[PATCH bpf-next 0/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[PATCH bpf-next 0/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[Matt Bobrowski]

bpf_path_d_path() can run on a file released from exit_task_work(),
after exit_fs() has cleared the task's fs, as a BPF LSM program attached
to file_release does whenever a task exits holding the last reference to
an open file. d_path() then dereferences the NULL fs.

Patch 1 returns -EINVAL from bpf_path_d_path() when current->fs is NULL.
Patch 2 adds a selftest that crashes the kernel without it.

Matt Bobrowski (2):
  bpf: fix crash in bpf_path_d_path() for exiting tasks
  selftests/bpf: test bpf_path_d_path() on a file released at task exit

 fs/bpf_fs_kfuncs.c                            |  8 +++
 .../selftests/bpf/prog_tests/fs_kfuncs.c      | 51 +++++++++++++++++++
 .../selftests/bpf/progs/test_path_d_path.c    | 35 +++++++++++++
 3 files changed, 94 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/test_path_d_path.c


base-commit: 0007654a7bf13ef2a19afd398bbc6507087f19fd
-- 
2.47.3

[PATCH bpf-next 1/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[Matt Bobrowski]

bpf_path_d_path() passes its path to d_path(), which resolves it
against current->fs through get_fs_root_rcu() without checking that
the calling task still has one. An exiting task does not. do_exit()
calls exit_fs(), which clears tsk->fs, before exit_task_work(), and a
file whose last reference the task drops in exit_files() has its final
__fput() deferred to that task work.

A BPF LSM program attached to security_file_release, which may pass
file->f_path to bpf_path_d_path() as a trusted argument, dereferences a
NULL fs whenever a task exits with a file still open:

  Call trace:
   lock_acquire+0x8c/0x218
   d_path+0x270/0x378
   bpf_path_d_path+0x28/0x80
   bpf_prog_9801cf2b982f05b2_file_release+0x1d0/0xc60
   bpf_trampoline_6442515074+0x7c/0x100
   bpf_lsm_file_release+0x8/0x20
   security_file_release+0x58/0x90
   __fput+0xcc/0x320
   ____fput+0x1c/0x30
   task_work_run+0x88/0x100
   do_exit+0x210/0x3e8
   do_group_exit+0x3c/0xa0
   __arm64_sys_exit_group+0x20/0x30

With lockdep enabled the fault is taken in lock_acquire(), from the
seqcount read in get_fs_root_rcu(); without it, on the read of
fs->seq itself.

Fix this by returning -EINVAL from bpf_path_d_path() when current->fs
is NULL. Only the task itself clears its fs, in exit_fs(), so the check
cannot race with that.

Fixes: d08e2045ebf0 ("bpf: introduce new VFS based BPF kfuncs")
Cc: stable@vger.kernel.org
Signed-off-by: Matt Bobrowski <matt@bobrowski.net>
---
 fs/bpf_fs_kfuncs.c | 8 ++++++++
 1 file changed, 8 insertions(+)

diff --git a/fs/bpf_fs_kfuncs.c b/fs/bpf_fs_kfuncs.c
index 08f0847c4970..6665a78218e1 100644
--- a/fs/bpf_fs_kfuncs.c
+++ b/fs/bpf_fs_kfuncs.c
@@ -87,6 +87,14 @@ __bpf_kfunc int bpf_path_d_path(const struct path *path, char *buf, size_t buf__
 	if (!buf__sz)
 		return -EINVAL;
 
+	/*
+	 * An exiting task releases its last files from exit_task_work(), after
+	 * exit_fs() has already dropped its ->fs, which d_path() uses to
+	 * resolve against.
+	 */
+	if (!current->fs)
+		return -EINVAL;
+
 	ret = d_path(path, buf, buf__sz);
 	if (IS_ERR(ret))
 		return PTR_ERR(ret);
-- 
2.47.3

[PATCH bpf-next 2/2] selftests/bpf: test bpf_path_d_path() on a file released at task exit

[Matt Bobrowski]

Add a fs_kfuncs subtest for the crash fixed by the previous commit. A
child opens a file and exits without closing it, so exit_files() drops
the last reference and the file is released from exit_task_work(),
after exit_fs() has cleared the task's fs. A file_release program calls
bpf_path_d_path() on that release, matched by the file's inode number,
and records what it returned.

Without the fix the subtest crashes the kernel in d_path(). With it,
bpf_path_d_path() returns -EINVAL.

Fixes: d08e2045ebf0 ("bpf: introduce new VFS based BPF kfuncs")
Cc: stable@vger.kernel.org
Signed-off-by: Matt Bobrowski <matt@bobrowski.net>
---
 .../selftests/bpf/prog_tests/fs_kfuncs.c      | 51 +++++++++++++++++++
 .../selftests/bpf/progs/test_path_d_path.c    | 35 +++++++++++++
 2 files changed, 86 insertions(+)
 create mode 100644 tools/testing/selftests/bpf/progs/test_path_d_path.c

diff --git a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
index 43a26ec69a8e..510cf11c8bd8 100644
--- a/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
+++ b/tools/testing/selftests/bpf/prog_tests/fs_kfuncs.c
@@ -4,12 +4,15 @@
 #include <stdlib.h>
 #include <sys/types.h>
 #include <sys/xattr.h>
+#include <sys/stat.h>
+#include <sys/wait.h>
 #include <linux/fsverity.h>
 #include <unistd.h>
 #include <test_progs.h>
 #include "test_get_xattr.skel.h"
 #include "test_set_remove_xattr.skel.h"
 #include "test_fsverity.skel.h"
+#include "test_path_d_path.skel.h"
 
 static const char testfile[] = "/tmp/test_progs_fs_kfuncs";
 
@@ -268,6 +271,51 @@ static void test_fsverity(void)
 	remove(testfile);
 }
 
+static void test_path_d_path_on_exit(void)
+{
+	struct test_path_d_path *skel = NULL;
+	struct stat st;
+	int fd, status;
+	pid_t pid;
+
+	fd = open(testfile, O_CREAT | O_RDONLY, 0644);
+	if (!ASSERT_GE(fd, 0, "create_file"))
+		return;
+	close(fd);
+
+	if (!ASSERT_OK(stat(testfile, &st), "stat"))
+		goto out;
+
+	skel = test_path_d_path__open_and_load();
+	if (!ASSERT_OK_PTR(skel, "test_path_d_path__open_and_load"))
+		goto out;
+
+	skel->bss->monitored_ino = st.st_ino;
+
+	if (!ASSERT_OK(test_path_d_path__attach(skel), "test_path_d_path__attach"))
+		goto out;
+
+	/*
+	 * The child exits with testfile still open, so exit_files() drops
+	 * the last reference and the file is released from exit_task_work(),
+	 * after exit_fs().
+	 */
+	pid = fork();
+	if (!ASSERT_GE(pid, 0, "fork"))
+		goto out;
+	if (pid == 0)
+		_exit(open(testfile, O_RDONLY) < 0);
+
+	if (!ASSERT_EQ(waitpid(pid, &status, 0), pid, "waitpid"))
+		goto out;
+	ASSERT_TRUE(WIFEXITED(status) && !WEXITSTATUS(status), "child_open");
+	ASSERT_EQ(skel->bss->called, 1, "called");
+	ASSERT_EQ(skel->bss->ret, -EINVAL, "bpf_path_d_path");
+out:
+	test_path_d_path__destroy(skel);
+	remove(testfile);
+}
+
 void test_fs_kfuncs(void)
 {
 	/* Matches xattr_names in progs/test_get_xattr.c */
@@ -288,4 +336,7 @@ void test_fs_kfuncs(void)
 
 	if (test__start_subtest("fsverity"))
 		test_fsverity();
+
+	if (test__start_subtest("path_d_path_on_exit"))
+		test_path_d_path_on_exit();
 }
diff --git a/tools/testing/selftests/bpf/progs/test_path_d_path.c b/tools/testing/selftests/bpf/progs/test_path_d_path.c
new file mode 100644
index 000000000000..4ea50a067819
--- /dev/null
+++ b/tools/testing/selftests/bpf/progs/test_path_d_path.c
@@ -0,0 +1,35 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include "vmlinux.h"
+#include <errno.h>
+#include <bpf/bpf_helpers.h>
+#include <bpf/bpf_tracing.h>
+#include "bpf_misc.h"
+#include "bpf_experimental.h"
+
+#define PF_EXITING 0x00000004
+
+char _license[] SEC("license") = "GPL";
+
+__u64 monitored_ino;
+int called;
+int ret;
+
+/*
+ * A task that exits with a file still open releases it from
+ * exit_task_work(), after exit_fs() has already cleared its fs.
+ */
+SEC("lsm/file_release")
+int BPF_PROG(path_d_path_on_exit, struct file *file)
+{
+	struct task_struct *task = bpf_get_current_task_btf();
+	char buf[64];
+
+	if (file->f_inode->i_ino != monitored_ino ||
+	    !(task->flags & PF_EXITING))
+		return 0;
+
+	ret = bpf_path_d_path(&file->f_path, buf, sizeof(buf));
+	called = 1;
+	return 0;
+}
-- 
2.47.3

Re: [PATCH bpf-next 1/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[Song Liu]

On Tue, Oct 20, 2026 at 3:50 AM Matt Bobrowski <matt@bobrowski.net> wrote:
>
[...]
>
> With lockdep enabled the fault is taken in lock_acquire(), from the
> seqcount read in get_fs_root_rcu(); without it, on the read of
> fs->seq itself.
>
> Fix this by returning -EINVAL from bpf_path_d_path() when current->fs
> is NULL. Only the task itself clears its fs, in exit_fs(), so the check
> cannot race with that.
>
> Fixes: d08e2045ebf0 ("bpf: introduce new VFS based BPF kfuncs")
> Cc: stable@vger.kernel.org
> Signed-off-by: Matt Bobrowski <matt@bobrowski.net>

Acked-by: Song Liu <song@kernel.org>

Re: [PATCH bpf-next 2/2] selftests/bpf: test bpf_path_d_path() on a file released at task exit

[Song Liu]

On Tue, Oct 20, 2026 at 3:50 AM Matt Bobrowski <matt@bobrowski.net> wrote:
>
> Add a fs_kfuncs subtest for the crash fixed by the previous commit. A
> child opens a file and exits without closing it, so exit_files() drops
> the last reference and the file is released from exit_task_work(),
> after exit_fs() has cleared the task's fs. A file_release program calls
> bpf_path_d_path() on that release, matched by the file's inode number,
> and records what it returned.
>
> Without the fix the subtest crashes the kernel in d_path(). With it,
> bpf_path_d_path() returns -EINVAL.
>
> Fixes: d08e2045ebf0 ("bpf: introduce new VFS based BPF kfuncs")
> Cc: stable@vger.kernel.org
> Signed-off-by: Matt Bobrowski <matt@bobrowski.net>

Acked-by: Song Liu <song@kernel.org>

Re: [PATCH bpf-next 1/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[Alexei Starovoitov]

On Tue, Oct 20, 2026 at 03:50 AM Matt Bobrowski <matt@bobrowski.net> wrote:
> Fix this by returning -EINVAL from bpf_path_d_path() when current->fs
> is NULL. Only the task itself clears its fs, in exit_fs(), so the check
> cannot race with that.

The crash is gone, but so is the path. file_release progs will get
-EINVAL for every file that a task still has open when it exits.

It also breaks what works today. The d_dname branch in d_path() returns
before it looks at current->fs, so at task exit a pipe or a socket
resolves to "pipe:[123]" now and to -EINVAL with this patch.

And it's not specific to bpf. The comment in fs/cachefiles/namei.c says
that current->fs == NULL in do_exit() breaks d_path() called from ext4
open.

can we fix it in d_path() instead ?
With an empty root prepend_path() goes up to the global root, the same
way d_absolute_path() does. Then the check in bpf_path_d_path() is not
needed and the selftest can expect the path. A chrooted task gets the
path from the global root, but it doesn't have a root anymore.

Something like this (compile tested only):

diff --git a/fs/d_path.c b/fs/d_path.c
index c253090..ea49d33 100644
--- a/fs/d_path.c
+++ b/fs/d_path.c
@@ -265,7 +265,7 @@ static void get_fs_root_rcu(struct fs_struct *fs, struct path *root)
 char *d_path(const struct path *path, char *buf, int buflen)
 {
 	DECLARE_BUFFER(b, buf, buflen);
-	struct path root;
+	struct path root = {};

 	/*
 	 * We have various synthetic filesystems that never get mounted.  On
@@ -284,7 +284,9 @@ char *d_path(const struct path *path, char *buf, int buflen)
 		return path->dentry->d_op->d_dname(path->dentry, buf, buflen);

 	rcu_read_lock();
-	get_fs_root_rcu(current->fs, &root);
+	/* Past exit_fs() there is no root to stop at, go up to the global one. */
+	if (current->fs)
+		get_fs_root_rcu(current->fs, &root);
 	if (unlikely(d_unlinked(path->dentry)))
 		prepend(&b, " (deleted)", 11);
 	else

pw-bot: cr

Re: [PATCH bpf-next 1/2] bpf: fix crash in bpf_path_d_path() for exiting tasks

[Al Viro]

On Tue, Oct 20, 2026 at 06:24:42PM +0000, Alexei Starovoitov wrote:

> And it's not specific to bpf. The comment in fs/cachefiles/namei.c says
> that current->fs == NULL in do_exit() breaks d_path() called from ext4
> open.
> 
> can we fix it in d_path() instead ?

Once more, with feeling: if you want a pathname relative to your chroot,
use d_path(); otherwise, use d_absolute_path().  If you don't know what
the difference is, just use "I don't give a fuck" and save the cycles
(and avoid potential infoleaks, while we are at it).

Folks, it's _not_ just about an oops - specific callers need to decide
which string do they want.