LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
From: Jens Axboe <axboe@kernel.dk>
To: "Krzysztof Olędzki" <ole@ans.pl>,
	"Sinan Kaya" <sinan.kaya@microsoft.com>,
	"Karel Zak" <kzak@redhat.com>
Cc: util-linux@vger.kernel.org, linux-block@vger.kernel.org,
	Linux Kernel Mailing List <linux-kernel@vger.kernel.org>
Subject: Re: Commit d5fd456c88aba4fcf77d35fe38024a8d5c814686 - "loopdev: use LOOP_CONFIG ioctl" broke loop on x86-64 w/ 32 bit userspace
Date: Tue, 27 Jul 2021 19:24:18 -0600	[thread overview]
Message-ID: <c1c9d728-c4d9-eaf4-63c3-d13b99da3a3d@kernel.dk> (raw)
In-Reply-To: <7de1bd0b-b8ea-daf0-b677-f92db1c1cdff@ans.pl>

On 7/27/21 4:56 PM, Krzysztof Olędzki wrote:
> On 2021-07-27 at 15:39, Krzysztof Olędzki wrote:
>> On 2021-07-27 at 14:53, Krzysztof Olędzki wrote:
>>> Hi,
>>>
>>> I have a number of (older) systems that are still based on 32 bit
>>> userspace but are running a relatively modern 64 bit kernel -
>>> 5.4-stable, where BTW - LOOP_CONFIGURE is not yet available.
>>>
>>> I noticed that starting with util-linux-2.37 it is no longer possible to
>>> mount images using loop:
>>>
>>> # mount /usr/install/iso/systemrescue-8.04-amd64.iso /mnt/cdrom
>>> mount: /mnt/cdrom: failed to setup loop device for
>>> /usr/install/iso/systemrescue-8.04-amd64.iso.
>>>
>>> Reverting d5fd456c88aba4fcf77d35fe38024a8d5c814686 fixes the problem:
>>>
>>> /tmp/util-linux-2.37# ./mount
>>> /usr/install/iso/systemrescue-8.04-amd64.iso /mnt/cdrom
>>> mount: /mnt/cdrom: WARNING: source write-protected, mounted read-only.
>>>
>>> I have not tested if 32 bit kernel + 32 bit userspace is also affected,
>>> but 64 bit kernel + 64 bit userspace works.
>>
>> Some debugging data:
>>
>> 30399: loopdev:      CXT: [0xff8d0f98]: using loop-control
>> 30399: loopdev:      CXT: [0xff8d0f98]: loop0 name assigned
>> 30399: loopdev:      CXT: [0xff8d0f98]: find_unused by loop-control [rc=0]
>> 30399: libmount:     LOOP: [0x57cbbcb0]: trying to use /dev/loop0
>> 30399: loopdev:      CXT: [0xff8d0f98]: set backing file=/usr/install/iso/systemrescue-8.04-amd64.iso
>> 30399: loopdev:      CXT: [0xff8d0f98]: set flags=4
>> 30399: loopdev:    SETUP: [0xff8d0f98]: device setup requested
>> 30399: loopdev:    SETUP: [0xff8d0f98]: backing file open: OK
>> 30399: loopdev:      CXT: [0xff8d0f98]: open /dev/loop0 [rw]: Success
>> 30399: loopdev:    SETUP: [0xff8d0f98]: device open: OK
>> 30399: loopdev:    SETUP: [0xff8d0f98]: LOOP_CONFIGURE failed: Inappropriate ioctl for device
>> 30399: loopdev:    SETUP: [0xff8d0f98]: failed [rc=-25]
>> 30399: libmount:     LOOP: [0x57cbbcb0]: failed to setup device
>> 30399: loopdev:      CXT: [0xff8d0f98]: de-initialize
>> 30399: loopdev:      CXT: [0xff8d0f98]: closing old open fd
>> 30399: loopdev:     ITER: [0xff8d1168]: de-initialize
>> 30399: libmount:      CXT: [0x57cbbcb0]: mount: preparing failed
>> 30399: libmount:      CXT: [0x57cbbcb0]: excode: rc=32 message="failed to setup loop device for /usr/install/iso/systemrescue-8.04-amd64.iso"
>> mount: /mnt/cdrom: failed to setup loop device for /usr/install/iso/systemrescue-8.04-amd64.iso.
>> 30399: libmount:      CXT: [0x57cbbcb0]: <---- reset [status=0] ---->
>>
>> Seems like the code expects EINVAL (-22) but gets ENOTTY (-25), confirmed with strace:
>> ioctl(4, LOOP_CONFIGURE, {fd=3, block_size=0, info={lo_offset=0, lo_number=0, lo_flags=LO_FLAGS_AUTOCLEAR, lo_file_name="/usr/install/iso/systemrescue-8.04-amd64.iso", ...}}) = -1 ENOTTY (Inappropriate ioctl for device)
>>
>> Indeed, changing the code from:
>>     if (errno != EINVAL)
>> to:
>>     if (errno != EINVAL && errno != ENOTTY)
>> allows it to work.
>>
>> Not that with 64-bit userspace, kernel returns EINVAL:
>>
>> ioctl(4, LOOP_CONFIGURE, {fd=3, block_size=0, info={lo_offset=0, lo_number=0, lo_flags=LO_FLAGS_AUTOCLEAR, lo_file_name="/usr/src/PACKAGES/systemrescue-8.04-amd64.iso", ...}}) = -1 EINVAL (Invalid argument)
> 
> ... which is because lo_compat_ioctl returns -ENOIOCTLCMD for 
> unsupported cmds, while lo_ioctl returns -EINVAL via lo_simple_ioctl.
> 
> And vfs_ioctl returns -ENOTTY for -ENOIOCTLCMD.
> 
> Now the question is if this inconsistency is intended? :)

That's unfortunate, but probably not something that can get corrected at
this time. The correct return value for an unknown ioctl is -ENOTTY
(ENOIOCTLCMD isn't user visible, should get turned into that). But
current behavior is set in stone at this point, even if it is
technically incorrect.

-- 
Jens Axboe


  reply	other threads:[~2021-07-28  1:24 UTC|newest]

Thread overview: 5+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
     [not found] <a797f527-4599-e986-a326-4bb141487f2c@ans.pl>
     [not found] ` <e7f64d43-2a26-e386-b208-5c35d6a56ed4@ans.pl>
2021-07-27 22:56   ` Krzysztof Olędzki
2021-07-28  1:24     ` Jens Axboe [this message]
2021-07-28  5:46       ` Krzysztof Olędzki
2021-07-28  9:14         ` Karel Zak
2021-07-28 18:00           ` Sinan Kaya

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=c1c9d728-c4d9-eaf4-63c3-d13b99da3a3d@kernel.dk \
    --to=axboe@kernel.dk \
    --cc=kzak@redhat.com \
    --cc=linux-block@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=ole@ans.pl \
    --cc=sinan.kaya@microsoft.com \
    --cc=util-linux@vger.kernel.org \
    --subject='Re: Commit d5fd456c88aba4fcf77d35fe38024a8d5c814686 - "loopdev: use LOOP_CONFIG ioctl" broke loop on x86-64 w/ 32 bit userspace' \
    /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).