Skip to content

v1.18

Latest

Choose a tag to compare

@github-actions github-actions released this 17 Nov 13:12
· 3 commits to master since this release

Based on upstream DFRoot 4.6 - 4.7, plus one commit cherry-picked from upstream master (7a6513a6,
PR #185). versionCode 35, versionName 1.18.

Simplified Chinese (Removal of hardcoded language)

  • PR #145 (@edgeval), restored after the branch behind it was deleted.
  • It adds res/values/strings.xml, the file this app never had, with values-b+zh+Hans/strings.xml beside it: every android:text and contentDescription in activity_main.xml, and every literal MainActivity used to paint with, now resolves through R.string. res/xml/locales_config.xml plus android:localeConfig names the two locales the app answers in. The app's own word stays put through it: android:label reads @string/app_name, marked translatable="false", so the headline the update pill sizes itself against is the same word in either locale.
    r Screenshot_20261117_135721_DirtyFrag1

Android 12 can install: minSdk 32 -> 31

  • The floor of 32 came down with upstream's initial commit rather than from anything this fork needs, and
    nothing in the app calls an API above 31 - so it kept every Android 12 phone out at install time for no
    reason. IpSecManager and UdpEncapsulationSocket are API 28, the device-protected storage context the
    bootstrap reads is 24, and both splash-screen theme attributes arrived in API 31. The only place the old
    floor had actually been load-bearing was readLastLog(), which 1.17 moved off the API 34 Matcher
    overloads onto the ones carried since the first API level.

Falls back to libbase.so when the libc++ hook target is unusable

  • Cherry-picked from upstream (7a6513a6, PR #185) and it applied cleanly, because this fork's exp.c
    never touched patch_hook, struct PatchRestore or the tail of exploit(). When the hook target's
    offsets cannot be located, patch_hook returns 1 and the run stopped there; it now retries the same
    shellcode against android::base::LogMessage::~LogMessage() in libbase.so, a destructor that also
    runs when init reaps an orphan and often sits in a stretch with more executable slack than the
    libc++ ostream constructor.

Disabling modules no longer costs the root on a device that has none

  • disable_modules() aborted the bootstrap when opendir("/data/adb/modules") failed. A manager that
    has never had a module installed has no such directory - the ordinary state of a fresh KernelSU - and
    the abort was not a warning: main() touched /dev/dfme1 and returned before ksud was ever started,
    so the run ended with no root at all. That made the Modules card's OFF switch unusable on exactly the
    devices that reach for it, since disabling modules is what someone whose module is bootlooping the
    phone asks for first. ENOENT now reads as "there was nothing to disable", which is what the switch
    wants anyway, and the chain continues to ksud; any other errno is still a real failure (permission,
    I/O) and still aborts with dfme1.

"Hooked" stopped being reported as "rooted"

  • /dev/df is not a root marker. The shellcode creates it with O_CREAT|O_EXCL (libcxx.S) as a
    one-shot mutex and returns without running its payload on every later hit of the hook, so the node
    proves the payload started and says nothing about the stages behind it - module load, bootstrap, ksud,
    the su bind - any of which can fail on a boot where the node exists. Such a boot was described as
    rooted: ExploitRunner.run() returned 0 the moment it saw the node, which wrote last_run_success,
    which put "Verified" on the two-step bar. canRun() blocked the same boot on the same node with no
    explanation, so the pill read as "try again" for a retry that cannot work - an unprivileged app cannot
    unlink a node in root-owned devtmpfs, and nothing in this tree unlinks it.
  • RootState is now the single reader of the node and states the boot as ROOTED, HOOKED_NO_ROOT or
    NOT_HOOKED, and the UI stops collapsing the middle one into either neighbour: the bar draws patch 100%
    with root "No su" instead of an empty strip or a false Verified, the Run pill reads Reboot and stays
    locked with the reason in the log, and the verdict after a run weighs the return code from exp.c
    against what the boot can prove instead of trusting one of the two. BootReceiver says which half of
    "already hooked" it is before spending a wakelock on a payload that will not run.