mds: fix rstat.rctime propagation

This patch fixes two issues.
- if (do_parent_mtime || linkunlink) is true, we need to call
  CDir::resync_accounted_rstat() before updating rstat.rctime.
- when primary_dn is false, we still need to propagate dirfrag's
  rstat to inode. Because rstat.rctime may have been changed.

Signed-off-by: Yan, Zheng <zyan@redhat.com>

diff --git a/src/mds/MDCache.cc b/src/mds/MDCache.cc
index 152b47b85ae4afd6ba19ef3c69c32cae234ea5dc..1ea9d8d3c715c7e9da52e7a6e2fd2735fcb1525c 100644 (file)
--- a/src/mds/MDCache.cc
+++ b/src/mds/MDCache.cc
@@ -2119,10 +2119,12 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
 
     if (do_parent_mtime || linkunlink) {
       assert(mut->wrlocks.count(&pin->filelock));
+      assert(mut->wrlocks.count(&pin->nestlock));
       assert(cfollows == CEPH_NOSNAP);
       
-      // update stale fragstat?
+      // update stale fragstat/rstat?
       parent->resync_accounted_fragstat();
+      parent->resync_accounted_rstat();
 
       if (do_parent_mtime) {
        pf->fragstat.mtime = mut->get_op_stamp();
@@ -2145,35 +2147,34 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
       }
     }
 
-    
     // rstat
     if (!primary_dn) {
       // don't update parent this pass
-    } else if (!linkunlink && !(parent->inode->nestlock.can_wrlock(-1) &&
-                               parent->inode->versionlock.can_wrlock())) {
-      dout(20) << " unwritable parent nestlock " << parent->inode->nestlock
-              << ", marking dirty rstat on " << *cur << dendl;
+    } else if (!linkunlink && !(pin->nestlock.can_wrlock(-1) &&
+                               pin->versionlock.can_wrlock())) {
+      dout(20) << " unwritable parent nestlock " << pin->nestlock
+       << ", marking dirty rstat on " << *cur << dendl;
       cur->mark_dirty_rstat();
-   } else {
+    } else {
       // if we don't hold a wrlock reference on this nestlock, take one,
       // because we are about to write into the dirfrag fnode and that needs
       // to commit before the lock can cycle.
-     if (linkunlink) {
-       assert(parent->inode->nestlock.get_num_wrlocks() || mut->is_slave());
-     }
+      if (linkunlink) {
+       assert(pin->nestlock.get_num_wrlocks() || mut->is_slave());
+      }
 
-      if (mut->wrlocks.count(&parent->inode->nestlock) == 0) {
-       dout(10) << " taking wrlock on " << parent->inode->nestlock << " on " << *parent->inode << dendl;
-       mds->locker->wrlock_force(&parent->inode->nestlock, mut);
+      if (mut->wrlocks.count(&pin->nestlock) == 0) {
+       dout(10) << " taking wrlock on " << pin->nestlock << " on " << *pin << dendl;
+       mds->locker->wrlock_force(&pin->nestlock, mut);
       }
 
       // now we can project the inode rstat diff the dirfrag
-      SnapRealm *prealm = parent->inode->find_snaprealm();
-      
+      SnapRealm *prealm = pin->find_snaprealm();
+
       snapid_t follows = cfollows;
       if (follows == CEPH_NOSNAP)
        follows = prealm->get_newest_seq();
-      
+
       snapid_t first = follows+1;
 
       // first, if the frag is stale, bring it back in sync.
@@ -2291,36 +2292,33 @@ void MDCache::predirty_journal_parents(MutationRef mut, EMetaBlob *blob,
     assert(parentdn);
 
     // rstat
-    if (primary_dn) {
+    dout(10) << "predirty_journal_parents frag->inode on " << *parent << dendl;
 
-      dout(10) << "predirty_journal_parents frag->inode on " << *parent << dendl;
+    // first, if the frag is stale, bring it back in sync.
+    parent->resync_accounted_rstat();
 
-      // first, if the frag is stale, bring it back in sync.
-      parent->resync_accounted_rstat();
+    if (g_conf->mds_snap_rstat) {
+      for (compact_map<snapid_t,old_rstat_t>::iterator p = parent->dirty_old_rstat.begin();
+          p != parent->dirty_old_rstat.end();
+          ++p)
+       project_rstat_frag_to_inode(p->second.rstat, p->second.accounted_rstat, p->second.first,
+                                   p->first, pin, true);//false);
+    }
+    parent->dirty_old_rstat.clear();
+    project_rstat_frag_to_inode(pf->rstat, pf->accounted_rstat, parent->first, CEPH_NOSNAP, pin, true);//false);
 
-      if (g_conf->mds_snap_rstat) {
-       for (compact_map<snapid_t,old_rstat_t>::iterator p = parent->dirty_old_rstat.begin();
-            p != parent->dirty_old_rstat.end();
-            ++p)
-         project_rstat_frag_to_inode(p->second.rstat, p->second.accounted_rstat, p->second.first,
-                                     p->first, pin, true);//false);
-      }
-      parent->dirty_old_rstat.clear();
-      project_rstat_frag_to_inode(pf->rstat, pf->accounted_rstat, parent->first, CEPH_NOSNAP, pin, true);//false);
+    pf->accounted_rstat = pf->rstat;
 
-      pf->accounted_rstat = pf->rstat;
+    if (parent->get_frag() == frag_t()) { // i.e., we are the only frag
+      if (pi->rstat.rbytes != pf->rstat.rbytes) {
+       mds->clog->error() << "unmatched rstat rbytes on single dirfrag "
+         << parent->dirfrag() << ", inode has " << pi->rstat
+         << ", dirfrag has " << pf->rstat << "\n";
 
-      if (parent->get_frag() == frag_t()) { // i.e., we are the only frag
-       if (pi->rstat.rbytes != pf->rstat.rbytes) { 
-         mds->clog->error() << "unmatched rstat rbytes on single dirfrag "
-             << parent->dirfrag() << ", inode has " << pi->rstat
-             << ", dirfrag has " << pf->rstat << "\n";
-         
-         // trust the dirfrag for now
-         pi->rstat = pf->rstat;
+       // trust the dirfrag for now
+       pi->rstat = pf->rstat;
 
-         assert(!"unmatched rstat rbytes" == g_conf->mds_verify_scatter);
-       }
+       assert(!"unmatched rstat rbytes" == g_conf->mds_verify_scatter);
       }
     }
 

diff --git a/src/mds/mdstypes.h b/src/mds/mdstypes.h
index e789856b36d7cb70e1ddb401383fdaf0021f3109..730b49f8704239773a490b1389c604df816dc86c 100644 (file)
--- a/src/mds/mdstypes.h
+++ b/src/mds/mdstypes.h
@@ -182,7 +182,7 @@ struct frag_info_t : public scatter_info_t {
 
   // *this += cur - acc;
   void add_delta(const frag_info_t &cur, frag_info_t &acc, bool& touched_mtime) {
-    if (!(cur.mtime == acc.mtime)) {
+    if (cur.mtime > mtime) {
       mtime = cur.mtime;
       touched_mtime = true;
     }