ceph: more precise CEPH_CLIENT_CAPS_PENDING_CAPSNAP

Client uses this flag to tell mds if there is more cap snap need to
flush. It's mainly for the case that client needs to re-send cap/snap
flushes after mds failover, but CEPH_CAP_ANY_FILE_WR on corresponding
inodes are all released before mds failover.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
Reviewed-by: Jeff Layton <jlayton@redhat.com>
Signed-off-by: Ilya Dryomov <idryomov@gmail.com>

1 files changed, 30 insertions, 11 deletions

diff --git a/fs/ceph/caps.c b/fs/ceph/caps.c
index f9055cdec3c7..d98dcd976c80 100644
--- a/fs/ceph/caps.c
+++ b/fs/ceph/caps.c
@@ -1295,7 +1295,7 @@ void __ceph_remove_caps(struct ceph_inode_info *ci)
  * caller should hold snap_rwsem (read), s_mutex.
  */
 static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
-		      int op, bool sync, int used, int want, int retain,
+		      int op, int flags, int used, int want, int retain,
 		      int flushing, u64 flush_tid, u64 oldest_flush_tid)
 	__releases(cap->ci->i_ceph_lock)
 {
@@ -1393,12 +1393,19 @@ static int __send_cap(struct ceph_mds_client *mdsc, struct ceph_cap *cap,
 	arg.mode = inode->i_mode;
 
 	arg.inline_data = ci->i_inline_version != CEPH_INLINE_NONE;
-	if (list_empty(&ci->i_cap_snaps))
-		arg.flags = CEPH_CLIENT_CAPS_NO_CAPSNAP;
-	else
-		arg.flags = CEPH_CLIENT_CAPS_PENDING_CAPSNAP;
-	if (sync)
-		arg.flags |= CEPH_CLIENT_CAPS_SYNC;
+	if (!(flags & CEPH_CLIENT_CAPS_PENDING_CAPSNAP) &&
+	    !list_empty(&ci->i_cap_snaps)) {
+		struct ceph_cap_snap *capsnap;
+		list_for_each_entry_reverse(capsnap, &ci->i_cap_snaps, ci_item) {
+			if (capsnap->cap_flush.tid)
+				break;
+			if (capsnap->need_flush) {
+				flags |= CEPH_CLIENT_CAPS_PENDING_CAPSNAP;
+				break;
+			}
+		}
+	}
+	arg.flags = flags;
 
 	spin_unlock(&ci->i_ceph_lock);
 
@@ -2085,7 +2092,7 @@ ack:
 		sent++;
 
 		/* __send_cap drops i_ceph_lock */
-		delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, false,
+		delayed += __send_cap(mdsc, cap, CEPH_CAP_OP_UPDATE, 0,
 				cap_used, want, retain, flushing,
 				flush_tid, oldest_flush_tid);
 		goto retry; /* retake i_ceph_lock and restart our cap scan. */
@@ -2155,7 +2162,8 @@ retry_locked:
 						&flush_tid, &oldest_flush_tid);
 
 		/* __send_cap drops i_ceph_lock */
-		delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH, true,
+		delayed = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
+				     CEPH_CLIENT_CAPS_SYNC,
 				     __ceph_caps_used(ci),
 				     __ceph_caps_wanted(ci),
 				     (cap->issued | cap->implemented),
@@ -2328,9 +2336,17 @@ static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
 	struct ceph_cap_flush *cf;
 	int ret;
 	u64 first_tid = 0;
+	u64 last_snap_flush = 0;
 
 	ci->i_ceph_flags &= ~CEPH_I_KICK_FLUSH;
 
+	list_for_each_entry_reverse(cf, &ci->i_cap_flush_list, i_list) {
+		if (!cf->caps) {
+			last_snap_flush = cf->tid;
+			break;
+		}
+	}
+
 	list_for_each_entry(cf, &ci->i_cap_flush_list, i_list) {
 		if (cf->tid < first_tid)
 			continue;
@@ -2348,10 +2364,13 @@ static void __kick_flushing_caps(struct ceph_mds_client *mdsc,
 			dout("kick_flushing_caps %p cap %p tid %llu %s\n",
 			     inode, cap, cf->tid, ceph_cap_string(cf->caps));
 			ci->i_ceph_flags |= CEPH_I_NODELAY;
+
 			ret = __send_cap(mdsc, cap, CEPH_CAP_OP_FLUSH,
-					  false, __ceph_caps_used(ci),
+					 (cf->tid < last_snap_flush ?
+					  CEPH_CLIENT_CAPS_PENDING_CAPSNAP : 0),
+					  __ceph_caps_used(ci),
 					  __ceph_caps_wanted(ci),
-					  cap->issued | cap->implemented,
+					  (cap->issued | cap->implemented),
 					  cf->caps, cf->tid, oldest_flush_tid);
 			if (ret) {
 				pr_err("kick_flushing_caps: error sending "