Currently if a process sets the sys.powerctl property, init adds this
property change into the event queue, just like any other property.
The actual logic to shutdown the device is not executed until init
gets to the action associated with the property change.
This is bad for multiple reasons, but explicitly causes deadlock in
the follow scenario:
A service is started with `exec` or `exec_start`
The same service sets sys.powerctl indicating to the system to
shutdown
The same service then waits infinitely
In this case, init doesn't process any further commands until the exec
service completes, including the command to reboot the device.
This change causes init to immediately handle sys.powerctl and reboot
the device regardless of the state of the event queue, wait for exec,
or wait for property conditions.
Bug: 37209359
Bug: 37415192
Test: Init reboots normally
Test: Update verifier can reboot the system
Change-Id: Iff2295aed970840f47e56c4bacc93001b791fa35
Starting zygote early requires cpuset to be initialized to all cores for
foreground cpuset. Change to expolit all cores by default at boot and
let device manufacturers override to proper values in device specific
init script.
Bug: 36576280
Test: marlin boot fast and checked cpuset during early boot
Change-Id: I2c1ce0630e58a7b04d1a453c6740d3f0bce9de9f
This CL disables module loading by writing 1 to
/proc/sys/kernel/modules_disabled when the property sys.boot_completed
is set to 1 by ActivityManagerService (at the broadcast of
PHASE_BOOT_COMPLETED).
Bug: 36515654
Test: tested on sailfish and verified that module loading is disabled in
userdebug and enabled in eng mode
Merged-In: Id38d34a6395966ab21e440614337c0cfca791ad0
(cherry picked from commit 6ed19d1675)
Change-Id: I2faa459b450b3a64f854c832c6b91cbf682bbffa
The class early_hal is essentially for the keymaster hal which needs
to be up before vold tries to unlock a storage encryption key (FDE or
FBE). The current position is too early in the boot process, because
on devices with legacy HAL the wrapper service uses system properties
to find the legacy HAL.
This patch moves the start of the early_hal class to the late-fs trigger
action which runs right after the system property action.
Test: Manually tested and update tested on bullhead, sailfish, and
another device.
Bug: 35764921
Change-Id: I34b45b85f8450e9ef18861535fdb2ee963df8c9b
(cherry picked from commit 1ad8d21947)
This class is used to start hals which are required in order to mount
data (for instance keymaster).
Test: works to start early_hal in internal
Bug: 36278706
Change-Id: If06908135e59b187683d8cf4cc4a00b490559081
This reverts commit 5011270225.
Now starting even earlier.
Reason for revert: Needed change, reverted b/c broken device.
Bug: 36278706
Test: original DOA device boots
Test: angler, bullhead, fugu, marlin, ryu
Test: all these devices boot with wipe
Test: all these devices boot with w/o wipe
Test: lshal shows all included services
Change-Id: Ic639aedf7834b1bd3a26d23d109727f5559317e9
Vendor owns /data/vendor.
HAL data must go in /data/vendor/hardware/.
Bug: 34980020
Test: build and boot AOSP Marlin. Observe /data/vendor and
/data/vendor/hardware exist and are empty.
Change-Id: I6fe96e3c76a10a5eb480ba10e10d4d006de56c12
Also start hals where hwservicemanager was started before.
Bug: 36278706
Test: internal marlin+angler boots
Change-Id: Ia55d2ef747fcbd086a09e1bb856824b14343118b
We have seen cases when threads in this cgroup not scheduled for more than
a few seconds in heavy workload situation and causing device freeze.
In Linux, multiple threads placed in ROOT cgroup cause the CPU resource to
be split per thread, rather than per group.
Currently we have many threads in ROOT cgroup, which makes threads in
bg_non_interactive cgroup to have "tiny" CPU resource other than 5%
quota defined.
Bug: 34193533
Test: on marlin
Change-Id: I7721f6196560fbedf6265e8b6db130cec9edefd7
Current init doesn't order the triggeres it scaned, and there is no
guarantee that general event trigger exec first and then event+property
triggers.
This CL will make sure netd started after post-fs-data trigger is done.
Bug: 35110957
Test: marlin boots
Change-Id: I7bb55af4e00f336682388abfa8a06eac2136b7d4
This change makes the init process to always attempts to enable
transient trigger for vibrator. This allows the exported properties to
change the ownership later at the on boot stage.
Test: device vibrates with the driver supports ledtrig-transient
Change-Id: If5eb7b7feaefe803f2ead634fbe4fc7b48da84ea
Signed-off-by: David Lin <dtwlin@google.com>
When native coverage is enabled, add a global GCOV_PREFIX
environment variable specifying that gcda files be output
with path prefix /data/local/tmp.
Bug: 35635587
Test: make NATIVE_COVERAGE=true; check init.environ.rc
Change-Id: I40972aea3ca3168d0687bdc93e9d4b7b3a1071b9
When vold mounts the encrypted /data partition, it first checks for and
kills processes that have open fds to the tmpfs placeholder at /data.
This resulted in a 20 second boot-time regression (vold's timeout period)
when tombstoned was started before vold.
Bug: http://b/34461270
Test: boot is faster, no messages from vold in console spew
Test: tombstoned still started by init
Change-Id: Ib5e9ddb05f40c9da852f00e103861c6ff2d94888
Also ensure that it uses the highest supported value, and
abort if the value is not above a minimum threshold.
Test: Tested against the curent kernel (maximum value of 2,
set to 0 by the kernel initially) and against a
modified kernel (maximum value of 4, set to 4 by the
kernel initially)
Bug: 30368199
Change-Id: I608db577258b68b390ffe96f452e1f7c0bc9ad8a
Since splitting tracefs out from debugfs, we now need to set the
attributes explicitly on tracing folder.
Test: Run adb shell atrace -c -b 16000 -t 5 gfx
Bug: 34197733
Change-Id: If2a962332b6d1ec227e289bdf952213756e26186
The SO_RCVBUFFORCE option requires the caller of uevent_create_socket()
to have net_admin capabilities.
Set platform default rcv/snd buffer sizes to 256kb that will always be
overridden by the device/target. However, it will allow
ueventd / healthd to use the uevent_create_socket() API w/o requiring
the net_admin capability.
Note: All devices override the buffer sizes according to the technology
maximum to at least ~8MB. So, the init.rc change here is to make sure
platform code can work w/o any overrides.
Test: no SELinux failures for healthd with 'net_admin' removed.
Bug: https://b/32733887
Change-Id: Ida346468cd550ad07901bf3a78ad508939849906
Signed-off-by: Sandeep Patil <sspatil@google.com>
Most notably, there's no longer any need to guess an end time.
Bug: http://b/23478578
Bug: http://b/33450491
Test: rebooted with bootcharting on/off
Change-Id: Icb7d6859581da5526d77dfc5aa4d57c9bfbfd7e2
It's 5 characters shorter, has no runtime costs, can be stored on a
read-only partition, and avoids problems like b/27262109 and b/27204904.
It allows makes some security hardening easier.
Bug: 27262109
Bug: 27204904
Bug: 32799236
Test: verified new symlink created and old one not present
Change-Id: Ief362e13569ad9c868a7f0f9c4dbd6a328c96c6b
Use to solve the problem of tracefs conditionally being mounted
under debugfs and needing restorecon'd without boot performance
penalty.
Also move skip-ce to a flag for consistency.
Test: Check that trace_mount has correct attributes after boot
Bug: 32849675
Change-Id: Ib6731f502b6afc393ea5ada96fa95b339f14da49
Start update_verifier in cache group to avoid dac_override
Bug: 30020920
Change-Id: I227b4a0e1c07e7b9bf209b432e9db02275ffe660
Test: mma
(cherry picked from commit 1e4635f0a8)
restorecon_recursive doesn't traverse filesystem boundaries. On
tracefs systems, tracing is a separate filesystem, so restorecon
this as well
Bug: 30963384
Test: Boot hikey, and check that there are no debugfs_tracing denials in dmesg
Change-Id: I24abd3ad80d2cfdab4f64fecee799fc0c24ed238
