OSD,PG,ObjectStore: handle messages with old hobject_t encoding

Messages that embed an hobject_t need to have the pool field fixed
on messages from old peers.

Signed-off-by: Samuel Just <sam.just@inktank.com>

diff --git a/src/messages/MOSDPGBackfill.h b/src/messages/MOSDPGBackfill.h
index 7e6c45e1e9a705d55bee17ab3d4991da142cb936..bc2a696b6ce79759a3eb776c45bf842216b73abd 100644 (file)
--- a/src/messages/MOSDPGBackfill.h
+++ b/src/messages/MOSDPGBackfill.h
@@ -48,6 +48,11 @@ public:
     ::decode(pgid, p);
     ::decode(last_backfill, p);
     ::decode(stats, p);
+
+    // Handle hobject_t format change
+    if (!last_backfill.is_max() &&
+       last_backfill.pool == -1)
+      last_backfill.pool = pgid.pool();
   }
 
   virtual void encode_payload(uint64_t features) {

diff --git a/src/messages/MOSDPGLog.h b/src/messages/MOSDPGLog.h
index b296fb3997d393ae154847c028a19129cc5be09c..53358d68cdf4115f71f89d911f06c07fcd9d7803 100644 (file)
--- a/src/messages/MOSDPGLog.h
+++ b/src/messages/MOSDPGLog.h
@@ -70,8 +70,8 @@ public:
     bufferlist::iterator p = payload.begin();
     ::decode(epoch, p);
     ::decode(info, p);
-    ::decode(log, p);
-    ::decode(missing, p);
+    log.decode(p, info.pgid.pool());
+    missing.decode(p, info.pgid.pool());
     if (header.version >= 2) {
       ::decode(query_epoch, p);
     }

diff --git a/src/messages/MOSDPGMissing.h b/src/messages/MOSDPGMissing.h
index 024414b37913f918724fed359e3f4cecff702cfe..b931b36353fe2f3716e2265b1b4d00fa5e28537f 100644 (file)
--- a/src/messages/MOSDPGMissing.h
+++ b/src/messages/MOSDPGMissing.h
@@ -50,7 +50,7 @@ public:
     bufferlist::iterator p = payload.begin();
     ::decode(epoch, p);
     ::decode(info, p);
-    ::decode(missing, p);
+    missing.decode(p, info.pgid.pool());
   }
 };
 

diff --git a/src/messages/MOSDPGScan.h b/src/messages/MOSDPGScan.h
index cc69b92e170b7add8e19ba8072c1eeadfc86c728..7df3a966dba7a641757ec37c474b087692b2bd61 100644 (file)
--- a/src/messages/MOSDPGScan.h
+++ b/src/messages/MOSDPGScan.h
@@ -45,6 +45,12 @@ public:
     ::decode(pgid, p);
     ::decode(begin, p);
     ::decode(end, p);
+
+    // handle hobject_t format upgrade
+    if (begin.pool == -1)
+      begin.pool = pgid.pool();
+    if (end.pool == -1)
+      end.pool = pgid.pool();
   }
 
   virtual void encode_payload(uint64_t features) {

diff --git a/src/messages/MOSDSubOp.h b/src/messages/MOSDSubOp.h
index a92449a930950f6b2dee70419e4d273bfe743cf5..3da997dc8df2f0ed3f52c7dddb9f49b9128b06f4 100644 (file)
--- a/src/messages/MOSDSubOp.h
+++ b/src/messages/MOSDSubOp.h
@@ -25,7 +25,7 @@
 
 class MOSDSubOp : public Message {
 
-  static const int HEAD_VERSION = 6;
+  static const int HEAD_VERSION = 7;
   static const int COMPAT_VERSION = 1;
 
 public:
@@ -83,6 +83,9 @@ public:
   map<string,bufferlist> omap_entries;
   bufferlist omap_header;
 
+  // indicates that we must fix hobject_t encoding
+  bool hobject_incorrect_pool;
+
   virtual void decode_payload() {
     bufferlist::iterator p = payload.begin();
     ::decode(map_epoch, p);
@@ -125,7 +128,7 @@ public:
       ::decode(oloc, p);
     if (header.version >= 4) {
       ::decode(data_included, p);
-      ::decode(recovery_info, p);
+      recovery_info.decode(p, pgid.pool());
       ::decode(recovery_progress, p);
       ::decode(current_progress, p);
     }
@@ -133,6 +136,13 @@ public:
       ::decode(omap_entries, p);
     if (header.version >= 6)
       ::decode(omap_header, p);
+
+    if (header.version < 7) {
+      // Handle hobject_t format change
+      if (poid.pool == -1)
+       poid.pool = pgid.pool();
+      hobject_incorrect_pool = true;
+    }
   }
 
   virtual void encode_payload(uint64_t features) {
@@ -192,7 +202,8 @@ public:
       noop(noop_),   
       old_exists(false), old_size(0),
       version(v),
-      first(false), complete(false) {
+      first(false), complete(false),
+      hobject_incorrect_pool(false) {
     memset(&peer_stat, 0, sizeof(peer_stat));
     set_tid(rtid);
   }

diff --git a/src/messages/MOSDSubOpReply.h b/src/messages/MOSDSubOpReply.h
index fc2d7717eab72e974784138c512f1514c8fe48c4..d577dd3c2d30fd4c1642b175cc38ae6bb2b15b0d 100644 (file)
--- a/src/messages/MOSDSubOpReply.h
+++ b/src/messages/MOSDSubOpReply.h
@@ -68,6 +68,9 @@ public:
     ::decode(last_complete_ondisk, p);
     ::decode(peer_stat, p);
     ::decode(attrset, p);
+
+    if (poid.pool == -1)
+      poid.pool = pgid.pool();
   }
   virtual void encode_payload(uint64_t features) {
     ::encode(map_epoch, payload);

diff --git a/src/os/ObjectStore.h b/src/os/ObjectStore.h
index 56f5401a5afee9e03b16f4fbdecef9945825f98e..53f365773db69b90db9ef8ece4d2d7b89b0da214 100644 (file)
--- a/src/os/ObjectStore.h
+++ b/src/os/ObjectStore.h
@@ -158,8 +158,13 @@ public:
     uint32_t largest_data_len, largest_data_off, largest_data_off_in_tbl;
     bufferlist tbl;
     bool sobject_encoding;
+    int64_t pool_override;
+    bool use_pool_override;
 
   public:
+    void set_pool_override(int64_t pool) {
+      pool_override = pool;
+    }
 
     void swap(Transaction& other) {
       std::swap(ops, other.ops);
@@ -225,10 +230,14 @@ public:
     class iterator {
       bufferlist::iterator p;
       bool sobject_encoding;
+      int64_t pool_override;
+      bool use_pool_override;
 
       iterator(Transaction *t)
        : p(t->tbl.begin()),
-         sobject_encoding(t->sobject_encoding) {}
+         sobject_encoding(t->sobject_encoding),
+         pool_override(t->pool_override),
+         use_pool_override(t->use_pool_override) {}
 
       friend class Transaction;
 
@@ -253,6 +262,10 @@ public:
          hoid.oid = soid.oid;
        } else {
          ::decode(hoid, p);
+         if (use_pool_override && pool_override != -1 &&
+             hoid.pool == -1) {
+           hoid.pool = pool_override;
+         }
        }
        return hoid;
       }
@@ -531,7 +544,7 @@ public:
     // etc.
     Transaction() :
       ops(0), pad_unused_bytes(0), largest_data_len(0), largest_data_off(0), largest_data_off_in_tbl(0),
-      sobject_encoding(false) {}
+      sobject_encoding(false), pool_override(-1), use_pool_override(false) {}
     Transaction(bufferlist::iterator &dp) :
       ops(0), pad_unused_bytes(0), largest_data_len(0), largest_data_off(0), largest_data_off_in_tbl(0),
       sobject_encoding(false) {
@@ -545,7 +558,7 @@ public:
     }
 
     void encode(bufferlist& bl) const {
-      ENCODE_START(5, 5, bl);
+      ENCODE_START(6, 5, bl);
       ::encode(ops, bl);
       ::encode(pad_unused_bytes, bl);
       ::encode(largest_data_len, bl);
@@ -555,7 +568,7 @@ public:
       ENCODE_FINISH(bl);
     }
     void decode(bufferlist::iterator &bl) {
-      DECODE_START_LEGACY_COMPAT_LEN(5, 5, 5, bl);
+      DECODE_START_LEGACY_COMPAT_LEN(6, 5, 5, bl);
       DECODE_OLDEST(2);
       if (struct_v < 4)
        sobject_encoding = true;
@@ -570,6 +583,9 @@ public:
       }
       ::decode(tbl, bl);
       DECODE_FINISH(bl);
+      if (struct_v < 6) {
+       use_pool_override = true;
+      }
     }
 
     void dump(ceph::Formatter *f);

diff --git a/src/osd/PG.cc b/src/osd/PG.cc
index a63ab28779c0e6c7d7cd7f9547e52ae8dec1a55a..d3c670aa77dc2f0ee3d3796f4c45e1883626a90f 100644 (file)
--- a/src/osd/PG.cc
+++ b/src/osd/PG.cc
@@ -2622,12 +2622,12 @@ void PG::sub_op_scrub_map(OpRequestRef op)
   bufferlist::iterator p = m->get_data().begin();
   if (scrub_received_maps.count(from)) {
     ScrubMap incoming;
-    incoming.decode(p);
+    incoming.decode(p, info.pgid.pool());
     dout(10) << "from replica " << from << dendl;
     dout(10) << "map version is " << incoming.valid_through << dendl;
     scrub_received_maps[from].merge_incr(incoming);
   } else {
-    scrub_received_maps[from].decode(p);
+    scrub_received_maps[from].decode(p, info.pgid.pool());
   }
 
   if (--scrub_waiting_on == 0) {

diff --git a/src/osd/ReplicatedPG.cc b/src/osd/ReplicatedPG.cc
index 2473be6264584a7d4cc80a31fb5ab5920d09bfa3..a2a0efcdce98ed6af6acb476e8af1875944964c1 100644 (file)
--- a/src/osd/ReplicatedPG.cc
+++ b/src/osd/ReplicatedPG.cc
@@ -1140,6 +1140,20 @@ void ReplicatedPG::do_scan(OpRequestRef op)
       bufferlist::iterator p = m->get_data().begin();
       ::decode(bi.objects, p);
 
+      // handle hobject_t encoding change
+      if (bi.objects.size() && bi.objects.begin()->first.pool == -1) {
+       map<hobject_t, eversion_t> tmp;
+       tmp.swap(bi.objects);
+       for (map<hobject_t, eversion_t>::iterator i = tmp.begin();
+            i != tmp.end();
+            ++i) {
+         hobject_t first(i->first);
+         if (first.pool == -1)
+           first.pool = info.pgid.pool();
+         bi.objects[first] = i->second;
+       }
+      }
+
       backfill_pos = backfill_info.begin > peer_backfill_info.begin ?
        peer_backfill_info.begin : backfill_info.begin;
       dout(10) << " backfill_pos now " << backfill_pos << dendl;
@@ -4178,9 +4192,19 @@ void ReplicatedPG::sub_op_modify(OpRequestRef op)
       vector<pg_log_entry_t> log;
       
       bufferlist::iterator p = m->get_data().begin();
+
       ::decode(rm->opt, p);
       p = m->logbl.begin();
       ::decode(log, p);
+      if (m->hobject_incorrect_pool) {
+       for (vector<pg_log_entry_t>::iterator i = log.begin();
+            i != log.end();
+            ++i) {
+         if (i->soid.pool == -1)
+           i->soid.pool = info.pgid.pool();
+       }
+       rm->opt.set_pool_override(info.pgid.pool());
+      }
       
       info.stats = m->pg_stats;
       update_snap_collections(log, rm->localt);

diff --git a/src/osd/osd_types.cc b/src/osd/osd_types.cc
index ecb42ce33d77f2b60bef37ce506c756f061b52ec..cdc7d67a31cc4fda79f77541e6cbdc0c0dc09e26 100644 (file)
--- a/src/osd/osd_types.cc
+++ b/src/osd/osd_types.cc
@@ -1528,16 +1528,16 @@ ostream& operator<<(ostream& out, const pg_log_entry_t& e)
 
 void pg_log_t::encode(bufferlist& bl) const
 {
-  ENCODE_START(3, 3, bl);
+  ENCODE_START(4, 3, bl);
   ::encode(head, bl);
   ::encode(tail, bl);
   ::encode(log, bl);
   ENCODE_FINISH(bl);
 }
  
-void pg_log_t::decode(bufferlist::iterator &bl)
+void pg_log_t::decode(bufferlist::iterator &bl, int64_t pool)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(3, 3, 3, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(4, 3, 3, bl);
   ::decode(head, bl);
   ::decode(tail, bl);
   if (struct_v < 2) {
@@ -1546,6 +1546,17 @@ void pg_log_t::decode(bufferlist::iterator &bl)
   }
   ::decode(log, bl);
   DECODE_FINISH(bl);
+
+  // handle hobject_t format change
+  if (struct_v < 4) {
+    assert(pool != -1);
+    for (list<pg_log_entry_t>::iterator i = log.begin();
+        i != log.end();
+        ++i) {
+      if (i->soid.pool == -1)
+       i->soid.pool = pool;
+    }
+  }
 }
 
 void pg_log_t::dump(Formatter *f) const
@@ -1642,17 +1653,36 @@ ostream& pg_log_t::print(ostream& out) const
 
 void pg_missing_t::encode(bufferlist &bl) const
 {
-  ENCODE_START(2, 2, bl);
+  ENCODE_START(3, 2, bl);
   ::encode(missing, bl);
   ENCODE_FINISH(bl);
 }
 
-void pg_missing_t::decode(bufferlist::iterator &bl)
+void pg_missing_t::decode(bufferlist::iterator &bl, int64_t pool)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(3, 2, 2, bl);
   ::decode(missing, bl);
   DECODE_FINISH(bl);
 
+  if (struct_v < 3) {
+    assert(pool != -1);
+    // Handle hobject_t upgrade
+    map<hobject_t, item> tmp;
+    for (map<hobject_t, item>::iterator i = missing.begin();
+        i != missing.end();
+      ) {
+      if (i->first.pool == -1) {
+       hobject_t to_insert(i->first);
+       to_insert.pool = pool;
+       tmp[to_insert] = i->second;
+       missing.erase(i++);
+      } else {
+       ++i;
+      }
+    }
+    missing.insert(tmp.begin(), tmp.end());
+  }
+
   for (map<hobject_t,item>::iterator it = missing.begin();
        it != missing.end();
        ++it)
@@ -2286,7 +2316,7 @@ void ObjectRecoveryProgress::dump(Formatter *f) const
 
 void ObjectRecoveryInfo::encode(bufferlist &bl) const
 {
-  ENCODE_START(1, 1, bl);
+  ENCODE_START(2, 1, bl);
   ::encode(soid, bl);
   ::encode(version, bl);
   ::encode(size, bl);
@@ -2297,9 +2327,10 @@ void ObjectRecoveryInfo::encode(bufferlist &bl) const
   ENCODE_FINISH(bl);
 }
 
-void ObjectRecoveryInfo::decode(bufferlist::iterator &bl)
+void ObjectRecoveryInfo::decode(bufferlist::iterator &bl,
+                               int64_t pool)
 {
-  DECODE_START(1, bl);
+  DECODE_START(2, bl);
   ::decode(soid, bl);
   ::decode(version, bl);
   ::decode(size, bl);
@@ -2308,6 +2339,22 @@ void ObjectRecoveryInfo::decode(bufferlist::iterator &bl)
   ::decode(copy_subset, bl);
   ::decode(clone_subset, bl);
   DECODE_FINISH(bl);
+
+  if (struct_v < 2) {
+    assert(pool != -1);
+    if (soid.pool == -1)
+      soid.pool = pool;
+    map<hobject_t, interval_set<uint64_t> > tmp;
+    tmp.swap(clone_subset);
+    for (map<hobject_t, interval_set<uint64_t> >::iterator i = tmp.begin();
+        i != tmp.end();
+        ++i) {
+      hobject_t first(i->first);
+      if (first.pool == -1)
+       first.pool = pool;
+      clone_subset[first].swap(i->second);
+    }
+  }
 }
 
 void ObjectRecoveryInfo::generate_test_instances(
@@ -2378,7 +2425,7 @@ void ScrubMap::merge_incr(const ScrubMap &l)
 
 void ScrubMap::encode(bufferlist& bl) const
 {
-  ENCODE_START(2, 2, bl);
+  ENCODE_START(3, 2, bl);
   ::encode(objects, bl);
   ::encode(attrs, bl);
   ::encode(logbl, bl);
@@ -2387,15 +2434,30 @@ void ScrubMap::encode(bufferlist& bl) const
   ENCODE_FINISH(bl);
 }
 
-void ScrubMap::decode(bufferlist::iterator& bl)
+void ScrubMap::decode(bufferlist::iterator& bl, int64_t pool)
 {
-  DECODE_START_LEGACY_COMPAT_LEN(2, 2, 2, bl);
+  DECODE_START_LEGACY_COMPAT_LEN(3, 2, 2, bl);
   ::decode(objects, bl);
   ::decode(attrs, bl);
   ::decode(logbl, bl);
   ::decode(valid_through, bl);
   ::decode(incr_since, bl);
   DECODE_FINISH(bl);
+
+  // handle hobject_t upgrade
+  if (struct_v < 3) {
+    assert(pool != -1);
+    map<hobject_t, object> tmp;
+    tmp.swap(objects);
+    for (map<hobject_t, object>::iterator i = tmp.begin();
+        i != tmp.end();
+        ++i) {
+      hobject_t first(i->first);
+      if (first.pool == -1)
+       first.pool = pool;
+      objects[first] = i->second;
+    }
+  }
 }
 
 void ScrubMap::dump(Formatter *f) const

diff --git a/src/osd/osd_types.h b/src/osd/osd_types.h
index 8d11aea2e30e700a49b376f1c9ee9907421bc188..d3033c344424fe0585833f783663a80462771c4b 100644 (file)
--- a/src/osd/osd_types.h
+++ b/src/osd/osd_types.h
@@ -1330,7 +1330,7 @@ struct pg_log_t {
   ostream& print(ostream& out) const;
 
   void encode(bufferlist &bl) const;
-  void decode(bufferlist::iterator &bl);
+  void decode(bufferlist::iterator &bl, int64_t pool = -1);
   void dump(Formatter *f) const;
   static void generate_test_instances(list<pg_log_t*>& o);
 };
@@ -1401,7 +1401,7 @@ struct pg_missing_t {
   }
 
   void encode(bufferlist &bl) const;
-  void decode(bufferlist::iterator &bl);
+  void decode(bufferlist::iterator &bl, int64_t pool = -1);
   void dump(Formatter *f) const;
   static void generate_test_instances(list<pg_missing_t*>& o);
 };
@@ -1705,7 +1705,7 @@ struct ObjectRecoveryInfo {
 
   static void generate_test_instances(list<ObjectRecoveryInfo*>& o);
   void encode(bufferlist &bl) const;
-  void decode(bufferlist::iterator &bl);
+  void decode(bufferlist::iterator &bl, int64_t pool = -1);
   ostream &print(ostream &out) const;
   void dump(Formatter *f) const;
 };
@@ -1768,7 +1768,7 @@ struct ScrubMap {
   void merge_incr(const ScrubMap &l);
 
   void encode(bufferlist& bl) const;
-  void decode(bufferlist::iterator& bl);
+  void decode(bufferlist::iterator& bl, int64_t pool=-1);
   void dump(Formatter *f) const;
   static void generate_test_instances(list<ScrubMap*>& o);
 };