lkml.org 
[lkml]   [2018]   [Jul]   [28]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: Remounting filesystem read-only
On Fri, Jul 27, 2018 at 08:18:23PM -0400, Theodore Y. Ts'o wrote:
> On Fri, Jul 27, 2018 at 01:34:31PM -0700, Sodagudi Prasad wrote:
> > > The error should be pretty clear: "Inode table for bg 0 marked as
> > > needing zeroing". That should never happen.
> >
> > Can you provide any debug patch to detect when this corruption is happening?
> > Source of this corruption and how this is partition getting corrupted?
> > Or which file system operation lead to this corruption?
>
> Do you have a reliable repro? If it's a one-off, it can be caused by
> *anything*. Crappy hardware, a bug in some proprietary, binary-only
> GPU driver dereferencing some wild pointer that corrupts kernel
> memory, etc.
>
> Asking for a debug patch is like asking for "can you create technology
> that can detect when a cockroach enter my house?"

Well, ext4 *could* add metadata read and write verifiers to complain
loudly in dmesg about stuff that shouldn't be there, so at least we'd
know when we're writing cockroaches into the house... :)

--D

> So if you have a reliable repro, then we know what operations might be
> triggering the corruption, and then you work on creating a minimal
> repro, and only *then* when we have a restricted set of possibilities
> that might be the cause (for example, if removing a GPU call makes the
> problem go away, then the patch would need to be in the proprietary
> GPU driver....)
>
> > I am digging code a bit around this warning to understand more.
>
> The warning means that a flag in block group descriptor #0 is set
> that should never be set. How did the flag get set? There is any
> number of things that could cause that.
>
> You might want to look at the block group descriptor via dumpe2fs or
> debugfs, to see if it's just a single bit getting flipped, or if the
> entire block group descriptor is garbage. Note that under normal code
> paths, the flag *never* gets set by ext4 kernel code. The flag will
> get set on non-block group 0 block group descriptors by ext4, and the
> ext4 kernel code will only clear the flag.
>
> Of course, if there is a bug in some driver that dereferences a
> pointer widely, all bets are off.
>
> - Ted

\
 
 \ /
  Last update: 2018-07-28 09:48    [W:0.835 / U:0.792 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site