mds: allow rdlock on replica to request sync from auth

This fixes a ping-pong readdir between inode and dirfrag auth.

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index f3eef4ba2f6de1b11ee3b4b639c4f11ff5f256de..179df1e2fbff891dbf756d17c5e1c539861913ee 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -200,6 +200,11 @@ bool Locker::acquire_locks(MDRequest *mdr,
     sorted.insert(*p);
     if ((*p)->get_parent()->is_auth())
       mustpin.insert(*p);
+    else if ((*p)->get_type() == CEPH_LOCK_IFILE &&
+            !(*p)->get_parent()->is_auth() && !(*p)->can_rdlock(client)) { // we might have to request an rdlock
+      dout(15) << " will also auth_pin " << *(*p)->get_parent() << " in case we need to request a rdlock" << dendl;
+      mustpin.insert(*p);
+    }
   }
 
  
@@ -677,16 +682,24 @@ void Locker::eval(SimpleLock *lock, bool *need_issue)
 bool Locker::_rdlock_kick(SimpleLock *lock)
 {
   // kick the lock
-  if (lock->is_stable() &&
-      lock->get_parent()->is_auth()) {
-    if (lock->get_sm() == &sm_scatterlock) {
-      if (lock->get_parent()->is_replicated())
-       scatter_tempsync((ScatterLock*)lock);
-      else
+  if (lock->is_stable()) {
+    if (lock->get_parent()->is_auth()) {
+      if (lock->get_sm() == &sm_scatterlock) {
+       if (lock->get_parent()->is_replicated())
+         scatter_tempsync((ScatterLock*)lock);
+       else
+         simple_sync(lock);
+      } else 
        simple_sync(lock);
-    } else 
-      simple_sync(lock);
-    return true;
+      return true;
+    } else {
+      // request rdlock state change from auth
+      int auth = lock->get_parent()->authority().first;
+      dout(10) << "requesting rdlock from auth on " 
+              << *lock << " on " << *lock->get_parent() << dendl;
+      mds->send_message_mds(new MLock(lock, LOCK_AC_REQRDLOCK, mds->get_nodeid()), auth);
+      return false;
+    }
   }
   return false;
 }
@@ -737,8 +750,13 @@ bool Locker::rdlock_start(SimpleLock *lock, MDRequest *mut)
   }
 
   // wait!
+  int wait_on;
+  if (lock->get_parent()->is_auth())
+    wait_on = SimpleLock::WAIT_RD;
+  else
+    wait_on = SimpleLock::WAIT_STABLE;  // REQRDLOCK is ignored if lock is unstable, so we need to retry.
   dout(7) << "rdlock_start waiting on " << *lock << " on " << *lock->get_parent() << dendl;
-  lock->add_waiter(SimpleLock::WAIT_RD, new C_MDS_RetryRequest(mdcache, mut));
+  lock->add_waiter(wait_on, new C_MDS_RetryRequest(mdcache, mut));
   nudge_log(lock);
   return false;
 }
@@ -3564,6 +3582,19 @@ void Locker::handle_file_lock(ScatterLock *lock, MLock *m)
     }
     break;
 
+  case LOCK_AC_REQRDLOCK:
+    if (lock->is_stable()) {
+      dout(7) << "handle_file_lock got rdlock request on " << *lock
+             << " on " << *lock->get_parent() << dendl;
+      assert(in->is_auth()); // replica auth pinned if they're doing this!
+      simple_sync(lock);
+    } else {
+      dout(7) << "handle_file_lock ignoring rdlock request on " << *lock
+             << " on " << *lock->get_parent() << dendl;
+      // replica will retry.
+    }
+    break;
+
   case LOCK_AC_NUDGE:
     if (lock->get_parent()->is_auth()) {
       dout(7) << "handle_file_lock trying nudge on " << *lock
@@ -3576,7 +3607,6 @@ void Locker::handle_file_lock(ScatterLock *lock, MLock *m)
     }    
     break;
 
-
   default:
     assert(0);
   }  

diff --git a/src/mds/SimpleLock.h b/src/mds/SimpleLock.h
index 3561f1857f728213435116d9827add56168dabf2..43fb40d18cd48fe64b87554156e60a99cac254e7 100644 (file)
--- a/src/mds/SimpleLock.h
+++ b/src/mds/SimpleLock.h
@@ -315,6 +315,9 @@ public:
   bool fw_rdlock_to_auth() {
     return get_sm()->states[state].can_rdlock == FW;
   }
+  bool req_rdlock_from_auth() {
+    return get_sm()->states[state].can_rdlock == REQ;
+  }
 
   // gather set
   static set<int> empty_gather_set;

diff --git a/src/mds/locks.c b/src/mds/locks.c
index aa46bae2d1dc5c287a8d0bd4fc7d5f6679e84dbb..3ffdf75884972e2cef5589e2dd39150db1fd5c0d 100644 (file)
--- a/src/mds/locks.c
+++ b/src/mds/locks.c
@@ -18,7 +18,7 @@ struct sm_state_t simplelock[LOCK_MAX] = {
     [LOCK_LOCK_SYNC] = { LOCK_SYNC, false, LOCK_SYNC, ANY,  XCL, XCL, 0,   XCL, 0,   0,0,0,0 },
     [LOCK_EXCL_SYNC] = { LOCK_SYNC, true,  LOCK_LOCK, 0,    0,   0,   0,   0,   0,   0,CEPH_CAP_GSHARED,0,0 },
 
-    [LOCK_LOCK]      = { 0,         false, LOCK_LOCK, AUTH, 0,   0,   0,   0,   0,   0,0,0,0 },
+    [LOCK_LOCK]      = { 0,         false, LOCK_LOCK, AUTH, 0,   FW,  0,   0,   0,   0,0,0,0 },
     [LOCK_SYNC_LOCK] = { LOCK_LOCK, false, LOCK_LOCK, ANY,  0,   0,   0,   0,   0,   0,0,0,0 }, 
     [LOCK_EXCL_LOCK] = { LOCK_LOCK, false, LOCK_LOCK, 0,    0,   0,   0,   0,   0,   0,0,0,0 },
 
@@ -85,22 +85,22 @@ struct sm_state_t filelock[LOCK_MAX] = {
     [LOCK_MIX_SYNC]  = { LOCK_SYNC, false, LOCK_MIX,  0,    0,   0,   0,   0,   0,   CEPH_CAP_GRD,0,0,CEPH_CAP_GRD },
     [LOCK_MIX_SYNC2] = { LOCK_SYNC, false, 0,         0,    0,   0,   0,   0,   0,   CEPH_CAP_GRD,0,0,CEPH_CAP_GRD },
     
-    [LOCK_LOCK]      = { 0,         false, LOCK_LOCK, AUTH, 0,   FW,  AUTH,0,   0,   CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,0 },
-    [LOCK_SYNC_LOCK] = { LOCK_LOCK, false, LOCK_LOCK, AUTH, 0,   FW,  0,   0,   0,   CEPH_CAP_GCACHE,0,0,CEPH_CAP_GCACHE },
+    [LOCK_LOCK]      = { 0,         false, LOCK_LOCK, AUTH, 0,   REQ, AUTH,0,   0,   CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,0 },
+    [LOCK_SYNC_LOCK] = { LOCK_LOCK, false, LOCK_LOCK, AUTH, 0,   REQ, 0,   0,   0,   CEPH_CAP_GCACHE,0,0,CEPH_CAP_GCACHE },
     [LOCK_EXCL_LOCK] = { LOCK_LOCK, false, LOCK_LOCK, 0,    0,   0,   0,   0,   0,   CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,CEPH_CAP_GCACHE },
-    [LOCK_MIX_LOCK]  = { LOCK_LOCK, false, LOCK_LOCK, AUTH, 0,   FW,  0,   0,   0,   0,0,0,0 },
+    [LOCK_MIX_LOCK]  = { LOCK_LOCK, false, LOCK_LOCK, AUTH, 0,   REQ, 0,   0,   0,   0,0,0,0 },
 
     [LOCK_PREXLOCK]  = { LOCK_SYNC, false, LOCK_LOCK, 0,    XCL, 0,   0,   0,   ANY, CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,0 },
     [LOCK_XLOCK]     = { LOCK_SYNC, false, LOCK_LOCK, 0,    XCL, 0,   0,   0,   0,   CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,0 },
     [LOCK_XLOCKDONE] = { LOCK_SYNC, false, LOCK_LOCK, XCL,  XCL, XCL, 0,   XCL, 0,   CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,CEPH_CAP_GSHARED,0 },
     [LOCK_LOCK_XLOCK]= { LOCK_PREXLOCK,false,LOCK_LOCK,0,   XCL, 0,   0,   0,   XCL, CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0,0 },
 
-    [LOCK_MIX]       = { 0,         false, LOCK_MIX,  0,    0,   FW,  ANY, 0,   0,   CEPH_CAP_GRD|CEPH_CAP_GWR,0,0,CEPH_CAP_GRD },
-    [LOCK_SYNC_MIX]  = { LOCK_MIX,  false, LOCK_MIX,  ANY,  0,   FW,  0,   0,   0,   CEPH_CAP_GRD,0,0,CEPH_CAP_GRD },
+    [LOCK_MIX]       = { 0,         false, LOCK_MIX,  0,    0,   REQ, ANY, 0,   0,   CEPH_CAP_GRD|CEPH_CAP_GWR,0,0,CEPH_CAP_GRD },
+    [LOCK_SYNC_MIX]  = { LOCK_MIX,  false, LOCK_MIX,  ANY,  0,   REQ, 0,   0,   0,   CEPH_CAP_GRD,0,0,CEPH_CAP_GRD },
     [LOCK_SYNC_MIX2] = { LOCK_MIX,  false, 0,         ANY,  0,   0,   0,   0,   0,   CEPH_CAP_GRD,0,0,CEPH_CAP_GRD },
     [LOCK_EXCL_MIX]  = { LOCK_MIX,  true,  LOCK_LOCK, 0,    0,   0,   XCL, 0,   0,   0,CEPH_CAP_GRD|CEPH_CAP_GWR,0,0 },
     
-    [LOCK_EXCL]      = { 0,         true,  LOCK_LOCK, 0,    0,   FW,  XCL, 0,   0,   0,CEPH_CAP_GSHARED|CEPH_CAP_GEXCL|CEPH_CAP_GCACHE|CEPH_CAP_GRD|CEPH_CAP_GWR|CEPH_CAP_GBUFFER,0,0 },
+    [LOCK_EXCL]      = { 0,         true,  LOCK_LOCK, 0,    0,   AUTH,XCL, 0,   0,   0,CEPH_CAP_GSHARED|CEPH_CAP_GEXCL|CEPH_CAP_GCACHE|CEPH_CAP_GRD|CEPH_CAP_GWR|CEPH_CAP_GBUFFER,0,0 },
     [LOCK_SYNC_EXCL] = { LOCK_EXCL, true,  LOCK_LOCK, AUTH, 0,   0,   0,   0,   0,   0,CEPH_CAP_GSHARED|CEPH_CAP_GCACHE|CEPH_CAP_GRD,0,0 },
     [LOCK_MIX_EXCL]  = { LOCK_EXCL, true,  LOCK_LOCK, 0,    0,   0,   XCL, 0,   0,   0,CEPH_CAP_GRD|CEPH_CAP_GWR,0,0 },
     [LOCK_LOCK_EXCL] = { LOCK_EXCL, true,  LOCK_LOCK, AUTH, 0,   0,   0,   0,   0,   0,CEPH_CAP_GCACHE|CEPH_CAP_GBUFFER,0,0 },

diff --git a/src/mds/locks.h b/src/mds/locks.h
index 39f96fab56c127012a8123c5de495265b8ae5420..e6fba6a5468cef6f17479df494ca5aa91e1b41b1 100644 (file)
--- a/src/mds/locks.h
+++ b/src/mds/locks.h
@@ -30,6 +30,7 @@ struct sm_t {
 #define AUTH 2 // auth only
 #define XCL  3 // auth or exclusive client
 #define FW   4 // fw to auth, if replica
+#define REQ  5 // req state change from auth, if replica
 
 extern struct sm_t sm_simplelock;
 extern struct sm_t sm_filelock;
@@ -102,6 +103,7 @@ extern struct sm_t sm_locallock;
 #define LOCK_AC_REQSCATTER   7
 #define LOCK_AC_REQUNSCATTER 8
 #define LOCK_AC_NUDGE        9
+#define LOCK_AC_REQRDLOCK   10
 
 #define LOCK_AC_FOR_REPLICA(a)  ((a) < 0)
 #define LOCK_AC_FOR_AUTH(a)     ((a) > 0)
@@ -120,6 +122,7 @@ static inline const char *get_lock_action_name(int a) {
   case LOCK_AC_REQSCATTER: return "reqscatter";
   case LOCK_AC_REQUNSCATTER: return "requnscatter";
   case LOCK_AC_NUDGE: return "nudge";
+  case LOCK_AC_REQRDLOCK: return "reqrdlock";
   default: return "???";
   }
 }