Skip to content

Releases

NoMount Suite v1.3.212

Choose a tag to compare

@github-actions github-actions released this 13 Nov 16:43

v1.3.212 — engine v36 (needs a kernel rebuild)

  • Hide a rule from named apps: nomount vfs add --uid-excluded <pkg,...>. One entry covers that app under every user.
  • absorb can do the same for an absorbed APK. Opt in with /data/adb/nomount/apkscope.
  • A uid-scoped rule leaked to every uid once a permitted one looked it up. Legacy --uid too.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

ad5e378ba4cf0a11eba8b2bfcfdb58970c869e450a0c973637a06d1c9390d957  00_NoMount-Module-v1.3.212.zip
e2325130a43a40a2218b138001f2245615b70368e5a502a59a6be46c2271a06c  01_NoMount-Module-LKM-v1.3.212.zip

NoMount Suite v1.3.210

Choose a tag to compare

@Bouteillepleine Bouteillepleine released this 08 Nov 12:25

v1.3.210 — engine v34 (the engine fixes below need a kernel rebuild)

  • statmount/listmount handed a hide-listed app the module mounts that
    /proc/<pid>/mountinfo already hid. Both are filtered now, so the two views agree.
  • Every app granted root is concealed automatically, and a conceal entry naming a
    package that is gone is reported instead of sitting there.
  • A finding you have accepted stays out of the findings and is named once as a note.
  • The engine pins itself once a rule exists, so rmmod cannot free a live vtable.
  • A virtual parent another uid already owns is refused instead of stolen.
  • nm knobs reads the live knob state back out of the engine.
  • nomount check no longer calls a check unmeasured when it looked and there was
    nothing yet to look at.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

bb41e261375ea94c01cd54227c87f080bd588a2e68bca99c4ecaad5858bad200  00_NoMount-Module-v1.3.210.zip
e27c5f4eb5f7afdfd9f840d30de98178623f7a40a2dd75a269b32ff9176d23ba  01_NoMount-Module-LKM-v1.3.210.zip

NoMount Suite v1.3.208

Choose a tag to compare

@github-actions github-actions released this 06 Nov 05:59

v1.3.208 — engine v34 (unchanged)

  • nomount check from the WebUI could not run the four checks that need an app's uid
    — it crashed once per check instead. It now says why, and runs them through a
    watcher that can.
  • A check that measured nothing no longer overwrites one that measured everything.
  • The su probes use absolute paths, so they work when the caller has no PATH.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

ab2e5823b9d8e5ac0e553e6637241af3758e32cfe45bc277ec1f2ef98a23eab8  00_NoMount-Module-v1.3.208.zip
f311d25f67ace1f18cba517ac8d2e227fc1db829009bbd21e2e83329b5663181  01_NoMount-Module-LKM-v1.3.208.zip

NoMount Suite v1.3.207

Choose a tag to compare

@github-actions github-actions released this 01 Nov 12:08

v1.3.207 — engine v34 (the engine fixes below need a kernel rebuild)

  • The LKM's path and stat guards no longer fail open under load: the kretprobes that
    carry them have room for 64 concurrent calls per core, not 8.
  • unbind restores a SELinux label longer than 256 bytes instead of writing the
    adb_data_file default over it.
  • The package watcher no longer removes a lock another watcher owns.
  • boot.log survives a rotation a full /data could not write.
  • The LKM no longer calls path_put() from a kretprobe handler. Dropping the last
    reference there can sleep, and the dentries these two guards fire for are the
    ones dput() kills rather than keeps, so it could.
  • /proc/<pid>/maps and /proc/<pid>/map_files are matched by one renderer and
    one rule scan instead of two copies of the same substring test.
  • A path carrying any Unicode default-ignorable code point is refused for a hidden
    caller, not just the eight that had been seen in use. The built-in engine needs a
    kernel rebuild for this one.
  • The maps spoof parks the answer rather than a pointer to the inode it came from.
  • Uninstalling clears the engine's rule table, so it takes effect without a reboot,
    and it keeps binds.list when it could not find a binary to unbind with.
  • An engine module that loads but whose nm probe does not answer says so, instead
    of reporting that no module loads on this kernel.
  • The lite zip says it carries no engine module, and names the LKM zip. It bundles
    none by design, so "no bundled module loads on it either" read as a verdict on
    the kernel and sent people to reflash one.
  • A netlink socket that was denied is told apart from one the kernel does not
    have. nm already reported the difference and the installer discarded it, so an
    SELinux denial was answered with "flash a kernel built with CONFIG_NOMOUNT".
  • Another metamodule is recognised when its module.prop says metamodule=true,
    not only 1. The manager accepts either spelling, so the guard had been blind to
    the one upstream NoMount uses, and two installed at once is silent - whichever
    was flashed last takes the role and the loser's metamount.sh is never called.
  • Installing no longer deletes the bundled engine modules. It decided they were
    surplus from lsmod showing nothing, which is also what a loaded-and-hidden
    module shows.
  • The engine module is loaded on the metamodule path too, so its ordering against
    post-fs-data.sh no longer decides whether the first boot serves anything.
  • The LKM symbol gate reads the module instead of the modpost log, which stops
    after ten warnings - it had been checking ten symbols of a hundred and more.
  • The LKM resolves kallsyms_lookup_name by kprobe on every kernel. Below 5.7 it
    took the address directly, which android-4.19-stable does not export at all - so
    that build could never have linked there. Found by the gate above.
  • Neither build writes the module's name to the kernel log any more. The LKM's
    guards printed their "armed" lines a moment before unlinking the module from
    /proc/modules, and the Suite's own boot logging went to /dev/kmsg on every
    boot of both builds. boot.log carries the same lines; touch /data/adb/nomount/kmsg puts them back in dmesg, and hide=0 restores the
    kernel-side ones.
  • The subtree scan that places an injected inode can be preempted, instead of
    walking up to 8192 directories without yielding. The built-in engine needs a
    kernel rebuild for this one.
  • The engine declares its write on the mount it writes to. Nothing changes on a
    read-only stock mount, which is every rule today. The built-in engine needs a
    kernel rebuild for this one.
  • Releases publish SHA256SUMS. The install log says the payload is intact rather
    than unmodified, because the manifest ships inside the zip it checks.
  • hookless/README.md names what the four integration hunks cover and what they
    do not.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

569231a34f178e907dfa3877208360ee28cc9ff672a7c3e0591b8115965ba1bf  00_NoMount-Module-v1.3.207.zip
eb7ead27cf4de0633e4d71896ca9af6d9c009e9c77041ae7113a2cb7d7b5a4b6  01_NoMount-Module-LKM-v1.3.207.zip

NoMount Suite v1.3.206

Choose a tag to compare

@github-actions github-actions released this 31 Oct 07:47

v1.3.206 — engine v34 (unchanged)

  • Bundled engine modules ship without .BTF; some kernels reject it.
  • ko-loader output reaches the install log instead of being dropped.
  • A self-mount with a runtime target is seen, not just a literal path.
  • A self-mount note no longer promises absorb will take it over.
  • Install ends on INSTALLED BUT INACTIVE when no engine answers and no bundled module loads.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

NoMount Suite v1.3.205

Choose a tag to compare

@github-actions github-actions released this 28 Oct 21:16

v1.3.205 — engine v34 (unchanged)

  • The LKM keeps a hidden mapping out of maps, smaps, pagemap and mincore.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

NoMount Suite v1.3.204

Choose a tag to compare

@github-actions github-actions released this 26 Oct 20:19

v1.3.204 — engine v34

  • Module-rooted binds no longer appear in any mount table.
  • A path carrying a zero-width or ignorable code point is refused for a hidden caller, closes an Android/data bypass. The built-in engine needs a kernel rebuild for this one.
  • uidconceal and ghostpaths survive a module update.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

NoMount Suite v1.3.203

Choose a tag to compare

@github-actions github-actions released this 23 Oct 20:46

v1.3.203 — engine v34 (unchanged)

  • uidconceal: apps on this list read as idle to the callers already hidden from —
    getpriority answers -ESRCH, the same as a user owning no task.
  • ghostpaths: extra cloak entries beside the synced ones — absolute paths, and @name
    for abstract sockets, which refuse with -ECONNREFUSED.
  • Both need a kernel built with the _ghost patches; the LKM build cannot carry them.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

NoMount Suite v1.3.202

Choose a tag to compare

@github-actions github-actions released this 19 Oct 17:53

v1.3.202 — engine v34 (unchanged)

  • A third bottom bar: Compact, a short floating pill with icons only. Checks › Appearance
    picks between Dock, Capsule and Compact.

00_NoMount-Module for a kernel with NoMount built in, 01_NoMount-Module-LKM for one without.

NoMount Suite v1.3.201

Choose a tag to compare

@github-actions github-actions released this 18 Oct 18:31
  • New dock bottom bar. The floating capsule is still there, under Checks › Appearance.
  • Module rows line up, and mountless is quiet so a module that owns a mount stands out.