revlog.rs
401 lines
| 12.0 KiB
| application/rls-services+xml
|
RustLexer
|
r46097 | use std::borrow::Cow; | ||
use std::io::Read; | ||||
use std::ops::Deref; | ||||
use std::path::Path; | ||||
use byteorder::{BigEndian, ByteOrder}; | ||||
use flate2::read::ZlibDecoder; | ||||
use micro_timer::timed; | ||||
|
r48171 | use sha1::{Digest, Sha1}; | ||
|
r46097 | use zstd; | ||
use super::index::Index; | ||||
|
r47160 | use super::node::{NodePrefix, NODE_BYTES_LENGTH, NULL_NODE}; | ||
|
r46706 | use super::nodemap; | ||
|
r47172 | use super::nodemap::{NodeMap, NodeMapError}; | ||
|
r46706 | use super::nodemap_docket::NodeMapDocket; | ||
|
r46097 | use super::patch; | ||
|
r47172 | use crate::errors::HgError; | ||
|
r46782 | use crate::repo::Repo; | ||
|
r46097 | use crate::revlog::Revision; | ||
|
r48776 | use crate::NULL_REVISION; | ||
|
r46097 | |||
|
r47172 | #[derive(derive_more::From)] | ||
|
r46097 | pub enum RevlogError { | ||
InvalidRevision, | ||||
|
r47577 | /// Working directory is not supported | ||
WDirUnsupported, | ||||
|
r46646 | /// Found more than one entry whose ID match the requested prefix | ||
AmbiguousPrefix, | ||||
|
r47172 | #[from] | ||
Other(HgError), | ||||
|
r46097 | } | ||
|
r47172 | impl From<NodeMapError> for RevlogError { | ||
fn from(error: NodeMapError) -> Self { | ||||
match error { | ||||
NodeMapError::MultipleResults => RevlogError::AmbiguousPrefix, | ||||
NodeMapError::RevisionNotInIndex(_) => RevlogError::corrupted(), | ||||
} | ||||
} | ||||
} | ||||
impl RevlogError { | ||||
fn corrupted() -> Self { | ||||
RevlogError::Other(HgError::corrupted("corrupted revlog")) | ||||
|
r47119 | } | ||
} | ||||
|
r46097 | /// Read only implementation of revlog. | ||
pub struct Revlog { | ||||
/// When index and data are not interleaved: bytes of the revlog index. | ||||
/// When index and data are interleaved: bytes of the revlog index and | ||||
/// data. | ||||
|
r47478 | pub(crate) index: Index, | ||
|
r46097 | /// When index and data are not interleaved: bytes of the revlog data | ||
data_bytes: Option<Box<dyn Deref<Target = [u8]> + Send>>, | ||||
|
r46706 | /// When present on disk: the persistent nodemap for this revlog | ||
nodemap: Option<nodemap::NodeTree>, | ||||
|
r46097 | } | ||
impl Revlog { | ||||
/// Open a revlog index file. | ||||
/// | ||||
/// It will also open the associated data file if index and data are not | ||||
/// interleaved. | ||||
#[timed] | ||||
|
r46408 | pub fn open( | ||
|
r46782 | repo: &Repo, | ||
index_path: impl AsRef<Path>, | ||||
|
r46408 | data_path: Option<&Path>, | ||
|
r48777 | ) -> Result<Self, HgError> { | ||
|
r46782 | let index_path = index_path.as_ref(); | ||
|
r47172 | let index_mmap = repo.store_vfs().mmap_open(&index_path)?; | ||
|
r46097 | |||
let version = get_version(&index_mmap); | ||||
if version != 1 { | ||||
|
r47172 | // A proper new version should have had a repo/store requirement. | ||
|
r48777 | return Err(HgError::corrupted("corrupted revlog")); | ||
|
r46097 | } | ||
|
r46176 | let index = Index::new(Box::new(index_mmap))?; | ||
|
r46097 | |||
|
r46408 | let default_data_path = index_path.with_extension("d"); | ||
|
r46097 | // type annotation required | ||
// won't recognize Mmap as Deref<Target = [u8]> | ||||
let data_bytes: Option<Box<dyn Deref<Target = [u8]> + Send>> = | ||||
|
r46175 | if index.is_inline() { | ||
|
r46097 | None | ||
} else { | ||||
|
r46408 | let data_path = data_path.unwrap_or(&default_data_path); | ||
|
r47172 | let data_mmap = repo.store_vfs().mmap_open(data_path)?; | ||
|
r46097 | Some(Box::new(data_mmap)) | ||
}; | ||||
|
r46782 | let nodemap = NodeMapDocket::read_from_file(repo, index_path)?.map( | ||
|
r46706 | |(docket, data)| { | ||
nodemap::NodeTree::load_bytes( | ||||
Box::new(data), | ||||
docket.data_length, | ||||
) | ||||
}, | ||||
); | ||||
Ok(Revlog { | ||||
index, | ||||
data_bytes, | ||||
nodemap, | ||||
}) | ||||
|
r46097 | } | ||
|
r46105 | /// Return number of entries of the `Revlog`. | ||
pub fn len(&self) -> usize { | ||||
|
r46175 | self.index.len() | ||
|
r46105 | } | ||
/// Returns `true` if the `Revlog` has zero `entries`. | ||||
pub fn is_empty(&self) -> bool { | ||||
|
r46175 | self.index.is_empty() | ||
|
r46105 | } | ||
/// Return the full data associated to a node. | ||||
#[timed] | ||||
|
r46647 | pub fn get_node_rev( | ||
&self, | ||||
|
r47160 | node: NodePrefix, | ||
|
r46647 | ) -> Result<Revision, RevlogError> { | ||
|
r48776 | if node.is_prefix_of(&NULL_NODE) { | ||
return Ok(NULL_REVISION); | ||||
} | ||||
|
r46706 | if let Some(nodemap) = &self.nodemap { | ||
return nodemap | ||||
|
r47172 | .find_bin(&self.index, node)? | ||
|
r46706 | .ok_or(RevlogError::InvalidRevision); | ||
} | ||||
// Fallback to linear scan when a persistent nodemap is not present. | ||||
// This happens when the persistent-nodemap experimental feature is not | ||||
// enabled, or for small revlogs. | ||||
// | ||||
// TODO: consider building a non-persistent nodemap in memory to | ||||
// optimize these cases. | ||||
|
r46646 | let mut found_by_prefix = None; | ||
|
r46105 | for rev in (0..self.len() as Revision).rev() { | ||
let index_entry = | ||||
|
r47172 | self.index.get_entry(rev).ok_or(HgError::corrupted( | ||
"revlog references a revision not in the index", | ||||
))?; | ||||
|
r46647 | if node == *index_entry.hash() { | ||
|
r46105 | return Ok(rev); | ||
} | ||||
|
r46647 | if node.is_prefix_of(index_entry.hash()) { | ||
|
r46646 | if found_by_prefix.is_some() { | ||
return Err(RevlogError::AmbiguousPrefix); | ||||
} | ||||
found_by_prefix = Some(rev) | ||||
} | ||||
|
r46105 | } | ||
|
r46646 | found_by_prefix.ok_or(RevlogError::InvalidRevision) | ||
|
r46105 | } | ||
|
r47162 | /// Returns whether the given revision exists in this revlog. | ||
pub fn has_rev(&self, rev: Revision) -> bool { | ||||
self.index.get_entry(rev).is_some() | ||||
} | ||||
|
r46097 | /// Return the full data associated to a revision. | ||
/// | ||||
/// All entries required to build the final data out of deltas will be | ||||
/// retrieved as needed, and the deltas will be applied to the inital | ||||
/// snapshot to rebuild the final data. | ||||
#[timed] | ||||
pub fn get_rev_data(&self, rev: Revision) -> Result<Vec<u8>, RevlogError> { | ||||
// Todo return -> Cow | ||||
let mut entry = self.get_entry(rev)?; | ||||
let mut delta_chain = vec![]; | ||||
while let Some(base_rev) = entry.base_rev { | ||||
delta_chain.push(entry); | ||||
|
r47172 | entry = self | ||
.get_entry(base_rev) | ||||
.map_err(|_| RevlogError::corrupted())?; | ||||
|
r46097 | } | ||
|
r46102 | // TODO do not look twice in the index | ||
|
r46175 | let index_entry = self | ||
.index | ||||
.get_entry(rev) | ||||
.ok_or(RevlogError::InvalidRevision)?; | ||||
|
r46102 | |||
let data: Vec<u8> = if delta_chain.is_empty() { | ||||
entry.data()?.into() | ||||
} else { | ||||
Revlog::build_data_from_deltas(entry, &delta_chain)? | ||||
}; | ||||
if self.check_hash( | ||||
index_entry.p1(), | ||||
index_entry.p2(), | ||||
|
r46647 | index_entry.hash().as_bytes(), | ||
|
r46102 | &data, | ||
) { | ||||
Ok(data) | ||||
|
r46097 | } else { | ||
|
r47172 | Err(RevlogError::corrupted()) | ||
|
r46097 | } | ||
} | ||||
|
r46102 | /// Check the hash of some given data against the recorded hash. | ||
pub fn check_hash( | ||||
&self, | ||||
p1: Revision, | ||||
p2: Revision, | ||||
expected: &[u8], | ||||
data: &[u8], | ||||
) -> bool { | ||||
|
r46175 | let e1 = self.index.get_entry(p1); | ||
|
r46102 | let h1 = match e1 { | ||
Some(ref entry) => entry.hash(), | ||||
|
r46647 | None => &NULL_NODE, | ||
|
r46102 | }; | ||
|
r46175 | let e2 = self.index.get_entry(p2); | ||
|
r46102 | let h2 = match e2 { | ||
Some(ref entry) => entry.hash(), | ||||
|
r46647 | None => &NULL_NODE, | ||
|
r46102 | }; | ||
|
r48171 | &hash(data, h1.as_bytes(), h2.as_bytes()) == expected | ||
|
r46102 | } | ||
|
r46097 | /// Build the full data of a revision out its snapshot | ||
/// and its deltas. | ||||
#[timed] | ||||
fn build_data_from_deltas( | ||||
snapshot: RevlogEntry, | ||||
deltas: &[RevlogEntry], | ||||
) -> Result<Vec<u8>, RevlogError> { | ||||
let snapshot = snapshot.data()?; | ||||
let deltas = deltas | ||||
.iter() | ||||
.rev() | ||||
.map(RevlogEntry::data) | ||||
.collect::<Result<Vec<Cow<'_, [u8]>>, RevlogError>>()?; | ||||
let patches: Vec<_> = | ||||
deltas.iter().map(|d| patch::PatchList::new(d)).collect(); | ||||
let patch = patch::fold_patch_lists(&patches); | ||||
Ok(patch.apply(&snapshot)) | ||||
} | ||||
/// Return the revlog data. | ||||
fn data(&self) -> &[u8] { | ||||
match self.data_bytes { | ||||
Some(ref data_bytes) => &data_bytes, | ||||
|
r46175 | None => panic!( | ||
"forgot to load the data or trying to access inline data" | ||||
), | ||||
|
r46097 | } | ||
} | ||||
/// Get an entry of the revlog. | ||||
fn get_entry(&self, rev: Revision) -> Result<RevlogEntry, RevlogError> { | ||||
|
r46175 | let index_entry = self | ||
.index | ||||
.get_entry(rev) | ||||
.ok_or(RevlogError::InvalidRevision)?; | ||||
|
r46097 | let start = index_entry.offset(); | ||
let end = start + index_entry.compressed_len(); | ||||
|
r46175 | let data = if self.index.is_inline() { | ||
self.index.data(start, end) | ||||
} else { | ||||
&self.data()[start..end] | ||||
}; | ||||
|
r46097 | let entry = RevlogEntry { | ||
rev, | ||||
|
r46175 | bytes: data, | ||
|
r46097 | compressed_len: index_entry.compressed_len(), | ||
uncompressed_len: index_entry.uncompressed_len(), | ||||
base_rev: if index_entry.base_revision() == rev { | ||||
None | ||||
} else { | ||||
Some(index_entry.base_revision()) | ||||
}, | ||||
}; | ||||
Ok(entry) | ||||
} | ||||
} | ||||
/// The revlog entry's bytes and the necessary informations to extract | ||||
/// the entry's data. | ||||
#[derive(Debug)] | ||||
pub struct RevlogEntry<'a> { | ||||
rev: Revision, | ||||
bytes: &'a [u8], | ||||
compressed_len: usize, | ||||
uncompressed_len: usize, | ||||
base_rev: Option<Revision>, | ||||
} | ||||
impl<'a> RevlogEntry<'a> { | ||||
/// Extract the data contained in the entry. | ||||
pub fn data(&self) -> Result<Cow<'_, [u8]>, RevlogError> { | ||||
if self.bytes.is_empty() { | ||||
return Ok(Cow::Borrowed(&[])); | ||||
} | ||||
match self.bytes[0] { | ||||
// Revision data is the entirety of the entry, including this | ||||
// header. | ||||
b'\0' => Ok(Cow::Borrowed(self.bytes)), | ||||
// Raw revision data follows. | ||||
b'u' => Ok(Cow::Borrowed(&self.bytes[1..])), | ||||
// zlib (RFC 1950) data. | ||||
|
r46169 | b'x' => Ok(Cow::Owned(self.uncompressed_zlib_data()?)), | ||
|
r46097 | // zstd data. | ||
|
r46169 | b'\x28' => Ok(Cow::Owned(self.uncompressed_zstd_data()?)), | ||
|
r47172 | // A proper new format should have had a repo/store requirement. | ||
_format_type => Err(RevlogError::corrupted()), | ||||
|
r46097 | } | ||
} | ||||
|
r46169 | fn uncompressed_zlib_data(&self) -> Result<Vec<u8>, RevlogError> { | ||
|
r46097 | let mut decoder = ZlibDecoder::new(self.bytes); | ||
if self.is_delta() { | ||||
let mut buf = Vec::with_capacity(self.compressed_len); | ||||
|
r46169 | decoder | ||
.read_to_end(&mut buf) | ||||
|
r47172 | .map_err(|_| RevlogError::corrupted())?; | ||
|
r46169 | Ok(buf) | ||
|
r46097 | } else { | ||
let mut buf = vec![0; self.uncompressed_len]; | ||||
|
r46169 | decoder | ||
.read_exact(&mut buf) | ||||
|
r47172 | .map_err(|_| RevlogError::corrupted())?; | ||
|
r46169 | Ok(buf) | ||
|
r46097 | } | ||
} | ||||
|
r46169 | fn uncompressed_zstd_data(&self) -> Result<Vec<u8>, RevlogError> { | ||
|
r46097 | if self.is_delta() { | ||
let mut buf = Vec::with_capacity(self.compressed_len); | ||||
zstd::stream::copy_decode(self.bytes, &mut buf) | ||||
|
r47172 | .map_err(|_| RevlogError::corrupted())?; | ||
|
r46169 | Ok(buf) | ||
|
r46097 | } else { | ||
let mut buf = vec![0; self.uncompressed_len]; | ||||
let len = zstd::block::decompress_to_buffer(self.bytes, &mut buf) | ||||
|
r47172 | .map_err(|_| RevlogError::corrupted())?; | ||
|
r46169 | if len != self.uncompressed_len { | ||
|
r47172 | Err(RevlogError::corrupted()) | ||
|
r46169 | } else { | ||
Ok(buf) | ||||
} | ||||
|
r46097 | } | ||
} | ||||
/// Tell if the entry is a snapshot or a delta | ||||
/// (influences on decompression). | ||||
fn is_delta(&self) -> bool { | ||||
self.base_rev.is_some() | ||||
} | ||||
} | ||||
/// Format version of the revlog. | ||||
pub fn get_version(index_bytes: &[u8]) -> u16 { | ||||
BigEndian::read_u16(&index_bytes[2..=3]) | ||||
} | ||||
|
r46102 | /// Calculate the hash of a revision given its data and its parents. | ||
|
r48171 | fn hash( | ||
data: &[u8], | ||||
p1_hash: &[u8], | ||||
p2_hash: &[u8], | ||||
) -> [u8; NODE_BYTES_LENGTH] { | ||||
|
r46102 | let mut hasher = Sha1::new(); | ||
let (a, b) = (p1_hash, p2_hash); | ||||
if a > b { | ||||
|
r48171 | hasher.update(b); | ||
hasher.update(a); | ||||
|
r46102 | } else { | ||
|
r48171 | hasher.update(a); | ||
hasher.update(b); | ||||
|
r46102 | } | ||
|
r48171 | hasher.update(data); | ||
*hasher.finalize().as_ref() | ||||
|
r46102 | } | ||
|
r46097 | #[cfg(test)] | ||
mod tests { | ||||
use super::*; | ||||
use super::super::index::IndexEntryBuilder; | ||||
#[test] | ||||
fn version_test() { | ||||
|
r46175 | let bytes = IndexEntryBuilder::new() | ||
.is_first(true) | ||||
|
r46097 | .with_version(1) | ||
|
r46175 | .build(); | ||
|
r46097 | |||
assert_eq!(get_version(&bytes), 1) | ||||
} | ||||
} | ||||