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

[Dmitry Vyukov <dvyukov@google.com>]

On Tue, 23 Nov 2021 at 06:17, Peter Collingbourne <pcc@google.com> wrote:
>
> Add documentation for the uaccess logging feature.
>
> Link: https://linux-review.googlesource.com/id/Ia626c0ca91bc0a3d8067d7f28406aa40693b65a2
> Signed-off-by: Peter Collingbourne <pcc@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.

> +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?

> +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".

> +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.

> +  struct utsname un;
> +  uname(&un);
> +
> +  struct uaccess_buffer_entry* entries_end = (struct uaccess_buffer_entry*)desc.addr;
> +  for (struct uaccess_buffer_entry* entry = entries; entry != entries_end; ++entry) {
> +    printf("%s at 0x%lx size 0x%lx\n", entry->flags & UACCESS_BUFFER_FLAG_WRITE ? "WRITE" : "READ",
> +           (unsigned long)entry->addr, (unsigned long)entry->size);
> +  }
> +
> +Limitations
> +-----------
> +
> +This feature is currently only supported on the arm64, s390 and x86
> +architectures.
> +
> +Uaccess buffers are a "best-effort" mechanism for logging uaccesses. Of
> +course, not all of the accesses may fit in the buffer, but aside from
> +that, not all internal kernel APIs that access userspace memory are
> +covered. Therefore, userspace programs should tolerate unreported
> +accesses.
> +
> +On the other hand, the kernel guarantees that it will not
> +(intentionally) report accessing more data than it is specified
> +to read. For example, if the kernel implements a syscall that is
> +specified to read a data structure of size ``N`` bytes by first
> +reading a page's worth of data and then only using the first ``N``
> +bytes from it, the kernel will either report reading ``N`` bytes or
> +not report the access at all.
> --
> 2.34.0.rc2.393.gf8c9666880-goog
>