Re: [PATCH 1/2 v2] kprobe: Do not use uaccess functions to access kernel memory that can fault

[Nadav Amit <namit@vmware.com>]

> On Feb 22, 2019, at 3:59 PM, Andy Lutomirski <luto@kernel.org> wrote:
> 
> On Fri, Feb 22, 2019 at 3:02 PM Jann Horn <jannh@google.com> wrote:
>> On Fri, Feb 22, 2019 at 11:39 PM Nadav Amit <namit@vmware.com> wrote:
>>>> On Feb 22, 2019, at 2:21 PM, Nadav Amit <namit@vmware.com> wrote:
>>>> 
>>>>> On Feb 22, 2019, at 2:17 PM, Jann Horn <jannh@google.com> wrote:
>>>>> 
>>>>> On Fri, Feb 22, 2019 at 11:08 PM Nadav Amit <namit@vmware.com> wrote:
>>>>>>> On Feb 22, 2019, at 1:43 PM, Jann Horn <jannh@google.com> wrote:
>>>>>>> 
>>>>>>> (adding some people from the text_poke series to the thread, removing stable@)
>>>>>>> 
>>>>>>> On Fri, Feb 22, 2019 at 8:55 PM Andy Lutomirski <luto@amacapital.net> wrote:
>>>>>>>>> On Feb 22, 2019, at 11:34 AM, Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
>>>>>>>>>> On Fri, Feb 22, 2019 at 02:30:26PM -0500, Steven Rostedt wrote:
>>>>>>>>>> On Fri, 22 Feb 2019 11:27:05 -0800
>>>>>>>>>> Alexei Starovoitov <alexei.starovoitov@gmail.com> wrote:
>>>>>>>>>> 
>>>>>>>>>>>> On Fri, Feb 22, 2019 at 09:43:14AM -0800, Linus Torvalds wrote:
>>>>>>>>>>>> 
>>>>>>>>>>>> Then we should still probably fix up "__probe_kernel_read()" to not
>>>>>>>>>>>> allow user accesses. The easiest way to do that is actually likely to
>>>>>>>>>>>> use the "unsafe_get_user()" functions *without* doing a
>>>>>>>>>>>> uaccess_begin(), which will mean that modern CPU's will simply fault
>>>>>>>>>>>> on a kernel access to user space.
>>>>>>>>>>> 
>>>>>>>>>>> On bpf side the bpf_probe_read() helper just calls probe_kernel_read()
>>>>>>>>>>> and users pass both user and kernel addresses into it and expect
>>>>>>>>>>> that the helper will actually try to read from that address.
>>>>>>>>>>> 
>>>>>>>>>>> If __probe_kernel_read will suddenly start failing on all user addresses
>>>>>>>>>>> it will break the expectations.
>>>>>>>>>>> How do we solve it in bpf_probe_read?
>>>>>>>>>>> Call probe_kernel_read and if that fails call unsafe_get_user byte-by-byte
>>>>>>>>>>> in the loop?
>>>>>>>>>>> That's doable, but people already complain that bpf_probe_read() is slow
>>>>>>>>>>> and shows up in their perf report.
>>>>>>>>>> 
>>>>>>>>>> We're changing kprobes to add a specific flag to say that we want to
>>>>>>>>>> differentiate between kernel or user reads. Can this be done with
>>>>>>>>>> bpf_probe_read()? If it's showing up in perf report, I doubt a single
>>>>>>>>> 
>>>>>>>>> so you're saying you will break existing kprobe scripts?
>>>>>>>>> I don't think it's a good idea.
>>>>>>>>> It's not acceptable to break bpf_probe_read uapi.
>>>>>>>> 
>>>>>>>> If so, the uapi is wrong: a long-sized number does not reliably identify an address if you don’t separately know whether it’s a user or kernel address. s390x and 4G:4G x86_32 are the notable exceptions. I have lobbied for RISC-V and future x86_64 to join the crowd.  I don’t know whether I’ll win this fight, but the uapi will probably have to change for at least s390x.
>>>>>>>> 
>>>>>>>> What to do about existing scripts is a different question.
>>>>>>> 
>>>>>>> This lack of logical separation between user and kernel addresses
>>>>>>> might interact interestingly with the text_poke series, specifically
>>>>>>> "[PATCH v3 05/20] x86/alternative: Initialize temporary mm for
>>>>>>> patching" (https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Flkml%2F20190221234451.17632-6-rick.p.edgecombe%40intel.com%2F&amp;data=02%7C01%7Cnamit%40vmware.com%7Cbab53e52cc5c4ac4419008d69921d1f1%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636864767879731941&amp;sdata=2tqD7udTCfNbcNLcj5SFpZt8WwK5NwtgaWMKm1Ye1EE%3D&amp;reserved=0)
>>>>>>> and "[PATCH v3 06/20] x86/alternative: Use temporary mm for text
>>>>>>> poking" (https://na01.safelinks.protection.outlook.com/?url=https%3A%2F%2Flore.kernel.org%2Flkml%2F20190221234451.17632-7-rick.p.edgecombe%40intel.com%2F&amp;data=02%7C01%7Cnamit%40vmware.com%7Cbab53e52cc5c4ac4419008d69921d1f1%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636864767879731941&amp;sdata=7%2BLShgLKnra6xzSkxdJrCclCacfdE5IdczwScW83nuE%3D&amp;reserved=0),
>>>>>>> right? If someone manages to get a tracing BPF program to trigger in a
>>>>>>> task that has switched to the patching mm, could they use
>>>>>>> bpf_probe_write_user() - which uses probe_kernel_write() after
>>>>>>> checking that KERNEL_DS isn't active and that access_ok() passes - to
>>>>>>> overwrite kernel text that is mapped writable in the patching mm?
>>>>>> 
>>>>>> Yes, this is a good point. I guess text_poke() should be defined with
>>>>>> “__kprobes” and open-code memcpy.
>>>>>> 
>>>>>> Does it sound reasonable?
>>>>> 
>>>>> Doesn't __text_poke() as implemented in the proposed patch use a
>>>>> couple other kernel functions, too? Like switch_mm_irqs_off() and
>>>>> pte_clear() (which can be a call into a separate function on paravirt
>>>>> kernels)?
>>>> 
>>>> I will move the pte_clear() to be done after the poking mm was unloaded.
>>>> Give me a few minutes to send a sketch of what I think should be done.
>>> 
>>> Err.. You are right, I don’t see an easy way of preventing a kprobe from
>>> being set on switch_mm_irqs_off(), and open-coding this monster is too ugly.
>>> 
>>> The reasonable solution seems to me as taking all the relevant pieces of
>>> code (and data) that might be used during text-poking and encapsulating them, so they
>>> will be set in a memory area which cannot be kprobe'd. This can also be
>>> useful to write-protect data structures of code that calls text_poke(),
>>> e.g., static-keys. It can also protect data on that stack that is used
>>> during text_poke() from being overwritten from another core.
>>> 
>>> This solution is somewhat similar to Igor Stoppa’s idea of using “enclaves”
>>> when doing write-rarely operations.
>>> 
>>> Right now, I think that text_poke() will keep being susceptible to such
>>> an attack, unless you have a better suggestion.
>> 
>> A relatively simple approach might be to teach BPF not to run kprobe
>> programs and such in contexts where current->mm isn't the active mm?
>> Maybe using nmi_uaccess_okay(), or something like that? It looks like
>> perf_callchain_user() also already uses that. Except that a lot of
>> this code is x86-specific...
> 
> This sounds like exactly the right solution.  If you're running from
> some unknown context (like NMI or tracing), then you should check
> nmi_uaccess_okay().  I think we should just promote that to be a
> non-arch-specific function (that returns true by default) and check it
> the relevant bpf_probe_xyz() functions.

I can do that, but notice that switch_mm_irqs_off() writes to
cpu_tlbstate.loaded_mm before it actually writes to CR3. So there are still
a couple of instructions (and the load_new_mm_cr3()) in between that a
kprobe can be set on, no?

I can mark them as non-kprobable.