nullfs with double mount is broken in FreeBSD 10 Stable

[Andreas Longwitz]

Thanks for quick answer !

Konstantin Belousov wrote:
> On Tue, Sep 19, 2017 at 11:33:18AM +0200, Andreas Longwitz wrote:
>> The described problem does not exist in V8, but in V10 Stable r317936.
>>
>> I use a device /dev/md2 defined in /etc/rc.conf:
>>
>> mdconfig_md2="-t swap -s 384m"
>> mdconfig_md2_newfs="-n"
>> mdconfig_md2_owner="root:wheel"
>> mdconfig_md2_perms="750"
>>
>> In /etc/fstab I have:
>>
>>  /dev/md2   /tmp1       ufs     rw,async,noatime,noauto 0  0
>>  /tmp1      /home/tmp1  nullfs  rw,noauto               0  0
>>  /tmp1      /var/tmp1   nullfs  rw,noauto               0  0
>>
>> Now I run:
>>
>> mount /home/tmp1
>> mount /var/tmp1
>> while true; do
>>    rm -f /home/tmp1/*
>>    cpdup /boot/kernel /home/tmp1
>>    cp /var/tmp1/kernel /dev/null
>>    df -h /tmp1
>> done
>>
>> The output is
>>
>> Filesystem    Size    Used   Avail Capacity  Mounted on
>> /dev/md2      372M     88M    254M    26%    /tmp1
>> Filesystem    Size    Used   Avail Capacity  Mounted on
>> /dev/md2      372M    100M    242M    29%    /tmp1
>> Filesystem    Size    Used   Avail Capacity  Mounted on
>> /dev/md2      372M    112M    229M    33%    /tmp1
>> Filesystem    Size    Used   Avail Capacity  Mounted on
>> /dev/md2      372M    125M    217M    36%    /tmp1
>> .....
>>
>> The "Used" space grows quickly until "No space available".
>> After umount /var/tmp1 the space is back again.
>>
>> I see the same behaviour when md is replaced by tmpfs.
>>
>> I have tried the patch from commit 317936 (V11) without success.
>> The problem exists on i386 and amd64. In production (V8) I use this
>> double mount construction for jail communication without any problems.
> 
> Can you try the _latest_ stable/11 ?  It is enough to only install the
> kernel, keep the userspace.
> 
> If it does not help, perhaps you will need the -o nocache mount option.

Yes, the option -o nocache worked for me.
Please can you tell me whether the commit 305659 (nullfs: plug vnode ref
 leak in null_vptocnp) should be used in stable/10 (in stable/11 it is
commit 306682) ?

-- 
Andreas Longwitz