Merge branch 'next'

8 files changed, 157 insertions, 9 deletions

diff --git a/ceph.spec.in b/ceph.spec.in
index 3c9ac20902a..9ba972e5671 100644
--- a/ceph.spec.in
+++ b/ceph.spec.in
@@ -325,7 +325,7 @@ rm -rf $RPM_BUILD_ROOT
 
 %post
 /sbin/ldconfig
-#/sbin/chkconfig --add ceph
+/sbin/chkconfig --add ceph
 
 %preun
 %if %{defined suse_version}
@@ -333,7 +333,7 @@ rm -rf $RPM_BUILD_ROOT
 %endif
 if [ $1 = 0 ] ; then
     /sbin/service ceph stop >/dev/null 2>&1
-#    /sbin/chkconfig --del ceph
+    /sbin/chkconfig --del ceph
 fi
 
 %postun
@@ -398,6 +398,7 @@ fi
 %{_libdir}/rados-classes/libcls_refcount.so*
 %{_libdir}/ceph
 /lib/udev/rules.d/50-rbd.rules
+/lib/udev/rules.d/60-ceph-partuuid-workaround.rules
 /lib/udev/rules.d/95-ceph-osd.rules
 %dir %{_sysconfdir}/ceph/
 %config %{_sysconfdir}/bash_completion.d/ceph
diff --git a/src/ceph-disk b/src/ceph-disk
index 127d809902d..bd7e6206ae8 100755
--- a/src/ceph-disk
+++ b/src/ceph-disk
@@ -14,6 +14,48 @@ import tempfile
 import uuid
 import lockfile
 
+"""
+Prepare:
+ - create GPT partition
+ - mark the partition with the ceph type uuid
+ - create a file system
+ - mark the fs as ready for ceph consumption
+ - entire data disk is used (one big partition)
+ - a new partition is added to the journal disk (so it can be easily shared)
+
+ - triggered by administrator or ceph-deploy, e.g.  'ceph-disk <data disk> [journal disk]
+
+Activate:
+ - mount the volume in a temp loation
+ - allocate an osd id (if needed)
+ - remount in the correct location /var/lib/ceph/osd/$cluster-$id
+ - start ceph-osd
+
+ - triggered by udev when it sees the OSD gpt partition type
+ - triggered by admin 'ceph-disk activate <path>'
+ - triggered on ceph service startup with 'ceph-disk activate-all'
+
+We rely on /dev/disk/by-partuuid to find partitions by their UUID;
+this is what the journal symlink inside the osd data volume normally
+points to.
+
+activate-all relies on /dev/disk/by-parttype-uuid/$typeuuid.$uuid to
+find all partitions.  We install special udev rules to create these
+links.
+
+udev triggers 'ceph-disk activate <dev>' or 'ceph-disk
+activate-journal <dev>' based on the partition type.
+
+On old distros (e.g., RHEL6), the blkid installed does not recognized
+GPT partition metadata and the /dev/disk/by-partuuid etc. links aren't
+present.  We have a horrible hack in the form of ceph-disk-udev that
+parses gparted output to create the symlinks above and triggers the
+'ceph-disk activate' etc commands that udev normally would do if it
+knew the GPT partition type.
+
+"""
+
+
 CEPH_OSD_ONDISK_MAGIC = 'ceph osd volume v026'
 
 JOURNAL_UUID =         '45b0969e-9b03-4f30-b4c6-b4b80ceff106'
@@ -57,6 +99,11 @@ INIT_SYSTEMS = [
     'auto',
     ]
 
+# Nuke the TERM variable to avoid confusing any subprocesses we call.
+# For example, libreadline will print weird control sequences for some
+# TERM values.
+if 'TERM' in os.environ:
+    del os.environ['TERM']
 
 LOG_NAME = __name__
 if LOG_NAME == '__main__':
@@ -1689,6 +1736,46 @@ def main_activate_journal(args):
 
 ###########################
 
+def main_activate_all(args):
+    dir = '/dev/disk/by-parttypeuuid'
+    LOG.debug('Scanning %s', dir)
+    if not os.path.exists(dir):
+        return
+    err = False
+    for name in os.listdir(dir):
+        if name.find('.') < 0:
+            continue
+        (tag, uuid) = name.split('.')
+        if tag == OSD_UUID:
+            path = os.path.join(dir, name)
+            LOG.info('Activating %s', path)
+            activate_lock.acquire()
+            try:
+                (cluster, osd_id) = mount_activate(
+                    dev=path,
+                    activate_key_template=args.activate_key_template,
+                    init=args.mark_init,
+                    )
+                start_daemon(
+                    cluster=cluster,
+                    osd_id=osd_id,
+                    )
+
+            except Exception as e:
+                print >> sys.stderr, '{prog}: {msg}'.format(
+                    prog=args.prog,
+                    msg=e,
+                    )
+                err = True
+
+            finally:
+                activate_lock.release()
+    if err:
+        raise Error('One or more partitions failed to activate')
+
+
+###########################
+
 def is_swap(dev):
     dev = os.path.realpath(dev)
     with file('/proc/swaps', 'rb') as proc_swaps:
@@ -2090,6 +2177,25 @@ def parse_args():
         func=main_activate_journal,
         )
 
+    activate_all_parser = subparsers.add_parser('activate-all', help='Activate all tagged OSD partitions')
+    activate_all_parser.add_argument(
+        '--activate-key',
+        metavar='PATH',
+        help='bootstrap-osd keyring path template (%(default)s)',
+        dest='activate_key_template',
+        )
+    activate_all_parser.add_argument(
+        '--mark-init',
+        metavar='INITSYSTEM',
+        help='init system to manage this dir',
+        default='auto',
+        choices=INIT_SYSTEMS,
+        )
+    activate_all_parser.set_defaults(
+        activate_key_template='/var/lib/ceph/bootstrap-osd/{cluster}.keyring',
+        func=main_activate_all,
+        )
+
     list_parser = subparsers.add_parser('list', help='List disks, partitions, and Ceph OSDs')
     list_parser.set_defaults(
         func=main_list,
diff --git a/src/ceph-disk-udev b/src/ceph-disk-udev
index 885f638a1e0..bdf524e6aea 100755
--- a/src/ceph-disk-udev
+++ b/src/ceph-disk-udev
@@ -1,8 +1,9 @@
 #! /bin/sh
 
-# Wrapper for the ceph udev rules.  Since older versions of udev do not support gpt label fields, this shell
-# script is invoked from the udev rule to read the needed gpt label fields and call the appropriate ceph
-# OSD functions.
+# Wrapper for the ceph udev rules.  Since older versions of udev+blkid
+# do not support gpt label fields, this shell script is invoked from
+# the udev rule to read the needed gpt label fields and call the
+# appropriate ceph OSD functions.
 
 PARTNO=$1
 NAME=$2
@@ -10,6 +11,19 @@ PARENT_NAME=$3
 
 # Get GPT partition type guid
 ID_PART_ENTRY_TYPE=$(/usr/sbin/sgdisk --info=${PARTNO} /dev/${PARENT_NAME} | grep "Partition GUID code" | awk '{print $4}' | tr '[:upper:]' '[:lower:]')
+
+if [ -z "$ID_PART_ENTRY_TYPE" ]; then
+    exit
+fi
+
+ID_PART_ENTRY_UUID=$(/usr/sbin/sgdisk --info=${PARTNO} /dev/${PARENT_NAME} | grep "Partition unique GUID" | awk '{print $4}' | tr '[:upper:]' '[:lower:]')
+
+# set up the symlinks
+mkdir -p /dev/disk/by-partuuid
+ln -sf ../../${NAME} /dev/disk/by-partuuid/$ID_PART_ENTRY_UUID
+mkdir -p /dev/disk/by-parttypeuuid
+ln -sf ../../${NAME} /dev/disk/by-parttypeuuid/${ID_PART_ENTRY_TYPE}.${ID_PART_ENTRY_UUID}
+
 case $ID_PART_ENTRY_TYPE in
 
 45b0969e-9b03-4f30-b4c6-b4b80ceff106)
@@ -21,7 +35,6 @@ case $ID_PART_ENTRY_TYPE in
 45b0969e-9b03-4f30-b4c6-5ec00ceff106)
     # DMCRYPT_JOURNAL_UUID
     # Map journal if using dm-crypt
-    ID_PART_ENTRY_UUID=$(/usr/sbin/sgdisk --info=${PARTNO} /dev/${PARENT_NAME} | grep "Partition unique GUID" | awk '{print $4}' | tr '[:upper:]' '[:lower:]')
     /sbin/cryptsetup --key-file /etc/ceph/dmcrypt-keys/${ID_PART_ENTRY_UUID} --key-size 256 create ${ID_PART_ENTRY_UUID} /dev/${NAME}
     ;;
 
@@ -35,7 +48,6 @@ case $ID_PART_ENTRY_TYPE in
     # DMCRYPT_OSD_UUID
     # Map data device and activate ceph-tagged partitions
     # for dm-crypted data devices
-    ID_PART_ENTRY_UUID=$(/usr/sbin/sgdisk --info=${PARTNO} /dev/${PARENT_NAME} | grep "Partition unique GUID" | awk '{print $4}' | tr '[:upper:]' '[:lower:]')
     /sbin/cryptsetup --key-file /etc/ceph/dmcrypt-keys/${ID_PART_ENTRY_UUID} --key-size 256 create ${ID_PART_ENTRY_UUID} /dev/${NAME}
     bash -c 'while [ ! -e /dev/mapper/${ID_PART_ENTRY_UUID} ];do sleep 1; done'
     /usr/sbin/ceph-disk-activate /dev/mapper/${ID_PART_ENTRY_UUID}
diff --git a/src/client/Client.cc b/src/client/Client.cc
index dd27261be23..eeb46eef2fc 100644
--- a/src/client/Client.cc
+++ b/src/client/Client.cc
@@ -7892,9 +7892,22 @@ void Client::ms_handle_remote_reset(Connection *con)
 	}
       }
       if (mds >= 0) {
-	if (s->state == MetaSession::STATE_CLOSING) {
+	switch (s->state) {
+	case MetaSession::STATE_CLOSING:
 	  ldout(cct, 1) << "reset from mds we were closing; we'll call that closed" << dendl;
 	  _closed_mds_session(s);
+	  break;
+
+	case MetaSession::STATE_OPENING:
+	  {
+	    ldout(cct, 1) << "reset from mds we were opening; retrying" << dendl;
+	    list<Cond*> waiters;
+	    waiters.swap(s->waiting_for_open);
+	    _closed_mds_session(s);
+	    MetaSession *news = _get_or_open_mds_session(mds);
+	    news->waiting_for_open.swap(waiters);
+	  }
+	  break;
 	}
       }
     }
diff --git a/src/init-ceph.in b/src/init-ceph.in
index aea2dc37680..0ed906be987 100644
--- a/src/init-ceph.in
+++ b/src/init-ceph.in
@@ -430,4 +430,11 @@ for name in $what; do
     esac
 done
 
+# activate latent osds?
+if [ "$command" = "start" ]; then
+    if [ "$*" = "" ] || echo $* | grep -q ^osd\$ ; then
+	ceph-disk activate-all
+    fi
+fi
+
 exit $EXIT_STATUS
diff --git a/src/mon/MonCommands.h b/src/mon/MonCommands.h
index 2c805362255..5e6f9d49198 100644
--- a/src/mon/MonCommands.h
+++ b/src/mon/MonCommands.h
@@ -272,7 +272,9 @@ COMMAND("mon remove " \
  * OSD commands
  */
 COMMAND("osd stat", "print summary of OSD map")
-COMMAND("osd dump", "print summary of OSD map")
+COMMAND("osd dump " \
+	"name=epoch,type=CephInt,range=0,req=false",
+	"print summary of OSD map")
 COMMAND("osd tree " \
 	"name=epoch,type=CephInt,range=0,req=false", \
 	"print OSD tree")
diff --git a/src/upstart/ceph-osd-all-starter.conf b/src/upstart/ceph-osd-all-starter.conf
index 0311716cdb4..eeb64bca567 100644
--- a/src/upstart/ceph-osd-all-starter.conf
+++ b/src/upstart/ceph-osd-all-starter.conf
@@ -6,6 +6,10 @@ task
 
 script
   set -e
+
+  # first activate any partitions
+  ceph-disk activate-all
+
   # TODO what's the valid charset for cluster names and osd ids?
   find /var/lib/ceph/osd/ -mindepth 1 -maxdepth 1 -regextype posix-egrep -regex '.*/[A-Za-z0-9]+-[A-Za-z0-9._-]+' -printf '%P\n' \
   | while read f; do
diff --git a/udev/60-ceph-partuuid-workaround.rules b/udev/60-ceph-partuuid-workaround.rules
index a1aa060d452..c41a2720440 100644
--- a/udev/60-ceph-partuuid-workaround.rules
+++ b/udev/60-ceph-partuuid-workaround.rules
@@ -31,4 +31,7 @@ KERNEL!="sr*", IMPORT{program}="/sbin/blkid -o udev -p $tempnode"
 ENV{ID_PART_ENTRY_SCHEME}=="gpt", ENV{ID_PART_ENTRY_UUID}=="?*", SYMLINK+="disk/by-partuuid/$env{ID_PART_ENTRY_UUID}"
 ENV{ID_PART_ENTRY_SCHEME}=="gpt", ENV{ID_PART_ENTRY_NAME}=="?*", SYMLINK+="disk/by-partlabel/$env{ID_PART_ENTRY_NAME}"
 
+# NEW: by-parttypeuuid links (type.id)
+ENV{ID_PART_ENTRY_SCHEME}=="gpt", ENV{ID_PART_ENTRY_TYPE}=="?*", ENV{ID_PART_ENTRY_UUID}=="?*", SYMLINK+="disk/by-parttypeuuid/$env{ID_PART_ENTRY_TYPE}.$env{ID_PART_ENTRY_UUID}"
+
 LABEL="persistent_storage_end_two"