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
(cherry picked from commit 98ad32a967)
Set ro.boot.avb_version to "AVB_VERSION_MAJOR.AVB_VERSION_MINOR".
During Treble OTA match, the major version must be the same as that in
the avb metadata on disk, while the minor version can be equal or
greater to that in the avb metadata on disk.
See how avb versioning work on the following link:
https://android-review.googlesource.com/#/c/342757/
Also renames AvbHashtreeDisabled() -> hashtree_disabled().
Bug: 35322304
Test: Early mount with AVB, checks [ro.boot.avb_version]: [1.0] exists.
Test: Not enable AVB, checks [ro.boot.avb_version] doesn't exists.
Change-Id: I5aaf476ca53c4fe817779518ba14b68ebcfdc6d6
Merged-In: I5aaf476ca53c4fe817779518ba14b68ebcfdc6d6
(cherry picked from commit 1a898c25f9)
libavb requires verifying AVB metadata on all verified partitions at
once. For example, /vbmeta, /boot, /system and /vendor. We need to
invoke device_init() for those partitions even if we only want to early
mount some of them, like /vendor and /system.
This CL gets all AVB partitions and the early mount partitions from
device tree through "firmware/android/vbmeta" and "firmware/fstab",
respectively. The following is an example to early mount /vendor
partition on bullhead:
firmware {
android {
compatible = "android,firmware";
vbmeta {
compatible = "android,vbmeta";
parts = "boot,system,vendor";
by_name_prefix="/dev/block/platform/soc.0/f9824900.sdhci/by-name"
};
fstab {
compatible = "android,fstab";
vendor {
compatible = "android,vendor";
dev = "/dev/block/platform/soc.0/f9824900.sdhci/by-name/vendor";
type = "ext4";
mnt_flags = "ro,barrier=1,inode_readahead_blks=8";
fsmgr_flags = "wait,avb";
};
};
};
};
Bug: 33254008
Test: early mount /vendor with vboot 2.0 (AVB) on bullhead
Test: early mount /system without dm-verity on bullhead
Test: early mount /vendor with vboot 1.0 on sailfish
Change-Id: I89a1f77c97124f309346b33d9e700544b92ecf05
Merged-In: I89a1f77c97124f309346b33d9e700544b92ecf05
(cherry picked from commit 8bba52fc4b)
It's possible, in the event of a platform update, for the platform
SELinux policy to change from the policy on which the vendor SELinux
policy was originally based. In this case, a different mapping file
to bridge the differences between the new policy and the old needs to
be selected.
Make init choose which mapping policy file to use based on the version
reported in /vendor/etc/selinux/plat_sepolicy_vers.txt.
Bug: 36783775
Test: Force compilation of sepolicy on-device with mapping file changed
to new location and name, using the value reported on /vendor.
Change-Id: I63c883ccb79dd31c92dabe44a55c4ab50a3735e6
Bug: 36783775
Test: boot device with matching sha256 and non-matching and verify that
device boots and uses either precompiled or compiled policy as needed. Also
verify that mapping_sepolicy.cil has moved.
Change-Id: I9c3df2e7ed3f0eadd98e98529ebed360fec66dba
Slowly try to decouple property_service.cpp from the rest of init.
Test: Boot bullhead
Change-Id: I267ae0b057bca0bf657b97cb8bfbb18199282729
(cherry picked from commit ccf23537ee)
Exec services may also want to set other service flags such as
priority. Instead of expanding the exec syntax to handle this, create
a new command, exec_start, that will treat an existing service
definition as an exec service. The new exec_start command will start
the service then halt init from executing further commands until the
service has exited.
This change additionally encapsulates the waiting_for_exec logic into
ServiceManager and removes the ambiguous 'bool' return value from
Reap() which previously indicated if a Reaped service was an exec
service or not.
Bug: 36511808
Bug: 36102163
Test: Bullhead boots, services run with exec_start as they do exec.
Change-Id: I44f775cf1c1dd81d5c715f44fdc150c651a2c80a
(cherry picked from commit b27004aa05)
With system and vendor split, the files under them get labelled
differently as part of b/36527360. These files also include selinux
policy files if the policy is split. In order to consistently label
them, all sepolicy files on rootfs also are labelled with labels
matching to that of the ones in system and vendor. So, this now requires
init to explicitly label *all* selinux files in rootfs after
initializing selinux.
Test: Boot angler and ensure servicemananger continues to work (uses
/{plat,nonplat}_service_contexts).
Test: Sideloaded OTA to angler to ensure recovery continues to work.
Test: Launch and load a website in Chrome.
Test: Launch camera and take picture, record video.
Change-Id: I4134eae80b5b51f381b549eb18e604cf271f9eb5
Signed-off-by: Sandeep Patil <sspatil@google.com>
Currently, when we are waiting for an exec service or a property, if
there are either any services to be restarted or any more commands to
be run, we set the epoll_timeout to handle these events. However, we
don't actually restart and processes or execute any commands while
waiting, so this essentially turns this waiting into a spin loop,
particularly in the common case of having more commands to execute,
where epoll_timeout is set to 0.
The change only sets epoll_timeout if we're not waiting.
Note that the only way to stop waiting for an exec service or a
property is for a signal or property to be delivered to init, which
happens through the epoll fds, so it's safe to indefinitely wait for
epoll to return.
Test: Boot bullhead
Change-Id: Iae3b217eb28182038b464fd39df8e7d27b5e23ff
Currently, if init crashes, the kernel panics. During development, we
would like to catch this crash before the kernel panics and reboot
into bootloader. This will prevent boot looping bad configurations,
particularly desired in test labs where manual intervention would
otherwise be required to reset the devices.
Keep the existing behavior for user builds, as init crashes should be
rare for production builds and rebooting the device is the correct
behavior for end users.
Bug: 34147472
Test: Boot bullhead userdebug, force init to crash, check that the
device is in bootloader
Test: Boot bullhead user, force init to crash, check that the kernel
panics and the device reboots as it did previously
Change-Id: Iab3d45ed0d1f82ffaad2a0835d9ca537c0516421
A recent change to the is_first_stage conditionals created a unneeded
else { } block as both the code in the else { } block and any code
that runs after it are both in the second stage of init. A first step
to clean this up is to remove this else block.
Secondly, given the above confusion, it makes sense to simplify the two
if (is_first_stage) conditions into one, which only now requires
duplicating one line to initialize logging and the actual "init
first/second stage started!" logs.
Lastly, there are a few commands ran at the beginning of both init
stages that do not need to be,
* boot_clock::time_point start_time = boot_clock::now();
This is only used in the first stage so keep it there
* umask(0);
umasks are preserved across execve() so it only needs to be set in the
first stage
* chmod("/proc/cmdline", 0440);
This needs to be moved until after /proc is mounted in the first
stage, but otherwise only needs to be done once
Test: Boot bullhead, check umask, check cmdline permissions, check
boot time property
Change-Id: Idb7df1d4330960ce282d9609f5c62281ee2638b9
Fix two formatting issues in Action::DumpState(),
Old:
on ro.crypto.state=encrypted ro.crypto.type=file zygote-start
New:
on ro.crypto.state=encrypted && ro.crypto.type=file && zygote-start
Old:
on boot
%sifup lo
%shostname localhost
%sdomainname localdomain
%swrite /proc/sys/vm/overcommit_memory 1
New:
on boot
ifup lo
hostname localhost
domainname localdomain
write /proc/sys/vm/overcommit_memory 1
Also, now that we're importing many small rc files, it no longer makes
sense to call Parser::DumpState() after each import. Therefore, move
the conditional to call Parser::DumpState() to after /init.rc and its
imports are parsed and after the late imports are parsed.
Test: Boot bullhead with DumpState() enabled and check the output
Change-Id: I0b81305b8938aa1a7133d7dd2055f34f47609cf9
This change makes init's SELinux policy compilation step target the
highest SELinux policy language version supported by the kernel.
Prior to this change the version was simply hard-coded in init.
P. S. clang-format (part of presubmit for this change) is being
counter-productive trying to format the section with secilc
parameters. The resulting layout is harder to read. This commit thus
disables clang-format for this section of code and formats the code
for improved readability.
Test: Remove precompiled policy, device boots up, no new denials.
Added log statement to print out the policy version passed
into secilc -- it printed 30, as expected.
Bug: 31363362
Change-Id: I151017b5211712861bafb662525e794a44026dd2
NOTE: This change affects only devices which use SELinux kernel policy
split over system and vendor directories/partitions.
Prior to this change, init compiled sepolicy from *.cil files on every
boot, thus slowing boot down by about 400 ms. This change enables init
to skip the step compilation and thus avoid spending the 400 ms. The
skipping occurs only if the device's vendor partition includes an
acceptable precompiled policy file. If no acceptable policy is found,
the compilation step takes place same as before.
Because such devices support updating system and vendor partitions
independently of each other, the vendor partition's precompiled policy
is only used if it was compiled against the system partition's policy.
The exact mechanism is that both partitions include a file containing
the SHA-256 digest of the system partition's policy
(plat_sepolicy.cil) and the precompiled policy is considered usable
only if the two digests are identical.
Test: Device with monolithic policy boots up just fine
Test: Device with split policy and with matching precompiled policy
boots up just fine and getprop ro.boottime.init.selinux returns
a number below 100 ms. No "Compiling SELinux policy" message in
dmesg.
Test: Device with split policy and with non-matching precompiled
policy boots up just fine and getpropr ro.boottime.init.selinux
returns a number above 400 ms. There is a "Compiling SELinux
policy" message in dmesg. The non-matching policy was obtained
by adding an allow rule to system/sepolicy, building a new
system image using make systemimage and then flashing it onto
the device.
Bug: 31363362
Change-Id: Ic2e81a83051689b5cd5ef1299ba6aaa1b1df1bdc
This makes init log stderr of secilc invoked to compile SELinux
policy. Having an explanation for why secilc failed is very useful for
debugging boot issues.
Test: Device with PRODUCT_FULL_TREBLE boots up just fine
Test: Modified init.cpp to reference non-existent .cil file on a device
with PRODUCT_FULL_TREBLE and confirmed that dmesg now contains the
error message from secilc saying that the file was not found.
Bug: 31363362
Change-Id: I6a3b3576daf0d6fd09e2c79bc43ae63850f44a00
The three CIL files comprising split sepolicy are being moved from the
root directory to system and vendor directories based on whether the
file is for platform/system policy or non-platform/vendor policy.
Test: Device boots, no additional SELinux denials. This test was run
for a device which has split policy and for a device which has
monolithic policy.
Bug: 31363362
Change-Id: Ica49f0beae56be0f1cea7117e48bf2f6af8b848b
This modifies init's loading of SELinux policy into the kernel to
load the split (platform/system vs non-platform/vendor) policy if it's
present. If the split policy is not present, the usual monolithic
policy is loaded into the kernel, same as before.
Split policy is loaded by first compiling it from CIL form using
secilc compiler into the conventional monolithic/compiled form which
is then loaded into the kernel.
The build system has not yet been modified to place split policy onto
devices. Thus, this commit currently has no effect. For testing split
policy, build plat_sepolicy.cil, nonplat_sepolicy.cil, and
mapping_sepolicy.cil, and place them into the root directory of the
device.
The following tests were performed for a device with monolithic policy
and for the same device with split policy.
Test: Device boots, no new denials
Test: Play Movies plays back movies
Test: Load ip6.me im Chrome
Bug: 31363362
Change-Id: I9a75a48ac88f3392abc36669f91b0803e88cd147
We don't need early mount in recovery mode for security considerations,
e.g., users should explicitly select 'mount /system' from the recovery
menu. This CL checks the existence of file "/sbin/recovery" and skip
early mount when it is found.
Bug: 35853576
Test: early mount /vendor without dm-verity on sailfish
Test: early mount not happen in recovery mode on sailfish
Change-Id: I69cc96f6fd0de6ce493082921738a958dd571115
Current early mount has some hard-coded paths that are not easy to extend
when we switch verified boot to AVB (external/avb/libavb). This CL uses some
C++ containers to replace those fixed paths.
Bug: 33254008
Test: early mount /vendor with dm-verity on sailfish
Test: early mount /vendor without dm-verity on sailfish
Test: early mount /vendor with dm-verity on bullhead
Change-Id: I32a22fe486d7649e33bb23c9018ddc0428df6069
Treblization requires to locate partner-specific modules in its own
partition. So their own init.rc file could be located in /odm or
/vendor.
This CL is to support those locations for the init.rc.
Additionally thic CL modified import parser to support importing a
relative path.
Test: building succeeded and tested on sailfish with enabling early
mount.
Bug: 35269867
Change-Id: I1bce924a32c8a2b53fb5d981d35d758cf9ddd9a6
With init parsing fstab fragments from kernel separately, the fs_mgr
would completely miss the device tree entries. That leads to things like
'adb remount' to go through without warning for verity even if /system
is verified. This happens because 'verity_update_state' completely
misses the partitions passed to android through the device tree.
solution is to teach fs_mgr about device tree fstab entries and add 2
new public APIs.
1. fs_mgr_read_fstab_dt() - reads device tree and returns fstab
generated from it.
2. fs_mgr_read_fstab_default() - reads both device tree fstab and
/fstab.{ro.hardware} and returns the combined table.
This also reduces the hardcoded /fstab.{ro.hardware} occurence only to
fs_mgr and for eveyone who wants to read the "default" fstab must be
changed to call fs_mgr_read_fstab_default() instead. e.g. adb.
b/27805372
Test: Angler was used since it has 2 early mounted partitions instead of
one. 1 verified and 1 unverified.
- Boot angler successfully without early mount
- Boot angler successfully with /vendor early mount and test if 'adb
remount' warns us about verity
- Boot angler successfully with both /system and /vendor early mounted
and ensure 'adb remount' warns us about verity.
- check partitions.system.verified status after /system early mount ot
ensure it is set to VERITY_MODE_DEFAULT.
- 'adb disable-verity' with early mounted /system doesn't work due to
missing changes in adb
TODO:
change adb to use the new fs_mgr_read_fstab_default() API
Change-Id: I82038d87c7a44488e938acce2cc1082c08f6f73a
Signed-off-by: Sandeep Patil <sspatil@google.com>
Most devices pass the veritymode through 'androidboot.veritymode' kernel
cmdline partition. However, for those who don't, the verity state is
read from a different block device whose path it passed to "verify="
fs_mgr option in fstab.
This change add support for such a case if the partition that needs to
be mounted early requires this additional block device to load the
verity state from.
Note that, there can only be 1 partition to get the verity state
regardless of the number of partitions we enable verity for.
Bug: 27805372
Test: Test angler by removing the metdata argument in fstab when it
boots fine. Tested by adding the argument when it fails to boot as
veritymode gets set to EIO during early mount due to lack of access to
properties as expected.
TODO: fs_mgr must pull the veritymode from kernel cmdline or device tree
by itself
Change-Id: I9e62b8c1bf2c0ae0365677df697a0cbe9e5810c1
Signed-off-by: Sandeep Patil <sspatil@google.com>
While technically possible, the verification at boot basically will
block init for as long as the entire partition is read while nothing
else is running. Disallow that as this is not going to be used anywhere.
Bug: 27805372
Test: boot angler with verifyatboot fs_mgr option for early mounted
vendor partition. That resulted in a panic() as expected.
Change-Id: I9da5caa163cae8bce6dbfb630f0ed5605ea044a0
Signed-off-by: Sandeep Patil <sspatil@google.com>
support mounting partitions early regardless of their "verified" status.
uses the newly exported fs_mgr APIs to split verity setup and mount
operations.
b/27805372
Test:
Angler:
- Early mount /vendor without dm-verity
Sailfish:
- Early mount /vendor without dm-verity
- Early mount /vendor with dm-verity
TODO:
add support for metadata partition used in angler
to load dm-verity data
Change-Id: Ie2768d4d895c19d045293c573773ee7bb03fff99
Signed-off-by: Sandeep Patil <sspatil@google.com>
Treble allows framework-only (system.img) OTA. To libavb statically
linked in init/fs_mgr cannot parse the AVB metadata of system.img, add
this property for the Treble OTA match process.
Bug: 35236019
Test: Normal boots, use 'adb shell getprop' to check the property is set.
Test: Recovery boots, press 'Mount system' and use 'adb shell getprop'
to check the property is set
Change-Id: I552be229c4efd45088d93252ac67176606f75d4d
* changes:
init: fstab: add support to read fstab entries from device tree
init: early_mount: add support to mount non-verity partitions early
init: remove the existing early_mount code
init: refactor: add support for doing early coldboot
ueventd: make selinux labeling optional for device creation
Use this for bootstat and init. This replaces the custom uptime parser in
bootstat.
This is a reland of aosp/332854 with a fix for Darwin.
Bug: 34352037
Test: chrono_utils_test
Change-Id: Ib2567d8df0e460ab59753ac1c053dd7f9f1008a7
for early mount, we need a way to tell init where to find vendor,
odm partitions (also system in case of non-A/B devices). Also, that
needs to be independent of kernel cmdline since the cmdline will likely
exceed its limit.
The change adds support for parse and create fstab entries that can be
directly sent to the fs_mgr for mounting partitions early in init first
stage.
Sample DT entry to mount vendor partition early on angler-
firmware {
android {
compatible = "android,firmware";
fstab {
compatible = "android,fstab";
vendor {
compatible = "android,vendor";
dev = "/dev/block/platform/soc.0/f9824900.sdhci/by-name/vendor";
type = "ext4";
mnt_flags = "ro,barrier=1,inode_readahead_blks=8";
fsmgr_flags = "wait";
};
};
};
};
b/27805372
Test: Boot angler and sailfish with early "vendor" partition mount by
adding aforementioned DT node and enable CONFIG_PROC_DEVICETREE in kernel
Change-Id: I669013e3fdb157e88719436534f63989dec95d60
Signed-off-by: Sandeep Patil <sspatil@google.com>
This is done by parsing 'androidboot.fstab=<fstab>' kernel cmdline
option to get the fstab file that *only* specifies partitions to be
mounted early (i.e. in init's first stage).
Note that, the same fstab file may not be used as an argument to
mount_all later in the boot as that will cause fs_mgr to fail with
EBUSY.
TODO:
- Possibly add a new mount_mode so the same fstab can be used for
early_mount, 'mount_all --early/--late' etc.
- Add support for dm-verity enabled partitions to mount early.
- Add support for getting fstab arguments through DT instead of kernel
cmdline.
Bug: 27805372
Test:
Boot angler by passing a seperate fstab file using the kernel
cmdline option to mount vendor partition early, remove the vendor
partition entry from the main fstab file for the test.
Boot sailfish by passing a seperate fstab entry via device tree
to mount vendor partition early. Remove vendor partition entry from
the main fstab file for the test
Change-Id: I18785b893c54c8cee960ab44d5e8f83e5d624aa8
Signed-off-by: Sandeep Patil <sspatil@google.com>
keeps parts of the code that are still needed for the fs_mgr
+ dt based implementation
b/27805372
Test: boot angler, sailfish without regressions
Change-Id: I1b08f8b7b4f2e67118d328443a5011c0f5ead919
Signed-off-by: Sandeep Patil <sspatil@google.com>
Add wait time log for exec service execution time
Fix memory leak when exec service failed to start
Test: on marlin
Bug: 34518530
Change-Id: I01736bd9b1429414f3dc91dd5d02d88a681f0985
This reverts commit db929bf9b7.
Seccomp is now inserted at the zygote level, not in init
Bug: 34710876
Test: Boots, seccomp policy in zygote & zygote64 but not init
Change-Id: I9075a79793171a4eaccf6228e9ff3398c791f8bd
There are many use cases from vendors to exec service in background and then
use a shell scriprt to wait for the command done.
This CL is to add a wait_for_prop command to suppor those use cases.
Bug: 34746108
Test: on marlin
Change-Id: Ia81290b0928f9d375710d2daa546714f0cd65b72
* Nanosecond precision ended up being harder to grok.
* This change modifies the Timer class to have duration_ms instead of
duration_ns.
Bug: 34466121
Test: adb logcat | grep bootstat
Change-Id: Ibd1c27dc3cb29d838a956e342281b2fb98d752a6
Bug: 33746484
Test: Successfully boot with original service and property contexts.
Test: Successfully boot with split serivce and property contexts.
Test: 'getprop -Z'
Change-Id: Ib9c087115040c1609aa64a3ed66d4e67e937e33d
Signed-off-by: Sandeep Patil <sspatil@google.com>
