summaryrefslogtreecommitdiff
path: root/fs/ceph/snap.c
diff options
context:
space:
mode:
authorYan, Zheng <zyan@redhat.com>2015-05-01 16:57:16 +0800
committerIlya Dryomov <idryomov@gmail.com>2015-06-25 11:49:28 +0300
commit860560904962d08fd38666207c910065fe53e074 (patch)
tree88a9532f9652a11bb393a4496130c532705282b9 /fs/ceph/snap.c
parent5dda377cf0a6bd43f64a3c1efb670d7c668e7b29 (diff)
ceph: avoid sending unnessesary FLUSHSNAP message
when a snap notification contains no new snapshot, we can avoid sending FLUSHSNAP message to MDS. But we still need to create cap_snap in some case because it's required by write path and page writeback path Signed-off-by: Yan, Zheng <zyan@redhat.com>
Diffstat (limited to 'fs/ceph/snap.c')
-rw-r--r--fs/ceph/snap.c49
1 files changed, 37 insertions, 12 deletions
diff --git a/fs/ceph/snap.c b/fs/ceph/snap.c
index 5bfdab9a465e..ba708017d60b 100644
--- a/fs/ceph/snap.c
+++ b/fs/ceph/snap.c
@@ -436,6 +436,14 @@ static int dup_array(u64 **dst, __le64 *src, u32 num)
return 0;
}
+static bool has_new_snaps(struct ceph_snap_context *o,
+ struct ceph_snap_context *n)
+{
+ if (n->num_snaps == 0)
+ return false;
+ /* snaps are in descending order */
+ return n->snaps[0] > o->seq;
+}
/*
* When a snapshot is applied, the size/mtime inode metadata is queued
@@ -455,7 +463,7 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci)
{
struct inode *inode = &ci->vfs_inode;
struct ceph_cap_snap *capsnap;
- struct ceph_snap_context *old_snapc;
+ struct ceph_snap_context *old_snapc, *new_snapc;
int used, dirty;
capsnap = kzalloc(sizeof(*capsnap), GFP_NOFS);
@@ -469,6 +477,7 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci)
dirty = __ceph_caps_dirty(ci);
old_snapc = ci->i_head_snapc;
+ new_snapc = ci->i_snap_realm->cached_context;
/*
* If there is a write in progress, treat that as a dirty Fw,
@@ -486,20 +495,37 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci)
dout("queue_cap_snap %p already pending\n", inode);
goto update_snapc;
}
- if (ci->i_snap_realm->cached_context == ceph_empty_snapc) {
- dout("queue_cap_snap %p empty snapc\n", inode);
- goto update_snapc;
- }
- if (!(dirty & (CEPH_CAP_AUTH_EXCL|CEPH_CAP_XATTR_EXCL|
- CEPH_CAP_FILE_EXCL|CEPH_CAP_FILE_WR))) {
+ if (ci->i_wrbuffer_ref_head == 0 &&
+ !(dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))) {
dout("queue_cap_snap %p nothing dirty|writing\n", inode);
goto update_snapc;
}
BUG_ON(!old_snapc);
- dout("queue_cap_snap %p cap_snap %p queuing under %p %s\n",
- inode, capsnap, old_snapc, ceph_cap_string(dirty));
+ /*
+ * There is no need to send FLUSHSNAP message to MDS if there is
+ * no new snapshot. But when there is dirty pages or on-going
+ * writes, we still need to create cap_snap. cap_snap is needed
+ * by the write path and page writeback path.
+ *
+ * also see ceph_try_drop_cap_snap()
+ */
+ if (has_new_snaps(old_snapc, new_snapc)) {
+ if (dirty & (CEPH_CAP_ANY_EXCL|CEPH_CAP_FILE_WR))
+ capsnap->need_flush = true;
+ } else {
+ if (!(used & CEPH_CAP_FILE_WR) &&
+ ci->i_wrbuffer_ref_head == 0) {
+ dout("queue_cap_snap %p "
+ "no new_snap|dirty_page|writing\n", inode);
+ goto update_snapc;
+ }
+ }
+
+ dout("queue_cap_snap %p cap_snap %p queuing under %p %s %s\n",
+ inode, capsnap, old_snapc, ceph_cap_string(dirty),
+ capsnap->need_flush ? "" : "no_flush");
ihold(inode);
atomic_set(&capsnap->nref, 1);
@@ -549,9 +575,8 @@ void ceph_queue_cap_snap(struct ceph_inode_info *ci)
update_snapc:
if (ci->i_head_snapc) {
- ci->i_head_snapc = ceph_get_snap_context(
- ci->i_snap_realm->cached_context);
- dout(" new snapc is %p\n", ci->i_head_snapc);
+ ci->i_head_snapc = ceph_get_snap_context(new_snapc);
+ dout(" new snapc is %p\n", new_snapc);
}
spin_unlock(&ci->i_ceph_lock);