Gentle poke about this patch, which I haven't seen any feedback for.
It's causing us a problem in terms of having a string of kexecs result
in the kernel no longer being able to talk to the TPM with an HMAC
session, and the rng random data fetches from the TPM then failing.
On Fri, Mar 07, 2025 at 12:25:23PM +0000, Jonathan McDowell wrote:
From: Jonathan McDowell <noodles@xxxxxxxx>
Lazy flushing of TPM auth sessions can interact badly with IMA + kexec,
resulting in loaded session handles being leaked across the kexec and
not cleaned up. Fix by ensuring any active auth session is ended before
the TPM is told about the shutdown, matching what is done when
suspending.
Before:
root@debian-qemu-efi:~# tpm2_getcap handles-loaded-session
root@debian-qemu-efi:~# tpm2_getcap handles-saved-session
root@debian-qemu-efi:~# kexec --load --kexec-file-syscall …
root@debian-qemu-efi:~# systemctl kexec
…
root@debian-qemu-efi:~# tpm2_getcap handles-loaded-session
- 0x2000000
root@debian-qemu-efi:~# tpm2_getcap handles-saved-session
root@debian-qemu-efi:~#
(repeat kexec steps)
root@debian-qemu-efi:~# tpm2_getcap handles-loaded-session
- 0x2000000
- 0x2000001
root@debian-qemu-efi:~# tpm2_getcap handles-saved-session
root@debian-qemu-efi:~#
After:
root@debian-qemu-efi:~# tpm2_getcap handles-loaded-session
root@debian-qemu-efi:~# tpm2_getcap handles-saved-session
root@debian-qemu-efi:~# kexec --load --kexec-file-syscall …
root@debian-qemu-efi:~# systemctl kexec
…
root@debian-qemu-efi:~# tpm2_getcap handles-loaded-session
root@debian-qemu-efi:~# tpm2_getcap handles-saved-session
root@debian-qemu-efi:~#
Signed-off-by: Jonathan McDowell <noodles@xxxxxxxx>
---
drivers/char/tpm/tpm-chip.c | 1 +
1 file changed, 1 insertion(+)
diff --git a/drivers/char/tpm/tpm-chip.c b/drivers/char/tpm/tpm-chip.c
index 7df7abaf3e52..87f01269b9b5 100644
--- a/drivers/char/tpm/tpm-chip.c
+++ b/drivers/char/tpm/tpm-chip.c
@@ -300,6 +300,7 @@ int tpm_class_shutdown(struct device *dev)
down_write(&chip->ops_sem);
if (chip->flags & TPM_CHIP_FLAG_TPM2) {
if (!tpm_chip_start(chip)) {
+ tpm2_end_auth_session(chip);
tpm2_shutdown(chip, TPM2_SU_CLEAR);
tpm_chip_stop(chip);
}
--
2.48.1
J.
--
I've got a trigger inside.
