Re: [RFC PATCH v2 0/5] fs: interface for directly reading/writing compressed data

On Tue, Oct 15, 2019 at 11:42:38AM -0700, Omar Sandoval wrote:

From: Omar Sandoval <redacted>

Hello,

This series adds an API for reading compressed data on a filesystem
without decompressing it as well as support for writing compressed data
directly to the filesystem. It is based on my previous series which
added a Btrfs-specific ioctl [1], but it is now an extension to
preadv2()/pwritev2() as suggested by Dave Chinner [2]. I've included a
man page patch describing the API in detail. Test cases and examples
programs are available [3].

The use case that I have in mind is Btrfs send/receive: currently, when
sending data from one compressed filesystem to another, the sending side
decompresses the data and the receiving side recompresses it before
writing it out. This is wasteful and can be avoided if we can just send
and write compressed extents. The send part will be implemented in a
separate series, as this API can stand alone.

Patches 1 and 2 add the VFS support. Patch 3 is a Btrfs prep patch.
Patch 4 implements encoded reads for Btrfs, and patch 5 implements
encoded writes.

Changes from v1 [4]:

- Encoded reads are now also implemented.
- The encoded_iov structure now includes metadata for referring to a
  subset of decoded data. This is required to handle certain cases where
  a compressed extent is truncated, hole punched, or otherwise sliced up
  and Btrfs chooses to reflect this in metadata instead of decompressing
  the whole extent and rewriting the pieces. We call these "bookend
  extents" in Btrfs, but any filesystem supporting transparent encoding
  is likely to have a similar concept.

Where's the in-kernel documentation for this API? You're encoding a
specific set of behaviours into the user API, so this needs a whole
heap of documentation in the generic code to describe how it works
so that other filesystems implementing have a well defined guideline
to what they need to support.

Also, I don't see any test code for this - can you please add
support for RWF_ENCODED to xfs_io and write a suite of unit tests
for fstests that exercise the user API fully?  Given our history of
screwing up new user APIs, this absolutely should not be merged
until there is a full set of generic unit tests written and reviewed
for it and support has been added to fsstress, fsx, and other test
utilities to fuzz and stress the implementation as part of normal
day-to-day filesystem development...

Cheers,

Dave.
-- 
Dave Chinner
david@fromorbit.com