* Add flag to enable/disable timstamps for process log.
solves #2615
* Remove timestamp argument from log.Printf()
* Add log-timestamps=false to systemd unit
* Copy log flags
* Fix argument list
Clear rc_flags in start precmd. If these flags are still present during
start command, they are passed to daemon(8) instead of caddy(8).
Extract all options into $caddy_options environment variable.
In systemd 231
(4f10b80786/NEWS (L3558-L3565)),
ReadWriteDirectories was renamed ReadWritePaths.
In https://github.com/caddyserver/caddy/pull/2620/files, @aspeteRakete
renamed the directive in Caddy's example systemd unit.
However, this means that if anyone runs this sytemd unit on a version of
systemd older than 231, Caddy will go into a crash loop that hammers
Let's Encrypt's servers. That's because the ProtectSystem=full directive
prevents writes to all paths that aren't explicitly permitted, and older
systemd doesn't see any paths being permitted.
To maximize compatibility, I re-add the original ReadWriteDirectories
directive. Older systemd will read that; newer systemd will read the
newer directive. Both should ignore the directive they do not recognize.
Another approach to solve this problem would be to remove
ProtectSystem=true, originally introduced in da8ae9e5. That would reduce
the risk of similar breakages in the future. It would make for a slightly
less "exemplary" systemd unit, but I think it would still be adequate,
given that this unit runs caddy as "www-data", a user the presumably has
low privileges.
The Handler interface expects a first return value of 0 if headers have
already been written.
(cf. https://godoc.org/github.com/caddyserver/caddy/caddyhttp/httpserver#Handler)
When using http.Redirect, this is the case as http.Redirect does write
headers. When using Caddy with nested handlers, returning
http.StatusMovedPermanently could cause a wrong (200) response on a
non-slashy request on a directory name. Returning 0 ensures the
redirection will always take place.
Issue #2716 was caused by a misunderstanding regarding the filename to use when creating a custom build from source. These changes aim to make the instructions more explicit.
* Use syslog to manage caddy std{out,err} on FreeBSD
There is no good way to rotate the logfile created by the previous
FreeBSD rc.d script (it's the result of redirecting std{out,err} and
is held open by the shell).
This solves the problem by sending caddy's std{out,err} stream to
syslog, using the daemon command's builtin functionality.
It replaces the old `caddy_logfile` rc.conf variable with
`caddy_syslog_facility` (which defaults to 'local7') and
`caddy_syslog_level` (which defaults to 'notice').
By default, these messages will end up in /var/log/messages but can
be redirected as documented in the script's comments.
* Add info about rotating log with newsyslog
If you create a caddy specific logfile in `/var/log`, you should
rotate it.
This adds a bit of info to the dist/init/freebsd/README.md about
rotating that log file with newsyslog.
Currently, the instance waitgroup is decremented twice in `startServers()`: once when `Serve()` is finished and once when `ServePacket()` is finished. However, with a graceful shutdown, `Serve()` returns before the server has actually finished shutting down all active connections. This patch increases the wait group by one when the server is shut down so that the program only exits when all the server instances have finished serving their connections.
A random config is intended to be used only for solving TLS-ALPN
challenges; so we have to be sure to disable on-demand TLS so that
arbitrary names can't request certificates with another name's
on-demand config.
Slightly inconvenient because it uses int type and we use string, but
oh well. This fixes a bug related to setting -http-port and -https-port
flags which weren't being used by CertMagic in some cases.
This should help prevent hanging in some cases when the process is
restarted and tries to obtain or renew a certificate, for example, but
the lock remains from the previous shutdown (which was during the same
operation). Only works if the process is cleanly shut down with a signal
it can capture.
Each time the Caddyfile reloads and Caddy is restarted,
caddytls.NewConfig starts a goroutine for cleaning the
certificate storage. This goroutine ranges over a time.Ticker
channel; although Stop is called on this ticker, Stop does
not close the underlying channel so the goroutine never exits.
This change adds an additional channel that is listened to
in the certificate cleaning goroutine so it can exit
on restarts.
