crimson/os/seastore: make pladdr_t only store the local clone id instead of full...

Signed-off-by: Zhang Song <zhangsong02@qianxin.com>
Signed-off-by: Xuehan Xu <xuxuehan@qianxin.com>

diff --git a/src/crimson/os/seastore/backref/backref_tree_node.h b/src/crimson/os/seastore/backref/backref_tree_node.h
index f8e0612900680a3c71f459d05ae4e8b22a4269cb..d8d5961a9ad2a26d0eb5b97184eca982a46022eb 100644 (file)
--- a/src/crimson/os/seastore/backref/backref_tree_node.h
+++ b/src/crimson/os/seastore/backref/backref_tree_node.h
@@ -33,14 +33,14 @@ constexpr size_t INTERNAL_NODE_CAPACITY = 254;
  *   checksum   : ceph_le32[1]                    4B
  *   size       : ceph_le32[1]                    4B
  *   meta       : backref_node_meta_le_t[1]       20B
- *   keys       : paddr_le_t[CAPACITY]            (193*8)B
- *   values     : backref_map_val_le_t[CAPACITY]  (193*13)B
- *                                                = 4081B
+ *   keys       : paddr_le_t[CAPACITY]            (140*8)B
+ *   values     : backref_map_val_le_t[CAPACITY]  (140*21)B
+ *                                                = 4088B
  *
  * TODO: update FixedKVNodeLayout to handle the above calculation
  * TODO: the above alignment probably isn't portable without further work
  */
-constexpr size_t LEAF_NODE_CAPACITY = 193;
+constexpr size_t LEAF_NODE_CAPACITY = 140;
 
 using BackrefNode = FixedKVNode<paddr_t>;
 

diff --git a/src/crimson/os/seastore/lba/btree_lba_manager.cc b/src/crimson/os/seastore/lba/btree_lba_manager.cc
index c836b5d3b7f824d9289866d2677fd5645b03ea6f..34a065da75b89e077251bb7554515710d725e92e 100644 (file)
--- a/src/crimson/os/seastore/lba/btree_lba_manager.cc
+++ b/src/crimson/os/seastore/lba/btree_lba_manager.cc
@@ -252,7 +252,7 @@ BtreeLBAManager::reserve_region(
   auto iter = btree.make_partial_iter(c, *cursor);
   lba_map_val_t val{
     len,
-    P_ADDR_ZERO,
+    pladdr_t{P_ADDR_ZERO},
     EXTENT_DEFAULT_REF_COUNT,
     0,
     extent_types_t::NONE};
@@ -287,7 +287,7 @@ BtreeLBAManager::alloc_extents(
       ext->get_laddr(),
       lba_map_val_t{
        ext->get_length(),
-       ext->get_paddr(),
+       pladdr_t{ext->get_paddr()},
        EXTENT_DEFAULT_REF_COUNT,
        ext->get_last_committed_crc(),
         ext->get_type()},
@@ -333,13 +333,14 @@ BtreeLBAManager::clone_mapping(
   auto btree = co_await get_btree<LBABtree>(cache, c);
   co_await pos->refresh();
   assert(laddr + len <= pos->get_laddr());
+  assert(inter_key.get_clone_prefix() != laddr.get_clone_prefix());
   auto p = co_await btree.insert(
     c,
     btree.make_partial_iter(c, *pos),
     laddr,
     lba_map_val_t{
       len,
-      inter_key,
+      pladdr_t{inter_key.get_local_clone_id()},
       EXTENT_DEFAULT_REF_COUNT,
       0,
       mapping->get_extent_type()},
@@ -924,7 +925,7 @@ BtreeLBAManager::scan_mapped_space(
            pos.get_val().type);
     ceph_assert(pos.get_val().len > 0 &&
                 pos.get_val().len % block_size == 0);
-    ceph_assert(pos.get_val().pladdr != L_ADDR_NULL);
+    ceph_assert(pos.get_val().pladdr != pladdr_t{LOCAL_CLONE_ID_NULL});
     scan_visitor(
         pos.get_val().pladdr.get_paddr(),
         pos.get_val().len,
@@ -1020,9 +1021,10 @@ BtreeLBAManager::remap_mappings(
     auto new_key = (orig_laddr + remap.offset).checked_to_laddr();
     val.len = remap.len;
     if (val.pladdr.is_laddr()) {
-      auto laddr = val.pladdr.get_laddr();
-      DEBUGT("{} + {:#x}", t, laddr, remap.offset);
-      val.pladdr = (laddr + remap.offset).checked_to_laddr();
+      DEBUGT("{} + {:#x}",
+        t,
+        val.pladdr.get_local_clone_id(),
+        remap.offset);
     } else {
       auto paddr = val.pladdr.get_paddr();
       val.pladdr = paddr + remap.offset;

diff --git a/src/crimson/os/seastore/lba/btree_lba_manager.h b/src/crimson/os/seastore/lba/btree_lba_manager.h
index 35c2de364798d0e1eeeb8ba723dfc44446cfa810..ae0f135d74e1732384fb76c4589d4129cf54ed9f 100644 (file)
--- a/src/crimson/os/seastore/lba/btree_lba_manager.h
+++ b/src/crimson/os/seastore/lba/btree_lba_manager.h
@@ -290,7 +290,7 @@ private:
        laddr,
        {
          len,
-         pladdr_t(intermediate_key),
+         pladdr_t(intermediate_key.get_local_clone_id()),
          EXTENT_DEFAULT_REF_COUNT,
          0,    // crc will only be used and checked with LBA direct mappings
                // also see pin_to_extent(_by_type)

diff --git a/src/crimson/os/seastore/lba/lba_btree_node.h b/src/crimson/os/seastore/lba/lba_btree_node.h
index 0a5f56b62cab48c296be9354d99eba3d30e390d7..e242061d3f45c39e99b68c3726710152479f59f2 100644 (file)
--- a/src/crimson/os/seastore/lba/lba_btree_node.h
+++ b/src/crimson/os/seastore/lba/lba_btree_node.h
@@ -46,15 +46,15 @@ using lba_node_meta_le_t = fixed_kv_node_meta_le_t<laddr_le_t>;
  * Layout (4KiB):
  *   checksum   : ceph_le32[1]               4B
  *   size       : ceph_le32[1]               4B
- *   meta       : lba_node_meta_le_t[1]      20B
- *   keys       : laddr_le_t[CAPACITY]       (254*8)B
- *   values     : paddr_le_t[CAPACITY]       (254*8)B
- *                                           = 4092B
+ *   meta       : lba_node_meta_le_t[1]      36B
+ *   keys       : laddr_le_t[CAPACITY]       (168*16)B
+ *   values     : paddr_le_t[CAPACITY]       (168*8)B
+ *                                           = 4076B
 
  * TODO: make the above capacity calculation part of FixedKVNodeLayout
  * TODO: the above alignment probably isn't portable without further work
  */
-constexpr size_t INTERNAL_NODE_CAPACITY = 254;
+constexpr size_t INTERNAL_NODE_CAPACITY = 168;
 struct LBAInternalNode
   : FixedKVInternalNode<
       INTERNAL_NODE_CAPACITY,
@@ -89,15 +89,15 @@ using LBAInternalNodeRef = LBAInternalNode::Ref;
  * Layout (4KiB):
  *   checksum   : ceph_le32[1]                4B
  *   size       : ceph_le32[1]                4B
- *   meta       : lba_node_meta_le_t[1]       20B
- *   keys       : laddr_le_t[CAPACITY]        (140*8)B
- *   values     : lba_map_val_le_t[CAPACITY]  (140*21)B
- *                                            = 4088B
+ *   meta       : lba_node_meta_le_t[1]       36B
+ *   keys       : laddr_le_t[CAPACITY]        (109*16)B
+ *   values     : lba_map_val_le_t[CAPACITY]  (109*21)B
+ *                                            = 4077B
  *
  * TODO: update FixedKVNodeLayout to handle the above calculation
  * TODO: the above alignment probably isn't portable without further work
  */
-constexpr size_t LEAF_NODE_CAPACITY = 135;
+constexpr size_t LEAF_NODE_CAPACITY = 109;
 
 struct LBALeafNode
   : FixedKVLeafNode<
@@ -425,7 +425,7 @@ struct LBACursor : BtreeCursor<laddr_t, lba::lba_map_val_t, LBALeafNode> {
     assert(is_viewable());
     assert(is_indirect());
     assert(!is_end());
-    return iter.get_val().pladdr.get_laddr();
+    return iter.get_val().pladdr.build_laddr(key);
   }
   checksum_t get_checksum() const {
     assert(is_viewable());

diff --git a/src/crimson/os/seastore/seastore_types.cc b/src/crimson/os/seastore/seastore_types.cc
index 929d68d1526d689bd05fae1629abdc23491912bb..0d599436265459f46c18c4ef43588ab5b8357950 100644 (file)
--- a/src/crimson/os/seastore/seastore_types.cc
+++ b/src/crimson/os/seastore/seastore_types.cc
@@ -156,11 +156,16 @@ std::ostream &operator<<(std::ostream &out, const laddr_offset_t &laddr_offset)
 
 std::ostream &operator<<(std::ostream &out, const pladdr_t &pladdr)
 {
+  out << "pladdr(";
   if (pladdr.is_laddr()) {
-    return out << pladdr.get_laddr();
+    // pladdr(local_clone_id=0x...)
+    out << "local_clone_id=0x" << std::hex
+       << pladdr.get_local_clone_id() << std::dec;
   } else {
-    return out << pladdr.get_paddr();
+    // pladdr(paddr<...>)
+    out << pladdr.get_paddr();
   }
+  return out << ")";
 }
 
 std::ostream &operator<<(std::ostream &out, const paddr_t &rhs)

diff --git a/src/crimson/os/seastore/seastore_types.h b/src/crimson/os/seastore/seastore_types.h
index 56b316a145c45d891c729edd065bdb6eeb37d773..9bbb35495c87f85abe3d27247beb94f7f6f57141 100644 (file)
--- a/src/crimson/os/seastore/seastore_types.h
+++ b/src/crimson/os/seastore/seastore_types.h
@@ -1647,13 +1647,20 @@ struct __attribute__((packed)) laddr_le_t {
   bool operator==(const laddr_le_t&) const = default;
 };
 
+/**
+ * pladdr_t
+ *
+ * The value of LBA tree leaf node entries, stores either the physical address
+ * of the logical extent, or the value of local_clone_id field of the intermediate
+ * key which points to the physical lba mapping.
+ */
 struct pladdr_t {
-  std::variant<laddr_t, paddr_t> pladdr;
+  std::variant<local_clone_id_t, paddr_t> pladdr;
 
   pladdr_t() = default;
-  pladdr_t(const pladdr_t &) noexcept = default;
-  explicit pladdr_t(laddr_t laddr)
-    : pladdr(laddr) {}
+  pladdr_t(const pladdr_t &) = default;
+  explicit pladdr_t(local_clone_id_t id)
+    : pladdr(id) {}
   constexpr explicit pladdr_t(paddr_t paddr)
     : pladdr(paddr) {}
 
@@ -1670,8 +1677,8 @@ struct pladdr_t {
     return *this;
   }
 
-  pladdr_t& operator=(laddr_t laddr) {
-    pladdr = laddr;
+  pladdr_t& operator=(local_clone_id_t id) {
+    pladdr = id;
     return *this;
   }
 
@@ -1682,11 +1689,16 @@ struct pladdr_t {
     return paddr_t(std::get<1>(pladdr));
   }
 
-  laddr_t get_laddr() const {
+  local_clone_id_t get_local_clone_id() const {
     assert(pladdr.index() == 0);
-    return laddr_t(std::get<0>(pladdr));
+    return std::get<0>(pladdr);
   }
 
+  // The corresponding lba key with stored local clone id is the real
+  // intermediate key.
+  laddr_t build_laddr(laddr_t key) const {
+    return key.with_local_clone_id(get_local_clone_id());
+  }
 };
 
 constexpr pladdr_t PL_ADDR_NULL = pladdr_t(P_ADDR_NULL);
@@ -1699,29 +1711,26 @@ enum class addr_type_t : uint8_t {
 };
 
 struct __attribute__((packed)) pladdr_le_t {
-  ceph_le64 pladdr = ceph_le64(0);
-  addr_type_t addr_type = addr_type_t::MAX;
+  ceph_le64 addr;
+  addr_type_t addr_type;
 
-  pladdr_le_t() = default;
+  pladdr_le_t() : pladdr_le_t(PL_ADDR_NULL) {}
   pladdr_le_t(const pladdr_le_t &) = default;
   explicit pladdr_le_t(const pladdr_t &addr)
-    : pladdr(
-       ceph_le64(
-         addr.is_laddr() ?
-           std::get<0>(addr.pladdr).value :
-           std::get<1>(addr.pladdr).internal_paddr)),
-      addr_type(
-       addr.is_laddr() ?
-         addr_type_t::LADDR :
-         addr_type_t::PADDR)
+    : addr(ceph_le64(addr.is_laddr()
+                    ? addr.get_local_clone_id()
+                    : addr.get_paddr().internal_paddr)),
+      addr_type(addr.is_laddr()
+               ? addr_type_t::LADDR
+               : addr_type_t::PADDR)
   {}
 
   operator pladdr_t() const {
     if (addr_type == addr_type_t::LADDR) {
-      return pladdr_t(laddr_t(pladdr));
+      return pladdr_t(static_cast<local_clone_id_t>(addr));
     } else {
       assert(addr_type == addr_type_t::PADDR);
-      return pladdr_t(paddr_t(pladdr));
+      return pladdr_t(paddr_t(addr));
     }
   }
 };

diff --git a/src/test/crimson/seastore/test_transaction_manager.cc b/src/test/crimson/seastore/test_transaction_manager.cc
index d792d412ed9f78dbfe91b7e313472bbafb3abe35..1579ae110c7a95744d78b754cc8f82b491be225f 100644 (file)
--- a/src/test/crimson/seastore/test_transaction_manager.cc
+++ b/src/test/crimson/seastore/test_transaction_manager.cc
@@ -1860,10 +1860,10 @@ TEST_P(tm_random_block_device_test_t, scatter_allocation)
     laddr_t ADDR = get_laddr_hint(0xFF * 4096);
     epm->prefill_fragmented_devices();
     auto t = create_transaction();
-    for (int i = 0; i < 1958; i++) {
+    for (int i = 0; i < 1975; i++) {
       auto extents = alloc_extents(t, (ADDR + i * 16384).checked_to_laddr(), 16384, 'a');
     }
-    alloc_extents_deemed_fail(t, (ADDR + 1958 * 16384).checked_to_laddr(), 16384, 'a');
+    alloc_extents_deemed_fail(t, (ADDR + 1975 * 16384).checked_to_laddr(), 16384, 'a');
     check_mappings(t);
     check();
     submit_transaction(std::move(t));