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

[Jann Horn]

On Fri, Feb 22, 2019 at 11:08 PM Nadav Amit <namit@xxxxxxxxxx> wrote:
> > On Feb 22, 2019, at 1:43 PM, Jann Horn <jannh@xxxxxxxxxx> 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@xxxxxxxxxxxxxx> wrote:
> >>> On Feb 22, 2019, at 11:34 AM, Alexei Starovoitov <alexei.starovoitov@xxxxxxxxx> 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@xxxxxxxxx> 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 donât separately know whether itâs 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 donât know whether Iâll 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%7Cd44d6f0765dd49b20db708d6990ee7e8%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636864686717142892&amp;sdata=gVALdkEULEhj4iJNEWAGxyYWe2lxnHRdamW5ZA2A5RQ%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%7Cd44d6f0765dd49b20db708d6990ee7e8%7Cb39138ca3cee4b4aa4d6cd83d9dd62f0%7C0%7C0%7C636864686717142892&amp;sdata=nu2J1FtJsZJmt53SKJz8C8ktWE9eycwdAA%2BiCi1TfCc%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
> â__kprobesâ 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)?