path: root/src/lib-mail/message-parser.c

/* Copyright (c) 2002-2018 Dovecot authors, see the included COPYING file */

#include "lib.h"
#include "array.h"
#include "str.h"
#include "istream.h"
#include "rfc822-parser.h"
#include "rfc2231-parser.h"
#include "message-parser-private.h"

message_part_header_callback_t *null_message_part_header_callback = NULL;

static int parse_next_header_init(struct message_parser_ctx *ctx,
				  struct message_block *block_r);
static int parse_next_body_to_boundary(struct message_parser_ctx *ctx,
				       struct message_block *block_r);
static int parse_next_body_to_eof(struct message_parser_ctx *ctx,
				  struct message_block *block_r);

static struct message_boundary *
boundary_find(struct message_boundary *boundaries,
	      const unsigned char *data, size_t len, bool trailing_dashes)
{
	struct message_boundary *best = NULL;

	/* As MIME spec says: search from latest one to oldest one so that we
	   don't break if the same boundary is used in nested parts. Also the
	   full message line doesn't have to match the boundary, only the
	   beginning. However, if there are multiple prefixes whose beginning
	   matches, use the longest matching one. */
	while (boundaries != NULL) {
		if (boundaries->len <= len &&
		    memcmp(boundaries->boundary, data, boundaries->len) == 0 &&
		    (best == NULL || best->len < boundaries->len)) {
			best = boundaries;
			/* If we see "foo--", it could either mean that there
			   is a boundary named "foo" that ends now or there's
			   a boundary "foo--" which continues. */
			if (best->len == len ||
			    (best->len == len-2 && trailing_dashes)) {
				/* This is exactly the wanted boundary. There
				   can't be a better one. */
				break;
			}
		}

		boundaries = boundaries->next;
	}

	return best;
}

static void parse_body_add_block(struct message_parser_ctx *ctx,
				 struct message_block *block)
{
	unsigned int missing_cr_count = 0;
	const unsigned char *cur, *next, *data = block->data;

	i_assert(block->size > 0);

	block->hdr = NULL;

	/* check if we have NULs */
	if (memchr(data, '\0', block->size) != NULL)
		ctx->part->flags |= MESSAGE_PART_FLAG_HAS_NULS;

	/* count number of lines and missing CRs */
	if (*data == '\n') {
		ctx->part->body_size.lines++;
		if (ctx->last_chr != '\r')
			missing_cr_count++;
	}

	cur = data + 1;
	while ((next = memchr(cur, '\n', block->size - (cur - data))) != NULL) {
		ctx->part->body_size.lines++;
		if (next[-1] != '\r')
			missing_cr_count++;

		cur = next + 1;
	}
	ctx->last_chr = data[block->size - 1];
	ctx->skip += block->size;

	ctx->part->body_size.physical_size += block->size;
	ctx->part->body_size.virtual_size += block->size + missing_cr_count;
}

int message_parser_read_more(struct message_parser_ctx *ctx,
			     struct message_block *block_r, bool *full_r)
{
	int ret;

	if (ctx->skip > 0) {
		i_stream_skip(ctx->input, ctx->skip);
		ctx->skip = 0;
	}

	*full_r = FALSE;
	ret = i_stream_read_bytes(ctx->input, &block_r->data,
				  &block_r->size, ctx->want_count + 1);
	if (ret <= 0) {
		switch (ret) {
		case 0:
			if (!ctx->input->eof) {
				i_assert(!ctx->input->blocking);
				return 0;
			}
			break;
		case -1:
			i_assert(ctx->input->eof ||
				 ctx->input->stream_errno != 0);
			ctx->eof = TRUE;
			if (block_r->size != 0) {
				/* EOF, but we still have some data.
				   return it. */
				return 1;
			}
			return -1;
		case -2:
			*full_r = TRUE;
			break;
		default:
			i_unreached();
		}
	}

	if (!*full_r) {
		/* reset number of wanted characters if we actually got them */
		ctx->want_count = 1;
	}
	return 1;
}

static void
message_part_append(struct message_parser_ctx *ctx)
{
	struct message_part *parent = ctx->part;
	struct message_part *part;

	i_assert(!ctx->preparsed);
	i_assert(parent != NULL);
	i_assert((parent->flags & (MESSAGE_PART_FLAG_MULTIPART |
				   MESSAGE_PART_FLAG_MESSAGE_RFC822)) != 0);

	part = p_new(ctx->part_pool, struct message_part, 1);
	part->parent = parent;

	/* set child position */
	part->physical_pos =
		parent->physical_pos +
		parent->body_size.physical_size +
		parent->header_size.physical_size;

	/* add to parent's linked list */
	*ctx->next_part = part;
	/* update the parent's end-of-linked-list pointer */
	struct message_part **next_part = &part->next;
	array_push_back(&ctx->next_part_stack, &next_part);
	/* This part is now the new parent for the next message_part_append()
	   call. Its linked list begins with the children pointer. */
	ctx->next_part = &part->children;

	ctx->part = part;
	ctx->nested_parts_count++;
	ctx->total_parts_count++;
	i_assert(ctx->nested_parts_count < ctx->max_nested_mime_parts);
	i_assert(ctx->total_parts_count <= ctx->max_total_mime_parts);
}

static void message_part_finish(struct message_parser_ctx *ctx)
{
	struct message_part **const *parent_next_partp;

	if (!ctx->preparsed) {
		i_assert(ctx->nested_parts_count > 0);
		ctx->nested_parts_count--;

		parent_next_partp = array_back(&ctx->next_part_stack);
		array_pop_back(&ctx->next_part_stack);
		ctx->next_part = *parent_next_partp;
	}

	message_size_add(&ctx->part->parent->body_size, &ctx->part->body_size);
	message_size_add(&ctx->part->parent->body_size, &ctx->part->header_size);
	ctx->part->parent->children_count += 1 + ctx->part->children_count;
	ctx->part = ctx->part->parent;
}

static void message_boundary_free(struct message_boundary *b)
{
	i_free(b->boundary);
	i_free(b);
}

static void
boundary_remove_until(struct message_parser_ctx *ctx,
		      struct message_boundary *boundary)
{
	while (ctx->boundaries != boundary) {
		struct message_boundary *cur = ctx->boundaries;

		i_assert(cur != NULL);
		ctx->boundaries = cur->next;
		message_boundary_free(cur);

	}
	ctx->boundaries = boundary;
}

static void parse_next_body_multipart_init(struct message_parser_ctx *ctx)
{
	struct message_boundary *b;

	b = i_new(struct message_boundary, 1);
	b->part = ctx->part;
	b->boundary = ctx->last_boundary;
	ctx->last_boundary = NULL;
	b->len = strlen(b->boundary);

	b->next = ctx->boundaries;
	ctx->boundaries = b;
}

static int parse_next_body_message_rfc822_init(struct message_parser_ctx *ctx,
					       struct message_block *block_r)
{
	message_part_append(ctx);
	return parse_next_header_init(ctx, block_r);
}

static int
boundary_line_find(struct message_parser_ctx *ctx,
		   const unsigned char *data, size_t size, bool full,
		   struct message_boundary **boundary_r)
{
	*boundary_r = NULL;

	if (size < 2) {
		i_assert(!full);

		if (ctx->input->eof)
			return -1;
		ctx->want_count = 2;
		return 0;
	}

	if (data[0] != '-' || data[1] != '-') {
		/* not a boundary, just skip this line */
		return -1;
	}

	if (ctx->total_parts_count >= ctx->max_total_mime_parts) {
		/* can't add any more MIME parts. just stop trying to find
		   more boundaries. */
		ctx->part->flags |= MESSAGE_PART_FLAG_OVERFLOW;
		return -1;
	}

	/* need to find the end of line */
	data += 2;
	size -= 2;
	const unsigned char *lf_pos = memchr(data, '\n', size);
	if (lf_pos == NULL &&
	    size+2 < BOUNDARY_END_MAX_LEN &&
	    !ctx->input->eof && !full) {
		/* no LF found */
		ctx->want_count = BOUNDARY_END_MAX_LEN;
		return 0;
	}
	size_t find_size = size;
	bool trailing_dashes = FALSE;

	if (lf_pos != NULL) {
		find_size = lf_pos - data;
		if (find_size > 0 && data[find_size-1] == '\r')
			find_size--;
		if (find_size > 2 && data[find_size-1] == '-' &&
		    data[find_size-2] == '-')
			trailing_dashes = TRUE;
	} else if (find_size > BOUNDARY_END_MAX_LEN)
		find_size = BOUNDARY_END_MAX_LEN;

	*boundary_r = boundary_find(ctx->boundaries, data, find_size,
				    trailing_dashes);
	if (*boundary_r == NULL)
		return -1;

	(*boundary_r)->epilogue_found =
		size >= (*boundary_r)->len + 2 &&
		memcmp(data + (*boundary_r)->len, "--", 2) == 0;
	return 1;
}

static int parse_next_mime_header_init(struct message_parser_ctx *ctx,
				       struct message_block *block_r)
{
	message_part_append(ctx);
	ctx->part->flags |= MESSAGE_PART_FLAG_IS_MIME;

	return parse_next_header_init(ctx, block_r);
}

static int parse_next_body_skip_boundary_line(struct message_parser_ctx *ctx,
					      struct message_block *block_r)
{
	const unsigned char *ptr;
	int ret;
	bool full;

	if ((ret = message_parser_read_more(ctx, block_r, &full)) <= 0)
		return ret;

	ptr = memchr(block_r->data, '\n', block_r->size);
	if (ptr == NULL) {
		parse_body_add_block(ctx, block_r);
		if (block_r->size > 0 &&
		    (ctx->flags & MESSAGE_PARSER_FLAG_INCLUDE_BOUNDARIES) != 0)
			return 1;
		return 0;
	}

	/* found the LF */
	block_r->size = (ptr - block_r->data) + 1;
	parse_body_add_block(ctx, block_r);

	if (ctx->boundaries == NULL || ctx->boundaries->part != ctx->part) {
		/* epilogue */
		if (ctx->boundaries != NULL)
			ctx->parse_next_block = parse_next_body_to_boundary;
		else
			ctx->parse_next_block = parse_next_body_to_eof;
	} else {
		/* a new MIME part begins */
		ctx->parse_next_block = parse_next_mime_header_init;
	}
	if (block_r->size > 0 &&
	    (ctx->flags & MESSAGE_PARSER_FLAG_INCLUDE_BOUNDARIES) != 0)
		return 1;
	return ctx->parse_next_block(ctx, block_r);
}

static int parse_part_finish(struct message_parser_ctx *ctx,
			     struct message_boundary *boundary,
			     struct message_block *block_r, bool first_line)
{
	size_t line_size;
	size_t boundary_len = boundary->len;
	bool boundary_epilogue_found = boundary->epilogue_found;

	i_assert(ctx->last_boundary == NULL);

	/* get back to parent MIME part, summing the child MIME part sizes
	   into parent's body sizes */
	while (ctx->part != boundary->part) {
		message_part_finish(ctx);
		i_assert(ctx->part != NULL);
	}

	if (boundary->epilogue_found) {
		/* this boundary isn't needed anymore */
		boundary_remove_until(ctx, boundary->next);
	} else {
		/* forget about the boundaries we possibly skipped */
		boundary_remove_until(ctx, boundary);
	}

	/* the boundary itself should already be in buffer. add that. */
	block_r->data = i_stream_get_data(ctx->input, &block_r->size);
	i_assert(block_r->size >= ctx->skip);
	block_r->data += ctx->skip;
	/* [[\r]\n]--<boundary>[--] */
	if (first_line)
		line_size = 0;
	else if (block_r->data[0] == '\r') {
		i_assert(block_r->data[1] == '\n');
		line_size = 2;
	} else {
		i_assert(block_r->data[0] == '\n');
		line_size = 1;
	}
	line_size += 2 + boundary_len + (boundary_epilogue_found ? 2 : 0);
	i_assert(block_r->size >= ctx->skip + line_size);
	block_r->size = line_size;
	parse_body_add_block(ctx, block_r);

	ctx->parse_next_block = parse_next_body_skip_boundary_line;

	if ((ctx->flags & MESSAGE_PARSER_FLAG_INCLUDE_BOUNDARIES) != 0)
		return 1;
	return ctx->parse_next_block(ctx, block_r);
}

static int parse_next_body_to_boundary(struct message_parser_ctx *ctx,
				       struct message_block *block_r)
{
	struct message_boundary *boundary = NULL;
	const unsigned char *data, *cur, *next, *end;
	size_t boundary_start;
	int ret;
	bool full;

	if ((ret = message_parser_read_more(ctx, block_r, &full)) <= 0)
		return ret;

	data = block_r->data;
	if (ctx->last_chr == '\n') {
		/* handle boundary in first line of message. alternatively
		   it's an empty line. */
		ret = boundary_line_find(ctx, block_r->data,
					 block_r->size, full, &boundary);
		if (ret >= 0) {
			return ret == 0 ? 0 :
				parse_part_finish(ctx, boundary, block_r, TRUE);
		}
	}

	i_assert(block_r->size > 0);
	boundary_start = 0;

	/* skip to beginning of the next line. the first line was
	   handled already. */
	cur = data; end = data + block_r->size;
	while ((next = memchr(cur, '\n', end - cur)) != NULL) {
		cur = next + 1;

		boundary_start = next - data;
		if (next > data && next[-1] == '\r')
			boundary_start--;

		if (boundary_start != 0) {
			/* we can at least skip data until the first [CR]LF.
			   input buffer can't be full anymore. */
			full = FALSE;
		}

		ret = boundary_line_find(ctx, cur, end - cur, full, &boundary);
		if (ret >= 0) {
			/* found / need more data */
			if (ret == 0 && boundary_start == 0)
				ctx->want_count += cur - block_r->data;
			break;
		}
	}

	if (next != NULL) {
		/* found / need more data */
		i_assert(ret >= 0);
		i_assert(!(ret == 0 && full));
	} else if (boundary_start == 0) {
		/* no linefeeds in this block. we can just skip it. */
		ret = 0;
		if (block_r->data[block_r->size-1] == '\r' && !ctx->eof) {
			/* this may be the beginning of the \r\n--boundary */
			block_r->size--;
		}
		boundary_start = block_r->size;
	} else {
		/* the boundary wasn't found from this data block,
		   we'll need more data. */
		ret = 0;
		ctx->want_count = (block_r->size - boundary_start) + 1;
	}

	if (ret > 0 || (ret == 0 && !ctx->eof)) {
		/* a) we found the boundary
		   b) we need more data and haven't reached EOF yet
		   so leave CR+LF + last line to buffer */
		block_r->size = boundary_start;
	}
	if (block_r->size != 0) {
		parse_body_add_block(ctx, block_r);

		if ((ctx->part->flags & MESSAGE_PART_FLAG_MULTIPART) != 0 &&
		    (ctx->flags & MESSAGE_PARSER_FLAG_INCLUDE_MULTIPART_BLOCKS) == 0)
			return 0;

		return 1;
	}
	return ret <= 0 ? ret :
		parse_part_finish(ctx, boundary, block_r, FALSE);
}

static int parse_next_body_to_eof(struct message_parser_ctx *ctx,
				  struct message_block *block_r)
{
	bool full;
	int ret;

	if ((ret = message_parser_read_more(ctx, block_r, &full)) <= 0)
		return ret;

	parse_body_add_block(ctx, block_r);

	if ((ctx->part->flags & MESSAGE_PART_FLAG_MULTIPART) != 0 &&
	    (ctx->flags & MESSAGE_PARSER_FLAG_INCLUDE_MULTIPART_BLOCKS) == 0)
		return 0;

	return 1;
}

static void parse_content_type(struct message_parser_ctx *ctx,
			       struct message_header_line *hdr)
{
	struct rfc822_parser_context parser;
	const char *const *results;
	string_t *content_type;
	int ret;

	if (ctx->part_seen_content_type)
		return;
	ctx->part_seen_content_type = TRUE;

	rfc822_parser_init(&parser, hdr->full_value, hdr->full_value_len, NULL);
	rfc822_skip_lwsp(&parser);

	content_type = t_str_new(64);
	ret = rfc822_parse_content_type(&parser, content_type);

	if (strcasecmp(str_c(content_type), "message/rfc822") == 0)
		ctx->part->flags |= MESSAGE_PART_FLAG_MESSAGE_RFC822;
	else if (strncasecmp(str_c(content_type), "text", 4) == 0 &&
		 (str_len(content_type) == 4 ||
		  str_data(content_type)[4] == '/'))
		ctx->part->flags |= MESSAGE_PART_FLAG_TEXT;
	else if (strncasecmp(str_c(content_type), "multipart/", 10) == 0) {
		ctx->part->flags |= MESSAGE_PART_FLAG_MULTIPART;

		if (strcasecmp(str_c(content_type)+10, "digest") == 0)
			ctx->part->flags |= MESSAGE_PART_FLAG_MULTIPART_DIGEST;
	}

	if (ret < 0 ||
	    (ctx->part->flags & MESSAGE_PART_FLAG_MULTIPART) == 0 ||
	    ctx->last_boundary != NULL) {
		rfc822_parser_deinit(&parser);
		return;
	}

	rfc2231_parse(&parser, &results);
	for (; *results != NULL; results += 2) {
		if (strcasecmp(results[0], "boundary") == 0) {
			/* truncate excessively long boundaries */
			i_free(ctx->last_boundary);
			ctx->last_boundary =
				i_strndup(results[1], BOUNDARY_STRING_MAX_LEN);
			break;
		}
	}
	rfc822_parser_deinit(&parser);
}

static bool block_is_at_eoh(const struct message_block *block)
{
	if (block->size < 1)
		return FALSE;
	if (block->data[0] == '\n')
		return TRUE;
	if (block->data[0] == '\r') {
		if (block->size < 2)
			return FALSE;
		if (block->data[1] == '\n')
			return TRUE;
	}
	return FALSE;
}

static bool parse_too_many_nested_mime_parts(struct message_parser_ctx *ctx)
{
	return ctx->nested_parts_count+1 >= ctx->max_nested_mime_parts;
}

#define MUTEX_FLAGS \
	(MESSAGE_PART_FLAG_MESSAGE_RFC822 | MESSAGE_PART_FLAG_MULTIPART)

static int parse_next_header(struct message_parser_ctx *ctx,
			     struct message_block *block_r)
{
	struct message_part *part = ctx->part;
	struct message_header_line *hdr;
	struct message_boundary *boundary;
	bool full;
	int ret;

	if ((ret = message_parser_read_more(ctx, block_r, &full)) == 0)
		return ret;

	if (ret > 0 && block_is_at_eoh(block_r) &&
	    ctx->last_boundary != NULL &&
	    (part->flags & MESSAGE_PART_FLAG_IS_MIME) != 0) {
		/* we are at the end of headers and we've determined that we're
		   going to start a multipart. add the boundary already here
		   at this point so we can reliably determine whether the
		   "\n--boundary" belongs to us or to a previous boundary.
		   this is a problem if the boundary prefixes are identical,
		   because MIME requires only the prefix to match. */
		if (!parse_too_many_nested_mime_parts(ctx)) {
			parse_next_body_multipart_init(ctx);
			ctx->multipart = TRUE;
		} else {
			part->flags |= MESSAGE_PART_FLAG_OVERFLOW;
			part->flags &= ENUM_NEGATE(MESSAGE_PART_FLAG_MULTIPART);
		}
	}

	/* before parsing the header see if we can find a --boundary from here.
	   we're guaranteed to be at the beginning of the line here. */
	if (ret > 0) {
		ret = ctx->boundaries == NULL ? -1 :
			boundary_line_find(ctx, block_r->data,
					   block_r->size, full, &boundary);
		if (ret > 0 && boundary->part == ctx->part) {
			/* our own body begins with our own --boundary.
			   we don't want to handle that yet. */
			ret = -1;
		}
	}
	if (ret < 0) {
		/* no boundary */
		size_t headers_available =
			ctx->all_headers_max_size > ctx->all_headers_total_size ?
			ctx->all_headers_max_size - ctx->all_headers_total_size : 0;
		message_parse_header_lower_limit(ctx->hdr_parser_ctx, headers_available);
		ret = message_parse_header_next(ctx->hdr_parser_ctx, &hdr);
		if (ret > 0) {
			if (!hdr->continues) {
				ctx->all_headers_total_size += hdr->name_len;
				ctx->all_headers_total_size += hdr->middle_len;
			}
			ctx->all_headers_total_size += hdr->value_len;
		}
		if (ret == 0 || (ret < 0 && ctx->input->stream_errno != 0)) {
			ctx->want_count = i_stream_get_data_size(ctx->input) + 1;
			return ret;
		}
	} else if (ret == 0) {
		/* need more data */
		return 0;
	} else {
		/* boundary found. stop parsing headers here. The previous
		   [CR]LF belongs to the MIME boundary though. */
		if (ctx->prev_hdr_newline_size > 0) {
			i_assert(ctx->part->header_size.lines > 0);
			/* remove the newline size from the MIME header */
			ctx->part->header_size.lines--;
			ctx->part->header_size.physical_size -=
				ctx->prev_hdr_newline_size;
			ctx->part->header_size.virtual_size -= 2;
			/* add the newline size to the parent's body */
			ctx->part->parent->body_size.lines++;
			ctx->part->parent->body_size.physical_size +=
				ctx->prev_hdr_newline_size;
			ctx->part->parent->body_size.virtual_size += 2;
		}
		hdr = NULL;
	}

	if (hdr != NULL) {
		if (hdr->eoh)
			;
		else if (strcasecmp(hdr->name, "Mime-Version") == 0) {
			/* it's MIME. Content-* headers are valid */
			part->flags |= MESSAGE_PART_FLAG_IS_MIME;
		} else if (strcasecmp(hdr->name, "Content-Type") == 0) {
			if ((ctx->flags &
			     MESSAGE_PARSER_FLAG_MIME_VERSION_STRICT) == 0)
				part->flags |= MESSAGE_PART_FLAG_IS_MIME;

			if (hdr->continues)
				hdr->use_full_value = TRUE;
			else T_BEGIN {
				parse_content_type(ctx, hdr);
			} T_END;
		}

		block_r->hdr = hdr;
		block_r->size = 0;
		ctx->prev_hdr_newline_size = hdr->no_newline ? 0 :
			(hdr->crlf_newline ? 2 : 1);
		return 1;
	}

	/* end of headers */
	if ((part->flags & MESSAGE_PART_FLAG_IS_MIME) == 0) {
		/* It's not MIME. Reset everything we found from
		   Content-Type. */
		i_assert(!ctx->multipart);
		part->flags = 0;
	}
	i_free(ctx->last_boundary);

	if (!ctx->part_seen_content_type ||
	    (part->flags & MESSAGE_PART_FLAG_IS_MIME) == 0) {
		if (part->parent != NULL &&
		    (part->parent->flags &
		     MESSAGE_PART_FLAG_MULTIPART_DIGEST) != 0) {
			/* when there's no content-type specified and we're
			   below multipart/digest, assume message/rfc822
			   content-type */
			part->flags |= MESSAGE_PART_FLAG_MESSAGE_RFC822;
		} else {
			/* otherwise we default to text/plain */
			part->flags |= MESSAGE_PART_FLAG_TEXT;
		}
	}

	if (message_parse_header_has_nuls(ctx->hdr_parser_ctx))
		part->flags |= MESSAGE_PART_FLAG_HAS_NULS;
	message_parse_header_deinit(&ctx->hdr_parser_ctx);

	i_assert((part->flags & MUTEX_FLAGS) != MUTEX_FLAGS);

	ctx->last_chr = '\n';
	if (ctx->multipart) {
		i_assert(ctx->last_boundary == NULL);
		ctx->multipart = FALSE;
		ctx->parse_next_block = parse_next_body_to_boundary;
	} else if ((part->flags & MESSAGE_PART_FLAG_MESSAGE_RFC822) == 0) {
		/* Not message/rfc822 */
		if (ctx->boundaries != NULL)
			ctx->parse_next_block = parse_next_body_to_boundary;
		else
			ctx->parse_next_block = parse_next_body_to_eof;
	} else if (!parse_too_many_nested_mime_parts(ctx) &&
		   ctx->total_parts_count < ctx->max_total_mime_parts) {
		/* message/rfc822 - not reached MIME part limits yet */
		ctx->parse_next_block = parse_next_body_message_rfc822_init;
	} else {
		/* message/rfc822 - already reached MIME part limits */
		part->flags |= MESSAGE_PART_FLAG_OVERFLOW;
		part->flags &= ENUM_NEGATE(MESSAGE_PART_FLAG_MESSAGE_RFC822);
		if (ctx->boundaries != NULL)
			ctx->parse_next_block = parse_next_body_to_boundary;
		else
			ctx->parse_next_block = parse_next_body_to_eof;
	}

	ctx->want_count = 1;

	/* return empty block as end of headers */
	block_r->hdr = NULL;
	block_r->size = 0;
	return 1;
}

static int parse_next_header_init(struct message_parser_ctx *ctx,
				  struct message_block *block_r)
{
	i_assert(ctx->hdr_parser_ctx == NULL);

	ctx->hdr_parser_ctx =
		message_parse_header_init(ctx->input, &ctx->part->header_size,
					  ctx->hdr_flags);
	ctx->part_seen_content_type = FALSE;
	ctx->prev_hdr_newline_size = 0;

	ctx->parse_next_block = parse_next_header;
	return parse_next_header(ctx, block_r);
}

struct message_parser_ctx *
message_parser_init_int(struct istream *input,
			const struct message_parser_settings *set)
{
	struct message_parser_ctx *ctx;

	ctx = i_new(struct message_parser_ctx, 1);
	ctx->hdr_flags = set->hdr_flags;
	ctx->flags = set->flags;
	ctx->max_nested_mime_parts = set->max_nested_mime_parts != 0 ?
		set->max_nested_mime_parts :
		MESSAGE_PARSER_DEFAULT_MAX_NESTED_MIME_PARTS;
	ctx->max_total_mime_parts = set->max_total_mime_parts != 0 ?
		set->max_total_mime_parts :
		MESSAGE_PARSER_DEFAULT_MAX_TOTAL_MIME_PARTS;
	ctx->all_headers_max_size = set->all_headers_max_size != 0 ?
		set->all_headers_max_size :
		MESSAGE_PARSER_DEFAULT_ALL_HEADERS_MAX_SIZE;
	ctx->input = input;
	i_stream_ref(input);
	return ctx;
}

struct message_parser_ctx *
message_parser_init(pool_t part_pool, struct istream *input,
		    const struct message_parser_settings *set)
{
	struct message_parser_ctx *ctx;

	ctx = message_parser_init_int(input, set);
	ctx->part_pool = part_pool;
	ctx->parts = ctx->part = p_new(part_pool, struct message_part, 1);
	ctx->next_part = &ctx->part->children;
	ctx->parse_next_block = parse_next_header_init;
	ctx->total_parts_count = 1;
	ctx->all_headers_total_size = 0;
	i_array_init(&ctx->next_part_stack, 4);
	return ctx;
}

void message_parser_deinit(struct message_parser_ctx **_ctx,
			  struct message_part **parts_r)
{
	const char *error;

	i_assert((**_ctx).preparsed == FALSE);
	if (message_parser_deinit_from_parts(_ctx, parts_r, &error) < 0)
		i_panic("message_parser_deinit_from_parts: %s", error);
}

int message_parser_deinit_from_parts(struct message_parser_ctx **_ctx,
				     struct message_part **parts_r,
				     const char **error_r)
{
        struct message_parser_ctx *ctx = *_ctx;
	int ret = ctx->broken_reason != NULL ? -1 : 0;

	*_ctx = NULL;
	*parts_r = ctx->parts;
	*error_r = ctx->broken_reason;

	if (ctx->hdr_parser_ctx != NULL)
		message_parse_header_deinit(&ctx->hdr_parser_ctx);
	if (ctx->part != NULL) {
		/* If the whole message has been parsed, the parts are
		   usually finished in message_parser_parse_next_block().
		   However, it's possible that the caller finishes reading
		   through the istream without calling
		   message_parser_parse_next_block() afterwards. In that case
		   we still need to finish these parts. */
		while (ctx->part->parent != NULL)
			message_part_finish(ctx);
	}
	boundary_remove_until(ctx, NULL);
	i_assert(ctx->nested_parts_count == 0);

	i_stream_unref(&ctx->input);
	array_free(&ctx->next_part_stack);
	i_free(ctx->last_boundary);
	i_free(ctx);
	i_assert(ret < 0 || *parts_r != NULL);
	return ret;
}

int message_parser_parse_next_block(struct message_parser_ctx *ctx,
				    struct message_block *block_r)
{
	int ret;
	bool eof = FALSE, full;

	i_zero(block_r);

	while ((ret = ctx->parse_next_block(ctx, block_r)) == 0) {
		ret = message_parser_read_more(ctx, block_r, &full);
		if (ret == 0) {
			i_assert(!ctx->input->blocking);
			return 0;
		}
		if (ret == -1) {
			i_assert(!eof);
			eof = TRUE;
		}
	}

	block_r->part = ctx->part;

	if (ret < 0 && ctx->part != NULL) {
		/* Successful EOF or unexpected failure */
		i_assert(ctx->input->eof || ctx->input->closed ||
			 ctx->input->stream_errno != 0 ||
			 ctx->broken_reason != NULL);
		while (ctx->part->parent != NULL)
			message_part_finish(ctx);
	}

	if (block_r->size == 0) {
		/* data isn't supposed to be read, so make sure it's NULL */
		block_r->data = NULL;
	}
	return ret;
}

#undef message_parser_parse_header
void message_parser_parse_header(struct message_parser_ctx *ctx,
				 struct message_size *hdr_size,
				 message_part_header_callback_t *callback,
				 void *context)
{
	struct message_block block;
	int ret;

	while ((ret = message_parser_parse_next_block(ctx, &block)) > 0) {
		callback(block.part, block.hdr, context);

		if (block.hdr == NULL)
			break;
	}
	i_assert(ret != 0);
	i_assert(ctx->part != NULL);

	if (ret < 0) {
		/* well, can't return error so fake end of headers */
		callback(ctx->part, NULL, context);
	}

        *hdr_size = ctx->part->header_size;
}

#undef message_parser_parse_body
void message_parser_parse_body(struct message_parser_ctx *ctx,
			       message_part_header_callback_t *hdr_callback,
			       void *context)
{
	struct message_block block;
	int ret;

	while ((ret = message_parser_parse_next_block(ctx, &block)) > 0) {
		if (block.size == 0 && hdr_callback != NULL)
			hdr_callback(block.part, block.hdr, context);
	}
	i_assert(ret != 0);
}