mds: some remote rename fixes, more to come

diff --git a/src/TODO b/src/TODO
index 6c78835711039fc8412e6a3d6ea4244c5429d39c..e0a9be2cfe753b8ee875881771e18e9e002b4252 100644 (file)
--- a/src/TODO
+++ b/src/TODO
@@ -70,6 +70,10 @@ mds nested
 - fix rejoin vs updated dirfrag nested/dirlocks
 
 mds mustfix
+- rollback vs accounting is just broken!  can we avoid it.
+  - even with _just_ parent dir mtime, the rollback metablob isn't sufficient during replay.. we need special handling regardless.
+- when renaming a dir, how do we know if we need to update the anchor table?
+
 - replay of dir fragmentation  (dont want dir frozen, pins, etc.?)
 - fix rm -r vs mds exports
 - rename slave in-memory rollback on failure

diff --git a/src/mds/Locker.cc b/src/mds/Locker.cc
index 6693336598477add4bf8960e18a99148570e581d..4b7cd3efee0f26d287a0c95361786de22e30ec63 100644 (file)
--- a/src/mds/Locker.cc
+++ b/src/mds/Locker.cc
@@ -1232,6 +1232,14 @@ void Locker::predirty_nested(Mutation *mut, EMetaBlob *blob,
     parent = in->get_projected_parent_dn()->get_dir();
   }
 
+  if (flags == 0) {
+    dout(10) << " no flags, just adding dir context to blob(s)" << dendl;
+    blob->add_dir_context(parent);
+    if (rollback)
+      rollback->add_dir_context(parent);
+    return;
+  }
+
   inode_t *curi = in->get_projected_inode();
 
   __s64 drbytes = 1, drfiles = 0, drsubdirs = 0;
@@ -1343,7 +1351,7 @@ void Locker::predirty_nested(Mutation *mut, EMetaBlob *blob,
   blob->add_dir_context(parent);
   blob->add_dir(parent, true);
   if (rollback)
-    blob->add_dir_context(parent);
+    rollback->add_dir_context(parent);
   for (list<CInode*>::iterator p = lsi.begin();
        p != lsi.end();
        p++) {

diff --git a/src/mds/Server.cc b/src/mds/Server.cc
index 375571c678cc882c81ea41d6be211e0027f4beb8..3507b2e47c80a0b5b54c46d5941e6105696cde64 100644 (file)
--- a/src/mds/Server.cc
+++ b/src/mds/Server.cc
@@ -3223,7 +3223,7 @@ void Server::handle_client_rename(MDRequest *mdr)
   }
   
   // -- prepare anchor updates -- 
-  if (!linkmerge) {
+  if (!linkmerge || srcdn->is_primary()) {
     C_Gather *anchorgather = 0;
 
     if (srcdn->is_primary() && srcdn->inode->is_anchored() &&
@@ -3371,7 +3371,8 @@ version_t Server::_rename_prepare_import(MDRequest *mdr, CDentry *srcdn, bufferl
 
 void Server::_rename_prepare(MDRequest *mdr,
                             EMetaBlob *metablob, bufferlist *client_map_bl,
-                            CDentry *srcdn, CDentry *destdn, CDentry *straydn)
+                            CDentry *srcdn, CDentry *destdn, CDentry *straydn,
+                            EMetaBlob *rollback)
 {
   dout(10) << "_rename_prepare " << *mdr << " " << *srcdn << " " << *destdn << dendl;
   if (straydn) dout(10) << " straydn " << *straydn << dendl;
@@ -3458,18 +3459,18 @@ void Server::_rename_prepare(MDRequest *mdr,
     // sub off target
     int predirty_dir = silent ? 0:PREDIRTY_DIR;
     if (!linkmerge && destdn->is_primary())
-      mds->locker->predirty_nested(mdr, metablob, destdn->inode, destdn->dir, PREDIRTY_PRIMARY|predirty_dir, -1);
+      mds->locker->predirty_nested(mdr, metablob, destdn->inode, destdn->dir, PREDIRTY_PRIMARY|predirty_dir, -1,
+                                  rollback);
     if (destdn->dir == srcdn->dir) {
       // same dir.  don't update nested info or adjust counts.
-      mds->locker->predirty_nested(mdr, metablob, srcdn->inode, srcdn->dir,
-                                  false, true);
+      mds->locker->predirty_nested(mdr, metablob, srcdn->inode, srcdn->dir, predirty_dir, 0, rollback);
     } else {
       // different dir.  update nested accounting.
       int flags = srcdn->is_primary() ? PREDIRTY_PRIMARY:0;
       flags |= predirty_dir;
       if (srcdn->is_auth())
-       mds->locker->predirty_nested(mdr, metablob, srcdn->inode, srcdn->dir, flags, -1);
-      mds->locker->predirty_nested(mdr, metablob, srcdn->inode, destdn->dir, flags, 1);
+       mds->locker->predirty_nested(mdr, metablob, srcdn->inode, srcdn->dir, flags, -1, rollback);
+      mds->locker->predirty_nested(mdr, metablob, srcdn->inode, destdn->dir, flags, 1, rollback);
     }
   }
 
@@ -3682,8 +3683,10 @@ void Server::handle_slave_rename_prep(MDRequest *mdr)
   mdr->pin(srcdn->inode);
 
   // stray?
+  bool linkmerge = (srcdn->inode == destdn->inode &&
+                   (srcdn->is_primary() || destdn->is_primary()));
   CDentry *straydn = 0;
-  if (destdn->is_primary()) {
+  if (destdn->is_primary() && !linkmerge) {
     assert(mdr->slave_request->stray.length() > 0);
     straydn = mdcache->add_replica_stray(mdr->slave_request->stray, 
                                         destdn->inode, mdr->slave_to_mds);
@@ -3746,22 +3749,20 @@ void Server::handle_slave_rename_prep(MDRequest *mdr)
     mdr->ls = mdlog->get_current_segment();
     ESlaveUpdate *le = new ESlaveUpdate(mdlog, "slave_rename_prep", mdr->reqid, mdr->slave_to_mds, ESlaveUpdate::OP_PREPARE);
 
+    // commit case
+    bufferlist blah;
+    _rename_prepare(mdr, &le->commit, &blah, srcdn, destdn, straydn, &le->rollback);
+
     // rollback case
     if (destdn->inode && destdn->inode->is_auth()) {
       assert(destdn->is_remote());
-      le->rollback.add_dir_context(destdn->dir);
       le->rollback.add_dentry(destdn, true);
     }
     if (srcdn->is_auth() ||
        (srcdn->inode && srcdn->inode->is_auth())) {
-      le->rollback.add_dir_context(srcdn->dir);
       le->rollback.add_dentry(srcdn, true);
     }
 
-    // commit case
-    bufferlist blah;
-    _rename_prepare(mdr, &le->commit, &blah, srcdn, destdn, straydn);
-
     mdlog->submit_entry(le, new C_MDS_SlaveRenamePrep(this, mdr, srcdn, destdn, straydn));
   } else {
     // don't journal.

diff --git a/src/mds/Server.h b/src/mds/Server.h
index 12c7b85606523b0578281ca014683019894c8646..01ebc3b941226212605246b0db6a2ae8b1dd74da 100644 (file)
--- a/src/mds/Server.h
+++ b/src/mds/Server.h
@@ -169,7 +169,8 @@ public:
   version_t _rename_prepare_import(MDRequest *mdr, CDentry *srcdn, bufferlist *client_map_bl);
   void _rename_prepare(MDRequest *mdr,
                       EMetaBlob *metablob, bufferlist *client_map_bl,
-                      CDentry *srcdn, CDentry *destdn, CDentry *straydn);
+                      CDentry *srcdn, CDentry *destdn, CDentry *straydn,
+                      EMetaBlob *rollback=0);
   void _rename_apply(MDRequest *mdr, CDentry *srcdn, CDentry *destdn, CDentry *straydn); 
 
   // slaving

diff --git a/src/vstartnew.sh b/src/vstartnew.sh
index 6fc0eab805bb5b45409becb21c90230a1d932e9d..73f7b03b845f9f049c8bbfe53400f146928c60e7 100755 (executable)
--- a/src/vstartnew.sh
+++ b/src/vstartnew.sh
@@ -49,7 +49,7 @@ done
 
 # mds
 $CEPH_BIN/cmds $ARGS --debug_ms 1 --debug_mds 20 --mds_thrash_fragments 0 --mds_thrash_exports 0 #--debug_ms 20
-#$CEPH_BIN/cmds $ARGS --debug_ms 1 --debug_mds 20 --mds_thrash_fragments 0 --mds_thrash_exports 0 #--debug_ms 20
+$CEPH_BIN/cmds $ARGS --debug_ms 1 --debug_mds 20 --mds_thrash_fragments 0 --mds_thrash_exports 0 #--debug_ms 20
 ./cmonctl mds set_max_mds 2
 
 echo "started.  stop.sh to stop.  see out/* (e.g. 'tail -f out/????') for debug output."