Merge remote-tracking branch 'gh/next'

19 files changed, 387 insertions, 188 deletions

diff --git a/README b/README
index d8f03f52eb4..2ac68504d87 100644
--- a/README
+++ b/README
@@ -102,8 +102,10 @@ To build the source code, you must install the following:
 
 - automake
 - autoconf
+- pkg-config
 - gcc
 - g++
+- make
 - libboost-dev
 - libedit-dev
 - libssl-dev
@@ -123,7 +125,8 @@ To build the source code, you must install the following:
 - libexpat1-dev
 - libleveldb-dev
 - libsnappy-dev
+- libcurl4-gnutls-dev
 
 For example:
 
-	$ apt-get install automake autoconf gcc g++ libboost-dev libedit-dev libssl-dev libtool libfcgi libfcgi-dev libfuse-dev linux-kernel-headers libcrypto++-dev libaio-dev libgoogle-perftools-dev libkeyutils-dev uuid-dev libatomic-ops-dev libboost-program-options-dev libboost-thread-dev libexpat1-dev libleveldb-dev libsnappy-dev
+	$ apt-get install automake autoconf pkg-config gcc g++ make libboost-dev libedit-dev libssl-dev libtool libfcgi libfcgi-dev libfuse-dev linux-kernel-headers libcrypto++-dev libaio-dev libgoogle-perftools-dev libkeyutils-dev uuid-dev libatomic-ops-dev libboost-program-options-dev libboost-thread-dev libexpat1-dev libleveldb-dev libsnappy-dev libcurl4-gnutls-dev
diff --git a/qa/workunits/mon/crush_ops.sh b/qa/workunits/mon/crush_ops.sh
index 748d00bc363..4f66e552153 100755
--- a/qa/workunits/mon/crush_ops.sh
+++ b/qa/workunits/mon/crush_ops.sh
@@ -27,10 +27,15 @@ ceph osd crush add-bucket foo root
 o1=`ceph osd create`
 o2=`ceph osd create`
 ceph osd crush add $o1 1 host=host1 root=foo
+ceph osd crush add $o1 1 host=host1 root=foo  # idemptoent
 ceph osd crush add $o2 1 host=host2 root=foo
+ceph osd crush add $o2 1 host=host2 root=foo  # idempotent
 ceph osd crush add-bucket bar root
+ceph osd crush add-bucket bar root  # idempotent
 ceph osd crush link host1 root=bar
+ceph osd crush link host1 root=bar  # idempotent
 ceph osd crush link host2 root=bar
+ceph osd crush link host2 root=bar  # idempotent
 
 ceph osd tree | grep -c osd.$o1 | grep -q 2
 ceph osd tree | grep -c host1 | grep -q 2
diff --git a/src/ceph-disk b/src/ceph-disk
index 6c1b3703847..0389b5ce55b 100755
--- a/src/ceph-disk
+++ b/src/ceph-disk
@@ -1085,7 +1085,7 @@ def main_prepare(args):
             args.cluster_uuid = get_fsid(cluster=args.cluster)
             if args.cluster_uuid is None:
                 raise Error(
-                    'must have fsid in config or pass --cluster--uuid=',
+                    'must have fsid in config or pass --cluster-uuid=',
                     )
 
         if args.fs_type is None:
diff --git a/src/ceph.in b/src/ceph.in
index 857197185f4..7d6c62a7303 100755
--- a/src/ceph.in
+++ b/src/ceph.in
@@ -345,8 +345,9 @@ class CephPgid(CephArgtype):
 
 class CephName(CephArgtype):
     """
-    Name, or type.id, where type is osd|mon|client|mds, and id is a base10 int,
-    or just id.
+    Name (type.id) where:
+    type is osd|mon|client|mds
+    id is a base10 int, if type == osd, or a string otherwise
 
     Also accept '*'
     """
@@ -357,16 +358,17 @@ class CephName(CephArgtype):
             self.nameid = None
             return True
         if s.find('.') == -1:
-            i = s
+            raise ArgumentFormat('CephName: no . in {0}'.format(s))
         else:
             t, i = s.split('.')
             if not t in ('osd', 'mon', 'client', 'mds'):
                 raise ArgumentValid('unknown type ' + self.t)
             if t == 'osd':
-                try:
-                    i = int(i)
-                except:
-                    raise ArgumentFormat('osd id ' + i + ' not integer')
+                if i != '*':
+                    try:
+                        i = int(i)
+                    except:
+                        raise ArgumentFormat('osd id ' + i + ' not integer')
             self.nametype = t
         self.val = s
         self.nameid = i
@@ -375,6 +377,36 @@ class CephName(CephArgtype):
     def __str__(self):
         return '<name (type.id)>'
 
+class CephOsdName(CephArgtype):
+    """
+    Like CephName, but specific to osds: allow <id> alone
+
+    osd.<id>, or <id>, or *, where id is a base10 int
+    """
+    def valid(self, s, partial=False):
+        if s == '*':
+            self.val = s
+            self.nametype = None
+            self.nameid = None
+            return True
+        if s.find('.') != -1:
+            t, i = s.split('.')
+        else:
+            t = 'osd'
+            i = s
+        if t != 'osd':
+            raise ArgumentValid('unknown type ' + self.t)
+        try:
+            i = int(i)
+        except:
+            raise ArgumentFormat('osd id ' + i + ' not integer')
+        self.nametype = t
+        self.nameid = i
+        self.val = i
+        return True
+
+    def __str__(self):
+        return '<osdname (id|osd.id)>'
 
 class CephChoices(CephArgtype):
     """
@@ -778,7 +810,7 @@ def osdids():
     ret, outbuf, outs = json_command(prefix='osd ls')
     if ret:
         raise RuntimeError('Can\'t contact mon for osd list')
-    return [i for i in outbuf.split('\n')]
+    return [i for i in outbuf.split('\n') if i != '']
 
 def monids():
     ret, outbuf, outs = json_command(prefix='mon dump',
@@ -1107,8 +1139,10 @@ def send_command(target=('mon', ''), cmd=[], inbuf='', timeout=0):
             if verbose:
                 print >> sys.stderr, '{0} to {1}'.\
                     format(cmd, target[0])
-            ret, outbuf, outs = cluster_handle.mon_command(cmd, inbuf,
-                                                           timeout)
+            if target[1] == '':
+                ret, outbuf, outs = cluster_handle.mon_command(cmd, inbuf, timeout)
+            else:
+                ret, outbuf, outs = cluster_handle.mon_command(cmd, inbuf, timeout, target[1])
 
     except Exception as e:
         raise RuntimeError('"{0}": exception {1}'.format(cmd, e))
@@ -1266,7 +1300,7 @@ def find_cmd_target(childargs):
     if len(valid_dict) == 2:
         name = CephName()
         name.valid(valid_dict['target'])
-        return 'osd', name.nameid
+        return name.nametype, name.nameid
 
     sig = parse_funcsig(['pg', {'name':'pgid','type':'CephPgid'}])
     valid_dict = validate(childargs, sig, partial=True);
@@ -1508,56 +1542,76 @@ def main():
     # of the form 'cmdNNN' followed by an array of argument descriptors)
     # as part of the validated argument JSON object
 
-    ret, outbuf, outs = json_command(target=target,
-                                     prefix='get_command_descriptions')
-    if ret == -errno.EINVAL:
-        # send command to old monitor or OSD
-        if verbose:
-            print '{0} to old {1}'.format(' '.join(childargs), target[0])
-        ret, outbuf, outs = send_command(target, childargs, inbuf)
-        # combine nonerror outbuf and outs; either may have cmd output
-        if ret == 0:
-            outbuf += outs
-            # clear outs so generic code below doesn't print it to stderr
-            outs = ''
-    elif ret:
+    targets = [target]
+
+    if target[1] == '*':
+        targets = [(target[0], o) for o in osdids()]
+
+    final_ret = 0
+    for target in targets:
+        ret, outbuf, outs = json_command(target=target,
+                                         prefix='get_command_descriptions')
+        if ret == -errno.EINVAL:
+            # send command to old monitor or OSD
+            if verbose:
+                print '{0} to old {1}'.format(' '.join(childargs), target[0])
+            ret, outbuf, outs = send_command(target, childargs, inbuf)
+            # combine nonerror outbuf and outs; either may have cmd output
+            if ret == 0:
+                outbuf += outs
+                # clear outs so generic code below doesn't print it to stderr
+                outs = ''
+        elif ret:
+            if ret < 0:
+                ret = -ret
+                print >> sys.stderr, \
+                    'Problem getting command descriptions from {0}, {1}'.\
+                    format(target, errno.errorcode[ret])
+        else:
+            sigdict = parse_json_funcsigs(outbuf)
+
+            if parsed_args.completion:
+                return complete(sigdict, childargs, target)
+
+            ret, outbuf, outs = new_style_command(parsed_args, childargs, target,
+                                                  sigdict, inbuf, verbose)
+
         if ret < 0:
             ret = -ret
-        print >> sys.stderr, \
-            'Problem getting command descriptions from {0}, {1}'.\
-            format(target, errno.errorcode[ret])
-        return ret
-    else:
-        sigdict = parse_json_funcsigs(outbuf)
+            if len(targets) > 1:
+                sys.stderr.write('{0}.{1}: '.format(*target))
+            print >> sys.stderr, 'Error {0}: {1}'.format(errno.errorcode[ret], outs)
+            if len(targets) > 1:
+                final_ret = ret
+            else:
+                return ret
 
-        if parsed_args.completion:
-            return complete(sigdict, childargs, target)
+        # this assumes outs never has useful command output, only status
+        if outs:
+            print >> sys.stderr, outs
 
-        ret, outbuf, outs = new_style_command(parsed_args, childargs, target,
-                                              sigdict, inbuf, verbose)
+        if (parsed_args.output_file):
+            outf.write(outbuf)
+        else:
+            # hack: old code printed status line before many json outputs
+            # (osd dump, etc.) that consumers know to ignore.  Add blank line
+            # to satisfy consumers that skip the first line, but not annoy
+            # consumers that don't.
+            if parsed_args.output_format and \
+               parsed_args.output_format.startswith('json'):
+                sys.stdout.write('\n');
 
-    if ret < 0:
-        ret = -ret
-        print >> sys.stderr, 'Error {0}: {1}'.format(errno.errorcode[ret], outs)
-        return ret
+            # prefix output with target, if there was a wildcard used
+            if len(targets) > 1:
+                sys.stdout.write('{0}.{1}: '.format(*target))
 
-    # this assumes outs never has useful command output, only status
-    if outs:
-        print >> sys.stderr, outs
+            sys.stdout.write(outbuf)
 
     if (parsed_args.output_file):
-        outf.write(outbuf)
         outf.close()
-    else:
-        # hack: old code printed status line before many json outputs
-        # (osd dump, etc.) that consumers know to ignore.  Add blank line
-        # to satisfy consumers that skip the first line, but not annoy
-        # consumers that don't.
-        if parsed_args.output_format and \
-           parsed_args.output_format.startswith('json'):
-            sys.stdout.write('\n');
-
-        sys.stdout.write(outbuf)
+
+    if final_ret:
+        return ret
 
     return 0
 
diff --git a/src/crush/CrushWrapper.cc b/src/crush/CrushWrapper.cc
index f0ae6dd9488..8a9addfb6c2 100644
--- a/src/crush/CrushWrapper.cc
+++ b/src/crush/CrushWrapper.cc
@@ -255,14 +255,14 @@ int CrushWrapper::get_full_location_ordered(int id, vector<pair<string, string>
   string high_type_name = type_map[high_type];
 
   path.push_back(parent_coord);
-  parent_id = get_item_id( (parent_coord.second).c_str() );
+  parent_id = get_item_id(parent_coord.second);
 
 
   while (parent_coord.first != high_type_name) {
     parent_coord = get_immediate_parent(parent_id);
     path.push_back(parent_coord);
     if ( parent_coord.first != high_type_name ){
-      parent_id = get_item_id( (parent_coord.second).c_str() );
+      parent_id = get_item_id(parent_coord.second);
     }
   }
 
@@ -291,7 +291,7 @@ map<int, string> CrushWrapper::get_parent_hierarchy(int id)
       high_type = (*it).first;
   }
 
-  parent_id = get_item_id((parent_coord.second).c_str());
+  parent_id = get_item_id(parent_coord.second);
 
   while (type_counter < high_type) {
     type_counter++;
@@ -300,7 +300,7 @@ map<int, string> CrushWrapper::get_parent_hierarchy(int id)
     if (type_counter < high_type){
       // get the coordinate information for the next parent
       parent_coord = get_immediate_parent(parent_id);
-      parent_id = get_item_id(parent_coord.second.c_str());
+      parent_id = get_item_id(parent_coord.second);
     }
   }
 
diff --git a/src/crush/CrushWrapper.h b/src/crush/CrushWrapper.h
index c123da3d9d5..f5a88c8bdd5 100644
--- a/src/crush/CrushWrapper.h
+++ b/src/crush/CrushWrapper.h
@@ -633,7 +633,7 @@ private:
     pair<string, string> bucket_location = get_immediate_parent(item);
 
     // get the id of the parent bucket
-    int parent_id = get_item_id( (bucket_location.second).c_str() );
+    int parent_id = get_item_id(bucket_location.second);
 
     // get the parent bucket
     crush_bucket *parent_bucket = get_bucket(parent_id);
diff --git a/src/crushtool.cc b/src/crushtool.cc
index 054c4df8698..75c26c098b6 100644
--- a/src/crushtool.cc
+++ b/src/crushtool.cc
@@ -609,11 +609,11 @@ int main(int argc, const char **argv)
   if (!reweight_name.empty()) {
     cout << me << " reweighting item " << reweight_name << " to " << reweight_weight << std::endl;
     int r;
-    if (!crush.name_exists(reweight_name.c_str())) {
+    if (!crush.name_exists(reweight_name)) {
       cerr << " name " << reweight_name << " dne" << std::endl;
       r = -ENOENT;
     } else {
-      int item = crush.get_item_id(reweight_name.c_str());
+      int item = crush.get_item_id(reweight_name);
       r = crush.adjust_item_weightf(g_ceph_context, item, reweight_weight);
     }
     if (r >= 0)
@@ -627,11 +627,11 @@ int main(int argc, const char **argv)
   if (!remove_name.empty()) {
     cout << me << " removing item " << remove_name << std::endl;
     int r;
-    if (!crush.name_exists(remove_name.c_str())) {
+    if (!crush.name_exists(remove_name)) {
       cerr << " name " << remove_name << " dne" << std::endl;
       r = -ENOENT;
     } else {
-      int remove_item = crush.get_item_id(remove_name.c_str());
+      int remove_item = crush.get_item_id(remove_name);
       r = crush.remove_item(g_ceph_context, remove_item, false);
     }
     if (r == 0)
diff --git a/src/include/rados/librados.h b/src/include/rados/librados.h
index 69eeccb323b..eb1a30632b7 100644
--- a/src/include/rados/librados.h
+++ b/src/include/rados/librados.h
@@ -1712,7 +1712,7 @@ int rados_break_lock(rados_ioctx_t io, const char *o, const char *name,
  * @note Takes command string in carefully-formatted JSON; must match
  * defined commands, types, etc.
  *
- * The result buffers are allocated on the heapt; the caller is
+ * The result buffers are allocated on the heap; the caller is
  * expected to release that memory with rados_buffer_free().  The
  * buffer and length pointers can all be NULL, in which case they are
  * not filled in.
@@ -1733,6 +1733,34 @@ int rados_mon_command(rados_t cluster, const char **cmd, size_t cmdlen,
 		      char **outs, size_t *outslen);
 
 /**
+ * Send monitor command to a specific monitor.
+ *
+ * @note Takes command string in carefully-formatted JSON; must match
+ * defined commands, types, etc.
+ *
+ * The result buffers are allocated on the heap; the caller is
+ * expected to release that memory with rados_buffer_free().  The
+ * buffer and length pointers can all be NULL, in which case they are
+ * not filled in.
+ *
+ * @param cluster cluster handle
+ * @param name target monitor's name
+ * @param cmd an array of char *'s representing the command
+ * @param cmdlen count of valid entries in cmd
+ * @param inbuf any bulk input data (crush map, etc.)
+ * @param outbuf double pointer to output buffer
+ * @param outbuflen pointer to output buffer length
+ * @param outs double pointer to status string
+ * @param outslen pointer to status string length
+ * @returns 0 on success, negative error code on failure
+ */
+int rados_mon_command_target(rados_t cluster, const char *name,
+			     const char **cmd, size_t cmdlen,
+			     const char *inbuf, size_t inbuflen,
+			     char **outbuf, size_t *outbuflen,
+			     char **outs, size_t *outslen);
+
+/**
  * free a rados-allocated buffer
  *
  * Release memory allocated by librados calls like rados_mon_command().
diff --git a/src/librados/RadosClient.cc b/src/librados/RadosClient.cc
index a2b54abf058..f5dccaffc09 100644
--- a/src/librados/RadosClient.cc
+++ b/src/librados/RadosClient.cc
@@ -201,11 +201,6 @@ int librados::RadosClient::connect()
   objecter->init_locked();
   monclient.renew_subs();
 
-  while (osdmap.get_epoch() == 0) {
-    ldout(cct, 1) << "waiting for osdmap" << dendl;
-    cond.Wait(lock);
-  }
-
   finisher.start();
 
   state = CONNECTED;
@@ -583,6 +578,44 @@ int librados::RadosClient::mon_command(const vector<string>& cmd,
   return rval;
 }
 
+int librados::RadosClient::mon_command(int rank, const vector<string>& cmd,
+					      bufferlist &inbl,
+					      bufferlist *outbl, string *outs)
+{
+  Mutex mylock("RadosClient::mon_command::mylock");
+  Cond cond;
+  bool done;
+  int rval;
+  lock.Lock();
+  monclient.start_mon_command(rank, cmd, inbl, outbl, outs,
+			       new C_SafeCond(&mylock, &cond, &done, &rval));
+  lock.Unlock();
+  mylock.Lock();
+  while (!done)
+    cond.Wait(mylock);
+  mylock.Unlock();
+  return rval;
+}
+
+int librados::RadosClient::mon_command(string name, const vector<string>& cmd,
+					      bufferlist &inbl,
+					      bufferlist *outbl, string *outs)
+{
+  Mutex mylock("RadosClient::mon_command::mylock");
+  Cond cond;
+  bool done;
+  int rval;
+  lock.Lock();
+  monclient.start_mon_command(name, cmd, inbl, outbl, outs,
+			       new C_SafeCond(&mylock, &cond, &done, &rval));
+  lock.Unlock();
+  mylock.Lock();
+  while (!done)
+    cond.Wait(mylock);
+  mylock.Unlock();
+  return rval;
+}
+
 int librados::RadosClient::osd_command(int osd, vector<string>& cmd,
 					bufferlist& inbl,
 					bufferlist *poutbl, string *prs)
diff --git a/src/librados/RadosClient.h b/src/librados/RadosClient.h
index 9374553b0ad..337beff5750 100644
--- a/src/librados/RadosClient.h
+++ b/src/librados/RadosClient.h
@@ -108,6 +108,12 @@ public:
   void watch_notify(MWatchNotify *m);
   int mon_command(const vector<string>& cmd, bufferlist &inbl,
 	          bufferlist *outbl, string *outs);
+  int mon_command(int rank,
+		  const vector<string>& cmd, bufferlist &inbl,
+	          bufferlist *outbl, string *outs);
+  int mon_command(string name,
+		  const vector<string>& cmd, bufferlist &inbl,
+	          bufferlist *outbl, string *outs);
   int osd_command(int osd, vector<string>& cmd, bufferlist& inbl,
                   bufferlist *poutbl, string *prs);
   int pg_command(pg_t pgid, vector<string>& cmd, bufferlist& inbl,
diff --git a/src/librados/librados.cc b/src/librados/librados.cc
index 1dda70828fd..43c2584c390 100644
--- a/src/librados/librados.cc
+++ b/src/librados/librados.cc
@@ -1857,6 +1857,45 @@ extern "C" int rados_mon_command(rados_t cluster, const char **cmd,
   return ret;
 }
 
+extern "C" int rados_mon_command_target(rados_t cluster, const char *name,
+					const char **cmd,
+					size_t cmdlen,
+					const char *inbuf, size_t inbuflen,
+					char **outbuf, size_t *outbuflen,
+					char **outs, size_t *outslen)
+{
+  librados::RadosClient *client = (librados::RadosClient *)cluster;
+  bufferlist inbl;
+  bufferlist outbl;
+  string outstring;
+  vector<string> cmdvec;
+
+  // is this a numeric id?
+  char *endptr;
+  errno = 0;
+  long rank = strtol(name, &endptr, 10);
+  if ((errno == ERANGE && (rank == LONG_MAX || rank == LONG_MIN)) ||
+      (errno != 0 && rank == 0) ||
+      endptr == name ||    // no digits
+      *endptr != '\0') {   // extra characters
+    rank = -1;
+  }
+
+  for (size_t i = 0; i < cmdlen; i++)
+    cmdvec.push_back(cmd[i]);
+
+  inbl.append(inbuf, inbuflen);
+  int ret;
+  if (rank >= 0)
+    ret = client->mon_command(rank, cmdvec, inbl, &outbl, &outstring);
+  else
+    ret = client->mon_command(name, cmdvec, inbl, &outbl, &outstring);
+
+  do_out_buffer(outbl, outbuf, outbuflen);
+  do_out_buffer(outstring, outs, outslen);
+  return ret;
+}
+
 extern "C" int rados_osd_command(rados_t cluster, int osdid, const char **cmd,
 				 size_t cmdlen,
 				 const char *inbuf, size_t inbuflen,
diff --git a/src/mon/MonCap.cc b/src/mon/MonCap.cc
index bf8eb6fb79d..6f1055091e0 100644
--- a/src/mon/MonCap.cc
+++ b/src/mon/MonCap.cc
@@ -143,6 +143,8 @@ void MonCapGrant::expand_profile(entity_name_t name) const
     profile_grants.push_back(MonCapGrant("config-key delete", "key", StringConstraint("", prefix)));
   }
   if (profile == "bootstrap-osd") {
+    profile_grants.push_back(MonCapGrant("mon", MON_CAP_R));  // read monmap
+    profile_grants.push_back(MonCapGrant("osd", MON_CAP_R));  // read osdmap
     profile_grants.push_back(MonCapGrant("mon getmap"));
     profile_grants.push_back(MonCapGrant("osd create"));
     profile_grants.push_back(MonCapGrant("osd crush set"));  // FIXME: constraint this further?
@@ -152,6 +154,8 @@ void MonCapGrant::expand_profile(entity_name_t name) const
     profile_grants.back().command_args["caps_osd"] = StringConstraint("allow *", "");
   }
   if (profile == "bootstrap-mds") {
+    profile_grants.push_back(MonCapGrant("mon", MON_CAP_R));  // read monmap
+    profile_grants.push_back(MonCapGrant("osd", MON_CAP_R));  // read osdmap
     profile_grants.push_back(MonCapGrant("mon getmap"));
     profile_grants.push_back(MonCapGrant("auth get-or-create"));  // FIXME: this can expose other mds keys
     profile_grants.back().command_args["name"] = StringConstraint("", "mds.");
diff --git a/src/mon/MonClient.cc b/src/mon/MonClient.cc
index 181d0688311..c0d2b5183c5 100644
--- a/src/mon/MonClient.cc
+++ b/src/mon/MonClient.cc
@@ -479,11 +479,18 @@ string MonClient::_pick_random_mon()
   }
 }
 
-void MonClient::_pick_new_mon()
+void MonClient::_reopen_session(int rank, string name)
 {
   assert(monc_lock.is_locked());
+  ldout(cct, 10) << "_reopen_session rank " << rank << " name " << name << dendl;
 
-  cur_mon = _pick_random_mon();
+  if (rank < 0 && name.length() == 0) {
+    cur_mon = _pick_random_mon();
+  } else if (name.length()) {
+    cur_mon = name;
+  } else {
+    cur_mon = monmap.get_name(rank);
+  }
 
   if (cur_con) {
     messenger->mark_down(cur_con);
@@ -491,18 +498,9 @@ void MonClient::_pick_new_mon()
   }
   cur_con = messenger->get_connection(monmap.get_inst(cur_mon));
 	
-  ldout(cct, 10) << "_pick_new_mon picked mon." << cur_mon << " con " << cur_con
+  ldout(cct, 10) << "picked mon." << cur_mon << " con " << cur_con
 		 << " addr " << cur_con->get_peer_addr()
 		 << dendl;
-}
-
-
-void MonClient::_reopen_session()
-{
-  assert(monc_lock.is_locked());
-  ldout(cct, 10) << "_reopen_session" << dendl;
-
-  _pick_new_mon();
 
   // throw out old queued messages
   while (!waiting_for_session.empty()) {
@@ -724,6 +722,37 @@ int MonClient::wait_auth_rotating(double timeout)
 void MonClient::_send_command(MonCommand *r)
 {
   version_t last_seen_version = 0;
+
+  if (r->target_rank >= 0 &&
+      r->target_rank != monmap.get_rank(cur_mon)) {
+    ldout(cct, 10) << "_send_command " << r->tid << " " << r->cmd
+		   << " wants rank " << r->target_rank
+		   << ", reopening session"
+		   << dendl;
+    if (r->target_rank >= (int)monmap.size()) {
+      ldout(cct, 10) << " target " << r->target_rank << " >= max mon " << monmap.size() << dendl;
+      _finish_command(r, -ENOENT, "mon rank dne");
+      return;
+    }
+    _reopen_session(r->target_rank, string());
+    return;
+  }
+
+  if (r->target_name.length() &&
+      r->target_name != cur_mon) {
+    ldout(cct, 10) << "_send_command " << r->tid << " " << r->cmd
+		   << " wants mon " << r->target_name
+		   << ", reopening session"
+		   << dendl;
+    if (!monmap.contains(r->target_name)) {
+      ldout(cct, 10) << " target " << r->target_name << " not present in monmap" << dendl;
+      _finish_command(r, -ENOENT, "mon dne");
+      return;
+    }
+    _reopen_session(-1, r->target_name);
+    return;
+  }
+
   ldout(cct, 10) << "_send_command " << r->tid << " " << r->cmd << dendl;
   MMonCommand *m = new MMonCommand(monmap.fsid, last_seen_version);
   m->set_tid(r->tid);
@@ -798,6 +827,43 @@ int MonClient::start_mon_command(const vector<string>& cmd, bufferlist& inbl,
   return 0;
 }
 
+int MonClient::start_mon_command(string name,
+				 const vector<string>& cmd, bufferlist& inbl,
+				 bufferlist *outbl, string *outs,
+				 Context *onfinish)
+{
+  Mutex::Locker l(monc_lock);
+  MonCommand *r = new MonCommand(++last_mon_command_tid);
+  r->target_name = name;
+  r->cmd = cmd;
+  r->inbl = inbl;
+  r->poutbl = outbl;
+  r->prs = outs;
+  r->onfinish = onfinish;
+  mon_commands[r->tid] = r;
+  _send_command(r);
+  // can't fail
+  return 0;
+}
+
+int MonClient::start_mon_command(int rank,
+				 const vector<string>& cmd, bufferlist& inbl,
+				 bufferlist *outbl, string *outs,
+				 Context *onfinish)
+{
+  Mutex::Locker l(monc_lock);
+  MonCommand *r = new MonCommand(++last_mon_command_tid);
+  r->target_rank = rank;
+  r->cmd = cmd;
+  r->inbl = inbl;
+  r->poutbl = outbl;
+  r->prs = outs;
+  r->onfinish = onfinish;
+  mon_commands[r->tid] = r;
+  _send_command(r);
+  return 0;
+}
+
 // ---------
 
 void MonClient::get_version(string map, version_t *newest, version_t *oldest, Context *onfinish)
diff --git a/src/mon/MonClient.h b/src/mon/MonClient.h
index 04e6b0e44b5..7d4a0548f9d 100644
--- a/src/mon/MonClient.h
+++ b/src/mon/MonClient.h
@@ -129,8 +129,10 @@ private:
 
   string _pick_random_mon();
   void _finish_hunting();
-  void _reopen_session();
-  void _pick_new_mon();
+  void _reopen_session(int rank, string name);
+  void _reopen_session() {
+    _reopen_session(-1, string());
+  }
   void _send_mon_message(Message *m, bool force=false);
 
 public:
@@ -270,6 +272,8 @@ public:
 private:
   uint64_t last_mon_command_tid;
   struct MonCommand {
+    string target_name;
+    int target_rank;
     uint64_t tid;
     vector<string> cmd;
     bufferlist inbl;
@@ -279,7 +283,8 @@ private:
     Context *onfinish;
 
     MonCommand(uint64_t t)
-      : tid(t),
+      : target_rank(-1),
+	tid(t),
 	poutbl(NULL), prs(NULL), prval(NULL), onfinish(NULL)
     {}
   };
@@ -294,6 +299,14 @@ public:
   int start_mon_command(const vector<string>& cmd, bufferlist& inbl,
 			bufferlist *outbl, string *outs,
 			Context *onfinish);
+  int start_mon_command(int mon_rank,
+			const vector<string>& cmd, bufferlist& inbl,
+			bufferlist *outbl, string *outs,
+			Context *onfinish);
+  int start_mon_command(const string mon_name,  ///< mon name, with mon. prefix
+			const vector<string>& cmd, bufferlist& inbl,
+			bufferlist *outbl, string *outs,
+			Context *onfinish);
 
   // version requests
 public:
diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
index 4e64a0a70d4..2c805362255 100644
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -55,6 +55,7 @@
  * CephObjectname: Another plainold string
  * CephPgid: n.xxx where n is an int > 0, xxx is a hex number > 0
  * CephName: daemon name, '*' or '<type>.<id>' (id must be int for type osd)
+ * CephOsdName: osd name, '*' or '<id> or 'osd.<id>' (id must be int)
  * CephChoices: strings="foo|bar" means this param can be either
  * CephFilepath: openable file
  * CephFragment: cephfs 'fragID': val/bits, val in hex 0xnnn, bits in dec
@@ -259,10 +260,6 @@ COMMAND("mon stat", "summarize monitor status")
 COMMAND("mon getmap " \
 	"name=epoch,type=CephInt,range=0,req=false", \
 	"get monmap")
-COMMAND("mon tell " \
-	"name=who,type=CephString " \
-	"name=args,type=CephString,n=N", \
-	"send command to specific monitor(s)")
 COMMAND("mon add " \
 	"name=name,type=CephString " \
 	"name=addr,type=CephIPAddr", \
@@ -289,10 +286,6 @@ COMMAND("osd getcrushmap " \
 	"name=epoch,type=CephInt,range=0,req=false", \
 	"get CRUSH map")
 COMMAND("osd getmaxosd", "show largest OSD id")
-COMMAND("osd tell " \
-	"name=who,type=CephString " \
-	"name=args,type=CephString,n=N", \
-	"send command to particular osd")
 COMMAND("osd find " \
 	"name=id,type=CephInt,range=0", \
 	"find osd <id> in the CRUSH map and show its location")
@@ -324,24 +317,22 @@ COMMAND("osd crush add-bucket " \
 	"name=type,type=CephString", \
 	"add no-parent (probably root) crush bucket <name> of type <type>")
 COMMAND("osd crush set " \
-	"name=id,type=CephInt,range=0 " \
-	"name=name,type=CephName,req=false " \
+	"name=id,type=CephOsdName " \
 	"name=weight,type=CephFloat,range=0.0 " \
 	"name=args,type=CephString,n=N", \
-	"set crushmap entry for <id> to <weight> with location <args>")
+	"set crushmap entry for <name> to <weight> with location <args>")
 COMMAND("osd crush add " \
-	"name=id,type=CephInt,range=0 " \
-	"name=name,type=CephName,req=false " \
+	"name=id,type=CephOsdName " \
 	"name=weight,type=CephFloat,range=0.0 " \
 	"name=args,type=CephString,n=N", \
-	"add crushmap entry for <id> with <weight> and location <args>")
+	"add crushmap entry for <name> with <weight> and location <args>")
 COMMAND("osd crush create-or-move " \
-	"name=id,type=CephInt,range=0 " \
+	"name=id,type=CephOsdName " \
 	"name=weight,type=CephFloat,range=0.0 " \
 	"name=args,type=CephString,n=N", \
-	"create entry or move existing entry for <id> <weight> at/to location <args>")
+	"create entry or move existing entry for <name> <weight> at/to location <args>")
 COMMAND("osd crush move " \
-	"name=name,type=CephString " \
+	"name=id,type=CephOsdName " \
 	"name=args,type=CephString,n=N", \
 	"move existing entry for <name> to location <args>")
 COMMAND("osd crush link " \
diff --git a/src/mon/MonmapMonitor.cc b/src/mon/MonmapMonitor.cc
index 90839b7706f..cbcefae104f 100644
--- a/src/mon/MonmapMonitor.cc
+++ b/src/mon/MonmapMonitor.cc
@@ -264,60 +264,12 @@ bool MonmapMonitor::preprocess_command(MMonCommand *m)
       if (p != mon->monmap)
 	delete p;
     }
-  } else if (prefix == "mon tell") {
-    dout(20) << "got tell: " << m->cmd << dendl;
-    string whostr;
-    cmd_getval(g_ceph_context, cmdmap, "who", whostr);
-    vector<string> argvec;
-    cmd_getval(g_ceph_context, cmdmap, "args", argvec);
-
-    if (whostr == "*") { // send to all mons and do myself
-      for (unsigned i = 0; i < mon->monmap->size(); ++i) {
-	MMonCommand *newm = new MMonCommand(m->fsid, m->version);
-	newm->cmd.insert(newm->cmd.begin(), argvec.begin(), argvec.end());
-	mon->messenger->send_message(newm, mon->monmap->get_inst(i));
-      }
-      ss << "bcast to all mons";
-      r = 0;
-    } else {
-      // find target.  Ignore error from parsing long as we probably
-      // have a string instead
-      long who = parse_pos_long(whostr.c_str(), NULL);
-      EntityName name;
-      if (who < 0) {
-
-	// not numeric; try as name or id, and see if in monmap
-	if (!name.from_str(whostr))
-	  name.set("mon", whostr);
-
-	if (mon->monmap->contains(name.get_id())) {
-	  who = mon->monmap->get_rank(name.get_id());
-	} else {
-	  ss << "bad mon name \"" << whostr << "\"";
-	  r = -ENOENT;
-	  goto out;
-	}
-      } else if (who >= (long)mon->monmap->size()) {
-	ss << "mon." << whostr << " does not exist";
-	r = -ENOENT;
-	goto out;
-      }
-
-      // send to target, or handle if it's me
-      stringstream ss;
-      MMonCommand *newm = new MMonCommand(m->fsid, m->version);
-      newm->cmd.insert(newm->cmd.begin(), argvec.begin(), argvec.end());
-      mon->messenger->send_message(newm, mon->monmap->get_inst(who));
-      ss << "fw to mon." << whostr;
-      r = 0;
-    }
   }
   else if (prefix == "mon add")
     return false;
   else if (prefix == "mon remove")
     return false;
 
- out:
   if (r != -1) {
     string rs;
     getline(ss, rs);
diff --git a/src/mon/OSDMonitor.cc b/src/mon/OSDMonitor.cc
index d6e2e1cb560..d785cf626c3 100644
--- a/src/mon/OSDMonitor.cc
+++ b/src/mon/OSDMonitor.cc
@@ -1697,7 +1697,7 @@ void OSDMonitor::tick()
 
 	// is this an entire large subtree down?
 	if (g_conf->mon_osd_down_out_subtree_limit.length()) {
-	  int type = osdmap.crush->get_type_id(g_conf->mon_osd_down_out_subtree_limit.c_str());
+	  int type = osdmap.crush->get_type_id(g_conf->mon_osd_down_out_subtree_limit);
 	  if (type > 0) {
 	    if (osdmap.containing_subtree_is_down(g_ceph_context, o, type, &down_cache)) {
 	      dout(10) << "tick entire containing " << g_conf->mon_osd_down_out_subtree_limit
@@ -2482,7 +2482,13 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
   cmd_getval(g_ceph_context, cmdmap, "prefix", prefix);
 
   int64_t id;
+  string name;
   bool osdid_present = cmd_getval(g_ceph_context, cmdmap, "id", id);
+  if (osdid_present) {
+    ostringstream oss;
+    oss << "osd." << id;
+    name = oss.str();
+  }
 
   if (prefix == "osd setcrushmap" ||
       (prefix == "osd crush set" && !osdid_present)) {
@@ -2544,23 +2550,19 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
     ss << "added bucket " << name << " type " << typestr
        << " to crush map";
     goto update;
-  } else if (osdid_present && 
+  } else if (osdid_present &&
 	     (prefix == "osd crush set" || prefix == "osd crush add")) {
     do {
-      // osd crush set <osd-id> [<osd.* name>] <weight> <loc1> [<loc2> ...]
-      // osd crush add <osd-id> [<osd.* name>] <weight> <loc1> [<loc2> ...]
+      // <OsdName> is 'osd.<id>' or '<id>', passed as int64_t id
+      // osd crush set <OsdName> <weight> <loc1> [<loc2> ...]
+      // osd crush add <OsdName> <weight> <loc1> [<loc2> ...]
+
       if (!osdmap.exists(id)) {
 	err = -ENOENT;
-	ss << "osd." << id << " does not exist.  create it before updating the crush map";
+	ss << name << " does not exist.  create it before updating the crush map";
 	goto reply;
       }
 
-      string name;
-      if (!cmd_getval(g_ceph_context, cmdmap, "name", name)) {
-	// new usage; infer name
-	name = "osd." + stringify(id);
-      }
-
       double weight;
       cmd_getval(g_ceph_context, cmdmap, "weight", weight);
 
@@ -2577,7 +2579,8 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
       _get_pending_crush(newcrush);
 
       string action;
-      if (prefix == "osd crush set") {
+      if (prefix == "osd crush set" ||
+	  newcrush.check_item_loc(g_ceph_context, id, loc, (int *)NULL)) {
 	action = "set";
 	err = newcrush.update_item(g_ceph_context, id, weight, name, loc);
       } else {
@@ -2604,16 +2607,13 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
 
   } else if (prefix == "osd crush create-or-move") {
     do {
-      // osd crush create-or-move <id> <initial_weight> <loc1> [<loc2> ...]
-      int64_t id;
-      cmd_getval(g_ceph_context, cmdmap, "id", id);
+      // osd crush create-or-move <OsdName> <initial_weight> <loc1> [<loc2> ...]
       if (!osdmap.exists(id)) {
 	err = -ENOENT;
-	ss << "osd." << id << " does not exist.  create it before updating the crush map";
+	ss << name << " does not exist.  create it before updating the crush map";
 	goto reply;
       }
 
-      string name = "osd." + stringify(id);
       double weight;
       cmd_getval(g_ceph_context, cmdmap, "weight", weight);
 
@@ -2623,7 +2623,7 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
       map<string,string> loc;
       parse_loc_map(argvec, &loc);
 
-      dout(0) << "create-or-move crush item id " << id << " name '" << name << "' initial_weight " << weight
+      dout(0) << "create-or-move crush item name '" << name << "' initial_weight " << weight
 	      << " at location " << loc << dendl;
 
       CrushWrapper newcrush;
@@ -2631,14 +2631,14 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
 
       err = newcrush.create_or_move_item(g_ceph_context, id, weight, name, loc);
       if (err == 0) {
-	ss << "create-or-move updated item id " << id << " name '" << name << "' weight " << weight
+	ss << "create-or-move updated item name '" << name << "' weight " << weight
 	   << " at location " << loc << " to crush map";
 	break;
       }
       if (err > 0) {
 	pending_inc.crush.clear();
 	newcrush.encode(pending_inc.crush);
-	ss << "create-or-move updating item id " << id << " name '" << name << "' weight " << weight
+	ss << "create-or-move updating item name '" << name << "' weight " << weight
 	   << " at location " << loc << " to crush map";
 	getline(ss, rs);
 	wait_for_finished_proposal(new Monitor::C_Command(mon, m, 0, rs, get_version()));
@@ -2649,8 +2649,6 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
   } else if (prefix == "osd crush move") {
     do {
       // osd crush move <name> <loc1> [<loc2> ...]
-      string name;
-      cmd_getval(g_ceph_context, cmdmap, "name", name);
 
       string args;
       vector<string> argvec;
@@ -2662,12 +2660,12 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
       CrushWrapper newcrush;
       _get_pending_crush(newcrush);
 
-      if (!newcrush.name_exists(name.c_str())) {
+      if (!newcrush.name_exists(name)) {
 	err = -ENOENT;
 	ss << "item " << name << " does not exist";
 	break;
       }
-      int id = newcrush.get_item_id(name.c_str());
+      int id = newcrush.get_item_id(name);
 
       if (!newcrush.check_item_loc(g_ceph_context, id, loc, (int *)NULL)) {
 	err = newcrush.move_bucket(g_ceph_context, id, loc);
@@ -2699,12 +2697,12 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
       CrushWrapper newcrush;
       _get_pending_crush(newcrush);
 
-      if (!newcrush.name_exists(name.c_str())) {
+      if (!newcrush.name_exists(name)) {
 	err = -ENOENT;
 	ss << "item " << name << " does not exist";
 	break;
       }
-      int id = newcrush.get_item_id(name.c_str());
+      int id = newcrush.get_item_id(name);
 
       if (!newcrush.check_item_loc(g_ceph_context, id, loc, (int *)NULL)) {
 	err = newcrush.link_bucket(g_ceph_context, id, loc);
@@ -2732,12 +2730,12 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
       string name;
       cmd_getval(g_ceph_context, cmdmap, "name", name);
 
-      if (!newcrush.name_exists(name.c_str())) {
+      if (!newcrush.name_exists(name)) {
 	err = 0;
 	ss << "device '" << name << "' does not appear in the crush map";
 	break;
       }
-      int id = newcrush.get_item_id(name.c_str());
+      int id = newcrush.get_item_id(name);
       bool unlink_only = prefix == "osd crush unlink";
       string ancestor_str;
       if (cmd_getval(g_ceph_context, cmdmap, "ancestor", ancestor_str)) {
@@ -2776,13 +2774,13 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
 
       string name;
       cmd_getval(g_ceph_context, cmdmap, "name", name);
-      if (!newcrush.name_exists(name.c_str())) {
+      if (!newcrush.name_exists(name)) {
 	err = -ENOENT;
 	ss << "device '" << name << "' does not appear in the crush map";
 	break;
       }
 
-      int id = newcrush.get_item_id(name.c_str());
+      int id = newcrush.get_item_id(name);
       if (id < 0) {
 	ss << "device '" << name << "' is not a leaf in the crush map";
 	break;
@@ -3082,7 +3080,7 @@ bool OSDMonitor::prepare_command(MMonCommand *m)
 	// osd already exists
 	err = 0;
 	ss << i;
-	getline(ss, rs);
+	rdata.append(ss);
 	goto reply;
       }
       i = pending_inc.identify_osd(uuid);
diff --git a/src/pybind/rados.py b/src/pybind/rados.py
index 4a5ed961603..c4000465272 100644
--- a/src/pybind/rados.py
+++ b/src/pybind/rados.py
@@ -518,9 +518,9 @@ Rados object in state %s." % (self.state))
             raise make_ex(ret, "error opening ioctx '%s'" % ioctx_name)
         return Ioctx(ioctx_name, self.librados, ioctx)
 
-    def mon_command(self, cmd, inbuf, timeout=0):
+    def mon_command(self, cmd, inbuf, timeout=0, target=None):
         """
-        mon_command(cmd, inbuf, outbuf, outbuflen, outs, outslen)
+        mon_command[_target](cmd, inbuf, outbuf, outbuflen, outs, outslen)
         returns (int ret, string outbuf, string outs)
         """
         import sys
@@ -531,11 +531,18 @@ Rados object in state %s." % (self.state))
         outslen = c_long()
         cmdarr = (c_char_p * len(cmd))(*cmd)
 
-        ret = run_in_thread(self.librados.rados_mon_command,
-                            (self.cluster, cmdarr, len(cmd),
-                            c_char_p(inbuf), len(inbuf),
-                            outbufp, byref(outbuflen), outsp, byref(outslen)),
-                            timeout)
+        if target:
+            ret = run_in_thread(self.librados.rados_mon_command_target,
+                                (self.cluster, target, cmdarr, len(cmd),
+                                 c_char_p(inbuf), len(inbuf),
+                                 outbufp, byref(outbuflen), outsp, byref(outslen)),
+                                timeout)
+        else:
+            ret = run_in_thread(self.librados.rados_mon_command,
+                                (self.cluster, cmdarr, len(cmd),
+                                 c_char_p(inbuf), len(inbuf),
+                                 outbufp, byref(outbuflen), outsp, byref(outslen)),
+                                timeout)
 
         if ret == 0:
             # copy returned memory (ctypes makes a copy, not a reference)
diff --git a/src/vstart.sh b/src/vstart.sh
index 27a750cbbeb..f179b0b1bf1 100755
--- a/src/vstart.sh
+++ b/src/vstart.sh
@@ -423,7 +423,7 @@ EOF
 	    uuid=`uuidgen`
 	    echo "add osd$osd $uuid"
 	    $SUDO $CEPH_ADM osd create $uuid
-	    $SUDO $CEPH_ADM osd crush set $osd osd.$osd 1.0 host=localhost rack=localrack root=default
+	    $SUDO $CEPH_ADM osd crush set osd.$osd 1.0 host=localhost rack=localrack root=default
 	    $SUDO $CEPH_BIN/ceph-osd -i $osd $ARGS --mkfs --mkkey --osd-uuid $uuid
 
 	    key_fn=dev/osd$osd/keyring