Re: [PATCH v8 1/4] fs/proc/kcore: avoid bounce buffer for ktext data

[Jiri Olsa]

On Mon, Jul 31, 2023 at 10:12:08PM +0100, Lorenzo Stoakes wrote:
> On Mon, Jul 31, 2023 at 10:34:21PM +0200, Jiri Olsa wrote:
> > On Mon, Jul 31, 2023 at 08:40:24PM +0100, Lorenzo Stoakes wrote:
> > > On Mon, Jul 31, 2023 at 09:24:50PM +0200, David Hildenbrand wrote:
> > > > On 31.07.23 21:21, Lorenzo Stoakes wrote:
> > > > > On Mon, Jul 24, 2023 at 08:23:55AM +0200, David Hildenbrand wrote:
> > > > > > Hi,
> > > > > >
> > > > > > >
> > > > > > > I met this too when I executed below command to trigger a kcore reading.
> > > > > > > I wanted to do a simple testing during system running and got this.
> > > > > > >
> > > > > > >     makedumpfile --mem-usage /proc/kcore
> > > > > > >
> > > > > > > Later I tried your above objdump testing, it corrupted system too.
> > > > > > >
> > > > > >
> > > > > > What do you mean with "corrupted system too" --  did it not only fail to
> > > > > > dump the system, but also actually harmed the system?
> > > > > >
> > > > > > @Lorenzo do you plan on reproduce + fix, or should we consider reverting
> > > > > > that change?
> > > > > >
> > > > > > --
> > > > > > Cheers,
> > > > > >
> > > > > > David / dhildenb
> > > > > >
> > > > >
> > > > > Apologies I mised this, I have been very busy lately not least with book :)
> > > > >
> > > > > Concerning, I will take a look as I get a chance. I think the whole series
> > > > > would have to be reverted which would be... depressing... as other patches
> > > > > in series eliminates the bounce buffer altogether.
> > > > >
> > > >
> > > > I spotted
> > > >
> > > > https://lkml.kernel.org/r/069dd40aa71e634b414d07039d72467d051fb486.camel@xxxxxx
> > > >
> > >
> > > Find that slightly confusing, they talk about just reveritng the patch but then
> > > also add a kern_addr_valid()?
> > >
> > > I'm also confused about people talking about just reverting the patch, as
> > > 4c91c07c93bb drops the bounce buffer altogether... presumably they mean
> > > reverting both?
> > >
> > > Clearly this is an arm64 thing (obviously), I have some arm64 hardware let me
> > > see if I can repro...
> >
> > I see the issue on x86
> 
> Ummmm what? I can't! What repro are you seeing on x86, exactly?

# cat /proc/kallsyms | grep ksys_read
ffffffff8151e450 T ksys_read

# objdump -d  --start-address=0xffffffff8151e450 --stop-address=0xffffffff8151e460 /proc/kcore 

/proc/kcore:     file format elf64-x86-64

objdump: Reading section load1 failed because: Bad address


jirka

> 
> >
> > >
> > > Baoquan, Jiri - are you reverting more than just the one commit? And does doing
> > > this go from not working -> working? Or from not working (worst case oops) ->
> > > error?
> >
> > yes, I used to revert all 4 patches
> >
> > I did quick check and had to revert 2 more patches to get clean revert
> >
> > 38b138abc355 Revert "fs/proc/kcore: avoid bounce buffer for ktext data"
> > e2c3b418d365 Revert "fs/proc/kcore: convert read_kcore() to read_kcore_iter()"
> > d8bc432cb314 Revert "iov_iter: add copy_page_to_iter_nofault()"
> > bf2c6799f68c Revert "iov_iter: Kill ITER_PIPE"
> > ccf4b2c5c5ce Revert "mm: vmalloc: convert vread() to vread_iter()"
> > de400d383a7e Revert "mm/vmalloc: replace the ternary conditional operator with min()"
> >
> > jirka
> 
> That's quite a few more reverts and obviously not an acceptable solution here.
> 
> Looking at
> https://lore.kernel.org/all/CAA5enKaUYehLZGL3abv4rsS7caoUG-pN9wF3R+qek-DGNZufbA@xxxxxxxxxxxxxx
> a parallel thread on this, it looks like the issue is that we are no longer
> using a no-fault kernel copy in KCORE_TEXT path and arm64 doesn't map everything
> in the text range.
> 
> Solution would be to reinstate the bounce buffer in this case (ugh). Longer term
> solution I think would be to create some iterator helper that does no fault
> copies from the kernel.
> 
> I will try to come up with a semi-revert that keeps the iterator stuff but
> keeps a hideous bounce buffer for the KCORE_TEXT bit with a comment
> explaining why...