mirror of
https://github.com/moparisthebest/k-9
synced 2024-11-15 22:15:15 -05:00
416 lines
13 KiB
Java
416 lines
13 KiB
Java
|
|
package com.fsck.k9.mail.internet;
|
|
|
|
import android.util.Log;
|
|
import com.fsck.k9.K9;
|
|
import com.fsck.k9.mail.*;
|
|
import org.apache.commons.io.IOUtils;
|
|
import org.apache.james.mime4j.decoder.Base64InputStream;
|
|
import org.apache.james.mime4j.decoder.DecoderUtil;
|
|
import org.apache.james.mime4j.decoder.QuotedPrintableInputStream;
|
|
|
|
import java.io.ByteArrayOutputStream;
|
|
import java.io.IOException;
|
|
import java.io.InputStream;
|
|
import java.io.OutputStream;
|
|
import java.util.ArrayList;
|
|
import java.util.regex.Pattern;
|
|
import java.nio.charset.Charset;
|
|
|
|
|
|
public class MimeUtility
|
|
{
|
|
public static final String DEFAULT_ATTACHMENT_MIME_TYPE = "application/octet-stream";
|
|
|
|
public static String unfold(String s)
|
|
{
|
|
if (s == null)
|
|
{
|
|
return null;
|
|
}
|
|
return s.replaceAll("\r|\n", "");
|
|
}
|
|
|
|
public static String decode(String s)
|
|
{
|
|
if (s == null)
|
|
{
|
|
return null;
|
|
}
|
|
return DecoderUtil.decodeEncodedWords(s);
|
|
}
|
|
|
|
public static String unfoldAndDecode(String s)
|
|
{
|
|
return decode(unfold(s));
|
|
}
|
|
|
|
// TODO implement proper foldAndEncode
|
|
public static String foldAndEncode(String s)
|
|
{
|
|
return s;
|
|
}
|
|
|
|
/**
|
|
* Returns the named parameter of a header field. If name is null the first
|
|
* parameter is returned, or if there are no additional parameters in the
|
|
* field the entire field is returned. Otherwise the named parameter is
|
|
* searched for in a case insensitive fashion and returned. If the parameter
|
|
* cannot be found the method returns null.
|
|
*
|
|
* @param header
|
|
* @param name
|
|
* @return
|
|
*/
|
|
public static String getHeaderParameter(String header, String name)
|
|
{
|
|
if (header == null)
|
|
{
|
|
return null;
|
|
}
|
|
header = header.replaceAll("\r|\n", "");
|
|
String[] parts = header.split(";");
|
|
if (name == null)
|
|
{
|
|
return parts[0];
|
|
}
|
|
for (String part : parts)
|
|
{
|
|
if (part.trim().toLowerCase().startsWith(name.toLowerCase()))
|
|
{
|
|
String parameter = part.split("=", 2)[1].trim();
|
|
if (parameter.startsWith("\"") && parameter.endsWith("\""))
|
|
{
|
|
return parameter.substring(1, parameter.length() - 1);
|
|
}
|
|
else
|
|
{
|
|
return parameter;
|
|
}
|
|
}
|
|
}
|
|
return null;
|
|
}
|
|
|
|
public static Part findFirstPartByMimeType(Part part, String mimeType)
|
|
throws MessagingException
|
|
{
|
|
if (part.getBody() instanceof Multipart)
|
|
{
|
|
Multipart multipart = (Multipart)part.getBody();
|
|
for (int i = 0, count = multipart.getCount(); i < count; i++)
|
|
{
|
|
BodyPart bodyPart = multipart.getBodyPart(i);
|
|
Part ret = findFirstPartByMimeType(bodyPart, mimeType);
|
|
if (ret != null)
|
|
{
|
|
return ret;
|
|
}
|
|
}
|
|
}
|
|
else if (part.getMimeType().equalsIgnoreCase(mimeType))
|
|
{
|
|
return part;
|
|
}
|
|
return null;
|
|
}
|
|
|
|
public static Part findPartByContentId(Part part, String contentId) throws Exception
|
|
{
|
|
if (part.getBody() instanceof Multipart)
|
|
{
|
|
Multipart multipart = (Multipart)part.getBody();
|
|
for (int i = 0, count = multipart.getCount(); i < count; i++)
|
|
{
|
|
BodyPart bodyPart = multipart.getBodyPart(i);
|
|
Part ret = findPartByContentId(bodyPart, contentId);
|
|
if (ret != null)
|
|
{
|
|
return ret;
|
|
}
|
|
}
|
|
}
|
|
String[] header = part.getHeader("Content-ID");
|
|
if (header != null)
|
|
{
|
|
for (String s : header)
|
|
{
|
|
if (s.equals(contentId))
|
|
{
|
|
return part;
|
|
}
|
|
}
|
|
}
|
|
return null;
|
|
}
|
|
|
|
/**
|
|
* Reads the Part's body and returns a String based on any charset conversion that needed
|
|
* to be done.
|
|
* @param part The part containing a body
|
|
* @return a String containing the converted text in the body, or null if there was no text
|
|
* or an error during conversion.
|
|
*/
|
|
public static String getTextFromPart(Part part)
|
|
{
|
|
try
|
|
{
|
|
if ((part != null) && (part.getBody() != null))
|
|
{
|
|
final Body body = part.getBody();
|
|
if (body instanceof TextBody)
|
|
{
|
|
return ((TextBody)body).getText();
|
|
}
|
|
|
|
final String mimeType = part.getMimeType();
|
|
if ((mimeType != null) && MimeUtility.mimeTypeMatches(mimeType, "text/*"))
|
|
{
|
|
/*
|
|
* We've got a text part, so let's see if it needs to be processed further.
|
|
*/
|
|
final String originalCharset = getHeaderParameter(part.getContentType(), "charset");
|
|
String charset = "ASCII"; // No encoding, so use us-ascii, which is the standard.
|
|
if ((originalCharset != null) && ("0".equals(originalCharset) == false))
|
|
{
|
|
/*
|
|
* See if there is conversion from the MIME charset to the Java one.
|
|
*/
|
|
|
|
charset = Charset.forName(fixupCharset(originalCharset)).name();
|
|
|
|
if (charset == null)
|
|
{
|
|
Log.e(K9.LOG_TAG,"I don't know how to deal with the charset "+originalCharset+". Falling back to US-ASCII");
|
|
charset = "US-ASCII";
|
|
}
|
|
}
|
|
|
|
/*
|
|
* Now we read the part into a buffer for further processing. Because
|
|
* the stream is now wrapped we'll remove any transfer encoding at this point.
|
|
*/
|
|
InputStream in = part.getBody().getInputStream();
|
|
ByteArrayOutputStream out = new ByteArrayOutputStream(in.available());
|
|
IOUtils.copy(in, out);
|
|
in.close();
|
|
in = null; // we want all of our memory back, and close might not release
|
|
// Cargo culted from AOSP - This disagrees with the platform docs
|
|
|
|
/*
|
|
* Convert and return as new String
|
|
*/
|
|
final String result = out.toString(charset);
|
|
out.close();
|
|
return result;
|
|
}
|
|
}
|
|
|
|
}
|
|
catch (OutOfMemoryError oom)
|
|
{
|
|
/*
|
|
* If we are not able to process the body there's nothing we can do about it. Return
|
|
* null and let the upper layers handle the missing content.
|
|
*/
|
|
Log.e(K9.LOG_TAG, "Unable to getTextFromPart " + oom.toString());
|
|
}
|
|
catch (Exception e)
|
|
{
|
|
/*
|
|
* If we are not able to process the body there's nothing we can do about it. Return
|
|
* null and let the upper layers handle the missing content.
|
|
*/
|
|
Log.e(K9.LOG_TAG, "Unable to getTextFromPart", e);
|
|
}
|
|
return null;
|
|
}
|
|
|
|
/**
|
|
* Returns true if the given mimeType matches the matchAgainst specification.
|
|
* @param mimeType A MIME type to check.
|
|
* @param matchAgainst A MIME type to check against. May include wildcards such as image/* or
|
|
* * /*.
|
|
* @return
|
|
*/
|
|
public static boolean mimeTypeMatches(String mimeType, String matchAgainst)
|
|
{
|
|
Pattern p = Pattern.compile(matchAgainst.replaceAll("\\*", "\\.\\*"),
|
|
Pattern.CASE_INSENSITIVE);
|
|
return p.matcher(mimeType).matches();
|
|
}
|
|
|
|
/**
|
|
* Returns true if the given mimeType matches any of the matchAgainst specifications.
|
|
* @param mimeType A MIME type to check.
|
|
* @param matchAgainst An array of MIME types to check against. May include wildcards such
|
|
* as image/* or * /*.
|
|
* @return
|
|
*/
|
|
public static boolean mimeTypeMatches(String mimeType, String[] matchAgainst)
|
|
{
|
|
for (String matchType : matchAgainst)
|
|
{
|
|
if (mimeTypeMatches(mimeType, matchType))
|
|
{
|
|
return true;
|
|
}
|
|
}
|
|
return false;
|
|
}
|
|
|
|
/**
|
|
* Removes any content transfer encoding from the stream and returns a Body.
|
|
*/
|
|
public static Body decodeBody(InputStream in, String contentTransferEncoding)
|
|
throws IOException
|
|
{
|
|
/*
|
|
* We'll remove any transfer encoding by wrapping the stream.
|
|
*/
|
|
if (contentTransferEncoding != null)
|
|
{
|
|
contentTransferEncoding =
|
|
MimeUtility.getHeaderParameter(contentTransferEncoding, null);
|
|
if ("quoted-printable".equalsIgnoreCase(contentTransferEncoding))
|
|
{
|
|
in = new QuotedPrintableInputStream(in);
|
|
}
|
|
else if ("base64".equalsIgnoreCase(contentTransferEncoding))
|
|
{
|
|
in = new Base64InputStream(in);
|
|
}
|
|
}
|
|
|
|
BinaryTempFileBody tempBody = new BinaryTempFileBody();
|
|
OutputStream out = tempBody.getOutputStream();
|
|
IOUtils.copy(in, out);
|
|
out.close();
|
|
return tempBody;
|
|
}
|
|
|
|
/**
|
|
* An unfortunately named method that makes decisions about a Part (usually a Message)
|
|
* as to which of it's children will be "viewable" and which will be attachments.
|
|
* The method recursively sorts the viewables and attachments into seperate
|
|
* lists for further processing.
|
|
* @param part
|
|
* @param viewables
|
|
* @param attachments
|
|
* @throws MessagingException
|
|
*/
|
|
public static void collectParts(Part part, ArrayList<Part> viewables,
|
|
ArrayList<Part> attachments) throws MessagingException
|
|
{
|
|
/*
|
|
* If the part is Multipart but not alternative it's either mixed or
|
|
* something we don't know about, which means we treat it as mixed
|
|
* per the spec. We just process it's pieces recursively.
|
|
*/
|
|
if (part.getBody() instanceof Multipart)
|
|
{
|
|
Multipart mp = (Multipart)part.getBody();
|
|
for (int i = 0; i < mp.getCount(); i++)
|
|
{
|
|
collectParts(mp.getBodyPart(i), viewables, attachments);
|
|
}
|
|
}
|
|
/*
|
|
* If the part is an embedded message we just continue to process
|
|
* it, pulling any viewables or attachments into the running list.
|
|
*/
|
|
else if (part.getBody() instanceof Message)
|
|
{
|
|
Message message = (Message)part.getBody();
|
|
collectParts(message, viewables, attachments);
|
|
}
|
|
/*
|
|
* If the part is HTML and it got this far it's part of a mixed (et
|
|
* al) and should be rendered inline.
|
|
*/
|
|
else if (isPartTextualBody(part))
|
|
{
|
|
viewables.add(part);
|
|
}
|
|
else
|
|
{
|
|
attachments.add(part);
|
|
}
|
|
|
|
}
|
|
|
|
|
|
public static Boolean isPartTextualBody(Part part) throws MessagingException
|
|
{
|
|
String disposition = part.getDisposition();
|
|
String dispositionType = null;
|
|
String dispositionFilename = null;
|
|
if (disposition != null)
|
|
{
|
|
dispositionType = MimeUtility.getHeaderParameter(disposition, null);
|
|
dispositionFilename = MimeUtility.getHeaderParameter(disposition, "filename");
|
|
}
|
|
|
|
/*
|
|
* A best guess that this part is intended to be an attachment and not inline.
|
|
*/
|
|
boolean attachment = ("attachment".equalsIgnoreCase(dispositionType)
|
|
|| (dispositionFilename != null));
|
|
|
|
if ((!attachment) && (part.getMimeType().equalsIgnoreCase("text/html")))
|
|
{
|
|
return true;
|
|
}
|
|
/*
|
|
* If the part is plain text and it got this far it's part of a
|
|
* mixed (et al) and should be rendered inline.
|
|
*/
|
|
else if ((!attachment) && (part.getMimeType().equalsIgnoreCase("text/plain")))
|
|
{
|
|
return true;
|
|
}
|
|
/*
|
|
* Finally, if it's nothing else we will include it as an attachment.
|
|
*/
|
|
else
|
|
{
|
|
return false;
|
|
}
|
|
}
|
|
|
|
|
|
public static String getMimeTypeByExtension(String filename)
|
|
{
|
|
String returnedType = null;
|
|
if (filename != null
|
|
&& filename.lastIndexOf('.') != -1)
|
|
{
|
|
String extension = filename.substring(filename.lastIndexOf('.')+1).toLowerCase();
|
|
returnedType = android.webkit.MimeTypeMap.getSingleton().getMimeTypeFromExtension(extension);
|
|
}
|
|
if (returnedType != null)
|
|
{
|
|
return returnedType;
|
|
}
|
|
else
|
|
{
|
|
return DEFAULT_ATTACHMENT_MIME_TYPE;
|
|
}
|
|
}
|
|
|
|
|
|
private static String fixupCharset(String charset)
|
|
{
|
|
charset = charset.toLowerCase();
|
|
if (charset.equals("cp932"))
|
|
return "shift-jis";
|
|
else if (charset.equals("koi8-u"))
|
|
return "koi8-r";
|
|
|
|
return charset;
|
|
|
|
}
|
|
|
|
|
|
}
|