X-Git-Url: https://vcs.fsf.org/?p=squirrelmail.git;a=blobdiff_plain;f=functions%2Fmime.php;h=aacc7268f47f126ad83f2a0159a4aced13cb419c;hp=13a453a9ef89983746d790814344a98011d778c9;hb=ae3fac8dab1e33dc841922b522a9027bd45f6c32;hpb=f9b3e5d9902a047bfa8995c1d795cae2226a857a diff --git a/functions/mime.php b/functions/mime.php index 13a453a9..aacc7268 100644 --- a/functions/mime.php +++ b/functions/mime.php @@ -4,73 +4,70 @@ ** This contains the functions necessary to detect and decode MIME ** messages. ** + ** $Id$ **/ - $debug_mime = false; - $mime_php = true; - - if (!isset($i18n_php)) - include "../functions/i18n.php"; - if (!isset($imap_php)) - include "../functions/imap.php"; - if (!isset($config_php)) - include "../config/config.php"; + if (defined('mime_php')) + return; + define('mime_php', true); + require_once('../functions/imap.php'); /** Setting up the objects that have the structure for the message **/ class msg_header { /** msg_header contains generic variables for values that **/ /** could be in a header. **/ - - var $type0, $type1, $boundary, $charset, $encoding, $size; - var $to, $from, $date, $cc, $bcc, $reply_to, $subject; - var $id, $mailbox, $description; - var $entity_id, $message_id, $charset; + + var $type0 = '', $type1 = '', $boundary = '', $charset = ''; + var $encoding = '', $size = 0, $to = array(), $from = '', $date = ''; + var $cc = array(), $bcc = array(), $reply_to = '', $subject = ''; + var $id = 0, $mailbox = '', $description = '', $filename = ''; + var $entity_id = 0, $message_id = 0, $name = ''; } - + class message { /** message is the object that contains messages. It is a recursive - object in that through the $entities variable, it can contain + object in that through the $entities variable, it can contain more objects of type message. See documentation in mime.txt for a better description of how this works. - **/ - var $header; - var $entities; - + **/ + var $header = ''; + var $entities = array(); + function addEntity ($msg) { - $this->entities[count($this->entities)] = $msg; + $this->entities[] = $msg; } } - - /* --------------------------------------------------------------------------------- */ /* MIME DECODING */ /* --------------------------------------------------------------------------------- */ - + // This function gets the structure of a message and stores it in the "message" class. // It will return this object for use with all relevant header information and // fully parsed into the standard "message" object format. function mime_structure ($imap_stream, $header) { - global $debug_mime; - sqimap_messages_flag ($imap_stream, $header->id, $header->id, "Seen"); - + + sqimap_messages_flag ($imap_stream, $header->id, $header->id, 'Seen'); + $ssid = sqimap_session_id(); + $lsid = strlen( $ssid ); $id = $header->id; - fputs ($imap_stream, "a001 FETCH $id BODYSTRUCTURE\r\n"); + fputs ($imap_stream, "$ssid FETCH $id BODYSTRUCTURE\r\n"); + // + // This should use sqimap_read_data instead of reading it itself + // $read = fgets ($imap_stream, 10000); - $response = substr($read, 0, 4); - while ($response != "a001") { + $bodystructure = ''; + while( substr($read, 0, $lsid) <> $ssid && + !feof( $imap_stream ) ) { $bodystructure .= $read; $read = fgets ($imap_stream, 10000); - $response = substr($read, 0, 4); } - // $read = strtolower($bodystructure); $read = $bodystructure; - if ($debug_mime) echo "$read

\n"; // isolate the body structure and remove beginning and end parenthesis - $read = trim(substr ($read, strpos(strtolower($read), "bodystructure") + 13)); + $read = trim(substr ($read, strpos(strtolower($read), 'bodystructure') + 13)); $read = trim(substr ($read, 0, -1)); $end = mime_match_parenthesis(0, $read); while ($end == strlen($read)-1) { @@ -79,8 +76,6 @@ $end = mime_match_parenthesis(0, $read); } - if ($debug_mime) echo "$read

\n"; - $msg = mime_parse_structure ($read, 0); $msg->header = $header; return $msg; @@ -94,38 +89,32 @@ // then it parses it and adds the necessary header information (by calling out // to mime_get_elements() function mime_parse_structure ($structure, $ent_id) { - global $debug_mime; - if ($debug_mime) echo "START: mime_parse_structure()
\n"; + $msg = new message(); - if (substr($structure, 0, 1) == "(") { + if ($structure{0} == '(') { $ent_id = mime_new_element_level($ent_id); $start = $end = -1; - if ($debug_mime) echo "
$structure
"; do { - if ($debug_mime) echo "Found entity...
"; $start = $end+1; $end = mime_match_parenthesis ($start, $structure); - + $element = substr($structure, $start+1, ($end - $start)-1); $ent_id = mime_increment_id ($ent_id); $newmsg = mime_parse_structure ($element, $ent_id); $msg->addEntity ($newmsg); - } while (substr($structure, $end+1, 1) == "("); + } while ($structure{$end+1} == '('); } else { // parse the elements - if ($debug_mime) echo "
$structure
"; - $msg = mime_get_element (&$structure, $msg, $ent_id); - if ($debug_mime) echo "
"; + $msg = mime_get_element ($structure, $msg, $ent_id); } return $msg; - if ($debug_mime) echo "  END: mime_parse_structure()
"; } // Increments the element ID. An element id can look like any of // the following: 1, 1.2, 4.3.2.4.1, etc. This function increments // the last number of the element id, changing 1.2 to 1.3. function mime_increment_id ($id) { - global $debug_mime; + if (strpos($id, ".")) { $first = substr($id, 0, strrpos($id, ".")); $last = substr($id, strrpos($id, ".")+1); @@ -134,47 +123,51 @@ } else { $new = $id + 1; } - if ($debug_mime) echo "INCREMENT: $new
"; + return $new; } // See comment for mime_increment_id(). // This adds another level on to the entity_id changing 1.3 to 1.3.0 - // NOTE: 1.3.0 is not a valid element ID. It MUST be incremented + // NOTE: 1.3.0 is not a valid element ID. It MUST be incremented // before it can be used. I left it this way so as not to have // to make a special case if it is the first entity_id. It // always increments it, and that works fine. function mime_new_element_level ($id) { - if (!$id) $id = 0; - else $id = $id . ".0"; - return $id; + if (!$id) { + $id = 0; + } else { + $id = $id . '.0'; + } + + return( $id ); } function mime_get_element (&$structure, $msg, $ent_id) { - global $debug_mime; + $elem_num = 1; $msg->header = new msg_header(); $msg->header->entity_id = $ent_id; - + $properties = array(); + while (strlen($structure) > 0) { $structure = trim($structure); - $char = substr($structure, 0, 1); + $char = $structure{0}; - if (strtolower(substr($structure, 0, 3)) == "nil") { - $text = ""; + if (strtolower(substr($structure, 0, 3)) == 'nil') { + $text = ''; $structure = substr($structure, 3); - } else if ($char == "\"") { + } else if ($char == '"') { // loop through until we find the matching quote, and return that as a string $pos = 1; - $char = substr($structure, $pos, 1); - while ($char != "\"" && $pos < strlen($structure)) { + $text = ''; + while ( ($char = $structure{$pos} ) <> '"' && $pos < strlen($structure)) { $text .= $char; $pos++; - $char = substr($structure, $pos, 1); - } + } $structure = substr($structure, strlen($text) + 2); - } else if ($char == "(") { + } else if ($char == '(') { // comment me $end = mime_match_parenthesis (0, $structure); $sub = substr($structure, 1, $end-1); @@ -183,61 +176,59 @@ } else { // loop through until we find a space or an end parenthesis $pos = 0; - $char = substr($structure, $pos, 1); - while ($char != " " && $char != ")" && $pos < strlen($structure)) { + $char = $structure{$pos}; + $text = ''; + while ($char != ' ' && $char != ')' && $pos < strlen($structure)) { $text .= $char; $pos++; - $char = substr($structure, $pos, 1); + $char = $structure{$pos}; } $structure = substr($structure, strlen($text)); } - if ($debug_mime) echo "$elem_num : $text
"; // This is where all the text parts get put into the header switch ($elem_num) { - case 1: + case 1: $msg->header->type0 = strtolower($text); - if ($debug_mime) echo "type0 = ".strtolower($text)."
"; break; - case 2: + case 2: $msg->header->type1 = strtolower($text); - if ($debug_mime) echo "type1 = ".strtolower($text)."
"; break; + case 4: // Id + // Invisimail enclose images with <> + $msg->header->id = str_replace( '<', '', str_replace( '>', '', $text ) ); + break; case 5: $msg->header->description = $text; - if ($debug_mime) echo "description = $text
"; break; case 6: $msg->header->encoding = strtolower($text); - if ($debug_mime) echo "encoding = ".strtolower($text)."
"; break; case 7: $msg->header->size = $text; - if ($debug_mime) echo "size = $text
"; break; default: - if ($msg->header->type0 == "text" && $elem_num == 8) { + if ($msg->header->type0 == 'text' && $elem_num == 8) { // This is a plain text message, so lets get the number of lines // that it contains. $msg->header->num_lines = $text; - if ($debug_mime) echo "num_lines = $text
"; - } else if ($msg->header->type0 == "message" && $msg->header->type1 == "rfc822" && $elem_num == 8) { - // This is an encapsulated message, so lets start all over again and + } else if ($msg->header->type0 == 'message' && $msg->header->type1 == 'rfc822' && $elem_num == 8) { + // This is an encapsulated message, so lets start all over again and // parse this message adding it on to the existing one. $structure = trim($structure); - if (substr($structure, 0, 1) == "(") { + if ( $structure{0} == '(' ) { $e = mime_match_parenthesis (0, $structure); $structure = substr($structure, 0, $e); $structure = substr($structure, 1); $m = mime_parse_structure($structure, $msg->header->entity_id); - + // the following conditional is there to correct a bug that wasn't // incrementing the entity IDs correctly because of the special case // that message/rfc822 is. This fixes it fine. - if (substr($structure, 1, 1) != "(") + if (substr($structure, 1, 1) != '(') $m->header->entity_id = mime_increment_id(mime_new_element_level($ent_id)); - + // Now we'll go through and reformat the results. if ($m->entities) { for ($i=0; $i < count($m->entities); $i++) { @@ -246,7 +237,7 @@ } else { $msg->addEntity($m); } - $structure = ""; + $structure = ""; } } break; @@ -255,10 +246,9 @@ $text = ""; } // loop through the additional properties and put those in the various headers - if ($msg->header->type0 != "message") { + if ($msg->header->type0 != 'message') { for ($i=0; $i < count($properties); $i++) { - $msg->header->{$properties[$i]["name"]} = $properties[$i]["value"]; - if ($debug_mime) echo "".$properties[$i]["name"]." = " . $properties[$i]["value"] . "
"; + $msg->header->{$properties[$i]['name']} = $properties[$i]['value']; } } @@ -275,48 +265,50 @@ // This gets properties in a nested parenthesisized list. For example, // this would get passed something like: ("attachment" ("filename" "luke.tar.gz")) // This returns an array called $props with all paired up properties. - // It ignores the "attachment" for now, maybe that should change later + // It ignores the "attachment" for now, maybe that should change later // down the road. In this case, what is returned is: // $props[0]["name"] = "filename"; // $props[0]["value"] = "luke.tar.gz"; function mime_get_props ($props, $structure) { - global $debug_mime; + while (strlen($structure) > 0) { $structure = trim($structure); - $char = substr($structure, 0, 1); + $char = $structure{0}; - if ($char == "\"") { + if ($char == '"') { $pos = 1; - $char = substr($structure, $pos, 1); - while ($char != "\"" && $pos < strlen($structure)) { + $tmp = ''; + while ( ( $char = $structure{$pos} ) != '"' && + $pos < strlen($structure)) { $tmp .= $char; $pos++; - $char = substr($structure, $pos, 1); - } + } $structure = trim(substr($structure, strlen($tmp) + 2)); - $char = substr($structure, 0, 1); + $char = $structure{0}; - if ($char == "\"") { + if ($char == '"') { $pos = 1; - $char = substr($structure, $pos, 1); - while ($char != "\"" && $pos < strlen($structure)) { + $value = ''; + while ( ( $char = $structure{$pos} ) != '"' && + $pos < strlen($structure) ) { $value .= $char; $pos++; - $char = substr($structure, $pos, 1); - } + } $structure = trim(substr($structure, strlen($tmp) + 2)); - + $k = count($props); - $props[$k]["name"] = strtolower($tmp); - $props[$k]["value"] = $value; - } else if ($char == "(") { + $props[$k]['name'] = strtolower($tmp); + $props[$k]['value'] = $value; + } else if ($char == '(') { $end = mime_match_parenthesis (0, $structure); $sub = substr($structure, 1, $end-1); + if (! isset($props)) + $props = array(); $props = mime_get_props($props, $sub); $structure = substr($structure, strlen($sub) + 2); } return $props; - } else if ($char == "(") { + } else if ($char == '(') { $end = mime_match_parenthesis (0, $structure); $sub = substr($structure, 1, $end-1); $props = mime_get_props($props, $sub); @@ -334,42 +326,129 @@ // x x // then this would return 42 to match up those two. function mime_match_parenthesis ($pos, $structure) { - $char = substr($structure, $pos, 1); + + $j = strlen( $structure ); // ignore all extra characters // If inside of a string, skip string -- Boundary IDs and other // things can have ) in them. - while ($pos < strlen($structure)) { + if( $structure{$pos} != '(' ) + return( $j ); + + while( $pos < $j ) { $pos++; - $char = substr($structure, $pos, 1); - if ($char == ")") { + if ($structure{$pos} == ')') { return $pos; - } else if ($char == '"') { - $pos ++; - while (substr($structure, $pos, 1) != '"' && - $pos < strlen($structure)) { - $pos ++; + } elseif ($structure{$pos} == '"') { + $pos++; + while( $structure{$pos} != '"' && + $pos < $j ) { + if (substr($structure, $pos, 2) == '\\"') + $pos++; + elseif (substr($structure, $pos, 2) == '\\\\') + $pos++; + $pos++; } - } else if ($char == "(") { + } elseif ( $structure{$pos} == '(' ) { $pos = mime_match_parenthesis ($pos, $structure); } } + echo "Error decoding mime structure. Report this as a bug!
\n"; + return( $pos ); } - function mime_fetch_body ($imap_stream, $id, $ent_id) { + function mime_fetch_body ($imap_stream, $id, $ent_id ) { + // do a bit of error correction. If we couldn't find the entity id, just guess + // that it is the first one. That is usually the case anyway. + if (!$ent_id) + $ent_id = 1; + $sid = sqimap_session_id(); + fputs ($imap_stream, "$sid FETCH $id BODY[$ent_id]\r\n"); + $data = sqimap_read_data ($imap_stream, $sid, true, $response, $message); + $topline = array_shift($data); + while (! ereg('\\* [0-9]+ FETCH ', $topline) && $data) + $topline = array_shift($data); + $wholemessage = implode('', $data); + if (ereg('\\{([^\\}]*)\\}', $topline, $regs)) { + $ret = substr( $wholemessage, 0, $regs[1] ); + /* + There is some information in the content info header that could be important + in order to parse html messages. Let's get them here. + */ + if( $ret{0} == '<' ) { + fputs ($imap_stream, "$sid FETCH $id BODY[$ent_id.MIME]\r\n"); + $data = sqimap_read_data ($imap_stream, $sid, true, $response, $message); + $base = ''; + $k = 10; + foreach( $data as $d ) { + if( substr( $d, 0, 13 ) == 'Content-Base:' ) { + $j = strlen( $d ); + $i = 13; + $base = ''; + while( $i < $j && + ( !isNoSep( $d{$i} ) || $d{$i} == '"' ) ) + $i++; + while( $i < $j ) { + if( isNoSep( $d{$i} ) ) + $base .= $d{$i}; + $i++; + } + $k = 0; + } elseif( $k == 1 && !isnosep( $d{0} ) ) { + $base .= substr( $d, 1 ); + } + $k++; + } + if( $base <> '' ) + $ret = "" . $ret; + } + } else if (ereg('"([^"]*)"', $topline, $regs)) { + $ret = $regs[1]; + } else { + $ret = "Body retrieval error. Please report this bug!\n" . + "Response: $response\n" . + "Message: $message\n" . + "FETCH line: $topline" . + "---------------\n$wholemessage"; + + foreach ($data as $d) { + $ret .= htmlspecialchars($d) . "\n"; + } + } + return( $ret ); + } + + function mime_print_body_lines ($imap_stream, $id, $ent_id, $encoding) { // do a bit of error correction. If we couldn't find the entity id, just guess // that it is the first one. That is usually the case anyway. if (!$ent_id) $ent_id = 1; - - fputs ($imap_stream, "a001 FETCH $id BODY[$ent_id]\r\n"); - $topline = fgets ($imap_stream, 1024); - if (ereg('\{([^\}]*)\}', $topline, $regs)) { - return fread ($imap_stream, $regs[1]); - } - else if (ereg('"([^"]*)"', $topline, $regs)) { - return $regs[1]; + $sid = sqimap_session_id(); + // Don't kill the connection if the browser is over a dialup + // and it would take over 30 seconds to download it. + set_time_limit(0); + + fputs ($imap_stream, "$sid FETCH $id BODY[$ent_id]\r\n"); + $cnt = 0; + $continue = true; + $read = fgets ($imap_stream,4096); + // This could be bad -- if the section has sqimap_session_id() . ' OK' + // or similar, it will kill the download. + while (!ereg("^".$sid." (OK|BAD|NO)(.*)$", $read, $regs)) { + if (trim($read) == ')==') { + $read1 = $read; + $read = fgets ($imap_stream,4096); + if (ereg("^".$sid." (OK|BAD|NO)(.*)$", $read, $regs)) { + return; + } else { + echo decodeBody($read1, $encoding) . + decodeBody($read, $encoding); + } + } else if ($cnt) { + echo decodeBody($read, $encoding); + } + $read = fgets ($imap_stream,4096); + $cnt++; } - return "Body retrival error, please report this bug!\n\nTop line is \"$topline\"\n"; } /* -[ END MIME DECODING ]----------------------------------------------------------- */ @@ -379,17 +458,18 @@ /** This is the first function called. It decides if this is a multipart message or if it should be handled as a single entity **/ - function decodeMime ($imap_stream, $body, $header) { + function decodeMime ($imap_stream, &$header) { global $username, $key, $imapServerAddress, $imapPort; return mime_structure ($imap_stream, $header); } // This is here for debugging purposese. It will print out a list // of all the entity IDs that are in the $message object. + /* function listEntities ($message) { if ($message) { if ($message->header->entity_id) - echo "" . $message->header->entity_id . " : " . $message->header->type0 . "/" . $message->header->type1 . "
"; + echo "" . $message->header->entity_id . ' : ' . $message->header->type0 . '/' . $message->header->type1 . '
'; for ($i = 0; $message->entities[$i]; $i++) { $msg = listEntities($message->entities[$i], $ent_id); if ($msg) @@ -397,6 +477,7 @@ } } } + */ // returns a $message object for a particular entity id function getEntity ($message, $ent_id) { @@ -404,30 +485,61 @@ if ($message->header->entity_id == $ent_id && strlen($ent_id) == strlen($message->header->entity_id)) { return $message; } else { - for ($i = 0; $message->entities[$i]; $i++) { + for ($i = 0; isset($message->entities[$i]); $i++) { $msg = getEntity ($message->entities[$i], $ent_id); if ($msg) return $msg; } - } + } } } - // figures out what entity to display and returns the $message object - // for that entity. - function findDisplayEntity ($message) { - if ($message) { - if ($message->header->type0 == "text") { - if ($message->header->type1 == "plain" || - $message->header->type1 == "html") { - return $message->header->entity_id; + // figures out what entity to display and returns the $message object + // for that entity. + function findDisplayEntity ($message, $textOnly = 1) { + global $show_html_default; + + $entity = 0; + + if ($message) { + if ( $message->header->type0 == 'multipart' && + ( $message->header->type1 == 'alternative' || + $message->header->type1 == 'related' ) && + $show_html_default && ! $textOnly ) { + $entity = findDisplayEntityHTML($message); } - } else { - for ($i=0; $message->entities[$i]; $i++) { - return findDisplayEntity($message->entities[$i]); - } - } + + // Show text/plain or text/html -- the first one we find. + if ( $entity == 0 && + $message->header->type0 == 'text' && + ( $message->header->type1 == 'plain' || + $message->header->type1 == 'html' ) && + isset($message->header->entity_id) ) { + $entity = $message->header->entity_id; + } + + $i = 0; + while ($entity == 0 && isset($message->entities[$i]) ) { + $entity = findDisplayEntity($message->entities[$i], $textOnly); + $i++; + } + } + + return( $entity ); + } + + // Shows the HTML version + function findDisplayEntityHTML ($message) { + if ($message->header->type0 == 'text' && + $message->header->type1 == 'html' && + isset($message->header->entity_id)) + return $message->header->entity_id; + for ($i = 0; isset($message->entities[$i]); $i ++) { + $entity = findDisplayEntityHTML($message->entities[$i]); + if ($entity != 0) + return $entity; } + return 0; } /** This returns a parsed string called $body. That string can then @@ -439,37 +551,49 @@ // this if statement checks for the entity to show as the // primary message. To add more of them, just put them in the // order that is their priority. - global $startMessage, $username, $key, $imapServerAddress, $imapPort; + global $startMessage, $username, $key, $imapServerAddress, $imapPort, + $show_html_default; $id = $message->header->id; $urlmailbox = urlencode($message->header->mailbox); // Get the right entity and redefine message to be this entity - $ent_num = findDisplayEntity ($message); + // Pass the 0 to mean that we want the 'best' viewable one + $ent_num = findDisplayEntity ($message, 0); $body_message = getEntity($message, $ent_num); - if (($body_message->header->type0 == "text") || - ($body_message->header->type0 == "rfc822")) { - - $body = mime_fetch_body ($imap_stream, $id, $ent_num); + if (($body_message->header->type0 == 'text') || + ($body_message->header->type0 == 'rfc822')) { + + $body = mime_fetch_body ($imap_stream, $id, $ent_num); $body = decodeBody($body, $body_message->header->encoding); - + $hookResults = do_hook("message_body", $body); + $body = $hookResults[1]; + // If there are other types that shouldn't be formatted, add - // them here - if ($body_message->header->type1 != "html") { + // them here + if ($body_message->header->type1 == 'html') { + if( $show_html_default <> 1 ) { + $body = strip_tags( $body ); + translateText($body, $wrap_at, $body_message->header->charset); + } else { + $body = MagicHTML( $body, $id ); + } + } else { translateText($body, $wrap_at, $body_message->header->charset); - } - - $body .= "
". _("Download this as a file") ."

"; - + } + + $body .= "
". _("Download this as a file") ."

"; + /** Display the ATTACHMENTS: message if there's more than one part **/ $body .= ""; - if ($message->entities) { + if (isset($message->entities[0])) { $body .= formatAttachments ($message, $ent_num, $message->header->mailbox, $id); } + $body .= ""; } else { - $body .= formatAttachments ($message, -1, $message->header->mailbox, $id); + $body = formatAttachments ($message, -1, $message->header->mailbox, $id); } - return $body; + return( $body ); } // A recursive function that returns a list of attachments with links @@ -477,94 +601,96 @@ function formatAttachments ($message, $ent_id, $mailbox, $id) { global $where, $what; global $startMessage, $color; - static $ShownHTML; - - if ($ShownHTML == 0) - { + static $ShownHTML = 0; + + $body = ""; + if ($ShownHTML == 0) { $ShownHTML = 1; - - $body .= "\n"; - $body .= "
\n"; - $body .= _('Attachments') . ':'; - $body .= "
\n"; - - $body .= "\n"; - - $body .= formatAttachments ($message, $ent_id, $mailbox, $id); - - $body .= "
"; - - return $body; + + $body .= "\n" . + "
\n" . + _("Attachments") . ':' . + "
\n" . + "\n" . + formatAttachments ($message, $ent_id, $mailbox, $id) . + "
"; + + return( $body ); } - + if ($message) { if (!$message->entities) { $type0 = strtolower($message->header->type0); $type1 = strtolower($message->header->type1); - + $name = decodeHeader($message->header->name); + if ($message->header->entity_id != $ent_id) { $filename = decodeHeader($message->header->filename); - if (trim($filename) == "") { - $display_filename = "untitled-".$message->header->entity_id; + if (trim($filename) == '') { + if (trim($name) == '') { + if( trim( $message->header->id ) == '' ) + $display_filename = 'untitled-[' . $message->header->entity_id . ']' ; + else + $display_filename = 'cid: ' . $message->header->id; + // $display_filename = 'untitled-[' . $message->header->entity_id . ']' ; + } else { + $display_filename = $name; + $filename = $name; + } } else { $display_filename = $filename; } - + $urlMailbox = urlencode($mailbox); $ent = urlencode($message->header->entity_id); - - $DefaultLink = + + $DefaultLink = "../src/download.php?startMessage=$startMessage&passed_id=$id&mailbox=$urlMailbox&passed_ent_id=$ent"; if ($where && $what) $DefaultLink .= '&where=' . urlencode($where) . '&what=' . urlencode($what); - $Links['download link']['text'] = _('download'); - $Links['download link']['href'] = + $Links['download link']['text'] = _("download"); + $Links['download link']['href'] = "../src/download.php?absolute_dl=true&passed_id=$id&mailbox=$urlMailbox&passed_ent_id=$ent"; $ImageURL = ''; - + $HookResults = do_hook("attachment $type0/$type1", $Links, - $startMessage, $id, $urlMailbox, $ent, $DefaultLink, + $startMessage, $id, $urlMailbox, $ent, $DefaultLink, $display_filename, $where, $what); $Links = $HookResults[1]; $DefaultLink = $HookResults[6]; - $body .= '  '; - $body .= "$display_filename "; - $body .= '' . show_readable_size($message->header->size) . - '  '; - $body .= "[ $type0/$type1 ] "; - $body .= ''; + $body .= '  ' . + "$display_filename " . + '' . show_readable_size($message->header->size) . + '  ' . + "[ $type0/$type1 ] " . + ''; if ($message->header->description) $body .= '' . htmlspecialchars($message->header->description) . ''; $body .= ' '; - - + + $SkipSpaces = 1; - foreach ($Links as $Val) - { - if ($SkipSpaces) - { + foreach ($Links as $Val) { + if ($SkipSpaces) { $SkipSpaces = 0; - } - else - { + } else { $body .= '  |  '; } $body .= '' . $Val['text'] . ''; } - + unset($Links); - + $body .= "\n"; } - return $body; } else { for ($i = 0; $i < count($message->entities); $i++) { $body .= formatAttachments ($message->entities[$i], $ent_id, $mailbox, $id); } - return $body; } + return( $body ); } } @@ -574,12 +700,20 @@ $body = str_replace("\r\n", "\n", $body); $encoding = strtolower($encoding); - if ($encoding == "quoted-printable") { - $body = quoted_printable_decode($body); + global $show_html_default; + if ($encoding == 'quoted-printable') { + $body = quoted_printable_decode($body); + + + /* + Following code has been comented as I see no reason for it. + If there is any please tell me a mingo@rotedic.com + while (ereg("=\n", $body)) $body = ereg_replace ("=\n", "", $body); - } else if ($encoding == "base64") { + */ + } else if ($encoding == 'base64') { $body = base64_decode($body); } @@ -588,31 +722,32 @@ } - // This functions decode strings that is encoded according to + // This functions decode strings that is encoded according to // RFC1522 (MIME Part Two: Message Header Extensions for Non-ASCII Text). function decodeHeader ($string) { - if (eregi('=\?([^?]+)\?(q|b)\?([^?]+)\?=', + if (eregi('=\\?([^?]+)\\?(q|b)\\?([^?]+)\\?=', $string, $res)) { if (ucfirst($res[2]) == "B") { $replace = base64_decode($res[3]); } else { $replace = ereg_replace("_", " ", $res[3]); - // Convert lowercase Quoted Printable to uppercase for - // quoted_printable_decode to understand it. - while (ereg("(=([0-9][a-f])|([a-f][0-9])|([a-f][0-9]))", $replace, $res)) { - $replace = str_replace($res[1], strtoupper($res[1]), $replace); - } + // Convert lowercase Quoted Printable to uppercase for + // quoted_printable_decode to understand it. + while (ereg("(=(([0-9][abcdef])|([abcdef][0-9])|([abcdef][abcdef])))", $replace, $res)) { + $replace = str_replace($res[1], strtoupper($res[1]), $replace); + } $replace = quoted_printable_decode($replace); } $replace = charset_decode ($res[1], $replace); - $string = eregi_replace - ('=\?([^?]+)\?(q|b)\?([^?]+)\?=', + // Remove the name of the character set. + $string = eregi_replace ('=\\?([^?]+)\\?(q|b)\\?([^?]+)\\?=', $replace, $string); + // In case there should be more encoding in the string: recurse return (decodeHeader($string)); - } else + } else return ($string); } @@ -622,29 +757,360 @@ function encodeHeader ($string) { global $default_charset; - // Encode only if the string contains 8-bit characters or =? - if (ereg("([\200-\377]|=\\?)", $string)) { - $newstring = "=?$default_charset?Q?"; - - // First the special characters - $string = str_replace("=", "=3D", $string); - $string = str_replace("?", "=3F", $string); - $string = str_replace("_", "=5F", $string); - $string = str_replace(" ", "_", $string); + // Encode only if the string contains 8-bit characters or =? + $j = strlen( $string ); + $l = FALSE; // Must be encoded ? + $ret = ''; + for( $i=0; $i < $j; ++$i) { + switch( $string{$i} ) { + case '=': + $ret .= '=3D'; + break; + case '?': + $l = TRUE; + $ret .= '=3F'; + break; + case '_': + $ret .= '=5F'; + break; + case ' ': + $ret .= '_'; + break; + default: + $k = ord( $string{$i} ); + if( $k > 126 ) { + $ret .= sprintf("=%02X", $k); + $l = TRUE; + } else + $ret .= $string{$i}; + } + } + if( $l ) + $string = "=?$default_charset?Q?$ret?="; - while (ereg("([\200-\377])", $string, $regs)) { - $replace = $regs[1]; - $insert = "=" . strtoupper(bin2hex($replace)); - $string = str_replace($replace, $insert, $string); - } + return( $string ); + } - $newstring = "=?$default_charset?Q?".$string."?="; - - return $newstring; - } + /* + Strips dangerous tags from html messages. + */ + + function MagicHTML( $body, $id ) { + + global $message, $PHP_SELF, $HTTP_SERVER_VARS; + + $j = strlen( $body ); // Legnth of the HTML + $ret = ''; // Returned string + $bgcolor = '#ffffff'; // Background style color (defaults to white) + $leftmargin = ''; // Left margin style + $title = ''; // HTML title if any + + $i = 0; + while( $i < $j ) { + if( $body{$i} == '<' ) { + $tag = $body{$i+1}.$body{$i+2}.$body{$i+3}.$body{$i+4}; + switch( strtoupper( $tag ) ) { + // Strips the entire tag and contents + case 'APPL': + case 'EMBB': + case 'FRAM': + case 'SCRI': + case 'OBJE': + $etg = '/' . $tag; + while( $body{$i+1}.$body{$i+2}.$body{$i+3}.$body{$i+4}.$body{$i+5} <> $etg && + $i < $j ) $i++; + while( $i < $j && $body{++$i} <> '>' ); + // $ret .= ""; + break; + // Substitute Title + case 'TITL': + $i += 5; + while( $body{$i} <> '>' && // + $i < $j ) + $i++; + $i++; + $title = ''; + while( $body{$i} <> '<' && // + $i < $j ) { + $title .= $body{$i}; + $i++; + } + $i += 7; + break; + // Destroy these tags + case 'HTML': + case 'HEAD': + case '/HTM': + case '/HEA': + case '!DOC': + case 'META': + case 'DIV ': + case '/DIV': + case '!-- ': + $i += 4; + while( $body{$i} <> '>' && + $i < $j ) + $i++; + // $i++; + break; + case 'STYL': + $i += 5; + while( $body{$i} <> '>' && // + $i < $j ) + $i++; + $i++; + // We parse the style to look for interesting stuff + $styleblk = ''; + while( $body{$i} <> '>' && + $i < $j ) { + // First we get the name of the style + $style = ''; + while( $body{$i} <> '>' && + $body{$i} <> '<' && + $body{$i} <> '{' && + $i < $j ) { + if( isnoSep( $body{$i} ) ) + $style .= $body{$i}; + $i++; + } + stripComments( &$i, $j, &$body ); + $style = strtoupper( trim( $style ) ); + if( $style == 'BODY' ) { + // Next we look into the definitions of the body style + while( $body{$i} <> '>' && + $body{$i} <> '}' && + $i < $j ) { + // We look for the background color if any. + if( substr( $body, $i, 17 ) == 'BACKGROUND-COLOR:' ) { + $i += 17; + $bgcolor = getStyleData( $i, $j, $body ); + } elseif ( substr( $body, $i, 12 ) == 'MARGIN-LEFT:' ) { + $i += 12; + $leftmargin = getStyleData( $i, $j, $body ); + } + $i++; + } + } else { + // Other style are mantained + $styleblk .= "$style "; + while( $body{$i} <> '>' && + $body{$i} <> '<' && + $body{$i} <> '}' && + $i < $j ) { + $styleblk .= $body{$i}; + $i++; + } + $styleblk .= $body{$i}; + } + stripComments( &$i, $j, &$body ); + if( $body{$i} <> '>' ) + $i++; + } + if( $styleblk <> '' ) + $ret .= "