tpm_lite: Set O_CLOEXEC when opening TPM device.stabilize-10682.B

This sets O_CLOEXEC when opening the TPM device to make sure the file
descriptor isn't shared across processes. The TPM character device
exposes the raw communication channel to send/receive commands to/from
the TPM. The TPM is not designed for concurrent access by multiple
users and the kernel driver already returns EBUSY on open when a
different process has already opened it. Consequently, it only makes
sense to have the /dev/tpm0 file descriptor be closed automatically on
exec().

None of the callers I'm aware of need to share the TPM file descriptor
across processes, and mount-encrypted has some ad-hoc code to close the
descriptor when it does fork+exec to spawn a helper. The existing code
isn't consistent and comprehensive (mount-encrypted spawns other
helpers where it forgets to close the file descriptor), so the plan is
to set O_CLOEXEC and remove the ad-hoc code.

BRANCH=None
BUG=None
TEST=Compiles, passes tests, image boots.

Change-Id: Ia6e73fb12e8f2ed8fe99b4c53ea6eb8cda4a21f5
Reviewed-on: https://chromium-review.googlesource.com/1055569
Commit-Ready: Mattias Nissler <mnissler@chromium.org>
Tested-by: Mattias Nissler <mnissler@chromium.org>
Reviewed-by: Andrey Pronin <apronin@chromium.org>

1 files changed, 1 insertions, 1 deletions

diff --git a/firmware/stub/tpm_lite_stub.c b/firmware/stub/tpm_lite_stub.c
index 4c846312..004eeaf5 100644
--- a/firmware/stub/tpm_lite_stub.c
+++ b/firmware/stub/tpm_lite_stub.c
@@ -172,7 +172,7 @@ VbError_t VbExTpmOpen(void)
 	/* Retry TPM opens on EBUSY failures. */
 	for (retries = 0; retries < OPEN_RETRY_MAX_NUM; ++ retries) {
 		errno = 0;
-		tpm_fd = open(device_path, O_RDWR);
+		tpm_fd = open(device_path, O_RDWR | O_CLOEXEC);
 		saved_errno = errno;
 		if (tpm_fd >= 0)
 			return VBERROR_SUCCESS;