[Devel] [PATCH RHEL7 COMMIT] ve/fs/sync: don't skip dirty metadata on filtered sync

[Konstantin Khorenko]

The commit is pushed to "branch-rh7-3.10.0-327.3.1-vz7.10.x-ovz" and will appear at https://src.openvz.org/scm/ovz/vzkernel.git
after rh7-3.10.0-327.3.1.vz7.10.12
------>
commit 43cb9daab86c674bc435fff988f36b1fad996247
Author: Andrey Ryabinin <aryabinin at virtuozzo.com>
Date:   Thu Feb 18 15:01:05 2016 +0400

    ve/fs/sync: don't skip dirty metadata on filtered sync
    
    Port of diff-ubc-dont-skip-dirty-metadata-on-filtered-sync patch
    from PCS6.
    
    Don't skip dirty metadata during per container sync[fs].
    
    	date: 2012/01/11 19:59:46; author: xemul; state: Exp;
    	Author: Konstantin Khlebnikov
    	Email: khlebnikov at openvz.org
    	Subject: ubc: disable filtered sync for too dirty inodes
    	Date: Tue, 10 Jan 2012 17:45:25 +0400
    
    	Filter only inodes with clean meta-data.
    
    	We cannot filter writes in this case. Actually we only move writes to
    	the journal thread, and it submit them in not very effective way.
    
    	Found-by: Alexey Kuznetsov <kuznet at parallels.com>
    	Signed-off-by: Konstantin Khlebnikov <khlebnikov at openvz.org>
    
    https://jira.sw.ru/browse/PSBM-39583
    
    And the issue below is caused because we don't account metadata =>
    inodes with dirty metadata could be detached from UB
    (mapping->dirtied_ub == NULL). So sync() will skip such inodes.
    With this patch sync() will skip inodes with dirty data only.
    
    https://jira.sw.ru/browse/PSBM-44125
    
    Signed-off-by: Andrey Ryabinin <aryabinin at virtuozzo.com>
    Reviewed-by: Vladimir Davydov <vdavydov at virtuozzo.com>
    
    khorenko@: we still have 2 cases:
    1) bindmount from host into a CT => dirtied_ub will be set, but not "ours",
    but we still want to sync inodes with dirty metadata because of the single
    journal
    
    2) we've dropped vzfs, but still have simfs case: again a single journal =>
    we should commit inodes with dirty metadata even in case they (inodes) do not
    belong to "our" Container.
---
 fs/fs-writeback.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

diff --git a/fs/fs-writeback.c b/fs/fs-writeback.c
index b6b33d0..101892a 100644
--- a/fs/fs-writeback.c
+++ b/fs/fs-writeback.c
@@ -714,6 +714,7 @@ static long writeback_sb_inodes(struct super_block *sb,
 			continue;
 		}
 		if ((work->ub || work->filter_ub) &&
+		    ((inode->i_state & I_DIRTY) == I_DIRTY_PAGES) &&
 		     ub_should_skip_writeback(work->ub, inode)) {
 			spin_unlock(&inode->i_lock);
 			redirty_tail(inode, wb);
@@ -1411,7 +1412,8 @@ static void wait_sb_inodes(struct super_block *sb, struct user_beancounter *ub)
 			spin_unlock(&inode->i_lock);
 			continue;
 		}
-		if (ub && (mapping->dirtied_ub != ub)) {
+		if (ub && (mapping->dirtied_ub != ub) &&
+		    ((inode->i_state & I_DIRTY) == I_DIRTY_PAGES)) {
 			spin_unlock(&inode->i_lock);
 			continue;
 		}