pjs/modules/libjar/nsZipArchive.cpp

1337 строки
32 KiB
C++
Исходник Обычный вид История

/* -*- Mode: C; tab-width: 4; indent-tabs-mode: nil; c-basic-offset: 2 -*- */
/*
* The contents of this file are subject to the Netscape Public
* License Version 1.1 (the "License"); you may not use this file
* except in compliance with the License. You may obtain a copy of
* the License at http://www.mozilla.org/NPL/
*
* Software distributed under the License is distributed on an "AS
* IS" basis, WITHOUT WARRANTY OF ANY KIND, either express or
* implied. See the License for the specific language governing
* rights and limitations under the License.
*
* The Original Code is Mozilla Communicator client code,
* released March 31, 1998.
*
* The Initial Developer of the Original Code is Netscape Communications
* Corporation. Portions created by Netscape are
* Copyright (C) 1998 Netscape Communications Corporation. All
* Rights Reserved.
*
* Contributor(s):
* Daniel Veditz <dveditz@netscape.com>
* Samir Gehani <sgehani@netscape.com>
* Mitch Stoltz <mstoltz@netscape.com>
*/
/*
* This module implements a simple archive extractor for the PKZIP format.
1999-05-19 07:21:03 +04:00
*
* The underlying nsZipArchive is NOT thread-safe. Do not pass references
* or pointers to it across thread boundaries.
*/
#ifndef STANDALONE
1999-09-01 05:28:22 +04:00
#include "nscore.h"
#include "prmem.h"
#include "prio.h"
#include "plstr.h"
#include "xp_regexp.h"
#define ZFILE_CREATE PR_WRONLY | PR_CREATE_FILE
#define READTYPE PRInt32
1999-09-08 02:03:53 +04:00
#include "zlib.h"
1999-09-01 05:28:22 +04:00
1999-05-19 07:21:03 +04:00
#else
1999-09-01 05:28:22 +04:00
#ifdef WIN32
#include "windows.h"
#endif
1999-09-08 02:03:53 +04:00
#undef MOZILLA_CLIENT // undoes prtypes damage in zlib.h
#define ZFILE_CREATE "wb"
#define READTYPE PRUint32
1999-09-08 02:03:53 +04:00
#include "zlib.h"
#undef PR_PUBLIC_API
#include "zipstub.h"
1999-09-01 05:28:22 +04:00
1999-05-19 07:21:03 +04:00
#endif /* STANDALONE */
1999-04-28 04:34:30 +04:00
#ifdef XP_UNIX
#include <sys/stat.h>
1999-11-03 01:36:59 +03:00
#elif defined(XP_PC)
#include <io.h>
#endif
#include "zipfile.h"
#include "zipstruct.h"
#include "nsZipArchive.h"
static PRUint16 xtoint(unsigned char *ii);
static PRUint32 xtolong(unsigned char *ll);
static PRUint16 ExtractMode(PRUint32 ext_attr);
/*---------------------------------------------
* C API wrapper for nsZipArchive
*--------------------------------------------*/
#ifdef STANDALONE
/**
* ZIP_OpenArchive
*
* opens the named zip/jar archive and returns a handle that
* represents the archive in other ZIP_ calls.
*
* @param zipname archive filename
* @param hZip receives handle if archive opened OK
* @return status code
*/
PR_PUBLIC_API(PRInt32) ZIP_OpenArchive( const char * zipname, void** hZip )
{
PRInt32 status;
/*--- error check args ---*/
1999-05-19 07:21:03 +04:00
if ( hZip == 0 )
return ZIP_ERR_PARAM;
1999-04-28 04:34:30 +04:00
/*--- NULL output to prevent use by bozos who don't check errors ---*/
1999-05-19 07:21:03 +04:00
*hZip = 0;
/*--- create and open the archive ---*/
nsZipArchive* zip = new nsZipArchive();
1999-05-19 07:21:03 +04:00
if ( zip == 0 )
return ZIP_ERR_MEMORY;
status = zip->OpenArchive(zipname);
if ( status == ZIP_OK )
*hZip = NS_STATIC_CAST(void*,zip);
1999-04-28 04:34:30 +04:00
else
delete zip;
return status;
}
/**
* ZIP_CloseArchive
*
* closes zip archive and frees memory
* @param hZip handle obtained from ZIP_OpenArchive
* @return status code
*/
PR_PUBLIC_API(PRInt32) ZIP_CloseArchive( void** hZip )
{
/*--- error check args ---*/
1999-05-19 07:21:03 +04:00
if ( hZip == 0 || *hZip == 0 )
return ZIP_ERR_PARAM;
nsZipArchive* zip = NS_STATIC_CAST(nsZipArchive*,*hZip);
if ( zip->kMagic != ZIP_MAGIC )
return ZIP_ERR_PARAM; /* whatever it is isn't one of ours! */
/*--- close the archive ---*/
1999-05-19 07:21:03 +04:00
*hZip = 0;
delete zip;
return ZIP_OK;
}
/**
* ZIP_ExtractFile
*
* extracts named file from an opened archive
*
* @param hZip handle obtained from ZIP_OpenArchive
* @param filename name of file in archive
* @param outname filename to extract to
*/
PR_PUBLIC_API(PRInt32) ZIP_ExtractFile( void* hZip, const char * filename, const char * outname )
{
/*--- error check args ---*/
1999-05-19 07:21:03 +04:00
if ( hZip == 0 )
return ZIP_ERR_PARAM;
nsZipArchive* zip = NS_STATIC_CAST(nsZipArchive*,hZip);
if ( zip->kMagic != ZIP_MAGIC )
return ZIP_ERR_PARAM; /* whatever it is isn't one of ours! */
/*--- extract the file ---*/
return zip->ExtractFile( filename, outname );
}
1999-04-28 04:34:30 +04:00
/**
* ZIP_FindInit
*
* Initializes an enumeration of files in the archive
*
* @param hZip handle obtained from ZIP_OpenArchive
* @param pattern regexp to match files in archive, the usual shell expressions.
* NULL pattern also matches all files, faster than "*"
*/
1999-05-19 07:21:03 +04:00
PR_PUBLIC_API(void*) ZIP_FindInit( void* hZip, const char * pattern )
1999-04-28 04:34:30 +04:00
{
/*--- error check args ---*/
1999-05-19 07:21:03 +04:00
if ( hZip == 0 )
return 0;
1999-04-28 04:34:30 +04:00
nsZipArchive* zip = NS_STATIC_CAST(nsZipArchive*,hZip);
if ( zip->kMagic != ZIP_MAGIC )
1999-05-19 07:21:03 +04:00
return 0; /* whatever it is isn't one of ours! */
1999-04-28 04:34:30 +04:00
/*--- initialize the pattern search ---*/
return zip->FindInit( pattern );
}
/**
* ZIP_FindNext
1999-04-28 04:34:30 +04:00
*
* Puts the next name in the passed buffer. Returns ZIP_ERR_SMALLBUF when
* the name is too large for the buffer, and ZIP_ERR_FNF when there are no
* more files that match the pattern
*
1999-05-19 07:21:03 +04:00
* @param hFind handle obtained from ZIP_FindInit
1999-04-28 04:34:30 +04:00
* @param outbuf buffer to receive next filename
* @param bufsize size of allocated buffer
*/
1999-05-19 07:21:03 +04:00
PR_PUBLIC_API(PRInt32) ZIP_FindNext( void* hFind, char * outbuf, PRUint16 bufsize )
1999-04-28 04:34:30 +04:00
{
PRInt32 status;
1999-04-28 04:34:30 +04:00
/*--- error check args ---*/
1999-05-19 07:21:03 +04:00
if ( hFind == 0 )
1999-04-28 04:34:30 +04:00
return ZIP_ERR_PARAM;
1999-05-19 07:21:03 +04:00
nsZipFind* find = NS_STATIC_CAST(nsZipFind*,hFind);
if ( find->kMagic != ZIPFIND_MAGIC )
1999-04-28 04:34:30 +04:00
return ZIP_ERR_PARAM; /* whatever it is isn't one of ours! */
/*--- return next filename file ---*/
nsZipItem* item;
status = find->mArchive->FindNext( find, &item );
if ( status == ZIP_OK )
{
if ( bufsize > item->namelen )
{
PL_strcpy( outbuf, item->name );
}
else
status = ZIP_ERR_SMALLBUF;
}
return status;
1999-04-28 04:34:30 +04:00
}
1999-05-19 07:21:03 +04:00
/**
* ZIP_FindFree
*
* Releases allocated memory associated with the find token
*
* @param hFind handle obtained from ZIP_FindInit
*/
PR_PUBLIC_API(PRInt32) ZIP_FindFree( void* hFind )
{
/*--- error check args ---*/
if ( hFind == 0 )
return ZIP_ERR_PARAM;
nsZipFind* find = NS_STATIC_CAST(nsZipFind*,hFind);
if ( find->kMagic != ZIPFIND_MAGIC )
return ZIP_ERR_PARAM; /* whatever it is isn't one of ours! */
/* free the find structure */
return find->mArchive->FindFree( find );
}
#endif /* STANDALONE */
1999-05-19 07:21:03 +04:00
//***********************************************************
// nsZipArchive -- public methods
//***********************************************************
//---------------------------------------------
// nsZipArchive::OpenArchive
//---------------------------------------------
PRInt32 nsZipArchive::OpenArchive( const char * aArchiveName )
{
//-- validate arguments
1999-05-19 07:21:03 +04:00
if ( aArchiveName == 0 || *aArchiveName == '\0')
return ZIP_ERR_PARAM;
//-- not allowed to do two opens on the same object!
1999-05-19 07:21:03 +04:00
if ( mFd != 0 )
return ZIP_ERR_GENERAL;
//-- open the physical file
mFd = PR_Open( aArchiveName, PR_RDONLY, 0 );
1999-05-19 07:21:03 +04:00
if ( mFd == 0 )
return ZIP_ERR_DISK;
//-- get table of contents for archive
return BuildFileList();
}
//---------------------------------------------
// nsZipArchive::CloseArchive
//---------------------------------------------
PRInt32 nsZipArchive::CloseArchive()
{
// close the file if open
if ( mFd != 0 ) {
PR_Close(mFd);
}
// delete nsZipItems in table
nsZipItem* pItem;
for ( int i = 0; i < ZIP_TABSIZE; ++i)
{
pItem = mFiles[i];
while ( pItem != 0 )
{
mFiles[i] = pItem->next;
delete pItem;
pItem = mFiles[i];
}
}
return ZIP_OK;
}
//---------------------------------------------
// nsZipArchive::ItemSize
//---------------------------------------------
PRInt32 nsZipArchive::GetItem( const char * aFilename, nsZipItem *result)
{
//-- Parameter validity check
if (aFilename == 0)
return ZIP_ERR_PARAM;
//PRInt32 result;
nsZipItem* aItem;
//-- find file information
aItem = GetFileItem( aFilename );
if ( aItem == 0 )
{
return ZIP_ERR_FNF;
}
memcpy(result, &aItem, sizeof(aItem)); // copy struct
return ZIP_OK;
}
//---------------------------------------------
// nsZipArchive::ReadInit
//---------------------------------------------
PRInt32 nsZipArchive::ReadInit(const char* aFilename, nsZipRead** aRead)
{
//-- Parameter validity check
if (aFilename == 0 || aRead == 0)
return ZIP_ERR_PARAM;
PRInt32 result;
//-- find item and seek to file position
nsZipItem* item;
result = ReadInitImpl(aFilename, &item);
if (result != ZIP_OK) return result;
//-- Create nsZipRead object
*aRead = new nsZipRead(this, item);
if (aRead == 0)
return ZIP_ERR_MEMORY;
//-- initialize crc
(*aRead)->mCRC32 = crc32(0L, Z_NULL, 0);
//-- If file is deflated, do the inflation now.
// We save the complete inflated item in mInflatedFileBuffer
// to be copied later.
if (item->compression == DEFLATED)
{
char* buf = (char*)PR_Malloc(item->realsize);
result = InflateItem(item, 0, buf);
if (result != ZIP_OK) return result;
(*aRead)->mInflatedFileBuffer = buf;
}
return ZIP_OK;
}
//---------------------------------------------
// nsZipArchive::Read
//---------------------------------------------
PRInt32 nsZipArchive::Read(nsZipRead* aRead, char* aBuf,
PRUint32 aCount, PRUint32* aBytesRead)
{
//-- Parameter validity check
/*
* Note the nsZipRead passed in must have been created
* by this nsZipArchive.
*/
if (aBytesRead == 0 || aRead == 0 || aBuf == 0 ||
aRead->mArchive != this)
return ZIP_ERR_PARAM;
//-- extract the file using appropriate method
switch( aRead->mItem->compression )
{
case STORED:
//-- Read from the zip file directly into the caller's buffer
return ReadItem( aRead, aBuf, aCount, aBytesRead );
case DEFLATED:
//-- We've already done the inflation; copy from mInflatedFileBuffer
// into the caller's buffer
return ReadInflatedItem( aRead->mItem, aRead->mInflatedFileBuffer, aBuf,
&(aRead->mCurPos), aCount, aBytesRead);
default:
//-- unsupported compression type
return ZIP_ERR_UNSUPPORTED;
}
}
//---------------------------------------------
// nsZipArchive::Available
//---------------------------------------------
PRUint32 nsZipArchive::Available(nsZipRead* aRead)
{
if (aRead == 0)
return 0;
nsZipItem* item = aRead->mItem;
if (item->compression == DEFLATED)
return item->realsize - aRead->mCurPos;
else
return item->size - aRead->mCurPos;
}
1999-04-28 04:34:30 +04:00
//---------------------------------------------
// nsZipArchive::ExtractFile
//---------------------------------------------
PRInt32 nsZipArchive::ExtractFile(const char* aFilename, const char* aOutname)
{
//-- sanity check arguments
1999-05-19 07:21:03 +04:00
if ( aFilename == 0 || aOutname == 0 )
return ZIP_ERR_PARAM;
PRInt32 status;
nsZipItem* item;
//-- Find item in archive and seek to it
status = ReadInitImpl( aFilename, &item );
if (status != ZIP_OK)
return status;
//-- extract the file using the appropriate method
switch( item->compression )
{
case STORED:
status = CopyItemToDisk( item, aOutname );
break;
case DEFLATED:
status = InflateItem( item, aOutname, 0 );
break;
default:
//-- unsupported compression type
return ZIP_ERR_UNSUPPORTED;
}
if (status != ZIP_OK)
PR_Delete(aOutname);
#if defined(XP_UNIX) || defined(XP_PC)
else
{
//-- set extracted file permissions
chmod(aOutname, item->mode);
}
#endif
return status;
}
1999-04-28 04:34:30 +04:00
//---------------------------------------------
// nsZipArchive::FindInit
//---------------------------------------------
1999-05-19 07:21:03 +04:00
nsZipFind* nsZipArchive::FindInit( const char * aPattern )
1999-04-28 04:34:30 +04:00
{
1999-05-19 07:21:03 +04:00
PRBool regExp;
char* pattern = 0;
1999-04-28 04:34:30 +04:00
// validate the pattern
1999-05-19 07:21:03 +04:00
if ( aPattern )
{
1999-04-28 04:34:30 +04:00
switch (XP_RegExpValid( (char*)aPattern ))
{
case INVALID_SXP:
1999-05-19 07:21:03 +04:00
return 0;
1999-04-28 04:34:30 +04:00
case NON_SXP:
1999-05-19 07:21:03 +04:00
regExp = PR_FALSE;
1999-04-28 04:34:30 +04:00
break;
case VALID_SXP:
1999-05-19 07:21:03 +04:00
regExp = PR_TRUE;
1999-04-28 04:34:30 +04:00
break;
default:
1999-05-19 07:21:03 +04:00
// undocumented return value from RegExpValid!
1999-04-28 04:34:30 +04:00
PR_ASSERT( PR_FALSE );
1999-05-19 07:21:03 +04:00
return 0;
1999-04-28 04:34:30 +04:00
}
1999-05-19 07:21:03 +04:00
pattern = PL_strdup( aPattern );
if ( !pattern )
return 0;
1999-04-28 04:34:30 +04:00
}
1999-05-19 07:21:03 +04:00
return new nsZipFind( this, pattern, regExp );
1999-04-28 04:34:30 +04:00
}
//---------------------------------------------
// nsZipArchive::FindNext
//---------------------------------------------
PRInt32 nsZipArchive::FindNext( nsZipFind* aFind, nsZipItem** aResult)
1999-04-28 04:34:30 +04:00
{
PRInt32 status;
PRBool found = PR_FALSE;
1999-05-19 07:21:03 +04:00
PRUint16 slot = aFind->mSlot;
nsZipItem* item = aFind->mItem;
if ( aFind->mArchive != this )
1999-05-19 07:21:03 +04:00
return ZIP_ERR_PARAM;
1999-04-28 04:34:30 +04:00
// we start from last match, look for next
while ( slot < ZIP_TABSIZE && !found )
{
1999-05-19 07:21:03 +04:00
if ( item != 0 )
1999-04-28 04:34:30 +04:00
item = item->next; // move to next in current chain
else
item = mFiles[slot]; // starting a new slot
1999-05-19 07:21:03 +04:00
if ( item == 0 )
1999-04-28 04:34:30 +04:00
{ // no more in this chain, move to next slot
++slot;
continue;
}
1999-05-19 07:21:03 +04:00
else if ( aFind->mPattern == 0 )
1999-04-28 04:34:30 +04:00
found = PR_TRUE; // always match
1999-05-19 07:21:03 +04:00
else if ( aFind->mRegExp )
1999-06-10 05:58:06 +04:00
found = (XP_RegExpMatch( item->name, aFind->mPattern, PR_FALSE ) == MATCH);
1999-04-28 04:34:30 +04:00
else
#if defined(STANDALONE) && defined(XP_MAC)
// simulate <regexp>* matches
found = ( strncmp( item->name, aFind->mPattern, strlen(aFind->mPattern) ) == 0 );
#else
1999-05-19 07:21:03 +04:00
found = ( PL_strcmp( item->name, aFind->mPattern ) == 0 );
#endif
1999-04-28 04:34:30 +04:00
}
if ( found )
{
*aResult = item;
aFind->mSlot = slot;
aFind->mItem = item;
1999-04-28 04:34:30 +04:00
status = ZIP_OK;
}
else
status = ZIP_ERR_FNF;
return status;
}
1999-05-19 07:21:03 +04:00
//---------------------------------------------
// nsZipArchive::FindFree
//---------------------------------------------
PRInt32 nsZipArchive::FindFree( nsZipFind* aFind )
{
if ( aFind->mArchive != this )
return ZIP_ERR_PARAM;
delete aFind;
return ZIP_OK;
}
1999-04-28 04:34:30 +04:00
//***********************************************************
// nsZipArchive -- private implementation
//***********************************************************
//---------------------------------------------
// nsZipArchive::BuildFileList
//---------------------------------------------
PRInt32 nsZipArchive::BuildFileList()
{
PRInt32 status = ZIP_OK;
PRUint32 sig = 0L;
PRUint32 namelen, extralen, commentlen;
1999-05-19 07:21:03 +04:00
PRUint32 hash;
PRUint32 size;
ZipLocal Local;
ZipCentral Central;
nsZipItem* item;
1999-04-28 04:34:30 +04:00
//-----------------------------------------------------------------------
// skip to the central directory
1999-04-28 04:34:30 +04:00
//-----------------------------------------------------------------------
PRInt32 pos = 0L;
while ( status == ZIP_OK )
{
#ifndef STANDALONE
if ( PR_Seek( mFd, pos, PR_SEEK_SET ) != (PRInt32)pos )
#else
// For standalone, PR_Seek() is stubbed with fseek(), which returns 0
// if successfull, otherwise a non-zero.
if ( PR_Seek( mFd, pos, PR_SEEK_SET ) != 0 )
#endif
{
//-- couldn't seek to next position
status = ZIP_ERR_CORRUPT;
break;
}
if ( PR_Read( mFd, (char*)&Local, sizeof(ZipLocal) ) != sizeof(ZipLocal) )
{
//-- file ends prematurely
status = ZIP_ERR_CORRUPT;
break;
}
//-- check if we hit the central directory
sig = xtolong( Local.signature );
if ( sig == LOCALSIG )
{
//-- check length of this file and its metadata so we can skip over it
namelen = xtoint( Local.filename_len );
extralen = xtoint( Local.extrafield_len );
size = xtolong( Local.size );
//-- reposition file mark to next expected local header
pos += sizeof(ZipLocal) + namelen + extralen + size;
}
else if ( sig == CENTRALSIG )
{
// file mark set to start of central directory
break;
}
else
{
//-- otherwise expected to find a local header
status = ZIP_ERR_CORRUPT;
break;
}
} /* while reading local headers */
//-------------------------------------------------------
// read the central directory headers
//-------------------------------------------------------
while ( status == ZIP_OK )
{
#ifndef STANDALONE
if ( PR_Seek( mFd, pos, PR_SEEK_SET ) != (PRInt32)pos )
#else
if ( PR_Seek( mFd, pos, PR_SEEK_SET ) != 0 )
#endif
{
//-- couldn't seek to next position
status = ZIP_ERR_CORRUPT;
break;
}
if ( PR_Read( mFd, (char*)&Central, sizeof(Central) ) != sizeof(ZipCentral) )
{
//-- central dir end record is smaller than a central dir header
sig = xtolong( Central.signature );
if ( sig == ENDSIG )
{
//-- we've reached the end of the central dir
break;
}
else
{
status = ZIP_ERR_CORRUPT;
break;
}
}
sig = xtolong( Central.signature );
if ( sig == ENDSIG )
{
//-- we've reached the end of the central dir
break;
}
else if ( sig != CENTRALSIG )
{
//-- otherwise expected to find a central dir header
status = ZIP_ERR_CORRUPT;
break;
}
namelen = xtoint( Central.filename_len );
extralen = xtoint( Central.extrafield_len );
commentlen = xtoint( Central.commentfield_len );
item = new nsZipItem();
if (item != 0)
{
item->name = new char[namelen+1];
1999-05-19 07:21:03 +04:00
if ( item->name != 0 )
{
1999-05-19 07:21:03 +04:00
if ( PR_Read( mFd, item->name, namelen ) == (READTYPE)namelen )
{
item->name[namelen] = 0;
1999-04-28 04:34:30 +04:00
item->namelen = namelen;
item->headerloc = xtolong( Central.localhdr_offset );
//-- seek to local header
#ifndef STANDALONE
if ( PR_Seek( mFd, item->headerloc, PR_SEEK_SET ) != (PRInt32)item->headerloc )
#else
if ( PR_Seek( mFd, item->headerloc, PR_SEEK_SET ) != 0 )
#endif
{
//-- couldn't seek to next position
status = ZIP_ERR_CORRUPT;
break;
}
//-- read local header to extract local extralen
//-- NOTE: extralen is different in central header and local header
//-- for archives created using the Unix "zip" utility. To set
//-- the offset accurately we need the local extralen.
if ( PR_Read( mFd, (char*)&Local, sizeof(ZipLocal) ) != (READTYPE)sizeof(ZipLocal) )
{
//-- expected a complete local header
status = ZIP_ERR_CORRUPT;
break;
}
item->offset = item->headerloc +
sizeof(ZipLocal) +
namelen +
xtoint( Local.extrafield_len );
item->compression = xtoint( Central.method );
item->size = xtolong( Central.size );
item->realsize = xtolong( Central.orglen );
item->crc32 = xtolong( Central.crc32 );
item->mode = ExtractMode(xtolong( Central.external_attributes ));
//-- add item to file table
hash = HashName( item->name );
item->next = mFiles[hash];
mFiles[hash] = item;
//-- set pos to next expected central dir header
pos += sizeof(ZipCentral) + namelen + extralen + commentlen;
}
else
{
//-- file is truncated
status = ZIP_ERR_CORRUPT;
delete item;
}
}
else
{
//-- couldn't allocate for the filename
status = ZIP_ERR_MEMORY;
delete item;
}
}
else
{
//-- couldn't create an nsZipItem
status = ZIP_ERR_MEMORY;
}
} /* while reading central directory records */
return status;
}
//---------------------------------------------
// nsZipArchive::GetFileItem
//---------------------------------------------
nsZipItem* nsZipArchive::GetFileItem( const char * aFilename )
{
PR_ASSERT( aFilename != 0 );
nsZipItem* item = mFiles[ HashName(aFilename) ];
for ( ; item != 0; item = item->next )
{
if ( 0 == PL_strcmp( aFilename, item->name ) )
break; //-- found it
}
return item;
}
//---------------------------------------------
// nsZipArchive::HashName
//---------------------------------------------
PRUint32 nsZipArchive::HashName( const char* aName )
{
PRUint32 val = 0;
PRUint8* c;
PR_ASSERT( aName != 0 );
for ( c = (PRUint8*)aName; *c != 0; c++ ) {
val = val*37 + *c;
}
return (val % ZIP_TABSIZE);
}
//---------------------------------------------
// nsZipArchive::ReadInitImpl
//---------------------------------------------
PRInt32 nsZipArchive::ReadInitImpl(const char* aFilename, nsZipItem** aItem)
{
PR_ASSERT (aFilename != 0);
//-- find file information
*aItem = GetFileItem( aFilename );
if ( *aItem == 0 )
return ZIP_ERR_FNF;
//-- find start of file in archive
#ifndef STANDALONE
if ( PR_Seek( mFd, (*aItem)->offset, PR_SEEK_SET ) != (PRInt32)(*aItem)->offset )
#else
// For standalone, PR_Seek() is stubbed with fseek(), which returns 0
// if successfull, otherwise a non-zero.
if ( PR_Seek( mFd, (*aItem)->offset, PR_SEEK_SET ) != 0)
#endif
return ZIP_ERR_CORRUPT;
return ZIP_OK;
}
//---------------------------------------------
// nsZipArchive::ReadItem
//---------------------------------------------
PRInt32 nsZipArchive::ReadItem( nsZipRead* aRead, char* aBuf,
PRUint32 aCount, PRUint32* aActual)
{
PRUint32 chunk;
nsZipItem *item = aRead->mItem;
PRUint32 *curpos = &(aRead->mCurPos);
PRUint32 *crc = &(aRead->mCRC32);
PR_ASSERT(aBuf != 0 && item != 0);
// Seek to position in file
if ( PR_Seek( mFd, (PROffset32)(item->offset + *curpos), PR_SEEK_SET ) !=
(PROffset32)(item->offset + *curpos) )
return ZIP_ERR_CORRUPT;
//-- Read from file
chunk = (aCount <= ((item->size) - *curpos) ) ? aCount
: ((item->size) - *curpos);
*aActual = PR_Read( mFd, aBuf, chunk );
*curpos += *aActual;
//-- incrementally update crc32
*crc = crc32(*crc,(const unsigned char*)aBuf, chunk);
//-- verify crc32
if (0 == (item->size - *curpos)) // eof
{
if (*crc != item->crc32)
return ZIP_ERR_CORRUPT;
}
return ZIP_OK;
}
//---------------------------------------------
// nsZipArchive::CopyItemToDisk
//---------------------------------------------
PRInt32 nsZipArchive::CopyItemToDisk( const nsZipItem* aItem, const char* aOutname )
{
PRInt32 status = ZIP_OK;
PRUint32 chunk, pos, size, crc;
1999-05-19 07:21:03 +04:00
PRFileDesc* fOut = 0;
1999-05-19 07:21:03 +04:00
PR_ASSERT( aItem != 0 && aOutname != 0 );
char* buf = (char*)PR_Malloc(ZIP_BUFLEN);
1999-05-19 07:21:03 +04:00
if ( buf == 0 )
return ZIP_ERR_MEMORY;
//-- We should already be at the correct spot in the archive.
//-- ReadInitImpl did the seek().
//-- initialize crc
crc = crc32(0L, Z_NULL, 0);
//-- open output file
1999-05-19 07:21:03 +04:00
fOut = PR_Open( aOutname, ZFILE_CREATE , 0644);
if ( fOut == 0 )
{
status = ZIP_ERR_DISK;
goto cleanup;
}
//-- copy chunks until file is done
size = aItem->size;
for ( pos=0; pos < size; pos += chunk )
{
chunk = (pos+ZIP_BUFLEN <= size) ? ZIP_BUFLEN : size - pos;
1999-05-19 07:21:03 +04:00
if ( PR_Read( mFd, buf, chunk ) != (READTYPE)chunk )
{
//-- unexpected end of data in archive
status = ZIP_ERR_CORRUPT;
break;
}
//-- incrementally update crc32
crc = crc32(crc, (const unsigned char*)buf, chunk);
1999-05-19 07:21:03 +04:00
if ( PR_Write( fOut, buf, chunk ) < (READTYPE)chunk )
{
//-- Couldn't write all the data (disk full?)
status = ZIP_ERR_DISK;
break;
}
}
//-- verify crc32
if ( (status == ZIP_OK) && (crc != aItem->crc32) )
status = ZIP_ERR_CORRUPT;
cleanup:
1999-05-19 07:21:03 +04:00
if ( fOut != 0 )
PR_Close( fOut );
PR_FREEIF( buf );
return status;
}
//---------------------------------------------
// nsZipArchive::InflateItem
//---------------------------------------------
PRInt32 nsZipArchive::InflateItem( const nsZipItem* aItem, const char* aOutname,
char* bigBuf )
/*
* This function either inflates an archive item to disk, to the
* file specified by aOutname, or into a buffer specified by
* bigBuf. bigBuf then gets copied into the "real" output
* buffer a chunk at a time by ReadInflatedItem(). Memory-wise,
* this is inefficient, since it stores an entire copy of the
* decompressed item in memory, then copies it to the caller's
* buffer. A more memory-efficient implementation is possible,
* in which the outbuf in this function could be returned
* directly, but implementing it would be complex.
*/
{
PRInt32 status = ZIP_OK;
PRUint32 chunk, inpos, outpos, size, crc;
PRUint32 bigBufSize;
1999-05-19 07:21:03 +04:00
PRFileDesc* fOut = 0;
z_stream zs;
int zerr;
PRBool bInflating = PR_FALSE;
PRBool bRead;
PRBool bWrote;
PRBool bToFile;
#if defined STANDALONE && defined WIN32
MSG msg;
#endif /* STANDALONE */
// -- if aOutname is null, we'll be writing to a buffer instead of a file
if (aOutname != 0)
{
PR_ASSERT( aItem != 0 );
bToFile = PR_TRUE;
}
else
{
// -- Writing to a buffer, so bigBuf must not be null.
PR_ASSERT( aItem != 0 && bigBuf != 0 );
bToFile = PR_FALSE;
bigBufSize = aItem->realsize;
}
//-- allocate deflation buffers
Bytef *inbuf = (Bytef*)PR_Malloc(ZIP_BUFLEN);
Bytef *outbuf = (Bytef*)PR_Malloc(ZIP_BUFLEN);
1999-05-19 07:21:03 +04:00
if ( inbuf == 0 || outbuf == 0 )
{
status = ZIP_ERR_MEMORY;
goto cleanup;
}
//-- initialize crc
crc = crc32(0L, Z_NULL, 0);
//-- We should already be at the correct spot in the archive.
//-- ReadInitImpl did the seek().
if (bToFile)
{
//-- open output file
fOut = PR_Open( aOutname, ZFILE_CREATE, 0644);
if ( fOut == 0 )
{
status = ZIP_ERR_DISK;
goto cleanup;
}
}
//-- set up the inflate
memset( &zs, 0, sizeof(zs) );
zerr = inflateInit2( &zs, -MAX_WBITS );
if ( zerr != Z_OK )
{
status = ZIP_ERR_GENERAL;
goto cleanup;
}
bInflating = PR_TRUE;
//-- inflate loop
size = aItem->size;
outpos = inpos = 0;
zs.next_out = outbuf;
zs.avail_out = ZIP_BUFLEN;
while ( zerr == Z_OK )
{
bRead = PR_FALSE;
bWrote = PR_FALSE;
if ( zs.avail_in == 0 && zs.total_in < size )
{
//-- no data to inflate yet still more in file:
//-- read another chunk of compressed data
inpos = zs.total_in; // input position
chunk = ( inpos + ZIP_BUFLEN <= size ) ? ZIP_BUFLEN : size - inpos;
1999-05-19 07:21:03 +04:00
if ( PR_Read( mFd, inbuf, chunk ) != (READTYPE)chunk )
{
//-- unexpected end of data
status = ZIP_ERR_CORRUPT;
break;
}
zs.next_in = inbuf;
zs.avail_in = chunk;
bRead = PR_TRUE;
}
if ( zs.avail_out == 0 )
{
if (bToFile)
{
//-- write inflated buffer to disk and make space
if ( PR_Write( fOut, outbuf, ZIP_BUFLEN ) < ZIP_BUFLEN )
{
//-- Couldn't write all the data (disk full?)
status = ZIP_ERR_DISK;
break;
}
}
else
{
//-- copy inflated buffer to our big buffer
// Assertion makes sure we don't overflow bigBuf
PR_ASSERT( outpos + ZIP_BUFLEN <= bigBufSize);
char* copyStart = bigBuf + outpos;
memcpy(copyStart, outbuf, ZIP_BUFLEN);
}
outpos = zs.total_out;
zs.next_out = outbuf;
zs.avail_out = ZIP_BUFLEN;
bWrote = PR_TRUE;
}
if(bRead || bWrote)
zerr = inflate( &zs, Z_PARTIAL_FLUSH );
else
zerr = Z_STREAM_END;
//-- incrementally update crc32
crc = crc32(crc, (const unsigned char*)outbuf, zs.total_out - outpos);
#if defined STANDALONE && defined WIN32
while(PeekMessage(&msg, 0, 0, 0, PM_REMOVE))
{
TranslateMessage(&msg);
DispatchMessage(&msg);
}
#endif /* STANDALONE */
} // while
//-- verify crc32
if ( (status == ZIP_OK) && (crc != aItem->crc32) )
{
status = ZIP_ERR_CORRUPT;
goto cleanup;
}
//-- write last inflated bit to disk
1999-03-10 02:06:11 +03:00
if ( zerr == Z_STREAM_END && outpos < zs.total_out )
{
chunk = zs.total_out - outpos;
if (bToFile)
{
if ( PR_Write( fOut, outbuf, chunk ) < (READTYPE)chunk )
status = ZIP_ERR_DISK;
}
else
{
PR_ASSERT( (outpos + chunk) <= bigBufSize );
char* copyStart = bigBuf + outpos;
memcpy(copyStart, outbuf, chunk);
}
}
//-- convert zlib error to return value
if ( status == ZIP_OK && zerr != Z_OK && zerr != Z_STREAM_END )
{
status = (zerr == Z_MEM_ERROR) ? ZIP_ERR_MEMORY : ZIP_ERR_CORRUPT;
}
//-- if found no errors make sure we've converted the whole thing
PR_ASSERT( status != ZIP_OK || zs.total_in == aItem->size );
PR_ASSERT( status != ZIP_OK || zs.total_out == aItem->realsize );
cleanup:
if ( bInflating )
{
//-- free zlib internal state
inflateEnd( &zs );
}
if (fOut != 0 )
PR_Close( fOut );
PR_FREEIF( inbuf );
PR_FREEIF( outbuf );
return status;
}
//------------------------------------------
// nsZipArchive::ReadInflatedItem
//------------------------------------------
PRInt32 nsZipArchive::ReadInflatedItem( const nsZipItem* aItem,
char* inflatedBuf,
char* outbuf,
PRUint32* aCurPos,
PRUint32 count,
PRUint32* actual)
{
//-- Parameter check
PR_ASSERT(inflatedBuf != 0 && outbuf != 0 && aItem != 0);
//-- Set up the copy
PRUint32 bigBufSize = aItem->realsize;
PRUint32 c = ((*aCurPos + count) < bigBufSize) ? count : bigBufSize - *aCurPos;
char* src = inflatedBuf + (*aCurPos);
//-- Do the copy and record number of bytes copied
memcpy(outbuf, src, c);
*actual = c;
*aCurPos += *actual;
return ZIP_OK;
}
//------------------------------------------
// nsZipArchive constructor and destructor
//------------------------------------------
nsZipArchive::nsZipArchive()
1999-05-19 07:21:03 +04:00
: kMagic(ZIP_MAGIC), mFd(0)
{
// initialize the table to NULL
for ( int i = 0; i < ZIP_TABSIZE; ++i) {
1999-05-19 07:21:03 +04:00
mFiles[i] = 0;
}
}
nsZipArchive::~nsZipArchive()
{
(void)CloseArchive();
}
1999-05-19 07:21:03 +04:00
//------------------------------------------
// nsZipItem constructor and destructor
//------------------------------------------
1999-05-19 07:21:03 +04:00
nsZipItem::nsZipItem() : name(0),next(0) {}
nsZipItem::~nsZipItem()
{
1999-05-19 07:21:03 +04:00
if (name != 0 )
delete [] name;
}
PRInt32
nsZipItem::Init(nsZipItem* other)
{
if (name) delete[] name;
name = (char*)PR_Malloc(other->namelen);
if (name == 0)
return ZIP_ERR_MEMORY;
memcpy(name, other->name, other->namelen);
namelen = other->namelen;
offset = other->offset;
headerloc = other->headerloc;
compression = other->compression;
size = other->size;
realsize = other->realsize;
crc32 = other->crc32;
mode = other->mode;
next = 0; // don't copy next
return ZIP_OK;
}
//------------------------------------------
// nsZipRead constructor and destructor
//------------------------------------------
nsZipRead::nsZipRead( nsZipArchive* aZipArchive, nsZipItem* aZipItem )
: mArchive(aZipArchive),
mItem(aZipItem),
mCurPos(0),
mInflatedFileBuffer(0)
{}
nsZipRead::~nsZipRead()
{
PR_FREEIF(mInflatedFileBuffer);
}
1999-05-19 07:21:03 +04:00
//------------------------------------------
// nsZipFind constructor and destructor
//------------------------------------------
nsZipFind::nsZipFind( nsZipArchive* aZip, char* aPattern, PRBool aRegExp )
: kMagic(ZIPFIND_MAGIC),
mArchive(aZip),
mPattern(aPattern),
mSlot(0),
mItem(0),
mRegExp(aRegExp)
{}
nsZipFind::~nsZipFind()
{
if (mPattern != 0 )
PL_strfree( mPattern );
}
//------------------------------------------
// nsZipFind::GetArchive
//------------------------------------------
nsZipArchive* nsZipFind::GetArchive()
{
if (!mArchive)
return NULL;
return mArchive;
}
1999-05-19 07:21:03 +04:00
//------------------------------------------
1999-04-28 04:34:30 +04:00
// helper functions
//------------------------------------------
/*
* x t o i n t
*
* Converts a two byte ugly endianed integer
* to our platform's integer.
*/
static PRUint16 xtoint (unsigned char *ii)
{
1999-05-19 07:21:03 +04:00
return (PRUint16) ((ii [0]) | (ii [1] << 8));
}
/*
* x t o l o n g
*
* Converts a four byte ugly endianed integer
* to our platform's integer.
*/
static PRUint32 xtolong (unsigned char *ll)
{
PRUint32 ret;
ret = (
(((PRUint32) ll [0]) << 0) |
(((PRUint32) ll [1]) << 8) |
(((PRUint32) ll [2]) << 16) |
(((PRUint32) ll [3]) << 24)
);
return ret;
}
/*
* ExtractMode
*
* Extracts bits 17-24 from a 32-bit unsigned long
* representation of the external attributes field.
* Subsequently it tacks on the implicit user-read
* bit.
*/
static PRUint16 ExtractMode(PRUint32 ext_attr)
{
ext_attr &= 0x00FF0000;
ext_attr >>= 16;
ext_attr |= 0x00000100;
return (PRUint16) ext_attr;
}