42d676db47
Because DSUs mount userdata via a fiemap, libfiemap has trouble creating additional fiemaps on top of it. The complex stacking of dm-linear is not supported. For other libfiemap tests we've hacked around this limitation. If LpMetadata is in a folder named "test", we allow the backing device search to stop at a dm node, whereas otherwise it would need to stop at a physical device. However this was not quite enough for vts_libsnapshot_test, because (1) the test folder was not included in the pattern match, and (2) CreateLogicalPartition() could not handle device-mapper names, as it expects a named physical partition. Addressing both of these allows the tests to pass on DSUs. Bug: 156713441 Test: vts_libsnapshot_test on DSU Change-Id: Ie7ee70e31dff0809a5f0c402ed132d80dd03d9b1 Merged-In: Ie7ee70e31dff0809a5f0c402ed132d80dd03d9b1 |
||
|---|---|---|
| .. | ||
| include/libfiemap | ||
| testdata | ||
| Android.bp | ||
| binder.cpp | ||
| fiemap_status.cpp | ||
| fiemap_writer.cpp | ||
| fiemap_writer_test.cpp | ||
| image_manager.cpp | ||
| image_test.cpp | ||
| metadata.cpp | ||
| metadata.h | ||
| passthrough.cpp | ||
| README.md | ||
| split_fiemap_writer.cpp | ||
| utility.cpp | ||
| utility.h | ||
libfiemap
libfiemap is a library for creating block-devices that are backed by
storage in read-write partitions. It exists primary for gsid. Generally, the
library works by using libfiemap_writer to allocate large files within
filesystem, and then tracks their extents.
There are three main uses for libfiemap:
- Creating images that will act as block devices. For example, gsid needs to
create a
system_gsiimage to store Dynamic System Updates. - Mapping the image as a block device while /data is mounted. This is fairly tricky and is described in more detail below.
- Mapping the image as a block device during first-stage init. This is simple because it uses the same logic from dynamic partitions.
Image creation is done through SplitFiemap. Depending on the file system,
a large image may have to be split into multiple files. On Ext4 the limit is
16GiB and on FAT32 it's 4GiB. Images are saved into /data/gsi/<name>/
where <name> is chosen by the process requesting the image.
At the same time, a file called /metadata/gsi/<name>/lp_metadata is created.
This is a super partition header that allows first-stage init to create dynamic
partitions from the image files. It also tracks the canonical size of the image,
since the file size may be larger due to alignment.
Mapping
It is easy to make block devices out of blocks on /data when it is not
mounted, so first-stage init has no issues mapping dynamic partitions from
images. After /data is mounted however, there are two problems:
/datais encrypted./dev/block/by-name/datamay be marked as in-use.
We break the problem down into three scenarios.
FDE and Metadata Encrypted Devices
When FDE or metadata encryption is used, /data is not mounted from
/dev/block/by-name/data. Instead, it is mounted from an intermediate
dm-crypt or dm-default-key device. This means the underlying device is
not marked in use, and we can create new dm-linear devices on top of it.
On these devices, a block device for an image will consist of a single
device-mapper device with a dm-linear table entry for each extent in the
backing file.
Unencrypted and FBE-encrypted Devices
When a device is unencrypted, or is encrypted with FBE but not metadata
encryption, we instead use a loop device with LOOP_SET_DIRECT_IO enabled.
Since /data/gsi has encryption disabled, this means the raw blocks will be
unencrypted as well.
Split Images
If an image was too large to store a single file on the underlying filesystem,
on an FBE/unencrypted device we will have multiple loop devices. In this case,
we create a device-mapper device as well. For each loop device it will have one
dm-linear table entry spanning the length of the device.
State Tracking
It's important that we know whether or not an image is currently in-use by a
block device. It could be catastrophic to write to a dm-linear device if the
underlying blocks are no longer owned by the original file. Thus, when mapping
an image, we create a property called gsid.mapped_image.<name> and set it to
the path of the block device.
Additionally, we create a /metadata/gsi/<subdir>/<name>.status file. Each
line in this file denotes a dependency on either a device-mapper node or a loop
device. When deleting a block device, this file is used to release all
resources.