Re: [patch 1/2] fuse: delete dead .write_begin and .write_end aops

[Miklos Szeredi <miklos@szeredi.hu>]

Christoph Hellwig <hch@infradead.org> writes:

> On Mon, Aug 08, 2011 at 05:05:20PM +0200, Miklos Szeredi wrote:
>> > The loop code still calls them uncondtionally.  This actually is a big
>> > as write_begin and write_end require filesystems specific locking,
>> > and might require code in the filesystem to e.g. update the ctime
>> > properly.  I'll let Miklos chime in if leaving them in was intentional,
>> > and if it was a comment is probably justified.
>> 
>> Loop checks for ->write_begin() and falls back to ->write if the former
>> isn't defined.
>> 
>> So I think the patch is fine.  I tested loop over fuse, and it still
>> works after the patch.
>
> It works, but it involves another data copy, which will slow down
> various workloads that people at least historically cared about.

AFAICS, normally there isn't an additional copy.  If ->write_begin is
defined the copy from the bio_vec to the filesystem page is done with
transfer_none() in the loop driver.

Otherwise the copy is done by ->write() itself on the kmapped bio.

If there's a crypto transfer function then a temporary page will be used
in the no write_begin case.  But I don't think there the additional copy
makes much difference or that anyone cares.

Thanks,
Miklos

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>