chrootenv: rewrite on top of GLib

Changes:

* doesn't handle root user separately
* doesn't chdir("/") which makes using it seamless
* only bind mounts, doesn't symlink (i.e. files)

Incidentally, fixes #33106.

It's about two times shorter than the previous version, and much
easier to read/follow through. It uses GLib quite heavily, along with
RAII (available in GCC/Clang).
This commit is contained in:
Yegor Timoshenko 2017-12-28 05:49:35 +00:00
parent 25b35f4ffb
commit 4b1cf5afb8
No known key found for this signature in database
GPG Key ID: C34BF9DCC9DF8210
4 changed files with 160 additions and 247 deletions

View File

@ -1,238 +0,0 @@
#define _GNU_SOURCE
#include <errno.h>
#include <error.h>
#define errorf(status, fmt, ...) \
error_at_line(status, errno, __FILE__, __LINE__, fmt, ##__VA_ARGS__)
#include <dirent.h>
#include <ftw.h>
#include <sched.h>
#include <stdarg.h>
#include <stdbool.h>
#include <stdio.h>
#include <stdlib.h>
#include <string.h>
#include <sysexits.h>
#include <unistd.h>
#include <sys/mount.h>
#include <sys/stat.h>
#include <sys/wait.h>
#define LEN(x) (sizeof(x) / sizeof(*x))
// TODO: fill together with @abbradar when he gets better
const char *environ_blacklist[] = {};
void environ_blacklist_filter() {
for (size_t i = 0; i < LEN(environ_blacklist); i++) {
if (unsetenv(environ_blacklist[i]) < 0)
errorf(EX_OSERR, "unsetenv(%s)", environ_blacklist[i]);
}
}
void bind(const char *from, const char *to) {
if (mkdir(to, 0755) < 0)
errorf(EX_IOERR, "mkdir(%s)", to);
if (mount(from, to, "bind", MS_BIND | MS_REC, NULL) < 0)
errorf(EX_OSERR, "mount(%s, %s)", from, to);
}
const char *bind_blacklist[] = {".", "..", "bin", "etc", "host", "usr"};
bool str_contains(const char *needle, const char **haystack, size_t len) {
for (size_t i = 0; i < len; i++) {
if (!strcmp(needle, haystack[i]))
return true;
}
return false;
}
bool is_dir(const char *path) {
struct stat buf;
if (stat(path, &buf) < 0)
errorf(EX_IOERR, "stat(%s)", path);
return S_ISDIR(buf.st_mode);
}
void bind_to_cwd(const char *prefix) {
DIR *prefix_dir = opendir(prefix);
if (prefix_dir == NULL)
errorf(EX_IOERR, "opendir(%s)", prefix);
struct dirent *prefix_dirent;
while (prefix_dirent = readdir(prefix_dir)) {
if (str_contains(prefix_dirent->d_name, bind_blacklist,
LEN(bind_blacklist)))
continue;
char *prefix_dirent_path;
if (asprintf(&prefix_dirent_path, "%s%s", prefix, prefix_dirent->d_name) <
0)
errorf(EX_IOERR, "asprintf");
if (is_dir(prefix_dirent_path)) {
bind(prefix_dirent_path, prefix_dirent->d_name);
} else {
char *host_target;
if (asprintf(&host_target, "host/%s", prefix_dirent->d_name) < 0)
errorf(EX_IOERR, "asprintf");
if (symlink(host_target, prefix_dirent->d_name) < 0)
errorf(EX_IOERR, "symlink(%s, %s)", host_target, prefix_dirent->d_name);
free(host_target);
}
free(prefix_dirent_path);
}
bind(prefix, "host");
if (closedir(prefix_dir) < 0)
errorf(EX_IOERR, "closedir(%s)", prefix);
}
void spitf(const char *path, char *fmt, ...) {
va_list args;
va_start(args, fmt);
FILE *f = fopen(path, "w");
if (f == NULL)
errorf(EX_IOERR, "spitf(%s): fopen", path);
if (vfprintf(f, fmt, args) < 0)
errorf(EX_IOERR, "spitf(%s): vfprintf", path);
if (fclose(f) < 0)
errorf(EX_IOERR, "spitf(%s): fclose", path);
}
int nftw_remove(const char *path, const struct stat *sb, int type,
struct FTW *ftw) {
return remove(path);
}
#define REQUIREMENTS \
"Requires Linux version >= 3.19 built with CONFIG_USER_NS option.\n"
int main(int argc, char *argv[]) {
const char *self = *argv++;
if (argc < 2) {
fprintf(stderr, "Usage: %s command [arguments...]\n" REQUIREMENTS, self);
exit(EX_USAGE);
}
if (getenv("NIX_CHROOTENV") != NULL) {
fputs("Can't create chrootenv inside chrootenv!\n", stderr);
exit(EX_USAGE);
}
if (setenv("NIX_CHROOTENV", "1", false) < 0)
errorf(EX_OSERR, "setenv(NIX_CHROOTENV, 1)");
const char *temp = getenv("TMPDIR");
if (temp == NULL)
temp = "/tmp";
char *root;
if (asprintf(&root, "%s/chrootenvXXXXXX", temp) < 0)
errorf(EX_IOERR, "asprintf");
root = mkdtemp(root);
if (root == NULL)
errorf(EX_IOERR, "mkdtemp(%s)", root);
// Don't make root private so that privilege drops inside chroot are possible:
if (chmod(root, 0755) < 0)
errorf(EX_IOERR, "chmod(%s, 0755)", root);
pid_t cpid = fork();
if (cpid < 0)
errorf(EX_OSERR, "fork");
if (cpid == 0) {
uid_t uid = getuid();
gid_t gid = getgid();
// If we are root, no need to create new user namespace.
if (uid == 0) {
if (unshare(CLONE_NEWNS) < 0) {
fputs(REQUIREMENTS, stderr);
errorf(EX_OSERR, "unshare");
}
// Mark all mounted filesystems as slave so changes
// don't propagate to the parent mount namespace.
if (mount(NULL, "/", NULL, MS_REC | MS_SLAVE, NULL) < 0)
errorf(EX_OSERR, "mount");
} else {
// Create new mount and user namespaces. CLONE_NEWUSER
// requires a program to be non-threaded.
if (unshare(CLONE_NEWNS | CLONE_NEWUSER) < 0) {
fputs(access("/proc/sys/kernel/unprivileged_userns_clone", F_OK)
? REQUIREMENTS
: "Run: sudo sysctl -w kernel.unprivileged_userns_clone=1\n",
stderr);
errorf(EX_OSERR, "unshare");
}
// Map users and groups to the parent namespace.
// setgroups is only available since Linux 3.19:
spitf("/proc/self/setgroups", "deny");
spitf("/proc/self/uid_map", "%d %d 1", uid, uid);
spitf("/proc/self/gid_map", "%d %d 1", gid, gid);
}
if (chdir(root) < 0)
errorf(EX_IOERR, "chdir(%s)", root);
bind_to_cwd("/");
if (chroot(root) < 0)
errorf(EX_OSERR, "chroot(%s)", root);
if (chdir("/") < 0)
errorf(EX_IOERR, "chdir(/)");
environ_blacklist_filter();
if (execvp(*argv, argv) < 0)
errorf(EX_OSERR, "execvp(%s)", *argv);
}
int status;
if (waitpid(cpid, &status, 0) < 0)
errorf(EX_OSERR, "waitpid(%d)", cpid);
if (nftw(root, nftw_remove, getdtablesize(),
FTW_DEPTH | FTW_MOUNT | FTW_PHYS) < 0)
errorf(EX_IOERR, "nftw(%s)", root);
free(root);
if (WIFEXITED(status)) {
return WEXITSTATUS(status);
} else if (WIFSIGNALED(status)) {
kill(getpid(), WTERMSIG(status));
}
return EX_OSERR;
}

View File

@ -0,0 +1,139 @@
#define _GNU_SOURCE
#include <glib.h>
#include <glib/gstdio.h>
#include <errno.h>
#include <sched.h>
#include <unistd.h>
#define fail(s, err) g_error("%s: %s: %s", __func__, s, g_strerror(err))
#define fail_if(expr) \
if (expr) \
fail(#expr, errno);
#include <ftw.h>
#include <sys/mount.h>
#include <sys/stat.h>
#include <sys/types.h>
#include <sys/wait.h>
const gchar *bind_blacklist[] = {"bin", "etc", "host", "usr", NULL};
void bind_mount(const gchar *source, const gchar *target) {
fail_if(g_mkdir(target, 0755));
fail_if(mount(source, target, "bind", MS_BIND | MS_REC, NULL));
}
void bind_mount_host(const gchar *host, const gchar *guest) {
g_autofree gchar *point = g_build_filename(guest, "host", NULL);
bind_mount(host, point);
}
void bind_mount_item(const gchar *host, const gchar *guest, const gchar *name) {
g_autofree gchar *source = g_build_filename(host, name, NULL);
g_autofree gchar *target = g_build_filename(guest, name, NULL);
if (G_LIKELY(g_file_test(source, G_FILE_TEST_IS_DIR)))
bind_mount(source, target);
}
void bind(const gchar *host, const gchar *guest) {
g_autoptr(GError) err = NULL;
g_autoptr(GDir) dir = g_dir_open(host, 0, &err);
if (err != NULL)
fail("g_dir_open", errno);
const gchar *item;
while (item = g_dir_read_name(dir))
if (!g_strv_contains(bind_blacklist, item))
bind_mount_item(host, guest, item);
bind_mount_host(host, guest);
}
void spit(const char *path, char *fmt, ...) {
va_list args;
va_start(args, fmt);
FILE *f = g_fopen(path, "w");
if (f == NULL)
fail("g_fopen", errno);
g_vfprintf(f, fmt, args);
fclose(f);
}
int nftw_remove(const char *path, const struct stat *sb, int type,
struct FTW *ftw) {
return remove(path);
}
int main(gint argc, gchar **argv) {
const gchar *self = *argv++;
if (argc < 2) {
g_message("%s command [arguments...]", self);
return 1;
}
if (g_getenv("NIX_CHROOTENV"))
g_warning("chrootenv doesn't stack!");
else
g_setenv("NIX_CHROOTENV", "", TRUE);
g_autofree gchar *prefix =
g_build_filename(g_get_tmp_dir(), "chrootenvXXXXXX", NULL);
fail_if(!g_mkdtemp_full(prefix, 0755));
pid_t cpid = fork();
if (cpid < 0)
fail("fork", errno);
else if (cpid == 0) {
uid_t uid = getuid();
gid_t gid = getgid();
if (unshare(CLONE_NEWNS | CLONE_NEWUSER) < 0) {
int unshare_errno = errno;
g_message("Requires Linux version >= 3.19 built with CONFIG_USER_NS");
if (g_file_test("/proc/sys/kernel/unprivileged_userns_clone",
G_FILE_TEST_EXISTS))
g_message("Run: sudo sysctl -w kernel.unprivileged_userns_clone=1");
fail("unshare", unshare_errno);
}
spit("/proc/self/setgroups", "deny");
spit("/proc/self/uid_map", "%d %d 1", uid, uid);
spit("/proc/self/gid_map", "%d %d 1", gid, gid);
bind("/", prefix);
fail_if(chroot(prefix));
fail_if(execvp(*argv, argv));
}
else {
int status;
fail_if(waitpid(cpid, &status, 0) != cpid);
fail_if(nftw(prefix, nftw_remove, getdtablesize(),
FTW_DEPTH | FTW_MOUNT | FTW_PHYS));
if (WIFEXITED(status))
return WEXITSTATUS(status);
else if (WIFSIGNALED(status))
kill(getpid(), WTERMSIG(status));
return 1;
}
}

View File

@ -0,0 +1,19 @@
{ stdenv, pkgconfig, glib }:
stdenv.mkDerivation {
name = "chrootenv";
nativeBuildInputs = [ pkgconfig ];
buildInputs = [ glib ];
buildCommand = ''
cc ${./chrootenv.c} $(pkg-config --cflags --libs glib-2.0) -o $out
'';
meta = with stdenv.lib; {
description = "Setup mount/user namespace for FHS emulation";
license = licenses.free;
maintainers = with maintainers; [ yegortimoshenko ];
platforms = platforms.linux;
};
}

View File

@ -1,4 +1,4 @@
{ callPackage, runCommand, lib, writeScript, stdenv, coreutils, ruby }:
{ callPackage, runCommand, lib, writeScript, stdenv, coreutils }:
let buildFHSEnv = callPackage ./env.nix { }; in
@ -7,14 +7,7 @@ args@{ name, runScript ? "bash", extraInstallCommands ? "", meta ? {}, passthru
let
env = buildFHSEnv (removeAttrs args [ "runScript" "extraInstallCommands" "meta" "passthru" ]);
chrootenv = stdenv.mkDerivation {
name = "chrootenv";
unpackPhase = "cp ${./chrootenv.c} chrootenv.c";
installPhase = "cp chrootenv $out";
makeFlags = [ "chrootenv" ];
};
chrootenv = callPackage ./chrootenv {};
init = run: writeScript "${name}-init" ''
#! ${stdenv.shell}