rfjakob

With this PR plus the change below against loopback I was able to get 1 MiB writes [1] but not 1 MiB reads [2]. @darthShadow did you manage to get...

PS: As for the public API, I think we should do what libfuse does: calculate MaxPages from MaxWrite ( https://github.com/libfuse/libfuse/commit/027d0d17c8a4605109f09d9c988e255b64a2c19a ) and don't expose MaxPages to the user at all

I went down the rabbit hole about the read sizes a little bit: A normal (=not direct i/o) read ends up in [filemap_read()](https://github.com/torvalds/linux/blob/75acfdb6fd922598a408a0d864486aeb167c1a97/mm/filemap.c#L2608), and, quoting from the function comment: ```...

Pushed my attempt: https://review.gerrithub.io/c/hanwen/go-fuse/+/530724

Oh, looks like this got stuck on me. Maybe I'll find time to pick this up again.

Yeah you can update the patch to address Han-Wen's comments at https://review.gerrithub.io/c/hanwen/go-fuse/+/530724

Can you push this to https://gerrithub.io/ acc. to https://github.com/hanwen/go-fuse/blob/master/CONTRIBUTING ?

@stevegt some comments accumulated on gerrithub. Do you want to work on this? The unittest could look like TestMknodNotSupported: https://github.com/hanwen/go-fuse/blob/934a183ed91446d218b5471c4df9f93db039f6e1/fs/simple_test.go#L360

(I may pick it up if not)

@stevegt Yes gocryptfs has . and .. , check out https://github.com/rfjakob/gocryptfs/commit/18befda0e6f1a690fa000951df8f9a4a61daebbb @hanwen `xfstests generic/401` breaks with . and .. missing, but I guess other things will explode too, as everything...