Version 3.16.0

Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>
CHANGELOG.md: Add entry for PR #446
2026-07-03 14:05:11 +02:00 · 2025-05-04 18:15:02 -04:00 · 2025-05-04 18:14:25 -04:00 · 2025-05-04 18:06:59 -04:00 · 2025-05-04 16:57:03 -04:00 · 2025-05-04 16:55:38 -04:00
31 changed files with 1867 additions and 849 deletions
@@ -34,8 +34,7 @@ jobs:
              - aarch64-apple-darwin
              - x86_64-apple-darwin
            combine: lipo
-          # ubuntu-latest is not 24.04 yet and 22.04's qemu-user-static segfaults.
-          - os: ubuntu-24.04
+          - os: ubuntu-latest
            name: aarch64-linux-android31
            targets:
              - aarch64-linux-android
@@ -65,7 +64,7 @@ jobs:
        run: |
          cargo install \
              --git https://github.com/chenxiaolong/cargo-android \
-              --tag v0.1.1
+              --tag v0.1.3

      - name: Get version
        id: get_version
@@ -84,7 +83,7 @@ jobs:
          done

      - name: Cache Rust dependencies
-        uses: Swatinem/rust-cache@82a92a6e8fbeee089604da2575dc567ae9ddeaab # v2.7.5
+        uses: Swatinem/rust-cache@9d47c6ad4b02e050fd481d890b2ea34778fd09d6 # v2.7.8
        with:
          key: ${{ matrix.artifact.name }}

@@ -155,7 +154,7 @@ jobs:
        run: cp LICENSE README.md target/output/

      - name: Archive executable
-        uses: actions/upload-artifact@6f51ac03b9356f520e9adb1b1b7802705f340c2b # v4.5.0
+        uses: actions/upload-artifact@ea165f8d65b6e75b540449e92b4886f43607fa02 # v4.6.2
        with:
          name: avbroot-${{ steps.get_version.outputs.version }}-${{ matrix.artifact.name }}
          path: |
@@ -13,4 +13,4 @@ jobs:
        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

      - name: Run cargo-deny
-        uses: EmbarkStudios/cargo-deny-action@e2f4ede4a4e60ea15ff31bc0647485d80c66cfba # v2.0.4
+        uses: EmbarkStudios/cargo-deny-action@34899fc7ba81ca6268d5947a7a16b4649013fea1 # v2.0.11
@@ -27,7 +27,7 @@ jobs:
        uses: actions/checkout@11bd71901bbe5b1630ceea73d27597364c9af683 # v4.2.2

      - name: Create release
-        uses: softprops/action-gh-release@7b4da11513bf3f43f9999e90eabced41ab8bb048 # v2.2.0
+        uses: softprops/action-gh-release@da05d552573ad5aba039eaac05058a918a7bf631 # v2.2.2
        with:
          tag_name: v${{ steps.get_version.outputs.version }}
          name: Version ${{ steps.get_version.outputs.version }}
@@ -7,6 +7,55 @@
    to update the actual links at the bottom of the file.
 -->

+### Version 3.16.0
+
+* Add support for CoW version 3 for virtual A/B ([Issue #441], [PR #442], [PR #445])
+  * This was recently introduced with the Pixel 9a. Previous devices all used CoW version 2.
+* Add support for uncompressed CoW for virtual A/B ([PR #443])
+  * This is not used on actual devices, but is very useful for testing the CoW estimation logic.
+  * All differences between avbroot's and AOSP delta_generator's estimation logic are now fixed.
+* Add support for custom CoW compression levels for virtual A/B ([PR #444])
+  * This is also not used on actual devices, but is supported by AOSP, so avbroot should support it too.
+* Update dependencies ([PR #446])
+
+### Version 3.15.0
+
+* Add support for changing the virtual A/B compression algorithm ([PR #437])
+    * For devices that launched with Android <14, `--vabc-algo lz4` can significantly increase OTA installation speed when using a custom OTA updater app (with caveats). There is no difference when sideloading from recovery mode.
+    * See [the documentation](./README.md#changing-virtual-ab-cow-compression-algorithm) for more details.
+* Switch back to the ring library now that it is maintained again ([PR #438])
+* Update dependencies ([PR #439])
+
+### Version 3.14.0
+
+* Report as many errors as possible before failing in `avbroot ota verify` and improve error messages ([Discussion #426], [PR #428], [PR #430])
+* Fix new clippy warnings introduced in Rust 1.85 ([PR #429])
+* Fix massive performance regression introduced in 3.13.0 for OTAs that use gzip for virtual A/B CoW compression ([Issue #433], [PR #434])
+* Update dependencies ([PR #435])
+
+### Version 3.13.0
+
+* Fix parsing Samsung `super.img` files in `avbroot lp` due to Samsung putting their own data structures in a region that's supposed to be filled with zeros ([PR #415])
+* Add advanced option to skip replacing the OTA certificate in the system image ([Discussion #417], [PR #418])
+* Switch to stable bzip2-rs release and use zlib-rs as the backend for flate2 ([PR #421])
+* Switch to the aws-lc cryptography library for SHA1 and SHA2 hashing ([PR #422])
+    * The ring library is no longer maintained
+* Fix incorrect `Partitions aren't protected by AVB: system` warning when using `--skip-system-ota-cert` ([PR #423])
+* Discard unneeded temp file sooner when using `--skip-system-ota-cert` ([PR #424])
+* Make `avbroot lp`'s parser less strict so that it can load on-device `super` partitions ([PR #425])
+    * The on-disk layout on virtual A/B devices violates some requirements stated in AOSP's documentation
+* Update dependencies ([PR #427])
+
+### Version 3.12.0
+
+* Add new `-p <name>` option to `avbroot ota extract` for extracting specific partitions ([PR #408])
+* Deprecate the `--boot-only` option in `avbroot ota extract` ([PR #408])
+    * The option will remain indefinitely for backwards compatibility, but is hidden from `--help`
+* Add support for extracting the embedded OTA certificate and AVB public key in `avbroot ota extract` ([PR #409])
+* Rename `avbroot key extract-avb` to `avbroot key encode-avb` for consistency with `avbroot key decode-avb` ([PR #410])
+    * The old syntax will remain supported indefinitely for backwards compatibility, but is hidden from `--help`
+* Update dependencies ([PR #411])
+
 ### Version 3.11.0

 * Fix crash when ignoring warning about `--magisk-preinit-device` not being specified ([PR #394])
@@ -270,6 +319,8 @@ Behind-the-scenes changes:
 [Discussion #235]: https://github.com/chenxiaolong/avbroot/discussions/235
 [Discussion #286]: https://github.com/chenxiaolong/avbroot/discussions/286
 [Discussion #294]: https://github.com/chenxiaolong/avbroot/discussions/294
+[Discussion #417]: https://github.com/chenxiaolong/avbroot/discussions/417
+[Discussion #426]: https://github.com/chenxiaolong/avbroot/discussions/426
 [Issue #138]: https://github.com/chenxiaolong/avbroot/issues/138
 [Issue #144]: https://github.com/chenxiaolong/avbroot/issues/144
 [Issue #145]: https://github.com/chenxiaolong/avbroot/issues/145
@@ -296,6 +347,8 @@ Behind-the-scenes changes:
 [Issue #356]: https://github.com/chenxiaolong/avbroot/issues/356
 [Issue #366]: https://github.com/chenxiaolong/avbroot/issues/366
 [Issue #393]: https://github.com/chenxiaolong/avbroot/issues/393
+[Issue #433]: https://github.com/chenxiaolong/avbroot/issues/433
+[Issue #441]: https://github.com/chenxiaolong/avbroot/issues/441
 [PR #130]: https://github.com/chenxiaolong/avbroot/pull/130
 [PR #132]: https://github.com/chenxiaolong/avbroot/pull/132
 [PR #133]: https://github.com/chenxiaolong/avbroot/pull/133
@@ -430,3 +483,28 @@ Behind-the-scenes changes:
 [PR #402]: https://github.com/chenxiaolong/avbroot/pull/402
 [PR #403]: https://github.com/chenxiaolong/avbroot/pull/403
 [PR #404]: https://github.com/chenxiaolong/avbroot/pull/404
+[PR #408]: https://github.com/chenxiaolong/avbroot/pull/408
+[PR #409]: https://github.com/chenxiaolong/avbroot/pull/409
+[PR #410]: https://github.com/chenxiaolong/avbroot/pull/410
+[PR #411]: https://github.com/chenxiaolong/avbroot/pull/411
+[PR #415]: https://github.com/chenxiaolong/avbroot/pull/415
+[PR #418]: https://github.com/chenxiaolong/avbroot/pull/418
+[PR #421]: https://github.com/chenxiaolong/avbroot/pull/421
+[PR #422]: https://github.com/chenxiaolong/avbroot/pull/422
+[PR #423]: https://github.com/chenxiaolong/avbroot/pull/423
+[PR #424]: https://github.com/chenxiaolong/avbroot/pull/424
+[PR #425]: https://github.com/chenxiaolong/avbroot/pull/425
+[PR #427]: https://github.com/chenxiaolong/avbroot/pull/427
+[PR #428]: https://github.com/chenxiaolong/avbroot/pull/428
+[PR #429]: https://github.com/chenxiaolong/avbroot/pull/429
+[PR #430]: https://github.com/chenxiaolong/avbroot/pull/430
+[PR #434]: https://github.com/chenxiaolong/avbroot/pull/434
+[PR #435]: https://github.com/chenxiaolong/avbroot/pull/435
+[PR #437]: https://github.com/chenxiaolong/avbroot/pull/437
+[PR #438]: https://github.com/chenxiaolong/avbroot/pull/438
+[PR #439]: https://github.com/chenxiaolong/avbroot/pull/439
+[PR #442]: https://github.com/chenxiaolong/avbroot/pull/442
+[PR #443]: https://github.com/chenxiaolong/avbroot/pull/443
+[PR #444]: https://github.com/chenxiaolong/avbroot/pull/444
+[PR #445]: https://github.com/chenxiaolong/avbroot/pull/445
+[PR #446]: https://github.com/chenxiaolong/avbroot/pull/446
@@ -4,9 +4,9 @@ members = ["avbroot", "e2e", "fuzz", "xtask"]
 resolver = "2"

 [workspace.package]
-version = "3.11.0"
+version = "3.16.0"
 license = "GPL-3.0-only"
-edition = "2021"
+edition = "2024"
 repository = "https://github.com/chenxiaolong/avbroot"

 [workspace.lints.clippy]
@@ -2,12 +2,10 @@

 (This page is also available in: [Russian (Русский)](./README.ru.md).)

-avbroot is a program for patching Android A/B-style OTA images for root access while preserving AVB (Android Verified Boot) using custom signing keys. It is compatible with both Magisk and KernelSU. If desired, it can also just re-sign an OTA without enabling root access.
+avbroot is a tool for modifying Android A/B OTA images reproducibly and re-signing them with custom keys. It also includes a [collection of subcommands](./README.extra.md) for packing and unpacking numerous Android image formats.

 Having a good understanding of how AVB and A/B OTAs work is recommended prior to using avbroot. At the very least, please make sure the [warnings and caveats](#warnings-and-caveats) are well-understood to avoid the risk of hard bricking.

-**NOTE:** avbroot 2.0 has been rewritten in Rust and no longer relies on any AOSP code. The CLI is fully backwards compatible, but the old Python implementation can be found in the `python` branch if needed.
-
 ## Requirements

 * Only devices that use modern A/B partitioning are supported. This is the case for most non-Samsung devices launched with Android 10 or newer. To check if a device uses this partitioning scheme, open the OTA zip file and check that:
@@ -23,7 +21,7 @@ Having a good understanding of how AVB and A/B OTAs work is recommended prior to

 avbroot applies the following patches to the partition images:

-* The `boot` or `init_boot` image, depending on device, is patched to enable root access. For Magisk, the patch is equivalent to what would be normally done by the Magisk app.
+* The `boot` or `init_boot` image, depending on device, is patched to enable root access if requested.

 * The `boot`, `recovery`, or `vendor_boot` image, depending on device, is patched to replace the OTA signature verification certificates with the custom OTA signing certificate. This allows future patched OTAs to be sideloaded from recovery mode after the bootloader has been locked. It also prevents accidental flashing of the original unpatched OTA.

@@ -31,7 +29,7 @@ avbroot applies the following patches to the partition images:

 ## Warnings and Caveats

-* **Always leave the `OEM unlocking` checkbox enabled when using a locked bootloader with root.** This is critically important. Root access allows the boot partition to potentially be overwritten, either accidentally or intentionally, with an image that is not properly signed. In this scenario, if the checkbox is turned off, both the OS and recovery mode will be made unbootable and `fastboot flashing unlock` will not be allowed. This effectively renders the device **_hard bricked_**.
+* **Always leave the `OEM unlocking` checkbox enabled when using a locked bootloader while rooted.** This is critically important. Root access allows the boot partition to potentially be overwritten, either accidentally or intentionally, with an image that is not properly signed. In this scenario, if the checkbox is turned off, both the OS and recovery mode will be made unbootable and `fastboot flashing unlock` will not be allowed. This effectively renders the device **_hard bricked_**.

    Repeat: **_ALWAYS leave `OEM unlocking` enabled if rooted._**

@@ -121,7 +119,7 @@ When patching OTAs for multiple devices, generating unique keys for each device
 2. Convert the public key portion of the AVB signing key to the AVB public key metadata format. This is the format that the bootloader requires when setting the custom root of trust.

    ```bash
-    avbroot key extract-avb -k avb.key -o avb_pkmd.bin
+    avbroot key encode-avb -k avb.key -o avb_pkmd.bin
    ```

 3. Generate a self-signed certificate for the OTA signing key. This is used by recovery to verify OTA updates when sideloading.
@@ -297,8 +295,7 @@ Magisk versions 25211 and newer require a writable partition for storing custom
    ```bash
    avbroot ota extract \
        --input /path/to/ota.zip \
-        --directory . \
-        --boot-only
+        --partition <name> # init_boot or boot, depending on device
    ```

 2. Patch the boot image via the Magisk app. This **MUST** be done on the target device or a device of the same model! The partition name will be incorrect if patched from Magisk on a different device model.
@@ -385,11 +382,28 @@ Note that avbroot will validate that the prepatched image is compatible with the

 avbroot can be used for just re-signing an OTA by specifying `--rootless` instead of `--magisk`/`--prepatched`. With this option, the patched OTA will not be rooted. The only modification applied is the replacement of the OTA verification certificate so that the OS can be upgraded with future (patched) OTAs.

-### Skipping recovery OTA certificate patches
+### Skipping OTA certificate patches

-avbroot can skip modifying `otacerts.zip` in the recovery image with the `--skip-recovery-ota-cert` option. **Do not do this unless you have a good reason to do so.** (For example, if you've already manually inserted the OTA certificate into a boot image specified with `--prepatched` or `--replace`.) When this option is used with `--rootless` (and `--dsu` is not specified), then no modifications are performed on any boot image besides ensuring they are properly signed.
+avbroot can skip modifying `otacerts.zip` with the `--skip-system-ota-cert` and `--skip-recovery-ota-cert` options. **Do not use these unless you have a good reason to do so.**

-When manually adding the OTA certificate to a boot image, [verifying the patched OTA](#verifying-otas) afterwards is recommended to ensure that it was properly done.
+When `--skip-system-ota-cert` is used, the OTA certificates in the `system` partition will not be modified. This prevents custom OTA updater apps from installing further patched OTAs while booted into Android.
+
+When `--skip-recovery-ota-cert` is used, the OTA certificates in the `vendor_boot` or `recovery` partition will not be modified. **This prevents sideloading further patched OTAs from recovery mode.**
+
+If `--skip-recovery-ota-cert` is used because the OTA certificate was already manually added to the boot image, then [verifying the patched OTA](#verifying-otas) afterwards is recommended to ensure that it was properly done. The verification process is only capable of checking the boot image's copy of the OTA certificates, not the system image's copy of them.
+
+### Skipping all patches
+
+To have avbroot make the absolute minimal changes:
+
+* Specify `--skip-system-ota-cert`
+* Specify `--skip-recovery-ota-cert`
+* Specify `--rootless`
+* Omit `--dsu`
+
+This will re-sign the `vbmeta` partition and the OTA with the custom keys, but leave all other partitions untouched.
+
+**This should only be used for advanced troubleshooting.** Without the OTA certificate patches, the resulting OTA will not be able to install further updates.

 ### Replacing partitions

@@ -415,6 +429,18 @@ Verified boot is disabled by vbmeta's header flags: 0x3

 To forcibly enable AVB (by clearing the flags), pass in `--clear-vbmeta-flags`.

+### Changing virtual A/B CoW compression algorithm
+
+The virtual A/B CoW compression algorithm can be changed by passing in `--vabc-algo <algo>` with `gz` or `lz4`. OTAs normally use an algorithm that is compatible with the initial version of Android shipped on the device.
+
+* Devices launching with Android 12 support `gz` and `brotli` (unsupported by avbroot)
+* Devices launching with Android 14 support `lz4`
+* Devices launching with Android 15 support `zstd` (unsupported by avbroot)
+
+Picking a fast algorithm, like lz4, can speed up OTA installation significantly when installing via a custom OTA updater app. However, there is no performance difference when sideloading an OTA from recovery mode.
+
+Note that the currently running version of Android must support the specified compression algorithm or else the OTA will fail to install. For example, trying to install an Android 14 OTA that uses lz4 CoW compression will fail if the running system is Android 13.
+
 ### Non-interactive use

 avbroot prompts for the private key passphrases interactively by default. To run avbroot non-interactively, either:
@@ -451,17 +477,20 @@ avbroot prompts for the private key passphrases interactively by default. To run

 * Use unencrypted private keys. This is strongly discouraged.

-### Extracting the entire OTA
+### Extracting an OTA

-To extract all images contained within the OTA's `payload.bin`, run:
+To extract the partition images contained within an OTA's `payload.bin`, run:

 ```bash
 avbroot ota extract \
    --input /path/to/ota.zip \
-    --directory extracted \
-    --all
+    --directory extracted
 ```

+By default, this only extracts the images that could potentially be patched by avbroot. To extract all images, use the `--all` option. To extract specific images, use the `--partition <name>` option, which can be specified multiple times.
+
+This command also supports extracting the embedded OTA certificate and AVB public key using the `--cert-ota` and `--public-key-avb` options. To extract only these components, pass in `--none` to skip extracting partition images.
+
 ### Zip write mode

 By default, avbroot uses streaming writes for the output OTA during patching. This means it computes the sha256 digest for the digital signature as the file is being written. This mode causes the zip file to contain data descriptors, which is part of the zip standard and works on the vast majority of devices. However, some devices may have broken zip file parsers and fail to properly read OTA zip files containing data descriptors. If this is the case, pass in `--zip-mode seekable` when patching.
@@ -1,11 +1,9 @@
 # avbroot

-avbroot – это программа для модификации OTA-образов Android A/B-формата с целью получения root-прав при сохранении прохождения AVB (Android Verified Boot) с использованием кастомных (пользовательских) ключей подписи. Она совместима как с Magisk, так и с KernelSU. При необходимости можно просто переподписать OTA, без получения root-доступа.
+avbroot – это утилита для воспроизводимой модификации OTA-образов Android A/B-формата и их переподписания пользовательскими ключами. Она также включает в себя [набор подкоманд](./README.extra.md) для упаковки и распаковки образов Android различных форматов.

 Прежде чем использовать avbroot, рекомендуется иметь хорошее понимание того, как работают AVB и OTA в формате A/B. Как минимум, следует ознакомиться с [разделом предостережений,](#предостережения) чтобы избежать хардбрика устройства.

-**ПРИМЕЧАНИЕ:** avbroot 2.0 была переписана на Rust и больше не имеет в основе никакого кода AOSP, а CLI полностью обратно совместим. Тем не менее, старую реализацию на Python можно найти в одноименной ветке `python`.
-
 ## Требования

 * Поддерживаются только устройства, использующие современную A/B-разметку. Это большинство девайсов, выпускаемых с Android 10 и новее (за исключением устройств от Samsung). Чтобы проверить, использует ли ваш телефон необходимую схему разметки, откройте zip-архив OTA и проверьте:
@@ -21,7 +19,7 @@ avbroot – это программа для модификации OTA-обра

 avbroot модифицирует следующие образы:

-* `boot` или `init_boot`, в зависимости от устройства, модифицируется для получения root-доступа. В случае с Magisk, патч будет эквивалентен тому, что производится в самом приложении Magisk.
+* `boot` или `init_boot`, в зависимости от устройства, модифицируется для получения root-доступа, если это запрашивается.

 * `boot`, `recovery` или `vendor_boot`, в зависимости от устройства, модифицируется для замены сертификата проверки подписи OTA на пользовательский. Это позволяет устанавливать будущие пропатченные OTA через режим Recovery уже после блокировки загрузчика, то есть в качестве обновления. Также это предотвращает случайную установку оригинального непропатченного OTA.

@@ -43,7 +41,7 @@ avbroot модифицирует следующие образы:

 ## Использование

-1. Убедитесь, что вы ознакомились и поняли указанные выше [предостережения.](#предостережения) 
+1. Убедитесь, что вы ознакомились и поняли указанные выше [предостережения.](#предостережения)

 2. Скачайте последнюю версию со страницы [релизов.](https://github.com/chenxiaolong/avbroot/releases) Чтобы сверить цифровую подпись, см. раздел [проверки цифровых подписей.](#проверка-цифровых-подписей)

@@ -107,7 +105,7 @@ avbroot модифицирует следующие образы:

 Первые два компонента подписываются ключом AVB, а последние два – ключом OTA. Можно использовать один и тот же ключ, однако в следующих шагах описано, как сгенерировать два отдельных.

-Если вы патчите OTA сразу для нескольких устройств, настоятельно рекомендуется генерировать уникальные ключи для каждого девайса – так вы защитите себя от случайной прошивки неподходящего OTA для другого телефона. 
+Если вы патчите OTA сразу для нескольких устройств, настоятельно рекомендуется генерировать уникальные ключи для каждого девайса – так вы защитите себя от случайной прошивки неподходящего OTA для другого телефона.

 1. Сгенерируйте ключи подписи для AVB и OTA.

@@ -119,10 +117,10 @@ avbroot модифицирует следующие образы:
 2. Преобразуйте публичную часть ключа подписи AVB в формат метаданных публичного ключа AVB. Именно этот формат используется в загрузчике устройства для установки пользовательского ключа.

    ```bash
-    avbroot key extract-avb -k avb.key -o avb_pkmd.bin
+    avbroot key encode-avb -k avb.key -o avb_pkmd.bin
    ```

-3. Сгенерируйте самоподписанный сертификат для ключа подписи OTA. Он используется режимом Recovery для проверки подписи OTA при сайдлоадинге обновления.
+3. Сгенерируйте самоподписанный сертификат для ключа подписи OTA. Он используется режимом Recovery для проверки подписи OTA при установке обновления.

    ```bash
    avbroot key generate-cert -k ota.key -o ota.crt
@@ -185,7 +183,7 @@ avbroot совместим с любым стандартным 4096-битны
    fastboot flashall --skip-reboot
    ```

-    Обратите внимание, что так прошиваются лишь те образы, что относятся к системе. Разделы загрузчика и модема же остаются нетронутыми из-за ограничений fastboot. Если они не обновлены до необходимой версии, или вы не уверены в этом, после прошивки перейдите к пункту [обновлений](#обновления) и установите пропатченный OTA-архив сайдлоадом в режиме Recovery. Прошивка полного OTA гарантирует, что абсолютно все разделы будут обновлены.
+    Обратите внимание, что так прошиваются лишь те образы, что относятся к системе. Разделы загрузчика и модема же остаются нетронутыми из-за ограничений fastboot. Если они не обновлены до необходимой версии, или вы не уверены в этом, после прошивки перейдите к пункту [обновлений](#обновления) и установите пропатченный OTA в режиме Recovery. Прошивка полного OTA гарантирует, что абсолютно все разделы будут обновлены.

    Для устройств Pixel есть ещё один вариант: запуск скрипта `flash-base.sh` из папки заводских образов (factory images) обновит загрузчик и модем.

@@ -255,11 +253,11 @@ avbroot совместим с любым стандартным 4096-битны

 ## OTA-обновления

-avbroot заменяет `/system/etc/security/otacerts.zip` в разделах системы и Recovery на новый архив, содержащий пользовательский сертификат подписи OTA. Это предотвращает случайную установку непропатченных OTA как при загрузке в Android, так и при сайдлоадинге через Recovery.
+avbroot заменяет `/system/etc/security/otacerts.zip` в разделах системы и Recovery на новый архив, содержащий пользовательский сертификат подписи OTA. Это предотвращает случайную установку непропатченных OTA как из-под загруженной системы, так и при прошивке через Recovery.

-Рекомендуется отключить приложение обновлений системы, чтобы оно не пыталось установить непропатченные OTA:
+Рекомендуется отключить системное приложение для обновлений, чтобы оно не пыталось установить непропатченные OTA:

-* Стоковая прошивка: Отключите `Автоматические обновления системы` (Automatic system updates в англ.) в настройках для разработчиков.
+* Стоковая (заводская) прошивка: Отключите `Автоматические обновления системы` (Automatic system updates в англ.) в настройках для разработчиков.
 * Кастомная прошивка: Отключите приложение обновлений системы (или запретите ему доступ к Интернету) через Настройки -> Приложения -> Все приложения -> (меню/три точки) -> Показать системные -> (найдите приложение обновлений, например Обновления системы/Updater).

 Это особенно важно для некоторых кастомных прошивок, поскольку их фирменное приложение для обновления системы может уйти в бесконечный цикл, загружая OTA-обновление, а затем повторяя попытку загрузки и установки при неудачной проверке подписи.
@@ -295,6 +293,7 @@ Magisk версии 25211 и новее требует наличие разде
        --input /path/to/ota.zip \
        --directory . \
        --boot-only
+        --partition <название раздела> # init_boot или boot, в зависимости от устройства
    ```

 2. Теперь нужно пропатчить загрузочный образ с помощью приложения Magisk. Это **ДОЛЖНО** быть сделано именно на целевом устройстве или устройстве той же модели! Имя раздела будет неверным и не подойдет, если пропатчить образ на устройстве иной модели.
@@ -381,17 +380,34 @@ avbroot может подменить используемый загрузоч

 avbroot можно использовать для простого переподписания OTA, указав аргумент `--rootless` вместо `--magisk`/`--prepatched`. В таком случае пропатченный OTA не будет рутирован. Единственная модификация, которая будет применена – это замена сертификата проверки OTA, чтобы систему можно было обновлять с помощью будущих пропатченных OTA.

-### Пропуск патчинга сертификата OTA в разделе Recovery
+### Пропуск патчинга сертификата OTA

-avbroot может пропустить изменение файла `otacerts.zip` в разделе Recovery с помощью опции `--skip-recovery-ota-cert`. **Не используйте эту функцию, если на то нет веской причины.** (Например, если вы уже самостоятельно встроили сертификат OTA в загрузочный образ (`boot.img`) и передаете его программе через опции `--prepatched` или `--replace`.) Если эта опция применяется совместно с `--rootless` (и без указания параметра `--dsu`), то в загрузочный образ не будут внесены никакие изменения, кроме обеспечения его корректной подписи.
+Вы можете пропустить изменение otacerts.zip, используя аргументы  `--skip-system-ota-cert` и `--skip-recovery-ota-cert`. **Не используйте их без веской причины.**

-Если вы вручную добавили сертификат OTA в загрузочный образ, рекомендуем [предварительно проверить пропатченный OTA.](#проверка-ota)
+При использовании `--skip-system-ota-cert`, сертификаты OTA в образе `system` изменены не будут. Это не позволит сторонним приложениям для OTA-обновлений устанавливать будущие пропатченные OTA из-под загруженной системы.

-### Подмена разделов
+При использовании `--skip-recovery-ota-cert`, сертификаты OTA в образах `vendor_boot` или `recovery` изменены не будут. **Это не позволит устанавливать будущие пропатченные OTA в режиме Recovery.**

-avbroot поддерживает подмену целых образов в OTA, даже тех, что не являются загрузочными (например, `vendor_dlkm`). Образ можно заменить, указав аргумент `--replace <имя раздела> /путь/к/образу.img`.
+Если вы используете аргумент `--skip-recovery-ota-cert`, потому что уже добавили сертификат OTA в загрузочный образ вручную, рекомендуетcя [проверить пропатченный OTA](#проверка-ota), дабы удостовериться, что замена произведена корректно. Процесс верификации проверяет только копию сертификатов OTA в загрузочном образе, не проверяя копию в образе системы.

-Единственное, что меняется – это то, откуда считывается раздел. При использовании `--replace` вместо использования образа раздела из оригинального `payload.bin` в OTA, он берется напрямую по указанному вами пути. Таким образом, заменяющие образы разделов должны иметь правильные колонтитулы vbmeta, соответствующие оригинальным.
+### Пропуск всех патчей
+ 
+Чтобы внести самый минимум изменений, укажите аргументы:
+ 
+* `--skip-system-ota-cert`
+* `--skip-recovery-ota-cert`
+* `--rootless`
+* не используйте аргумент `--dsu`.
+ 
+Так, пользовательскими ключами будут переподписаны лишь образ `vbmeta` и OTA, остальные разделы останутся нетронутыми.
+ 
+**Это следует использовать только для устранения неполадок.** Без патчей сертификатов, поверх полученного OTA не получится установить никакие обновления.	
+
+### Подмена образов
+
+avbroot поддерживает подмену целых образов в OTA, даже тех, что не являются загрузочными (например, `vendor_dlkm`). Образ можно заменить, используя аргумент `--replace <имя раздела> /путь/к/образу.img`.
+
+Единственное, что меняется – это то, откуда считывается файл. При использовании `--replace` вместо образа раздела из оригинального `payload.bin` в OTA, он берется напрямую по указанному вами пути. Заменяющие образы разделов должны иметь правильные колонтитулы vbmeta, соответствующие оригинальным.

 Это не влияет на ход применения пачтей. Например, при использовании Magisk, патч получения root-прав применяется к загрузочному образу одинаково, независимо от того, был ли он получен из оригинального `payload.bin` или это файл, указанный через `--replace`.

@@ -405,6 +421,18 @@ Verified boot is disabled by vbmeta's header flags: 0x3

 Чтобы принудительно включить AVB (очистив флаги), укажите аргумент `--clear-vbmeta-flags`.

+### Изменение алгоритма CoW сжатия для вирутального A/B
+ 
+Алгоритм CoW (copy-on-write) сжатия для виртуального A/B можно изменить, используя аргумент `--vabc-algo <алгоритм>`, указав `gz` или `lz4`. Как правило, по умолчанию OTA использует алгоритм, который совместим с изначальной версией Android, на которой поставлялось устройство.
+ 
+* Девайсы, поставляемые с Android 12, поддерживают `gz` и `brotli` (последний не поддерживается avbroot)
+* Девайсы, поставляемые с Android 14, поддерживают `lz4`
+* Девайсы, поставляемые с Android 15, поддерживают `zstd` (не поддерживается avbroot)
+ 
+Выбор быстрого алгоритма, такого как lz4, может значительно ускорить установку OTA из-под системы (при использованием стороннего приложения для OTA-обновлений). Однако, при установке OTA в режиме Recovery, разницы в скорости не будет.
+ 
+Обратите внимание, что текущая используемая версия Android должна поддерживать выбранный алгоритм сжатия. В противном случае установка завершится ошибкой. Например, попытка установить OTA-обновление с Android 14, использующее алгоритм lz4, приведет к ошибке, если установка производится из-под Android 13.
+
 ### Использование в неинтерактивном режиме

 По умолчанию avbroot интерактивно запрашивает пароли к приватным ключам. Чтобы запустить avbroot в неинтерактивном режиме, можно:
@@ -441,17 +469,20 @@ Verified boot is disabled by vbmeta's header flags: 0x3

 * Использовать незашифрованные приватные ключи. Крайне не рекомендуется.

-### Извлечение всей OTA
+### Извлечение образов из OTA

-Чтобы извлечь все образы, содержащиеся в `payload.bin`, используйте команду:
+Чтобы извлечь образы разделов, содержащихся в `payload.bin`, используйте команду:

 ```bash
 avbroot ota extract \
    --input /путь/к/ota.zip \
-    --directory extracted \
-    --all
+    --directory extracted
 ```

+По умолчанию извлекаются только те образы, которые потенциально могут быть пропатчены с помощью avbroot. Чтобы извлечь все образы, используйте опцию `--all`. Для извлечения конкретных образов используйте опцию `--partition <название раздела>`, которую можно указать несколько раз.
+
+Эта команда также поддерживает извлечение встроенного сертификата OTA и публичного ключа AVB с помощью опций `--cert-ota` и `--public-key-avb`. Чтобы извлечь только эти компоненты, укажите аргумент `--none`, чтобы пропустить извлечение образов разделов.
+
 ### Режим записи ZIP

 По умолчанию, avbroot использует потоковую запись для вывода OTA во время патчинга. Это означает, что он вычисляет дайджест sha256 для цифровой подписи одновременно с записью файла. Такой режим приводит к тому, что в ZIP-файле появляются описатели данных, что является частью стандарта ZIP и работает на подавляющем большинстве устройств. Однако некоторые устройства могут иметь некорректно работающие парсеры ZIP-файлов и не смогут правильно прочитать ZIP-файлы OTA, содержащие описатели данных. Если это так, используйте опцию `--zip-mode seekable` при патчинге.
@@ -480,6 +511,53 @@ avbroot поддерживает делегирование всех опера

 Обратите внимание, что avbroot проверит подпись, возвращенную внешней программой, на соответствие с публичным ключом. Это гарантирует, что процесс патчинга завершится ошибкой, если был использован неправильный приватный ключ.

+### Размер страницы 16 КБ в настройках для разработчиков
+
+На современных устройствах с Android 16 и выше, в настройках для разработчиков может появиться опция переключения на ядро с размером страницы 16 КБ. Однако, эта функция не будет работать в системе, пропатченной с помощью avbroot, поскольку переключение данной настройки осуществляется путём установки инкрементальной OTA:
+
+* `/vendor/boot_otas/boot_ota_16k.zip` — используется для переключения на ядро с размером страницы 16 КБ (в разделе `boot` уже должно быть прошито ядро с размером страницы 4K)  
+* `/vendor/boot_otas/boot_ota_4k.zip` — используется для переключения на ядро с размером страницы 4 КБ (в разделе `boot` уже должно быть прошито ядро с размером страницы 16K)
+
+Эти файлы (в `boot_otas`) невозможно прошить на системе, пропатченной avbroot, потому что `payload.bin` внутри них подписан ключом производителя. Кроме того, это неполноценные OTA-файлы: у них нет метаданных, характерных для OTA, а сам zip-файл не подписан. Это просто обычный архив, который содержит подписанный `payload.bin`.
+
+Поддержка `boot_otas` не планируется. Это потребует реализации функционала для модификации ФС в инкрементальных OTA и их дальнейшей обработки, что сделать очень непросто.
+
+Если вы всё же хотите завести эту функцию, можно попробовать вручную подписать файлы в `boot_otas` собственным ключом. Поскольку инкрементальные OTA не пересоздаются, раздел `boot` должен оставаться без изменений во время выполнения команды `avbroot ota patch`.
+
+1. Распакуйте `vendor.img` с помощью avbroot и [afsr](https://github.com/chenxiaolong/afsr):
+    ```bash
+    avbroot avb unpack -i vendor.img
+    afsr unpack -i raw.img
+    ```
+
+2. Извлеките `payload.bin` из `boot_otas/boot_ota_16k.zip`.
+
+3. Переподпишите `payload.bin` вашим OTA-ключом:
+    ```bash
+    avbroot payload repack \
+        -i payload.bin.orig \
+        -o payload.bin \
+        -k ota.key \
+        --output-properties payload_properties.txt
+    ```
+
+4. Создайте новый zip, включающий `payload.bin` и `payload_properties.txt`. Файлы должны быть добавлены без сжатия (например, с помощью `zip -0`).
+
+5. Повторите эту процедуру для `boot_otas/boot_ota_4k.zip`.
+
+6. Соберите `vendor.img` обратно и подпишите его вашим AVB-ключом:
+    ```bash
+    afsr pack -o raw.img
+    avbroot avb pack -o vendor.img -k avb.key --recompute-size
+    ```
+
+7. Пропатчите обычный OTA-архив с прошивкой, подменив `vendor` на модифицированный образ:
+    ```bash
+    avbroot ota patch \
+        --replace vendor <модифицированный vendor.img> \
+        <дальше указываются аргументы, как при обычном патчинге>
+    ```
+
 ## Сборка из исходного кода

 Убедитесь, что у вас установлен [набор инструментов Rust.](https://www.rust-lang.org/ru/) Затем выполните:
@@ -13,28 +13,31 @@ anyhow = "1.0.75"
 base64 = "0.22.1"
 bitflags = { version = "2.4.1", features = ["serde"] }
 bstr = "1.6.2"
+bzip2 = { version = "0.5.1", default-features = false, features = ["libbz2-rs-sys"] }
 cap-std = "3.0.0"
 cap-tempfile = "3.0.0"
 clap = { version = "4.4.1", features = ["derive"] }
 clap_complete = "4.4.0"
 cms = { version = "0.2.2", features = ["std"] }
+# We can't upgrade to 0.10.0 until x509-cert updates it too, since it's part of
+# the public API.
 const-oid = "0.9.5"
 crc32fast = "1.4.2"
 ctrlc = "3.4.0"
 dlv-list = "0.6.0"
-flate2 = "1.0.27"
+flate2 = { version = "1.0.29", features = ["zlib-rs"] }
 gf256 = { version = "0.3.0", features = ["rs"] }
 hex = { version = "0.4.3", features = ["serde"] }
-liblzma = "0.3.0"
+liblzma = "0.4.1"
 lz4_flex = "0.11.1"
 memchr = "2.6.0"
-miniz_oxide = "0.8.0"
 num-bigint-dig = "0.8.4"
 num-traits = "0.2.16"
 passterm = "2.0.3"
 phf = { version = "0.11.2", features = ["macros"] }
 pkcs8 = { version = "0.10.2", features = ["encryption", "pem"] }
 prost = "0.13.1"
+# We can't upgrade to 0.9.0 until rsa updates its rand_core dependency.
 rand = "0.8.5"
 rayon = "1.7.0"
 regex = { version = "1.9.4", default-features = false, features = ["perf", "std"] }
@@ -42,7 +45,7 @@ regex = { version = "1.9.4", default-features = false, features = ["perf", "std"
 # because sha2 is significantly slower on older x86_64 CPUs without the SHA-NI
 # instructions. sha2 is still used for signing purposes.
 # https://github.com/RustCrypto/hashes/issues/327
-ring = "0.17.0"
+ring = "0.17.14"
 rsa = { version = "0.9.2", features = ["sha1", "sha2"] }
 serde = { version = "1.0.188", features = ["derive"] }
 sha1 = "0.10.5"
@@ -57,13 +60,6 @@ x509-cert = { version = "0.2.4", features = ["builder"] }
 zerocopy = { version = "0.8.10", features = ["std"] }
 zerocopy-derive = "0.8.5"

-# Waiting for next stable release.
-[dependencies.bzip2]
-git = "https://github.com/trifectatechfoundation/bzip2-rs"
-rev = "15258feb0cdc1114d6fc6b936254c4f4e1d5730c"
-default-features = false
-features = ["libbz2-rs-sys"]
-
 # https://github.com/zip-rs/zip/pull/383
 [dependencies.zip]
 git = "https://github.com/chenxiaolong/zip"
@@ -73,12 +69,12 @@ features = ["deflate"]

 [target.'cfg(unix)'.dependencies]
 libc = "0.2.158"
-rustix = { version = "0.38.9", default-features = false, features = ["process"] }
+rustix = { version = "1.0.3", default-features = false, features = ["process"] }

 [build-dependencies]
-constcat = "0.5.0"
+constcat = "0.6.0"
 prost-build = "0.13.1"
-protox = "0.7.0"
+protox = "0.8.0"

 [dev-dependencies]
 assert_matches = "1.5.0"
@@ -30,10 +30,13 @@ pub enum Command {
    Payload(payload::PayloadCli),
    Sparse(sparse::SparseCli),
    /// (Deprecated: Use `avbroot ota patch` instead.)
+    #[command(hide = true)]
    Patch(ota::PatchCli),
    /// (Deprecated: Use `avbroot ota extract` instead.)
+    #[command(hide = true)]
    Extract(ota::ExtractCli),
    /// (Deprecated: Use `avbroot boot magisk-info` instead.)
+    #[command(hide = true)]
    MagiskInfo(boot::MagiskInfoCli),
 }

@@ -46,7 +46,7 @@ pub fn key_main(cli: &KeyCli) -> Result<()> {
            crypto::write_pem_cert_file(&c.output, &cert)
                .with_context(|| format!("Failed to write certificate: {:?}", c.output))?;
        }
-        KeyCommand::ExtractAvb(c) => {
+        KeyCommand::ExtractAvb(c) | KeyCommand::EncodeAvb(c) => {
            let public_key = if let Some(p) = &c.input.key {
                let passphrase = get_passphrase_source(&c.passphrase, p);
                let private_key = crypto::read_pem_key_file(p, &passphrase)
@@ -150,12 +150,12 @@ struct GenerateCertCli {
    validity: u64,
 }

-/// Extract the AVB public key from a private key or certificate.
+/// Convert a key or certificate to an AVB-encoded public key.
 ///
 /// The public key is stored in both the private key and the certificate. Either
 /// one can be used interchangeably.
 #[derive(Debug, Parser)]
-struct ExtractAvbCli {
+struct EncodeAvbCli {
    /// Path to output AVB public key.
    #[arg(short, long, value_name = "FILE", value_parser)]
    output: PathBuf,
@@ -183,7 +183,10 @@ struct DecodeAvbCli {
 enum KeyCommand {
    GenerateKey(GenerateKeyCli),
    GenerateCert(GenerateCertCli),
-    ExtractAvb(ExtractAvbCli),
+    /// (Deprecated: Use `avbroot key encode-avb` instead.)
+    #[command(hide = true)]
+    ExtractAvb(EncodeAvbCli),
+    EncodeAvb(EncodeAvbCli),
    DecodeAvb(DecodeAvbCli),
 }

@@ -122,13 +122,13 @@ fn find_allocated_regions(
    loop {
        stream::check_cancel(cancel_signal)?;

-        start = match rustix::fs::seek(reader, SeekFrom::Data(end as i64)) {
+        start = match rustix::fs::seek(reader, SeekFrom::Data(end)) {
            Ok(offset) => offset,
            Err(e) if e == Errno::NXIO => break,
            Err(e) => return Err(e).with_context(|| format!("Failed to seek to data: {path:?}")),
        };

-        end = rustix::fs::seek(reader, SeekFrom::Hole(start as i64))
+        end = rustix::fs::seek(reader, SeekFrom::Hole(start))
            .with_context(|| format!("Failed to seek to hole: {path:?}"))?;

        result.push(start..end);
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2023-2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2023-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{
@@ -144,6 +144,7 @@ pub enum PassphraseSource {

 impl PassphraseSource {
    pub fn new(key_file: &Path, pass_file: Option<&Path>, env_var: Option<&OsStr>) -> Self {
+        #[allow(clippy::option_if_let_else)]
        if let Some(v) = env_var {
            Self::EnvVar(v.to_owned())
        } else if let Some(p) = pass_file {
@@ -686,21 +687,17 @@ pub fn parse_cms(data: &[u8]) -> Result<SignedData> {
    Ok(sd)
 }

-/// Get a list of all standard X509 certificates contained within a
+/// Get an iterator to all standard X509 certificates contained within a
 /// [`SignedData`] structure.
-pub fn get_cms_certs(sd: &SignedData) -> Vec<Certificate> {
-    sd.certificates.as_ref().map_or_else(Vec::new, |certs| {
-        certs
-            .0
-            .iter()
-            .filter_map(|cc| {
-                if let CertificateChoices::Certificate(c) = cc {
-                    Some(c.clone())
-                } else {
-                    None
-                }
-            })
-            .collect()
+pub fn iter_cms_certs(sd: &SignedData) -> impl Iterator<Item = &Certificate> {
+    sd.certificates.iter().flat_map(|certs| {
+        certs.0.iter().filter_map(|cc| {
+            if let CertificateChoices::Certificate(c) = cc {
+                Some(c)
+            } else {
+                None
+            }
+        })
    })
 }

@@ -163,7 +163,7 @@ pub enum Error {

 type Result<T> = std::result::Result<T, Error>;

-pub(crate) fn ring_algorithm(name: &str, for_verify: bool) -> Result<&'static Algorithm> {
+pub(crate) fn digest_algorithm(name: &str, for_verify: bool) -> Result<&'static Algorithm> {
    match name {
        "sha1" if for_verify => Ok(&ring::digest::SHA1_FOR_LEGACY_USE_ONLY),
        "sha256" => Ok(&ring::digest::SHA256),
@@ -292,7 +292,7 @@ trait DescriptorTag {

 /// Raw on-disk layout for the AVB property descriptor after the prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawPropertyDescriptor {
    key_size: big_endian::U64,
    value_size: big_endian::U64,
@@ -395,7 +395,7 @@ impl<W: Write> ToWriter<W> for PropertyDescriptor {

 /// Raw on-disk layout for the AVB hash tree descriptor after the prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHashTreeDescriptor {
    dm_verity_version: big_endian::U32,
    image_size: big_endian::U64,
@@ -534,7 +534,7 @@ impl HashTreeDescriptor {
        ranges: Option<&[Range<u64>]>,
        cancel_signal: &AtomicBool,
    ) -> Result<()> {
-        let algorithm = ring_algorithm(&self.hash_algorithm, false)?;
+        let algorithm = digest_algorithm(&self.hash_algorithm, false)?;
        let hash_tree = HashTree::new(self.data_block_size, algorithm, &self.salt);
        let (root_digest, hash_tree_data) = match ranges {
            Some(r) => {
@@ -642,7 +642,7 @@ impl HashTreeDescriptor {
    ) -> Result<()> {
        self.check_offsets()?;

-        let algorithm = ring_algorithm(&self.hash_algorithm, true)?;
+        let algorithm = digest_algorithm(&self.hash_algorithm, true)?;

        util::check_bounds(self.tree_size, ..=HASH_TREE_MAX_SIZE)
            .map_err(|e| Error::IntOutOfBounds("HashTree::tree_size", e))?;
@@ -860,7 +860,7 @@ impl<W: Write> ToWriter<W> for HashTreeDescriptor {

 /// Raw on-disk layout for the AVB hash descriptor after the prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHashDescriptor {
    image_size: big_endian::U64,
    hash_algorithm: [u8; 32],
@@ -906,7 +906,7 @@ impl HashDescriptor {
        for_verify: bool,
        cancel_signal: &AtomicBool,
    ) -> Result<ring::digest::Digest> {
-        let algorithm = ring_algorithm(&self.hash_algorithm, for_verify)?;
+        let algorithm = digest_algorithm(&self.hash_algorithm, for_verify)?;
        let mut context = Context::new(algorithm);
        context.update(&self.salt);

@@ -1065,7 +1065,7 @@ impl<W: Write> ToWriter<W> for HashDescriptor {
 /// Raw on-disk layout for the AVB kernel command line descriptor after the
 /// prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawKernelCmdlineDescriptor {
    flags: big_endian::U32,
    cmdline_len: big_endian::U32,
@@ -1139,7 +1139,7 @@ impl<W: Write> ToWriter<W> for KernelCmdlineDescriptor {

 /// Raw on-disk layout for the AVB chain partition descriptor after the prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawChainPartitionDescriptor {
    rollback_index_location: big_endian::U32,
    partition_name_len: big_endian::U32,
@@ -1258,7 +1258,7 @@ impl<W: Write> ToWriter<W> for ChainPartitionDescriptor {

 /// Raw on-disk layout for the AVB descriptor prefix.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawDescriptor {
    tag: big_endian::U64,
    num_bytes_following: big_endian::U64,
@@ -1458,7 +1458,7 @@ impl<'a> TryFrom<&'a mut Descriptor> for AppendedDescriptorMut<'a> {

 /// Raw on-disk layout for the AVB header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`HEADER_MAGIC`].
    magic: [u8; 4],
@@ -1927,7 +1927,7 @@ impl<W: Write> ToWriter<W> for Header {

 /// Raw on-disk layout for the AVB footer.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawFooter {
    /// Magic value. This should be equal to [`FOOTER_MAGIC`].
    magic: [u8; 4],
@@ -2015,10 +2015,9 @@ impl<W: Write> ToWriter<W> for Footer {

 /// Raw on-disk layout for the AVB binary public key header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawPublicKey {
    key_num_bits: big_endian::U32,
-    #[expect(unused)]
    n0inv: big_endian::U32,
 }

@@ -117,7 +117,7 @@ pub trait BootImageExt {

 /// Raw on-disk layout for the v0 image header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawV0 {
    /// Magic value. This should be equal to [`BOOT_MAGIC`].
    magic: [u8; 8],
@@ -139,7 +139,7 @@ struct RawV0 {

 /// Raw on-disk layout for the extra v1 image header fields.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawV1Extra {
    recovery_dtbo_size: little_endian::U32,
    recovery_dtbo_offset: little_endian::U64,
@@ -148,7 +148,7 @@ struct RawV1Extra {

 /// Raw on-disk layout for the extra v2 image header fields.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawV2Extra {
    dtb_size: little_endian::U32,
    dtb_addr: little_endian::U64,
@@ -269,6 +269,7 @@ impl fmt::Display for BootImageV0Through2 {

 impl BootImageExt for BootImageV0Through2 {
    fn header_version(&self) -> u32 {
+        #[allow(clippy::bool_to_int_with_if)]
        if self.v2_extra.is_some() {
            2
        } else if self.v1_extra.is_some() {
@@ -595,7 +596,7 @@ impl<W: Write> ToWriter<W> for BootImageV0Through2 {

 /// Raw on-disk layout for the v3 image header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawV3 {
    /// Magic value. This should be equal to [`BOOT_MAGIC`].
    magic: [u8; 8],
@@ -610,7 +611,7 @@ struct RawV3 {

 /// Raw on-disk layout for the extra v4 image header fields.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawV4Extra {
    signature_size: little_endian::U32,
 }
@@ -961,7 +962,7 @@ impl<W: Write> ToWriter<W> for BootImageV3Through4 {

 /// Raw on-disk layout for the vendor v3 image header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawVendorV3 {
    /// Magic value. This should be equal to [`VENDOR_BOOT_MAGIC`].
    magic: [u8; 8],
@@ -980,7 +981,7 @@ struct RawVendorV3 {

 /// Raw on-disk layout for the extra vendor v4 image header fields.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawVendorV4Extra {
    vendor_ramdisk_table_size: little_endian::U32,
    vendor_ramdisk_table_entry_num: little_endian::U32,
@@ -990,7 +991,7 @@ struct RawVendorV4Extra {

 /// Raw on-disk layout for the vendor v4 ramdisk table entry.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawVendorV4RamdiskTableEntry {
    ramdisk_size: little_endian::U32,
    ramdisk_offset: little_endian::U32,
@@ -76,7 +76,7 @@ pub struct InvalidHexCharError(RawHexU32, char);

 /// ASCII-encoded hex integer value used in cpio header fields.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHexU32([u8; 8]);

 impl fmt::Debug for RawHexU32 {
@@ -121,7 +121,7 @@ impl TryFrom<RawHexU32> for u32 {

 /// Raw on-disk layout for the cpio header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`MAGIC_NEW`] or [`MAGIC_NEW_CRC`].
    magic: [u8; 6],
@@ -592,7 +592,7 @@ impl Fec {

 /// Raw on-disk layout for the FEC image header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`FEC_MAGIC`].
    magic: little_endian::U32,
@@ -431,7 +431,7 @@ impl HashTree {

 /// Raw on-disk layout for our custom hash tree image header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`HashTreeImage::MAGIC`].
    magic: [u8; 16],
@@ -492,8 +492,8 @@ impl HashTreeImage {
    const MAGIC: &'static [u8; 16] = b"avbroot!hashtree";
    const VERSION: u16 = 1;

-    fn ring_algorithm(name: &str) -> Result<&'static Algorithm> {
-        avb::ring_algorithm(name, false)
+    fn digest_algorithm(name: &str) -> Result<&'static Algorithm> {
+        avb::digest_algorithm(name, false)
            .map_err(|_| Error::UnsupportedHashAlgorithm(name.to_owned().into_bytes()))
    }

@@ -509,8 +509,8 @@ impl HashTreeImage {
            .reopen_boxed()
            .and_then(|mut f| f.seek(SeekFrom::End(0)))
            .map_err(Error::InputReopen)?;
-        let ring_algorithm = Self::ring_algorithm(algorithm)?;
-        let hash_tree = HashTree::new(block_size, ring_algorithm, salt);
+        let digest_algorithm = Self::digest_algorithm(algorithm)?;
+        let hash_tree = HashTree::new(block_size, digest_algorithm, salt);
        let (root_digest, hash_tree_data) = hash_tree.generate(input, image_size, cancel_signal)?;

        Ok(Self {
@@ -530,8 +530,8 @@ impl HashTreeImage {
        ranges: &[Range<u64>],
        cancel_signal: &AtomicBool,
    ) -> Result<()> {
-        let ring_algorithm = Self::ring_algorithm(&self.algorithm)?;
-        let hash_tree = HashTree::new(self.block_size, ring_algorithm, &self.salt);
+        let digest_algorithm = Self::digest_algorithm(&self.algorithm)?;
+        let hash_tree = HashTree::new(self.block_size, digest_algorithm, &self.salt);

        self.root_digest = hash_tree.update(
            input,
@@ -550,8 +550,8 @@ impl HashTreeImage {
        input: &(dyn ReadSeekReopen + Sync),
        cancel_signal: &AtomicBool,
    ) -> Result<()> {
-        let ring_algorithm = Self::ring_algorithm(&self.algorithm)?;
-        let hash_tree = HashTree::new(self.block_size, ring_algorithm, &self.salt);
+        let digest_algorithm = Self::digest_algorithm(&self.algorithm)?;
+        let hash_tree = HashTree::new(self.block_size, digest_algorithm, &self.salt);

        hash_tree.verify(
            input,
@@ -1,8 +1,7 @@
-// SPDX-FileCopyrightText: 2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2024-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{
-    cmp::Ordering,
    fmt,
    io::{self, Read, Seek, Write},
    mem,
@@ -152,10 +151,8 @@ pub enum Error {
    ExtentTypeZeroNotEmpty { index: usize },
    #[error("Extent #{index}: Invalid type: {extent_type}")]
    ExtentInvalidType { index: usize, extent_type: u32 },
-    #[error("Extent #{index}: Overlaps previous extent")]
-    ExtentOverlapsPrevious { index: usize },
-    #[error("Extent #{index}: Earlier block device index than previous extent")]
-    ExtentDeviceNotConsecutive { index: usize },
+    #[error("Extent #{index}: Overlaps another extent: #{other}")]
+    ExtentOverlapsAnother { index: usize, other: usize },
    #[error("Extent #{index}: Block device index too large")]
    ExtentDeviceIndexTooLarge { index: usize },
    // Partition group errors.
@@ -253,7 +250,7 @@ impl PartitionAttributes {

 /// Raw on-disk layout for the metadata geometry.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawGeometry {
    /// Magic value. This should be equal to [`GEOMETRY_MAGIC`].
    magic: little_endian::U32,
@@ -335,7 +332,7 @@ impl RawGeometry {

 /// Raw on-disk layout for a table descriptor within a [`RawHeader`].
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawTableDescriptor {
    /// Offset relative to the end of the [`RawHeader`].
    offset: little_endian::U32,
@@ -394,7 +391,7 @@ impl RawTableDescriptor {

 /// Raw on-disk layout for the metadata header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`HEADER_MAGIC`].
    magic: little_endian::U32,
@@ -572,7 +569,7 @@ impl RawHeader {

 /// A potentially invalid raw partition name string.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct PartitionName([u8; 36]);

 impl fmt::Debug for PartitionName {
@@ -598,17 +595,15 @@ impl PartitionName {

    fn validate(&self) -> Result<()> {
        let (prefix, suffix) = self.split();
-        let mut has_alnum = false;

-        for b in prefix {
-            match b {
-                b'a'..=b'z' | b'A'..=b'Z' | b'0'..=b'9' => has_alnum = true,
-                b'_' => {}
-                _ => return Err(Error::PartitionNameInvalid(DebugString::new(self))),
-            }
-        }
+        // AOSP liblp's metadata_format.h says "Characters may only be
+        // alphanumeric or _", but AOSP creates partitions named like
+        // "system_b-cow".
+        let prefix_valid = prefix
+            .iter()
+            .all(|b| matches!(*b, b'a'..=b'z' | b'A'..=b'Z' | b'0'..=b'9' | b'_' | b'-'));

-        if has_alnum && is_zero(suffix) {
+        if prefix_valid && is_zero(suffix) {
            Ok(())
        } else {
            Err(Error::PartitionNameInvalid(DebugString::new(self)))
@@ -644,7 +639,7 @@ impl FromStr for PartitionName {

 /// Raw on-disk layout for an entry in the logical partitions table.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawPartition {
    /// Partition name in ASCII. This must be unique across all partitions.
    name: PartitionName,
@@ -706,7 +701,7 @@ impl RawPartition {
                    .first_extent_index
                    .get()
                    .checked_add(self.num_extents.get())
-                    .map_or(true, |n| n as usize > extents.len())
+                    .is_none_or(|n| n as usize > extents.len())
                {
                    return Err(Error::PartitionExtentIndicesTooLarge {
                        name: DebugString::new(self.name),
@@ -735,7 +730,7 @@ impl RawPartition {

 /// Raw on-disk layout for an entry in the extent table.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawExtent {
    /// Number of [`SECTOR_SIZE`]-byte sectors in this extent.
    num_sectors: little_endian::U64,
@@ -827,7 +822,7 @@ impl RawExtent {

 /// Raw on-disk layout for an entry in the partition groups table.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawPartitionGroup {
    /// Partition group name in ASCII. This must be unique across all groups.
    name: PartitionName,
@@ -894,7 +889,7 @@ impl RawPartitionGroup {

 /// Raw on-disk layout for an entry in the block devices table.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawBlockDevice {
    /// The first [`SECTOR_SIZE`]-byte sector where actual data for the logical
    /// partitions can be allocated.
@@ -1010,23 +1005,29 @@ impl RawMetadataSlot {
            extent.validate(i, &self.block_devices)?;
        }

-        // Ensure that all extents are in increasing order and not overlapping.
-        let mut iter = self
+        // Ensure that all extents are not overlapping. We have to sort here
+        // because the extents may not be in order when loading the super
+        // partition on an actual device. Also, AOSP liblp's `metadata_format.h`
+        // says "Gaps between extents are not allowed", but AOSP frequently
+        // creates this situation after a virtual A/B CoW merge.
+        let mut sorted_extents = self
            .extents
            .iter()
-            .filter(|e| e.target_type.get() == RawExtent::TARGET_TYPE_LINEAR)
-            .enumerate();
-        while let (Some((_, a)), Some((i, b))) = (iter.next(), iter.next()) {
-            match a.target_source.get().cmp(&b.target_source.get()) {
-                Ordering::Equal => {
-                    if a.target_data.get() + a.num_sectors.get() > b.target_data.get() {
-                        return Err(Error::ExtentOverlapsPrevious { index: i });
-                    }
-                }
-                Ordering::Greater => {
-                    return Err(Error::ExtentDeviceNotConsecutive { index: i });
-                }
-                Ordering::Less => {}
+            .enumerate()
+            .filter(|(_, e)| e.target_type.get() == RawExtent::TARGET_TYPE_LINEAR)
+            .collect::<Vec<_>>();
+        sorted_extents.sort_by_key(|(_, e)| (e.target_source, e.target_data));
+
+        for window in sorted_extents.windows(2) {
+            let ((a_i, a), (b_i, b)) = (window[0], window[1]);
+
+            if a.target_source == b.target_source
+                && a.target_data.get() + a.num_sectors.get() > b.target_data.get()
+            {
+                return Err(Error::ExtentOverlapsAnother {
+                    index: b_i,
+                    other: a_i,
+                });
            }
        }

@@ -1060,14 +1061,19 @@ impl RawMetadata {
            .read_exact(&mut buf)
            .map_err(|e| Error::DataRead("geometry", e))?;

-        let image_type = if util::is_zero(&buf) {
-            ImageType::Normal
-        } else {
-            ImageType::Empty
-        };
+        // For non-empty images, AOSP says the first block is supposed to be
+        // filled with zeros, but Samsung puts their own SignerVer02 structure
+        // in there, so we can't rely on that.
+        let mut geometry = RawGeometry::ref_from_prefix(&buf).unwrap().0;

-        let geometry = match image_type {
-            ImageType::Normal => {
+        let image_type = match geometry.validate() {
+            Ok(()) => {
+                // This is an empty image for use with fastboot. These have no
+                // extra padding at the beginning of the file nor backup copies
+                // of the geometry and metadata structs.
+                ImageType::Empty
+            }
+            Err(Error::GeometryInvalidMagic(_)) => {
                // This is an normal non-empty image, which has extra padding at
                // the beginning to avoid having the geometry struct interpreted
                // as a boot sector.
@@ -1077,7 +1083,7 @@ impl RawMetadata {
                    .read_exact(&mut buf)
                    .map_err(|e| Error::DataRead("geometry_primary", e))?;

-                let mut geometry = RawGeometry::ref_from_prefix(&buf).unwrap().0;
+                geometry = RawGeometry::ref_from_prefix(&buf).unwrap().0;

                if geometry.validate().is_ok() {
                    // Skip the backup copy.
@@ -1094,17 +1100,9 @@ impl RawMetadata {
                    geometry.validate()?;
                }

-                geometry
-            }
-            ImageType::Empty => {
-                // This is an empty image for use with fastboot. These have no
-                // extra padding at the beginning of the file nor backup copies
-                // of the geometry and metadata structs.
-                let geometry = RawGeometry::ref_from_prefix(&buf).unwrap().0;
-                geometry.validate()?;
-
-                geometry
+                ImageType::Normal
            }
+            Err(e) => return Err(e),
        };

        Ok((image_type, geometry.to_owned()))
@@ -1642,7 +1640,7 @@ impl TryFrom<&RawMetadataSlot> for MetadataSlot {
    type Error = Error;

    fn try_from(raw_slot: &RawMetadataSlot) -> Result<Self> {
-        let mut slot = MetadataSlot {
+        let mut slot = Self {
            major_version: raw_slot.header.major_version.get(),
            minor_version: raw_slot.header.minor_version.get(),
            groups: Vec::with_capacity(raw_slot.groups.len()),
@@ -1722,7 +1720,7 @@ impl TryFrom<&MetadataSlot> for RawMetadataSlot {
    fn try_from(slot: &MetadataSlot) -> Result<Self> {
        let header_size = RawHeader::size_for_version(slot.major_version, slot.minor_version);

-        let mut raw_slot = RawMetadataSlot {
+        let mut raw_slot = Self {
            header: RawHeader {
                magic: HEADER_MAGIC.into(),
                major_version: slot.major_version.into(),
@@ -1939,7 +1937,7 @@ impl TryFrom<&Metadata> for RawMetadata {
        // We only do the bare minimum calculations needed here to fill out the
        // raw fields. There is no semantic validation.

-        let mut raw_metadata = RawMetadata {
+        let mut raw_metadata = Self {
            image_type: metadata.image_type,
            geometry: RawGeometry {
                magic: GEOMETRY_MAGIC.into(),
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2022-2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2022-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{
@@ -15,7 +15,7 @@ use cms::signed_data::SignedData;
 use const_oid::{db::rfc5912, ObjectIdentifier};
 use memchr::memmem;
 use prost::Message;
-use ring::digest::Context;
+use ring::digest::{Algorithm, Context};
 use thiserror::Error;
 use x509_cert::{der::Encode, Certificate};
 use zip::{result::ZipError, write::FileOptions, CompressionMethod, ZipArchive, ZipWriter};
@@ -377,7 +377,7 @@ fn compute_property_files(
        }

        let remain = l - joined.len();
-        joined.extend(iter::repeat(' ').take(remain));
+        joined.extend(iter::repeat_n(' ', remain));
    }

    Ok(joined)
@@ -565,11 +565,131 @@ pub fn verify_metadata(
    Ok(())
 }

-/// Parse the CMS signature from the OTA zip comment. Returns the decoded CMS
-/// [`SignedData`] structure and the length of the file (from the beginning)
-/// that's covered by the signature. This does not perform any parsing of zip
-/// data structures.
-fn parse_ota_sig(mut reader: impl Read + Seek) -> Result<(SignedData, u64)> {
+#[derive(Clone, Debug)]
+struct RawOtaSignature {
+    /// Decoded CMS structure.
+    signed_data: SignedData,
+    /// Length of the file (from the beginning) that's covered by the signature.
+    hashed_size: u64,
+}
+
+impl RawOtaSignature {
+    pub fn embedded_cert(&self) -> Result<&Certificate> {
+        let mut iter = crypto::iter_cms_certs(&self.signed_data);
+
+        let Some(cert) = iter.next() else {
+            return Err(Error::NotOneCmsCertificate(0));
+        };
+
+        let None = iter.next() else {
+            return Err(Error::NotOneCmsCertificate(2 + iter.count()));
+        };
+
+        Ok(cert)
+    }
+}
+
+#[derive(Clone, Debug)]
+pub struct OtaSignature {
+    pub cert: Certificate,
+    pub digest_algo: &'static Algorithm,
+    pub sig_algo: SignatureAlgorithm,
+    pub sig: Vec<u8>,
+    pub data_size: u64,
+}
+
+impl TryFrom<RawOtaSignature> for OtaSignature {
+    type Error = Error;
+
+    fn try_from(raw_ota_sig: RawOtaSignature) -> Result<Self> {
+        let cert = raw_ota_sig.embedded_cert()?;
+
+        // Make sure this is a signature scheme we can handle. There's currently
+        // no Rust library to verify arbitrary CMS signatures for large files
+        // without fully reading them into memory.
+        let signers_len = raw_ota_sig.signed_data.signer_infos.0.len();
+        if signers_len != 1 {
+            return Err(Error::NotOneCmsSignerInfo(signers_len));
+        }
+
+        let signer = raw_ota_sig.signed_data.signer_infos.0.get(0).unwrap();
+        if signer.digest_alg.oid != rfc5912::ID_SHA_256
+            && signer.digest_alg.oid != rfc5912::ID_SHA_1
+        {
+            return Err(Error::UnsupportedDigestAlgorithm(signer.digest_alg.oid));
+        } else if signer.signature_algorithm.oid != rfc5912::RSA_ENCRYPTION
+            && signer.signature_algorithm.oid != rfc5912::SHA_256_WITH_RSA_ENCRYPTION
+        {
+            return Err(Error::UnsupportedSignatureAlgorithm(
+                signer.signature_algorithm.oid,
+            ));
+        }
+
+        // We support SHA1 for verification only.
+        let (digest_algo, sig_algo) = if signer.digest_alg.oid == rfc5912::ID_SHA_256 {
+            (&ring::digest::SHA256, SignatureAlgorithm::Sha256WithRsa)
+        } else {
+            (
+                &ring::digest::SHA1_FOR_LEGACY_USE_ONLY,
+                SignatureAlgorithm::Sha1WithRsa,
+            )
+        };
+
+        Ok(Self {
+            cert: cert.clone(),
+            digest_algo,
+            sig_algo,
+            sig: signer.signature.as_bytes().to_vec(),
+            data_size: raw_ota_sig.hashed_size,
+        })
+    }
+}
+
+impl OtaSignature {
+    /// Verify an OTA zip against its embedded certificate. This function makes
+    /// no assertion about whether the certificate is actually trusted.
+    ///
+    /// CMS signed attributes are intentionally not supported because AOSP
+    /// recovery does not support them either. It expects the CMS [`SignedData`]
+    /// structure to be used for nothing more than a raw signature transport
+    /// mechanism.
+    pub fn verify_ota(
+        &self,
+        mut reader: impl Read + Seek,
+        cancel_signal: &AtomicBool,
+    ) -> Result<()> {
+        let public_key = crypto::get_public_key(&self.cert).map_err(Error::OtaCertExtractPubKey)?;
+
+        // Manually hash the parts of the file covered by the signature.
+        reader
+            .seek(SeekFrom::Start(0))
+            .map_err(|e| Error::DataRead("raw_data", e))?;
+
+        let mut hashing_reader = HashingReader::new(reader, Context::new(self.digest_algo));
+
+        stream::copy_n(
+            &mut hashing_reader,
+            io::sink(),
+            self.data_size,
+            cancel_signal,
+        )
+        .map_err(|e| Error::DataRead("raw_data", e))?;
+
+        let (_, context) = hashing_reader.finish();
+        let digest = context.finish();
+
+        // Verify the signature against the public key.
+        public_key
+            .verify_sig(self.sig_algo, digest.as_ref(), &self.sig)
+            .map_err(Error::CmsVerify)?;
+
+        Ok(())
+    }
+}
+
+/// Parse the CMS signature from the OTA zip comment. This does not perform any
+/// parsing of zip data structures.
+fn parse_raw_ota_sig(mut reader: impl Read + Seek) -> Result<RawOtaSignature> {
    let file_size = reader
        .seek(SeekFrom::End(0))
        .map_err(|e| Error::DataRead("file_size", e))?;
@@ -614,82 +734,22 @@ fn parse_ota_sig(mut reader: impl Read + Seek) -> Result<(SignedData, u64)> {
    }

    let sig_offset = eocd_size as usize - usize::from(abs_eoc_offset);
-    let sd =
+    let signed_data =
        crypto::parse_cms(&eocd[sig_offset..eocd_size as usize - 6]).map_err(Error::CmsLoad)?;
    // The signature covers everything aside from the archive comment and its
    // length field.
    let hashed_size = file_size - 2 - u64::from(comment_size);

-    Ok((sd, hashed_size))
+    Ok(RawOtaSignature {
+        signed_data,
+        hashed_size,
+    })
 }

-/// Verify an OTA zip against its embedded certificates. This function makes no
-/// assertion about whether the certificate is actually trusted. Returns the
-/// embedded certificate.
-///
-/// CMS signed attributes are intentionally not supported because AOSP recovery
-/// does not support them either. It expects the CMS [`SignedData`] structure to
-/// be used for nothing more than a raw signature transport mechanism.
-pub fn verify_ota(mut reader: impl Read + Seek, cancel_signal: &AtomicBool) -> Result<Certificate> {
-    let (sd, hashed_size) = parse_ota_sig(&mut reader)?;
-
-    // Make sure the certificate in the CMS structure matches the otacert zip
-    // entry.
-    let certs = crypto::get_cms_certs(&sd);
-    if certs.len() != 1 {
-        return Err(Error::NotOneCmsCertificate(certs.len()));
-    }
-
-    let cert = &certs[0];
-    let public_key = crypto::get_public_key(cert).map_err(Error::OtaCertExtractPubKey)?;
-
-    // Make sure this is a signature scheme we can handle. There's currently no
-    // Rust library to verify arbitrary CMS signatures for large files without
-    // fully reading them into memory.
-    if sd.signer_infos.0.len() != 1 {
-        return Err(Error::NotOneCmsSignerInfo(sd.signer_infos.0.len()));
-    }
-
-    let signer = sd.signer_infos.0.get(0).unwrap();
-    if signer.digest_alg.oid != rfc5912::ID_SHA_256 && signer.digest_alg.oid != rfc5912::ID_SHA_1 {
-        return Err(Error::UnsupportedDigestAlgorithm(signer.digest_alg.oid));
-    } else if signer.signature_algorithm.oid != rfc5912::RSA_ENCRYPTION
-        && signer.signature_algorithm.oid != rfc5912::SHA_256_WITH_RSA_ENCRYPTION
-    {
-        return Err(Error::UnsupportedSignatureAlgorithm(
-            signer.signature_algorithm.oid,
-        ));
-    }
-
-    // Manually hash the parts of the file covered by the signature.
-    reader
-        .seek(SeekFrom::Start(0))
-        .map_err(|e| Error::DataRead("raw_data", e))?;
-
-    // We support SHA1 for verification only.
-    let (algorithm, algo) = if signer.digest_alg.oid == rfc5912::ID_SHA_256 {
-        (&ring::digest::SHA256, SignatureAlgorithm::Sha256WithRsa)
-    } else {
-        (
-            &ring::digest::SHA1_FOR_LEGACY_USE_ONLY,
-            SignatureAlgorithm::Sha1WithRsa,
-        )
-    };
-
-    let mut hashing_reader = HashingReader::new(reader, Context::new(algorithm));
-
-    stream::copy_n(&mut hashing_reader, io::sink(), hashed_size, cancel_signal)
-        .map_err(|e| Error::DataRead("raw_data", e))?;
-
-    let (_, context) = hashing_reader.finish();
-    let digest = context.finish();
-
-    // Verify the signature against the public key.
-    public_key
-        .verify_sig(algo, digest.as_ref(), signer.signature.as_bytes())
-        .map_err(Error::CmsVerify)?;
-
-    Ok(cert.clone())
+/// Parse the signature information from the CMS signature embedded in the OTA
+/// zip archive comment.
+pub fn parse_ota_sig(reader: impl Read + Seek) -> Result<OtaSignature> {
+    parse_raw_ota_sig(reader)?.try_into()
 }

 /// Get and parse the protobuf-encoded OTA metadata, the PEM-encoded otacert,
@@ -1,22 +1,26 @@
-// SPDX-FileCopyrightText: 2022-2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2022-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{
    collections::{HashMap, HashSet},
    fmt,
    io::{self, Cursor, Read, Seek, SeekFrom, Write},
-    ops::Range,
+    num::NonZeroU32,
+    ops::{Add, Range},
+    str::FromStr,
    sync::atomic::AtomicBool,
 };

 use base64::engine::general_purpose::STANDARD;
 use base64::Engine;
 use bzip2::write::BzDecoder;
+use flate2::{write::GzEncoder, Compression};
 use liblzma::{
    stream::{Check, Stream},
    write::XzDecoder,
    write::XzEncoder,
 };
+use num_traits::CheckedAdd;
 use prost::Message;
 use rayon::{
    iter::{IndexedParallelIterator, IntoParallelRefMutIterator},
@@ -47,6 +51,11 @@ const PAYLOAD_VERSION: u64 = 2;

 const MANIFEST_MAX_SIZE: usize = 4 * 1024 * 1024;

+/// Size of each extent. This matches what AOSP's delta_generator does. We also
+/// require this to be a multiple of the block size and a multiple of the
+/// maximum CoW compression chunk size.
+const CHUNK_SIZE: u64 = 2 * 1024 * 1024;
+
 #[derive(Debug, Error)]
 pub enum Error {
    #[error("Unknown magic: {0:?}")]
@@ -76,6 +85,10 @@ pub enum Error {
        expected: Option<String>,
        actual: String,
    },
+    #[error("Invalid block size: {0}")]
+    InvalidBlockSize(u32),
+    #[error("Invalid maximum CoW compression chunk size: {0}")]
+    InvalidMaxCompressionChunkSize(u32),
    #[error("Size of {name} ({size}) is not aligned to the block size ({block_size})")]
    InvalidPartitionSize {
        name: String,
@@ -121,6 +134,8 @@ pub enum Error {
    InputOpen(String, #[source] io::Error),
    #[error("Failed to open output file for partition: {0}")]
    OutputOpen(String, #[source] io::Error),
+    #[error("Failed to GZ compress partition image chunk")]
+    GzCompress(#[source] io::Error),
    #[error("Failed to initialize XZ encoder")]
    XzInit(#[source] liblzma::stream::Error),
    #[error("Failed to XZ compress partition image chunk")]
@@ -135,7 +150,7 @@ type Result<T> = std::result::Result<T, Error>;

 /// Raw on-disk layout for the payload header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`PAYLOAD_MAGIC`].
    magic: [u8; 4],
@@ -919,6 +934,8 @@ pub fn extract_images<'a>(
        .collect()
 }

+/// Compress raw data into a chunk to be used with a [`Type::ReplaceXz`]
+/// [`InstallOperation`].
 fn compress_chunk(raw_data: &[u8], cancel_signal: &AtomicBool) -> Result<(Vec<u8>, Digest)> {
    let reader = Cursor::new(raw_data);
    let writer = Cursor::new(Vec::new());
@@ -943,51 +960,390 @@ fn compress_chunk(raw_data: &[u8], cancel_signal: &AtomicBool) -> Result<(Vec<u8
 }

 #[derive(Clone, Copy, Debug, PartialEq, Eq, Deserialize, Serialize)]
-pub enum VabcAlgo {
+pub enum CowVersion {
+    V2,
+    V3,
+}
+
+impl CowVersion {
+    /// Compute the size overhead required to store the headers and footers
+    /// needed for this version of the on-disk CoW format.
+    fn size_overhead(self, cow_replace_ops: u64, payload_install_ops: u64) -> u64 {
+        const BUFFER_REGION_DEFAULT_SIZE: u64 = 2 * 1024 * 1024;
+        const CLUSTER_OPS: u64 = 200;
+        const NUM_RESUME_POINTS: u64 = 4;
+        const SIZEOF_COW_FOOTER_V2: u64 = 84;
+        const SIZEOF_COW_HEADER_V2: u64 = 38;
+        const SIZEOF_COW_HEADER_V3: u64 = SIZEOF_COW_HEADER_V2 + 40;
+        const SIZEOF_COW_OPERATION_V2: u64 = 20;
+        const SIZEOF_COW_OPERATION_V3: u64 = 16;
+        const SIZEOF_RESUME_POINT_V3: u64 = 16;
+
+        let mut overhead = 0;
+
+        match self {
+            Self::V2 => {
+                // sizeof(CowHeader).
+                // AOSP: CowWriterV2::InitPos()
+                overhead += SIZEOF_COW_HEADER_V2;
+
+                // header_.buffer_size. update_engine uses the default value.
+                // AOSP: CowWriterV2::InitPos()
+                overhead += BUFFER_REGION_DEFAULT_SIZE;
+
+                // Add all the CoW operation headers:
+                //
+                // - There is a kCowReplaceOp for each compressed chunk.
+                // - There is a kCowLabelOp for each InstallOperation in the
+                //   payload. This is added by delta_generator in CowDryRun().
+                // - There is a kCowClusterOp at the end of each cluster of
+                //   operations (which includes the kCowClusterOp itself). The
+                //   cluster size used to be 200, but was changed to 1024 in
+                //   5e8e488c13cbff9e0a305ce7c22fd6a13aabb886. We'll use the
+                //   smaller value because it's better to overestimate.
+                //
+                // AOSP: CowWriterV2::EmitClusterIfNeeded()
+                let cow_label_ops = payload_install_ops;
+                let cow_cluster_ops = (cow_replace_ops + cow_label_ops).div_ceil(CLUSTER_OPS - 1);
+
+                // A cluster cannot be truncated, so round up to the nearest
+                // cluster boundary.
+                // AOSP: CowWriterV2::AddOperation()
+                overhead += cow_cluster_ops * CLUSTER_OPS * SIZEOF_COW_OPERATION_V2;
+
+                // sizeof(CowFooter).
+                // AOSP: CowWriterV2::GetCowSizeInfo()
+                overhead += SIZEOF_COW_FOOTER_V2;
+            }
+            Self::V3 => {
+                // AOSP: CowWriterV3::OpenForWrite() -> GetDataOffset()
+                overhead += SIZEOF_COW_HEADER_V3;
+                overhead += BUFFER_REGION_DEFAULT_SIZE;
+                overhead += NUM_RESUME_POINTS * SIZEOF_RESUME_POINT_V3;
+
+                // Add an operation header (sizeof(CowOperationV3)) for each
+                // chunk of compressed data.
+                // AOSP: CowWriterV3::WriteOperation()
+                overhead += cow_replace_ops * SIZEOF_COW_OPERATION_V3;
+            }
+        }
+
+        overhead
+    }
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+enum ChunkingMethod {
+    /// Compress data in block sized chunks each iteration.
+    Exact,
+    /// Compress data in chunks where each chunk is sized at the largest power
+    /// of 2 that's `<=` the specified size and the remaining input size.
+    MaxPowerOf2(NonZeroU32),
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+struct ChunkingParams {
+    block_size: u32,
+    method: ChunkingMethod,
+}
+
+impl ChunkingParams {
+    fn chunk_size(self, num_blocks: u64) -> u32 {
+        match self.method {
+            ChunkingMethod::Exact => self.block_size,
+            ChunkingMethod::MaxPowerOf2(max_chunk_size) => {
+                assert!(
+                    max_chunk_size.is_power_of_two() && max_chunk_size.get() % self.block_size == 0
+                );
+
+                let mut chunk_size = max_chunk_size.get();
+                while chunk_size > self.block_size {
+                    let min_blocks = chunk_size / self.block_size;
+                    if num_blocks >= u64::from(min_blocks) {
+                        return chunk_size;
+                    }
+
+                    chunk_size >>= 1;
+                }
+
+                self.block_size
+            }
+        }
+    }
+}
+
+#[derive(Clone, Copy, Debug, Default, PartialEq, Eq)]
+pub struct CowEstimate {
+    /// Size of estimate in bytes.
+    pub size: u64,
+    /// Number of CoW operations (v3 only).
+    pub num_ops: u64,
+}
+
+impl CowEstimate {
+    /// Add fudge factor to account for overhead.
+    fn fudged(&self, payload_install_ops: u64, cow_version: CowVersion) -> Option<Self> {
+        let version_overhead = cow_version.size_overhead(self.num_ops, payload_install_ops);
+
+        let mut size = self.size.checked_add(version_overhead)?;
+
+        // delta_generator adds 1% overhead to the original CoW size estimate,
+        // even if compression is disabled. We'll do the same too. For the
+        // compressed scenario, we rely on this more because lz4_flex and
+        // zlib-rs usually compress better than the lz4 and zlib implementations
+        // used by libsnapshot_cow.
+        size += size / 100;
+
+        // AOSP: PartitionProcessor::Run()
+        let num_ops = self.num_ops.max(25);
+
+        Some(Self { size, num_ops })
+    }
+}
+
+impl Add for CowEstimate {
+    type Output = Self;
+
+    fn add(self, rhs: Self) -> Self::Output {
+        Self {
+            size: self.size + rhs.size,
+            num_ops: self.num_ops + rhs.num_ops,
+        }
+    }
+}
+
+impl CheckedAdd for CowEstimate {
+    fn checked_add(&self, rhs: &Self) -> Option<Self> {
+        Some(Self {
+            size: self.size.checked_add(rhs.size)?,
+            num_ops: self.num_ops.checked_add(rhs.num_ops)?,
+        })
+    }
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Deserialize, Serialize)]
+pub enum VabcAlgoKind {
+    None,
    Lz4,
-    Gzip,
+    Gz,
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq, Deserialize, Serialize)]
+pub struct VabcAlgo {
+    /// Compression algorithm.
+    pub kind: VabcAlgoKind,
+    /// Compression level. AOSP allows this to be present even if the algorithm
+    /// can't use it.
+    pub level: Option<u32>,
 }

 impl VabcAlgo {
-    pub fn new(name: &str) -> Option<Self> {
-        match name {
-            "lz4" => Some(Self::Lz4),
-            "gz" => Some(Self::Gzip),
-            _ => None,
-        }
-    }
+    /// Compute the compressed size of the raw data when split into chunks based
+    /// on the specified [`ChunkingParams`]. The length of `raw_data` must be a
+    /// multiple of the block size or else this will panic. The compressed data
+    /// for each chunk is temporarily stored in memory, but discarded after each
+    /// loop iteration.
+    fn compressed_size(self, mut raw_data: &[u8], chunking: ChunkingParams) -> Result<CowEstimate> {
+        assert!(raw_data.len() as u64 % u64::from(chunking.block_size) == 0);

-    fn compressed_size(self, mut raw_data: &[u8], block_size: u32) -> u64 {
-        let mut total = 0;
+        let mut size = 0;
+        let mut num_ops = 0;

        while !raw_data.is_empty() {
-            let n = raw_data.len().min(block_size as usize);
-            let compressed = match self {
-                Self::Lz4 => lz4_flex::block::compress(&raw_data[..n]),
-                // We use the miniz_oxide backend for flate2, but flate2 doesn't
-                // expose a nice function for compressing to a vec, so just use
-                // miniz_oxide directly.
-                Self::Gzip => miniz_oxide::deflate::compress_to_vec_zlib(&raw_data[..n], 9),
+            let num_blocks = raw_data.len() as u64 / u64::from(chunking.block_size);
+            let chunk_size = chunking.chunk_size(num_blocks) as usize;
+            let (chunk, remaining) = raw_data.split_at(chunk_size);
+
+            // This should match CompressWorker::GetDefaultCompressionLevel() in
+            // AOSP's libsnapshot.
+            //
+            // CoW v3 uses the raw data instead of the compressed data if the
+            // raw data is smaller. Because we use a different implementation of
+            // the compression algorithms, we don't implement this. It's safer
+            // to just overestimate and use the (larger) compressed size.
+            size += match self.kind {
+                VabcAlgoKind::None => chunk_size as u64,
+                VabcAlgoKind::Lz4 => lz4_flex::block::compress(chunk).len() as u64,
+                VabcAlgoKind::Gz => {
+                    let level = self.level.map_or(Compression::best(), Compression::new);
+                    let mut encoder = GzEncoder::new(Vec::new(), level);
+                    encoder.write_all(chunk).map_err(Error::GzCompress)?;
+                    encoder.finish().map_err(Error::GzCompress)?.len() as u64
+                }
            };

-            total += compressed.len().min(n) as u64;
-
-            raw_data = &raw_data[n..];
+            num_ops += 1;
+            raw_data = remaining;
        }

-        total
+        Ok(CowEstimate { size, num_ops })
    }
 }

 impl fmt::Display for VabcAlgo {
    fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
-        match self {
-            Self::Lz4 => f.write_str("lz4"),
-            Self::Gzip => f.write_str("gz"),
+        let name = match self.kind {
+            VabcAlgoKind::None => "none",
+            VabcAlgoKind::Lz4 => "lz4",
+            VabcAlgoKind::Gz => "gz",
+        };
+
+        f.write_str(name)?;
+
+        if let Some(level) = self.level {
+            write!(f, ",{level}")?;
        }
+
+        Ok(())
    }
 }

+#[derive(Clone, Debug, Error)]
+#[error("Invalid VABC algorithm: {0:?} (must be {{none|lz4|gz}}[,<level>])")]
+pub struct InvalidVabcAlgo(String);
+
+impl FromStr for VabcAlgo {
+    type Err = InvalidVabcAlgo;
+
+    fn from_str(s: &str) -> std::result::Result<Self, Self::Err> {
+        let (prefix, suffix) = s.split_once(',').unwrap_or((s, ""));
+
+        // AOSP allows any algorithm to accept a level, even if it's unused.
+        let level = if !suffix.is_empty() {
+            Some(suffix.parse().map_err(|_| InvalidVabcAlgo(s.to_owned()))?)
+        } else {
+            None
+        };
+
+        let kind = match prefix {
+            "" | "none" => VabcAlgoKind::None,
+            "lz4" => VabcAlgoKind::Lz4,
+            "gz" => VabcAlgoKind::Gz,
+            _ => return Err(InvalidVabcAlgo(s.to_owned())),
+        };
+
+        Ok(Self { kind, level })
+    }
+}
+
+#[derive(Clone, Copy, Debug, PartialEq, Eq)]
+pub struct VabcParams {
+    /// CoW on-disk format version.
+    pub version: CowVersion,
+    /// CoW compression algorithm.
+    pub algo: VabcAlgo,
+    /// The maximum number of bytes to compress at a time.
+    pub compression_factor: u32,
+}
+
+/// Ensure that the partition size is aligned to the block size and that the
+/// block size and compression factor are factors of our [`CHUNK_SIZE`].
+fn validate_partition_size(
+    partition_name: &str,
+    file_size: u64,
+    block_size: u32,
+    compression_factor: u32,
+) -> Result<()> {
+    if block_size == 0 || !block_size.is_power_of_two() || CHUNK_SIZE % u64::from(block_size) != 0 {
+        return Err(Error::InvalidBlockSize(block_size));
+    }
+
+    if compression_factor == 0
+        || !compression_factor.is_power_of_two()
+        || CHUNK_SIZE % u64::from(compression_factor) != 0
+    {
+        return Err(Error::InvalidMaxCompressionChunkSize(compression_factor));
+    }
+
+    if file_size % u64::from(block_size) != 0 {
+        return Err(Error::InvalidPartitionSize {
+            name: partition_name.to_owned(),
+            size: file_size,
+            block_size,
+        });
+    }
+
+    Ok(())
+}
+
+/// Compute the VABC CoW size estimate. For a more accurate size estimate with
+/// CoW version 2, `payload_install_ops` must be equal to the number of
+/// [`InstallOperation`]s in the payload. The caller must update
+/// [`PartitionUpdate::estimate_cow_size`] and
+/// [`PartitionUpdate::estimate_op_count_max`] or else update_engine may fail to
+/// flash the partition due to running out of space on the CoW block device.
+pub fn compute_cow_estimate(
+    input: &(dyn ReadSeekReopen + Sync),
+    payload_install_ops: u64,
+    partition_name: &str,
+    block_size: u32,
+    vabc_params: VabcParams,
+    cancel_signal: &AtomicBool,
+) -> Result<CowEstimate> {
+    let file_size = input
+        .reopen_boxed()
+        .and_then(|mut r| r.seek(SeekFrom::End(0)))
+        .map_err(|e| Error::InputOpen(partition_name.to_owned(), e))?;
+    let final_chunk_different = file_size % CHUNK_SIZE != 0;
+
+    validate_partition_size(
+        partition_name,
+        file_size,
+        block_size,
+        vabc_params.compression_factor,
+    )?;
+
+    let chunking = ChunkingParams {
+        block_size,
+        method: match vabc_params.version {
+            CowVersion::V2 => ChunkingMethod::Exact,
+            CowVersion::V3 => {
+                ChunkingMethod::MaxPowerOf2(vabc_params.compression_factor.try_into().unwrap())
+            }
+        },
+    };
+
+    let chunks_total = file_size.div_ceil(CHUNK_SIZE);
+
+    let initial_estimate = (0..chunks_total)
+        .into_par_iter()
+        .map(|chunk| -> Result<CowEstimate> {
+            let data = (|| {
+                let mut reader = input.reopen_boxed()?;
+                reader.seek(SeekFrom::Start(chunk * CHUNK_SIZE))?;
+
+                let chunk_size = if final_chunk_different && chunk == chunks_total - 1 {
+                    file_size % CHUNK_SIZE
+                } else {
+                    CHUNK_SIZE
+                };
+
+                stream::check_cancel(cancel_signal)?;
+                reader.read_vec_exact(chunk_size as usize)
+            })()
+            .map_err(Error::ChunkRead)?;
+
+            vabc_params.algo.compressed_size(&data, chunking)
+        })
+        .try_fold(
+            CowEstimate::default,
+            |total, chunk_estimate| -> Result<CowEstimate> {
+                total
+                    .checked_add(&chunk_estimate?)
+                    .ok_or(Error::IntOverflow("initial_estimate"))
+            },
+        )
+        .try_reduce(CowEstimate::default, |total, partial| {
+            total
+                .checked_add(&partial)
+                .ok_or(Error::IntOverflow("initial_estimate"))
+        })?;
+
+    initial_estimate
+        .fudged(payload_install_ops, vabc_params.version)
+        .ok_or(Error::IntOverflow("fudged_estimate"))
+}
+
 /// Compress the image and return the corresponding information to insert into
 /// the payload manifest's [`PartitionUpdate`] instance. The uncompressed data
 /// is split into 2 MiB chunks, which are read and compressed in parallel, and
@@ -996,21 +1352,17 @@ impl fmt::Display for VabcAlgo {
 /// update [`InstallOperation::data_offset`] in each operation manually because
 /// the initial values are relative to 0.
 ///
-/// If `vabc_algo` is set, the VABC CoW v2 size estimate will be computed. The
-/// caller must update [`PartitionUpdate::estimate_cow_size`] with this value or
-/// else update_engine may fail to flash the partition due to running out of
-/// space on the CoW block device. CoW v2 + other algorithms and also CoW v3 are
-/// currently unsupported because there currently are no known OTAs that use
-/// those configurations.
+/// If `vabc_algo` is set, the VABC CoW size estimate will also be computed.
+/// This is more efficient than separately calling [`compute_cow_estimate`]
+/// since the input does not need to be read twice.
 pub fn compress_image(
    input: &(dyn ReadSeekReopen + Sync),
    output: &(dyn WriteSeekReopen + Sync),
    partition_name: &str,
    block_size: u32,
-    vabc_algo: Option<VabcAlgo>,
+    vabc_params: Option<VabcParams>,
    cancel_signal: &AtomicBool,
-) -> Result<(PartitionInfo, Vec<InstallOperation>, Option<u64>)> {
-    const CHUNK_SIZE: u64 = 2 * 1024 * 1024;
+) -> Result<(PartitionInfo, Vec<InstallOperation>, Option<CowEstimate>)> {
    const CHUNK_GROUP: u64 = 32;

    let file_size = input
@@ -1019,18 +1371,23 @@ pub fn compress_image(
        .map_err(|e| Error::InputOpen(partition_name.to_owned(), e))?;
    let final_chunk_different = file_size % CHUNK_SIZE != 0;

-    if file_size % u64::from(block_size) != 0 || CHUNK_SIZE % u64::from(block_size) != 0 {
-        return Err(Error::InvalidPartitionSize {
-            name: partition_name.to_owned(),
-            size: file_size,
-            block_size,
-        });
-    }
+    let compression_factor = vabc_params.map_or(block_size, |p| p.compression_factor);
+    validate_partition_size(partition_name, file_size, block_size, compression_factor)?;
+
+    let chunking = ChunkingParams {
+        block_size,
+        method: match vabc_params.map(|p| p.version) {
+            Some(CowVersion::V3) => {
+                ChunkingMethod::MaxPowerOf2(compression_factor.try_into().unwrap())
+            }
+            _ => ChunkingMethod::Exact,
+        },
+    };

    let chunks_total = file_size.div_ceil(CHUNK_SIZE);
-    let mut bytes_compressed = 0;
+    let mut bytes_compressed = 0u64;
    let mut context_uncompressed = Context::new(&ring::digest::SHA256);
-    let mut cow_estimate = 0;
+    let mut initial_estimate = CowEstimate::default();
    let mut operations = vec![];

    // Read the file one group at a time. This allows for some parallelization
@@ -1067,10 +1424,12 @@ pub fn compress_image(
        let mut compressed_data_group = uncompressed_data_group
            .into_par_iter()
            .map(
-                |(raw_offset, raw_data)| -> Result<(Vec<u8>, InstallOperation, u64)> {
+                |(raw_offset, raw_data)| -> Result<(Vec<u8>, InstallOperation, CowEstimate)> {
                    let (data, digest_compressed) = compress_chunk(&raw_data, cancel_signal)?;
-                    let cow_size =
-                        vabc_algo.map_or(0, |a| a.compressed_size(&raw_data, block_size));
+                    let cow_estimate = vabc_params
+                        .map(|p| p.algo.compressed_size(&raw_data, chunking))
+                        .transpose()?
+                        .unwrap_or_default();

                    let extent = Extent {
                        start_block: Some(raw_offset / u64::from(block_size)),
@@ -1083,15 +1442,19 @@ pub fn compress_image(
                    operation.dst_extents.push(extent);
                    operation.data_sha256_hash = Some(digest_compressed.as_ref().to_vec());

-                    Ok((data, operation, cow_size))
+                    Ok((data, operation, cow_estimate))
                },
            )
            .collect::<Result<Vec<_>>>()?;

-        for (data, operation, cow_size) in &mut compressed_data_group {
+        for (data, operation, cow_estimate) in &mut compressed_data_group {
            operation.data_offset = Some(bytes_compressed);
-            bytes_compressed += data.len() as u64;
-            cow_estimate += *cow_size;
+            bytes_compressed = bytes_compressed
+                .checked_add(data.len() as u64)
+                .ok_or(Error::IntOverflow("bytes_compressed"))?;
+            initial_estimate = initial_estimate
+                .checked_add(cow_estimate)
+                .ok_or(Error::IntOverflow("initial_estimate"))?;
        }

        let group_operations = compressed_data_group
@@ -1115,25 +1478,12 @@ pub fn compress_image(
        hash: Some(digest_uncompressed.as_ref().to_vec()),
    };

-    let cow_estimate = if vabc_algo.is_some() {
-        // lz4_flex and miniz_oxide usually compress better than the lz4 and
-        // zlib implementations used by libsnapshot_cow. Make up for this by
-        // adding percentage-based overhead.
-        cow_estimate += cow_estimate / 100;
-
-        // We also need to account for constant overhead, especially with
-        // smaller partitions. We can match what delta_generator normally adds
-        // in CowWriterV2::InitPos() exactly. Since we only ever create full
-        // OTAs, we can assume that all CoW operations are kCowReplaceOp.
-
-        // sizeof(CowHeader).
-        cow_estimate += 38;
-        // header_.buffer_size (equal to BUFFER_REGION_DEFAULT_SIZE).
-        cow_estimate += 2 * 1024 * 1024;
-        // CowOptions::cluster_ops * sizeof(CowOperationV2).
-        cow_estimate += 200 * 20;
-
-        Some(cow_estimate)
+    let cow_estimate = if let Some(p) = vabc_params {
+        Some(
+            initial_estimate
+                .fudged(operations.len() as u64, p.version)
+                .ok_or(Error::IntOverflow("fudged_estimate"))?,
+        )
    } else {
        None
    };
@@ -1280,6 +1630,8 @@ pub fn compress_modified_image(
                writer.seek(SeekFrom::Start(operation.data_offset.unwrap()))?;
                writer.write_all(&data)?;

+                // Clippy doesn't know we're returning a Range.
+                #[allow(clippy::range_plus_one)]
                Ok(i..i + 1)
            })
            .collect::<io::Result<Vec<_>>>()
@@ -106,7 +106,7 @@ type Result<T> = std::result::Result<T, Error>;

 /// Raw on-disk layout for the header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawHeader {
    /// Magic value. This should be equal to [`HEADER_MAGIC`].
    magic: little_endian::U32,
@@ -159,7 +159,7 @@ impl RawHeader {
            return Err(Error::UnsupportedMajorVersion(self.major_version.get()));
        }

-        if self.file_hdr_sz.get() < mem::size_of::<RawHeader>() as u16 {
+        if self.file_hdr_sz.get() < mem::size_of::<Self>() as u16 {
            return Err(Error::InvalidFileHeaderSize(self.file_hdr_sz.get()));
        } else if self.chunk_hdr_sz.get() < mem::size_of::<RawChunk>() as u16 {
            return Err(Error::InvalidChunkHeaderSize(self.chunk_hdr_sz.get()));
@@ -173,7 +173,7 @@ impl RawHeader {
    }

    fn excess_raw_header_bytes(&self) -> u16 {
-        self.file_hdr_sz.get() - mem::size_of::<RawHeader>() as u16
+        self.file_hdr_sz.get() - mem::size_of::<Self>() as u16
    }

    fn excess_raw_chunk_bytes(&self) -> u16 {
@@ -183,7 +183,7 @@ impl RawHeader {

 /// Raw on-disk layout for the chunk header.
 #[derive(Clone, Copy, FromBytes, IntoBytes, KnownLayout, Immutable, Unaligned)]
-#[repr(packed)]
+#[repr(C, packed)]
 struct RawChunk {
    /// Chunk type. Must be [`CHUNK_TYPE_RAW`], [`CHUNK_TYPE_FILL`],
    /// [`CHUNK_TYPE_DONT_CARE`], or [`CHUNK_TYPE_CRC32`].
@@ -811,7 +811,7 @@ impl<R: Read> SparseReader<R> {
                data = ChunkData::Crc32(expected.get());
            }
            _ => unreachable!(),
-        };
+        }

        let chunk = Chunk {
            bounds: ChunkBounds {
@@ -432,7 +432,7 @@ impl BootImagePatch for MagiskRootPatcher {
            targets.push("init_boot");
        } else if boot_images.contains_key("boot") {
            targets.push("boot");
-        };
+        }

        Ok(targets)
    }
@@ -781,7 +781,7 @@ impl DsuPubKeyPatcher {
            e.data = data;
        } else {
            entries.push(CpioEntry::new_file(Self::AVBROOT_KEY_PATH, 0o644, data));
-        };
+        }

        *ramdisk = save_ramdisk(&entries, ramdisk_format, cancel_signal)?;

@@ -971,7 +971,7 @@ impl BootImagePatch for PrepatchedImagePatcher {
            targets.push("init_boot");
        } else if boot_images.contains_key("boot") {
            targets.push("boot");
-        };
+        }

        Ok(targets)
    }
@@ -1,3 +1,4 @@
+#![allow(clippy::all)]
 #![allow(clippy::nursery)]
 #![allow(clippy::pedantic)]

@@ -37,8 +37,8 @@ allow = [
    "GPL-3.0",
    "ISC",
    "MIT",
-    "OpenSSL",
    "Unicode-3.0",
+    "Zlib",
 ]

 [[licenses.clarify]]
@@ -70,5 +70,4 @@ unknown-registry = "deny"
 unknown-git = "deny"
 allow-git = [
    "https://github.com/chenxiaolong/zip",
-    "https://github.com/trifectatechfoundation/bzip2-rs",
 ]
@@ -14,7 +14,7 @@ avbroot = { path = "../avbroot" }
 clap = { version = "4.4.1", features = ["derive"] }
 ctrlc = "3.4.0"
 hex = { version = "0.4.3", features = ["serde"] }
-ring = "0.17.0"
+ring = "0.17.14"
 rsa = { version = "0.9.6", features = ["hazmat"] }
 serde = { version = "1.0.188", features = ["derive"] }
 tempfile = "3.8.0"
@@ -12,8 +12,10 @@ security_patch_level = "2024-01-01"
 # Google Pixel 7 Pro
 # What's unique: init_boot (boot v4) + vendor_boot (vendor v4)

-[profile.pixel_v4_gki]
-vabc_algo = "Lz4"
+[profile.pixel_v4_gki.vabc]
+# CoW v3 is used starting with the Google Pixel 9a.
+version = "V3"
+algo = { kind = "Lz4" }

 [profile.pixel_v4_gki.partitions.boot]
 avb.signed = true
@@ -49,18 +51,19 @@ data.version = "vendor_v4"
 data.ramdisks = [["otacerts", "first_stage", "dsu_key_dir"]]

 [profile.pixel_v4_gki.hashes_streaming]
-original = "c00f891f941f3dddb28966f7b07f3acea773bee104dace82b37c2d1341f09422"
-patched = "ce9d8ee97828d233809742a5d3f23aa27b042675b1935ca9e3df0592c55788fd"
+original = "c83bff30dbd30ed8baf5a954a158a2c3a75b25175089809e52ea9052020b0627"
+patched = "20bd1835c6ea90104eab7ec51be48ee8aae76508f8a6501621d7a1f384103c2f"

 [profile.pixel_v4_gki.hashes_seekable]
-original = "96a6c366b5de1c3b10d4d6cb4ca503c83ac4cd9ca952a965cceb041990ba7022"
-patched = "e7b4609ba7a23609211dcae143bc43f091f286fbbb3a9301c02ee25614d35deb"
+original = "ffc5c7839dfa68d5ff7d888287f808259be34bd63c5df0a385e2e78e6fdcc647"
+patched = "90a51e890560486d4bd73685176c833c94f31f9d05d524186bb4c544c8eca32c"

 # Google Pixel 6a
 # What's unique: boot (boot v4, no ramdisk) + vendor_boot (vendor v4, 2 ramdisks)

-[profile.pixel_v4_non_gki]
-vabc_algo = "Lz4"
+[profile.pixel_v4_non_gki.vabc]
+version = "V2"
+algo = { kind = "Lz4" }

 [profile.pixel_v4_non_gki.partitions.boot]
 avb.signed = true
@@ -90,18 +93,19 @@ data.version = "vendor_v4"
 data.ramdisks = [["init", "otacerts", "first_stage", "dsu_key_dir"], ["dlkm"]]

 [profile.pixel_v4_non_gki.hashes_streaming]
-original = "4d692bc777b568b0626d3c08d2e6f83f1b472db5ad903486daaec6a78d0cc26e"
-patched = "e27673e4f30933710c11d51f0e73849068cbe9bc9f54e6076bdd93f9a5c8ea0a"
+original = "8d76f17b33949c8ca3d4d5872858bb725d755e315f67d61db0e0b7b58cf69685"
+patched = "7c7cb69087d4c8c54a7f4bac6525354b1e9c1550545bc954194aaf6882f7982d"

 [profile.pixel_v4_non_gki.hashes_seekable]
-original = "ea27ecd9718c17b63400b2548680bb3cee93ce63b4fc44ff9654ca0d9c5372a8"
-patched = "3456b14e014cf565a808a9e834d9105a23539f07b2c460db19c9384aadbc3b93"
+original = "6825bc02115e0c64f05651571138fbcc5cc04459e5c7e225dcc5e3096051c341"
+patched = "373309f28234866625462253d6f41f5a3b016b3676c7cb87a88bf55099ad6b13"

 # Google Pixel 4a 5G
 # What's unique: boot (boot v3) + vendor_boot (vendor v3)

-[profile.pixel_v3]
-vabc_algo = "Lz4"
+[profile.pixel_v3.vabc]
+version = "V2"
+algo = { kind = "Gz" }

 [profile.pixel_v3.partitions.boot]
 avb.signed = true
@@ -132,19 +136,16 @@ data.version = "vendor_v3"
 data.ramdisks = [["otacerts", "first_stage", "dsu_key_dir"]]

 [profile.pixel_v3.hashes_streaming]
-original = "f432dc7931520feb238474aa707dd5299747562ffe6129f3f763b5f11ac473ab"
-patched = "3850a2e73bd783a1ec4a70c59f37d2374e017c20df7ab4b591182b14d187c18e"
+original = "678f5ba8cff01802ce964d31abf7a2c77b63e9c06e911857b588a16872fb6860"
+patched = "3691c2cf89728307e143b7084b40bbbfb181e15be6f65914a6c40408e0c6eab3"

 [profile.pixel_v3.hashes_seekable]
-original = "7d29ecc6780953c22052a576b8dc85066c8667a875e918a786a08ff4545b47d1"
-patched = "9f6342940b7cfbeb27b0567f006bb35cbee910ef038ec535403c662d5252ca71"
+original = "202d447215f80919f464bd59c7b38f20cbe3a9625d38bc81a0b71e5a3da334eb"
+patched = "ec94739fdef7efc4930e96828761c2ebd4ece79014b735eab303c5cbdda2e529"

 # Google Pixel 4a
 # What's unique: boot (boot v2)

-[profile.pixel_v2]
-vabc_algo = "Gzip"
-
 [profile.pixel_v2.partitions.boot]
 avb.signed = false
 data.type = "boot"
@@ -168,9 +169,9 @@ data.type = "vbmeta"
 data.deps = ["system"]

 [profile.pixel_v2.hashes_streaming]
-original = "1b45235b58054009cc496f6c3ee11d3dc16ed5c388c861761e26a6fce83103a0"
-patched = "193b2dc70dd465d686f35c7b7f74d2cc1b06a55e48cf5c2e4df0f667e03032fc"
+original = "79bb20bd57f51a8a7b52d0ff59b7d096a1e4ada425a6230818ca25916a81e021"
+patched = "7f199cc4a6fd244034b597cb916651e1c3e0d6f59450617a5cadbd274f02ec7a"

 [profile.pixel_v2.hashes_seekable]
-original = "52284308fae10cbaf09ade14e92f3bbe6149751a42bff15432982fcef8d890ab"
-patched = "7ad74ac87ddcaf34938017e6149a646041d70926e31ecda93e156e9397467b3b"
+original = "c9b32282f247555fed92cc6d599b9005c81ffdc88d5f55d2248e8bf6fb66f1b5"
+patched = "c856f7f7245c02d32b3cec93c2d4365a2b3823fed9a9b27f94508637d79b1492"
@@ -1,14 +1,14 @@
-// SPDX-FileCopyrightText: 2023-2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2023-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{collections::BTreeMap, fs, path::Path};

 use anyhow::{Context, Result};
-use avbroot::format::payload::VabcAlgo;
+use avbroot::format::payload::{CowVersion, VabcAlgo};
 use serde::{Deserialize, Serialize};
 use toml_edit::DocumentMut;

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Copy, Serialize, Deserialize)]
 pub struct Sha256Hash(
    #[serde(
        serialize_with = "hex::serialize",
@@ -17,7 +17,7 @@ pub struct Sha256Hash(
    pub [u8; 32],
 );

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct OtaInfo {
    pub device: String,
@@ -29,7 +29,7 @@ pub struct OtaInfo {
    pub security_patch_level: String,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Copy, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct Avb {
    pub signed: bool,
@@ -55,7 +55,7 @@ pub enum BootVersion {
    VendorV4,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct BootData {
    pub version: BootVersion,
@@ -71,19 +71,19 @@ pub enum DmVerityContent {
    SystemOtacerts,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Copy, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct DmVerityData {
    pub content: DmVerityContent,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct VbmetaData {
    pub deps: Vec<String>,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(tag = "type", rename_all = "snake_case")]
 pub enum Data {
    Boot(BootData),
@@ -91,30 +91,37 @@ pub enum Data {
    Vbmeta(VbmetaData),
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Copy, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct Hashes {
    pub original: Sha256Hash,
    pub patched: Sha256Hash,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct Partition {
    pub avb: Avb,
    pub data: Data,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Copy, Serialize, Deserialize)]
+#[serde(deny_unknown_fields)]
+pub struct VabcSettings {
+    pub version: CowVersion,
+    pub algo: VabcAlgo,
+}
+
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct Profile {
-    pub vabc_algo: Option<VabcAlgo>,
+    pub vabc: Option<VabcSettings>,
    pub partitions: BTreeMap<String, Partition>,
    pub hashes_streaming: Hashes,
    pub hashes_seekable: Hashes,
 }

-#[derive(Serialize, Deserialize)]
+#[derive(Clone, Serialize, Deserialize)]
 #[serde(deny_unknown_fields)]
 pub struct Config {
    pub ota_info: OtaInfo,
@@ -1,4 +1,4 @@
-// SPDX-FileCopyrightText: 2023-2024 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2023-2025 Andrew Gunnerson
 // SPDX-FileCopyrightText: 2023 Pascal Roeleven
 // SPDX-License-Identifier: GPL-3.0-only

@@ -36,7 +36,7 @@ use avbroot::{
        cpio::{self, CpioEntry, CpioEntryData},
        ota::{self, SigningWriter, ZipEntry, ZipMode},
        padding,
-        payload::{self, PayloadHeader, PayloadWriter},
+        payload::{self, CowVersion, PayloadHeader, PayloadWriter, VabcParams},
    },
    patch::otacert::{self, OtaCertBuildFlags},
    protobuf::{
@@ -97,7 +97,7 @@ fn verify_hash(path: &Path, sha256: &[u8; 32], cancel_signal: &AtomicBool) -> Re
 fn append_avb(
    file: &mut PSeekFile,
    name: &str,
-    avb: &Avb,
+    avb: Avb,
    hash_tree: bool,
    ota_info: &OtaInfo,
    key_avb: &RsaSigningKey,
@@ -295,7 +295,7 @@ fn create_ramdisk(
 fn create_boot_image(
    file: &mut PSeekFile,
    name: &str,
-    avb: &Avb,
+    avb: Avb,
    boot_data: &BootData,
    ota_info: &OtaInfo,
    key_avb: &RsaSigningKey,
@@ -422,8 +422,8 @@ fn create_boot_image(
 fn create_dm_verity_image(
    file: &mut PSeekFile,
    name: &str,
-    avb: &Avb,
-    dm_verity_data: &DmVerityData,
+    avb: Avb,
+    dm_verity_data: DmVerityData,
    ota_info: &OtaInfo,
    key_avb: &RsaSigningKey,
    cert_ota: &Certificate,
@@ -451,7 +451,7 @@ fn create_dm_verity_image(
 fn create_vbmeta_image(
    file: &mut PSeekFile,
    name: &str,
-    avb: &Avb,
+    avb: Avb,
    vbmeta_data: &VbmetaData,
    inputs: &BTreeMap<String, PSeekFile>,
    key: &RsaSigningKey,
@@ -537,7 +537,7 @@ fn create_partition_images(
                create_boot_image(
                    &mut file,
                    name,
-                    &partition.avb,
+                    partition.avb,
                    data,
                    ota_info,
                    key_avb,
@@ -550,8 +550,8 @@ fn create_partition_images(
                create_dm_verity_image(
                    &mut file,
                    name,
-                    &partition.avb,
-                    data,
+                    partition.avb,
+                    *data,
                    ota_info,
                    key_avb,
                    cert_ota,
@@ -560,7 +560,7 @@ fn create_partition_images(
                .with_context(|| format!("Failed to create dm-verity image: {name}"))?;
            }
            Data::Vbmeta(data) => {
-                create_vbmeta_image(&mut file, name, &partition.avb, data, &files, key_avb)
+                create_vbmeta_image(&mut file, name, partition.avb, data, &files, key_avb)
                    .with_context(|| format!("Failed to create vbmeta image: {name}"))?;
            }
        }
@@ -580,6 +580,8 @@ fn create_payload(
    key_ota: &RsaSigningKey,
    cancel_signal: &AtomicBool,
 ) -> Result<(String, u64)> {
+    const COMPRESSION_FACTOR: u32 = 64 * 1024;
+
    let dynamic_partitions_names = partitions
        .iter()
        .filter(|(_, p)| matches!(&p.data, Data::DmVerity(_)))
@@ -594,17 +596,23 @@ fn create_payload(
            .map(PSeekFile::new)
            .with_context(|| format!("Failed to create temp file for: {name}"))?;

-        let vabc_algo = if dynamic_partitions_names.contains(name) {
-            profile.vabc_algo
+        let vabc_params = if dynamic_partitions_names.contains(name) {
+            profile.vabc.map(|v| VabcParams {
+                version: v.version,
+                algo: v.algo,
+                compression_factor: COMPRESSION_FACTOR,
+            })
        } else {
            None
        };

        let (partition_info, operations, cow_estimate) =
-            payload::compress_image(file, &writer, name, 4096, vabc_algo, cancel_signal)?;
+            payload::compress_image(file, &writer, name, 4096, vabc_params, cancel_signal)?;

        compressed.insert(name, writer);

+        let is_v3 = profile.vabc.is_some_and(|e| e.version == CowVersion::V3);
+
        payload_partitions.push(PartitionUpdate {
            partition_name: name.clone(),
            run_postinstall: None,
@@ -624,8 +632,8 @@ fn create_payload(
            fec_roots: None,
            version: None,
            merge_operations: vec![],
-            estimate_cow_size: cow_estimate,
-            estimate_op_count_max: None,
+            estimate_cow_size: cow_estimate.map(|e| e.size),
+            estimate_op_count_max: cow_estimate.and_then(|e| is_v3.then_some(e.num_ops)),
        });
    }

@@ -645,11 +653,16 @@ fn create_payload(
                    partition_names: dynamic_partitions_names,
                }],
                snapshot_enabled: Some(true),
-                vabc_enabled: Some(true),
-                vabc_compression_param: profile.vabc_algo.map(|a| a.to_string()),
-                cow_version: Some(2),
+                // Everything below is meant to be unset if VABC is not
+                // supported.
+                vabc_enabled: profile.vabc.map(|_| true),
+                vabc_compression_param: profile.vabc.map(|v| v.algo.to_string()),
+                cow_version: profile.vabc.map(|v| match v.version {
+                    CowVersion::V2 => 2,
+                    CowVersion::V3 => 3,
+                }),
                vabc_feature_set: None,
-                compression_factor: None,
+                compression_factor: profile.vabc.map(|_| COMPRESSION_FACTOR.into()),
            }),
            partial_update: None,
            apex_info: vec![],
@@ -1189,6 +1202,7 @@ fn test_subcommand(cli: &TestCli, cancel_signal: &AtomicBool) -> Result<()> {
        Some(_) => None,
        None => Some(TempDir::new().context("Failed to create temp directory")?),
    };
+    #[allow(clippy::option_if_let_else)]
    let work_dir = match &cli.config.work_dir {
        Some(w) => w.as_path(),
        None => work_temp_dir.as_ref().unwrap().path(),
@@ -1324,7 +1338,7 @@ fn helper_mode() -> Result<()> {
    let cli = HelperCli::parse();

    let private_key_path = {
-        let parent = cli.public_key.parent().unwrap_or(Path::new("."));
+        let parent = cli.public_key.parent().unwrap_or_else(|| Path::new("."));
        let name = cli
            .public_key
            .file_name()
@@ -1398,7 +1412,10 @@ fn main() -> Result<()> {
    if env::var_os(ENV_HELPER_MODE).is_some() {
        return helper_mode();
    }
-    env::set_var(ENV_HELPER_MODE, "true");
+    // SAFETY: No multithreading at this point.
+    unsafe {
+        env::set_var(ENV_HELPER_MODE, "true");
+    }

    // Set up a cancel signal so we can properly clean up any temporary files.
    let cancel_signal = Arc::new(AtomicBool::new(false));
@@ -1,9 +1,9 @@
-// SPDX-FileCopyrightText: 2023 Andrew Gunnerson
+// SPDX-FileCopyrightText: 2023-2025 Andrew Gunnerson
 // SPDX-License-Identifier: GPL-3.0-only

 use std::{
    collections::BTreeMap,
-    fmt,
+    fmt::{self, Write as _},
    fs::{self, File},
    io::{BufRead, BufReader},
    path::Path,
@@ -108,7 +108,7 @@ fn update_changelog_links(path: &Path, base_url: &str) -> Result<()> {
    }

    for (link_ref, link) in links {
-        result.push_str(&format!("{link_ref}: {link}\n"));
+        let _ = writeln!(result, "{link_ref}: {link}");
    }

    fs::write(path, result)?;
Author	SHA1	Message	Date
Andrew Gunnerson	19129ae927	Version 3.16.0 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 18:15:02 -04:00
Andrew Gunnerson	fb2aaed042	CHANGELOG.md: Add entry for PR #446 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 18:14:25 -04:00
Andrew Gunnerson	182d937d34	Update all dependencies This also fixes a number of disabled-by-default clippy warnings and updates the Rust edition to 2024. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 18:06:59 -04:00
Andrew Gunnerson	113bdec6dc	CHANGELOG.md: Add entry for PR #445 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 16:57:03 -04:00
Andrew Gunnerson	1f0d012ac0	e2e: Fix incorrect metadata when generating a non-CoW-V2 payload It was hardcoded to set the CoW version to v2 in the payload manifest. This commit also updates the pixel_v2 profile to disable VABC so that scenario gets tested. The pixel_v3 profile now uses CoW v2 with gz. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 16:55:38 -04:00
Andrew Gunnerson	e4fbe00ea2	CHANGELOG.md: Add entry for PR #444 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-04 16:29:40 -04:00
Andrew Gunnerson	ad0b3d5aa8	payload: Add support for custom CoW compression levels No known device uses this functionality, but AOSP supports it, so we should too. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-03 23:34:25 -04:00
Andrew Gunnerson	f4394b0c21	CHANGELOG.md: Add entry for PR #443 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-03 23:03:44 -04:00
Andrew Gunnerson	9dd98fd715	payload: Add support for uncompressed CoW and fix more estimation bugs No known device uses this, but it's very useful for testing that our overhead calculation is identical to AOSP's. A few more bugs were found in our overhead calculation logic: * The CowHeaderV3 size was missing the original CowHeader (v2) fields that are supposed to be included due to inheritance in the C++ class. * The additional 1% overhead was incorrectly calculated against the initial CoW estimate before static overhead for CoW headers was added. * The V3 num_ops estimation did not set a minimum of 25 to match delta_generator. * The V2 size estimation did not take into account that a cluster of CoW operations cannot be truncated. It must be a multiple of cluster_ops (200 for avbroot). With these fixes, the CoW estimation when compression is disabled matches AOSP exactly. This means all the overhead calculation is now correct and the only difference when compression is enabled is in the compression ratios of the various lz4/gz implementations. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-03 22:59:15 -04:00
Andrew Gunnerson	b6e3c68241	CHANGELOG.md: Add entry for PR #442 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-03 20:39:57 -04:00
Andrew Gunnerson	6fee5346bb	payload: Add support for CoW version 3 AOSP has long supported CoW version 3, but it wasn't used by the stock OS on any Pixel devices until the new Pixel 9a. CoW version 3 is fundamentally similar to version 2, though with differences in the main header and how the operation headers are stored. The compression is no longer done in fixed-size chunks equal to the block size. Instead, the payload specifies a "compression factor", which is the maximum chunk size to pass to the compressor. The actual chunk size is the largest power of 2 <= the compression factor and the remaining input size. Additionally, for version 3, the payload stores an additional estimate_op_count_max field containing the number of CoW operations. While working on support for version 3, a few bugs in the version 2 estimation logic were found and fixed: * The cluster_ops * sizeof(CowOperationV2) overhead incorrectly assumed that cluster_ops was a constant 200 instead of the actual number of CoW operations. * The overhead did not account for kCowLabelOp headers, which delta_generator emits once for every InstallOperation in the payload. * The overhead did not account for kCowClusterOp headers, which batch CoW operations into groups of 200. * The overhead did not account for the CowFooter. The version 3 overhead is much simpler and easier to calculate compared to version 2. Fixes: #441 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-05-03 20:28:50 -04:00
Ivan	7d786150cf	README.ru.md: update translation * https://github.com/chenxiaolong/avbroot/commit/de433a27245adb26e578bdeb34a404552c79ea85 * https://github.com/chenxiaolong/avbroot/commit/5d66774d131505928b31e1968da0498732ac28ab * https://github.com/chenxiaolong/avbroot/commit/b922f0d23ce841ac0bd14bfd33da46273f183f68	2025-04-21 19:51:43 +03:00
Andrew Gunnerson	b922f0d23c	README.md: Deemphasize rooting in project description While avbroot initially started as a way to allow a rooted boot image to be used with a locked bootloader, it has evolved much since then. Nowadays, many folks use it to make modifications to their OTAs that don't involve enabling root access. avbroot also has many subcommands for packing and unpacking various Android image formats that people use without ever using avbroot's main OTA patching functionality. This commit updates the project description to reflect this and simplifies the wording a bit. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-18 20:51:54 -04:00
Andrew Gunnerson	1f2b2170b3	CHANGELOG.md: Fix incorrect link to VABC compression algorithm documentation Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-11 21:13:17 -04:00
Andrew Gunnerson	da124e4e05	Version 3.15.0 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-06 00:09:21 -04:00
Andrew Gunnerson	7b778515d1	CHANGELOG.md: Add entry for PR #439 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-06 00:02:48 -04:00
Andrew Gunnerson	98745afe33	Update dependencies Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-05 23:54:20 -04:00
Andrew Gunnerson	a47c501211	CHANGELOG.md: Add entry for PR #438 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-05 23:53:50 -04:00
Andrew Gunnerson	f2e47a65d4	Switch back to ring This reverts commit `e929ecbe44`. Ring is back to being maintained again, so let's switch back to it since it has fewer build dependencies and is much faster to compile. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-05 23:52:53 -04:00
Andrew Gunnerson	e11ddd2ba7	CHANGELOG.md: Add entry for PR #437 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-05 23:38:51 -04:00
Andrew Gunnerson	5d66774d13	Add support for changing VABC CoW compression algorithm Devices that launch with Android <14 generally use gzip as the CoW compression algorithm. This never changes because future full OTAs always need to be installable from the version of Android the device launched with. However, for users that don't care about the upgrade path from old versions of Android, a new --vabc-algo option can be used to switch from gz to lz4 compression. This can cut down the OTA installation time by more than 2/3rds when installing via a custom OTA updater app. On my Pixel Tablet, the installation time for the update_engine DOWNLOADING phase decreased from 32:05 to 9:41. Note that this has absolutely no effect on the performance when sideloading from recovery mode because that does not use CoW. When this new option is used, all dynamic partitions need to be extracted from the OTA during patching so that the CoW estimates can be recomputed. This will slow down the patching process and use up more temporary disk space. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-04-05 23:06:13 -04:00
Andrew Gunnerson	3f09a506a0	Version 3.14.0 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-23 19:27:05 -04:00
Andrew Gunnerson	4664f8ea37	CHANGELOG.md: Add entry for PR #435 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-23 19:26:29 -04:00
Andrew Gunnerson	796e2a4fa2	Update dependencies Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-23 19:17:15 -04:00
Andrew Gunnerson	e6b60d5d0f	CHANGELOG.md: Add entry for PR #434 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-23 19:02:39 -04:00
Andrew Gunnerson	e8cb4a8d53	Fix incorrect compression input for gzip CoW size estimation Instead of compressing the 64 MiB input in 2 MiB chunks, each loop iteration was compressing the full 64 MiB. This massively slowed down the patching process from seconds to potentially hours and would temporarily waste a bunch of space during OTA installation. Fixes: #433 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-23 18:50:44 -04:00
Andrew Gunnerson	b2d280eb20	CHANGELOG.md: Add entry for PR #430 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 20:23:48 -04:00
Andrew Gunnerson	e2dc5174b4	format/ota: Decouple OTA signature parsing from verification This way, we can fail hard for parsing errors, but not for verification errors in `avbroot ota verify`. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 20:22:16 -04:00
Andrew Gunnerson	80c47e9a02	CHANGELOG.md: Add entry for PR #429 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 19:49:54 -04:00
Andrew Gunnerson	65ba3ad5cc	Fix clippy 1.85 warnings Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 19:48:32 -04:00
Andrew Gunnerson	a7438876ce	CHANGELOG.md: Add entry for PR #428 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 19:38:16 -04:00
Andrew Gunnerson	c039901a85	cli/ota: Improve `ota verify` otacerts.zip handling * Try to avoid fail-fast behavior to show as many errors as possible. Parsing errors always fail immediately, but verification errors don't. * Move the recovery otacerts.zip check to the end to let more important checks run first. * Improve error message when otacerts.zip does not contain the signing certificate for the OTA to make it clear the issue is not that the zip contains no certificates at all. * Always run the recovery otacerts.zip check, but just log the error as a warning when running with --skip-recovery-ota-cert. * Fix unformatted error context string when parsing a boot image's otacerts.zip file fails. Discussion: #426 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-13 19:34:42 -04:00
Andrew Gunnerson	0cdc7172dd	Version 3.13.0 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-12 18:39:59 -04:00
Andrew Gunnerson	1296275418	CHANGELOG.md: Add entry for PR #427 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-12 18:38:43 -04:00
Andrew Gunnerson	a62d0a5c91	Update dependencies Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-12 18:25:35 -04:00
Andrew Gunnerson	de433a2724	README.md: Clarify what --skip-{system,recovery}-ota-cert affects Discussion: #426 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-03-12 18:07:17 -04:00
Andrew Gunnerson	c08af33343	CHANGELOG.md: Add entry for PR #425 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-24 21:17:44 -05:00
Andrew Gunnerson	e224884591	format/lp: Relax validation checks to parse on-device super partition There are two documented behaviors in liblp that are violated with the on-disk super partition layout after a virtual A/B CoW merge: 1. The partition name includes a `-` due to the `-cow` suffix. This is not meant to be a valid character. 2. The extent list is likely to have many gaps and not be sorted. The format documentation says that gaps are not allowed. This commit updates avbroot's LP parser to be less strict so that it can load real on-device super partitions. The extent allocator for the `pack` subcommand remains unchanged though, so avbroot will always produce LP images with sorted, gapless extents. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-24 21:05:28 -05:00
Andrew Gunnerson	a525fc4550	CHANGELOG.md: Add entry for PR #424 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 21:34:30 -05:00
Andrew Gunnerson	8c6b0fdfc5	cli/ota: Discard unmodified system image temp file earlier This is the same optimization as is currently done for boot images. There's no reason to keep the temp file around for the entire patching process if it's unmodified and we're not going to be copying its data into the payload. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 21:23:10 -05:00
Andrew Gunnerson	36d4ed19ad	CHANGELOG.md: Add entry for PR #423 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 21:18:12 -05:00
Andrew Gunnerson	f9b06b33e6	cli/ota: Fix incorrect unprotected partition warning with --skip-system-ota-cert The filtering out of partitions was done at the wrong scope, causing avbroot to warn that extracted-but-unmodified partitions were not protected by AVB. We never encountered this before because the system image was always patched and unmodified boot images got filtered out at an earlier phase during patching. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 21:00:59 -05:00
Andrew Gunnerson	620c873be5	CHANGELOG.md: Add entry for PR #422 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 21:00:20 -05:00
Andrew Gunnerson	e929ecbe44	Replace ring with aws-lc The author of ring recently announced that the library is no longer being maintained and fixes for security issues may be significantly delayed. Big thanks to Brian Smith for creating and maintaining the library for so long! This commit replaces ring with aws-lc, a cryptography library maintained by Amazon AWS. It seems to be well-regarded and is used by high-profile projects like rustls. It is also API-compatible with ring, so it is effectively a drop-in replacement. Unfortunately, we still cannot switch back to the RustCrypto SHA1 and SHA2 implementations because they are still significantly slower than ring and aws-lc on systems that do not support the SHA-NI extensions. https://rustsec.org/advisories/RUSTSEC-2025-0007 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 20:44:19 -05:00
Andrew Gunnerson	31685713ef	CHANGELOG.md: Add entry for PR #421 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 19:33:58 -05:00
Andrew Gunnerson	0dab7aa32c	Switch to stable bzip2-rs release and use zlib-rs flate2 backend * There is now a stable release of bzip2-rs with the fix for both the C and Rust versions of bzip2 being compiled. * The zlib-rs deflate implementation is faster than the default miniz_oxide. Changing this requires updating the checksums in the e2e tests due to slight differences in compression levels between the two implementations. * Temporarily silence RUSTSEC-2025-0007 to avoid blocking CI. The ring library is no longer maintained. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-21 19:27:18 -05:00
Ivan	ba6c1b1400	README.ru.md: update translation * https://github.com/chenxiaolong/avbroot/commit/c198646c7c9b3eb8013dca43df6df8502e619e3c * https://github.com/chenxiaolong/avbroot/commit/d4eb231dd49e7e6c7320135785cde7cd57634a50 * https://github.com/chenxiaolong/avbroot/commit/1484cd47c354197e1179a554f9f001020297aac8 * https://github.com/chenxiaolong/avbroot/commit/84fa6c6bc63b62d1b3b96dcb50b18b13dea0076f Signed-off-by: Ivan <reddxae@proton.me>	2025-02-11 14:20:11 +03:00
Andrew Gunnerson	1ecbf1144d	CHANGELOG.md: Add entry for PR #418 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-09 21:59:13 -05:00
Andrew Gunnerson	84fa6c6bc6	Add option to skip replacing OTA cert in system image This is analogous to the existing --skip-recovery-ota-cert option, except for the system image. Discussion: #417 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-09 21:42:18 -05:00
Andrew Gunnerson	15b7db4631	CHANGELOG.md: Add entry for PR #415 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-04 00:16:38 -05:00
Andrew Gunnerson	1d9c1574da	lp: Stop checking for zeroed first block AOSP says that for non-empty images, the first 4 KiB block is supposed to be filled with zeros to prevent it from being interpreted as an old BIOS boot sector. The previous implementation relied on that to distinguish between empty and non-empty images. However, Samsung decided to use this region for their own SignerVer02 structure, so the heuristic doesn't work. AOSP's liblp tries to parse the input file as an empty image before falling back to parsing as a normal image. We'll do the same. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-02-04 00:13:27 -05:00
Andrew Gunnerson	43db728b9d	Version 3.12.0 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 18:20:06 -05:00
Andrew Gunnerson	fbe9f629ab	CHANGELOG.md: Add entry for PR #411 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 18:11:40 -05:00
Andrew Gunnerson	80d5f19223	Update dependencies Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 18:10:42 -05:00
Andrew Gunnerson	cff5ac6b2e	CHANGELOG.md: Add entry for PR #410 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 18:04:35 -05:00
Andrew Gunnerson	2e3b5db9fe	cli/key: Rename extract-avb to encode-avb For consistency with decode-avb. The old syntax will remain supported indefinitely for backwards compatibility. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 18:01:24 -05:00
Andrew Gunnerson	58a279f3a4	CHANGELOG.md: Add entry for PR #409 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 17:53:08 -05:00
Andrew Gunnerson	1484cd47c3	cli/ota: Add support for extracting OTA certificate and AVB public key This also adds a new --none option so that these two components can be extracted without extracting any partition images. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 17:50:09 -05:00
Andrew Gunnerson	2ee38b716b	CHANGELOG.md: Add entry for PR #408 Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 16:48:08 -05:00
Andrew Gunnerson	d4eb231dd4	cli/ota: Add support for extracting specific images Previously, we only supported extracting all images or the subset of images that could potentially be patched by avbroot. This was unnecessarily slow if the user only needed to extract a specific image. This commit also deprecates and hides the `--boot-only` option, though the functionality will remain indefinitely. Signed-off-by: Andrew Gunnerson <accounts+github@chiller3.com>	2025-01-26 16:43:35 -05:00