github DanielLavrushin/b4 v1.93.0

33 minutes ago

[1.93.0] - 2026-11-21

Added

  • IP Block Detection can Collect detected IPs: when two connections in a row to an address stall, the address joins a pool for sets with Use collected IPs, listed in Collected IPs on the Dashboard. Before installing an earlier version, turn off IP Block Detection in the sets that collect, and turn off sets with a Port Filter whose only target is Use collected IPs: that version would block and ban detected addresses, or apply such a set to every connection on those ports. (#618, thanks @SoloIl)
  • Reuse WSSIZE connection warms up a connection to a set's IP target with its Real TCP window (wssize) from zapret2, then sends the device's HTTPS from the same router port with the normal window; if the warm-up fails, the set's own strategy applies and Traffic marks the connection fallback. (#618, thanks @SoloIl)
  • A Telegram set under Telegram upstream (shared) takes b4's own connections to Telegram on every route, the CF proxy domains and the Worker included; Create a Telegram set adds one, and a Telegram set badge marks it on the Sets page. #615, #639, #640
  • After an Update or Upload on the Geo Databases page, the card lists the sets with categories that match nothing in the new file, and the set editor marks a category absent from the file in red with the reason on hover. #620
  • Discovery and the watchdog's heals run under the TUN engine, routers without NFQUEUE included. Under TUN they test IPv4 only, skip strategies with Incoming Mode or RST Injection Protection, and cannot test a set with its own DSCP value. #528
  • The Payloads page shows what each payload is (its kind, size, server name, whether it fits one packet, where it came from and the sets that use it), lists the payloads built into b4 under Built In, and Regenerate rebuilds a generated payload under the same name; in a set, the My own Payload File list is sorted and described, and the UDP tab offers Use the file size. #651
  • A set created on the Sets page can start from your own values instead of b4's defaults, set under Settings, Core, Other or with Use for new sets in a set's menu; imported and shared sets keep b4's defaults.

Changed

  • Settings left out when a set is sent through the API get their default values, and a change that leaves out the set's name, targets or on/off state is refused. #527
  • The built-in fake payload Preset: DuckDuckGo is replaced by Preset: VK, a 517-byte ClientHello for vk.com, and sets that used it send the new one; MCP names it vk, and the Payloads page no longer calls generated payloads "optimized". #648
  • Randomize TLS Random also gives each fake a random session ID, as zapret's rnd does. #649
  • Fake Packet Size on a set's UDP tab goes up to 1472 bytes, what one 1500-byte packet carries, and a saved value above that is lowered to 1472 when b4 starts. #645
  • With a splitting strategy such as Combo, the second packet of a ClientHello sent in two packets waits for the first, which adds one First Segment Delay (30 ms by default). A set that worked only while the second packet overtook the first, such as one with a Preset: STUN fake and no splitting, can stop working; run Discovery again for its sites. #653
  • Settings in the web interface keep one short line of help; the rest opens from an info button beside the setting, with Learn more and Ask AI, warnings show only while they apply, rarely changed settings sit under Advanced, and a set's TCP, Faking tab shows only the settings its chosen modes use.
  • In a set's Targets tab, Bypass Domains and Bypass IPs are renamed Domains and IP Addresses, and the panels inside drop the word Bypass.

Removed

  • The /api/capture/probe API, unused by the web interface, is removed, and with it a slow memory leak when clients sent their ClientHello in two packets. #646
  • Let sets process Worker connections under Telegram upstream (shared) is gone: a Telegram set without routing applies its strategy to b4's Worker connections, and without one no set processes them. #639

Fixed

  • After b4 stopped, a router with its own connection-tracking settings, such as OpenWrt, was left with the kernel defaults instead, and in rare cases with b4's settings. (#613, thanks @mvanhorn; #575)
  • Test connection under Telegram upstream (shared) checked its routes one after another, so with CF proxy fallback on a run took about a minute and showed nothing until it ended. #616
  • A set added or changed through the API by a script, not in the web interface, could start sending fake packets nobody had turned on, and changing just one setting this way wiped the set's name and targets. #527
  • The log misreported where a set's traffic went: when the set's interface went away it said the traffic would not leave by the normal uplink, although without Kill switch it did, and the Logs page showed routed-> for a tunnelled set's connection that left by another interface, as for a KeeneticOS device in an internet access policy.
  • b4's own connections that a set processes, such as its direct connections to Telegram, broke when a set routed through an interface held their address.
  • b4 could crash, again after every restart, at the first packet of a set whose UDP Fake Packet Size was below zero (set through the API or the config file) or whose fake from a zapret import was close to 64 KB. #645
  • On a 1500-byte link a fake that did not fit one packet was never sent, with at most an occasional log line: a UDP Fake Packet Size over 1472, a payload file over 1448 bytes without Match Fake Size, or an MD5 fake with Match Fake Size on a ClientHello of two packets. #645
  • Randomize TLS Random left the fakes of Enable Decoy, SYN Fake Packets and Match Fake Size unchanged, and Duplicate Session ID did nothing for a ClientHello longer than one packet, as current browsers send. #649
  • A Past Sequence or Random Sequence fake with a Sequence Offset smaller than the fake reached the server as the start of the connection's data, a Match Fake Size fake for a site whose name ends the payload's own, such as google.com with Preset: Google (classic), copied the real ClientHello, and a zapret import reported rndsni as matched exactly and left out zapret's default rnd and dupsid. #649
  • A payload generated on the Payloads page or by Generated from Domain had three malformed fields that a strict parser rejects. Payloads generated before this version keep them until Regenerate on the Payloads page rebuilds them. #648
  • When a browser sent its ClientHello in two packets, as current Chrome and Firefox do, the second packet often left before the fake and the first one, so whether a set worked depended on timing and on the router's load. #653
  • Deleting, replacing or uploading a payload on the Payloads page reached the sets that use it only after the next configuration save or restart. #647
  • A payload whose file was gone looked like any other, the set editor showed no warning for a missing or unchosen payload file while the set sent Preset: Google (classic), the My own Payload File list offered QUIC files that then went out inside TCP, and Fake Packet TLS Modification was shown for fakes it cannot change, such as All Zeros. #650
  • On the Payloads page an upload could overwrite another payload, because names differing only in punctuation or Cyrillic letters shared one file and a second file kept the first one's name; deleting a name with &, #, + or % could delete another payload, a blank name could not be deleted, a 64 KB file was refused, errors gave no reason, and Generate kept a typed port and a raw Cyrillic domain. #647, #650
  • A payload file cut short by a full disk replaced the old one, an empty payload file was sent as Preset: Google (classic) with no error in the log, and after a backup restore the next payload change lost the restored payload list. #647
  • Descriptions on a set's TCP, Splitting tab did not match what b4 does: TLS Record Splitting was said to write several TLS records, First-Byte Desync to wait at least 100 ms, and Combo with every split point off to send the packet whole.
  • In the set editor, the TCP tab's warning that the Port Filter leaves out port 443 could appear when the saved filter covered 443 and stay hidden when it did not, a set in Block (blackhole) mode with Domain-only matching was warned that it routes nothing, and the First-Byte Desync diagram showed the delay as typed, not the one b4 uses.

Security

  • A restored backup with a crafted payload list could make the Payloads page show and delete files outside its folder. #647

What's Changed

  • fix(tables): keep each network namespace's conntrack sysctl snapshot to itself by @mvanhorn in #613
  • fix(mtproto): the Telegram connection test checks its routes three at a time (#616) by @DanielLavrushin in #617
  • fix(ui): say which set categories a replaced geo file lacks (#620) by @DanielLavrushin in #621
  • fix(sets): a partial REST set body takes the set defaults, not empty values (#527) by @DanielLavrushin in #622
  • feat(mtproto): a Telegram set for b4's own Telegram connections (#615, phases 1 and 2) by @DanielLavrushin in #619
  • ci(release): pre-releases take their notes from [Unreleased], checked before the build by @DanielLavrushin in #637
  • fix(routing): tag b4's routes on newer kernels, and correct the gone-interface warning and the routed-> label by @DanielLavrushin in #632
  • feat(labs): a RouterOS guest with the MikroTik community b4 container by @DanielLavrushin in #638
  • feat(discovery): run Discovery and watchdog heals under the TUN engine without NFQUEUE by @DanielLavrushin in #628
  • test(tables): a KeeneticOS policy stand-in, tests for policy devices and a lost output interface by @DanielLavrushin in #624
  • feat(telegram): sets may process b4's CF proxy connections, and the Sets page marks the Telegram set (#639, #640) by @DanielLavrushin in #644
  • fix(fakes): fake sizes, the TLS generator, tls_mod coverage and split-hello order by @DanielLavrushin in #657
  • feat(telegram): the Telegram set handles all of b4's Telegram connections in NFQUEUE and TUN mode by @DanielLavrushin in #656
  • feat(toolkit): b4-mcp.mjs, a command-line client for b4's MCP server by @DanielLavrushin in #661
  • fix(payloads): the Payloads page, payload storage and the payload pickers by @DanielLavrushin in #660
  • Add collected IP targets and WSSIZE connection reuse by @SoloIl in #618
  • feat(sets): new sets start from the user's own values, the set defaults stay frozen by @DanielLavrushin in #663
  • fix(ui): version block in the side menu fits both locales by @DanielLavrushin in #664
  • feat(ui): one help button per setting, short helper lines and messages shown when they apply by @DanielLavrushin in #662
  • Refactor code structure for improved readability and maintainability by @DanielLavrushin in #665
  • fix(tests): update stubBinaries calls to include "ipset" for accurate… by @DanielLavrushin in #666

New Contributors

Full Changelog: v1.92.0...v1.93.0

Don't miss a new b4 release

NewReleases is sending notifications on new releases.