Although we do sometimes use single-line if blocks, I'm expanding this one to
make it easier to compare with the duplicated (but modified) code in
mod_websocket that we plan to de-duplicate one day.
This allows certain session-specific code that needs to run in the async
context, but is itself triggered outside of that context (e.g. timers), to
be queued.
An example of this is the session destruction code of mod_smacks, when the
hibernation timeout is reached.
This is the time after liveness checks are performed via the respective
read-timeout event, which by default involves sending a space character
but could be overridden e.g. as is done by mod_smacks.
Only advertised, unsure what we would do with it.
Ensures unavailable presence and other outgoing stanzas are sent.
Waiting for c2s sessions to close first before proceeding to disable and
close s2s ensures that unavailable presence can go out, even if it
requires dialback to complete first.
This channel binding method is now enabled when a hash is manually set in the
config, or it attempts to discover the hash automatically if the value is the
special string "auto".
A related change to mod_c2s prevents complicated certificate lookups in the
client connection hot path - this work now happens only when this channel
binding method is used. I'm not aware of anything else that uses ssl_cfg (vs
ssl_ctx).
Rationale for disabling by default:
- Minor performance impact in automatic cert detection
- This method is weak against a leaked/stolen private key (other methods such
as 'tls-exporter' would not be compromised in such a case)
Rationale for keeping the implementation:
- For some deployments, this may be the only method available (e.g. due to
TLS offloading in another process/server).
Many of these fall into a few categories:
- util.cache size, must be >= 1
- byte or item counts that logically can't be negative
- port numbers that should be in 1..0xffff
Closing the session invokes ondisconnect and session close logic,
including mod_smacks hibernation and the timer that destroys the session
after a timeout.
By closing the connection after it has been detached from the sessions
table it will no longer invoke the ondetach handler, which should
prevent the above problem.
The (still not published) XEP-xxxx: Stream Limits Advertisement uses the
element <max-bytes/> to advertise the maximum octet size of top level
stream elements. "size" was probably a leftover of an even earlier
version of the (Proto)XEP.
We decided that at the first stage, accounts that are disabled should
simply be prevented from authenticating, thus they should also be
prevented from having connected sessions. Since this is aimed to be a
moderation action for cases of abuse, they shouldn't be allowed to
continue being connected.
These changes to the API (hopefully the last) introduce a cleaner separation
between the user's primary (default) role, and their secondary (optional)
roles.
To keep the code sane and reduce complexity, a data migration is needed for
people using stored roles in 0.12. This can be performed with
prosodyctl mod_authz_internal migrate <host>
For this, various accessor functions are now provided directly on the
sockets, which reach down into the LuaSec implementation to obtain the
information.
While this may seem of little gain at first, it hides the implementation
detail of the LuaSec+LuaSocket combination that the actual socket and
the TLS layer are separate objects.
The net gain here is that an alternative implementation does not have to
emulate that specific implementation detail and "only" has to expose
LuaSec-compatible data structures on the new functions.
c2s/bosh/ws streams will default to 256KB, s2s and components to 512KB.
These values are aligned with ejabberd's default settings, which should reduce
issues related to inconsistent size limits between servers on the XMPP network.
The previous default (10MB) is excessive for any production server, and allows
significant memory usage by even unauthenticated sessions.
With mod_smacks, a session can outlive the connection, so whether the
event is fired should not hinge on whether the session is connected or
not.
Helps mod_smacks remove some state.
To allow modules to do things just before the session is closed. The
'(pre-)resource-unbind' events are fired after the stream is closed,
leaving no way to send something.
The :info() method has been observed to return nothing ... sometimes.
Unclear what causes it. Perhaps the TLS connection was shut down or
hasn't fully settled?
The LuaSec code has code paths that return nothing or nil, error, so it
is best to guard against it.
Could happen with the 'opportunistic_writes' setting, since then the
stream opening is written directly to the socket, which can in turn
trigger session destruction if the socket somehow got closed just after
the other sent their stream header.
Error happens later when it tries to `hosts[session.host == nil].events`
Matthew Wild