/** @file Parser: String class part: untaint mechanizm. Copyright (c) 2001-2026 Art. Lebedev Studio (https://www.artlebedev.com) Authors: Konstantin Morshnev , Alexandr Petrosian */ volatile const char * IDENT_UNTAINT_C="$Id: untaint.C,v 1.183 2026/10/13 20:02:01 moko Exp $"; #include "pa_string.h" #include "pa_hash.h" #include "pa_exception.h" #include "pa_table.h" #include "pa_globals.h" #include "pa_dictionary.h" #include "pa_common.h" #include "pa_charset.h" #include "pa_request_charsets.h" #include "pa_sapi.h" extern "C" { // author forgot to do that #include "../lib/cord/include/ec.h" } #include "pa_sql_connection.h" // defines #undef CORD_ec_append // redefining to intercept flushes and implement whitespace optimization // of all consequent white space chars leaving only first one #define CORD_ec_append(x, c) \ { \ bool skip=false; \ if(optimize) switch(c) { \ case ' ': case '\n': case '\t': \ if(whitespace) \ skip=true; /*skipping subsequent*/ \ else \ whitespace=true; \ break; \ default: \ whitespace=false; \ break; \ } \ if(!skip) { \ if ((x)[0].ec_bufptr == (x)[0].ec_buf + CORD_BUFSZ) { \ CORD_ec_flush_buf(x); \ } \ *((x)[0].ec_bufptr)++ = (c); \ } \ } #define escape_fragment(action) \ for(; fragment_length--; CORD_next(info->pos)) { \ char c=CORD_pos_fetch(info->pos); \ action \ } #define encode(need_encode_func, prefix, otherwise) \ if(need_encode_func(c)) { \ CORD_ec_append(info->result, prefix); \ to_hex(c); \ } else \ CORD_ec_append(info->result, otherwise); #define to_hex(c) \ { \ CORD_ec_append(info->result, hex_digits[((unsigned char)c) >> 4]); \ CORD_ec_append(info->result, hex_digits[((unsigned char)c) & 0x0F]); \ } #define to_char(c) { CORD_ec_append(info->result, c); whitespace=false; } #define to_string(s) { CORD_ec_append_cord(info->result, (CORD)(s)); whitespace=false; } #define _default CORD_ec_append(info->result, c) /// \u2028 and \u2029 (line/paragraph separators) and the next ones in utf-8 as \u20XX: they end a js string, see bug #1023; /// other bytes after 0xE2 are not taken here, so the loop escapes them as usual #define escape_u20XX { \ size_t index=CORD_pos_to_index(info->pos); \ unsigned char c2; \ if(info->charsets && info->charsets->source().isUTF8() && fragment_length>=2 \ && (unsigned char)info->body->fetch(index+1)==0x80 \ && (c2=(unsigned char)info->body->fetch(index+2))>=0xA8 && c2<=0xAF){ \ to_string("\\u20"); \ to_hex(c2-0x80); \ CORD_next(info->pos); \ CORD_next(info->pos); \ fragment_length-=2; \ } else { \ _default; \ } \ } inline bool need_file_encode(unsigned char c){ // russian letters and space ENABLED // encoding only these... return strchr( "*?\"<>|" #ifndef WIN32 ":" // not allowed in windows file names; '\' is a path separator everywhere since #1300 #endif , c)!=0; } inline bool need_uri_encode(unsigned char c){ return !(pa_isalnum(c) || strchr("_-./*", c)); } inline bool need_regex_escape(unsigned char c){ return strchr("\\^$.[]|()?*+{}-", c)!=0; } inline bool need_parser_code_escape(unsigned char c){ return strchr("^$;@()[]{}:#\"", c)!=0; } // String /* HTTP-header = field-name ":" [ field-value ] CRLF field-name = token field-value = *( field-content | LWS ) field-content = token = 1* word = token | quoted-string quoted-string = ( <"> *(qdtext | quoted-pair ) <"> ) qdtext = > quoted-pair = "\" CHAR OCTET = CHAR = tspecials = "(" | ")" | "<" | ">" | "@" | "," | ";" | ":" | "\" | <"> | "/" | "[" | "]" | "?" | "=" | "{" | "}" | SP | HT SP = HT = LWS = [CRLF] 1*( SP | HT ) TEXT = CTL = if(strchr("()<>@,;:\\\"/[]?={} \t", *ptr)) */ inline bool need_quote_http_header(const char* ptr, size_t size) { for(; size--; ptr++) if(strchr(";\\\"= \t" /* excluded ()<>@, :/ ? []{} */, *ptr)) return true; return false; } #ifndef DOXYGEN struct Append_fragment_info { String::Language lang; String::Languages* dest_languages; size_t dest_body_plan_length; }; #endif int append_fragment_optimizing(char alang, size_t asize, Append_fragment_info* info) { const String::Language lang=(String::Language)(unsigned char)alang; // main idea here: // tainted piece would get OPTIMIZED bit from 'lang' // clean piece would be marked OPTIMIZED manually // pieces with determined languages [not tainted|clean] would retain theirs langs info->dest_languages->append(info->dest_body_plan_length, lang==String::L_TAINTED? info->lang :lang==String::L_CLEAN? (String::Language)(String::L_CLEAN|String::L_OPTIMIZE_BIT) // ORing with OPTIMIZED flag :lang, asize); info->dest_body_plan_length+=asize; return 0; // 0=continue } int append_fragment_nonoptimizing(char alang, size_t asize, Append_fragment_info* info) { const String::Language lang=(String::Language)(unsigned char)alang; // The core idea: tainted pieces got marked with context's lang info->dest_languages->append(info->dest_body_plan_length, lang==String::L_TAINTED? info->lang :lang, asize); info->dest_body_plan_length+=asize; return 0; // 0=continue } /** appends to other String without language change */ String& String::append_to(String& dest) const { if(is_empty()) return dest; // first: fragment infos dest.langs.appendHelper(dest.body, langs, body); // next: letters dest.body<. An 'encoded-word' that appears within a 'phrase' MUST be separated from any adjacent 'word', 'text' or 'special' by 'linear-white-space'. ... (2) The 8-bit hexadecimal value 20 (e.g., ISO-8859-1 SPACE) may be represented as "_" (underscore, ASCII 95.). (This character may not pass through some internetwork mail gateways, but its use will greatly enhance readability of "Q" encoded data with mail readers that do not support this encoding.) Note that the "_" always represents hexadecimal 20, even if the SPACE character occupies a different code position in the character set in use. paf: obviously, without "=", or one could not differ "=E0" and "russian letter a" and without "_", or in would mean 0x20 */ inline bool mail_header_char_valid_within_Qencoded(char c) { return (pa_isalnum((unsigned char)c) || strchr("!*+-/", c)); } inline bool addr_spec_soon(const char *src) { for(char c; (c=*src); src++) if(c=='<') return true; else if(!(c==' ' || c=='\t')) return false; return false; } /** RFC Upper case should be used for hexadecimal digits "A" through "F" The 8-bit hexadecimal value 20 (e.g., ISO-8859-1 SPACE) may be represented as "_" */ inline bool mail_header_nonspace_char(char c) { return c != 0x20; } inline void ec_append(CORD_ec& result, bool& optimize, bool& whitespace, CORD_pos pos, size_t size) { while(size--) { CORD_ec_append(result, CORD_pos_fetch(pos)); CORD_next(pos); } } inline void pa_CORD_pos_advance(CORD_pos pos, size_t n) { while(true) { long avail=CORD_pos_chars_left(pos); if(avail<=0) { CORD_next(pos); if(!--n) break; } else if((size_t)avail=n CORD_pos_advance(pos, n); break; } } } #ifndef DOXYGEN struct Cstr_to_string_body_block_info { // input String::Language lang; SQL_Connection* connection; const Request_charsets* charsets; const String::Body* body; // output CORD_ec result; // private CORD_pos pos; size_t fragment_begin; bool whitespace; const char* exception; }; #endif // @todo: replace info->body->mid with something that uses info->pos int cstr_to_string_body_block(String::Language to_lang, size_t fragment_length, Cstr_to_string_body_block_info* info) { bool& whitespace=info->whitespace; size_t fragment_end=info->fragment_begin+fragment_length; //fprintf(stderr, "%d, %d =%s=\n", to_lang, fragment_length, info->body->cstr()); bool optimize=(to_lang & String::L_OPTIMIZE_BIT)!=0; if(!optimize) whitespace=false; switch(to_lang & ~String::L_OPTIMIZE_BIT) { case String::L_CLEAN: case String::L_TAINTED: case String::L_AS_IS: // clean piece // tainted piece, but undefined untaint language // for VString.as_double of tainted values // for ^process{body} evaluation // tainted, untaint language: as-is ec_append(info->result, optimize, whitespace, info->pos, fragment_length); break; case String::L_FILE_SPEC: // tainted, untaint language: file [name] { escape_fragment( encode(need_file_encode, '_', c); ); } break; case String::L_URI: // tainted, untaint language: uri { const char *fragment_str=info->body->mid(info->fragment_begin, fragment_length).cstr(); // skip source [we use recoded version] pa_CORD_pos_advance(info->pos, fragment_length); String::C output(fragment_str, fragment_length); if(info->charsets) output=Charset::transcode(output, info->charsets->source(), info->charsets->client()); char c; for(const char* src=output.str; (c=*src++); ) encode(need_uri_encode, '%', c); } break; case String::L_HTTP_HEADER: // tainted, untaint language: http-field-content-text escape_fragment(switch(c) { case '\n': case '\r': to_string(" "); break; default: _default; break; }); break; case String::L_MAIL_HEADER: // tainted, untaint language: mail-header // http://www.ietf.org/rfc/rfc2047.txt if(info->charsets) { size_t mail_size; const char *mail_ptr= info->body->mid(info->fragment_begin, mail_size=fragment_length).cstr(); // skip source [we use recoded version] pa_CORD_pos_advance(info->pos, mail_size); const char* charset_name=info->charsets->mail().NAME().cstr(); // Subject: Re: parser3: =?koi8-r?Q?=D3=C5=CD=C9=CE=C1=D2?= bool to_quoted_printable=false; bool email=false; uchar c; for(const char* src=mail_ptr; c=(uchar)*src; src++) { if(c=='\r' || c=='\n') c=' '; if(to_quoted_printable && (c==',' || c == '"' || addr_spec_soon(src))) { email=c=='<'; to_string("?="); to_quoted_printable=false; } //RFC + An 'encoded-word' MUST NOT appear in any portion of an 'addr-spec'. if(!email && ( ( !to_quoted_printable && (c & 0x80 || (c == ' ' && src == mail_ptr) ) ) // starting quote-printable-encoding on first 8bit char or leading space (issue #123) || ( to_quoted_printable && !mail_header_char_valid_within_Qencoded(c) ) )) { if(!to_quoted_printable) { to_string("=?"); to_string(charset_name); to_string("?Q?"); to_quoted_printable=true; } encode(mail_header_nonspace_char, '=', '_'); } else to_char(c); if(c=='>') email=false; } if(to_quoted_printable) // close to_string("?="); } else ec_append(info->result, optimize, whitespace, info->pos, fragment_length); break; case String::L_SQL: // tainted, untaint language: sql if(info->connection) { const char *fragment_str=info->body->mid(info->fragment_begin, fragment_length).cstr(); // skip source [we use recoded version] pa_CORD_pos_advance(info->pos, fragment_length); to_string(info->connection->quote(fragment_str, fragment_length)); } else { info->exception="untaint in SQL language failed - no connection specified"; info->fragment_begin=fragment_end; return 1; // stop processing. can't throw exception here } break; case String::L_JS: // quotes, < > & as \xXX: safe both inside