3ea911bc06
The flow of I/O path is as follows:
1: When there is a I/O request for a given sector, we first
check the in-memory COW operation mapping for that sector.
2: If the mapping of sector to COW operation is found, then the
existing I/O path will work seamlessly. Even if the COW operation
encodes multiple blocks, we will discard the remaining data.
3: If the mapping of sector to COW operation is not found:
a: Find the previous COW operation as the vector has sorted sectors.
b: If the previous COW operation is a REPLACE op:
i: Check if the current sector is encoded in the previous COW
operations compressed block.
ii: If the sector falls within the range of compressed blocks,
retrieve the block offset.
iii: De-compress the COW operation based on the compression
factor.
iv: memcpy the data based on the block offset.
v: cache the COW operation pointer as subsequent I/O requests
are sequential and can just be a memcpy at the correct offset.
c: If the previous COW operation is not a REPLACE op or if the
requested sector does not fall within the compression factor
of the previous COW operation, then fallback and read the data
from base device.
Snapshot-merge:
During merge of REPLACE ops, read the entire op in one shot, de-compress
multiple blocks and write all the blocks in one shot.
Performance:
go/variable-block-vabc-perf covers detail performance runs
on Pixel 6 for full and incremental OTA.
Bug: 319309466
Test: snapuserd_test covers all the I/O path with various block sizes.
About 252 cases with all combinations and tunables.
[==========] 252 tests from 4 test suites ran. (702565 ms total)
[ PASSED ] 252 tests.
On Pixel 6:
=======================================
COW Writer V3:
for i in full, incremental OTA
for j in 4k, 16k, 32k, 64k, 128, 256k
for k in lz4, zstd, gz
install OTA, reboot, verify merge
=======================================
COW Writer V2:
for i in full, incremental OTA
for j in 4k
for k in lz4, zstd, gz
install OTA, reboot, verity merge
=====================================
Change-Id: I4c3b5c3efa0d09677568b4396cc53db0e74e7c99
Signed-off-by: Akilesh Kailash <akailash@google.com>
|
||
|---|---|---|
| .. | ||
| include | ||
| libdm | ||
| libfiemap | ||
| libfs_avb | ||
| libfstab | ||
| liblp | ||
| libsnapshot | ||
| libstorage_literals | ||
| libvbmeta | ||
| tests | ||
| tools | ||
| .clang-format | ||
| Android.bp | ||
| blockdev.cpp | ||
| blockdev.h | ||
| clean_scratch_files.rc | ||
| file_wait.cpp | ||
| fs_mgr.cpp | ||
| fs_mgr_dm_linear.cpp | ||
| fs_mgr_format.cpp | ||
| fs_mgr_overlayfs_control.cpp | ||
| fs_mgr_overlayfs_control.h | ||
| fs_mgr_overlayfs_mount.cpp | ||
| fs_mgr_overlayfs_mount.h | ||
| fs_mgr_priv.h | ||
| fs_mgr_remount.cpp | ||
| fs_mgr_roots.cpp | ||
| fs_mgr_vendor_overlay.cpp | ||
| NOTICE | ||
| OWNERS | ||
| README.overlayfs.md | ||
| TEST_MAPPING | ||
Android OverlayFS Integration with adb Remount
Introduction
Users working with userdebug or eng builds expect to be able to remount the system partition as read-write and then add or modify any number of files without reflashing the system image, which is efficient for a development cycle.
Limited memory systems use read-only types of file systems or dynamic Android partitions (DAPs). These file systems land system partition images right-sized, and have been deduped at the block level to compress the content. This means that a remount either isn’t possible, or isn't useful because of space limitations or support logistics.
OverlayFS resolves these debug scenarios with the adb disable-verity and adb remount commands, which set up backing storage for a writable file system as an upper reference, and mount the lower reference on top.
Performing a remount
Use the following sequence to perform the remount.
$ adb root
$ adb disable-verity
$ adb reboot
$ adb wait-for-device
$ adb root
$ adb remount
Then enter one of the following sequences:
$ adb shell stop
$ adb sync
$ adb shell start
$ adb reboot
or
$ adb push <source> <destination>
$ adb reboot
Note that you can replace these two lines in the above sequence:
$ adb disable-verity
$ adb reboot
with this line:
$ adb remount -R
Note: adb remount -R won’t reboot if the device is already in the adb remount state.
None of this changes if OverlayFS needs to be engaged. The decisions whether to use traditional direct file-system remount, or one wrapped by OverlayFS is automatically determined based on a probe of the file-system types and space remaining.
Backing Storage
When OverlayFS logic is feasible, it uses either the /cache/overlay/ directory for non-A/B devices, or the /mnt/scratch/overlay directory for A/B devices that have access to LRAP. It is also possible for an A/B device to use the system_ partition for backing storage. eg: if booting off system_a+vendor_a, use system_b. The backing store is used as soon as possible in the boot process and can occur at first stage init, or when the mount_all commands are run in init RC scripts.
By attaching OverlayFS early, SEpolicy or init can be pushed and used after the exec phases of each stage.
Caveats
- Backing storage requires more space than immutable storage, as backing is done file by file. Be mindful of wasted space. For example, defining BOARD_IMAGE_PARTITION_RESERVED_SIZE has a negative impact on the right-sizing of images and requires more free dynamic partition space.
- The kernel requires CONFIG_OVERLAY_FS=y. If the kernel version is higher than 4.4, it requires source to be in line with android-common kernels. The patch series is available on the upstream mailing list and the latest as of Sep 5 2019 is https://www.spinics.net/lists/linux-mtd/msg08331.html This patch adds an override_creds mount option to OverlayFS that permits legacy behavior for systems that do not have overlapping sepolicy rules, principals of least privilege, which is how Android behaves. For 4.19 and higher a rework of the xattr handling to deal with recursion is required. https://patchwork.kernel.org/patch/11117145/ is a start of that adjustment.
- adb enable-verity frees up OverlayFS and reverts the device to the state prior to content updates. The update engine performs a full OTA.
- adb remount overrides are incompatible with OTA resources, so the update engine may not run if fs_mgr_overlayfs_is_setup() returns true.
- If a dynamic partition runs out of space, making a logical partition larger may fail because of the scratch partition. If this happens, clear the scratch storage by running either either fastboot flashall or adb enable-verity. Then reinstate the overrides and continue.
- For implementation simplicity on retrofit dynamic partition devices, take the whole alternate super (eg: if "a" slot, then the whole of "system_b"). Since landing a filesystem on the alternate super physical device without differentiating if it is setup to support logical or physical, the alternate slot metadata and previous content will be lost.
- There are other subtle caveats requiring complex logic to solve.
Have evaluated them as too complex or not worth the trouble, please
File a bug if a use case needs to be covered.
- The backing storage is treated fragile, if anything else has issue with the space taken, the backing storage will be cleared out and we reserve the right to not inform, if the layering does not prevent any messaging.
- Space remaining threshold is hard coded. If 1% or more space still remains, OverlayFS will not be used, yet that amount of space remaining is problematic.
- Flashing a partition via bootloader fastboot, as opposed to user space fastbootd, is not detected, thus a partition may have override content remaining. adb enable-verity to wipe.
- Space is limited, there is near unlimited space on userdata, we have made an architectural decision to not utilize /data/overlay/ at this time. Acquiring space to use for backing remains an ongoing battle.
- First stage init, or ramdisk, can not be overriden.
- Backing storage will be discarded or ignored on errors, leading to confusion. When debugging using adb remount it is currently advised to confirm update is present after a reboot to develop confidence.
- File bugs or submit fixes for review.