[parted-devel] Partition move function

[Curtis Gedak]

Phillip Susi wrote:
> On 4/29/2010 4:22 PM, Curtis Gedak wrote:
>   
>> For all other file systems, GParted has implemented a block by block
>> copy operation.  Please note that I believe early releases of GParted
>> used 'dd' only for copying partitions, not _moving_ partitions.
>>     
>
> Does it currently use large ( 4-8 MB maybe ) buffers with O_DIRECT?
> That is much faster than transferring 4kb at a time through the kernel
> buffer cache.
>   

The GParted code in the git repository will benchmark block sizes from 1 
MB up to 16 MB doubling the size for each test.  It will choose the copy 
operation that takes the shortest amount of time.  You can review the 
code in GParted_Core.cc by searching for "optimal block size".

The libparted functions ped_device_read and ped_device_write are used to 
perform the read and write operations respectively.

>> The GParted move code does not contain intelligence regarding the file
>> system.  While performing a move GParted copies each and every block in
>> the partition regardless of whether the block contains data.
>>     
>
> Would you be interested in improving this to skip the unused blocks?
> Asking e2fslibs for the allocation bitmap is pretty straight forward.
>   

Interested?  Yes.
Ready to do it?  No, as I have many more higher priority bug fixes and 
enhancements to make to the GParted code.
Perhaps you would consider writing a patch?  ;-)

>> GParted first performs a read of the entire partition, prior to moving
>> the partition in an attempt to avoid moving a partition containing bad
>> blocks.
>>     
>
> Outch, so it reads the entire thing twice?  Why?  Why not just warn the
> user that there were some bad sectors detected and fill the destination
> with zeroes?
>   

Two reasons for performing a test read come to mind:

1)  If bad sectors are detected on a drive, then there is still a chance 
for the user to recover the data.  This can be done by repeatedly trying 
to read the bad sector until the read operation finally succeeds.  I 
have personally done this to recover data from a commercial database.  A 
higher number of recovery schemes are available if the partition is not 
moved.

2)  If bad sectors are detected, then I personally would suggest copying 
the data to another drive to avoid further data loss.  Moving the 
partition does not address the underlying problem of a hard drive failing.

3)  We try our best in GParted to protect data as best we can.  The 
extra step of reading first prior to performing a move operation does 
provide some additional security because we avoid manipulating data on a 
partition with bad sectors.

Your alternative suggestion is certainly a valid, if not desirable 
option.  The challenge comes down to which enhancements to make in the 
time that we have.  At the moment I have my hands full with other 
changes, with support for devices with sector sizes greater than 512 
bytes foremost on my mind.