On 10/23/25 10:39 AM, Benjamin Marzinski wrote:
On Wed, Oct 22, 2025 at 04:11:58PM +0200, Martin Wilck wrote:
On Tue, 2025-10-21 at 15:16 -0400, Benjamin Marzinski wrote:
On Fri, Oct 10, 2025 at 12:19:51PM +0200, Martin Wilck wrote:
I find Bart's approach very attractive; freezing might not be
necessary
at all in that case. We'dd just need to avoid a race where paths
get
reinstated while the operation that would normally have required a
freeze is ongoing.
I agree. Even just the timing out of freezes, his
"[PATCH 2/3] block: Restrict the duration of sysfs attribute changes"
would be enough to keep this from deadlocking the system.
OK, let's see how it goes. Given your explanations, I'm ok with your
patch, too.
I see Mikulas pulled this commit into linux-dm. Bart, does this solve
your issue? Looking at your hang, it should. Also, do you have any
interest in attempting again to get your fixes upstream?
I think that I have done what I could to get this regression fixed. As
one can see here I reported on June 30 that a regression got introduced
in the block layer:
https://lore.kernel.org/linux-block/[email protected]/
Although Linus Torvalds does not tolerate regressions in the Linux
kernel, the people who introduced this regression were not impressed
by my email, refused my patches, did not revert their own patches that
introduced the regression and did not suggest an alternative approach
for fixing the regression.
Bart.
