gfs2: Inode dirtying fix (706cb549) · Commits · 戴 / test

With the recent iomap write page reclaim deadlock fix, it turns out that the GLF_DIRTY flag isn't always set when it needs to be anymore: previously, this happened as a side effect of always adding the inode buffer head to the current transaction with gfs2_trans_add_meta, but this isn't happening consistently anymore. Fix by removing an additional unnecessary gfs2_trans_add_meta call and by setting the GLF_DIRTY flag in gfs2_iomap_end. (The GLF_DIRTY flag causes inode_go_sync to flush the transaction log when syncing out the glock of that inode. When the flag isn't set, inode_go_sync will skip inodes, including ones with an i_state of I_DIRTY_PAGES, which will lead to cluster incoherency.) In addition, in gfs2_iomap_page_done, if the metadata has changed, mark the inode as I_DIRTY_DATASYNC to have the inode added to the current transaction: we don't expect metadata to change here, but let's err on the safe side. Fixes: ("gfs2: Fix iomap write page reclaim deadlock"); Signed-off-by:

Andreas Gruenbacher <agruenba@redhat.com>

fs/gfs2/bmap.c

+12 −3

Original line number	Diff line number	Diff line
		@@ -1002,11 +1002,16 @@ static void gfs2_iomap_page_done(struct inode *inode, loff_t pos,
		unsigned copied, struct page *page,
		struct iomap *iomap)
		{
		struct gfs2_trans *tr = current->journal_info;
		struct gfs2_inode *ip = GFS2_I(inode);
		struct gfs2_sbd *sdp = GFS2_SB(inode);

		if (page && !gfs2_is_stuffed(ip))
		gfs2_page_add_databufs(ip, page, offset_in_page(pos), copied);

		if (tr->tr_num_buf_new)
		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);

		gfs2_trans_end(sdp);
		}

		@@ -1099,8 +1104,6 @@ static int gfs2_iomap_begin_write(struct inode *inode, loff_t pos,
		tr = current->journal_info;
		if (tr->tr_num_buf_new)
		__mark_inode_dirty(inode, I_DIRTY_DATASYNC);
		else
		gfs2_trans_add_meta(ip->i_gl, mp->mp_bh[0]);

		gfs2_trans_end(sdp);
		}
		@@ -1181,10 +1184,16 @@ static int gfs2_iomap_end(struct inode *inode, loff_t pos, loff_t length,

		if (ip->i_qadata && ip->i_qadata->qa_qd_num)
		gfs2_quota_unlock(ip);

		if (unlikely(!written))
		goto out_unlock;

		if (iomap->flags & IOMAP_F_SIZE_CHANGED)
		mark_inode_dirty(inode);
		gfs2_write_unlock(inode);
		set_bit(GLF_DIRTY, &ip->i_gl->gl_flags);

		out_unlock:
		gfs2_write_unlock(inode);
		out:
		return 0;
		}

Admin message