[PATCH v2 5/5] Documentation: document uaccess logging

[Peter Collingbourne]

On Mon, Nov 22, 2021 at 11:46 PM Dmitry Vyukov <dvyukov at google.com> wrote:
>
> On Tue, 23 Nov 2021 at 06:17, Peter Collingbourne <pcc at google.com> wrote:
> >
> > Add documentation for the uaccess logging feature.
> >
> > Link: https://linux-review.googlesource.com/id/Ia626c0ca91bc0a3d8067d7f28406aa40693b65a2
> > Signed-off-by: Peter Collingbourne <pcc at google.com>
> > ---
> >  Documentation/admin-guide/index.rst           |   1 +
> >  Documentation/admin-guide/uaccess-logging.rst | 149 ++++++++++++++++++
> >  2 files changed, 150 insertions(+)
> >  create mode 100644 Documentation/admin-guide/uaccess-logging.rst
> >
> > diff --git a/Documentation/admin-guide/index.rst b/Documentation/admin-guide/index.rst
> > index 1bedab498104..4f6ee447ab2f 100644
> > --- a/Documentation/admin-guide/index.rst
> > +++ b/Documentation/admin-guide/index.rst
> > @@ -54,6 +54,7 @@ ABI will be found here.
> >     :maxdepth: 1
> >
> >     sysfs-rules
> > +   uaccess-logging
> >
> >  The rest of this manual consists of various unordered guides on how to
> >  configure specific aspects of kernel behavior to your liking.
> > diff --git a/Documentation/admin-guide/uaccess-logging.rst b/Documentation/admin-guide/uaccess-logging.rst
> > new file mode 100644
> > index 000000000000..4b2b297afc00
> > --- /dev/null
> > +++ b/Documentation/admin-guide/uaccess-logging.rst
> > @@ -0,0 +1,149 @@
> > +.. SPDX-License-Identifier: GPL-2.0
> > +
> > +===============
> > +Uaccess Logging
> > +===============
> > +
> > +Background
> > +----------
> > +
> > +Userspace tools such as sanitizers (ASan, MSan, HWASan) and tools
> > +making use of the ARM Memory Tagging Extension (MTE) need to
> > +monitor all memory accesses in a program so that they can detect
> > +memory errors. Furthermore, fuzzing tools such as syzkaller need to
> > +monitor all memory accesses so that they know which parts of memory
> > +to fuzz. For accesses made purely in userspace, this is achieved
> > +via compiler instrumentation, or for MTE, via direct hardware
> > +support. However, accesses made by the kernel on behalf of the user
> > +program via syscalls (i.e. uaccesses) are normally invisible to
> > +these tools.
> > +
> > +Traditionally, the sanitizers have handled this by interposing the libc
> > +syscall stubs with a wrapper that checks the memory based on what we
> > +believe the uaccesses will be. However, this creates a maintenance
> > +burden: each syscall must be annotated with its uaccesses in order
> > +to be recognized by the sanitizer, and these annotations must be
> > +continuously updated as the kernel changes.
> > +
> > +The kernel's uaccess logging feature provides userspace tools with
> > +the address and size of each userspace access, thereby allowing these
> > +tools to report memory errors involving these accesses without needing
> > +annotations for every syscall.
> > +
> > +By relying on the kernel's actual uaccesses, rather than a
> > +reimplementation of them, the userspace memory safety tools may
> > +play a dual role of verifying the validity of kernel accesses. Even
> > +a sanitizer whose syscall wrappers have complete knowledge of the
> > +kernel's intended API may vary from the kernel's actual uaccesses due
> > +to kernel bugs. A sanitizer with knowledge of the kernel's actual
> > +uaccesses may produce more accurate error reports that reveal such
> > +bugs. For example, a kernel that accesses more memory than expected
> > +by the userspace program could indicate that either userspace or the
> > +kernel has the wrong idea about which kernel functionality is being
> > +requested -- either way, there is a bug.
> > +
> > +Interface
> > +---------
> > +
> > +The feature may be used via the following prctl:
> > +
> > +.. code-block:: c
> > +
> > +  uint64_t addr = 0; /* Generally will be a TLS slot or equivalent */
> > +  prctl(PR_SET_UACCESS_DESCRIPTOR_ADDR_ADDR, &addr, 0, 0, 0);
> > +
> > +Supplying a non-zero address as the second argument to ``prctl``
>
> Is it possible to unregister it? Is it what happens when 0 is passed
> as addr? If so, please describe.
> It may be handy to do one-off tracing with the address on stack.

Yes, done in v3.

> > +will cause the kernel to read an address (referred to as the *uaccess
> > +descriptor address*) from that address on each kernel entry.
> > +
> > +When entering the kernel with a non-zero uaccess descriptor address
> > +to handle a syscall, the kernel will read a data structure of type
> > +``struct uaccess_descriptor`` from the uaccess descriptor address,
> > +which is defined as follows:
> > +
> > +.. code-block:: c
> > +
> > +  struct uaccess_descriptor {
> > +    uint64_t addr, size;
> > +  };
>
> Want to double check the extension story. If we ever want flags in
> uaccess_descriptor, we can add a flag to prctl that would say that
> address must point to uaccess_descriptor_v2 that contains flags,
> right?
> And similarly we can extend uaccess_buffer_entry, right?

Yes, we can specify a flag bit in e.g. the third argument to prctl,
which could switch us to using new struct definitions for the uaccess
descriptor and uaccess buffer entries.

> > +This data structure contains the address and size (in array elements)
> > +of a *uaccess buffer*, which is an array of data structures of type
> > +``struct uaccess_buffer_entry``. Before returning to userspace, the
> > +kernel will log information about uaccesses to sequential entries
> > +in the uaccess buffer. It will also store ``NULL`` to the uaccess
> > +descriptor address, and store the address and size of the unused
> > +portion of the uaccess buffer to the uaccess descriptor.
> > +
> > +The format of a uaccess buffer entry is defined as follows:
> > +
> > +.. code-block:: c
> > +
> > +  struct uaccess_buffer_entry {
> > +    uint64_t addr, size, flags;
> > +  };
> > +
> > +The meaning of ``addr`` and ``size`` should be obvious. On arm64,
>
> I would say explicitly "addr and size contain address and size of the
> user memory access".

Done in v3.

> > +tag bits are preserved in the ``addr`` field. There is currently
> > +one flag bit assignment for the ``flags`` field:
> > +
> > +.. code-block:: c
> > +
> > +  #define UACCESS_BUFFER_FLAG_WRITE 1
> > +
> > +This flag is set if the access was a write, or clear if it was a
> > +read. The meaning of all other flag bits is reserved.
> > +
> > +When entering the kernel with a non-zero uaccess descriptor
> > +address for a reason other than a syscall (for example, when
> > +IPI'd due to an incoming asynchronous signal), any signals other
> > +than ``SIGKILL`` and ``SIGSTOP`` are masked as if by calling
> > +``sigprocmask(SIG_SETMASK, set, NULL)`` where ``set`` has been
> > +initialized with ``sigfillset(set)``. This is to prevent incoming
> > +signals from interfering with uaccess logging.
> > +
> > +Example
> > +-------
> > +
> > +Here is an example of a code snippet that will enumerate the accesses
> > +performed by a ``uname(2)`` syscall:
> > +
> > +.. code-block:: c
> > +
> > +  struct uaccess_buffer_entry entries[64];
> > +  struct uaccess_descriptor desc;
> > +  uint64_t desc_addr = 0;
> > +  prctl(PR_SET_UACCESS_DESCRIPTOR_ADDR_ADDR, &desc_addr, 0, 0, 0);
> > +
> > +  desc.addr = (uint64_t)&entries;
> > +  desc.size = 64;
> > +  desc_addr = (uint64_t)&desc;
>
> We don't need any additional compiler barriers here, right?
> It seems that we only need to prevent re-ordering of these writes with
> the next and previous syscalls, which the compiler should do already.

Right. From the compiler's perspective the address of desc_addr is
leaked at the prctl call site, so any external function call
(including syscalls) could read or write to it.

Peter