linux-kernel - Re: [PATCH] mm/filemap: do not allocate cache pages beyond end of file at read

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives

Hash Suite: Windows password security audit tool. GUI, reports in PDF.

[<prev] [next>] [<thread-prev] [day] [month] [year] [list]

Message-ID: <CAHk-=wh5yeAj47x5EAuec7NCgFyYV1VtB9A+nBq+bVP9HmeepQ@mail.gmail.com>
Date:   Wed, 27 Nov 2019 09:29:44 -0800
From:   Linus Torvalds <torvalds@...ux-foundation.org>
To:     Steven Whitehouse <swhiteho@...hat.com>
Cc:     Andreas Grünbacher <andreas.gruenbacher@...il.com>,
        Konstantin Khlebnikov <khlebnikov@...dex-team.ru>,
        "Kirill A. Shutemov" <kirill@...temov.name>,
        Linux-MM <linux-mm@...ck.org>,
        Andrew Morton <akpm@...ux-foundation.org>,
        Linux Kernel Mailing List <linux-kernel@...r.kernel.org>,
        linux-fsdevel <linux-fsdevel@...r.kernel.org>,
        Alexander Viro <viro@...iv.linux.org.uk>,
        Johannes Weiner <hannes@...xchg.org>,
        "cluster-devel@...hat.com" <cluster-devel@...hat.com>,
        Ronnie Sahlberg <lsahlber@...hat.com>,
        Steve French <sfrench@...ba.org>,
        Andreas Gruenbacher <agruenba@...hat.com>,
        Bob Peterson <rpeterso@...hat.com>
Subject: Re: [PATCH] mm/filemap: do not allocate cache pages beyond end of
 file at read

On Wed, Nov 27, 2019 at 7:42 AM Steven Whitehouse <swhiteho@...hat.com> wrote:
>
> I'll leave the finer details to Andreas here, since it is his patch, and
> hopefully we can figure out a good path forward.

As mentioned, I don't _hate_ that patch (ok, I seem to have typoed it
and said that I don't "gate" it ;), so if that's what you guys really
want to do, I'm ok with it. But..

I do think you already get the data with the current case, from the
"short read" thing. So just changing the current generic read function
to check against the size first:

  --- a/mm/filemap.c
  +++ b/mm/filemap.c
  @@ -2021,9 +2021,9 @@ static ssize_t
generic_file_buffered_read(struct kiocb *iocb,
        unsigned int prev_offset;
        int error = 0;

  -     if (unlikely(*ppos >= inode->i_sb->s_maxbytes))
  +     if (unlikely(*ppos >= inode->i_size))
                return 0;
  -     iov_iter_truncate(iter, inode->i_sb->s_maxbytes);
  +     iov_iter_truncate(iter, inode->i_size);

        index = *ppos >> PAGE_SHIFT;
        prev_index = ra->prev_pos >> PAGE_SHIFT;

and you're done. Nice and clean.

Then in gfs2 you just notice the short read, and check at that point.
Sure, you'll also cut read-ahead to the old size boundary, but does
anybody _seriously_ believe that read-ahead matters when you hit the
"some other node write more data, we're reading past the old end"
case? I don't think that's the case.

But I _can_ live with the patch that adds the extra "cached only" bit.
It just honestly feels pointless.

               Linus