On Tue, May 19, 2015 at 03:30:12PM +0800, Herbert Xu wrote:

On Tue, May 19, 2015 at 12:27:40AM -0700, Jaegeuk Kim wrote:

quoted

So, IMHO, it can consume memory too much, since tfm should be allocated for
every inodes and be alive until inode eviction.

Are you sure this is a real problem? Have you actually looked at
how much memory it consumes?

I didn't measure the memory footprint. I can't tell it is a real problem,
but at least, I hope it can reduce somewhat memory consumption.
Because, already per-inode crypto structure consumes at least 84 bytes.

quoted

Apart from giving GFP_NOFS, do you mean that it is a wrong approach?

Allocating the tfm and setting a key on the data path is not
acceptable.

Ok, I'll check it again and investigate in more detail how not to break
this rule.

Thank you very much for the comments.

Cheers,
-- 
Email: Herbert Xu [off-list ref]
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt