f2fs: fix to add refcount once page is tagged PG_private
authorChao Yu <yuchao0@huawei.com>
Wed, 6 Mar 2019 09:30:59 +0000 (17:30 +0800)
committerJaegeuk Kim <jaegeuk@kernel.org>
Wed, 13 Mar 2019 01:59:19 +0000 (18:59 -0700)
commit240a59156d9bcfabceddb66be449e7b32fb5dc4a
tree83fdc1204a92dccad9b2dd895c56e75f2a8970ca
parent25720cc05e492467099a2d4d21a50f6ee8555cfd
f2fs: fix to add refcount once page is tagged PG_private

As Gao Xiang reported in bugzilla:

https://bugzilla.kernel.org/show_bug.cgi?id=202749

f2fs may skip pageout() due to incorrect page reference count.

The problem here is that MM defined the rule [1] very clearly that
once page was set with PG_private flag, we should increment the
refcount in that page, also main flows like pageout(), migrate_page()
will assume there is one additional page reference count if
page_has_private() returns true.

But currently, f2fs won't add/del refcount when changing PG_private
flag. Anyway, f2fs should follow MM's rule to make MM's related flows
running as expected.

[1] https://lore.kernel.org/lkml/2b19b3c4-2bc4-15fa-15cc-27a13e5c7af1@aol.com/

Reported-by: Gao Xiang <gaoxiang25@huawei.com>
Signed-off-by: Chao Yu <yuchao0@huawei.com>
Signed-off-by: Jaegeuk Kim <jaegeuk@kernel.org>
fs/f2fs/checkpoint.c
fs/f2fs/data.c
fs/f2fs/dir.c
fs/f2fs/f2fs.h
fs/f2fs/node.c
fs/f2fs/segment.c