[RFC,v2,0/3] Convert NFS with fscache to the netfs API

Message

David Wysochanski Aug. 28, 2022, 8:39 p.m. UTC

This patchset converts NFS with fscache non-direct READ IO paths to
use the netfs API with a non-invasive approach.  The existing NFS pgio
layer does not need extensive changes, and is the best way so far I've
found to address Trond's concerns about modifying the IO path [1] as
well as only enabling netfs when fscache is configured and enabled [2].
I have not yet attempted performance comparisions to address Chuck
Lever's concern [3], though this is probably less of a concern now,
since netfs is not used unless fscache is configured and enabled.

The main patch to be reviewed is patch #3 which converts nfs_read_folio
and nfs_readahead.

Changes since v1
- PATCH 2,3: #ifdef CONFIG_NFS_FSCACHE all netfs bits
- PATCH 3: Minimize changes to nfs_read_folio
- PATCH 3: Add spinlock around updating of RPC accounting info inside
  nfs_complete_read (this resolved previous hang issue #3 on generic/075)

The patches are moving in the right direction, and is the best I have
come up with so far to move things forward.  The patches are fairly
stable as evidenced with xfstests generic with various servers.  The
known issues are as follows:

1. Unit test setting rsize < readahead does not properly read from
fscache but re-reads data from the NFS server
* This will be fixed once another linux-cachefs [4] patch to resolve
"Stop read optimisation when folio removed from pagecache"

2. "Cache volume key already in use" after xfstest runs
* xfstests (hammerspace with vers=4.2,fsc) shows the following on the
console after some tests:
"NFS: Cache volume key already in use (nfs,4.1,2,c50,cfe0100a,3,,,8000,100000,100000,bb8,ea60,7530,ea60,1)"
* This may be fixed with another patch [4] that is in progress

3. Hang with xfstests involving directIO writes + reads
* seen with NFSv4.1, NetApp (ontap9) NFS server
* seen one time with generic/209, one time with generic/451
* root of the problem is the added netfs RPC accounting is not
handling read errors such as BAD_STATEID properly; might need
to move the logic to anothe location such as 
nfs_initiate_read() and nfs_readpage_done()

4. Data corruption seen with unit test where rsize < readahead
* Seen with vanilla 6.0-rc2 (did not occur on 5.19); likely unrelated
to this patchset
 mount -o vers=4.2,fsc,rsize=8192 127.0.0.1:/export /mnt
 dd if=/dev/urandom of=/tmp/integrity-rsize-file1.bin bs=16k count=1
 ./nfs-readahead.sh set /mnt 16384
 dd if=/tmp/integrity-rsize-file1.bin of=/mnt/integrity-rsize-file1.bin bs=16k count=1
 echo 3 > /proc/sys/vm/drop_caches
 md5sum /mnt/integrity-rsize-file1.bin /tmp/integrity-rsize-file1.bin
 md5sums don't match, MD5_NFS = 00eaf1a5bc1b3dfd54711db551619afa != MD5_LOCAL = e8d835c83ba1f1264869dc40673fa20c

The patchset is based on 6.0-rc2 and has been pushed to github at:
https://github.com/DaveWysochanskiRH/kernel/commits/nfs-fscache-netfs

[1] https://lore.kernel.org/linux-nfs/[e-mail]/
[2] https://lore.kernel.org/linux-nfs/[e-mail]/
[3] https://marc.info/?l=linux-nfs&m=160597917525083&w=4
[4] https://www.mail-archive.com/[e-mail]/msg03043.html
[5] https://marc.info/?l=linux-nfs&m=165962662200679&w=4

Dave Wysochanski (3):
  NFS: Rename readpage_async_filler to nfs_pageio_add_page
  NFS: Add support for netfs in struct nfs_inode and Kconfig
  NFS: Convert nfs_read_folio and nfs_readahead to netfs APIs

 fs/nfs/Kconfig           |   1 +
 fs/nfs/delegation.c      |   2 +-
 fs/nfs/dir.c             |   2 +-
 fs/nfs/fscache.c         | 191 ++++++++++++++++++---------------------
 fs/nfs/fscache.h         |  77 ++++++++--------
 fs/nfs/inode.c           |   8 +-
 fs/nfs/internal.h        |  10 +-
 fs/nfs/pagelist.c        |  14 +++
 fs/nfs/pnfs.c            |  12 +--
 fs/nfs/read.c            | 117 ++++++++----------------
 fs/nfs/write.c           |   2 +-
 include/linux/nfs_fs.h   |  19 +---
 include/linux/nfs_page.h |   1 +
 include/linux/nfs_xdr.h  |   1 +
 14 files changed, 210 insertions(+), 247 deletions(-)