From: Steven Rostedt <rostedt@goodmis.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Masami Hiramatsu <mhiramat@kernel.org>,
linux-kernel@vger.kernel.org, linux-trace-kernel@vger.kernel.org
Subject: Re: [PATCH 5/6] eventfs: get rid of dentry pointers without refcounts
Date: Wed, 31 Jan 2024 00:33:17 -0500 [thread overview]
Message-ID: <20240131003317.7a63e799@gandalf.local.home> (raw)
In-Reply-To: <CAHk-=wjH+k47je4YbP=D+KOiNYp8cJh8C_gZFzSOa8HPDm=AQw@mail.gmail.com>
On Tue, 30 Jan 2024 21:25:30 -0800
Linus Torvalds <torvalds@linux-foundation.org> wrote:
> > Does this work:
> >
> > d_invalidate(dentry);
>
> It does, but it's basically irrelevant with the d_revalidate approach.
>
> Basically, once you have d_revalidate(), the unhashing happens there,
> and it's just extra work and pointless to do it elsewhere.
>
> So if you look at the "clean up dentry ops and add revalidate
> function" patch, you'll see that it just does
>
> - simple_recursive_removal(dentry, NULL);
>
> and the thing is just history.
With even the last patch included, without the d_invalidate() I get errors
with simply doing:
# cd /sys/kernel/tracing
# mkdir instances/foo
# ls instances/foo/events
# rmdir instances/foo
As the rmdir calls tracefs_remove() that calls simple_recursive_removal()
that then walks into the "events" directory. Without that d_invalidate, it
walks beyond just the top directory and then splats on the dentries that
are cached.
>
> So really, that final patch is the one that fixes the whole eventfs
> mess for good (knock wood). But you can't do it first, because it
> basically depends on all the refcount fixes.
I'm running my full suite with the final patch included, plus some of the
updates I mentioned in replies to other patches, as well as including this
"d_invalidate()" as it doesn't pass without it.
>
> It might be possible to re-organize the patches so that the refcount
> changes go first, then the d_revalidate(), and then the rest. But I
> suspect they all really end up depending on each other some way,
> because the basic issue was that the whole "keep unrefcounted dentry
> pointers around" was just wrong. So it doesn't end up right until
> it's _all_ fixed, because every step of the way exposes some problem.
>
> At least that was my experience. Fix one thing, and it exposes the
> hack that another thing depended on.
>
> This is actually something that Al is a master at. You sometimes see
> him send one big complicated patch where he talks about all the
> problems in some area and it's one huge "fix up everything patch" that
> looks very scary.
>
> And then a week later he sends a series of 19 patches that all make
> sense and all look "obvious" and all make small progress.
>
> And magically they end up matching that big cleanup patch in the end.
> And you just *know* that it didn't start out as that beautiful logical
> series, because you saw the big messy patch first...
I'll take a look at breaking the patches up further, as I now have a much
better understanding of dentries then I did before this discussion.
-- Steve
next prev parent reply other threads:[~2024-01-31 5:33 UTC|newest]
Thread overview: 43+ messages / expand[flat|nested] mbox.gz Atom feed top
2024-01-30 19:03 [PATCH 1/6] tracefs: avoid using the ei->dentry pointer unnecessarily Linus Torvalds
2024-01-30 19:03 ` [PATCH 2/6] eventfsfs: initialize the tracefs inode properly Linus Torvalds
2024-01-30 19:48 ` Steven Rostedt
2024-01-30 19:56 ` Steven Rostedt
2024-01-30 19:49 ` Steven Rostedt
2024-01-30 19:03 ` [PATCH 3/6] tracefs: dentry lookup crapectomy Linus Torvalds
2024-01-30 23:26 ` Al Viro
2024-01-30 23:55 ` Steven Rostedt
2024-01-31 0:07 ` Al Viro
2024-01-31 0:23 ` Al Viro
2024-01-31 0:39 ` Linus Torvalds
2024-01-30 19:03 ` [PATCH 4/6] eventfs: remove unused 'd_parent' pointer field Linus Torvalds
2024-01-30 19:03 ` [PATCH 5/6] eventfs: get rid of dentry pointers without refcounts Linus Torvalds
2024-01-30 20:55 ` Steven Rostedt
2024-01-30 21:52 ` Linus Torvalds
2024-01-30 22:56 ` Steven Rostedt
2024-01-30 22:58 ` Linus Torvalds
2024-01-30 23:04 ` Steven Rostedt
2024-01-30 22:56 ` Linus Torvalds
2024-01-30 23:06 ` Linus Torvalds
2024-01-30 23:10 ` Steven Rostedt
2024-01-30 23:25 ` Linus Torvalds
2024-01-31 0:48 ` Al Viro
2024-01-31 5:09 ` Steven Rostedt
2024-01-31 5:25 ` Linus Torvalds
2024-01-31 5:33 ` Steven Rostedt [this message]
2024-01-31 5:57 ` Linus Torvalds
2024-01-31 6:20 ` Linus Torvalds
2024-01-31 12:57 ` Steven Rostedt
2024-01-31 13:14 ` Steven Rostedt
2024-01-31 18:08 ` Linus Torvalds
2024-01-31 18:39 ` Steven Rostedt
2024-01-30 19:03 ` [PATCH 6/6] eventfs: clean up dentry ops and add revalidate function Linus Torvalds
2024-01-30 21:25 ` Steven Rostedt
2024-01-30 21:36 ` Linus Torvalds
2024-01-31 1:12 ` Al Viro
2024-01-31 2:37 ` Linus Torvalds
2024-01-31 2:46 ` Al Viro
2024-01-31 3:39 ` Linus Torvalds
2024-01-31 4:28 ` Al Viro
2024-01-31 18:00 ` Steven Rostedt
2024-01-31 23:07 ` Masami Hiramatsu
2024-01-31 23:23 ` Steven Rostedt
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to=20240131003317.7a63e799@gandalf.local.home \
--to=rostedt@goodmis.org \
--cc=linux-kernel@vger.kernel.org \
--cc=linux-trace-kernel@vger.kernel.org \
--cc=mhiramat@kernel.org \
--cc=torvalds@linux-foundation.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).