zapret/docs/bsd.eng.md

## Supported versions

FreeBSD 11.x+ , OpenBSD 6.x+, partially MacOS Sierra+

Older versions may work or not. pfSense is not supported.

## BSD features

BSD does not have NFQUEUE. Similar mechanism - divert sockets.
In BSD compiling the source from nfq directory result in dvtws binary instead of nfqws.
dvtws shares most of the code with nfqws and offers almost identical parameters.

FreeBSD has 3 firewalls : IPFilter, ipfw and Packet Filter (PF). OpenBSD has only PF.

To compile sources in FreeBSD use `make`, in OpenBSD - use `make bsd`, in MacOS - use `make mac`.

Compile all programs : `make -C /opt/zapret`

Compile all programs with PF support : `make -C /opt/zapret CFLAGS=-DUSE_PF`

In FreeBSD enable PF only if you use it. Its undesirable if you don't.
PF is enabled automatically in OpenBSD and MacOS.

Divert sockets are internal type sockets in the BSD kernel. They have no relation to network addresses
or network packet exchange. They are identified by a port number `1..65535`. Its like queue number in NFQUEUE.
Traffic can be diverted to a divert socket using firewall rule.
If nobody listens on the specified divert port packets are dropped. Its similar to NFQUEUE without `--queue-bypass`.

`ipset/*.sh` scripts work with ipfw lookup tables if ipfw is present.

ipfw table is analog to linux ipset. Unlike ipsets ipfw tables share v4 an v6 addresses and subnets.

- If ipfw is absent scripts check LISTS_RELOAD config variable.
- If its present then scripts execute a command from LISTS_RELOAD.
- If LISTS_RELOAD=- scripts do not load tables even if ipfw exists.

PF can load ip tables from a file. To use this feature with ipset/*.sh scripts disable gzip file creation
using `GZIP_LISTS=0` directive in the `/opt/zapret/config' file.

BSD kernel doesn't implement splice syscall. tpws uses regular recv/send operations with data copying to user space.
Its slower but not critical.
tpws uses nonblocking sockets with linux specific epoll feature.
In BSD systems epoll is emulated by epoll-shim library on top of kqueue.

dvtws uses some programming HACKs, assumptions and knowledge of discovered bugs and limitations.
BSD systems have many limitations, version specific features and bugs in low level networking, especially for ipv6.
Many years have passed but BSD code still has 15-20 year artificial limiters in the code.
dvtws uses additinal divert socket(s) for layer 3 packet injection if raw sockets do not allow it.
It works for the moment but who knows. Such a usage is not very documented.

mdig and ip2net are fully compatible with BSD.

## FreeBSD

Divert sockets require special kernel module 'ipdivert'.
Write the following to config files :

/boot/loader.conf (create if absent) :
```
ipdivert_load="YES"
net.inet.ip.fw.default_to_accept=1
```

/etc/rc.conf :
```
firewall_enable="YES"
firewall_script="/etc/rc.firewall.my"
```

/etc/rc.firewall.my :

```
ipfw -q -f flush
```

Later you will add ipfw commands to `/etc/rc.firewall.my` to be reapplied after reboot.
You can also run zapret daemons from there. Start them with `--daemon` options, for example :
```
pkill ^dvtws$
/opt/zapret/nfq/dvtws --port=989 --daemon --dpi-desync=split2
```

To restart firewall and daemons run : `/etc/rc.d/ipfw restart`

Assume `LAN=em1`, `WAN="em0"`.

tpws transparent mode quick start.

For all traffic:
```
ipfw delete 100
ipfw add 100 fwd 127.0.0.1,988 tcp from me to any 80,443 proto ip4 xmit em0 not uid daemon
ipfw add 100 fwd ::1,988 tcp from me to any 80,443 proto ip6 xmit em0 not uid daemon
ipfw add 100 fwd 127.0.0.1,988 tcp from any to any 80,443 proto ip4 recv em1
ipfw add 100 fwd ::1,988 tcp from any to any 80,443 proto ip6 recv em1
/opt/zapret/tpws/tpws --port=988 --user=daemon --bind-addr=::1 --bind-addr=127.0.0.1
```

Process only table zapret with the exception of table nozapret :
```
ipfw delete 100
ipfw add 100 allow tcp from me to table\(nozapret\) 80,443
ipfw add 100 fwd 127.0.0.1,988 tcp from me to table\(zapret\) 80,443 proto ip4 xmit em0 not uid daemon
ipfw add 100 fwd ::1,988 tcp from me to table\(zapret\) 80,443 proto ip6 xmit em0 not uid daemon
ipfw add 100 allow tcp from any to table\(nozapret\) 80,443 recv em1
ipfw add 100 fwd 127.0.0.1,988 tcp from any to any 80,443 proto ip4 recv em1
ipfw add 100 fwd ::1,988 tcp from any to any 80,443 proto ip6 recv em1
/opt/zapret/tpws/tpws --port=988 --user=daemon --bind-addr=::1 --bind-addr=127.0.0.1
```

Tables zapret, nozapret, ipban are created by ipset/*.sh scripts the same way as in Linux.
Its a good idea to update tables periodically :
```
 crontab -e
```
 write the line : `0 12 */2 * * /opt/zapret/ipset/get_config.sh`

When using ipfw tpws does not require special permissions for transparent mode.
However without root its not possible to bind to ports <1024 and change UID/GID. Without changing UID tpws
will run into recursive loop, and that's why its necessary to write ipfw rules with the right UID.
Redirecting to ports >=1024 is dangerous. If tpws is not running any unprivileged process can
listen to that port and intercept traffic.


dvtws quick start.

For all traffic:
```
ipfw delete 100
ipfw add 100 divert 989 tcp from any to any 80,443 out not diverted not sockarg xmit em0
/opt/zapret/nfq/dvtws --port=989 --dpi-desync=split2
```

Process only table zapret with the exception of table nozapret :
```
ipfw delete 100
ipfw add 100 allow tcp from me to table\(nozapret\) 80,443
ipfw add 100 divert 989 tcp from any to table\(zapret\) 80,443 out not diverted not sockarg xmit em0
/opt/zapret/nfq/dvtws --port=989 --dpi-desync=split2
```

Reinjection loop avoidance.
FreeBSD artificially ignores sockarg for ipv6 in the kernel.
This limitation is coming from the ipv6 early age. Code is still in "testing" state. 10-20 years. Everybody forgot about it.
dvtws sends ipv6 forged frames using another divert socket (HACK). they can be filtered out using 'diverted'.
ipv4 frames are filtered using 'sockarg'.

PF in FreeBSD:
The setup is similar to OpenBSD, but there are important nuances.
1) Don't forget to build special PF-enabled version of tpws : make CFLAGS=-DUSE_PF
2) It's not possible to redirect to ::1. Need to redirect to the link-local address of the incoming interface.
Look for fe80:... address in ifconfig and use it for redirection target.
3) pf.conf syntax is a bit different from OpenBSD.
4) How to set maximum table size : sysctl net.pf.request_maxcount=2000000
5) The word 'divert-packet' is absent in the pfctl binary, divert-packet rules are not working.
'divert-to' is not the same thing. Looks like its not possible to use dvtws with PF in FreeBSD.

/etc/pf.conf
```
rdr pass on em1 inet6 proto tcp to port {80,443} -> fe80::31c:29ff:dee2:1c4d port 988
rdr pass on em1 inet  proto tcp to port {80,443} -> 127.0.0.1 port 988
```
then
```
/opt/zapret/tpws/tpws --port=988 --bind-addr=127.0.0.1 --bind-iface6=em1 --bind-linklocal=force
```

Its not clear how to do rdr-to outgoing traffic. I could not make route-to scheme work.


## OpenBSD

In OpenBSD default tpws bind is ipv6 only. to bind to ipv4 specify --bind-addr=0.0.0.0
Use --bind-addr=0.0.0.0 --bind-addr=::  to achieve the same default bind as in others OSes.

tpws for forwarded traffic only :

/etc/pf.conf
```
pass in quick on em1 inet  proto tcp to port {80,443} rdr-to 127.0.0.1 port 988 
pass in quick on em1 inet6 proto tcp to port {80,443} rdr-to ::1 port 988 
```
then
```
pfctl -f /etc/pf.conf
tpws --port=988 --user=daemon --bind-addr=::1 --bind-addr=127.0.0.1
```

Its not clear how to do rdr-to outgoing traffic. I could not make route-to scheme work.
rdr-to support is done using /dev/pf, that's why transparent mode requires root.

dvtws for all traffic:

/etc/pf.conf
```
pass in  quick on em0 proto tcp from port {80,443} no state
pass out quick on em0 proto tcp to   port {80,443} divert-packet port 989
```
then
```
pfctl -f /etc/pf.conf
./dvtws --port=989 --dpi-desync=split2
```

dwtws only for table zapret with the exception of table nozapret :

/etc/pf.conf
```
set limit table-entries 2000000
table <zapret> file "/opt/zapret/ipset/zapret-ip.txt"
table <zapret-user> file "/opt/zapret/ipset/zapret-ip-user.txt"
table <nozapret> file "/opt/zapret/ipset/zapret-ip-exclude.txt"
pass out quick on em0 inet  proto tcp to   <nozapret> port {80,443}
pass in  quick on em0 inet  proto tcp from <zapret>  port {80,443} no state
pass out quick on em0 inet  proto tcp to   <zapret>  port {80,443} divert-packet port 989 no state
pass in  quick on em0 inet  proto tcp from <zapret-user>  port {80,443} no state
pass out quick on em0 inet  proto tcp to   <zapret-user>  port {80,443} divert-packet port 989 no state
table <zapret6> file "/opt/zapret/ipset/zapret-ip6.txt"
table <zapret6-user> file "/opt/zapret/ipset/zapret-ip-user6.txt"
table <nozapret6> file "/opt/zapret/ipset/zapret-ip-exclude6.txt"
pass out quick on em0 inet6 proto tcp to   <nozapret6> port {80,443}
pass in  quick on em0 inet6 proto tcp from <zapret6>  port {80,443} no state
pass out quick on em0 inet6 proto tcp to   <zapret6> port {80,443} divert-packet port 989 no state
pass in  quick on em0 inet6 proto tcp from <zapret6-user>  port {80,443} no state
pass out quick on em0 inet6 proto tcp to   <zapret6-user> port {80,443} divert-packet port 989 no state
```
then
```
pfctl -f /etc/pf.conf
./dvtws --port=989 --dpi-desync=split2
```

divert-packet automatically adds the reverse rule. By default also incoming traffic will be passwed to dvtws.
This is highly undesired because it is waste of cpu resources and speed limiter.
The trick with "no state" and "in" rules allows to bypass auto reverse rule.

dvtws in OpenBSD sends all fakes through a divert socket because raw sockets have critical artificial limitations.
Looks like pf automatically prevent reinsertion of diverted frames. Loop problem does not exist.

OpenBSD forcibly recomputes tcp checksum after divert. Thats why most likely
dpi-desync-fooling=badsum will not work. dvtws will warn if you specify this parameter.

ipset scripts do not reload PF by default. To enable reload specify command in `/opt/zapret/config` :
```
LISTS_RELOAD="pfctl -f /etc/pf.conf"
```

Newer pfctl versions can reload tables only : 
```
pfctl -Tl -f /etc/pf.conf
```
But OpenBSD 6.8 pfctl is old enough and does not support that. Newer FreeBSD do.

Don't forget to disable gzip compression :
```
GZIP_LISTS=0
```
If some list files do not exist and have references in pf.conf it leads to error.
You need to exclude those tables from pf.conf and referencing them rules.
After configuration is done you can put ipset script :
```
 crontab -e
```
 then write the line : `0 12 */2 * * /opt/zapret/ipset/get_config.sh`

## MacOS

Initially, the kernel of this OS was based on BSD. That's why it is still BSD but a lot was modified by Apple.
As usual a mass commercial project priorities differ from their free counterparts.
Apple guys do what they want.
What everyone have updated long ago they keep old like a mammoth. But who cares ?

MacOS used to have ipfw but it was removed later and replaced by PF.
It looks like divert sockets are internally replaced with raw. Its possible to request a divert socket
but it behaves exactly as raw socket with all its BSD inherited + apple specific bugs and feature.
The fact is that divert-packet in `/etc/pf.conf` does not work. pfctl binary does not contain the word `divert`.

dvtws does compile but is useless.

After some efforts tpws works. Apple has removed some important stuff from their newer SDKs (DIOCNATLOOK) making
them undocumented and unsupported.

With important definitions copied from an older SDK it was possible to make
transparent mode working again. But this is not guaranteed to work in the future versions.

Another MacOS unique feature is root requirement while polling `/dev/pf`.

By default tpws drops root. Its necessary to specify `--user=root` to stay with root.

In other aspects PF behaves very similar to FreeBSD and shares the same pf.conf syntax.

In MacOS redirection works both for passthrough and outgoing traffic. Outgoing redirection requires route-to rule.
Because tpws is forced to run as root to avoid loop its necessary to exempt root from the redirection.
That's why DPI bypass will not work for local requests from root.

If you do ipv6 routing you have to get rid of "secured" ipv6 address assignment.

"secured" addresses are designed to be permanent and not related to the MAC address.

And they really are. Except for link-locals.

If you just reboot the system link-locals will not change. But next day they will change.Not necessary to wait so long.

Just change the system time to tomorrow and reboot. Link-locals will change. (at least they change in vmware guest)
Looks like its a kernel bug. Link locals should not change. Its useless and can be harmful. Cant use LL as a gateway.

The easiest solution is to disable "secured" addresses.

Outgoing connections prefer randomly generated temporary addressesas like in other systems. 

Put the string `net.inet6.send.opmode=0` to `/etc/sysctl.conf`.  If not present - create it.

Then reboot the system.

If you dont like this solution you can assign an additional static ipv6 address from `fd00::/8` range with `/128` prefix
to your LAN interface and use it as the gateway address.

tpws transparent mode only for outgoing connections.

/etc/pf.conf
```
rdr pass on lo0 inet  proto tcp from !127.0.0.0/8 to any port {80,443} -> 127.0.0.1 port 988
rdr pass on lo0 inet6 proto tcp from !::1 to any port {80,443} -> fe80::1 port 988
pass out route-to (lo0 127.0.0.1) inet proto tcp from any to any port {80,443} user { >root }
pass out route-to (lo0 fe80::1) inet6 proto tcp from any to any port {80,443} user { >root }
```
then
```
pfctl -ef /etc/pf.conf
/opt/zapret/tpws/tpws --user=root --port=988 --bind-addr=127.0.0.1 --bind-iface6=lo0 --bind-linklocal=force
```

tpws transparent mode for both passthrough and outgoing connections. en1 - LAN.

```
ifconfig en1 | grep fe80
        inet6 fe80::bbbb:bbbb:bbbb:bbbb%en1 prefixlen 64 scopeid 0x8 
```
/etc/pf.conf
```
rdr pass on en1 inet  proto tcp from any to any port {80,443} -> 127.0.0.1 port 988
rdr pass on en1 inet6 proto tcp from any to any port {80,443} -> fe80::bbbb:bbbb:bbbb:bbbb port 988
rdr pass on lo0 inet  proto tcp from !127.0.0.0/8 to any port {80,443} -> 127.0.0.1 port 988
rdr pass on lo0 inet6 proto tcp from !::1 to any port {80,443} -> fe80::1 port 988
pass out route-to (lo0 127.0.0.1) inet proto tcp from any to any port {80,443} user { >root }
pass out route-to (lo0 fe80::1) inet6 proto tcp from any to any port {80,443} user { >root }
```
then
```
pfctl -ef /etc/pf.conf
/opt/zapret/tpws/tpws --user=root --port=988 --bind-addr=127.0.0.1 --bind-iface6=lo0 --bind-linklocal=force --bind-iface6=en1 --bind-linklocal=force
```

Build from source : `make -C /opt/zapret mac`

`ipset/*.sh` scripts work.


### MacOS easy install

`install_easy.sh` supports MacOS

Shipped precompiled binaries are built for 64-bit MacOS with `-mmacosx-version-min=10.8` option.
They should run on all supported MacOS versions.
If no - its easy to build your own. Running `make` automatically installs developer tools.

!! **Internet sharing is not supported** !!

Routing is supported but only manually configured through PF.
If you enable internet sharing tpws stops functioning. When you disable internet sharing you may lose web site access.

To fix : `pfctl -f /etc/pf.conf`

If you need internet sharing use tpws socks mode.

launchd is used for autostart (`/Library/LaunchDaemons/zapret.plist`)

Control script : `/opt/zapret/init.d/macos/zapret`

The following commands fork with both tpws and firewall (if `INIT_APPLY_FW=1` in config)
```
/opt/zapret/init.d/macos/zapret start
/opt/zapret/init.d/macos/zapret stop
/opt/zapret/init.d/macos/zapret restart
```
Work with tpws only :
```
/opt/zapret/init.d/macos/zapret start-daemons
/opt/zapret/init.d/macos/zapret stop-daemons
/opt/zapret/init.d/macos/zapret restart-daemons
```
Work with PF only :
```
/opt/zapret/init.d/macos/zapret start-fw
/opt/zapret/init.d/macos/zapret stop-fw
/opt/zapret/init.d/macos/zapret restart-fw
```
Reloading PF tables :
```
/opt/zapret/init.d/macos/zapret reload-fw-tables
```

Installer configures `LISTS_RELOAD` in the config so `ipset/*.sh` scripts automatically reload PF tables.
Installer creates cron job for `ipset/get_config.sh`, as in OpenWRT.

start-fw script automatically patches `/etc/pf.conf` inserting there `zapret` anchors.
Auto patching requires pf.conf with apple anchors preserved.
If your `pf.conf` is highly customized and patching fails you will see the warning. Do not ignore it.
In that case you need to manually insert "zapret" anchors to your `pf.conf` (keeping the right rule type ordering) :
```
rdr-anchor "zapret"
anchor "zapret"
unistall_easy.sh unpatches pf.conf
```
start-fw creates 3 anchor files in `/etc/pf.anchors` : zapret,zapret-v4,zapret-v6.

- Last 2 are referenced by anchor `zapret`.
- Tables `nozapret`,`nozapret6` belong to anchor `zapret`.
- Tables `zapret`,`zapret-user` belong to anchor `zapret-v4`.
- Tables `zapret6`,`apret6-user` belong to anchor `zapret-v6`.

If an ip version is disabled then corresponding anchor is empty and is not referenced from the anchor `zapret`.
Tables are only created for existing list files in the ipset directory.