Re: [linux-yocto] Question about aufs in kernel 5.10

2021-04-13 Thread Bruce Ashfield
On Tue, Apr 13, 2021 at 4:36 AM Zhang, Qiang  wrote:
>
> Hello Bruce
>
> commit 06fd68d75bd1fd41105e6726fee39594e134c558
> Author: Bruce Ashfield 
> Date:   Fri Oct 30 16:39:57 2020 -0400
>
> aufs: initial port to v5.10
>
> get_fs/set_fs are only around for nommu systems, we need to
> replace them with:
>
> mm_segment_t old_fs = force_uaccess_begin();
> force_uaccess_end(old_fs);
>
> Pairs. This may or may not be the last porting effort.
>
> Signed-off-by: Bruce Ashfield 
>
> There is a problem with this patch,  in standard kernel 5.10, bsp is 
> xilinx-zynq
> enable CONFIG_SET_FS=y,  due to remove set_fs(KERNEL_DS) , when mount aufs, 
> the vfs_write(file, ubuf, count, ppos) will be called, because the ubuf is 
> kernel space address, but remove set_fs(KERNEL_DS), the ubuf will treated as 
> user mode address processing, when call raw_copy_from_user, the error
>
> "aufs au_xino_set:1635:mount[373]: I/O Error, failed setting xino(-14).
> mount: /tmp/aufs: mount(2) system call failed: Bad address."
>
> will trigger.
>
> in aufs5, the vfs_write replace by kernel_write, this problem is solved, I 
> find the aufs be update in kernel-next.
>

Paul Gortmaker was mentioning on Monday that another aufs issue was
being looked at, can you coordinate with him ?

At this point in the 5.10 cycle, I'd prefer not to do a full aufs
update, but see if we can pull a fix out of the latest aufs branches
(not really as a commit, but as some code we identify and manually
port).

I always need the new version before the aufs code base is ready, so I
do end up doing the initial port, and revisit if issues are found.

As you mentioned, I've already done an update in the -dev kernel
through the aufs5.x-rcN branch, so we do have that reference to check
for fixes.

Bruce

> Thanks
> Qiang



--
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II

-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#9746): 
https://lists.yoctoproject.org/g/linux-yocto/message/9746
Mute This Topic: https://lists.yoctoproject.org/mt/82060140/21656
Group Owner: linux-yocto+ow...@lists.yoctoproject.org
Unsubscribe: https://lists.yoctoproject.org/g/linux-yocto/unsub 
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-



[linux-yocto] Question about aufs in kernel 5.10

2021-04-13 Thread Zhang, Qiang
Hello Bruce

commit 06fd68d75bd1fd41105e6726fee39594e134c558
Author: Bruce Ashfield 
Date:   Fri Oct 30 16:39:57 2020 -0400

aufs: initial port to v5.10

get_fs/set_fs are only around for nommu systems, we need to
replace them with:

mm_segment_t old_fs = force_uaccess_begin();
force_uaccess_end(old_fs);

Pairs. This may or may not be the last porting effort.

Signed-off-by: Bruce Ashfield 

There is a problem with this patch,  in standard kernel 5.10, bsp is xilinx-zynq
enable CONFIG_SET_FS=y,  due to remove set_fs(KERNEL_DS) , when mount aufs, the 
vfs_write(file, ubuf, count, ppos) will be called, because the ubuf is kernel 
space address, but remove set_fs(KERNEL_DS), the ubuf will treated as user mode 
address processing, when call raw_copy_from_user, the error

"aufs au_xino_set:1635:mount[373]: I/O Error, failed setting xino(-14).
mount: /tmp/aufs: mount(2) system call failed: Bad address."
 
will trigger. 

in aufs5, the vfs_write replace by kernel_write, this problem is solved, I find 
the aufs be update in kernel-next. 

Thanks
Qiang
-=-=-=-=-=-=-=-=-=-=-=-
Links: You receive all messages sent to this group.
View/Reply Online (#9743): 
https://lists.yoctoproject.org/g/linux-yocto/message/9743
Mute This Topic: https://lists.yoctoproject.org/mt/82060140/21656
Group Owner: linux-yocto+ow...@lists.yoctoproject.org
Unsubscribe: https://lists.yoctoproject.org/g/linux-yocto/unsub 
[arch...@mail-archive.com]
-=-=-=-=-=-=-=-=-=-=-=-