Re: [PATCH v6] tee: optee: Add SMC for loading OP-TEE image

[Sumit Garg]

On Mon, 13 Mar 2023 at 22:47, Jeffrey Kardatzke <jkardatzke@xxxxxxxxxxxx> wrote:
>
> On Mon, Mar 13, 2023 at 1:03 AM Sumit Garg <sumit.garg@xxxxxxxxxx> wrote:
> >
> > On Fri, 10 Mar 2023 at 01:13, Jeffrey Kardatzke <jkardatzke@xxxxxxxxxxxx> wrote:
> > >
> > > Adds an SMC call that will pass an OP-TEE binary image to EL3 and
> > > instruct it to load it as the BL32 payload. This works in conjunction
> > > with a feature added to Trusted Firmware for ARMv8 and above
> > > architectures that supports this.
> > >
> > > The main purpose of this change is to facilitate updating the OP-TEE
> > > component on devices via a rootfs change rather than having to do a
> > > firmware update. Further details are linked to in the Kconfig file.
> > >
> > > Signed-off-by: Jeffrey Kardatzke <jkardatzke@xxxxxxxxxxxx>
> > > Signed-off-by: Jeffrey Kardatzke <jkardatzke@xxxxxxxxxx>
> > > ---
> > >
> > > Changes in v6:
> > > - Expanded Kconfig documentation
> > >
> > > Changes in v5:
> > > - Renamed config option
> > > - Added runtime warning when config is used
> > >
> > > Changes in v4:
> > > - Update commit message
> > > - Added more documentation
> > > - Renamed config option, added ARM64 dependency
> > >
> > > Changes in v3:
> > > - Removed state tracking for driver reload
> > > - Check UID of service to verify it needs image load
> > >
> > > Changes in v2:
> > > - Fixed compile issue when feature is disabled
> > > - Addressed minor comments
> > > - Added state tracking for driver reload
> > >
> > >  drivers/tee/optee/Kconfig     | 29 +++++++++++
> > >  drivers/tee/optee/optee_msg.h | 12 +++++
> > >  drivers/tee/optee/optee_smc.h | 24 +++++++++
> > >  drivers/tee/optee/smc_abi.c   | 97 +++++++++++++++++++++++++++++++++++
> > >  4 files changed, 162 insertions(+)
> > >
> > > diff --git a/drivers/tee/optee/Kconfig b/drivers/tee/optee/Kconfig
> > > index f121c224e682..8d4836c58486 100644
> > > --- a/drivers/tee/optee/Kconfig
> > > +++ b/drivers/tee/optee/Kconfig
> > > @@ -7,3 +7,32 @@ config OPTEE
> > >         help
> > >           This implements the OP-TEE Trusted Execution Environment (TEE)
> > >           driver.
> > > +
> > > +config OPTEE_INSECURE_LOAD_IMAGE
> > > +       bool "Load OP-TEE image as firmware"
> > > +       default n
> > > +       depends on OPTEE && ARM64
> > > +       help
> > > +         This loads the BL32 image for OP-TEE as firmware when the driver is
> > > +         probed. This returns -EPROBE_DEFER until the firmware is loadable from
> > > +         the filesystem which is determined by checking the system_state until
> > > +         it is in SYSTEM_RUNNING. This also requires enabling the corresponding
> > > +         option in Trusted Firmware for Arm. The documentation there explains
> > > +         the security threat associated with enabling this as well as
> > > +         mitigations at the firmware and platform level.
> > > +         https://trustedfirmware-a.readthedocs.io/en/latest/threat_model/threat_model.html
> > > +
> > > +         When utilizing this option, the following mitigations should be
> > > +         implemented to prevent attacks at the kernel level.
> > > +         1. There must be boot chain security that verifies the kernel and
> > > +            rootfs, otherwise an attacker can modify the loaded OP-TEE binary.
> > > +         2. It is recommended to build it as an included driver rather than
> > > +            a module to prevent exploits that may cause the module to not be
> > > +            loaded.
> > > +         3. If there are alternate methods of booting the device, such as a
> > > +            recovery mode, it should be ensured that the same mitigations are
> > > +            applied in that mode.
> > > +         4. The OP-TEE driver must be loaded before any potential attack
> > > +            vectors are opened up. This should include mounting of any
> > > +            modifiable filesystems, opening of network ports or communicating
> > > +            with external devices (such a USB).
> >
> > This detailed threat model documentation belongs here [1] and it
> > should rather be in following format for every bullet point:
> >
> > Attack vector: <>
> > Mitigation: <>
> >
> > [1] https://docs.kernel.org/staging/tee.html?highlight=tee#op-tee-driver
> >
>
> Done in v7 patch set.
>
> > > diff --git a/drivers/tee/optee/optee_msg.h b/drivers/tee/optee/optee_msg.h
> > > index 70e9cc2ee96b..e8840a82b983 100644
> > > --- a/drivers/tee/optee/optee_msg.h
> > > +++ b/drivers/tee/optee/optee_msg.h
> > > @@ -241,11 +241,23 @@ struct optee_msg_arg {
> > >   * 384fb3e0-e7f8-11e3-af63-0002a5d5c51b.
> > >   * Represented in 4 32-bit words in OPTEE_MSG_UID_0, OPTEE_MSG_UID_1,
> > >   * OPTEE_MSG_UID_2, OPTEE_MSG_UID_3.
> > > + *
> > > + * In the case where the OP-TEE image is loaded by the kernel, this will
> > > + * initially return an alternate UID to reflect that we are communicating with
> > > + * the TF-A image loading service at that time instead of OP-TEE. That UID is:
> > > + * a3fbeab1-1246-315d-c7c4-06b9c03cbea4.
> > > + * Represented in 4 32-bit words in OPTEE_MSG_IMAGE_LOAD_UID_0,
> > > + * OPTEE_MSG_IMAGE_LOAD_UID_1, OPTEE_MSG_IMAGE_LOAD_UID_2,
> > > + * OPTEE_MSG_IMAGE_LOAD_UID_3.
> > >   */
> > >  #define OPTEE_MSG_UID_0                        0x384fb3e0
> > >  #define OPTEE_MSG_UID_1                        0xe7f811e3
> > >  #define OPTEE_MSG_UID_2                        0xaf630002
> > >  #define OPTEE_MSG_UID_3                        0xa5d5c51b
> > > +#define OPTEE_MSG_IMAGE_LOAD_UID_0     0xa3fbeab1
> > > +#define OPTEE_MSG_IMAGE_LOAD_UID_1     0x1246315d
> > > +#define OPTEE_MSG_IMAGE_LOAD_UID_2     0xc7c406b9
> > > +#define OPTEE_MSG_IMAGE_LOAD_UID_3     0xc03cbea4
> > >  #define OPTEE_MSG_FUNCID_CALLS_UID     0xFF01
> > >
> > >  /*
> > > diff --git a/drivers/tee/optee/optee_smc.h b/drivers/tee/optee/optee_smc.h
> > > index 73b5e7760d10..7d9fa426505b 100644
> > > --- a/drivers/tee/optee/optee_smc.h
> > > +++ b/drivers/tee/optee/optee_smc.h
> > > @@ -104,6 +104,30 @@ struct optee_smc_call_get_os_revision_result {
> > >         unsigned long reserved1;
> > >  };
> > >
> > > +/*
> > > + * Load Trusted OS from optee/tee.bin in the Linux firmware.
> > > + *
> > > + * WARNING: Use this cautiously as it could lead to insecure loading of the
> > > + * Trusted OS.
> > > + * This SMC instructs EL3 to load a binary and execute it as the Trusted OS.
> > > + *
> > > + * Call register usage:
> > > + * a0 SMC Function ID, OPTEE_SMC_CALL_LOAD_IMAGE
> > > + * a1 Upper 32bit of a 64bit size for the payload
> > > + * a2 Lower 32bit of a 64bit size for the payload
> > > + * a3 Upper 32bit of the physical address for the payload
> > > + * a4 Lower 32bit of the physical address for the payload
> > > + *
> > > + * The payload is in the OP-TEE image format.
> > > + *
> > > + * Returns result in a0, 0 on success and an error code otherwise.
> > > + */
> > > +#define OPTEE_SMC_FUNCID_LOAD_IMAGE 2
> > > +#define OPTEE_SMC_CALL_LOAD_IMAGE \
> > > +       ARM_SMCCC_CALL_VAL(ARM_SMCCC_FAST_CALL, ARM_SMCCC_SMC_32, \
> > > +                          ARM_SMCCC_OWNER_TRUSTED_OS_END, \
> > > +                          OPTEE_SMC_FUNCID_LOAD_IMAGE)
> > > +
> > >  /*
> > >   * Call with struct optee_msg_arg as argument
> > >   *
> > > diff --git a/drivers/tee/optee/smc_abi.c b/drivers/tee/optee/smc_abi.c
> > > index a1c1fa1a9c28..00b6b69b6f79 100644
> > > --- a/drivers/tee/optee/smc_abi.c
> > > +++ b/drivers/tee/optee/smc_abi.c
> > > @@ -8,9 +8,11 @@
> > >
> > >  #include <linux/arm-smccc.h>
> > >  #include <linux/errno.h>
> > > +#include <linux/firmware.h>
> > >  #include <linux/interrupt.h>
> > >  #include <linux/io.h>
> > >  #include <linux/irqdomain.h>
> > > +#include <linux/kernel.h>
> > >  #include <linux/mm.h>
> > >  #include <linux/module.h>
> > >  #include <linux/of.h>
> > > @@ -1149,6 +1151,22 @@ static bool optee_msg_api_uid_is_optee_api(optee_invoke_fn *invoke_fn)
> > >         return false;
> > >  }
> > >
> > > +#ifdef CONFIG_OPTEE_INSECURE_LOAD_IMAGE
> > > +static bool optee_msg_api_uid_is_optee_image_load(optee_invoke_fn *invoke_fn)
> > > +{
> > > +       struct arm_smccc_res res;
> > > +
> > > +       invoke_fn(OPTEE_SMC_CALLS_UID, 0, 0, 0, 0, 0, 0, 0, &res);
> > > +
> > > +       if (res.a0 == OPTEE_MSG_IMAGE_LOAD_UID_0 &&
> > > +          res.a1 == OPTEE_MSG_IMAGE_LOAD_UID_1 &&
> > > +          res.a2 == OPTEE_MSG_IMAGE_LOAD_UID_2 &&
> > > +          res.a3 == OPTEE_MSG_IMAGE_LOAD_UID_3)
> > > +               return true;
> > > +       return false;
> > > +}
> > > +#endif
> > > +
> > >  static void optee_msg_get_os_revision(optee_invoke_fn *invoke_fn)
> > >  {
> > >         union {
> > > @@ -1354,6 +1372,81 @@ static void optee_shutdown(struct platform_device *pdev)
> > >                 optee_disable_shm_cache(optee);
> > >  }
> > >
> > > +#ifdef CONFIG_OPTEE_INSECURE_LOAD_IMAGE
> > > +
> > > +#define OPTEE_FW_IMAGE "optee/tee.bin"
> > > +
> > > +static int optee_load_fw(struct platform_device *pdev,
> > > +                        optee_invoke_fn *invoke_fn)
> > > +{
> > > +       const struct firmware *fw = NULL;
> > > +       struct arm_smccc_res res;
> > > +       phys_addr_t data_pa;
> > > +       u8 *data_buf = NULL;
> > > +       u64 data_size;
> > > +       u32 data_pa_high, data_pa_low;
> > > +       u32 data_size_high, data_size_low;
> > > +       int rc;
> > > +
> > > +       if (!optee_msg_api_uid_is_optee_image_load(invoke_fn))
> > > +               return 0;
> > > +
> > > +       rc = request_firmware(&fw, OPTEE_FW_IMAGE, &pdev->dev);
> > > +       if (rc) {
> > > +               /*
> > > +                * The firmware in the rootfs will not be accessible until we
> > > +                * are in the SYSTEM_RUNNING state, so return EPROBE_DEFER until
> > > +                * that point.
> > > +                */
> > > +               if (system_state < SYSTEM_RUNNING)
> > > +                       return -EPROBE_DEFER;
> > > +               goto fw_err;
> > > +       }
> > > +
> > > +       data_size = fw->size;
> > > +       /*
> > > +        * This uses the GFP_DMA flag to ensure we are allocated memory in the
> > > +        * 32-bit space since TF-A cannot map memory beyond the 32-bit boundary.
> > > +        */
> > > +       data_buf = kmalloc(fw->size, GFP_KERNEL | GFP_DMA);
> > > +       if (!data_buf) {
> > > +               rc = -ENOMEM;
> > > +               goto fw_err;
> > > +       }
> > > +       memcpy(data_buf, fw->data, fw->size);
> > > +       data_pa = virt_to_phys(data_buf);
> > > +       reg_pair_from_64(&data_pa_high, &data_pa_low, data_pa);
> > > +       reg_pair_from_64(&data_size_high, &data_size_low, data_size);
> > > +       goto fw_load;
> > > +
> > > +fw_err:
> > > +       pr_warn("image loading failed\n");
> > > +       data_pa_high = data_pa_low = data_size_high = data_size_low = 0;
> > > +
> > > +fw_load:
> > > +       /*
> > > +        * Always invoke the SMC, even if loading the image fails, to indicate
> > > +        * to EL3 that we have passed the point where it should allow invoking
> > > +        * this SMC.
> > > +        */
> > > +       pr_warn("OP-TEE image loaded from kernel, this can be insecure");
> > > +       invoke_fn(OPTEE_SMC_CALL_LOAD_IMAGE, data_size_high, data_size_low,
> > > +                 data_pa_high, data_pa_low, 0, 0, 0, &res);
> >
> > Apart from the security considerations discussed, I see an issue with
> > the implementation here. Here you only initialize OP-TEE on *CPUX*
> > that is performing OP-TEE probe. IIRC, it is required for that CPUX to
> > be primary CPU0. How do we ensure that here?
>
> I'm not aware of any restrictions that require OP-TEE to be loaded on
> CPU0, do you have a point to something that indicates such a
> requirement? (it's always worked fine for me no matter what core it
> inits on)

You can grep in OP-TEE OS code base for keywords like "primary" and
"secondary". This will give you an idea how initialization differs
among primary and secondary CPUs. I would be interested to see the
OP-TEE init log (LOG_LEVEL = 4) when the primary CPU is *not* CPU0.
You can try to force that environment using taskset [1] while
installing the OP-TEE module as an out-of-tree build.

The one thing that essentially pops out to me is that, with the
default SPI interrupt routing model (routed to CPU0), OP-TEE won't be
able to handle the secure interrupts until it has been initialized on
CPU0.

[1] https://man7.org/linux/man-pages/man1/taskset.1.html

> >
> > Also, I observe from the TF-A patch that you are doing lazy OP-TEE
> > initialization for other secondary CPUs. IOW, if there is an OP-TEE
> > SMC call invoked for a particular CPU then only you invoke
> > opteed_cpu_on_finish_handler(0) once. This can be a bit unsafe as
> > OP-TEE hasn't setup its context for those CPU which may involve some
> > CPU specific security bits too such as:
> > - GIC CPU interface initialization, secure vs non-secure interrupts.
> > - Any platform and CPU specific TrustZone configuration.
>
> When opteed_cpu_on_finish_handler(0) is invoked...that invokes
> opteed_init_optee_ep_state(...) which then sets up the CPU context for
> that CPU. It then invokes opteed_synchronous_sp_entry for the
> cpu_on_entry handler which should then do any other type of platform
> specific initialization in the OP-TEE code itself. What setup in
> OP-TEE are you referring to that isn't going to be invoked when doing
> it the lazy way?

No, I am not worried about any missing setup but rather deferred
OP-TEE initialization for secondary CPUs after you open up the Linux
attack surface.

>
> >
> > I would have rather expected you to utilize cpuhp_setup_state() and
> > friends to initialize OP-TEE for secondary CPUs during boot instead
> > which is safe as per your platform threat model.
> >
> That could be another way to do it (I'm not familiar with that kernel
> code currently)..

AFAIK, there are many Linux kernel experts within Google who can guide
you through.

> but I'd rather stick with what I have unless there is
> something technically wrong with it since it's already been approved
> in TF-A.

We can very well amend the TF-A implementation with a followup patch.
Given above comments, the approach to initialize OP-TEE on all CPUs at
once during boot should be the correct approach.

-Sumit

>
> > -Sumit
> >
> > > +       if (!rc)
> > > +               rc = res.a0;
> > > +       if (fw)
> > > +               release_firmware(fw);
> > > +       kfree(data_buf);
> > > +
> > > +       return rc;
> > > +}
> > > +#else
> > > +static inline int optee_load_fw(struct platform_device *__unused1,
> > > +               optee_invoke_fn *__unused2) {
> > > +       return 0;
> > > +}
> > > +#endif
> > > +
> > >  static int optee_probe(struct platform_device *pdev)
> > >  {
> > >         optee_invoke_fn *invoke_fn;
> > > @@ -1372,6 +1465,10 @@ static int optee_probe(struct platform_device *pdev)
> > >         if (IS_ERR(invoke_fn))
> > >                 return PTR_ERR(invoke_fn);
> > >
> > > +       rc = optee_load_fw(pdev, invoke_fn);
> > > +       if (rc)
> > > +               return rc;
> > > +
> > >         if (!optee_msg_api_uid_is_optee_api(invoke_fn)) {
> > >                 pr_warn("api uid mismatch\n");
> > >                 return -EINVAL;
> > > --
> > > 2.40.0.rc1.284.g88254d51c5-goog
> > >