xref: /netbsd-current/doc/roadmaps/storage

$NetBSD: storage,v 1.20 2016/11/10 21:28:15 jdolecek Exp $

NetBSD Storage Roadmap
======================

This is a small roadmap document, and deals with the storage and file
systems side of the operating system. It discusses elements, projects,
and goals that are under development or under discussion; and it is
divided into three categories based on perceived priority.

The following elements, projects, and goals are considered strategic
priorities for the project:

 1. Improving iscsi
 2. nfsv4 support
 3. A better journaling file system solution
 4. Getting zfs working for real
 5. Seamless full-disk encryption
 6. Finish tls-maxphys

The following elements, projects, and goals are not strategic
priorities but are still important undertakings worth doing:

 7. nvme support
 8. lfs64
 9. Per-process namespaces
 10. lvm tidyup
 11. Flash translation layer
 12. Shingled disk support
 13. ext3/ext4 support
 14. Port hammer from Dragonfly
 15. afs maintenance
 16. execute-in-place
 17. extended attributes for acl and capability storage

The following elements, projects, and goals are perhaps less pressing;
this doesn't mean one shouldn't work on them but the expected payoff
is perhaps less than for other things:

 18. coda maintenance


Explanations
============

1. Improving iscsi
------------------

Both the existing iscsi target and initiator are fairly bad code, and
neither works terribly well. Fixing this is fairly important as iscsi
is where it's at for remote block devices. Note that there appears to
be no compelling reason to move the target to the kernel or otherwise
make major architectural changes.

 - As of November 2015 nobody is known to be working on this.
 - There is currently no clear timeframe or release target.
 - Contact agc for further information.


2. nfsv4 support
----------------

nfsv4 is at this point the de facto standard for FS-level (as opposed
to block-level) network volumes in production settings. The legacy nfs
code currently in NetBSD only supports nfsv2 and nfsv3.

The intended plan is to port FreeBSD's nfsv4 code, which also includes
nfsv2 and nfsv3 support, and eventually transition to it completely,
dropping our current nfs code. (Which is kind of a mess.) So far the
only step that has been taken is to import the code from FreeBSD. The
next step is to update that import (since it was done a while ago now)
and then work on getting it to configure and compile.

 - As of November 2015 nobody is working on this, and a volunteer to
   take charge is urgently needed.
 - There is no clear timeframe or release target, although having an
   experimental version ready for -8 would be great.
 - Contact dholland for further information.


3. A better journaling file system solution
-------------------------------------------

WAPBL, the journaling FFS that NetBSD rolled out some time back, has a
critical problem: it does not address the historic ffs behavior of
allowing stale on-disk data to leak into user files in crashes. And
because it runs faster, this happens more often and with more data.
This situation is both a correctness and a security liability. Fixing
it has turned out to be difficult. It is not really clear what the
best option at this point is:

+ Fixing WAPBL (e.g. to flush newly allocated/newly written blocks to
disk early) has been examined by several people who know the code base
and judged difficult. Also, some other problems have come to light
more recently; e.g. PR 50725, and 45676. Still, it might be the best
way forward.

+ There is another journaling FFS; the Harvard one done by Margo
Seltzer's group some years back. We have a copy of this, but as it was
written in BSD/OS circa 1999 it needs a lot of merging, and then will
undoubtedly also need a certain amount of polishing to be ready for
production use. It does record-based rather than block-based
journaling and does not share the stale data problem.

+ We could bring back softupdates (in the softupdates-with-journaling
form found today in FreeBSD) -- this code is even more complicated
than the softupdates code we removed back in 2009, and it's not clear
that it's any more robust either. However, it would solve the stale
data problem if someone wanted to port it over. It isn't clear that
this would be any less work than getting the Harvard journaling FFS
running... or than writing a whole new file system either.

+ We could write a whole new journaling file system. (That is, not
FFS. Doing a new journaling FFS implementation is probably not
sensible relative to merging the Harvard journaling FFS.) This is a
big project.

Right now it is not clear which of these avenues is the best way
forward. Given the general manpower shortage, it may be that the best
way is whatever looks best to someone who wants to work on the
problem.

 - There has been some interest in the Harvard journaling FFS but no
   significant progress. Nobody is known to be working on or particularly
   interested in porting softupdates-with-journaling. And, while
   dholland has been mumbling for some time about a plan for a
   specific new file system to solve this problem, there isn't any
   realistic prospect of significant progress on that in the
   foreseeable future, and nobody else is known to have or be working
   on even that much.
 - There is no clear timeframe or release target; but given that WAPBL
   has been disabled by default for new installs in -7 this problem
   can reasonably be said to have become critical.
 - jdolecek is working on fixing WAPBL, goal is to get WAPBL fixed
   enough to be safe to re-enable as default for -8
 - Contact joerg or martin regarding WAPBL; contact dholland regarding
   the Harvard journaling FFS.


4. Getting zfs working for real
-------------------------------

ZFS has been almost working for years now. It is high time we got it
really working. One of the things this entails is updating the ZFS
code, as what we have is rather old. The Illumos version is probably
what we want for this.

 - There has been intermittent work on zfs, but as of November 2015
   nobody is known to be actively working on it
 - There is no clear timeframe or release target.
 - Contact riastradh or ?? for further information.


5. Seamless full-disk encryption
--------------------------------

(This is only sort of a storage issue.) We have cgd, and it is
believed to still be cryptographically suitable, at least for the time
being. However, we don't have any of the following things:

+ An easy way to install a machine with full-disk encryption. It
should really just be a checkbox item in sysinst, or not much more
than that.

+ Ideally, also an easy way to turn on full-disk encryption for a
machine that's already been installed, though this is harder.

+ A good story for booting off a disk that is otherwise encrypted;
obviously one cannot encrypt the bootblocks, but it isn't clear where
in boot the encrypted volume should take over, or how to make a best
effort at protecting the unencrypted elements needed to boot. (At
least, in the absence of something like UEFI secure boot combined with
an cryptographic oracle to sign your bootloader image so UEFI will
accept it.) There's also the question of how one runs cgdconfig(8) and
where the cgdconfig binary comes from.

+ A reasonable way to handle volume passphrases. MacOS apparently uses
login passwords for this (or as passphrases for secondary keys, or
something) and this seems to work well enough apart from the somewhat
surreal experience of sometimes having to log in twice. However, it
will complicate the bootup story.

Given the increasing regulatory-level importance of full-disk
encryption, this is at least a de facto requirement for using NetBSD
on laptops in many circumstances.

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - Contact dholland for further information.


6. Finish tls-maxphys
---------------------

The tls-maxphys branch changes MAXPHYS (the maximum size of a single
I/O request) from a global fixed constant to a value that's probed
separately for each particular I/O channel based on its
capabilities. Large values are highly desirable for e.g. feeding large
disk arrays but do not work with all hardware.

The code is nearly done and just needs more testing and support in
more drivers.

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - Contact tls for further information.


7. nvme suppport
----------------

nvme ("NVM Express") is a hardware interface standard for PCI-attached
SSDs. NetBSD now has a driver for these.

Driver is now MPSAFE and uses bufq fcfs (i.e. no disksort()) already,
so the most obvious software bottlenecks were treated. It still needs
more testing on real hardware, and it may be good to investigate some further
optimizations, such as DragonFly pbuf(9) or something similar.

Semi-relatedly, it is also time for scsipi to become MPSAFE.

 - As of May 2016 a port of OpenBSD's driver has been commited. This
   will be in -8.
 - The nvme driver is a backend to ld(4) which is MPSAFE, but we still
   need to attend to I/O path bottlenecks. Better instrumentation
   is needed.
 - Flush cache commands via DIOCCACHESYNC currently doesn't wait for completion;
   it must not poll since that corrupts command queue, but it should use
   a condition variable to wait for the flush to actually finish
 - NVMe controllers supports write cache administration via GET/SET FEATURE, but
   driver doesn't currently implement the cache ioctls, leading to somewhat
   ugly dkctl(1) output; it would be fairly simple to add this, but would
   require ld(4) attachment code changed to support passing arbitrary ioctls
   to attachments
 - There is no clear timeframe or release target for these points.
 - Contact msaitoh or agc for further information.


8. lfs64
--------

LFS currently only supports volumes up to 2 TB. As LFS is of interest
for use on shingled disks (which are larger than 2 TB) and also for
use on disk arrays (ditto) this is something of a problem. A 64-bit
version of LFS for large volumes is in the works.

 - As of November 2015 dholland is working on this.
 - It is close to being ready for at least experimental use and is
   expected to be in 8.0.
 - Responsible: dholland


9. Per-process namespaces
-------------------------

Support for per-process variation of the file system namespace enables
a number of things; more flexible chroots, for example, and also
potentially more efficient pkgsrc builds. dholland thought up a
somewhat hackish but low-footprint way to implement this.

 - As of November 2015 dholland is working on this.
 - It is scheduled to be in 8.0.
 - Responsible: dholland


10. lvm tidyup
--------------

[agc says someone should look at our lvm stuff; XXX fill this in]

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - Contact agc for further information.


11. Flash translation layer
---------------------------

SSDs ship with firmware called a "flash translation layer" that
arbitrates between the block device software expects to see and the
raw flash chips. FTLs handle wear leveling, lifetime management, and
also internal caching, striping, and other performance concerns. While
NetBSD has a file system for raw flash (chfs), it seems that given
things NetBSD is often used for it ought to come with a flash
translation layer as well.

Note that this is an area where writing your own is probably a bad
plan; it is a complicated area with a lot of prior art that's also
reportedly full of patent mines. There are a couple of open FTL
implementations that we might be able to import.

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - Contact dholland for further information.


12. Shingled disk support
-------------------------

Shingled disks (or more technically, disks with "shingled magnetic
recording" or SMR) can only write whole tracks at once. Thus, to
operate effectively they require translation support similar to the
flash translation layers found in SSDs. The nature and structure of
shingle translation layers is still being researched; however, at some
point we will want to support these things in NetBSD.

 - As of November 2015 one of dholland's coworkers is looking at this.
 - There is no clear timeframe or release target.
 - Contact dholland for further information.


13. ext3/ext4 support
---------------------

We would like to be able to read and write Linux ext3fs and ext4fs
volumes. (We can already read clean ext3fs volumes as they're the same
as ext2fs, modulo volume features our ext2fs code does not support;
but we can't write them.)

Ideally someone would write ext3 and/or ext4 code, whether integrated
with or separate from the ext2 code we already have. It might also
make sense to port or wrap the Linux ext3 or ext4 code so it can be
loaded as a GPL'd kernel module; it isn't clear if that would be more
or less work than doing an implementation.

Note however that implementing ext3 has already defeated several
people; this is a harder project than it looks.

 - GSoc 2016 brought support for extents, and also ro support for dir
   hashes; jdolecek also implemented several frequently used ext4 features
   so most contemporary ext filesystems should be possible to mount
   read-write
 - still need rw dir_nhash and xattr (semi-easy), and eventually journalling
   (hard)
 - There is no clear timeframe or release target.
 - jdolecek is working on improving ext3/ext4 support (particularily
   journalling) 


14. Port hammer from Dragonfly
------------------------------

While the motivation for and role of hammer isn't perhaps super
persuasive, it would still be good to have it. Porting it from
Dragonfly is probably not that painful (compared to, say, zfs) but as
the Dragonfly and NetBSD VFS layers have diverged in different
directions from the original 4.4BSD, may not be entirely trivial
either.

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - There probably isn't any particular person to contact; for VFS
   concerns contact dholland or hannken.


15. afs maintenance
-------------------

AFS needs periodic care and feeding to continue working as NetBSD
changes, because the kernel-level bits aren't kept in the NetBSD tree
and don't get updated with other things. This is an ongoing issue that
always seems to need more manpower than it gets. It might make sense
to import some of the kernel AFS code, or maybe even just some of the
glue layer that it uses, in order to keep it more current.

 - jakllsch sometimes works on this.
 - We would like every release to have working AFS by the time it's
   released.
 - Contact jakllsch or gendalia about AFS; for VFS concerns contact
   dholland or hannken.


16. execute-in-place
--------------------

It is likely that the future includes non-volatile storage (so-called
"nvram") that looks like RAM from the perspective of software. Most
importantly: the storage is memory-mapped rather than looking like a
disk controller. There are a number of things NetBSD ought to have to
be ready for this, of which probably the most important is
"execute-in-place": when an executable is run from such storage, and
mapped into user memory with mmap, the storage hardware pages should
be able to appear directly in user memory. Right now they get
gratuitously copied into RAM, which is slow and wasteful. There are
also other reasons (e.g. embedded device ROMs) to want execute-in-
place support.

Note that at the implementation level this is a UVM issue rather than
strictly a storage issue. 

Also note that one does not need access to nvram hardware to work on
this issue; given the performance profiles touted for nvram
technologies, a plain RAM disk like md(4) is sufficient both
structurally and for performance analysis.

 - As of November 2015 nobody is known to be working on this. Some
   time back, uebayasi wrote some preliminary patches, but they were
   rejected by the UVM maintainers.
 - There is no clear timeframe or release target.
 - Contact dholland for further information.


17. use extended attributes for ACL and capability storage
----------------------------------------------------------

Currently there is some support for extended attributes in ffs,
but nothing really uses it. I would be nice if we came up with
a standard format to store ACL's and capabilities like Linux has.
The various tools must be modified to understand this and be able
to copy them if requested. Also tools to manipulate the data will
need to be written.

18. coda maintenance
--------------------

Coda only sort of works. [And I think it's behind relative to
upstream, or something of the sort; XXX fill this in.] Also the code
appears to have an ugly incestuous relationship with FFS. This should
really be cleaned up. That or maybe it's time to remove Coda.

 - As of November 2015 nobody is known to be working on this.
 - There is no clear timeframe or release target.
 - There isn't anyone in particular to contact.
 - Circa 2012 christos made it work read-write and split it
   into modules. Since then christos has not tested it.

Alistair Crooks, David Holland
Fri Nov 20 02:17:53 EST 2015
Sun May  1 16:50:42 EDT 2016 (some updates)