From 008a98e1440bb41a010c0acbfb9d40073602757b Mon Sep 17 00:00:00 2001
From: sageweil <sageweil@29311d96-e01e-0410-9327-a35deaab8ce9>
Date: Fri, 2 Nov 2007 18:00:32 +0000
Subject: [PATCH] migrate_stray; filepath fixup; messenger _myaddr/inst
 cleanup; server rename pin bugfix

git-svn-id: https://ceph.svn.sf.net/svnroot/ceph@2013 29311d96-e01e-0410-9327-a35deaab8ce9
---
 branches/sage/mds/TODO                        |  37 +++---
 branches/sage/mds/client/Client.cc            |  29 +++--
 branches/sage/mds/client/Client.h             |   4 +-
 branches/sage/mds/client/SyntheticClient.cc   |  16 +--
 branches/sage/mds/include/filepath.h          |  77 +++++++-----
 branches/sage/mds/include/types.h             |  22 +---
 branches/sage/mds/mds/CDentry.cc              |  28 +++--
 branches/sage/mds/mds/CDentry.h               |   5 +-
 branches/sage/mds/mds/CDir.cc                 |   2 +-
 branches/sage/mds/mds/CInode.cc               |  18 ++-
 branches/sage/mds/mds/CInode.h                |   7 +-
 branches/sage/mds/mds/ClientMap.h             |  21 ++--
 branches/sage/mds/mds/Locker.cc               |  11 +-
 branches/sage/mds/mds/MDBalancer.cc           |   2 +-
 branches/sage/mds/mds/MDCache.cc              | 114 +++++++++++++++---
 branches/sage/mds/mds/MDCache.h               |   3 +-
 branches/sage/mds/mds/MDS.cc                  |  12 +-
 branches/sage/mds/mds/MDS.h                   |   4 +
 branches/sage/mds/mds/Migrator.cc             |  15 ++-
 branches/sage/mds/mds/Migrator.h              |   4 +-
 branches/sage/mds/mds/Server.cc               |  82 +++++++------
 branches/sage/mds/mds/journal.cc              |   4 +-
 branches/sage/mds/mds/mdstypes.h              |  24 ++--
 branches/sage/mds/messages/MClientFileCaps.h  |   8 +-
 branches/sage/mds/messages/MClientReply.h     |   4 +
 branches/sage/mds/messages/MClientRequest.h   |  38 +++---
 branches/sage/mds/messages/MDirUpdate.h       |  13 +-
 .../sage/mds/messages/MExportDirDiscover.h    |   8 +-
 branches/sage/mds/messages/MExportStrays.h    |  43 -------
 branches/sage/mds/messages/MMDSSlaveRequest.h |  12 +-
 branches/sage/mds/msg/FakeMessenger.cc        |   9 +-
 branches/sage/mds/msg/FakeMessenger.h         |   9 --
 branches/sage/mds/msg/Message.cc              |   1 +
 branches/sage/mds/msg/Message.h               |   6 -
 branches/sage/mds/msg/Messenger.h             |  20 +--
 branches/sage/mds/msg/SimpleMessenger.cc      |  37 +-----
 branches/sage/mds/msg/SimpleMessenger.h       |   6 +-
 branches/sage/mds/msg/msg_types.h             |   2 +-
 38 files changed, 417 insertions(+), 340 deletions(-)
 delete mode 100644 branches/sage/mds/messages/MExportStrays.h

diff --git a/branches/sage/mds/TODO b/branches/sage/mds/TODO
index 5e1a37b7441a4..5c670b9694ce1 100644
--- a/branches/sage/mds/TODO
+++ b/branches/sage/mds/TODO
@@ -53,36 +53,26 @@ mdsmon
 
 
 
-- clean up client mds session vs mdsmap behavior
 
 
 mds bugs
-- open file rejournaling vs capped log...
-  - open files vs shutdown in general!  need to export any caps on replicated metadata
-- stray purge on shutdown
-
+- stray migration
+ - purge on shutdown
 - rename slave in-memory rollback on failure
-
 - fix purge_stray bug
 - try_remove_unlinked_dn thing
-
 - proper handling of cache expire messages during rejoin phase?
 
 mds
+- stray reintegration
 - extend/clean up filepath to allow paths relative to an ino
   - fix path_traverse
   - fix reconnect/rejoin open file weirdness
-
-- get rid of C*Discover objects for replicate_to .. encode to bufferlists directly
-
-- stray reintegration
-
 - real chdir (directory "open")
   - relative metadata ops
-
+- get rid of C*Discover objects for replicate_to .. encode to bufferlists directly?
 - consistency points/snapshots
   - dentry versions vs dirfrags...
-
 - detect and deal with client failure
   - failure during reconnect vs clientmap.  although probalby the whole thing needs a larger overhaul...
 
@@ -113,6 +103,13 @@ mds
 - delayed replica caps release... we need to set a timer event? (and cancel it when appropriate?)
 
 
+client
+- clean up client mds session vs mdsmap behavior?
+- client caps migration races
+  - caps need a seq number; reap logic needs to be a bit smarter
+  - also needs cope with mds failures
+- fstat
+
 
 osdmon
 - allow fresh replacement osds.  add osd_created in osdmap, probably
@@ -202,11 +199,14 @@ reliability
 
 ebofs
 - allow holes
+- allow btree sets
+- optionally scrub deallocated extents
+- clone()
+
+- map ObjectStore
 
 - verify proper behavior of conflicting/overlapping reads of clones
 - combine inodes and/or cnodes into same blocks
-- allow btree sets instead of maps
-- eliminate nodepools
 - nonblocking write on missing onodes?
 - fix bug in node rotation on insert (and reenable)
 - fix NEAR_LAST_FWD (?)
@@ -255,11 +255,6 @@ crush
 
 
 
-client
-- fstat
-- mixed lazy and non-lazy io will clobber each others' caps in the buffer cache.. how to isolate..
-- test client caps migration w/ mds exports
-- some heuristic behavior to consolidate caps to inode auth?
 
 
 
diff --git a/branches/sage/mds/client/Client.cc b/branches/sage/mds/client/Client.cc
index 117b2b08e401d..08f431bf84835 100644
--- a/branches/sage/mds/client/Client.cc
+++ b/branches/sage/mds/client/Client.cc
@@ -508,8 +508,7 @@ Dentry *Client::lookup(filepath& path)
   Dentry *dn = 0;
   for (unsigned i=0; i<path.depth(); i++) {
     dout(14) << " seg " << i << " = " << path[i] << dendl;
-    if (cur->inode.mode & INODE_MODE_DIR &&
-        cur->dir) {
+    if (cur->inode.is_dir() && cur->dir) {
       // dir, we can descend
       Dir *dir = cur->dir;
       if (dir->dentries.count(path[i])) {
@@ -548,7 +547,7 @@ int Client::choose_target_mds(MClientRequest *req)
   unsigned i;
   for (i=0; i<depth; i++) {
     // dir?
-    if (diri && diri->inode.mode & INODE_MODE_DIR && diri->dir) {
+    if (diri && diri->inode.is_dir() && diri->dir) {
       Dir *dir = diri->dir;
       
       // do we have the next dentry?
@@ -803,6 +802,8 @@ void Client::send_request(MetaRequest *request, int mds)
   }
   request->request = 0;
 
+  r->set_mdsmap_epoch(mdsmap->get_epoch());
+
   dout(10) << "send_request " << *r << " to mds" << mds << dendl;
   messenger->send_message(r, mdsmap->get_inst(mds), MDS_PORT_SERVER);
   
@@ -1107,6 +1108,17 @@ void Client::handle_file_caps(MClientFileCaps *m)
   if (m->get_op() == MClientFileCaps::OP_IMPORT) {
     int other = m->get_mds();
 
+    /*
+     * FIXME: there is a race here.. if the caps are exported twice in succession,
+     *  you may get the second import before the first, in which case the middle MDS's
+     *  import and then export won't be handled properly.
+     *  there should be a sequence number attached to the cap, incremented each time
+     *  it is exported... 
+     */
+    /*
+     * FIXME: handle mds failures
+     */
+
     if (in && in->stale_caps.count(other)) {
       dout(5) << "handle_file_caps on ino " << m->get_ino() << " from mds" << mds << " imported from mds" << other << dendl;
 
@@ -1196,6 +1208,7 @@ void Client::handle_file_caps(MClientFileCaps *m)
             << " seq " << m->get_seq() 
             << " " << cap_string(m->get_caps()) 
             << ", which we don't want caps for, releasing." << dendl;
+    m->set_op(MClientFileCaps::OP_ACK);
     m->set_caps(0);
     m->set_wanted(0);
     messenger->send_message(m, m->get_source_inst(), MDS_PORT_LOCKER);
@@ -1585,7 +1598,7 @@ int Client::_link(const char *existing, const char *newname)
 
   MClientRequest *req = new MClientRequest(MDS_OP_LINK, messenger->get_myinst());
   req->set_path(newname);
-  req->set_sarg(existing);
+  req->set_path2(existing);
   
   // FIXME where does FUSE maintain user information
   req->set_caller_uid(getuid());
@@ -1664,7 +1677,7 @@ int Client::_rename(const char *from, const char *to)
 {
   MClientRequest *req = new MClientRequest(MDS_OP_RENAME, messenger->get_myinst());
   req->set_path(from);
-  req->set_sarg(to);
+  req->set_path2(to);
  
   // FIXME where does FUSE maintain user information
   req->set_caller_uid(getuid());
@@ -1792,7 +1805,7 @@ int Client::_symlink(const char *target, const char *link)
 {
   MClientRequest *req = new MClientRequest(MDS_OP_SYMLINK, messenger->get_myinst());
   req->set_path(link);
-  req->set_sarg(target);
+  req->set_path2(target);
  
   // FIXME where does FUSE maintain user information
   req->set_caller_uid(getuid());
@@ -1879,7 +1892,7 @@ int Client::_do_lstat(const char *path, int mask, Inode **in)
     
     req = new MClientRequest(MDS_OP_LSTAT, messenger->get_myinst());
     req->args.stat.mask = mask;
-    req->set_path(fpath);
+    req->set_filepath(fpath);
 
     MClientReply *reply = make_request(req);
     res = reply->get_result();
@@ -2297,7 +2310,7 @@ int Client::_readdir_get_frag(DirResult *dirp)
     diri = inode_map[ino];
     dout(10) << "_readdir_get_frag got diri " << diri << " " << diri->inode.ino << dendl;
     assert(diri);
-    assert(diri->inode.mode & INODE_MODE_DIR);
+    assert(diri->inode.is_dir());
   }
   
   if (!dirp->inode && diri) {
diff --git a/branches/sage/mds/client/Client.h b/branches/sage/mds/client/Client.h
index 727098906c617..fff9829c90ec5 100644
--- a/branches/sage/mds/client/Client.h
+++ b/branches/sage/mds/client/Client.h
@@ -210,9 +210,7 @@ class Inode {
 
   inodeno_t ino() { return inode.ino; }
 
-  bool is_dir() {
-    return (inode.mode & INODE_TYPE_MASK) == INODE_MODE_DIR;
-  }
+  bool is_dir() { return inode.is_dir(); }
 
   int file_caps() {
     int c = 0;
diff --git a/branches/sage/mds/client/SyntheticClient.cc b/branches/sage/mds/client/SyntheticClient.cc
index 270ebafec010f..4ac6ab356081a 100644
--- a/branches/sage/mds/client/SyntheticClient.cc
+++ b/branches/sage/mds/client/SyntheticClient.cc
@@ -264,7 +264,7 @@ string SyntheticClient::get_sarg(int seq)
   }
   if (a.length() == 0 || a == "~") {
     char s[20];
-    sprintf(s,"syn.%d.%d", client->whoami, seq);
+    sprintf(s,"/syn.%d.%d", client->whoami, seq);
     a = s;
   } 
   return a;
@@ -1365,7 +1365,7 @@ int SyntheticClient::clean_dir(string& basedir)
       continue;
     }
 
-    if ((st.st_mode & INODE_TYPE_MASK) == INODE_MODE_DIR) {
+    if ((st.st_mode & S_IFMT) == S_IFDIR) {
       clean_dir(file);
       client->rmdir(file.c_str());
     } else {
@@ -1434,7 +1434,7 @@ int SyntheticClient::full_walk(string& basedir)
 	     file.c_str());
 
       
-      if ((st.st_mode & INODE_TYPE_MASK) == INODE_MODE_DIR) {
+      if ((st.st_mode & S_IFMT) == S_IFDIR) {
 	dirq.push_back(file);
       }
     }
@@ -1650,10 +1650,10 @@ int SyntheticClient::open_shared(int num, int count)
     for (int n=0; n<num; n++) {
       sprintf(d,"file.%d", n);
       int fd = client->open(d,O_RDONLY);
-      fds.push_back(fd);
+      if (fd > 0) fds.push_back(fd);
     }
 
-    if (1)
+    if (false && client->get_nodeid() == 0)
       for (int n=0; n<num; n++) {
 	sprintf(d,"file.%d", n);
 	client->unlink(d);
@@ -2825,10 +2825,10 @@ void SyntheticClient::import_find(const char *base, const char *find, bool data)
     assert(modestring.length() == 10);
     mode_t mode = 0;
     switch (modestring[0]) {
-    case 'd': mode |= INODE_MODE_DIR; break;
-    case 'l': mode |= INODE_MODE_SYMLINK; break;
+    case 'd': mode |= S_IFDIR; break;
+    case 'l': mode |= S_IFLNK; break;
     default:
-    case '-': mode |= INODE_MODE_FILE; break;
+    case '-': mode |= S_IFREG; break;
     }
     if (modestring[1] == 'r') mode |= 0400;
     if (modestring[2] == 'w') mode |= 0200;
diff --git a/branches/sage/mds/include/filepath.h b/branches/sage/mds/include/filepath.h
index 4425e1d7c5b3a..c92663049bf1f 100644
--- a/branches/sage/mds/include/filepath.h
+++ b/branches/sage/mds/include/filepath.h
@@ -30,13 +30,13 @@
 using namespace std;
 
 #include "buffer.h"
-
+#include "encodable.h"
 
 class filepath {
   /** path
-   * can be relative "a/b/c" or absolute "/a/b/c".
    */
-  string path;
+  inodeno_t ino;   // base inode
+  string path;     // relative path
 
   /** bits - path segemtns
    * this is ['a', 'b', 'c'] for both the aboslute and relative case.
@@ -74,40 +74,29 @@ class filepath {
   }
 
  public:
-  filepath() {}
-  filepath(const string& s) {
-    set_path(s);
-  }
-  filepath(const char* s) {
-    set_path(s);
-  }
+  filepath() : ino(0) {}
+  filepath(const string& s, inodeno_t i=1) : ino(i), path(s) { }
+  filepath(const char* s, inodeno_t i=1) : ino(i), path(s) { }
   filepath(const filepath& o) {
-    set_path(o.get_path());
+    ino = o.ino;
+    path = o.path;
+    bits = o.bits;
   }
 
-
   // accessors
-  const string& get_path() const {
-    return path;
-  }
-  const char *c_str() const {
-    return path.c_str();
-  }
+  inodeno_t get_ino() const { return ino; }
+  const string& get_path() const { return path; }
+  const char *c_str() const { return path.c_str(); }
 
-  int length() const {
-    return path.length();
-  }
+  int length() const { return path.length(); }
   unsigned depth() const {
     if (bits.empty() && path.length() > 0) parse_bits();
     return bits.size();
   }
-  bool empty() const {
-    return path.length() == 0;
-  }
+  bool empty() const { return path.length() == 0; }
 
-  // FIXME: const-edness
-  bool absolute() { return path.length() && path[0] == '/'; }
-  bool relative() { return !absolute(); }
+  bool absolute() const { return ino > 0; }
+  bool relative() const { return !absolute(); }
   
   const string& operator[](int i) const {
     if (bits.empty() && path.length() > 0) parse_bits();
@@ -121,6 +110,7 @@ class filepath {
 
   filepath prefixpath(int s) const {
     filepath t;
+    t.ino = ino;    
     for (int i=0; i<s; i++)
       t.push_dentry(bits[i]);
     return t;
@@ -133,17 +123,31 @@ class filepath {
   }
 
 
-
   // modifiers
+  //  string can be relative "a/b/c" (ino=0) or absolute "/a/b/c" (ino=1)
   void set_path(const string& s) {
-    path = s;
+    if (s.length() && s[0] == '/') {
+      ino = 1;  // relative to root
+      path = string(s, 1);
+    } else {
+      ino = 0;  // strictly relative   
+      path = s;
+    }
     bits.clear();
   }
   void set_path(const char *s) {
-    path = s;
+    if (s[0] == '/') {
+      ino = 1;  // relative to root
+      path = s + 1;
+    } else {
+      ino = 0;  // strictly relative   
+      path = s;
+    }
     bits.clear();
   }
+  void set_ino(inodeno_t i) { ino = i; }
   void clear() {
+    ino = 0;
     path = "";
     bits.clear();
   }
@@ -167,17 +171,28 @@ class filepath {
 
   // encoding
   void _encode(bufferlist& bl) {
-    ::_encode(path, bl);
+    ::_encode_simple(ino, bl);
+    ::_encode_simple(path, bl);
   }
   void _decode(bufferlist& bl, int& off) {
+    bits.clear();
+    ::_decode(ino, bl, off);
     ::_decode(path, bl, off);
+  }
+  void _decode(bufferlist::iterator& blp) {
     bits.clear();
+    ::_decode_simple(ino, blp);
+    ::_decode_simple(path, blp);
   }
 
 };
 
 inline ostream& operator<<(ostream& out, filepath& path)
 {
+  if (path.get_ino() > 1)
+    out << '#' << hex << path.get_ino() << dec;
+  if (path.get_ino() > 0 && path.depth())
+    out << '/';
   return out << path.get_path();
 }
 
diff --git a/branches/sage/mds/include/types.h b/branches/sage/mds/include/types.h
index cf8374d329a77..39f6afdf2376f 100644
--- a/branches/sage/mds/include/types.h
+++ b/branches/sage/mds/include/types.h
@@ -150,11 +150,6 @@ namespace __gnu_cxx {
 }
 
 
-#define INODE_MODE_FILE     0100000 // S_IFREG
-#define INODE_MODE_SYMLINK  0120000 // S_IFLNK
-#define INODE_MODE_DIR      0040000 // S_IFDIR
-#define INODE_TYPE_MASK     0170000
-
 #define FILE_MODE_R          1
 #define FILE_MODE_W          2
 #define FILE_MODE_RW         (1|2)
@@ -179,6 +174,9 @@ namespace __gnu_cxx {
 inline int DT_TO_MODE(int dt) {
   return dt << 12;
 }
+inline unsigned char MODE_TO_DT(int mode) {
+  return mode >> 12;
+}
 
 struct inode_t {
   // base (immutable)
@@ -208,19 +206,11 @@ struct inode_t {
   version_t file_data_version; // auth only
 
   // file type
-  bool is_symlink() { return (mode & INODE_TYPE_MASK) == INODE_MODE_SYMLINK; }
-  bool is_dir()     { return (mode & INODE_TYPE_MASK) == INODE_MODE_DIR; }
-  bool is_file()    { return (mode & INODE_TYPE_MASK) == INODE_MODE_FILE; }
-
-  // corresponding d_types
-  static const unsigned char DT_REG = 8;
-  static const unsigned char DT_DIR = 4;
-  static const unsigned char DT_LNK = 10;
+  bool is_symlink() { return (mode & S_IFMT) == S_IFLNK; }
+  bool is_dir()     { return (mode & S_IFMT) == S_IFDIR; }
+  bool is_file()    { return (mode & S_IFMT) == S_IFREG; }
 };
 
-inline unsigned char MODE_TO_DT(int mode) {
-  return mode >> 12;
-}
 
 
 
diff --git a/branches/sage/mds/mds/CDentry.cc b/branches/sage/mds/mds/CDentry.cc
index 2b6bb3470e8a8..0ffedb20bc330 100644
--- a/branches/sage/mds/mds/CDentry.cc
+++ b/branches/sage/mds/mds/CDentry.cc
@@ -41,7 +41,7 @@ ostream& CDentry::print_db_line_prefix(ostream& out)
 
 ostream& operator<<(ostream& out, CDentry& dn)
 {
-  string path;
+  filepath path;
   dn.make_path(path);
   
   out << "[dentry " << path;
@@ -59,10 +59,10 @@ ostream& operator<<(ostream& out, CDentry& dn)
   if (dn.is_null()) out << " NULL";
   if (dn.is_remote()) {
     out << " REMOTE(";
-    switch (dn.get_remote_d_type()) {
-    case inode_t::DT_REG: out << "reg"; break;
-    case inode_t::DT_DIR: out << "dir"; break;
-    case inode_t::DT_LNK: out << "lnk"; break;
+    switch (dn.get_remote_d_type() << 12) {
+    case S_IFREG: out << "reg"; break;
+    case S_IFDIR: out << "dir"; break;
+    case S_IFLNK: out << "lnk"; break;
     default: assert(0);
     }
     out << ")";
@@ -186,10 +186,10 @@ void CDentry::mark_new()
   state_set(STATE_NEW);
 }
 
-void CDentry::make_path(string& s)
+void CDentry::make_path_string(string& s)
 {
   if (dir) {
-    dir->inode->make_path(s);
+    dir->inode->make_path_string(s);
   } else {
     s = "???";
   }
@@ -197,6 +197,19 @@ void CDentry::make_path(string& s)
   s += name;
 }
 
+void CDentry::make_path(filepath& fp)
+{
+  assert(dir);
+  if (dir->inode->is_base())
+    fp.set_ino(dir->inode->ino());               // base case
+  else if (dir->inode->get_parent_dn())
+    dir->inode->get_parent_dn()->make_path(fp);  // recurse
+  else
+    fp.set_ino(dir->inode->ino());               // relative but not base?  hrm!
+  fp.push_dentry(name);
+}
+
+/*
 void CDentry::make_path(string& s, inodeno_t tobase)
 {
   assert(dir);
@@ -211,6 +224,7 @@ void CDentry::make_path(string& s, inodeno_t tobase)
   }
   s += name;
 }
+*/
 
 /** make_anchor_trace
  * construct an anchor trace for this dentry, as if it were linked to *in.
diff --git a/branches/sage/mds/mds/CDentry.h b/branches/sage/mds/mds/CDentry.h
index 416792beb8778..e28aa2e380b09 100644
--- a/branches/sage/mds/mds/CDentry.h
+++ b/branches/sage/mds/mds/CDentry.h
@@ -26,6 +26,7 @@ using namespace std;
 #include "include/buffer.h"
 #include "include/lru.h"
 #include "include/xlist.h"
+#include "include/filepath.h"
 #include "mdstypes.h"
 
 #include "SimpleLock.h"
@@ -184,8 +185,8 @@ public:
   const CDentry& operator= (const CDentry& right);
 
   // misc
-  void make_path(string& p);
-  void make_path(string& p, inodeno_t tobase);
+  void make_path_string(string& s);
+  void make_path(filepath& fp);
   void make_anchor_trace(vector<class Anchor>& trace, CInode *in);
 
   // -- version --
diff --git a/branches/sage/mds/mds/CDir.cc b/branches/sage/mds/mds/CDir.cc
index b4663b269c659..adaf5fa6c0d4f 100644
--- a/branches/sage/mds/mds/CDir.cc
+++ b/branches/sage/mds/mds/CDir.cc
@@ -45,7 +45,7 @@
 
 ostream& operator<<(ostream& out, CDir& dir)
 {
-  string path;
+  filepath path;
   dir.get_inode()->make_path(path);
   out << "[dir " << dir.dirfrag() << " " << path << "/";
   if (dir.is_auth()) {
diff --git a/branches/sage/mds/mds/CInode.cc b/branches/sage/mds/mds/CInode.cc
index d5981eea5ca3b..cb436254de13e 100644
--- a/branches/sage/mds/mds/CInode.cc
+++ b/branches/sage/mds/mds/CInode.cc
@@ -46,7 +46,7 @@ ostream& CInode::print_db_line_prefix(ostream& out)
 
 ostream& operator<<(ostream& out, CInode& in)
 {
-  string path;
+  filepath path;
   in.make_path(path);
   out << "[inode " << in.inode.ino << " " << path << (in.is_dir() ? "/ ":" ");
   if (in.is_auth()) {
@@ -351,10 +351,10 @@ CInode *CInode::get_parent_inode()
 
 
 
-void CInode::make_path(string& s)
+void CInode::make_path_string(string& s)
 {
   if (parent) {
-    parent->make_path(s);
+    parent->make_path_string(s);
   } 
   else if (is_root()) {
     s = "";  // root
@@ -370,6 +370,14 @@ void CInode::make_path(string& s)
   }
 }
 
+void CInode::make_path(filepath& fp)
+{
+  if (parent) 
+    parent->make_path(fp);
+  else
+    fp.set_ino(ino());
+}
+
 void CInode::make_anchor_trace(vector<Anchor>& trace)
 {
   if (parent) {
@@ -738,8 +746,8 @@ void CInode::adjust_nested_auth_pins(int a)
 
 pair<int,int> CInode::authority() 
 {
-  if (force_auth.first >= 0) 
-    return force_auth;
+  if (inode_auth.first >= 0) 
+    return inode_auth;
 
   if (parent)
     return parent->dir->authority();
diff --git a/branches/sage/mds/mds/CInode.h b/branches/sage/mds/mds/CInode.h
index ef05b81b2c6cf..2677f3d227881 100644
--- a/branches/sage/mds/mds/CInode.h
+++ b/branches/sage/mds/mds/CInode.h
@@ -185,7 +185,7 @@ public:
   CDentry         *parent;             // primary link
   set<CDentry*>    remote_parents;     // if hard linked
 
-  pair<int,int> force_auth;
+  pair<int,int> inode_auth;
 
   // -- distributed state --
 protected:
@@ -229,7 +229,7 @@ public:
     last_journaled(0), last_open_journaled(0), 
     //hack_accessed(true),
     stickydir_ref(0),
-    parent(0), force_auth(CDIR_AUTH_DEFAULT),
+    parent(0), inode_auth(CDIR_AUTH_DEFAULT),
     replica_caps_wanted(0),
     xlist_dirty(this), xlist_open_file(this), 
     xlist_dirty_inode_mtime(this), xlist_purging_inode(this),
@@ -280,7 +280,8 @@ public:
   }
 
   // -- misc -- 
-  void make_path(string& s);
+  void make_path_string(string& s);
+  void make_path(filepath& s);
   void make_anchor_trace(vector<class Anchor>& trace);
   void name_stray_dentry(string& dname);
 
diff --git a/branches/sage/mds/mds/ClientMap.h b/branches/sage/mds/mds/ClientMap.h
index b16abdbef7b91..59b3dde49b6be 100644
--- a/branches/sage/mds/mds/ClientMap.h
+++ b/branches/sage/mds/mds/ClientMap.h
@@ -125,19 +125,18 @@ public:
 
 private:
   // -- completed requests --
-  // client id -> tid -> result code
-  map<int, set<tid_t> > completed_requests;  // completed client requests
-  map<int, map<tid_t, Context*> > waiting_for_trim;
+  // who -> { tid set ... }
+  map<entity_name_t, set<tid_t> >           completed_requests;
+  map<entity_name_t, map<tid_t, Context*> > waiting_for_trim;
   version_t requestmapv;
  
 public:
   void add_completed_request(metareqid_t ri) {
-    completed_requests[ri.client].insert(ri.tid);
+    completed_requests[ri.name].insert(ri.tid);
     requestmapv++;
   }
-  void trim_completed_requests(int client, 
-			       tid_t mintid) {  // zero means trim all!
-    map<int, set<tid_t> >::iterator p = completed_requests.find(client);
+  void trim_completed_requests(entity_name_t who, tid_t mintid) {  // zero means trim all!
+    map<entity_name_t, set<tid_t> >::iterator p = completed_requests.find(who);
     if (p == completed_requests.end())
       return;
 
@@ -148,7 +147,7 @@ public:
       completed_requests.erase(p);
 
     // kick waiters
-    map<int, map<tid_t,Context*> >::iterator q = waiting_for_trim.find(client);
+    map<entity_name_t, map<tid_t,Context*> >::iterator q = waiting_for_trim.find(who);
     if (q != waiting_for_trim.end()) {
       list<Context*> fls;
       while (!q->second.empty() &&
@@ -162,11 +161,11 @@ public:
     }
   }
   void add_trim_waiter(metareqid_t ri, Context *c) {
-    waiting_for_trim[ri.client][ri.tid] = c;
+    waiting_for_trim[ri.name][ri.tid] = c;
   }
   bool have_completed_request(metareqid_t ri) {
-    return completed_requests.count(ri.client) &&
-      completed_requests[ri.client].count(ri.tid);
+    return completed_requests.count(ri.name) &&
+      completed_requests[ri.name].count(ri.tid);
   }
 
 
diff --git a/branches/sage/mds/mds/Locker.cc b/branches/sage/mds/mds/Locker.cc
index f5503d1345f31..f9b29bf23fee8 100644
--- a/branches/sage/mds/mds/Locker.cc
+++ b/branches/sage/mds/mds/Locker.cc
@@ -237,6 +237,7 @@ bool Locker::acquire_locks(MDRequest *mdr,
 	MDSCacheObjectInfo info;
 	(*q)->set_object_info(info);
 	req->get_authpins().push_back(info);      
+	mdr->pin(*q);
       }
       mds->send_message_mds(req, p->first, MDS_PORT_SERVER);
 
@@ -452,7 +453,7 @@ Capability* Locker::issue_new_caps(CInode *in,
   dout(7) << "issue_new_caps for mode " << mode << " on " << *in << dendl;
   
   // my needs
-  int my_client = req->get_client();
+  int my_client = req->get_client().num();
   int my_want = 0;
   if (mode & FILE_MODE_R) my_want |= CAP_FILE_RDCACHE  | CAP_FILE_RD;
   if (mode & FILE_MODE_W) my_want |= CAP_FILE_WRBUFFER | CAP_FILE_WR;
@@ -1565,6 +1566,14 @@ void Locker::scatter_writebehind(ScatterLock *lock)
   CInode *in = (CInode*)lock->get_parent();
   dout(10) << "scatter_writebehind " << in->inode.mtime << " on " << *lock << " on " << *in << dendl;
 
+  // hack:
+  if (in->is_base()) {
+    dout(10) << "scatter_writebehind just clearing updated flag for base inode " << *in << dendl;
+    lock->clear_updated();
+    scatter_eval_gather(lock);
+    return;
+  }
+
   // journal write-behind.
   inode_t *pi = in->project_inode();
   pi->mtime = in->inode.mtime;   // make sure an intermediate version isn't goofing us up
diff --git a/branches/sage/mds/mds/MDBalancer.cc b/branches/sage/mds/mds/MDBalancer.cc
index 8e9d0e2dd46fa..933c8306a7526 100644
--- a/branches/sage/mds/mds/MDBalancer.cc
+++ b/branches/sage/mds/mds/MDBalancer.cc
@@ -1001,7 +1001,7 @@ void MDBalancer::dump_pop_map()
 
 	// filename last
 	string p;
-	in->make_path(p);
+	in->make_path_string(p);
 	myfile << "." << p;
 	if (dir->get_frag() != frag_t()) 
 	  myfile << "___" << (unsigned)dir->get_frag();
diff --git a/branches/sage/mds/mds/MDCache.cc b/branches/sage/mds/mds/MDCache.cc
index 8456b5bb555b1..25ae8e6030379 100644
--- a/branches/sage/mds/mds/MDCache.cc
+++ b/branches/sage/mds/mds/MDCache.cc
@@ -211,7 +211,7 @@ CInode *MDCache::create_root_inode()
   root->inode.ino = MDS_INO_ROOT;
   
   // make it up (FIXME)
-  root->inode.mode = 0755 | INODE_MODE_DIR;
+  root->inode.mode = 0755 | S_IFDIR;
   root->inode.size = 0;
   root->inode.ctime = 
     root->inode.mtime = g_clock.now();
@@ -219,7 +219,7 @@ CInode *MDCache::create_root_inode()
   root->inode.nlink = 1;
   root->inode.layout = g_OSD_MDDirLayout;
   
-  root->force_auth = pair<int,int>(0, CDIR_AUTH_UNKNOWN);
+  root->inode_auth = pair<int,int>(0, CDIR_AUTH_UNKNOWN);
 
   add_inode( root );
 
@@ -262,7 +262,7 @@ CInode *MDCache::create_stray_inode(int whose)
   in->inode.ino = MDS_INO_STRAY(whose);
   
   // make it up (FIXME)
-  in->inode.mode = 0755 | INODE_MODE_DIR;
+  in->inode.mode = 0755 | S_IFDIR;
   in->inode.size = 0;
   in->inode.ctime = 
     in->inode.mtime = g_clock.now();
@@ -2224,7 +2224,7 @@ void MDCache::handle_cache_rejoin_strong(MMDSCacheRejoin *strong)
       if (!in) in = rejoin_invent_inode(p->first.ino);
       if (!in->is_dir()) {
 	assert(in->state_test(CInode::STATE_REJOINUNDEF));
-	in->inode.mode = INODE_MODE_DIR;
+	in->inode.mode = S_IFDIR;
       }
       dir = in->get_or_open_dirfrag(this, p->first.frag);
     } else {
@@ -3570,6 +3570,61 @@ bool MDCache::shutdown_pass()
     }
   }
 
+  // export caps?
+  //  note: this runs more often than it should.
+  static bool exported_caps = false;
+  static set<CDir*> exported_caps_in;
+  if (!exported_caps) {
+    dout(7) << "searching for caps to export" << dendl;
+    exported_caps = true;
+
+    list<CDir*> dirq;
+    for (map<CDir*,set<CDir*> >::iterator p = subtrees.begin();
+	 p != subtrees.end();
+	 ++p) {
+      if (exported_caps_in.count(p->first)) continue;
+      if (p->first->is_auth() ||
+	  p->first->is_ambiguous_auth())
+	exported_caps = false; // we'll have to try again
+      else {
+	dirq.push_back(p->first);
+	exported_caps_in.insert(p->first);
+      }
+    }
+    while (!dirq.empty()) {
+      CDir *dir = dirq.front();
+      dirq.pop_front();
+      for (CDir::map_t::iterator p = dir->items.begin();
+	   p != dir->items.end();
+	   ++p) {
+	CDentry *dn = p->second;
+	if (!dn->is_primary()) continue;
+	CInode *in = dn->get_inode();
+	if (in->is_dir())
+	  in->get_nested_dirfrags(dirq);
+	if (in->is_any_caps() &&
+	    !in->state_test(CInode::STATE_EXPORTINGCAPS))
+	  migrator->export_caps(in);
+      }
+    }
+  }
+
+  static bool exported_strays = false;
+  if (!exported_strays && stray && mds->get_nodeid() > 0) {
+    list<CDir*> dfs;
+    stray->get_dirfrags(dfs);
+    while (!dfs.empty()) {
+      CDir *dir = dfs.front();
+      dfs.pop_front();
+      for (CDir::map_t::iterator p = dir->items.begin();
+	   p != dir->items.end();
+	   p++) {
+	CDentry *dn = p->second;
+	migrate_stray(dn, 0);  // send to root!
+      }
+    }
+    exported_strays = true;
+  }
 
   // subtrees map not empty yet?
   if (!subtrees.empty()) {
@@ -3730,7 +3785,7 @@ Context *MDCache::_get_waiter(MDRequest *mdr, Message *req)
 }
 
 int MDCache::path_traverse(MDRequest *mdr, Message *req,     // who
-			   CInode *base, filepath& origpath, // what
+			   filepath& origpath,               // what
                            vector<CDentry*>& trace,          // result
                            bool follow_trailing_symlink,     // how
                            int onfail)
@@ -3746,11 +3801,17 @@ int MDCache::path_traverse(MDRequest *mdr, Message *req,     // who
   set< pair<CInode*, string> > symlinks_resolved; 
 
   // root
-  CInode *cur = base;
-  if (!cur) cur = get_root();
+  CInode *cur = get_inode(origpath.get_ino());
   if (cur == NULL) {
-    dout(7) << "traverse: i don't have root" << dendl;
-    open_root(_get_waiter(mdr, req));
+    dout(7) << "traverse: opening base ino " << origpath.get_ino() << dendl;
+    if (origpath.get_ino() == MDS_INO_ROOT)
+      open_root(_get_waiter(mdr, req));
+    else if (MDS_INO_IS_STRAY(origpath.get_ino())) 
+      open_foreign_stray(origpath.get_ino() - MDS_INO_STRAY_OFFSET, _get_waiter(mdr, req));
+    else {
+      assert(0);  // hrm.. broken
+      return -EIO;
+    }
     return 1;
   }
 
@@ -3766,7 +3827,6 @@ int MDCache::path_traverse(MDRequest *mdr, Message *req,     // who
   while (depth < path.depth()) {
     dout(12) << "traverse: path seg depth " << depth << " = " << path[depth] << dendl;
     
-    // ENOTDIR?
     if (!cur->is_dir()) {
       dout(7) << "traverse: " << *cur << " not a dir " << dendl;
       return -ENOTDIR;
@@ -4739,8 +4799,20 @@ void MDCache::reintegrate_stray(CDentry *dn, CDentry *rlink)
 
 void MDCache::migrate_stray(CDentry *dn, int dest)
 {
-  dout(10) << "migrate_stray to mds" << dest << " " << *dn << dendl;
+  dout(10) << "migrate_stray to mds" << dest << " " << *dn << " " << *dn->inode << dendl;
+
+  // rename it to another mds.
+  string dname;
+  dn->get_inode()->name_stray_dentry(dname);
+  filepath src(dname, MDS_INO_STRAY(mds->get_nodeid()));
+  filepath dst(dname, MDS_INO_STRAY(dest));
+
+  MClientRequest *req = new MClientRequest(MDS_OP_RENAME, mds->messenger->get_myinst());
+  req->set_filepath(src);
+  req->set_filepath2(dst);
+  req->set_tid(mds->issue_tid());
 
+  mds->send_message_mds(req, dest, MDS_PORT_SERVER);
 }
 
 
@@ -5216,7 +5288,6 @@ void MDCache::handle_discover_reply(MDiscoverReply *m)
 
     // add base inode
     cur = add_replica_inode(m->get_inode(0), NULL, finished);
-    cur->force_auth = pair<int,int>(m->get_source().num(), CDIR_AUTH_UNKNOWN);
 
     dout(7) << "discover_reply got base inode " << *cur << dendl;
     
@@ -5460,12 +5531,20 @@ CInode *MDCache::add_replica_inode(CInodeDiscover& dis, CDentry *dn, list<Contex
     dis.update_inode(in);
     dis.init_inode_locks(in);
     add_inode(in);
-    dout(10) << "add_replica_inode had " << *in << dendl;
+    if (in->is_base()) {
+      if (in->ino() == MDS_INO_ROOT)
+	in->inode_auth.first = 0;
+      else if (MDS_INO_IS_STRAY(in->ino())) 
+	in->inode_auth.first = in->ino() - MDS_INO_STRAY_OFFSET;
+      else
+	assert(0);
+    }
+    dout(10) << "add_replica_inode added " << *in << dendl;
     if (dn && dn->is_null()) 
       dn->dir->link_primary_inode(dn, in);
   } else {
     dis.update_inode(in);
-    dout(10) << "add_replica_inode added " << *in << dendl;
+    dout(10) << "add_replica_inode had " << *in << dendl;
   }
 
   if (dn) {
@@ -5488,7 +5567,6 @@ CDentry *MDCache::add_replica_stray(bufferlist &bl, CInode *in, int from)
   CInodeDiscover indis;
   indis._decode(bl, off);
   CInode *strayin = add_replica_inode(indis, NULL, finished);
-  strayin->force_auth = pair<int,int>(from, CDIR_AUTH_UNKNOWN);
   dout(15) << "strayin " << *strayin << dendl;
   
   // dir
@@ -5579,7 +5657,7 @@ int MDCache::send_dir_updates(CDir *dir, bool bcast)
   
   dout(7) << "sending dir_update on " << *dir << " bcast " << bcast << " to " << who << dendl;
 
-  string path;
+  filepath path;
   dir->inode->make_path(path);
 
   int whoami = mds->get_nodeid();
@@ -5619,7 +5697,7 @@ void MDCache::handle_dir_update(MDirUpdate *m)
       dout(5) << "trying discover on dir_update for " << path << dendl;
 
       int r = path_traverse(0, m,
-			    0, path, trace, true,
+			    path, trace, true,
                             MDS_TRAVERSE_DISCOVER);
       if (r > 0)
         return;
@@ -5687,7 +5765,7 @@ void MDCache::handle_dentry_unlink(MDentryUnlink *m)
 	// send caps to auth (if we're not already)
 	if (in->is_any_caps() &&
 	    !in->state_test(CInode::STATE_EXPORTINGCAPS))
-	  migrator->export_caps(in, in->authority().first);
+	  migrator->export_caps(in);
 	
 	lru.lru_bottouch(straydn);  // move stray to end of lru
 
diff --git a/branches/sage/mds/mds/MDCache.h b/branches/sage/mds/mds/MDCache.h
index 896ee791c367b..61a38f98049a7 100644
--- a/branches/sage/mds/mds/MDCache.h
+++ b/branches/sage/mds/mds/MDCache.h
@@ -582,8 +582,7 @@ public:
   CDentry *get_or_create_stray_dentry(CInode *in);
 
   Context *_get_waiter(MDRequest *mdr, Message *req);
-  int path_traverse(MDRequest *mdr, Message *req, 
-		    CInode *base, filepath& path, 
+  int path_traverse(MDRequest *mdr, Message *req, filepath& path, 
 		    vector<CDentry*>& trace, bool follow_trailing_sym,
                     int onfail);
   bool path_is_mine(filepath& path);
diff --git a/branches/sage/mds/mds/MDS.cc b/branches/sage/mds/mds/MDS.cc
index bf9f41cfac851..0ec62b7f4ed05 100644
--- a/branches/sage/mds/mds/MDS.cc
+++ b/branches/sage/mds/mds/MDS.cc
@@ -75,6 +75,8 @@ MDS::MDS(int whoami, Messenger *m, MonMap *mm) :
 
   this->whoami = whoami;
 
+  last_tid = 0;
+
   monmap = mm;
   messenger = m;
 
@@ -251,7 +253,8 @@ void MDS::send_message_mds(Message *m, int mds, int port, int fromport)
 void MDS::forward_message_mds(Message *req, int mds, int port)
 {
   // client request?
-  if (req->get_type() == MSG_CLIENT_REQUEST) {
+  if (req->get_type() == MSG_CLIENT_REQUEST &&
+      ((MClientRequest*)req)->get_client_inst().name.is_client()) {
     MClientRequest *creq = (MClientRequest*)req;
     creq->inc_num_fwd();    // inc forward counter
 
@@ -260,8 +263,13 @@ void MDS::forward_message_mds(Message *req, int mds, int port)
 			    creq->get_client_inst());
     
     if (!creq->is_idempotent()) {
+      /* don't actually forward if non-idempotent!
+       * client has to do it.  although the MDS will ignore duplicate requests,
+       * the affected metadata may migrate, in which case the new authority
+       * won't have the metareq_id in the completed request map.
+       */
       delete req;
-      return;  // don't actually forward if non-idempotent!  client has to do it.
+      return; 
     }
   }
   
diff --git a/branches/sage/mds/mds/MDS.h b/branches/sage/mds/mds/MDS.h
index 6d3f90dc465af..89f6e6d278dd3 100644
--- a/branches/sage/mds/mds/MDS.h
+++ b/branches/sage/mds/mds/MDS.h
@@ -118,6 +118,8 @@ class MDS : public Dispatcher {
 
   map<int,version_t> peer_mdsmap_epoch;
 
+  tid_t last_tid;    // for mds-initiated requests (e.g. stray rename)
+
  public:
   void wait_for_active(Context *c) { 
     waiting_for_active.push_back(c); 
@@ -142,6 +144,8 @@ class MDS : public Dispatcher {
 
   void set_want_state(int s);
 
+  tid_t issue_tid() { return ++last_tid; }
+    
 
   // -- waiters --
   list<Context*> finished_queue;
diff --git a/branches/sage/mds/mds/Migrator.cc b/branches/sage/mds/mds/Migrator.cc
index bc7c82396fe55..ea1e9f1b216c2 100644
--- a/branches/sage/mds/mds/Migrator.cc
+++ b/branches/sage/mds/mds/Migrator.cc
@@ -52,6 +52,10 @@
 #include "messages/MExportCaps.h"
 #include "messages/MExportCapsAck.h"
 
+
+
+
+
 #include "config.h"
 
 #define  dout(l)    if (l<=g_conf.debug || l <= g_conf.debug_mds || l <= g_conf.debug_mds_migrator) *_dout << dbeginl << g_clock.now() << " mds" << mds->get_nodeid() << ".migrator "
@@ -1394,9 +1398,7 @@ void Migrator::handle_export_discover(MExportDirDiscover *m)
     // must discover it!
     filepath fpath(m->get_path());
     vector<CDentry*> trace;
-    int r = cache->path_traverse(0, m,
-				 0, fpath, trace, true,
-				 MDS_TRAVERSE_DISCOVER);
+    int r = cache->path_traverse(0, m, fpath, trace, true, MDS_TRAVERSE_DISCOVER);
     if (r > 0) return; // wait
     if (r < 0) {
       dout(7) << "handle_export_discover_2 failed to discover or not dir " << m->get_path() << ", NAK" << dendl;
@@ -2223,8 +2225,9 @@ void Migrator::handle_export_notify(MExportDirNotify *m)
 
 
 
-void Migrator::export_caps(CInode *in, int dest)
+void Migrator::export_caps(CInode *in)
 {
+  int dest = in->authority().first;
   dout(7) << "export_caps to mds" << dest << " " << *in << dendl;
 
   assert(in->is_any_caps());
@@ -2306,3 +2309,7 @@ void Migrator::logged_import_caps(CInode *in,
   mds->send_message_mds(new MExportCapsAck(in->ino()), from, MDS_PORT_MIGRATOR);
 }
 
+
+
+
+
diff --git a/branches/sage/mds/mds/Migrator.h b/branches/sage/mds/mds/Migrator.h
index ca84968ac5d57..b32a6a1a4f33f 100644
--- a/branches/sage/mds/mds/Migrator.h
+++ b/branches/sage/mds/mds/Migrator.h
@@ -200,7 +200,7 @@ public:
   }
   void clear_export_proxy_pins(CDir *dir);
 
-  void export_caps(CInode *in, int dest);
+  void export_caps(CInode *in);
 
  protected:
   void handle_export_discover_ack(MExportDirDiscoverAck *m);
@@ -216,6 +216,7 @@ public:
 
   void handle_export_caps_ack(MExportCapsAck *m);
 
+
   friend class C_MDC_ExportFreeze;
   friend class C_MDS_ExportFinishLogged;
   friend class C_M_ExportGo;
@@ -270,7 +271,6 @@ protected:
   // bystander
   void handle_export_notify(MExportDirNotify *m);
 
-
 };
 
 
diff --git a/branches/sage/mds/mds/Server.cc b/branches/sage/mds/mds/Server.cc
index 3a45b625f8a53..56d6960bba33c 100644
--- a/branches/sage/mds/mds/Server.cc
+++ b/branches/sage/mds/mds/Server.cc
@@ -93,7 +93,7 @@ void Server::dispatch(Message *m)
   }
 
   // active?
-  if (!mds->is_active()) {
+  if (!mds->is_active() && !mds->is_stopping()) {
     dout(3) << "not active yet, waiting" << dendl;
     mds->wait_for_active(new C_MDS_RetryMessage(mds, m));
     return;
@@ -196,7 +196,7 @@ void Server::_session_logged(entity_inst_t client_inst, bool open, version_t cma
     mds->clientmap.close_session(from);
     
     // purge completed requests from clientmap
-    mds->clientmap.trim_completed_requests(from, 0);
+    mds->clientmap.trim_completed_requests(client_inst.name, 0);
   } else {
     // close must have been canceled (by an import?) ...
     assert(!open);
@@ -459,9 +459,14 @@ void Server::reply_request(MDRequest *mdr, MClientReply *reply, CInode *tracei)
   if (tracei) {
     reply->set_trace_dist( tracei, mds->get_nodeid() );
   }
+
+  reply->set_mdsmap_epoch(mds->mdsmap->get_epoch());
   
   // send reply
-  messenger->send_message(reply, req->get_client_inst());
+  if (req->get_client_inst().name.is_mds())
+    delete reply;   // mds doesn't need a reply
+  else
+    messenger->send_message(reply, req->get_client_inst());
   
   // finish request
   mdcache->request_finish(mdr);
@@ -477,12 +482,12 @@ void Server::reply_request(MDRequest *mdr, MClientReply *reply, CInode *tracei)
 void Server::handle_client_request(MClientRequest *req)
 {
   dout(4) << "handle_client_request " << *req << dendl;
-  int client = req->get_client();
 
   if (logger) logger->inc("hcreq");
 
-  if (!mds->is_active()) {
-    dout(5) << " not active, discarding client request." << dendl;
+  if (!mds->is_active() &&
+      !(mds->is_stopping() && req->get_client_inst().name.is_mds())) {
+    dout(5) << " not active (or stopping+mds), discarding request." << dendl;
     delete req;
     return;
   }
@@ -494,12 +499,18 @@ void Server::handle_client_request(MClientRequest *req)
   }
 
   // active session?
-  if (!mds->clientmap.have_session(client)) {
-    dout(5) << "no session for client" << client << ", dropping" << dendl;
+  if (req->get_client_inst().name.is_client() && 
+      !mds->clientmap.have_session(req->get_client_inst().name.num())) {
+    dout(5) << "no session for " << req->get_client_inst().name << ", dropping" << dendl;
     delete req;
     return;
   }
 
+  // old mdsmap?
+  if (req->get_mdsmap_epoch() < mds->mdsmap->get_epoch()) {
+    // send it?  hrm, this isn't ideal; they may get a lot of copies if
+    // they have a high request rate.
+  }
 
   // okay, i want
   CInode *ref = 0;
@@ -508,8 +519,7 @@ void Server::handle_client_request(MClientRequest *req)
   if (req->get_retry_attempt()) {
     if (mds->clientmap.have_completed_request(req->get_reqid())) {
       dout(5) << "already completed " << req->get_reqid() << dendl;
-      mds->messenger->send_message(new MClientReply(req, 0),
-				   req->get_client_inst());
+      mds->messenger->send_message(new MClientReply(req, 0), req->get_client_inst());
       delete req;
       return;
     }
@@ -517,7 +527,7 @@ void Server::handle_client_request(MClientRequest *req)
   // trim completed_request list
   if (req->get_oldest_client_tid() > 0) {
     dout(15) << " oldest_client_tid=" << req->get_oldest_client_tid() << dendl;
-    mds->clientmap.trim_completed_requests(client,
+    mds->clientmap.trim_completed_requests(req->get_client_inst().name, 
 					   req->get_oldest_client_tid());
   }
 
@@ -1046,7 +1056,7 @@ CDir *Server::traverse_to_auth_dir(MDRequest *mdr, vector<CDentry*> &trace, file
 
   // traverse to parent dir
   int r = mdcache->path_traverse(mdr, mdr->client_request,
-				 0, refpath, trace, true,
+				 refpath, trace, true,
 				 MDS_TRAVERSE_FORWARD);
   if (r > 0) return 0; // delayed
   if (r < 0) {
@@ -1057,7 +1067,7 @@ CDir *Server::traverse_to_auth_dir(MDRequest *mdr, vector<CDentry*> &trace, file
   // open inode
   CInode *diri;
   if (trace.empty())
-    diri = mdcache->get_root();
+    diri = mdcache->get_inode(refpath.get_ino());
   else
     diri = mdcache->get_dentry_inode(trace[trace.size()-1], mdr);
   if (!diri) 
@@ -1086,7 +1096,7 @@ CInode* Server::rdlock_path_pin_ref(MDRequest *mdr, bool want_auth)
   filepath refpath = req->get_filepath();
   vector<CDentry*> trace;
   int r = mdcache->path_traverse(mdr, req,
-				 0, refpath, 
+				 refpath, 
 				 trace, req->follow_trailing_symlink(),
 				 MDS_TRAVERSE_FORWARD);
   if (r > 0) return false; // delayed
@@ -1296,7 +1306,7 @@ version_t Server::predirty_dn_diri(MDRequest *mdr, CDentry *dn, EMetaBlob *blob)
   version_t dirpv = 0;
   CInode *diri = dn->dir->inode;
   
-  if (diri->is_root()) return 0;
+  if (diri->is_base()) return 0;
 
   if (diri->is_auth()) {
     assert(mdr->wrlocks.count(&diri->dirlock));
@@ -1717,8 +1727,8 @@ void Server::handle_client_mknod(MDRequest *mdr)
   // it's a file.
   newi->inode.rdev = req->args.mknod.rdev;
   newi->inode.mode = req->args.mknod.mode;
-  newi->inode.mode &= ~INODE_TYPE_MASK;
-  newi->inode.mode |= INODE_MODE_FILE;
+  newi->inode.mode &= ~S_IFMT;
+  newi->inode.mode |= S_IFREG;
   newi->inode.version = dn->pre_dirty() - 1;
   
   // prepare finisher
@@ -1752,8 +1762,8 @@ void Server::handle_client_mkdir(MDRequest *mdr)
 
   // it's a directory.
   newi->inode.mode = req->args.mkdir.mode;
-  newi->inode.mode &= ~INODE_TYPE_MASK;
-  newi->inode.mode |= INODE_MODE_DIR;
+  newi->inode.mode &= ~S_IFMT;
+  newi->inode.mode |= S_IFDIR;
   newi->inode.layout = g_OSD_MDDirLayout;
   newi->inode.version = dn->pre_dirty() - 1;
 
@@ -1808,9 +1818,9 @@ void Server::handle_client_symlink(MDRequest *mdr)
   assert(newi);
 
   // it's a symlink
-  newi->inode.mode &= ~INODE_TYPE_MASK;
-  newi->inode.mode |= INODE_MODE_SYMLINK;
-  newi->symlink = req->get_sarg();
+  newi->inode.mode &= ~S_IFMT;
+  newi->inode.mode |= S_IFLNK;
+  newi->symlink = req->get_path2();
   newi->inode.version = dn->pre_dirty() - 1;
 
   // prepare finisher
@@ -1837,7 +1847,7 @@ void Server::handle_client_link(MDRequest *mdr)
   MClientRequest *req = mdr->client_request;
 
   dout(7) << "handle_client_link " << req->get_filepath()
-	  << " to " << req->get_sarg()
+	  << " to " << req->get_filepath2()
 	  << dendl;
 
   // traverse to dest dir, make sure it's ours.
@@ -1849,11 +1859,11 @@ void Server::handle_client_link(MDRequest *mdr)
   dout(7) << "handle_client_link link " << dname << " in " << *dir << dendl;
   
   // traverse to link target
-  filepath targetpath = req->get_sarg();
+  filepath targetpath = req->get_filepath2();
   dout(7) << "handle_client_link discovering target " << targetpath << dendl;
   vector<CDentry*> targettrace;
   int r = mdcache->path_traverse(mdr, req,
-				 0, targetpath, targettrace, false,
+				 targetpath, targettrace, false,
 				 MDS_TRAVERSE_DISCOVER);
   if (r > 0) return; // wait
   if (targettrace.empty()) r = -EINVAL; 
@@ -2267,7 +2277,7 @@ void Server::handle_client_unlink(MDRequest *mdr)
   // traverse to path
   vector<CDentry*> trace;
   int r = mdcache->path_traverse(mdr, req, 
-				 0, req->get_filepath(), trace, false,
+				 req->get_filepath(), trace, false,
 				 MDS_TRAVERSE_FORWARD);
   if (r > 0) return;
   if (trace.empty()) r = -EINVAL;   // can't unlink root
@@ -2697,7 +2707,7 @@ void Server::handle_client_rename(MDRequest *mdr)
   // traverse to dest dir (not dest)
   //  we do this FIRST, because the rename should occur on the 
   //  destdn's auth.
-  const filepath &destpath = req->get_sarg();
+  const filepath &destpath = req->get_filepath2();
   const string &destname = destpath.last_dentry();
   vector<CDentry*> desttrace;
   CDir *destdir = traverse_to_auth_dir(mdr, desttrace, destpath);
@@ -2709,7 +2719,7 @@ void Server::handle_client_rename(MDRequest *mdr)
   filepath srcpath = req->get_filepath();
   vector<CDentry*> srctrace;
   int r = mdcache->path_traverse(mdr, req,
-				 0, srcpath, srctrace, false,
+				 srcpath, srctrace, false,
 				 MDS_TRAVERSE_DISCOVER);
   if (r > 0) return;
   if (srctrace.empty()) r = -EINVAL;  // can't rename root
@@ -3257,7 +3267,6 @@ void Server::_rename_apply(MDRequest *mdr, CDentry *srcdn, CDentry *destdn, CDen
       // srcdn inode import?
       if (!srcdn->is_auth() && destdn->is_auth()) {
 	assert(mdr->more()->inode_import.length() > 0);
-	assert(destdn->inode->is_dirty());
 
 	// finish cap imports
 	finish_force_open_sessions(mdr->more()->imported_client_map);
@@ -3328,7 +3337,7 @@ void Server::handle_slave_rename_prep(MDRequest *mdr)
   dout(10) << " dest " << destpath << dendl;
   vector<CDentry*> trace;
   int r = mdcache->path_traverse(mdr, mdr->slave_request, 
-				 0, destpath, trace, false,
+				 destpath, trace, false,
 				 MDS_TRAVERSE_DISCOVERXLOCK);
   if (r > 0) return;
   assert(r == 0);  // we shouldn't get an error here!
@@ -3342,7 +3351,7 @@ void Server::handle_slave_rename_prep(MDRequest *mdr)
   filepath srcpath(mdr->slave_request->srcdnpath);
   dout(10) << " src " << srcpath << dendl;
   r = mdcache->path_traverse(mdr, mdr->slave_request,
-			     0, srcpath, trace, false,  
+			     srcpath, trace, false,  
 			     MDS_TRAVERSE_DISCOVERXLOCK);
   if (r > 0) return;
   assert(r == 0);  // we shouldn't get an error here!
@@ -3740,11 +3749,16 @@ void Server::handle_client_open(MDRequest *mdr)
   if (!cur) return;
   
   // regular file?
-  if ((cur->inode.mode & INODE_TYPE_MASK) != INODE_MODE_FILE) {
-    dout(7) << "not a regular file " << *cur << dendl;
+  if (!cur->inode.is_file() && !cur->inode.is_dir()) {
+    dout(7) << "not a file or dir " << *cur << dendl;
     reply_request(mdr, -EINVAL);                 // FIXME what error do we want?
     return;
   }
+  // can only open a dir rdonly, no flags.
+  if (cur->inode.is_dir() && (cmode != FILE_MODE_R || flags != 0)) {
+    reply_request(mdr, -EINVAL);
+    return;
+  }
 
   // hmm, check permissions or something.
 
@@ -4007,7 +4021,7 @@ void Server::handle_client_openc(MDRequest *mdr)
   
   // it's a file.
   in->inode.mode = req->args.open.mode;
-  in->inode.mode |= INODE_MODE_FILE;
+  in->inode.mode |= S_IFREG;
   in->inode.version = dn->pre_dirty() - 1;
   
   // prepare finisher
diff --git a/branches/sage/mds/mds/journal.cc b/branches/sage/mds/mds/journal.cc
index 2cf32d99ca337..0d08b4e6aa456 100644
--- a/branches/sage/mds/mds/journal.cc
+++ b/branches/sage/mds/mds/journal.cc
@@ -756,7 +756,7 @@ void ESession::replay(MDS *mds)
 
     // hrm, this isn't very pretty.
     if (!open)
-      mds->clientmap.trim_completed_requests(client_inst.name.num(), 0);
+      mds->clientmap.trim_completed_requests(client_inst.name, 0);
 
   } else {
     dout(10) << "ESession.replay clientmap " << mds->clientmap.get_version() 
@@ -766,7 +766,7 @@ void ESession::replay(MDS *mds)
       mds->clientmap.open_session(client_inst);
     } else {
       mds->clientmap.close_session(client_inst.name.num());
-      mds->clientmap.trim_completed_requests(client_inst.name.num(), 0);
+      mds->clientmap.trim_completed_requests(client_inst.name, 0);
     }
     mds->clientmap.reset_projected(); // make it follow version.
   }
diff --git a/branches/sage/mds/mds/mdstypes.h b/branches/sage/mds/mds/mdstypes.h
index a2f779757255e..f64bb908e9498 100644
--- a/branches/sage/mds/mds/mdstypes.h
+++ b/branches/sage/mds/mds/mdstypes.h
@@ -54,30 +54,30 @@ using namespace std;
 
 
 struct metareqid_t {
+  entity_name_t name;
   uint64_t tid;
-  int32_t client;
-  int32_t _pad;
-  metareqid_t() : tid(0), client(-1), _pad(0) {}
-  metareqid_t(int c, tid_t t) : tid(t), client(c), _pad(0) {}
+  metareqid_t() : tid(0) {}
+  //metareqid_t(int c, tid_t t) : tid(t) { name = entity_name_t::CLIENT(c); }
+  metareqid_t(entity_name_t n, tid_t t) : name(n), tid(t) {}
 };
 
 inline ostream& operator<<(ostream& out, const metareqid_t& r) {
-  return out << "client" << r.client << ":" << r.tid;
+  return out << r.name << ":" << r.tid;
 }
 
 inline bool operator==(const metareqid_t& l, const metareqid_t& r) {
-  return (l.client == r.client) && (l.tid == r.tid);
+  return (l.name == r.name) && (l.tid == r.tid);
 }
 inline bool operator!=(const metareqid_t& l, const metareqid_t& r) {
-  return (l.client != r.client) || (l.tid != r.tid);
+  return (l.name != r.name) || (l.tid != r.tid);
 }
 inline bool operator<(const metareqid_t& l, const metareqid_t& r) {
-  return (l.client < r.client) || 
-    (l.client == r.client && l.tid < r.tid);
+  return (l.name < r.name) || 
+    (l.name == r.name && l.tid < r.tid);
 }
 inline bool operator<=(const metareqid_t& l, const metareqid_t& r) {
-  return (l.client < r.client) ||
-    (l.client == r.client && l.tid <= r.tid);
+  return (l.name < r.name) ||
+    (l.name == r.name && l.tid <= r.tid);
 }
 inline bool operator>(const metareqid_t& l, const metareqid_t& r) { return !(l <= r); }
 inline bool operator>=(const metareqid_t& l, const metareqid_t& r) { return !(l < r); }
@@ -86,7 +86,7 @@ namespace __gnu_cxx {
   template<> struct hash<metareqid_t> {
     size_t operator()(const metareqid_t &r) const { 
       hash<uint64_t> H;
-      return H(r.client) ^ H(r.tid);
+      return H(r.name.num()) ^ H(r.name.type()) ^ H(r.tid);
     }
   };
 }
diff --git a/branches/sage/mds/messages/MClientFileCaps.h b/branches/sage/mds/messages/MClientFileCaps.h
index e6fe60b37343b..758205902b41b 100644
--- a/branches/sage/mds/messages/MClientFileCaps.h
+++ b/branches/sage/mds/messages/MClientFileCaps.h
@@ -18,13 +18,19 @@
 #include "msg/Message.h"
 #include "mds/Capability.h"
 
+
 class MClientFileCaps : public Message {
  public:
   static const int OP_GRANT   = 0;  // mds->client grant.
   static const int OP_ACK     = 1;  // client->mds ack (if prior grant was a recall)
-  static const int OP_RELEASE = 2;  // mds closed the cap
+  static const int OP_RELEASE = 2;  // mds->client release cap (*)
   static const int OP_EXPORT  = 3;  // mds has exported the cap
   static const int OP_IMPORT  = 4;  // mds has imported the cap from get_mds()
+  /* 
+   * (*) it's a bit counterintuitive, but the mds has to 
+   *  close the cap because the client isn't able to tell
+   *  if a concurrent open() would map to the same inode.
+   */
   static const char* get_opname(int op) {
     switch (op) {
     case OP_GRANT: return "grant";
diff --git a/branches/sage/mds/messages/MClientReply.h b/branches/sage/mds/messages/MClientReply.h
index 760dcc971ebad..f50384000f2f5 100644
--- a/branches/sage/mds/messages/MClientReply.h
+++ b/branches/sage/mds/messages/MClientReply.h
@@ -127,6 +127,7 @@ class MClientReply : public Message {
   // reply data
   struct st_ {
     long tid;
+    epoch_t mdsmap_epoch;
     int op;
     int result;  // error code
     unsigned char file_caps;  // for open
@@ -150,6 +151,9 @@ class MClientReply : public Message {
   long get_tid() { return st.tid; }
   int get_op() { return st.op; }
 
+  void set_mdsmap_epoch(epoch_t e) { st.mdsmap_epoch = e; }
+  epoch_t get_mdsmap_epoch() { return st.mdsmap_epoch; }
+
   int get_result() { return st.result; }
   const string& get_path() { return path; }
 
diff --git a/branches/sage/mds/messages/MClientRequest.h b/branches/sage/mds/messages/MClientRequest.h
index 8f03044cf5a4f..d0c4a57a4af8a 100644
--- a/branches/sage/mds/messages/MClientRequest.h
+++ b/branches/sage/mds/messages/MClientRequest.h
@@ -81,15 +81,16 @@ class MClientRequest : public Message {
     inodeno_t  mds_wants_replica_in_dirino;
 
     entity_inst_t client_inst;
+    epoch_t mdsmap_epoch;
 
     int op;
     int caller_uid, caller_gid;
     inodeno_t cwd_ino;
+
   } st;
 
   // path arguments
-  filepath path;
-  string sarg;
+  filepath path, path2;
 
  public:
   // fixed size arguments.  in a union.
@@ -146,9 +147,12 @@ class MClientRequest : public Message {
     this->st.client_inst = ci;
   }
 
+  void set_mdsmap_epoch(epoch_t e) { st.mdsmap_epoch = e; }
+  epoch_t get_mdsmap_epoch() { return st.mdsmap_epoch; }
+
   metareqid_t get_reqid() {
     // FIXME: for now, assume clients always have 1 incarnation
-    return metareqid_t(st.client_inst.name.num(), st.tid); 
+    return metareqid_t(st.client_inst.name, st.tid); 
   }
 
   int get_open_file_mode() {
@@ -214,18 +218,19 @@ class MClientRequest : public Message {
   void set_retry_attempt(int a) { st.retry_attempt = a; }
   void set_path(string& p) { path.set_path(p); }
   void set_path(const char *p) { path.set_path(p); }
-  void set_path(const filepath& fp) { path = fp; }
+  void set_filepath(const filepath& fp) { path = fp; }
+  void set_path2(string& p) { path2.set_path(p); }
+  void set_path2(const char *p) { path2.set_path(p); }
+  void set_filepath2(const filepath& fp) { path2 = fp; }
   void set_caller_uid(int u) { st.caller_uid = u; }
   void set_caller_gid(int g) { st.caller_gid = g; }
-  void set_sarg(string& arg) { this->sarg = arg; }
-  void set_sarg(const char *arg) { this->sarg = arg; }
   void set_mds_wants_replica_in_dirino(inodeno_t dirino) { 
     st.mds_wants_replica_in_dirino = dirino; }
   
   void set_client_inst(const entity_inst_t& i) { st.client_inst = i; }
   const entity_inst_t& get_client_inst() { return st.client_inst; }
+  entity_name_t get_client() { return st.client_inst.name; }
 
-  int get_client() { return st.client_inst.name.num(); }
   tid_t get_tid() { return st.tid; }
   tid_t get_oldest_client_tid() { return st.oldest_client_tid; }
   int get_num_fwd() { return st.num_fwd; }
@@ -233,10 +238,11 @@ class MClientRequest : public Message {
   int get_op() { return st.op; }
   int get_caller_uid() { return st.caller_uid; }
   int get_caller_gid() { return st.caller_gid; }
-  //inodeno_t get_ino() { return st.ino; }
   const string& get_path() { return path.get_path(); }
   filepath& get_filepath() { return path; }
-  string& get_sarg() { return sarg; }
+  const string& get_path2() { return path.get_path(); }
+  filepath& get_filepath2() { return path2; }
+
   inodeno_t get_mds_wants_replica_in_dirino() { 
     return st.mds_wants_replica_in_dirino; }
 
@@ -249,19 +255,19 @@ class MClientRequest : public Message {
     payload.copy(off, sizeof(args), (char*)&args);
     off += sizeof(args);
     path._decode(payload, off);
-    ::_decode(sarg, payload, off);
+    path2._decode(payload, off);
   }
 
   void encode_payload() {
     payload.append((char*)&st, sizeof(st));
     payload.append((char*)&args, sizeof(args));
     path._encode(payload);
-    ::_encode(sarg, payload);
+    path2._encode(payload);
   }
 
   char *get_type_name() { return "creq"; }
   void print(ostream& out) {
-    out << "clientreq(client" << get_client() 
+    out << "clientreq(" << get_client() 
 	<< "." << get_tid() 
 	<< " ";
     switch(get_op()) {
@@ -311,10 +317,10 @@ class MClientRequest : public Message {
       out << "unknown=" << get_op();
       assert(0);
     }
-    if (get_path().length()) 
-      out << " " << get_path();
-    if (get_sarg().length())
-      out << " " << get_sarg();
+    if (!get_filepath().empty()) 
+      out << " " << get_filepath();
+    if (!get_filepath2().empty())
+      out << " " << get_filepath2();
     if (st.retry_attempt)
       out << " RETRY=" << st.retry_attempt;
     out << ")";
diff --git a/branches/sage/mds/messages/MDirUpdate.h b/branches/sage/mds/messages/MDirUpdate.h
index 0db32208efd45..87d7e4fa7389b 100644
--- a/branches/sage/mds/messages/MDirUpdate.h
+++ b/branches/sage/mds/messages/MDirUpdate.h
@@ -25,14 +25,14 @@ class MDirUpdate : public Message {
     int discover;
   } st;
   set<int> dir_rep_by;
-  string path;
+  filepath path;
 
  public:
   dirfrag_t get_dirfrag() { return st.dirfrag; }
   int get_dir_rep() { return st.dir_rep; }
   set<int>& get_dir_rep_by() { return dir_rep_by; } 
   bool should_discover() { return st.discover > 0; }
-  string& get_path() { return path; }
+  filepath& get_path() { return path; }
 
   void tried_discover() {
     if (st.discover) st.discover--;
@@ -42,7 +42,7 @@ class MDirUpdate : public Message {
   MDirUpdate(dirfrag_t dirfrag,
              int dir_rep,
              set<int>& dir_rep_by,
-             string& path,
+             filepath& path,
              bool discover = false) :
     Message(MSG_MDS_DIRUPDATE) {
     this->st.dirfrag = dirfrag;
@@ -52,19 +52,22 @@ class MDirUpdate : public Message {
     this->path = path;
   }
   virtual char *get_type_name() { return "dir_update"; }
+  void print(ostream& out) {
+    out << "dir_update(" << get_dirfrag() << ")";
+  }
 
   virtual void decode_payload() {
     int off = 0;
     payload.copy(off, sizeof(st), (char*)&st);
     off += sizeof(st);
     ::_decode(dir_rep_by, payload, off);
-    ::_decode(path, payload, off);
+    path._decode(payload, off);
   }
 
   virtual void encode_payload() {
     payload.append((char*)&st, sizeof(st));
     ::_encode(dir_rep_by, payload);
-    ::_encode(path, payload);
+    path._encode(payload);
   }
 };
 
diff --git a/branches/sage/mds/messages/MExportDirDiscover.h b/branches/sage/mds/messages/MExportDirDiscover.h
index c311d1e87e940..01c61a67648c3 100644
--- a/branches/sage/mds/messages/MExportDirDiscover.h
+++ b/branches/sage/mds/messages/MExportDirDiscover.h
@@ -21,12 +21,12 @@
 
 class MExportDirDiscover : public Message {
   dirfrag_t dirfrag;
-  string path;
+  filepath path;
 
  public:
   inodeno_t get_ino() { return dirfrag.ino; }
   dirfrag_t get_dirfrag() { return dirfrag; }
-  string& get_path() { return path; }
+  filepath& get_path() { return path; }
 
   bool started;
 
@@ -47,12 +47,12 @@ class MExportDirDiscover : public Message {
   virtual void decode_payload() {
     bufferlist::iterator p = payload.begin();
     ::_decode_simple(dirfrag, p);
-    ::_decode_simple(path, p);
+    path._decode(p);
   }
 
   virtual void encode_payload() {
     ::_encode_simple(dirfrag, payload);
-    ::_encode_simple(path, payload);
+    path._encode(payload);
   }
 };
 
diff --git a/branches/sage/mds/messages/MExportStrays.h b/branches/sage/mds/messages/MExportStrays.h
deleted file mode 100644
index be055c0312c89..0000000000000
--- a/branches/sage/mds/messages/MExportStrays.h
+++ /dev/null
@@ -1,43 +0,0 @@
-// -*- mode:C++; tab-width:8; c-basic-offset:2; indent-tabs-mode:t -*- 
-// vim: ts=8 sw=2 smarttab
-/*
- * Ceph - scalable distributed file system
- *
- * Copyright (C) 2004-2006 Sage Weil <sage@newdream.net>
- *
- * This is free software; you can redistribute it and/or
- * modify it under the terms of the GNU Lesser General Public
- * License version 2.1, as published by the Free Software 
- * Foundation.  See file COPYING.
- * 
- */
-
-
-#ifndef __MEXPORTSTRAYS_H
-#define __MEXPORTSTRAYS_H
-
-#include "msg/Message.h"
-
-
-class MExportStrays : public Message {
- public:  
-  bufferlist state;
-
-  MExportStrays() :
-    Message(MSG_MDS_EXPORTSTRAYS) {}
-
-  virtual char *get_type_name() { return "SEx"; }
-  void print(ostream& o) {
-    o << "export_strays";
-  }
-
-  virtual void decode_payload() {
-    state = payload;
-  }
-  virtual void encode_payload() {
-    payload = state;
-  }
-
-};
-
-#endif
diff --git a/branches/sage/mds/messages/MMDSSlaveRequest.h b/branches/sage/mds/messages/MMDSSlaveRequest.h
index 5ef65223ec1c9..a5c2339fd4cd5 100644
--- a/branches/sage/mds/messages/MMDSSlaveRequest.h
+++ b/branches/sage/mds/messages/MMDSSlaveRequest.h
@@ -77,8 +77,8 @@ class MMDSSlaveRequest : public Message {
 
  public:
   // for rename prep
-  string srcdnpath;
-  string destdnpath;
+  filepath srcdnpath;
+  filepath destdnpath;
   set<int> witnesses;
   bufferlist inode_export;
   version_t inode_export_v;
@@ -110,8 +110,8 @@ public:
     ::_encode(lock_type, payload);
     object_info._encode(payload);
     ::_encode_complex(authpins, payload);
-    ::_encode(srcdnpath, payload);
-    ::_encode(destdnpath, payload);
+    srcdnpath._encode(payload);
+    destdnpath._encode(payload);
     ::_encode(witnesses, payload);
     ::_encode(now, payload);
     ::_encode(inode_export, payload);
@@ -126,8 +126,8 @@ public:
     ::_decode_simple(lock_type, p);
     object_info._decode(p);
     ::_decode_complex(authpins, p);
-    ::_decode_simple(srcdnpath, p);
-    ::_decode_simple(destdnpath, p);
+    srcdnpath._decode(p);
+    destdnpath._decode(p);
     ::_decode_simple(witnesses, p);
     ::_decode_simple(now, p);
     ::_decode_simple(inode_export, p);
diff --git a/branches/sage/mds/msg/FakeMessenger.cc b/branches/sage/mds/msg/FakeMessenger.cc
index ee80df3dc0626..77434c35e64da 100644
--- a/branches/sage/mds/msg/FakeMessenger.cc
+++ b/branches/sage/mds/msg/FakeMessenger.cc
@@ -291,7 +291,8 @@ FakeMessenger::FakeMessenger(entity_name_t me)  : Messenger(me)
   lock.Unlock();
 
 
-  dout(0) << "fakemessenger " << get_myname() << " messenger is " << this << " at " << _myinst << dendl;
+  dout(0) << "fakemessenger " << get_myname() << " messenger is " << this 
+	  << " at " << get_myaddr() << dendl;
 
   qlen = 0;
 
@@ -341,12 +342,8 @@ int FakeMessenger::shutdown()
 void FakeMessenger::reset_myname(entity_name_t m)
 {
   dout(1) << "reset_myname from " << get_myname() << " to " << m << dendl;
-  _set_myname(m);
-
-  directory.erase(_myinst.addr);
   _myinst.name = m;
-  directory[_myinst.addr] = this;
-  
+
   // put myself in the fail queue?
   if (g_fake_kill_after.count(m)) {
     utime_t w = start_time;
diff --git a/branches/sage/mds/msg/FakeMessenger.h b/branches/sage/mds/msg/FakeMessenger.h
index 2284ea110b51f..57379133812a9 100644
--- a/branches/sage/mds/msg/FakeMessenger.h
+++ b/branches/sage/mds/msg/FakeMessenger.h
@@ -32,8 +32,6 @@ class FakeMessenger : public Messenger {
   int    qlen;
   list<Message*>       incoming;        // incoming queue
 
-  entity_inst_t _myinst;
-
  public:
   bool failed;
 
@@ -42,13 +40,6 @@ class FakeMessenger : public Messenger {
 
   virtual int shutdown();
 
-  const entity_inst_t& get_myinst() {
-    return _myinst;
-  };
-  const entity_addr_t& get_myaddr() {
-    return _myinst.addr;
-  }
-
   void reset_myname(entity_name_t m);
 
   // msg interface
diff --git a/branches/sage/mds/msg/Message.cc b/branches/sage/mds/msg/Message.cc
index ebffe647c62f0..e3de6ce0677ba 100644
--- a/branches/sage/mds/msg/Message.cc
+++ b/branches/sage/mds/msg/Message.cc
@@ -84,6 +84,7 @@ using namespace std;
 #include "messages/MExportCaps.h"
 #include "messages/MExportCapsAck.h"
 
+
 #include "messages/MDentryUnlink.h"
 
 #include "messages/MHeartbeat.h"
diff --git a/branches/sage/mds/msg/Message.h b/branches/sage/mds/msg/Message.h
index b570722ad1888..cb106251a425b 100644
--- a/branches/sage/mds/msg/Message.h
+++ b/branches/sage/mds/msg/Message.h
@@ -127,12 +127,6 @@
 #define MSG_MDS_EXPORTDIRNOTIFYACK    159
 #define MSG_MDS_EXPORTDIRFINISH       160
 
-#define MSG_MDS_EXPORTSTRAY           161
-#define MSG_MDS_EXPORTSTRAYNOTIFY     162
-#define MSG_MDS_EXPORTSTRAYNOTIFYACK  163
-#define MSG_MDS_EXPORTSTRAYACK        164
-#define MSG_MDS_EXPORTSTRAYFINISH     165
-
 #define MSG_MDS_EXPORTCAPS            166
 #define MSG_MDS_EXPORTCAPSACK         167
 
diff --git a/branches/sage/mds/msg/Messenger.h b/branches/sage/mds/msg/Messenger.h
index d29441a744ca0..85ac2745ad0d7 100644
--- a/branches/sage/mds/msg/Messenger.h
+++ b/branches/sage/mds/msg/Messenger.h
@@ -34,21 +34,23 @@ class Timer;
 class Messenger {
  private:
   Dispatcher          *dispatcher;
-  entity_name_t           _myname;
+
+protected:
+  entity_inst_t _myinst;
 
  public:
-  Messenger(entity_name_t w) : dispatcher(0), _myname(w) { }
+  Messenger(entity_name_t w) : dispatcher(0) {
+    _myinst.name = w;
+  }
   virtual ~Messenger() { }
   
   // accessors
-  entity_name_t get_myname() { return _myname; }
-  void _set_myname(entity_name_t m) { _myname = m; }
-
+  entity_name_t get_myname() { return _myinst.name; }
+  const entity_addr_t& get_myaddr() { return _myinst.addr; }
+  const entity_inst_t& get_myinst() { return _myinst; }
+  
+  void _set_myname(entity_name_t m) { _myinst.name = m; }
   virtual void reset_myname(entity_name_t m) = 0;
-
-  virtual const entity_addr_t &get_myaddr() = 0;
-
-  entity_inst_t get_myinst() { return entity_inst_t(_myname, get_myaddr()); }
   
   // hrmpf.
   virtual int get_dispatch_queue_len() { return 0; };
diff --git a/branches/sage/mds/msg/SimpleMessenger.cc b/branches/sage/mds/msg/SimpleMessenger.cc
index 7e29f033d83b5..51ff661ebb7ec 100644
--- a/branches/sage/mds/msg/SimpleMessenger.cc
+++ b/branches/sage/mds/msg/SimpleMessenger.cc
@@ -1379,13 +1379,6 @@ int Rank::EntityMessenger::send_first_message(Dispatcher *d,
   return 0;
 }
 
-
-const entity_addr_t &Rank::EntityMessenger::get_myaddr()
-{
-  return rank.my_addr;
-}
-
-
 void Rank::EntityMessenger::reset_myname(entity_name_t newname)
 {
   rank.lock.Lock();
@@ -1402,8 +1395,6 @@ void Rank::EntityMessenger::reset_myname(entity_name_t newname)
 }
 
 
-
-
 void Rank::EntityMessenger::mark_down(entity_addr_t a)
 {
   rank.mark_down(a);
@@ -1411,34 +1402,8 @@ void Rank::EntityMessenger::mark_down(entity_addr_t a)
 
 void Rank::mark_down(entity_addr_t addr)
 {
-  //if (my_rank == 0) return;   // ugh.. rank0 already handles this stuff in the namer
   lock.Lock();
-  /*
-  if (entity_map.count(a) &&
-      entity_map[a] > inst) {
-    dout(10) << "mark_down " << a << " inst " << inst << " < " << entity_map[a] << dendl;
-    derr(10) << "mark_down " << a << " inst " << inst << " < " << entity_map[a] << dendl;
-    // do nothing!
-  } else {
-    if (entity_map.count(a) == 0) {
-      // don't know it
-      dout(10) << "mark_down " << a << " inst " << inst << " ... unknown by me" << dendl;
-      derr(10) << "mark_down " << a << " inst " << inst << " ... unknown by me" << dendl;
-    } else {
-      // know it
-      assert(entity_map[a] <= inst);
-      dout(10) << "mark_down " << a << " inst " << inst << dendl;
-      derr(10) << "mark_down " << a << " inst " << inst << dendl;
-      
-      entity_map.erase(a);
-      
-      if (rank_pipe.count(inst)) {
-	rank_pipe[inst]->close();
-	rank_pipe.erase(inst);
-      }
-    }
-  }
-  */
+  // FIXME
   lock.Unlock();
 }
 
diff --git a/branches/sage/mds/msg/SimpleMessenger.h b/branches/sage/mds/msg/SimpleMessenger.h
index 6bd417adc8e10..ae888a8f461b6 100644
--- a/branches/sage/mds/msg/SimpleMessenger.h
+++ b/branches/sage/mds/msg/SimpleMessenger.h
@@ -197,8 +197,8 @@ private:
     }
 
   public:
-    EntityMessenger(entity_name_t myaddr) : 
-      Messenger(myaddr),
+    EntityMessenger(entity_name_t name) : 
+      Messenger(name),
       stop(false),
       qlen(0), pqlen(0),
       dispatch_thread(this) { }
@@ -215,8 +215,6 @@ private:
       dispatch_thread.join();
     }
     
-    const entity_addr_t &get_myaddr();
-
     int get_dispatch_queue_len() { return qlen + pqlen; }
 
     void reset_myname(entity_name_t m);
diff --git a/branches/sage/mds/msg/msg_types.h b/branches/sage/mds/msg/msg_types.h
index 52b1e69c8886c..d5ce108583564 100644
--- a/branches/sage/mds/msg/msg_types.h
+++ b/branches/sage/mds/msg/msg_types.h
@@ -33,7 +33,7 @@ public:
 
   // cons
   entity_name_t() { v.type = v.num = 0; }
-  entity_name_t(int t, int n=NEW) { v.type = t; v.num = n; }
+  entity_name_t(int t, int n) { v.type = t; v.num = n; }
 
   // static cons
   static entity_name_t MON(int i=NEW) { return entity_name_t(TYPE_MON, i); }
-- 
2.39.5