Create 2020-05-21--design-discussion-subdomains-dns-label-limit.md

meeting-notes/2020/Q2/2020-05-21--design-discussion-subdomains-dns-label-limit.md

@@ -0,0 +1,105 @@
+# Design Discussion: Subdomain support for CIDs longer than 63 characters
+
+## Context
+
+Full explanation: https://github.com/ipfs/go-ipfs/issues/7318
+
+**tl;dr**  
+
+[RFC 1034](https://tools.ietf.org/html/rfc1034#page-7) states that "each node has a label, 
+which is zero to 63 octets in length", this means CID in subdomain can't be longer than 63 characters.
+
+If we split at 63rd character that will produce TLS error because it is impossible 
+to have TLS cert for more than one wildcard level (`*.*.ipns.dweb.link`), as clarified in [RFC6125](https://tools.ietf.org/html/rfc6125#page-27)
+and implemented by browser vendors.
+
+DNS length limits were known, but we did not include inlined ed25519 in evaluation at a time, and this recent development changes the timeline, so we need to have a fixes for longer CIDs and TLS for IPNS websites.
+
+The Goal of this meeting is to find the least invasive path forward.
+
+
+## Participants
+
+- @Stebalien
+- @lidel
+- @ribasushi
+- @aschmahmann
+
+
+
+## Options vs Tradeoffs
+
+### Generic fix
+
+Split after 63rd char: "Do it anyway" ?
+add support, if not for ed25519, it will be useful for longer CIDs (they won't have TLS anyway) 
+
+
+### IPNS/ed25519-specific fixes
+
+Anything we can do to give IPNS websites created with ed25519 TLS without warnings?
+
+| Fix    | TLS for IPNS | TLS for long CID  | touches libp2p | different text representation in subdomain than base32 |
+| --- | --- | --- | --- | --- |
+| custom multicodec for ed25519 keys | YES | NO | YES | NO |
+| Allow PeerIDs as multihash in base32 (only in `*.ipns` subdomains) | YES | NO | NO | YES |
+| PeerIDs as CIDv1 in base36 | YES | NO | YES | YES |
+| "sha256 split in two" idea [missing notes, sorry!] | YES | ? | NO | YES |
+
+## Notes
+
+- changing the way we represent keys in libp2p is expensive and may introduce problems in the future
+  - right now all key types represented as CIDs get `libp2p-key` codec
+    - this makes it easy to add support for different key types in the future without the need for updating CID/multicodec-related libraries
+  - if we introduce a custom multicodec for ed25519 this creates bad precedent of moving key type to multicodec layer: leaking abstractions and making it harder to support across implementations 
+- leaking browser limitations down the stack is something we want to avoid, looking for a surgical fix that is compatible with CIDv1 specs
+
+- base36 makes ed25519 fit in a single DNS label
+  - missing from multibase - keeps need to register 
+  - could be used only for IPNS
+- AI: test 63 character domain on Cloudflare
+    - OK: confirmed 63 works: https://abc012345678901234567890123456789012345678901234567890123456789.lidel.org/ 
+      ("github pages" proxied via Cloudflare that gives automatic TLS cert)
+- **Plan for enabling TLS for IPNS with ed25519 on subdomain gateways**
+    - implement support for base36 in go and js (multibase support)
+    - ship with go-ipfs 0.6 ad js-ipfs, giving us one release overlap
+    -  in parallel implement splitting after 63rd char (for longer CIDs that are not ed25519 keys)    
+    - switch to ed25519 keys in go-ipfs 0.7 by default
+    - mvp: make subdomain gateway redirect upgrade ed25519 to base36
+
+
+### Open questions
+
+Things TBD, with provisional decisions for MVP:
+
+- should we use base36:
+  - only for ed25519? (yes for mvp)
+  - only on subdomain gateways? (yes for mvp)
+    - always, or only when libp2p-key can't fit in a single DNS label? (only ed25519 for mvp)
+  - or maybe we should change the default text representation of ipns keys everywhere in IPFS?
+  - .. should this get bubbled up to libp2p? 
+    - if so, should we update https://github.com/libp2p/specs/blob/master/RFC/0001-text-peerid-cid.md ?
+    - or should we phase out base32 completely, and not use it in any API/CLI UX going forward
+
+### Side-discussions    
+
+-  ot: dnslink – how to add security?
+    -  (a) DNSSec - no-cigar: relies on root keys, hard to set up
+    -  (b) sign dnslink record and add TXT record with CID of signature
+        -  at some point in the future start redirecting to `.ipns-insecure.` if not signed
+
+-  ot: solve "what is IPNS" problem
+   - mostly renaming things in docs, to remove confusion in ecosystem
+   - propose a split on next core impl. meeting:  
+     - "IPNS" as a generic mutable namespace
+     - pick a new name for libp2p-key based subset of "IPNS" ("signed records", "IPRS", etc)
+
+-----
+
+### Update from 2020-07-14
+
+We decided to simplify our approach and merged [ED25519 support in subdomains](https://github.com/ipfs/go-ipfs/pull/7441)
+  - works with TLS certs
+  - defaults to base36 (instead of base32) in subdomains for all libp2p-keys
+  - b36 fix will be attempted for all long CIDs if original base is too long
+  - returns error if CID is still too long to be loaded via subdomain (this might be the case if you use custom hash functions eg. `sha-512`)