On 1/24/14, Zameer Manji [off-list ref] wrote:

Hey,

My colleague Will and I are working on improving eCryptfs, an encrypted
file system that ships with linux. We are trying to add a new cipher mode
and we have run into a problem [1]. When the user calls `ftruncate` on a
file and increases the file size, eCryptfs attempts fetch new pages for the
file by calling `read_mapping_page` and which calls `read_cache_page`. This
calls eCryptfs' `readpage` implementation. We believe `read_cache_page`
calls `readpage` with a page that we have not written to before (since the
user is increasing the file size via `ftruncate`).

What function can we use to identify when we are given a page to our
`readpage` implementation that is a page we have never written to before?

Why do you think that will happen? I mean why would you get a page in your
readpage implmentation that you already wrote? Unless you unlock the recently
written page calling thread would wait for it. Be it a thread or reclaim by VM.

As long as your readpage doesn't want to touch a page which is being
handled by another
readpage call already i think you are safe to assume that your
readpage is getting
only a NOT_UPTODATE page.

We need to do this so we know if should check the integrity of the data in
the page (if we wrote to it before) or just ignore the contents (because it
is a fresh page with garbage data). For more information our
ecryptfs_readpage implementation is available on github [2].

Please note that I am not subscribed to kernelnewbies so please include me
directly in any replies.

[1]: http://marc.info/?l=ecryptfs-users&m=139007357027191&w=2
[2]:
https://github.com/zmanji/ecryptfs/blob/next-patch/fs/ecryptfs/mmap.c#L193

--
Zameer Manji

-- 
Pranay Srivastava