph.this_bucket = *block_no;
+ p->ph.this_bucket = block_no;
p->dirty = 1;
return p;
}
-int cf_lookup (CFile cf, int no)
+static zint cf_lookup_flat (CFile cf, zint no)
+{
+ zint hno = (no*sizeof(zint))/HASH_BSIZE;
+ int off = (int) ((no*sizeof(zint)) - hno*HASH_BSIZE);
+ zint vno = 0;
+
+ mf_read (cf->hash_mf, hno+cf->head.next_bucket, off, sizeof(zint), &vno);
+ return vno;
+}
+
+static zint cf_lookup_hash (CFile cf, zint no)
{
int hno = cf_hash (cf, no);
struct CFile_hash_bucket *hb;
- int block_no, i;
+ zint block_no;
+ int i;
for (hb = cf->parray[hno]; hb; hb = hb->h_next)
{
for (i = 0; i<HASH_BUCKET && hb->ph.vno[i]; i++)
if (hb->ph.no[i] == no)
+ {
+ (cf->no_hits)++;
return hb->ph.vno[i];
+ }
}
for (block_no = cf->array[hno]; block_no; block_no = hb->ph.next_bucket)
{
}
if (hb)
continue;
+#if 0
+ /* extra check ... */
+ for (hb = cf->bucket_lru_back; hb; hb = hb->lru_next)
+ {
+ if (hb->ph.this_bucket == block_no)
+ {
+ yaz_log (YLOG_FATAL, "Found hash bucket on other chain (1)");
+ abort ();
+ }
+ for (i = 0; i<HASH_BUCKET && hb->ph.vno[i]; i++)
+ if (hb->ph.no[i] == no)
+ {
+ yaz_log (YLOG_FATAL, "Found hash bucket on other chain (2)");
+ abort ();
+ }
+ }
+#endif
+ (cf->no_miss)++;
hb = get_bucket (cf, block_no, hno);
for (i = 0; i<HASH_BUCKET && hb->ph.vno[i]; i++)
if (hb->ph.no[i] == no)
return 0;
}
-int cf_new (CFile cf, int no)
+static void cf_write_flat (CFile cf, zint no, zint vno)
+{
+ zint hno = (no*sizeof(zint))/HASH_BSIZE;
+ int off = (int) ((no*sizeof(zint)) - hno*HASH_BSIZE);
+
+ hno += cf->head.next_bucket;
+ if (hno >= cf->head.flat_bucket)
+ cf->head.flat_bucket = hno+1;
+ cf->dirty = 1;
+ mf_write (cf->hash_mf, hno, off, sizeof(zint), &vno);
+}
+
+static void cf_moveto_flat (CFile cf)
+{
+ struct CFile_hash_bucket *p;
+ int j;
+ zint i;
+
+ yaz_log (YLOG_DEBUG, "cf: Moving to flat shadow: %s", cf->rmf->name);
+ yaz_log (YLOG_DEBUG, "cf: hits=%d miss=%d bucket_in_memory=" ZINT_FORMAT " total="
+ ZINT_FORMAT,
+ cf->no_hits, cf->no_miss, cf->bucket_in_memory,
+ cf->head.next_bucket - cf->head.first_bucket);
+ assert (cf->head.state == 1);
+ flush_bucket (cf, -1);
+ assert (cf->bucket_in_memory == 0);
+ p = (struct CFile_hash_bucket *) xmalloc (sizeof(*p));
+ for (i = cf->head.first_bucket; i < cf->head.next_bucket; i++)
+ {
+ if (!mf_read (cf->hash_mf, i, 0, 0, &p->ph))
+ {
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "read bucket moveto flat");
+ exit (1);
+ }
+ for (j = 0; j < HASH_BUCKET && p->ph.vno[j]; j++)
+ cf_write_flat (cf, p->ph.no[j], p->ph.vno[j]);
+ }
+ xfree (p);
+ xfree (cf->array);
+ cf->array = NULL;
+ xfree (cf->parray);
+ cf->parray = NULL;
+ cf->head.state = 2;
+ cf->dirty = 1;
+}
+
+static zint cf_lookup (CFile cf, zint no)
+{
+ if (cf->head.state > 1)
+ return cf_lookup_flat (cf, no);
+ return cf_lookup_hash (cf, no);
+}
+
+static zint cf_new_flat (CFile cf, zint no)
+{
+ zint vno = (cf->head.next_block)++;
+
+ cf_write_flat (cf, no, vno);
+ return vno;
+}
+
+static zint cf_new_hash (CFile cf, zint no)
{
int hno = cf_hash (cf, no);
struct CFile_hash_bucket *hbprev = NULL, *hb = cf->parray[hno];
- int *bucketpp = &cf->array[hno];
+ zint *bucketpp = &cf->array[hno];
int i;
- int vno = (cf->head.next_block)++;
-
+ zint vno = (cf->head.next_block)++;
+
for (hb = cf->parray[hno]; hb; hb = hb->h_next)
if (!hb->ph.vno[HASH_BUCKET-1])
for (i = 0; i<HASH_BUCKET; i++)
if (!hb->ph.vno[i])
{
+ (cf->no_hits)++;
hb->ph.no[i] = no;
hb->ph.vno[i] = vno;
hb->dirty = 1;
}
if (hb)
continue;
+
+#if 0
+ /* extra check ... */
+ for (hb = cf->bucket_lru_back; hb; hb = hb->lru_next)
+ {
+ if (hb->ph.this_bucket == *bucketpp)
+ {
+ yaz_log (YLOG_FATAL, "Found hash bucket on other chain");
+ abort ();
+ }
+ }
+#endif
+ (cf->no_miss)++;
hb = get_bucket (cf, *bucketpp, hno);
assert (hb);
for (i = 0; i<HASH_BUCKET; i++)
return vno;
}
-int cf_read (CFile cf, int no, int offset, int num, void *buf)
+zint cf_new (CFile cf, zint no)
{
- int block;
+ if (cf->head.state > 1)
+ return cf_new_flat (cf, no);
+ if (cf->no_miss*2 > cf->no_hits)
+ {
+ cf_moveto_flat (cf);
+ assert (cf->head.state > 1);
+ return cf_new_flat (cf, no);
+ }
+ return cf_new_hash (cf, no);
+}
+
+
+int cf_read (CFile cf, zint no, int offset, int nbytes, void *buf)
+{
+ zint block;
assert (cf);
+ zebra_mutex_lock (&cf->mutex);
if (!(block = cf_lookup (cf, no)))
+ {
+ zebra_mutex_unlock (&cf->mutex);
return -1;
- if (!mf_read (cf->block_mf, block, offset, num, buf))
+ }
+ zebra_mutex_unlock (&cf->mutex);
+ if (!mf_read (cf->block_mf, block, offset, nbytes, buf))
{
- logf (LOG_FATAL|LOG_ERRNO, "cf_read no=%d, block=%d", no, block);
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "cf_read no=" ZINT_FORMAT " block=" ZINT_FORMAT, no, block);
exit (1);
}
return 1;
}
-int cf_write (CFile cf, int no, int offset, int num, const void *buf)
+int cf_write (CFile cf, zint no, int offset, int nbytes, const void *buf)
{
- int block;
+ zint block;
assert (cf);
+ zebra_mutex_lock (&cf->mutex);
if (!(block = cf_lookup (cf, no)))
{
block = cf_new (cf, no);
- if (offset || num)
+ if (offset || nbytes)
{
mf_read (cf->rmf, no, 0, 0, cf->iobuf);
- memcpy (cf->iobuf + offset, buf, num);
+ memcpy (cf->iobuf + offset, buf, nbytes);
buf = cf->iobuf;
offset = 0;
- num = 0;
+ nbytes = 0;
}
}
- if (mf_write (cf->block_mf, block, offset, num, buf))
+ zebra_mutex_unlock (&cf->mutex);
+ if (mf_write (cf->block_mf, block, offset, nbytes, buf))
{
- logf (LOG_FATAL|LOG_ERRNO, "cf_write no=%d, block=%d", no, block);
+ yaz_log (YLOG_FATAL|YLOG_ERRNO, "cf_write no=" ZINT_FORMAT
+ " block=" ZINT_FORMAT, no, block);
exit (1);
}
return 0;
int cf_close (CFile cf)
{
+ yaz_log (YLOG_DEBUG, "cf: close hits=%d miss=%d bucket_in_memory=" ZINT_FORMAT
+ " total=" ZINT_FORMAT,
+ cf->no_hits, cf->no_miss, cf->bucket_in_memory,
+ cf->head.next_bucket - cf->head.first_bucket);
flush_bucket (cf, -1);
if (cf->dirty)
{
xfree (cf->array);
xfree (cf->parray);
xfree (cf->iobuf);
+ zebra_mutex_destroy (&cf->mutex);
xfree (cf);
return 0;
}
projects / idzebra-moved-to-github.git / blobdiff