Merge pull request #690 from ceph/wip-fuse

a few ceph-fuse fixes

Reviewed-by: Sage Weil <sage@inktank.com>
Reviewed-by: Yan, Zheng <zheng.z.yan@intel.com>

5 files changed, 128 insertions, 22 deletions

diff --git a/fusetrace/fusetrace_ll.cc b/fusetrace/fusetrace_ll.cc
index eb7100a867f..7f2b8438f1f 100644
--- a/fusetrace/fusetrace_ll.cc
+++ b/fusetrace/fusetrace_ll.cc
@@ -11,7 +11,7 @@
     gcc -Wall `pkg-config fuse --cflags --libs` -lulockmgr fusexmp_fh.c -o fusexmp_fh
 */
 
-#define FUSE_USE_VERSION 26
+#define FUSE_USE_VERSION 30
 
 #ifdef HAVE_CONFIG_H
 #include <config.h>
diff --git a/src/client/Client.cc b/src/client/Client.cc
index 77fd2084cf1..60a5e4550b8 100644
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -148,9 +148,12 @@ Client::Client(Messenger *m, MonClient *mc)
     timer(m->cct, client_lock),
     ino_invalidate_cb(NULL),
     ino_invalidate_cb_handle(NULL),
+    dentry_invalidate_cb(NULL),
+    dentry_invalidate_cb_handle(NULL),
     getgroups_cb(NULL),
     getgroups_cb_handle(NULL),
     async_ino_invalidator(m->cct),
+    async_dentry_invalidator(m->cct),
     tick_event(NULL),
     monclient(mc), messenger(m), whoami(m->get_myname().num()),
     initialized(false), mounted(false), unmounting(false),
@@ -410,11 +413,17 @@ void Client::shutdown()
   admin_socket->unregister_command("dump_cache");
 
   if (ino_invalidate_cb) {
-    ldout(cct, 10) << "shutdown stopping invalidator finisher" << dendl;
+    ldout(cct, 10) << "shutdown stopping cache invalidator finisher" << dendl;
     async_ino_invalidator.wait_for_empty();
     async_ino_invalidator.stop();
   }
 
+  if (dentry_invalidate_cb) {
+    ldout(cct, 10) << "shutdown stopping dentry invalidator finisher" << dendl;
+    async_dentry_invalidator.wait_for_empty();
+    async_dentry_invalidator.stop();
+  }
+
   objectcacher->stop();  // outside of client_lock! this does a join.
 
   client_lock.Lock();
@@ -1532,7 +1541,7 @@ void Client::_closed_mds_session(MetaSession *s)
   signal_context_list(s->waiting_for_open);
   mount_cond.Signal();
   remove_session_caps(s);
-  kick_requests(s, true);
+  kick_requests_closed(s);
   mds_sessions.erase(s->mds_num);
   delete s;
 }
@@ -1905,7 +1914,7 @@ void Client::handle_mds_map(MMDSMap* m)
 
     if (newstate >= MDSMap::STATE_ACTIVE) {
       if (oldstate < MDSMap::STATE_ACTIVE) {
-	kick_requests(p->second, false);
+	kick_requests(p->second);
 	kick_flushing_caps(p->second);
 	signal_context_list(p->second->waiting_for_open);
 	kick_maxsize_requests(p->second);
@@ -1989,25 +1998,16 @@ void Client::send_reconnect(MetaSession *session)
 }
 
 
-void Client::kick_requests(MetaSession *session, bool signal)
+void Client::kick_requests(MetaSession *session)
 {
   ldout(cct, 10) << "kick_requests for mds." << session->mds_num << dendl;
-
   for (map<tid_t, MetaRequest*>::iterator p = mds_requests.begin();
        p != mds_requests.end();
-       ++p) 
+       ++p) {
     if (p->second->mds == session->mds_num) {
-      if (signal) {
-	// only signal caller if there is a caller
-	// otherwise, let resend_unsafe handle it
-	if (p->second->caller_cond) {
-	  p->second->kick = true;
-	  p->second->caller_cond->Signal();
-	}
-      } else {
-	send_request(p->second, session);
-      }
+      send_request(p->second, session);
     }
+  }
 }
 
 void Client::resend_unsafe_requests(MetaSession *session)
@@ -2018,6 +2018,25 @@ void Client::resend_unsafe_requests(MetaSession *session)
     send_request(*iter, session);
 }
 
+void Client::kick_requests_closed(MetaSession *session)
+{
+  ldout(cct, 10) << "kick_requests_closed for mds." << session->mds_num << dendl;
+  for (map<tid_t, MetaRequest*>::iterator p = mds_requests.begin();
+       p != mds_requests.end();
+       ++p) {
+    if (p->second->mds == session->mds_num) {
+      if (p->second->caller_cond) {
+	p->second->kick = true;
+	p->second->caller_cond->Signal();
+      }
+      p->second->item.remove_myself();
+      p->second->unsafe_item.remove_myself();
+    }
+  }
+  assert(session->requests.empty());
+  assert(session->unsafe_requests.empty());
+}
+
 
 
 
@@ -3551,6 +3570,45 @@ void Client::handle_cap_flushsnap_ack(MetaSession *session, Inode *in, MClientCa
   m->put();
 }
 
+class C_Client_DentryInvalidate : public Context  {
+private:
+  Client *client;
+  vinodeno_t dirino;
+  vinodeno_t ino;
+  string name;
+public:
+  C_Client_DentryInvalidate(Client *c, Dentry *dn) :
+			    client(c), dirino(dn->dir->parent_inode->vino()),
+			    ino(dn->inode->vino()), name(dn->name) { }
+  void finish(int r) {
+    client->_async_dentry_invalidate(dirino, ino, name);
+  }
+};
+
+void Client::_async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name)
+{
+  ldout(cct, 10) << "_async_dentry_invalidate '" << name << "' ino " << ino
+		 << " in dir " << dirino << dendl;
+  dentry_invalidate_cb(dentry_invalidate_cb_handle, dirino, ino, name);
+}
+
+void Client::_schedule_invalidate_dentry_callback(Dentry *dn)
+{
+  if (dentry_invalidate_cb && dn->inode->ll_ref > 0)
+    async_dentry_invalidator.queue(new C_Client_DentryInvalidate(this, dn));
+}
+
+void Client::_invalidate_inode_parents(Inode *in)
+{
+  set<Dentry*>::iterator q = in->dn_set.begin();
+  while (q != in->dn_set.end()) {
+    Dentry *dn = *q++;
+    // FIXME: we play lots of unlink/link tricks when handling MDS replies,
+    //        so in->dn_set doesn't always reflect the state of kernel's dcache.
+    _schedule_invalidate_dentry_callback(dn);
+    unlink(dn, false);
+  }
+}
 
 void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, MClientCaps *m)
 {
@@ -3578,8 +3636,12 @@ void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, MClient
     in->uid = m->head.uid;
     in->gid = m->head.gid;
   }
+  bool deleted_inode = false;
   if ((issued & CEPH_CAP_LINK_EXCL) == 0) {
     in->nlink = m->head.nlink;
+    if (in->nlink == 0 &&
+	(new_caps & (CEPH_CAP_LINK_SHARED | CEPH_CAP_LINK_EXCL)))
+      deleted_inode = true;
   }
   if ((issued & CEPH_CAP_XATTR_EXCL) == 0 &&
       m->xattrbl.length() &&
@@ -3633,6 +3695,10 @@ void Client::handle_cap_grant(MetaSession *session, Inode *in, Cap *cap, MClient
   if (new_caps)
     signal_cond_list(in->waitfor_caps);
 
+  // may drop inode's last ref
+  if (deleted_inode)
+    _invalidate_inode_parents(in);
+
   m->put();
 }
 
@@ -6319,6 +6385,17 @@ void Client::ll_register_ino_invalidate_cb(client_ino_callback_t cb, void *handl
   async_ino_invalidator.start();
 }
 
+void Client::ll_register_dentry_invalidate_cb(client_dentry_callback_t cb, void *handle)
+{
+  Mutex::Locker l(client_lock);
+  ldout(cct, 10) << "ll_register_dentry_invalidate_cb cb " << (void*)cb << " p " << (void*)handle << dendl;
+  if (cb == NULL)
+    return;
+  dentry_invalidate_cb = cb;
+  dentry_invalidate_cb_handle = handle;
+  async_dentry_invalidator.start();
+}
+
 void Client::ll_register_getgroups_cb(client_getgroups_callback_t cb, void *handle)
 {
   Mutex::Locker l(client_lock);
diff --git a/src/client/Client.h b/src/client/Client.h
index c7c9cef0e0c..df59f235de4 100644
--- a/src/client/Client.h
+++ b/src/client/Client.h
@@ -120,6 +120,9 @@ struct MetaRequest;
 
 typedef void (*client_ino_callback_t)(void *handle, vinodeno_t ino, int64_t off, int64_t len);
 
+typedef void (*client_dentry_callback_t)(void *handle, vinodeno_t dirino,
+					 vinodeno_t ino, string& name);
+
 typedef int (*client_getgroups_callback_t)(void *handle, uid_t uid, gid_t **sgids);
 
 // ========================================================
@@ -211,10 +214,14 @@ class Client : public Dispatcher {
   client_ino_callback_t ino_invalidate_cb;
   void *ino_invalidate_cb_handle;
 
+  client_dentry_callback_t dentry_invalidate_cb;
+  void *dentry_invalidate_cb_handle;
+
   client_getgroups_callback_t getgroups_cb;
   void *getgroups_cb_handle;
 
   Finisher async_ino_invalidator;
+  Finisher async_dentry_invalidator;
 
   Context *tick_event;
   utime_t last_cap_renew;
@@ -270,7 +277,8 @@ public:
   void connect_mds_targets(int mds);
   void send_request(MetaRequest *request, MetaSession *session);
   MClientRequest *build_client_request(MetaRequest *request);
-  void kick_requests(MetaSession *session, bool signal);
+  void kick_requests(MetaSession *session);
+  void kick_requests_closed(MetaSession *session);
   void handle_client_request_forward(MClientRequestForward *reply);
   void handle_client_reply(MClientReply *reply);
 
@@ -357,6 +365,7 @@ protected:
 
   friend class C_Client_PutInode; // calls put_inode()
   friend class C_Client_CacheInvalidate;  // calls ino_invalidate_cb
+  friend class C_Client_DentryInvalidate;  // calls dentry_invalidate_cb
 
   //int get_cache_size() { return lru.lru_get_size(); }
   //void set_cache_size(int m) { lru.lru_set_max(m); }
@@ -459,6 +468,10 @@ protected:
   void finish_cap_snap(Inode *in, CapSnap *capsnap, int used);
   void _flushed_cap_snap(Inode *in, snapid_t seq);
 
+  void _schedule_invalidate_dentry_callback(Dentry *dn);
+  void _async_dentry_invalidate(vinodeno_t dirino, vinodeno_t ino, string& name);
+  void _invalidate_inode_parents(Inode *in);
+
   void _schedule_invalidate_callback(Inode *in, int64_t off, int64_t len, bool keep_caps);
   void _invalidate_inode_cache(Inode *in, bool keep_caps);
   void _invalidate_inode_cache(Inode *in, int64_t off, int64_t len, bool keep_caps);
@@ -735,6 +748,8 @@ public:
 
   void ll_register_ino_invalidate_cb(client_ino_callback_t cb, void *handle);
 
+  void ll_register_dentry_invalidate_cb(client_dentry_callback_t cb, void *handle);
+
   void ll_register_getgroups_cb(client_getgroups_callback_t cb, void *handle);
 };
 
diff --git a/src/client/fuse_ll.cc b/src/client/fuse_ll.cc
index 6bf5ea3d34f..88f727e454e 100644
--- a/src/client/fuse_ll.cc
+++ b/src/client/fuse_ll.cc
@@ -12,7 +12,7 @@
  * 
  */
 
-#define FUSE_USE_VERSION 26
+#define FUSE_USE_VERSION 30
 
 #include <fuse/fuse.h>
 #include <fuse/fuse_lowlevel.h>
@@ -551,7 +551,7 @@ static int getgroups_cb(void *handle, uid_t uid, gid_t **sgids)
 }
 #endif
 
-static void invalidate_cb(void *handle, vinodeno_t vino, int64_t off, int64_t len)
+static void ino_invalidate_cb(void *handle, vinodeno_t vino, int64_t off, int64_t len)
 {
 #if FUSE_VERSION >= FUSE_MAKE_VERSION(2, 8)
   CephFuse::Handle *cfuse = (CephFuse::Handle *)handle;
@@ -560,6 +560,19 @@ static void invalidate_cb(void *handle, vinodeno_t vino, int64_t off, int64_t le
 #endif
 }
 
+static void dentry_invalidate_cb(void *handle, vinodeno_t dirino,
+				 vinodeno_t ino, string& name)
+{
+  CephFuse::Handle *cfuse = (CephFuse::Handle *)handle;
+  fuse_ino_t fdirino = cfuse->make_fake_ino(dirino.ino, dirino.snapid);
+#if FUSE_VERSION >= FUSE_MAKE_VERSION(2, 9)
+  fuse_ino_t fino = cfuse->make_fake_ino(ino.ino, ino.snapid);
+  fuse_lowlevel_notify_delete(cfuse->ch, fdirino, fino, name.c_str(), name.length());
+#elif FUSE_VERSION >= FUSE_MAKE_VERSION(2, 8)
+  fuse_lowlevel_notify_inval_entry(cfuse->ch, fdirino, name.c_str(), name.length());
+#endif
+}
+
 static void do_init(void *data, fuse_conn_info *bar)
 {
   CephFuse::Handle *cfuse = (CephFuse::Handle *)data;
@@ -743,9 +756,10 @@ int CephFuse::Handle::init(int argc, const char *argv[])
   client->ll_register_getgroups_cb(getgroups_cb, this);
 
    */
+  client->ll_register_dentry_invalidate_cb(dentry_invalidate_cb, this);
 
   if (client->cct->_conf->fuse_use_invalidate_cb)
-    client->ll_register_ino_invalidate_cb(invalidate_cb, this);
+    client->ll_register_ino_invalidate_cb(ino_invalidate_cb, this);
 
 done:
   fuse_opt_free_args(&args);
diff --git a/src/rbd_fuse/rbd-fuse.c b/src/rbd_fuse/rbd-fuse.c
index eea6edb9eb8..2a6a8d22e81 100644
--- a/src/rbd_fuse/rbd-fuse.c
+++ b/src/rbd_fuse/rbd-fuse.c
@@ -1,7 +1,7 @@
 /*
  * rbd-fuse
  */
-#define FUSE_USE_VERSION 26
+#define FUSE_USE_VERSION 30
 
 #include "include/int_types.h"