index : cgit

/*
 * The plain page, which hands over a repository's own bytes rather than
 * rendering a view of them. A file is written out whole under a content type
 * guessed from its name, though a repository that has not enabled html serving
 * keeps only the types a browser will not act on. A directory, or a request
 * carrying no path, is answered with a bare document of links to the entries
 * below it rather than with one of cgit's themed pages.
 */

#define USE_THE_REPOSITORY_VARIABLE

#include "cgit.h"
#include "html.h"
#include "shared.h"
#include "ui-plain.h"
#include "ui-shared.h"

/*
 * A listing is opened by the entry that matched and closed only once the walk
 * is over, so the end of the page has to tell the three cases apart.
 */
enum response {
	RESPONSE_NONE,
	RESPONSE_BLOB,
	RESPONSE_LISTING
};

struct walk_tree_context {
	// Length of the directory part of the requested path, slash included,
	// and -1 when no path was requested so that no base length can equal
	// it.
	int dir_len;
	enum response response;
};

/*
 * Everything below text/ and application/ can carry markup or script that a
 * browser would run against the site, so only PDF is let back through.
 */
static int is_unsafe_type(const char *mimetype)
{
	return (starts_with(mimetype, "text/") ||
		starts_with(mimetype, "application/")) &&
		strcmp(mimetype, "application/pdf");
}

/*
 * A nonzero return says the response has been written, error pages included,
 * so the walk does not go on to report the path as missing.
 */
static int print_object(const struct object_id *oid, const char *path)
{
	enum object_type type;
	char *buf, *mimetype;
	unsigned long size;

	type = odb_read_object_info(the_repository->objects, oid, &size);
	if (type == OBJ_BAD) {
		cgit_print_error_page(404, "Not found", "Not found");
		return 1;
	}

	// The limit counts kilobytes and is checked before the read, so a huge
	// blob is kept out of memory rather than noticed once it is there.
	if (ctx.cfg.max_blob_size &&
	    size / 1024 > (unsigned long)ctx.cfg.max_blob_size) {
		cgit_print_error_page(413, "Too large",
			"Object size (%luKB) exceeds limit (%dKB)",
			size / 1024, ctx.cfg.max_blob_size);
		return 1;
	}

	buf = odb_read_object(the_repository->objects, oid, &type, &size);
	if (!buf) {
		cgit_print_error_page(404, "Not found", "Not found");
		return 1;
	}

	mimetype = cgit_get_mimetype_for_filename(path);
	ctx.page.mimetype = mimetype;

	if (!ctx.repo->enable_html_serving) {
		// The bytes are whatever the repository holds, so the browser
		// is told not to guess a type of its own and not to load
		// anything they reference. Both lines must go out before
		// cgit_print_http_headers, which closes the header block.
		html("X-Content-Type-Options: nosniff\n");
		html("Content-Security-Policy: default-src 'none'\n");
		if (mimetype && is_unsafe_type(mimetype))
			ctx.page.mimetype = NULL;
	}

	if (!ctx.page.mimetype) {
		if (buffer_is_binary(buf, size)) {
			ctx.page.mimetype = "application/octet-stream";
			ctx.page.charset = NULL;
		} else {
			ctx.page.mimetype = "text/plain";
		}
	}
	ctx.page.filename = path;
	ctx.page.size = size;
	cgit_print_http_headers();
	html_raw(buf, size);
	free(mimetype);
	free(buf);
	return 1;
}

static char *build_path(const char *base, int baselen, const char *path)
{
	if (path[0])
		return cgit_fmtalloc("%.*s%s/", baselen, base, path);
	else
		return cgit_fmtalloc("%.*s/", baselen, base);
}

static void print_dir(const struct object_id *oid, const char *base,
		      int baselen, const char *path)
{
	char *fullpath;
	const char *leading_slash;
	size_t len;

	fullpath = build_path(base, baselen, path);
	leading_slash = (fullpath[0] == '/' ? "" : "/");
	cgit_print_http_headers();
	// The listing is a full document of its own, so it carries the same
	// doctype and charset as the layout pages or the browser would parse
	// it in quirks mode.
	html("<!DOCTYPE html>\n<html lang='en'>\n<head>\n");
	html("<meta charset='utf-8'>\n");
	htmlf("<title>%s", leading_slash);
	html_txt(fullpath);
	htmlf("</title>\n</head>\n<body>\n<h2>%s", leading_slash);
	html_txt(fullpath);
	html("</h2>\n<ul>\n");
	len = strlen(fullpath);
	if (len > 1) {
		char *slash;

		// Nothing left to drop means the parent is the root, which
		// cgit_plain_link is asked for with a null path.
		fullpath[len - 1] = 0;
		slash = strrchr(fullpath, '/');
		if (slash)
			*(slash + 1) = 0;
		else {
			free(fullpath);
			fullpath = NULL;
		}
		html("<li>");
		cgit_plain_link("../", NULL, NULL, ctx.qry.head, ctx.qry.oid,
				fullpath);
		html("</li>\n");
	}
	free(fullpath);
}

static void print_dir_entry(const struct object_id *oid, const char *base,
			    int baselen, const char *path, unsigned mode)
{
	char *fullpath;

	fullpath = build_path(base, baselen, path);
	if (!S_ISDIR(mode) && !S_ISGITLINK(mode))
		fullpath[strlen(fullpath) - 1] = 0;
	html("<li>");
	if (S_ISGITLINK(mode))
		cgit_submodule_link(NULL, fullpath, oid_to_hex(oid));
	else
		cgit_plain_link(path, NULL, NULL, ctx.qry.head, ctx.qry.oid,
				fullpath);
	html("</li>\n");
	free(fullpath);
}

static void print_dir_tail(void)
{
	html("</ul>\n</body>\n</html>\n");
}

/*
 * read_tree reads the return value as a direction rather than a status, so
 * READ_TREE_RECURSIVE means step into this entry and zero means step over it.
 */
static int walk_tree(const struct object_id *oid, struct strbuf *base,
		const char *pathname, unsigned mode, void *context)
{
	struct walk_tree_context *walk = context;

	if (walk->dir_len >= 0 && base->len == (size_t)walk->dir_len) {
		if (S_ISREG(mode) || S_ISLNK(mode)) {
			if (print_object(oid, pathname))
				walk->response = RESPONSE_BLOB;
		} else if (S_ISDIR(mode)) {
			print_dir(oid, base->buf, base->len, pathname);
			walk->response = RESPONSE_LISTING;
			return READ_TREE_RECURSIVE;
		}
	} else if (base->len < INT_MAX && (int)base->len > walk->dir_len) {
		print_dir_entry(oid, base->buf, base->len, pathname, mode);
		walk->response = RESPONSE_LISTING;
	} else if (S_ISDIR(mode)) {
		return READ_TREE_RECURSIVE;
	}

	return 0;
}

static int dir_prefix_len(const char *path)
{
	const char *slash = strrchr(path, '/');
	if (slash)
		return slash - path + 1;
	return 0;
}

void cgit_print_plain(void)
{
	const char *rev = ctx.qry.oid;
	struct object_id oid;
	struct commit *commit;
	int path_len = ctx.qry.path ? strlen(ctx.qry.path) : 0;
	// A hand built pathspec leaves nowildcard_len at zero, which tells git
	// the match may be a glob. It would then hand this walk every entry a
	// pattern like * matches, and each one would be answered with its own
	// set of HTTP headers inside the body of the first.
	struct pathspec_item path_items = {
		.match = ctx.qry.path,
		.len = path_len,
		.nowildcard_len = path_len
	};
	struct pathspec paths = {
		.nr = 1,
		.items = &path_items
	};
	struct walk_tree_context walk = {
		.response = RESPONSE_NONE
	};

	if (!rev)
		rev = ctx.qry.head;

	if (repo_get_oid(the_repository, rev, &oid)) {
		cgit_print_error_page(404, "Not found", "Not found");
		return;
	}
	commit = lookup_commit_reference(the_repository, &oid);
	if (!commit || repo_parse_commit(the_repository, commit)) {
		cgit_print_error_page(404, "Not found", "Not found");
		return;
	}
	if (!path_items.match) {
		// The walk is never handed an entry for the top of the tree
		// itself, so the listing it would have opened is opened here.
		path_items.match = "";
		walk.dir_len = -1;
		print_dir(get_commit_tree_oid(commit), "", 0, "");
		walk.response = RESPONSE_LISTING;
	} else {
		walk.dir_len = dir_prefix_len(path_items.match);
	}
	read_tree(the_repository, repo_get_commit_tree(the_repository, commit),
		  &paths, walk_tree, &walk);
	if (walk.response == RESPONSE_NONE)
		cgit_print_error_page(404, "Not found", "Not found");
	else if (walk.response == RESPONSE_LISTING)
		print_dir_tail();
}