[haiku-commits] Re: haiku: hrev43290 - src/add-ons/kernel/file_systems/btrfs
- From: Jérôme Duval <korli@xxxxxxxxxxxxxxxx>
- To: haiku-commits@xxxxxxxxxxxxx
- Date: Mon, 21 Nov 2011 00:36:49 +0100
2011/11/20 Ingo Weinhold <ingo_weinhold@xxxxxx>:
> The data must be decompressed before adding them to the file cache. That has
> to happen in the FS's io() hook. The short story is that you need to manually
> satisfy the io_request with data using the write_to_io_request() function
> (<drivers/io_request.h>) instead of just letting do_iterative_fd_io() do that.
>
> For an example you can have a look at packagefs. There the control flow
> eventually ends up in PackageFile::Read(io_request*) which in turn lets the
> inner classes DataWriter and IORequestOutput do the work. Due to the plugable
> reader and output classes it might not be the easiest to follow, but AFAIK
> it's the only FS actually doing such a thing. Virtual FSs would have to do
> that too, but I believe none is using the file cache yet.
Thanks for the pointer. AFAICT the btrfs module doesn't use the io()
hook at the moment, while using the file cache.
Should I get rid of VM file access hooks?
I understand file_map is useless for compressed files because it only
knows the disk length. Could it be extended to differenciate disk
length and extent length?
Bye,
Jérôme
Other related posts:
