Re: [PATCH v2 0/2] Replace and improve "mcsafe" with copy_safe()

[Andy Lutomirski]

> On Apr 30, 2020, at 12:51 PM, Dan Williams <dan.j.williams@xxxxxxxxx> wrote:
> 
> ïOn Thu, Apr 30, 2020 at 12:23 PM Luck, Tony <tony.luck@xxxxxxxxx> wrote:
>> 
>>> On Thu, Apr 30, 2020 at 11:42:20AM -0700, Andy Lutomirski wrote:
>>> I suppose there could be a consistent naming like this:
>>> 
>>> copy_from_user()
>>> copy_to_user()
>>> 
>>> copy_from_unchecked_kernel_address() [what probe_kernel_read() is]
>>> copy_to_unchecked_kernel_address() [what probe_kernel_write() is]
>>> 
>>> copy_from_fallible() [from a kernel address that can fail to a kernel
>>> address that can't fail]
>>> copy_to_fallible() [the opposite, but hopefully identical to memcpy() on x86]
>>> 
>>> copy_from_fallible_to_user()
>>> copy_from_user_to_fallible()
>>> 
>>> These names are fairly verbose and could probably be improved.
>> 
>> How about
>> 
>>        try_copy_catch(void *dst, void *src, size_t count, int *fault)
>> 
>> returns number of bytes not-copied (like copy_to_user etc).
>> 
>> if return is not zero, "fault" tells you what type of fault
>> cause the early stop (#PF, #MC).
> 
> I do like try_copy_catch() for the case when neither of the buffers
> are __user (like in the pmem driver) and _copy_to_iter_fallible()
> (plus all the helpers it implies) for the other cases.
> 
> copy_to_user_fallible
> copy_fallible_to_page
> copy_pipe_to_iter_fallible
> 
> ...because the mmu-fault handling is an aspect of "_user" and fallible
> implies other source fault reasons. It does leave a gap if an
> architecture has a concept of a fallible write, but that seems like
> something that will be handled asynchronously and not subject to this
> interface.


Iâm suspicious that, as a practical matter, x86 does have a fallible write. In particular, if a page fails and the memory failure code is notified, the page will be unmapped. At that point, an attempt to write to the failed fallible page will get #PF, and code that writes to it needs to be prepared to handle #PF.  Perhaps copy_to_fallible(), etc can still return void, but Iâm unconvinced
the implementation can be the same as memcpy.