mds: put notable caps at the front of session's caps list

Notable Capabilities are ones that are being revoked, ones that
have writeable ranges and ones that want exclusive caps or want
file read/write.

Signed-off-by: "Yan, Zheng" <zyan@redhat.com>
(cherry picked from commit cb6e7184458e64720ad83c266e5e393a80c32697)

 Conflicts:
	src/mds/Capability.cc

diff --git a/src/mds/CInode.cc b/src/mds/CInode.cc
index 0c89617aa7c4a6384c62d995238e375daa87b5e9..c834fc427e46c00ea7ddfa30b0edc2bf4efaee9b 100644 (file)
--- a/src/mds/CInode.cc
+++ b/src/mds/CInode.cc
@@ -2816,7 +2816,6 @@ Capability *CInode::add_client_cap(client_t client, Session *session, SnapRealm
   assert(client_caps.count(client) == 0);
   client_caps[client] = cap;
 
-  session->add_cap(cap);
   cap->client_follows = first-1;
   
   containing_realm->add_cap(client, cap);

diff --git a/src/mds/Capability.cc b/src/mds/Capability.cc
index 5c50b59eb7add645cd80c3b09b9ba049104aee5d..98be60f079f4b5c21de86b4aedef740e93b36106 100644 (file)
--- a/src/mds/Capability.cc
+++ b/src/mds/Capability.cc
@@ -152,6 +152,9 @@ Capability::Capability(CInode *i, Session *s, uint64_t id) :
   _pending(0), _issued(0), last_sent(0), last_issue(0), mseq(0),
   suppress(0), state(0)
 {
+  if (session) {
+    session->touch_cap_bottom(this);
+  }
 }
 
 client_t Capability::get_client() const
@@ -164,6 +167,36 @@ bool Capability::is_stale() const
   return session ? session->is_stale() : false;
 }
 
+void Capability::mark_notable()
+{
+  state |= STATE_NOTABLE;
+  session->touch_cap(this);
+}
+
+void Capability::maybe_clear_notable()
+{
+  if ((_issued == _pending) &&
+      !is_clientwriteable() &&
+      !is_wanted_notable(_wanted)) {
+    ceph_assert(is_notable());
+    state &= ~STATE_NOTABLE;
+    session->touch_cap_bottom(this);
+  }
+}
+
+void Capability::set_wanted(int w) {
+  CInode *in = get_inode();
+  if (in) {
+    if (!is_wanted_notable(_wanted) && is_wanted_notable(w)) {
+      if (!is_notable())
+       mark_notable();
+    } else if (is_wanted_notable(_wanted) && !is_wanted_notable(w)) {
+      maybe_clear_notable();
+    }
+  }
+  _wanted = w;
+}
+
 void Capability::encode(bufferlist& bl) const
 {
   ENCODE_START(2, 2, bl)
@@ -187,7 +220,7 @@ void Capability::decode(bufferlist::iterator &bl)
   ::decode(_revokes, bl);
   DECODE_FINISH(bl);
   
-  _calc_issued();
+  calc_issued();
 }
 
 void Capability::dump(Formatter *f) const

diff --git a/src/mds/Capability.h b/src/mds/Capability.h
index af7ca015a4664948ef1486619b2b9ae9c4ef6ef1..4ff5b811fc852eb91f8c41e8d180eeeb882a4c65 100644 (file)
--- a/src/mds/Capability.h
+++ b/src/mds/Capability.h
@@ -110,6 +110,7 @@ public:
     static void generate_test_instances(list<revoke_info*>& ls);
   };
 
+  const static unsigned STATE_NOTABLE          = (1<<0);
   const static unsigned STATE_NEW              = (1<<1);
   const static unsigned STATE_IMPORTING                = (1<<2);
   const static unsigned STATE_CLIENTWRITEABLE  = (1<<4);
@@ -129,6 +130,8 @@ public:
       _revokes.emplace_back(_pending, last_sent, last_issue);
       _pending = c;
       _issued |= c;
+      if (!is_notable())
+       mark_notable();
     } else if (~_pending & c) {
       // adding bits only.  remove obsolete revocations?
       _pending |= c;
@@ -152,13 +155,8 @@ public:
     inc_last_seq();
     return last_sent;
   }
-  void _calc_issued() {
-    _issued = _pending;
-    for (const auto &r : _revokes) {
-      _issued |= r.before;
-    }
-  }
   void confirm_receipt(ceph_seq_t seq, unsigned caps) {
+    bool was_revoking = (_issued & ~_pending);
     if (seq == last_sent) {
       _revokes.clear();
       _issued = caps;
@@ -171,16 +169,17 @@ public:
       if (!_revokes.empty()) {
        if (_revokes.front().seq == seq)
          _revokes.begin()->before = caps;
-       _calc_issued();
+       calc_issued();
       } else {
        // seq < last_sent
        _issued = caps | _pending;
       }
     }
 
-    if (_issued == _pending) {
+    if (was_revoking && _issued == _pending) {
       item_revoking_caps.remove_myself();
       item_client_revoking_caps.remove_myself();
+      maybe_clear_notable();
     }
     //check_rdcaps_list();
   }
@@ -193,10 +192,12 @@ public:
       changed = true;
     }
     if (changed) {
-      _calc_issued();
-      if (_issued == _pending) {
+      bool was_revoking = (_issued & ~_pending);
+      calc_issued();
+      if (was_revoking && _issued == _pending) {
        item_revoking_caps.remove_myself();
        item_client_revoking_caps.remove_myself();
+       maybe_clear_notable();
       }
     }
   }
@@ -223,6 +224,11 @@ public:
   void inc_suppress() { suppress++; }
   void dec_suppress() { suppress--; }
 
+  static bool is_wanted_notable(int wanted) {
+    return wanted & (CEPH_CAP_ANY_WR|CEPH_CAP_FILE_WR|CEPH_CAP_FILE_RD);
+  }
+  bool is_notable() const { return state & STATE_NOTABLE; }
+
   bool is_stale() const;
   bool is_new() const { return state & STATE_NEW; }
   void mark_new() { state |= STATE_NEW; }
@@ -235,11 +241,14 @@ public:
   void mark_clientwriteable() {
     if (!is_clientwriteable()) {
       state |= STATE_CLIENTWRITEABLE;
+      if (!is_notable())
+       mark_notable();
     }
   }
   void clear_clientwriteable() {
     if (is_clientwriteable()) {
       state &= ~STATE_CLIENTWRITEABLE;
+      maybe_clear_notable();
     }
   }
 
@@ -248,11 +257,8 @@ public:
   client_t get_client() const;
 
   // caps this client wants to hold
-  int wanted() { return _wanted; }
-  void set_wanted(int w) {
-    _wanted = w;
-    //check_rdcaps_list();
-  }
+  int wanted() const { return _wanted; }
+  void set_wanted(int w);
 
   void inc_last_seq() { last_sent++; }
   ceph_seq_t get_last_seq() { return last_sent; }
@@ -283,7 +289,7 @@ public:
     client_follows = other.client_follows;
 
     // wanted
-    _wanted = _wanted | other.wanted;
+    set_wanted(wanted() | other.wanted);
     if (auth_cap)
       mseq = other.mseq;
   }
@@ -300,7 +306,7 @@ public:
     }
 
     // wanted
-    _wanted = _wanted | otherwanted;
+    set_wanted(wanted() | otherwanted);
   }
 
   void revoke() {
@@ -350,6 +356,16 @@ private:
 
   int suppress;
   unsigned state;
+
+  void calc_issued() {
+    _issued = _pending;
+    for (const auto &r : _revokes) {
+      _issued |= r.before;
+    }
+  }
+
+  void mark_notable();
+  void maybe_clear_notable();
 };
 
 WRITE_CLASS_ENCODER(Capability::Export)

diff --git a/src/mds/SessionMap.h b/src/mds/SessionMap.h
index eee382f4af7849033909b0610f5182c00fa91ef8..404c456d770237e9657206d0d4e90be75e81a49d 100644 (file)
--- a/src/mds/SessionMap.h
+++ b/src/mds/SessionMap.h
@@ -265,7 +265,10 @@ public:
     }
   }
 
-  void add_cap(Capability *cap) {
+  void touch_cap(Capability *cap) {
+    caps.push_front(&cap->item_session_caps);
+  }
+  void touch_cap_bottom(Capability *cap) {
     caps.push_back(&cap->item_session_caps);
   }
   void touch_lease(ClientLease *r) {