The non-A/B package installation is subject to TOC/TOU flaw if the
attacker can switch the package in the middle of installation. And the
most pratical case is to store the package on an external device, e.g. a
sdcard, and swap the device in the middle.
To prevent that, we can adopt the same protection as used in sideloading
a package with FUSE. Specifically, when we install the package with FUSE,
we read the entire package to cryptographically verify its signature.
The hash for each transfer block is recorded in the memory (TOC), and
the subsequent reads (TOU) will be rejected upon dectecting a mismatch.
This CL forces the package installation with FUSE when the package stays
on a removable media.
Bug: 136498130
Test: Run bin/recovery --update_package with various paths;
and packages are installed from FUSE as expected
Change-Id: Ibc9b095036a2fa624e8edf6c347ed4f12aef072f
A number of utility functions are intended for serving recovery's own
use. Exposing them via libotautil (which is a static lib) would pass the
dependencies onto libotautil's users (e.g. recovery image, updater, host
simulator, device-specific recovery UI/updater extensions etc). This CL
finds a new home for the utils that are private to recovery.
Test: mmma bootable/recovery
Change-Id: I575e97ad099b85fe1c1c8c7c9458a5a43d4e11e1
Since this was combined into libhidlbase.
Bug: 135686713
Test: build only (libhidltransport is empty)
Change-Id: I253e50726967044714275ab995fb8a8a57bcde36
Some global variables are only used for recovery.cpp and
recovery_main.cpp, remove them from common.h and handle their usage
accordingly. Variables include:
static constexpr int kRecoveryApiVersion;
extern struct selabel_handle* sehandle;
extern RecoveryUI* ui;
extern bool has_cache;
bool is_ro_debuggable();
Test: unit tests pass, boot into recovery mode and run graphic tests
Change-Id: If83a005786c9b38412731da97aaf85af69a3b917
We may fail to memory map the package on 32 bit builds for packages with
2GiB+ size. This cl tries to install the package with fuse when memory map
fails in such cases.
Bug: 127071893
Test: build 32 bit version sailfish, push package and block.map, reboot into recovery with
the corresponding update_package argument.
Change-Id: I5dae4f3e27ccaf8d64ff3657d36f0e75db2330b0
We start minadbd and rescue services in two processes. In particular,
minadbd handles the requests from host, then communicates with rescue
service to do install/wipe works. When resuce service doesn't see any
request in a pre-defined timeout (currently 300s), rescue service will
exit to avoid endless waiting.
This CL changes minadbd to additionally send a no-op command to rescue
service as a heartbeat signal, so that host side can finish
time-consuming operations (e.g. downloading over network) while keeping
rescue service alive.
Bug: 136457446
Test: Enter resuce mode on blueline. Send `adb rescue getprop
ro.build.fingerprint` and check that rescue service doesn't exit.
Test: Stop sending the getprop command. Check that rescue service exits
after 300s.
Change-Id: Ib9d5ed710cfa94ecfe6cf393a71a0b67b2539531
Factor out a new function from ApplyFromSdcard that installs a package
from a local path. Inside this function, we start the fuse and choose the
type of data provider depending on the path string. And similar to the
existing logic, we treat the package as a block map if the path starts
with a '@'.
This is part of the effort to install larger than 2GiB packages on ILP32
devices.
Bug: 127071893
Test: Build a 32 bit sailfish and create a 3GiB OTA package. Sideload
the package, uncrypt and install the package from sdcard.
Change-Id: I328ea34fa530731acbce7554bfc3059313ad6ece
Therefore InstallPackage() doesn't need to worry about the details of a
given Package.
Bug: 127071893
Test: run update from /bin/recovery --update_package=@path, sideload a package
Change-Id: I0caa36785b43924f884ee398e7ea640d7472a92e
The results from these functions have boolean semantics. They're
returning `int` prior to this CL, with some of them mixing 0 and
InstallResult. Note that SetUpNonAbUpdateCommands() was returning
INSTALL_CORRUPT / INSTALL_ERROR / 0 prior to this change, but all the
callers handle INSTALL_CORRUPT and INSTALL_ERROR the same way.
This CL changes them to return bool instead.
Test: `mmma -j bootable/recovery`
Test: TreeHugger
Test: Sideload on taimen.
Change-Id: Ic1b5dbf79aaca68b53ab8ea2c8ba3d19f988c571
Prior to this CL, GetWipePartitionList was declared in install.h
(libinstall) but defined in recovery.cpp (librecovery). This CL
addresses the issue by refactoring wipe-device related functions into
install/wipe_device.cpp.
Test: atest recovery_component_test
Change-Id: I7ebe04ccfda3d793e085403560a0a202752d9ee3
bootloader will set `boot-rescue` in BCB command field to indicate
booting into rescue mode. This CL adds the matching parsing code.
This CL changes the on-screen UI to display the default image while
waiting for each sideload / rescue command.
It also changes the minadbd reboot handlers to use REBOOT_ instead of
the previous ENTER_ actions. This ensures a reboot going through
bootloader, which may load a newly installed bootloader/recovery.
Bug: 128505466
Test: Boot into rescue mode. Run `adb rescue getprop` and `adb rescue
install`. Check the UI. Then run `adb reboot rescue`.
Change-Id: I5b7de9dfd898ed8e14bea0d4ad7385a9bae26e94
We perform these steps to perserve the recovery logs when wiping
/cache partition. Move them to logging.cpp to keep the actually
EraseVolume function concise.
Bug: 130166585
Test: unit tests pass, mount cache and check last log after cache
Change-Id: Idc52833817a446f3a0148a3dd2112f911c9ef48d
Therefore, libinstall becomes the sole owner to handle the request
from minadbd service.
The change also includes
1. move logging.cpp out of librecovery
2. drop the dependency on common.h
3. now it's more sensible to move the wipe_cache as part of
install_package. move the wipe_cache to the end of the function.
Bug: 130166585
Test: wipe data and cache from menu
Change-Id: I6f356dccdb38015c50acf756bac246f87c30fc1f
This cl adds a socket pair to support the communication between recovery
and minadbd. Therefore, minadbd will be able to issue multiple commands
to recovery and get back the status of each command.
This cl also switches the adb sideload from the recovery menu to use
this protocol; and moves minadbd to a separate binary.
Bug: 130166585
Test: sideload a package
Change-Id: I80d36d5c4e6fe1ae3ea23640907bc50c0dc0d482
It was once considered to be shared between recovery and minadbd, so
that the latter can start an install on its own. The plan has been
changed, since package install -- including device wipe operations --
could be device-specific, which should be done by recovery only.
This CL moves libinstall back to a static library, which also saves the
overall size (reducing from 140256 + 660576 to 555880 bytes on
aosp_taimen-userdebug).
Bug: 130166585
Test: Run recovery_component_test.
Test: `adb sideload` on taimen.
Change-Id: Ib1f5f79f235df4682c0bd104425c9c122f6091ba
otautil/roots.h includes <fstab/fstab.h>, but users of otautil/roots.h
don't need to explicitly depend on libfstab unless they have a real
need.
Also remove the unneeded include of <fstab/fstab.h> from
fsck_unshare_blocks.cpp.
Test: mmma -j bootable/recovery
Change-Id: Id3dc995a4769e631ab242843ee439bd94b2bf0bc
Build libinstall as a shared library. Also drop the dependency on the
global variables in common.h.
Test: unit tests pass, sideload an OTA
Change-Id: I30a20047768ce00689fc0e7851c1c5d712a365a0
