Re: [Bug-ddrescue] Mapfile update mechanism

[Antonio Diaz Diaz]

Hello Eliyahu,

Eliyahu Saks wrote:
> Hi, thanks for your great work.

You are welcome.


> I've writing a FUSE driver that presents a damaged drive as a file that's
> copy-on-read: any reads to the file are checked against a mapfile; if they
> have already been read the data is returned immediately from the ddrescue
> image file, if not, ddrescue is invoked on the requested sectors, and then
> the data is returned from the image file.
>
> Understandably, invoking ddrescue on every read is very inefficient
> (parsing arguments, loading the mapfile, logging, etc.).
>
> I've looked through the source code to see where I can 'hook' in to an
> existing rescuebook and rescue parts of the file on demand.

I guess the function you are looking for is 
'Rescuebook::copy_and_update' in line 202 of rescuebook.cc.


> I see that the constructor for Mapbook calls:
>
>      split_by_domain_borders( domain_ );
>
> which seems to limit the mapfile sblock_vector to blocks in the domain.
> When updating the original mapfile, how/where are these blocks merged into
> the entire (pre-domain) mapfile? The code in update_mapfile seems the write
> just the existing (domain-limited) sblock_vector?

'update_mapfile' always writes the whole block vector (maybe with 
domain-limited annotations). 'split_by_domain_borders' just splits the 
blocks to assure that no block is partially in domain. Before the final 
update, 'Mapfile::compact_sblock_vector' is called to merge the blocks 
again.


Hope this helps,
Antonio.