Ceph » CephFS

Been working on this today; I haven't run across any big architectural issues so far, just a lot of asserts and bad assumptions. Continuing to work through them.

Yay, so close!

Pushed a couple cleanups that drop the projected xattr/snaprealm pointers.

I'm worried about this hunk

@@ -507,8 +507,10 @@ private:
   inode_t *add_primary_dentry(CDentry *dn, bool dirty, 
                   CInode *in=0, fragtree_t *pdft=0, bufferlist *psnapbl=0,
                   map<string,bufferptr> *px=0) {
-    return add_primary_dentry(add_dir(dn->get_dir(), false),
+    if (dn)
+      return add_primary_dentry(add_dir(dn->get_dir(), false),
                   dn, dirty, in, pdft, psnapbl, px);
+    else return NULL;
   }
   inode_t *add_primary_dentry(dirlump& lump, CDentry *dn, bool dirty, 
                   CInode *in=0, fragtree_t *pdft=0, bufferlist *psnapbl=0,

The real fix, I think, is for mksnap and rmsnap to not do all the metablob updates at all, but to just call journal_dirty_inode() and be done with it. See the _setattr() method:

  ...all the projectiong and predirtying...

  // log + wait
  le->metablob.add_client_req(req->get_reqid(), req->get_oldest_client_tid());
  mdcache->predirty_journal_parents(mdr, &le->metablob, cur, 0, PREDIRTY_PRIMARY, false);
  mdcache->journal_dirty_inode(mdr, &le->metablob, cur);

Merged by Sage in commit:bc8ee67cc77ab687e735c936ccc229585845effb

Bulk updating project=ceph category=mds bugs so that I can remove the MDS category from the Ceph project to avoid confusion.