diff --git a/include/dwarfs/block_compressor.h b/include/dwarfs/block_compressor.h
index 40662913..582f776e 100644
--- a/include/dwarfs/block_compressor.h
+++ b/include/dwarfs/block_compressor.h
@@ -61,12 +61,12 @@ class block_compressor {
   block_compressor(block_compressor&& bc) = default;
   block_compressor& operator=(block_compressor&& rhs) = default;
 
-  std::vector<uint8_t> compress(std::span<uint8_t const> data) const {
+  shared_byte_buffer compress(shared_byte_buffer const& data) const {
     return impl_->compress(data, nullptr);
   }
 
-  std::vector<uint8_t>
-  compress(std::span<uint8_t const> data, std::string const& metadata) const {
+  shared_byte_buffer
+  compress(shared_byte_buffer const& data, std::string const& metadata) const {
     return impl_->compress(data, &metadata);
   }
 
@@ -91,9 +91,8 @@ class block_compressor {
 
     virtual std::unique_ptr<impl> clone() const = 0;
 
-    virtual std::vector<uint8_t>
-    compress(std::span<uint8_t const> data,
-             std::string const* metadata) const = 0;
+    virtual shared_byte_buffer compress(shared_byte_buffer const& data,
+                                        std::string const* metadata) const = 0;
 
     virtual compression_type type() const = 0;
     virtual std::string describe() const = 0;
@@ -152,7 +151,7 @@ class compression_info {
 
   virtual std::string_view name() const = 0;
   virtual std::string_view description() const = 0;
-  virtual std::vector<std::string> const& options() const = 0;
+  virtual std::vector<std::string> const& options() const = 0; // TODO: span?
   virtual std::set<std::string> library_dependencies() const = 0;
 };
 
diff --git a/include/dwarfs/byte_buffer.h b/include/dwarfs/byte_buffer.h
index 4073e718..96c4fa73 100644
--- a/include/dwarfs/byte_buffer.h
+++ b/include/dwarfs/byte_buffer.h
@@ -25,6 +25,7 @@
 #include <concepts>
 #include <memory>
 #include <span>
+#include <vector>
 
 namespace dwarfs {
 
@@ -45,15 +46,23 @@ class byte_buffer_interface {
  public:
   virtual ~byte_buffer_interface() = default;
 
+  virtual uint8_t const* data() const = 0;
+  virtual size_t size() const = 0;
   virtual std::span<uint8_t const> span() const = 0;
 };
 
 class mutable_byte_buffer_interface : public byte_buffer_interface {
  public:
+  virtual uint8_t* mutable_data() = 0;
   virtual std::span<uint8_t> mutable_span() = 0;
   virtual void clear() = 0;
   virtual void reserve(size_t size) = 0;
   virtual void resize(size_t size) = 0;
+  virtual void shrink_to_fit() = 0;
+
+  // TODO: See if we can do without this. This will *only* be implemented
+  //       in the vector_byte_buffer, other implementations will throw.
+  virtual std::vector<uint8_t>& raw_vector() = 0;
 };
 
 class shared_byte_buffer {
@@ -63,14 +72,16 @@ class shared_byte_buffer {
   explicit shared_byte_buffer(std::shared_ptr<byte_buffer_interface const> bb)
       : bb_{std::move(bb)} {}
 
-  uint8_t const* data() const { return span().data(); }
+  uint8_t const* data() const { return bb_->data(); }
 
-  size_t size() const { return span().size(); }
+  size_t size() const { return bb_->size(); }
 
-  bool empty() const { return span().empty(); }
+  bool empty() const { return bb_->size() == 0; }
 
   std::span<uint8_t const> span() const { return bb_->span(); }
 
+  void swap(shared_byte_buffer& other) noexcept { std::swap(bb_, other.bb_); }
+
   template <detail::byte_range T>
   friend bool operator==(shared_byte_buffer const& lhs, T const& rhs) {
     return detail::compare_spans(lhs.span(), {rhs.data(), rhs.size()}) ==
@@ -78,6 +89,7 @@ class shared_byte_buffer {
   }
 
   template <detail::byte_range T>
+    requires(!std::same_as<T, shared_byte_buffer>)
   friend bool operator==(T const& lhs, shared_byte_buffer const& rhs) {
     return detail::compare_spans({lhs.data(), lhs.size()}, rhs.span()) ==
            std::strong_ordering::equal;
@@ -90,6 +102,7 @@ class shared_byte_buffer {
   }
 
   template <detail::byte_range T>
+    requires(!std::same_as<T, shared_byte_buffer>)
   friend std::strong_ordering
   operator<=>(T const& lhs, shared_byte_buffer const& rhs) {
     return detail::compare_spans({lhs.data(), lhs.size()}, rhs.span());
@@ -105,13 +118,13 @@ class mutable_byte_buffer {
       std::shared_ptr<mutable_byte_buffer_interface> bb)
       : bb_{std::move(bb)} {}
 
-  uint8_t const* data() const { return span().data(); }
+  uint8_t const* data() const { return bb_->data(); }
 
-  uint8_t* data() { return span().data(); }
+  uint8_t* data() { return bb_->mutable_data(); }
 
-  size_t size() const { return span().size(); }
+  size_t size() const { return bb_->size(); }
 
-  bool empty() const { return span().empty(); }
+  bool empty() const { return bb_->size() == 0; }
 
   std::span<uint8_t const> span() const { return bb_->span(); }
 
@@ -123,6 +136,12 @@ class mutable_byte_buffer {
 
   void resize(size_t size) { bb_->resize(size); }
 
+  void shrink_to_fit() { bb_->shrink_to_fit(); }
+
+  std::vector<uint8_t>& raw_vector() { return bb_->raw_vector(); }
+
+  void swap(mutable_byte_buffer& other) noexcept { std::swap(bb_, other.bb_); }
+
   template <detail::byte_range T>
   friend bool operator==(mutable_byte_buffer const& lhs, T const& rhs) {
     return detail::compare_spans(lhs.span(), {rhs.data(), rhs.size()}) ==
@@ -130,6 +149,7 @@ class mutable_byte_buffer {
   }
 
   template <detail::byte_range T>
+    requires(!std::same_as<T, mutable_byte_buffer>)
   friend bool operator==(T const& lhs, mutable_byte_buffer const& rhs) {
     return detail::compare_spans({lhs.data(), lhs.size()}, rhs.span()) ==
            std::strong_ordering::equal;
@@ -142,6 +162,7 @@ class mutable_byte_buffer {
   }
 
   template <detail::byte_range T>
+    requires(!std::same_as<T, mutable_byte_buffer>)
   friend std::strong_ordering
   operator<=>(T const& lhs, mutable_byte_buffer const& rhs) {
     return detail::compare_spans({lhs.data(), lhs.size()}, rhs.span());
diff --git a/include/dwarfs/history.h b/include/dwarfs/history.h
index 9ee4e018..e9ebe948 100644
--- a/include/dwarfs/history.h
+++ b/include/dwarfs/history.h
@@ -30,6 +30,7 @@
 
 #include <nlohmann/json.hpp>
 
+#include <dwarfs/byte_buffer.h>
 #include <dwarfs/history_config.h>
 
 namespace dwarfs {
@@ -50,7 +51,7 @@ class history {
   thrift::history::history const& get() const { return *history_; }
   void append(std::optional<std::vector<std::string>> args);
   size_t size() const;
-  std::vector<uint8_t> serialize() const;
+  shared_byte_buffer serialize() const;
   void dump(std::ostream& os) const;
   nlohmann::json as_json() const;
 
diff --git a/include/dwarfs/vector_byte_buffer.h b/include/dwarfs/vector_byte_buffer.h
index c2e54094..f80a4288 100644
--- a/include/dwarfs/vector_byte_buffer.h
+++ b/include/dwarfs/vector_byte_buffer.h
@@ -21,6 +21,8 @@
 
 #pragma once
 
+#include <string>
+
 #include <dwarfs/byte_buffer.h>
 
 namespace dwarfs {
@@ -28,6 +30,10 @@ namespace dwarfs {
 class vector_byte_buffer {
  public:
   static mutable_byte_buffer create();
+  static mutable_byte_buffer create(size_t size);
+  static mutable_byte_buffer create(std::string_view data);
+  static mutable_byte_buffer create(std::span<uint8_t const> data);
+  static mutable_byte_buffer create(std::vector<uint8_t>&& data);
 };
 
 } // namespace dwarfs
diff --git a/include/dwarfs/writer/internal/block_data.h b/include/dwarfs/writer/internal/block_data.h
deleted file mode 100644
index e9ea346b..00000000
--- a/include/dwarfs/writer/internal/block_data.h
+++ /dev/null
@@ -1,54 +0,0 @@
-/* vim:set ts=2 sw=2 sts=2 et: */
-/**
- * \author     Marcus Holland-Moritz (github@mhxnet.de)
- * \copyright  Copyright (c) Marcus Holland-Moritz
- *
- * This file is part of dwarfs.
- *
- * dwarfs is free software: you can redistribute it and/or modify
- * it under the terms of the GNU General Public License as published by
- * the Free Software Foundation, either version 3 of the License, or
- * (at your option) any later version.
- *
- * dwarfs is distributed in the hope that it will be useful,
- * but WITHOUT ANY WARRANTY; without even the implied warranty of
- * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
- * GNU General Public License for more details.
- *
- * You should have received a copy of the GNU General Public License
- * along with dwarfs.  If not, see <https://www.gnu.org/licenses/>.
- */
-
-#pragma once
-
-#include <cstdint>
-#include <string_view>
-#include <vector>
-
-namespace dwarfs::writer::internal {
-
-class block_data {
- public:
-  block_data() = default;
-  explicit block_data(std::vector<uint8_t>&& vec)
-      : vec_{std::move(vec)} {}
-  explicit block_data(std::string_view str)
-      : vec_{str.begin(), str.end()} {}
-
-  std::vector<uint8_t> const& vec() const { return vec_; }
-  std::vector<uint8_t>& vec() { return vec_; }
-
-  void reserve(size_t size) { vec_.reserve(size); }
-
-  uint8_t const* data() const { return vec_.data(); }
-  uint8_t* data() { return vec_.data(); }
-
-  size_t size() const { return vec_.size(); }
-
-  bool empty() const { return vec_.empty(); }
-
- private:
-  std::vector<uint8_t> vec_;
-};
-
-} // namespace dwarfs::writer::internal
diff --git a/include/dwarfs/writer/internal/filesystem_writer_detail.h b/include/dwarfs/writer/internal/filesystem_writer_detail.h
index f8eff04d..cecd3677 100644
--- a/include/dwarfs/writer/internal/filesystem_writer_detail.h
+++ b/include/dwarfs/writer/internal/filesystem_writer_detail.h
@@ -31,6 +31,7 @@
 #include <vector>
 
 #include <dwarfs/block_compressor.h>
+#include <dwarfs/byte_buffer.h>
 #include <dwarfs/compression_constraints.h>
 #include <dwarfs/fstypes.h>
 #include <dwarfs/writer/fragment_category.h>
@@ -45,8 +46,6 @@ class fs_section;
 
 namespace writer::internal {
 
-class block_data;
-
 class filesystem_writer_detail {
  public:
   virtual ~filesystem_writer_detail() = default;
@@ -71,14 +70,13 @@ class filesystem_writer_detail {
   virtual void
   configure_rewrite(size_t filesystem_size, size_t block_count) = 0;
   virtual void copy_header(std::span<uint8_t const> header) = 0;
-  virtual void
-  write_block(fragment_category cat, std::shared_ptr<block_data>&& data,
-              physical_block_cb_type physical_block_cb,
-              std::optional<std::string> meta = std::nullopt) = 0;
+  virtual void write_block(fragment_category cat, shared_byte_buffer data,
+                           physical_block_cb_type physical_block_cb,
+                           std::optional<std::string> meta = std::nullopt) = 0;
   virtual void finish_category(fragment_category cat) = 0;
-  virtual void write_metadata_v2_schema(std::shared_ptr<block_data>&& data) = 0;
-  virtual void write_metadata_v2(std::shared_ptr<block_data>&& data) = 0;
-  virtual void write_history(std::shared_ptr<block_data>&& data) = 0;
+  virtual void write_metadata_v2_schema(shared_byte_buffer data) = 0;
+  virtual void write_metadata_v2(shared_byte_buffer data) = 0;
+  virtual void write_history(shared_byte_buffer data) = 0;
   virtual void check_block_compression(
       compression_type compression, std::span<uint8_t const> data,
       std::optional<fragment_category::value_type> cat = std::nullopt) = 0;
diff --git a/include/dwarfs/writer/internal/metadata_freezer.h b/include/dwarfs/writer/internal/metadata_freezer.h
index 59fdea10..ed8aa36f 100644
--- a/include/dwarfs/writer/internal/metadata_freezer.h
+++ b/include/dwarfs/writer/internal/metadata_freezer.h
@@ -23,7 +23,8 @@
 
 #include <cstdint>
 #include <utility>
-#include <vector>
+
+#include <dwarfs/byte_buffer.h>
 
 namespace dwarfs {
 
@@ -35,7 +36,7 @@ namespace writer::internal {
 
 class metadata_freezer {
  public:
-  static std::pair<std::vector<uint8_t>, std::vector<uint8_t>>
+  static std::pair<shared_byte_buffer, shared_byte_buffer>
   freeze(thrift::metadata::metadata const& data);
 };
 
diff --git a/include/dwarfs/writer/segmenter.h b/include/dwarfs/writer/segmenter.h
index 9d49e2ea..ebb8c122 100644
--- a/include/dwarfs/writer/segmenter.h
+++ b/include/dwarfs/writer/segmenter.h
@@ -26,6 +26,8 @@
 #include <memory>
 #include <vector>
 
+#include <dwarfs/byte_buffer.h>
+
 namespace dwarfs {
 
 struct compression_constraints;
@@ -38,7 +40,6 @@ class writer_progress;
 
 namespace internal {
 
-class block_data;
 class block_manager;
 class chunkable;
 
@@ -55,8 +56,8 @@ class segmenter {
     unsigned block_size_bits{22};
   };
 
-  using block_ready_cb = std::function<void(
-      std::shared_ptr<internal::block_data>, size_t logical_block_num)>;
+  using block_ready_cb =
+      std::function<void(shared_byte_buffer, size_t logical_block_num)>;
 
   segmenter(logger& lgr, writer_progress& prog,
             std::shared_ptr<internal::block_manager> blkmgr, config const& cfg,
diff --git a/src/compression/brotli.cpp b/src/compression/brotli.cpp
index 6d62329e..31aff400 100644
--- a/src/compression/brotli.cpp
+++ b/src/compression/brotli.cpp
@@ -31,6 +31,7 @@
 #include <dwarfs/fstypes.h>
 #include <dwarfs/option_map.h>
 #include <dwarfs/varint.h>
+#include <dwarfs/vector_byte_buffer.h>
 
 namespace dwarfs {
 
@@ -48,10 +49,9 @@ class brotli_block_compressor final : public block_compressor::impl {
     return std::make_unique<brotli_block_compressor>(*this);
   }
 
-  std::vector<uint8_t>
-  compress(std::span<uint8_t const> data,
-           std::string const* /*metadata*/) const override {
-    std::vector<uint8_t> compressed;
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* /*metadata*/) const override {
+    auto compressed = vector_byte_buffer::create(); // TODO: make configurable
     compressed.resize(varint::max_size +
                       ::BrotliEncoderMaxCompressedSize(data.size()));
     size_t size_size = varint::encode(data.size(), compressed.data());
@@ -66,7 +66,7 @@ class brotli_block_compressor final : public block_compressor::impl {
       throw bad_compression_ratio_error();
     }
     compressed.shrink_to_fit();
-    return compressed;
+    return compressed.share();
   }
 
   compression_type type() const override { return compression_type::BROTLI; }
diff --git a/src/compression/flac.cpp b/src/compression/flac.cpp
index 4b34de7c..de2ac87b 100644
--- a/src/compression/flac.cpp
+++ b/src/compression/flac.cpp
@@ -38,6 +38,7 @@
 #include <dwarfs/option_map.h>
 #include <dwarfs/pcm_sample_transformer.h>
 #include <dwarfs/varint.h>
+#include <dwarfs/vector_byte_buffer.h>
 
 #include <dwarfs/gen-cpp2/compression_types.h>
 
@@ -53,7 +54,7 @@ constexpr size_t const kBlockSize{65536};
 
 class dwarfs_flac_stream_encoder final : public FLAC::Encoder::Stream {
  public:
-  explicit dwarfs_flac_stream_encoder(std::vector<uint8_t>& data)
+  explicit dwarfs_flac_stream_encoder(mutable_byte_buffer& data)
       : data_{data}
       , pos_{data_.size()} {}
 
@@ -90,7 +91,7 @@ class dwarfs_flac_stream_encoder final : public FLAC::Encoder::Stream {
   }
 
  private:
-  std::vector<uint8_t>& data_;
+  mutable_byte_buffer& data_;
   size_t pos_;
 };
 
@@ -208,8 +209,8 @@ class flac_block_compressor final : public block_compressor::impl {
     return std::make_unique<flac_block_compressor>(*this);
   }
 
-  std::vector<uint8_t> compress(std::span<uint8_t const> data,
-                                std::string const* metadata) const override {
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* metadata) const override {
     if (!metadata) {
       DWARFS_THROW(runtime_error,
                    "internal error: flac compression requires metadata");
@@ -265,7 +266,7 @@ class flac_block_compressor final : public block_compressor::impl {
       pcm_pad = pcm_sample_padding::Msb;
     }
 
-    std::vector<uint8_t> compressed;
+    auto compressed = vector_byte_buffer::create(); // TODO: make configurable
 
     {
       using namespace ::apache::thrift;
@@ -286,7 +287,7 @@ class flac_block_compressor final : public block_compressor::impl {
       CompactSerializer::serialize(hdr, &hdrbuf);
 
       compressed.resize(pos + hdrbuf.size());
-      ::memcpy(&compressed[pos], hdrbuf.data(), hdrbuf.size());
+      ::memcpy(compressed.data() + pos, hdrbuf.data(), hdrbuf.size());
     }
 
     dwarfs_flac_stream_encoder encoder(compressed);
@@ -341,7 +342,7 @@ class flac_block_compressor final : public block_compressor::impl {
 
     compressed.shrink_to_fit();
 
-    return compressed;
+    return compressed.share();
   }
 
   compression_type type() const override { return compression_type::FLAC; }
diff --git a/src/compression/lz4.cpp b/src/compression/lz4.cpp
index f20efbbb..715ebd29 100644
--- a/src/compression/lz4.cpp
+++ b/src/compression/lz4.cpp
@@ -29,6 +29,7 @@
 #include <dwarfs/error.h>
 #include <dwarfs/fstypes.h>
 #include <dwarfs/option_map.h>
+#include <dwarfs/vector_byte_buffer.h>
 
 namespace dwarfs {
 
@@ -69,18 +70,18 @@ class lz4_block_compressor final : public block_compressor::impl {
     return std::make_unique<lz4_block_compressor>(*this);
   }
 
-  std::vector<uint8_t>
-  compress(std::span<uint8_t const> data,
-           std::string const* /*metadata*/) const override {
-    std::vector<uint8_t> compressed(sizeof(uint32_t) +
-                                    LZ4_compressBound(to<int>(data.size())));
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* /*metadata*/) const override {
+    auto compressed = vector_byte_buffer::create(); // TODO: make configurable
+    compressed.resize(sizeof(uint32_t) +
+                      LZ4_compressBound(to<int>(data.size())));
     // TODO: this should have been a varint; also, if we ever support
     //       big-endian systems, we'll have to properly convert this
     uint32_t size = data.size();
     std::memcpy(compressed.data(), &size, sizeof(size));
-    auto csize = Policy::compress(data.data(), &compressed[sizeof(uint32_t)],
-                                  data.size(),
-                                  compressed.size() - sizeof(uint32_t), level_);
+    auto csize = Policy::compress(
+        data.data(), compressed.data() + sizeof(uint32_t), data.size(),
+        compressed.size() - sizeof(uint32_t), level_);
     if (csize == 0) {
       DWARFS_THROW(runtime_error, "error during compression");
     }
@@ -88,7 +89,7 @@ class lz4_block_compressor final : public block_compressor::impl {
       throw bad_compression_ratio_error();
     }
     compressed.resize(sizeof(uint32_t) + csize);
-    return compressed;
+    return compressed.share();
   }
 
   compression_type type() const override { return compression_type::LZ4; }
diff --git a/src/compression/lzma.cpp b/src/compression/lzma.cpp
index d00eeeac..a35d23b8 100644
--- a/src/compression/lzma.cpp
+++ b/src/compression/lzma.cpp
@@ -37,6 +37,7 @@
 #include <dwarfs/option_map.h>
 #include <dwarfs/sorted_array_map.h>
 #include <dwarfs/types.h>
+#include <dwarfs/vector_byte_buffer.h>
 
 namespace dwarfs {
 
@@ -111,8 +112,8 @@ class lzma_block_compressor final : public block_compressor::impl {
     return std::make_unique<lzma_block_compressor>(*this);
   }
 
-  std::vector<uint8_t> compress(std::span<uint8_t const> data,
-                                std::string const* metadata) const override;
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* metadata) const override;
 
   compression_type type() const override { return compression_type::LZMA; }
 
@@ -126,8 +127,8 @@ class lzma_block_compressor final : public block_compressor::impl {
   }
 
  private:
-  std::vector<uint8_t>
-  compress(std::span<uint8_t const> data, lzma_filter const* filters) const;
+  shared_byte_buffer
+  compress(shared_byte_buffer const& data, lzma_filter const* filters) const;
 
   static uint32_t get_preset(unsigned level, bool extreme) {
     uint32_t preset = level;
@@ -197,8 +198,8 @@ lzma_block_compressor::lzma_block_compressor(option_map& om) {
   }
 }
 
-std::vector<uint8_t>
-lzma_block_compressor::compress(std::span<uint8_t const> data,
+shared_byte_buffer
+lzma_block_compressor::compress(shared_byte_buffer const& data,
                                 lzma_filter const* filters) const {
   lzma_stream s = LZMA_STREAM_INIT;
 
@@ -210,7 +211,8 @@ lzma_block_compressor::compress(std::span<uint8_t const> data,
 
   lzma_action action = LZMA_FINISH;
 
-  std::vector<uint8_t> compressed(data.size() - 1);
+  auto compressed = vector_byte_buffer::create(); // TODO: make configurable
+  compressed.resize(data.size() - 1);
 
   s.next_in = data.data();
   s.avail_in = data.size();
@@ -234,21 +236,21 @@ lzma_block_compressor::compress(std::span<uint8_t const> data,
                                             lzma_error_string(ret)));
   }
 
-  return compressed;
+  return compressed.share();
 }
 
-std::vector<uint8_t>
-lzma_block_compressor::compress(std::span<uint8_t const> data,
+shared_byte_buffer
+lzma_block_compressor::compress(shared_byte_buffer const& data,
                                 std::string const* /*metadata*/) const {
   auto lzma_opts = opt_lzma_;
   std::array<lzma_filter, 3> filters{{{binary_vli_, nullptr},
                                       {LZMA_FILTER_LZMA2, &lzma_opts},
                                       {LZMA_VLI_UNKNOWN, nullptr}}};
 
-  std::vector<uint8_t> best = compress(data, &filters[1]);
+  auto best = compress(data, &filters[1]);
 
   if (filters[0].id != LZMA_VLI_UNKNOWN) {
-    std::vector<uint8_t> compressed = compress(data, filters.data());
+    auto compressed = compress(data, filters.data());
 
     if (compressed.size() < best.size()) {
       best.swap(compressed);
diff --git a/src/compression/null.cpp b/src/compression/null.cpp
index cf5f8681..97e58c47 100644
--- a/src/compression/null.cpp
+++ b/src/compression/null.cpp
@@ -41,11 +41,9 @@ class null_block_compressor final : public block_compressor::impl {
     return std::make_unique<null_block_compressor>(*this);
   }
 
-  // TODO: we should not have to copy the data here...
-  std::vector<uint8_t>
-  compress(std::span<uint8_t const> data,
-           std::string const* /*metadata*/) const override {
-    return std::vector<uint8_t>(data.begin(), data.end());
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* /*metadata*/) const override {
+    return data;
   }
 
   compression_type type() const override { return compression_type::NONE; }
diff --git a/src/compression/ricepp.cpp b/src/compression/ricepp.cpp
index 22dc58df..17872fbb 100644
--- a/src/compression/ricepp.cpp
+++ b/src/compression/ricepp.cpp
@@ -52,8 +52,8 @@ class ricepp_block_compressor final : public block_compressor::impl {
     return std::make_unique<ricepp_block_compressor>(*this);
   }
 
-  std::vector<uint8_t> compress(std::span<uint8_t const> data,
-                                std::string const* metadata) const override {
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* metadata) const override {
     if (!metadata) {
       DWARFS_THROW(runtime_error,
                    "internal error: ricepp compression requires metadata");
@@ -88,8 +88,10 @@ class ricepp_block_compressor final : public block_compressor::impl {
         .unused_lsb_count = static_cast<unsigned>(unused_lsb_count),
     });
 
-    std::vector<uint8_t> compressed;
+    auto compressed = vector_byte_buffer::create(); // TODO: make configurable
 
+    // TODO: see if we can resize just once...
+    // TODO: maybe the mutable_byte_buffer interface can have .append()?
     {
       using namespace ::apache::thrift;
 
@@ -111,7 +113,7 @@ class ricepp_block_compressor final : public block_compressor::impl {
       CompactSerializer::serialize(hdr, &hdrbuf);
 
       compressed.resize(pos + hdrbuf.size());
-      ::memcpy(&compressed[pos], hdrbuf.data(), hdrbuf.size());
+      ::memcpy(compressed.data() + pos, hdrbuf.data(), hdrbuf.size());
     }
 
     std::span<pixel_type const> input{
@@ -121,13 +123,11 @@ class ricepp_block_compressor final : public block_compressor::impl {
     size_t header_size = compressed.size();
     compressed.resize(header_size + codec->worst_case_encoded_bytes(input));
 
-    std::span<uint8_t> buffer(compressed);
-
-    auto output = codec->encode(buffer.subspan(header_size), input);
+    auto output = codec->encode(compressed.span().subspan(header_size), input);
     compressed.resize(header_size + output.size());
     compressed.shrink_to_fit();
 
-    return compressed;
+    return compressed.share();
   }
 
   compression_type type() const override { return compression_type::RICEPP; }
diff --git a/src/compression/zstd.cpp b/src/compression/zstd.cpp
index def9c289..b3fd88ec 100644
--- a/src/compression/zstd.cpp
+++ b/src/compression/zstd.cpp
@@ -29,6 +29,7 @@
 #include <dwarfs/error.h>
 #include <dwarfs/fstypes.h>
 #include <dwarfs/option_map.h>
+#include <dwarfs/vector_byte_buffer.h>
 #include <dwarfs/zstd_context_manager.h>
 
 #if ZSTD_VERSION_MAJOR > 1 ||                                                  \
@@ -54,8 +55,8 @@ class zstd_block_compressor final : public block_compressor::impl {
     return std::make_unique<zstd_block_compressor>(*this);
   }
 
-  std::vector<uint8_t> compress(std::span<uint8_t const> data,
-                                std::string const* metadata) const override;
+  shared_byte_buffer compress(shared_byte_buffer const& data,
+                              std::string const* metadata) const override;
 
   compression_type type() const override { return compression_type::ZSTD; }
 
@@ -87,10 +88,11 @@ class zstd_block_compressor final : public block_compressor::impl {
   int const level_;
 };
 
-std::vector<uint8_t>
-zstd_block_compressor::compress(std::span<uint8_t const> data,
+shared_byte_buffer
+zstd_block_compressor::compress(shared_byte_buffer const& data,
                                 std::string const* /*metadata*/) const {
-  std::vector<uint8_t> compressed(ZSTD_compressBound(data.size()));
+  auto compressed = vector_byte_buffer::create(); // TODO: make configurable
+  compressed.resize(ZSTD_compressBound(data.size()));
   auto ctx = ctxmgr_->make_context();
   auto size = ZSTD_compressCCtx(ctx.get(), compressed.data(), compressed.size(),
                                 data.data(), data.size(), level_);
@@ -103,7 +105,7 @@ zstd_block_compressor::compress(std::span<uint8_t const> data,
   }
   compressed.resize(size);
   compressed.shrink_to_fit();
-  return compressed;
+  return compressed.share();
 }
 
 class zstd_block_decompressor final : public block_decompressor::impl {
diff --git a/src/history.cpp b/src/history.cpp
index cfbdc6a8..edde7c34 100644
--- a/src/history.cpp
+++ b/src/history.cpp
@@ -29,6 +29,7 @@
 #include <dwarfs/config.h>
 #include <dwarfs/history.h>
 #include <dwarfs/library_dependencies.h>
+#include <dwarfs/vector_byte_buffer.h>
 #include <dwarfs/version.h>
 
 #include <dwarfs/gen-cpp2/history_types.h>
@@ -80,10 +81,10 @@ void history::append(std::optional<std::vector<std::string>> args) {
 
 size_t history::size() const { return history_->entries()->size(); }
 
-std::vector<uint8_t> history::serialize() const {
+shared_byte_buffer history::serialize() const {
   std::string buf;
   ::apache::thrift::CompactSerializer::serialize(*history_, &buf);
-  return {buf.begin(), buf.end()};
+  return vector_byte_buffer::create(buf).share();
 }
 
 void history::dump(std::ostream& os) const {
diff --git a/src/mapped_byte_buffer.cpp b/src/mapped_byte_buffer.cpp
index 4736bd3c..5ca91935 100644
--- a/src/mapped_byte_buffer.cpp
+++ b/src/mapped_byte_buffer.cpp
@@ -32,6 +32,10 @@ class mapped_byte_buffer_impl : public byte_buffer_interface {
       : data_{data}
       , mm_{std::move(mm)} {}
 
+  size_t size() const override { return data_.size(); }
+
+  uint8_t const* data() const override { return data_.data(); }
+
   std::span<uint8_t const> span() const override {
     return {data_.data(), data_.size()};
   }
diff --git a/src/reader/filesystem_v2.cpp b/src/reader/filesystem_v2.cpp
index 72a1e925..da0fe919 100644
--- a/src/reader/filesystem_v2.cpp
+++ b/src/reader/filesystem_v2.cpp
@@ -53,7 +53,6 @@
 #include <dwarfs/reader/internal/filesystem_parser.h>
 #include <dwarfs/reader/internal/inode_reader_v2.h>
 #include <dwarfs/reader/internal/metadata_v2.h>
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/filesystem_writer_detail.h>
 
 namespace dwarfs::reader {
diff --git a/src/reader/internal/cached_block.cpp b/src/reader/internal/cached_block.cpp
index 1b40e028..0f037b45 100644
--- a/src/reader/internal/cached_block.cpp
+++ b/src/reader/internal/cached_block.cpp
@@ -78,6 +78,9 @@ class cached_block_ final : public cached_block {
   // This can be called from any thread
   size_t range_end() const override { return range_end_.load(); }
 
+  // TODO: The code relies on the fact that the data_ buffer is never
+  // reallocated once block decompression has started. I would like to
+  // somehow enforce that this cannot happen.
   uint8_t const* data() const override { return data_.data(); }
 
   void decompress_until(size_t end) override {
diff --git a/src/utility/rewrite_filesystem.cpp b/src/utility/rewrite_filesystem.cpp
index 85acf331..47efd17f 100644
--- a/src/utility/rewrite_filesystem.cpp
+++ b/src/utility/rewrite_filesystem.cpp
@@ -28,7 +28,6 @@
 #include <dwarfs/writer/filesystem_writer.h>
 
 #include <dwarfs/reader/internal/filesystem_parser.h>
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/filesystem_writer_detail.h>
 
 namespace dwarfs::utility {
@@ -38,7 +37,6 @@ void rewrite_filesystem(logger& lgr, dwarfs::reader::filesystem_v2 const& fs,
                         dwarfs::writer::category_resolver const& cat_resolver,
                         rewrite_options const& opts) {
   using dwarfs::writer::fragment_category;
-  using dwarfs::writer::internal::block_data;
 
   LOG_PROXY(debug_logger_policy, lgr);
 
@@ -174,7 +172,7 @@ void rewrite_filesystem(logger& lgr, dwarfs::reader::filesystem_v2 const& fs,
     case section_type::HISTORY:
       if (opts.enable_history) {
         history hist{opts.history};
-        hist.parse(fs.get_history().serialize());
+        hist.parse(fs.get_history().serialize().span());
         hist.append(opts.command_line_arguments);
 
         LOG_VERBOSE << "updating " << get_section_name(s->type()) << " ("
@@ -182,7 +180,7 @@ void rewrite_filesystem(logger& lgr, dwarfs::reader::filesystem_v2 const& fs,
                     << "), compressing using '"
                     << writer.get_compressor(s->type()).describe() << "'";
 
-        writer.write_history(std::make_shared<block_data>(hist.serialize()));
+        writer.write_history(hist.serialize());
       } else {
         LOG_VERBOSE << "removing " << get_section_name(s->type());
       }
diff --git a/src/vector_byte_buffer.cpp b/src/vector_byte_buffer.cpp
index 35b7a634..a007da6d 100644
--- a/src/vector_byte_buffer.cpp
+++ b/src/vector_byte_buffer.cpp
@@ -29,6 +29,22 @@ namespace {
 
 class vector_byte_buffer_impl : public mutable_byte_buffer_interface {
  public:
+  vector_byte_buffer_impl() = default;
+  explicit vector_byte_buffer_impl(size_t size)
+      : data_(size) {}
+  explicit vector_byte_buffer_impl(std::string_view data)
+      : data_{data.begin(), data.end()} {}
+  explicit vector_byte_buffer_impl(std::span<uint8_t const> data)
+      : data_{data.begin(), data.end()} {}
+  explicit vector_byte_buffer_impl(std::vector<uint8_t>&& data)
+      : data_{std::move(data)} {}
+
+  size_t size() const override { return data_.size(); }
+
+  uint8_t const* data() const override { return data_.data(); }
+
+  uint8_t* mutable_data() override { return data_.data(); }
+
   std::span<uint8_t const> span() const override {
     return {data_.data(), data_.size()};
   }
@@ -43,6 +59,10 @@ class vector_byte_buffer_impl : public mutable_byte_buffer_interface {
 
   void resize(size_t size) override { data_.resize(size); }
 
+  void shrink_to_fit() override { data_.shrink_to_fit(); }
+
+  std::vector<uint8_t>& raw_vector() override { return data_; }
+
  private:
   std::vector<uint8_t> data_;
 };
@@ -53,4 +73,21 @@ mutable_byte_buffer vector_byte_buffer::create() {
   return mutable_byte_buffer{std::make_shared<vector_byte_buffer_impl>()};
 }
 
+mutable_byte_buffer vector_byte_buffer::create(size_t size) {
+  return mutable_byte_buffer{std::make_shared<vector_byte_buffer_impl>(size)};
+}
+
+mutable_byte_buffer vector_byte_buffer::create(std::string_view data) {
+  return mutable_byte_buffer{std::make_shared<vector_byte_buffer_impl>(data)};
+}
+
+mutable_byte_buffer vector_byte_buffer::create(std::span<uint8_t const> data) {
+  return mutable_byte_buffer{std::make_shared<vector_byte_buffer_impl>(data)};
+}
+
+mutable_byte_buffer vector_byte_buffer::create(std::vector<uint8_t>&& data) {
+  return mutable_byte_buffer{
+      std::make_shared<vector_byte_buffer_impl>(std::move(data))};
+}
+
 } // namespace dwarfs
diff --git a/src/writer/filesystem_writer.cpp b/src/writer/filesystem_writer.cpp
index 6382f34b..60047825 100644
--- a/src/writer/filesystem_writer.cpp
+++ b/src/writer/filesystem_writer.cpp
@@ -48,7 +48,6 @@
 
 #include <dwarfs/internal/fs_section.h>
 #include <dwarfs/internal/worker_group.h>
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/filesystem_writer_detail.h>
 #include <dwarfs/writer/internal/multi_queue_block_merger.h>
 #include <dwarfs/writer/internal/progress.h>
@@ -122,8 +121,7 @@ class compression_progress : public progress::context {
 class fsblock {
  public:
   fsblock(section_type type, block_compressor const& bc,
-          std::shared_ptr<block_data>&& data,
-          std::shared_ptr<compression_progress> pctx,
+          shared_byte_buffer data, std::shared_ptr<compression_progress> pctx,
           folly::Function<void(size_t)> set_block_cb = nullptr);
 
   fsblock(section_type type, compression_type compression,
@@ -198,12 +196,12 @@ class fsblock_merger_policy {
 class raw_fsblock : public fsblock::impl {
  public:
   raw_fsblock(section_type type, block_compressor const& bc,
-              std::shared_ptr<block_data>&& data,
+              shared_byte_buffer data,
               std::shared_ptr<compression_progress> pctx,
               folly::Function<void(size_t)> set_block_cb)
       : type_{type}
       , bc_{bc}
-      , uncompressed_size_{data->size()}
+      , uncompressed_size_{data.size()}
       , data_{std::move(data)}
       , comp_type_{bc_.type()}
       , pctx_{std::move(pctx)}
@@ -215,30 +213,30 @@ class raw_fsblock : public fsblock::impl {
     std::promise<void> prom;
     future_ = prom.get_future();
 
-    wg.add_job([this, prom = std::move(prom),
-                meta = std::move(meta)]() mutable {
-      try {
-        std::shared_ptr<block_data> tmp;
+    wg.add_job(
+        [this, prom = std::move(prom), meta = std::move(meta)]() mutable {
+          try {
+            shared_byte_buffer tmp;
 
-        if (meta) {
-          tmp = std::make_shared<block_data>(bc_.compress(data_->vec(), *meta));
-        } else {
-          tmp = std::make_shared<block_data>(bc_.compress(data_->vec()));
-        }
+            if (meta) {
+              tmp = bc_.compress(data_, *meta);
+            } else {
+              tmp = bc_.compress(data_);
+            }
 
-        pctx_->bytes_in += data_->vec().size();
-        pctx_->bytes_out += tmp->vec().size();
+            pctx_->bytes_in += data_.size();
+            pctx_->bytes_out += tmp.size();
 
-        {
-          std::lock_guard lock(mx_);
-          data_.swap(tmp);
-        }
-      } catch (bad_compression_ratio_error const&) {
-        comp_type_ = compression_type::NONE;
-      }
+            {
+              std::lock_guard lock(mx_);
+              data_.swap(tmp);
+            }
+          } catch (bad_compression_ratio_error const&) {
+            comp_type_ = compression_type::NONE;
+          }
 
-      prom.set_value();
-    });
+          prom.set_value();
+        });
   }
 
   void wait_until_compressed() override { future_.wait(); }
@@ -249,13 +247,13 @@ class raw_fsblock : public fsblock::impl {
 
   std::string description() const override { return bc_.describe(); }
 
-  std::span<uint8_t const> data() const override { return data_->vec(); }
+  std::span<uint8_t const> data() const override { return data_.span(); }
 
   size_t uncompressed_size() const override { return uncompressed_size_; }
 
   size_t size() const override {
     std::lock_guard lock(mx_);
-    return data_->size();
+    return data_.size();
   }
 
   void set_block_no(uint32_t number) override {
@@ -291,7 +289,7 @@ class raw_fsblock : public fsblock::impl {
   block_compressor const& bc_;
   size_t const uncompressed_size_;
   mutable std::recursive_mutex mx_;
-  std::shared_ptr<block_data> data_;
+  shared_byte_buffer data_;
   std::future<void> future_;
   std::optional<uint32_t> number_;
   std::optional<section_header_v2> mutable header_;
@@ -382,7 +380,7 @@ class rewritten_fsblock : public fsblock::impl {
     wg.add_job(
         [this, prom = std::move(prom), meta = std::move(meta)]() mutable {
           try {
-            std::vector<uint8_t> block;
+            shared_byte_buffer block;
 
             {
               // TODO: we don't have to do this for uncompressed blocks
@@ -398,9 +396,9 @@ class rewritten_fsblock : public fsblock::impl {
 
               try {
                 if (meta) {
-                  block = bc_.compress(buffer.span(), *meta);
+                  block = bc_.compress(buffer.share(), *meta);
                 } else {
-                  block = bc_.compress(buffer.span());
+                  block = bc_.compress(buffer.share());
                 }
               } catch (bad_compression_ratio_error const&) {
                 comp_type_ = compression_type::NONE;
@@ -411,7 +409,7 @@ class rewritten_fsblock : public fsblock::impl {
 
             {
               std::lock_guard lock(mx_);
-              block_data_.swap(block);
+              block_data_.emplace(std::move(block));
             }
 
             prom.set_value();
@@ -429,13 +427,18 @@ class rewritten_fsblock : public fsblock::impl {
 
   std::string description() const override { return bc_.describe(); }
 
-  std::span<uint8_t const> data() const override { return block_data_; }
+  std::span<uint8_t const> data() const override {
+    std::lock_guard lock(mx_);
+    return block_data_.value().span();
+  }
 
   size_t uncompressed_size() const override { return data_.size(); }
 
   size_t size() const override {
     std::lock_guard lock(mx_);
-    return block_data_.size();
+    // TODO: this should not be called when block_data_ is not set, figure
+    //       out who calls this
+    return block_data_.has_value() ? block_data_->size() : 0;
   }
 
   void set_block_no(uint32_t number) override {
@@ -467,7 +470,7 @@ class rewritten_fsblock : public fsblock::impl {
   block_compressor const& bc_;
   mutable std::recursive_mutex mx_;
   std::span<uint8_t const> data_;
-  std::vector<uint8_t> block_data_;
+  std::optional<shared_byte_buffer> block_data_;
   std::future<void> future_;
   std::optional<uint32_t> number_;
   std::optional<section_header_v2> mutable header_;
@@ -477,7 +480,7 @@ class rewritten_fsblock : public fsblock::impl {
 };
 
 fsblock::fsblock(section_type type, block_compressor const& bc,
-                 std::shared_ptr<block_data>&& data,
+                 shared_byte_buffer data,
                  std::shared_ptr<compression_progress> pctx,
                  folly::Function<void(size_t)> set_block_cb)
     : impl_(std::make_unique<raw_fsblock>(type, bc, std::move(data),
@@ -573,13 +576,13 @@ class filesystem_writer_ final : public filesystem_writer_detail {
                  size_t max_active_slots) override;
   void configure_rewrite(size_t filesystem_size, size_t block_count) override;
   void copy_header(std::span<uint8_t const> header) override;
-  void write_block(fragment_category cat, std::shared_ptr<block_data>&& data,
+  void write_block(fragment_category cat, shared_byte_buffer data,
                    physical_block_cb_type physical_block_cb,
                    std::optional<std::string> meta) override;
   void finish_category(fragment_category cat) override;
-  void write_metadata_v2_schema(std::shared_ptr<block_data>&& data) override;
-  void write_metadata_v2(std::shared_ptr<block_data>&& data) override;
-  void write_history(std::shared_ptr<block_data>&& data) override;
+  void write_metadata_v2_schema(shared_byte_buffer data) override;
+  void write_metadata_v2(shared_byte_buffer data) override;
+  void write_history(shared_byte_buffer data) override;
   void check_block_compression(
       compression_type compression, std::span<uint8_t const> data,
       std::optional<fragment_category::value_type> cat) override;
@@ -600,12 +603,11 @@ class filesystem_writer_ final : public filesystem_writer_detail {
   block_compressor const&
   compressor_for_category(fragment_category::value_type cat) const;
   void
-  write_block_impl(fragment_category cat, std::shared_ptr<block_data>&& data,
+  write_block_impl(fragment_category cat, shared_byte_buffer data,
                    block_compressor const& bc, std::optional<std::string> meta,
                    physical_block_cb_type physical_block_cb);
   void on_block_merged(block_holder_type holder);
-  void
-  write_section_impl(section_type type, std::shared_ptr<block_data>&& data);
+  void write_section_impl(section_type type, shared_byte_buffer data);
   void write(fsblock const& fsb);
   void write(char const* data, size_t size);
   template <typename T>
@@ -779,9 +781,8 @@ filesystem_writer_<LoggerPolicy>::compressor_for_category(
 
 template <typename LoggerPolicy>
 void filesystem_writer_<LoggerPolicy>::write_block_impl(
-    fragment_category cat, std::shared_ptr<block_data>&& data,
-    block_compressor const& bc, std::optional<std::string> meta,
-    physical_block_cb_type physical_block_cb) {
+    fragment_category cat, shared_byte_buffer data, block_compressor const& bc,
+    std::optional<std::string> meta, physical_block_cb_type physical_block_cb) {
   if (!merger_) {
     DWARFS_THROW(runtime_error, "filesystem_writer not configured");
   }
@@ -840,7 +841,7 @@ void filesystem_writer_<LoggerPolicy>::finish_category(fragment_category cat) {
 
 template <typename LoggerPolicy>
 void filesystem_writer_<LoggerPolicy>::write_section_impl(
-    section_type type, std::shared_ptr<block_data>&& data) {
+    section_type type, shared_byte_buffer data) {
   auto& bc = get_compressor(type, std::nullopt);
 
   uint32_t number;
@@ -1071,7 +1072,7 @@ void filesystem_writer_<LoggerPolicy>::copy_header(
 
 template <typename LoggerPolicy>
 void filesystem_writer_<LoggerPolicy>::write_block(
-    fragment_category cat, std::shared_ptr<block_data>&& data,
+    fragment_category cat, shared_byte_buffer data,
     physical_block_cb_type physical_block_cb, std::optional<std::string> meta) {
   write_block_impl(cat, std::move(data), compressor_for_category(cat.value()),
                    std::move(meta), std::move(physical_block_cb));
@@ -1079,19 +1080,18 @@ void filesystem_writer_<LoggerPolicy>::write_block(
 
 template <typename LoggerPolicy>
 void filesystem_writer_<LoggerPolicy>::write_metadata_v2_schema(
-    std::shared_ptr<block_data>&& data) {
+    shared_byte_buffer data) {
   write_section_impl(section_type::METADATA_V2_SCHEMA, std::move(data));
 }
 
 template <typename LoggerPolicy>
 void filesystem_writer_<LoggerPolicy>::write_metadata_v2(
-    std::shared_ptr<block_data>&& data) {
+    shared_byte_buffer data) {
   write_section_impl(section_type::METADATA_V2, std::move(data));
 }
 
 template <typename LoggerPolicy>
-void filesystem_writer_<LoggerPolicy>::write_history(
-    std::shared_ptr<block_data>&& data) {
+void filesystem_writer_<LoggerPolicy>::write_history(shared_byte_buffer data) {
   write_section_impl(section_type::HISTORY, std::move(data));
 }
 
diff --git a/src/writer/internal/metadata_freezer.cpp b/src/writer/internal/metadata_freezer.cpp
index 924a39fa..e8fd2958 100644
--- a/src/writer/internal/metadata_freezer.cpp
+++ b/src/writer/internal/metadata_freezer.cpp
@@ -22,6 +22,8 @@
 #include <thrift/lib/cpp2/frozen/FrozenUtil.h>
 #include <thrift/lib/cpp2/protocol/Serializer.h>
 
+#include <dwarfs/vector_byte_buffer.h>
+
 #include <dwarfs/writer/internal/metadata_freezer.h>
 
 #include <dwarfs/gen-cpp2/metadata_layouts.h>
@@ -34,8 +36,7 @@ namespace dwarfs::writer::internal {
 namespace {
 
 template <class T>
-std::pair<std::vector<uint8_t>, std::vector<uint8_t>>
-freeze_to_buffer(T const& x) {
+std::pair<shared_byte_buffer, shared_byte_buffer> freeze_to_buffer(T const& x) {
   using namespace ::apache::thrift::frozen;
 
   Layout<T> layout;
@@ -44,24 +45,22 @@ freeze_to_buffer(T const& x) {
   std::string schema;
   serializeRootLayout(layout, schema);
 
-  size_t schema_size = schema.size();
-  auto schema_begin = reinterpret_cast<uint8_t const*>(schema.data());
-  std::vector<uint8_t> schema_buffer(schema_begin, schema_begin + schema_size);
+  auto schema_buffer = vector_byte_buffer::create(schema);
 
-  std::vector<uint8_t> data_buffer;
-  data_buffer.resize(content_size, 0);
+  auto data_buffer = vector_byte_buffer::create(content_size);
 
   folly::MutableByteRange content_range(data_buffer.data(), data_buffer.size());
   ByteRangeFreezer::freeze(layout, x, content_range);
 
   data_buffer.resize(data_buffer.size() - content_range.size());
+  data_buffer.shrink_to_fit();
 
-  return {schema_buffer, data_buffer};
+  return {schema_buffer.share(), data_buffer.share()};
 }
 
 } // namespace
 
-std::pair<std::vector<uint8_t>, std::vector<uint8_t>>
+std::pair<shared_byte_buffer, shared_byte_buffer>
 metadata_freezer::freeze(thrift::metadata::metadata const& data) {
   return freeze_to_buffer(data);
 }
diff --git a/src/writer/scanner.cpp b/src/writer/scanner.cpp
index c940483b..27e1ce24 100644
--- a/src/writer/scanner.cpp
+++ b/src/writer/scanner.cpp
@@ -63,7 +63,6 @@
 #include <dwarfs/internal/features.h>
 #include <dwarfs/internal/string_table.h>
 #include <dwarfs/internal/worker_group.h>
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/block_manager.h>
 #include <dwarfs/writer/internal/entry.h>
 #include <dwarfs/writer/internal/file_scanner.h>
@@ -1107,13 +1106,13 @@ void scanner_<LoggerPolicy>::scan(
 
   LOG_VERBOSE << "uncompressed metadata size: " << size_with_unit(data.size());
 
-  fsw.write_metadata_v2_schema(std::make_shared<block_data>(std::move(schema)));
-  fsw.write_metadata_v2(std::make_shared<block_data>(std::move(data)));
+  fsw.write_metadata_v2_schema(schema);
+  fsw.write_metadata_v2(data);
 
   if (options_.enable_history) {
     history hist(options_.history);
     hist.append(options_.command_line_arguments);
-    fsw.write_history(std::make_shared<block_data>(hist.serialize()));
+    fsw.write_history(hist.serialize());
   }
 
   LOG_INFO << "waiting for compression to finish...";
diff --git a/src/writer/segmenter.cpp b/src/writer/segmenter.cpp
index 3b047ccb..9f6fc6d8 100644
--- a/src/writer/segmenter.cpp
+++ b/src/writer/segmenter.cpp
@@ -46,10 +46,10 @@
 #include <dwarfs/error.h>
 #include <dwarfs/logger.h>
 #include <dwarfs/util.h>
+#include <dwarfs/vector_byte_buffer.h>
 #include <dwarfs/writer/segmenter.h>
 #include <dwarfs/writer/writer_progress.h>
 
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/block_manager.h>
 #include <dwarfs/writer/internal/chunkable.h>
 #include <dwarfs/writer/internal/cyclic_hash.h>
@@ -576,23 +576,23 @@ class active_block : private GranularityPolicy {
       , filter_(bloom_filter_size)
       , repseqmap_{repseqmap}
       , repeating_collisions_{repcoll}
-      , data_{std::make_shared<block_data>()} {
+      , data_{vector_byte_buffer::create()} {
     DWARFS_CHECK((window_step & window_step_mask_) == 0,
                  "window step size not a power of two");
-    data_->reserve(this->frames_to_bytes(capacity_in_frames_));
+    data_.reserve(this->frames_to_bytes(capacity_in_frames_));
   }
 
   DWARFS_FORCE_INLINE size_t num() const { return num_; }
 
   DWARFS_FORCE_INLINE size_t size_in_frames() const {
-    return this->bytes_to_frames(data_->size());
+    return this->bytes_to_frames(data_.size());
   }
 
   DWARFS_FORCE_INLINE bool full() const {
     return size_in_frames() == capacity_in_frames_;
   }
 
-  DWARFS_FORCE_INLINE std::shared_ptr<block_data> data() const { return data_; }
+  DWARFS_FORCE_INLINE mutable_byte_buffer data() const { return data_; }
 
   DWARFS_FORCE_INLINE void
   append_bytes(std::span<uint8_t const> data, bloom_filter& global_filter);
@@ -637,7 +637,7 @@ class active_block : private GranularityPolicy {
   fast_multimap<hash_t, offset_t, num_inline_offsets> offsets_;
   repeating_sequence_map_type const& repseqmap_;
   repeating_collisions_map_type& repeating_collisions_;
-  std::shared_ptr<block_data> data_;
+  mutable_byte_buffer data_;
 };
 
 class segmenter_progress : public progress::context {
@@ -841,7 +841,7 @@ DWARFS_FORCE_INLINE bool
 active_block<LoggerPolicy, GranularityPolicy>::is_existing_repeating_sequence(
     hash_t hashval, size_t offset) {
   if (auto it = repseqmap_.find(hashval); it != repseqmap_.end()) [[unlikely]] {
-    auto& raw = data_->vec();
+    auto& raw = data_.raw_vector();
     auto winbeg = raw.begin() + frames_to_bytes(offset);
     auto winend = winbeg + frames_to_bytes(window_size_);
     auto byte = *winbeg;
@@ -881,7 +881,7 @@ active_block<LoggerPolicy, GranularityPolicy>::append_bytes(
       granular_span_adapter<uint8_t const, GranularityPolicy>>(data);
 
   auto v = this->template create<
-      granular_vector_adapter<uint8_t, GranularityPolicy>>(data_->vec());
+      granular_vector_adapter<uint8_t, GranularityPolicy>>(data_.raw_vector());
 
   auto offset = v.size();
 
@@ -920,7 +920,7 @@ void segment_match<LoggerPolicy, GranularityPolicy>::verify_and_extend(
     size_t pos, size_t len, size_t begin, size_t end) {
   auto v = this->template create<
       granular_vector_adapter<uint8_t, GranularityPolicy>>(
-      block_->data()->vec());
+      block_->data().raw_vector());
 
   // First, check if the regions actually match
   if (v.compare(offset_, data.subspan(pos, len)) == 0) {
@@ -1034,7 +1034,7 @@ DWARFS_FORCE_INLINE void
 segmenter_<LoggerPolicy, SegmentingPolicy>::block_ready() {
   auto& block = blocks_.back();
   block.finalize(stats_);
-  block_ready_(block.data(), block.num());
+  block_ready_(block.data().share(), block.num());
   ++prog_.block_count;
 }
 
diff --git a/test/flac_compressor_test.cpp b/test/flac_compressor_test.cpp
index c3cf5ff4..0ee4afc4 100644
--- a/test/flac_compressor_test.cpp
+++ b/test/flac_compressor_test.cpp
@@ -60,7 +60,7 @@ std::vector<T> multiplex(std::vector<std::vector<T>> const& in) {
 }
 
 template <typename T = int32_t>
-std::vector<uint8_t>
+shared_byte_buffer
 make_test_data(int channels, int samples, int bytes, int bits,
                pcm_sample_endianness end, pcm_sample_signedness sig,
                pcm_sample_padding pad) {
@@ -70,10 +70,11 @@ make_test_data(int channels, int samples, int bytes, int bits,
         make_sine<T>(bits, samples, 3.1 * ((599 * (c + 1)) % 256)));
   }
   auto muxed = multiplex(data);
-  std::vector<uint8_t> out(bytes * channels * samples);
+  auto out = vector_byte_buffer::create();
+  out.resize(bytes * channels * samples);
   pcm_sample_transformer<T> xfm(end, sig, pad, bytes, bits);
-  xfm.pack(out, muxed);
-  return out;
+  xfm.pack(out.span(), muxed);
+  return out.share();
 }
 
 struct data_params {
@@ -151,7 +152,7 @@ TEST(flac_compressor, basic) {
   EXPECT_LT(compressed.size(), data.size() / 2);
 
   auto decompressed =
-      block_decompressor::decompress(compression_type::FLAC, compressed);
+      block_decompressor::decompress(compression_type::FLAC, compressed.span());
 
   EXPECT_EQ(data, decompressed);
 }
@@ -184,7 +185,7 @@ TEST_P(flac_param, combinations) {
   EXPECT_LT(compressed.size(), data.size() / 2);
 
   auto decompressed =
-      block_decompressor::decompress(compression_type::FLAC, compressed);
+      block_decompressor::decompress(compression_type::FLAC, compressed.span());
 
   EXPECT_EQ(data, decompressed);
 }
diff --git a/test/ricepp_compressor_test.cpp b/test/ricepp_compressor_test.cpp
index fad9ac5c..12ceb18f 100644
--- a/test/ricepp_compressor_test.cpp
+++ b/test/ricepp_compressor_test.cpp
@@ -37,6 +37,7 @@
 #include <range/v3/view/concat.hpp>
 
 #include <dwarfs/block_compressor.h>
+#include <dwarfs/vector_byte_buffer.h>
 
 using namespace dwarfs;
 
@@ -61,8 +62,7 @@ generate_random_data(std::mt19937_64& rng, size_t count,
 }
 
 template <std::unsigned_integral ValueType>
-std::vector<uint8_t>
-make_test_data(int components, int pixels, int unused_lsb) {
+shared_byte_buffer make_test_data(int components, int pixels, int unused_lsb) {
   std::mt19937_64 rng(42);
   std::uniform_int_distribution<ValueType> any_value(
       0, std::numeric_limits<ValueType>::max());
@@ -93,11 +93,11 @@ make_test_data(int components, int pixels, int unused_lsb) {
     }
   }
 
-  std::vector<uint8_t> out;
+  auto out = vector_byte_buffer::create();
   out.resize(tmp.size() * sizeof(ValueType));
   std::memcpy(out.data(), tmp.data(), out.size());
 
-  return out;
+  return out.share();
 }
 
 struct data_params {
@@ -151,8 +151,8 @@ TEST_P(ricepp_param, combinations) {
 
   EXPECT_LT(compressed.size(), 7 * data.size() / 10);
 
-  auto decompressed =
-      block_decompressor::decompress(compression_type::RICEPP, compressed);
+  auto decompressed = block_decompressor::decompress(compression_type::RICEPP,
+                                                     compressed.span());
 
   ASSERT_EQ(data.size(), decompressed.size());
   EXPECT_EQ(data, decompressed);
diff --git a/test/segmenter_benchmark.cpp b/test/segmenter_benchmark.cpp
index 2ff07170..5779024a 100644
--- a/test/segmenter_benchmark.cpp
+++ b/test/segmenter_benchmark.cpp
@@ -28,7 +28,6 @@
 #include <dwarfs/writer/segmenter.h>
 #include <dwarfs/writer/writer_progress.h>
 
-#include <dwarfs/writer/internal/block_data.h>
 #include <dwarfs/writer/internal/block_manager.h>
 #include <dwarfs/writer/internal/chunkable.h>
 
@@ -145,13 +144,12 @@ void run_segmenter_test(unsigned iters, unsigned granularity,
     dwarfs::writer::writer_progress prog;
     auto blkmgr = std::make_shared<dwarfs::writer::internal::block_manager>();
 
-    std::vector<std::shared_ptr<dwarfs::writer::internal::block_data>> written;
+    std::vector<dwarfs::shared_byte_buffer> written;
 
     dwarfs::writer::segmenter seg(
         lgr, prog, blkmgr, cfg, cc, total_size,
-        [&written,
-         blkmgr](std::shared_ptr<dwarfs::writer::internal::block_data> blk,
-                 auto logical_block_num) {
+        [&written, blkmgr](dwarfs::shared_byte_buffer blk,
+                           auto logical_block_num) {
           auto physical_block_num = written.size();
           written.push_back(blk);
           blkmgr->set_written_block(logical_block_num, physical_block_num, 0);
@@ -167,7 +165,7 @@ void run_segmenter_test(unsigned iters, unsigned granularity,
     size_t segmented [[maybe_unused]]{0};
 
     for (auto const& blk : written) {
-      segmented += blk->size();
+      segmented += blk.size();
     }
 
     // std::cerr << total_size << " -> " << segmented << fmt::format("