rspamd/src/lua/lua_text.c

/*-
 * Copyright 2019 Vsevolod Stakhov
 *
 * Licensed under the Apache License, Version 2.0 (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.apache.org/licenses/LICENSE-2.0
 *
 * Unless required by applicable law or agreed to in writing, software
 * distributed under the License is distributed on an "AS IS" BASIS,
 * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 * See the License for the specific language governing permissions and
 * limitations under the License.
 */

#include "lua_common.h"
#include "libcryptobox/cryptobox.h"
#include "contrib/fastutf8/fastutf8.h"
#include "unix-std.h"

/***
 * @module rspamd_text
 * This module provides access to opaque text structures used widely to prevent
 * copying between Lua and C for various concerns: performance, security etc...
 *
 * You can convert rspamd_text into string but it will copy data.
 */

/***
 * @function rspamd_text.fromstring(str)
 * Creates rspamd_text from Lua string (copied to the text)
 * @param {string} str string to use
 * @return {rspamd_text} resulting text
 */
LUA_FUNCTION_DEF (text, fromstring);

/***
 * @function rspamd_text.null()
 * Creates rspamd_text with NULL pointer for testing purposes
 * @param {string} str string to use
 * @return {rspamd_text} resulting text
 */
LUA_FUNCTION_DEF (text, null);
/***
 * @function rspamd_text.randombytes(nbytes)
 * Creates rspamd_text with random bytes inside (raw bytes)
 * @param {number} nbytes number of random bytes generated
 * @return {rspamd_text} random bytes text
 */
LUA_FUNCTION_DEF (text, randombytes);

/***
 * @function rspamd_text.fromtable(tbl[, delim])
 * Same as `table.concat` but generates rspamd_text instead of the Lua string
 * @param {table} tbl table to use
 * @param {string} delim optional delimiter
 * @return {rspamd_text} resulting text
 */
LUA_FUNCTION_DEF (text, fromtable);
/***
 * @method rspamd_text:byte(pos[, pos2])
 * Returns a byte at the position `pos` or bytes from `pos` to `pos2` if specified
 * @param {integer} pos index
 * @param {integer} pos2 index
 * @return {integer} byte at the position `pos` or varargs of bytes
 */
LUA_FUNCTION_DEF (text, byte);
/***
 * @method rspamd_text:len()
 * Returns length of a string
 * @return {number} length of string in **bytes**
 */
LUA_FUNCTION_DEF (text, len);
/***
 * @method rspamd_text:str()
 * Converts text to string by copying its content
 * @return {string} copy of text as Lua string
 */
LUA_FUNCTION_DEF (text, str);
/***
 * @method rspamd_text:ptr()
 * Converts text to lightuserdata
 * @return {lightuserdata} pointer value of rspamd_text
 */
LUA_FUNCTION_DEF (text, ptr);
/***
 * @method rspamd_text:save_in_file(fname[, mode])
 * Saves text in file
 * @return {boolean} true if save has been completed
 */
LUA_FUNCTION_DEF (text, save_in_file);
/***
 * @method rspamd_text:span(start[, len])
 * Returns a span for lua_text starting at pos [start] (1 indexed) and with
 * length `len` (or to the end of the text)
 * @param {integer} start start index
 * @param {integer} len length of span
 * @return {rspamd_text} new rspamd_text with span (must be careful when using with owned texts...)
 */
LUA_FUNCTION_DEF (text, span);
/***
 * @method rspamd_text:sub(start[, len])
 * Returns a substrin for lua_text similar to string.sub from Lua
 * @return {rspamd_text} new rspamd_text with span (must be careful when using with owned texts...)
 */
LUA_FUNCTION_DEF (text, sub);
/***
 * @method rspamd_text:lines([stringify])
 * Returns an iter over all lines as rspamd_text objects or as strings if `stringify` is true
 * @param {boolean} stringify stringify lines
 * @return {iterator} iterator triplet
 */
LUA_FUNCTION_DEF (text, lines);
/***
 * @method rspamd_text:split(regexp, [stringify])
 * Returns an iter over all encounters of the specific regexp as rspamd_text objects or as strings if `stringify` is true
 * @param {rspamd_regexp} regexp regexp (pcre syntax) used for splitting
 * @param {boolean} stringify stringify lines
 * @return {iterator} iterator triplet
 */
LUA_FUNCTION_DEF (text, split);
/***
 * @method rspamd_text:at(pos)
 * Returns a byte at the position `pos`
 * @param {integer} pos index
 * @return {integer} byte at the position `pos` or nil if pos out of bound
 */
LUA_FUNCTION_DEF (text, at);
/***
 * @method rspamd_text:memchr(chr, [reverse])
 * Returns the first or the last position of the character `chr` in the text or
 * -1 in case if a character has not been found. Indexes start from `1`
 * @param {string/number} chr character or a character code to find
 * @param {boolean} reverse last character if `true`
 * @return {integer} position of the character or `-1`
 */
LUA_FUNCTION_DEF (text, memchr);
/***
 * @method rspamd_text:bytes()
 * Converts text to an array of bytes
 * @return {table|integer} bytes in the array (as unsigned char)
 */
LUA_FUNCTION_DEF (text, bytes);
/***
 * @method rspamd_text:lower([is_utf, [inplace]])
 * Return a new text with lowercased characters, if is_utf is true then Rspamd applies utf8 lowercase
 * @param {boolean} is_utf apply utf8 lowercase
 * @param {boolean} inplace lowercase the original text
 * @return {rspamd_text} new rspamd_text (or the original text if inplace) with lowercased letters
 */
LUA_FUNCTION_DEF (text, lower);
LUA_FUNCTION_DEF (text, take_ownership);
/***
 * @method rspamd_text:exclude_chars(set_to_exclude, [always_copy])
 * Returns a text (if owned, then the original text is modified, if not, then it is copied and owned)
 * where all chars from `set_to_exclude` are removed
 * Patterns supported:
 *
 * - %s - all space characters
 * - %n - all newline characters
 * - %c - all control characters (it includes 8bit characters and spaces)
 * - %8 - all 8 bit characters
 * - %% - just a percent character
 *
 * @param {string} set_to_exclude characters to exclude
 * @param {boolean} always_copy always copy the source text
 * @return {rspamd_text} modified or copied text
 */
LUA_FUNCTION_DEF (text, exclude_chars);
/***
 * @method rspamd_text:oneline([always_copy])
 * Returns a text (if owned, then the original text is modified, if not, then it is copied and owned)
 * where the following transformations are made:
 * - All spaces sequences are replaced with a single space
 * - All newlines sequences are replaced with a single space
 * - Trailing and leading spaces are removed
 * - Control characters are excluded
 * - UTF8 sequences are normalised
 *
 * @param {boolean} always_copy always copy the source text
 * @return {rspamd_text} modified or copied text
 */
LUA_FUNCTION_DEF (text, oneline);
/***
 * @method rspamd_text:base32([b32type])
 * Returns a text encoded in base32 (new rspamd_text is allocated)
 *
 * @param {string} b32type base32 type (default, bleach, rfc)
 * @return {rspamd_text} new text encoded in base32
 */
LUA_FUNCTION_DEF (text, base32);
/***
 * @method rspamd_text:base64([line_length, [nline, [fold]]])
 * Returns a text encoded in base64 (new rspamd_text is allocated)
 *
 * @param {number} line_length return text splited with newlines up to this attribute
 * @param {string} nline newline type: `cr`, `lf`, `crlf`
 * @param {boolean} fold use folding when splitting into lines (false by default)
 * @return {rspamd_text} new text encoded in base64
 */
LUA_FUNCTION_DEF (text, base64);
/***
 * @method rspamd_text:hex()
 * Returns a text encoded in hex (new rspamd_text is allocated)
 *
 * @return {rspamd_text} new text encoded in hex
 */
LUA_FUNCTION_DEF (text, hex);
/***
 * @method rspamd_text:find(pattern [, init])
 * Looks for the first match of pattern in the string s.
 * If it finds a match, then find returns the indices of s where this occurrence
 * starts and ends; otherwise, it returns nil. A third,
 * optional numerical argument init specifies where to start the search;
 * its default value is 1 and can be negative.
 * This method currently supports merely a plain search, no patterns.
 *
 * @param {string} pattern pattern to find
 * @param {number} init specifies where to start the search (1 default)
 * @return {number,number/nil} If it finds a match, then find returns the indices of s where this occurrence starts and ends; otherwise, it returns nil
 */
LUA_FUNCTION_DEF (text, find);
LUA_FUNCTION_DEF (text, gc);
LUA_FUNCTION_DEF (text, eq);
LUA_FUNCTION_DEF (text, lt);
LUA_FUNCTION_DEF (text, concat);
LUA_FUNCTION_DEF (text, strtoul);

static const struct luaL_reg textlib_f[] = {
		LUA_INTERFACE_DEF (text, fromstring),
		{"from_string", lua_text_fromstring},
		LUA_INTERFACE_DEF (text, fromtable),
		{"from_table", lua_text_fromtable},
		LUA_INTERFACE_DEF (text, null),
		LUA_INTERFACE_DEF (text, randombytes),
		{NULL, NULL}
};

static const struct luaL_reg textlib_m[] = {
		LUA_INTERFACE_DEF (text, len),
		LUA_INTERFACE_DEF (text, str),
		LUA_INTERFACE_DEF (text, ptr),
		LUA_INTERFACE_DEF (text, take_ownership),
		LUA_INTERFACE_DEF (text, save_in_file),
		LUA_INTERFACE_DEF (text, span),
		LUA_INTERFACE_DEF (text, sub),
		LUA_INTERFACE_DEF (text, lines),
		LUA_INTERFACE_DEF (text, split),
		LUA_INTERFACE_DEF (text, at),
		LUA_INTERFACE_DEF (text, memchr),
		LUA_INTERFACE_DEF (text, byte),
		LUA_INTERFACE_DEF (text, bytes),
		LUA_INTERFACE_DEF (text, lower),
		LUA_INTERFACE_DEF (text, exclude_chars),
		LUA_INTERFACE_DEF (text, oneline),
		LUA_INTERFACE_DEF (text, base32),
		LUA_INTERFACE_DEF (text, base64),
		LUA_INTERFACE_DEF (text, hex),
		LUA_INTERFACE_DEF (text, find),
		LUA_INTERFACE_DEF (text, strtoul),
		{"write", lua_text_save_in_file},
		{"__len", lua_text_len},
		{"__tostring", lua_text_str},
		{"__gc", lua_text_gc},
		{"__eq", lua_text_eq},
		{"__lt", lua_text_lt},
		{"__concat", lua_text_concat},
		{NULL, NULL}
};

struct rspamd_lua_text *
lua_check_text (lua_State * L, gint pos)
{
	void *ud = rspamd_lua_check_udata (L, pos, "rspamd{text}");
	luaL_argcheck (L, ud != NULL, pos, "'text' expected");
	return ud ? (struct rspamd_lua_text *)ud : NULL;
}

struct rspamd_lua_text *
lua_check_text_or_string (lua_State * L, gint pos)
{
	gint pos_type = lua_type (L, pos);

	if (pos_type == LUA_TUSERDATA) {
		void *ud = rspamd_lua_check_udata (L, pos, "rspamd{text}");
		luaL_argcheck (L, ud != NULL, pos, "'text' expected");
		return ud ? (struct rspamd_lua_text *) ud : NULL;
	}
	else if (pos_type == LUA_TSTRING) {
		/*
		 * Fake static lua_text, we allow to use this function multiple times
		 * by having a small array of static structures.
		 */
		static int cur_txt_idx = 0;
		static struct rspamd_lua_text fake_text[4];
		gsize len;
		int sel_idx;

		sel_idx = cur_txt_idx++ % G_N_ELEMENTS (fake_text);
		fake_text[sel_idx].start = lua_tolstring (L, pos, &len);

		if (len >= G_MAXUINT) {
			return NULL;
		}

		fake_text[sel_idx].len = len;
		fake_text[sel_idx].flags = RSPAMD_TEXT_FLAG_FAKE;

		return &fake_text[sel_idx];
	}

	return NULL;
}

struct rspamd_lua_text *
lua_new_text (lua_State *L, const gchar *start, gsize len, gboolean own)
{
	struct rspamd_lua_text *t;

	t = lua_newuserdata (L, sizeof (*t));
	t->flags = 0;

	if (own) {
		gchar *storage;

		if (len > 0) {
			storage = g_malloc (len);

			if (start != NULL) {
				memcpy (storage, start, len);
			}

			t->start = storage;
			t->flags = RSPAMD_TEXT_FLAG_OWN;
		}
		else {
			t->start = "";
		}
	}
	else {
		t->start = start;
	}

	t->len = len;
	rspamd_lua_setclass (L, "rspamd{text}", -1);

	return t;
}


static gint
lua_text_fromstring (lua_State *L)
{
	LUA_TRACE_POINT;
	const gchar *str;
	gsize l = 0;
	gboolean transparent = FALSE;

	str = luaL_checklstring (L, 1, &l);

	if (str) {
		if (lua_isboolean (L, 2)) {
			transparent = lua_toboolean (L, 2);
		}

		lua_new_text (L, str, l, !transparent);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}


	return 1;
}

static gint
lua_text_null (lua_State *L)
{
	LUA_TRACE_POINT;

	lua_new_text (L, NULL, 0, false);

	return 1;
}

static gint
lua_text_randombytes (lua_State *L)
{
	LUA_TRACE_POINT;
	guint nbytes = luaL_checkinteger (L, 1);
	struct rspamd_lua_text *out;

	out = lua_new_text (L, NULL, nbytes, TRUE);
	randombytes_buf ((char *)out->start, nbytes);
	out->len = nbytes;

	return 1;
}

#define MAX_REC 10

static void
lua_text_tbl_length (lua_State *L, gsize dlen, gsize *dest, guint rec)
{
	gsize tblen, stlen;
	struct rspamd_lua_text *elt;

	if (rec > MAX_REC) {
		luaL_error (L, "lua_text_tbl_length: recursion limit exceeded");

		return;
	}

	tblen = rspamd_lua_table_size (L, -1);

	for (gsize i = 0; i < tblen; i ++) {
		lua_rawgeti (L, -1, i + 1);

		if (lua_type (L, -1) == LUA_TSTRING) {
#if LUA_VERSION_NUM >= 502
			stlen = lua_rawlen (L, -1);
#else
			stlen = lua_objlen (L, -1);
#endif
			(*dest) += stlen;
		}
		else if (lua_type (L, -1) == LUA_TUSERDATA){
			elt = (struct rspamd_lua_text *)lua_touserdata (L, -1);

			if (elt) {
				(*dest) += elt->len;
			}
		}
		else if (lua_type (L, -1) == LUA_TTABLE) {
			lua_text_tbl_length (L, dlen, dest, rec + 1);
		}

		if (i != tblen - 1) {
			(*dest) += dlen;
		}

		lua_pop (L, 1);
	}
}

static void
lua_text_tbl_append (lua_State *L,
					 const gchar *delim,
					 gsize dlen,
					 gchar **dest,
					 guint rec)
{
	const gchar *st;
	gsize tblen, stlen;
	struct rspamd_lua_text *elt;

	if (rec > MAX_REC) {
		luaL_error (L, "lua_text_tbl_length: recursion limit exceeded");

		return;
	}

	tblen = rspamd_lua_table_size (L, -1);

	for (guint i = 0; i < tblen; i ++) {
		lua_rawgeti (L, -1, i + 1);

		if (lua_type (L, -1) == LUA_TSTRING) {
			st = lua_tolstring (L, -1, &stlen);
			memcpy ((*dest), st, stlen);
			(*dest) += stlen;
		}
		else if (lua_type (L, -1) == LUA_TUSERDATA){
			elt = (struct rspamd_lua_text *)lua_touserdata (L, -1);

			if (elt) {
				memcpy ((*dest), elt->start, elt->len);
				(*dest) += elt->len;
			}
		}
		else if (lua_type (L, -1) == LUA_TTABLE) {
			lua_text_tbl_append (L, delim, dlen, dest, rec + 1);
		}

		if (dlen && i != tblen - 1) {
			memcpy ((*dest), delim, dlen);
			(*dest) += dlen;
		}

		lua_pop (L, 1);
	}
}

static gint
lua_text_fromtable (lua_State *L)
{
	LUA_TRACE_POINT;
	const gchar *delim = "";
	struct rspamd_lua_text *t;
	gsize textlen = 0, dlen, oldtop = lua_gettop (L);
	gchar *dest;

	if (!lua_istable (L, 1)) {
		return luaL_error (L, "invalid arguments");
	}

	if (lua_type (L, 2) == LUA_TSTRING) {
		delim = lua_tolstring (L, 2, &dlen);
	}
	else {
		dlen = 0;
	}

	/* Calculate length needed */
	lua_pushvalue (L, 1);
	lua_text_tbl_length (L, dlen, &textlen, 0);
	lua_pop (L, 1);

	/* Allocate new text */
	t = lua_newuserdata (L, sizeof (*t));
	dest = g_malloc (textlen);
	t->start = dest;
	t->len = textlen;
	t->flags = RSPAMD_TEXT_FLAG_OWN;
	rspamd_lua_setclass (L, "rspamd{text}", -1);

	lua_pushvalue (L, 1);
	lua_text_tbl_append (L, delim, dlen, &dest, 0);
	lua_pop (L, 1); /* Table arg */

	gint newtop = lua_gettop (L);
	g_assert ( newtop== oldtop + 1);

	return 1;
}

static gint
lua_text_len (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gsize l = 0;

	if (t != NULL) {
		l = t->len;
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	lua_pushinteger (L, l);

	return 1;
}

static gint
lua_text_str (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t != NULL) {
		lua_pushlstring (L, t->start, t->len);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_ptr (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t != NULL) {
		lua_pushlightuserdata (L, (gpointer)t->start);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_take_ownership (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gchar *dest;

	if (t != NULL) {
		if (t->flags & RSPAMD_TEXT_FLAG_OWN) {
			/* We already own it */
			lua_pushboolean (L, true);
		}
		else {
			dest = g_malloc (t->len);
			memcpy (dest, t->start, t->len);
			t->start = dest;
			t->flags |= RSPAMD_TEXT_FLAG_OWN;
			lua_pushboolean (L, true);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_span (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gint64 start = lua_tointeger (L, 2), len = -1;

	if (t && start >= 1 && start <= t->len) {
		if (lua_isnumber (L, 3)) {
			len = lua_tonumber (L, 3);
		}

		if (len == -1) {
			len = t->len - (start - 1);
		}

		if (len < 0 || (len > (t->len - (start - 1)))) {
			return luaL_error (L, "invalid length");
		}

		lua_new_text (L, t->start + (start - 1), len, FALSE);
	}
	else {
		if (!t) {
			return luaL_error (L, "invalid arguments, text required");
		}
		else {
			return luaL_error (L, "invalid arguments: start offset %d "
						 "is larger than text len %d", (int)start, (int)t->len);
		}
	}

	return 1;
}

/* Helpers to behave exactly as Lua does */
static inline gsize
relative_pos_start (gint pos, gsize len)
{
	if (pos > 0) {
		return pos;
	}
	else if (pos == 0) {
		return 1;
	}
	else if (pos < -((gint) len)) {
		return 1;
	}

	/* Negative pos inside str */
	return len + ((gsize)pos) + 1;
}

static inline gsize
relative_pos_end (gint pos, gsize len)
{
	if (pos > (gint)len) {
		return len;
	}
	else if (pos >= 0) {
		return (size_t) pos;
	}
	else if (pos < -((gint)len)) {
		return 0;
	}

	return len + ((gsize)pos) + 1;
}

static gint
lua_text_sub (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t) {
		size_t start = relative_pos_start (luaL_checkinteger (L, 2),
				t->len);
		size_t end = relative_pos_end (luaL_optinteger (L, 3, -1),
				t->len);


		if (start <= end) {
			lua_new_text (L, t->start + (start - 1),
					(end - start) + 1, FALSE);
		}
		else {
			lua_new_text (L, "", 0, TRUE);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint64
rspamd_lua_text_push_line (lua_State *L,
						   struct rspamd_lua_text *t,
						   gint64 start_offset,
						   const gchar *sep_pos,
						   gboolean stringify)
{
	const gchar *start;
	gsize len;
	gint64 ret;

	start = t->start + start_offset;
	len = sep_pos ? (sep_pos - start) : (t->len - start_offset);
	ret = start_offset + len;

	/* Trim line */
	while (len > 0) {
		if (start[len - 1] == '\r' || start[len - 1] == '\n') {
			len --;
		}
		else {
			break;
		}
	}

	if (stringify) {
		lua_pushlstring (L, start, len);
	}
	else {
		struct rspamd_lua_text *ntext;

		ntext = lua_newuserdata (L, sizeof (*ntext));
		rspamd_lua_setclass (L, "rspamd{text}", -1);
		ntext->start = start;
		ntext->len = len;
		ntext->flags = 0; /* Not own as it must be owned by a top object */
	}

	return ret;
}

static gint
rspamd_lua_text_readline (lua_State *L)
{
	struct rspamd_lua_text *t = lua_touserdata (L, lua_upvalueindex (1));
	gboolean stringify = lua_toboolean (L, lua_upvalueindex (2));
	gint64 pos = lua_tointeger (L, lua_upvalueindex (3));

	if (pos < 0) {
		return luaL_error (L, "invalid pos: %d", (gint)pos);
	}

	if (pos >= t->len) {
		/* We are done */
		return 0;
	}

	const gchar *sep_pos;

	/* We look just for `\n` ignoring `\r` as it is very rare nowadays */
	sep_pos = memchr (t->start + pos, '\n', t->len - pos);

	if (sep_pos == NULL) {
		/* Either last `\n` or `\r` separated text */
		sep_pos = memchr (t->start + pos, '\r', t->len - pos);
	}

	pos = rspamd_lua_text_push_line (L, t, pos, sep_pos, stringify);

	/* Skip separators */
	while (pos < t->len) {
		if (t->start[pos] == '\n' || t->start[pos] == '\r') {
			pos ++;
		}
		else {
			break;
		}
	}

	/* Update pos */
	lua_pushinteger (L, pos);
	lua_replace (L, lua_upvalueindex (3));

	return 1;
}

static gint
lua_text_lines (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gboolean stringify = FALSE;

	if (t) {
		if (lua_isboolean (L, 2)) {
			stringify = lua_toboolean (L, 2);
		}

		lua_pushvalue (L, 1);
		lua_pushboolean (L, stringify);
		lua_pushinteger (L, 0); /* Current pos */
		lua_pushcclosure (L, rspamd_lua_text_readline, 3);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
rspamd_lua_text_regexp_split (lua_State *L) {
	struct rspamd_lua_text *t = lua_touserdata (L, lua_upvalueindex (1)),
			*new_t;
	struct rspamd_lua_regexp *re = *(struct rspamd_lua_regexp **)
			lua_touserdata (L, lua_upvalueindex (2));
	gboolean stringify = lua_toboolean (L, lua_upvalueindex (3));
	gint64 pos = lua_tointeger (L, lua_upvalueindex (4));
	gboolean matched;

	if (pos < 0) {
		return luaL_error (L, "invalid pos: %d", (gint) pos);
	}

	if (pos >= t->len) {
		/* We are done */
		return 0;
	}

	const gchar *start, *end, *old_start;

	end = t->start + pos;

	for (;;) {
		old_start = end;

		matched = rspamd_regexp_search (re->re, t->start, t->len, &start, &end, FALSE,
				NULL);

		if (matched) {
			if (start - old_start > 0) {
				if (stringify) {
					lua_pushlstring (L, old_start, start - old_start);
				}
				else {
					new_t = lua_newuserdata (L, sizeof (*t));
					rspamd_lua_setclass (L, "rspamd{text}", -1);
					new_t->start = old_start;
					new_t->len = start - old_start;
					new_t->flags = 0;
				}

				break;
			}
			else {
				if (start == end) {
					matched = FALSE;
					break;
				}
				/*
				 * All match separators (e.g. starting separator,
				 * we need to skip it). Continue iterations.
				 */
			}
		}
		else {
			/* No match, stop */
			break;
		}
	}

	if (!matched && (t->len > 0 && (end == NULL || end < t->start + t->len))) {
		/* No more matches, but we might need to push the last element */
		if (end == NULL) {
			end = t->start;
		}
		/* No separators, need to push the whole remaining part */
		if (stringify) {
			lua_pushlstring (L, end, (t->start + t->len) - end);
		}
		else {
			new_t = lua_newuserdata (L, sizeof (*t));
			rspamd_lua_setclass (L, "rspamd{text}", -1);
			new_t->start = end;
			new_t->len = (t->start + t->len) - end;
			new_t->flags = 0;
		}

		pos = t->len;
	}
	else {

		pos = end - t->start;
	}

	/* Update pos */
	lua_pushinteger (L, pos);
	lua_replace (L, lua_upvalueindex (4));

	return 1;
}

static gint
lua_text_split (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	struct rspamd_lua_regexp *re;
	gboolean stringify = FALSE, own_re = FALSE;

	if (t == NULL) {
		return luaL_error (L, "invalid arguments");
	}

	if (lua_type (L, 2) == LUA_TUSERDATA) {
		re = lua_check_regexp (L, 2);
	}
	else {
		rspamd_regexp_t *c_re;
		GError *err = NULL;

		c_re = rspamd_regexp_new (lua_tostring (L, 2), NULL, &err);
		if (c_re == NULL) {

			gint ret = luaL_error (L, "cannot parse regexp: %s, error: %s",
					lua_tostring (L, 2),
					err == NULL ? "undefined" : err->message);
			if (err) {
				g_error_free (err);
			}

			return ret;
		}

		re = g_malloc0 (sizeof (struct rspamd_lua_regexp));
		re->re = c_re;
		re->re_pattern = g_strdup (lua_tostring (L, 2));
		re->module = rspamd_lua_get_module_name (L);
		own_re = TRUE;
	}

	if (re) {
		if (lua_isboolean (L, 3)) {
			stringify = lua_toboolean (L, 3);
		}

		/* Upvalues */
		lua_pushvalue (L, 1); /* text */

		if (own_re) {
			struct rspamd_lua_regexp **pre;
			pre = lua_newuserdata (L, sizeof (struct rspamd_lua_regexp *));
			rspamd_lua_setclass (L, "rspamd{regexp}", -1);
			*pre = re;
		}
		else {
			lua_pushvalue (L, 2); /* regexp */
		}

		lua_pushboolean (L, stringify);
		lua_pushinteger (L, 0); /* Current pos */
		lua_pushcclosure (L, rspamd_lua_text_regexp_split, 4);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}


static gint
lua_text_at (lua_State *L)
{
	return lua_text_byte(L);
}

static gint
lua_text_byte (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	if (!t) {
		return luaL_error (L, "invalid arguments");
	}

	gsize start = relative_pos_start (luaL_optinteger (L, 2, 1), t->len);
	gsize end = relative_pos_end (luaL_optinteger (L, 3, start), t->len);
	start--;

	if (start >= end) {
		return 0;
	}

	for (gsize i = start; i < end; i++) {
		lua_pushinteger (L, t->start[i]);
	}
	return end - start;
}

static gint
lua_text_memchr (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	int c;
	bool reverse = false;

	if (lua_isnumber (L, 2)) {
		c = lua_tonumber (L, 2);
	}
	else {
		gsize l;
		const gchar *str = lua_tolstring (L, 2, &l);

		if (str) {
			c = str[0];

			if (l != 1) {
				return luaL_error (L, "need exactly one character to search");
			}
		}
		else {
			return luaL_error (L, "invalid arguments");
		}
	}

	if (t) {
		void *f;

		if (lua_isboolean (L, 3)) {
			reverse = lua_toboolean (L, 3);
		}

		if (reverse) {
			f = rspamd_memrchr (t->start, c, t->len);
		}
		else {
			f = memchr (t->start, c, t->len);
		}

		if (f) {
			lua_pushinteger (L, ((const char *)f) - t->start + 1);
		}
		else {
			lua_pushinteger (L, -1);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_bytes (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t) {
		lua_createtable (L, t->len, 0);

		for (gsize i = 0; i < t->len; i ++) {
			lua_pushinteger (L, (guchar)t->start[i]);
			lua_rawseti (L, -2, i + 1);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_save_in_file (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	const gchar *fname = NULL;
	guint mode = 00644;
	gint fd = -1;
	gboolean need_close = FALSE;

	if (t != NULL) {
		if (lua_type (L, 2) == LUA_TSTRING) {
			fname = luaL_checkstring (L, 2);

			if (lua_type (L, 3) == LUA_TNUMBER) {
				mode = lua_tonumber (L, 3);
			}
		}
		else if (lua_type (L, 2) == LUA_TNUMBER) {
			/* Created fd */
			fd = lua_tonumber (L, 2);
		}

		if (fd == -1) {
			if (fname) {
				fd = rspamd_file_xopen (fname, O_CREAT | O_WRONLY | O_EXCL, mode, 0);

				if (fd == -1) {
					lua_pushboolean (L, false);
					lua_pushstring (L, strerror (errno));

					return 2;
				}
				need_close = TRUE;
			}
			else {
				fd = STDOUT_FILENO;
			}
		}

		if (write (fd, t->start, t->len) == -1) {
			if (fd != STDOUT_FILENO) {
				close (fd);
			}

			lua_pushboolean (L, false);
			lua_pushstring (L, strerror (errno));

			return 2;
		}

		if (need_close) {
			close (fd);
		}

		lua_pushboolean (L, true);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_gc (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t != NULL) {
		g_assert (!(t->flags & RSPAMD_TEXT_FLAG_FAKE));

		if (t->flags & RSPAMD_TEXT_FLAG_OWN) {
			if (t->flags & RSPAMD_TEXT_FLAG_WIPE) {
				rspamd_explicit_memzero ((guchar *)t->start, t->len);
			}

			if (t->flags & RSPAMD_TEXT_FLAG_MMAPED) {
				munmap ((gpointer)t->start, t->len);
			}
			else {
				if (t->flags & RSPAMD_TEXT_FLAG_SYSMALLOC) {
					free ((gpointer) t->start);
				}
				else {
					g_free ((gpointer) t->start);
				}
			}
		}

	}

	return 0;
}

static gint
lua_text_eq (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t1 = lua_check_text_or_string (L, 1),
			*t2 = lua_check_text_or_string (L, 2);

	if (t1->len == t2->len) {
		lua_pushboolean (L, memcmp (t1->start, t2->start, t1->len) == 0);
	}
	else {
		lua_pushboolean (L, false);
	}

	return 1;
}

static gint
lua_text_lt (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t1 = lua_check_text_or_string (L, 1),
			*t2 = lua_check_text_or_string (L, 2);

	if (t1 && t2) {
		if (t1->len == t2->len) {
			lua_pushboolean (L, memcmp (t1->start, t2->start, t1->len) < 0);
		}
		else {
			lua_pushboolean (L, t1->len < t2->len);
		}
	}

	return 1;
}

static gint
lua_text_concat (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t1 = lua_check_text_or_string (L, 1),
			*t2 = lua_check_text_or_string (L, 2);

	if (t1 && t2) {
		struct rspamd_lua_text *final;

		final = lua_new_text (L, NULL, t1->len + t2->len, TRUE);
		memcpy ((void *)final->start, t1->start, t1->len);
		memcpy ((void *)(final->start + t1->len), t2->start, t2->len);
	}

	return 1;
}

static gint
lua_text_wipe (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t != NULL) {
		if (t->flags & RSPAMD_TEXT_FLAG_OWN) {
			rspamd_explicit_memzero ((guchar *)t->start, t->len);
		}
		else {
			return luaL_error (L, "cannot wipe not owned text");
		}

	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 0;
}

static gint
lua_text_base32 (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1), *out;
	enum rspamd_base32_type btype = RSPAMD_BASE32_DEFAULT;

	if (t != NULL) {
		if (lua_type (L, 2) == LUA_TSTRING) {
			btype = rspamd_base32_decode_type_from_str (lua_tostring (L, 2));

			if (btype == RSPAMD_BASE32_INVALID) {
				return luaL_error (L, "invalid b32 type: %s", lua_tostring (L, 2));
			}
		}

		out = lua_new_text (L, NULL, t->len * 8 / 5 + 2, TRUE);
		out->len = rspamd_encode_base32_buf (t->start, t->len, (gchar *)out->start,
				out->len, btype);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_base64 (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1), *out;
	gsize line_len = 0;
	gboolean fold = FALSE;

	if (t != NULL) {
		if (lua_type (L, 2) == LUA_TNUMBER) {
			line_len = lua_tointeger (L, 2);

			if (line_len <= 8) {
				return luaL_error (L, "too small line length (at least 8 is required)");
			}
		}

		enum rspamd_newlines_type how = RSPAMD_TASK_NEWLINES_CRLF;

		if (lua_type (L, 3) == LUA_TSTRING) {
			const gchar *how_str = lua_tostring (L, 3);

			if (g_ascii_strcasecmp (how_str, "cr") == 0) {
				how = RSPAMD_TASK_NEWLINES_CR;
			}
			else if (g_ascii_strcasecmp (how_str, "lf") == 0) {
				how = RSPAMD_TASK_NEWLINES_LF;
			}
			else if (g_ascii_strcasecmp (how_str, "crlf") != 0) {
				return luaL_error (L, "invalid newline style: %s", how_str);
			}
		}

		if (lua_type (L, 4) == LUA_TBOOLEAN) {
			fold = lua_toboolean (L, 4);
		}

		gsize sz_len;

		out = lua_newuserdata (L, sizeof (*t));
		out->flags = RSPAMD_TEXT_FLAG_OWN;
		out->start = rspamd_encode_base64_common (t->start, t->len,
				line_len, &sz_len, fold, how);
		out->len = sz_len;
		rspamd_lua_setclass (L, "rspamd{text}", -1);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_hex (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1), *out;

	if (t != NULL) {

		out = lua_new_text (L, NULL, t->len * 2, TRUE);
		out->len = rspamd_encode_hex_buf (t->start, t->len, (gchar *)out->start,
				out->len);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_find (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gsize patlen, init = 1;
	const gchar *pat = luaL_checklstring (L, 2, &patlen);

	if (t != NULL && pat != NULL) {

		if (lua_isnumber (L, 3)) {
			init = relative_pos_start (lua_tointeger (L, 3), t->len);
		}

		init --;

		if (init > t->len) {
			return luaL_error (L, "invalid arguments to find: init too large");
		}

		goffset pos = rspamd_substring_search (t->start + init,
				t->len - init,
				pat, patlen);

		if (pos == -1) {
			lua_pushnil (L);

			return 1;
		}

		lua_pushinteger (L, pos + 1);
		lua_pushinteger (L, pos + patlen);
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 2;
}

#define BITOP(a,b,op) \
		((a)[(gsize)(b)/(8*sizeof *(a))] op (gsize)1<<((gsize)(b)%(8*sizeof *(a))))

static gint
lua_text_exclude_chars (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	gssize patlen;
	const gchar *pat = lua_tolstring (L, 2, &patlen), *p, *end;
	gchar *dest, *d;
	gsize byteset[32 / sizeof(gsize)]; /* Bitset for ascii */
	gboolean copy = TRUE;
	guint *plen;

	if (t != NULL && pat && patlen > 0) {
		if (lua_isboolean (L, 3)) {
			copy = lua_toboolean (L, 3);
		}
		else if (t->flags & RSPAMD_TEXT_FLAG_OWN) {
			copy = FALSE;
		}

		if (!copy) {
			dest = (gchar *)t->start;
			plen = &t->len;
			lua_pushvalue (L, 1); /* Push text as a result */
		}
		else {
			/* We need to copy read only text */
			struct rspamd_lua_text *nt;

			dest = g_malloc (t->len);
			nt = lua_newuserdata (L, sizeof (*nt));
			rspamd_lua_setclass (L, "rspamd{text}", -1);
			nt->len = t->len;
			nt->flags = RSPAMD_TEXT_FLAG_OWN;
			memcpy (dest, t->start, t->len);
			nt->start = dest;
			plen = &nt->len;
		}

		/* Fill pattern bitset */
		memset (byteset, 0, sizeof byteset);

		while (patlen > 0) {
			if (*pat == '%') {
				pat ++;
				patlen --;

				if (patlen > 0) {
					/*
					 * This stuff assumes little endian, but GSIZE_FROM_LE should
					 * deal with proper conversion
					 */
					switch (*pat) {
					case '%':
						BITOP (byteset, *(guchar *) pat, |=);
						break;
					case 's':
						/* "\r\n\t\f " */
						byteset[0] |= GSIZE_FROM_LE (0x100003600);
						break;
					case 'n':
						/* newlines: "\r\n" */
						byteset[0] |= GSIZE_FROM_LE (0x2400);
						break;
					case '8':
						/* 8 bit characters */
						byteset[2] |= GSIZE_FROM_LE (0xffffffffffffffffLLU);
						byteset[3] |= GSIZE_FROM_LE (0xffffffffffffffffLLU);
						break;
					case 'c':
						/* Non printable (control) characters */
						byteset[0] |= GSIZE_FROM_LE (0xffffffff);
						/* Del character */
						byteset[1] |= GSIZE_FROM_LE (0x8000000000000000);
						break;
					}
				}
				else {
					/* Last '%' */
					BITOP (byteset, (guchar)'%', |=);
				}
			}
			else {
				BITOP (byteset, *(guchar *)pat, |=);
			}

			pat ++;
			patlen --;
		}
		for (; patlen > 0 && BITOP (byteset, *(guchar *)pat, |=); pat++, patlen --);

		p = t->start;
		end = t->start + t->len;
		d = dest;

		while (p < end) {
			if (!BITOP (byteset, *(guchar *)p, &)) {
				*d++ = *p;
			}

			p ++;
		}

		*(plen) = d - dest;
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_oneline (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);
	const gchar *p, *end;
	gchar *dest, *d;
	gsize byteset[32 / sizeof(gsize)]; /* Bitset for ascii */
	gboolean copy = TRUE, seen_8bit = FALSE;
	guint *plen;

	if (t != NULL) {
		if (lua_isboolean (L, 2)) {
			copy = lua_toboolean (L, 2);
		}
		else if (t->flags & RSPAMD_TEXT_FLAG_OWN) {
			copy = FALSE;
		}

		if (!copy) {
			dest = (gchar *)t->start;
			plen = &t->len;
			lua_pushvalue (L, 1); /* Push text as a result */
		}
		else {
			/* We need to copy read only text */
			struct rspamd_lua_text *nt;

			dest = g_malloc (t->len);
			nt = lua_newuserdata (L, sizeof (*nt));
			rspamd_lua_setclass (L, "rspamd{text}", -1);
			nt->len = t->len;
			nt->flags = RSPAMD_TEXT_FLAG_OWN;
			memcpy (dest, t->start, t->len);
			nt->start = dest;
			plen = &nt->len;
		}

		/* Fill pattern bitset */
		memset (byteset, 0, sizeof byteset);
		/* All spaces */
		byteset[0] |= GSIZE_FROM_LE (0x100003600);
		/* Control characters */
		byteset[0] |= GSIZE_FROM_LE (0xffffffff);
		/* Del character */
		byteset[1] |= GSIZE_FROM_LE (0x8000000000000000);
		/* 8 bit characters */
		byteset[2] |= GSIZE_FROM_LE (0xffffffffffffffffLLU);
		byteset[3] |= GSIZE_FROM_LE (0xffffffffffffffffLLU);

		p = t->start;
		end = t->start + t->len;
		d = dest;

		while (p < end) {
			if (!BITOP (byteset, *(guchar *)p, &)) {
				*d++ = *p;
			}
			else {
				if ((*(guchar *)p) & 0x80) {
					seen_8bit = TRUE;
					*d++ = *p;
				}
				else {
					if (*p == ' ') {
						if (d != dest) {
							*d++ = *p++;
						}

						while (p < end && g_ascii_isspace (*p)) {
							p ++;
						}

						continue; /* To avoid p++ */
					}
					else if (*p == '\r' || *p == '\n') {
						if (d != dest) {
							*d++ = ' ';
							p ++;
						}

						while (p < end && g_ascii_isspace (*p)) {
							p ++;
						}

						continue; /* To avoid p++ */
					}
				}
			}

			p ++;
		}

		while (d > dest && g_ascii_isspace (*(d - 1))) {
			d --;
		}

		if (seen_8bit) {
			if (rspamd_fast_utf8_validate (dest, d - dest) != 0) {
				/* Need to make it valid :( */
				UChar32 uc;
				goffset err_offset;
				gsize remain = d - dest;
				gchar *nd = dest;

				while (remain > 0 && (err_offset = rspamd_fast_utf8_validate (nd, remain)) > 0) {
					gint i = 0;

					err_offset --; /* As it returns it 1 indexed */
					nd += err_offset;
					remain -= err_offset;

					/* Each invalid character of input requires 3 bytes of output (+2 bytes) */
					while (i < remain) {
						gint old_pos = i;
						U8_NEXT (nd, i, remain, uc);

						if (uc < 0) {
							nd[old_pos] = '?';
						}
						else {
							break;
						}
					}

					nd += i;
					remain -= i;
				}
			}
		}

		*(plen) = d - dest;
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_lower (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1), *nt;
	gboolean is_utf8 = FALSE, is_inplace = FALSE;

	if (t != NULL) {
		if (lua_isboolean (L, 2)) {
			is_utf8 = lua_toboolean (L, 2);
		}
		if (lua_isboolean (L, 3)) {
			is_inplace = lua_toboolean (L, 3);
		}

		if (is_inplace) {
			nt = t;
			lua_pushvalue (L, 1);
		}
		else {
			nt = lua_new_text (L, t->start, t->len, TRUE);
		}

		if (!is_utf8) {
			rspamd_str_lc ((gchar *) nt->start, nt->len);
		}
		else {
			rspamd_str_lc_utf8 ((gchar *) nt->start, nt->len);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

static gint
lua_text_strtoul (lua_State *L)
{
	LUA_TRACE_POINT;
	struct rspamd_lua_text *t = lua_check_text (L, 1);

	if (t) {
		unsigned long ll;

		if (rspamd_strtoul (t->start, t->len, &ll)) {
			lua_pushinteger (L, ll);
		}
		else {
			lua_pushnil (L);
		}
	}
	else {
		return luaL_error (L, "invalid arguments");
	}

	return 1;
}

/* Used to distinguish lua text metatable */
static const guint rspamd_lua_text_cookie = 0x2b21ef6fU;

static gint
lua_load_text (lua_State * L)
{
	lua_newtable (L);
	lua_pushstring (L, "cookie");
	lua_pushnumber (L, rspamd_lua_text_cookie);
	lua_settable (L, -3);
	luaL_register (L, NULL, textlib_f);

	return 1;
}

void
luaopen_text (lua_State *L)
{
	rspamd_lua_new_class (L, "rspamd{text}", textlib_m);
	lua_pushstring (L, "cookie");
	lua_pushnumber (L, rspamd_lua_text_cookie);
	lua_settable (L, -3);
	lua_pop (L, 1);

	rspamd_lua_add_preload (L, "rspamd_text", lua_load_text);
}