From: Sage Weil <sage@inktank.com>
Date: Wed, 23 Oct 2013 01:44:03 +0000 (-0700)
Subject: librados, osd: add IGNORE_OVERLAY flag
X-Git-Tag: v0.75~45^2~60
X-Git-Url: http://git-server-git.apps.pok.os.sepia.ceph.com/?a=commitdiff_plain;h=181cb8e83cc010884bda4d650ca1b08a22e716f9;p=ceph.git

librados, osd: add IGNORE_OVERLAY flag

Add a flag that will make the OSD bypass the cache overlay logic.  This is
needed in order to handle operations like CACHE_EVICT and CACHE_FLUSH.

Signed-off-by: Sage Weil <sage@inktank.com>
---

diff --git a/src/include/rados.h b/src/include/rados.h
index 2ab293daa689..484f22175165 100644
--- a/src/include/rados.h
+++ b/src/include/rados.h
@@ -346,6 +346,7 @@ enum {
 	CEPH_OSD_FLAG_EXEC_PUBLIC =    0x1000,  /* DEPRECATED op may exec (public) */
 	CEPH_OSD_FLAG_LOCALIZE_READS = 0x2000,  /* read from nearby replica, if any */
 	CEPH_OSD_FLAG_RWORDERED =      0x4000,  /* order wrt concurrent reads */
+	CEPH_OSD_FLAG_IGNORE_OVERLAY = 0x8000,  /* ignore pool overlay */
 };
 
 enum {
diff --git a/src/include/rados/librados.hpp b/src/include/rados/librados.hpp
index e66ef224f2ef..7e7f069ff2f4 100644
--- a/src/include/rados/librados.hpp
+++ b/src/include/rados/librados.hpp
@@ -141,12 +141,17 @@ namespace librados
    * ORDER_READS_WRITES will order reads the same way writes are
    * ordered (e.g., waiting for degraded objects).  In particular, it
    * will make a write followed by a read sequence be preserved.
+   *
+   * IGNORE_OVERLAY will ignore the pool overlay tiering metadata and
+   * process the op directly on the destination pool.  This is useful
+   * for CACHE_FLUSH and CACHE_EVICT operations.
    */
   enum ObjectOperationGlobalFlags {
     OPERATION_NOFLAG         = 0,
     OPERATION_BALANCE_READS  = 1,
     OPERATION_LOCALIZE_READS = 2,
     OPERATION_ORDER_READS_WRITES = 4,
+    OPERATION_IGNORE_OVERLAY = 8,
   };
 
   /*
diff --git a/src/librados/librados.cc b/src/librados/librados.cc
index e58ccbdc92e1..44dbb40cc541 100644
--- a/src/librados/librados.cc
+++ b/src/librados/librados.cc
@@ -995,6 +995,8 @@ int librados::IoCtx::aio_operate(const std::string& oid, AioCompletion *c,
     op_flags |= CEPH_OSD_FLAG_LOCALIZE_READS;
   if (flags & OPERATION_ORDER_READS_WRITES)
     op_flags |= CEPH_OSD_FLAG_RWORDERED;
+  if (flags & OPERATION_IGNORE_OVERLAY)
+    op_flags |= CEPH_OSD_FLAG_IGNORE_OVERLAY;
 
   return io_ctx_impl->aio_operate_read(obj, (::ObjectOperation*)o->impl, c->pc,
 				       op_flags, pbl);
diff --git a/src/messages/MOSDOp.h b/src/messages/MOSDOp.h
index 3855670f8ebf..11c8bd95f6f1 100644
--- a/src/messages/MOSDOp.h
+++ b/src/messages/MOSDOp.h
@@ -375,6 +375,8 @@ struct ceph_osd_request_head {
       out << " localize_reads";
     if (get_flags() & CEPH_OSD_FLAG_RWORDERED)
       out << " rwordered";
+    if (get_flags() & CEPH_OSD_FLAG_IGNORE_OVERLAY)
+      out << " ignore_overlay";
     out << " e" << osdmap_epoch;
     out << ")";
   }
diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index fe6d948d403b..0c25393524d4 100644
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -1035,7 +1035,8 @@ void ReplicatedPG::do_op(OpRequestRef op)
     }
   }
 
-  if (maybe_handle_cache(op, obc, r))
+  if ((m->get_flags() & CEPH_OSD_FLAG_IGNORE_OVERLAY) == 0 &&
+      maybe_handle_cache(op, obc, r))
     return;
 
   if (r) {