]> git.ipfire.org Git - thirdparty/kernel/linux.git/commit
make 'user_access_begin()' do 'access_ok()'
authorLinus Torvalds <torvalds@linux-foundation.org>
Fri, 4 Jan 2019 20:56:09 +0000 (12:56 -0800)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 4 Jan 2019 20:56:09 +0000 (12:56 -0800)
commit594cc251fdd0d231d342d88b2fdff4bc42fb0690
tree259269a399e6504a7cf8739201cf172d1cbb197a
parent0b2c8f8b6b0c7530e2866c95862546d0da2057b0
make 'user_access_begin()' do 'access_ok()'

Originally, the rule used to be that you'd have to do access_ok()
separately, and then user_access_begin() before actually doing the
direct (optimized) user access.

But experience has shown that people then decide not to do access_ok()
at all, and instead rely on it being implied by other operations or
similar.  Which makes it very hard to verify that the access has
actually been range-checked.

If you use the unsafe direct user accesses, hardware features (either
SMAP - Supervisor Mode Access Protection - on x86, or PAN - Privileged
Access Never - on ARM) do force you to use user_access_begin().  But
nothing really forces the range check.

By putting the range check into user_access_begin(), we actually force
people to do the right thing (tm), and the range check vill be visible
near the actual accesses.  We have way too long a history of people
trying to avoid them.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
arch/x86/include/asm/uaccess.h
drivers/gpu/drm/i915/i915_gem_execbuffer.c
include/linux/uaccess.h
kernel/compat.c
kernel/exit.c
lib/strncpy_from_user.c
lib/strnlen_user.c