go-ipfs 0.4.0 (opens new window) has been released! Among the many changes are a revamped implementation of the IPFS communication protocols, increased performance, improvements to IPNS (the Interplanetary Naming System), many bugfixes, and lots of new features to make IPFS even more powerful.

0.4.0

# The IPFS Files API

In 0.4.0, we've added a new feature, the Files API, available through the subcommand: ipfs files. This subcommand allows a program to interact with IPFS using familiar filesystem operations, namely: creating directories, reading, writing and deleting files, listing out different directories, and so on.

This feature enables any application that uses a filesystem-like backend to use IPFS for storage without changing the application logic at all.

It is used like so:

> ipfs files mkdir /cats

> ipfs files ls /
cats

> echo bar | ipfs files write --create /cats/foo

> ipfs files ls /cats
foo

> ipfs files read /cats/foo
bar

> ipfs files stat /
QmNU8HmaeRa8VtfqAoQRJhSE5Zx54vyYf2nT1bDGcYXaNv
...

# /ipfs/QmNU8HmaeRa8VtfqAoQRJhSE5Zx54vyYf2nT1bDGcYXaNv is a snapshot
# of this file system!

# You can see it locally or on the public gateway:
#  https://ipfs.io/ipfs/QmNU8HmaeRa8VtfqAoQRJhSE5Zx54vyYf2nT1bDGcYXaNv

# Subsequent edits will produce a new /ipfs address for the root directory.

One great example of this is ipfs-blob-store (opens new window), an IPFS-backed storage driver that implements the blob-store-interface (opens new window), so that any app that uses any other blob-store storage driver (S3, IndexedDB, LevelDB, etc) can now use IPFS. registry-mirror (opens new window) uses this approach to mirror the npm registry onto IPFS.

We are looking forward to seeing more use cases of this new convenient and powerful API.

# Why we're changing the protocol

The most important change allows IPFS implementations to use pluggable stream multiplexers, such as yamux (opens new window), spdystream (opens new window), or muxado (opens new window). Instead of locking IPFS permanently into a single multiplexer that won't work for every language or situation, this change allows each IPFS implementation to implement the multiplexers that they choose to, and then negotiate which to use during the initial connection handshake.

This modularity with stream muxing makes it easier for certain languages to improve performance. For example, the Go programming language may have muxado and yamux implementations that are really good, but many languages lack good (or any) implementations of them. Or take Node.js, which works well with spdy-transport, and it would be nice to take advantage of that. And then there are options like multiplex (opens new window), which may not have the same performance, but are much easier to implement.

So, by supporting as many muxers as we can, we get to choose the best multiplexers for the job. It also makes it much easier to implement the IPFS protocols in a new language. And of course, if a better multiplexers standard comes along, it will be easier to upgrade IPFS to support it in the future.

The same code that allows us to easily select a stream muxer is also being used to select which IPFS sub-protocol to use for any given stream between peers. Now, if we need to make a breaking protocol change to any of those (like the DHT or bitswap) we can do so seamlessly, and provide easy backwards compatibility. We won't need to "break" anything because we will be able to have nodes run multiple protocols at the same time for compatibility.

In addition to the multiplexers changes, the protocol revamp has also improved efficiency and performance in a few important ways, including the elimination of a double wrapping of the length prefixer, and the removal of some unnecessary round trips between nodes.

# Breaking changes

This release contains a breaking change to the network wire protocol in the form of a major refactor and upgrade to the libp2p handshake protocol. Because of the refactor, all IPFS daemons earlier than 0.4.0 will not be able to communicate with the newest version. It is strongly recommended that everyone running an IPFS node upgrades to the latest version as soon as possible, as these nodes will, after a certain time, no longer be able to communicate with the majority of the network until they are upgraded. There are instructions on how to update below.

Refactoring the protocol is not something to be done lightly. But at this early stage, this is necessary to ensure we have the right design for IPFS in place for the future. It's better to improve the protocol now during this alpha stage of the project than when there are a lot more people running nodes across a lot of different implementations.

One of the important changes that's been made to the protocol means that there should never be a breaking change like this again. This is due to a change to allow nodes to announce the version of the protocol they are using when connecting to other IPFS nodes. The goal is to roll any future protocol changes into the implementations gradually, so that we can still support legacy protocols for a period of time, making it easier to deprecate old versions over time.

# The public gateway and bootstrappers

We provide two essential public services to the IPFS community: the public gateway at https://ipfs.io and the default bootstrap nodes. We're making sure that despite the breaking changes, both will continue to work with 0.4.x and 0.3.x for a while. You can read more about the details of this in an earlier blog post: Migrating ipfs.io from go-ipfs 0.3.x to 0.4.0. We expect to keep this grace period open until the end of April 2016.

# Changes to the repo

We made a few changes to our on disk storage format (called the 'repo'). The way object pinning (ipfs pin add) works has also been upgraded to be much more efficient, which will improve the overall speed of adding and downloading IPFS data. Previously, when you pinned a file recursively, it would add a recursive type pin to the root, and then add an indirect type pin to each child node. This made enumeration for GC convenient, but was very slow and made a lot of things needlessly complicated. We have switched to just adding the recursive pin to the root object, and then doing the enumeration of child nodes when we actually run a garbage collection.

Because of this change, you will need to run a migration (from repo version 2 to version 3). If you update with the ipfs-update tool, this will be done for you automatically. If you updated manually, and did not run the migration, ipfs will fail to run, and print a message saying that there is a mismatch in the repo versions.

# Other improvements and fixes

In addition to a major protocol improvement and upgrade, this release adds a lot of new functionality, performance speedups, and stability fixes that make this the best version of IPFS to date.

This release also includes performance and usability improvements to IPNS, which is IPFS's mutability layer. IPNS creates a link from an IPFS node's public key to the hash of an arbitrary objects hash, in a way that is cryptographically verifiable. We call this "publishing" a hash, and you can try it out using the ipfs publish command. By allowing users to change what the pubkey hash points to, we provide users with a single hash they can give to their users to get the latest version of their data. This creates a seamless way to use IPFS to verify content, and to distribute content via trustless nodes in a smart, safe way. This brings IPFS closer to the goal of being a global filesystem of data, that can allow everyone in the world to help serve the world's data in a way that enriches and empowers everybody.

# How to upgrade

Depending on how you initially installed IPFS, there are several ways to upgrade. If you installed IPFS with a pre-built binary, you can either head over to dist.ipfs.tech (opens new window) and grab the latest version from there. Or alternatively, from the same page you can grab the ipfs-update binary, and use it to perform the upgrade for you. If you installed from source, you can simply update your git repo to the v0.4.0 tag, run make toolkit_upgrade && make install.

Please upgrade your IPFS nodes as soon as you can, so you can take advantage of the improvements!

# Changelog

This is a major release with plenty of new features and bugfixes. It also includes breaking changes which make it incompatible with v0.3.x on the networking layer.

  • Major Changes
    • Multistream
      • The addition of multistream is a breaking change on the networking layer, but gives IPFS implementations the ability to mix and match different stream multiplexers, e.g. yamux, spdystream, or muxado. This adds a ton of flexibility on one of the lower layers of the protocol, and will help us avoid further breaking protocol changes in the future.
    • Files API
      • The new files command and API allow a program to interact with IPFS using familiar filesystem operations, namely: creating directories, reading, writing, and deleting files, listing out different directories, and so on. This feature enables any other application that uses a filesystem-like backend for storage, to use IPFS as its storage driver without having change the application logic at all.
    • Gx
      • go-ipfs now uses gx (opens new window) to manage its dependencies. This means that under the hood, go-ipfs's dependencies are backed by IPFS itself! It also means that go-ipfs is no longer installed using go get. Use make install instead.
  • New Features
    • Web UI
      • Update to new version which is compatible with 0.4.0. (@dignifiedquire)
    • Networking
      • Implement uTP transport. (@whyrusleeping)
      • Allow multiple addresses per configured bootstrap node. (@whyrusleeping)
    • IPNS
      • Improve IPNS resolution performance. (@whyrusleeping)
      • Have dnslink prefer TXT _dnslink.example.com, allows usage of CNAME records. (@Kubuxu)
      • Prevent ipfs name publish when /ipns is mounted. (@noffle)
    • Repo
      • Improve performance of ipfs add. (@whyrusleeping)
      • Add Datastore.NoSync config option for flatfs. (@rht)
      • Implement mark-and-sweep GC. (@whyrusleeping)
      • Allow for GC during ipfs add. (@whyrusleeping)
      • Add ipfs repo stat command. (@tmg, @diasdavid)
    • General
      • Add support for HTTP OPTIONS requests. (@lidel)
      • Add ipfs diag cmds to view active API requests (@whyrusleeping)
      • Add an IPFS_LOW_MEM environment veriable which relaxes Bitswap's memory usage. (@whyrusleeping)
      • The Docker image now lives at ipfs/go-ipfs and has been completely reworked. (@lgierth)
  • Security fixes
    • The gateway path prefix added in v0.3.10 was vulnerable to cross-site scripting attacks. This release introduces a configurable list of allowed path prefixes. It's called Gateway.PathPrefixes and takes a list of strings, e.g. ["/blog", "/foo/bar"]. The v0.3.x line will not receive any further updates, so please update to v0.4.0 as soon as possible. (@lgierth)
  • Incompatible Changes
    • Install using make install instead of go get (@whyrusleeping)
    • Rewrite pinning to store pins in IPFS objects. (@tv42)
    • Bump fs-repo version to 3. (@whyrusleeping)
    • Use multistream muxer (@whyrusleeping)
    • The default for --type in ipfs pin ls is now all. (@chriscool)
  • Bug Fixes
    • Remove msgio double wrap. (@jbenet)
    • Buffer msgio. (@whyrusleeping)
    • Perform various fixes to the FUSE code. (@tv42)
    • Compute ipfs add size in background to not stall add operation. (@whyrusleeping)
    • Add option to have ipfs add include top-level hidden files. (@noffle)
    • Fix CORS checks on the API. (@rht)
    • Fix ipfs update error message. (@tomgg)
    • Resolve paths in ipfs pin rm without network lookup. (@noffle)
    • Detect FUSE unmounts and track mount state. (@noffle)
    • Fix go1.6rc2 panic caused by CloseNotify being called from wrong goroutine. (@rwcarlsen)
    • Bump DHT kvalue from 10 to 20. (@whyrusleeping)
    • Put public key and IPNS entry to DHT in parallel. (@whyrusleeping)
    • Fix panic in CLI argument parsing. (@whyrusleeping)
    • Fix range error by using larger-than-zero-length buffer. (@noffle)
    • Fix yamux hanging issue by increasing AcceptBacklog. (@whyrusleeping)
    • Fix double Transport-Encoding header bug. (@whyrusleeping)
    • Fix uTP panic and file descriptor leak. (@whyrusleeping)
  • Tool Changes
    • Add --pin option to ipfs add, which defaults to true and allows --pin=false. (@eminence)
    • Add arguments to ipfs pin ls. (@chriscool)
    • Add dns and resolve commands to read-only API. (@Kubuxu)
    • Add option to display headers for ipfs object links. (@palkeo)
  • General Codebase Changes
    • Check Golang version in Makefile. (@chriscool)
    • Improve Makefile. (@tomgg)
    • Remove dead Jenkins CI code. (@lgierth)
    • Add locking interface to blockstore. (@whyrusleeping)
    • Add Merkledag FetchGraph and EnumerateChildren. (@whyrusleeping)
    • Rename Lock/RLock to GCLock/PinLock. (@jbenet)
    • Implement pluggable datastore types. (@tv42)
    • Record datastore metrics for non-default datastores. (@tv42)
    • Allow multistream to have zero-rtt stream opening. (@whyrusleeping)
    • Refactor ipnsfs into a more generic and well tested mfs. (@whyrusleeping)
    • Grab more peers if bucket doesn't contain enough. (@whyrusleeping)
    • Use CloseNotify in gateway. (@whyrusleeping)
    • Flatten multipart file transfers. (@whyrusleeping)
    • Send updated DHT record fixes to peers who sent outdated records. (@whyrusleeping)
    • Replace go-psutil with go-sysinfo. (@whyrusleeping)
    • Use ServeContent for index.html. (@AtnNn)
    • Refactor object patch API to not store data in URL. (@whyrusleeping)
    • Use mfs for ipfs add. (@whyrusleeping)
    • Add Server header to API responses. (@Kubuxu)
    • Wire context directly into HTTP requests. (@rht)
    • Wire context directly into GetDAG operations within GC. (@rht)
    • Vendor libp2p using gx. (@whyrusleeping)
    • Use gx vendored packages instead of Godeps. (@whyrusleeping)
    • Simplify merkledag package interface to ease IPLD inclusion. (@mildred)
    • Add default option value support to commands lib. (@whyrusleeping)
    • Refactor merkledag fetching methods. (@whyrusleeping)
    • Use net/url to escape paths within Web UI. (@noffle)
    • Deprecated key.Pretty(). (@MichealMure)
  • Documentation
    • Fix and update help text for every ipfs command. (@RichardLitt)
    • Change sample API origin settings from wildcard (*) to example.com. (@Kubuxu)
    • Improve documentation of installation process in README. (@whyrusleeping)
    • Improve windows.md. (@chriscool)
    • Clarify instructions for installing from source. (@noffle)
    • Make version checking more robust. (@jedahan)
    • Assert the source code is located within GOPATH. (@whyrusleeping)
    • Remove mentions of /dns from ipfs dns command docs. (@lgierth)
  • Testing
    • Refactor iptb tests. (@chriscool)
    • Improve t0240 sharness test. (@chriscool)
    • Make bitswap tests less flaky. (@whyrusleeping)
    • Use TCP port zero for ipfs daemon in sharness tests. (@whyrusleeping)
    • Improve sharness tests on AppVeyor. (@chriscool)
    • Add a pause to fix timing on t0065. (@whyrusleeping)
    • Add support for arbitrary TCP ports to t0060-daemon.sh. (@noffle)
    • Make t0060 sharness test use TCP port zero. (@whyrusleeping)
    • Randomized ipfs stress testing via randor (@dignifiedquire)
    • Stress test pinning and migrations (@whyrusleeping)