Snap for 10453563 from 84ad640730426064d81e25ce95a73f1c2ff59c64 to mainline-uwb-releaseaml_uwb_341513070aml_uwb_341511050aml_uwb_341310300aml_uwb_341310030aml_uwb_341111010aml_uwb_341011000android14-mainline-uwb-release

Change-Id: I1c56e37cadad9df7a274c735ba874d19f5f6873e

28 files changed, 1728 insertions, 254 deletions

diff --git a/.cargo_vcs_info.json b/.cargo_vcs_info.json
index 64b65d2..b4c4c4e 100644
--- a/.cargo_vcs_info.json
+++ b/.cargo_vcs_info.json
@@ -1,5 +1,6 @@
 {
   "git": {
-    "sha1": "ebc61e5af14cd9b436ba880cf19e849b05a04c29"
-  }
-}
+    "sha1": "21ed3328364716fa30a4bf7502c913bbf0a90f45"
+  },
+  "path_in_vcs": ""
+}
\ No newline at end of file
diff --git a/.github/workflows/ci.yml b/.github/workflows/ci.yml
index 00a4414..a4f7b1d 100644
--- a/.github/workflows/ci.yml
+++ b/.github/workflows/ci.yml
@@ -11,7 +11,7 @@ on:
 env:
   RUSTFLAGS: -Dwarnings
   RUST_BACKTRACE: 1
-  nightly: nightly-2021-04-13
+  nightly: nightly-2022-11-12
 
 defaults:
   run:
@@ -23,7 +23,7 @@ jobs:
     name: rustfmt
     runs-on: ubuntu-latest
     steps:
-    - uses: actions/checkout@v2
+    - uses: actions/checkout@v3
     - name: Install Rust
       run: rustup update stable && rustup default stable
     - name: Check formatting
@@ -35,7 +35,7 @@ jobs:
   #   name: clippy
   #   runs-on: ubuntu-latest
   #   steps:
-  #   - uses: actions/checkout@v2
+  #   - uses: actions/checkout@v3
   #   - name: Apply clippy lints
   #     run: cargo clippy --all-features
 
@@ -48,7 +48,7 @@ jobs:
     name: minrust
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         run: rustup update 1.39.0 && rustup default 1.39.0
       - name: Check
@@ -65,7 +65,7 @@ jobs:
           - windows-latest
     runs-on: ${{ matrix.os }}
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         # --no-self-update is necessary because the windows environment cannot self-update rustup.exe.
         run: rustup update stable --no-self-update && rustup default stable
@@ -77,7 +77,7 @@ jobs:
     name: nightly
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         run: rustup update $nightly && rustup default $nightly
       - name: Test
@@ -96,7 +96,7 @@ jobs:
           - wasm32-unknown-unknown
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         run: rustup update stable && rustup default stable
       - name: cross build --target ${{ matrix.target }}
@@ -116,7 +116,7 @@ jobs:
     name: tsan
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         run: rustup update $nightly && rustup default $nightly
       - name: Install rust-src
@@ -127,7 +127,7 @@ jobs:
     name: miri
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Miri
         run: ci/miri.sh
 
@@ -136,7 +136,7 @@ jobs:
     name: loom
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
         run: rustup update $nightly && rustup default $nightly
       - name: Loom tests
@@ -155,9 +155,9 @@ jobs:
       - loom
     runs-on: ubuntu-latest
     steps:
-      - uses: actions/checkout@v2
+      - uses: actions/checkout@v3
       - name: Install Rust
-        run: rustup update stable && rustup default stable
+        run: rustup update $nightly && rustup default $nightly
       - name: Build documentation
         run: cargo doc --no-deps --all-features
         env:
diff --git a/Android.bp b/Android.bp
index 1909ec1..05bfcd5 100644
--- a/Android.bp
+++ b/Android.bp
@@ -22,7 +22,7 @@ rust_defaults {
     name: "bytes_test_defaults",
     crate_name: "bytes",
     cargo_env_compat: true,
-    cargo_pkg_version: "1.1.0",
+    cargo_pkg_version: "1.4.0",
     test_suites: ["general-tests"],
     auto_gen_config: true,
     edition: "2018",
@@ -153,7 +153,7 @@ rust_library {
     host_supported: true,
     crate_name: "bytes",
     cargo_env_compat: true,
-    cargo_pkg_version: "1.1.0",
+    cargo_pkg_version: "1.4.0",
     srcs: ["src/lib.rs"],
     edition: "2018",
     features: [
@@ -166,12 +166,13 @@ rust_library {
     ],
     apex_available: [
         "//apex_available:platform",
-        "com.android.bluetooth",
+        "com.android.btservices",
         "com.android.compos",
         "com.android.resolv",
         "com.android.uwb",
         "com.android.virt",
     ],
+    product_available: true,
     vendor_available: true,
     min_sdk_version: "29",
 }
diff --git a/CHANGELOG.md b/CHANGELOG.md
index 636d36b..a1bad4a 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -1,3 +1,66 @@
+# 1.4.0 (January 31, 2023)
+
+### Added
+
+- Make `IntoIter` constructor public (#581)
+
+### Fixed
+
+- Avoid large reallocations when freezing `BytesMut` (#592)
+
+### Documented
+
+- Document which functions require `std` (#591)
+- Fix duplicate "the the" typos (#585)
+
+# 1.3.0 (November 20, 2022)
+
+### Added
+
+- Rename and expose `BytesMut::spare_capacity_mut` (#572)
+- Implement native-endian get and put functions for `Buf` and `BufMut` (#576)
+
+### Fixed
+
+- Don't have important data in unused capacity when calling reserve (#563)
+
+### Documented
+
+- `Bytes::new` etc should return `Self` not `Bytes` (#568)
+
+# 1.2.1 (July 30, 2022)
+
+### Fixed
+
+- Fix unbounded memory growth when using `reserve` (#560)
+
+# 1.2.0 (July 19, 2022)
+
+### Added
+
+- Add `BytesMut::zeroed` (#517)
+- Implement `Extend<Bytes>` for `BytesMut` (#527)
+- Add conversion from `BytesMut` to `Vec<u8>` (#543, #554)
+- Add conversion from `Bytes` to `Vec<u8>` (#547)
+- Add `UninitSlice::as_uninit_slice_mut()` (#548)
+- Add const to `Bytes::{len,is_empty}` (#514)
+
+### Changed
+
+- Reuse vector in `BytesMut::reserve` (#539, #544)
+
+### Fixed
+
+- Make miri happy (#515, #523, #542, #545, #553)
+- Make tsan happy (#541)
+- Fix `remaining_mut()` on chain (#488)
+- Fix amortized asymptotics of `BytesMut` (#555)
+
+### Documented
+
+- Redraw layout diagram with box drawing characters (#539)
+- Clarify `BytesMut::unsplit` docs (#535)
+
 # 1.1.0 (August 25, 2021)
 
 ### Added
diff --git a/Cargo.toml b/Cargo.toml
index 065f0f1..d0acb34 100644
--- a/Cargo.toml
+++ b/Cargo.toml
@@ -3,36 +3,52 @@
 # When uploading crates to the registry Cargo will automatically
 # "normalize" Cargo.toml files for maximal compatibility
 # with all versions of Cargo and also rewrite `path` dependencies
-# to registry (e.g., crates.io) dependencies
+# to registry (e.g., crates.io) dependencies.
 #
-# If you believe there's an error in this file please file an
-# issue against the rust-lang/cargo repository. If you're
-# editing this file be aware that the upstream Cargo.toml
-# will likely look very different (and much more reasonable)
+# If you are reading this file be aware that the original Cargo.toml
+# will likely look very different (and much more reasonable).
+# See Cargo.toml.orig for the original contents.
 
 [package]
 edition = "2018"
 name = "bytes"
-version = "1.1.0"
-authors = ["Carl Lerche <me@carllerche.com>", "Sean McArthur <sean@seanmonstar.com>"]
+version = "1.4.0"
+authors = [
+    "Carl Lerche <me@carllerche.com>",
+    "Sean McArthur <sean@seanmonstar.com>",
+]
 description = "Types and traits for working with bytes"
 readme = "README.md"
-keywords = ["buffers", "zero-copy", "io"]
-categories = ["network-programming", "data-structures"]
+keywords = [
+    "buffers",
+    "zero-copy",
+    "io",
+]
+categories = [
+    "network-programming",
+    "data-structures",
+]
 license = "MIT"
 repository = "https://github.com/tokio-rs/bytes"
+
 [package.metadata.docs.rs]
-rustdoc-args = ["--cfg", "docsrs"]
+rustdoc-args = [
+    "--cfg",
+    "docsrs",
+]
+
 [dependencies.serde]
 version = "1.0.60"
 features = ["alloc"]
 optional = true
 default-features = false
+
 [dev-dependencies.serde_test]
 version = "1.0"
 
 [features]
 default = ["std"]
 std = []
+
 [target."cfg(loom)".dev-dependencies.loom]
 version = "0.5"
diff --git a/Cargo.toml.orig b/Cargo.toml.orig
index 2b7e32b..4a96ec1 100644
--- a/Cargo.toml.orig
+++ b/Cargo.toml.orig
@@ -4,7 +4,7 @@ name = "bytes"
 # When releasing to crates.io:
 # - Update CHANGELOG.md.
 # - Create "v1.x.y" git tag.
-version = "1.1.0"
+version = "1.4.0"
 license = "MIT"
 authors = [
     "Carl Lerche <me@carllerche.com>",
diff --git a/METADATA b/METADATA
index 6723e08..8f9837f 100644
--- a/METADATA
+++ b/METADATA
@@ -1,3 +1,7 @@
+# This project was upgraded with external_updater.
+# Usage: tools/external_updater/updater.sh update rust/crates/bytes
+# For more info, check https://cs.android.com/android/platform/superproject/+/master:tools/external_updater/README.md
+
 name: "bytes"
 description: "Types and traits for working with bytes"
 third_party {
@@ -7,13 +11,13 @@ third_party {
   }
   url {
     type: ARCHIVE
-    value: "https://static.crates.io/crates/bytes/bytes-1.1.0.crate"
+    value: "https://static.crates.io/crates/bytes/bytes-1.4.0.crate"
   }
-  version: "1.1.0"
+  version: "1.4.0"
   license_type: NOTICE
   last_upgrade_date {
-    year: 2021
-    month: 9
-    day: 22
+    year: 2023
+    month: 2
+    day: 15
   }
 }
diff --git a/README.md b/README.md
index 468485d..be46642 100644
--- a/README.md
+++ b/README.md
@@ -36,6 +36,15 @@ Serde support is optional and disabled by default. To enable use the feature `se
 bytes = { version = "1", features = ["serde"] }
 ```
 
+## Building documentation
+
+When building the `bytes` documentation the `docsrs` option should be used, otherwise
+feature gates will not be shown. This requires a nightly toolchain:
+
+```
+RUSTDOCFLAGS="--cfg docsrs" cargo +nightly doc
+```
+
 ## License
 
 This project is licensed under the [MIT license](LICENSE).
diff --git a/TEST_MAPPING b/TEST_MAPPING
index f298357..0e8ed79 100644
--- a/TEST_MAPPING
+++ b/TEST_MAPPING
@@ -2,6 +2,9 @@
 {
   "imports": [
     {
+      "path": "external/rust/crates/async-stream"
+    },
+    {
       "path": "external/rust/crates/futures-util"
     },
     {
@@ -12,66 +15,66 @@
     },
     {
       "path": "external/rust/crates/tokio-test"
-    }
-  ],
-  "presubmit": [
+    },
     {
-      "name": "authfs_device_test_src_lib"
+      "path": "external/uwb/src"
     },
     {
-      "name": "bytes_test_tests_test_buf"
+      "path": "packages/modules/DnsResolver"
     },
     {
-      "name": "bytes_test_tests_test_buf_mut"
+      "path": "packages/modules/Virtualization/apkdmverity"
     },
     {
-      "name": "bytes_test_tests_test_bytes"
+      "path": "packages/modules/Virtualization/authfs"
     },
     {
-      "name": "bytes_test_tests_test_bytes_odd_alloc"
+      "path": "packages/modules/Virtualization/avmd"
     },
     {
-      "name": "bytes_test_tests_test_bytes_vec_alloc"
+      "path": "packages/modules/Virtualization/libs/apkverify"
     },
     {
-      "name": "bytes_test_tests_test_chain"
+      "path": "packages/modules/Virtualization/microdroid_manager"
     },
     {
-      "name": "bytes_test_tests_test_debug"
+      "path": "packages/modules/Virtualization/virtualizationmanager"
+    }
+  ],
+  "presubmit": [
+    {
+      "name": "bytes_test_tests_test_buf"
     },
     {
-      "name": "bytes_test_tests_test_iter"
+      "name": "bytes_test_tests_test_buf_mut"
     },
     {
-      "name": "bytes_test_tests_test_reader"
+      "name": "bytes_test_tests_test_bytes"
     },
     {
-      "name": "bytes_test_tests_test_take"
+      "name": "bytes_test_tests_test_bytes_odd_alloc"
     },
     {
-      "name": "doh_unit_test"
+      "name": "bytes_test_tests_test_bytes_vec_alloc"
     },
     {
-      "name": "libapkverify.integration_test"
+      "name": "bytes_test_tests_test_chain"
     },
     {
-      "name": "libapkverify.test"
+      "name": "bytes_test_tests_test_debug"
     },
     {
-      "name": "microdroid_manager_test"
+      "name": "bytes_test_tests_test_iter"
     },
     {
-      "name": "rustBinderTest"
+      "name": "bytes_test_tests_test_reader"
     },
     {
-      "name": "virtualizationservice_device_test"
+      "name": "bytes_test_tests_test_take"
     }
   ],
   "presubmit-rust": [
     {
-      "name": "authfs_device_test_src_lib"
-    },
-    {
       "name": "bytes_test_tests_test_buf"
     },
     {
@@ -100,24 +103,6 @@
     },
     {
       "name": "bytes_test_tests_test_take"
-    },
-    {
-      "name": "doh_unit_test"
-    },
-    {
-      "name": "libapkverify.integration_test"
-    },
-    {
-      "name": "libapkverify.test"
-    },
-    {
-      "name": "microdroid_manager_test"
-    },
-    {
-      "name": "rustBinderTest"
-    },
-    {
-      "name": "virtualizationservice_device_test"
     }
   ]
 }
diff --git a/benches/buf.rs b/benches/buf.rs
index 6dc8516..616d187 100644
--- a/benches/buf.rs
+++ b/benches/buf.rs
@@ -46,7 +46,7 @@ impl TestBuf {
 }
 impl Buf for TestBuf {
     fn remaining(&self) -> usize {
-        return self.buf.len() - self.pos;
+        self.buf.len() - self.pos
     }
     fn advance(&mut self, cnt: usize) {
         self.pos += cnt;
diff --git a/benches/bytes.rs b/benches/bytes.rs
index c5b8412..61d1e83 100644
--- a/benches/bytes.rs
+++ b/benches/bytes.rs
@@ -88,6 +88,7 @@ fn from_long_slice(b: &mut Bencher) {
 #[bench]
 fn slice_empty(b: &mut Bencher) {
     b.iter(|| {
+        // `clone` is to convert to ARC
         let b = Bytes::from(vec![17; 1024]).clone();
         for i in 0..1000 {
             test::black_box(b.slice(i % 100..i % 100));
diff --git a/cargo2android.json b/cargo2android.json
index 3f01463..a77ea80 100644
--- a/cargo2android.json
+++ b/cargo2android.json
@@ -1,7 +1,7 @@
 {
   "apex-available": [
     "//apex_available:platform",
-    "com.android.bluetooth",
+    "com.android.btservices",
     "com.android.compos",
     "com.android.resolv",
     "com.android.uwb",
diff --git a/ci/miri.sh b/ci/miri.sh
index 88d2b6a..0158756 100755
--- a/ci/miri.sh
+++ b/ci/miri.sh
@@ -1,11 +1,11 @@
 #!/bin/bash
 set -e
 
-MIRI_NIGHTLY=nightly-$(curl -s https://rust-lang.github.io/rustup-components-history/x86_64-unknown-linux-gnu/miri)
-echo "Installing latest nightly with Miri: $MIRI_NIGHTLY"
-rustup set profile minimal
-rustup default "$MIRI_NIGHTLY"
-rustup component add miri
+rustup toolchain install nightly --component miri
+rustup override set nightly
+cargo miri setup
+
+export MIRIFLAGS="-Zmiri-strict-provenance"
 
 cargo miri test
 cargo miri test --target mips64-unknown-linux-gnuabi64
diff --git a/clippy.toml b/clippy.toml
new file mode 100644
index 0000000..53095b1
--- /dev/null
+++ b/clippy.toml
@@ -0,0 +1 @@
+msrv = "1.39"
diff --git a/src/buf/buf_impl.rs b/src/buf/buf_impl.rs
index a33c8a4..366cfc9 100644
--- a/src/buf/buf_impl.rs
+++ b/src/buf/buf_impl.rs
@@ -160,6 +160,7 @@ pub trait Buf {
     ///
     /// [`writev`]: http://man7.org/linux/man-pages/man2/readv.2.html
     #[cfg(feature = "std")]
+    #[cfg_attr(docsrs, doc(cfg(feature = "std")))]
     fn chunks_vectored<'a>(&'a self, dst: &mut [IoSlice<'a>]) -> usize {
         if dst.is_empty() {
             return 0;
@@ -354,6 +355,29 @@ pub trait Buf {
         buf_get_impl!(self, u16::from_le_bytes);
     }
 
+    /// Gets an unsigned 16 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09 hello",
+    ///     false => b"\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809, buf.get_u16_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u16_ne(&mut self) -> u16 {
+        buf_get_impl!(self, u16::from_ne_bytes);
+    }
+
     /// Gets a signed 16 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 2.
@@ -394,6 +418,29 @@ pub trait Buf {
         buf_get_impl!(self, i16::from_le_bytes);
     }
 
+    /// Gets a signed 16 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09 hello",
+    ///     false => b"\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809, buf.get_i16_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i16_ne(&mut self) -> i16 {
+        buf_get_impl!(self, i16::from_ne_bytes);
+    }
+
     /// Gets an unsigned 32 bit integer from `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -434,6 +481,29 @@ pub trait Buf {
         buf_get_impl!(self, u32::from_le_bytes);
     }
 
+    /// Gets an unsigned 32 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09\xA0\xA1 hello",
+    ///     false => b"\xA1\xA0\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809A0A1, buf.get_u32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u32_ne(&mut self) -> u32 {
+        buf_get_impl!(self, u32::from_ne_bytes);
+    }
+
     /// Gets a signed 32 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -474,6 +544,29 @@ pub trait Buf {
         buf_get_impl!(self, i32::from_le_bytes);
     }
 
+    /// Gets a signed 32 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x08\x09\xA0\xA1 hello",
+    ///     false => b"\xA1\xA0\x09\x08 hello",
+    /// };
+    /// assert_eq!(0x0809A0A1, buf.get_i32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i32_ne(&mut self) -> i32 {
+        buf_get_impl!(self, i32::from_ne_bytes);
+    }
+
     /// Gets an unsigned 64 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -514,6 +607,29 @@ pub trait Buf {
         buf_get_impl!(self, u64::from_le_bytes);
     }
 
+    /// Gets an unsigned 64 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08 hello",
+    ///     false => b"\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x0102030405060708, buf.get_u64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u64_ne(&mut self) -> u64 {
+        buf_get_impl!(self, u64::from_ne_bytes);
+    }
+
     /// Gets a signed 64 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -554,6 +670,29 @@ pub trait Buf {
         buf_get_impl!(self, i64::from_le_bytes);
     }
 
+    /// Gets a signed 64 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08 hello",
+    ///     false => b"\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x0102030405060708, buf.get_i64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i64_ne(&mut self) -> i64 {
+        buf_get_impl!(self, i64::from_ne_bytes);
+    }
+
     /// Gets an unsigned 128 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -594,6 +733,29 @@ pub trait Buf {
         buf_get_impl!(self, u128::from_le_bytes);
     }
 
+    /// Gets an unsigned 128 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16 hello",
+    ///     false => b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x01020304050607080910111213141516, buf.get_u128_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_u128_ne(&mut self) -> u128 {
+        buf_get_impl!(self, u128::from_ne_bytes);
+    }
+
     /// Gets a signed 128 bit integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -634,6 +796,29 @@ pub trait Buf {
         buf_get_impl!(self, i128::from_le_bytes);
     }
 
+    /// Gets a signed 128 bit integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16 hello",
+    ///     false => b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x01020304050607080910111213141516, buf.get_i128_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_i128_ne(&mut self) -> i128 {
+        buf_get_impl!(self, i128::from_ne_bytes);
+    }
+
     /// Gets an unsigned n-byte integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -674,6 +859,33 @@ pub trait Buf {
         buf_get_impl!(le => self, u64, nbytes);
     }
 
+    /// Gets an unsigned n-byte integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03 hello",
+    ///     false => b"\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x010203, buf.get_uint_ne(3));
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_uint_ne(&mut self, nbytes: usize) -> u64 {
+        if cfg!(target_endian = "big") {
+            self.get_uint(nbytes)
+        } else {
+            self.get_uint_le(nbytes)
+        }
+    }
+
     /// Gets a signed n-byte integer from `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -714,6 +926,33 @@ pub trait Buf {
         buf_get_impl!(le => self, i64, nbytes);
     }
 
+    /// Gets a signed n-byte integer from `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x01\x02\x03 hello",
+    ///     false => b"\x03\x02\x01 hello",
+    /// };
+    /// assert_eq!(0x010203, buf.get_int_ne(3));
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_int_ne(&mut self, nbytes: usize) -> i64 {
+        if cfg!(target_endian = "big") {
+            self.get_int(nbytes)
+        } else {
+            self.get_int_le(nbytes)
+        }
+    }
+
     /// Gets an IEEE754 single-precision (4 bytes) floating point number from
     /// `self` in big-endian byte order.
     ///
@@ -756,6 +995,30 @@ pub trait Buf {
         f32::from_bits(Self::get_u32_le(self))
     }
 
+    /// Gets an IEEE754 single-precision (4 bytes) floating point number from
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x3F\x99\x99\x9A hello",
+    ///     false => b"\x9A\x99\x99\x3F hello",
+    /// };
+    /// assert_eq!(1.2f32, buf.get_f32_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_f32_ne(&mut self) -> f32 {
+        f32::from_bits(Self::get_u32_ne(self))
+    }
+
     /// Gets an IEEE754 double-precision (8 bytes) floating point number from
     /// `self` in big-endian byte order.
     ///
@@ -798,6 +1061,30 @@ pub trait Buf {
         f64::from_bits(Self::get_u64_le(self))
     }
 
+    /// Gets an IEEE754 double-precision (8 bytes) floating point number from
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::Buf;
+    ///
+    /// let mut buf: &[u8] = match cfg!(target_endian = "big") {
+    ///     true => b"\x3F\xF3\x33\x33\x33\x33\x33\x33 hello",
+    ///     false => b"\x33\x33\x33\x33\x33\x33\xF3\x3F hello",
+    /// };
+    /// assert_eq!(1.2f64, buf.get_f64_ne());
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining data in `self`.
+    fn get_f64_ne(&mut self) -> f64 {
+        f64::from_bits(Self::get_u64_ne(self))
+    }
+
     /// Consumes `len` bytes inside self and returns new instance of `Bytes`
     /// with this data.
     ///
@@ -897,6 +1184,7 @@ pub trait Buf {
     /// assert_eq!(&dst[..11], &b"hello world"[..]);
     /// ```
     #[cfg(feature = "std")]
+    #[cfg_attr(docsrs, doc(cfg(feature = "std")))]
     fn reader(self) -> Reader<Self>
     where
         Self: Sized,
@@ -948,6 +1236,10 @@ macro_rules! deref_forward_buf {
             (**self).get_u16_le()
         }
 
+        fn get_u16_ne(&mut self) -> u16 {
+            (**self).get_u16_ne()
+        }
+
         fn get_i16(&mut self) -> i16 {
             (**self).get_i16()
         }
@@ -956,6 +1248,10 @@ macro_rules! deref_forward_buf {
             (**self).get_i16_le()
         }
 
+        fn get_i16_ne(&mut self) -> i16 {
+            (**self).get_i16_ne()
+        }
+
         fn get_u32(&mut self) -> u32 {
             (**self).get_u32()
         }
@@ -964,6 +1260,10 @@ macro_rules! deref_forward_buf {
             (**self).get_u32_le()
         }
 
+        fn get_u32_ne(&mut self) -> u32 {
+            (**self).get_u32_ne()
+        }
+
         fn get_i32(&mut self) -> i32 {
             (**self).get_i32()
         }
@@ -972,6 +1272,10 @@ macro_rules! deref_forward_buf {
             (**self).get_i32_le()
         }
 
+        fn get_i32_ne(&mut self) -> i32 {
+            (**self).get_i32_ne()
+        }
+
         fn get_u64(&mut self) -> u64 {
             (**self).get_u64()
         }
@@ -980,6 +1284,10 @@ macro_rules! deref_forward_buf {
             (**self).get_u64_le()
         }
 
+        fn get_u64_ne(&mut self) -> u64 {
+            (**self).get_u64_ne()
+        }
+
         fn get_i64(&mut self) -> i64 {
             (**self).get_i64()
         }
@@ -988,6 +1296,10 @@ macro_rules! deref_forward_buf {
             (**self).get_i64_le()
         }
 
+        fn get_i64_ne(&mut self) -> i64 {
+            (**self).get_i64_ne()
+        }
+
         fn get_uint(&mut self, nbytes: usize) -> u64 {
             (**self).get_uint(nbytes)
         }
@@ -996,6 +1308,10 @@ macro_rules! deref_forward_buf {
             (**self).get_uint_le(nbytes)
         }
 
+        fn get_uint_ne(&mut self, nbytes: usize) -> u64 {
+            (**self).get_uint_ne(nbytes)
+        }
+
         fn get_int(&mut self, nbytes: usize) -> i64 {
             (**self).get_int(nbytes)
         }
@@ -1004,6 +1320,10 @@ macro_rules! deref_forward_buf {
             (**self).get_int_le(nbytes)
         }
 
+        fn get_int_ne(&mut self, nbytes: usize) -> i64 {
+            (**self).get_int_ne(nbytes)
+        }
+
         fn copy_to_bytes(&mut self, len: usize) -> crate::Bytes {
             (**self).copy_to_bytes(len)
         }
diff --git a/src/buf/buf_mut.rs b/src/buf/buf_mut.rs
index 4c2bd2c..685fcc7 100644
--- a/src/buf/buf_mut.rs
+++ b/src/buf/buf_mut.rs
@@ -56,6 +56,10 @@ pub unsafe trait BufMut {
     /// Implementations of `remaining_mut` should ensure that the return value
     /// does not change unless a call is made to `advance_mut` or any other
     /// function that is documented to change the `BufMut`'s current position.
+    ///
+    /// # Note
+    ///
+    /// `remaining_mut` may return value smaller than actual available space.
     fn remaining_mut(&self) -> usize;
 
     /// Advance the internal cursor of the BufMut
@@ -382,6 +386,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes an unsigned 16 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_u16_ne(0x0809);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x08\x09");
+    /// } else {
+    ///     assert_eq!(buf, b"\x09\x08");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_u16_ne(&mut self, n: u16) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes a signed 16 bit integer to `self` in big-endian byte order.
     ///
     /// The current position is advanced by 2.
@@ -426,6 +456,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes a signed 16 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 2.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_i16_ne(0x0809);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x08\x09");
+    /// } else {
+    ///     assert_eq!(buf, b"\x09\x08");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_i16_ne(&mut self, n: i16) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes an unsigned 32 bit integer to `self` in big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -470,6 +526,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes an unsigned 32 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_u32_ne(0x0809A0A1);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x08\x09\xA0\xA1");
+    /// } else {
+    ///     assert_eq!(buf, b"\xA1\xA0\x09\x08");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_u32_ne(&mut self, n: u32) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes a signed 32 bit integer to `self` in big-endian byte order.
     ///
     /// The current position is advanced by 4.
@@ -514,6 +596,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes a signed 32 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_i32_ne(0x0809A0A1);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x08\x09\xA0\xA1");
+    /// } else {
+    ///     assert_eq!(buf, b"\xA1\xA0\x09\x08");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_i32_ne(&mut self, n: i32) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes an unsigned 64 bit integer to `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -558,6 +666,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes an unsigned 64 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_u64_ne(0x0102030405060708);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03\x04\x05\x06\x07\x08");
+    /// } else {
+    ///     assert_eq!(buf, b"\x08\x07\x06\x05\x04\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_u64_ne(&mut self, n: u64) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes a signed 64 bit integer to `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 8.
@@ -602,6 +736,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes a signed 64 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_i64_ne(0x0102030405060708);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03\x04\x05\x06\x07\x08");
+    /// } else {
+    ///     assert_eq!(buf, b"\x08\x07\x06\x05\x04\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_i64_ne(&mut self, n: i64) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes an unsigned 128 bit integer to `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -646,6 +806,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes an unsigned 128 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_u128_ne(0x01020304050607080910111213141516);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16");
+    /// } else {
+    ///     assert_eq!(buf, b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_u128_ne(&mut self, n: u128) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes a signed 128 bit integer to `self` in the big-endian byte order.
     ///
     /// The current position is advanced by 16.
@@ -690,6 +876,32 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes())
     }
 
+    /// Writes a signed 128 bit integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 16.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_i128_ne(0x01020304050607080910111213141516);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03\x04\x05\x06\x07\x08\x09\x10\x11\x12\x13\x14\x15\x16");
+    /// } else {
+    ///     assert_eq!(buf, b"\x16\x15\x14\x13\x12\x11\x10\x09\x08\x07\x06\x05\x04\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_i128_ne(&mut self, n: i128) {
+        self.put_slice(&n.to_ne_bytes())
+    }
+
     /// Writes an unsigned n-byte integer to `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -734,6 +946,36 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes()[0..nbytes]);
     }
 
+    /// Writes an unsigned n-byte integer to `self` in the native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_uint_ne(0x010203, 3);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03");
+    /// } else {
+    ///     assert_eq!(buf, b"\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_uint_ne(&mut self, n: u64, nbytes: usize) {
+        if cfg!(target_endian = "big") {
+            self.put_uint(n, nbytes)
+        } else {
+            self.put_uint_le(n, nbytes)
+        }
+    }
+
     /// Writes low `nbytes` of a signed integer to `self` in big-endian byte order.
     ///
     /// The current position is advanced by `nbytes`.
@@ -778,6 +1020,36 @@ pub unsafe trait BufMut {
         self.put_slice(&n.to_le_bytes()[0..nbytes]);
     }
 
+    /// Writes low `nbytes` of a signed integer to `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by `nbytes`.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_int_ne(0x010203, 3);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x01\x02\x03");
+    /// } else {
+    ///     assert_eq!(buf, b"\x03\x02\x01");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self` or if `nbytes` is greater than 8.
+    fn put_int_ne(&mut self, n: i64, nbytes: usize) {
+        if cfg!(target_endian = "big") {
+            self.put_int(n, nbytes)
+        } else {
+            self.put_int_le(n, nbytes)
+        }
+    }
+
     /// Writes  an IEEE754 single-precision (4 bytes) floating point number to
     /// `self` in big-endian byte order.
     ///
@@ -824,6 +1096,33 @@ pub unsafe trait BufMut {
         self.put_u32_le(n.to_bits());
     }
 
+    /// Writes an IEEE754 single-precision (4 bytes) floating point number to
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 4.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_f32_ne(1.2f32);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x3F\x99\x99\x9A");
+    /// } else {
+    ///     assert_eq!(buf, b"\x9A\x99\x99\x3F");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_f32_ne(&mut self, n: f32) {
+        self.put_u32_ne(n.to_bits());
+    }
+
     /// Writes  an IEEE754 double-precision (8 bytes) floating point number to
     /// `self` in big-endian byte order.
     ///
@@ -870,6 +1169,33 @@ pub unsafe trait BufMut {
         self.put_u64_le(n.to_bits());
     }
 
+    /// Writes  an IEEE754 double-precision (8 bytes) floating point number to
+    /// `self` in native-endian byte order.
+    ///
+    /// The current position is advanced by 8.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut buf = vec![];
+    /// buf.put_f64_ne(1.2f64);
+    /// if cfg!(target_endian = "big") {
+    ///     assert_eq!(buf, b"\x3F\xF3\x33\x33\x33\x33\x33\x33");
+    /// } else {
+    ///     assert_eq!(buf, b"\x33\x33\x33\x33\x33\x33\xF3\x3F");
+    /// }
+    /// ```
+    ///
+    /// # Panics
+    ///
+    /// This function panics if there is not enough remaining capacity in
+    /// `self`.
+    fn put_f64_ne(&mut self, n: f64) {
+        self.put_u64_ne(n.to_bits());
+    }
+
     /// Creates an adaptor which can write at most `limit` bytes to `self`.
     ///
     /// # Examples
@@ -913,6 +1239,7 @@ pub unsafe trait BufMut {
     /// assert_eq!(*buf, b"hello world"[..]);
     /// ```
     #[cfg(feature = "std")]
+    #[cfg_attr(docsrs, doc(cfg(feature = "std")))]
     fn writer(self) -> Writer<Self>
     where
         Self: Sized,
@@ -982,6 +1309,10 @@ macro_rules! deref_forward_bufmut {
             (**self).put_u16_le(n)
         }
 
+        fn put_u16_ne(&mut self, n: u16) {
+            (**self).put_u16_ne(n)
+        }
+
         fn put_i16(&mut self, n: i16) {
             (**self).put_i16(n)
         }
@@ -990,6 +1321,10 @@ macro_rules! deref_forward_bufmut {
             (**self).put_i16_le(n)
         }
 
+        fn put_i16_ne(&mut self, n: i16) {
+            (**self).put_i16_ne(n)
+        }
+
         fn put_u32(&mut self, n: u32) {
             (**self).put_u32(n)
         }
@@ -998,6 +1333,10 @@ macro_rules! deref_forward_bufmut {
             (**self).put_u32_le(n)
         }
 
+        fn put_u32_ne(&mut self, n: u32) {
+            (**self).put_u32_ne(n)
+        }
+
         fn put_i32(&mut self, n: i32) {
             (**self).put_i32(n)
         }
@@ -1006,6 +1345,10 @@ macro_rules! deref_forward_bufmut {
             (**self).put_i32_le(n)
         }
 
+        fn put_i32_ne(&mut self, n: i32) {
+            (**self).put_i32_ne(n)
+        }
+
         fn put_u64(&mut self, n: u64) {
             (**self).put_u64(n)
         }
@@ -1014,6 +1357,10 @@ macro_rules! deref_forward_bufmut {
             (**self).put_u64_le(n)
         }
 
+        fn put_u64_ne(&mut self, n: u64) {
+            (**self).put_u64_ne(n)
+        }
+
         fn put_i64(&mut self, n: i64) {
             (**self).put_i64(n)
         }
@@ -1021,6 +1368,10 @@ macro_rules! deref_forward_bufmut {
         fn put_i64_le(&mut self, n: i64) {
             (**self).put_i64_le(n)
         }
+
+        fn put_i64_ne(&mut self, n: i64) {
+            (**self).put_i64_ne(n)
+        }
     };
 }
 
diff --git a/src/buf/chain.rs b/src/buf/chain.rs
index 9ce5f23..78979a1 100644
--- a/src/buf/chain.rs
+++ b/src/buf/chain.rs
@@ -198,8 +198,7 @@ where
     fn remaining_mut(&self) -> usize {
         self.a
             .remaining_mut()
-            .checked_add(self.b.remaining_mut())
-            .unwrap()
+            .saturating_add(self.b.remaining_mut())
     }
 
     fn chunk_mut(&mut self) -> &mut UninitSlice {
diff --git a/src/buf/iter.rs b/src/buf/iter.rs
index 8914a40..c694e3d 100644
--- a/src/buf/iter.rs
+++ b/src/buf/iter.rs
@@ -2,8 +2,6 @@ use crate::Buf;
 
 /// Iterator over the bytes contained by the buffer.
 ///
-/// This struct is created by the [`iter`] method on [`Buf`].
-///
 /// # Examples
 ///
 /// Basic usage:
@@ -43,7 +41,7 @@ impl<T> IntoIter<T> {
     /// assert_eq!(iter.next(), Some(b'c'));
     /// assert_eq!(iter.next(), None);
     /// ```
-    pub(crate) fn new(inner: T) -> IntoIter<T> {
+    pub fn new(inner: T) -> IntoIter<T> {
         IntoIter { inner }
     }
 
diff --git a/src/buf/uninit_slice.rs b/src/buf/uninit_slice.rs
index fb67c0a..3161a14 100644
--- a/src/buf/uninit_slice.rs
+++ b/src/buf/uninit_slice.rs
@@ -22,6 +22,10 @@ use core::ops::{
 pub struct UninitSlice([MaybeUninit<u8>]);
 
 impl UninitSlice {
+    pub(crate) fn from_slice(slice: &mut [MaybeUninit<u8>]) -> &mut UninitSlice {
+        unsafe { &mut *(slice as *mut [MaybeUninit<u8>] as *mut UninitSlice) }
+    }
+
     /// Create a `&mut UninitSlice` from a pointer and a length.
     ///
     /// # Safety
@@ -44,7 +48,7 @@ impl UninitSlice {
     pub unsafe fn from_raw_parts_mut<'a>(ptr: *mut u8, len: usize) -> &'a mut UninitSlice {
         let maybe_init: &mut [MaybeUninit<u8>] =
             core::slice::from_raw_parts_mut(ptr as *mut _, len);
-        &mut *(maybe_init as *mut [MaybeUninit<u8>] as *mut UninitSlice)
+        Self::from_slice(maybe_init)
     }
 
     /// Write a single byte at the specified offset.
@@ -124,6 +128,32 @@ impl UninitSlice {
         self.0.as_mut_ptr() as *mut _
     }
 
+    /// Return a `&mut [MaybeUninit<u8>]` to this slice's buffer.
+    ///
+    /// # Safety
+    ///
+    /// The caller **must not** read from the referenced memory and **must not** write
+    /// **uninitialized** bytes to the slice either. This is because `BufMut` implementation
+    /// that created the `UninitSlice` knows which parts are initialized. Writing uninitalized
+    /// bytes to the slice may cause the `BufMut` to read those bytes and trigger undefined
+    /// behavior.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BufMut;
+    ///
+    /// let mut data = [0, 1, 2];
+    /// let mut slice = &mut data[..];
+    /// unsafe {
+    ///     let uninit_slice = BufMut::chunk_mut(&mut slice).as_uninit_slice_mut();
+    /// };
+    /// ```
+    #[inline]
+    pub unsafe fn as_uninit_slice_mut<'a>(&'a mut self) -> &'a mut [MaybeUninit<u8>] {
+        &mut *(self as *mut _ as *mut [MaybeUninit<u8>])
+    }
+
     /// Returns the number of bytes in the slice.
     ///
     /// # Examples
diff --git a/src/bytes.rs b/src/bytes.rs
index d0be0d2..0404a72 100644
--- a/src/bytes.rs
+++ b/src/bytes.rs
@@ -2,12 +2,18 @@ use core::iter::FromIterator;
 use core::ops::{Deref, RangeBounds};
 use core::{cmp, fmt, hash, mem, ptr, slice, usize};
 
-use alloc::{borrow::Borrow, boxed::Box, string::String, vec::Vec};
+use alloc::{
+    alloc::{dealloc, Layout},
+    borrow::Borrow,
+    boxed::Box,
+    string::String,
+    vec::Vec,
+};
 
 use crate::buf::IntoIter;
 #[allow(unused)]
 use crate::loom::sync::atomic::AtomicMut;
-use crate::loom::sync::atomic::{self, AtomicPtr, AtomicUsize, Ordering};
+use crate::loom::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 use crate::Buf;
 
 /// A cheaply cloneable and sliceable chunk of contiguous memory.
@@ -26,7 +32,7 @@ use crate::Buf;
 /// All `Bytes` implementations must fulfill the following requirements:
 /// - They are cheaply cloneable and thereby shareable between an unlimited amount
 ///   of components, for example by modifying a reference count.
-/// - Instances can be sliced to refer to a subset of the the original buffer.
+/// - Instances can be sliced to refer to a subset of the original buffer.
 ///
 /// ```
 /// use bytes::Bytes;
@@ -55,7 +61,7 @@ use crate::Buf;
 /// # Sharing
 ///
 /// `Bytes` contains a vtable, which allows implementations of `Bytes` to define
-/// how sharing/cloneing is implemented in detail.
+/// how sharing/cloning is implemented in detail.
 /// When `Bytes::clone()` is called, `Bytes` will call the vtable function for
 /// cloning the backing storage in order to share it behind between multiple
 /// `Bytes` instances.
@@ -65,7 +71,7 @@ use crate::Buf;
 ///
 /// For `Bytes` implementations which point to a reference counted shared storage
 /// (e.g. an `Arc<[u8]>`), sharing will be implemented by increasing the
-/// the reference count.
+/// reference count.
 ///
 /// Due to this mechanism, multiple `Bytes` instances may point to the same
 /// shared memory region.
@@ -78,18 +84,18 @@ use crate::Buf;
 ///
 /// ```text
 ///
-///    Arc ptrs                   +---------+
-///    ________________________ / | Bytes 2 |
-///   /                           +---------+
-///  /          +-----------+     |         |
-/// |_________/ |  Bytes 1  |     |         |
-/// |           +-----------+     |         |
+///    Arc ptrs                   ┌─────────┐
+///    ________________________ / │ Bytes 2 │
+///   /                           └─────────┘
+///  /          ┌───────────┐     |         |
+/// |_________/ │  Bytes 1  │     |         |
+/// |           └───────────┘     |         |
 /// |           |           | ___/ data     | tail
 /// |      data |      tail |/              |
 /// v           v           v               v
-/// +-----+---------------------------------+-----+
-/// | Arc |     |           |               |     |
-/// +-----+---------------------------------+-----+
+/// ┌─────┬─────┬───────────┬───────────────┬─────┐
+/// │ Arc │     │           │               │     │
+/// └─────┴─────┴───────────┴───────────────┴─────┘
 /// ```
 pub struct Bytes {
     ptr: *const u8,
@@ -103,6 +109,10 @@ pub(crate) struct Vtable {
     /// fn(data, ptr, len)
     pub clone: unsafe fn(&AtomicPtr<()>, *const u8, usize) -> Bytes,
     /// fn(data, ptr, len)
+    ///
+    /// takes `Bytes` to value
+    pub to_vec: unsafe fn(&AtomicPtr<()>, *const u8, usize) -> Vec<u8>,
+    /// fn(data, ptr, len)
     pub drop: unsafe fn(&mut AtomicPtr<()>, *const u8, usize),
 }
 
@@ -121,7 +131,7 @@ impl Bytes {
     /// ```
     #[inline]
     #[cfg(not(all(loom, test)))]
-    pub const fn new() -> Bytes {
+    pub const fn new() -> Self {
         // Make it a named const to work around
         // "unsizing casts are not allowed in const fn"
         const EMPTY: &[u8] = &[];
@@ -129,7 +139,7 @@ impl Bytes {
     }
 
     #[cfg(all(loom, test))]
-    pub fn new() -> Bytes {
+    pub fn new() -> Self {
         const EMPTY: &[u8] = &[];
         Bytes::from_static(EMPTY)
     }
@@ -149,7 +159,7 @@ impl Bytes {
     /// ```
     #[inline]
     #[cfg(not(all(loom, test)))]
-    pub const fn from_static(bytes: &'static [u8]) -> Bytes {
+    pub const fn from_static(bytes: &'static [u8]) -> Self {
         Bytes {
             ptr: bytes.as_ptr(),
             len: bytes.len(),
@@ -159,7 +169,7 @@ impl Bytes {
     }
 
     #[cfg(all(loom, test))]
-    pub fn from_static(bytes: &'static [u8]) -> Bytes {
+    pub fn from_static(bytes: &'static [u8]) -> Self {
         Bytes {
             ptr: bytes.as_ptr(),
             len: bytes.len(),
@@ -179,7 +189,7 @@ impl Bytes {
     /// assert_eq!(b.len(), 5);
     /// ```
     #[inline]
-    pub fn len(&self) -> usize {
+    pub const fn len(&self) -> usize {
         self.len
     }
 
@@ -194,7 +204,7 @@ impl Bytes {
     /// assert!(b.is_empty());
     /// ```
     #[inline]
-    pub fn is_empty(&self) -> bool {
+    pub const fn is_empty(&self) -> bool {
         self.len == 0
     }
 
@@ -225,7 +235,7 @@ impl Bytes {
     ///
     /// Requires that `begin <= end` and `end <= self.len()`, otherwise slicing
     /// will panic.
-    pub fn slice(&self, range: impl RangeBounds<usize>) -> Bytes {
+    pub fn slice(&self, range: impl RangeBounds<usize>) -> Self {
         use core::ops::Bound;
 
         let len = self.len();
@@ -262,7 +272,7 @@ impl Bytes {
         let mut ret = self.clone();
 
         ret.len = end - begin;
-        ret.ptr = unsafe { ret.ptr.offset(begin as isize) };
+        ret.ptr = unsafe { ret.ptr.add(begin) };
 
         ret
     }
@@ -292,7 +302,7 @@ impl Bytes {
     ///
     /// Requires that the given `sub` slice is in fact contained within the
     /// `Bytes` buffer; otherwise this function will panic.
-    pub fn slice_ref(&self, subset: &[u8]) -> Bytes {
+    pub fn slice_ref(&self, subset: &[u8]) -> Self {
         // Empty slice and empty Bytes may have their pointers reset
         // so explicitly allow empty slice to be a subslice of any slice.
         if subset.is_empty() {
@@ -308,15 +318,15 @@ impl Bytes {
         assert!(
             sub_p >= bytes_p,
             "subset pointer ({:p}) is smaller than self pointer ({:p})",
-            sub_p as *const u8,
-            bytes_p as *const u8,
+            subset.as_ptr(),
+            self.as_ptr(),
         );
         assert!(
             sub_p + sub_len <= bytes_p + bytes_len,
             "subset is out of bounds: self = ({:p}, {}), subset = ({:p}, {})",
-            bytes_p as *const u8,
+            self.as_ptr(),
             bytes_len,
-            sub_p as *const u8,
+            subset.as_ptr(),
             sub_len,
         );
 
@@ -349,7 +359,7 @@ impl Bytes {
     ///
     /// Panics if `at > len`.
     #[must_use = "consider Bytes::truncate if you don't need the other half"]
-    pub fn split_off(&mut self, at: usize) -> Bytes {
+    pub fn split_off(&mut self, at: usize) -> Self {
         assert!(
             at <= self.len(),
             "split_off out of bounds: {:?} <= {:?}",
@@ -398,7 +408,7 @@ impl Bytes {
     ///
     /// Panics if `at > len`.
     #[must_use = "consider Bytes::advance if you don't need the other half"]
-    pub fn split_to(&mut self, at: usize) -> Bytes {
+    pub fn split_to(&mut self, at: usize) -> Self {
         assert!(
             at <= self.len(),
             "split_to out of bounds: {:?} <= {:?}",
@@ -501,7 +511,7 @@ impl Bytes {
         // should already be asserted, but debug assert for tests
         debug_assert!(self.len >= by, "internal: inc_start out of bounds");
         self.len -= by;
-        self.ptr = self.ptr.offset(by as isize);
+        self.ptr = self.ptr.add(by);
     }
 }
 
@@ -604,7 +614,7 @@ impl<'a> IntoIterator for &'a Bytes {
     type IntoIter = core::slice::Iter<'a, u8>;
 
     fn into_iter(self) -> Self::IntoIter {
-        self.as_slice().into_iter()
+        self.as_slice().iter()
     }
 }
 
@@ -686,7 +696,7 @@ impl PartialOrd<Bytes> for str {
 
 impl PartialEq<Vec<u8>> for Bytes {
     fn eq(&self, other: &Vec<u8>) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -710,7 +720,7 @@ impl PartialOrd<Bytes> for Vec<u8> {
 
 impl PartialEq<String> for Bytes {
     fn eq(&self, other: &String) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -797,8 +807,36 @@ impl From<&'static str> for Bytes {
 
 impl From<Vec<u8>> for Bytes {
     fn from(vec: Vec<u8>) -> Bytes {
-        let slice = vec.into_boxed_slice();
-        slice.into()
+        let mut vec = vec;
+        let ptr = vec.as_mut_ptr();
+        let len = vec.len();
+        let cap = vec.capacity();
+
+        // Avoid an extra allocation if possible.
+        if len == cap {
+            return Bytes::from(vec.into_boxed_slice());
+        }
+
+        let shared = Box::new(Shared {
+            buf: ptr,
+            cap,
+            ref_cnt: AtomicUsize::new(1),
+        });
+        mem::forget(vec);
+
+        let shared = Box::into_raw(shared);
+        // The pointer should be aligned, so this assert should
+        // always succeed.
+        debug_assert!(
+            0 == (shared as usize & KIND_MASK),
+            "internal: Box<Shared> should have an aligned pointer",
+        );
+        Bytes {
+            ptr,
+            len,
+            data: AtomicPtr::new(shared as _),
+            vtable: &SHARED_VTABLE,
+        }
     }
 }
 
@@ -815,18 +853,18 @@ impl From<Box<[u8]>> for Bytes {
         let ptr = Box::into_raw(slice) as *mut u8;
 
         if ptr as usize & 0x1 == 0 {
-            let data = ptr as usize | KIND_VEC;
+            let data = ptr_map(ptr, |addr| addr | KIND_VEC);
             Bytes {
                 ptr,
                 len,
-                data: AtomicPtr::new(data as *mut _),
+                data: AtomicPtr::new(data.cast()),
                 vtable: &PROMOTABLE_EVEN_VTABLE,
             }
         } else {
             Bytes {
                 ptr,
                 len,
-                data: AtomicPtr::new(ptr as *mut _),
+                data: AtomicPtr::new(ptr.cast()),
                 vtable: &PROMOTABLE_ODD_VTABLE,
             }
         }
@@ -839,6 +877,13 @@ impl From<String> for Bytes {
     }
 }
 
+impl From<Bytes> for Vec<u8> {
+    fn from(bytes: Bytes) -> Vec<u8> {
+        let bytes = mem::ManuallyDrop::new(bytes);
+        unsafe { (bytes.vtable.to_vec)(&bytes.data, bytes.ptr, bytes.len) }
+    }
+}
+
 // ===== impl Vtable =====
 
 impl fmt::Debug for Vtable {
@@ -854,6 +899,7 @@ impl fmt::Debug for Vtable {
 
 const STATIC_VTABLE: Vtable = Vtable {
     clone: static_clone,
+    to_vec: static_to_vec,
     drop: static_drop,
 };
 
@@ -862,6 +908,11 @@ unsafe fn static_clone(_: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Bytes {
     Bytes::from_static(slice)
 }
 
+unsafe fn static_to_vec(_: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    let slice = slice::from_raw_parts(ptr, len);
+    slice.to_vec()
+}
+
 unsafe fn static_drop(_: &mut AtomicPtr<()>, _: *const u8, _: usize) {
     // nothing to drop for &'static [u8]
 }
@@ -870,11 +921,13 @@ unsafe fn static_drop(_: &mut AtomicPtr<()>, _: *const u8, _: usize) {
 
 static PROMOTABLE_EVEN_VTABLE: Vtable = Vtable {
     clone: promotable_even_clone,
+    to_vec: promotable_even_to_vec,
     drop: promotable_even_drop,
 };
 
 static PROMOTABLE_ODD_VTABLE: Vtable = Vtable {
     clone: promotable_odd_clone,
+    to_vec: promotable_odd_to_vec,
     drop: promotable_odd_drop,
 };
 
@@ -883,25 +936,57 @@ unsafe fn promotable_even_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize
     let kind = shared as usize & KIND_MASK;
 
     if kind == KIND_ARC {
-        shallow_clone_arc(shared as _, ptr, len)
+        shallow_clone_arc(shared.cast(), ptr, len)
     } else {
         debug_assert_eq!(kind, KIND_VEC);
-        let buf = (shared as usize & !KIND_MASK) as *mut u8;
+        let buf = ptr_map(shared.cast(), |addr| addr & !KIND_MASK);
         shallow_clone_vec(data, shared, buf, ptr, len)
     }
 }
 
+unsafe fn promotable_to_vec(
+    data: &AtomicPtr<()>,
+    ptr: *const u8,
+    len: usize,
+    f: fn(*mut ()) -> *mut u8,
+) -> Vec<u8> {
+    let shared = data.load(Ordering::Acquire);
+    let kind = shared as usize & KIND_MASK;
+
+    if kind == KIND_ARC {
+        shared_to_vec_impl(shared.cast(), ptr, len)
+    } else {
+        // If Bytes holds a Vec, then the offset must be 0.
+        debug_assert_eq!(kind, KIND_VEC);
+
+        let buf = f(shared);
+
+        let cap = (ptr as usize - buf as usize) + len;
+
+        // Copy back buffer
+        ptr::copy(ptr, buf, len);
+
+        Vec::from_raw_parts(buf, len, cap)
+    }
+}
+
+unsafe fn promotable_even_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    promotable_to_vec(data, ptr, len, |shared| {
+        ptr_map(shared.cast(), |addr| addr & !KIND_MASK)
+    })
+}
+
 unsafe fn promotable_even_drop(data: &mut AtomicPtr<()>, ptr: *const u8, len: usize) {
     data.with_mut(|shared| {
         let shared = *shared;
         let kind = shared as usize & KIND_MASK;
 
         if kind == KIND_ARC {
-            release_shared(shared as *mut Shared);
+            release_shared(shared.cast());
         } else {
             debug_assert_eq!(kind, KIND_VEC);
-            let buf = (shared as usize & !KIND_MASK) as *mut u8;
-            drop(rebuild_boxed_slice(buf, ptr, len));
+            let buf = ptr_map(shared.cast(), |addr| addr & !KIND_MASK);
+            free_boxed_slice(buf, ptr, len);
         }
     });
 }
@@ -914,38 +999,49 @@ unsafe fn promotable_odd_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize)
         shallow_clone_arc(shared as _, ptr, len)
     } else {
         debug_assert_eq!(kind, KIND_VEC);
-        shallow_clone_vec(data, shared, shared as *mut u8, ptr, len)
+        shallow_clone_vec(data, shared, shared.cast(), ptr, len)
     }
 }
 
+unsafe fn promotable_odd_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    promotable_to_vec(data, ptr, len, |shared| shared.cast())
+}
+
 unsafe fn promotable_odd_drop(data: &mut AtomicPtr<()>, ptr: *const u8, len: usize) {
     data.with_mut(|shared| {
         let shared = *shared;
         let kind = shared as usize & KIND_MASK;
 
         if kind == KIND_ARC {
-            release_shared(shared as *mut Shared);
+            release_shared(shared.cast());
         } else {
             debug_assert_eq!(kind, KIND_VEC);
 
-            drop(rebuild_boxed_slice(shared as *mut u8, ptr, len));
+            free_boxed_slice(shared.cast(), ptr, len);
         }
     });
 }
 
-unsafe fn rebuild_boxed_slice(buf: *mut u8, offset: *const u8, len: usize) -> Box<[u8]> {
+unsafe fn free_boxed_slice(buf: *mut u8, offset: *const u8, len: usize) {
     let cap = (offset as usize - buf as usize) + len;
-    Box::from_raw(slice::from_raw_parts_mut(buf, cap))
+    dealloc(buf, Layout::from_size_align(cap, 1).unwrap())
 }
 
 // ===== impl SharedVtable =====
 
 struct Shared {
-    // holds vec for drop, but otherwise doesnt access it
-    _vec: Vec<u8>,
+    // Holds arguments to dealloc upon Drop, but otherwise doesn't use them
+    buf: *mut u8,
+    cap: usize,
     ref_cnt: AtomicUsize,
 }
 
+impl Drop for Shared {
+    fn drop(&mut self) {
+        unsafe { dealloc(self.buf, Layout::from_size_align(self.cap, 1).unwrap()) }
+    }
+}
+
 // Assert that the alignment of `Shared` is divisible by 2.
 // This is a necessary invariant since we depend on allocating `Shared` a
 // shared object to implicitly carry the `KIND_ARC` flag in its pointer.
@@ -954,6 +1050,7 @@ const _: [(); 0 - mem::align_of::<Shared>() % 2] = []; // Assert that the alignm
 
 static SHARED_VTABLE: Vtable = Vtable {
     clone: shared_clone,
+    to_vec: shared_to_vec,
     drop: shared_drop,
 };
 
@@ -966,9 +1063,42 @@ unsafe fn shared_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Byte
     shallow_clone_arc(shared as _, ptr, len)
 }
 
+unsafe fn shared_to_vec_impl(shared: *mut Shared, ptr: *const u8, len: usize) -> Vec<u8> {
+    // Check that the ref_cnt is 1 (unique).
+    //
+    // If it is unique, then it is set to 0 with AcqRel fence for the same
+    // reason in release_shared.
+    //
+    // Otherwise, we take the other branch and call release_shared.
+    if (*shared)
+        .ref_cnt
+        .compare_exchange(1, 0, Ordering::AcqRel, Ordering::Relaxed)
+        .is_ok()
+    {
+        let buf = (*shared).buf;
+        let cap = (*shared).cap;
+
+        // Deallocate Shared
+        drop(Box::from_raw(shared as *mut mem::ManuallyDrop<Shared>));
+
+        // Copy back buffer
+        ptr::copy(ptr, buf, len);
+
+        Vec::from_raw_parts(buf, len, cap)
+    } else {
+        let v = slice::from_raw_parts(ptr, len).to_vec();
+        release_shared(shared);
+        v
+    }
+}
+
+unsafe fn shared_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    shared_to_vec_impl(data.load(Ordering::Relaxed).cast(), ptr, len)
+}
+
 unsafe fn shared_drop(data: &mut AtomicPtr<()>, _ptr: *const u8, _len: usize) {
     data.with_mut(|shared| {
-        release_shared(*shared as *mut Shared);
+        release_shared(shared.cast());
     });
 }
 
@@ -1006,9 +1136,9 @@ unsafe fn shallow_clone_vec(
     // updated and since the buffer hasn't been promoted to an
     // `Arc`, those three fields still are the components of the
     // vector.
-    let vec = rebuild_boxed_slice(buf, offset, len).into_vec();
     let shared = Box::new(Shared {
-        _vec: vec,
+        buf,
+        cap: (offset as usize - buf as usize) + len,
         // Initialize refcount to 2. One for this reference, and one
         // for the new clone that will be returned from
         // `shallow_clone`.
@@ -1082,10 +1212,40 @@ unsafe fn release_shared(ptr: *mut Shared) {
     // > "acquire" operation before deleting the object.
     //
     // [1]: (www.boost.org/doc/libs/1_55_0/doc/html/atomic/usage_examples.html)
-    atomic::fence(Ordering::Acquire);
+    //
+    // Thread sanitizer does not support atomic fences. Use an atomic load
+    // instead.
+    (*ptr).ref_cnt.load(Ordering::Acquire);
 
     // Drop the data
-    Box::from_raw(ptr);
+    drop(Box::from_raw(ptr));
+}
+
+// Ideally we would always use this version of `ptr_map` since it is strict
+// provenance compatible, but it results in worse codegen. We will however still
+// use it on miri because it gives better diagnostics for people who test bytes
+// code with miri.
+//
+// See https://github.com/tokio-rs/bytes/pull/545 for more info.
+#[cfg(miri)]
+fn ptr_map<F>(ptr: *mut u8, f: F) -> *mut u8
+where
+    F: FnOnce(usize) -> usize,
+{
+    let old_addr = ptr as usize;
+    let new_addr = f(old_addr);
+    let diff = new_addr.wrapping_sub(old_addr);
+    ptr.wrapping_add(diff)
+}
+
+#[cfg(not(miri))]
+fn ptr_map<F>(ptr: *mut u8, f: F) -> *mut u8
+where
+    F: FnOnce(usize) -> usize,
+{
+    let old_addr = ptr as usize;
+    let new_addr = f(old_addr);
+    new_addr as *mut u8
 }
 
 // compile-fails
diff --git a/src/bytes_mut.rs b/src/bytes_mut.rs
index 147484d..70613b2 100644
--- a/src/bytes_mut.rs
+++ b/src/bytes_mut.rs
@@ -1,5 +1,5 @@
 use core::iter::{FromIterator, Iterator};
-use core::mem::{self, ManuallyDrop};
+use core::mem::{self, ManuallyDrop, MaybeUninit};
 use core::ops::{Deref, DerefMut};
 use core::ptr::{self, NonNull};
 use core::{cmp, fmt, hash, isize, slice, usize};
@@ -8,6 +8,7 @@ use alloc::{
     borrow::{Borrow, BorrowMut},
     boxed::Box,
     string::String,
+    vec,
     vec::Vec,
 };
 
@@ -15,7 +16,7 @@ use crate::buf::{IntoIter, UninitSlice};
 use crate::bytes::Vtable;
 #[allow(unused)]
 use crate::loom::sync::atomic::AtomicMut;
-use crate::loom::sync::atomic::{self, AtomicPtr, AtomicUsize, Ordering};
+use crate::loom::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 use crate::{Buf, BufMut, Bytes};
 
 /// A unique reference to a contiguous slice of memory.
@@ -252,12 +253,28 @@ impl BytesMut {
 
             let ptr = self.ptr.as_ptr();
             let len = self.len;
-            let data = AtomicPtr::new(self.data as _);
+            let data = AtomicPtr::new(self.data.cast());
             mem::forget(self);
             unsafe { Bytes::with_vtable(ptr, len, data, &SHARED_VTABLE) }
         }
     }
 
+    /// Creates a new `BytesMut`, which is initialized with zero.
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BytesMut;
+    ///
+    /// let zeros = BytesMut::zeroed(42);
+    ///
+    /// assert_eq!(zeros.len(), 42);
+    /// zeros.into_iter().for_each(|x| assert_eq!(x, 0));
+    /// ```
+    pub fn zeroed(len: usize) -> BytesMut {
+        BytesMut::from_vec(vec![0; len])
+    }
+
     /// Splits the bytes into two at the given index.
     ///
     /// Afterwards `self` contains elements `[0, at)`, and the returned
@@ -494,11 +511,20 @@ impl BytesMut {
     /// reallocations. A call to `reserve` may result in an allocation.
     ///
     /// Before allocating new buffer space, the function will attempt to reclaim
-    /// space in the existing buffer. If the current handle references a small
-    /// view in the original buffer and all other handles have been dropped,
-    /// and the requested capacity is less than or equal to the existing
-    /// buffer's capacity, then the current view will be copied to the front of
-    /// the buffer and the handle will take ownership of the full buffer.
+    /// space in the existing buffer. If the current handle references a view
+    /// into a larger original buffer, and all other handles referencing part
+    /// of the same original buffer have been dropped, then the current view
+    /// can be copied/shifted to the front of the buffer and the handle can take
+    /// ownership of the full buffer, provided that the full buffer is large
+    /// enough to fit the requested additional capacity.
+    ///
+    /// This optimization will only happen if shifting the data from the current
+    /// view to the front of the buffer is not too expensive in terms of the
+    /// (amortized) time required. The precise condition is subject to change;
+    /// as of now, the length of the data being shifted needs to be at least as
+    /// large as the distance that it's shifted by. If the current view is empty
+    /// and the original buffer is large enough to fit the requested additional
+    /// capacity, then reallocations will never happen.
     ///
     /// # Examples
     ///
@@ -562,17 +588,34 @@ impl BytesMut {
             // space.
             //
             // Otherwise, since backed by a vector, use `Vec::reserve`
+            //
+            // We need to make sure that this optimization does not kill the
+            // amortized runtimes of BytesMut's operations.
             unsafe {
                 let (off, prev) = self.get_vec_pos();
 
                 // Only reuse space if we can satisfy the requested additional space.
-                if self.capacity() - self.len() + off >= additional {
-                    // There's space - reuse it
+                //
+                // Also check if the value of `off` suggests that enough bytes
+                // have been read to account for the overhead of shifting all
+                // the data (in an amortized analysis).
+                // Hence the condition `off >= self.len()`.
+                //
+                // This condition also already implies that the buffer is going
+                // to be (at least) half-empty in the end; so we do not break
+                // the (amortized) runtime with future resizes of the underlying
+                // `Vec`.
+                //
+                // [For more details check issue #524, and PR #525.]
+                if self.capacity() - self.len() + off >= additional && off >= self.len() {
+                    // There's enough space, and it's not too much overhead:
+                    // reuse the space!
                     //
                     // Just move the pointer back to the start after copying
                     // data back.
                     let base_ptr = self.ptr.as_ptr().offset(-(off as isize));
-                    ptr::copy(self.ptr.as_ptr(), base_ptr, self.len);
+                    // Since `off >= self.len()`, the two regions don't overlap.
+                    ptr::copy_nonoverlapping(self.ptr.as_ptr(), base_ptr, self.len);
                     self.ptr = vptr(base_ptr);
                     self.set_vec_pos(0, prev);
 
@@ -580,13 +623,14 @@ impl BytesMut {
                     // can gain capacity back.
                     self.cap += off;
                 } else {
-                    // No space - allocate more
+                    // Not enough space, or reusing might be too much overhead:
+                    // allocate more space!
                     let mut v =
                         ManuallyDrop::new(rebuild_vec(self.ptr.as_ptr(), self.len, self.cap, off));
                     v.reserve(additional);
 
                     // Update the info
-                    self.ptr = vptr(v.as_mut_ptr().offset(off as isize));
+                    self.ptr = vptr(v.as_mut_ptr().add(off));
                     self.len = v.len() - off;
                     self.cap = v.capacity() - off;
                 }
@@ -596,7 +640,7 @@ impl BytesMut {
         }
 
         debug_assert_eq!(kind, KIND_ARC);
-        let shared: *mut Shared = self.data as _;
+        let shared: *mut Shared = self.data;
 
         // Reserving involves abandoning the currently shared buffer and
         // allocating a new vector with the requested capacity.
@@ -619,29 +663,65 @@ impl BytesMut {
                 // sure that the vector has enough capacity.
                 let v = &mut (*shared).vec;
 
-                if v.capacity() >= new_cap {
-                    // The capacity is sufficient, reclaim the buffer
-                    let ptr = v.as_mut_ptr();
+                let v_capacity = v.capacity();
+                let ptr = v.as_mut_ptr();
 
-                    ptr::copy(self.ptr.as_ptr(), ptr, len);
+                let offset = offset_from(self.ptr.as_ptr(), ptr);
+
+                // Compare the condition in the `kind == KIND_VEC` case above
+                // for more details.
+                if v_capacity >= new_cap + offset {
+                    self.cap = new_cap;
+                    // no copy is necessary
+                } else if v_capacity >= new_cap && offset >= len {
+                    // The capacity is sufficient, and copying is not too much
+                    // overhead: reclaim the buffer!
+
+                    // `offset >= len` means: no overlap
+                    ptr::copy_nonoverlapping(self.ptr.as_ptr(), ptr, len);
 
                     self.ptr = vptr(ptr);
                     self.cap = v.capacity();
+                } else {
+                    // calculate offset
+                    let off = (self.ptr.as_ptr() as usize) - (v.as_ptr() as usize);
 
-                    return;
-                }
+                    // new_cap is calculated in terms of `BytesMut`, not the underlying
+                    // `Vec`, so it does not take the offset into account.
+                    //
+                    // Thus we have to manually add it here.
+                    new_cap = new_cap.checked_add(off).expect("overflow");
 
-                // The vector capacity is not sufficient. The reserve request is
-                // asking for more than the initial buffer capacity. Allocate more
-                // than requested if `new_cap` is not much bigger than the current
-                // capacity.
-                //
-                // There are some situations, using `reserve_exact` that the
-                // buffer capacity could be below `original_capacity`, so do a
-                // check.
-                let double = v.capacity().checked_shl(1).unwrap_or(new_cap);
+                    // The vector capacity is not sufficient. The reserve request is
+                    // asking for more than the initial buffer capacity. Allocate more
+                    // than requested if `new_cap` is not much bigger than the current
+                    // capacity.
+                    //
+                    // There are some situations, using `reserve_exact` that the
+                    // buffer capacity could be below `original_capacity`, so do a
+                    // check.
+                    let double = v.capacity().checked_shl(1).unwrap_or(new_cap);
+
+                    new_cap = cmp::max(double, new_cap);
 
-                new_cap = cmp::max(cmp::max(double, new_cap), original_capacity);
+                    // No space - allocate more
+                    //
+                    // The length field of `Shared::vec` is not used by the `BytesMut`;
+                    // instead we use the `len` field in the `BytesMut` itself. However,
+                    // when calling `reserve`, it doesn't guarantee that data stored in
+                    // the unused capacity of the vector is copied over to the new
+                    // allocation, so we need to ensure that we don't have any data we
+                    // care about in the unused capacity before calling `reserve`.
+                    debug_assert!(off + len <= v.capacity());
+                    v.set_len(off + len);
+                    v.reserve(new_cap - v.len());
+
+                    // Update the info
+                    self.ptr = vptr(v.as_mut_ptr().add(off));
+                    self.cap = v.capacity() - off;
+                }
+
+                return;
             } else {
                 new_cap = cmp::max(new_cap, original_capacity);
             }
@@ -659,7 +739,7 @@ impl BytesMut {
 
         // Update self
         let data = (original_capacity_repr << ORIGINAL_CAPACITY_OFFSET) | KIND_VEC;
-        self.data = data as _;
+        self.data = invalid_ptr(data);
         self.ptr = vptr(v.as_mut_ptr());
         self.len = v.len();
         self.cap = v.capacity();
@@ -686,11 +766,11 @@ impl BytesMut {
         self.reserve(cnt);
 
         unsafe {
-            let dst = self.uninit_slice();
+            let dst = self.spare_capacity_mut();
             // Reserved above
             debug_assert!(dst.len() >= cnt);
 
-            ptr::copy_nonoverlapping(extend.as_ptr(), dst.as_mut_ptr() as *mut u8, cnt);
+            ptr::copy_nonoverlapping(extend.as_ptr(), dst.as_mut_ptr().cast(), cnt);
         }
 
         unsafe {
@@ -700,10 +780,11 @@ impl BytesMut {
 
     /// Absorbs a `BytesMut` that was previously split off.
     ///
-    /// If the two `BytesMut` objects were previously contiguous, i.e., if
-    /// `other` was created by calling `split_off` on this `BytesMut`, then
-    /// this is an `O(1)` operation that just decreases a reference
-    /// count and sets a few indices. Otherwise this method degenerates to
+    /// If the two `BytesMut` objects were previously contiguous and not mutated
+    /// in a way that causes re-allocation i.e., if `other` was created by
+    /// calling `split_off` on this `BytesMut`, then this is an `O(1)` operation
+    /// that just decreases a reference count and sets a few indices.
+    /// Otherwise this method degenerates to
     /// `self.extend_from_slice(other.as_ref())`.
     ///
     /// # Examples
@@ -754,7 +835,7 @@ impl BytesMut {
             ptr,
             len,
             cap,
-            data: data as *mut _,
+            data: invalid_ptr(data),
         }
     }
 
@@ -801,7 +882,7 @@ impl BytesMut {
         // Updating the start of the view is setting `ptr` to point to the
         // new start and updating the `len` field to reflect the new length
         // of the view.
-        self.ptr = vptr(self.ptr.as_ptr().offset(start as isize));
+        self.ptr = vptr(self.ptr.as_ptr().add(start));
 
         if self.len >= start {
             self.len -= start;
@@ -825,7 +906,7 @@ impl BytesMut {
             return Ok(());
         }
 
-        let ptr = unsafe { self.ptr.as_ptr().offset(self.len as isize) };
+        let ptr = unsafe { self.ptr.as_ptr().add(self.len) };
         if ptr == other.ptr.as_ptr()
             && self.kind() == KIND_ARC
             && other.kind() == KIND_ARC
@@ -875,7 +956,7 @@ impl BytesMut {
         // always succeed.
         debug_assert_eq!(shared as usize & KIND_MASK, KIND_ARC);
 
-        self.data = shared as _;
+        self.data = shared;
     }
 
     /// Makes an exact shallow clone of `self`.
@@ -908,16 +989,45 @@ impl BytesMut {
         debug_assert_eq!(self.kind(), KIND_VEC);
         debug_assert!(pos <= MAX_VEC_POS);
 
-        self.data = ((pos << VEC_POS_OFFSET) | (prev & NOT_VEC_POS_MASK)) as *mut _;
+        self.data = invalid_ptr((pos << VEC_POS_OFFSET) | (prev & NOT_VEC_POS_MASK));
     }
 
+    /// Returns the remaining spare capacity of the buffer as a slice of `MaybeUninit<u8>`.
+    ///
+    /// The returned slice can be used to fill the buffer with data (e.g. by
+    /// reading from a file) before marking the data as initialized using the
+    /// [`set_len`] method.
+    ///
+    /// [`set_len`]: BytesMut::set_len
+    ///
+    /// # Examples
+    ///
+    /// ```
+    /// use bytes::BytesMut;
+    ///
+    /// // Allocate buffer big enough for 10 bytes.
+    /// let mut buf = BytesMut::with_capacity(10);
+    ///
+    /// // Fill in the first 3 elements.
+    /// let uninit = buf.spare_capacity_mut();
+    /// uninit[0].write(0);
+    /// uninit[1].write(1);
+    /// uninit[2].write(2);
+    ///
+    /// // Mark the first 3 bytes of the buffer as being initialized.
+    /// unsafe {
+    ///     buf.set_len(3);
+    /// }
+    ///
+    /// assert_eq!(&buf[..], &[0, 1, 2]);
+    /// ```
     #[inline]
-    fn uninit_slice(&mut self) -> &mut UninitSlice {
+    pub fn spare_capacity_mut(&mut self) -> &mut [MaybeUninit<u8>] {
         unsafe {
-            let ptr = self.ptr.as_ptr().offset(self.len as isize);
+            let ptr = self.ptr.as_ptr().add(self.len);
             let len = self.cap - self.len;
 
-            UninitSlice::from_raw_parts_mut(ptr, len)
+            slice::from_raw_parts_mut(ptr.cast(), len)
         }
     }
 }
@@ -934,7 +1044,7 @@ impl Drop for BytesMut {
                 let _ = rebuild_vec(self.ptr.as_ptr(), self.len, self.cap, off);
             }
         } else if kind == KIND_ARC {
-            unsafe { release_shared(self.data as _) };
+            unsafe { release_shared(self.data) };
         }
     }
 }
@@ -991,7 +1101,7 @@ unsafe impl BufMut for BytesMut {
         if self.capacity() == self.len() {
             self.reserve(64);
         }
-        self.uninit_slice()
+        UninitSlice::from_slice(self.spare_capacity_mut())
     }
 
     // Specialize these methods so they can skip checking `remaining_mut`
@@ -1016,7 +1126,7 @@ unsafe impl BufMut for BytesMut {
     fn put_bytes(&mut self, val: u8, cnt: usize) {
         self.reserve(cnt);
         unsafe {
-            let dst = self.uninit_slice();
+            let dst = self.spare_capacity_mut();
             // Reserved above
             debug_assert!(dst.len() >= cnt);
 
@@ -1161,7 +1271,7 @@ impl<'a> IntoIterator for &'a BytesMut {
     type IntoIter = core::slice::Iter<'a, u8>;
 
     fn into_iter(self) -> Self::IntoIter {
-        self.as_ref().into_iter()
+        self.as_ref().iter()
     }
 }
 
@@ -1190,7 +1300,18 @@ impl<'a> Extend<&'a u8> for BytesMut {
     where
         T: IntoIterator<Item = &'a u8>,
     {
-        self.extend(iter.into_iter().map(|b| *b))
+        self.extend(iter.into_iter().copied())
+    }
+}
+
+impl Extend<Bytes> for BytesMut {
+    fn extend<T>(&mut self, iter: T)
+    where
+        T: IntoIterator<Item = Bytes>,
+    {
+        for bytes in iter {
+            self.extend_from_slice(&bytes)
+        }
     }
 }
 
@@ -1202,7 +1323,7 @@ impl FromIterator<u8> for BytesMut {
 
 impl<'a> FromIterator<&'a u8> for BytesMut {
     fn from_iter<T: IntoIterator<Item = &'a u8>>(into_iter: T) -> Self {
-        BytesMut::from_iter(into_iter.into_iter().map(|b| *b))
+        BytesMut::from_iter(into_iter.into_iter().copied())
     }
 }
 
@@ -1243,10 +1364,13 @@ unsafe fn release_shared(ptr: *mut Shared) {
     // > "acquire" operation before deleting the object.
     //
     // [1]: (www.boost.org/doc/libs/1_55_0/doc/html/atomic/usage_examples.html)
-    atomic::fence(Ordering::Acquire);
+    //
+    // Thread sanitizer does not support atomic fences. Use an atomic load
+    // instead.
+    (*ptr).ref_count.load(Ordering::Acquire);
 
     // Drop the data
-    Box::from_raw(ptr);
+    drop(Box::from_raw(ptr));
 }
 
 impl Shared {
@@ -1392,7 +1516,7 @@ impl PartialOrd<BytesMut> for str {
 
 impl PartialEq<Vec<u8>> for BytesMut {
     fn eq(&self, other: &Vec<u8>) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -1416,7 +1540,7 @@ impl PartialOrd<BytesMut> for Vec<u8> {
 
 impl PartialEq<String> for BytesMut {
     fn eq(&self, other: &String) -> bool {
-        *self == &other[..]
+        *self == other[..]
     }
 }
 
@@ -1482,13 +1606,51 @@ impl PartialOrd<BytesMut> for &str {
 
 impl PartialEq<BytesMut> for Bytes {
     fn eq(&self, other: &BytesMut) -> bool {
-        &other[..] == &self[..]
+        other[..] == self[..]
     }
 }
 
 impl PartialEq<Bytes> for BytesMut {
     fn eq(&self, other: &Bytes) -> bool {
-        &other[..] == &self[..]
+        other[..] == self[..]
+    }
+}
+
+impl From<BytesMut> for Vec<u8> {
+    fn from(mut bytes: BytesMut) -> Self {
+        let kind = bytes.kind();
+
+        let mut vec = if kind == KIND_VEC {
+            unsafe {
+                let (off, _) = bytes.get_vec_pos();
+                rebuild_vec(bytes.ptr.as_ptr(), bytes.len, bytes.cap, off)
+            }
+        } else if kind == KIND_ARC {
+            let shared = bytes.data as *mut Shared;
+
+            if unsafe { (*shared).is_unique() } {
+                let vec = mem::replace(unsafe { &mut (*shared).vec }, Vec::new());
+
+                unsafe { release_shared(shared) };
+
+                vec
+            } else {
+                return bytes.deref().to_vec();
+            }
+        } else {
+            return bytes.deref().to_vec();
+        };
+
+        let len = bytes.len;
+
+        unsafe {
+            ptr::copy(bytes.ptr.as_ptr(), vec.as_mut_ptr(), len);
+            vec.set_len(len);
+        }
+
+        mem::forget(bytes);
+
+        vec
     }
 }
 
@@ -1501,6 +1663,35 @@ fn vptr(ptr: *mut u8) -> NonNull<u8> {
     }
 }
 
+/// Returns a dangling pointer with the given address. This is used to store
+/// integer data in pointer fields.
+///
+/// It is equivalent to `addr as *mut T`, but this fails on miri when strict
+/// provenance checking is enabled.
+#[inline]
+fn invalid_ptr<T>(addr: usize) -> *mut T {
+    let ptr = core::ptr::null_mut::<u8>().wrapping_add(addr);
+    debug_assert_eq!(ptr as usize, addr);
+    ptr.cast::<T>()
+}
+
+/// Precondition: dst >= original
+///
+/// The following line is equivalent to:
+///
+/// ```rust,ignore
+/// self.ptr.as_ptr().offset_from(ptr) as usize;
+/// ```
+///
+/// But due to min rust is 1.39 and it is only stablised
+/// in 1.47, we cannot use it.
+#[inline]
+fn offset_from(dst: *mut u8, original: *mut u8) -> usize {
+    debug_assert!(dst >= original);
+
+    dst as usize - original as usize
+}
+
 unsafe fn rebuild_vec(ptr: *mut u8, mut len: usize, mut cap: usize, off: usize) -> Vec<u8> {
     let ptr = ptr.offset(-(off as isize));
     len += off;
@@ -1513,6 +1704,7 @@ unsafe fn rebuild_vec(ptr: *mut u8, mut len: usize, mut cap: usize, off: usize)
 
 static SHARED_VTABLE: Vtable = Vtable {
     clone: shared_v_clone,
+    to_vec: shared_v_to_vec,
     drop: shared_v_drop,
 };
 
@@ -1520,10 +1712,32 @@ unsafe fn shared_v_clone(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> By
     let shared = data.load(Ordering::Relaxed) as *mut Shared;
     increment_shared(shared);
 
-    let data = AtomicPtr::new(shared as _);
+    let data = AtomicPtr::new(shared as *mut ());
     Bytes::with_vtable(ptr, len, data, &SHARED_VTABLE)
 }
 
+unsafe fn shared_v_to_vec(data: &AtomicPtr<()>, ptr: *const u8, len: usize) -> Vec<u8> {
+    let shared: *mut Shared = data.load(Ordering::Relaxed).cast();
+
+    if (*shared).is_unique() {
+        let shared = &mut *shared;
+
+        // Drop shared
+        let mut vec = mem::replace(&mut shared.vec, Vec::new());
+        release_shared(shared);
+
+        // Copy back buffer
+        ptr::copy(ptr, vec.as_mut_ptr(), len);
+        vec.set_len(len);
+
+        vec
+    } else {
+        let v = slice::from_raw_parts(ptr, len).to_vec();
+        release_shared(shared);
+        v
+    }
+}
+
 unsafe fn shared_v_drop(data: &mut AtomicPtr<()>, _ptr: *const u8, _len: usize) {
     data.with_mut(|shared| {
         release_shared(*shared as *mut Shared);
diff --git a/src/fmt/debug.rs b/src/fmt/debug.rs
index a854551..83de695 100644
--- a/src/fmt/debug.rs
+++ b/src/fmt/debug.rs
@@ -25,7 +25,7 @@ impl Debug for BytesRef<'_> {
             } else if b == b'\0' {
                 write!(f, "\\0")?;
             // ASCII printable
-            } else if b >= 0x20 && b < 0x7f {
+            } else if (0x20..0x7f).contains(&b) {
                 write!(f, "{}", b as char)?;
             } else {
                 write!(f, "\\x{:02x}", b)?;
@@ -38,12 +38,12 @@ impl Debug for BytesRef<'_> {
 
 impl Debug for Bytes {
     fn fmt(&self, f: &mut Formatter<'_>) -> Result {
-        Debug::fmt(&BytesRef(&self.as_ref()), f)
+        Debug::fmt(&BytesRef(self.as_ref()), f)
     }
 }
 
 impl Debug for BytesMut {
     fn fmt(&self, f: &mut Formatter<'_>) -> Result {
-        Debug::fmt(&BytesRef(&self.as_ref()), f)
+        Debug::fmt(&BytesRef(self.as_ref()), f)
     }
 }
diff --git a/src/lib.rs b/src/lib.rs
index 706735e..af436b3 100644
--- a/src/lib.rs
+++ b/src/lib.rs
@@ -4,6 +4,7 @@
     attr(deny(warnings, rust_2018_idioms), allow(dead_code, unused_variables))
 ))]
 #![no_std]
+#![cfg_attr(docsrs, feature(doc_cfg))]
 
 //! Provides abstractions for working with bytes.
 //!
diff --git a/src/loom.rs b/src/loom.rs
index 1cae881..9e6b2d5 100644
--- a/src/loom.rs
+++ b/src/loom.rs
@@ -1,7 +1,7 @@
 #[cfg(not(all(test, loom)))]
 pub(crate) mod sync {
     pub(crate) mod atomic {
-        pub(crate) use core::sync::atomic::{fence, AtomicPtr, AtomicUsize, Ordering};
+        pub(crate) use core::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
         pub(crate) trait AtomicMut<T> {
             fn with_mut<F, R>(&mut self, f: F) -> R
@@ -23,7 +23,7 @@ pub(crate) mod sync {
 #[cfg(all(test, loom))]
 pub(crate) mod sync {
     pub(crate) mod atomic {
-        pub(crate) use loom::sync::atomic::{fence, AtomicPtr, AtomicUsize, Ordering};
+        pub(crate) use loom::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
         pub(crate) trait AtomicMut<T> {}
     }
diff --git a/tests/test_bytes.rs b/tests/test_bytes.rs
index 402017b..da1b074 100644
--- a/tests/test_bytes.rs
+++ b/tests/test_bytes.rs
@@ -4,8 +4,8 @@ use bytes::{Buf, BufMut, Bytes, BytesMut};
 
 use std::usize;
 
-const LONG: &'static [u8] = b"mary had a little lamb, little lamb, little lamb";
-const SHORT: &'static [u8] = b"hello world";
+const LONG: &[u8] = b"mary had a little lamb, little lamb, little lamb";
+const SHORT: &[u8] = b"hello world";
 
 fn is_sync<T: Sync>() {}
 fn is_send<T: Send>() {}
@@ -412,8 +412,8 @@ fn freeze_after_split_off() {
 fn fns_defined_for_bytes_mut() {
     let mut bytes = BytesMut::from(&b"hello world"[..]);
 
-    bytes.as_ptr();
-    bytes.as_mut_ptr();
+    let _ = bytes.as_ptr();
+    let _ = bytes.as_mut_ptr();
 
     // Iterator
     let v: Vec<u8> = bytes.as_ref().iter().cloned().collect();
@@ -444,7 +444,7 @@ fn reserve_growth() {
     let _ = bytes.split();
 
     bytes.reserve(65);
-    assert_eq!(bytes.capacity(), 128);
+    assert_eq!(bytes.capacity(), 117);
 }
 
 #[test]
@@ -517,6 +517,34 @@ fn reserve_in_arc_unique_doubles() {
 }
 
 #[test]
+fn reserve_in_arc_unique_does_not_overallocate_after_split() {
+    let mut bytes = BytesMut::from(LONG);
+    let orig_capacity = bytes.capacity();
+    drop(bytes.split_off(LONG.len() / 2));
+
+    // now bytes is Arc and refcount == 1
+
+    let new_capacity = bytes.capacity();
+    bytes.reserve(orig_capacity - new_capacity);
+    assert_eq!(bytes.capacity(), orig_capacity);
+}
+
+#[test]
+fn reserve_in_arc_unique_does_not_overallocate_after_multiple_splits() {
+    let mut bytes = BytesMut::from(LONG);
+    let orig_capacity = bytes.capacity();
+    for _ in 0..10 {
+        drop(bytes.split_off(LONG.len() / 2));
+
+        // now bytes is Arc and refcount == 1
+
+        let new_capacity = bytes.capacity();
+        bytes.reserve(orig_capacity - new_capacity);
+    }
+    assert_eq!(bytes.capacity(), orig_capacity);
+}
+
+#[test]
 fn reserve_in_arc_nonunique_does_not_overallocate() {
     let mut bytes = BytesMut::with_capacity(1000);
     let _copy = bytes.split();
@@ -528,6 +556,25 @@ fn reserve_in_arc_nonunique_does_not_overallocate() {
     assert_eq!(2001, bytes.capacity());
 }
 
+/// This function tests `BytesMut::reserve_inner`, where `BytesMut` holds
+/// a unique reference to the shared vector and decide to reuse it
+/// by reallocating the `Vec`.
+#[test]
+fn reserve_shared_reuse() {
+    let mut bytes = BytesMut::with_capacity(1000);
+    bytes.put_slice(b"Hello, World!");
+    drop(bytes.split());
+
+    bytes.put_slice(b"!123ex123,sadchELLO,_wORLD!");
+    // Use split_off so that v.capacity() - self.cap != off
+    drop(bytes.split_off(9));
+    assert_eq!(&*bytes, b"!123ex123");
+
+    bytes.reserve(2000);
+    assert_eq!(&*bytes, b"!123ex123");
+    assert_eq!(bytes.capacity(), 2009);
+}
+
 #[test]
 fn extend_mut() {
     let mut bytes = BytesMut::with_capacity(0);
@@ -546,6 +593,13 @@ fn extend_from_slice_mut() {
 }
 
 #[test]
+fn extend_mut_from_bytes() {
+    let mut bytes = BytesMut::with_capacity(0);
+    bytes.extend([Bytes::from(LONG)]);
+    assert_eq!(*bytes, LONG[..]);
+}
+
+#[test]
 fn extend_mut_without_size_hint() {
     let mut bytes = BytesMut::with_capacity(0);
     let mut long_iter = LONG.iter();
@@ -875,7 +929,7 @@ fn from_iter_no_size_hint() {
 
 fn test_slice_ref(bytes: &Bytes, start: usize, end: usize, expected: &[u8]) {
     let slice = &(bytes.as_ref()[start..end]);
-    let sub = bytes.slice_ref(&slice);
+    let sub = bytes.slice_ref(slice);
     assert_eq!(&sub[..], expected);
 }
 
@@ -895,7 +949,7 @@ fn slice_ref_empty() {
     let bytes = Bytes::from(&b""[..]);
     let slice = &(bytes.as_ref()[0..0]);
 
-    let sub = bytes.slice_ref(&slice);
+    let sub = bytes.slice_ref(slice);
     assert_eq!(&sub[..], b"");
 }
 
@@ -1003,3 +1057,155 @@ fn box_slice_empty() {
     let b = Bytes::from(empty);
     assert!(b.is_empty());
 }
+
+#[test]
+fn bytes_into_vec() {
+    // Test kind == KIND_VEC
+    let content = b"helloworld";
+
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(content);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, content);
+
+    // Test kind == KIND_ARC, shared.is_unique() == True
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(b"abcdewe23");
+    bytes.put_slice(content);
+
+    // Overwrite the bytes to make sure only one reference to the underlying
+    // Vec exists.
+    bytes = bytes.split_off(9);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, content);
+
+    // Test kind == KIND_ARC, shared.is_unique() == False
+    let prefix = b"abcdewe23";
+
+    let mut bytes = BytesMut::new();
+    bytes.put_slice(prefix);
+    bytes.put_slice(content);
+
+    let vec: Vec<u8> = bytes.split_off(prefix.len()).into();
+    assert_eq!(&vec, content);
+
+    let vec: Vec<u8> = bytes.into();
+    assert_eq!(&vec, prefix);
+}
+
+#[test]
+fn test_bytes_into_vec() {
+    // Test STATIC_VTABLE.to_vec
+    let bs = b"1b23exfcz3r";
+    let vec: Vec<u8> = Bytes::from_static(bs).into();
+    assert_eq!(&*vec, bs);
+
+    // Test bytes_mut.SHARED_VTABLE.to_vec impl
+    eprintln!("1");
+    let mut bytes_mut: BytesMut = bs[..].into();
+
+    // Set kind to KIND_ARC so that after freeze, Bytes will use bytes_mut.SHARED_VTABLE
+    eprintln!("2");
+    drop(bytes_mut.split_off(bs.len()));
+
+    eprintln!("3");
+    let b1 = bytes_mut.freeze();
+    eprintln!("4");
+    let b2 = b1.clone();
+
+    eprintln!("{:#?}", (&*b1).as_ptr());
+
+    // shared.is_unique() = False
+    eprintln!("5");
+    assert_eq!(&*Vec::from(b2), bs);
+
+    // shared.is_unique() = True
+    eprintln!("6");
+    assert_eq!(&*Vec::from(b1), bs);
+
+    // Test bytes_mut.SHARED_VTABLE.to_vec impl where offset != 0
+    let mut bytes_mut1: BytesMut = bs[..].into();
+    let bytes_mut2 = bytes_mut1.split_off(9);
+
+    let b1 = bytes_mut1.freeze();
+    let b2 = bytes_mut2.freeze();
+
+    assert_eq!(Vec::from(b2), bs[9..]);
+    assert_eq!(Vec::from(b1), bs[..9]);
+}
+
+#[test]
+fn test_bytes_into_vec_promotable_even() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}
+
+#[test]
+fn test_bytes_vec_conversion() {
+    let mut vec = Vec::with_capacity(10);
+    vec.extend(b"abcdefg");
+    let b = Bytes::from(vec);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 7);
+    assert_eq!(v.capacity(), 10);
+
+    let mut b = Bytes::from(v);
+    b.advance(1);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 6);
+    assert_eq!(v.capacity(), 10);
+    assert_eq!(v.as_slice(), b"bcdefg");
+}
+
+#[test]
+fn test_bytes_mut_conversion() {
+    let mut b1 = BytesMut::with_capacity(10);
+    b1.extend(b"abcdefg");
+    let b2 = Bytes::from(b1);
+    let v = Vec::from(b2);
+    assert_eq!(v.len(), 7);
+    assert_eq!(v.capacity(), 10);
+
+    let mut b = Bytes::from(v);
+    b.advance(1);
+    let v = Vec::from(b);
+    assert_eq!(v.len(), 6);
+    assert_eq!(v.capacity(), 10);
+    assert_eq!(v.as_slice(), b"bcdefg");
+}
+
+#[test]
+fn test_bytes_capacity_len() {
+    for cap in 0..100 {
+        for len in 0..=cap {
+            let mut v = Vec::with_capacity(cap);
+            v.resize(len, 0);
+            let _ = Bytes::from(v);
+        }
+    }
+}
diff --git a/tests/test_bytes_odd_alloc.rs b/tests/test_bytes_odd_alloc.rs
index 04ba7c2..27ed877 100644
--- a/tests/test_bytes_odd_alloc.rs
+++ b/tests/test_bytes_odd_alloc.rs
@@ -24,8 +24,7 @@ unsafe impl GlobalAlloc for Odd {
             };
             let ptr = System.alloc(new_layout);
             if !ptr.is_null() {
-                let ptr = ptr.offset(1);
-                ptr
+                ptr.offset(1)
             } else {
                 ptr
             }
@@ -67,3 +66,32 @@ fn test_bytes_clone_drop() {
     let b1 = Bytes::from(vec);
     let _b2 = b1.clone();
 }
+
+#[test]
+fn test_bytes_into_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}
diff --git a/tests/test_bytes_vec_alloc.rs b/tests/test_bytes_vec_alloc.rs
index 418a9cd..107e56e 100644
--- a/tests/test_bytes_vec_alloc.rs
+++ b/tests/test_bytes_vec_alloc.rs
@@ -1,61 +1,87 @@
 use std::alloc::{GlobalAlloc, Layout, System};
-use std::{mem, ptr};
+use std::ptr::null_mut;
+use std::sync::atomic::{AtomicPtr, AtomicUsize, Ordering};
 
 use bytes::{Buf, Bytes};
 
 #[global_allocator]
-static LEDGER: Ledger = Ledger;
+static LEDGER: Ledger = Ledger::new();
 
-struct Ledger;
+const LEDGER_LENGTH: usize = 2048;
 
-const USIZE_SIZE: usize = mem::size_of::<usize>();
+struct Ledger {
+    alloc_table: [(AtomicPtr<u8>, AtomicUsize); LEDGER_LENGTH],
+}
 
-unsafe impl GlobalAlloc for Ledger {
-    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
-        if layout.align() == 1 && layout.size() > 0 {
-            // Allocate extra space to stash a record of
-            // how much space there was.
-            let orig_size = layout.size();
-            let size = orig_size + USIZE_SIZE;
-            let new_layout = match Layout::from_size_align(size, 1) {
-                Ok(layout) => layout,
-                Err(_err) => return ptr::null_mut(),
-            };
-            let ptr = System.alloc(new_layout);
-            if !ptr.is_null() {
-                (ptr as *mut usize).write(orig_size);
-                let ptr = ptr.offset(USIZE_SIZE as isize);
-                ptr
-            } else {
-                ptr
+impl Ledger {
+    const fn new() -> Self {
+        const ELEM: (AtomicPtr<u8>, AtomicUsize) =
+            (AtomicPtr::new(null_mut()), AtomicUsize::new(0));
+        let alloc_table = [ELEM; LEDGER_LENGTH];
+
+        Self { alloc_table }
+    }
+
+    /// Iterate over our table until we find an open entry, then insert into said entry
+    fn insert(&self, ptr: *mut u8, size: usize) {
+        for (entry_ptr, entry_size) in self.alloc_table.iter() {
+            // SeqCst is good enough here, we don't care about perf, i just want to be correct!
+            if entry_ptr
+                .compare_exchange(null_mut(), ptr, Ordering::SeqCst, Ordering::SeqCst)
+                .is_ok()
+            {
+                entry_size.store(size, Ordering::SeqCst);
+                break;
             }
-        } else {
-            System.alloc(layout)
         }
     }
 
-    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
-        if layout.align() == 1 && layout.size() > 0 {
-            let off_ptr = (ptr as *mut usize).offset(-1);
-            let orig_size = off_ptr.read();
-            if orig_size != layout.size() {
-                panic!(
-                    "bad dealloc: alloc size was {}, dealloc size is {}",
-                    orig_size,
-                    layout.size()
-                );
+    fn remove(&self, ptr: *mut u8) -> usize {
+        for (entry_ptr, entry_size) in self.alloc_table.iter() {
+            // set the value to be something that will never try and be deallocated, so that we
+            // don't have any chance of a race condition
+            //
+            // dont worry, LEDGER_LENGTH is really long to compensate for us not reclaiming space
+            if entry_ptr
+                .compare_exchange(
+                    ptr,
+                    invalid_ptr(usize::MAX),
+                    Ordering::SeqCst,
+                    Ordering::SeqCst,
+                )
+                .is_ok()
+            {
+                return entry_size.load(Ordering::SeqCst);
             }
+        }
+
+        panic!("Couldn't find a matching entry for {:x?}", ptr);
+    }
+}
+
+unsafe impl GlobalAlloc for Ledger {
+    unsafe fn alloc(&self, layout: Layout) -> *mut u8 {
+        let size = layout.size();
+        let ptr = System.alloc(layout);
+        self.insert(ptr, size);
+        ptr
+    }
+
+    unsafe fn dealloc(&self, ptr: *mut u8, layout: Layout) {
+        let orig_size = self.remove(ptr);
 
-            let new_layout = match Layout::from_size_align(layout.size() + USIZE_SIZE, 1) {
-                Ok(layout) => layout,
-                Err(_err) => std::process::abort(),
-            };
-            System.dealloc(off_ptr as *mut u8, new_layout);
+        if orig_size != layout.size() {
+            panic!(
+                "bad dealloc: alloc size was {}, dealloc size is {}",
+                orig_size,
+                layout.size()
+            );
         } else {
             System.dealloc(ptr, layout);
         }
     }
 }
+
 #[test]
 fn test_bytes_advance() {
     let mut bytes = Bytes::from(vec![10, 20, 30]);
@@ -77,3 +103,41 @@ fn test_bytes_truncate_and_advance() {
     bytes.advance(1);
     drop(bytes);
 }
+
+/// Returns a dangling pointer with the given address. This is used to store
+/// integer data in pointer fields.
+#[inline]
+fn invalid_ptr<T>(addr: usize) -> *mut T {
+    let ptr = std::ptr::null_mut::<u8>().wrapping_add(addr);
+    debug_assert_eq!(ptr as usize, addr);
+    ptr.cast::<T>()
+}
+
+#[test]
+fn test_bytes_into_vec() {
+    let vec = vec![33u8; 1024];
+
+    // Test cases where kind == KIND_VEC
+    let b1 = Bytes::from(vec.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 1
+    let b1 = Bytes::from(vec.clone());
+    drop(b1.clone());
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where kind == KIND_ARC, ref_cnt == 2
+    let b1 = Bytes::from(vec.clone());
+    let b2 = b1.clone();
+    assert_eq!(Vec::from(b1), vec);
+
+    // Test cases where vtable = SHARED_VTABLE, kind == KIND_ARC, ref_cnt == 1
+    assert_eq!(Vec::from(b2), vec);
+
+    // Test cases where offset != 0
+    let mut b1 = Bytes::from(vec.clone());
+    let b2 = b1.split_off(20);
+
+    assert_eq!(Vec::from(b2), vec[20..]);
+    assert_eq!(Vec::from(b1), vec[..20]);
+}
diff --git a/tests/test_chain.rs b/tests/test_chain.rs
index affaf7a..cfda6b8 100644
--- a/tests/test_chain.rs
+++ b/tests/test_chain.rs
@@ -134,6 +134,28 @@ fn vectored_read() {
 }
 
 #[test]
+fn chain_growing_buffer() {
+    let mut buff = [' ' as u8; 10];
+    let mut vec = b"wassup".to_vec();
+
+    let mut chained = (&mut buff[..]).chain_mut(&mut vec).chain_mut(Vec::new()); // Required for potential overflow because remaining_mut for Vec is isize::MAX - vec.len(), but for chain_mut is usize::MAX
+
+    chained.put_slice(b"hey there123123");
+
+    assert_eq!(&buff, b"hey there1");
+    assert_eq!(&vec, b"wassup23123");
+}
+
+#[test]
+fn chain_overflow_remaining_mut() {
+    let mut chained = Vec::<u8>::new().chain_mut(Vec::new()).chain_mut(Vec::new());
+
+    assert_eq!(chained.remaining_mut(), usize::MAX);
+    chained.put_slice(&[0; 256]);
+    assert_eq!(chained.remaining_mut(), usize::MAX);
+}
+
+#[test]
 fn chain_get_bytes() {
     let mut ab = Bytes::copy_from_slice(b"ab");
     let mut cd = Bytes::copy_from_slice(b"cd");