Date:   Mon, 8 Feb 2021 17:03:41 +0100
From:   Edward Shishkin <edward.shishkin@...il.com>
To:     Metztli Information Technology <jose.r.r@...ztli.com>,
        reiserfs-devel@...r.kernel.org, hch@....de,
        linux-kernel@...r.kernel.org
Subject: Re: [reiser4 SFRN 5.1.3] kernel [5.10.x] read not supported for file
 /test-exec \(pid: 10094 comm: debootstrap\)

On 02/08/2021 01:54 PM, Metztli Information Technology wrote:
> On Wed, Dec 23, 2020 at 3:40 PM Edward Shishkin <edward.shishkin@...il.com> wrote:
>>
>> On 12/23/2020 05:01 PM, Metztli Information Technology wrote:
>>> Niltze [Ð—Ð´Ñ€Ð°Ð²Ñ Ñ‚Ð²ÑƒÐ¹Ñ‚Ðµ : Hello], Ed-
>>>
>>> I built Linux kernel 5.10.1-1 within the 'Debian way' -- as usual -- to generate a kernel component for my Debian-Installer (d-i).
>>> The patch I applied is reiser4-for-5.10-rc3.patch.gz from v5-unstable.
>>>
>>> Once I built the proper reiser4progs-2.0.4.tar.gz and generated one set of components for d-i I built the d-i image.
>>>
>>> Fact is, the installer throws an error in *both* bare metal and VirtualBox 6.1.16:
>>> ...
>>> Dec 22 20:19:56 main-menu[330]: INFO: Menu item 'bootstrap-base' selected
>>> Dec 22 20:19:56 debootstrap: /usr/sbin/debootstrap --components=main --debian-installer --resolve-deps --keyring=/usr/share/keyrings/archive.gpg buster /target http://deb.debian.org/debian/
>>> Dec 22 20:19:56 debootstrap: /usr/sbin/debootstrap: line 1596: /target/test-exec: Invalid argument
>>> Dec 22 20:19:56 kernel: [ 1018.632648] kernel read not supported for file /test-exec (pid: 10077 comm: debootstrap)
>>> Dec 22 20:19:56 debootstrap: E: NOEXEC
>>> Dec 22 20:19:56 debootstrap: EF: Cannot install into target '/target' mounted with noexec or nodev
>>> Dec 22 20:20:12 base-installer: error: exiting on error base-installer/debootstrap-failed
>>> Dec 22 20:20:14 main-menu[330]: WARNING **: Configuring 'bootstrap-base' failed with error code 1
>>> Dec 22 20:20:14 main-menu[330]: WARNING **: Menu item 'bootstrap-base' failed.
>>> Dec 22 20:20:15 main-menu[330]: INFO: Falling back to the package description for brltty-udeb
>>>
>>
>> [...]
>>
>>>
>>> Apparently, d-i [Debian-installer] complains about being unable to set the test file executable and causes the error when 1 is returned.
>>> Notwithstanding, I manually verified that I am able to touch a file and set it +x executable.
>>>
>>> Furthermore, tricking the function return value to 0 I am able to make d-i continue with the latest SFRN5 installation (see [*trick*] below); yet, subsequently halts again with
>>> an apparently related error --can not proceed any further.
>>>
>>> Digging deeper with dmesg, we can see that apparently it is the kernel which cannot 'read' properly. Please find a partial dmesg log with relevant output
>>> from an attempt on my physical development machine.
>>> ...
>>> [  508.614488] Loading Reiser4 (Software Framework Release: 5.1.3). See reiser4.wiki.kernel.org for a description of Reiser4.
>>> [  508.661951] SGI XFS with ACLs, security attributes, realtime, quota, no debug enabled
>>> [  509.326270] device-mapper: uevent: version 1.0.3
>>> [  509.326505] device-mapper: ioctl: 4.43.0-ioctl (2020-10-01) initialised: dm-devel@...hat.com
>>> [  509.902828]  sda: sda1 sda2 sda3 sda4 sda5 sda6
>>> [  509.915300]  sdb: sdb1 sdb2 sdb3
>>> [  511.973360]  sdb: sdb1 sdb2 sdb3
>>> [  627.525371] Adding 9765884k swap on /dev/sda3.  Priority:-2 extents:1 across:9765884k FS
>>> [  636.240812] reiser4[mount(9430)]: reiser4_register_subvol (fs/reiser4/init_volume.c:222)[edward-1932]:
>>> [  636.240812] NOTICE: brick /dev/sda6 has been registered
>>> [  636.243003] reiser4 (sda6): found disk format 5.1.3.
>>> [  643.759971] reiser4 (/dev/sda6): using Hybrid Transaction Model.
>>> [  643.759980] reiser4: brick /dev/sda6 activated
>>> [  643.788537] EXT4-fs (sda1): mounting ext2 file system using the ext4 subsystem
>>> [  643.813474] EXT4-fs (sda1): mounted filesystem without journal. Opts: (null)
>>> [  643.813488] ext2 filesystem being mounted at /target/boot supports timestamps until 2038 (0x7fffffff)
>>> [  648.168730] kernel read not supported for file /test-exec (pid: 9876 comm: debootstrap) [*trick*]
>>> [  898.761385] reiser4: brick /dev/sda6 deactivated
>>> [  991.001332] reiser4 (sda6): found disk format 5.1.3.
>>> [  999.093471] reiser4 (/dev/sda6): using Hybrid Transaction Model.
>>> [  999.093480] reiser4: brick /dev/sda6 activated
>>> [ 1009.340117] EXT4-fs (sda1): mounting ext2 file system using the ext4 subsystem
>>> [ 1009.362722] EXT4-fs (sda1): mounted filesystem without journal. Opts: (null)
>>> [ 1009.362737] ext2 filesystem being mounted at /target/boot supports timestamps until 2038 (0x7fffffff)
>>> [ 6373.748413] kernel read not supported for file /test-exec (pid: 10094 comm: debootstrap)
>>> [ 6413.169920] kernel read not supported for file /usr/bin/true (pid: 15960 comm: chroot)
>>
>>
>> Hello.
>>
>> This is because of VFS changes in Linux-5.10.X.
>> Specifically, because of the following patch:
>> https://lkml.org/lkml/2020/8/17/174
>> In the upstream git repository it is commit 4d03e3cc59828c82ee89ea6e2
>>
>> So, Christoph, what to do now for file systems which implement
>> ->read() method of file operations?
> 
> *deafening silence* it appears that -- in the best of cases -- Christoph engaged in an act of _iter masturbation [1];
> and in the worst of cases, the gentleman was aiming straight at reiser4.
> 
>> ... It seems that chroot doesn't work
>> for them. And people are not able to release distros with upgraded
>> kernels..
> 
> Not only 'chroot doesn't work' for us, but even after replacing the kernel in a reiser4 (proper SFRN ;) instance and
>   upon an initial Linux 5.10.x kernel boot:
> ...
> kernel read not supported for file usr/lib/systemd/systemd (pid: 1 comm: run-init)
> kernel panic -- not syncing: Attempted to kill init! exitcod=0x00000100
> ...
> 
> Fact is some of us have commercial interests when deploying reiser4, both in cloud instances, baremetal, and on-premises.
> 
> In the future if -- and only if -- our reiser4 efforts come to successful fruition, quite likely in due time we will be
>   able to financially commit to the Penguin's Linux Foundation temple, just like large corporations do
>   in exchange for indulgences[2] which virtue-wash their past and/or current corp. officers' *substantially darker deeds*;
>   heck, 'indulgence trafficking' seems to assuage->numb->arg(STFU) that 'virtuous' cult of GNU/Linux
>   developers/gatekeepers/maintainers' frivolous, *narcissist*, ethics and/or moralities so often piled up against
>   Reiser's work --which, paradoxically(!?), actually was largely implemented by Russian developers ;)
> 
> In the meantime, I hacked a reverse patch that undoes some(all) of the surreptitious lethal attack on reiser4 fs
>   -- at least on AMD64 architectures (I did away with other arch/Kconfigs).
> And no, I am not a git pro-, undoing what I could of commit 4d03e3cc59828c82ee89ea6e27a2f3cdf95aaadf (as your hinted, Ed)
>   does not fix the 'kernel read' issue.
> 
> Notwithstanding, I would appreciate if you can take a look at the attached patch. Probably it can be streamlined and/or improved
>   further to minimize pain on subsequent Linux kernel upgrades.


That patch is an attempt to swim against the current ;)

I no longer remember, why they want to get rid of set_fs for already 15
years, but ->read() and ->write() methods seem to be deprecated, and the
correct way would be to implement the new ->read_iter() and write_iter()
methods, where reiser4 works with "chunked" streams, represented by
iov_iter structure, rather than with "continuous" streams, represented
by char __user *buf. The task is not that difficult, but rather time
consuming - I don't have a time for this right now..

Thanks,
Edward.

> 
> The patch has been tested in my local development machine environment --
>   as I intalled the generated reiser4 -enabled linux 5.10.13/14 meta/kernel into a Debian Bullseye already running reiser4 (with proper SFRN)
>   and the kernel booted nicely. Subsequently, after installing the linux headers, etc. I built a couple of upgraded kernels
>   in Debian Buster GCC-8 and Bullseye GCC-10 environments -- thus the hack seems to be stable.
> 
> Additionally, I have just made a Debian-Installer (d-i) with a 'kernel read' -patched Linux 5.10.14.1 which successfully installed
>   into a VirtualBox 6.1.18 VM:
> < https://metztli.it/buster/reiser4_0_2-linux-5.10.14-kernel-read-patched.png >
> 
>>
>> Thanks,
>> Edward.
> 
> Best Professional Regards.
> 
> [1]
> "The bug was fixed, again way back in 2010, and over time chip-designers have moved on to improved memory management techniques.
> Torvalds wrote that this sort of memory space override has been banished from the x86, powerpc, s390 and RISC-V architectures."
> < https://www.theregister.com/2020/10/25/linux_5_10_rc1/ >
> 
> [2] https://www.britannica.com/topic/indulgence
> 

Powered by blists - more mailing lists