From: Catalin Marinas <catalin.marinas@arm.com>
To: Andreas Gruenbacher <agruenba@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>,
Paul Mackerras <paulus@ozlabs.org>,
Alexander Viro <viro@zeniv.linux.org.uk>,
Christoph Hellwig <hch@infradead.org>,
"Darrick J. Wong" <djwong@kernel.org>, Jan Kara <jack@suse.cz>,
Matthew Wilcox <willy@infradead.org>,
cluster-devel <cluster-devel@redhat.com>,
linux-fsdevel <linux-fsdevel@vger.kernel.org>,
Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
ocfs2-devel@oss.oracle.com, kvm-ppc@vger.kernel.org,
linux-btrfs <linux-btrfs@vger.kernel.org>
Subject: Re: [PATCH v8 00/17] gfs2: Fix mmap + page fault deadlocks
Date: Tue, 26 Oct 2021 19:24:18 +0100 [thread overview]
Message-ID: <YXhH0sBSyTyz5Eh2@arm.com> (raw)
In-Reply-To: <CAHc6FU7BEfBJCpm8wC3P+8GTBcXxzDWcp6wAcgzQtuaJLHrqZA@mail.gmail.com>
On Mon, Oct 25, 2021 at 09:00:43PM +0200, Andreas Gruenbacher wrote:
> On Fri, Oct 22, 2021 at 9:23 PM Linus Torvalds
> <torvalds@linux-foundation.org> wrote:
> > On Fri, Oct 22, 2021 at 8:06 AM Catalin Marinas <catalin.marinas@arm.com> wrote:
> > > Probing only the first byte(s) in fault_in() would be ideal, no need to
> > > go through all filesystems and try to change the uaccess/probing order.
> >
> > Let's try that. Or rather: probing just the first page - since there
> > are users like that btrfs ioctl, and the direct-io path.
>
> For direct I/O, we actually only want to trigger page fault-in so that
> we can grab page references with bio_iov_iter_get_pages. Probing for
> sub-page error domains will only slow things down. If we hit -EFAULT
> during the actual copy-in or copy-out, we know that the error can't be
> page fault related. Similarly, in the buffered I/O case, we only
> really care about the next byte, so any probing beyond that is
> unnecessary.
>
> So maybe we should split the sub-page error domain probing off from
> the fault-in functions. Or at least add an argument to the fault-in
> functions that specifies the amount of memory to probe.
My preferred option is not to touch fault-in for sub-page faults (though
I have some draft patches, they need testing).
All this fault-in and uaccess with pagefaults_disabled() is needed to
avoid a deadlock when the uaccess fault handling would take the same
lock. With sub-page faults, the kernel cannot fix it up anyway, so the
arch code won't even attempt call handle_mm_fault() (it is not an mm
fault). But the problem is the copy_*_user() etc. API that can only
return the number of bytes not copied. That's what I think should be
fixed. fault_in() feels like the wrong place to address this when it's
not an mm fault.
As for fault_in() getting another argument with the amount of sub-page
probing to do, I think the API gets even more confusing. I was also
thinking, with your patches for fault_in() now returning size_t, is the
expectation to be precise in what cannot be copied? We don't have such
requirement for copy_*_user().
While more intrusive, I'd rather change copy_page_from_iter_atomic()
etc. to take a pointer where to write back an error code. If it's
-EFAULT, retry the loop. If it's -EACCES/EPERM just bail out. Or maybe
simply a bool set if there was an mm fault to be retried. Yet another
option to return an -EAGAIN if it could not process the mm fault due to
page faults being disabled.
Happy to give this a try, unless there's a strong preference for the
fault_in() fix-up (well, I can do both options and post them).
--
Catalin
next prev parent reply other threads:[~2021-10-26 18:24 UTC|newest]
Thread overview: 47+ messages / expand[flat|nested] mbox.gz Atom feed top
2021-10-19 13:41 [PATCH v8 00/17] gfs2: Fix mmap + page fault deadlocks Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 01/17] iov_iter: Fix iov_iter_get_pages{,_alloc} page fault return value Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 02/17] powerpc/kvm: Fix kvm_use_magic_page Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 03/17] gup: Turn fault_in_pages_{readable,writeable} into fault_in_{readable,writeable} Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 04/17] iov_iter: Turn iov_iter_fault_in_readable into fault_in_iov_iter_readable Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 05/17] iov_iter: Introduce fault_in_iov_iter_writeable Andreas Gruenbacher
2021-10-20 16:25 ` Catalin Marinas
2021-10-19 13:41 ` [PATCH v8 06/17] gfs2: Add wrapper for iomap_file_buffered_write Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 07/17] gfs2: Clean up function may_grant Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 08/17] gfs2: Introduce flag for glock holder auto-demotion Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 09/17] gfs2: Move the inode glock locking to gfs2_file_buffered_write Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 10/17] gfs2: Eliminate ip->i_gh Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 11/17] gfs2: Fix mmap + page fault deadlocks for buffered I/O Andreas Gruenbacher
2021-10-19 13:41 ` [PATCH v8 12/17] iomap: Fix iomap_dio_rw return value for user copies Andreas Gruenbacher
2021-10-19 13:42 ` [PATCH v8 13/17] iomap: Support partial direct I/O on user copy failures Andreas Gruenbacher
2021-10-19 13:42 ` [PATCH v8 14/17] iomap: Add done_before argument to iomap_dio_rw Andreas Gruenbacher
2021-10-19 15:51 ` Darrick J. Wong
2021-10-19 19:30 ` Andreas Gruenbacher
2021-10-20 1:57 ` Darrick J. Wong
2021-10-19 13:42 ` [PATCH v8 15/17] gup: Introduce FOLL_NOFAULT flag to disable page faults Andreas Gruenbacher
2021-10-19 13:42 ` [PATCH v8 16/17] iov_iter: Introduce nofault " Andreas Gruenbacher
2021-10-19 13:42 ` [PATCH v8 17/17] gfs2: Fix mmap + page fault deadlocks for direct I/O Andreas Gruenbacher
2021-10-19 15:40 ` [PATCH v8 00/17] gfs2: Fix mmap + page fault deadlocks Linus Torvalds
2021-10-19 16:00 ` Bob Peterson
2021-10-20 16:36 ` Catalin Marinas
2021-10-20 20:11 ` Linus Torvalds
2021-10-20 22:44 ` Catalin Marinas
2021-10-21 6:19 ` Linus Torvalds
2021-10-22 18:06 ` Catalin Marinas
2021-10-22 19:22 ` Linus Torvalds
2021-10-25 19:00 ` Andreas Gruenbacher
2021-10-26 18:24 ` Catalin Marinas [this message]
2021-10-26 18:50 ` Linus Torvalds
2021-10-26 19:18 ` Linus Torvalds
2021-10-27 19:13 ` Catalin Marinas
2021-10-27 21:14 ` Linus Torvalds
2021-10-28 21:20 ` Catalin Marinas
2021-10-28 21:40 ` Catalin Marinas
2021-10-28 22:15 ` Andreas Grünbacher
2021-10-29 12:50 ` Catalin Marinas
2021-10-28 22:32 ` Linus Torvalds
2021-10-29 17:50 ` Catalin Marinas
2021-10-29 18:47 ` Linus Torvalds
2021-10-25 18:24 ` Andreas Gruenbacher
2021-10-26 5:12 ` Theodore Ts'o
2021-10-26 9:44 ` Andreas Gruenbacher
2021-10-27 21:21 ` Andreas Gruenbacher
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=YXhH0sBSyTyz5Eh2@arm.com \
--to=catalin.marinas@arm.com \
--cc=agruenba@redhat.com \
--cc=cluster-devel@redhat.com \
--cc=djwong@kernel.org \
--cc=hch@infradead.org \
--cc=jack@suse.cz \
--cc=kvm-ppc@vger.kernel.org \
--cc=linux-btrfs@vger.kernel.org \
--cc=linux-fsdevel@vger.kernel.org \
--cc=linux-kernel@vger.kernel.org \
--cc=ocfs2-devel@oss.oracle.com \
--cc=paulus@ozlabs.org \
--cc=torvalds@linux-foundation.org \
--cc=viro@zeniv.linux.org.uk \
--cc=willy@infradead.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).