LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Andreas Gruenbacher <agruenba@redhat.com>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Alexander Viro <viro@zeniv.linux.org.uk>,
	Christoph Hellwig <hch@infradead.org>,
	"Darrick J. Wong" <djwong@kernel.org>,
	Paul Mackerras <paulus@ozlabs.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
Subject: Re: [PATCH v5 00/12] gfs2: Fix mmap + page fault deadlocks
Date: Mon, 16 Aug 2021 21:14:09 +0200	[thread overview]
Message-ID: <CAHc6FU6H5q20qiQ5FX1726i0FJHyh=Y46huWkCBZTR3sk+3Dhg@mail.gmail.com> (raw)
In-Reply-To: <CAHk-=wj+_Y7NQ-NhhE0jk52c9ZB0VJbO1AjtMJFB8wP=PO+bdw@mail.gmail.com>

On Tue, Aug 3, 2021 at 9:45 PM Linus Torvalds
<torvalds@linux-foundation.org> wrote:
> On Tue, Aug 3, 2021 at 12:18 PM Andreas Gruenbacher <agruenba@redhat.com> wrote:
> > With this patch queue, fstest generic/208 (aio-dio-invalidate-failure.c)
> > endlessly spins in gfs2_file_direct_write.  It looks as if there's a bug
> > in get_user_pages_fast when called with FOLL_FAST_ONLY:
> >
> >  (1) The test case performs an aio write into a 32 MB buffer.
> >
> >  (2) The buffer is initially not in memory, so when iomap_dio_rw() ->
> >      ... -> bio_iov_iter_get_pages() is called with the iter->noio flag
> >      set, we get to get_user_pages_fast() with FOLL_FAST_ONLY set.
> >      get_user_pages_fast() returns 0, which causes
> >      bio_iov_iter_get_pages to return -EFAULT.
> >
> >  (3) Then gfs2_file_direct_write faults in the entire buffer with
> >      fault_in_iov_iter_readable(), which succeeds.
> >
> >  (4) With the buffer in memory, we retry the iomap_dio_rw() ->
> >      ... -> bio_iov_iter_get_pages() -> ... -> get_user_pages_fast().
> >      This should succeed now, but get_user_pages_fast() still returns 0.
>
> Hmm. Have you tried to figure out why that "still returns 0" happens?

The call stack is:

gup_pte_range
gup_pmd_range
gup_pud_range
gup_p4d_range
gup_pgd_range
lockless_pages_from_mm
internal_get_user_pages_fast
get_user_pages_fast
iov_iter_get_pages
__bio_iov_iter_get_pages
bio_iov_iter_get_pages
iomap_dio_bio_actor
iomap_dio_actor
iomap_apply
iomap_dio_rw
gfs2_file_direct_write

In gup_pte_range, pte_special(pte) is true and so we return 0.

> One option - for debugging only - would be to introduce a new flag to
> get_user_pages_fast() that says "print out reason if failed" and make
> the retry (but not the original one) have that flag set.
>
> There are a couple of things of note when it comes to "get_user_pages_fast()":
>
>  (a) some architectures don't even enable it
>
>  (b) it can be very picky about the page table contents, and wants the
> accessed bit to already be set (or the dirty bit, in the case of a
> write).
>
> but (a) shouldn't be an issue on any common platform and (b) shouldn't
> be an issue with  fault_in_iov_iter_readable() that actually does a
> __get_user() so it will access through the page tables.
>
> (It might be more of an issue with fault_in_iov_iter_writable() due to
> walking the page tables by hand - if we don't do the proper
> access/dirty setting, I could see get_user_pages_fast() failing).
>
> Anyway, for reason (a) I do think that eventually we should probably
> introduce FOLL_NOFAULT, and allow the full "slow" page table walk -
> just not calling down to handle_mm_fault() if it fails.
>
> But (a) should be a non-issue in your test environment, and so it
> would be interesting to hear what it is that fails. Because scanning
> through the patches, they all _look_ fine to me (apart from the one
> comment about return values, which is more about being consistent with
> copy_to/from_user() and making the code simpler - not about
> correctness)
>
>                        Linus
>

Thanks,
Andreas


  reply	other threads:[~2021-08-16 19:14 UTC|newest]

Thread overview: 23+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2021-08-03 19:18 Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 01/12] iov_iter: Fix iov_iter_get_pages{,_alloc} page fault return value Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 02/12] powerpc/kvm: Fix kvm_use_magic_page Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 03/12] Turn fault_in_pages_{readable,writeable} into fault_in_{readable,writeable} Andreas Gruenbacher
2021-08-03 19:43   ` Linus Torvalds
2021-08-03 20:57   ` Al Viro
2021-08-03 21:38     ` Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 04/12] Turn iov_iter_fault_in_readable into fault_in_iov_iter_readable Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 05/12] iov_iter: Introduce fault_in_iov_iter_writeable Andreas Gruenbacher
2021-08-04 15:24   ` Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 06/12] gfs2: Add wrapper for iomap_file_buffered_write Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 07/12] gfs2: Fix mmap + page fault deadlocks for buffered I/O Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 08/12] iomap: Fix iomap_dio_rw return value for user copies Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 09/12] iomap: Support restarting direct I/O requests after user copy failures Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 10/12] iomap: Add done_before argument to iomap_dio_rw Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 11/12] iov_iter: Introduce noio flag to disable page faults Andreas Gruenbacher
2021-08-03 19:18 ` [PATCH v5 12/12] gfs2: Fix mmap + page fault deadlocks for direct I/O Andreas Gruenbacher
2021-08-03 19:45 ` [PATCH v5 00/12] gfs2: Fix mmap + page fault deadlocks Linus Torvalds
2021-08-16 19:14   ` Andreas Gruenbacher [this message]
2021-08-18 21:49     ` Linus Torvalds
2021-08-19 19:40       ` Andreas Gruenbacher
2021-08-19 20:14         ` Linus Torvalds
2021-08-19 21:39           ` 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='CAHc6FU6H5q20qiQ5FX1726i0FJHyh=Y46huWkCBZTR3sk+3Dhg@mail.gmail.com' \
    --to=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-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 \
    --subject='Re: [PATCH v5 00/12] gfs2: Fix mmap + page fault deadlocks' \
    /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

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).