ceph: invalidate pages that beyond EOF in ceph_writepages_start()

Commit Message

Yan, Zheng Nov. 1, 2017, 1:43 p.m. UTC

It's possible there are several pending ceph_cap_snap, and they have
different i_size. For a given dirty page, value of EOF depends on which
snap context it's associated. calling truncate_pagecache() in
__ceph_do_pending_vmtruncate() can not handle such complex situation.

Signed-off-by: "Yan, Zheng" <[e-mail]>
---
 fs/ceph/addr.c  | 32 +++++++++++++++++++-------------
 fs/ceph/inode.c | 14 +-------------
 2 files changed, 20 insertions(+), 26 deletions(-)

Patch

diff --git a/fs/ceph/addr.c b/fs/ceph/addr.c
index e57e9d37bf2d..938acb15b4d3 100644
--- a/fs/ceph/addr.c
+++ b/fs/ceph/addr.c
@@ -892,16 +892,30 @@  static int ceph_writepages_start(struct address_space *mapping,
 				unlock_page(page);
 				continue;
 			}
-			if (strip_unit_end && (page->index > strip_unit_end)) {
-				dout("end of strip unit %p\n", page);
+			/* only if matching snap context */
+			pgsnapc = page_snap_context(page);
+			if (pgsnapc != snapc) {
+				dout("page snapc %p %lld != oldest %p %lld\n",
+				     pgsnapc, pgsnapc->seq, snapc, snapc->seq);
 				unlock_page(page);
-				break;
+				continue;
 			}
 			if (page_offset(page) >= ceph_wbc.i_size) {
 				dout("%p page eof %llu\n",
 				     page, ceph_wbc.i_size);
-				/* not done if range_cyclic */
-				stop = true;
+				if (ceph_wbc.size_stable) {
+					mapping->a_ops->invalidatepage(page,
+								0, PAGE_SIZE);
+				} else {
+					stop = true;
+				}
+				unlock_page(page);
+				if (stop)
+					break;
+				continue;
+			}
+			if (strip_unit_end && (page->index > strip_unit_end)) {
+				dout("end of strip unit %p\n", page);
 				unlock_page(page);
 				break;
 			}
@@ -915,14 +929,6 @@  static int ceph_writepages_start(struct address_space *mapping,
 				wait_on_page_writeback(page);
 			}
 
-			/* only if matching snap context */
-			pgsnapc = page_snap_context(page);
-			if (pgsnapc != snapc) {
-				dout("page snapc %p %lld != oldest %p %lld\n",
-				     pgsnapc, pgsnapc->seq, snapc, snapc->seq);
-				unlock_page(page);
-				continue;
-			}
 
 			if (!clear_page_dirty_for_io(page)) {
 				dout("%p !clear_page_dirty_for_io\n", page);
diff --git a/fs/ceph/inode.c b/fs/ceph/inode.c
index 8100f8a1a12d..e3515f759998 100644
--- a/fs/ceph/inode.c
+++ b/fs/ceph/inode.c
@@ -1838,22 +1838,10 @@  void __ceph_do_pending_vmtruncate(struct inode *inode)
 	 * possibly truncate them.. so write AND block!
 	 */
 	if (ci->i_wrbuffer_ref_head < ci->i_wrbuffer_ref) {
-		struct ceph_cap_snap *capsnap;
-		to = ci->i_truncate_size;
-		list_for_each_entry(capsnap, &ci->i_cap_snaps, ci_item) {
-			// MDS should have revoked Frw caps
-			WARN_ON_ONCE(capsnap->writing);
-			if (capsnap->dirty_pages && capsnap->size > to)
-				to = capsnap->size;
-		}
 		spin_unlock(&ci->i_ceph_lock);
 		dout("__do_pending_vmtruncate %p flushing snaps first\n",
 		     inode);
-
-		truncate_pagecache(inode, to);
-
-		filemap_write_and_wait_range(&inode->i_data, 0,
-					     inode->i_sb->s_maxbytes);
+		filemap_write_and_wait_range(&inode->i_data, 0, LLONG_MAX);
 		goto retry;
 	}