[HICN-690] Transport Library Major Refactory

The current patch provides a major refactory of the transportlibrary.
A summary of the different components that underwent major modifications is
reported below.

- Transport protocol updates
The hierarchy of classes has been optimized to have common transport services
across different transport protocols. This can allow to customize a transport
protocol with new features.

- A new real-time communication protocol
The RTC protocol has been optimized in terms of algorithms to reduce
consumer-producer synchronization latency.

- A novel socket API
The API has been reworked to be easier to consumer but also to have a more
efficient integration in L4 proxies.

- Several performance improvements
A large number of performance improvements have been included in
particular to make the entire stack zero-copy and optimize cache miss.

- New memory buffer framework
Memory management has been reworked entirely to provide a more efficient infra
with a richer API. Buffers are now allocated in blocks and a single buffer
holds the memory for (1) the shared_ptr control block, (2) the metadata of the
packet (e.g. name, pointer to other buffers if buffer is chained and relevant
offsets), and (3) the packet itself, as it is sent/received over the network.

- A new slab allocator
Dynamic memory allocation is now managed by a novel slab allocator that is
optimised for packet processing and connection management. Memory is organized
in pools of blocks all of the same size which are used during the processing of
outgoing/incoming packets. When a memory block Is allocated is always taken
from a global pool and when it is deallocated is returned to the pool, thus
avoiding the cost of any heap allocation in the data path.

- New transport connectors
Consumer and producer end-points can communication either using an hicn packet
forwarder or with direct connector based on shared memories or sockets.
The usage of transport connectors typically for unit and funcitonal
testing but may have additional usage.

- Support for FEC/ECC for transport services
FEC/ECC via reed solomon is supported by default and made available to
transport services as a modular component. Reed solomon block codes is a
default FEC model that can be replaced in a modular way by many other
codes including RLNC not avaiable in this distribution.
The current FEC framework support variable size padding and efficiently
makes use of the infra memory buffers to avoid additiona copies.

- Secure transport framework for signature computation and verification
Crypto support is nativelty used in hICN for integrity and authenticity.
Novel support that includes RTC has been implemented and made modular
and reusable acrosso different transport protocols.

- TLS - Transport layer security over hicn
Point to point confidentiality is provided by integrating TLS on top of
hICN reliable and non-reliable transport. The integration is common and
makes a different use of the TLS record.

- MLS - Messaging layer security over hicn
MLS integration on top of hICN is made by using the MLSPP implemetation
open sourced by Cisco. We have included instrumentation tools to deploy
performance and functional tests of groups of end-points.

- Android support
The overall code has been heavily tested in Android environments and
has received heavy lifting to better run natively in recent Android OS.

Co-authored-by: Mauro Sardara <msardara@cisco.com>
Co-authored-by: Michele Papalini <micpapal@cisco.com>
Co-authored-by: Olivier Roques <oroques+fdio@cisco.com>
Co-authored-by: Giulio Grassi <gigrassi@cisco.com>
Change-Id: If477ba2fa686e6f47bdf96307ac60938766aef69
Signed-off-by: Luca Muscariello <muscariello@ieee.org>

1 files changed, 144 insertions, 65 deletions

diff --git a/libtransport/includes/hicn/transport/utils/fixed_block_allocator.h b/libtransport/includes/hicn/transport/utils/fixed_block_allocator.h
index 1ade1516e..b1df36493 100644
--- a/libtransport/includes/hicn/transport/utils/fixed_block_allocator.h
+++ b/libtransport/includes/hicn/transport/utils/fixed_block_allocator.h
@@ -5,104 +5,99 @@
 #pragma once
 
 #include <hicn/transport/portability/c_portability.h>
+#include <hicn/transport/utils/singleton.h>
 #include <hicn/transport/utils/spinlock.h>
-
 #include <stdint.h>
+
+#include <cassert>
 #include <cstdlib>
 #include <memory>
-#include <cassert>
 
 namespace utils {
-template <std::size_t DEFAULT_SIZE = 512, std::size_t OBJECTS = 4096>
-class FixedBlockAllocator {
-  FixedBlockAllocator(std::size_t size = DEFAULT_SIZE,
-                      std::size_t objects = OBJECTS)
-      : block_size_(size < sizeof(void*) ? sizeof(long*) : size),
-        object_size_(size),
-        max_objects_(objects),
-        p_head_(NULL),
-        pool_index_(0),
-        block_count_(0),
-        blocks_in_use_(0),
-        allocations_(0),
-        deallocations_(0) {
-    p_pool_ = (uint8_t*)new uint8_t[block_size_ * max_objects_];
-  }
+template <std::size_t SIZE = 512, std::size_t OBJECTS = 4096>
+class FixedBlockAllocator
+    : public utils::Singleton<FixedBlockAllocator<SIZE, OBJECTS>> {
+  friend class utils::Singleton<FixedBlockAllocator<SIZE, OBJECTS>>;
 
  public:
-  static FixedBlockAllocator* getInstance() {
-    if (!instance_) {
-      instance_ = std::unique_ptr<FixedBlockAllocator>(
-          new FixedBlockAllocator(DEFAULT_SIZE, OBJECTS));
+  ~FixedBlockAllocator() {
+    for (auto& p : p_pools_) {
+      delete[] p;
     }
-
-    return instance_.get();
   }
 
-  ~FixedBlockAllocator() { delete[] p_pool_; }
-
-  TRANSPORT_ALWAYS_INLINE void* allocateBlock(size_t size = DEFAULT_SIZE) {
-    assert(size <= DEFAULT_SIZE);
+  void* allocateBlock(size_t size = SIZE) {
+    assert(size <= SIZE);
     uint32_t index;
 
+    SpinLock::Acquire locked(lock_);
     void* p_block = pop();
     if (!p_block) {
-      if (pool_index_ < max_objects_) {
-        {
-          SpinLock::Acquire locked(lock_);
-          index = pool_index_++;
-        }
-        p_block = (void*)(p_pool_ + (index * block_size_));
-      } else {
-        // TODO Consider increasing pool here instead of throwing an exception
-        throw std::runtime_error("No more memory available from packet pool!");
+      if (TRANSPORT_EXPECT_FALSE(current_pool_index_ >= max_objects_)) {
+        // Allocate new memory block
+        TRANSPORT_LOGV("Allocating new block of %zu size", SIZE * OBJECTS);
+        p_pools_.emplace_front(
+            new typename std::aligned_storage<SIZE>::type[max_objects_]);
+        // reset current_pool_index_
+        current_pool_index_ = 0;
       }
-    }
 
-    blocks_in_use_++;
-    allocations_++;
+      auto& latest = p_pools_.front();
+      index = current_pool_index_++;
+      blocks_in_use_++;
+      allocations_++;
+      p_block = (void*)&latest[index];
+    }
 
     return p_block;
   }
 
-  TRANSPORT_ALWAYS_INLINE void deallocateBlock(void* pBlock) {
+  void deallocateBlock(void* pBlock) {
+    SpinLock::Acquire locked(lock_);
     push(pBlock);
-    {
-      SpinLock::Acquire locked(lock_);
-      blocks_in_use_--;
-      deallocations_++;
-    }
+    blocks_in_use_--;
+    deallocations_++;
   }
 
-  TRANSPORT_ALWAYS_INLINE std::size_t blockSize() { return block_size_; }
+ public:
+  std::size_t blockSize() { return block_size_; }
 
-  TRANSPORT_ALWAYS_INLINE uint32_t blockCount() { return block_count_; }
+  uint32_t blockCount() { return block_count_; }
 
-  TRANSPORT_ALWAYS_INLINE uint32_t blocksInUse() { return blocks_in_use_; }
+  uint32_t blocksInUse() { return blocks_in_use_; }
 
-  TRANSPORT_ALWAYS_INLINE uint32_t allocations() { return allocations_; }
+  uint32_t allocations() { return allocations_; }
 
-  TRANSPORT_ALWAYS_INLINE uint32_t deallocations() { return deallocations_; }
+  uint32_t deallocations() { return deallocations_; }
 
  private:
-  TRANSPORT_ALWAYS_INLINE void push(void* p_memory) {
+  FixedBlockAllocator()
+      : block_size_(SIZE),
+        object_size_(SIZE),
+        max_objects_(OBJECTS),
+        p_head_(NULL),
+        current_pool_index_(0),
+        block_count_(0),
+        blocks_in_use_(0),
+        allocations_(0),
+        deallocations_(0) {
+    static_assert(SIZE >= sizeof(long*), "SIZE must be at least 8 bytes");
+    p_pools_.emplace_front(
+        new typename std::aligned_storage<SIZE>::type[max_objects_]);
+  }
+
+  void push(void* p_memory) {
     Block* p_block = (Block*)p_memory;
-    {
-      SpinLock::Acquire locked(lock_);
-      p_block->p_next = p_head_;
-      p_head_ = p_block;
-    }
+    p_block->p_next = p_head_;
+    p_head_ = p_block;
   }
 
-  TRANSPORT_ALWAYS_INLINE void* pop() {
+  void* pop() {
     Block* p_block = nullptr;
 
-    {
-      SpinLock::Acquire locked(lock_);
-      if (p_head_) {
-        p_block = p_head_;
-        p_head_ = p_head_->p_next;
-      }
+    if (p_head_) {
+      p_block = p_head_;
+      p_head_ = p_head_->p_next;
     }
 
     return (void*)p_block;
@@ -119,8 +114,8 @@ class FixedBlockAllocator {
   const std::size_t max_objects_;
 
   Block* p_head_;
-  uint8_t* p_pool_;
-  uint32_t pool_index_;
+  uint32_t current_pool_index_;
+  std::list<typename std::aligned_storage<SIZE>::type*> p_pools_;
   uint32_t block_count_;
   uint32_t blocks_in_use_;
   uint32_t allocations_;
@@ -133,4 +128,88 @@ template <std::size_t A, std::size_t B>
 std::unique_ptr<FixedBlockAllocator<A, B>>
     FixedBlockAllocator<A, B>::instance_ = nullptr;
 
+/**
+ * STL Allocator trait to be used with allocate_shared.
+ */
+template <typename T, typename Pool>
+class STLAllocator {
+  /**
+   * If STLAllocator is rebound to another type (!= T) using copy constructor,
+   * we may need to access private members of the source allocator to copy
+   * memory and pool.
+   */
+  template <typename U, typename P>
+  friend class STLAllocator;
+
+ public:
+  using size_type = std::size_t;
+  using difference_type = ptrdiff_t;
+  using pointer = T*;
+  using const_pointer = const T*;
+  using reference = T&;
+  using const_reference = const T&;
+  using value_type = T;
+
+  STLAllocator(pointer memory, Pool* memory_pool)
+      : memory_(memory), pool_(memory_pool) {
+    TRANSPORT_LOGV("Creating allocator. This: %p, memory: %p, memory_pool: %p",
+                   this, memory, memory_pool);
+  }
+
+  ~STLAllocator() {}
+
+  template <typename U>
+  STLAllocator(const STLAllocator<U, Pool>& other) {
+    memory_ = other.memory_;
+    pool_ = other.pool_;
+  }
+
+  template <typename U>
+  struct rebind {
+    typedef STLAllocator<U, Pool> other;
+  };
+
+  pointer address(reference x) const { return &x; }
+  const_pointer address(const_reference x) const { return &x; }
+
+  pointer allocate(size_type n, pointer hint = 0) {
+    TRANSPORT_LOGV(
+        "Allocating memory (%zu). This: %p, memory: %p, memory_pool: %p", n,
+        this, memory_, pool_);
+    return static_cast<pointer>(memory_);
+  }
+
+  void deallocate(pointer p, size_type n) {
+    TRANSPORT_LOGV("Deallocating memory. This: %p, memory: %p, memory_pool: %p",
+                   this, memory_, pool_);
+    pool_->deallocateBlock(memory_);
+  }
+
+  template <typename... Args>
+  void construct(pointer p, Args&&... args) {
+    new (static_cast<pointer>(p)) T(std::forward<Args>(args)...);
+  }
+
+  void destroy(pointer p) {
+    TRANSPORT_LOGV("Destroying object. This: %p, memory: %p, memory_pool: %p",
+                   this, memory_, pool_);
+    p->~T();
+  }
+
+ private:
+  void* memory_;
+  Pool* pool_;
+};
+
+template <typename T, typename U, typename V>
+inline bool operator==(const STLAllocator<T, V>&, const STLAllocator<U, V>&) {
+  return true;
+}
+
+template <typename T, typename U, typename V>
+inline bool operator!=(const STLAllocator<T, V>& a,
+                       const STLAllocator<U, V>& b) {
+  return !(a == b);
+}
+
 }  // namespace utils
\ No newline at end of file