taoensso.nippy
High-performance serialization library for Clojure.
*auto-freeze-compressor*
dynamic
(fn [byte-array])->compressor used by `(freeze <x> {:compressor :auto}),
nil => default
*custom-readers*
dynamic
{<hash-or-byte-id> (fn [data-input])->read}
*freeze-fallback*
dynamic
Controls Nippy's behaviour when trying to freeze an item for which Nippy
doesn't currently have a native freeze/thaw implementation.
Possible values:
1. `nil` (no freeze-fallback, default)
Tries the following in order:
- Freeze with Java's `Serializable` interface if possible
- Freeze with Clojure's reader if possible
- Throw
2. `:write-unfreezable` keyword
Tries the following in order:
- Freeze with Java's `Serializable` interface if possible
- Freeze with Clojure's reader if possible
- Freeze a {:nippy/unfreezable {:type _}} placeholder value
3. [Advanced] Custom (fn [^java.io.DataOutput out item]) that must
write exactly one value to the given `DataOutput` stream
*freeze-serializable-allowlist*
dynamic
Used when attempting to <freeze/thaw> an object that:
- Does NOT implement Nippy's `Freezable` protocol.
- DOES implement Java's `Serializable` interface.
In this case, an allowlist will be checked to see if Java's
`Serializable` interface may be used.
This is a security measure to prevent possible Remote Code Execution
(RCE) when thawing malicious payloads. See [1] for details.
If `freeze` encounters a disallowed `Serializable` class, it will throw.
If `thaw` encounters a disallowed `Serializable` class, it will:
- Throw if it's not possible to safely quarantine the object
(object was frozen with Nippy < v2.15.0-final).
- Otherwise it will return a safely quarantined object of form
`{:nippy/unthawable {:class-name <> :content <quarantined-ba>}}`.
- Quarantined objects may be manually unquarantined with
`read-quarantined-serializable-object-unsafe!`.
There are 2x allowlists:
- `*freeze-serializable-allowlist*` ; Checked when freezing
- `*thaw-serializable-allowlist*` ; Checked when thawing
Example allowlist values:
- `(fn allow-class? [class-name] true)` ; Arbitrary predicate fn
- `#{"java.lang.Throwable", "clojure.lang.*"}` ; Set of class-names
- `"allow-and-record"` ; Special value, see [2]
Note that class-names in sets may contain "*" wildcards.
Default allowlist values are:
- default-freeze-serializable-allowlist ; `{"*"}` => allow any class
- default-thaw-serializable-allowlist ; A set of common safe classes
Allowlist values may be overridden with `binding`, `alter-var-root`, or:
- `taoensso.nippy.<freeze/thaw>-serializable-allowlist-base` JVM property value
- `taoensso.nippy.<freeze/thaw>-serializable-allowlist-add` JVM property value
- `TAOENSSO_NIPPY_<FREEZE/THAW>_SERIALIZABLE_ALLOWLIST_BASE` Environment variable value
- `TAOENSSO_NIPPY_<FREEZE/THAW>_SERIALIZABLE_ALLOWLIST_ADD` Environment variable value
If present, these will be read as comma-separated lists of class names
and formed into sets. Each initial allowlist value will then be:
(into (or <?base> <default>) <?additions>).
I.e. you can use:
- The "base" property/var to REPLACE Nippy's default allowlists.
- The "add" property/var to ADD TO Nippy's default allowlists.
The special `"allow-and-record"` value is also possible, see [2].
Upgrading from an older version of Nippy and unsure whether you've been
using Nippy's `Serializable` support, or which classes to allow? See [2].
See also `taoensso.encore/name-filter` for a util to help easily build
more advanced predicate functions.
Thanks to Timo Mihaljov (@solita-timo-mihaljov) for an excellent report
identifying this vulnerability.
[1] https://github.com/ptaoussanis/nippy/issues/130
[2] See `allow-and-record-any-serializable-class-unsafe`.
*thaw-serializable-allowlist*
dynamic
Used when attempting to <freeze/thaw> an object that:
- Does NOT implement Nippy's `Freezable` protocol.
- DOES implement Java's `Serializable` interface.
In this case, an allowlist will be checked to see if Java's
`Serializable` interface may be used.
This is a security measure to prevent possible Remote Code Execution
(RCE) when thawing malicious payloads. See [1] for details.
If `freeze` encounters a disallowed `Serializable` class, it will throw.
If `thaw` encounters a disallowed `Serializable` class, it will:
- Throw if it's not possible to safely quarantine the object
(object was frozen with Nippy < v2.15.0-final).
- Otherwise it will return a safely quarantined object of form
`{:nippy/unthawable {:class-name <> :content <quarantined-ba>}}`.
- Quarantined objects may be manually unquarantined with
`read-quarantined-serializable-object-unsafe!`.
There are 2x allowlists:
- `*freeze-serializable-allowlist*` ; Checked when freezing
- `*thaw-serializable-allowlist*` ; Checked when thawing
Example allowlist values:
- `(fn allow-class? [class-name] true)` ; Arbitrary predicate fn
- `#{"java.lang.Throwable", "clojure.lang.*"}` ; Set of class-names
- `"allow-and-record"` ; Special value, see [2]
Note that class-names in sets may contain "*" wildcards.
Default allowlist values are:
- default-freeze-serializable-allowlist ; `{"*"}` => allow any class
- default-thaw-serializable-allowlist ; A set of common safe classes
Allowlist values may be overridden with `binding`, `alter-var-root`, or:
- `taoensso.nippy.<freeze/thaw>-serializable-allowlist-base` JVM property value
- `taoensso.nippy.<freeze/thaw>-serializable-allowlist-add` JVM property value
- `TAOENSSO_NIPPY_<FREEZE/THAW>_SERIALIZABLE_ALLOWLIST_BASE` Environment variable value
- `TAOENSSO_NIPPY_<FREEZE/THAW>_SERIALIZABLE_ALLOWLIST_ADD` Environment variable value
If present, these will be read as comma-separated lists of class names
and formed into sets. Each initial allowlist value will then be:
(into (or <?base> <default>) <?additions>).
I.e. you can use:
- The "base" property/var to REPLACE Nippy's default allowlists.
- The "add" property/var to ADD TO Nippy's default allowlists.
The special `"allow-and-record"` value is also possible, see [2].
Upgrading from an older version of Nippy and unsure whether you've been
using Nippy's `Serializable` support, or which classes to allow? See [2].
See also `taoensso.encore/name-filter` for a util to help easily build
more advanced predicate functions.
Thanks to Timo Mihaljov (@solita-timo-mihaljov) for an excellent report
identifying this vulnerability.
[1] https://github.com/ptaoussanis/nippy/issues/130
[2] See `allow-and-record-any-serializable-class-unsafe`.
-cache-proxy
{[<x> <meta>] <idx>} for freezing, {<idx> <x-with-meta>} for thawing.
aes128-cbc-encryptor
Default 128bit AES-CBC encryptor with many-round SHA-512 key-gen.
See also `aes-128-cbc-encryptor`.
aes128-encryptor
Default 128bit AES-GCM encryptor with many-round SHA-512 key-gen.
Password form [:salted "my-password"]
---------------------------------------
USE CASE: You want more than a small, finite number of passwords (e.g. each
item encrypted will use a unique user-provided password).
IMPLEMENTATION: Uses a relatively cheap key hash, but automatically salts
every key.
PROS: Each key is independent so would need to be attacked independently.
CONS: Key caching impossible, so there's an inherent trade-off between
encryption/decryption speed and the difficulty of attacking any
particular key.
Slower than `aes128-cached`, and easier to attack any particular key - but
keys are independent.
Password form [:cached "my-password"]
---------------------------------------
USE CASE: You want only a small, finite number of passwords (e.g. a limited
number of staff/admins, or you'll be using a single password to
encrypt many items).
IMPLEMENTATION: Uses a _very_ expensive (but cached) key hash, and no salt.
PROS: Great amortized encryption/decryption speed. Expensive key hash makes
attacking any particular key very difficult.
CONS: Using a small number of keys for many encrypted items means that if any
key _is_ somehow compromised, _all_ items encrypted with that key are
compromised.
Faster than `aes128-salted`, and harder to attack any particular key - but
increased danger if a key is somehow compromised.
aes128-gcm-encryptor
Default 128bit AES-GCM encryptor with many-round SHA-512 key-gen.
Password form [:salted "my-password"]
---------------------------------------
USE CASE: You want more than a small, finite number of passwords (e.g. each
item encrypted will use a unique user-provided password).
IMPLEMENTATION: Uses a relatively cheap key hash, but automatically salts
every key.
PROS: Each key is independent so would need to be attacked independently.
CONS: Key caching impossible, so there's an inherent trade-off between
encryption/decryption speed and the difficulty of attacking any
particular key.
Slower than `aes128-cached`, and easier to attack any particular key - but
keys are independent.
Password form [:cached "my-password"]
---------------------------------------
USE CASE: You want only a small, finite number of passwords (e.g. a limited
number of staff/admins, or you'll be using a single password to
encrypt many items).
IMPLEMENTATION: Uses a _very_ expensive (but cached) key hash, and no salt.
PROS: Great amortized encryption/decryption speed. Expensive key hash makes
attacking any particular key very difficult.
CONS: Using a small number of keys for many encrypted items means that if any
key _is_ somehow compromised, _all_ items encrypted with that key are
compromised.
Faster than `aes128-salted`, and harder to attack any particular key - but
increased danger if a key is somehow compromised.
allow-and-record-any-serializable-class-unsafe
(allow-and-record-any-serializable-class-unsafe class-name)
A predicate (fn allow-class? [class-name]) fn that can be assigned
to `*freeze-serializable-allowlist*` and/or
`*thaw-serializable-allowlist*` that:
- Will allow ANY class to use Nippy's `Serializable` support (unsafe).
- And will record {<class-name> <frequency-allowed>} for the <=1000
classes that ~most frequently made use of this support.
`get-recorded-serializable-classes` returns the recorded state.
This predicate is provided as a convenience for users upgrading from
previous versions of Nippy that allowed the use of `Serializable` for all
classes by default.
While transitioning from an unsafe->safe configuration, you can use
this predicate (unsafe) to record information about which classes have
been using Nippy's `Serializable` support in your environment.
Once some time has passed, you can check the recorded state. If you're
satisfied that all recorded classes are safely `Serializable`, you can
then merge the recorded classes into Nippy's default allowlist/s, e.g.:
(alter-var-root #'thaw-serializable-allowlist*
(fn [_] (into default-thaw-serializable-allowlist
(keys (get-recorded-serializable-classes)))))
cache
(cache x)
Wraps value so that future writes of the same wrapped value with same
metadata will be efficiently encoded as references to this one.
(freeze [(cache "foo") (cache "foo") (cache "foo")])
will incl. a single "foo", plus 2x single-byte references to "foo".
compress
(compress compressor ba)
decompress
(decompress compressor ba)
decrypt
(decrypt encryptor pwd ba)
default-freeze-serializable-allowlist
Allows *any* class-name to be frozen using Java's `Serializable` interface.
This is generally safe since RCE risk is present only when thawing.
See also `*freeze-serializable-allowlist*`.
default-thaw-serializable-allowlist
A set of common safe class-names to allow to be frozen using Java's
`Serializable` interface. PRs welcome for additions.
See also `*thaw-serializable-allowlist*`.
encrypt
(encrypt encryptor pwd ba)
extend-freeze
macro
(extend-freeze type custom-type-id [x out] & body)
Extends Nippy to support freezing of a custom type (ideally concrete) with
given id of form:
* ℕ∈[1, 128] - 0 byte overhead. You are responsible for managing ids.
* (Namespaced) keyword - 2 byte overhead. Keyword will be hashed to 16 bit int,
collisions will throw at compile-time.
NB: be careful about extending to interfaces, Ref. <http://goo.gl/6gGRlU>.
(defrecord MyRec [data])
(extend-freeze MyRec :foo/my-type [x data-output] ; Keyword id
(.writeUTF [data-output] (:data x)))
;; or
(extend-freeze MyRec 1 [x data-output] ; Byte id
(.writeUTF [data-output] (:data x)))
extend-thaw
macro
(extend-thaw custom-type-id [in] & body)
Extends Nippy to support thawing of a custom type with given id:
(extend-thaw :foo/my-type [data-input] ; Keyword id
(MyRec. (.readUTF data-input)))
;; or
(extend-thaw 1 [data-input] ; Byte id
(MyRec. (.readUTF data-input)))
fast-freeze
(fast-freeze x)
Like `freeze` but:
- Writes data without a Nippy header
- Drops all support for compression and encryption
- Must be thawed with `fast-thaw`
Equivalent to (but a little faster than) `freeze` with opts
{:no-header? true, :compressor nil, :encryptor nil}.
Intended for use only by advanced users that clearly understand the tradeoffs.
I STRONGLY recommend that most users prefer the standard `freeze` since:
- The Nippy header is useful for data portability and preservation
- Compression is often benefitial at little/no cost
- The performance difference between `freeze` and `fast-freeze` is
often negligible in practice.
fast-thaw
(fast-thaw ba)
Like `thaw` but:
- Supports only data frozen with `fast-freeze`
- Drops all support for compression and encryption
Equivalent to (but a little faster than) `thaw` with opts:
{:no-header? true, :compressor nil, :encryptor nil}.
freezable?
(freezable? x)
(freezable? x {:as opts, :keys [recursive? allow-clojure-reader? allow-java-serializable?], :or {recursive? true}})
Alpha, subject to change.
Returns ∈ #{:native :maybe-clojure-reader :maybe-java-serializable nil},
truthy iff Nippy seems to support freezing for the given argument.
Important: result can be inaccurate in some cases. To be completely sure you
unfortunately need to try freeze then thaw the argument, and check the thawed
value.
Options include:
`recursive?` - Check recursively into given arg?
`allow-clojure-reader?` - Allow freezing with Clojure's reader?
`allow-java-serializable?` - Allow freezing with Java's `Serializable`?
freeze
(freeze x)
(freeze x {:as opts, :keys [compressor encryptor password serializable-allowlist incl-metadata?], :or {compressor :auto, encryptor aes128-gcm-encryptor}})
Serializes arg (any Clojure data type) to a byte array.
To freeze custom types, extend the Clojure reader or see `extend-freeze`.
freeze-to-file
(freeze-to-file file x)
(freeze-to-file file x freeze-opts)
Convenience util: like `freeze`, but writes to `(clojure.java.io/file <file>)`.
freeze-to-out!
(freeze-to-out! data-output x)
Serializes arg (any Clojure data type) to a DataOutput.
This is a low-level util: in most cases you'll want `freeze` instead.
freeze-to-string
(freeze-to-string x)
(freeze-to-string x freeze-opts)
Convenience util: like `freeze`, but returns a Base64-encoded string.
See also `thaw-from-string`.
get-recorded-serializable-classes
(get-recorded-serializable-classes)
Returns {<class-name> <frequency>} of the <=1000 classes that ~most
frequently made use of Nippy's `Serializable` support via
`allow-and-record-any-serializable-class-unsafe`.
See that function's docstring for more info.
IFreezable
protocol
Private implementation detail.
Protocol that types must implement to support native freezing by Nippy.
Don't use this directly, instead see `extend-freeze`.
members
-freezable?
(-freezable? _)
inspect-ba
(inspect-ba ba)
(inspect-ba ba thaw-opts)
Experimental, subject to change. Feedback welcome!
lz4-compressor
Default `LZ4` compressor:
- Compression ratio: `C` (0.58 on reference benchmark).
- Compression speed: `A` (240 msecs on reference benchmark).
- Decompression speed: `A+` (30 msecs on reference benchmark).
Good general-purpose compressor, favours speed.
See `taoensso.nippy-benchmarks` for detailed comparative benchmarks.
lzma2-compressor
Default `LZMA2` compressor:
- Compression ratio: `A+` (0.4 on reference benchmark).
- Compression speed: `E` (18.5 secs on reference benchmark).
- Decompression speed: `D` (12 secs on reference benchmark).
Specialized compressor, strongly favours ratio.
See `taoensso.nippy-benchmarks` for detailed comparative benchmarks.
public-types-spec
Public representation of Nippy's internal type schema.
For use by tooling and advanced users.
**HIGHLY EXPERIMENTAL!**
Subject to breaking change without notice.
Currently completely untested, may contain bugs.
Intended for use only by early adopters to give design feedback.
Format:
{<type-id> {:keys [type-id type-kw payload-spec deprecated?]}},
- `type-id`: A +ive single-byte identifier like `110`.
-ive type ids are reserved for custom user-defined types.
- `type-kw`: A keyword like `:kw-sm`,
suffixes used to differentiate subtypes of different sizes:
-0 ; Empty => 0 byte payload / element-count
-sm ; Small => 1 byte (byte) payload / element-count
-md ; Medium => 2 byte (short) payload / element-count
-lg ; Large => 4 byte (int) payload / element-count
-xl ; Extra large => 8 byte (long) payload / element-count
- `payload-spec` examples:
- nil ; No spec available (e.g. unpredictable payload)
- [] ; Type has no payload
- [[:bytes 4]] ; Type has payload of exactly 4 bytes
- [[:bytes 2] [:elements 2]] ; Type has payload of exactly 2 bytes,
; followed by 2 elements
- [[:bytes {:read 2}]
[:elements {:read 4 :multiplier 2 :unsigned? true}]]
; Type has payload of <short-count> bytes, followed by
; <unsigned-int-count>*2 (multiplier) elements
Note that `payload-spec` can be handy for skipping over items in
data stream without fully reading every item.
read-quarantined-serializable-object-unsafe!
(read-quarantined-serializable-object-unsafe! m)
Given a quarantined Serializable object like
{:nippy/unthawable {:class-name <> :content <quarantined-ba>}}, reads and
returns the object WITHOUT regard for `*thaw-serializable-allowlist*`.
**MAY BE UNSAFE!** Don't call this unless you absolutely trust the payload
to not contain any malicious code.
See `*thaw-serializable-allowlist*` for more info.
snappy-compressor
Default `Snappy` compressor:
- Compression ratio: `C` (0.58 on reference benchmark).
- Compression speed: `A+` (210 msecs on reference benchmark).
- Decompression speed: `B` (130 msecs on reference benchmark).
Good general-purpose compressor, favours speed.
See `taoensso.nippy-benchmarks` for detailed comparative benchmarks.
stress-data
(stress-data {:keys [comparable?], :as opts})
Returns map of reference stress data for use by tests, benchmarks, etc.
thaw
(thaw ba)
(thaw ba {:as opts, :keys [v1-compatibility? compressor encryptor password serializable-allowlist incl-metadata? thaw-xform], :or {compressor :auto, encryptor :auto}})
Deserializes a frozen Nippy byte array to its original Clojure data type.
To thaw custom types, extend the Clojure reader or see `extend-thaw`.
** By default, supports data frozen with Nippy v2+ ONLY **
Add `{:v1-compatibility? true}` option to support thawing of data frozen with
legacy versions of Nippy.
Options include:
:v1-compatibility? - support data frozen by legacy versions of Nippy?
:compressor - :auto (checks header, default) an ICompressor, or nil
:encryptor - :auto (checks header, default), an IEncryptor, or nil
thaw-from-file
(thaw-from-file file)
(thaw-from-file file thaw-opts)
Convenience util: like `thaw`, but reads from `(clojure.java.io/file <file>)`.
thaw-from-in!
(thaw-from-in! data-input)
Deserializes a frozen object from given DataInput to its original Clojure
data type.
This is a low-level util: in most cases you'll want `thaw` instead.
thaw-from-resource
(thaw-from-resource res)
(thaw-from-resource res thaw-opts)
Convenience util: like `thaw`, but reads from `(clojure.java.io/resource <res>)`.
thaw-from-string
(thaw-from-string s)
(thaw-from-string s thaw-opts)
Convenience util: like `thaw`, but takes a Base64-encoded string.
See also `freeze-to-string`.
throw-unfreezable
(throw-unfreezable x)
try-write-readable
(try-write-readable out x)
try-write-serializable
(try-write-serializable out x)
with-cache
macro
(with-cache & body)
Executes body with support for freezing/thawing cached values.
This is a low-level util: you won't need to use this yourself unless
you're using `freeze-to-out!` or `thaw-from-in!` (also low-level utils).
See also `cache`.
write-id
macro
(write-id out id)
write-unfreezable
(write-unfreezable out x)
zstd-compressor
Default `Zstd` (`Zstandard`) compressor:
- Compression ratio: `B` (0.53 on reference benchmark).
- Compression speed: `C` (1300 msecs on reference benchmark).
- Decompression speed: `B` (400 msecs on reference benchmark).
Good general-purpose compressor, balances ratio & speed.
See `taoensso.nippy-benchmarks` for detailed comparative benchmarks.