Bypasses Googlevideo detection systems that relies on SNI. The package is for Linux only. The package is fully compatible with routers running OpenWRT.
The program offers binaries via Github Actions. You can find packages for OpenWRT under this link. You can check the architecture of your device with command grep ARCH /etc/openwrt_release. Also static binaries for PCs are available here.
The program is also compatible with routers driven by Entware (Keenetics/some Asuses). You can find binaries here. And here is an installation guide. Install the package with opkg. If you got read-only filesystem error you may unpack the binary manually or specify opkg path opkg -o <destdir>.
For Windows use GoodbyeDPI from ValdikSS (you can find how to use it for YouTube here) The same behavior is also implemented in zapret package for linux.
When you got the packet, you should install it. Go to your router interface and put it in via System-Software-install_package. Now the package is on the router. Goto System-Startup, restart firewall and start youtubeUnblock. You are done! To make it work you should register an iptables rule and install required kernel modules. The list of modules depends on your the version of OpenWRT and which firewall do you use (iptables or nftables). The common dependency id kmod-nfnetlink-queue but it is provided as dependency for another firewall packets.
So, if you are on iptables you should install: kmod-ipt-nfqueue, iptables-mod-nfqueue, kmod-ipt-conntrack-extra, iptables-mod-conntrack-extra and of course iptables user-space app should be available.
On nftables the dependencies are: kmod-nft-queue and kmod-nf-conntrack.
Next step is to add required rules. For nftables on OpenWRT rules comes out-of-the-box and stored under /usr/share/nftables.d/ruleset-post/537-youtubeUnblock.nft. All you need is install requirements and do /etc/init.d/firewall reload. If no, go to Firewall configuration.
Now we are ready to daemonize the application.
If you installed package from Github Actions or built it yourself with OpenWRT SDK, rc scripts are preinstalled. All you need is to do /etc/init.d/youtubeUnblock start.
Else copy owrt/youtubeUnblock.owrt to /etc/init.d/youtubeUnblock and put the program into /usr/bin/. (Don't forget to chmod +x both). Now run /etc/init.d/youtubeUnblock start.
You can also run /etc/init.d/youtubeUnblock enable to force OpenWRT autostart the program on boot, but I don't recommend this since if the packet has bug you may lose access to the router (I think you will be able to reset it with reset settings tricks documented for your router).
On local host make sure to change FORWARD to OUTPUT chain in the Firewall rulesets.
Copy youtubeUnblock.service to /usr/lib/systemd/system (you should change the path inside the file to the program position, for example /usr/bin/youtubeUnblock, also you may want to delete default iptables rule addition in systemd file to controll it manually). And run systemctl start youtubeUnblock.
On nftables you should put next nftables rules:
nft add rule inet fw4 mangle_forward tcp dport 443 ct original "packets < 20" counter queue num 537 bypass
nft insert rule inet fw4 output mark and 0x8000 == 0x8000 counter accept
On iptables you should put next iptables rules:
iptables -t mangle -A FORWARD -p tcp --dport 443 -m connbytes --connbytes-dir original --connbytes-mode packets --connbytes 0:19 -j NFQUEUE --queue-num 537 --queue-bypass
iptables -I OUTPUT -m mark --mark 32768/32768 -j ACCEPT
Note that above rules use conntrack to route only first 20 packets from the connection to youtubeUnblock. If you got some troubles with it, for example youtubeUnblock doesn't detect youtube, try to delete connbytes from rules. But it is an unlikely behavior and you should probably check your ruleset.
You can use --queue-balance with multiple instances of youtubeUnblock for performance. This behavior is supported via multithreading. Just pass --threads=n where n stands for an amount of threads you want to be enabled. The n defaults to 1. The maximum threads defaults to 16 but may be altered programatically. Note, that if you are about to increase it, here is 100% chance that you are on the wrong way.
Also DNS over HTTPS (DOH) is preferred for additional anonimity.
Here is a command aims to help you deterime whether it works or no:
curl -o/dev/null -k --connect-to ::google.com -k -L -H Host:\ mirror.gcr.io https://test.googlevideo.com/v2/cimg/android/blobs/sha256:6fd8bdac3da660bde7bd0b6f2b6a46e1b686afb74b9a4614def32532b73f5eaa
It should return bad speed without youtubeUnblock and good with it. With youtubeUnblock the speed should be the same as with next the command:
curl -o/dev/null -k --connect-to ::google.com -k -L -H Host:\ mirror.gcr.io https://mirror.gcr.io/v2/cimg/android/blobs/sha256:6fd8bdac3da660bde7bd0b6f2b6a46e1b686afb74b9a4614def32532b73f5eaa
Available flags:
--queue-num=<number of netfilter queue>- The number of netfilter queue youtubeUnblock will be linked to. Defaults to 537.--sni-domains=<comma separated domain list>|all- List of domains you want to be handled by sni. Use this string if you want to change default domains. Defaults togooglevideo.com,youtube.com,ggpht.com,ytimg.com. You can pass all if you want for every Client Hello to be handled.--fake-sni={0|1}This flag enables fake-sni which forces youtubeUnblock to send at least three packets instead of one with TLS ClientHello: Fake ClientHello, 1st part of original ClientHello, 2nd part of original ClientHello. This flag may be related to some Operation not permitted error messages, so befor open an issue refer to Troubleshooting for EPERMS. Defaults to 1.--fake-sni-seq-len=<length>This flag specifies youtubeUnblock to build a complicated construction of fake client hello packets. length determines how much fakes will be sent. Defaults to 1.--faking-strategy={ack,ttl}This flag determines the strategy of fake packets invalidation.ackspecifies that random sequence/acknowledgemend random will be set. This options may be handled by provider which uses conntrack with drop on invalid conntrack state firewall rule enabled.ttlspecifies that packet will be invalidated after--faking-ttl=nhops.ttlis better but may cause issues if unconfigured. Defaults toack--faking-ttl=<ttl>Tunes the time to live of fake sni messages. TTL is specified like that the packet will go through the TSPU and captured by it, but will not reach the destination server. Defaults to 8.--frag={tcp,ip,none}Specifies the fragmentation strategy for the packet. tcp is used by default. Ip fragmentation may be blocked by TSPU. None specifies no fragmentation. Probably this won't work, but may be will work for some fake sni strategies.--frag-sni-reverse={0|1}Specifies youtubeUnblock to send Client Hello fragments in the reverse order. Defaults to 1.--frag-sni-faked={0|1}Specifies youtubeUnblock to send fake packets near Client Hello (fills payload with zeroes). Defaults to 0.--fk-winsize=<winsize>Specifies window size for the fragmented TCP packet. Applicable if you want for response to be fragmented. May slowdown connection initialization.--seg2delay=<delay>- This flag forces youtubeUnblock to wait little bit before send the 2nd part of the split packet.--silent- Disables verbose mode.--no-gsoDisables support for Google Chrome fat packets which uses GSO. This feature is well tested now, so this flag probably won't fix anything.--threads=<threads number>Specifies the amount of threads you want to be running for your program. This defaults to 1 and shouldn't be edited for normal use. If you have performance issues, consult performance chaptr
If you have troubles with some sites being proxied, you can play with flags. For example, for someone --fake-sni=ttl works. You should specify proper --fake-sni-ttl=<ttl value> where ttl is amount of hops between you and DPI.
If you are on Chromium you may have to disable kyber (the feature that makes the TLS ClientHello very fat). I've got the problem with it on router, so to escape possibly errors it is better to just disable it: in chrome://flags search for kyber and switch it to disabled state. If your browser is using quic it may not work properly. Disable it in chrome in chrome://flags and in Firefox network.http.http{2,3}.enable(d) in about:config
EPERM may occur in a lot of places but generally here are two: mnl_cb_run and when sending the packet via rawsocket (raw_frags_send and send fake sni).
- mnl_cb_run Operation not permitted indicates that another instance of youtubeUnblock is running on the specified queue-num.
- rawsocket Operation not permitted indicates that the packet is being dropped by nefilter rules. In fact this is a hint from the kernel that something wrong is going on and we should check the firewall rules. Before dive into the problem let's make it clean how the mangled packets are being sent. Nefilter queue provides us with the ability to mangle the packet on fly but that is not suitable for this program because we need to split the packet to at least two independent packets. So we are using linux raw sockets which allows us to send any ipv4 packet. The packet goes from the OUTPUT chain even when NFQUEUE is set up on FORWARD (suitable for OpenWRT). So we need to escape packet rejects here.
- raw_frags_send EPERM: just make sure outgoing traffic is allowed (RELATED,ESTABLISHED should work, if not, go to step 3)
- send fake sni EPERM: Fake SNI is out-of-state thing and will likely corrupt the connection (the behavior is expected). conntrack considers it as an invalid packet. By default OpenWRT set up to drop outgoing packets like this one. You may delete nftables/iptables rule that drops packets with invalid conntrack state, but I don't recommend to do this. The step 3 is better solution.
- Step 3, ultimate solution. Use mark (don't confuse with connmark). The youtubeUnblock uses mark internally to avoid infinity packet loops (when the packet is sent by youtubeUnblock but on next step handled by itself). Currently it uses mark (1 << 15) = 32768. You should put iptables/nftables that ultimately accepts such marks at the very start of the filter OUTPUT chain:
iptables -I OUTPUT -m mark --mark 32768/32768 -j ACCEPTornft insert rule inet fw4 output mark and 0x8000 == 0x8000 counter accept.
Lets look from the DPIses side of view: All they have is ip and tcp information, higher-level data is encrypted. So from the IP header only IP address might be helpful for them. In tcp here is basically nothing. So they may handle IP addresses and process it. What's wrong? Google servers are on the way: It is very hard to handle all that infrastracture. One server may host multiple websites and it is very bad if them block, say Google Search trying to block googlevideo. But even if googlevideo servers have their own ip for only googlevideo purposes, here is a problem about how large is Google infrastracture and how much servers are here. The DPIs can't even parse normally all the servers, because each video may live on it's cache server. So what's else? Let's take a look at a TLS level. All information here is encrypted. All... Except hello messages! They are used to initialize handshake connections and hold tons of helpful information. If we talk about TLS v1.3, it is optimized to transfer as less information as possible unencrypted. But here is only one thing that may point us which domain the user wants to connect, the SNI extension. It transfers all domain names unencrypted. Exactly what we need! And DPIs may use this thing to detect google video connections and slow down them (In fact they are corrupting a tcp connection with bad packets).
So we aims to somehow hide the SNI from them. How?
- We can alter the SNI name in the tls packet to something else. But what's wrong with this? The server also uses SNI name for certificates. And if we change it, the server will return an invalid certificate which browser can't normally process, which may turn out to the MITM problem.
- We can encrypt it. Here are a lot of investigations about SNI, but the server should support the technique. Also ISPs may block encrypted SNI. Check this Wikipedia page
- So what else can we do with the SNI info? If we can't hide it, let's rely on DPIs weak spots. The DPI is an extremly high loaded machine that analyzes every single packet sent to the Internet. And every performance-impacted feature should be avoided for them. One of this features is IP packet fragmentation. We can split the packet in the middle of SNI message and post it. For DPI fragmentation involves too much overhead: they should store a very big mapping table which maps IP id, Source ip and Destination ip. Also note that some packets may be lost and DPI should support auto-clean of that table. So just imagine how much memory and CPU time will this cost for DPI. But fragments are ok for clients and hosts. And that's the base idea behind this package. I have to mention here that the idea isn't mine, I get in here after some research for this side. Here already was a solution for Windows, GoodbyeDPI. I just made an alternative for Linux.
You may read further in an yt-dlp issue page and in ntc party forum.
When the packet is joining the queue, the application checks sni payload to be googlevideo (right how the DPIs do), segmentates/fragmentates (both TCP and IP fragmentation techniques are supported) and posts the packet. Note that it is impossible to post two fragmented packets from one netfilter queue verdict. Instead, the application drops an original packet and makes another linux raw socket to post the packets in the network. To escape infinity loops the socket marks outgoing packets and the application automatically accepts it.
Before compilation make sure gcc, make, autoconf, automake, pkg-config and libtool is installed. For Fedora glibc-static should be installed as well.
Compile with make. Install with make install. The package include libnetfilter_queue, libnfnetlink and libmnl as static dependencies. The package requires linux-headers and kernel built with netfilter nfqueue support.
The package is also compatible with routers. The router should be running by linux-based system such as OpenWRT. You can build under openwrt with two options: first - through the SDK, which is preferred way and second is cross-compile manually with openwrt toolchain.
OpenWRT provides a high-level SDK for the package builds.
First step is to download or compile OpenWRT SDK for your specific platform. The SDK can be compiled according to this tutorial. Beside of raw source code of SDK, OpenWRT also offers precompiled SDKs for your router. You can find it on the router page. For example, I have ramips/mt76x8 based router so for me the sdk is on https://downloads.openwrt.org/releases/23.05.3/targets/ramips/mt76x8/ and called openwrt-sdk-23.05.3-ramips-mt76x8_gcc-12.3.0_musl.Linux-x86_64. You will need to install sdk requirements on your system If you have any problems, use docker ubuntu:24.04 image. Make sure to be a non-root user since some makesystem fails with it. Next, untar the SDK and cd into it. Do echo "src-git youtubeUnblock https://github.com/Waujito/youtubeUnblock.git;openwrt" >> feeds.conf, ./scripts/feeds update youtubeUnblock, ./scripts/feeds install -a -p youtubeUnblock, make package/youtubeUnblock/compile. Now the packet is built and you can import it to the router. Find it in bin/packages/<target>/youtubeUnblock/youtubeUnblock-<version>.ipk.
The precompiled toolchain located near the SDK. For me it is called openwrt-toolchain-23.05.3-ramips-mt76x8_gcc-12.3.0_musl.Linux-x86_64.tar.xz. When you download the toolchain, untar it somewhere. Now we are ready for compilation. My cross gcc asked me to create a staging dir for it and pass it as an environment variable. Also you should notice toolsuite packages and replace my make command with yours. STAGING_DIR=temp make CC=/usr/bin/mipsel-openwrt-linux-gcc LD=/usr/bin/mipsel-openwrt-linux-ld AR=/usr/bin/mipsel-openwrt-linux-ar OBJDUMP=/usr/bin/mipsel-openwrt-linux-objdump NM=/usr/bin/mipsel-openwrt-linux-nm STRIP=/usr/bin/mipsel-openwrt-linux-strip CROSS_COMPILE_PLATFORM=mipsel-buildroot-linux-gnu. Take a look at CROSS_COMPILE_PLATFORM It is required by autotools but I think it is not necessary. Anyways I put mipsel-buildroot-linux-gnu in here. For your model may be an automake cross-compile manual will be helpful. When compilation is done, the binary file will be in build directory. Copy it to your router. Note that an ssh access is likely to be required to proceed. sshfs don't work on my model so I injected the application to the router via Software Upload Package page. It has given me an error, but also a /tmp/upload.ipk file which I copied in root directory, chmod +x-ed and run.