mbox series

[RFC,v2,0/14] Convert NFS to new FS-Cache iter API

Message ID 1596031949-26793-1-git-send-email-dwysocha@redhat.com
Headers show
Series Convert NFS to new FS-Cache iter API | expand


David Wysochanski July 29, 2020, 2:12 p.m. UTC
These patches update the nfs client to use the new FS-Cache API and are at:

They are based on David Howells fscache-iter tree at 757ac8b16a0edd3befa15c9bdcb2ab3811be945d

The first 5 patches refactor some of the NFS read code to facilitate
re-use, while the last 9 patches do the conversion to the new FS-Cache

Changes since v1
* Refactor and general cleanup of read code paths
* Fixes a few build errors when NFS_FSCACHE is not configured
* Fixes directIO data corruption (needed fscache_invalidate() on directIO write)

* Takes a "least invasive to existing code" approach
  * most fscache bits stay fs/nfs/fscache.[ch]
  * only enable fscache for files open for READ (disable for WRITE)
  * may not be the best approach (see future patcheset items below)
* Basically works and passes a series of tests (see below)
  * No kernel oopses or hangs seen with tests run

Future patchset items
* Call fscache_read_helper_* directly rather than isolation into
  fs/nfs/fscache.c, similar to the AFS conversion
* Add write-through support
  * Eliminate on/off switching of fscache based on whether a
  file is open for read or write
  * TODO: Work out any limitations of NFS versions
* Rework dfprintks and/or add ftrace points
  * fscache/cachefiles has 'debug' logging similar to rpcdebug
  * convert IO path to ftrace, leave non-IO path as dfprintk?

Tests run
* A few individual NFS/fscache unit tests: PASS
* cthon04 (fsc/non-fsc, vers=3,4.0,4.1,4.2, sec=sys): PASS
* iozone tests (fsc, vers=3,4.0,4.1,4.2, sec=sys): PASS
* xfstests/generic (fsc,vers=4.2): 17/151 (Failed/Ran) 595/444 (Total/NotRan)
Failures: generic/029 generic/030 generic/240 generic/294 generic/306 generic/356 generic/357 generic/452 generic/472 generic/493 generic/494 generic/495 generic/496 generic/497 generic/554 generic/568 generic/569
Failed 17 of 595 tests
* kernel build: FAIL (linking module fails; truncate / invalidate related?)

Test not run
* error injections (for example, connection disruptions, server errors during IO, etc)
* pNFS
* many process mixed read/write on same file
* sec=krb5

Dave Wysochanski (14):
  NFS: Clean up nfs_readpage() and nfs_readpages()
  NFS: In nfs_readpage() only increment NFSIOS_READPAGES when read
  NFS: Refactor nfs_readpage() and nfs_readpage_async() to use
  NFS: Call readpage_async_filler() from nfs_readpage_async()
  NFS: Add nfs_pageio_complete_read() and remove nfs_readpage_async()
  NFS: Allow internal use of read structs and functions
  NFS: Convert nfs_readpage() and readpages() to new fscache API
  NFS: Convert fscache_acquire_cookie and fscache_relinquish_cookie
  NFS: Only use and unuse an fscache cookie a single time based on
  NFS: Convert fscache invalidation and update aux_data and i_size
  NFS: Call nfs_fscache_invalidate() when write extends the size of the
  NFS: Invalidate fscache for direct writes
  NFS: Call fscache_resize_cookie() when inode size changes due to
  NFS: Allow NFS use of new fscache API in build

 fs/nfs/Kconfig           |   2 +-
 fs/nfs/direct.c          |   2 +
 fs/nfs/file.c            |  20 +--
 fs/nfs/fscache-index.c   |  94 --------------
 fs/nfs/fscache.c         | 309 +++++++++++++++++++++++------------------------
 fs/nfs/fscache.h         |  99 ++++++---------
 fs/nfs/inode.c           |   4 +-
 fs/nfs/internal.h        |   9 ++
 fs/nfs/nfs4proc.c        |   2 +-
 fs/nfs/pagelist.c        |   1 +
 fs/nfs/read.c            | 217 +++++++++++++++------------------
 fs/nfs/write.c           |   3 +-
 include/linux/nfs_fs.h   |   2 -
 include/linux/nfs_page.h |   1 +
 include/linux/nfs_xdr.h  |   1 +
 15 files changed, 316 insertions(+), 450 deletions(-)