From b29d1ae372bb0794cc8425ced7986b3d059a2be5 Mon Sep 17 00:00:00 2001 From: Florian Klink Date: Mon, 27 Feb 2023 13:03:53 +0100 Subject: feat(tvix/store): add import::import_path This imports the contents at a given Path into the tvix store. It doesn't register the contents at a Path in the store itself, that's up to the PathInfoService. Change-Id: I2c493532d65b90f199ddb7dfc90249f5c2957dee Reviewed-on: https://cl.tvl.fyi/c/depot/+/8159 Reviewed-by: raitobezarius Tested-by: BuildkiteCI --- tvix/store/src/import.rs | 248 +++++++++++++++++++++++++++++++++++++++++++++++ 1 file changed, 248 insertions(+) create mode 100644 tvix/store/src/import.rs (limited to 'tvix/store/src/import.rs') diff --git a/tvix/store/src/import.rs b/tvix/store/src/import.rs new file mode 100644 index 000000000000..d2abb292fcce --- /dev/null +++ b/tvix/store/src/import.rs @@ -0,0 +1,248 @@ +use crate::{proto, BlobWriter}; +use std::{ + collections::HashMap, + fmt::Debug, + fs, + fs::File, + io::BufReader, + os::unix::prelude::PermissionsExt, + path::{Path, PathBuf}, +}; +use tracing::instrument; +use walkdir::WalkDir; + +use crate::{ + blobservice::BlobService, chunkservice::ChunkService, directoryservice::DirectoryService, +}; + +#[derive(Debug, thiserror::Error)] +pub enum Error { + #[error("failed to upload directory at {0}: {1}")] + UploadDirectoryError(PathBuf, crate::Error), + + #[error("invalid encoding encountered for entry {0:?}")] + InvalidEncoding(PathBuf), + + #[error("unable to stat {0}: {1}")] + UnableToStat(PathBuf, std::io::Error), + + #[error("unable to open {0}: {1}")] + UnableToOpen(PathBuf, std::io::Error), + + #[error("unable to read {0}: {1}")] + UnableToRead(PathBuf, std::io::Error), +} + +impl From for Error { + fn from(value: super::Error) -> Self { + match value { + crate::Error::InvalidRequest(_) => panic!("tvix bug"), + crate::Error::StorageError(_) => panic!("error"), + } + } +} + +// This processes a given [walkdir::DirEntry] and returns a +// proto::node::Node, depending on the type of the entry. +// +// If the entry is a file, its contents are uploaded. +// If the entry is a directory, the Directory is uploaded as well. +// For this to work, it relies on the caller to provide the directory object +// with the previously returned (child) nodes. +// +// It assumes entries to be returned in "contents first" order, means this +// will only be called with a directory if all children of it have been +// visited. If the entry is indeed a directory, it'll also upload that +// directory to the store. For this, the so-far-assembled Directory object for +// this path needs to be passed in. +// +// It assumes the caller adds returned nodes to the directories it assembles. +#[instrument(skip_all, fields(entry.file_type=?&entry.file_type(),entry.path=?entry.path()))] +fn process_entry( + blob_service: &mut BS, + chunk_service: &mut CS, + directory_service: &mut DS, + entry: &walkdir::DirEntry, + maybe_directory: Option, +) -> Result { + let file_type = entry.file_type(); + + let entry_path: PathBuf = entry.path().to_path_buf(); + + if file_type.is_dir() { + let directory = maybe_directory + .expect("tvix bug: must be called with some directory in the case of directory"); + let directory_digest = directory.digest(); + let directory_size = directory.size(); + + // upload this directory + directory_service + .put(directory) + .map_err(|e| Error::UploadDirectoryError(entry.path().to_path_buf(), e))?; + + return Ok(proto::node::Node::Directory(proto::DirectoryNode { + name: entry + .file_name() + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(entry.path().to_path_buf())))?, + digest: directory_digest, + size: directory_size, + })); + } + + if file_type.is_symlink() { + let target = std::fs::read_link(&entry_path) + .map_err(|e| Error::UnableToStat(entry_path.clone(), e))?; + + return Ok(proto::node::Node::Symlink(proto::SymlinkNode { + name: entry + .file_name() + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(entry.path().to_path_buf())))?, + target: target + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(entry.path().to_path_buf())))?, + })); + } + + if file_type.is_file() { + let metadata = entry + .metadata() + .map_err(|e| Error::UnableToStat(entry_path.clone(), e.into()))?; + + // hash the file contents, upload chunks if not there yet + let (blob_digest, blob_meta) = { + let mut blob_writer = BlobWriter::new(chunk_service); + + let file = File::open(entry_path.clone()) + .map_err(|e| Error::UnableToOpen(entry_path.clone(), e))?; + + let mut file_reader = BufReader::new(file); + + std::io::copy(&mut file_reader, &mut blob_writer) + .map_err(|e| Error::UnableToRead(entry_path, e))?; + + // TODO: handle errors + blob_writer.finalize().unwrap() + }; + + // upload blobmeta if not there yet + if blob_service + .stat(&proto::StatBlobRequest { + digest: blob_digest.to_vec(), + include_chunks: false, + include_bao: false, + })? + .is_none() + { + // upload blobmeta + blob_service.put(&blob_digest, blob_meta)?; + } + + return Ok(proto::node::Node::File(proto::FileNode { + name: entry + .file_name() + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(entry.path().to_path_buf())))?, + digest: blob_digest, + size: metadata.len() as u32, + // If it's executable by the user, it'll become executable. + // This matches nix's dump() function behaviour. + executable: metadata.permissions().mode() & 64 != 0, + })); + } + todo!("handle other types") +} + +/// Imports the contents at a given Path into the tvix store. +/// +/// It doesn't register the contents at a Path in the store itself, that's up +/// to the PathInfoService. +// +// returns the root node, or an error. +#[instrument(skip(blob_service, chunk_service, directory_service), fields(path=?p))] +pub fn import_path< + BS: BlobService, + CS: ChunkService, + DS: DirectoryService, + P: AsRef + Debug, +>( + blob_service: &mut BS, + chunk_service: &mut CS, + directory_service: &mut DS, + p: P, +) -> Result { + // Probe if the path points to a symlink. If it does, we process it manually, + // due to https://github.com/BurntSushi/walkdir/issues/175. + let symlink_metadata = fs::symlink_metadata(p.as_ref()) + .map_err(|e| Error::UnableToStat(p.as_ref().to_path_buf(), e))?; + if symlink_metadata.is_symlink() { + let target = std::fs::read_link(p.as_ref()) + .map_err(|e| Error::UnableToStat(p.as_ref().to_path_buf(), e))?; + return Ok(proto::node::Node::Symlink(proto::SymlinkNode { + name: p + .as_ref() + .file_name() + .unwrap_or_default() + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(p.as_ref().to_path_buf())))?, + target: target + .to_str() + .map(|s| Ok(s.to_owned())) + .unwrap_or(Err(Error::InvalidEncoding(p.as_ref().to_path_buf())))?, + })); + } + + let mut directories: HashMap = HashMap::default(); + + // TODO: make sure we traverse in sorted order, or insert to parent_directory in sorted order at least. + for entry in WalkDir::new(p).follow_links(false).contents_first(true) { + let entry = entry.unwrap(); + + // process_entry wants an Option in case the entry points to a directory. + // make sure to provide it. + let maybe_directory: Option = { + if entry.file_type().is_dir() { + Some( + directories + .entry(entry.path().to_path_buf()) + .or_default() + .clone(), + ) + } else { + None + } + }; + + let node = process_entry( + blob_service, + chunk_service, + directory_service, + &entry, + maybe_directory, + )?; + + if entry.depth() == 0 { + return Ok(node); + } else { + // calculate the parent path, and make sure we register the node there. + // NOTE: entry.depth() > 0 + let parent_path = entry.path().parent().unwrap().to_path_buf(); + + // record node in parent directory, creating a new [proto:Directory] if not there yet. + let parent_directory = directories.entry(parent_path).or_default(); + match node { + proto::node::Node::Directory(e) => parent_directory.directories.push(e), + proto::node::Node::File(e) => parent_directory.files.push(e), + proto::node::Node::Symlink(e) => parent_directory.symlinks.push(e), + } + } + } + // unreachable, we already bailed out before if root doesn't exist. + panic!("tvix bug") +} -- cgit 1.4.1