Revision control

Copy as Markdown

/* This Source Code Form is subject to the terms of the Mozilla Public
* License, v. 2.0. If a copy of the MPL was not distributed with this
* file, You can obtain one at http://mozilla.org/MPL/2.0/.
*/
pub mod address;
mod common;
pub mod credit_card;
pub mod engine;
pub(crate) use crate::db::models::Metadata;
use crate::error::Result;
use interrupt_support::Interruptee;
use rusqlite::Transaction;
use sync15::bso::{IncomingBso, IncomingContent, IncomingEnvelope, IncomingKind, OutgoingBso};
use sync15::ServerTimestamp;
use sync_guid::Guid;
use types::Timestamp;
// This type is used as a snazzy way to capture all unknown fields from the payload
// upon deserialization without having to work with a concrete type
type UnknownFields = serde_json::Map<String, serde_json::Value>;
// The fact that credit-card numbers are encrypted makes things a little tricky
// for sync in various ways - and one non-obvious way is that the tables that
// store sync payloads can't just store them directly as they are not encrypted
// in that form.
// ie, in the database, an address record's "payload" column looks like:
// > '{"entry":{"address-level1":"VIC", "street-address":"2/25 Somewhere St","timeCreated":1497567116554, "version":1},"id":"29ac67adae7d"}'
// or a tombstone: '{"deleted":true,"id":"6544992973e6"}'
// > (Note a number of fields have been removed from 'entry' for clarity)
// and in the database a credit-card's "payload" looks like:
// > 'eyJhbGciOiJkaXIiLCJlbmMiOiJBMjU2R0NNIn0..<snip>-<snip>.<snip lots more>'
// > while a tombstone here remains encrypted but has the 'deleted' entry after decryption.
// (Note also that the address entry, and the decrypted credit-card json both have an "id" in
// the JSON, but we ignore that when deserializing and will stop persisting that soon)
// Some traits that help us abstract away much of the sync functionality.
// A trait that abstracts the *storage* implementation of the specific record
// types, and must be implemented by the concrete record owners.
// Note that it doesn't assume a SQL database or anything concrete about the
// storage, although objects implementing this trait will live only long enough
// to perform the sync "incoming" steps - ie, a transaction is likely to live
// exactly as long as this object.
// XXX - *sob* - although each method has a `&Transaction` param, which in
// theory could be avoided if the concrete impls could keep the ref (ie, if
// it was held behind `self`), but markh failed to make this work due to
// lifetime woes.
pub trait ProcessIncomingRecordImpl {
type Record;
fn stage_incoming(
&self,
tx: &Transaction<'_>,
incoming: Vec<IncomingBso>,
signal: &dyn Interruptee,
) -> Result<()>;
/// Finish the incoming phase. This will typically caused staged records
// to be written to the mirror.
fn finish_incoming(&self, tx: &Transaction<'_>) -> Result<()>;
fn fetch_incoming_states(
&self,
tx: &Transaction<'_>,
) -> Result<Vec<IncomingState<Self::Record>>>;
/// Returns a local record that has the same values as the given incoming record (with the exception
/// of the `guid` values which should differ) that will be used as a local duplicate record for
/// syncing.
fn get_local_dupe(
&self,
tx: &Transaction<'_>,
incoming: &Self::Record,
) -> Result<Option<Self::Record>>;
fn update_local_record(
&self,
tx: &Transaction<'_>,
record: Self::Record,
was_merged: bool,
) -> Result<()>;
fn insert_local_record(&self, tx: &Transaction<'_>, record: Self::Record) -> Result<()>;
fn change_record_guid(
&self,
tx: &Transaction<'_>,
old_guid: &Guid,
new_guid: &Guid,
) -> Result<()>;
fn remove_record(&self, tx: &Transaction<'_>, guid: &Guid) -> Result<()>;
fn remove_tombstone(&self, tx: &Transaction<'_>, guid: &Guid) -> Result<()>;
}
pub trait ProcessOutgoingRecordImpl {
type Record;
fn fetch_outgoing_records(&self, tx: &Transaction<'_>) -> anyhow::Result<Vec<OutgoingBso>>;
fn finish_synced_items(
&self,
tx: &Transaction<'_>,
records_synced: Vec<Guid>,
) -> anyhow::Result<()>;
}
// A trait that abstracts the functionality in the record itself.
pub trait SyncRecord {
fn record_name() -> &'static str; // "addresses" or similar, for logging/debuging.
fn id(&self) -> &Guid;
fn metadata(&self) -> &Metadata;
fn metadata_mut(&mut self) -> &mut Metadata;
// Merge or fork multiple copies of the same record. The resulting record
// might have the same guid as the inputs, meaning it was truly merged, or
// a different guid, in which case it was forked due to conflicting changes.
fn merge(incoming: &Self, local: &Self, mirror: &Option<Self>) -> MergeResult<Self>
where
Self: Sized;
}
impl Metadata {
/// Merge the metadata from `other`, and possibly `mirror`, into `self`
/// (which must already have valid metadata).
/// Note that mirror being None is an edge-case and typically means first
/// sync since a "reset" (eg, disconnecting and reconnecting.
pub fn merge(&mut self, other: &Metadata, mirror: Option<&Metadata>) {
match mirror {
Some(m) => {
fn get_latest_time(t1: Timestamp, t2: Timestamp, t3: Timestamp) -> Timestamp {
std::cmp::max(t1, std::cmp::max(t2, t3))
}
fn get_earliest_time(t1: Timestamp, t2: Timestamp, t3: Timestamp) -> Timestamp {
std::cmp::min(t1, std::cmp::min(t2, t3))
}
self.time_created =
get_earliest_time(self.time_created, other.time_created, m.time_created);
self.time_last_used =
get_latest_time(self.time_last_used, other.time_last_used, m.time_last_used);
self.time_last_modified = get_latest_time(
self.time_last_modified,
other.time_last_modified,
m.time_last_modified,
);
self.times_used = m.times_used
+ std::cmp::max(other.times_used - m.times_used, 0)
+ std::cmp::max(self.times_used - m.times_used, 0);
}
None => {
fn get_latest_time(t1: Timestamp, t2: Timestamp) -> Timestamp {
std::cmp::max(t1, t2)
}
fn get_earliest_time(t1: Timestamp, t2: Timestamp) -> Timestamp {
std::cmp::min(t1, t2)
}
self.time_created = get_earliest_time(self.time_created, other.time_created);
self.time_last_used = get_latest_time(self.time_last_used, other.time_last_used);
self.time_last_modified =
get_latest_time(self.time_last_modified, other.time_last_modified);
// No mirror is an edge-case that almost certainly means the
// client was disconnected and this is the first sync after
// reconnection. So we can't really do a simple sum() of the
// times_used values as if the disconnection was recent, it will
// be double the expected value.
// So we just take the largest.
self.times_used = std::cmp::max(other.times_used, self.times_used);
}
}
}
}
// A local record can be in any of these 5 states.
#[derive(Debug)]
enum LocalRecordInfo<T> {
Unmodified { record: T },
Modified { record: T },
// encrypted data was scrubbed from the local record and needs to be resynced from the server
Scrubbed { record: T },
Tombstone { guid: Guid },
Missing,
}
// An enum for the return value from our "merge" function, which might either
// update the record, or might fork it.
#[derive(Debug)]
pub enum MergeResult<T> {
Merged { merged: T },
Forked { forked: T },
}
// This ties the 3 possible records together and is what we expect the
// implementations to put together for us.
#[derive(Debug)]
pub struct IncomingState<T> {
incoming: IncomingContent<T>,
local: LocalRecordInfo<T>,
// We don't have an enum for the mirror - an Option<> is fine because
// although we do store tombstones there, we ignore them when reconciling
// (ie, we ignore tombstones in the mirror)
// don't store tombstones there.
mirror: Option<T>,
}
/// The distinct incoming sync actions to be performed for incoming records.
#[derive(Debug, PartialEq)]
enum IncomingAction<T> {
// Remove the local record with this GUID.
DeleteLocalRecord { guid: Guid },
// Insert a new record.
Insert { record: T },
// Update an existing record. If `was_merged` was true, then the updated
// record isn't identical to the incoming one, so needs to be flagged as
// dirty.
Update { record: T, was_merged: bool },
// We forked a record because we couldn't merge it. `forked` will have
// a new guid, while `incoming` is the unmodified version of the incoming
// record which we need to apply.
Fork { forked: T, incoming: T },
// An existing record with old_guid needs to be replaced with this record.
UpdateLocalGuid { old_guid: Guid, record: T },
// There's a remote tombstone, but our copy of the record is dirty. The
// remote tombstone should be replaced with this.
ResurrectRemoteTombstone { record: T },
// There's a local tombstone - it should be removed and replaced with this.
ResurrectLocalTombstone { record: T },
// Nothing to do.
DoNothing,
}
/// Convert a IncomingState to an IncomingAction - this is where the "policy"
/// lives for when we resurrect, or merge etc.
fn plan_incoming<T: std::fmt::Debug + SyncRecord>(
rec_impl: &dyn ProcessIncomingRecordImpl<Record = T>,
tx: &Transaction<'_>,
staged_info: IncomingState<T>,
) -> Result<IncomingAction<T>> {
log::trace!("plan_incoming: {:?}", staged_info);
let IncomingState {
incoming,
local,
mirror,
} = staged_info;
let state = match incoming.kind {
IncomingKind::Tombstone => {
match local {
LocalRecordInfo::Unmodified { .. } | LocalRecordInfo::Scrubbed { .. } => {
// Note: On desktop, when there's a local record for an incoming tombstone, a local tombstone
// would created. But we don't actually need to create a local tombstone here. If we did it would
// immediately be deleted after being uploaded to the server.
IncomingAction::DeleteLocalRecord {
guid: incoming.envelope.id,
}
}
LocalRecordInfo::Modified { record } => {
// Incoming tombstone with local changes should cause us to "resurrect" the local.
// At a minimum, the implementation will need to ensure the record is marked as
// dirty so it's uploaded, overwriting the server's tombstone.
IncomingAction::ResurrectRemoteTombstone { record }
}
LocalRecordInfo::Tombstone {
guid: tombstone_guid,
} => {
assert_eq!(incoming.envelope.id, tombstone_guid);
IncomingAction::DoNothing
}
LocalRecordInfo::Missing => IncomingAction::DoNothing,
}
}
IncomingKind::Content(mut incoming_record) => {
match local {
LocalRecordInfo::Unmodified {
record: local_record,
}
| LocalRecordInfo::Scrubbed {
record: local_record,
} => {
// The local record was either unmodified, or scrubbed of its encrypted data.
// Either way we want to:
// - Merge the metadata
// - Update the local record using data from the server
// - Don't flag the local item as dirty. We don't want to reupload for just
// metadata changes.
let metadata = incoming_record.metadata_mut();
metadata.merge(
local_record.metadata(),
mirror.as_ref().map(|m| m.metadata()),
);
// a micro-optimization here would be to `::DoNothing` if
// the metadata was actually identical and the local data wasn't scrubbed, but
// this seems like an edge-case on an edge-case?
IncomingAction::Update {
record: incoming_record,
was_merged: false,
}
}
LocalRecordInfo::Modified {
record: local_record,
} => {
match SyncRecord::merge(&incoming_record, &local_record, &mirror) {
MergeResult::Merged { merged } => {
// The record we save locally has material differences
// from the incoming one, so we are going to need to
// reupload it.
IncomingAction::Update {
record: merged,
was_merged: true,
}
}
MergeResult::Forked { forked } => IncomingAction::Fork {
forked,
incoming: incoming_record,
},
}
}
LocalRecordInfo::Tombstone { .. } => IncomingAction::ResurrectLocalTombstone {
record: incoming_record,
},
LocalRecordInfo::Missing => {
match rec_impl.get_local_dupe(tx, &incoming_record)? {
None => IncomingAction::Insert {
record: incoming_record,
},
Some(local_dupe) => {
// local record is missing but we found a dupe - so
// the dupe must have a different guid (or we wouldn't
// consider the local record missing!)
assert_ne!(incoming_record.id(), local_dupe.id());
// The existing item is identical except for the metadata, so
// we still merge that metadata.
let metadata = incoming_record.metadata_mut();
metadata.merge(
local_dupe.metadata(),
mirror.as_ref().map(|m| m.metadata()),
);
IncomingAction::UpdateLocalGuid {
old_guid: local_dupe.id().clone(),
record: incoming_record,
}
}
}
}
}
}
IncomingKind::Malformed => {
log::warn!("skipping incoming record: {}", incoming.envelope.id);
IncomingAction::DoNothing
}
};
log::trace!("plan_incoming resulted in {:?}", state);
Ok(state)
}
/// Apply the incoming action
fn apply_incoming_action<T: std::fmt::Debug + SyncRecord>(
rec_impl: &dyn ProcessIncomingRecordImpl<Record = T>,
tx: &Transaction<'_>,
action: IncomingAction<T>,
) -> Result<()> {
log::trace!("applying action: {:?}", action);
match action {
IncomingAction::Update { record, was_merged } => {
rec_impl.update_local_record(tx, record, was_merged)?;
}
IncomingAction::Fork { forked, incoming } => {
// `forked` exists in the DB with the same guid as `incoming`, so fix that.
// change_record_guid will also update the mirror (if it exists) to prevent
// the server from overriding the forked mirror record (and losing any unknown fields)
rec_impl.change_record_guid(tx, incoming.id(), forked.id())?;
// `incoming` has the correct new guid.
rec_impl.insert_local_record(tx, incoming)?;
}
IncomingAction::Insert { record } => {
rec_impl.insert_local_record(tx, record)?;
}
IncomingAction::UpdateLocalGuid { old_guid, record } => {
// expect record to have the new guid.
assert_ne!(old_guid, *record.id());
rec_impl.change_record_guid(tx, &old_guid, record.id())?;
// the item is identical with the item with the new guid
// *except* for the metadata - so we still need to update, but
// don't need to treat the item as dirty.
rec_impl.update_local_record(tx, record, false)?;
}
IncomingAction::ResurrectLocalTombstone { record } => {
rec_impl.remove_tombstone(tx, record.id())?;
rec_impl.insert_local_record(tx, record)?;
}
IncomingAction::ResurrectRemoteTombstone { record } => {
// This is just "ensure local record dirty", which
// update_local_record conveniently does.
rec_impl.update_local_record(tx, record, true)?;
}
IncomingAction::DeleteLocalRecord { guid } => {
rec_impl.remove_record(tx, &guid)?;
}
IncomingAction::DoNothing => {}
}
Ok(())
}
// Helpers for tests
#[cfg(test)]
mod tests; // pull in our integration tests
// and a module for unit test utilities.
#[cfg(test)]
pub mod test {
use crate::db::{schema::create_empty_sync_temp_tables, test::new_mem_db, AutofillDb};
pub fn new_syncable_mem_db() -> AutofillDb {
let _ = env_logger::try_init();
let db = new_mem_db();
create_empty_sync_temp_tables(&db).expect("should work");
db
}
}