Re: [PATCH RFC v9 00/25] pkeys-based page table hardening

On Thu, Sep 03, 2026 at 06:47:50PM +0200, Kevin Brodsky wrote:

On 01/09/2026 16:24, Linu Cherian wrote:

quoted

Hi Kevin,

On Tue, Aug 18, 2026 at 03:08:42PM +0100, Kevin Brodsky wrote:

quoted

[Sending during the merge window in case reviewers have spare
cycles; I'm not aiming to have this series merged in v7.3.]

This is a proposal to leverage protection keys (pkeys) to harden
critical kernel data, by making it mostly read-only. The series includes
a simple framework called "kpkeys" to manipulate pkeys for in-kernel use,
as well as a page table hardening feature based on that framework,
"kpkeys_hardened_pgtables". Both are implemented on arm64 as a proof of
concept, but they are designed to be compatible with any architecture
that supports pkeys.

The proposed approach is a typical use of pkeys: the data to protect is
mapped with a given pkey P, and the pkey register is initially
configured to grant read-only access to P. Where the protected data
needs to be written to, the pkey register is temporarily switched to
grant write access to P on the current CPU.

The key fact this approach relies on is that the target data is
only written to via a limited and well-defined API. This makes it
possible to explicitly switch the pkey register where needed, without
introducing excessively invasive changes, and only for a small amount of
trusted code.

Page tables are chosen as an initial target because of their especially
critical nature - a single write may result in arbitrary pages becoming
accessible to any context (including userspace). In order to keep the
series digestible for reviewers, this version focuses on functionality
rather than performance, making it most suitable as a debug feature. The
key trade-off is the requirement to PTE-map the linear map - see section
"Protected page table allocation" for details.

This series has similarities with the "PKS write protected page tables"
series posted by Rick Edgecombe a few years ago [1] but it is not
specific to x86/PKS - the approach is meant to be generic.

This proposal (as of RFC v5) was presented at Linux Security Summit
Europe 2025 [2].

[Table of contents]

* kpkeys
  - pkey register management

* kpkeys_hardened_pgtables
  - Protected page table allocation
  - kpkeys context switching
  - Performance
  - Limitations

* This series
  - Branches

* Threat model

* Further use-cases

* Open questions

kpkeys
======

The use of pkeys involves two separate mechanisms: assigning a pkey to
pages, and defining the pkeys -> permissions mapping via the pkey
register. This is implemented through the following interface:

- Pages are assigned a pkey in the linear map using set_memory_pkey().
  This is sufficient for this series, but it is also plausible for
  higher-level allocators to support marking allocations with a given
  pkey.

- The pkey register is configured based on a *kpkeys context*. kpkeys
  contexts are represented as simple integers that correspond to a given
  configuration, for instance:

  KPKEYS_CTX_DEFAULT:
        RW access to KPKEYS_PKEY_DEFAULT
        RO access to any other KPKEYS_PKEY_*

  KPKEYS_CTX_<FEAT>:
        RW access to KPKEYS_PKEY_DEFAULT
        RW access to KPKEYS_PKEY_<FEAT>
        RO access to any other KPKEYS_PKEY_*

  Only pkeys that are managed by the kpkeys framework are impacted;
  permissions for other pkeys are left unchanged (this allows for other
  schemes using pkeys to be used in parallel, and arch-specific use of
  certain pkeys).

- Adding some basic details on what a scheme and context is quite helpful.

 - Giving some hints (may be an example) on how multiple schemes and multiple contexts
  play together would be quite helpful.

"scheme" doesn't mean anything precise, it's only the notion that pkeys
that aren't reserved for kpkeys (i.e. anything but 0 or 1 in this
series) may be used for other purposes. Happy to reword if you have a
suggestion.

Got it. IMHO, adding two definitions towards the start would make it easier to follow.

kpkeys: Set of pkeys reserved and managed by the kpkeys framework.
        Pkeys outside this set are left untouched.

kpkeys context: A permission state that defines the permissions for each pkey owned by
 	kpkeys

Or something better.

"kpkeys context" is what is described above this paragraph, it's really
just a set of permissions for the managed pkeys. Transitioning between
context is described below.

Its clear to me now.

quoted

Adding a documentation that covers these aspects would be much
appreciated.

For sure, I am planning to have a documentation patch in a subsequent
version.

That would be great.

quoted

My understanding is that pkeys are being partitioned across different
contexts. But then the introduction of the term "scheme" looks bit confusing to me.

I wouldn't say pkeys are partitioned across contexts. Every context has
a set of permissions for all the pkeys managed by kpkeys. Any other pkey
is ignored (permissions left unchanged) by this framework.

Ack.

quoted

quoted

  The current kpkeys context is changed by calling
  kpkeys_enter_context(), which will set the pkey register
  accordingly and return the original state. A
  subsequent call to kpkeys_leave_context() restores the original
  state (and thus the original kpkeys context). The numeric value of
  KPKEYS_CTX_* (kpkeys context) is purely symbolic and thus generic,
  however each architecture is free to define non-default pkeys
  values (KPKEYS_PKEY_*).

..snip

quoted

Open questions
==============

A few aspects in this RFC that are debatable and/or worth discussing:

- There is currently no restriction on how kpkeys contexts map to pkeys
  permissions. A typical approach is to allocate one pkey per context and
  make it writable in that context only. As the number of contexts

Probably to avoid the assumption, may be we can we have something like
below 

For a pkey P, we could define
PKEY_P_PERM_CTXT_OTHERS	 //permission for pkey p in other contexts
PKEY_P_PERM_CTXT_SELF	 //permission for pkey p in self context

With the assumption of one pkey mapped for every context,
the permission for the default context would look something like,

PKEY_DEF_PERM_CTXT_SELF << PKEY_DEF_PKEY_SHIFT |
PKEY_CT0_PERM_CTXT_OTHERS << PKEY_CT0_PKEY_SHIFT | 
PKEY_CT1_PERM_CTXT_OTHERS << PKEY_CT1_PKEY_SHIFT |
...(for all valid contexts)

where,
Permission key, PKEY_DEF is associated with context DEFAULT,
Permission key, PKEY_CT0 is associated with context CT0,
Permission key, PKEY_CT1 is associated with context CT1

This adds assumptions rather than avoiding them. *Typically* when adding
a context you'd allocate a pkey that's only writable by this context,
but it doesn't have to be this way.

Okay agree. Then may be something like

Define permissions:

For default context,
KPKEYS_CTX_DEFAULT_PERM_PKEY_DEF
KPKEYS_CTX_DEFAULT_PERM_PKEY_CT0

For CT0 context,
KPKEYS_CTX_CT0_PERM_PKEY_DEF
KPKEYS_CTX_CT0_PERM_PKEY_CT0

Define POR_EL1:

For default context,
KPKEYS_POR_EL1_DEFAULT

For CT0 context,
KPKEYS_POR_EL1_CT0

Finally,
#define POR_EL1_INIT KPKEYS_POR_EL1_DEFAULT

Probably using something similar would make the idea of kpkeys context 
more evident in the code as well ?

The configuration space is more easily understood by considering the
other use-cases we've investigated (struct cred protection and eBPF
isolation, linked further down). For instance, for cred protection, we
had KPKEYS_LVL_UNRESTRICTED with write access to all pkeys, and for eBPF
isolation, we need a level that is less privileged and therefore does
*not* have write access to pkey 0.

quoted

quoted

  increases, we may however run out of pkeys, especially on arm64 (just
  8 pkeys with POE). Depending on the use-cases, it may be acceptable to
  use the same pkey for the data associated to multiple contexts.

Lets say two contexts A and B, use the same pkey P as their permission matches.
But then, when we enter context A, permission for pkey P gets
relaxed, then that would relax permission for pages associated with
context B as well which is unintended ?

That may be exactly what is intended, it all depends on the use-case. C1
may have a private pkey P1, and C2 P2, and then P3 that is shared by C1
and C2 (writable by both)

Got it. With each context defining permissions for each pkey owned by
kpkeys makes sense. 

Also do we need to assume that nesting of different contexts is not valid ?
For example,
Default context:
	enter CTX 0
		enter CTX 1
		leave CTX 1
	leave CTX 0
Default context:
  

quoted

As the hardware supports 16 pkeys, should we consider removing the limit
of 8 pkeys so that we can have unique pkeys for each context ?

FEAT_S1POE only supports 4-bit pkeys when using 128-bit page tables.

Ack.

--
Linu Cherian