Unexpected behavior involving file attributes and snapshots.

Unexpected behavior involving file attributes and snapshots.

[Austin S. Hemmelgarn]

I was just experimenting with snapshots on 4.9.0, and came across some 
unexpected behavior.

The simple explanation is that if you snapshot a subvolume, any files in 
the subvolume that have the NOCOW attribute will not have that attribute 
in the snapshot.  Some further testing indicates that this is the only 
file attribute that isn't preserved (I checked all the chattr flags that 
BTRFS supports).

I'm kind of curious whether:
1. This is actually documented somewhere, as it's somewhat unexpected 
given that everything else is preserved when snapshotting.
2. This is intended behavior, or just happens to be a side effect of the 
implementation.

Re: Unexpected behavior involving file attributes and snapshots.

[Roman Mamedov]

On Tue, 14 Feb 2017 10:30:43 -0500
"Austin S. Hemmelgarn" <ahferroin7@gmail.com> wrote:

> I was just experimenting with snapshots on 4.9.0, and came across some 
> unexpected behavior.
> 
> The simple explanation is that if you snapshot a subvolume, any files in 
> the subvolume that have the NOCOW attribute will not have that attribute 
> in the snapshot.  Some further testing indicates that this is the only 
> file attribute that isn't preserved (I checked all the chattr flags that 
> BTRFS supports).
> 
> I'm kind of curious whether:
> 1. This is actually documented somewhere, as it's somewhat unexpected 
> given that everything else is preserved when snapshotting.
> 2. This is intended behavior, or just happens to be a side effect of the 
> implementation.

I don't seem to get this on 4.4.45 and 4.4.47.

$ btrfs sub create test
Create subvolume './test'
$ touch test/abc
$ chattr +C test/abc 
$ echo def > test/abc
$ ls -la test/abc 
-rw-r--r-- 1 rm rm 4 Feb 14 20:52 test/abc
$ lsattr test/abc 
---------------C test/abc
$ btrfs sub snap test test2
Create a snapshot of 'test' in './test2'
$ ls -la test2/abc 
-rw-r--r-- 1 rm rm 4 Feb 14 20:52 test2/abc
$ lsattr test2/abc 
---------------C test2/abc

-- 
With respect,
Roman

Re: Unexpected behavior involving file attributes and snapshots.

[Chris Murphy]

On Tue, Feb 14, 2017 at 8:30 AM, Austin S. Hemmelgarn
<ahferroin7@gmail.com> wrote:
> I was just experimenting with snapshots on 4.9.0, and came across some
> unexpected behavior.
>
> The simple explanation is that if you snapshot a subvolume, any files in the
> subvolume that have the NOCOW attribute will not have that attribute in the
> snapshot.  Some further testing indicates that this is the only file
> attribute that isn't preserved (I checked all the chattr flags that BTRFS
> supports).

Huh, I can't reproduce this with 4.9.8 or 4.10rc7. systemd sets
journal files with chattr +C, and I do manual snapshots of rootfs
periodically, and those snapshots have journal files that have +C
still set.


-- 
Chris Murphy

Re: Unexpected behavior involving file attributes and snapshots.

[Austin S. Hemmelgarn]

On 2017-02-14 11:07, Chris Murphy wrote:
> On Tue, Feb 14, 2017 at 8:30 AM, Austin S. Hemmelgarn
> <ahferroin7@gmail.com> wrote:
>> I was just experimenting with snapshots on 4.9.0, and came across some
>> unexpected behavior.
>>
>> The simple explanation is that if you snapshot a subvolume, any files in the
>> subvolume that have the NOCOW attribute will not have that attribute in the
>> snapshot.  Some further testing indicates that this is the only file
>> attribute that isn't preserved (I checked all the chattr flags that BTRFS
>> supports).
>
> Huh, I can't reproduce this with 4.9.8 or 4.10rc7. systemd sets
> journal files with chattr +C, and I do manual snapshots of rootfs
> periodically, and those snapshots have journal files that have +C
> still set.
>
Just tested on a different filesystem, and I'm not seeing it there 
either, I'll take a closer look at the FS I saw this on and see if I can 
figure out what's up.

Re: Unexpected behavior involving file attributes and snapshots.

[Austin S. Hemmelgarn]

On 2017-02-14 11:46, Austin S. Hemmelgarn wrote:
> On 2017-02-14 11:07, Chris Murphy wrote:
>> On Tue, Feb 14, 2017 at 8:30 AM, Austin S. Hemmelgarn
>> <ahferroin7@gmail.com> wrote:
>>> I was just experimenting with snapshots on 4.9.0, and came across some
>>> unexpected behavior.
>>>
>>> The simple explanation is that if you snapshot a subvolume, any files
>>> in the
>>> subvolume that have the NOCOW attribute will not have that attribute
>>> in the
>>> snapshot.  Some further testing indicates that this is the only file
>>> attribute that isn't preserved (I checked all the chattr flags that
>>> BTRFS
>>> supports).
>>
>> Huh, I can't reproduce this with 4.9.8 or 4.10rc7. systemd sets
>> journal files with chattr +C, and I do manual snapshots of rootfs
>> periodically, and those snapshots have journal files that have +C
>> still set.
>>
> Just tested on a different filesystem, and I'm not seeing it there
> either, I'll take a closer look at the FS I saw this on and see if I can
> figure out what's up.
>
After poking around a bit further, the system crashed, and it looks like 
there was some memory corruption scattered throughout the kernel from 
one of the other modules I had loaded (now I get to spend a day or more 
figuring out which one and reporting that bug).  Given the state of the 
kernel crash dump though, I'm actually somewhat surprised that things 
weren't misbehaving more spectacularly than this.