fsopen() records the caller's user namespace in fc->user_ns and hands back an ordinary file descriptor. Nothing ties the task that calls fsconfig(FSCONFIG_CMD_CREATE) to the task that created the context. The fd is inherited across fork() and exec() and it can be passed over a unix socket. Completing a context from another user namespace is allowed on purpose. vfs_cmd_create() authorizes the create with mount_capable(), which for FS_USERNS_MOUNT checks ns_capable(fc->user_ns, CAP_SYS_ADMIN), and that succeeds for a task holding CAP_SYS_ADMIN in an ancestor of fc->user_ns. So an unprivileged task can reach the WARN_ON() in ovl_fill_super(): create a user and a mount namespace in a child, call fsopen("overlay") there, send the fscontext fd to the parent and let the parent issue FSCONFIG_CMD_CREATE. Both namespaces come from a plain unshare(1) and no capability is needed anywhere: WARNING: fs/overlayfs/super.c:1551 at ovl_fill_super+0x7b9/0x1e20 [overlay] CPU: 3 UID: 1000 PID: 3243376 Comm: fswarn Call Trace: get_tree_nodev+0x71/0xa0 ovl_get_tree+0x15/0x20 [overlay] vfs_get_tree+0x2a/0x100 vfs_cmd_create+0x60/0xf0 __do_sys_fsconfig+0x4b2/0x500 The child needs the mount namespace because fsopen() itself gates on may_mount(), which asks for CAP_SYS_ADMIN in the user namespace owning the caller's mount namespace. fsconfig() doesn't repeat that check. It is a WARN_ON() and not a WARN_ON_ONCE(), so the condition can be raised in a loop to taint the kernel and flood the log, and it panics a kernel booted with panic_on_warn. Keep refusing the mount and stop warning about it. ovl_parse_param() already spells a user namespace check this way for Opt_override_creds. Fixes: 1784fbc2ed9c ("ovl: port to new mount api") Cc: stable@vger.kernel.org # v6.5+ Signed-off-by: Christian Brauner (Amutable) --- fs/overlayfs/super.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/fs/overlayfs/super.c b/fs/overlayfs/super.c index 60f0b7ceef0a..60b808b85fc4 100644 --- a/fs/overlayfs/super.c +++ b/fs/overlayfs/super.c @@ -1544,7 +1544,8 @@ int ovl_fill_super(struct super_block *sb, struct fs_context *fc) int err; err = -EIO; - if (WARN_ON(fc->user_ns != current_user_ns())) + /* The fscontext fd may have been passed to another user namespace. */ + if (fc->user_ns != current_user_ns()) goto out_err; ovl_set_d_op(sb); -- 2.53.0 fsopen() records the caller's user namespace in fc->user_ns and hands back an ordinary file descriptor. Nothing ties the task that calls fsconfig(FSCONFIG_CMD_CREATE) to the task that created the context. The fd is inherited across fork() and exec() and it can be passed over a unix socket. Completing a context from another user namespace is allowed on purpose. vfs_cmd_create() authorizes the create with mount_capable(), which for FS_USERNS_MOUNT checks ns_capable(fc->user_ns, CAP_SYS_ADMIN), and that succeeds for a task holding CAP_SYS_ADMIN in an ancestor of fc->user_ns. So an unprivileged task can reach the WARN_ON() in bm_fill_super(): create a user and a mount namespace in a child, call fsopen("binfmt_misc") there, send the fscontext fd to the parent and let the parent issue FSCONFIG_CMD_CREATE. Both namespaces come from a plain unshare(1) and no capability is needed anywhere: WARNING: fs/binfmt_misc.c:938 at bm_fill_super+0xa2/0xc0 [binfmt_misc] CPU: 15 UID: 1000 PID: 3243382 Comm: fswarn Call Trace: get_tree_keyed+0x7d/0xb0 bm_get_tree+0x34/0x90 [binfmt_misc] vfs_get_tree+0x2a/0x100 vfs_cmd_create+0x60/0xf0 __do_sys_fsconfig+0x4b2/0x500 The child needs the mount namespace because fsopen() itself gates on may_mount(), which asks for CAP_SYS_ADMIN in the user namespace owning the caller's mount namespace. fsconfig() doesn't repeat that check. It is a WARN_ON() and not a WARN_ON_ONCE(), so the condition can be raised in a loop to taint the kernel and flood the log, and it panics a kernel booted with panic_on_warn. Keep refusing the mount and stop warning about it. Nothing in bm_fill_super() depends on the two namespaces matching, it derives everything from sb->s_user_ns. Fixes: 21ca59b365c0 ("binfmt_misc: enable sandboxed mounts") Cc: stable@vger.kernel.org # v6.7+ Signed-off-by: Christian Brauner (Amutable) --- fs/binfmt_misc.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/fs/binfmt_misc.c b/fs/binfmt_misc.c index c97f10b48b5b..613dd28e3f1a 100644 --- a/fs/binfmt_misc.c +++ b/fs/binfmt_misc.c @@ -937,7 +937,8 @@ static int bm_fill_super(struct super_block *sb, struct fs_context *fc) /* last one */ {""} }; - if (WARN_ON(user_ns != current_user_ns())) + /* The fscontext fd may have been passed to another user namespace. */ + if (user_ns != current_user_ns()) return -EINVAL; /* Never exec off this instance and never let anything stack on it. */ -- 2.53.0 fsopen() records the caller's user namespace in fc->user_ns and hands back an ordinary file descriptor, so the task that issues FSCONFIG_CMD_CREATE need not be the one that created the context. mount_capable() authorizes that for a caller holding CAP_SYS_ADMIN in an ancestor of fc->user_ns, which any unprivileged user has over a user namespace it just created. binfmt_misc and overlayfs used to WARN_ON() the mismatch. Add a test for both. Also cover the handover within one user namespace. That is a supported thing to do and has to keep working. Signed-off-by: Christian Brauner (Amutable) --- tools/testing/selftests/Makefile | 1 + .../selftests/filesystems/fscontext_ns/Makefile | 10 + .../filesystems/fscontext_ns/fscontext_ns_test.c | 239 +++++++++++++++++++++ 3 files changed, 250 insertions(+) diff --git a/tools/testing/selftests/Makefile b/tools/testing/selftests/Makefile index 8d4db2241cc2..b622052ec3e9 100644 --- a/tools/testing/selftests/Makefile +++ b/tools/testing/selftests/Makefile @@ -42,6 +42,7 @@ TARGETS += filesystems/fuse TARGETS += filesystems/move_mount TARGETS += filesystems/empty_mntns TARGETS += filesystems/fsmount_ns +TARGETS += filesystems/fscontext_ns TARGETS += firmware TARGETS += fpu TARGETS += ftrace diff --git a/tools/testing/selftests/filesystems/fscontext_ns/Makefile b/tools/testing/selftests/filesystems/fscontext_ns/Makefile new file mode 100644 index 000000000000..7e3506294757 --- /dev/null +++ b/tools/testing/selftests/filesystems/fscontext_ns/Makefile @@ -0,0 +1,10 @@ +# SPDX-License-Identifier: GPL-2.0 +TEST_GEN_PROGS := fscontext_ns_test + +CFLAGS += -Wall -O0 -g $(KHDR_INCLUDES) $(TOOLS_INCLUDES) +LDLIBS := -lcap + +include ../../lib.mk + +$(OUTPUT)/fscontext_ns_test: fscontext_ns_test.c ../utils.c + $(CC) $(CFLAGS) -o $@ $^ $(LDLIBS) diff --git a/tools/testing/selftests/filesystems/fscontext_ns/fscontext_ns_test.c b/tools/testing/selftests/filesystems/fscontext_ns/fscontext_ns_test.c new file mode 100644 index 000000000000..0f30a3e6e197 --- /dev/null +++ b/tools/testing/selftests/filesystems/fscontext_ns/fscontext_ns_test.c @@ -0,0 +1,239 @@ +// SPDX-License-Identifier: GPL-2.0 +/* + * Copyright (C) 2026 Christian Brauner + * + * Test that completing a filesystem context from another user namespace + * doesn't warn. + * + * fsopen() records the caller's user namespace in fc->user_ns and hands + * back an ordinary file descriptor. The task that issues + * FSCONFIG_CMD_CREATE need not be the one that created the context: the fd + * is inherited across fork() and exec() and it can be passed over a unix + * socket. vfs_cmd_create() authorizes the create with mount_capable(), + * which for FS_USERNS_MOUNT checks ns_capable(fc->user_ns, CAP_SYS_ADMIN), + * and that succeeds for a task holding CAP_SYS_ADMIN in an ancestor of + * fc->user_ns. + * + * binfmt_misc and overlayfs used to WARN_ON() that mismatch, which let an + * unprivileged user taint the kernel, flood the log and panic a kernel + * booted with panic_on_warn. The mount must still be refused, but it must + * not warn. + */ +#define _GNU_SOURCE + +#include +#include +#include +#include +#include +#include +#include +#include + +#include "../wrappers.h" +#include "../utils.h" +#include "../../kselftest_harness.h" + +#ifndef FSCONFIG_CMD_CREATE +#define FSCONFIG_CMD_CREATE 6 +#endif + +/* TAINT_WARN, i.e. bit 9 of /proc/sys/kernel/tainted. */ +#define TAINT_WARN_BIT 9 + +static bool taint_warn_set(void) +{ + unsigned long taint = 0; + FILE *f; + + f = fopen("/proc/sys/kernel/tainted", "r"); + if (!f) + return false; + if (fscanf(f, "%lu", &taint) != 1) + taint = 0; + fclose(f); + + return taint & (1UL << TAINT_WARN_BIT); +} + +static int send_fd(int sock, int fd) +{ + char cmsgbuf[CMSG_SPACE(sizeof(int))] = {}; + char b[1] = { 'x' }; + struct iovec iov = { .iov_base = b, .iov_len = sizeof(b) }; + struct msghdr msg = { + .msg_iov = &iov, + .msg_iovlen = 1, + .msg_control = cmsgbuf, + .msg_controllen = sizeof(cmsgbuf), + }; + struct cmsghdr *cmsg; + + cmsg = CMSG_FIRSTHDR(&msg); + cmsg->cmsg_level = SOL_SOCKET; + cmsg->cmsg_type = SCM_RIGHTS; + cmsg->cmsg_len = CMSG_LEN(sizeof(int)); + memcpy(CMSG_DATA(cmsg), &fd, sizeof(int)); + + return sendmsg(sock, &msg, 0) < 0 ? -1 : 0; +} + +static int recv_fd(int sock) +{ + char cmsgbuf[CMSG_SPACE(sizeof(int))] = {}; + char b[1]; + struct iovec iov = { .iov_base = b, .iov_len = sizeof(b) }; + struct msghdr msg = { + .msg_iov = &iov, + .msg_iovlen = 1, + .msg_control = cmsgbuf, + .msg_controllen = sizeof(cmsgbuf), + }; + struct cmsghdr *cmsg; + int fd = -1; + + if (recvmsg(sock, &msg, 0) <= 0) + return -1; + + cmsg = CMSG_FIRSTHDR(&msg); + if (!cmsg || cmsg->cmsg_type != SCM_RIGHTS) + return -1; + memcpy(&fd, CMSG_DATA(cmsg), sizeof(int)); + + return fd; +} + +/* + * Create a context for @fsname in a child and complete it here. With @nest + * the child first creates its own user namespace, so that the context is + * created in a descendant of the namespace completing it. The child needs a + * mount namespace of its own as well: fsopen() gates on may_mount(), which + * asks for CAP_SYS_ADMIN in the user namespace owning the caller's mount + * namespace. + * + * Returns the result of FSCONFIG_CMD_CREATE with errno set, or -ENODATA if + * the child could not create the context at all. + */ +static int create_from_child(const char *fsname, bool nest) +{ + int sock[2], fd, ret, status; + pid_t pid; + + if (socketpair(AF_UNIX, SOCK_STREAM, 0, sock)) + return -ENODATA; + + pid = fork(); + if (pid < 0) { + close(sock[0]); + close(sock[1]); + return -ENODATA; + } + + if (pid == 0) { + close(sock[0]); + + if (nest && unshare(CLONE_NEWUSER | CLONE_NEWNS)) + _exit(1); + + fd = sys_fsopen(fsname, 0); + if (fd < 0) + _exit(1); + if (send_fd(sock[1], fd)) + _exit(1); + _exit(0); + } + + close(sock[1]); + fd = recv_fd(sock[0]); + close(sock[0]); + wait_for_pid(pid); + waitpid(pid, &status, WNOHANG); + + if (fd < 0) + return -ENODATA; + + errno = 0; + ret = sys_fsconfig(fd, FSCONFIG_CMD_CREATE, NULL, NULL, 0); + status = errno; + close(fd); + errno = status; + + return ret; +} + +FIXTURE(fscontext_ns) { + bool warn_before; +}; + +FIXTURE_SETUP(fscontext_ns) +{ + self->warn_before = taint_warn_set(); + + if (setup_userns() != 0) + SKIP(return, "setup_userns failed"); +} + +FIXTURE_TEARDOWN(fscontext_ns) +{ +} + +/* + * The condition the kernel used to WARN about. It has to be refused, and it + * has to be refused quietly: an unprivileged task reaches this. + */ +FIXTURE_VARIANT(fscontext_ns) { + const char *fsname; + int expected_errno; +}; + +FIXTURE_VARIANT_ADD(fscontext_ns, binfmt_misc) { + .fsname = "binfmt_misc", + .expected_errno = EINVAL, +}; + +FIXTURE_VARIANT_ADD(fscontext_ns, overlay) { + .fsname = "overlay", + .expected_errno = EIO, +}; + +TEST_F(fscontext_ns, create_from_descendant_userns) +{ + int ret; + + ret = create_from_child(variant->fsname, true); + if (ret == -ENODATA) + SKIP(return, "%s unavailable", variant->fsname); + + ASSERT_EQ(-1, ret); + ASSERT_EQ(variant->expected_errno, errno); + + /* + * Only meaningful if nothing had warned before us. Note that an + * unrelated warning racing this test would look like a failure. + */ + if (self->warn_before) + TH_LOG("TAINT_WARN already set, not checking for a new warning"); + else + ASSERT_FALSE(taint_warn_set()); +} + +/* + * The same handover within one user namespace is a supported thing to do and + * has to keep working. binfmt_misc takes no options, so the create succeeds + * outright and this also shows the test really drives the create path. + */ +TEST(create_from_same_userns) +{ + int ret; + + if (setup_userns() != 0) + SKIP(return, "setup_userns failed"); + + ret = create_from_child("binfmt_misc", false); + if (ret == -ENODATA) + SKIP(return, "binfmt_misc unavailable"); + + ASSERT_EQ(0, ret); +} + +TEST_HARNESS_MAIN -- 2.53.0