mds: fix bug filelock stuck at LOCK_XSYN leading client can't read data

The file's wanted_loner_cap will be -1 when more than one client open
it.Sometimes the loner can't be dropped successfully which will leading
filelock stuck at LOCK_XSYN.However,this state doesn't allow others any
caps so the client open the file by O_RDONLY can't read data all the time
untill someone else open it by O_RDWR or O_WRONLY.We should change the
state to LOCK_MIX or LOCK_SYNC instead of stucking all the time.

Fixes: http://tracker.ceph.com/issues/37333
Signed-off-by: Guan yunfei <yunfei.guan@xtaotech.com>
(cherry picked from commit c3339d93f04030197d075e8a9c7235dba08530cf)

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 058fcae7642c5aa9fe89750b007b541fc3001126..feb00ac83c49823eddbf187777a2dd012df20de5 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -4871,7 +4871,7 @@ void Locker::file_eval(ScatterLock *lock, bool *need_issue)
   else if (lock->get_state() != LOCK_SYNC &&
           !lock->is_wrlocked() &&   // drain wrlocks first!
           !lock->is_waiter_for(SimpleLock::WAIT_WR) &&
-          !(wanted & (CEPH_CAP_GWR|CEPH_CAP_GBUFFER)) &&
+          !(wanted & CEPH_CAP_GWR) &&
           !((lock->get_state() == LOCK_MIX) &&
             in->is_dir() && in->has_subtree_or_exporting_dirfrag())  // if we are a delegation point, stay where we are
           //((wanted & CEPH_CAP_RD) ||