|
[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index] Re: [PATCH] xen/ucode: Fix buffer under-run when parsing AMD containers
On Fri, Sep 13, 2024 at 12:09:07PM +0100, Andrew Cooper wrote:
> From: Demi Marie Obenour <demi@xxxxxxxxxxxxxxxxxxxxxx>
>
> The AMD container format has no formal spec. It is, at best, precision
> guesswork based on AMD's prior contributions to open source projects. The
> Equivalence Table has both an explicit length, and an expectation of having a
> NULL entry at the end.
>
> Xen was sanity checking the NULL entry, but without confirming that an entry
> was present, resulting in a read off the front of the buffer. With some
> manual debugging/annotations this manifests as:
>
> (XEN) *** Buf ffff83204c00b19c, eq ffff83204c00b194
> (XEN) *** eq: 0c 00 00 00 44 4d 41 00 00 00 00 00 00 00 00 00 aa aa aa aa
> ^-Actual buffer-------------------^
> (XEN) *** installed_cpu: 000c
> (XEN) microcode: Bad equivalent cpu table
> (XEN) Parsing microcode blob error -22
>
> When loaded by hypercall, the 4 bytes interpreted as installed_cpu happen to
> be the containing struct ucode_buf's len field, and luckily will be nonzero.
>
> When loaded at boot, it's possible for the access to #PF if the module happens
> to have been placed on a 2M boundary by the bootloader. Under Linux, it will
> commonly be the end of the CPIO header.
>
> Drop the probe of the NULL entry; Nothing else cares. A container without one
> is well formed, insofar that we can still parse it correctly. With this
> dropped, the same container results in:
>
> (XEN) microcode: couldn't find any matching ucode in the provided blob!
>
> Fixes: 4de936a38aa9 ("x86/ucode/amd: Rework parsing logic in
> cpu_request_microcode()")
> Signed-off-by: Demi Marie Obenour <demi@xxxxxxxxxxxxxxxxxxxxxx>
> Signed-off-by: Andrew Cooper <andrew.cooper3@xxxxxxxxxx>
> ---
> CC: Jan Beulich <JBeulich@xxxxxxxx>
> CC: Roger Pau Monné <roger.pau@xxxxxxxxxx>
> CC: Demi Marie Obenour <demi@xxxxxxxxxxxxxxxxxxxxxx>
>
> Split out of joint patch, and analyse.
>
> I couldn't trigger any of the sanitisers with this, hence the manual
> debugging.
>
> This patch intentionally doesn't include patch 2's extra hunk changing:
>
> @@ -364,7 +364,8 @@ static struct microcode_patch *cf_check
> cpu_request_microcode(
> if ( size < sizeof(*mc) ||
> (mc = buf)->type != UCODE_UCODE_TYPE ||
> size - sizeof(*mc) < mc->len ||
> - mc->len < sizeof(struct microcode_patch) )
> + mc->len < sizeof(struct microcode_patch) ||
> + mc->len % 4 != 0 )
> {
> printk(XENLOG_ERR "microcode: Bad microcode data\n");
> error = -EINVAL;
>
> Intel have a spec saying the length is mutliple of 4. AMD do not, and have
> microcode which genuinely isn't a multiple of 4.
In this case the structs at the top should be __attribute__((packed)) to
avoid undefined behavior. That can be a separate patch, though.
> ---
> xen/arch/x86/cpu/microcode/amd.c | 3 +--
> 1 file changed, 1 insertion(+), 2 deletions(-)
>
> diff --git a/xen/arch/x86/cpu/microcode/amd.c
> b/xen/arch/x86/cpu/microcode/amd.c
> index d2a26967c6db..32490c8b7d2a 100644
> --- a/xen/arch/x86/cpu/microcode/amd.c
> +++ b/xen/arch/x86/cpu/microcode/amd.c
> @@ -338,8 +338,7 @@ static struct microcode_patch *cf_check
> cpu_request_microcode(
> if ( size < sizeof(*et) ||
> (et = buf)->type != UCODE_EQUIV_CPU_TABLE_TYPE ||
> size - sizeof(*et) < et->len ||
> - et->len % sizeof(et->eq[0]) ||
> - et->eq[(et->len / sizeof(et->eq[0])) - 1].installed_cpu )
> + et->len % sizeof(et->eq[0]) )
> {
> printk(XENLOG_ERR "microcode: Bad equivalent cpu table\n");
> error = -EINVAL;
> --
> 2.39.2
>
Reviewed-by: Demi Marie Obenour <demi@xxxxxxxxxxxxxxxxxxxxxx>
--
Sincerely,
Demi Marie Obenour (she/her/hers)
Invisible Things Lab
Attachment:
signature.asc
|
![]() |
Lists.xenproject.org is hosted with RackSpace, monitoring our |