[RFC PATCH v3 00/22] Rust PuzzleFS filesystem driver

From: Ariel Miculas
Date: Thu May 16 2024 - 15:07:18 EST


Hello everyone,

This series is the third version of the proof-of-concept PuzzleFS
filesystem driver, an open-source next-generation container filesystem
[1], designed to address the limitation of the existing OCI format. It
supports direct mounting of container filesystems without an extraction
step, and it uses content defined chunking (CDC) in order to often
achieve substantial disk space savings. It is similar to casync [9] in
this regard. For more details, I had a presentation last year at Open
Source Summit Europe [2]. There's also a dedicated PuzzleFS page on the
rust-for-linux website [3] and an LWN article about PuzzleFS [4].

This patch series is based on the latest filesystem abstraction patches
[5] sent to LKML a few days ago. I've been keeping PuzzleFS up-to-date
with the multiple versions of the filesystem abstractions, even though I
haven't sent patches after each change. However, Cisco is still
interested in upstreaming PuzzleFS and I would like to gather another
round of feedback.

I have described my kernel setup and how to get started with the
PuzzleFS kernel driver in this blog post [7]. Below I've included a
short demo of the PuzzleFS kernel driver.

To mount a PuzzleFS driver, we must pass the `oci_root_dir` and
`image_manifest` as parameters.
```
~ # mount -t puzzlefs -o oci_root_dir="/home/puzzlefs_xattr" -o \
image_manifest="ed63ace21eccceabab08d89afb75e94dae47973f82a17a172396a19ea953c8ab" \
none /mnt
[ 6789.110219] PuzzleFS: opened puzzlefs [/home/puzzlefs_xattr]:[ed63ace21eccceabab08d89afb75e94dae47973f82a17a172396a19ea953c8ab]

~ # cd /mnt/
/mnt # ls -la
total 1
drwxr-xr-x 2 1000 1000 0 Jan 1 1970 test
drwxr-xr-x 2 1000 1000 0 Jan 1 1970 test-2
-rw-r--r-- 1 1000 1000 20 Jan 1 1970 test.txt

/mnt # cat test.txt
The quick brown fox
```

In this example, `/home/puzzlefs_xattr` is the puzzlefs oci directory,
obtained by building a PuzzleFS image via the `puzzlefs build` command.
Here's the disk layout of the image (for more details about the image
format, consult the README in the github repository [1]):
```
/mnt # ls -lR /home/puzzlefs_xattr/
/home/puzzlefs_xattr/:
total 8
drwxr-xr-x 3 1000 1000 0 May 15 20:24 blobs
-rw-r--r-- 1 1000 1000 263 May 15 20:24 index.json
-rw-r--r-- 1 1000 1000 37 May 15 20:24 oci-layout

/home/puzzlefs_xattr/blobs:
total 0
drwxr-xr-x 2 1000 1000 0 May 15 20:24 sha256

/home/puzzlefs_xattr/blobs/sha256:
total 12
-rw------- 1 1000 1000 20 May 15 20:24 35fb7cc2337d10d618a1bad35c7a9e957c213f00d0ed32f2454b2a99a9718
-rw------- 1 1000 1000 752 May 15 20:24 d1cbb5b11a9044ecb3889397e0df8c594484cc630294cf208e57ad9dcf6ca
-rw------- 1 1000 1000 272 May 15 20:24 ed63ace21eccceabab08d89afb75e94dae47973f82a17a172396a19ea953b
```

To get the PuzzleFS image manifest, we can look into the `index.json`
file. The following digest, corresponding to the `xattr` tag, must be
passed as a parameter when mounting the filesystem:
ed63ace21eccceabab08d89afb75e94dae47973f82a17a172396a19ea953c8ab

In userspace we can pass the `xattr` tag directly, but we don't want to
parse the json format in the kernel.
```
$ cat index.json | jq
{
"schemaVersion": -1,
"manifests": [
{
"digest": "sha256:ed63ace21eccceabab08d89afb75e94dae47973f82a17a172396a19ea953c8ab",
"size": 272,
"media_type": "application/vnd.puzzlefs.image.rootfs.v1",
"annotations": {
"org.opencontainers.image.ref.name": "xattr"
}
}
],
"annotations": {}
}
```

For testing extended attributes, I've written a small cli program called
`xattr-cli` which is available here [8]. No reason to use this over
`getfattr`, except it is not available in a busybox environment.

```
/mnt # /home/xattr-cli test
Extended attributes of test:
user.comment -> this is a dir
user.groceries -> want to buy a file
/mnt # /home/xattr-cli test-2/
Extended attributes of test-2/:
user.whatever -> want to buy a file
```

A git tree for this patch series is available here:
git://github.com/ariel-miculas/linux.git puzzlefs_rfc_v3

Web:
https://github.com/ariel-miculas/linux/tree/puzzlefs_rfc_v3

[1] https://github.com/project-machine/puzzlefs
[2] https://osseu2023.sched.com/event/1OGjk/puzzlefs-the-next-generation-container-filesystem-armand-ariel-miculas-cisco-systems
[3] https://rust-for-linux.com/puzzlefs-filesystem-driver
[4] https://lwn.net/Articles/945320/
[5] https://lore.kernel.org/rust-for-linux/20240514131711.379322-1-wedsonaf@xxxxxxxxx/
[6] https://lore.kernel.org/rust-for-linux/20230726164535.230515-1-amiculas@xxxxxxxxx/
[7] https://machinerun.io/puzzlefs/2023/10/30/Linux-kernel-setup.html
[8] https://github.com/ariel-miculas/xattr-cli
[9] https://0pointer.net/blog/casync-a-tool-for-distributing-file-system-images.html

---

Main changes since v2 [6]:
* PuzzleFS now uses the latest filesystem abstractions [5]
* added extended attributes support

---

Alice Ryhl (1):
rust: add improved version of `ForeignOwnable::borrow_mut`

Ariel Miculas (19):
kernel: configs: add qemu-busybox-min.config
rust: hex: import crate
rust: hex: add SPDX license identifiers
rust: Kbuild: enable `hex`
rust: hex: add encode_hex_iter and encode_hex_upper_iter methods
rust: capnp: import crate
rust: capnp: add SPDX License Identifiers
rust: capnp: return an error when trying to display floating point
values
rust: Kbuild: enable `capnp`
rust: kernel: add an abstraction over vfsmount to allow cloning a new
private mount
rust: file: Add support for reading files using their path
fs: puzzlefs: Implement the initial version of PuzzleFS
rust: kernel: add from_iter_fallible for Vec<T>
kernel: configs: add puzzlefs config fragment
scripts: add fs directory to rust-analyzer
fs: puzzlefs: add extended attributes support
Add borrow_mut implementation to a ForeignOwnable CString
fs: puzzlefs: add oci_root_dir and image_manifest mount parameters
fs: puzzlefs: implement statfs for puzzlefs

Wedson Almeida Filho (2):
rust: file: add bindings for `struct file`
rust: add support for file system parameters

arch/x86/configs/qemu-busybox-min.config | 11 +
fs/Kconfig | 1 +
fs/Makefile | 1 +
fs/puzzlefs/Kconfig | 14 +
fs/puzzlefs/Makefile | 8 +
fs/puzzlefs/puzzle.rs | 5 +
fs/puzzlefs/puzzle/error.rs | 105 +
fs/puzzlefs/puzzle/inode.rs | 122 +
fs/puzzlefs/puzzle/oci.rs | 70 +
fs/puzzlefs/puzzle/types.rs | 377 +
fs/puzzlefs/puzzle/types/manifest.capnp | 15 +
fs/puzzlefs/puzzle/types/manifest_capnp.rs | 757 +
fs/puzzlefs/puzzle/types/metadata.capnp | 65 +
fs/puzzlefs/puzzle/types/metadata_capnp.rs | 4049 +++++
fs/puzzlefs/puzzlefs.rs | 330 +
kernel/configs/puzzlefs.config | 4 +
kernel/configs/qemu-busybox-min.config | 56 +
rust/Makefile | 41 +-
rust/bindings/bindings_helper.h | 3 +
rust/capnp/any_pointer.rs | 315 +
rust/capnp/any_pointer_list.rs | 210 +
rust/capnp/capability.rs | 365 +
rust/capnp/capability_list.rs | 299 +
rust/capnp/constant.rs | 56 +
rust/capnp/data.rs | 97 +
rust/capnp/data_list.rs | 220 +
rust/capnp/dynamic_list.rs | 410 +
rust/capnp/dynamic_struct.rs | 784 +
rust/capnp/dynamic_value.rs | 319 +
rust/capnp/enum_list.rs | 239 +
rust/capnp/introspect.rs | 284 +
rust/capnp/io.rs | 204 +
rust/capnp/lib.rs | 653 +
rust/capnp/list_list.rs | 298 +
rust/capnp/message.rs | 880 +
rust/capnp/primitive_list.rs | 281 +
rust/capnp/private/arena.rs | 419 +
rust/capnp/private/capability.rs | 157 +
rust/capnp/private/layout.rs | 4212 +++++
rust/capnp/private/layout_test.rs | 194 +
rust/capnp/private/mask.rs | 64 +
rust/capnp/private/mod.rs | 38 +
rust/capnp/private/primitive.rs | 121 +
rust/capnp/private/read_limiter.rs | 115 +
rust/capnp/private/units.rs | 70 +
rust/capnp/private/zero.rs | 48 +
rust/capnp/raw.rs | 71 +
rust/capnp/schema.rs | 432 +
rust/capnp/schema_capnp.rs | 14450 ++++++++++++++++
rust/capnp/serialize.rs | 963 +
.../serialize/no_alloc_buffer_segments.rs | 629 +
rust/capnp/serialize_packed.rs | 620 +
rust/capnp/stringify.rs | 187 +
rust/capnp/struct_list.rs | 302 +
rust/capnp/text.rs | 296 +
rust/capnp/text_list.rs | 218 +
rust/capnp/traits.rs | 235 +
rust/helpers.c | 16 +
rust/hex/error.rs | 61 +
rust/hex/lib.rs | 495 +
rust/hex/serde.rs | 104 +
rust/kernel/alloc/vec_ext.rs | 12 +
rust/kernel/error.rs | 4 +-
rust/kernel/file.rs | 298 +
rust/kernel/fs.rs | 212 +-
rust/kernel/fs/inode.rs | 66 +-
rust/kernel/fs/param.rs | 576 +
rust/kernel/lib.rs | 9 +
rust/kernel/mount.rs | 73 +
rust/kernel/str.rs | 24 +
rust/kernel/sync/arc.rs | 25 +-
rust/kernel/types.rs | 93 +-
samples/rust/rust_rofs.rs | 32 +-
scripts/Makefile.build | 2 +-
scripts/generate_rust_analyzer.py | 2 +-
75 files changed, 37843 insertions(+), 50 deletions(-)
create mode 100644 arch/x86/configs/qemu-busybox-min.config
create mode 100644 fs/puzzlefs/Kconfig
create mode 100644 fs/puzzlefs/Makefile
create mode 100644 fs/puzzlefs/puzzle.rs
create mode 100644 fs/puzzlefs/puzzle/error.rs
create mode 100644 fs/puzzlefs/puzzle/inode.rs
create mode 100644 fs/puzzlefs/puzzle/oci.rs
create mode 100644 fs/puzzlefs/puzzle/types.rs
create mode 100644 fs/puzzlefs/puzzle/types/manifest.capnp
create mode 100644 fs/puzzlefs/puzzle/types/manifest_capnp.rs
create mode 100644 fs/puzzlefs/puzzle/types/metadata.capnp
create mode 100644 fs/puzzlefs/puzzle/types/metadata_capnp.rs
create mode 100644 fs/puzzlefs/puzzlefs.rs
create mode 100644 kernel/configs/puzzlefs.config
create mode 100644 kernel/configs/qemu-busybox-min.config
create mode 100644 rust/capnp/any_pointer.rs
create mode 100644 rust/capnp/any_pointer_list.rs
create mode 100644 rust/capnp/capability.rs
create mode 100644 rust/capnp/capability_list.rs
create mode 100644 rust/capnp/constant.rs
create mode 100644 rust/capnp/data.rs
create mode 100644 rust/capnp/data_list.rs
create mode 100644 rust/capnp/dynamic_list.rs
create mode 100644 rust/capnp/dynamic_struct.rs
create mode 100644 rust/capnp/dynamic_value.rs
create mode 100644 rust/capnp/enum_list.rs
create mode 100644 rust/capnp/introspect.rs
create mode 100644 rust/capnp/io.rs
create mode 100644 rust/capnp/lib.rs
create mode 100644 rust/capnp/list_list.rs
create mode 100644 rust/capnp/message.rs
create mode 100644 rust/capnp/primitive_list.rs
create mode 100644 rust/capnp/private/arena.rs
create mode 100644 rust/capnp/private/capability.rs
create mode 100644 rust/capnp/private/layout.rs
create mode 100644 rust/capnp/private/layout_test.rs
create mode 100644 rust/capnp/private/mask.rs
create mode 100644 rust/capnp/private/mod.rs
create mode 100644 rust/capnp/private/primitive.rs
create mode 100644 rust/capnp/private/read_limiter.rs
create mode 100644 rust/capnp/private/units.rs
create mode 100644 rust/capnp/private/zero.rs
create mode 100644 rust/capnp/raw.rs
create mode 100644 rust/capnp/schema.rs
create mode 100644 rust/capnp/schema_capnp.rs
create mode 100644 rust/capnp/serialize.rs
create mode 100644 rust/capnp/serialize/no_alloc_buffer_segments.rs
create mode 100644 rust/capnp/serialize_packed.rs
create mode 100644 rust/capnp/stringify.rs
create mode 100644 rust/capnp/struct_list.rs
create mode 100644 rust/capnp/text.rs
create mode 100644 rust/capnp/text_list.rs
create mode 100644 rust/capnp/traits.rs
create mode 100644 rust/hex/error.rs
create mode 100644 rust/hex/lib.rs
create mode 100644 rust/hex/serde.rs
create mode 100644 rust/kernel/file.rs
create mode 100644 rust/kernel/fs/param.rs
create mode 100644 rust/kernel/mount.rs

--
2.34.1