Magisk's policy is to never allow 3rd party code to be loaded in the
zygote daemon process so we have 100% control over injection and hiding.
However, this makes it impossible for 3rd party modules to run anything
before process specialization, which includes the ability to modify the
arguments being sent to these original nativeForkAndXXX methods.
The trick here is to fork before calling the original nativeForkAndXXX
methods, and hook `fork` in libandroid_runtime.so to skip the next
invocation; basically, we're moving the responsibility of process
forking to our own hands.
On devices where the primary storage is slow to probe it makes sense to
wait forever for the system partition to mount, this emulates the
kernel's behaviour when waiting for rootfs on SAR if the rootwait
parameter is supplied.
This issue was encountered with some SD cards on the Nintendo Switch.
Previously, Magisk uses persist or cache for storing modules' custom
sepolicy rules. In this commit, we significantly broaden its
compatibility and also prevent mounting errors.
The persist partition is non-standard and also critical for Snapdragon
devices, so we prefer not to use it by default.
We will go through the following logic to find the best suitable
non-volatile, writable location to store and load sepolicy.rule files:
Unencrypted data -> FBE data unencrypted dir -> cache -> metadata -> persist
This should cover almost all possible cases: very old devices have
cache partitions; newer devices will use FBE; latest devices will use
metadata FBE (which guarantees a metadata parition); and finally,
all Snapdragon devices have the persist partition (as a last resort).
Fix#3179
This commit adds support for kernel initialized dm-verity on legacy SAR
devices.
Tested on a Pixel 2 XL with a kernel patch to initialize mappings
specified via the `dm=` kernel parameter even when an initramfs is used.
Due to changes in ec3705f2ed187863efc34af5415495e1ee7775d2, the app can
no longer communicate with the dameon through a socket opened on the
daemon side due to SELinux restrictions. The workaround here is to have
the daemon decide a socket name, send it to the app, have the app create
the socket server, then finally the daemon connects to the app through
the socket.
Introduce new domain `magisk_client` and new file type `magisk_exec`.
Connection to magiskd's always-on socket is restricted to magisk_client
only. Whitelisted process domains can transit to magisk_client through
executing files labelled magisk_exec. The main magisk binary shall be
the only file labelled as magisk_exec throughout the whole system.
All processes thus are no longer allowed to connect to magiskd directly
without going through the proper magisk binary.
Connection failures are silenced from audit logs with dontaudit rules,
so crazy processes which traverse through all unix domain sockets to try
connection can no longer check logcat to know the actual reason behind
EACCES, leaking the denied process policy (which is u:r:magisk:s0).
This also allows us to remove many rules that open up holes in
untrusted_app domains that were used to make remote shell work properly.
Since all processes establishing the remote shell are now restricted to
the magisk_client domain, all these rules are moved to magisk_client.
This makes Magisk require fewer compromises in Android's security model.
Note: as of this commit, requesting new root access via Magisk Manager
will stop working as Magisk Manager can no longer communicate with
magiskd directly. This will be addressed in a future commit that
involves changes in both native and application side.
- legacy devices brought up to Android 10 may now use a compressed dt in a hdr_v0 AOSP dt variant extra section, so detect, decompress and recompress this
- so far these have only been done using lz4 compression (latest format revision magic), e.g. LOS 17.1 victara (Moto X)
For match-all-type rules (e.g. "allow magisk * * *" used in Magisk),
we used to iterate and apply rules on all existing types. However, this
is actually unnecessary as all selinux types should have at least 1
attributes assigned to it (process types "domain", file context types
"file_type" etc.). This means in order to create rules that applies to
all types, we actually only need to create rules for all attributes.
This optimization SIGNIFICANTLY reduces the patched sepolicy that is
loaded into the kernel when running Magisk. For example on Pixel 4 XL
running Android R DP4, the sepolicy sizes are
patched (before) : 3455948
patched (after) : 843176
stock : 630229
The active sepolicy size actually impacts the performance of every single
operation in the operating system, because the larger the policies gets,
the longer it takes for the kernel to lookup and match rules.
It is possible that a module is breaking the device so bad that zygote
cannot even be started. In this case, system_server cannot start and
detect the safe mode key combo, set the persist property, and reboot.
Also on old Android versions, the system directly goes to safe mode
after detecting a key combo without rebooting, defeating the purpose of
Magisk's safe mode protection if we only check for the persist property.
Directly adding key combo check natively in magiskd allows us to enter
Magisk safe mode before the system is even aware of it.
When detecting device is booting as Safe Mode, disable all modules and
MagiskHide and skip all operations. The only thing that'll be available
in this state is root (Magisk Manager will also be disabled by system).
Since the next normal boot will also have all modules disabled, this can
be used to rescue a device in the case when a rogue module causes
bootloop and no custom recovery is available (or recoveries without
the ability to decrypt data).
Since we no longer need to add new properties in the device tree, and
all the patches we do removes strings, we can just directly patch
the flat device tree in-place, ignoring basically all the higher level
DTB structure and format to accomplish 100% compatibility.
Patching DTBs is proven to be difficult and problematic as there are
tons of different formats out there. Adding support for all the formats
in magiskboot has been quite an headache in the past year, and it still
definitely does not cover all possible cases of them out there.
There is another issue: fake dt fstabs. Some super old devices do not
have device trees in their boot images, so some custom ROM developers
had came up with a "genius" solution: hardcode fstab entries directly
in the kernel source code and create fake device tree nodes even if
Android 10+ init can graciously take fstab files instead (-_-) 。。。
And there is YET another issue: DTBs are not always in boot images!
Google is crazy enough to litter DTBs all over the place, it is like
they cannot make up their minds (duh). This means the dt fstabs can be
either concatnated after the kernel (1), in the DTB partition (2), in
the DTBO partition (3), in the recovery_dtbo section in boot images (4),
or in the dtb section in boot images (5). FIVE f**king places, how can
anyone keep up with that!
With Android 10+ that uses 2 stage inits, it is crutual for Magisk to
be able to modify fstab mount points in order to let the original init
mount partitions for us, but NOT switch root and continue booting. For
devices using dt for early mount fstab, we used to patch the DTB at
install time with magiskboot. However these changes are permanent and
cannot be restored back at reinstallation.
With this commit, Magisk will read dt fstabs and write them to ramdisk
at boot time. And in that case, the init binary will also be patched
to force it to NEVER use fstabs in device-tree. By doing so, we can
unify ramdisk based 2SI fstab patching as basically we are just patching
fstab files. This also means we can manipulate fstab whatever Magisk
needs in the future without the need to going through the headache that
is patching DTBs at installation.
Rewrite the whole module mounting logic from scratch.
Even the algorithm is different compared to the old one.
This new design focuses on a few key points:
- Modular: Custom nodes can be injected into the mount tree.
It's the main reason for starting the rewrite (needed for Android 11)
- Efficient: Compared to the existing implementation, this is the most
efficient (both in terms of computation and memory usage) design I
currently can come up with.
- Accurate: The old mounting logic relies on handling specifically every
edge case I can think of. During this rewrite I actually found some
cases that the old design does not handle properly. This new design is
architected in a way (node types and its rankings) that it should
handle edge cases all by itself when constructing mount trees.
Value of <dt>/fstab/<partition>/dev and <dt>/fstab/<partition>/type in official Android emulator ends with newline instead of \0, Magisk won’t be able to patch sepolicy and crash the system.
Signed-off-by: Shaka Huang <shakalaca@gmail.com>
The existing method for handling legacy SAR is:
1. Mount /sbin tmpfs overlay
2. Dump all patched/new files into /sbin
3. Magic mount root dir and re-exec patched stock init
With Android 11 removing the /sbin folder, it is quite obvious that
things completely break down right in step 1.
To overcome this issue, we have to find a way to swap out the init
binary AFTER we re-exec stock init. This is where 2SI comes to rescue!
2SI normal boot procedure is:
1st stage -> Load sepolicy -> 2nd stage -> boot continue...
2SI Magisk boot procedure is:
MagiskInit 1st stage -> Stock 1st stage -> MagiskInit 2nd Stage ->
-> Stock init load sepolicy -> Stock 2nd stage -> boot continue...
As you can see, the trick is to make stock 1st stage init re-exec back
into MagiskInit so we can do our setup. This is possible by manipulating
some ramdisk files on initramfs based 2SI devices (old ass non SAR
devices AND super modern devices like Pixel 3/4), but not possible
on device that are stuck using legacy SAR (device that are not that
modern but not too old, like Pixel 1/2. Fucking Google logic!!)
This commit introduces a new way to intercept stock init re-exec flow:
ptrace init with forked tracer, monitor PTRACE_EVENT_EXEC, then swap
out the init file with bind mounts right before execv returns!
Going through this flow however will lose some necessary backup files,
so some bookkeeping has to be done by making the tracer hold these
files in memory and act as a daemon. 2nd stage MagiskInit will ack the
daemon to release these files at the correct time.
It just works™ ¯\_(ツ)_/¯
Since SafetyNet CTS is impossible to achieve, leaving MagiskHide on
by default no longer serves a purpose.
For more details regarding the latest SafetyNet changes, please check:
https://twitter.com/topjohnwu/status/1237656703929180160https://twitter.com/topjohnwu/status/1237830555523149824
MagiskHide's functionality will continue to exist within the Magisk
project as it is still extremely effective to hide modifications in
userspace (including SafetyNet's basicIntegrity check).
Future MagiskHide improvements _may_ come, but since the holy grail
has been taken, any form of improvement is now a very low priority.
readlinkat() may return random value instead of the number of bytes placed in buf and crashing the system in two ways:
1. segmentation fault (buf[-7633350] = ‘\0’)
2. wrong link of watchdogd, resulting dog timeout
Confirmed working in ZenFone 2 x86 series, may fix#2247 and #2356
Signed-off-by: Shaka Huang <shakalaca@gmail.com>
Vendors are always adding “extra libraries” in /vendor/lib* for their own sake, in this case AS*S loaded with customized `libicuuc.so` for Zenf*ne 5z and led to the failure of dynamic loading libsqlite.so:
<quote>
db: dlopen failed: cannot locate symbol "UCNV_FROM_U_CALLBACK_ESCAPE_63" referenced by "/apex/com.android.runtime/lib64/libandroidicu.so"...
</quote>
Signed-off-by: Shaka Huang <shakalaca@gmail.com>
* Minor optimizations
Co-authored-by: John Wu <topjohnwu@gmail.com>
Some Motorola devices (Qualcomm kernel with CONFIG_MMI_DEVICE_DTBS
configuration enabled) need 1k of padding to the DTBs to allow for
environment variables to be runtime added by the bootloader.
Those extra paddings will be removed during the process of dtb patch,
devices won’t be able to boot-up and return to fastboot mode immediately
after flashed the flawed boot.img.
Credits to @shakalaca, close#2273
- some Samsung devices (e.g. Galaxy S5 SMG-900H) use a slightly different AOSP bootimg.h variant with `#define BOOT_NAME_SIZE 20` instead of 16
- since all known examples of these device images do not have anything in the NAME or CMDLINE fields, and the bootloader also accepts standard AOSP images, simply offset the SHA1/SHA256 detection by 4 bytes to avoid false positives from these images, remain an equally effective detection shortcut, and ensure a proper SHA1 checksum on repack
aosp-dtbhdt2-4offhash-seandroid-256sig-samsung_gs5-smg900h-boot.img
UNPACK CHECKSUM [00000000b11580f7d20f70297cdc31e02626def0356c82b90000000000000000]
REPACK CHECKSUM [73b18751202e56c433f89dfd1902c290eaf4eef3e167fcf03b814b59a5e984b6]
AIK CHECKSUM [b11580f7d20f70297cdc31e02626def0356c82b9000000000000000000000000]
This patch should result in a `magiskboot unpack -n boot.img; magiskboot repack boot.img` new-boot.img matching the AIK CHECKSUM above.
Previously, we use either BroadcastReceivers or Activities to receive
messages from our native daemon, but both have their own downsides.
Some OEMs blocks broadcasts if the app is not running in the background,
regardless of who the caller is. Activities on the other hand, despite
working 100% of the time, will steal the focus of the current foreground
app, even though we are just doing some logging and showing a toast.
In addition, since stubs for hiding Magisk Manager is introduced, our
only communication method is left with the broadcast option, as
only broadcasting allows targeting a specific package name, not a
component name (which will be obfuscated in the case of stubs).
To make sure root requests will work on all devices, Magisk had to do
some experiments every boot to test whether broadcast is deliverable or
not. This makes the whole thing even more complicated then ever.
So lets take a look at another kind of component in Android apps:
ContentProviders. It is a vital part of Android's ecosystem, and as far
as I know no OEMs will block requests to ContentProviders (or else
tons of functionality will break catastrophically). Starting at API 11,
the system supports calling a specific method in ContentProviders,
optionally sending extra data along with the method call. This is
perfect for the native daemon to start a communication with Magisk
Manager. Another cool thing is that we no longer need to know the
component name of the reciever, as ContentProviders identify themselves
with an "authority" name, which in Magisk Manager's case is tied to the
package name. We already have a mechanism to keep track of our current
manager package name, so this works out of the box.
So yay! No more flaky broadcast tests, no more stupid OEMs blocking
broadcasts for some bizzare reasons. This method should in theory
work on almost all devices and situations.
- support unpack without decompression to allow easy testing of magiskboot's header, structure and hashing handling by comparing repack checksum versus origbootimg
- make -n first to match repack
According to this comment in #1880:
https://github.com/topjohnwu/Magisk/issues/1880#issuecomment-546657588
If Linux recycled our PPID, and coincidentally the process that reused
the PPID is root, AND init wants to kill the whole process group,
magiskd will get killed as a result.
There is no real way to block a SIGKILL signal, so we simply make sure
our daemon PID is the process group leader by renaming the directory.
Close#1880
Usually, the communication between native and the app is done via
sending intents to either broadcast or activity. These communication
channels are for launching root requests dialogs, sending root request
notifications (the toast you see when an app gained root access), and
root request logging.
Sending intents by am (activity manager) usually requires specifying
the component name in the format of <pkg>/<class name>. This means parts
of Magisk Manager cannot be randomized or else the native daemon is
unable to know where to send data to the app.
On modern Android (not sure which API is it introduced), it is possible
to send broadcasts to a package, not a specific component. Which
component will receive the intent depends on the intent filter declared
in AndroidManifest.xml. Since we already have a mechanism in native code
to keep track of the package name of Magisk Manager, this makes it
perfect to pass intents to Magisk Manager that have components being
randomly obfuscated (stub APKs).
There are a few caveats though. Although this broadcasting method works
perfectly fine on AOSP and most systems, there are OEMs out there
shipping ROMs blocking broadcasts unexpectedly. In order to make sure
Magisk works in all kinds of scenarios, we run actual tests every boot
to determine which communication method should be used.
We have 3 methods in total, ordered in preference:
1. Broadcasting to a package
2. Broadcasting to a specific component
3. Starting a specific activity component
Method 3 will always work on any device, but the downside is anytime
a communication happens, Magisk Manager will steal foreground focus
regardless of whether UI is drawn. Method 1 is the only way to support
obfuscated stub APKs. The communication test will test method 1 and 2,
and if Magisk Manager is able to receive the messages, it will then
update the daemon configuration to use whichever is preferable. If none
of the broadcasts can be delivered, then the fallback method 3 will be
used.
Old Qualcomn devices have their own special QC table of DTB to
store device trees. Since patching fstab is now mandatory on Android 10,
and for older devices all early mount devices have to be included into
the fstab in DTBs, patching QCDT is crucial for rooting Android 10
on legacy devices.
Close#1876 (Thanks for getting me aware of this issue!)
The state of ROM A/B OTA addon.d-v2 support is an inconsistent mess currently:
- LineageOS builds userdebug with permissive update_engine domain, OmniROM builds userdebug with a more restricted update_engine domain, and CarbonROM builds user with a hybrid closer to Omni's
- addon.d-v2 scripts cannot function to the full extent they should when there is a more restricted update_engine domain sepolicy in place, which is likely why Lineage made update_engine completely permissive
Evidence for the above:
- many addon.d-v2 scripts only work (or fully work) on Lineage, see below
- Magisk's addon.d-v2 script would work on Lineage without issue, but would work on Carbon and Omni only if further allow rules were added for basic things like "file read" and "dir search" suggesting these ROMs' addon.d-v2 is severely limited
- Omni includes a /system/addon.d/69-gapps.sh script with the ROM itself (despite shipping without GApps), and with Magisk's more permissive sepolicy and no GApps installed it will remove important ROM files during OTA, resulting in a bootloop; the issue with shipping this script was therefore masked by Omni's overly restrictive update_engine sepolicy not allowing the script to function as intended
The solution:
- guarantee a consistent addon.d-v2 experience for users across ROMs when rooted with Magisk by making update_engine permissive as Lineage has
- hopefully ROMs can work together to come up with something standard for unrooted addon.d-v2 function
Directly read from urandom instead of using std::random_device.
libc++ will use iostream under-the-hood, which brings significant
binary size increase that is not welcomed, especially in magiskinit.
- while many newer devices cannot allow / (system partition) to be mounted rw due to compressed fs (e.g. erofs) or logical partitions, it should remain possible to alter rootfs files/directories on those that previously allowed it
The way how logical partition, or "Logical Resizable Android Partitions"
as they say in AOSP source code, is setup makes it impossible to early
mount the partitions from the shared super partition with just
a few lines of code; in fact, AOSP has a whole "fs_mgr" folder which
consist of multiple complex libraries, with 15K lines of code just
to deal with the device mapper shenanigans.
In order to keep the already overly complicated MagiskInit more
managable, I chose NOT to go the route of including fs_mgr directly
into MagiskInit. Luckily, starting from Android Q, Google decided to
split init startup into 3 stages, with the first stage doing _only_
early mount. This is great news, because we can simply let the stock
init do its own thing for us, and we intercept the bootup sequence.
So the workflow can be visualized roughly below:
Magisk First Stage --> First Stage Mount --> Magisk Second Stage --+
(MagiskInit) (Original Init) (MagiskInit) +
+
+
...Rest of the boot... <-- Second Stage <-- Selinux Setup <--+
(__________________ Original Init ____________________)
The catch here is that after doing all the first stage mounting, /init
will pivot /system as root directory (/), leaving us impossible to
regain control after we hand it over. So the solution here is to patch
fstab in /first_stage_ramdisk on-the-fly to redirect /system to
/system_root, making the original init do all the hard work for
us and mount required early mount partitions, but skips the step of
switching root directory. It will also conveniently hand over execution
back to MagiskInit, which we will reuse the routine for patching
root directory in normal system-as-root situations.
