filed
=====
Introduction
------------
"filed" is a simple HTTP server for serving local static files over HTTP from
Linux. It does the least amount of effort possible to get to the point of
handing the actual transfer over to the kernel.
It is multithreaded where every thread services a single concurrent client.
It attempts to reduce latency by caching open file descriptors as well.
It has no configuration file and supports only minimal configuration.
Operation is described in the manual page.
Usage
-----
The simplest usage of "filed" is to run it with no arguments. This will start
up an HTTP server listening on port 80 and sharing out the root filesystem for
the system as the current user. You probably do not want to share out your
root filesystem since that will expose information such as "/etc/passwd" to
anonymous access. You probably also don't want to do that as root since then
sensitive information such as "/etc/shadow" could be used to compromise the
system.
Therefore, the "--user" and "--root" options should be used to specify a user
for "filed" to run as in addition to a directory to change root (chroot(2)) to
prior to serving out files.
Example 1, insecure file sharing:
$ filed
Example 2, sharing sensitive information:
# filed
Example 3, sharing only a specific directory:
# filed --root /www --user nobody
Example 4, running as a daemon:
# filed --root /www --user nobody --daemon
Build Time Considerations
-------------------------
In general, it is best to leave all configuration to run-time so that a
compiled binary can do as much as possible without needing to be recompiled.
However, there are some things that are best done at compile time for
various reasons, especially in the name of performance or executable size.
filed admits those trade-offs are occasionally required and offers the
following tunables that can only be toggled during compile-time:
1. Logging (CFLAGS, -DFILED_DONT_LOG=1)
It is possible to disable ALL logging from filed. When logging is
completely disabled interlocks (mutexes) for the logging pointer are
not engaged and the logging functions are not compiled at all.
This results in a slightly smaller and faster binary
2. Debugging (CFLAGS, -DFILED_DEBUG=1)
This is an internal option and should only be used during development.
3. Differing HTTP semantics (CFLAGS, -DFILED_NONBLOCK_HTTP=1)
It is possible that some HTTP clients may not process the HTTP stream
being delivered if they cannot write to the HTTP stream itself. This
has not been observed yet, but it is possible. If these semantics are
needed (and they should not be) then they can be enabled with this
flag at the cost of performance
4. MIME Types (MIMETYPES)
For single-file convience "filed" compiles the mapping of file
extensions (the string in the filename following its last dot ("."))
into the executable. This mapping comes from a file in the format of
type1 type1_extension1 type1_extension2...
type2 type2_extension1 type2_extension2...
...
However it may not be desirable to include this mapping, or it may be
desirable to use your own mapping rather than the default one. This
can be done by specifying the MIMETYPES macro to "make". If no
mapping is desired, "/dev/null" may be specified.
Log Files
---------
Because "filed" relies on chroot(2) and setuid(2), log files cannot reliably
be re-opened. If you need log rotation then a second process, which can close
and re-open log files, must be used. Any process may be used for writing logs
but if the process does not support log rotation then it will not provide that
benefit. For example, if you wish to write logs to syslogd(8) you can use
logger(1), such as:
# ./filed --root /www --user nobody --log '|logger -t filed' --daemon