The loop which performs renewals in the background obtains a read lock
on the certificate cache map, so that it can be safely iterated. Before
this fix, it would obtain the renewals in the read lock. This has been
fine, except that the TLS-SNI challenge, when invoked after Caddy has
already started, requires adding a certificate to the cache. Doing this
requires an exclusive write lock. But it cannot obtain a write lock
because a read lock is obtained higher in the stack, while the loop
iterates. In other words, it's a deadlock.
I was able to reproduce this issue consistently locally, after jumping
through many hoops to force a renewal in a short time that bypasses
Let's Encrypt's authz caching. I was also able to verify that by queuing
renewals (like we do deletions and OCSP updates), lock contention is
relieved and the deadlock is avoided.
This only affects background renewals where the TLS-SNI(-01) challenge
are used. Users report seeing strange errors in the logs after this
happens ("tls: client offered an unsupported, maximum protocol version
of 301"), but I was not able to reproduce these locally. I was also not
able to reproduce the leak of sockets which are left in CLOSE_WAIT.
I am not sure if those are symptoms of running in production on Linux
and are related to this bug, or not.
Either way, this is an important fix. I do not yet know the ripple
effects this will have on other symptoms we've been chasing. But it
definitely resolves a deadlock during renewals.
* Initial concept for pluggable storage (sans tests and docs)
* Add TLS storage docs, test harness, and minor clean up from code review
* Fix issue with caddymain's temporary moveStorage
* Formatting improvement on struct array literal by removing struct name
* Pluggable storage changes:
* Change storage interface to persist all site or user data in one call
* Add lock/unlock calls for renewal and cert obtaining
* Key fields on composite literals
These changes span work from the last ~4 months in an effort to make
Caddy more extensible, reduce the coupling between its components, and
lay a more robust foundation of code going forward into 1.0. A bunch of
new features have been added, too, with even higher future potential.
The most significant design change is an overall inversion of
dependencies. Instead of the caddy package knowing about the server
and the notion of middleware and config, the caddy package exposes an
interface that other components plug into. This does introduce more
indirection when reading the code, but every piece is very modular and
pluggable. Even the HTTP server is pluggable.
The caddy package has been moved to the top level, and main has been
pushed into a subfolder called caddy. The actual logic of the main
file has been pushed even further into caddy/caddymain/run.go so that
custom builds of Caddy can be 'go get'able.
The HTTPS logic was surgically separated into two parts to divide the
TLS-specific code and the HTTPS-specific code. The caddytls package can
now be used by any type of server that needs TLS, not just HTTP. I also
added the ability to customize nearly every aspect of TLS at the site
level rather than all sites sharing the same TLS configuration. Not all
of this flexibility is exposed in the Caddyfile yet, but it may be in
the future. Caddy can also generate self-signed certificates in memory
for the convenience of a developer working on localhost who wants HTTPS.
And Caddy now supports the DNS challenge, assuming at least one DNS
provider is plugged in.
Dozens, if not hundreds, of other minor changes swept through the code
base as I literally started from an empty main function, copying over
functions or files as needed, then adjusting them to fit in the new
design. Most tests have been restored and adapted to the new API,
but more work is needed there.
A lot of what was "impossible" before is now possible, or can be made
possible with minimal disruption of the code. For example, it's fairly
easy to make plugins hook into another part of the code via callbacks.
Plugins can do more than just be directives; we now have plugins that
customize how the Caddyfile is loaded (useful when you need to get your
configuration from a remote store).
Site addresses no longer need be just a host and port. They can have a
path, allowing you to scope a configuration to a specific path. There is
no inheretance, however; each site configuration is distinct.
Thanks to amazing work by Lucas Clemente, this commit adds experimental
QUIC support. Turn it on using the -quic flag; your browser may have
to be configured to enable it.
Almost everything is here, but you will notice that most of the middle-
ware are missing. After those are transferred over, we'll be ready for
beta tests.
I'm very excited to get this out. Thanks for everyone's help and
patience these last few months. I hope you like it!!
