Re: [PATCH 1/5] Avoid to consider lvm snapshots when scanning devices.

[Goffredo Baroncelli <kreijack@inwind.it>]

On 12/09/2014 11:27 AM, David Sterba wrote:
>> > Today the lvm-snapshot and btrfs behave very poor: it is not
>> > predictable which device is pick (the original or the snapshot). 
>> > These patch *avoid* most problems skipping the snapshots, which
>> > to me seems a reasonable default.
>> > For the other case the user is still able to mount any disks
>> > [combination] passing them directly via command line (
>> > mount /dev/sdX -o device=/dev/sdY,device=/dev/sdz...  );

> Beware that passing 'device' does not mean that btrfs will use that
> device to assemble the filesystem. It only says to scan the device the
> same way any preceding 'btrfs dev scan' would do.

I thought a bit about your sentence, but I was unable to understand
the difference. Could you describe a case where it is different ?

I have quite clear that "btrfs scan <dev>" and "mount -o device=<dev>"
do the same thing: these fill a table with the devices information 
grouped by fsid. Then the kernel uses this table as hint to pick 
the devices for a filesystem. So except some strange case 
(like device "hot" removed) this shouldn't make any difference... 
Or not ?

The point is that when a btrfs scan is ran asynchronously,
a snapshot "may" hide the origin volume. Where the word
"may" means that it is not predictable. Passing the device solve only
this point: it becomes predictable which device is used.


 
-- 
gpg @keyserver.linux.it: Goffredo Baroncelli <kreijackATinwind.it>
Key fingerprint BBF5 1610 0B64 DAC6 5F7D  17B2 0EDA 9B37 8B82 E0B5