Shared pages should never return to the page allocator, or future usage of
the pages may allow for the contents to be exposed to the host. They may
also cause the guest to crash if the page is used in way disallowed by HW
(i.e. for executable code or as a page table).
Normally set_memory() call failures are rare. But on TDX
set_memory_XXcrypted() involves calls to the untrusted VMM, and an attacker
could fail these calls such that:
1. set_memory_encrypted() returns an error and leaves the pages fully
shared.
2. set_memory_decrypted() returns an error, but the pages are actually
full converted to shared.
This means that patterns like the below can cause problems:
void *addr = alloc();
int fail = set_memory_decrypted(addr, 1);
if (fail)
free_pages(addr, 0);
And:
void *addr = alloc();
int fail = set_memory_decrypted(addr, 1);
if (fail) {
set_memory_encrypted(addr, 1);
free_pages(addr, 0);
}
Unfortunately these patterns are all over the place. And what the
set_memory() callers should do in this situation is not clear either. They
shouldn’t use them as shared because something clearly went wrong, but
they also need to fully reset the pages to private to free them. But, the
kernel needs the VMMs help to do this and the VMM is already being
uncooperative around the needed operations. So this isn't guaranteed to
succeed and the caller is kind of stuck with unusable pages.
Looking at QEMU/KVM as an example, these VMM converstion failures either
indicates an attempt to attack the guest, or resource constraints on the
host. Preventing a DOS attack is out of scope for the coco threat model.
So this leaves the host resource constraint cause. When similar resource
constraints are encountered in the host, KVM punts the problem to
userspace and QEMU terminates the guest. When similar problems are
detected inside set_memory(), SEV issues a command to terminate the guest.
This all makes it appealing to simply panic (via tdx_panic() call
which informs the host what is happening) when observing troublesome VMM
behavior around the memory conversion. It is:
- Consistent with similar behavior on SEV side.
- Generally more consistent with how host resource constraints are handled
(at least in QEMU/KVM)
- Would be a more foolproof defense against the attack scenario.
Never-the-less, doing so would be an instance of the “crash the kernel for
security reasons” pattern. This is a big reason, and crashing is not fully
needed because the unusable pages could just be leaked (as they already
are in some cases). So instead, this series does a tree-wide search and
fixes the callers to handle the error by leaking the pages. Going forward
callers will need to handle the set_memory() errors correctly in order to
not reintroduce the issue.
I think there are some points for both sides, and we had some internal
discussion on the right way to handle it. So I've tried to characterize
both arguments. I'm interested to hear opinions on which is the best.
I’ve marked the hyperv guest parts in this as RFC, both because I can’t
test them and I believe Linux TDs can’t run on hyperv yet due to some
missing support. I would appreciate a correction on this if it’s wrong.
Rick Edgecombe (10):
mm: Add helper for freeing decrypted memory
x86/mm/cpa: Reject incorrect encryption change requests
kvmclock: Use free_decrypted_pages()
swiotlb: Use free_decrypted_pages()
ptp: Use free_decrypted_pages()
dma: Use free_decrypted_pages()
hv: Use free_decrypted_pages()
hv: Track decrypted status in vmbus_gpadl
hv_nstvsc: Don't free decrypted memory
uio_hv_generic: Don't free decrypted memory
arch/s390/include/asm/set_memory.h | 1 +
arch/x86/kernel/kvmclock.c | 2 +-
arch/x86/mm/pat/set_memory.c | 41 +++++++++++++++++++++++++++++-
drivers/hv/channel.c | 18 ++++++++-----
drivers/hv/connection.c | 13 +++++++---
drivers/net/hyperv/netvsc.c | 7 +++--
drivers/ptp/ptp_kvm_x86.c | 2 +-
drivers/uio/uio_hv_generic.c | 12 ++++++---
include/linux/dma-map-ops.h | 3 ++-
include/linux/hyperv.h | 1 +
include/linux/set_memory.h | 13 ++++++++++
kernel/dma/contiguous.c | 2 +-
kernel/dma/swiotlb.c | 11 +++++---
13 files changed, 101 insertions(+), 25 deletions(-)
--
2.34.1
