[PATCH 0/9 v2] scope GFP_NOFS api

From: Michal Hocko
Date: Thu Dec 15 2016 - 09:07:59 EST

I have posted the previous version here [1]. Since then I have added a
support to suppress reclaim lockdep warnings (__GFP_NOLOCKDEP) to allow
removing GFP_NOFS usage motivated by the lockdep false positives. On top
of that I've tried to convert few KM_NOFS usages to use the new flag in
the xfs code base. This would need a review from somebody familiar with
xfs of course.

Then I've added the new scope API to the jbd/ext transaction code +
reverted some explicit GFP_NOFS usages which are covered by the scope one
now. This also needs a deep review from ext developers. I have some more
patches which remove more explicit GFP_NOFS users but that is not really
ready yet. I would really appreciate if developers for other filesystems
joined me here as well. Maybe ext parts can help to show how to start.
Especially btrfs which uses GFP_NOFS a lot (and not with a good reason
in many cases I suspect).

The patchset is based on linux-next (next-20161214).

I think the GFP_NOIO should be seeing the same clean up but that is not
a part of this patchset.

Any feedback is highly appreciated of course.

Diffstat says
fs/ext4/acl.c | 6 +++---
fs/ext4/extents.c | 8 ++++----
fs/ext4/resize.c | 4 ++--
fs/ext4/xattr.c | 4 ++--
fs/jbd2/journal.c | 7 +++++++
fs/jbd2/transaction.c | 11 +++++++++++
fs/xfs/kmem.c | 10 +++++-----
fs/xfs/kmem.h | 6 +++++-
fs/xfs/libxfs/xfs_btree.c | 2 +-
fs/xfs/libxfs/xfs_da_btree.c | 4 ++--
fs/xfs/xfs_aops.c | 6 +++---
fs/xfs/xfs_buf.c | 10 +++++-----
fs/xfs/xfs_dir2_readdir.c | 2 +-
fs/xfs/xfs_trans.c | 12 ++++++------
include/linux/gfp.h | 18 +++++++++++++++++-
include/linux/jbd2.h | 2 ++
include/linux/sched.h | 32 ++++++++++++++++++++++++++------
kernel/locking/lockdep.c | 6 +++++-
lib/radix-tree.c | 2 ++
mm/page_alloc.c | 8 +++++---
mm/vmscan.c | 6 +++---
21 files changed, 117 insertions(+), 49 deletions(-)

Michal Hocko (9):
lockdep: allow to disable reclaim lockup detection
xfs: introduce and use KM_NOLOCKDEP to silence reclaim lockdep false positives
mm: introduce memalloc_nofs_{save,restore} API
xfs: use memalloc_nofs_{save,restore} instead of memalloc_noio*
jbd2: mark the transaction context with the scope GFP_NOFS context
jbd2: make the whole kjournald2 kthread NOFS safe
Revert "ext4: avoid deadlocks in the writeback path by using sb_getblk_gfp"
Revert "ext4: fix wrong gfp type under transaction"

[1] http://lkml.kernel.org/r/1461671772-1269-1-git-send-email-mhocko@xxxxxxxxxx