On Wed, Nov 29, 2017 at 03:21:14PM +0000, Maciej W. Rozycki wrote:
> Fix a commit d614fd58a283 ("mips/ptrace: Preserve previous registers for
> short regset write") bug and allow the last register requested with a
> ptrace(2) PTRACE_SETREGSET call to be partially written if supplied this
> way by the caller, like with other register sets.
>
> Cc: stable@xxxxxxxxxxxxxxx # v4.11+
> Fixes: d614fd58a283 ("mips/ptrace: Preserve previous registers for short regset write")
> Signed-off-by: Maciej W. Rozycki <macro@xxxxxxxx>
> ---
> arch/mips/kernel/ptrace.c | 2 +-
> 1 file changed, 1 insertion(+), 1 deletion(-)
>
> linux-mips-nt-prfpreg-count.diff
> Index: linux-sfr-test/arch/mips/kernel/ptrace.c
> ===================================================================
> --- linux-sfr-test.orig/arch/mips/kernel/ptrace.c 2017-11-21 22:12:00.000000000 +0000
> +++ linux-sfr-test/arch/mips/kernel/ptrace.c 2017-11-21 22:13:13.471970000 +0000
> @@ -484,7 +484,7 @@ static int fpr_set_msa(struct task_struc
> int err;
>
> BUILD_BUG_ON(sizeof(fpr_val) != sizeof(elf_fpreg_t));
> - for (i = 0; i < NUM_FPU_REGS && *count >= sizeof(elf_fpreg_t); i++) {
> + for (i = 0; i < NUM_FPU_REGS && *count > 0; i++) {
>
> err = user_regset_copyin(pos, count, kbuf, ubuf,
> &fpr_val, i * sizeof(elf_fpreg_t),
> (i + 1) * sizeof(elf_fpreg_t));
But mips*_regsets[REGSET_FPR].size == sizeof(elf_fpreg_t),
linux/kernel/regset.c:ptrace_regset() polices
iov_len % regset->size == 0, and each user_regset_copyout() call here
transfers sizeof(elf_fpreg_t) bytes, decrementing *count by that
amount unless something goest wrong in which case we return.
So how do we end up with *count > 0 && *count < sizeof(elf_fpreg_t)
here?
If we can't end up with that, then this patch doesn't change ABI-
observable behaviour, unless I've missed something.
If we can end up with that somehow, then this patch reintroduces the
issue d614fd58a283 aims to fix, whereby fpr_val can contain
uninitialised kernel stack which userspace can then obtain via
PTRACE_GETREGSET.
Cheers
---Dave