From: Stefan Hajnoczi Date: Tue, 6 Oct 2020 10:58:26 +0100 Subject: [PATCH] virtiofsd: avoid /proc/self/fd tempdir In order to prevent /proc/self/fd escapes a temporary directory is created where /proc/self/fd is bind-mounted. This doesn't work on read-only file systems. Avoid the temporary directory by bind-mounting /proc/self/fd over /proc. This does not affect other processes since we remounted / with MS_REC | MS_SLAVE. /proc must exist and virtiofsd does not use it so it's safe to do this. Path traversal can be tested with the following function: static void test_proc_fd_escape(struct lo_data *lo) { int fd; int level = 0; ino_t last_ino = 0; fd = lo->proc_self_fd; for (;;) { struct stat st; if (fstat(fd, &st) != 0) { perror("fstat"); return; } if (last_ino && st.st_ino == last_ino) { fprintf(stderr, "inode number unchanged, stopping\n"); return; } last_ino = st.st_ino; fprintf(stderr, "Level %d dev %lu ino %lu\n", level, (unsigned long)st.st_dev, (unsigned long)last_ino); fd = openat(fd, "..", O_PATH | O_DIRECTORY | O_NOFOLLOW); level++; } } Before and after this patch only Level 0 is displayed. Without /proc/self/fd bind-mount protection it is possible to traverse parent directories. Fixes: 397ae982f4df4 ("virtiofsd: jail lo->proc_self_fd") Cc: Miklos Szeredi Cc: Jens Freimann Signed-off-by: Stefan Hajnoczi Message-Id: <20201006095826.59813-1-stefanha@redhat.com> Reviewed-by: Dr. David Alan Gilbert Tested-by: Jens Freimann Reviewed-by: Jens Freimann Signed-off-by: Dr. David Alan Gilbert --- tools/virtiofsd/passthrough_ll.c | 34 +++++++++++--------------------- 1 file changed, 11 insertions(+), 23 deletions(-) diff --git a/tools/virtiofsd/passthrough_ll.c b/tools/virtiofsd/passthrough_ll.c index 50a164a599..b865dc564b 100644 --- a/tools/virtiofsd/passthrough_ll.c +++ b/tools/virtiofsd/passthrough_ll.c @@ -2386,8 +2386,6 @@ static void setup_wait_parent_capabilities(void) static void setup_namespaces(struct lo_data *lo, struct fuse_session *se) { pid_t child; - char template[] = "virtiofsd-XXXXXX"; - char *tmpdir; /* * Create a new pid namespace for *child* processes. We'll have to @@ -2451,33 +2449,23 @@ static void setup_namespaces(struct lo_data *lo, struct fuse_session *se) exit(1); } - tmpdir = mkdtemp(template); - if (!tmpdir) { - fuse_log(FUSE_LOG_ERR, "tmpdir(%s): %m\n", template); - exit(1); - } - - if (mount("/proc/self/fd", tmpdir, NULL, MS_BIND, NULL) < 0) { - fuse_log(FUSE_LOG_ERR, "mount(/proc/self/fd, %s, MS_BIND): %m\n", - tmpdir); + /* + * We only need /proc/self/fd. Prevent ".." from accessing parent + * directories of /proc/self/fd by bind-mounting it over /proc. Since / was + * previously remounted with MS_REC | MS_SLAVE this mount change only + * affects our process. + */ + if (mount("/proc/self/fd", "/proc", NULL, MS_BIND, NULL) < 0) { + fuse_log(FUSE_LOG_ERR, "mount(/proc/self/fd, MS_BIND): %m\n"); exit(1); } - /* Now we can get our /proc/self/fd directory file descriptor */ - lo->proc_self_fd = open(tmpdir, O_PATH); + /* Get the /proc (actually /proc/self/fd, see above) file descriptor */ + lo->proc_self_fd = open("/proc", O_PATH); if (lo->proc_self_fd == -1) { - fuse_log(FUSE_LOG_ERR, "open(%s, O_PATH): %m\n", tmpdir); + fuse_log(FUSE_LOG_ERR, "open(/proc, O_PATH): %m\n"); exit(1); } - - if (umount2(tmpdir, MNT_DETACH) < 0) { - fuse_log(FUSE_LOG_ERR, "umount2(%s, MNT_DETACH): %m\n", tmpdir); - exit(1); - } - - if (rmdir(tmpdir) < 0) { - fuse_log(FUSE_LOG_ERR, "rmdir(%s): %m\n", tmpdir); - } } /*