Re: Regression in squashfs mount option handling in v5.4

Re: Regression in squashfs mount option handling in v5.4

[Randy Dunlap]

[adding Cc-s]

On 11/30/19 10:15 AM, Jeremi Piotrowski wrote:
> Hi,
> 
> I'm working on an embedded project which uses 'rauc' as an updater. rauc mounts
> a squashfs image using
> 
>   mount -t squashfs -o ro,loop,sizelimit=xxx squashfs.img /mnt
> 
> On my system mount is busybox, and busybox does not know the sizelimit
> parameter, so it simply passes it on to the mount syscall. The syscall
> arguments end up being:
> 
>   mount("/dev/loop0", "dir", "squashfs", MS_RDONLY|MS_SILENT, "sizelimit=xxx")
> 
> Until kernel 5.4 this worked, since 5.4 this returns EINVAL and dmesg contains
> the line "squashfs: Unknown parameter 'sizelimit'". I believe this has to do
> with the conversion of squashfs to the new mount api. 
> 
> This is an unfortunate regression, and it does not seem like this can be simply
> reverted. What is the suggested course of action?
> 
> Please cc me on replies, I'm not subscribed to the list.
> 
> Thanks,
> Jeremi
> 


-- 
~Randy
Reported-by: Randy Dunlap <rdunlap@infradead.org>

Re: Regression in squashfs mount option handling in v5.4

[Jeremi Piotrowski]

On Sat, Nov 30, 2019 at 10:56:47AM -0800, Randy Dunlap wrote:
> [adding Cc-s]
> 
> On 11/30/19 10:15 AM, Jeremi Piotrowski wrote:
> > Hi,
> > 
> > I'm working on an embedded project which uses 'rauc' as an updater. rauc mounts
> > a squashfs image using
> > 
> >   mount -t squashfs -o ro,loop,sizelimit=xxx squashfs.img /mnt
> > 
> > On my system mount is busybox, and busybox does not know the sizelimit
> > parameter, so it simply passes it on to the mount syscall. The syscall
> > arguments end up being:
> > 
> >   mount("/dev/loop0", "dir", "squashfs", MS_RDONLY|MS_SILENT, "sizelimit=xxx")
> > 
> > Until kernel 5.4 this worked, since 5.4 this returns EINVAL and dmesg contains
> > the line "squashfs: Unknown parameter 'sizelimit'". I believe this has to do
> > with the conversion of squashfs to the new mount api. 
> > 
> > This is an unfortunate regression, and it does not seem like this can be simply
> > reverted. What is the suggested course of action?
> > 
> > Please cc me on replies, I'm not subscribed to the list.
> > 
> > Thanks,
> > Jeremi
> > 
> 
> 
> -- 
> ~Randy
> Reported-by: Randy Dunlap <rdunlap@infradead.org>

Ping. This is preventing me from updating the kernel on my systems.

Re: Regression in squashfs mount option handling in v5.4

[Laura Abbott]

On 12/10/19 1:50 PM, Jeremi Piotrowski wrote:
> On Sat, Nov 30, 2019 at 10:56:47AM -0800, Randy Dunlap wrote:
>> [adding Cc-s]
>>
>> On 11/30/19 10:15 AM, Jeremi Piotrowski wrote:
>>> Hi,
>>>
>>> I'm working on an embedded project which uses 'rauc' as an updater. rauc mounts
>>> a squashfs image using
>>>
>>>    mount -t squashfs -o ro,loop,sizelimit=xxx squashfs.img /mnt
>>>
>>> On my system mount is busybox, and busybox does not know the sizelimit
>>> parameter, so it simply passes it on to the mount syscall. The syscall
>>> arguments end up being:
>>>
>>>    mount("/dev/loop0", "dir", "squashfs", MS_RDONLY|MS_SILENT, "sizelimit=xxx")
>>>
>>> Until kernel 5.4 this worked, since 5.4 this returns EINVAL and dmesg contains
>>> the line "squashfs: Unknown parameter 'sizelimit'". I believe this has to do
>>> with the conversion of squashfs to the new mount api.
>>>
>>> This is an unfortunate regression, and it does not seem like this can be simply
>>> reverted. What is the suggested course of action?
>>>
>>> Please cc me on replies, I'm not subscribed to the list.
>>>
>>> Thanks,
>>> Jeremi
>>>
>>
>>
>> -- 
>> ~Randy
>> Reported-by: Randy Dunlap <rdunlap@infradead.org>
> 
> Ping. This is preventing me from updating the kernel on my systems.
> 

Fedora is hitting a similar issue https://bugzilla.redhat.com/show_bug.cgi?id=1781863

Dec 10 10:04:06 kernel: squashfs: version 4.0 (2009/01/31) Phillip Lougher
Dec 10 10:04:06 kernel: squashfs: Unknown parameter 'errors'

I don't think squashfs ever actually supported the errors parameter
but it was just silently ignored. It turns out you could can pass
whatever nonsense you want for parameters so I don't think we can
reject parameters in the generic case

diff --git a/fs/fs_context.c b/fs/fs_context.c
index 138b5b4d621d..7ec20b1f8a53 100644
--- a/fs/fs_context.c
+++ b/fs/fs_context.c
@@ -160,8 +160,7 @@ int vfs_parse_fs_param(struct fs_context *fc, struct fs_parameter *param)
  		return 0;
  	}
  
-	return invalf(fc, "%s: Unknown parameter '%s'",
-		      fc->fs_type->name, param->key);
+	return 0;
  }
  EXPORT_SYMBOL(vfs_parse_fs_param);