linux-kernel - About whether we should support the alignment in the generic_file_splice

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>] [day] [month] [year] [list]

Message-ID: <CAL+tcoD8o5A4vgLHHp8dyFV5PmJVL5tu0h-XQavLOiAexmVLRQ@mail.gmail.com>
Date:   Fri, 29 Nov 2019 15:50:22 +0800
From:   Jason Xing <kerneljasonxing@...il.com>
To:     viro@...iv.linux.org.uk
Cc:     linux-xfs@...r.kernel.org, linux-fsdevel@...r.kernel.org,
        linux-kernel@...r.kernel.org
Subject: About whether we should support the alignment in the
 generic_file_splice_read() ?

Hi all,

Sorry to send this email to you all. I recently noticed there're some
incompatibilities existing in the generic_file_splice_read() function
between 3.X and 4.X kernel. The result will goes wrong if we're using
sendfile() with unaligned offset in 4.X/5.X kernel. But if we do the
same in 3.X kernel, it will surely return success.

Here is the call trace:
1. Using sendfile() with unaligned offset
2. Then it runs into the kernel:
sendfile64()->do_sendfile()->do_splice_direct()->splice_direct_to_actor()->do_splice_to()->splice_read()
3. splice_read() calls the __generic_file_splice_read() in 3.X kernel,
generic_file_splice_read() in 4.X kernel.

In 3.X kernel, this function handles the alignment by using the
PAGE_SHIFT and PAGE_MASK. However, after applying this
commit(82c156f853840645604acd7c2cebcb75ed1b6652) the 4.X no longer
supports the unaligned data.

I'm wondering should we add the alignment process code back again?
Does anyone have idea about why this part got removed? Any information
and suggestions are welcome:-)

Thanks,
Jason