osd: Create EC Direct Read flag and pass through to EC.

This is in preperation for supporting sparse and sync reads in EC.
Such ops will only be supported for "balance reads".

Signed-off-by: Alex Ainscow <aainscow@uk.ibm.com>

diff --git a/src/crimson/osd/pg.cc b/src/crimson/osd/pg.cc
index 5154749bcaae311102358187c7bde9df65743106..09ad5f1946d73d5385117eb17da35cedca089d36 100644 (file)
--- a/src/crimson/osd/pg.cc
+++ b/src/crimson/osd/pg.cc
@@ -1604,8 +1604,7 @@ bool PG::can_discard_op(const MOSDOp& m) const {
     return true;
   }
 
-  if ((m.get_flags() & (CEPH_OSD_FLAG_BALANCE_READS |
-                        CEPH_OSD_FLAG_LOCALIZE_READS))
+  if ((m.get_flags() & CEPH_OSD_FLAG_DIRECT_READ)
     && !is_primary()
     && (m.get_map_epoch() <
         peering_state.get_info().history.same_interval_since))

diff --git a/src/include/rados.h b/src/include/rados.h
index e4c58faf59c886983e8fa3b7484c702392404d1a..7680fbdcb1d9b41334b739d6f7d8364d5a667308 100644 (file)
--- a/src/include/rados.h
+++ b/src/include/rados.h
@@ -481,8 +481,12 @@ enum {
        CEPH_OSD_FLAG_IGNORE_REDIRECT = 0x2000000,  /* ignore redirection */
        CEPH_OSD_FLAG_RETURNVEC = 0x4000000, /* allow overall result >= 0, and return >= 0 and buffer for each op in opvec */
        CEPH_OSD_FLAG_SUPPORTSPOOLEIO = 0x8000000,   /* client understands pool EIO flag */
+        CEPH_OSD_FLAG_EC_DIRECT_READ = 0x10000000,  /* Erasure code doing a partial read direct to OSD. */
 };
 
+// Indicates an IO which is direct-to-OSD and may not be on the primary.
+#define CEPH_OSD_FLAG_DIRECT_READ (CEPH_OSD_FLAG_BALANCE_READS | CEPH_OSD_FLAG_LOCALIZE_READS | CEPH_OSD_FLAG_EC_DIRECT_READ)
+
 enum {
        CEPH_OSD_OP_FLAG_EXCL = 0x1,      /* EXCL object create */
        CEPH_OSD_OP_FLAG_FAILOK = 0x2,    /* continue despite failure */

diff --git a/src/osd/OpRequest.h b/src/osd/OpRequest.h
index 2766a7aded5e28bc43cb80eaea612f884354c4bc..255aa59d5307be6b097600a236092285152c6769 100644 (file)
--- a/src/osd/OpRequest.h
+++ b/src/osd/OpRequest.h
@@ -48,6 +48,8 @@ public:
   bool need_skip_handle_cache() const { return op_info.need_skip_handle_cache(); }
   bool need_skip_promote() const { return op_info.need_skip_promote(); }
   bool allows_returnvec() const { return op_info.allows_returnvec(); }
+  bool ec_direct_read() const { return op_info.ec_direct_read(); }
+  void set_ec_direct_read() { return op_info.set_ec_direct_read(); }
 
   std::vector<OpInfo::ClassInfo> classes() const {
     return op_info.get_classes();

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index fe1c51e7595b1aa92ee5ed0274cafd43413bc646..1c57e4cddd82851a742e2aae3d51c438bff6ca2b 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -1938,8 +1938,7 @@ bool PG::can_discard_op(OpRequestRef& op)
     return true;
   }
 
-  if ((m->get_flags() & (CEPH_OSD_FLAG_BALANCE_READS |
-                        CEPH_OSD_FLAG_LOCALIZE_READS)) &&
+  if ((m->get_flags() & (CEPH_OSD_FLAG_DIRECT_READ)) &&
       !is_primary() &&
       m->get_map_epoch() < info.history.same_interval_since) {
     // Note: the Objecter will resend on interval change without the primary

diff --git a/src/osd/PrimaryLogPG.cc b/src/osd/PrimaryLogPG.cc
index 4f28cbbc6e15bf29b94ecb225fe10f88f3a81ee1..f5e7273ed3ab5cdaff09a3a8ea24b0ea5e26c0ec 100644 (file)
--- a/src/osd/PrimaryLogPG.cc
+++ b/src/osd/PrimaryLogPG.cc
@@ -2047,16 +2047,22 @@ void PrimaryLogPG::do_op(OpRequestRef& op)
   }
 
   // check for op with rwordered and rebalance or localize reads
-  if ((m->has_flag(CEPH_OSD_FLAG_BALANCE_READS) || m->has_flag(CEPH_OSD_FLAG_LOCALIZE_READS)) &&
-      op->rwordered()) {
+  if (m->has_flag(CEPH_OSD_FLAG_DIRECT_READ) && op->rwordered()) {
     dout(4) << __func__ << ": rebelance or localized reads with rwordered not allowed "
        << *m << dendl;
     osd->reply_op_error(op, -EINVAL);
     return;
   }
 
-  if ((m->get_flags() & (CEPH_OSD_FLAG_BALANCE_READS |
-                        CEPH_OSD_FLAG_LOCALIZE_READS)) &&
+  if (m->get_flags() & CEPH_OSD_FLAG_EC_DIRECT_READ) {
+    if (is_primary() || is_nonprimary()) {
+      op->set_ec_direct_read();
+    } else {
+      osd->handle_misdirected_op(this, op);
+      return;
+    }
+  } else if ((m->get_flags() & (CEPH_OSD_FLAG_BALANCE_READS |
+                                CEPH_OSD_FLAG_LOCALIZE_READS)) &&
       op->may_read() &&
       !(op->may_write() || op->may_cache())) {
     // balanced reads; any replica will do

diff --git a/src/osd/osd_op_util.cc b/src/osd/osd_op_util.cc
index b0fbe8f370a5cfb26a39e40153c5ed8739b29a4e..2c2ad8e3ec900bfe0c6983e828ed0187180856be 100644 (file)
--- a/src/osd/osd_op_util.cc
+++ b/src/osd/osd_op_util.cc
@@ -52,6 +52,9 @@ bool OpInfo::need_skip_promote() const {
 bool OpInfo::allows_returnvec() const {
   return check_rmw(CEPH_OSD_RMW_FLAG_RETURNVEC);
 }
+bool OpInfo::ec_direct_read() const {
+  return check_rmw(CEPH_OSD_RMW_FLAG_EC_DIRECT_READ);
+}
 /**
  * may_read_data()
  * 
@@ -79,6 +82,7 @@ void OpInfo::set_skip_promote() { set_rmw_flags(CEPH_OSD_RMW_FLAG_SKIP_PROMOTE);
 void OpInfo::set_force_rwordered() { set_rmw_flags(CEPH_OSD_RMW_FLAG_RWORDERED); }
 void OpInfo::set_returnvec() { set_rmw_flags(CEPH_OSD_RMW_FLAG_RETURNVEC); }
 void OpInfo::set_read_data() { set_rmw_flags(CEPH_OSD_RMW_FLAG_READ_DATA); }
+void OpInfo::set_ec_direct_read() { set_rmw_flags(CEPH_OSD_RMW_FLAG_EC_DIRECT_READ); }
 
 
 int OpInfo::set_from_op(

diff --git a/src/osd/osd_op_util.h b/src/osd/osd_op_util.h
index 118934206d8cfe3b045678229de2575495dadf75..ba1acae4c9e42c9e3ce438f0942602e8c9213705 100644 (file)
--- a/src/osd/osd_op_util.h
+++ b/src/osd/osd_op_util.h
@@ -59,6 +59,7 @@ public:
   bool need_skip_handle_cache() const;
   bool need_skip_promote() const;
   bool allows_returnvec() const;
+  bool ec_direct_read() const;
 
   void set_read();
   void set_write();
@@ -72,6 +73,7 @@ public:
   void set_force_rwordered();
   void set_returnvec();
   void set_read_data();
+  void set_ec_direct_read();
 
   int set_from_op(
     const MOSDOp *m,

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index 3e51bed17be59efaacf9e11a4f592bfbe14b0bcd..d3b39355d87f287c55b03092d06bed3a4b942700 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -103,6 +103,7 @@ const char *ceph_osd_flag_name(unsigned flag)
   case CEPH_OSD_FLAG_IGNORE_REDIRECT: return "ignore_redirect";
   case CEPH_OSD_FLAG_RETURNVEC: return "returnvec";
   case CEPH_OSD_FLAG_SUPPORTSPOOLEIO: return "supports_pool_eio";
+  case CEPH_OSD_FLAG_EC_DIRECT_READ: return "ec_direct_read";
   default: return "???";
   }
 }

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index f0495c6e2771ecd4a7233581991633c703e289bc..af58a9cb5e00f0d8103d20b394636f8a3cb38625 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -383,6 +383,7 @@ enum {
   CEPH_OSD_RMW_FLAG_RWORDERED         = (1 << 10),
   CEPH_OSD_RMW_FLAG_RETURNVEC = (1 << 11),
   CEPH_OSD_RMW_FLAG_READ_DATA  = (1 << 12),
+  CEPH_OSD_RMW_FLAG_EC_DIRECT_READ  = (1 << 13),
 };