This easy fix should improve performance with newer browsers especially
for applications that require many files to be sent over the wire
*cough*Nextcloud11*cough*
NGINX blog post about HTTP/2 support: https://www.nginx.com/blog/nginx-1-9-5/
All Cloudron containers need to have the nameserver 172.18.0.1. This was
being done at the daemon level, however since there are also iptables
rules restricting access to the nameserver from containers that aren't
on the Cloudron Docker network, this broke DNS for non-Cloudron
containers.
Since the DNS is only required for Cloudron containers in the first
place, this patch specifies 172.18.0.1 as the nameserver when Cloudron
creates a container and reverts the change at the daemon level
Explanation:
When proxying an HTTP request, nginx first fills up the memory buffers (set by proxy_buffer_size and proxy_buffers).
When these are full, it then writes them to a temporary file in batches of proxy_temp_file_write_size until it reaches proxy_max_temp_file_size.
When proxy_max_temp_file_size is not set, and a very large file is being served, it reaches the maximum of 1GB, and nginx begins to behave weirdly.
The goal here is to simply add a rate limit to prevent brute
force password attacks.
Covered services includes:
(public) http, https, ssh, smtp, msa, imap, sieve
(private) postgres, redis, mysql, ldap, mongodb. msa
The private limits are higher because some apps will create
a db connection for each page request. Some apps like mailtrain
will send out lots of emails etc.
Note that apps that use SSO are ratelimited by the ldap limit.
Part of #187
Also remove rate-limit middleware
Test using something like:
ab -v 1 -n 1000 -c 10 -s 5 -m POST https://my.<doamain>/api/v1/developer/login
Part of #187
some disk types do not contain proper partition tables like on time4vps
the type is simfs. On those fdisk fails to access the partition table,
thus being unable to determine the size of the volume.
df does only return the real usable disk space by the user, thus we
lower the 20GB threshold to 18
Fixes#275
Apps like nextcloud set their own security headers ending up with having
them set twice. I am not 100% sure if our headers should win or if we
should not inject headers with nginx if the upstream app sets them already.
This looks like the more permissive case where we simply enforce our
values, regardless what the apps sets.
This also fixes the nextcloud/owncloud security checks which were
failing because the header values were duplicated, which results in
string concatenation of values from same headers.
