index : cgit

/*
 * Reads a blob out of the object database and writes its bytes to the client.
 * The blob is named either directly by object id or by a path walked out of a
 * commit's tree. The blob page serves those bytes as the whole response, under
 * headers that stop a browser treating repository content as markup, while the
 * summary page instead drops a readme's contents into a page it is already
 * building.
 */

#define USE_THE_REPOSITORY_VARIABLE

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

struct walk_tree_context {
	const char *match_path;
	struct object_id *matched_oid;
	unsigned int found_path:1;
	unsigned int file_only:1;
};

/*
 * 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->file_only && !S_ISREG(mode))
		return READ_TREE_RECURSIVE;
	if (strncmp(base->buf, walk->match_path, base->len)
		|| strcmp(walk->match_path + base->len, pathname))
		return READ_TREE_RECURSIVE;
	oidcpy(walk->matched_oid, oid);
	walk->found_path = 1;
	return 0;
}

/*
 * oid comes in naming a commit and goes out naming the blob found at path,
 * which both callers rely on. The path has to be writable because a pathspec
 * item does not hold a const string.
 */
static int find_path_oid(struct object_id *oid, char *path, int file_only)
{
	struct commit *commit = lookup_commit_reference(the_repository, oid);
	// nowildcard_len matching len makes git treat the path as literal
	// rather than as a glob.
	struct pathspec_item item = {
		.match = path,
		.len = strlen(path),
		.nowildcard_len = strlen(path)
	};
	struct pathspec paths = {
		.nr = 1,
		.items = &item
	};
	struct walk_tree_context walk = {
		.match_path = path,
		.matched_oid = oid,
		.found_path = 0,
		.file_only = file_only
	};

	read_tree(the_repository, repo_get_commit_tree(the_repository, commit),
		  &paths, walk_tree, &walk);
	return walk.found_path;
}

/*
 * Callers ask before reading the object, because the point of the limit is to
 * keep a huge blob out of memory rather than to notice it once it is already
 * there.
 */
static int over_size_limit(unsigned long size)
{
	return ctx.cfg.max_blob_size &&
		size / 1024 > (unsigned long)ctx.cfg.max_blob_size;
}

int cgit_ref_path_exists(const char *path, const char *ref, int file_only)
{
	struct object_id oid;
	unsigned long size;
	char *path_copy = xstrdup(path);
	int found = 0;

	if (repo_get_oid(the_repository, ref, &oid))
		goto done;
	if (odb_read_object_info(the_repository->objects, &oid, &size) != OBJ_COMMIT)
		goto done;
	found = find_path_oid(&oid, path_copy, file_only);

done:
	free(path_copy);
	return found;
}

int cgit_print_file(char *path, const char *head, int file_only, int html_escape)
{
	struct object_id oid;
	enum object_type type;
	unsigned long size;
	char *buf;

	if (repo_get_oid(the_repository, head, &oid))
		return -1;
	type = odb_read_object_info(the_repository->objects, &oid, &size);
	if (type == OBJ_COMMIT) {
		if (!find_path_oid(&oid, path, file_only))
			return -1;
		type = odb_read_object_info(the_repository->objects, &oid, &size);
	}
	if (type == OBJ_BAD)
		return -1;
	if (over_size_limit(size))
		return -1;
	buf = odb_read_object(the_repository->objects, &oid, &type, &size);
	if (!buf)
		return -1;

	// html_txt wants a terminated string, and git leaves a spare byte
	// past every object it reads, so this write stays in the allocation.
	buf[size] = '\0';
	if (html_escape) {
		// html_txt stops at a NUL, so a blob holding one is written
		// segment by segment with a replacement character standing in
		// for each NUL byte, rather than silently cut short.
		const char *p = buf, *end = buf + size;

		while (p < end) {
			html_txt(p);
			p += strlen(p);
			while (p < end && !*p) {
				html("\xef\xbf\xbd");
				p++;
			}
		}
	} else {
		html_raw(buf, size);
	}
	free(buf);
	return 0;
}

void cgit_print_blob(const char *hex, char *path, const char *head, int file_only)
{
	struct object_id oid;
	enum object_type type;
	unsigned long size;
	char *buf;

	if (hex) {
		if (get_oid_hex(hex, &oid)) {
			cgit_print_error_page(400, "Bad request",
					"Bad hex value: %s", hex);
			return;
		}
	} else {
		if (repo_get_oid(the_repository, head, &oid)) {
			cgit_print_error_page(404, "Not found",
					"Bad ref: %s", head);
			return;
		}
	}

	type = odb_read_object_info(the_repository->objects, &oid, &size);

	if (!hex && type == OBJ_COMMIT && path) {
		if (!find_path_oid(&oid, path, file_only)) {
			cgit_print_error_page(404, "Not found",
					"Path not found: %s", path);
			return;
		}
		type = odb_read_object_info(the_repository->objects, &oid, &size);
	}

	if (type == OBJ_BAD) {
		cgit_print_error_page(404, "Not found",
				"Bad object name: %s", hex ? hex : path);
		return;
	}

	if (over_size_limit(size)) {
		cgit_print_error_page(413, "Too large",
				"Object size (%luKB) exceeds limit (%dKB)",
				size / 1024, ctx.cfg.max_blob_size);
		return;
	}

	buf = odb_read_object(the_repository->objects, &oid, &type, &size);
	if (!buf) {
		cgit_print_error_page(500, "Internal server error",
				"Error reading object %s", hex ? hex : path);
		return;
	}

	buf[size] = '\0';
	if (buffer_is_binary(buf, size))
		ctx.page.mimetype = "application/octet-stream";
	else
		ctx.page.mimetype = "text/plain";
	ctx.page.filename = path;

	// The bytes are whatever the repository holds, so the browser is told
	// not to guess a type of its own from them and not to load anything
	// they reference. Both 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");
	cgit_print_http_headers();
	html_raw(buf, size);
	free(buf);
}