// SPDX-License-Identifier: 0BSD

///////////////////////////////////////////////////////////////////////////////
//
/// \file       sandbox.c
/// \brief      Sandbox support
//
//  Author:     Lasse Collin
//
///////////////////////////////////////////////////////////////////////////////

#include "private.h"


#ifndef ENABLE_SANDBOX

// Prevent an empty translation unit when no sandboxing is supported.
typedef int dummy;

#else

/// If the conditions for strict sandboxing (described in main())
/// have been met, sandbox_allow_strict() can be called to set this
/// variable to true.
static bool strict_sandbox_allowed = false;


extern void
sandbox_allow_strict(void)
{
	strict_sandbox_allowed = true;
	return;
}


// Strict sandboxing prevents opening any files. This *tries* to ensure
// that any auxiliary files that might be required are already open.
//
// Returns true if strict sandboxing is allowed, false otherwise.
static bool
prepare_for_strict_sandbox(void)
{
	if (!strict_sandbox_allowed)
		return false;

	const char dummy_str[] = "x";

	// Try to ensure that both libc and xz locale files have been
	// loaded when NLS is enabled.
	snprintf(NULL, 0, "%s%s", _(dummy_str), strerror(EINVAL));

	// Try to ensure that iconv data files needed for handling multibyte
	// characters have been loaded. This is needed at least with glibc.
	tuklib_mbstr_width(dummy_str, NULL);

	return true;
}

#endif


#if defined(HAVE_PLEDGE)

///////////////
// pledge(2) //
///////////////

#include <unistd.h>


extern void
sandbox_init(void)
{
	if (pledge("stdio rpath wpath cpath fattr", "")) {
		// gettext hasn't been initialized yet so
		// there's no point to call it here.
		message_fatal("Failed to enable the sandbox");
	}

	return;
}


extern void
sandbox_enable_read_only(void)
{
	// We will be opening files for reading but
	// won't create or remove any files.
	if (pledge("stdio rpath", ""))
		message_fatal(_("Failed to enable the sandbox"));

	return;
}


extern void
sandbox_enable_strict_if_allowed(int src_fd lzma_attribute((__unused__)),
		int pipe_event_fd lzma_attribute((__unused__)),
		int pipe_write_fd lzma_attribute((__unused__)))
{
	if (!prepare_for_strict_sandbox())
		return;

	// All files that need to be opened have already been opened.
	if (pledge("stdio", ""))
		message_fatal(_("Failed to enable the sandbox"));

	return;
}


#elif defined(HAVE_LINUX_LANDLOCK)

//////////////
// Landlock //
//////////////

#include <linux/landlock.h>
#include <sys/syscall.h>
#include <sys/prctl.h>


// Highest Landlock ABI version supported by this file:
//   - For ABI versions 1-3 we don't need anything from <linux/landlock.h>
//     that isn't part of version 1.
//   - For ABI version 4 we need the larger struct landlock_ruleset_attr
//     with the handled_access_net member. That is bundled with the macros
//     LANDLOCK_ACCESS_NET_BIND_TCP and LANDLOCK_ACCESS_NET_CONNECT_TCP.
#ifdef LANDLOCK_ACCESS_NET_BIND_TCP
#	define LANDLOCK_ABI_MAX 4
#else
#	define LANDLOCK_ABI_MAX 3
#endif


/// Landlock ABI version supported by the kernel
static int landlock_abi;


// The required_rights should have those bits set that must not be restricted.
// This function will then bitwise-and ~required_rights with a mask matching
// the Landlock ABI version, leaving only those bits set that are supported
// by the ABI and allowed to be restricted by the function argument.
static void
enable_landlock(uint64_t required_rights)
{
	assert(landlock_abi <= LANDLOCK_ABI_MAX);

	if (landlock_abi <= 0)
		return;

	// We want to set all supported flags in handled_access_fs.
	// This way the ruleset will initially forbid access to all
	// actions that the available Landlock ABI version supports.
	// Exceptions can be added using landlock_add_rule(2) to
	// allow certain actions on certain files or directories.
	//
	// The same flag values are used on all archs. ABI v2 and v3
	// both add one new flag.
	//
	// First in ABI v1: LANDLOCK_ACCESS_FS_EXECUTE = 1ULL << 0
	// Last in ABI v1: LANDLOCK_ACCESS_FS_MAKE_SYM = 1ULL << 12
	// Last in ABI v2: LANDLOCK_ACCESS_FS_REFER = 1ULL << 13
	// Last in ABI v3: LANDLOCK_ACCESS_FS_TRUNCATE = 1ULL << 14
	//
	// This makes it simple to set the mask based on the ABI
	// version and we don't need to care which flags are #defined
	// in the installed <linux/landlock.h> for ABI versions 1-3.
	const struct landlock_ruleset_attr attr = {
		.handled_access_fs = ~required_rights
			& ((1ULL << (12 + my_min(3, landlock_abi))) - 1),
#if LANDLOCK_ABI_MAX >= 4
		.handled_access_net = landlock_abi < 4 ? 0 :
				(LANDLOCK_ACCESS_NET_BIND_TCP
				| LANDLOCK_ACCESS_NET_CONNECT_TCP),
#endif
	};

	const int ruleset_fd = syscall(SYS_landlock_create_ruleset,
			&attr, sizeof(attr), 0U);
	if (ruleset_fd < 0)
		message_fatal(_("Failed to enable the sandbox"));

	// All files we need should have already been opened. Thus,
	// we don't need to add any rules using landlock_add_rule(2)
	// before activating the sandbox.
	//
	// NOTE: It's possible that the hack prepare_for_strict_sandbox()
	// isn't be good enough. It tries to get translations and
	// libc-specific files loaded but if it's not good enough
	// then perhaps a Landlock rule to allow reading from /usr
	// and/or the xz installation prefix would be needed.
	//
	// prctl(PR_SET_NO_NEW_PRIVS, ...) was already called in
	// sandbox_init() so we don't do it here again.
	if (syscall(SYS_landlock_restrict_self, ruleset_fd, 0U) != 0)
		message_fatal(_("Failed to enable the sandbox"));

	return;
}


extern void
sandbox_init(void)
{
	// Prevent the process from gaining new privileges. This must be done
	// before landlock_restrict_self(2) but since we will never need new
	// privileges, this call can be done here already.
	//
	// This is supported since Linux 3.5. Ignore the return value to
	// keep compatibility with old kernels. landlock_restrict_self(2)
	// will fail if the no_new_privs attribute isn't set, thus if prctl()
	// fails here the error will still be detected when it matters.
	(void)prctl(PR_SET_NO_NEW_PRIVS, 1, 0, 0, 0);

	// Get the highest Landlock ABI version supported by the kernel.
	landlock_abi = syscall(SYS_landlock_create_ruleset,
			(void *)NULL, 0, LANDLOCK_CREATE_RULESET_VERSION);

	// The kernel might support a newer ABI than this file.
	if (landlock_abi > LANDLOCK_ABI_MAX)
		landlock_abi = LANDLOCK_ABI_MAX;

	// These are all in ABI version 1 already. We don't need truncate
	// rights because files are created with open() using O_EXCL and
	// without O_TRUNC.
	//
	// LANDLOCK_ACCESS_FS_READ_DIR is included here to get a clear error
	// message if xz is given a directory name. Without this permission
	// the message would be "Permission denied" but with this permission
	// it's "Is a directory, skipping". It could be worked around with
	// stat()/lstat() but just giving this permission is simpler and
	// shouldn't make the sandbox much weaker in practice.
	const uint64_t required_rights
			= LANDLOCK_ACCESS_FS_WRITE_FILE
			| LANDLOCK_ACCESS_FS_READ_FILE
			| LANDLOCK_ACCESS_FS_READ_DIR
			| LANDLOCK_ACCESS_FS_REMOVE_FILE
			| LANDLOCK_ACCESS_FS_MAKE_REG;

	enable_landlock(required_rights);
	return;
}


extern void
sandbox_enable_read_only(void)
{
	// We will be opening files for reading but
	// won't create or remove any files.
	const uint64_t required_rights
			= LANDLOCK_ACCESS_FS_READ_FILE
			| LANDLOCK_ACCESS_FS_READ_DIR;
	enable_landlock(required_rights);
	return;
}


extern void
sandbox_enable_strict_if_allowed(int src_fd lzma_attribute((__unused__)),
		int pipe_event_fd lzma_attribute((__unused__)),
		int pipe_write_fd lzma_attribute((__unused__)))
{
	if (!prepare_for_strict_sandbox())
		return;

	// Allow all restrictions that the kernel supports with the
	// highest Landlock ABI version that the kernel or xz supports.
	//
	// NOTE: LANDLOCK_ACCESS_FS_READ_DIR isn't needed here because
	// the only input file has already been opened.
	enable_landlock(0);
	return;
}


#elif defined(HAVE_CAP_RIGHTS_LIMIT)

//////////////
// Capsicum //
//////////////

#include <sys/capsicum.h>


extern void
sandbox_init(void)
{
	// Nothing to do.
	return;
}


extern void
sandbox_enable_read_only(void)
{
	// Nothing to do.
	return;
}


extern void
sandbox_enable_strict_if_allowed(
		int src_fd, int pipe_event_fd, int pipe_write_fd)
{
	if (!prepare_for_strict_sandbox())
		return;

	// Capsicum needs FreeBSD 10.2 or later.
	cap_rights_t rights;

	if (cap_enter())
		goto error;

	if (cap_rights_limit(src_fd, cap_rights_init(&rights,
			CAP_EVENT, CAP_FCNTL, CAP_LOOKUP, CAP_READ, CAP_SEEK)))
		goto error;

	// If not reading from stdin, remove all capabilities from it.
	if (src_fd != STDIN_FILENO && cap_rights_limit(
			STDIN_FILENO, cap_rights_clear(&rights)))
		goto error;

	if (cap_rights_limit(STDOUT_FILENO, cap_rights_init(&rights,
			CAP_EVENT, CAP_FCNTL, CAP_FSTAT, CAP_LOOKUP,
			CAP_WRITE, CAP_SEEK)))
		goto error;

	if (cap_rights_limit(STDERR_FILENO, cap_rights_init(&rights,
			CAP_WRITE)))
		goto error;

	if (cap_rights_limit(pipe_event_fd, cap_rights_init(&rights,
			CAP_EVENT)))
		goto error;

	if (cap_rights_limit(pipe_write_fd, cap_rights_init(&rights,
			CAP_WRITE)))
		goto error;

	return;

error:
	// If a kernel is configured without capability mode support or
	// used in an emulator that does not implement the capability
	// system calls, then the Capsicum system calls will fail and set
	// errno to ENOSYS. In that case xz will silently run without
	// the sandbox.
	if (errno == ENOSYS)
		return;

	message_fatal(_("Failed to enable the sandbox"));
}

#endif